Conference Agenda

.

Copernicus is the European system for monitoring the Earth. It consists of a complex set of instruments which collect data from multiple sources: earth observation satellites and in situ sensors. It processes these data and provides users with reliable and up-to-date information through a set of operational services related to environmental and security issues. The Copernicus Land service supports a wide range of applications including environment protection, management of urban areas, regional and local planning, agriculture, forestry, and sustainable development. Under its global component, the service delivers biophysical variables including land cover – land use information at different resolution. The main users of Copernicus services are policymakers and public authorities who need the information to develop and to monitor environmental legislation and policies including development policies. The Copernicus programme is coordinated and managed by the European Commission.

As part of the European Copernicus programme, the Sentinel-1 mission, based on a constellation of two SAR satellites, ensures continuity for Europe of C-band SAR observations. Sentinel-1A and Sentinel-1B were respectively launched from Kourou on 3rd April 2014 and 25th April 2016.

Full Operations Capacity (FOC) of the mission is expected to be achieved following the completion of the operational qualification of the Sentinel-1 constellation (Sentinel-1A and Sentinel-1B), by mid-2017 indicatively.

The presentation will give an overview of the overall mission status at the time of the workshop and will focus on the routine operations activities of the constellation. Topics including mission achievements, mission observation scenario of the constellation, ground segment operations performance, throughput and data access will be presented.

Copernicus is a joint programme of the European Commission (EC) and the European Space Agency (ESA), designed to establish a European capacity for the provision and use of operational monitoring information for environment and security applications.

Within the Copernicus programme, ESA is responsible for the development of the Space Component, a fully operational space-based capability to supply earth-observation data to sustain environmental information Services in Europe.

The Sentinel missions are Copernicus dedicated Earth Observation missions composing the essential elements of the Space Component. In the global Copernicus framework, they are complemented by other satellites made available by third-parties or by ESA and coordinated in the synergistic system through the Copernicus Data-Access system versus the Copernicus Services.

The Copernicus Sentinel-2 mission provides continuity to services relying on multi-spectral high-resolution optical observations over global terrestrial surfaces [1]. Sentinel-2 will capitalize on the technology and the vast experience acquired in Europe and the US to sustain the operational supply of data for services such as forest monitoring, land cover changes detection or natural disasters management.

The Sentinel-2 mission offers an unprecedented combination of the following capabilities:

○ Systematic global coverage of land surfaces: from 56°South to 84°North, coastal waters and Mediterranean sea;

○ High revisit: every 5 days at equator under the same viewing conditions with 2 satellites;

○ High spatial resolution: 10m, 20m and 60m;

○ Multi-spectral information with 13 bands in the visible, near infra-red and short wave infra-red part of the spectrum;

○ Wide field of view: 290 km.

The data from the Sentinel-2 mission are available openly and freely for all users with online easy access since December 2015. The presentation will give a status report on the Sentinel-2 mission, and outlook for the remaining ramp-up Phase, the completion of the constellation and a view to ongoing evolutions.

The first satellite of the Sentinel-3 constellation, Sentinel-3A, was launched in February 2016, with a launch of Sentinel-3B being expected at the end of 2017. The main objectives of the Sentinel-3 constellation, building on the heritage of ESA’s ERS and ENVISAT missions, are to measure sea-surface topography, sea- and land-surface temperature and ocean- and land-surface colour in support of ocean forecasting systems, and for environmental and climate monitoring. The series of Sentinel-3 satellites will ensure global, frequent and near-real time ocean, ice and land monitoring, with the provision of observation data in a routine, long-term (up to 20 years of operations) and continuous fashion, with a consistent quality and a high level of reliability and availability.
Sentinel-3A passed its commissioning phase in July 2016 and is now in the so called ramp-up phase, leading to full operational capacity in spring 2017. The Sentinel-3 missions is jointly operated by ESA and EUMETSAT. ESA will be responsible for the data acquisition, satellite and payload long-term performance monitoring and the operations, maintenance and evolution of the Sentinel-3 ground segment on land related products and EUMETSAT on the marine products and the satellite monitoring and control.
The presentation will give an overview of the overall mission status at the time of the workshop and provide in particular an overview on the status of the Sentinel-3 core data products and their provision in the mission’s ramp-up phase.

The US Geological Survey has a long land cover history, starting with the 1976 landmark A Land Use and Land Cover Classification System for use with Remote Sensor Data, and including global land cover mapping and the ongoing production of the National Land Cover Database. While these past projects have had a significant impact, land cover data needs are changing due to the demand for increasingly innovative and timely land cover products needed to meet the community’s insatiable appetite for science quality geospatial land cover and land change data. New strategies are needed that generate higher quality results that include additional land cover variables, more detailed legends, and more frequent land cover and land change geospatial and statistical information.

The USGS response to the growing requirements for land use, cover, and condition data, information, and knowledge, is the Land Change Monitoring, Assessment, and Projection (LCMAP) initiative. LCMAP is an end-to-end capability that uses the rich Landsat record to continuously track and characterize changes in land cover, use, and condition and translate such information into assessments of current and historical processes of cover and change. LCMAP aims to generate science-quality land cover and land change products from current and near-real time Landsat data. All available Landsat data for any given location are used to characterize land cover and change at any point across the full Landsat record and to detect and characterize land cover and land change as it occurs. The basis for this is the Continuous Change Detection and Classification (CCDC) algorithm developed by Zhu and Woodcock (2014). The initial LCMAP land cover and land change products are annual land cover maps, maps of key change attributes (e.g., date, type, and magnitude of change), corresponding accuracy assessments, and land cover and land change area estimates. This presentation focuses on the overall LCMAP strategy and reviews early results that demonstrates how an improved land and change information lead to a better understanding of the rates, causes, and consequences of land change.

Zhu, Z. and Woodcock, C.E., 2014. Continuous change detection and classification of land cover using all available Landsat data. Remote sensing of Environment, 144: 152-171.

The NASA Land-Cover/Land-Use Change (LCLUC) program is developing interdisciplinary research combining aspects of physical, social and economic sciences, with a high level of societal relevance, using remote sensing tools, methods and data.One of its stated goals is todevelop the capability for periodic satellite-based inventories of land cover and monitoring and characterizing land-cover and land-use change. Synergistic use of data from different satellites is an efficient way to get most out of current remote sensing capabilities to study land changes. This presentation will focus on the recent achievements in multi-source land imaging projects under LCLUC program, in which optical mid-resolution data from Landsat system and Sentinel-2, as well as radar data from Sentinel-1, are combined to develop higher-level medium spatial resolution (10-60m) satellite products in analyzing changes in land cover at the global and continental scales. This presentation will include illustrations of recent, significant results in the LCLUC program for various sectors, including forestry, agriculture and urban.

The Group on Earth Observations (GEO) is the organization building the Global Earth Observation Systems of Systems ( GEOSS). The organization is comprised of over 100 member countries and over 100 Participating Organizations that are coordinating their efforts to improve the use of Earth observations for decision making. GEO has eight Societal Benefit Areas (SBAs) and more than 70 different activities that are filling in observational gaps, increasing data sharing, and enhancing discovery, access and application of data contained in the GEOSS Common Infrastructure (GCI).

Land cover is one of GEO’s top-priority areas due to its importance to many SBAs. It is critically important for ecosystems, biodiversity, water and disasters, and plays a major role in climate change processes. Land cover and land use information is required by a variety of the Multilateral Environmental Agreements (MEAs) such as the Sustainable Development Goals (SDGs), the Convention on Biological Diversity (UNCBD), the Convention to Combat Desertification (UNCCD), the Ramsar Convention on Wetlands, and several others. GEO is actively involved with the SDGs, focused on providing the Earth observations and derived products that can support achieving them and generating the indicators that mark progress. And at the national level, governments need land cover and land use information, particularly change information, to meet their many internal needs.

Contributors to the GEO Land Cover (LC) and Land Cover Change (LCC) Task work to improve the availability and quality of LC and LCC data by helping to convene and coordinate various sectors of the LC community, including data providers and consumers. Stakeholders include environmental agencies, science communities, national mapping agencies, commercial users, and MEAs. Activities include work to improve access to existing LC and LCC information, development of shared tools to facilitate validation of LC datasets, and working towards sustainable systems for LC product generation.

Because LC products are needed for so many users and applications, meeting these varied needs is a big challenge. Most current LC product generation approaches are labour intensive and unable to meet the needs of all users. Consequently, GEO is facilitating discussions within the LC community on how to move towards automated, sustainable, operational systems that can support a wider range of user needs.

Land cover is one of the most easily detected indicators of human interventions on the land. Information on land cover is therefore critical for the implementation of environmental, food security and humanitarian programmes of UN, international and national institutions.

The Food and Agriculture Organisation (FAO) has long experiences in developing land cover data sets and in using space-technology-based data for: in situ monitoring, data collection, agriculture and environment monitoring and development of sustainable agriculture policy in the country members. During the last 3 decades FAO has devoted considerable attention to the development of techniques for land cover and land cover changes mapping using enhanced methodologies and tools underpinned by standards (LCML, LCCS).

FAO adopt two main approaches for producing estimates of land cover and land cover change: i) wall-to-wall approach, an analysis that covers the full spatial extent of the study area, derived from FAO GLCN programme; and ii) sample base approach derived from the methodology developed under the FAO’s EcoNet programme. Both methods take advantage of geospatial technology and FAO Land Cover Classification System/LCML language (ISO standard) for description of the land cover features by using a set of independent diagnostic criteria that allow correlation with existing classifications and legends. LCML provides a general framework of rules from which more exclusive conditions can be derived to create specific legends. The LCCS describes the “real world” according to specific needs of the end users, combining the LCML Elements, by means of a specific software, to form their own category (class).

The key application of FAO land cover mapping approach stand on the exploitation of the time series geospatial information for the assessment of land use and land cover change over time and generation of specific information in particular monitoring of changes on agriculture, forests, rural and urban communities. The assessment of the level of land with focus on crop changes in a long period of time can significantly support the analysing in the types, rates, and causes of change and the effect on the environment and climate.

The focus of this paper is to give an overview of the FAO commitment in developing state-of-the-art techniques and methods integrated with geospatial technology for the monitoring and analysing of the landscape evolution as well as the assessment of the impact of the land uses and land covers change in response to evolving economic, social, and biophysical conditions. Geospatial technology enables using the integrated assessment of biophysical and socioeconomic variables for a coherent, reliable science-based approach to address a number of related parameters that are key to food security and sustainable agricultural landscapes.

The latest, free or lower-cost generation of remote sensing data products (including thermal and radar imagery) has now achieved a level of spatial, temporal and spectral resolution that may be directly applied to tracking the location and performance of complex, fragmented, low-input smallholder farming systems and landscapes in various environments.

Land Cover (LC) and its bio-geo-physical feedbacks are important for the understanding of climate and its vulnerability to changes on the surface of the Earth. Recently ESA has published a new LC map derived by combining remotely sensed surface reflectance and ground-truth observations. For each grid-box at 300m resolution, an estimate of confidence is provided. This LC data set can be used in climate modelling to derive land surface boundary parameters for the respective Land Surface Model (LSM). However, the ESA LC classes are not directly suitable for LSMs, therefore they need to be converted into the model specific surface presentations. Due to different design and processes implemented in various climate models they might differ in the treatment of artificial, water bodies, ice, bare or vegetated surfaces. Nevertheless, usually vegetation distribution in models is presented by means of plant functional types (PFT), which is a classification system used to simplify vegetation representation and group different vegetation types according to their biophysical characteristics. The method of LC conversion into PFT is also called “cross-walking” (CW) procedure. The CW procedure is another source of uncertainty, since it depends on model design and processes implemented and resolved by LSMs. These two sources of uncertainty, (i) due to surface reflectance conversion into LC classes, (ii) due to CW procedure, have been studied by Hartley et al 2016 to investigate their impact on LSM state variables (albedo, evapotranspiration (ET) and primary productivity) by using three standalone LSMs. The present study is a follow up to that work and aims at quantifying the impact of these two uncertainties on climate simulations performed with the Max Planck Institute for Meteorology Earth System Model (MPI-ESM) using prescribed sea surface temperature and sea ice. The main focus is on the terrestrial water cycle, but the impacts on surface albedo, wind patterns, 2m temperatures, as well as plants productivity are also examined.

The analysis of vegetation covered area indicates that the range of uncertainty might be about the same order of magnitude as the estimated historical anthropogenic LC change. For example, the area covered with managed grasses (crops and pasture in MPI-ESM PFT classification) varies from 17 to 26 million km², and area covered with trees ranges from 15 million km² up to 51 million km². These uncertainties in vegetation distribution lead to noticeable variations in atmospheric temperature, humidity, cloud cover, circulation, and precipitation as well as local, regional and global climate forcing. For example, the amount of terrestrial ET ranges from 73 to 77 × 10³ km³yr^-1in MPI-ESM simulations and this range has about the same order of magnitude as the current estimate of the reduction of annual ET due to recent anthropogenic LC change. This and more impacts of LC uncertainty on the near surface climate will be presented and discussed in the context of LC change.

Hartley, A.J., MacBean, N., Georgievski, G., Bontemps, S.: Uncertainty in plant functional type distributions and its impact on land surface models (in review with Remote Sensing of Environment Special Issue)

The spatial distribution and fractional cover of plant functional types (PFTs) is a key uncertainty in land surface models (LSMs) that is closely linked to uncertainties in global carbon, hydrology and energy budgets. In this study, we assess the largest plausible range of PFT uncertainty derived from land cover maps produced by the European Space Agency (ESA) Land Cover Climate Change Initiative (LC_CCI) on simulations of land surface fluxes using 3 leading LSMs. PFT maps used in LSMs can be derived from a land cover (LC) class map and cross-walking (CW) table that allocates the fraction of each PFT that occurs within each LC class. We evaluate the impact of uncertainty due to both LC classification algorithms, and CW procedure.

We examined the impact of this PFT uncertainty on 3 key variables in the carbon, water and energy cycles (gross primary production (GPP), evapo-transpiration (ET), and albedo), for 3 LSMs (JSBACH, JULES and ORCHIDEE) at global and regional scales. Results showed a greater uncertainty in PFT fraction due to CW as opposed to LC uncertainty, for all three variables. CW uncertainty in tree fraction was found to be particularly important in the northern boreal forests for simulated LSM albedo. Uncertainty in the balance between grass and bare soil fraction in arid parts of Africa, central Asia, and central Australia was also found to influence albedo and ET in all models.

These results show that inter-model uncertainty for key variables in LSMs can be reduced by more accurate representation of PFT distributions. Future efforts in land cover mapping should therefore be focused on reducing CW uncertainty through better understanding of the fractional cover of PFTs within a land cover class. Efforts to reduce LC uncertainty should particularly be focused on more accurate mapping of grass and bare soil fractions in arid areas. We suggest that both issues can be significantly improved through the integration of very high spatial resolution satellite observations with more frequent and thematically detailed medium resolution observations.

We should rethink our common Earth Observation paradigm that one cannot observe land use and land use change (as compared to monitoring land cover) from remote sensing given the new opportunities we have at hand. Observation density from the Sentinel constellation plus the long-term legacy of the Landsat system are major cornerstones of this development. Open data policies push the use of data with a spatial grain from 10m to 30m and observation densities of a few days, which allow assessing land changes with a focus on how the observed land is used.

The climate change context has been one of the driving forces in global, regional and national monitoring of land cover and land changes. Relatively clear and documented user requirements (i.e. ECVs, REDD+, climate modelling) have stimulated dedicated observation programs. We here use the experiences and progress in this arena to highlight that observations are becoming available that lead to much more data-driven evidence/analysis of land changes and dynamics. At the same time, a dialog on opportunities and limitations of Earth Observations with the land change and system science community is emerging that will lead to a novel focus on monitoring land use with highly automated procedures. We will present several case studies on how that can become possible.

The recently signed Paris Climate agreement will lead to more investments in assessing land use effects on climate and to mitigation activities related to the land use sector. However, it seems essential and possible to move beyond purely climate change focused progress when monitoring land change with novel Earth Observation data. An important point of departure was created with the Sustainable Development Goals (SDGs) and related indicators and monitoring needs. Next to SDG 13 on “Climate Action”, specific opportunities may therefore arise in support of SDG 2 “Zero hunger” (e.g. on sustainable agriculture) and of SDG 15 “Life on Land” with a focus on the sustainable use of terrestrial ecosystems.

Ongoing climate change, biodiversity collapse and food insecurity are largely caused by the global accelerating land degradation process resulting from destruction of natural ecosystems and inappropriate land management and agricultural practices. The need for urgent and bold action to avoid, reduce and reverse land degradation, thereby achieving land degradation neutrality (LDN), became a firmly established global political target with the adoption of the 2030 Agenda for Sustainable Development and Sustainable Development Goals target 15.3. In close conjunction with the other Rio Conventions, the United Nations Convention to Combat Desertification (UNCCD) provides the international framework for LDN monitoring and implementation.

Building on lessons learned through a pilot project conducted in 14 voluntary countries and on recent efforts by the UNCCD Science-Policy Interface to design a scientific conceptual framework for LDN, the UNCCD is currently supporting over 100 countries in operationalizing the LDN concept. The approach consists in setting voluntary national LDN baselines as well as targets and associated measures, removing legal and economic barriers to sustainable land management, reinserting restored land in sustainable production and monitoring progress based on a harmonized set of three measurable indicators.

The three indicators used as proxies of the ecosystem services that LDN is intended to deliver are 1) land cover, 2) land productivity and 3) soil organic carbon. Earth observations from space have proven their reliability to track over long periods land cover change and biomass activity. As many countries face difficulties to access this type of information, the UNCCD has established partnerships with the European Space Agency (ESA), the Joint Research Centre of the European Commission (JRC) and ISRIC - World Soil Information to provide all interested countries with national estimates derived from global datasets as default information for their national LDN target setting processes.

Our presentation will provide an overview of the indicator framework and associated data requirements at various scales. It will focus on the UNCCD experience in using the ESA Climate Change Initiative Global Land Cover product to provide default data on land cover and land cover changes occurred between 2000 and 2010, aggregated in 6 main categories (forest, shrub and grassland, cropland, wetland, artificial areas and bare land) and analyzed in conjunction with: i) NDVI-based land productivity dynamics data made available by the JRC; and ii) soil organic carbon estimates extracted from the ISRIC SoilGrids250m. The presentation will highlight the limitations faced in different ecological contexts (climate, topography, hydrology, geology) with different categories of land cover such as tree-based cropping systems, complex mosaic of small plots of crops and natural vegetation, wetlands and finally in identifying expansion of urban areas over croplands, that should be better addressed in the next generation of higher resolution Global Land Cover databases. It will also highlight the need to count on a 1990 epoch in the new release.

Existing multi-purpose global land cover maps are of limited use in applications relating to biodiversity conservation. The reasons are twofold:

• Most current global land cover products are not operational or are infrequently updated, preventing their use for monitoring status and trends in biodiversity.
• Land cover maps that are frequently updated lack the detail in thematic classes required for conservation purposes.

Despite these shortcomings, multiple international environmental conventions cite land cover change as a key source of information to inform policy. The potential use cases for an annually-updated, synoptic view of land cover change are vast; however, the scientific community has yet to produce operational land cover monitoring systems that meet the thematic, spatial and temporal requirements of policy makers for tracking progress on global biodiversity commitments. For example, land use change is one of the four main causes of species extinction, linked to habitat loss, fragmentation and degradation and is a key pressure on biodiversity that needs to be tracked systematically. Yet land uses such as fertilizing a particular land cover (crop) can be very challenging to discern from the multispectral signal.

For scientific end users land cover can be both a discrete layer to monitor habitat cover and habitat change, as well as an input layer for species distribution models, the derivation of higher-level indicators of land use (in combination with contextual information) as well as for up-scaling Essential Biodiversity Variables (EBVs) on ecosystem structure, such as height, and function, such as the functional diversity of terrestrial plant communities.

Meanwhile, great advances have been made in the field of forest cover monitoring. The advent of freely available, multi-decadal and high resolution satellite imagery has been a game-changer for monitoring global forest land cover change. High performance, cloud computing has made on-the-fly analysis of huge volumes of pixels feasible. The global community can now evaluate the extent of habitat loss for rare and endangered biodiversity, from tropical to boreal forests. For example, over one million visitors (1,500 daily) have logged onto the interactive online Global Forest Watch platform to view and analyse forest cover change datasets.

Yet many non-forested habitats harbour biodiversity of key importance for global conservation and are being lost at alarming rates that often outstrip the rate of forest loss, e.g. temperate grasslands of Eurasia, montane grasslands of Africa, and global wetlands. The progress of satellite remote sensing for mapping these land covers and their changes periodically and systematically lags behind that for forest cover. The 2000-2005-2010 ESA Climate Change Initiative (CCI) 300m land cover datasets shows the potential for wall to wall land cover mapping using a standardised classification system.

This presentation will further explore these issues and present examples of international biodiversity and ecosystem service policy targets which urgently require timely and robust land cover data in order to track progress towards their achievement. These requirements will be framed in the context of the candidate set of EBVs proposed by the Group on Earth Observation Biodiversity Observation Network (GEO BON).

Improvements in the availability of earth observation satellite data, as well as increasing computation power and decreasing storage costs have made consistent, repeatable, global-scale forest change monitoring at medium resolutions a reality. The results of these monitoring efforts are delivered to the general public through the free, online Global Forest Watch platform. Since the launch of the Global Forest Watch platform in 2014, over 1 million unique users have accessed the platform from every country in the world, including users from national government agencies, commodity buying companies, journalists, researchers, and local communities. Forest change information from remotely sensed data is no longer simply a tool used by scientists, but increasingly an independent input into day-to-day forest management decisions by non-scientists.

The experience of Global Forest Watch in providing remotely-sensed data to the non-science community in usable, interactive formats has yielded many insights on how to build, disseminate, and maintain operational land cover monitoring systems. This presentation will reflect upon lessons learned from our over one million users to inform principles and criteria for future operational land cover monitoring.

This presentation will:

1. Illuminate challenges in the use of remotely senses data by non-scientists (e.g. communicating accuracy, data limitations, and appropriate applications of data to a variety of end-users)
2. Suggest key principals and criteria for data generation, including accuracy, timeliness, transparency, repeatability, and sustainability
3. Identify current data gaps related to land cover monitoring that are most relevant to our audiences

We will provide evidence for these discussions by reflecting on the myriad ways our audiences are using GFW data and tools. Successful examples include law enforcement officials in Peru and Uganda using near-real time alerts to find and respond to illegal activities, palm oil traders in Indonesia identifying risk in their supply chains using GFW tools, and journalists raising awareness about ecologically-important forest under threat with forest change data and satellite imagery.

The success of operational forest monitoring to add value to decision-making has given rise to increasing demand for new and improved remotely sensed products. For example, law enforcement agencies would benefit from higher spatial and temporal resolution of near-real time change detection. International policy-makers are interested in better information on forest recovery and regrowth, and differentiation of primary, secondary, degraded, and planted forests. We have also seen increasing interest in global, annual data on land cover and land cover change beyond forests to improve land-use planning by governments, civil society, and the private sector, as well as to provide an independent benchmark for progress on ambitious forest- and land-related commitments, like the New York Declaration, the Sustainable Development Goals and the Aichi Targets. In addition, users from different disciplines all require data to be accurate, easy to understand, and reliably updated into the future. This presentation will emphasize the essential task of considering end-user requirements in order to build, maintain, and disseminate land cover monitoring systems for maximum impact on decision-making at local-to-global scales.

One of the OECD’s core functions is the production of internationally harmonised data, statistics and indicators. Earth observation and other geospatial data offer opportunities to develop new or improved indicators, particularly in the domain of the environment and green growth. Earth observation data is often a unique source of relevant information that is commensurable across countries and at multiple spatial scales, and thus provides opportunities to help fill the many information gaps that OECD countries face at the national and sub-national levels (especially when it comes to monitoring natural resources and environmental sinks) and assessing environmental risks (to humans, built property and economic activity). Importantly, EO data can be combined with socio-demographic and economic data, thereby improving the policy relevance of the indicators. The OECD seeks to build on the rising body of EO and other geospatial data and develop internationally harmonised indicator methodologies to respond to the growing demands for a more granular and more policy-relevant information base and better targeted policy advice.

Applications of Earth observation and other geospatial data have been gaining momentum in OECD's work on the environment and green growth. Examples drawing on on-going work using geospatial data will be presented, such as air pollution exposure, the related economic costs, and distributional aspects. A specific current priority relates to land cover monitoring, aiming to quantify the rate of conversions from natural and semi-natural land cover types, to more anthropogenic land cover associations in a comparable way across all member countries as a proxy indicator of pressures on biodiversity and ecosystems. The associated user requirements for the underlying data will be discussed. Other data needs to support future developments include those arising from demands to better measure the environmental dimension of quality of life, the resilience to environmental risks, and the availability of natural assets.

A formal requirement from the United Nations Framework Convention on Climate Change (UNFCCC) for developing countries to implement the Reducing Emissions from Deforestation and Degradation (REDD+) policy process are national forest monitoring systems (NFMS). African countries committed to this process require national forest and land use class definitions as well as mapping/monitoring systems adjusted to their national circumstances. This paper examines the history of developing these components into operational Earth Observation (EO) based monitoring systems in the past 8 years with the support of the European Space Agency (ESA) projects in the Congo Basin and southern Africa; the user requirements from specific countries will be presented for Cameroon and Malawi. The challenges of transferring the policy and user requirements to technical specifications for EO-products and addressing the variability in forest types and land use change assessment will be noted. The NFMS in these countries started with the baseline year 2010 and aims to provide a continuous monitoring of forest cover and forest cover changes from 2015 onwards using multi-sensoral and -temporal satellite data (Sentinel-1/2, Landsat 8), which are validated with VHR optical data. The advent of the Sentinel-2 data series enhanced dramatically the utilisation of dense time series of multi-temporal satellite imagery to resolve problems caused by phenology changes of forest canopies between the seasons. Furthermore, this data is also needed to monitor forest degradation which can be better detected by assessing forest canopy disturbances with high frequency time series. The data availability improvement also addresses the problem of cloud cover in the tropics. Based on Sentinel 2 data and integration of Landsat 8 imagery the automatic processing chain for the NFMS is comprised of geometric, radiometric and topographic pre-processing steps and an iterative classification procedure that includes a rule based correction system which yields to thematic accuracies above 85%. It was noted that especially for the dry forest biomes these high accuracy levels could not be achieved when using global datasets. Due to the data volume generated with the application of Sentinel for the near real time forest monitoring there is a necessity for cloud processing in the operational systems and this further enables the user community to be directly involved in different aspects of the processing chain. The paper will emphasise the value and merit of user-driven approaches to developing national NFMS and land use monitoring systems in terms of in-country capacity and ownership of the processes.

The need and value of environmental accounting is well recognised internationally. However the development of environmental accounts has exposed some important gaps in the available spatial land cover information. The same issues often also limit the usefulness of land cover classification data in other land and water management applications. This presentation emphasises some of the key innovation needs into the future. Specifically, the requirements for land, water, carbon and environmental condition accounting will be discussed using Australian and global examples. Formal land accounting is currently complicated by the well-known confusion between land cover, land use and land tenure in available spatial data products. Water accounting specifically requires dynamic information on water bodies and irrigated crops, as well as delineation of other hydrological landscape elements such as floodplains, irrigable land and impermeable surfaces. Carbon accounting will require new land cover classification approaches that relate more closely to carbon stocks than to ecological communities. Finally, environmental condition accounting is a conceptually complex challenge but in essence requires new approaches to distinguish man-made from pre-existing land cover types with an indication of the degree of disturbance. In all cases, accounting demands that mapping occurs routinely on a regular (typically annual) basis using a consistent and transparent methodology. New technologies are rapidly changing the way in which land cover information is derived: data processing facilities such as Google Earth Engine empower a large community to develop bespoke land cover products, whereas new sensors (e.g. Sentinel 2) and sensor combinations relax the traditional trade-off between spatial and temporal resolution, supporting new classification approaches that simultaneously consider the spatial, temporal and spectral dimensions.

Existing global inventories of forest carbon stocks are up to debate because regionally strongly divergent. Inventory-based inference generally allows for estimates of carbon stocks at national or sub-national scales with low uncertainty in countries with established national forest inventories. The scarcity of such information across large forest areas, however, advises the use of spaceborne remote sensing imagery for obtaining wall-to-wall, and spatially explicit, carbon stock information. However, spaceborne measurements from optical or radar sensors are only indirectly related to the carbon variable of interest. Spaceborne radar has found limited use in global forest mapping applications so far, despite the availability of global observations from, by now, several missions and the proven sensitivity of, in particular, long wavelength radar backscatter observations to forest variables closely related to forest carbon stocks, e.g., the growing stock volume (GSV) or aboveground biomass (AGB). Large-scale applications of radar data face a number of specific challenges, such as the pronounced sensitivity of the radar measurements to changing environmental imaging conditions and forest structural differences altering the relationship of SAR backscatter observations to the forest biophysical variable of interest.

In this paper, we discuss options for using spaceborne radar data jointly with optical and lidar data, auxiliary datasets from forest inventories, climatological variables and ecosystem classifications, to map forest aboveground biomass globally, while minimizing the reliance on inventory data. A first set of global biomass maps as well as spatially explicit depictions of the associated uncertainties will be presented that have been produced from hyper-temporal stacks of ENVISAT ASAR C-band backscatter data (@1km resolution) and ALOS PALSAR L-band mosaics released by JAXA (@25m resolution). The maps are being produced in the frame of the ESA DUE GlobBiomass project. The spatially explicit datasets of forest aboveground biomass and carbon stocks are the first of its kind, obtained with a single, globally consistent, retrieval approach that allows for local tuning to account for the spatial variability of forest structure. We present current results from our investigations on the reliability of the estimates and compare with existing regional datasets, including inventory data and derived regional statistics and existing national or regional map products. While the systematic global assessment of the accuracy of the biomass estimates has not yet started, a few preliminary indications can be provided. The spatial distribution of aboveground biomass was well captured, with the largest values found in the tropical forests, in temperate forests of the US Pacific Northwest, Chile and South Australia. Carbon stocks of the northern hemisphere are in line with existing observations from on ground surveys. In the tropics, the estimates appear to be in agreement with previous mapping activities except for Southeast Asia where we are currently estimating less biomass. A systematic validation to be preformed in the coming year will help to identify strengths and limitations of carbon stock inventories relying on currently available global SAR datasets and will allow for further optimization of the retrieval approaches regionally.

Up-to-date information on land cover and how it is changing is required by many Sustainable Development Goals and other Multilateral Environmental Agreements. National governments need this information to meet their commitments to these agreements and for their internal regulations and applications. Various assessment bodies and other entities also have important needs. However, because most current approaches to generating land cover products are labor-intensive they have difficulty meeting the varied needs of these users. This results in a variety of important limitations that leave many user needs unmet. These limitations include but are not limited to: fixed number and types of classes; difficulty in generating products for large areas; infrequent and irregular updates; and long latency periods so the product may be out of date by the time it is available. Additionally, because products are generated by a variety of organizations with different mandates they are often inconsistent, making it difficult or impossible to combine or compare products. A better approach is needed.

Fortunately, advances in both science and technology now enable approaches that do not have these limitations. Specifically, improved algorithms that utilize multi-temporal and ancillary information are now practical, and increased data availability and decreased computing costs, among other advances, enable automated, on-demand systems that accept inputs from users to meet their specific needs. Several systems that take advantage of these advances and that support on-demand requests are already being developed. Developing on-demand land cover product generation systems has a variety of significant challenges, particularly for very large or, especially, global areas; reference data for training and validation is probably the most significant challenge at all scales but there are others. These topics were the focus of a workshop held in May, 2016 focused on exploring concepts for a sustainable land cover generation approach that can meet the varied needs of users. The outcome of that workshop and follow-on discussions has led to a suggested, generic architecture for land cover generation; while there are many good variants a “data cube like” approach is a common theme. In this presentation we discuss this new approach, its challenges, and some key steps forward to help it become more widespread so the needs of users can be better met.

In the last thirty years remarkable urban sprawl took place in Rome, especially along the consular roads, causing significant environmental impacts, whereby natural surfaces were replaced by buildings, with the consequence of disappearance of vegetated areas, increased soil sealing and atmospheric pollution. Luckily land use and land cover mapping have been core applications of Earth observation from its inception to today, and so data spanning over several decades is now available. Mainly based on the continuation of our past study, carried out since the early 1980s at the European Space Agency, the present analysis is conducted to examine past and current effects of the urbanization process occurred over the large Roman urban system. Technology for acquiring image data at higher and higher spatial and temporal resolution, finer and finer detail, has dramatically improved over the last three decades. That's enabled us to better explore and improve measurement of changes in land use. Overlays of OPTICAL/RADAR satellite images, collected over a long time period, as required by the spatial and temporal analysis of the urban sprawl, were validated via Geographic Information System (GIS) techniques, in a particular procedure applied to urban land/agricultural transformations. In particulare, the use of Copernicus Sentinel-2A imagery has improved the previous results on urban processes, by reducing the uncertainty of the discrimination of land cover classes and facilitating the photo-interpretation. Taking advantage of the recently launched radar missions Sentinel-1A and B, as well as of the available huge radar archive of the previous ESA missions such as ERS-2 and ENVISAT, Synthetic Aperture Radar (SAR) image data have been concurrently processed and integrated into the GIS multi-relational database thus allowing an easier discrimination of the urban landscapes, especially in a dense and heterogeneous city such as Rome. SAR imagery allowed detecting urban features in a complementary way with respect to the optical one and, the limitations of SAR and optical sensors for urban areas discrimination has been adequately compensated by appropriately combining the information extracted from each sensor. Statistical analysis was performed via the Urban Area Profile (UAP) index in order to quantify the sprawl phenomenon, by defining several landscape metrics: urban infill developments are expected to increase, as saturation trend of lower density areas has been spotted spreading over the municipality.

Soil erosion is one of the major causes of loss of soil and water resources. The sustainability of agriculture productivity and water resources are affected by soil and siltation activities. This study estimated the spatial soil erosion hazard potential and amount of soil lost in and around Greater Kinshasa. The study also quantified the amount of soil lost within different buffer zones from the main rivers within the study area and provided the appropriates measures that can be used by the policy makers for environment and water resources protection. The Water Observation and Information System (WOIS) 3.0 software was used to solve the Wischmeier and Smith (1960; 1978) universal soil loss equation (USLE) which is used land use, soil, slope, and rainfall products as mains inputs in the estimation of soil loss and erosion hazard. It was found that most of the south part of the urban area were prone to erosion. From the total area of Kinshasa (996 500 ha), 25 013ha (2.3%) is of very high (> 15tonnes/ha/year) risk of soil erosion. The Urban area consists of 4.3% of the area with very high (>15tonnes/ha/year) risk of soil erosion compared to a very high risk of 2.3% (>15tonnes/ha/year) in the rural area. The municipalities of Mont Ngafula, Ngaliema, Kisenso, Selembao are the most affected by risk to erosion. The rivers beds were also affected as the area under 1000 m from the Rivers (437046ha) the area with very high risk was of 12395 ha (2.8 %) (>15tonnes/ha/year). The most affected was the buffer region of 100 m with of 4.2% of very high risk (>15tonnes/ha/year). The study shows that the soil loss was mostly driven by the slope, elevation, and informal settlements, this because of the high percentage of soil losses where there were high slope and high percentage of soil losses where the informal settlements are located at the high elevated area in both the rural and urban areas. The study realised that the highest discharges of soil in rivers may affect the river beds, river streams, and aquatic biodiversity. The study provided the appropriate measures for environmental and water resources protections.

The main objective of AFORENSA is to generate relevant knowledge how the forest ecosystems in Croatia respond to observed extreme climatic variations, and what are possible future expectations on the progression of climate change and disturbances of the natural hydrologic cycle with the intensification of drought frequency and severity. This objective is aimed to be resolved by implementing the novel forest ecosystem assessment approach, applicable for the country scale, which incorporates: a) traditional in situ ecosystem survey (soil and vegetation), b) analysis of the functional state of the ecosystems by remote sensing, c) downscaling of the regional climatic models and d) coupling of scenario outputs with ecohydrological models on a site-specific environmental settings. The core system on which AFORENSA activities will be build up, presents “Dynamical Geoinformation system of forest ecosystems in Croatia” (DYN-CROFOR), a technological project financed by the Ministry of Science and Education from 2003 to 2007. Aforensa is aiming at the inventory, and integration of terrestrial forest ecosystem research on the Web-Gis (DYN-CROFOR) platform incorporating automatized procedures for retrieving remote sensing data (vegetation indices) together with data mining possibilities in R, a language and environment for statistical computing and graphics. Project workflow consists of four, interrelated, work packages: a) INVENTORY, b) AUTOMATIZATION, c) MINING, d) UPGRADING. AFORENSA will contribute with speeding-up of future exploratory forest ecosystem research, and with its capability of near real-time data processing will present a basis for the development of “Forest Watch”; an observatory for early determination of potentially harmful effects of Climate Change on forest ecosystem services such as biodiversity and water.

Land cover change is the result of complex interactions between social and environmental systems, systems that evolve over time. While climate and biophysical phenomena have long been the main drivers of changes in land surfaces, the human is now behind most of the changes affecting terrestrial ecosystems. The major observation now is that of a decrease in area of forest ecosystem, which is the consequence of the degradation sometimes extreme. This situation due to the phenomenon of deforestation in this region has caused degradation of biodiversity, including wildlife and flora, threatening to disappearance of natural resources hard renewable.

Thus, the last thirty years, there is a real dynamic change of land with intensive degradation of the natural vegetation cover especially in arid and semi-arid zone.

Indeed, the adverse effects of drought periods from the year 1970 combined with population growth and economic conditions experienced by the country in the

1990s have greatly upset the delicate balance of the natural environment. These adverse effects may result in partial or total disappearance of some natural ecosystems. However, the location of the most significant different changing sectors in space and time, allows specialists planning and local leaders understand these spatial changes that affect natural ecosystems in Algeria. In addition, there are relatively few studies using a long time series of Landsat data to determine land cover changes at local scale in Algeria.

The objective of this work is to study the degradation of land cover through a diachronic study. We used Support Vector Machines method for classification of Landsat data, and change detection technique to analyze change of land cover and land use from 1984 to 2014.

Our analysis showed that proportion of forest cover decreased from 41% in 1984 to 14% in 2014 that from approximately 190 hectares/year and agriculture land from 18 % to 1.5 %. The results showed that all land cover and lad use area have experienced structural changes in it's globally, Intensive regression of woody natural vegetation imposed by fires and unsustainable use of resources, a remarkable decline in land occupied by agriculture. Suggesting an immediate response to a policy based on priorities for the preservation, protection, development and rational use of land areas.

Land use land cover mapping and area estimation are two of the most fundamental application of remote sensing. Even though they are complementary, they respond to different needs with specific accuracy requirements. Compared to land cover mapping, area estimation has a more direct economic impact and responds to accuracy requirements clearly defined by the user community. The most elementary approach to estimate areas is to use pixel counting as estimator. Such an approach seemed to have been accepted in the early stage of remote sensing and is still widely reported. The accuracy of area estimates derived from these maps is known to be related to that of the maps: even with very accurate maps, errors in area estimates may occur. The pixel counting estimator is known to be biased because there is no guarantee that the omission error and the commission error will counterbalance each other. This paper proposes an end-to-end approach to define the classifier accuracy requirements for crop area estimation by simple pixel counting to be within the confidence interval of traditional area estimates. Basically, it seeks to define the applicability domain of pixel counting can be defined. The case study described within this study underlined that the approach is consistently applicable over a large territory such as South Africa. This framework can be used both to guide users in choosing the appropriate imagery to enhance the quality of their area estimates and as a diagnostic tool to evaluate the adequacy of existing satellite-based area estimation systems in different agrosystems. While explored in the context of estimation of cultivated areas, the findings presented here are generic to the problem of area estimation.

REDD+, which stands for 'Reducing Emissions from Deforestation and Forest Degradation in Developing Countries' - is a climate mitigation effort and aims to create a financial value for the carbon stored in forests, offering incentives for developing countries to reduce emissions from forested lands and invest in low-carbon paths to sustainable development.

The UN-REDD Programme, a collaborative partnership between FAO, UNDP and UNEP launched in September 2008, supports developing countries to develop capacity to REDD+ and to implement a future REDD mechanism in a post- 2012 climate regime.

The programme works at both the national and global scale, through support mechanisms for country-driven REDD+ strategies and international consensus-building on REDD processes.

The UN-REDD Programme gathers technical teams from around the world to develop common approaches, analyses and guidelines on issues such as measurement, reporting and verification (MRV) of carbon emissions and flows, remote sensing, and greenhouse gas inventories. Within the partnership, FAO supports countries on technical issues related to forestry and the development of cost effective and credible MRV processes for emission reductions. While at the international level, it fosters improved guidance on MRV approaches, including consensus on principles and guidelines for MRV and training programmes. It provides guidance on how best to design and implement REDD+, to ensure that forests continue to provide multiple benefits for livelihoods and biodiversity to societies while storing carbon at the same time. Other areas of work include national forest assessments and monitoring of in-country policy and institutional change. The outcomes about the role of satellite remote sensing technologies as a tool for monitoring, assessment, reporting and verification of carbon credits and co-benefits under the REDD+ mechanism are here presented.

The TanDEM-X mission (TerraSAR-X add-on for Digital Elevation Measurements), developed in public-private partnership between the German Aerospace Center (DLR) and Airbus Defence and Space, comprises two twin SAR satellites TerraSAR-X and TanDEM-X. They have been flying in a close orbit formation since 2010, with the main objective of generating a global and consistent digital elevation model (DEM) with unprecedented accuracy.

Moreover, the mission represents a highly valuable source for many scientific applications, among them land classification. In particular, the identification and monitoring of vegetated areas plays a key role in a large variety of different fields, such as agriculture, cartography, geology, forestry, global change research, and regional planning.
In this paper we present our activities for generating a global forest/non-forest map starting from TanDEM-X interferometric SAR data, operationally acquired by the TerraSAR-X and TanDEM-X satellites. The principle is to exploit the decorrelation contribution due to volume scattering, which results from the penetration of the radar wave within the foliage. This effect is highly related to the
presence of vegetation and depends on several parameters, such as the forest vertical profile and its density, the sensor frequency, and the viewing geometry. It is named volume decorrelation and can be estimated from the total interferometric coherence. A weighted clustering approach based on fuzzy logic is utilized for partitioning each pixel into two classes: forest or non-forest, by associating a membership value to it, which expresses the weighted probability of an observation to belong to each single class. The global input data set consists of quicklooks images of the total interferometric coherence, characterized by a ground resolution of 50 m x 50 m and generated as by-pass products from the full resolution ones, acquired for the generation of the global DEM. Furthermore, several subsequent overlapping coverages are globally available and have to be properly combined together to generate consistent large scale maps. The method for the mosaicking of overlapping scenes will also be presented. It is based on a weighted average of multiple membership values which takes into account indicators of their reliability, such as the dependency of volume decorrelation on the height of ambiguity, on the signal-to-noise ratio (SNR), and on the acquisition geometry. The delivered product will be characterized by a resolution of 50 m 50 m and global coverage. Full-resolution products (12 m x 12 m) will be initially used on a local scale only, in order to investigate the potentials of an increased resolution of classification purposes. The validation and performance measurement approach using ground truth maps and independent sources will be described and preliminary results will be discussed as well.

Finally, if input data covering a certain time span are available, the developed method can be used to detect temporal changes in the vegetation coverage. To conclude, we will present some examples for detecting on-going deforestation in the Amazon rain forest.

The suitability of Copernicus Global Land Service products for wind assessment is investigated using two approaches. In the first approach the CORINE land cover database and the pan-European high-resolution products were considered as input to atmospheric flow models. The CORINE data were used as input for modelling the wind conditions over a Danish near-coastal region. The flow model results were compared to alternative use of USGS land cover. Significant variations in the wind speed were found between the two atmospheric flow model results. Furthermore the wind speed from the flow model was compared to meteorological observations taken in a tall mast and from ground based remote-sensing wind profiling lidars. It is shown that simulations using CORINE provide better wind flow results close to the surface as compared to those using USGS on the investigated site.

The next step towards improvement of flow model inputs is to investigate in further detail applied use of satellite maps in forested areas. 75% of new land-based wind farms are planned in or near forests in Europe. In forested areas the near surface atmospheric flow is more challenging to calculate than in regions with low vegetation because the tall vegetation to a high degree influences the atmospheric flow. Also in many forests the variation in forest plant structure is high. The forest structure depends on the tree height, the tree density, the existence of clearings, the types of leafs and branches and their structure. So the method of assigning one typical roughness length for land cover type ‘forest’ is at many sites not sufficient. This method assumes that all land cover classes can be represented with one value each.

In our second approach, we look at a forested area in Northern Denmark, where an aerial lidar data observing terrain height, tree height and derived plant parameters provided a novel input for atmospheric flow modelling in forested areas. The flow model results were compared to horizontally scanning wind lidar observations and the results are very promising. Since, aerial lidar data are not available everywhere, we discuss the possibility of using similar Copernicus Global Land Service products as input to the flow model.

Acknowledgements: RUNE, NEWA.

Wetlands are one of the fastest declining ecosystem types worldwide, while at the same time they are hot spots of biodiversity and provide diverse and valuable ecosystem services; such as water supply, hydrological buffering against floods and droughts, and climate regulation through carbon storage. Information on wetlands extent, its ecological character and their services is often scattered, underestimated and difficult to find and access, which leads to the fact that wetlands are only partially covered worldwide by policies and management practices.

In this respect, SWOS (a Horizon-2020 project funded by the European Union) provides monitoring tools and information on wetland ecosystems, mainly derived from Earth Observation data.

The most important objective of SWOS is to prepare and install a service with users for users. Several user organizations are represented by the SWOS Consortium, and further users are identified at the global, regional, national and local level in order to support a multi-level user approach.

Because users are involved in the process from start to finish, the service will be widely used and accepted, harmonized with related activities and will have a long term impact on wetland management.

Service cases will demonstrate opportunities for improved wetland management, planning and decision making.

SWOS is using first of all free available satellite data from ESA’s Sentinel satellites and Landsat archives, and build on results of ESAs Globwetland projects. Near real-time observations via the SWOS service portal allow for dynamic monitoring of wetland status and changes (and the drivers of those changes) on a large spatial and temporal scale. The service will be demonstrated on a selection of wetlands in Europe, but also in Africa and Asia.

The SWOS service portal is a "knowledge hub", so that users can have access to a wide range of information for each wetland, not only the spatial mapping products, but also scientific papers, links to external webpages, documents, images, etc.. The portal provides in addition the on-demand processing of indicators or information based on the mapping products. Part of the SWOS Portal will be a mobile application as a citizen science tool for a wide range of people interested in wetlands and their dynamics. The app can be used in the field to get access to maps and information or to collect information about wetland delineation or the phenological status of the vegetation.

This presentation provides an overview of the SWOS approach and, gives insights into how the project is engaging users through the establishment of user groups and development of service cases. It will introduce the different service lines and products of SWOS, introduce the portal and mapping software and present monitoring results.

(for more information swos-service.eu)

Copernicus is a European Earth Observation (EO) programme headed by the European Commission (EC) in partnership with the European Space Agency (ESA) for the better understanding of the state and changes of our planet. Copernicus provides six operational services on the earth’s main sub-systems (i.e. Land, Atmosphere, Oceans) and on cross-cutting processes (i.e., Climate Change, Emergency and Security). These services are largely based on EO satellite data, which will rely on the fleet of ESA’s Sentinel satellites. The Copernicus Land Monitoring Service provides EO-based spatial information related to bio-geophysical variables and Land Cover/Land Use (LC/LU) characteristics as well as their changes over time. The related services are reflected in a Global, pan-European (Continental), Local Component, and an In-situ Component.

The Horizon2020 project, “Evolution of Copernicus Land Services based on Sentinel data” (ECoLaSS), which will be implemented from 2017-2019 aims at developing innovative methods, algorithms and prototypes to improve and invent future pre-operational Copernicus Land services from 2020 onwards, for the pan-European and Global Components. ECoLaSS will make full use of dense Sentinel time series of optical (S-2, S-3) and Synthetic Aperture Radar (SAR) data (S-1). Rapidly evolving scientific as well as user requirements will be analyzed in support of a future pan-European roll-out of new/improved Copernicus Land Monitoring services, and the transfer to global applications. This paper will describe the ECoLaSS concept, explain the current status of Copernicus Land services, the envisaged methods for their production and present first analyses and examples.

Service requirements assessment will be performed involving the main Copernicus Land stakeholders and users, and will thus steer methodological developments, such as: (i) Sentinel-1/-2/-3 time series integration, (ii) time series pre-processing methods, (iii) thematic classification and (iv) change detection from time series analysis, and (v) the development of an incremental update methodology for the Copernicus Land High Resolution Layers (HRLs). These methods will be applied on test sites, located both in Europe and Africa, prior to a prototyping phase. Larger demonstration sites representing various bio-geographic regions were selected to implement the following innovative prototypes: (i) indicators and variables from high spatial and temporal resolution data, for both the Continental and Global Component products; (ii) incremental update strategies for the main pan-European products (i.e. the HRLs); (iii) improved permanent grassland identification; (iv) crop area and crop status/parameters monitoring; (v) further novel LC/LU products. Finally, the main target to assess/benchmark all operational products in view of their innovation potential and technical excellence will be performed, leading to a strategy for an operationalization framework for a future pan-European roll-out of improved or newly developed Copernicus Land Monitoring services.

ECoLaSS will promote the innovation potential of new land monitoring services and applications to diverse user communities. The project will thus contribute to a growing “Copernicus Economy” by boosting (new) Copernicus Land Services and value-added applications (Downstream Services). It is expected that such new services will provide a variety of inter-linkages with other LC/LU projects, and bring new opportunities for a wide range of dedicated applications to the market from 2020 onwards.

Land cover and land use (LCLU) maps are extremely important for a lot of applied satellite monitoring problems. We have built high resolution land cover maps for the whole territory of Ukraine for three decades: 1990s, 2000s and 2010s [1]. For this, atmospherically corrected time-series of Landsat-4/5/7 images were classified using a neural network ensemble. These maps contain six main land cover classes of the European Land Use and Cover Area frame Survey (LUCAS) nomenclature: artificial surface, cropland, grassland, forest, bare land and water.

We consider three most common methods for reference data sampling design: pseudo-random sampling, systematic sampling on a regular 10 km grid and the approach on the base of segments. During the first approach, an expert selects samples that can be interpreted by him with minimal errors. In such a case, the accuracy of the map could be overestimated. Systematic sampling approach is more objective for reference data selection, but might be more difficult and resource consuming for photo-interpretation. Taking into account the impact of human subjectivity, two independent experts participated in reference data collecting within the second approach. This technique allows us to provide independent validation for land cover map and to compare it with the results based on random selection of reference samples. With the first pseudo-random sampling approach, the overall classification accuracy is approximately 95% for three different time periods (1990, 2000 and 2010). Within the second approach (regular grid), the overall accuracy of 84.5% was achieved [2]. We think this result is more objective due to regularity of grid and more independent selection of validation set. Third approach on the base of segments is the most difficult to realize because of a lot of so called “unknown” polygons which should be interpreted by expert with a low probability.

[1] M. Lavreniuk, N. Kussul, S. Skakun, A. Shelestov, B. Yailymov “Regional retrospective high resolution land cover for Ukraine: methodology and results,” IEEE International Geoscience and Remote Sensing Symposium (IGARSS), pp. 3965-3968, 2015.

[2] M. Lavreniuk, N. Kussul, A. Shelestov, B. Yailymov, T. Oliinyk, and A. Kosteckyi, “Validation methods for regional retrospective high resolution land cover for Ukraine,” IEEE International Geoscience and Remote Sensing Symposium (IGARSS), pp. 4502-4505, 2016.

Key words: land cover map, validation, sampling scheme.

Contour extraction methods are of fundamental importance in the context of mapping and updating for Geographic Information Systems applications. The International Society for Photogrammetry and Remote Sensing Commission VII/4 included Image classification techniques and new algorithms for the image extraction as a reference term. The motivation of this work is threefold: Firstly, the soybean is important to the economy of Brazil. Secondly, China-Brazil Earth Resources Satellite (CBERS) image allow rapid and efficient analyses of land cover, as well as it is possible to detect changes occurred during the period of image acquisition. Thirdly, and more important in the scope of our research, the task of contour extraction of soybean culture, due to scene complexity, requires the development of specific methods in the Remote Sensing image that permit to obtain the interest object. In this paper, a method for contour extraction of soybean culture from a CBERS-4 image is proposed. Our method is based on the recursive splitting technique using the quad tree structure consists of splitting the image into four homogeneous subregions of identical size. Each subregion is checked for homogeneity using a predefined threshold based on prior knowledge of objects presented in the scene. The splitting process proceeds recursively until no regions can be subdivided. Next, the resulting regions are classified using similarity criteria, in this case regions presenting high probability of similarity are merged. The algorithm for contour filling is applied to the regions. The soybeans contours are segmented using techniques such as, non-linear anisotropic diffusion via Partial Differential Equation that is used to previously focus the edge structure due to its notable characteristic in selectively smoothing the image, leaving the homogeneous regions strongly smoothed and mainly preserving the physical edges, i.e., those that are really related to objects presented on the image. The resulting regions are extracted by using techniques well-known, such as, vectorization, and polygonization. The preliminary results showed that the proposed methodology is promising for application involving extraction of cultures, because it has made possible the extraction of regions usually related to soybean culture.

A detailed and accurate knowledge of the land cover is crucial for many scientific and operational applications, and as such, it has been identified as an Essential Climate Variable. This accurate knowledge needs a frequent update of the information. For most applications, timeliness is more important than a detailed nomenclature. Timeliness can only be achieved with automatic methods which are robust and reliable.
The French Theia Land Data Centre has set up a Scientific Expertise Centre whose aim is to implement an operational fully automatic land cover map production system using mostly Sentinel-2 image time series. The map will be updated once a year and will contain 20 thematic classes mapped at 10 m. resolution. The system, which will be operational in 2017, is based on the iota2 processor, an open source framework based on Orfeo Toolbox for land cover map production developed by CESBIO.
iota2 uses all Sentinel-2 and Landsat-8 acquisitions available over the target region to feed a supervised classification system steered by an eco-climatic area stratification.
The reference data needed for supervised classification can present 2 main problems. The first one is the absence of reference data for one or several classes. For instance the crop grown in a field may change from one year to the following, and therefore, reference data older than one year is useless. However, obtaining fresh reference data over large areas may be difficult and can have high costs. The second problem are the errors in the reference data, which can have different origins. The databases can be much older than the remote sensing imagery and the land cover may have changed. This is different from the crop classes mentioned above, for which most of the changes occur within a limited number of classes. The images and the reference data can locally have inconsistent geometries. Finally, the databases can suffer from errors in the collection step. The strategies used to overcome the lack of reference data are different from those applied in the presence of errors.
The former is usually addressed by domain adaptation techniques. Although promising in the literature, they are difficult to scale up to systems mapping very large areas. We will present 2 pragmatic approaches which leverage historical reference data and imagery to map classes for which no reference data exists for the current mapping period. The first one is based on ensembles of classifiers and the second one is based on class modelling using clustering approaches. The latter has been implemented in the iota2 processor and used for the French national 2016 land cover map.
The errors in the reference data, depending on their nature and amount, can be addressed in different ways. For a limited amount of label noise, some classifiers are able to yield satisfactory performances. A comparison of supervised classification algorithms will be presented. In other cases, mislabelled training instances can be detected and even corrected using appropriate techniques in order to improve classifier accuracy/performance. Label noise detection methods will be outlined in the presentation.

Our goals were to evaluate consequences of climate warming on vegetation in central Siberia. We used our Siberian bioclimatic models: SiBCliM that simulates the zonobiome distribution and SiBCliMTree that simulates major forest-forming tree species distribution. Both models are of static envelope-type that predict zonobiomes and tree species from three bioclimatic indices characterizing warmth, cold, and moisture conditions: growing degree days, GDD₅, negative degree days, NDD₀, and annual moisture index, AMI. Additionally to climatic indices, both models included permafrost, a critical ecosystem determinant in Siberia occurring on 80% of Siberia. The permafrost border was substituted by the active layer depth (ALD) 2 m (R² = 0.7): ALD>2m explicitly allows all conifers to thrive, and ALD<2 m allows only one conifer Larix dahurica that can withstand ALD<2 m to grow.

Coupling our bioclimatic models with the climatic indices and the permafrost distributions we predicted potential distribution of vegetation zones and forest-forming tree species in current and the 2080s climates. Climatic anomalies by 2080s were derived from the two climate change ensemble scenarios of CMIP5: the rcp 2.6 and rcp 8.5 reflecting the smallest and the largest temperature increase correspondingly.

Siberian zonobiomes and tree species were simulated severely shifted northwards and forest-steppe and steppe would dominate 50% of central Siberia in the 2080s dryer climate. Light conifers (Larix spp. and Pinus sylvestris) may get an advantage before dark conifers (Pinus sibirica, Abies sibirica, and Picea obovata) in a predicted dry climate due to their resistance to water stress and wildfire. Fire and the thawing of permafrost are considered to be the principal mechanisms that will shape new vegetation physiognomies. These model predicted distributions of zonobiomes, major forest-forming tree species and permafrost need to be verified by current remote sensing observations.

Global scale land cover (LC) mapping has interested many researchers for last two decades as it is an input data source for various applications. However, current global land cover (GLC) maps often do not meet the accuracy and thematic requirements of multiple users. Along with the creation of new maps, current efforts for improving GLC maps focus on integrating existing maps. Such integration efforts may benefit from the use of multiple GLC reference datasets.

Using available reference datasets, we aimed to asses spatial accuracy of recent GLC maps and integrate GLC datasets for creating an improved map. We further attempted to address the thematic requirements of multiple users by demonstrating a concept of producing GLC maps with user-specific legends. Spatial correspondence with reference dataset was modelled for Globcover-2009, Land Cover-CCI-2010 v1.1, MODIS-2010 Collection-5 and Globeland30 maps for a continental scale, Africa. Five different methods that account for spatial variability in map accuracy and class probability were tested to integrate these GLC maps and available reference datasets. Based on cross-validation of these methods, a regression kriging method was selected to create an improved LC map of Africa. This method was further applied at global scale. Based on LC class probability maps produced from this integration, expected area fraction maps for LC classes at coarser resolution were created and used for characterizing additional mosaic classes that can be useful for users namely land system models and biodiversity assessments.

Comparison of the spatial correspondences showed that the preferences for GLC maps varied spatially and this supports the notion of integrating GLC maps based on their relative strengths such as spatial accuracy. An integrated GLC map was created and overall correspondence with reference LC was 80% based on 10-fold cross validation of 24681 sample sites. This was globally 10% and regionally 6-13% higher than the correspondence of the input GLC maps. Furthermore, two GLC maps with user-specific legends for land system models and biodiversity assessments were created using expected area fraction maps of LC classes.

Our results demonstrate the added value of using reference datasets and geostatistics for improving GLC maps. This finding further motivates the efforts of releasing available reference datasets to the public by international communities such as the GOFC-GOLD and Geo-Wiki portal. As more reference datasets are becoming available to the public, GLC mapping can be further improved by using the pool of all available reference datasets. Area fraction maps of LC classes make the translation to required user-specific legends namely mosaic classes easier and thus can address the thematic requirements of multiple users. Future efforts should focus on creating global land cover maps that take the requirements and perspectives GLC map users into account.

Environmental information is of crucial importance. It helps to understand how our planet and its climate are changing, the role human activities play in these changes and how these will influence our daily lives. To take the right actions, decision makers, businesses and citizens must be provided with reliable and up-to-date information on how our planet and its climate are changing. The European Earth monitoring program Copernicus has been set up to provide this information. Users are provided with information through services dedicated to a systematic monitoring and forecasting of the state of the Earth's subsystems. The scope of the Copernicus Global Land – Hot Spot Monitoring program is to provide detailed inventory of land cover and land cover change over identified areas of interest, with a change assessment frequency of 1 to 20 years, and based on high or very high resolution satellite data. Initial priority areas have been identified as Key Landscape for Conservation areas (KLC) in Africa comprising several protected areas and their surroundings, in support to the EU Wildlife Strategy to Africa and projects such as BIOPAMA among others. The production chain will be presented including image pre-processing, classification and collection of ancillary information followed by the systematic validation procedure and online distribution of the products. The classification scheme follows the Land Cover Classification System (LCCS) developed by FAO. The first series of mapped KLC’s in Africa cover a total area of about 380,000 km2. 150,000 km2 of this total area is mapped as dichotomous generic level LCCS legend (8 classes) and 230,000 km2 as detailed modular level of the LCCS legend (up to 30 classes). First results will be presented highlighting land cover changes and related indicators such as loss of natural vegetation and agriculture encroachment over selected KLC's.

Evapotranspiration over continents is a key process in the cooling of the Earth’surface and the hydrological processes in the soil and in the atmosphere. For many applications, evapotranspiration is seen as the process which leads to loss of soil water. Therefore, several environmental applications oriented towards agriculture and water management require mapping of evapotranspiration, sometimes in near-real time operations for fast response services.

At the Royal Meteorological Institute of Belgium, we develop and contribute to the implementation of such mapping services over entire continents, for EUMETSAT and soon for the European Copernicus program. In our algorithms, the knowledge of the type of land cover is crucial: it drives the selection of model and parameters to be applied. In operation, the algorithms make use of ECOCLIMAP, a land cover map specifically designed for land surface models. However, newest, less static maps with more spatial details do exist, although with different classifications that are sometimes less suited to our application. We have explored the possibility to use some of them, eg GlobCover 2005 & 2009 and MODIS LC.

In this contribution, we will discuss our ideal requirements for evapotranspiration mapping at different scales between 300 m and 5 km and the uncertainties arising from using existing land cover maps, by showing case studies.

The tropical forest is one of the more complex terrestrial ecosystems in the world. This biome presents a high level of biodiversity, which is a result of the great environmental variability providing different ecological niches and habitat preferences, a high complexity of relationships between biotic and abiotic factors and typical ecological processes. There are many challenges in characterising the floristic and structure of this kind of forest, such as forest complexity, forest heterogeneity and density, and forest stratification. The use of active remote sensing, especially Synthetic Aperture Radar (SAR), to characterise the tropical forest has been increasing greatly in the last 10 years and is based on three basic attributes: backscatter, coherence and phase-based approaches. The application of SAR images in tropical forest studies presents several advantages, such as independent imaging of the atmospheric conditions, increased penetration capacity of targets, iteration with this targets in terms of forest structure and the unique nature of the information based on electrical and geometrical proprieties of the target. The goal of this work is to use SAR images and the geomorphometric variables derived from Shuttle Radar Topography Mission (SRTM) to characterise the structure of tropical forest. The study area is in Tapajos National Forest, in the Brazilian Amazon. Linear regression analyses were conducted to characterise the biomass (BM), basal area (BA), height (H) and percentage canopy openness (CO) of the forest, based on geomorphometric variables (elevation, slope, aspect, plan curvature and profile curvature). The models revealed the coefficient of determination between observed and predicted values were r² = 0.32 for CO, r² = 0.26 for H, and r² = 0.52 for BA. It was also generated separate biomass estimate models using ALOS-PALSAR polarimetry attributes and geomorphometric variables showing r² of 0.35 and 0.58 respectively. When polarimetric and geomorphometric variables are integrated, the modelling improves to the r² of 0.74 with root-mean-square error (RMSE) of 20%. Although the geomorphometry variation in the study area is only about 150 m, the results show that geomorphometry has an important effect on the distribution of the forest characteristics and the integration of these variables with SAR attributes can improve the precision of the modelling. Currently, the project MF-RADAR (Marie Skłodowska-Curie grant agreement No 660020) is aiming to develop a methodology to derive information on forest structure and floristics. This is based on TanDEM-X (X band), PALSAR/ALOS (L band), Sentinel 1 (C band) integrated with geomorphometric data derived from SRTM (C band) to benefit the monitoring of forest communities and estimates of biomass and carbon stock to understand the carbon biogeochemical cycling in tropical forests. Preliminary results of this project show that the integration of those different SAR datasets allows a better characterisation of the tropical forest and has the potential to reduce uncertainties in the forest structure estimation using active remote sensing.

Wetlands are important providers of ecosystem services and several international initiatives make an effort to conserve these valuable environments. They are in the focus of the post 2015 development framework and its Sustainable Development Goals, the Ramsar Convention on the Wise Use of Wetlands as well as the Earth Observation domain within the recently launched GEO-Wetlands initiative. Consistent knowledge on the spatial extent of wetlands is still sparse and not well derivable from common Land Cover Classification systems. Even less is known about the temporal dynamics as well as land cover and (potential) use of these important ecosystems.

The objectives of the DeMo-Wetlands project are to develop improved methods and tools for detection and monitoring of tropical wetlands and to demonstrate and implement current remote sensing capabilities on the national scale in Rwanda. In DeMo-Wetlands, a special focus is set on tropical regions, where environmental and climatic conditions necessitate dense data input from different sensor systems. The working approach of DeMo-Wetlands is based on the Copernicus system: DEM data (if possible TanDEM-X) is used to identify wetland locations based on topographic conditions, followed by a verification of wetland sites based on a high spatial and temporal resolution optical dataset (Sentinel-2, RapidEye). This detection of wetlands is augmented by a monitoring component which considers inundation (Sentinel-1) and land cover/ land use of wetlands. The multisensoral working approach combined with a multitemporal analysis and continuous automation to enable a sustainable use of the demonstrator will be illustrated at the WorldCover 2017 Conference.

By presenting the DeMo-Wetlands project, its technical aims and its envisaged impacts on the monitoring of tropical wetlands, we contribute to the discussion on future Land Cover products: The requirements on data and processing capabilities as well as the integration of our specific Land Cover product into a community aiming for comprehensive products in the domain of Land Cover/ Land Use mapping are discussed. The work of DeMo-Wetlands is closely linked to GEO-Wetlands and other GEO related efforts.

Classifying land use / land cover (LULC) in heterogeneous environments with disturbed ecosystems and extensive human-induced LULC change can be challenging using only satellite data. Inclusion of auxiliary GIS datasets in the classification process to improve classification accuracy has been a standard method for a long time, but until recent years, auxiliary datasets of reasonable quality, accuracy and resolution have been non-existent on a global scale, especially in Sub-Saharan Africa. We wanted to test whether new global high-resolution datasets, namely SRTM v3 (http://www2.jpl.nasa.gov/srtm/), WorldPop (http://www.worldpop.org.uk/) and SoilGrids (https://www.soilgrids.org) can improve overall classification accuracy by including them as explanatory variables in a Random Forest (RF) LULC classification model.
Our study area, 1300 km2 in size, is located on the southern slopes of Mt. Kilimanjaro in Tanzania. It is characterized by a heterogeneous mosaic of disturbed savanna vegetation, croplands and built-up areas. The area is under heavy land use pressure, driven by human population growth, expansion and intensification of agriculture, urbanization, and grazing.
The classification was based on image segmentation derived from four geometrically co-registered, calibrated, atmospherically corrected and mosaicked Formosat-2 scenes from 2012 with eight meter spatial resolution. We used ground reference data (n=698) collected between 2012 and 2015 for defining 17 LULC classes following the LCCS v2 protocol, and for training and validation during the classification process.
A total of 63 potential explanatory variables were calculated for each segment. These included satellite-derived variables (mean reflectance and standard deviation of Formosat-2 bands and principal components, GLCM texture, band ratios and vegetation indices), elevation variables (mean elevation and slope range), Euclidian distance to visually interpreted rivers, mean population from WorldPop and soil variables from SoilGrids (Cation exchange capacity, C/N ratio, N, organic C concentration, soil pH in H2O, sum of exchangeable bases).
We used variable selection procedure implemented in the VSURF package in R to remove redundant variables. Selected variables were then employed in the RF model for training and prediction. From these results, we calculated confusion matrix and derived overall accuracy using out-of-bag samples.
The whole process of variable selection, RF model training and prediction was repeated five times. The first run used only satellite-derived variables and resulted in overall accuracy of 74.9%. In the four subsequent runs, we included more variables while still keeping all variables from previous runs. Inclusion of elevation variables increased overall accuracy to 79.5 %, Euclidian distance to rivers 82.1%, mean population 83.2% and soil variables 85.4%. With each run, the overall accuracy increased, yielding an overall improvement of 10.5 percentage points when using all explanatory variables compared to using only satellite-derived variables.
According to RF unscaled variable importance measure (mean decrease in accuracy), three most important variables for the whole model were mean elevation, mean reflectance of green band, and ratio of red band. There were, however, considerable differences in variable importance per LULC class.
Our study highlights the potential of these new global high-resolution datasets in improving overall accuracy of classification of image segments with the aid of a sufficiently large ground reference dataset.

The main objective of this study is to monitor, map and analyze land use/ land cover (LU/LC) changes between the years 2000 and 2015 for Dubai Emirate which is the most developing city in the world. To map and monitor LULC changes, we proposed a low cost approach based on support vector machine algorithm which uses Landsat images. We ran the algorithm and its proper parameters on each of the target years (2000, 2005, 2010 and 2015) using 235 training samples collected from Quikbird images with a spatial resolution of 0.6 m and field observation where the authors are living. After that post classification was performed. This includes filtering, combining classes, Sieving and Majority/ Minority. Finally, obtained LULC classes were separated using density slicing and converted into image classes and total area of each class was calculated. Results of change detection and image difference showed that from 2000 to 2015 a positive and a rapid change in built-up area, roads and water body classes. The results obtained indicate that from 2000 to 2005 approximately 233.721 km² (5.81%) of the Dubai Emirate was occupied by built-up area. The assessment of changes from periods 2000- 2005 and 2010-2015 confirmed that net vegetation area losses were more pronounced from 2000 to 2005 than from 2010 to 2015, dropping from 47,618 km² to 40,820 km² respectively.

River Chenab is formed by alluvial deposits, frequently facing riverbank and river channels dynamics due to floods. Present study is an attempt to compute the actual river bank dynamics within the Pakistan and India for a period of 3 years (2013-15) i.e. before and after flood of 2014. The entire course of Chenab River upstream of Jammu and Kashmir, Tandi to confluence with Indus River at Mithankot for a stretch of about 964 kilometer has been in to this study via an integrated approach of Geographic Information System (GIS) and Remote Sensing. Entire mapping of riverbank and river channels in the study area is from Jayotipuram to Kot Mithan. After image processing it was found that the River Chenab in highly snaking river with several prominent sections where the river has been suffering immensely with the shifting and erosion characteristics. This shift in riverbank in year 2013-14 before flood was 1029 m near city Gujrat and 2582 m in year 2014-15 after flood near Chiniot City.. These results reveals the dynamics of Chenab River.

Dead Sea Recession
Dead Sea is a salt lake bordered by Jordan to the east and West Bank to the west. Its surface and shores are 429m below sea level, Earth' slowest elevation on land. Dead Sea is 304m deep, the deepest hypersline lake in the world. Dead Sea is 50Km long and 15Km wide at its widest point. It lies in the Jordan Rift Valley.
Since 1930, when its surface was 1,050km2 and its level was 390m below sea level, Dead Sea has been monitored continuously. In recent decades, Dead Sea has been rapidly shrinking. From a water surface of 395m below sea level in 1970 it fell 22m to 418m below sea level in 2006, reaching a drop rate of 1m per year. As the water level decreases, the characteristics of the Sea and surrounding region may substantially change.
Remote Sensing technology become accessible tool and available to identify water bodies through satellite images which cover large areas of the regions of difficult access with short time. Satellite images provides information on earth surface, geographic area, and natural phenomena. Analyzing the images of the Dead Sea can help determining the water level and surface area of the Sea, its declining rate, the extension of water bodies and the appointment of flooded areas and monitor changes in water bodies over time.
In this study we use a combination of Land sat satellite images of Dead Sea for different years 1964-2014 with 30m resolution.

The urban landscape of Bucharest, the capital city of Romania (ca. 1.9 million inhabitants), shows dramatic transformations during the last five decades, from qualitative and quantitative points of view. Primary urban structures with house districts and gardens, with no urban planning, were replaced by systematically developed industrial districts with blocks of flats, in the context of an intensive industrial development during the socialist period. The paper proposes an integrated approach, by combining land cover derived data from a diachronic set of imagery: CORONA KH-4 panchromatic imagery from July 1968 (data source USGS, Declass-2 archive) and Sentinel-2 MSI multispectral imagery, from July 2016. The approach has three sections, as follows: a photogrammetric workflow applied to CORONA KH-4 imagery for orthophoto coverage production at ...m resolution, that was the subject of a visual interpretation and digitalization of land cover classes; a satellite image processing approach with Sentinel 2-MSI (source ESA Scientific hub) focused on a hybrid thematic classification of land cover classes (combined Support Vector Machine with image segmentation together with object-oriented approach); a GIS approach for urban land cover change mapping and standardized analysis, based on data from both sets of imagery.

After the evaluation of the land cover data accuracies (1968 and 2016), compared with ground truth data, derived from historical maps and recent orthophotos (2014), the analysis produced a set of maps and returned statistical data. The land cover change was quantified on five classes, adapted from the existing CORINE LAND COVER 2012 (data source EEA-European Environment Agency), by calculating a set of indices: the building density change, the industrial/traditional land cover transformation, the built-up area/vegetation transformation, the street pattern change index, the accessibility change index etc. Results reveal that both datasets can be used for urban land cover mapping with an admissable level of accuracy (higher than 85%). Also, the synergy of Sentinels is proven to be more reliable in depicting land cover classes than other similar datasets.

The first 30 m resolution global land cover maps - Finer Resolution Observation and Monitoring of Global Land Cover (FROM-GLC) were produced with supervised classification at the individual pixel level. This approach is relatively automatic once the training sample is collected. However, the first generation of training sample was compiled based on the interpretation of single date Landsat images assisted by the use of higher resolution images in Google Earth and various field images and literature. The lack of seasonality in the first generation training sample made it difficult to extend its use to other times of the year for global land cover mapping.

In this presentation, we introduce a new generation of training sample that is developed based on images acquired in the spring, summer, fall and winter seasons for the land area of the world, except Antarctica and Greenland. It is our hope that the spatial and temporal transferability of training samples can be done with such a training sample set. In addition, this new training set contains two set of land cover classification systems, the two-level 29-class FROM-GLC classification system and the 23-class GLC2000 classification system. To support our training sample collection, we developed a new software, named Global Mapper. This has been expanded into a web portal to allow accessibility of the training sample set and support crowdsourcing for data validation. We will also introduce this new portal in this presentation.

A preliminary analysis of this training sample set indicates that the all-season sample set can substantially improve classification accuracy when compared to the first generation FROM-GLC. Using Random Forest classifier as an example, the overall classification accuracy for all 29 classes in FROM-GLC can be improved by more than 10% as compared to that in the FROM-GLC. We will present detailed comparison results between different seasons and the consolidation of multi-seasonal data sets to produce annual land cover maps at the conference.

Semi-arid region of Mongolian Gobi is the most important dust source region in the world. In this region, precipitation supports the fragile ecosystem. Gobi's environmental driving forces are a limited rainfall. The distribution patterns of precipitation have indicated decreased amount and shifted location from east to west region. In Mongolian Gobi mainly distribution has an annual herbs and perennial shrubs. Annual plants strongly depend on rainfall, and perennial plants can live even in very drought years. In this study, we focuses on bare soil, no- photosynthesis vegetation (NPV) and photosynthesis vegetation (PV) to clarify the vegetation response to precipitation in the Asian Dust Storm (ADS) outbreaks area. Hovmoller diagrams are great for displaying large amounts of data in a meaningful and understandable form Hovmoller. Hovmoller (time longitude) diagrams were generated to summarize and examine the space-time features of seasonal evolution and the anomaly patterns for the entire monthly time series from 1985 to 2013. As a result of creating and analyzing the Hovmoller diagram using vegetation index and rainfall data, NDVI values tended to increase with increasing precipitation during the vegetation growth period (VGP) between May and September, vegetation showed high response to precipitation. It means, in the Gobi region, precipitations were concentrated in a period from May to September. The NDVI values, which represented vegetation mass levels, responded well with the precipitation in this period. The distribution patterns of precipitation have indicated decreased amount and shifted location from east to west region. The areas with fewer precipitations were more easily affected by the dynamics of precipitations than the area with more precipitations. The most degraded area was southwest region of Gobi with the least precipitations. The NDVI values responded to precipitations in wide scale regions including Inner Mongolia, China.

Castile and Leon Crops and Natural Land Map (MCSNCyL, Spanish acronym) is an operational country size (94.224 km2) land use layer, updated annually, obtained through satellite imagery and ancillary data. The goal of the project is to produce a land use map that represents the changes in annual arable crops as well as permanent crops and the areas of natural vegetation. The project began in 2013, and since then layers for the years 2011, 2012, 2013, 2014, 2015 and 2016 have been generated.

The procedure implies the use of images from Deimos-1 (2011-2016), Landsat 8 (2013-2016) and Sentinel-2A (2016) satellites. From 2017 onwards it is expected to improve the spatial resolution from 20 to 10 m as long as Sentinel-2 imagery becomes more reliable in terms of availability. The classification is performed using a machine learning algorithm trained with data retrieved from several sources, especially from Integrated Administration and Control System for Common Agricultural Policy subsidies database and some other Land use databases available in Spain and Europe (Land Parcel Identification System, LUCAS, etc.). There is not specific field work.

The project is led by the Agricultural Technological Institute of Castile and Leon (ITACyL) and has the support of the Duero River Basin District Authority and the National Geographical Institute of Spain for the image acquisition. The Regional Ministry of Public Works and Environment and the Regional Ministry of Agriculture cooperate in the supply of training cases. The project is an adaptation of the US Crop Data Layer from US Department of Agriculture using databases available within the European Framework.

This project is now included in the Horizon 2020 project Sentinels Synergy for Agriculture (SENSAGRI) that aims to exploit the unprecedented capacity of S1 and S2 to develop an innovative portfolio of prototypes agricultural monitoring services. SENSAGRI was proposed in response of the H2020 EO Work programme "EO-3-2016: Evaluation of Copernicus Services".

This paper describes the methodology and the accuracy of the current product as well as the strategy for the Pan-European version.

Glaciers are not only important indicators of the ongoing climate change, but are also highly relevant as water storages and in connection with natural hazards. Nevertheless, mid- or long-term field monitoring is only conducted on rather few glaciers, which are mainly concentrated in a few areas on the globe. Field work in remote mountain areas often time consuming and expensive but reaching remote glaciers, e.g. in the Himalayas or in Antarctica, can also be dangerous or even impossible. Therefore remote sensing has been a main instrument in glacier monitoring for more than a decade.

In this study, we investigate the use of multi-temporal high (Sentinel-2) and very high (Pléiades) resolution optical satellite data for the monitoring of glacier surfaces and their changes in the Khumbu-Himal (Nepal), the Ötztal Alps (Austria and Italy) and the Ortler-Cevedale Group (Italy). We classify the surfaces of different glaciers into classes relevant for mass balance calculations, which are snow, firn, ice, debris and water and try to detect changes in those surface classes over time.

The monitoring of the change of snow cover, firn bodies and glaciated area can help to investigate glacier mass balance, e.g. through the detection of the equilibrium line altitude or by providing conversion factors for the calculation for the calculation of the geodetic mass balance. The changes in debris cover and proglacial lakes are relevant, especially on the large Himalayan glaciers, where a significant amount of melt is assumed to take place at the exposed ice cliffs surrounding proglacial lakes on the heavily debris covered glacier tongues, where barely any ice is exposed.

Furthermore, we use the optical data to map avalanche cones on the glacier surfaces; since avalanches are believed to be a main source of mass input on many debris covered glaciers, although though this has seen very little investigation until now.

The cities of the United Arab Emirates (UAE) are, constantly changing and developing at a fast pace. The application presented in this paper is a continuation of the project “Smart Application for Feature extraction & 3D modelling using high resolution satellite Imagery”, called SAFIY2. SAFIY’s aim is to automate the production of geospatial information based on very high resolution satellite data to aid the planning and monitoring of urban change in support of UAE Government initiatives such as Smart Dubai. It will also establish a strategy to create a National Spatial Data Infrastructure (NSDI) to facilitate data sharing primarily amongst government departments.

The main aspects that we will be presenting in this paper are (a) a land cover map of the UAE based on the Food and Agriculture Organisation (FAO) codes and standards – which expands the classes used for the first part of SAFIY and (b) a palm tree detection algorithm, which goes a step beyond by looking at object detection as opposed to pixel detection.

For the first SAFIY project, traditional machine learning classification techniques were used such as Support Vector Machine. It was noted that the classification accuracy they can reach has a limit, around 80%, and it is very costly to improve on it. Furthermore, these techniques are not able to capture the complexity of the different colours and views of a unique object in different satellite acquisitions of areas, so the algorithms are often a bespoke design for a specific dataset.

Neural networks have been successfully used on image processing for classification in many different areas such as computer vision and can overcome the problems of traditional methods. In recent years, there have been big improvements in the results that these networks can achieve and this project explores how they can be applied to satellite imagery. Deimos Space UK uses recent deep learning techniques with convolutional neural networks for the classification of land and palm trees and will compare the results with more traditional classification techniques.

One of the challenges is the collection of ground data for training and evaluation. As part of the project, a good dataset will also be compiled using UAE data and manual labelling of satellite data. Several satellite data sources of varying resolution will be used, mainly Worldview-3 and Worldview-2 data at 30-50 cm resolution and Deimos-2 and DubaiSat-2 at 1 m resolution.

Quality of land cover mapping, as well as result of many other Earth observation studies can be highly affected by inherent terrain complexity. In order to cope with and overcome these difficulties there is a need for additional information, which can be incorporated in model formulation or preprocessing steps.

An example of such application is topographic or terrain correction in case of multispectral measurements, where correct information about surface slope enables more precise computation of irradiance received from the Sun. This is achieved by taking into account the local geometry of the scene. In addition to direct irradiance that is mostly affected by the terrain slope and Sun position, specific geometry can also produce or influence the overall amount of diffuse irradiances from the sky and surrounding ground.

As expected, such effect is present in regions with high terrain variability, which is the case in central and southern parts of Serbia. Therefore, availability of appropriate digital surface/elevation models (DSM/DEM) is highly desirable. Term appropriate suggests that in some cases there can be a compromise between quality of DSM, on one side, and the spatial resolution of satellite images, roughness of the terrain, and level of co-registration on the other. There are several parameters that determine quality of elevation data, such as: acquisition and production technique, which depends on instrument type (passive or active), spatial resolution and vertical accuracy of the measurements, presence of noise and production artifacts, rate and availability of data coverage. DEMs produced using digitized contours derived from topographic maps, or produced directly by photogrammetry from airphoto campaigns, are usually considered as very accurate.

However, very often these data are not available for some regions, have high cost, or they are just outdated. On the other hand, satellite products usually have regional or global coverage, and recently many of these high quality products became freely available due to new open data policies. In this paper we focus on two global datasets that were produced by optical stereoscopic observations carried out by Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) on board of NASA’s Terra satellite and Panchromatic Remote-sensing Instrument for Stereo Mapping (PRISM) on board of JAXA’s Advanced Land Observing Satellite (ALOS). More exactly, we compare specificities of two freely available global-scale 30 m products: improved version of ASTER Global Digital Elevation Model (GDEMv2) and 30 m derivative of original 5 m ALOS World 3D Topographic Data (AW3D) in the case of Serbia. Two datasets are also statistically compared over the larger region between 40 and 50 deg N, and 15 and 25 deg E. Since AW3D does not have full coverage for the given region, possible combination of two dataset is considered. Applicability of combined DEM is demonstrated over Serbia. In addition, an overview of alternative DSM products is also given, including: Shuttle Radar Topography Mission (SRTM), Digital Elevation Model over Europe (EU-DEM), and commercial products like WorldDEM and NEXTMap. Possibilities of utilizing Sentinel-1 and new initiatives like 3DEP (by USGS) in the future are also considered.

Land cover mapping has become an increasingly important source of information in agriculture. Farmers use it for on-field decision-making, retailers and stock traders for planning and governments for making agricultural strategies and setting subsidy levels. Besides agricultural stakeholders, there are biologists and environmental scientists who use this kind of information for monitoring the quality of habitats.

There are a number of optical EO satellites which offer images free of charge, with Sentinel-2, a part of ESA’s Copernicus programme, and Landsat-8, launched by NASA and USGS, being the most popular ones. This work focused on their application in land cover mapping in northern Serbia. Using joint information from these satellites we improved the system in many aspects. Data fusion allowed us to have images from more dates available. In this way we decreased the risk of misclassification due to missing data caused by high cloud coverage. Also, it allowed us to train a more accurate classifier compared to those trained on individual satellites. Finally, the spatial resolution of resulting maps was higher than the resolution of input images. This is extremely important for the observed region which is mostly constituted of small fields, 400 x 60 m in average.

The database included more than 3 billion pixels with around 200 features each, i.e. all image channels from key dates between January and September. Classifiers were trained to distinguish between following crops: maize, wheat, sunflower, sugar beet and soybean, as well as forest and water bodies. Random forest proved to be the best classification algorithm, in terms of accuracy, speed and ability to deal with missing data. In classification of forest and water bodies, accuracy of up to 97 - 98% was achieved even without data fusion. However, since crop classification is a more difficult problem, performances of Sentinel and Landsat based classifiers could not match the performance of the joint classifier. Data fusion increased the overall system accuracy through the increase of average accuracy over all classes, as well as through more equal distribution of accuracy values over categories, in addition to higher spatial resolution of final decisions. The most significant improvement was observed in soybean classification, where Sentinel, Landsat and joint classifiers achieved accuracies of 84%, 87%, 89%, respectively. Other crops, such as sugar beet and wheat, which could be accurately classified with Sentinel and Landsat, were not improved further.

This work is a step towards next year’s case, when besides these two satellites, Sentinel-2b will be available. It will cut the revisit time of Sentinels to only 6 days meaning that there will be even more data available and even better classification performance can be expected. The system developed in this research is intended to be a part of a broader geo service. This service would offer solutions customised for a vast variety of users, utilising the full potential of land cover mapping.

One of the main applications of ESA’s Sentinel-1 SAR data is land cover analysis and classification. The regular repeat cycle imaging and the resulting time series of overlapping Sentinel-1 SAR images prompted us to develop reliable image classification tools that allow a quantitative inter-comparison of selected land cover areas. Our aim was to identify pre-defined land cover categories, and to observe the temporal evolution of these categories versus time.

The understanding of our classification potential and results is based on the selection of representative data sets, and an initial study of typical SAR image data characteristics. To this end, we looked at the data acquisition and processing parameters of Sentinel-1 images, performed a first analysis of visually discernible land cover categories, studied the dynamic range and noise characteristics of the Sentinel-1 target areas, investigated spatial neighborhood and temporal relationships, and verified the signal-to-noise estimates of the given images. Then we applied classification-oriented feature extractors to local patches of the selected images containing a large variety of different target areas, and analyzed the characteristics of the resulting feature vectors, and their clustering potential and performance.

All these Sentinel-1 parameters were then compared with corresponding parameters derived from geometrically overlapping high-resolution X-band SAR data obtained by the TerraSAR-X mission. Here we have to consider the impact of the different target area resolutions, the peculiar effects of C-band and X-band imaging, and the role of the selected data product options that can be selected during TerraSAR-X product ordering. A comparison of the two SAR instruments mainly based on their imaging and feature domain parameters demonstrates typical differences constraining their potential classification performance.

Besides these technical instrument and processing parameters, the semantic classification performance of typical SAR images is usually improved by inclusion of additional knowledge obtained from external sources. In our case, we employed Google Earth, the in-situ LUCAS land cover / land use database, Sentinel-2 images, and our common knowledge about the vegetation cycle and common farming practices as quasi ground truth. In addition, we used active learning by an experienced image analyst to feed additional knowledge derived from these external data sources into a support vector machine being used as a classifier. This inclusion of knowledge was supported by an already existing catalogue of semantic annotations that had been established and validated during previous land cover studies over comparable target areas.

The use of this semantic catalogue was the basis for obtaining quantitative classification performance results including the basic confusion matrix approach as well as the widely known precision/recall figures-of-merit. Our results demonstrate that SAR images can be classified with good reliability once we know more about the general land use principles of the target area under study.

The detection of permafrost phenomena and its distribution in high mountain environments as well as the monitoring of changes in permafrost areas is essential for alpine risk, infrastructure, natural hazards and climate change studies. It is assumed that in Iceland less than ten percent of the land surface is underlain by permafrost, where most of it may disappear under further warming conditions in the 21^st century. These changes will cause sincere problems for the society in mountainous regions. But because of the complexity of permafrost detection, the knowledge about its distribution in Iceland is currently not very well evaluated and only based on small-scale observations.

As permafrost is at most not directly observable, different indicators, e.g. rock glaciers and perennial snow patches, can be mapped to identify the distribution of permafrost. The research for this study is conducted on the Tröllaskagi peninsula, in Northern Iceland. It is situated between Skagafjörður and Eyjafjörður and the highest summits and plateaus reach an altitude of about 1400 m.

For large-scale identification of perennial snow patches over the Tröllaskagi peninsula remote sensing techniques are a practicable technique. In the presented study, we are using Sentinel-2, Landsat-5/7/8 data as well as aerial images to map and analyse the spatial distribution of perennial snow patches, indicating a low or negative ground temperature underneath. Therefore, if a snow patch occurs annually in the same area, it can suggest the existence of permafrost.

After an atmospheric correction of the satellite data, pan sharpening of the Landsat data and resampling the Sentinel 2 data, and Normalized Difference Snow Index (NDSI) calculations, the perennial snow patches are classified in i) manly permafrost, ii) mainly wind and iii) manly avalanche induced origin. For that purpose, topographic information such as slope angle, aspect and curvature are determined from a DEM of Tröllaskagi peninsula. In a first step a digital elevation model with a grid size of 25 m is used, which will be replaced after the release of the ArcticDEM offering a grid size of 5 m. Additionally, the wind atlas from the Icelandic Met Office (IMO) is used to identify the main wind direction. The analysis of the satellite data in combination with topographic as well as wind atlas information will result in a permafrost distribution map of the Tröllaskagi peninsula.

The increased availability of SAR (Synthetic Aperture Radar) satellite images has led to new civil applications of these data. In this context, we show the use of Sentinel-1 data for efficient monitoring of coastlines within the framework of a coastal Thematic Exploitation Platform (TEP) development study under the auspices of ESA. We are interested in the classification of urban and agricultural areas as these two categories represent typical examples of near-coastal land cover apart from non-accessible and uninhabited remote areas.

In order to analyze these data, a cascaded active learning approach and a system incorporating multiple instance learning for SAR image mining and semantic catalogue creation has been developed by us. Based on a multi-scale and hierarchical patch-based image representation, a cascaded classifier is learned at different patch levels using for each level a Support Vector Machine classifier. A classification of the current level patches is applied only to relevant (i.e., positive) patches retrieved during previous level annotation, while the irrelevant (i.e., negative) patches are discarded from the initial training set. This approach reduces the computational effort, in particular, when a large data set is annotated. For example, from the first to the second level annotation we typically keep only about 20% to 40% of the entire data volume, while the remainder of the data is discarded. For the next levels, we preserve about 80% to 90% of the previously used data, as we have already singled out the most important relevant data.

We interactively trained the classifier, and compared the resulting classification accuracies for high and medium resolution SAR images of two space-borne SAR imaging instruments (TerraSAR-X and Sentinel-1A) taken over different near-coastal areas (such as the Adriatic Coast, the Black Sea Coast, the Dutch Coast, etc.), all characterized by a high diversity of target categories.

In order to evaluate the consistency among the categories retrieved from TerraSAR-X and Sentinel-1A, we selected pairs of images with identical polarization (e.g., VV) that cover the same area on ground. These images were pairwise co-registered using their WGS-84 map projection and were cropped with QGIS. The selected patch size depends on the image resolution and pixel spacing, and was chosen in order to obtain overlapping image pairs.

The accuracy of the annotation shows promising results. During a case study, we noticed that the precision/recall value is generally lower for Sentinel-1A (83% / 73%) compared with TerraSAR-X (89% / 83%) for our selected images in line with a higher number of categories identified for TerraSAR-X images. For example, for a typical coastal target area in Italy, the number of retrieved categories for TerraSAR-X is most often about nine categories (e.g., bridges, cemeteries, firth, harbor infrastructure, high-density residential areas, high-density residential areas and channels, mixed forest, railway tracks, and sea), while for Sentinel-1 the number of retrieved categories is typically five (e.g., bridges, harbor infrastructure, inhabited build-up areas, natural vegetation, and sea). These results demonstrate that SAR images can be considered for coastal area monitoring.

The activities of this new GEOBON working group focus on characterizing Ecosystem Structure. The WG develops new techniques and algorithms for earth observation from space in characterizing ecosystem structure and its change over time. In general, the WG focuses on measures of ecosystem state and how it is changing, including condition of and change in the structural components that maintain biodiversity characteristics. The specific objectives include Identifying the subset of EBVs for which satellite remote sensing can play a key role in characterizing Ecosystem Structure and its change over time (RS-EBVs) as well as coordinate the development of RS-EBVs with GEOBON working groups. An important activity is to engage user groups such as CBD SBSTTA, IPBES and NGOs on candidate RS-EBVs, solicit feedback, and get “buy-in”, with particular reference to the Aichi targets and sustainable development goals (SDGs).

The candidate Ecosystem Structure EBVs and associated RS-EBVs include Habitat structure (e.g. structural traits such as height, crown cover and density, SLA & LAI; structural biochemicals such as lignin, cellulose & protein; coral rugosity etc); Ecosystem extent and fragmentation; Ecosystem composition by functional type; and of course Land cover. The list contains RS-EBVs that are continuous and biophysical such as vegetation height, leaf area index and specific leaf area, as well as others that are categorical, such as land cover. Also, like some ECVs, some RS-EBVs are actually groups of related variables that describe a phenomenon of interest, such as forest disturbance.

With this list as a starting point, the next steps in the process can begin with the ultimate goal of putting a plan in place to acquire the needed RS observations to generate the related EBVs. Key organizations for this are the CBD, IPBES, CEOS, and GEO BON, with GEO playing a facilitative role, however the broader biodiversity community is also very important. A key goal is to meet as many as possible of the reporting needs that CBD signatory countries have for the Aichi targets.

Several ongoing initiatives aim to deliver land cover and land use change (LCLUC) information at global-scales with the aim to address different user needs. Such initiatives, through their differing evolutions and intent, presently follow different standards, methods and specifications in terms of spatial, time, and thematic resolutions. Such LCLUC datasets have the potential to support a number of the 17 Sustainable Development Goals (SDG) of the 2030 Agenda for Sustainable Development, ratified in 2015 by the United Nations.

The GOFC-GOLD Land Cover Project Office performed an analysis of the current list of SDGs with respect to existing LCLUC products. The analysis aimed to: 1) identify which goals, targets and indicators LCLUC data can support, and 2) identify gaps and needs, in terms of data, and indicators (are indicators specific enough to gather information relevant for the targets?). We conclude that some targets, and their commensurate global indicators, can be served already, but that some targets do not yet have sufficient data and methodological depth in their indicators to assess whether these are achievable or have been reached. Still some goals/ targets can benefit from already working/ anticipated programs. We also observe the need for continued dialogue between those experts responsible for developing the SDGs indicators, the technical scientific community, and the policy needs in order to assess gaps and further the development of indicators for national implementations.

The EU-funded ECOPOTENTIAL project shall demonstrate, among others, the application potential of satellite images for detailed ecological studies of environmental and ecological protected areas within pre-defined bio-geographical regions of Europe. These studies aim at the corroboration of essential variables for the monitoring of protected areas and ESA’s concepts for a thematic exploitation platform covering coastal areas.

The innovative character of our project is a systematic assessment of how to exploit SAR satellite images in combination with multispectral optical satellite images and/or local in-situ measurements. The results of the project shall demonstrate the scientific gain when combining SAR image data with optical instrument data. While standard products of Sentinel-2 and WorldView-2 provide a sound basis for multispectral analysis and interpretation of vegetated areas, any additional information contained in SAR images (of TerraSAR-X and Sentinel-1) can improve the classification results and the analysis of time series data. We estimate that about 10 essential protected area variables (out of 93 envisaged variables) can be extracted from these images.

In the meantime, we applied a number of data analysis and data mining tools to Landsat, Sentinel-1 and Sentinel-2 data. In particular, we analyzed and classified time series of Landsat 5 (acquired over Doñana Park in Spain) and Sentinel-1A data (acquired over the Dutch Wadden Sea and the Danube Delta in Romania). These activities resulted in reliable classification maps with more than 10 semantic categories, change maps for time series data, as well as the corresponding classification accuracy metrics together with statistical analytics. These can be used for detailed validation and long-term cross-comparisons.

In addition, we looked at Landsat 5, Sentinel-1, and TerraSAR-X data from the viewpoint of feature extraction, and tried to extract physically meaningful parameters such as tidal effects, noise levels, water currents, vegetation cycle phenomena in agricultural areas, and urban growth. These observed effects were detected by machine learning algorithms, supported by an experienced image interpreter, and ground truth comparisons. The quality of the retrieved results is quite encouraging.

In future, we plan to extend our tools to deep learning algorithms where we hope to gain an even higher degree of automation during routine operations (i.e., after having finished an initial training phase) combined with highly accurate classification results.

One of the primary bio-physical indicators of desertification and land degradation is the change in terrestrial ecosystems productivity. Productivity of land and vegetation is the source of the food, fiber and fuel for large part of humanity, so it represents the basis of the Earth habitability. The spatial variability of the vegetation productivity, measured as Net Primary Productivity (NPP), over the globe is enormous, from up to 1000 gC/m² year for evergreen tropical rain forests to less than 30 gC/m² year for deserts. Desertification, land degradation and climate change may affect NPP over large areas with impacts on land productivity. “Land Productivity Index” (LPI) is a key proxy indicator for NPP changes and related land degradation and desertification hot spots. The present work aims to test a remote sensing-based methodology to measure LPI from country to river basin scale on the Italian national territory with a focus on croplands, forestlands and grasslands.

Over the Italian territory, in general, the increasing LPI areas exceed decreasing LPI ones in all land cover classes: Southern Italian regions show a general “greening up” and only limited hot spots of LPI decline. In Central and Northern regions a general LPI decline prevails on both on croplands and forestlands. The LPI trends have higher statistical correlation with precipitation trends in southern regions and lower correlation in northern regions. The prevailing LPI decreasing trend in northern regions may be due to land management and/or land cover change. Land cover change and land productivity change are complementary to support decision makers.

The LPI trend analysis as well as the effect of the land cover information in the degradation assessment are showcased as a pilot activity within the EO Datacube service portal, a European Space Agency initiative to improve the accessibility and dissemination of the EO data and products currently available at the European Space Agency.

The paper will present results of ESAs EOEP-4 Data User Element Innovators III project on the development of Earth observation products and services for Vector Borne diseases (VecBorn).

During the last ten years Buruli Ulcer has affected more than 42.000 people, mainly in rural areas of Central and West Africa. Moreover, cases were reported in Japan, Australia and Papua New Guinea [1]. The tropical skin disease causes large ulcers, and without early treatment it needs extensive surgery leading to huge deformity. Since 1998 the World Health Organization (WHO) organizes the investigation of the disease, creating interdisciplinary Research priorities. Until today methodologies for appropriate medical treatment and diagnosis still need further investigation as well as the search for the mode of transmission of the pathogen. [1]. The German-based space company Jena-Optronik is the project leader and in cooperation with the Swiss Tropical and Public Health Institute endemic risk areas were located in stagnant or slow moving waterbodies and seasonal flooded environments, especially when they occur in combination with agricultural fields.

SAR satellite data improves the knowledge of environmental dynamics in tropical rural areas. Due to their temporal and spatial resolution Sentinel-1 GRD products have been chosen to determine risk areas and potential habitats of the bacterium. The following workflow was tested in Cameroon. Nearly the whole catchment area of the Mbam, Sanaga, Nyong and Dja River is covered by four Sentinel-1 footprints, each including 36 datasets from 2015-04-07 to 2016-10-10 with a twelve-day revisit time.

The data passes a workflow with basic SAR preprocessing steps. Multi-temporal statistics serve as data basis for the threshold method to delineate maximum and minimum water masks. After a selection of at least 14 scenes, which are spread throughout a year, a multi-temporal maximum of the backscatter values was calculated for the determination of the minimum water mask. To avoid misclassification in the maximum water mask the data basis for the multi-temporal minimum statistic was kept small with two scenes.

If available, optical data from Landsat or Sentinel-2 can easily support the adjustment of the thresholds by calculating the MNDWI. Misclassifications could be reduced by comparison to the corresponding slope of the Digital Elevation Model and excluding areas above a certain slope from the water mask.

Due to an increasing number of Buruli Ulcer cases and their increasing geographical distribution, the workflow is created for a global application.
The results of the study provide a fast and efficient approach to gain insights of environmental dynamics and delineate Buruli ulcer risk areas. This information supports epidemiologists to locate habitats and to understand the mode of transmission of the Mycobacterium Ulcerans.

[1] Röltgen K, Pluschke G (2015): Epidemiology and disease burden of Buruli ulcer: a review. <https://www.dovepress.com/epidemiology-and-disease-burden-of-buruli-ulcer-a-review-peer-reviewed-fulltext-article-RRTM>

Developing better agricultural monitoring capabilities based on Earth Observation (EO) data is critical for strengthening food production information. Since 2011, support for agricultural monitoring using satellite data has formal institutional support, objectives and timelines, e.g. with the Global Agricultural Monitoring Initiative (GEOGLAM) building on GEO’s Agricultural Community of Practice and the Joint Experiment of Crop Assessment and Monitoring.

Thanks to support of more than 15 champion users, including organizations such as FAO, WFP, CGIAR or JRC of the European Commission as well as national mandated institutions, the requirements and the technical specifications of relevant EO-derived products have been defined.

Launched in 2014, the Sentinel-2 for Agriculture (Sen2-Agri) project funded by the European Space Agency aims at developing an open-source processing system to convert Sentinel-2 (S2) and Landsat 8 (L8) data into relevant EO agricultural products in an operational way. With its 10/20-meter resolution, its forthcoming 5-day revisit frequency and its global coverage, the Sentinel-2 (S2) mission offers unprecedented properties for agriculture applications. Its synergy with the L8 sensor is an additional key factor which opens new opportunities for near-real time applications.

The open-source Sen2-Agri system makes full use of all S2 and L8 images available along a given growing season. It is designed to be fully automated, from the L1C S2 and L1T L8 data download from the ESA SciHub and USGS servers to the generation of the final EO agricultural products. These products consist of (i) monthly cloud-free composites, (ii) monthly cropland masks, (iii) cultivated crop type map and (iv) LAI and NDVI indicators describing on a 10-day basis the vegetative development of crops.

The algorithms selection based on site globally distributed and the system design and development are completed. This was achieved using S2-like time series, made of SPOT-Take5 and L8 images, complemented with in-situ data collected by the JECAM network teams.

The implementation of the Sen2Agri system allows to currently demonstrate the overall approach at full scale over the 2016-2017 growing seasons in near real-time with S2 and L8 data. This demonstration is completed at national scale over Ukraine and ongoing over Mali and South Africa. In addition, eight local demonstration sites are spread over Europe, Africa and Asia. This selection of sites aims to be representative from a large variety of agricultural practices and climate conditions.

The generation of the Sen2Agri products in near real-time highlights the synergy existing between the S2 and L8 sensors in terms of classification. It also provides valuable information about the joint S2 and L8 performance to deliver agricultural information at critical dates. The interest of using both sensors is discussed accordingly to the context and more particularly on the cloud cover frequency, on the agricultural practices and on the field size distribution.

The demonstration activities are carried out in close interactions with local teams, with the objective to transfer the system to their operations. These teams are involved in the products generation and validation, but also in the assessment of their relevance for varying agriculture applications.

The performance of a change detection technique depends on the radiometric quality of Earth Observation (EO) data. The Cross-Correlation Analysis (CCA) as change detection technique requires the use of only one multi-spectral image at time T2 and a Land Cover/Land Use (LC/LU) map at time T1. The radiometric accuracy of the time T2 image can be improved applying an atmospheric correction to obtain a surface reflectance suitable for the appropriate analysis. In the atmospheric correction processing many atmospheric variables are involved. The comprehensive representation of the local aerosol during the sensor acquisition are provided by the aerosol optical thickness at 550 nm and the micro-physical properties of the aerosol.

In this work, the CCA technique was applied to a recent Landsat 8 image for change detection. The CCA overall accuracy values obtained by selecting different aerosol types during the atmospheric correction performed by the OLI@CRI (OLI Atmospherically-Corrected Reflectance Imagery) algorithm (Bassani et al., 2016), are presented. The OLI@CRI algorithm was applied to the Landsat 8 image acquired August 10, 2014 on a protected Natura 2000 site in southern Italy. The microphysical properties of the aerosol provided by the AERONET station close to the site were used. The local aerosol was present in fine-dominated mode with absorption property similar to the water-soluble and dust-like.

The results highlighted the effectiveness of the atmospheric correction providing an Overall Accuracy (OA) of 95.47±0.31% using AERONET data vs. an OA of 91.10±0.43% with no atmospheric correction. When basic components were considered in the OLI@CRI algorithm, the OA decreased in the case of absorbing and fine-mode aerosol (soot) and with coarse aerosol (dust-like and oceanic); while in case of water-soluble component (similar properties to the local aerosol) the quality of the results was preserved as attested by the overall accuracy (95.52±0.31).

Glaciers are important climate indicators and water storages. The melt water from glaciated areas is an important water resource for human consumption, irrigation and hydropower generation. The correct and detailed identification of glacier areas and their changes have a high impact, e.g. on cryospheric, hydrological or biophysical applications. The total glacier areas are changing gradually as response to variations of regional climate. The class “permanent snow and ice” in land cover data such as GlobCover 2009 or GlobeLand30 need to be regularly updated at a high level of detail.

Glacier areas from optical satellite data have been widely mapped based on Landsat, SPOT, and other high resolution optical satellite data, applying different methods. Data bases from different sensors and years, used for generating glacier inventories, show considerable differences in completeness and detail for the different global glacier regions. The first Sentinel-2 satellite of the European Copernicus programme, launched in June 2015, provides a new comprehensive data base, combining high resolution with an increased spatial coverage and 13 bands in the VNIR and SWIR spectral range. At clear sky conditions, large glaciated areas can be observed by one scene. The improved repeat observation frequency, which will be further increased with the second Sentinel-2 satellite (planned to be launched in 2017), enhances the chances to acquire clear sky images at the end of the ablation period needed to monitor glacier outlines.

We exploit the spectral capabilities of the multispectral instrument (MSI) on board of the Sentinel-2A satellite to improve the processing chain for retrieving glacier outlines. We implemented an automated processing chain, including topographic and atmospheric correction applied to the spectral top of atmosphere reflectances, and combine multiple spectral bands and auxiliary data to generate an interim product on glacier area extent. The aim of the automated classification is to minimize the efforts for manual corrections of glacier areas, as e.g. by identifying debris-covered ice. We demonstrate the application of the processing line for monitoring glacier outlines for Sentinel-2 scenes over the Alps, acquired in summer 2015 and 2016. The output is used as contribution to update the alpine glacier inventory, and as additional land cover layer for monitoring snow.

Wetlands are complex and dynamic water – logged ecosystems, hosting a great amount of species of plant and animal communities, also supporting water regulation and peat related processes. Thus, developing accurate and robust techniques for mapping and monitoring changes in such is of crucial importance in environmental monitoring and management. Earth Observation (EO) offers great advantages in this respect thanks to its capability of acquiring - often at no cost - spectral data over large areas and at regular time intervals. The recent launch of EO instruments such as that of Sentinels series from the European Space Agency (ESA) opens up new opportunities for exploring the development of techniques that will allow improving our ability to map wetland ecosystems from space.

The aim of this study has been to evaluate the synergistic use of Sentinel-1 and Sentinel-2 data combined with the machine learning classification algorithm Support Vector Machines (SVMs) for mapping a typical wetland ecosystem in Greece. As a case study is selected the National Park of Koronia & Volvi lakes in Thessaloniki due to its high environmental interest. Additional spectral information created from the initial bands and of elevation to the thematic information accuracy was also examined. A Principal Component Analysis (PCA), a Minimum Noise Fraction (MNF) and a Grey Level Co-occurance Matrix (GLCM) were applied and their added value to the classification accuracy was evaluated. Accuracy assessment of the derived wetland maps was conducted on the basis of the estimation of standard classification error matrix statistics.

In overall, results exemplified the appropriateness of the Sentinel imagery combined with the SVMs in obtaining a mapping of the wetlands area. This is of considerable scientific and practical value, as it strengthens evidence on the suitability of synergistic use of Sentinel data for improving our ability to understand better Earth’s physical process and physical environment.

Starting from the already operational use of Sentinel 2A for the EU Common Agricultural Policy control (for instance Italy, through its agricultural payment Agency AGEA processed around 70 S2 images in 2016), several Member States are going to update their data bases and operate their controls fully exploiting the Copernicus imagery advantages.

The goal is to enable low cost and multiscale agronomic land use updating, be compliant with the agro-environmental CAP requirements, providing additional thematic layers for the private sector, both for farming support and insurance services. Main CAP Sentinel 2 forthcoming tasks could involve:

• The Land Parcel Identification Systems (LPIS) update: continuous updating of the different Reference Parcels subjected to EU subsidies through both automatic classifications and manual interpretation. This will allow following and marking any historical change concerning the requested aid eligibility by EU CAP (retrospective recovery) even between the cyclic periods of VHR aerial coverages (3-4 years)
• The Geo Spatial Aid Application (GSAA) support: helping each farm to better recognize his parcels, reducing errors and possible frauds through an updated Sentinel 2A layer allowing the farmer to better identify his current “land portion” to be declared
• Risk analysis of the overall agronomic situation support: along the season, and in near real time, multitemporal Sentinel 2 data will better address the selection of mandatory scattered samples of controls
• Controls with Remote Sensing (CwRS) improvement: multitemporal data availability over the area interested by subsidy declarations, will improve the phenological phase detection (early crops, crop types, crop rotation, environmental issues, etc)
• “Greening” policies better management: land changes on grassland and set-aside, nitrogen fixing crops and natural bio-strips (EFA Ecological Focus Areas) will be more reliably detected
• Farm Advisory System (FAS) support: multitemporal Sentinel 2 data will improve the operations by local (public or private) consultants providing a sound geographical “analytic” support to farmers
• Insurance sector support: providing reference maps and indicators in support to risk and damage estimation.

In this framework, EU Commission services, for the agricultural sector, should drive the Copernicus Sentinel technical improvement related to a new and powerful usage of Geo Information, also by redesigning regulations and reforms.

Invasive species (namely Prosopis juliflora and Parthenium hysterophorus) significantly impede rangeland and cropland productivity in Somaliland impacting the livelihoods of thousands of agro-pastoralists. Invasive species propagate well in overgrazed and poorly managed sites while their encroachment enables further land degradation processes such as loss of biodiversity and declining rangeland quality and quality. Their spread mechanisms are poorly understood and, moreover, information on containment areas for possible interventions is not readily available. In this study we used 250-meter Moderate Resolution Imaging Spectroradiometer (MODIS) Enhanced Vegetation Index (EVI) time-series data to compute vegetation productivity and phenometric trends (2001-2014) to map P. juliflora and P. hysterophorus propagation areas in western Somaliland. Field data collected during the invasive species growing seasons in 2014 and 2015 and bi-temporal (2001-2014) Landsat images (30-meter pixel resolution) were used to create an invasive species occurrence and propagation reference data set. The MODIS-based trends were linked to the Landsat-based reference data using a binomial logistics link function. Invasive species spread and containment zone maps were created for both invasive species, respectively, using the logistics model probability results. The propagation of P. hysterophorus could be modeled with a prediction accuracy of 0.89 while for P. juliflora the logistics prediction score was higher at 0.94. Areas of high propagation and spread for both species were mostly the cropland areas in the south-western part of the study area as well as the peri-urban zones (around Hargeysa most notable). Containment areas for P. juliflora could be effectively identified within buffer zones of riparian vegetation that were found to be highly infested. This study showed the potential and relevance of phenology metrics for the wide-area assessment of vegetation dynamics in relation to the spread of invasive species in African drylands.

Keywords: Moderate resolution imagery, phenometrics, Generalized Linear Modeling, invasive species, Somaliland

The work presents the intelligent platform that was developed for visualising, analysing and comparing global land coverage in Europe. The global data available in the webGIS are: GLC-2000 (2000), CORINE (1990, 2000, 2006 and 2016), GlobCover (2004-2006 and 2009), MODIS (2001-2012), CCI project (2000, 2005 and 2010), GlobeLand30 (2000 and 2010), Urban Atlas (2006 and 2012), GHS Settlement Grid (1975, 1990, 2000 and 2015) and Global Urban Footprint (2011-2013). The data storage consists in two major elements: the raster data storage, i.e. a collection of pre-processed GeoTIFF files optimized to increase the web map server performance; and the vector data storage, which is built on the PostgreSQL DBMS with PostGIS spatial extension.

The client, built upon the Bootstrap framework and Font Awesome icons is designed to have a similar navigability of a desktop GIS, where users can activate, order (e.g. drag and drop operation) and modify some of the layers’ properties (e.g. opacity). In addition, it is possible to upload external land coverage/land use maps or any other geospatial data via Web Map Services (WMS) provided that the user enters a valid GetCapabilities URL address. The results are parsed via a JavaScript function and rendered by Leaflet. Besides those functionalities, the platform provides the possibilities of processing on-the-fly selected data (in a rectangular area) in terms of change detection analysis and computation of the confusion matrix and related indexes (overall accuracy, allocation and quantity disagreements, user and producer accuracy). Finally, geotagged photos from some popular photo sharing services (Flickr, Panoramio and Geograph) that provide an Application Programming Interface (API) for extracting their data can be added to the maps.

One of the key goals of sustainable land management is achieving the environmentally-oriented landscape structure taking into account the watershed perspective, habitat connectivity, habitat core areas distribution, and ecosystem carrying capacity. Landscape fragmentation, habitat isolation andreduction the areas with natural vegetation are the main factors of biodiversity loss. To assess the landscape fragmentation dynamic over 15-year period in Ukraine the available historic land use land cover datasets with comparable classification were selected, in particular, GLC2000, GlobCover2005, GlobCover 2009. Proba-V satellite images were used to develop the land cover grid for 2016. Land cover classes dynamic was analyzed and landscape fragmentation indexes were derived to evaluate the changes in the land cover. Low spatial resolution of the input land cover datasets (about 300 m) was one of the essential limitation for landscape fragmentation analysis. Open street map vector data on road network, rivers and settlements boundaries was used to create the buffer zones and subset mask for land cover datasets, which helps to override the spatial resolution limitation of land cover data.

The use of artificial intelligence algorithms has set new perspectives in the classification of remotely sensed data. On one hand, they have become important because their non-parametric models are independent of a previous assumption about the type of statistical distribution of the data. On the other hand, they do not require any specific relationship between the predictive and response variables, and allow the adjustment of a response surface even for a complex and non-linear problem. However, there are only limited experiences of application of these methods in the Argentine Pampas, where Soybean and Corn are the mains crops. In the present work, we compared two machine learning algorithms against a conventional one for the mapping of lands dedicated to agriculture. To this end, we investigated the adjustment of associated parameters, as well as the effects of the nature and number of classes included in the models. In both cases we used a single cloud-free Landsat-8 scene, acquired in March 2016, when the crops were in the reproductive stages. Training and validation data were obtained from a field study, with a sampling date close to image acquisition. All the classification models were validated using the Global Accuracy (GA) and Kappa (k) statistics. In the first step we evaluated the widely used Maximum likelihood classifier (ML), generating predictable and comprehensible results when only two classes (Poaceae vs. Fabaceae) were separated (GA= 95%). However, due to the parametric feature of the maximum likelihood classifier, the high accuracy achieved in this study depended considerably on the size and reliability of the training set. ML approach made mistakes when the number of classes increased and the supra-classes were disaggregated. In practice, a size much larger than the ideal minimum is required to fully represent the distribution in a multidimensional space. Secondly, we tested Random Forest (RF) and Support Vector Machine (SVM) algorithms. The results showed that the reflectance images allow the mapping of the diversity of coverage present in the study area, before the end of the crops cycles. Its ability to maintain the separability of the classes was outstanding (GA > 95% and k > 0.94), even when the classes are numerous, similar one to each other, and with a great internal variability. We emphasize the need to manually adjust the parameters associated with each classifier, since this issue constitutes a specific problem. It is important to take into account the computational requirements of SVM, as concerning the optimization of the penalty values ​​(C) and the γ parameter. This cost increased considerably with the number of classes and the size of the training set. The strength of the method is highly dependent on the characteristics inherent to the local problem under study, such as the large size of the establishments (mean = 100ha), the limited number of cultivated species, and the morpho-physiological contrast among the major botanical families. These considerations make possible to produce precise and accurate maps prior to the completion of the campaigns, which is beneficial for planning, estimating and generating budgets.

Due to growing population and sparse land and water resources, the need grows for enhancing agricultural productivity to ensure food security. Accurate crop maps from earth observation can build the basis for agricultural monitoring at a range of scales. Such maps are one of the essential means to support sustainable land management. In this study, we exploited the intra-annual temporal signatures of remotely sensed observations and used the prior knowledge of crop calendars for the creation of sequential processing chain for crop classification. We applied the method to the study site in Central Ukraine as it has undergone profound changes during the last decades in the extent and intensity of land use. The area is characterized by volatility in agricultural production caused by several drivers such as weather conditions.

Landsat-based time-series metrics that capture within season phenological variation were preprocessed and analyzed using Google Earth Engine cloud computing platform. The development stage of each crop throughout the growing season was modeled using the harmonic regression. The model’s output was further used for the automatic generation of training samples. Sentinel-1 images were used as additional input of contextual feature information to classification. Two classification schemes were applied to discriminate the main crops in the study area, namely pixel- and object-based approaches, to classify the following crops classes: winter cereals (wheat, barely), winter rapeseed, maize, soy, and sunflower. Both methods yielded the acceptable levels of accuracies in the range of 80-86 %. By using seasonal composites, overall accuracy exceeded 80 %. Among crop classes, winter cereals were the most accurately classified (Producers accuracy 92 %, Users accuracy 86 %), while we observed misclassifications between soybean and maize. Furthermore, the combined use of the Landsat time series and Sentinel-1 data improved the classification accuracy by 4%. The method was tested in two years which enabled us to study the effect of inter-annual meteorological differences. Based on our results we recommend the use of seasonal composites based on harmonic regression and object-based classification to create accurate crop maps over several years.

SPOT World Heritage : images from the past for temporal analysis

SPOT 1-to-5 satellites have collected more than 15 million images all over the word during the last 30 years, from 1986 to 2015 which represent an unprecedented historical patrimony.

Today, these data are hosted in the central CNES long term archive system and in SPOT Receiving Stations.

Spot World Heritage (SWH) is the French initiative to perpetuate and valorize this archive by widely providing access to these data and providing new enhanced SPOT products.

A first step has been set off in 2016 with the start of the repatriation of all SPOT data to CNES central archive system for 3 years. Meanwhile, some preliminary SWH processing chains have been tested with the production of first orthorectified SPOT data accessible through CNES THEIA Web platform.

Now, SWH will enter a new phase in 2017 with the definition of enhanced SWH processing chains and the start of the mass reprocessing of the whole SPOT archive. Indeed, in addition to improved radiometric corrections, new SWH products will be geometrically compatible with ESA Sentinel-2 products in order to offer a better depth in time series analysis with images of the past.

SWH will thus offer 3 levels of products :

• SWH-L1A, the base archive product, including radiometric corrections (equalization, aberrant detectors/pixels, SPOTs specificities, etc.) and technical parameters (on-ground processing parameters, cloud/water masks, etc.) for a more convenient use without sensor specificities,
• SWH-L1B, the geometrically refined product using Sentinel-2 Global Reference Images (GRI) and a couple of enhanced radiometric corrections (new SPOT 5 denoising),
• SWH-L1C, the orthorectified product using Sentinel-2 configuration (DEM Planet Observer and UTM projection).

SWH-L1A processing should take place on CNES High Processing capabilities to take advantage of the location of SPOT data hosted in CNES long term archive system and thus optimize network transfer rates (export of current GERALD raw data / import of new SWH-L1A archive).

First enhanced SWH products are expected to be distributed in 2018 while the whole archive is expected to be reprocessed within 2 years until 2020, in line with SPOT data repatriation timing.

Free, open and multiple resources of remote sensing data provide the opportunity of urban expansion magnitude analysis from different dimensions. However, it remains unclear how critical the spatial resolution of the imagery is to pixel-based urban expansion magnitude estimates, and the spatial heterogeneity of multiple resolutions in measuring urban expansion magnitude is rarely discussed in geospatial analysis. This research therefore investigated how the urban expansion magnitude responds to different spatial resolutions. Multiple resources of remote sensing products collected from four main sources of earth observation satellite sensors: Night-Time Satellite data (1 km resolution), Global Human settlement layer images (300 meter and 30 meter resolution), Worldpop data (100 meter resolution) from 2010 to 2015 in the Pearl River Delta, China were used. These data were processed by fractal geometry analytics algorithm to demonstrate the urban expansion magnitude at different levels of resolutions, and then their spatial heterogeneity properties were characterized by the scaling law in fractal geometry modeling. Fractal geometry analytical algorithm in this research applies transferable scaling law across multiple resolutions to handle the complexity and heterogeneity of urban expansion spatial patterns. Finally, it offers a universal scaling pattern across multiple resource dataset at fine-, medium- and coarse-scales to provide a new overview that enables us to measure the urban expansion. The results showed that the differences of fractal dimension and self-similarity of urban expansion patterns based on multi-resolution images can be obtained from the scaling laws of fractal geometry algorithm. Our resulting patterns of urban expansion in Pearl River Delta at different spatial resolutions can be used as a basis for studying the spatial heterogeneity of other remote sensing datasets with variable spatial resolutions, especially for evaluating the capability of the multi-resolution dataset in reflecting spatial structure and spatial autocorrelation features in urban environment. This study also encourages employing the fractal geometry model as a way to effectively and systematically deal with spatial resolution issue in geospatial analyses.

The research project FLOWERED (de-FLuoridation technologies for imprOving quality of WatEr and agRo-animal products along the East African Rift Valley in the context of aDaptation to climate change) aims to address environmental and health (human and animal) issues related to the fluoride contamination in the African Rift Valley, in particular in three case study area located in Ethiopia, Tanzania and Kenya. Remote sensing data and GIS technologies are currently in exploitation for a systematic approach on data gathering and analysis for the whole interdisciplinary team.

In the framework of the project the development of a land cover mapping system is a primary key for the understanding and asses the fluoride pollution. Starting from bibliographic studies and taking into account existing international classifications and legends such as CLC and Africover projects, a land cover mapping system will be defined with a focus on the ground effects caused by pollution due to the high concentration of fluoride in water. Satellites data from Sentinel 1 and 2 will be considered as input base data for the implementation of automatic algorithms able to identify and monitor direct/indirect effects of high fluoride water concentration on vegetation (water stress, alterations of the phenological cycle of the crop, etc.) or on the ground (fluorescence due to the presence of fluoride salts). Sentinel multitemporal images will be classified by implementing supervised algorithms trained with information extracted from ancillary data or carried out from a photointerpreter.

It is provided in the FLOWERED project the transition from the land cover to the land use map through the development of a mobile application dedicated to the collection of local geo-information on land use, water uses, irrigation systems, household features, use of drinking water and the other information needful for the specific knowledge of water supply involving local communities through participative approach.

The ESA DUE Innovator III project “SAR for REDD” is an effort to provide African REDD countries with synthetic aperture radar (SAR) capabilities and supports in particular the Congolese NGO l’Observatoire Satellitale des Forêts d’Afrique Centrale. The potential of tropical forest and land cover monitoring by SAR is demonstrated on the entire Mai-Ndombe district in Democratic Republic of Congo (DRC) an area of about 127,000 km² with approximatively 75% forest cover.

The entire Sentinel-1 CSAR data set acquired since its launch has been processed into forest land cover products (FLC) over the area, classifying it into 6 dominant land/vegetation classes: forest, inundated forest, savannah, dry and wet grasslands, and river swamp. Three Sentinel-1A paths cover the whole area with a dense time series over the most western paths and a 1-2 yearly coverage of the two other eastern paths.

Results show clearly the advantage of a frequent data acquisition in regard to speckle noise reduction and less noisy classification results. Dense time series could even provide information on burn and slash activities that have been observed in the Kwamouth region during a field mission in September 2016.

A validation of the FLC products is done using very high resolution optical satellite data from the SPOT5/Take5 program and Pleiades images and data collected by a remotely controlled quadrocopter and ground observation.

We also investigate the detectability of forest change from Sentinel-1 and compare the results with L-band SAR data from the Japanese ALOS-2 PALSAR-2 and Landsat-based yearly loss data from the Global Forest Change archive (Hansen et al., 2013).

Studies of climate change indicate a significant increase in temperature in continental Siberia. In south-central Siberia, within 51-55°N and 88-92°E, the temperature of January and July temperature and annual moisture index (AMI, a ratio between GDD>5°C and annual precipitation) trends were analyzed from 1961 to 2010 based on instrumental data of 10 stations. The trends of both temperature series were positive and showed that north of 51°N, January temperatures increased 1-2°C and July temperatures increased 0.7-1.5°C over the last 50 years. The rain pattern was complicated by 2010 over the complex topography in the south. The AMI trends showed increased moisture or remained stable supporting the forest portion of the forest-steppe zone in south-central Siberia. The influence of climate warming on the natural and anthropogenic ecosystems includes alterations in the areas of forest and agricultural lands. To identify the forest/non forest boundary we used medium-resolution (30 m) imagery from Landsat 4, 5, 7 and 8 covering southern regions of Central Siberia. Overall temporal data coverage was from the late 1980-s until 2015 and 2016 including several periods 1989 – 1992, 1998 – 2003 and 2013 – 2016. For each of these periods we identified main vegetation classes, such as forest, non-forest (including steppes and agricultural lands) and mountain tundra using Landsat imagery. Several spectral features such as surface albedo in the near and short-wave infrared range, as well as the value of NDVI were calculated to distinguish vegetation classes. The main types of vegetation cover in the region showed generally good level of separability in the selected spectral feature space. Unsupervised clustering algorithm ISODATA was used to obtain separate land cover classes using surface albedo and NDVI as an input images. To validate the obtained interpretation results we compared our Landsat-based vegetation map to "Landscape map of the Altai-Sayan ecoregion" (Samoylova, 2001) based on geobotanical data. The geobotanical categories were generalized in three: tundra, forests and steppe (including agricultural lands). The kappa-based comparison of the Landsat image to the vegetation map showed their fair similarity (k= 0.4).

This study was supported by the Russian Foundation for Basic Research grant 16-05-00496.

Quality assurance is a key process in the development of land cover and land use products derived from remote sensing. This process involves comparing a reference database of sample sites with a land cover or land use product via a confusion matrix and the calculation of overall and class specific evaluation measures. The reference data can be field observations or derived from expert interpretation of very high resolution satellite imagery and aerial photographs. The latter are increasingly being used in the validation of land cover and land use products as a way of reducing the high costs associated with field data collection and because it is possible to gather much larger amounts of reference data due to the availability of imagery from Google Earth, Bing and an increasing amount of aerial photography provided through Web Map Services. Many in-house tools exist for land cover and land use validation, which have been developed across institutions around the world, but there is still a lack of open tools available for this task as well as a lack of reference data. For this reason, we have developed the LACO-Wiki tool for online land cover and land use validation (http://www.laco-wiki.net/). In addition to validation, the aim of the tool to create a community of map validators, who are happy to share their reference data collected via LACO-Wiki with the broader scientific community. LACO-Wiki will also facilitate the sharing of validated maps as well as the possibility to share aerial photographs for specific countries/regions. LACO-Wiki is aimed at a range of users including researchers and students as well as commercial map producers. The tool takes users through a simple four-step workflow as follows:

1. upload a raster or vector map for validation containing either categorical or continuous data;
2. create a random, stratified random or systematic validation sample, either point- or feature-based, or upload an existing sample that has been created using another system;
3. interpret the sample in a validation session using Google Earth, Bing imagery, OpenStreetMap, Sentinel-2 or images supplied by the user via a Web Map Service; and
4. produce a report, which contains the confusion matrix and the accuracy statistics, which includes overall accuracy, producer’s and user’s accuracy, among others. The raw data from the validation session can also be downloaded as well as the sample locations.

Users can specify the size of the validation sample or use the sample calculator, which can determine the number of samples needed based on the level of uncertainty specified, i.e. the confidence intervals. Once the validation is undertaken, LACO-Wiki can also be used to augment the sample, by determining which classes need additional samples in order to reduce the uncertainty further. Here we demonstrate the use of LACO-Wiki in validating GlobeLand30 using a stratified random sample.

Deimos Imaging is the Spanish subsidiary of UrtheCast, a Canadian Earth Observation company which owns and operates two EO satellites, Deimos-1 and Deimos-2, and launched and installed two cameras on the International Space Station (ISS).

UrtheCast is currently developing the 8-satellite UrtheDaily constellation (pronounced “Earth Daily”), designed to acquire a full coverage of the entire Earth’s landmass every day, with uncompromisingly high-quality multispectral imagery. The innovative wide-swath payload has been designed to provide 5-m data with scientific-grade radiometric and geometric quality, which will be constantly cross-calibrated with Sentinel-2 and Landsat-8.

A global network of ground stations, connected to our cloud infrastructure, will allow to download and process all the imagery within hours. The huge amount of data (140 million km2 of multispectral imagery every day, corresponding to more than 20 TB of imagery) will be available within 12 hours from acquisition through our API-based cloud platform (UrthePlatform), which is already operationally used for our current generation of sensors.

Scheduled to be launched in late 2018, and with a lifespan of 10 years, the UrtheDaily system will represent a huge leap for worldwide monitoring in terms of both spatial and temporal resolution: a daily coverage at 5m resolution provides 20 times more information than Sentinel-2, through a 5-fold improvements in temporal resolution and 4 times more pixels.

The end-to-end UrtheDaily system has been optimized for large-scale change detection and analysis applications, aimed at enabling powerful geoanalytics applications. The high-frequency, high-quality data will enable unprecedented global daily actionable insights for a wide number of verticals.

Among population and economics expanding as well as climate changing, Disaster is the major problem that seriously damages life and property around the world. Especially, when disaster occurs in Metropolitan City which has a characteristic high population density compared with surrounding areas and a number of activities in business, trade, communication, tourism and technology, it negatively impacts to tremendous people and spreads out through the country. To prepare and mitigate the effect of disaster, it is necessary to understand how risk and value are in each area.

The objective of this study is to investigate land use change as well as building height classification. We use satellite data from Corona in 1960’s, Landsat 1-7 from 1970 to 2010 for land use change study and Geoeye stereo measurements in 2013 to estimate building heights.

Land cover class was separated into five classes; (1) Urban (2) Plantation (3) Forest (4) Lake (5) Water. We applied supervised classification method for high accuracy. Mahalanobis distance method was used with Landsat multispectral images to classify. The sampling points are more than five hundred samples in each class. After three resultant of each classification; height of building, land cover, night time light, were already obtain. The hierarchy classification was used to classify the building type; (1) residential, (2) commercial, (3) industrial buildings.

The land cover change rules were employed to reduce the noises or to smooth the classification result. The first rule is urban expansion rule that other classes can change to urban class but the urban class cannot change to other classes. The second rule is deforestation rule which was defined that forest class can change to other classes but the other class cannot change to forest class. In additional, cloud effects were removed on this process by using the nearest time classification.

It was found that Yangon city land cover changes from 1960’s to 2010’s and has revealed drastic urban expansion from 38 km2 to 204 km2. A novel methodology was used to classify building type in urban area using remotely sensed data in Yangon, Myanmar. Stereo images of GeoEye was used to provide the height of building. Based on height of building, building class was obtain. Landsat-8 image was applied to provide land cover area. Using land cover, result of building class was improved. We combined building class with night time light data. Three building classes; residential, commercial, industrial buildings, were obtain. The rule of distance between commercial and industrial areas were defined to improve classification result. To validate the map of building classification, local images in residential and commercial buildings were compared as well as the industrial zone map in Yangon. The comparison between resultant map and validated data has high relationship.

In this study, spatio-temporal patterns of continuous paddy fields were examined using the patterns observed in metrics calculated from time-series of MODIS and AMSR-2 dataset over global scale. Four analytical approached were used; calculation of temporal mean, maximum and minimum layers for selected metrics showing significant spatial variability of channel 1, 2, NDVI to understand vegetation phenology over the area. MODIS 8 day composite normalized difference water index (NDWI) and AMSR-2 daily 16km normalized difference polarization index (NDPI) are used to map land surface water coverage (LSWC) which is effective to monitor agricultural irrigation and innundation.; linear discriminant for input into the un-mixing analysis was derived from the same multi-temporal metrics used for the classification product using ASTER, AVNIR2 and Landsat; the continuous percentage of paddy field was generated based on un-mixing technique with the training data derived from the above mentioned ASTER, AVNIR2 and Landsat data. The derived metrics were not sensitive to time of year or the seasonal cycle and can limit the inclusion of atmospheric contamination. The comparison of MODIS and AMSR2 product with the past efforts on global land cover product using AVHRR, SPOT-VEGETATION and MODIS sensors, and statistics by IRRI showed that the finer resolution and its un-mixing played a crucial role in depicting the paddy field cover over global scale.

Landcover maps from satellite data are of high demand for tundra regions. Of special interest are shrubs and wetlands, as well as the occurrence of mosses and lichens. Applications include up-scaling of carbon fluxes and pools, permafrost feature mapping, transition and pasture monitoring.

There is to date no commonly aggreed classification scheme but regional maps usually represent
shrub physiognomy (dwarf species or higher) and wetness patterns. Shrubs are included as a class in global land cover maps, but are either not present in their Arctic parts or patterns differ among them. Wetlands or wetness descriptions are in general under-represented in global maps. Mosses and lichens are omnipresent, growing along with grasses, sedges and shrub or as cryptogam crusts in varying fractions. They are usually only addressed locally as own classes when high spatial resolution satellite data are used.

The thematic content of existing global maps has been investigated by comparison to the
CAVM (circum arctic vegetation map, hand-drawn from AVHRR, Walker et al. 2002) and regional maps (Bartsch et al. 2016a). None of the global maps (GLC2000, ECOCLIMAP, CCI Landcover) provides the required thematic detail.

Spatial resolution has been compared to used classes for local to regional applications. The required thematic detail increases with spatial resolution since coarser datasets are usually applied over larger areas covering more relevant landscape units. This is especially of concern when the entire Arctic is addressed.

Circumpolar landcover is so far mostly addressed using optical data. Recent studies have shown that C-band SAR (Widhalm et al. 2015, Bartsch et al. 2016b), especially roughness and volume scattering behaviour, can reveal information about landscape units which are of relevance for a range of applications.

This contribution reviews existing schemes and discusses potential new strategies for landcover mapping North of the treeline. It contributes to the ESA DUE GlobPermafrost project. Especially user requirements for permafrost related applications are therefore addressed.

Walker, D.; Gould,W.; Maier, H.; Raynolds, M. The Circumpolar Arctic Vegetation Map: AVHRR-derived base maps, environmental controls, and integrated mapping procedures. Int. J. Remote Sens. 2002, 23, 4551–4570.

Bartsch, A.; Höfler, A.; Kroisleitner, C.; Trofaier, A.M. Land Cover Mapping in Northern High Latitude Permafrost Regions with Satellite Data: Achievements and Remaining Challenges. Remote Sens. 2016, 8, 979.

Widhalm, B.; Bartsch, A.; Heim, B. A novel approach for the characterization of tundra wetland regions with C-band SAR satellite data. Int. J. Remote Sens. 2015, 36, 5537–5556.

Bartsch, A., Widhalm, B., Kuhry, P., Hugelius, G., Palmtag, J., and Siewert, M. B.: Can C-band synthetic aperture radar be used to estimate soil organic carbon storage in tundra?, Biogeosciences, 13, 5453-5470, doi:10.5194/bg-13-5453-2016, 2016.

The identification of shrublands is of interest for various applications in tundra regions, including permafrost modelling. Not only the presence of shrubs is relevant but also the height. It serves as indicator for snow redistribution (and thus snow depth) as well as for monitoring vegetation succession on active layer detachments. Their occurence is often related to terrain, especially wetness.

Especially central Yamal is characterized by shrubs of 1,5 m height and larger. It is an area with continuous permafrost and many retrogressive thaw slumps in the proximity of lakes. Past studies revealed the suitability of L-band SAR to map shrubs using soil moisture as proxy (Dvornikov et al. 2016) as well as X-band due to volume scattering in canopy (Widhalm et al. 2016). C-band is expected to represent both moisture and volume scattering. In general shrubs are so far addressed using optical data for tundra landcover (Bartsch et al. 2016).

For this study, C-band SAR from Sentinel-1 as well as all bands of Sentinel-2 have been investigated in order to develop a shrub mapping scheme which represents shrub height. This contributes to the objectives of the joint Austrian-Russian project COLD Yamal (FWF and RFBR) as well as the ESA DUE project GlobPermafrost.

Validation data have been collected in summer 2014 and 2015. Measurements from various Sentinel-2 bands as well as common indices including NDVI and tassled cap have been investigated. C-band is assessed for frozen and unfrozen conditions as well as different incidence angles.

It can be shown that NDVI is not applicable for this purpose. Both Sentinel-1 and Sentinel-2 can support identification of shrublands but crucial are spatial resolution and acquisition timing.

Dvornikov, Y. , Leibmann, M. , Heim, B. , Bartsch, A. , Haas, A. , Khomutov, A. , Gubarkov, A. , Mikhaylova, M. , Mullanurov, D. , Widhalm, B. , Skorospekhova, T. and Fedorova, I. (2016). Geodatabase and WebGIS project for long-term permafrost monitoring at the Vaskiny Dachi research station, Yamal, Russia, Polarforschung, 85 (2), pp. 107-115 .
doi:10.2312/polfor.2016.007

Widhalm, B., Bartsch, A., Leibmann, M., and Khomutov, A.: Active Layer Thickness Estimation from X-Band SAR Backscatter Intensity, The Cryosphere Discuss., doi:10.5194/tc-2016-177, in review, 2016.

Bartsch, A.; Höfler, A.; Kroisleitner, C.; Trofaier, A.M. Land Cover Mapping in Northern High Latitude Permafrost Regions with Satellite Data: Achievements and Remaining Challenges. Remote Sens. 2016, 8, 979.

According to FAO, Land Cover (LC) can be represented using basic atomic elements, chosen using simple physiognomic criteria, rather than categories. Such elements (e.g., Tree, Shrub, Herb, Building, etc.) can be recombined into categories representing different ontologies. LC is then represented in a database by Basic Objects that can be further characterized by ‘Properties’ and ‘Characteristics’. This approache is fully described in the Land Cover Meta-Language (LCML) that became in 2012 an ISO/DIS standard (ISO 19144-2). The LCML Basic Elements, their relationship, inheritance and the properties and characteristics associated to them, are formalized in a Unified Modelling Language (UML) class diagram. Any user, applying the LCML rules and conditions, should be able to create compatible Object-Oriented Land Cover databases (Di Gregorio 2016). This can assure integration of local knowledge and basic semantic interoperability for re-interpretation and re-use of existing maps in combination with new updated maps in loing-term studies (change detection and validation).

The present paper describes an application of the LCML concept to improve the harmonization of existing LC national mapping and forest inventory in Bangladesh in support of forest ecosystem managementby the means of a new software framework. Such framework is called the Bangladesh Information System (BIS).. In particular the paper will describe a new “object based” methodology to make an automatic similarity assessment between, LCML derived classes, present in different data bases. The “object based” automatic semantic similarity assessment approach can be a first decisive step toward an automatic procedure for data bases harmonization.

The software architecture is composed by a web application (BIS client) and a web service (BIS service). The former can be accessed using a common browser and presents in a user-friendly way the core functionalities supplied by the latter. Underneath, the two components interact through the means of an HTTP API based on the LCML-based data model; the BIS service makes use of a geo-database to store georeferenced data and a native XML database to store and transform the LCML collections according to the user requests. The LCML-based data model is encoded using XML schema, in order to leverage the query capabilities of a native XML database (e.g. XQuery, XPath). The use of native XML technologies seems a reasonable choice to enable the system scaling, maintaining unaltered the full set of information that is available in LCCS3/LCML. Future steps of this work include the application of BIS functionalities on even larger LCCS3 legends (e.g. output of automatic classification systems).

In the frame of Sentinel-2 program, Earth observation (EO) images having thirteen spectral bands, acquired at 10m, 20m and 60m spatial resolutions, are available under a free for use license. Due to this fact, we can observe an accelerated development and increasing interest of methods and algorithms for EO image analysis. Since the first release of Sentinel-2 images, our interest was focused on developing feature extraction methods and algorithms that can be used in land-cover classification of multispectral data. The methods we propose are considering texture and spectral analysis, considering in the computation all the spectral bands available in the Sentinel-2 data product. By doing so, we increase the accuracy of the results and obtain good classification scores even for areas affected by clouds and their shadows. The feature extraction methods we present can be used on other EO multispectral data sets from high to low spatial resolution. Furthermore, our methods can be used for thematic map generation in which several classes like agriculture, vegetation, forest, water, urban, etc. are required at a reasonable accuracy.

Mount Etna (eastern Sicily, Italy) is one of the most active basaltic composite stratovolcanoes in the world and it has the ability to change its land field rapidly, vigorously and continuously. It is capable of producing both brief paroxysmal episodes and long-standing, relatively milder eruptions. During the last 100 years Mt. Etna has produced on average 10⁷ m³ of new lava per year, both from its summit craters and from its flank areas.

The aim of this study was to distinguish and map lava flows belonging to different effusive and explosive episodes within both the Mongibello (Torre del Filosofo formations: 122BC-1669AD, 1669AD-1971AC, 1971AD-2007) and the Ellittico (Portella Giumenta, Monte Calvario, Piano Provenzana, Pizzi Deneri formations) volcanic supersystems (Branca et al., 2011), combining TOPSAR polarimetry with spectral information from optical satellite data. Furthermore, we investigated the potential relations between backscatter coefficient (σ⁰) variations and spectral information to topographic (elevation, slope, aspect) settings.

For this purpose, we analyzed two Sentinel-1A scenes, GRD (Level-1), IW in single (VV) and dual (VH) polarization both in ascending and in descending geometry, acquired on 22/8/2016 and 16/8/2016, correspondingly, and one Sentinel-2 image (Level 1C) acquired on 21/8/2016. Pre-processing for polarimetric SAR datasets included radiometric calibration, orthorectification and despeckle filtering, whereas for optical data it included atmospheric correction, dimension and noise reduction (PCA).

Vegetation masking was performed on both datasets using the NDVI map calculated from Sentinel-2 data. Training sets for each lava flow were collected from chronologically different units according to the most recent geological map of Etna volcano (scale 1:50.000). For each formation, both the mean backscatter coefficient and the standard deviation were calculated for the three polarizations, namely VV, VH and the ratio VV/VH (σvv, σvh, and σvv/vh), both for ascending and descending geometry. For the same training sets, the mean spectral signature and the standard deviation were also retrieved from the Sentinel-2 dataset.

Preliminary results on the mean backscatter have shown that there are differences between the ascending and the descending pass for most of the lithostratigraphic units, especially when concerning the ratio VV/VH. These differences can be very significant for some lavas, whereas for others they seem to be less significant. This is probably due to differences between the satellite viewing geometry and the local topography orientation. For this reason, we compared the backscatter coefficients with the average slope and aspect of each unit using the DEM of the area.

On the other hand, spectral signatures present similarities between units with similar physicochemical properties. However, differences in the overall reflectance magnitude can help differentiating between younger and older lavas. Therefore, spectral signatures seem to be related more to petrology and grain size than to topography, as it is the case in the polarimetric backscatter.

The main conclusion of this study is that a synergistic approach using both optical (Sentinel 2) and SAR polarimetry data (Sentinel 1) can be a suitable tool for land cover mapping, especially in areas with significant topographic highs and with highly contrasting physicochemical characteristics of surface rocks.

The UN-REDD program uses satellite data of different sources for the worldwide mapping of forest cover and changes to it. Most of the technological approaches address forest cover loss and most of them use Landsat satellite imagery. Its spectral capabilities, suitable spatial resolution, huge data archive and not at least free availability make it a preferred information source for REDD. Although, for monitoring of forest degradation a higher spatial resolution is required. Sensors like RapidEye, Sentinel-2, SPOT and others can be used. The combined use of such data, especially for building up long time series, requires some special solutions for data processing.

The current paper is about the operational challenges of using long time series of Landsat, RapidEye, SPOT and Sentinel-2 data for forest degradation monitoring in REDD. It bases on the experiences collected during an ESA funded research project for “Forest Degradation Monitoring with Satellite Data – ForMoSa”. The project team of Planet Labs Germany, University of Wageningen and Forestry Department of FAO developed, tested and demonstrated a technology for forest degradation mapping in 3 natural environments: Ethiopia, Vietnam and Peru. For each of them long time series of Landsat and RapidEye imagery enriched with Spot-5 and Sentinel-2 were used to a) model the seasonal behavior of forest environments in these areas and b) identify forest areas, which behave differently.

Both tasks need to be solved for each pixel separately, using the information from all available satellite images from the past. This implicates, that all satellite images must be perfectly co-aligned and must be spectrally comparable to each other. To ensure this for each pixel for time serious of more than hundred multi-source satellite images is a data managerial and computational challenge, even if making use of high-performance cloud computing environments like the ESA rss cloudtoolbox service. It was successfully implemented for the mentioned areas and may serve as an example for other researchers having similar tasks in time series analysis.

Operational products for global agricultural monitoring will play an important role in addressing some challenges that the world will face in the coming years. The capabilities of the Sentinel satellite missions with short repeat frequency and large coverage offer new opportunities to study the whole diversity of the agricultural landscapes. These perspectives are important from multiple perspectives such as researchers', farmers', regional and national authorities' or policy makers'.

The recently launched Sentinels Synergy for Agriculture (SensAgri) project¹ aims to exploit the unprecedented capacity of the radar and optical payload of the Sentinel-1 and Sentinel-2 satellites to develop an innovative portfolio of prototype agricultural monitoring services. This project is financed by the European Commission under a H2020 program with the objective to propose and complement Copernicus land services. In the framework of this project, more robust, accurate, frequently updated and comprehensive crop maps are expected from the seldom exploited synergy of both types of measurements. Specifically, one major goal of the project is to develop a prototype service capable of mapping crops several times across the agriculture season.

In the past, few studies integrated both radar and optical dense image time series due to low data availability. However, the complementarity of these two types of information for crop classification has been widely proved in the literature. The most valuable advantage of radar data is that they are not affected by cloud coverage, thus allowing continuous multiple measurements along the short dynamic growing season of crops.

The preliminary results assessing the capabilities of combining Sentinel missions for crop monitoring will be the main focus of this presentation.

¹http://www.sensagri.eu

Sentinel-2 MSI (MultiSpectral Instrument) is super-spectral instrument of the European Space Agency (ESA), that can provide data continuity to Landsat data series for global land surface monitoring. Several simulation studies have been carried out during last years to show the potential of Sentinel-2 MSI. Now that one-year Sentinel-2 data are fully available, it is possible to assess this potentiality by comparing data from the two satellite sensors. In order to allow a proper comparison of the datasets, this paper focuses on the analysis of atmospheric correction workflows of Landsat 8 and Sentinel-2 data. In particular, two different workflows have been considered for each satellite: the standard workflow (USGS LaSRC for Landsat8 and Sen2Cor for Sentinel-2) and a proposed workflow based on the 6S radiative transfer model. The results of processing have been tested over agricultural sites by using Landsat 8 and Sentinel-2 images collected in the same day. In particular have been extracted and analyzed vegetation indices that are commonly used for vegetation monitoring

Sentinel-2 Global Land Cover (S2GLC) is an ESA SEOM project with the aim to define a scientific roadmap and recommendations for creating a Global Land Cover (GLC) database based on Sentinel-2 data. The database will bring a new quality of information to the Remote Sensing user community by exploiting Sentinnel-2 capabilities. The project began in 2016 comprising four partners: CBK PAN, IABG, EOXPLORE UG and FSU.

When fostering methods of satellite image analysis according to the increasing technological capabilities new satellite technology offers, these methods should systematically build on previously achieved efforts. Simultaneously, they should continue recent methods and take advantage of the new technology’s temporal, thematic and spatial capabilities. After an extensive review of currently available GLC databases, as an initial step of this project a preliminary hierarchical legend of land cover classes was defined, which intends to benefit from the high temporal, spatial and spectral resolution of Sentinel-2 in combination with Sentinel-1 data.

Subject of this presentation is the preparation of suitable reference data used in the classification and validation process performed within this study and beyond. For this purpose, five test sites were selected globally, enabling us to react sensible towards regional variety. The test sites are of significant extent and heterogeneous in their environmental character resulting in heterogeneous land cover classes. They are located in: Italy, Germany, Namibia, Colombia and China. Additionally, the quantity and quality of already existing classification results is very heterogeneous. Supplementary regional data sets (for example LUCAS, CORINE) with less accuracy than the envisaged data set but of higher accuracy than existing GLC data sets were considered as additional thematic information

Recent investigations took place on selected subsets. Overall goal is to identify and develop strategies to combine available GLC data and additional thematic layers in order to reduce manual effort. The latter will eventually support the set-up of an overall roadmap regarding a Global Sampling Database.

Investigations consider available global classification and validation results in combination with the coarse analysis of up-to-date Sentinel2 data supporting information extraction suitable for classifications and validation issues. Main aim within the project is to support the multi-temporal classification process, which is subject of a different presentation.

Sentinel-2 Global Land Cover (S2GLC) is ESA SEOM project with aim to define a scientific roadmap and recommendations for advancing towards the automatic production of Global Land Cover (GLC) maps based on Sentinel-2 data. The project began in 2016 by four partners: Space Research Centre of Polish Academy of Sciences (CBK PAN), Industrieanlagen-BetriebsgesellschaftmbH (IABG), EOXPLORE UG and Friedrich-Schiller-Universität.

Sentinel-2 data introduce new spatial, radiometric and temporal capabilities which can profit in more detailed global classification information. On the other hand, in order to maximize the profit, new challenges should be addressed: a user-driven legend definition, the collection and structure of reference training and validation samples into a S2 Global Reference Database, the different strategies to characterise the spectral, temporal and spatial information of pixels and parcels, the potential classification strategies, the potential synergies with other datasets (e.g., S1), the strategy for a reliable validation. The presentation will focus on overcoming some of the key problems.

A preliminary hierarchical regional/Eco-climate-based legend of land cover classes has been defined at prototype level (for the purpose of the project) including 3-levels after extensive review of the currently available GLC databases. Five test sites have been chosen to test and validate the applied classification techniques: Italy, Germany, Namibia, Colombia and China (each about 200 000 km²).

The proposed approach is based on the generation of a S2 Global Reference Database including multi-temporal samples of the different land cover classes considered in the legend for each of the eco-climatic regions defined. Different strategies for the generation of such a database are under investigation, including the potential to maximise the degree of automation of the sample identification from existing GLC databases vs. manually selected data and how the data could be filtered in order to increase its reliability.

The study of the different methodological strategies to perform the classification follow different paths in order to get a good overview of the potential problem areas and possible candidate solutions. Therefore the study concerns issues such as classification under a limited number of training samples, the exploitation of different features including multi-temporal information, the limits associated to the classification of areas where no training samples are available exploiting information from different regions, the potential to exploit multiple classifier systems, the combination of supervised and unsupervised techniques.

In this presentation we show the advantage of using multi-classifier approach over single classifier. As an examples of the experiments carried out, we present the multi-classifier aggregation workflow adjusted for cloud cover problems employing both single- and multi-temporal features.

Since few years, thanks to technological progress and free access policy development, thematic databases are becoming more and more accessible, gaining in resolution (temporal and spatial) as well as in spatial extent. These databases have been built either by compacting information from various sources either with the exploitation of EO data. Until recently, when using EO data, global databases were derived from low (ie SPOT Vegetation instrument) to moderate resolution data (ie MODIS, MERIS systems). Now with the opening of Landsat archive, ASAR ESA archive, SPOT World Heritage initiative, or with the on line direct access to OLI data, the global datasets are high resolution ones (30 to 90m). In a short future these would increase with the large scale availability of Sentinel constellation data.

In addition to providing information on specific life sciences domains, these databases can be also exploited within automatic selection of appropriate training samples as well as validation purposes (eg Sentinel exploitation). 42 databases containing multi-scale information on water bodies have been inventoried: 28 having a worldwide coverage, 9 a continental one (Europe), 4 a national one (France) and 1 regional level. This inventory provides for each database an index indicating the source, thematic water bodies nomenclature, geographic extent, publication date, update frequency etc.

Databases assessment has been performed over two tests sites corresponding to two flood plains over which studies based on EO data are carried since more than 15 to 30 years, the Alsace Flood plain and the Lorrain lakes (FR), representing a “normal” temperate landscape including semi-mountainous relief, the Vosges Mountains, and the Yangtze flood plain including the Poyang Lake and the Anhui lakes (PRC), monsoon lakes having a huge intra annual and inter annual dynamic in term of height and extent.

25 databases have been assessed. For each database, water bodies’ elements were compared with reference data and indicators combining the accuracy and level of detection were generated. Then, each database comparison received a certain number of point (1 to 3), depending on the following criteria: accuracy, spatial resolution, reliability/validation, updating frequency, age of used data, depth of the exploited data. A final global note was obtained by summing these results with a weighting factor, from 6 to1 depending of the given importance of the criteria knowing that accuracy of the database, resolution and reliability/validation were defined as the most crucial ones.

At global scale, the rankings obtained were for training purposes: Pekel 95% and Pekel 1% (not published), followed by Global inland water (Feng et al.2015) and From GLC (Gong et al., 2013) and for validation purposes: Pekel 1%, then From GLC, Pekel 95%, Global inland water, G3WBM (Yamazaki et al. 2015). Medium resolution products such as the ESA GlobCover were in the top ten database. At European level, the Permanent Water Bodies of the HR layers produced by the Copernicus Land Monitoring services is ranked at the first place, just before the Urban Atlas.

The availability of medium to high resolution optical imagery has increased considerably with the advent of the Sentinel-2 mission and the open data policies adopted by several earth observation data providers. The characterization of dynamic land surfaces can greatly benefit from the improved temporal information that dense, intra-annual time series provide. These data allow capturing seasonal variation of land surface reflectance and important stages of phenology, which allow for an improved mapping of many cover types. Especially the mapping of cropland, grassland and other semi-natural covers can greatly be improved. However, as the density of observations through the growing season is primarily subject to cloud cover, the combined use of Sentinel-2 and Landsat data is a valuable option for assembling dense, intra-annual time series. Temporal compositing is an important toolset to transform the irregular multi-sensor data stream into a set of equidistant and consistent cloud free image datasets. It remains to be determined if such intra-annual time series of reflectance composites are the most valuable input features for mapping or if alternative feature sets, e.g. such that are directly derived from the multi-sensor data stream, perform equally well or even better.

Here we evaluate the potential of almost two years’ worth of observations from Sentinel-2 and Landsat-7/8 in an experimental setup for land cover and crop type mapping for the state of Brandenburg, Germany. Temporal compositing is used to derive image composites at a 10-day interval based on bottom-of-atmosphere reflectance data. Compositing is performed with two sets of target spectral bands, the S2 MSI land bands, where missing bands in Landsat are spectrally interpolated, and “Landsat legacy” spectral bands. After compositing, a multi-step temporal gap filling is performed on the 10-day composite time series to further improve the observation density. We compare the 10-day interval time series against different feature sets: (1) monthly composites, (2) seasonal composites, (3) band-wise temporal-spectral metrics using different temporal windows derived directly from the input data and (4) the 10-day interval time series transformed into fractional cover estimates. We target a comprehensive class legend including 10 specific crop and grassland classes, three forest and three built-up classes and parameterize different Random Forest classification models using the various input features sets. We finally evaluated the different feature sets in terms of classification accuracy and variable importance. Results show that, in general, crop classes benefit greatly when data from all relevant phenological stages is included and achieved best results with the full 10-day time series. Including MSI red-edge bands led to a slight improvement over Landsat legacy bands for grassland and fodder crop classes. The second best performing feature sets were the monthly composites, followed by the three monthly band wise metrics. Less dynamic classes such as forests and built-up areas achieved comparable accuracies with less temporal features (e.g. seasonal composites and monthly composites, respectively). Overall the results provide valuable insights on the surplus value of dense intra annual information derived from Sentinel-2 and Landsat for crop and land cover mapping at high spatial resolution.

With the constantly increasing volume of remote sensing (RS) data, this becomes a powerful tool in addressing the challenges and improving our understanding of the Earth system. In particular, with launches of Sentinel-1, Sentinel-2, and Landsat-8 satellites, there will be generated up to petabyte of free high special resolution raw images per year. These images and derived products are extremely important for many applications in climate change, food security, and large-scale land cover and land use mapping. On the other hand, the increasing volume of remote sensing data, dubbed as a “Big Data” problem, creates new challenges in handling datasets that require new approaches to deal with it [1]. In past years there has been a large boost in developing advanced machine learning techniques, in particular deep learning (DL). DL is a powerful machine learning methodology for solving a wide range of tasks arising in image processing, computer vision, signal processing, and natural language processing. Within this technique, we propose a deep learning method, based on convolutional neural network (CNN) approach, using Tensorflow library and geospatial analysis [2]. Experiments are carried out for the Joint Experiment of Crop Assessment and Monitoring (JECAM) test site in Ukraine for classification of crops in a heterogeneous environment using time-series of images acquired by Sentinel-2 and Sentinel-1 remote sensing satellites.

[1] M. S. Lavreniuk, S. V. Skakun, A. J. Shelestov, B. Y. Yalimov, S. L. Yanchevskii, D. J. Yaschuk, and A. I. Kosteckiy, “Large-Scale Classification of Land Cover Using Retrospective Satellite Data,” Cybernetics and Systems Analysis, vol. 52, no. 1, pp. 127–138, 2016.

[2] N. Kussul, A. Shelestov, M. Lavreniuk, I. Butko, and S. Skakun, “Deep learning approach for large scale land cover mapping based on remote sensing data fusion,” IEEE International Geoscience and Remote Sensing Symposium (IGARSS), pp. 198-201, 2016.

Keywords: Deep learning, CNN, Tensorflow, Remote sensing, Crop mapping

Land cover is one of key terrestrial variables used in large-scale economic land-use and ecosystem models. Global Land Cover datasets provide critical information in global, regional and national scale applications. In the last decade, several global land- cover products were developed, including the Global Historical Land-Cover Change, Land-Use Conversions, the Globeland-30, Corine-2012, and GlobeCover-2009. Recent studies have demonstrated that when global land-cover products are compared there is a significant spatial disagreement across land-cover types. Also, there exist substantial semantic differences in the legend definitions. The use of different satellite sensors and classification methods, and the lack of sufficient in-situ data are some of the reasons for the disagreements.

The goals of the Global Observations of Forest Cover - Global Observations of Land Cover Dynamics (GOFC-GOLD) South Central and Eastern European Regional Information Network (SCERIN) include improving cooperation in developing methods for monitoring the dynamics, stability, and vulnerability of the major ecosystems for future effective sustainable management and preservation. The geographic domain of the SCERIN encompasses a large region of South Central and Eastern Europe (SCEE), which is characterised by extreme diversity of landscapes and environmental conditions. This project leverages on the network collaboration and expertise in the region.

The goals of the project are to evaluate the accuracy of the global medium-resolution (10-60m) land-cover products in the SCERIN geographic area, and to identify the specific regional differences and their underlining causes. In this presentation we will discuss the strategy for regional product inter-comparison and validation, demonstrate prototypes of the validation products, and provide preliminary findings using case studies from Greece and the Czech Republic.

The regional validation of the global land-cover products is of high importance for advancing the LCLUC research in SCERIN, since it will enable the inter-comparison of large-scale measurements and analysis of the land-cover parameters and natural processes. Accurate estimates of the state and dynamics of land-cover maps are needed for environmental change studies, land-resource management, climate modelling and sustainable development. The project is critical for achieving the goals of the international GOFC-GOLD program and aligns well with the agenda of the NASA Land-Cover/Land-Use Change Program to improve the quality of space-derived higher-level products.

Forest disturbances from wind, insects, drought, and wildfires have increased in Europe over the last century and are likely to continue to increase in the future. To successfully adapt management and policies to these future challenges will require improved monitoring of forest disturbances and their impacts on ecosystem structure and function. The frequent, high-resolution observations of the Sentinel-2 mission are expected to greatly enhance the detection and attribution of forest disturbances. Studies using Landsat data have shown that transient (e.g. defoliator, forest health) and abrupt (e.g. harvest, fire) forest changes can be identified with dense, long-term optical time series observations. However, most of these studies were conducted in North America and Australia; there have only been few applications in Europe, and those were mostly focused on Eastern Europe. Since forest ecosystems, disturbance agents, and forest management vary greatly in Europe, there is a strong need to build and test robust change detection approaches that are applicable over large areas.

The objective of this study is to test the generality and applicability of time-series based change detection across a range of forest disturbance and management regimes in Central Europe. We use all available Landsat data to construct cloud-free, annual time series for every year between 1985 and 2016. Producing consistent, annual time series was only feasible with the combined USGS and ESA data holdings. The reprocessing of the ESA Landsat archive greatly improved the combined usage of the two data sets, while it was still necessary to remove remaining geometric differences. To the Landsat time series data, we applied the LandTrendr algorithm to map and characterize annual forest disturbance trends, including the timing, duration and change magnitude. To assess map accuracy and to provide, statistically sound estimates of area and area changes at the country level, we developed a probability-based reference data set. The reference dataset consists of a random sample of Landsat pixels, for which a photo-interpreter records the disturbance and land-use history using Landsat image chips and aerial photography. At the ESA WorldCover conference, we will present first results from the mapping and country-level estimates of disturbance history in selected European countries. We conclude with a perspective for the Sentinel-2 mission. The results from this study will aid the on-going European-wide efforts, in line with the Copernicus Land Monitoring Service, to harmonize and operationalize forest monitoring for climate change mitigation and adaptation, sustainable forest management, and environmental protection.

Although the value of ecosystem services provided by wetlands is well demonstrated, they continue to disappear globally. Lack of spatial and temporal information to guide conservation and management strategies is a common challenge. The release of the Landsat archive and now the Sentinel satellites are allowing the generation of higher quality spatial and high cadence temporal information, and even optical and SAR data fusion. We use Landsat 5 and Sentinels-1 and -2 imagery to map and reveal Land Use Land cover trends that have been occurring in the Kilombero floodplain, Tanzania, since the 90’s. Farm encroachment has already transformed over 50% (350.000 ha) of the natural grasslands and wetlands in the floodplain during the last ~25 years, and the trend is expected to continue. Fusing mapping results from Sentinel-1 and Sentinel-2 we have been able to separate temporarily inundated grasslands from non-inundated grasslands. This is important for two reasons: inundated grasslands are vital for several endangered species of mammals (e.g. puku antelope), and these areas are unlikely to be transformed into farmland any time soon due to the high cadence of floods. However, climate change and water management upstream might change this in the future.

The systematic monitoring and map production service that SWOS is providing is proving essential, especially in areas where information is scarce such as tropical wetlands. SWOS mapping products of Kilombero are being currently used to support the development and implementation of strategies for the sustainable management of large wetland landscapes.

More work is still necessary to produce a systematic global coverage of wetland trends. Inconsistencies defining land cover-classes occurred using images from different years and different sensors, and more will be found when comparing products that use different nomenclatures (e.g. maps produced to fulfill national, European or international obligations). Building on standardization efforts like the Land Cover Meta Language for the definition of classes might overcome the inconsistencies derived from using different methodologies, sensors, or sources of information.

Arid wetlands are recognized as the temporary inundated territories, located in the zones of arid and semi-arid climate which are the habitats for very specific and diverse ecosystems. Representing the local biodiversity hotspots, the arid wetlands may be called as ‘the lands of contrasts’ where over-moistened territories are adjacent to deserted areas. Among all the wetland types, arid wetlands are the most sensitive to environmental changes and human impact, as they are often the only source of fresh water and other natural resources in arid regions. They are of a great social importance in terms of ecosystem services, and they are critical in regional and local biodiversity conservation in arid or semi-arid lands.

Desertification and freshwater ecosystem loss, occur in response to climate change, environmental disturbances and human impact are the most urgent problems in many arid wetlands, so its monitoring by remote sensing is crucial not only for understanding of their functioning and seasonality, but also for the development of the standards for the wetland estimation, land use regulation, and prediction of the response of wetland ecosystems to the future environmental changes.

Generally, the following basic research fields for satellite imagery application for wetlands exist: delineation and inventory of wetlands, habitat mapping and monitoring, ecosystem health estimation (seasonal and long-term dynamics, biomass estimation), desertification monitoring and land use monitoring (change detection).

This overview is based mainly on the methodologies of original researches on satellite data applications in the large arid and semi-arid wetlands located in lower flow of the rivers Volga, Terek, Sulak (Russia), lower Ural (Kazakhstan), and Indus River (Pakistan).

Remote sensing techniques we focus on, include the methods for delineation of wetlands, image classification and tasseled cap transformation methods adjusted to arid wetlands, the analysis of the correlation between water and vegetation indices, modeling of the dynamics of vegetation indices in different kinds of wetland vegetation, the evaluation of different parameters of the vegetation functions, and the study of their relationships with hydrological parameters and the weather. This allows predicting the inundated areas during floods and dry periods, and the ecosystem state, facilitating the local land use and environmental activities planning, based on the runoff parameters, at each specific site within the wetland.

Due to this extensive review, it is possible to conclude that in spite of the fact that EO satellites provide highly valuable datasets, there are many challenges and limitations exist in the use of such kind of data for the arid wetlands. In this framework, some complementary data, as hydrological parameters, is very helpful in arid wetland research. There is still not much information exist about the ways for satellite imagery applications in arid wetlands. Our experience also suggests that existing methodologies of optical satellite imagery applications used in the arid wetland researches are very laborious, but they are all-important in the context of climate change. Finally, the arid wetlands being extremely sensitive to water intake, are found to be an indicator of water regime, which can reflect the water scarcity in certain territories.

Forests in the Polish Carpathians are characterized by a significant heterogeneity of species composition, stand age and forest property. Furthermore, due to mountainous character of this area, the common problem here is occurrence of topographic effects. Forests mapping in this area is therefore a challenging task, even using satellite data with high spatial resolution. The main aim of this study was to develop approach which allows to classify forest stand species composition with overall accuracy above 80%. Two test sites were selected as the study area, both located in the Polish Carpathians: the Baligród Forest District and the Ujsoły Forest District. The first test site, Baligród Forest District (about 305 km²) is located in the eastern part of the Polish Carpathians. Over 70% of this area is covered by forests with the large proportion of mixed forests. Dominant species here are common beech (Fagus sylvatica) and silver fir (Abies alba). Other essential tree species found in this area are grey adler (Alnus incana), Norway spruce (Picea abies), Scots pine (Pinus sylvestris), common hazel (Corylus avellana) and sycamore (Acer pseudoplatanus). The second test site, Ujsoły Forest District (about 270 km²) is located in the western part of the Polish Carpathians and it is characterized by greater homogeneity in terms of species. Dominant species in this area are Norway spruce and common beech.

We used all available Sentinel-2A imagery from years 2015 and 2016 and performed classification in hierarchical approach. As the reference data for classification we used the National Forest Inventory Data (cycle II, 2010-2014; WISL) and the Forest Data Bank (BDL). Use of multitemporal and relatively dense series of data, hierarchical segmentation and classification slightly increase the overall accuracy of obtained tree species maps.

We gratefully acknowledge support by the National Science Centre, project RS4FOR [project no. 2015/19/B/ST10/02127].

Knowledge about land cover and land use (LULC) is important for advancing in earth system science, for decision making and natural resource management, as well as for environmental monitoring and reporting obligations. Such information is particularly required at high spatial resolution (<= 30m) for national to global scales. In this context, the data and acquisition characteristics of the Landsat and Sentinel-2 missions both following free and open data policies enable the derivation of high-resolution LULC information over large areas from optical remote sensing data. Furthermore, for Europe, abundant geo data on LULC characteristics are freely available, such as the LUCAS survey conducted by EUROSTATS or CORINE Land Cover provided within the framework of the Copernicus Land Monitoring Services. However, only little experience exists in using these open geodata sets for large scale land cover generation processes, particularly with respect to their usability as training data base for supervised classification approaches.

Against this background, we present results on our efforts to derive land cover information at 30m resolution using Landsat 7 and Landsat 8 imagery in combination with freely available open geodata. The results were derived based on a fully automated preprocessing chain that integrates data acquisition, radiometric, atmospheric and topographic correction, as well as spectral–temporal feature extraction for all Landsat surface reflectance bands, brightness temperature and various spectral indices. We also consider auxiliary layers as input features for the classification, e.g. elevation and slope. A Random Forest classifier is used for the land cover classification, due to its good performance in terms of computational cost and accuracy. Training of the classifier will be based on open reference data sets that provide area wide coverage over Europe, such as the EUROSTATS LUCAS survey. Furthermore, a methodological approach will be presented to enhance and extend the available training dataset by an automated and efficient allocation of a relatively small number of additional relevant samples to be labelled by an expert, thereby keeping the manual user interaction to a minimum. In addition, first research efforts will be presented to investigate how to best utilize freely available reference data at continental scales. Either reference data of the whole region can be used to train a single classification model or a range of local models are trained and applied for different sub regions.

The demonstrated approaches and findings contribute towards the advancement and enhancement of automated or semi-automated processing workflows in combination with open geodata, which are directed towards the generation of land cover products at regular intervals being of central importance to related land monitoring and reporting services.

CadasterENV Sweden is a project that aims at producing a national land cover (LC) database at a reasonable cost. To ensure long-term usefulness, a LC database must be updated on a regular basis in a cost-effective manner. Therefore, an effective LC change system is crucial to reaching the goal of a regularly updated land cover map.

The project incorporates a large user group consisting of both stakeholders/users and researchers and was conducted in close cooperation with the Swedish Environmental Protection Agency, Statistics Sweden and the Swedish Board of Agriculture. The Data User Element (DUE) program of the European Space Agency (ESA) funded the project, which started 2012 and resulted in the development of new and improved mapping methods, as well as production of LC maps within several test areas in different parts of Sweden.

The classes and attributes of the LC data model are based primarily on analyses of user requirements in combination with a determination of feasibility given robustness and repeatability. Data from Sentinel-2 have been used for several reasons including increasing the opportunity to acquire cloudless images, classifying images based on time series and conducting change analyses both within one growing season and between years. Sweden has a shorter vegetation season than most other European countries. This limits the possibility to acquire cloud free images during the vegetation season and, even more so, the optimal date (from a phenological perspective). Finding comparable dates from two different years is also greatly hindered. If time series from the vegetation period is available, this provides an improved opportunity to identify vegetation types by analyzing phenology. On the plus side, Sweden’s high northern latitude corresponds to a relatively high revisit frequency from Sentinel-2, resulting in a large amount of data that can potentially be utilized.

The product has a large variety of possible uses. Some examples are statistics, monitoring, planning and analysis and as a base for more detailed mapping in areas such as sustainable land use and biological diversity, ecosystem services, outdoor life and public health, climate and disaster risks and community planning. We foresee many more possible users for the product in both the private and public sectors. In the private sector, real estate, infrastructure and telecom markets are the major potential users.

Currently, Metria is integrating Sentinel 2 images into the production chain and coordinating the user group on behalf of the Swedish Environmental Protection Agency to prepare for a full size production at a national level. The production is planned to start in 2017 but the mapped test areas have already been used as a base for biotope mapping and landscape analysis. The methodology and the know-how is thoroughly tested and robust and can be implemented in other countries.

Already immerse in the Sentinel era, the ESA Research and Service Support (RSS) offers several services to make easier to users and researchers to get original and processed Sentinel data. The ESA RSS offers several service solutions to make bulk data processing using consolidated algorithms and algorithm development/ testing as well.

The RSS service offer is composed of several elements supporting different phases of the research process flow. It includes e-collaboration environments to find and share information, reference and sample datasets, access to a huge EO data archive without the need to download on scientist or developer “own” resources, customised cloud toolboxes where scientists and developers alike can fine-tune their algorithms on selected datasets, on-demand processing environment where fine-tuned algorithms can be integrated and made available as EO applications for on-demand massive processing, and results visualization tools.

As example of the consolidated algorithms, the ESA RSS at this moment offers services in its on-demand environment for manipulating the Sentinel-1 and Sentinel-2 data to provide processed Sentinel products.

These on-demand services are based on the ESA open source software such the Sentinel Application Platform (SNAP), making use of the Sentinel Toolboxes and also offering a EO tool developed by EUCENTRE and University of Pavia called SENSUM framed on a FP-7 funded project, which the current integrated algorithm on the ESA RSS infrastructure perform the Stacking of Sentinel-2 data for built-up area extraction.

In addition, the RSS CloudToolbox service offers Virtual Machines (VMs) with enough resources to work with Sentinels data, being the perfect solution for researchers and SME working in the algorithm development/testing and also for the researchers with need of getting short term extra resources for processing purposes.

Using this not exhaustive list of services offered by the ESA RSS, the researchers obtain benefit in terms of productivity, reducing the storage, timing and processing costs inherent to the research process.

Sugarcane plays an important role in the world’s food and energy market place; it is the number one crop in the world in terms of production quantity (FAOSTAT, 2014). Brazil is the world’s number one producer of sugarcane, which provides fuel for around 40% of the national gasoline market. It is projected that ethanol production in Brazil will increase from 22 billion (2011) to 46-65 billion liters in 2020.

However, an increased deployment of biomass for energy and materials could have significant adverse socio-economic and environmental impacts. Various initiatives for sustainability impact assessments related to bio-energy crop expansion have addressed the importance of the monitoring of direct and indirect land cover (LC) in order to identify and quantify their impacts such as GHG emission, food security, biodiversity, competition for water, etc. LC maps have proven to be essential as input for these models requiring spatial explicit information. Research activities conducted during the last decades led to the dissemination of a variety of LC products, each with different class taxonomy, temporal revisit, spatial resolution and geographic support, logically influenced by the satellite datasets used and the main targeted applications. It should however be noted that the existing products still cannot not meet every specific study requirement. This in particular applies when high demands in terms of temporal update and class discrimination are formulated.

We constructed a land cover change monitoring technique, based on a specifically designed Markov chain model digesting Landsat data. The classifier makes distinction between sugarcane, annual crops, forestry, pasture, urban and water. For the model construction, training and validation, more than one thousand fields in the São Paulo region were inspected, spread over two growing seasons in 2015. For the application on the whole São Paulo state, roughly the size of the United Kingdom, a Wide Area Processor was developed for acquiring, pre-processing and stacking of the remote sensing data over large regions. The Dutch national supercomputer is used to handle the processing; our resulting maps from 2003 to 2015 based on Landsat data can be found here: http://be-basic.grs.tudelft.nl/maps/316/view

An inter-comparison with global maps (GlobCover) and country-specific maps (TerraClass and Probio from Embrapa et al., and Canasat from INPE) will be presented to highlight benefits and challenges from the perspective of the sustainability analysis. Currently, the TUDelft group is at the stage of integrating radar data to improve the LC maps. Sentinel-1 Interferometric Wide and Extra Wide mode scenes (both HH+HV) are used as well as 20 Radarsat-2 Fine Quad and 16 Standard Dual Pol (HH+HV) scenes provided by ESA grants. Preliminary work shows that radar profiles of annual crops, sugarcane, pastures and forests provides clear potential for enhanced classification with respect to only optical-based classification suffering from cloud coverage-induced temporal gaps.

For our research, we work together with Utrecht University (the Netherlands) for integration with their socio-economic and sustainability impact models and with UniCamp (Brazil) for map comparisons, agronomic expert advice and ground campaigns.

Dependencies between vegetation cover dynamics and climate dynamics and change are described in many studies. Usually, data of climate monitoring are used to investigate and forecast the vegetation cover change and phytomass productivity, and climate change investigation is based upon observations at the meteorological stations and reanalysis data. It is obvious, on the other hand, that vegetation cover response to the climate change can be used as a marker when studying climate. This idea can be very helpful when studying peripheral areas in regional and local scales. In Russian Subarctic, for example, network of meteorological stations is too sparse, and it leads to significant errors in estimations of regional features of climate dynamics by traditional ways. Our study (conducted in 2013-2016) shows (as a number of other studies) the fundamental possibility of regional climate change monitoring based on vegetation dynamics observation and proposes methodology for such monitoring. One of the components of vegetation cover based climate change monitoring is the observation of vegetation cover parameters during the separated growing seasons. Namely, we used three growing seasons, which are the spring season (when surface air temperature exceeds +5 °C, and most of plants in studied region begin to grow their phytomass), summer season (when surface air temperature exceeds +10 °C, and the active phytomass growing begins), and autumn season (when surface air temperature remains between +10 °C and +5 °C, and phytomass growing processes degrades).

Main issue is how to allocate the framing calendar dates of the growing seasons in the case of absence of the ground observation data. To resolve this issue we proposed to use the Normalized Difference Water Index (NDWI). The technique is based on the analysis of the annual graphs of NDWI, which is sensitive to the liquid water in plant tissue. Having time series of satellite imagery applicable for NDWI maps production, we can produce the maps of spatial distribution of the growing season framing dates and duration on pixel-by-pixel basis. In our study we produced such maps for the test area located in the Komi Republic (Northern part of European Russia) for the period of 2000-2015.

Human activities have altered the global carbon cycle profoundly and non-reversibly. The necessity to avoid or at least lessen the negative effects of the ensuing environmental and climate change through mechanisms of adaption and mitigation has been acknowledged by international and national organizations and governments and by researchers of various disciplines. Since deforestation and degradation of tropical forests contribute about 10% to 15% of GHG emissions, there is a huge potential for mitigation through carbon sequestration in the tropical forests. Therefore the REDD mechanisms were created to support afforestation, reforestation, forest conservation and simultaneously an improved forest management to mitigate GHG emissions. Apart from being sinks for carbon there are additional reasons to protect forests from conversion to other land use types: They play important roles in climatic processes on different scales, they are a habitat to many animals and plants, thus taking an important role in biodiversity conservation, they prevent soil degradation, decrease flood and landslide risks, improve water availability and quality, in many places they have important recreational and cultural functions and they provide livelihoods and food in often poor rural regions. All this desired ecosystem services depend on the health and often on the spatial context of the forests. Due to that it is not only important how much forest is saved or restored, but also where and in which context these areas are located. Sometimes it might be possible to realize several benefits in one area. But since land is a limited resource competition is inevitable. To gain insights into possible future land-use changes and arising competition and synergies the global land-use model LandSHIFT was developed. LandSHIFT is a dynamic, integrated model that uses bottom-up and top-down mechanisms to simulate land-use change on the global and regional scale. Since LandSHIFT uses land cover maps as input the quality, resolution and availability of land cover maps is crucial for the quality of its output. We studied the influence of land cover map quality on output uncertainty. Furthermore we mapped where missing land cover maps or land cover maps lacking in quality prevented us from modelling on a higher resolution.

Madeira Island (Portugal), an European Outermost Region, constitutes a biodiversity hotspot due to its high number of endemic/native plant species. In this work we developed and assessed a methodological framework aiming to produce a Rapideye multispectral imagery-based vegetation map. Pixel and object-based Maximum Likelihood supervised classifications were applied to eight initial datasets stacking Rapideye multispectral bands from December 2009 (winter) and August 2011 (summer) and three derived vegetation indices: the classic NDVI, the already tested reNDVI and a newly tested NDVIre, respectively, in order to test the full potential of the Rapideye red-edge band for vegetation studies. The classification scheme to be used was categorized in three levels for a total of 7 first level (more generalized) classes, 12 second level classes and 26 third level (more detailed) classes. A total of 1244 ground truth points have been collected for the December 2009 dataset and a total of 1341 ground truth points have been collected for the August 2011 dataset. The random stratification procedure for the August 2011 datasets left respectively 931 points for classification and 410 points for accuracy assessment. This same procedure for the December 2009 datasets left respectively 851 points for classification and 393 points for accuracy assessment. The best results were obtained by applying the object-based approach to the Rapideye dataset of August 2011 including the Red Edge-NDVI band (best Kappa coefficient: 0.754) or the NDVI band (best Overall Accuracy: 74.77%). These methodological approaches applied to Rapideye high-resolution multispectral imagery were confirmed as cost-effective procedures for mapping and monitoring vegetation in Madeira Island.

Land cover mapping in Latvia is performed as part of the Corine Land Cover (CLC) initiative every six years. The advantage of CLC is the creation of a standardized nomenclature and mapping protocol comparable across all European countries, thereby making it a valuable information source at the European level. However, low spatial resolution and accuracy, infrequent updates and expensive manual production has limited its use at the national level. As of now, there is no remote sensing based high resolution land cover and land use (LC-LU) services designed specifically for Latvia which would account for the country’s natural and land use specifics and end-user interests.

The European Space Agency launched the Sentinel-2 satellite in 2015 aiming to provide continuity of free high resolution multispectral satellite data. On the one hand, Latvia as one of the Northern countries can benefit from more often (double or even triple) coverage of each ground point due to overlapping Sentinel orbit swaths, but on the other hand, the region’s frequently cloudy sky presents a challenge for work with optical data. For example, for monitoring of agricultural/grassland management practice there are 4 to 5 cloud free scene from the 2015 vegetation season, while in 2016 there is a huge data gap during the peak vegetation period with cloud free scenes available in April and them only in August 25 and September 14. Nevertheless, the algorithm for classification of land cover and land use from single date data and combination of data from spring and summer periods was developed accounting for national end-user needs.

It was observed that potential end-users have different background and experience working with remote sensing data. Some of them were used to work with high resolution (< 1 m/px) orthophotos and manual classification approaches and were sceptical about usability of relatively low resolution (10…20 m/px). Others pointed to the lack of experience working with Sentinel and other remote sensing data, but indicated their willingness to develop their competency. It was observed that demonstration of data product examples followed by discussion and end-user involvement is required for the development of attractive algorithm and data products. Most of Latvian end-users are interested in annual LC-LU data products with the highest possible spatial resolution.

The outcomes of the project “Simulation of Sentinel-2 Images for Land Cover/ Land Use Monitoring Using Hyperspectral Airborne Remote Sensing” financed by the Plan for European Cooperating States (PECS) programme of the European Space Agency (ESA) will be presented – the algorithm for high resolution LC-LU classification in Latvia from Sentinel-2 optical data, data product demonstrators according to Latvian user needs and lessons learnt.

The extent and condition of forest ecosystems in Armenia have decreased drastically since the disintegration of the Union of Soviet Socialist Republics in the early 1990s. This decline is not only a consequence of the recent history of the area, but also the result of decades of forest policies and management practices. To reverse the negative trends, it is important for stakeholders, scientists, resource managers and policy makers to have quantifiable information on forest area, yet there is distinct lack of a reliable official national statistics on forest area. This lack of reliable information has been identified as a key challenge in improving the current conditions of the forests and the forestry sector in Armenia. This paper explores the utility of combining local and global forest maps with post-stratified and model-assisted estimators to increase the precision of estimates of forest area in north and north-east Armenia. The post-stratified estimators produced estimates of greater precision than the model-assisted regression estimators for maps of categorical variables, but the model assisted estimators produced estimates of greater precision for maps of continuous variables. The Global Forest Change 2000-2014 map was the least accurate of all the maps, but it produced estimates of forest area that were similar to those for the other maps and that were more precise than if the map had not been used. It is hoped that this approach demonstrates how such processes can be integrated into national forest monitoring and reporting to increase precision and reduce costs.

Despite being recognized as a key baseline dataset for many applications, especially those related to biogeochemical cycles, operative and thematically oriented land cover products are limiting. Typically they lack either the thematic details necessary for driving the models that depend upon them or some information/ required resolution for minimizing the uncertainties of estimated parameters. This study has demonstrated the ability to produce a highly detailed (both spatially and thematically) land cover/land use dataset over Russia – by integrating existing datasets (regional and global; based on remote sensing and subnational statistics) into a hybrid dataset using a stepwise approach. By combining diverse data sources into a single land use/ land cover product, it became possible to produce a map that is consistent with national statistics and more accurate than the individual input layers.

In this paper we applied geographically weighted regression (GWR) at first step to integrate twenty different maps into a hybrid forest map at a 150 m resolution for the reference year 2010. Input products included global land cover, forest and agriculture maps at varying resolutions from 30m to 1km. The GWR was trained using crowdsourced data collected via the Geo‐Wiki platform. At the second step we calculated the probability of every land cover class and then calibrated class-by-class to regional statistics. The hybrid map contains 12 land use/land cover classes: forest, open woodland, arable, abandoned arable and other agricultural land, wetland, grassland, shrubs, burnt areas, water and unproductive areas. The hybrid map was then validated using an independent dataset collected via the same system. The obtained product had the best overall accuracy when compared with the individual input datasets. The hybrid land cover/land use map will be available at http://forest.geo-wiki.org/.

Keywords: land cover; land use; remote sensing; land statistics; Russia

Grassland is one of the most common agricultural land use land cover types. For example,1.88 million hectares (29%) of Latvia’s territory is covered by agricultural land of which 0.65 million hectares (34%) are covered by permanent grasslands (meadows and pastures). The Common Agriculture policy (CAP) stipulates that EU Member States have to designate permanent grasslands, ensure that farmers do not convert or plough them and that the ratio of permanent grasslands to the total agricultural area does not decrease by more than 5% in order to receive support payments. However, semi-natural grassland habitats require appropriate management activities to ensure their long-term conservation. The European Commission report (2015) required by the Birds and Habitats directives concludes that ‘grasslands and wetlands have the highest proportion of habitats with an unfavourable-bad and deteriorating status’ in the EU, while the midterm review of EU biodiversity strategy 2010-2020 highlighted that grassland habitat change presents high risk to biodiversity. Latvia’s rural development programme (2014-2020) has identified only 47 thousand hectares of biologically valuable grasslands. These grasslands are semi-natural meadows and pastures that include species and habitat types of EU importance. 70-90% of EU importance grassland habitats in Natura 2000 sites were in poor condition in Latvia during 2012. There is a clear interest from a number of end-users (e.g. the Rural Support Service, the Nature Conservation Agency) for grassland mapping and management practice monitoring solutions.

The Institute for Environmental Solutions has started the implementation of the ESA PECS project “Assessment of Grassland Quality and Quantity Parameters and Management Activities Using Sentinel-1&2 data (SentiGrass)” in 2016. It is aimed to explore the capability of Sentinel-1 radar and Sentinel-2 optical data use and fusion for the assessment of grassland management activities and quantitative/qualitative parameters, thus moving towards the development of multi-functional grassland surveillance and monitoring tool. The first stage of the project consisted of user need analysis (information gathering and exchange, on-site meetings as well as involvement in in situ data acquisition), definition of data product specification as well as identification of pilot territories and data acquisition (in situ, airborne) during vegetation season in 2016. Further involvement of potential end-users is planned after the first milestone at the beginning of 2017 in order to gain feedback and plan next activities. It is planned to present the progress of the project – analysis of user requirements and demonstrators of data products with feedback from potential end-users.

Land cover is a crucial variable that impacts on human life and environment. Currently the world has witnessed the importance of land cover and the changes of it that has significant efficacy on biogeochemical cycling which can cause natural consequences such as global warming. Understanding the importance of land cover and it’s changes is particularly limited to accurate data. Remote sensing knowledge proposes a unique mapping ability which covers wide range area while capturing the information on land, atmosphere and water resources of earth imagery processes. At the present, with the improvement of the integrated spatial techniques the procedure of land cover studies and classification dynamics has become quick, cost-effective and accurate. Due to this, remote sensing science which is an efficient source in order to achieve the target of land cover classification, can provide more accurate data toward past decades and is a powerful tool to improve land cover map accuracy. The classification process converts satellite data to operational maps and products that attracts the attention of natural resources managers, scientists and researchers. Also is important in current strategies and policies for natural resources management. This study focuses on classification of land cover in Nazlu district located in the west side of Urmia Lake in northwest of Iran. The classification was done by using two different satellite images derived from Landsat 7 (ETM+ 2005) and Landsat 8 (OLI 2013) data in order to compare the classification results. The method used is supervised classification, maximum likelihood tool using ENVI 4.5 software. Principal component analysis (PCA) and NDVI vegetation index was used to enhance the correctness and resolution of classified maps. Maximum likelihood classification method was done using ground truth points in order to extract more accurate maps. The results showed that in Landsat 8 data, the kappa coefficient rate is 0.8 and overall accuracy of the classification is 87.2 % and in ETM+ images kappa coefficient rate is 0.9 with overall accuracy of 90% which proofs that remote sensing is a trustworthy science that can provide accurate results and maximum likelihood classification is credible tool to extract the land cover maps. Comparing the changes of land cover area results of ETM+and OLI showed significant changes in agriculture that increases from 180 km² to 565 km² no cover areas has reduced considerably from 1253 km² to 664 km² and orchards has been increased from 59 km² to 277 km² during the study period. Results showed that both OLI/ETM+ are capable for LULC mapping.

Due to the continuous development of technological means, the quality of the delivered remote sensing data is always improving, better and better spatial resolution being available with each new system. This improvement in spatial resolution requires new methods and algorithms for processing and analyzing remote sensing data. With respect to low resolution PolSAR data, parametric methods have been widely used in the state-of-the-art literature. These methods rely on data's stationarity hypothesis, and in order to extract some parameters one has to assume an a-priori probability density model. Therefore, the results of different operations are strongly dependent on the validity of the stationarity hypothesis and on the goodness-of-fit of the applied pdf model. In the new context of high resolution data, the stationarity hypothesis does not hold anymore, so nonparametric methods have to be employed. In PolSAR data processing, the most common approach is applying coherent and incoherent target decompositions of the polarimetric representations (the scattering or covariance matrix), which would represent each backscattering mechanism as the sum of some simpler, canonical mechanisms. However, with the continuous decrease of the resolution cell, the spatial context of the recorded objects has become very important for detecting semantically meaningful categories. Polarimetric decompositions can be successfully applied to individual pixels or to small neighborhoods, but they are not suitable to large image patches, because the use of an averaging operator would cause a significant loss of information. In the present work we aim at presenting a number of methods and algorithms to overcome these issues, as follows:

• The Independent Component Analysis is employed for PolSAR data incoherent target decomposition. It is shown that applying this method on VHR PolSAR data can reveal the targets underlying backscattering mechanisms;
• A Convolutional Deep Belief Network is proposed for PolSAR image feature extraction. The stepwise discovery of higher level features leads to robust feature descriptors, which provide very good classification results;
• The BiQuaternion Fractional Fourier Transform and the Method of Moments are employed for PolSAR image feature extraction and classification. The joint use of the biquaternion algebra and of a patch-based approach enables both the spatial and polarimetric correlations embedded in PolSAR data, leading to very good classification results.
• A new feature descriptor is introduced, consisting of a Bag-of-Meaningful-Words. This feature descriptor is independent of the stationarity hypothesis of PolSAR data, extending the use of the EntropyAnisotropyAlpha classification method to VHR PolSAR image patches;
• Latent Dirichlet Allocation is used for unsupervised discovery of semantic relationships in PolSAR image patches. The discovered relationships help the image mining systems to adapt their results to human semantics, narrowing the so called semantic gap.

All these methods and algorithms are nonparametric, they are not biased by the data's stationarity hypothesis or by the goodness-of-fit of an applied pdf model, and they take into consideration the spatial context of the recorded objects, being able to detect semantically meaningful categories. These characteristics recommend the proposed methods for processing and analyzing high and very high resolution PolSAR images.

The estimation of forest aboveground biomass is important to understand carbon flow between trees and the atmosphere. Remote sensing plays an important role in making this possible in particular for relatively large or hard to reach areas such as mountainous regions. The main aim of this study is to compare and evaluate forest aboveground biomass models obtained using Sentinel 1 and/or 2 data. The test area covers the entire Polish Carpathians. In our study we will use all available Sentinel 1 or 2 images for two years 2015 and 2016 and their fusion. As reference data we will use forest aboveground biomass models obtained using LiDAR dataset from the ISOK project (year 2013; digital terrain model, digital surface model and canopy height model) and the National Forest Inventory data (cycle II, 2010-2014). The Random Forest algorithm will be used to build all models.

We gratefully acknowledge support by the National Science Centre, project RS4FOR [project no. 2015/19/B/ST10/02127].

In the context of natural or anthropic risk analysis at a global scale, worldwide information is needed on human settlements characterization, especially concerning population distribution and urban zones localization. This information is of prime importance for the knowledge and description of at-risk elements in modelling and risk assessment scenarios. Urban thematic information can be extracted from available global, regional or continental open sources geo-databases (e.g. ESA GlobCover, ESA CCI Land Cover, Natural Earth, Copernicus core services products, OSM, …) or derived from the exploitation of EO satellite images at low, medium, and high spatial resolution (e.g. MODIS, Landsat, Sentinel2, …) over territories where this type of information is not available or not sufficiently up to date.

Within the framework of the CNES POPSCAN ongoing study focused on space operations launch and re-entry risks assessments, which includes in particular the monitoring of safety requirements for people and property, a census with a critical review of existing open sources geo-information databases has been realized. Besides the analysis of existing global population databases, a particular focus has been made on global landcover geo-databases containing urban areas information and localization. The different characteristics of these databases, such as geographic coverage, reliability, consistency, accuracy and resolution, update frequency, availability and cost, or delivery format, have been recorded for a first qualitative assessment approach. Then, with the use of reference EO datasets, quantitative analysis of the thematic accuracy of the most pertinent geo-databases (eg. ESA CCI Land Cover, DMSP ISA, GLCNMO, FROM GCL) have been performed over 6 different test sites representing different types of landscape and urban settlements (Canada, SW France, French Guyana, Indonesia, Philippines, Africa). These comparative results highlight the potential and limits of these global geo-databases but it can be noticed that, among the different "urban areas" databases analyzed, and for the test zones covered, the one that offers the best compromise between accuracy rate and detection rate is the ESA CCI Land Cover 2010.

Urban land cover and land use (LUC) change is the most irreversible and human-dominated form of LUC. Urban Atlas data (UA) project allows the broad research community to use these data for the identification and evaluation of LUC change within cities and their surroundings. However, there is a general absence of UA data driven analysis and the assessment of urban LUC change. It is also not clear how the measured changes are supported by the other data sources. For this reason, the aims of this study are:

- to document the possibilities of identification, analysis, and assessment of LUC change using UA and cadastral data (CD),

- to identify and evaluate the LUC change in functional urban areas (FUAs) of Bratislava and Prague in the period of 2006–2012,

- to compare trends and intensity of change in Bratislava and Prague in relation to the social-economic development of both capitals (Pazur et al. 2016).

Results obtained via the analysis of the UA data layers were compared with the official statistical data of the real estate cadastre of the Slovak Republic and the Czech Republic. The differences of spatial (Urban Atlas) and non-spatial (cadastral data) datasets representing two time horizon periods 2006 and 2012 were evaluated.

The common phenomenon in the development of the FUAs Bratislava and Prague documented by UA and CD data is the marked increase of built-up areas and shrinkage of farmland. Both FUAs clearly show trends presented in modern industrial cities – extensive conversion of agricultural plots into residential or industrial areas and the development of such areas near high-quality infrastructure (motorways). The CD for FUA Bratislava 2006–2012 points to a considerably higher shrinkage of agricultural land than the UA data (3382 ha CD vs. 1316 ha UA). Contrary, in FUA Prague, the CD displays a substantially smaller uptake than UA (3408 ha vs. 4445 ha).

Both the CD and the UA identified a significant increase in built-up areas. Pursuing the CD, the size of built-up areas in FUA Bratislava was 15,402 ha (in 2006) and 16,434 (in 2012); meanwhile, the size of classes urban fabric, industrial, commercial, public, military and private units, and mine, dump and construction sites identified by the UA data, was 24,181 ha (in 2006) and 25,497 ha (in 2012). The size of built-up areas in FUA Prague according to CD was 18,818 ha (in 2006) and 19,454 ha (in 2012); meanwhile, the surface area of urban classes identified by means of the UA data was 86,913 ha (in 2006) and 91,025 ha (in 2012) (Pazur et al., 2016).

The UA data are an important output of the Copernicus Land Monitoring Services and results confirmed the utility and potentiality of the UA data for the analysis of changes that took place in two European cities in the period of 2006–2012.

References

R. Pazur, J. Feranec, P. Stych, M. Kopecka, L. Holman. 2016. Changes of urbanized landscape identified ba the Urban Atlas data: Case study of Prague and Bratislava. Land Use Policy, 61, 135-146. http://dx.doi.org/10.1016/j.landusepol.2016.11.022

Global land cover maps currently provide an urban mask, which differentiates between built-up and non-built-up areas. While there has recently been a great improvement in the resolution and precision of such global products, e.g. the Global Urban Footprint and the Global Human Settlement Layer, many applications require more detailed information about urban land cover since most human activities are concentrated in urban areas and the variation in structure, cover, materials and function is larger than for any other global land cover class.

A specific area in need of such data is urban climate science so that climate models can be applied to the development of climate change adaption and mitigation strategies. The World Urban Database and Access Portal Tools (WUDAPT) initiative was designed to address this data gap by collecting information about the form and function of urban areas on a global scale using remote sensing and crowdsourcing.

Although the variety of urban structures and the heterogeneity of materials cause large differentiation in climatic impacts, they also result in various spectral characteristics, which make them difficult to classify. This has led to numerous different approaches, mostly based on different typologies of urban structural types (UST), often specific to the application and area of interest. This limits the comparability of such studies across different locations. Thus a common and generic description of urban structures is an essential step towards a universal mapping scheme of detailed urban characteristics. The Local Climate Zone (LCZ) scheme was originally introduced for characterization of urban surface characteristics for urban heat island studies. It provides a good framework for the discretization of urban areas on a kilometric scale, is largely free of cultural or climatic bias, and additionally delivers a large number of quantitative properties for each of its constituent classes, which can be used directly by urban climate models. Furthermore, it has been demonstrated that LCZs can be classified using openly available, high resolution satellite data.

In this presentation, the need for more detailed land cover data for urban areas is highlighted, the LCZ scheme is outlined as a standard for UST classification, and the WUDAPT initiative is introduced.

The NextGEOSS H2020 project, starting in January 2017, proposes to develop an European centralised hub for Earth Observation data fully compatible with the GEOSS information system and data policies where the users can connect to access data and deploy EO-based applications. The concept revolves around providing the data and resources to the users communities, together with Cloud resources, seamlessly connected to provide an integrated ecosystem for supporting applications. A central component of NextGEOSS is the strong emphasis put on engaging the communities of providers and users, and bridging the space in between.

To provide a proof of concept of those capabilities several thematic community hubs, covering a wide range of EO related application areas such as Agriculture, Biodiversity and Energy, will develop and deploy specific services appliances within the advanced distributed NextGEOSS ICT infrastructure. These appliances are in different readiness level but are expected to be demonstrated in operational environment as system prototypes during the project. Those services, as well as their outputs, will be registered and published in the NextGEOSS associated Community Portals, together with the respective data and service policies.

Within those appliances there will be one dedicated to High Resolution Mapping for Territorial Planning, focused in urban environment Land Use/Land Cover (LULC) mapping. These appliance will make available a suite of tools, based on the legacy of the Deimos Engenharia and Deimos Space UK LULC applications, allowing the user will be able to customise the main steps needed for LULC classification and feature extraction, namely: a) define the type of satellite imagery to use; b) define the classification scheme; c) use already existent training and validation data or upload their own; d) compare different classification or feature detection algorithms; e) run change detection algorithms on an image time series; f) visualize all the outputs in map and graph form; g) edit the output to correct possible classification errors and; h) rerun the classification to include user corrections as new training data. The web based application will provide the user an integrated environment to access all these different functionalities required to produce a baseline LULC map, a change detection map or a detailed feature map (e.g. roundabouts, trees, roads, etc) with the added advantages of being able to access the NextGEOSS image catalogue and cloud processing capabilities. A demonstration of these suite of tools will be done for at least two pilot case study areas, Lisbon Metropolitan Area, based on a set of Sentinel, Landsat and Deimos-2 images, and the city of Dubai, taking advantage of the work already developed there by deimos Space UK and the respective available imagery

Since NextGEOSS is in its start, the poster will focus on presenting the concept of this tool in the context of the NextGEOSS activities, its proposed architecture and the presentation of past results, based on the Deimos Engenharia and Deimos Space UK previously developed applications.

The detailed and accurate land cover classification, mapping and change analysis emerge as essential for several scientific communities working on climate change studies, sustainable development and natural resources management. At the same time open data policies both in the USA and EU are delivering an unprecedented volume of satellite imagery data with an increasing level of detail and accuracy. Currently, the availability of Landsat-8 and Sentinel-2 data significantly increase the capabilities for high resolution land cover mapping with multitemporal, multispectral optical data. Once these datasets are fully combined and integrated in the same spatial scale, approximately one image per week will be available at most geographical regions. Under such a framework, critical features and information like vegetation phenology, can be adequately modeled by machine learning classifiers resulting into higher land cover mapping accuracy. To this end, in this paper we qualitatively and quantitatively compare land cover products derived from all annual (2016) Landsat-8 and Sentinel-2 data at two different spatial resolutions i.e., 30m and 10m. Firstly all annual surface reflectance data were interpolated at 30m or pansharpened at 10m spatial resolution. Then with a common training dataset which was manually annotated at 10m resolution, the same SVM and deep learning classifiers were employed for land cover classification. The two resulting land cover maps at 30m and 10m resolution are discussed and compared based on the resulting confusion matrices regarding the classification procedure and accuracy, based on the validation set which was also manually annotated at 10m resolution.

Throughout the years, Earth Observation satellite missions have gathered large amounts of data. These include multiple acquisitions over the same surface area creating SITS (Satellite Image Time Series), which represent an important resource in the process of monitoring land cover evolutions. Considering the large amount of data involved, these evolutions are almost impossible to be manually detected and thus the usage of automated systems is required. The problem with fully automated systems is that they usually don’t know what the user’s needs are. This can be resolved with the involvement of the user in the evolution detection process. In order to improve the evolution detection metrics a solution can be the creation of a relevance feedback loop which queries the user in regard to the proposed solutions and give him the option of indicating if the results are good or not which in turn will help the system become better at providing adequate results for the sought land cover evolutions. In this paper we test this concept on a data set which covers parts of the south-east Romania over a period of almost three decades. We are interested in the evolution of man-made land cover elements over the given period and the information generated can give an overview of the trend in which urban development has occurred in Romania.

In this paper we present the preliminary results obtained by the implementation of a processing scheme using satellite images and neural networks architectures to provide, and regularly update every three months, global land cover maps.

Aiming at keeping high both the level of automation and the final accuracy, the scheme consists of four steps. In the first step the area is divided in a certain number of tiles. In the second step for each tile a pixel based classification is performed using a multi-layer perceptron neural network (MLP-NN) algorithm [1]. By mosaicing all the classified tiles we obtain what we called the “Master” land cover map. In the third step the update of each classified tile is addressed using new satellite data. To this purpose a change detection algorithm based on Pulse Coupled NN (PCNN) is considered [2]. PCNN is a relatively new technique based on the implementation of the mechanisms underlying the visual cortex of small mammals. Only the changed pixels detected by the PCNN are reclassified with the MLP-NN and the whole updated land cover map is obtained. In the fourth step an accuracy evaluation of the final product is carried out.

This un-supervised and automatic land cover map update system is very challenging due to the amount of satellite data to be processed. Initially the algorithm will be deployed on flexible Cloud resources where it will be fine-tuned and re-engineered in order to support a scale-up and massive EO data processing., This approach allows to take full advantage of the dynamic virtualised resources during the engineering, verification and validation phase. For global applications, the service will run on a distributed computing system able to scale up adding new virtual resources to ensure the desired coverage of the maps.

In our study the “Master” land cover map has been produced using Landsat and Sentinel 2 acquisitions while for the updated versions of the map either Sentinel 2 or Sentinel 1 images have been considered. To assure enough robustness and accuracy a restricted number of land cover classes has been considered so far: forest, built areas, water, other natural surfaces. The methodology has been preliminarily applied to the Italian territory with the aim of extending it to all Europe. The results obtained on Italy are encouraging: first of all a consistent land cover map with a spatial resolution of 30 m has been produced with an overall accuracy of about 92%. Moreover the PCNN procedure allows us to update the maps using a very high level of automation and keeping the same final accuracy.

[1] Del Frate, F., F. Pacifici, G. Schiavon, C. Solimini, “Use of neural networks for automatic classification from high resolution imagery,” IEEE Transactions on Geoscience and Remote Sensing, vol. 45, n. 4, pp. 800-809, April 2007.

[2] F. Pacifici and F. Del Frate, “Automatic Change Detection in Very High Resolution Images with Pulse-Coupled Neural Networks,” IEEE Geoscience and Remote Sensing Letters, vol 7, n. 1, pp. 58-62, January 2010

Global land cover products are typically derived from satellite data acquired over a year or longer, and are often used to represent the nominal land cover types over that period. For many areas, however, the surface condition could change dramatically within that period due to precipitation, snowfall, vegetation phenological change, natural and anthropogenic disturbances, as well as land cover conversion. Near daily update of global products with these changes is necessary to provide more realistic representations of the land surface. Using satellite data sets being collected by the Visible Infrared Imaging Radiometer Suite (VIIRS) onboard the NOAA-NASA Joint Polar Satellite System (JPSS), we are producing global land cover classifications on an annual basis. Currently, the derived products have two classification systems: the widely used IGBP 17-class scheme and the Biome system required by a LAI/FPAR algorithm. These products are generated following conventional approaches using annual metrics and advanced machine learning algorithms. To provide land cover dynamic information in near real time, we are also exploring approaches for generating daily or near daily data products for a surface components that can change quickly, including snow cover, surface inundation, and vegetation fractional cover. When used together with the annual classifications, these products will allow more realistic representation of the land surface for individual dates. In this talk, we will first introduce the derivation of the annual global land cover classifications. We will then present the daily data products from VIIRS and demonstrate their usefulness in improving the representation of actual surface cover conditions on a daily basis. Potential implications of such improvements will be discussed, especially for climate, weather, and hydrological modeling.

Forest and land cover classes, site fertility, growing stock volume, stem basal area, main tree species, forest height, and leaf area index were predicted with Sentinel-2, Sentinel-1 and Suomi NPP data. The study area for the ten-meter resolution Sentinel data was the territory of Finland of 338,424 km² and for Suomi NPP data the boreal and arctic zone of two-million square kilometers from Iceland to the Ural Mountains.

The accuracy of the predictions was defined with 13249 sample plots of Finnish national forest inventory over the territory of Finland. For the boreal region the testing was done by applying two stage sampling design. All together 2690 plots from 43 VHR images were visually interpreted.

From the VHR plots the proportion of forest in the boreal region was 70.7% with 95% CI (61.5%, 79.9%) or 2 090 766 km2 including all the three classes Forest, Peat forest and Open forest, while Suomi NPP based land cover map the estimated forest proportion was 70.4% or 2 080 034.5 km². Thus, the difference of the VHR plot based estimate to the forest area and Suomi NPP map proportions was only 0.3 % or 10 731 km².

In Suomi NPP prediction the average growing stock volume of the three forest classes was 100.8 m³/ha and the predicted total growing stock volume 2 080 034.5 km² x 100.8 m³/ha = 21.0 billion m³ (), respectively. The difference between the Suomi NPP map and the expected value of the VHR plot based total growing stock volume prediction was -8.9 billion m³ or 29.9%. The difference to the lower limit of the 95 % confidence interval was -4.7 billion m³.

The total predicted growing stock volume from the VHR plots for the total area of interest was 2 080 034.5 km² x 143.8 m³/ha = 29.9 billion m³ () with 95% CI (25.7, 34.1). In Suomi NPP prediction the average growing stock volume of the three forest classes was 100.8 m³/ha and the predicted total growing stock volume 2 080 034.5 km² x 100.8 m³/ha = 21.0 billion m³ (), respectively. The difference between the Suomi NPP map and the expected value of the VHR plot based total growing stock volume prediction was thus -8.9 billion m³ or 29.9%. The difference to the lower limit of the 95 % confidence interval was -4.7 billion m³.

The Sentinel-2 predictions were combined with forest and non-forest map from 3521 Sentinel-1 images. All the predicted forest variables were within the 95 % confidence intervals that were computed from the field plots at 13 forestry regions in Finland.

The carbon flux variables were computed by University of Helsinki. All flux variables except Net Ecosystem Exchange were within the limits from the flux tower measurements.

Dryland savannahs are crucial for understanding carbon cycling and storage and for their provision of ecosystem services. Globally, the accurate mapping of the woody savannah component and its characteristics is especially important as it provides input to carbon emissions models. Moreover, in the southern African region, the encroachment of unpalatable woody species over large expanses of palatable grasses has received a lot of attention as it directly affects the livelihoods of local populations. Over these scales, Earth observation technologies are seen as the only viable means for mapping and monitoring the characteristics of woody vegetation. However, the commonly applied sampling and validation approach incorporating point woody samples identified over aerial photography or very-high resolution data (e.g. via Google Earth) is problematic as the satellite data used for the mapping, with a pixel size of 10 – 30 m, rarely consists of pure woody vegetation. To bridge this spatial gap between what is identified in the point-based samples and what is included in the 10-30m pixel, we employ a UAV-based 2D and 3D sampling strategy. We incorporate point samples collected from Google Earth in a 400km² area of the Northwest Province of South Africa together with UAV-collected RGB and 3D mosaics, in order to optimise the mapping of fractional woody cover. We test the approach using both Landsat-8 and Sentinel-2 data in order to assess the applicability at both 10 and 30m scales. We also test the accuracy of two different machine learning classification approaches: random forests and support vector machines. Our 2D/3D UAV-based sampling approach provides higher fractional woody cover classification results than simply incorporating the ‘traditional’ point samples from aerial photography or Google Earth. Our suggested methodology can provide much needed assistance to fractional woody vegetation monitoring efforts in Southern African savannahs where the process is partly related with bush encroachment and land degradation.

Forest cover is one of the key environmental parameters used as by scientists as by policy makers to address the global change challenges. With the advancement of nature and man-induced disturbances forest cover gained momentum in the scientific community and in general public as a reliable estimate for deforestation/afforestation processes. As such, the accuracy and reliability of this product, either released individually or as a part of global, regional or national land-use/land-cover (LULC) classifications, is of utmost importance.

Present study aims at validating the global and regional LULC products CORINE 2012, CCI LandCover 2010, GlobeLand30 (2010) and Hansen at al. (2013) and JAXA's ALOS PALSAR forest cover products for the territory of Bulgaria. For that purpose ground truth data has been collected using various data providers such as LUCAS database (JRC), the Ministry of Environment and Water (MEW) of Bulgaria, European Vegetation Archive (EVA) database and authors own field surveys. The validation study performed compares the accuracy assessment results of the thematic layers of forest cover and provides discussions and suggestions for improvement of the forest cover products over the territory of Bulgaria.

The study evaluates suitability of hyperspectral and multispectral data with different spatial and spectral resolutions for classifications of tundra vegetation cover in the Krkonoše Mts. National Park (Czechia). Two legends were proposed: the detailed one with eleven classes, and simplified legend with eight classes. Aerial orthorectified images (orthoimages) with very high spatial resolution (12.5 cm) and 4 spectral bands have been examined using the object based classification. Satellite multispectral data WorldView-2 (WV-2) with high spatial resolution (2 metres) and 8 spectral bands and also hyperspectral data APEX (288 bands; 2-5 m spatial resolution) and AISA Dual (494 bands, 1-3 m spatial resolution) have been examined using object based classification and per-pixel classification. Per-pixel classification has been applied also to the freely available Landsat 8 data (spatial resolution 30 metres, 7 spectral bands) and Sentinel-2A data (10 bands, 10-20 m spatial resolution).

Of the algorithms for per-pixel classification, the following classifiers were compared: maximum likelihood classification (MLC), support vector machine (SVM), and neural net (NN). The object based classification utilized the example-based approach and SVM algorithm (all available in ENVI 5.2).

The best classification results (overall accuracy 84.3%, Kappa coefficient = 0.81) were achieved for AISA Dual data using per-pixel SVM classifier for 40 PCA bands. The best classification results of APEX though were only 1.7 percentage points lower. Both legends (simplified and detailed ones) show very good results in the case of orthoimages (overall accuracy 83.56 % and 71.96 % respectively, Kappa coefficient 0.8 and 0.65 respectively). The WV-2 classification brought best results using the object based approach and simplified legend (68.4 %). Landsat data were best classified using the MLC (78.31%). For Sentinel-2A (the simplified legend) the accuracy using MLC classifier reached 77.7 %. Our research confirmed that Sentinel-2A and Landsat 8 data are sufficient to get a general overview of basic land cover classes above the tree line in the Krkonoše Mts. National Park. Our assumption that some grassland categories will not be distinguishable in Sentinel-2A and Landsat data was confirmed. This supports our conclusion that it is not appropriate to use the same classification legend for the data with significantly (order of magnitude) different spatial and spectral resolutions.

We suppose that Sentinel-2A and Landsat 8 classification accuracies for at least some categories can be improved using time series of images acquired during one season. This is a goal of our further research.

Currently within the EU’s Earth Observation (EO) monitoring framework, there is a need for low-cost methods for acquiring high quality in-situ data to create accurate and well-validated environmental monitoring products. To help address this need, a new four year large-scale collaborative project entitled LandSense will link remote sensing data with modern participatory data collection methods that involve citizen scientists. Through citizen-powered science LandSense aims to deliver concrete, measureable and quality-assured ground-based data that will complement existing satellite monitoring systems. LandSense will deploy advanced tools, services and resources to mobilize and engage citizens to collect in-situ observations (i.e. ground-based data and visual interpretations of EO imagery). Integrating these citizen-driven in-situ data collections with established authoritative and open access data sources will help reduce costs, extend GEOSS and Copernicus capacities, and support comprehensive environmental monitoring systems. Policy-relevant campaigns will be implemented in close collaboration with multiple stakeholders to ensure that citizen observations contribute to EU-wide environmental governance and decision-making. Campaigns for addressing local and regional Land Use and Land Cover (LULC) issues are planned for select areas in Austria, France, Germany, Spain, Slovenia and Serbia. Novel LandSense services (LandSense Campaigner, FarmLand Support, Change Detector and Quality Assurance & Control) will be deployed and tested in these areas to address critical LULC issues (i.e. urbanization, agricultural land use and forest/habitat monitoring). Such campaigns are facilitated through numerous pathways of citizen empowerment via the LandSense Engagement Platform, i.e. tools for discussion, online voting collaborative mapping, as well as events linked to various campaigns involving public consultation. In addition to creating tools for data collection, quality assurance, and interaction with the public, the project aims to drive innovation through collaboration with the private sector. LandSense will build an innovation marketplace to attract a vast community of users across numerous disciplines and sectors and boost Europe’s role in the business of ground-based monitoring. The anticipated outcomes of LandSense have considerable potential to lower expenditure costs on ground-based data collection and greatly extend the current sources of such data, thereby realizing innovations in the processing chain of LULC monitoring activities both within and beyond Europe.

Desertification and climate anomalies are challenge in the middle part of Asian Ecosystems (Central Asia) that needs to be understood and assumptive develop sustainable land use category, especially on degraded areas. A large-scaled datasets to assess and measure relationships between global climate anomalies and vegetation patterns are response to support land and water use management in this drought-prone region.Within applying time-series data on coarse and medium resolution: AVHRR (GIMMS 3g), MODIS NDVI (MOD44/MYD44) and Sentinel (2014-present), these precluding to understanding environmental changes with their causes and consequences. Results are presented using (a) GIMMS 3g (1982-2013) and MODIS (2001-2015) time series of open shrubland zones in Central Asia (Kazakhstan, Kyrgyzstan, Tajikistan, Turkmenistan and Uzbekistan), where we developed a real –time scenario; (b) change detection examined by Mann-Kendall trend analysis method; (c) validate and assess spatio-temporal accuracy with 3 month running mean. Preliminary results of using Copernicus-Sentinel time series for NDVI index will be presented for arid ecosystems and implementing to increase per-pixel missing datasets to improve a temporal accuracy. This suggests generally lead to regulate temporal anomalies in dry conditions in the sub-region on multi-scale years and it could be a beneficial management of ecosystem services in areas with similar climatic characteristics.

In a context of increasing availability of satellite images acquired by several recently-launched sensors (PROBA-V, Landsat-8, Sentinel-1 and -2), the number of land cover and cropland maps is expected to grow alongside. It is therefore critical for the user community to share a common reference data base to systematically assess the accuracy of the future and existing products, especially given their continued uncertainty. In addition, several shortcomings affect the currently existing validation data sets such as the GLC 2000, GlobCover 2005, STEP, VIIRS and GLCNMO 2008 datasets. When not directly obtained from in situ observations, reference data are traditionally collected by remote sensing experts with a deep knowledge and understanding of specific ecosystems through photointerpretation of very high spatial resolution imagery, often used in combination with temporal profiles of vegetation indices. As a result of recent technological developments, crowdsourcing appears now as an appealing alternative to collect a larger amount of reference samples with yet potentially a higher uncertainty. With this as backdrop, this research presents a new global validation dataset specifically targeting the cropland component of the land cover. In order to fully take advantage of recent global validation experiences (Geowiki, GlobCover and the Land Cover CCI campaigns), a two-tier sampling strategy was implemented: a large sample of around 40,000 samples was photo-interpreted on very high resolution (VHR) images through crowdsourcing whereas remote sensing experts photo-interpreted digitized segments for a subset of around 4,000 samples. The sampling itself is a stratified systematic sampling, adjusted for nonequal-area and based on a pattern of replicates projection which provides more flexibility compared to a traditional systematic sampling. Strata were derived from a cropland probability map resulting of a compilation of different thematic maps. Areas with very high or very low probability were considered easier to classify and thus requiring a lower sampling rate. Pixels with an attributed crop probability between 25% and 75% are sampled with a higher rate. In addition to presenting the overall sampling design, this work pursues three main objectives. First, it seeks to validate the two-tier approach and specify if and when the two sources of data can be integrated into a hybrid validation dataset. The second focus is the formal and spatially explicit comparison and validation of <300m land cover or cropland maps such as the Land Cover CCI, the Globeland 30, the FROM-GLC, and the Unified Cropland Layer products. Finally, a web-portal that allows user to validate their global cropland map is introduced.

Consistent land cover mapping implies solving conceptual issues with respect to the appropriate choice of legend and data model. Currently, all high resolution global land cover products can be classified in two groups : categorical map (e.g. GlobeLand30) or continuous fields (e.g. Global Forest Watch). In this study, we test a third data model in the frame of biodiversity modelling : a spatial region mapping that describes landscape with crisp boundaries and quantitative land cover characteristics. Advantages and incovenients of the three data models are therefore analysed from image analysis and GIS application perspectives.

As a case study, European regions with distinct ecosystems (in Belgium/Northern France, Spain and Finland) have been mapped using the spatial region concept at 10 meter resolution using Sentinel-1 and 2 data of the year 2016. The spatial regions were then used as input for the computation of ecological models. It was shown that i) quantitative information is easier to incorporate into bioinformatics frameworks than categorical values and ii) spatial regions can be used to easily integrate ancillary data that complement a pure land cover description.

From a technical point of view, spatial regions provide a good structure to fuse optical and SAR data. The Sentinel-2 images are used to delineate the spatial regions together with DEM information using automated image segmentation algorithms. Sentinel-2 image composites are then used to characterize vegetation types, and the land cover description is enriched and consolidated with Sentinel-1 data, which provide additional information about vertical structure in addition to complete time series in cloud prone regions.

From a conceptual point of view, the versatility of the spatial region can be highlighted by the use of several land cover legends. The main drawback is the dependency on a segmentation step that modifies the minimum mapping unit and creates arbitrary boundaries.

To sum up, the scope of use and the constraints on the validation framework of the three models has been analysed with examples from the case studies. None of the models are perfect, but spatial regions are well suited to create derived maps such as land use maps, biotopes, species habitat suitability maps or ecosystem services.This could actively contribute to achieve Europe biodiversity targets 1 and 2 thanks to enhanced monitoring and better understanding of the internal structure of ecosystems.

Despite the fact that many products are still lacking appropriate validation, a large set of good practices has been thoroughly described in the literature. These good practices mainly focus on the sampling design and on the estimation of the accuracy or quality indices. However, little attention has been paid on the impact of the response design on the estimation of accuracy and quality indices. The choice of rules for matching reference data with the map is rarely mentioned in validation papers because it is considered to be obvious and of little impact on the final results. However, mixed pixels can increase the uncertainty of the validation process depending on the response design. We therefore analyzed the interactions between selected legends and response designs with respect to their impact on the estimation of recommended accuracy indices.

Based on the 300-m ESA CCI land cover map validation experience, we first focus on a quantitative response design estimating the sub-pixel proportion of land cover components. We then aim to quantitatively analyze the interactions between this response design and the classification system by quantifying their impact on the estimation of overall accuracy. We therefore investigate three components of the response design: its conformity with the land cover classification system, its precision and its accuracy. These three components are assessed with respect to two types of legend: majority based legend and UN-Land Cover Classification System (LCCS)-based legend with composite classes.

Quantitative response designs are more costly to produce but they can be reused with any LCCS-based legend and reduce ambiguities for interpreters. We also demonstrate that unprecise estimates of sub-pixel proportions induce an underestimation of the actual overall accuracy, especially with large numbers of classes.

Based on simulated data, we then assessed the need for a quantitative response design in the case of a 10-m land cover map. For such a decametric scale, the type of response design could indeed differ, but a rigorous analysis shows that unbiased validation would also require a pixel-based response design instead of a point-based approach with a single category in the center of the pixels.

Two important facts are then discussed in this study: i) overall accuracy is underestimated (up to 10 %) because of unprecise sub-pixel proportions and ii) LCCS legends can be optimized to reduce the uncertainty of thematically precise legends. All the results highlight important points to consider in order to design accurate validation for decametric resolution maps.

Although occupying a small percentage of the Earth’s surface, urban land is home to more than half of the world’s population, driving most of the social economic activities globally and in many individual countries. While urban areas are among the most heterogeneous landscapes that are inherently difficult to map, reliable and detailed information of these areas is crucial for understanding the direct and indirect impact of urban growth and its roles in the coupled human-natural system. Based on a global sample of high resolution satellite images and near wall-to-wall coverage of Landsat images assembled through the Global Land Survey program, we have produced a global man-made impervious surface (GMIS) dataset at the 30 m spatial resolution. This circa-2010 dataset defines a human built-up and settlement extent (HBASE) mask at the 30 m resolution and provides an estimate of percent impervious surface cover within each HBASE pixel. This product will be distributed through a web portal at the NASA Socioeconomic Data and Applications Center (SEDAC) to the general public. In this presentation, we will provide an overview of the GMIS dataset, the data and methods used to derive this product, and accuracy assessment results. Lessons learned for global urban impervious mapping at sub-hectare spatial resolutions will be discussed.

Long-term data records of Earth Observation data are a key input for climate change analysis and climate models. The ESA Climate Change Initiative is providing Essential Climate Variables, defined by GCOS, based primarily on satellite based observations. Land Cover status and seasonality are ECVs supporting the understanding the earth system. Medium resolution (300m – 1000m) is suitable input for regional and global models. More focussed models require higher spatial resolution data for model validation and assimilation. The goal of the work presented here is to create land cover maps over Africa starting from Sentinel-2 L1C TOA products, in order to extent and enrich a time series from multiple sensors (MERIS, SPOT VGT, Proba-V and AVHRR). The input data for the medium resolution sensors, covering a period of more than 20 years, already exceeds 200 TB. However, the production challenge is increased for the high resolution sensors, namely Sentinel 2. The data volume for the African continent coverage is already estimated as 100 TB for a single year.

A processing system has been developed which comprises two parts: the preprocessing system which takes care of access and ingestion of all input data, performs automated quality control, thematic preprocessing steps (pixel classification & atmospheric correction), and finally the binning and compositing of surface reflectances. This steps reduces the data volume significantly. The global composites are input to the thematic processing into the different land cover products. The system has been developed originally for the medium resolution data and has been modified to work now also for Sentinel 2 data. One fundamental difference is caused by the fact that the medium resolution data were always static data sets, where processing and reprocessing could be managed as offline jobs, while Sentinel 2 data have to be processed continuously.

This presentation will show the infrastructure of the system with focus on the preprocessing part, the data flows, and the processing steps included. Input data quality control will be discussed and design of the system to mitigate processing failures. The preprocessing system is realised with the Calvalus EO processing system of Brockmann Consult. The different challenges of offline production of 20 years medium resolution data as well as continuous production of Sentinel 2 data will be discussed and the solutions found and lessons-learned by the Land Cover CCI team will be presented. At the time of abstract writing the team was still in the process of acquiring and processing S2 data over Africa in order to complete a full year of good data coverage. This will be achieved by the end of 2016, and we expect to have results of the surface reflectance product ready by the time of the conference.

Secondary forest succession on abandonment agriculture land represents one of the most significant process affecting currently European landscape in particular in the Central and Eastern Europe. In very mosaic landscapes, such as those in the Polish Carpathian Mountains, these changes occur often at small spatial scales and they have been difficult to track using traditional change-detection techniques, which typically use images from very sparse dates. The aim of this study is therefore to test and develop an approach for mapping forest succession on abandonment agriculture lands using dense series of Sentinel 2 data (all available images from two years 2015 and 2016). In our approach we test the use of different dataset configuration e.g. series of images with or without data from winter, spring and autumn. In addition we use ancillary data such as LiDAR dataset (digital terrain model, digital surface model and canopy height model), national topographic dataset, orthophotomaps and Landsat 8/OLI data. The study area is located in the northern part of the Carpathians and covers about 20000 km². Our approach contains the following steps: (1) image selection and acquisition, (2) image pre-processing (detecting and removing clouds and shadows, and applying atmospheric and topographic correction) (3) spatio-temporal segmentation and classification (hierarchical approach), (4) secondary forest succession on abandonment agricultural lands mapping, (5) secondary forest succession temporal and spatial pattern analyzing, (7) accuracy assessment. In our approach we evaluate also the performance of different spectral indices e.g. NDVI which can be used for secondary forest succession mapping. We generate spectral-temporal seasonal trajectories of changes associated with secondary forest succession using regression methods and point-to-point fitting. In our workflow, first, we distinguish segments with vegetation or without vegetation. Then we delineate three types of vegetation (low, medium and high). In final step we extracted the areas with secondary forest succession on abandonment agriculture lands. In general, the use of dense series of Sentinel 2 images, from different seasons, in particular from spring and autumn, increase the detection and delimitation accuracy of forest, semi-natural areas, agricultural areas and abandonment agriculture lands with secondary succession.

We gratefully acknowledge support by the National Science Centre, project RS4FOR [project no. 2015/19/B/ST10/02127].

Since recently, cropland mapping at global scale with high (HR) resolution EO products was facing data availability issue. Several satellite images recorded at specific moment of the growing season are required to accurately map cropland areas. In consequence, most of the operational applications have currently to proceed either by spatial sampling or by monitoring at medium resolution the overall agricultural landscape as HR EO images acquisitions over large area was associated with high cost. Sentinel constellation satellites is changing this game providing both optical and radar data at least every 10 or 12 days globally. However, over rainfed agricultural region, crop growth coincides with rainfall leading to dense cloud cover. Cropland mapping of such areas can only rely on few useful optical images during the growing season and SAR remote sensing is often considered as an alternative for its all-weather observation capability.

Sentinel-1a and 1b are currently the only satellites providing free-of-charge SAR time series continuity at decametric resolution with a revisit cycle of 12-days. While this cycle was systematic over Europe and some areas in the world since their launch, recent change in the acquisition planning strategy will provide such a revisit cycle for most terrestrial areas. Combining overlapping tracks as well as ascending and descending satellite overpasses further improve this revisiting capability but impose to combine images with very different viewing angles and therefore different local incident angles. As an example, any place in South Africa was covered by at least 5 images during the 2015-2016 growing season period. However, the angle constraint requires specific preprocessing method to build proper time series over a given area.

In this paper, the high spatial and temporal Sentinel-1 revisiting capability is investigated to develop an automatic method for regional cropland mapping for African agricultural landscapes. A stratified approach is adopted to isolate regions with similar agrosystems and similar crop calendar. Spatio-temporal features are computed from the SAR backscattering coefficient time series and associated polarimetric indices. The respective trajectories of the backscattering coefficient along the season from soil preparation to crop harvest and of the corresponding ratio between dual-pol signals are interpreted to discriminate agricultural cropland from the surrounding land cover.

The method is developed over the entire South Africa country thanks to the availability of an exhaustive training and validation data set. The robustness of the approach is then tested over another intertropical area to document the potential application domain. The robustness is characterized in term of (i) field size distribution, (ii) timeliness of the information delivery, (iii) number of SAR images in the time series, and (iv) accuracy of cropland map that can be obtained with optical series recorded during the corresponding period of time.

The Emissions Database for Global Atmospheric Research (EDGAR), managed by the Joint Research Centre (JRC) of the European Commission, provides global past and present day anthropogenic emissions of greenhouse gases (GHG) and air pollutants by country and on spatial grid. The JRC is working on updating the Land Use, Land Use Change, and Forestry (LULUCF) component of EDGAR to include global estimates of emissions and removals from the LULUCF sector calculated in a spatially-explicit way. The estimations make use of different sources of data, such as maps of land cover/land use, soils, climate, ecological zones, and primary forests. In particular land cover maps are an essential component of the system. At the moment the land cover maps produced within the ESA Climate Change Initiative (CCI) are reclassified to match the standard IPCC categories and monitor the different land uses considered within the GHG inventories (forest land, cropland, grassland, wetlands, settlements, other land). Estimations are based on the IPCC methodology at Tier 1. The poster presents the methodology and the preliminary results obtained.

Precise and free global/regional land cover (LC) products is an essential key for a wide range of environmental studies affecting biodiversity, climate, and human health.

We compared and evaluate over north-western Morocco freely abilable Global and regional land Cover datasets, each of them used different classification scheme and method, sensor, spatial and temporal resolutiontial resolution, overall accuracy, sensor used, Map projection and Datum which made the comparison difficult.

For harmonization, we first adopt Albert Equal area projection for Africa continent with latitude-longitude cordinate sytem(WGS_84) datum. Second we adopted a unified class legend for all LC and all products are resampled to match the lowlest spatioal resolution then compared in per-pixel level in order to test the level of agreement / disagreement area.

Accuracy assessment was conducted using systematic grid sampling strategy with avilable landsat archive and Historical Google Earth© images.

The findings of this study show agreement / disagreement across the different land cover products, even after harmonization, mainly attributed to differing classification algorithms and references data used by each land cover product,

Farthermore, some multitemporal fine resoltution LC products like ESA GlobeCover shows a very hight overal accuracy but not applicable for change detection over time, this is maybe due to the use of differente epoches to generate a landcover map.

Remote sensing satellite data have been widely used to classify land use and land cover in different scales. However, there are a few uncertainties in areas with complex land use e.g. traditional agriculture due to diversity and small sizes of farms. In this research, we have evaluated the potential of Rotation Forest algorithm (RoF) in traditional agricultural areas. On the other hand, advanced remote sensing classifiers such as Rotation Forest algorithm is one of the ensemble machine learning methods which is used to classify land use classes accurately. The study area is Moghan County located in the northwestern of Iran with complex agricultural landscapes; there are different land use and land cover classes in this area, including built-up areas, water bodies, rangelands, rain-fed crops, irrigated crops, fallow lands and orchards. Irrigated crops consists of alfalfa, sugar beet, wheat, barley, canola, grain maize, silage maize, cotton, double cropping and rain-fed crops are only wheat and barley. For the purpose of this paper, all available cloud free Landsat 8,7 and Sentinel 2 images in 2016 were used. Reference ground dataset obtained from agricultural administration of the county and field sampling in two levels includes semi-detailed and detailed. Kappa coefficient was used to evaluate the accuracy of RoF model output maps. We produced two levels of maps: semi - detailed and detailed which semi-detailed map separates six general land use classes: built-up, water bodies, rangelands, irrigated crops, rain-fed crops and orchards. Detailed map is the final product that shows all levels of details and classes mentioned earlier. Results showed that both semi-detailed and detailed levels have very accurate outputs, 0.89 and 0.74 Kappa coefficients, respectively. With regard to algorithm, these results indicated that machine learning algorithms and time-series remote sensing data are able to classify land use with different levels in areas with complex land uses.

In face of fast and widespread land degradation, habitat loss and biodiversity declines, protected areas are the backbone of today’s conservation strategies. They aim to preserve some of the most pristine species and ecoregions from further deterioration. But frequently protected areas suffer themselves from human encroachments and many are ecologically isolated from their surroundings. The efficiency of protected areas is often questioned.

Multilateral environmental agreements, like the UN Convention on Biological Diversity (CBD), are immense international efforts to halt habitat loss and safeguard biodiversity and human well-being. In 2010, parties to the CBD have renewed their commitment in biodiversity conservation around 20 targets. These so-called Aichi Biodiversity Targets define the priorities for the 2011-2020 period. The regular and accurate monitoring of land cover changes is one of the key information to assess progress towards achieving these targets. The European Commission’s Digital Observatory for Protected Areas (DOPA) is recognised by the CBD as reference information system to be used by Parties in preparation of the Conference of Parties. DOPA aims to provide a large variety of end uses with means to assess, monitor and possibly forecast the state of and pressures on protected areas from local to global scale.

In the frame of reinforcing the DOPA system, we develop a new indicator of land cover change to allow the ranking of protected areas according their exposure to anthropogenic pressures. Based on the ESA CCI land cover maps, we compute natural and anthropogenic change inside protected areas and their surrounding 10 km buffer zone. The analysis allows us to assess the efficiency of protected areas in maintaining natural land cover and map their ecological integrity into their wider landscape. Great attention is taken in assessing uncertainties in the obtained results. We present results from our analysis and discuss some requirements on land cover maps for monitoring habitat changes in and around protected areas, based on the experience gained from analysing the ESA CCI land cover maps.

Woody perennial vegetation is an integral part of savannah ecosystems and plays an important role in carbon cycling and ecosystem service provision. Accurately mapping its presence and its characteristics can provide useful input to global carbon emissions models as well regional policy decision making efforts regarding bush control or the overexploitation of fuelwood. Recent attempts to map the extent of savannah woody cover over the regional scale have employed Earth observation data either from optical or radar sensors, and most commonly from the dry season when the spectral difference from the ‘background’ grasses is maximised. By far the most common practice has been the use of Landsat optical bands, but some studies have also used vegetation indices or L-band or C-band SAR data. However, conflicting reports with regards to the effectiveness of the different approaches have emerged leaving the respective land cover mapping community with unclear methodological pathways to follow. We address this issue by employing Landsat and ALOS PALSAR data, together with colour aerial photography for training and validation of random forest regressions, to assess the accuracy of mapping woody vegetation when: (a) data from either or both seasons are considered; (b) annual PALSAR mosaics or the actual PALSAR data are used on their own or together with the optical data; (c) vegetation indices are calculated and are used either on their own or together with the Landsat bands; and (iv) thermal infrared information is not discarded but included in the parameterisation. We test our approach in an area of the Northwest Province of South Africa which spans over 6 Landsat scenes, covering an area of approximately 53,000 km². Our hard classification results (woody vegetation, non-woody vegetation and no-vegetation) show that the most accurate estimates are produced from the model that incorporates all 23 parameters: Landsat optical and thermal bands and three vegetation indices (NDVI, MSAVI and TNDVI) and HH polarised PALSAR data for the dry and wet seasons (overall accuracy: 89%; woody cover balanced accuracy: 91%, producer’s accuracy: 83% and user’s accuracy: 90%). The combination of either dry season Landsat bands with the HV polarised radar data, appears to be sufficient for achieving woody cover balanced accuracies of 89%. Dry season optical bands alone are able to map woody cover with more than 81% balanced accuracy and the accuracy increases by the inclusion of either the vegetation indices or the TIR band (to 83% and 84%, respectively). Our findings can provide much needed assistance to woody vegetation monitoring efforts in southern African savannahs where the process is partly related with bush encroachment and land degradation brought about by recent climatic changes and overgrazing.

Accurate and up-to-date maps of agriculture play an important role in the study of food security. However, mapping croplands over large areas with remote sensing using traditional image classification tools involving one time, one place approach requires significant computing and labor resources. Moreover, large area cropland mapping holds extra challenges because of the seasonal changes, variation in crop type and crop intensities. Here we report on a generalized image classifier based on a decision tree algorithm and the idea of signature extension applied to temporal statistics of vegetation indices derived from Landsat satellite data combined with climatic and topographic variables. We selected a number Landsat footprints spread over various climatic zones in Europe, Middle East, and Central Asia, covering the 2012-2015 period. We further applied the generalized classifier at three different levels: own level where training and testing data are extracted from the same footprint (no generalization); zone level where training data are extracted from all footprints in a zone; and the global level where all footprints form the source of training data. The results show that the generalized classifier is successful in identifying and mapping cropland pixels with comparable success across all three levels of signature extension, which shows the efficiency of the approach. Moreover, when the generalization/signature extension approach is applied to randomly selected footprints that were not involved in the original training process outside the core footprint locations, results are encouraging. The work has important implications for large area cropland mapping using the generalization/signature extension framework that requires very little user input and has the potential to map cultivated areas along with crop types across regional to global scales with 30m spatial resolution.

There are many classification systems for land cover (LC) and land use (LU), each filling a particular purpose and thus emphasizing particular aspects of LC and LU information. To date, attempts to establish an internationally accepted and operational standard on LC/LU have been unsuccessful. A more realistic approach is to establish a standard for the documentation of different classification systems, thus facilitating the data interchange between them.

Within the European context the EAGLE concept addresses a number of key criteria such as LC and LU separation, scale independence and the handling of temporal aspects of dynamic phenomena on both object and class level. The EAGLE data model thus consists of three main descriptive blocks; Land Cover Components (LCC), Land Use Attributes (LUA) and additional landscape Characteristics (CH). A land unit can therefore be described as a combination of one or more LCC, LUA and CH.

Two similar international systems are the Land Cover Classification System (LCCS) used by FAO, and the Land Cover Meta Language (LCML) which is an ISO standard. There are differences, but also important commonalities between LCCS, LCML and the EAGLE data model. LCCS is a classification system, but not in the traditional sense of a single nomenclature. Instead, LCCS offers a suite of diagnostic criteria and characteristics to describe classes within hierarchical nomenclatures (including the documentation of these classes).

LCML – not a classification system, but a meta-language – is taking LCCS one step further. The intended use of LCML is to describe and document land cover nomenclatures in general, including those that are developed independently from LCCS. The purpose is, in addition to documentation, to be able to compare nomenclatures and convert datasets from one nomenclature to another.

Attempts to document the European CORINE Land Cover using LCML, however, have revealed deficiencies with the meta-language. The EAGLE approach was therefore developed to support European land monitoring by facilitating documentation and comparison of land cover nomenclatures used in Europe. Similar to LCML, it also comprises land characteristics but, importantly, separates between land cover and land use aspects. It is therefore possible to describe classes in a nomenclature, not in a hierarchical manner, but based on an object-oriented approach. The strength of the EAGLE model thus lies in its application flexibility and the strict model semantics which reduces the degree of freedom and allows the derivation of machine readable encodings like ontologies or model application schemas.

Further development of the LCML ISO standard is required to make the meta-language operational and useful outside the LCCS environment. Major parts of this development have been undertaken by the EAGLE group. Perceiving that the EAGLE concept and LCML in many respects are fairly similar, a reasonable strategy would be to merge the two systems in a next version of the ISO standard (as proposed by the HELM project). Besides LC, also LU information could be integrated, to expand the scope of the LCML standard, as many nomenclatures are a mix of LC and LU aspects.

Accurate and reliable information on agricultural production, crop and livestock conditions is needed for monitoring agricultural drought and for reacting to food security crises. Early warning systems can provide the relevant evidence in and operational manner, but need timely geospatial information on crop and rangeland location and conditions. Land cover/land use information based on satellite imagery have been extensively used for agricultural monitoring over the last two decades and there is a multitude of global and regional datasets available. Global land cover datasets provide information based on homogeneous input data and methods for large areas, but pay the price of lower spatial resolution (usually from 300 to 1 km). On the other hand, regional and national datasets are able to provide a higher spatial detail but are affected by lower level of availability and geographic comparability. Comparative analysis among datasets in Africa, showed large discrepancies and inconsistencies due to spectral similarity between cropland and grassland in arid and semi-arid areas, complex cropping patterns and landscape structure. Therefore harmonization and hybridization approaches have been proposed, involving the integration of data from different sources and reducing the inconsistencies of single products.

This work follows on previous products based on land cover hybridization and targeting the specific information needs of crop and rangeland monitoring for food security early warning by incorporating the latest high-resolution datasets and by identifying the most adequate low-resolution land cover products in those countries where high-resolution products are not available. The proposed methodology relies on a multi-criteria decision analysis (MCDA), where each dataset is evaluated and weighted according to a series of criteria including among others validation based on existing and new crowd sourced data sets centered on areas of highest disagreement. The final product is an agricultural and pastoral mask at the continental level which represents for each country the most recent and highest quality land cover/use data for agricultural monitoring.

Land use, land cover changes, and ecosystem-specific management practices are recognized for their roles in mediating the climatic effects on ecosystem structure and function. A major challenge is that our understanding and forecasting of ecosystem C fluxes cannot rely solely on conventional biophysical regulations at any scale, from the local ecosystem to the globe. A second challenge is to quantify the magnitude of the C fluxes from managed ecosystems and landscapes over the lifetime of the C cycle, and to deduct the various energy inputs during management. In this long term EO data series can give a pivotal contribution. Our objective is to quantify the landscape-scale C footprint of both managed agricultural-forest landscapes and people, using the Kalamazoo watershed in southwestern Michigan as testbed. The mechanisms from both human activities and biophysical changes on ecosystem C dynamics at different temporal and spatial scales will be explored by modeling total net ecosystem C production (physical and social C fluxes), exploring the relationships through Bayesian SEM, and performing a spatially-explicit LCA on the total C production.

Earth Observation data (including Sentinel 1 and 2 data), available geospatial data, records of management practices, survey of historical practices, a land surface model (CLM), in situ measurements of C fluxes will be used to achieve our objectives. Research tasks are: (1) model the long term dynamics of the physical C fluxes of the watershed and landscapes; (2) estimate the social C fluxes for the same time period; and (3) diagnose the mechanisms from land use, land cover changes, management practices, climatic change, and climatic extremes on the total CO2eq fluxes at the two spatial and temporal scales through LCA. The physical C will be quantified through the CLM Community Land Model after ecosystem-specific parameterization and independent validation. The activity is undertaken under a 3 years NASA Grant n. NNX17AE16G.

Having a transboundary approach for modelling land-cover / land-use (LCLU) is a strategic territorial planning tool because LCLU is a driver of climate change, environmental and health risks. A LCLU model is then a prospective tool for risks monitoring. It allows simulating the impact of new or existing policies and projects. Regional and national authorities need such operational decision support tools to plan the evolution and scenarios of changing their territory in a sustainable way. Since risks are not stopped by regional borders, integrated transboundary modelling tool can help to develop sustainable common strategies to deal with environmental and health risks.

The SmartPop Project spatially simulates the population growth in Wallonia and, in particular, in the city of Liège in the context of planning and monitoring Smart Cities. In Flanders/Brussels, the “RuimteModel Vlaanderen”, which is a Cellular Automata (CA)-based land use model, has been used in support of several land planning decision making actions. CA are discrete computational systems defined by a regular grid of cells, each in one of a finite number of cell states and changing according to rules regarding the state in their neighboring cells. CA have perhaps been the most popular way to model LCLU change and spatially explicit population density because they are (1) intrinsically dynamic, (2) able to deal with high resolutions and thus produce results with a useful amount of detail and (3) they outperform other models in realistically modelling LCLU change. Within the SmartPop project, Wallonia can benefit from this long term research and experience within Flanders/Brussels and both regions can work toward a common understanding of LCLU changes through interregional modelling studies.

The model application to Wallonia needs some contextual and methodological choices such as LCLU map legend definition, model refinement, calibration, scenarios definition and validation of inputs. The model outputs are annual LCLU maps from 2007 to 2050 and spatial indicators (population dynamics/density, urban growth, natural habitat fragmentation, residential area growth vs. croplands area decrease ...

Involving end-users at the start of the model implementation guarantees future use and valorization of the model and of the outputs generated. The key results of our end-users analysis are the strong demand for updated and separate LU & LC maps over Wallonia, with a high degree of semantic and geometric accuracies. More than half of the respondents are interested in thematic outputs such as LCLU change, predictive modelling and spatial criteria (urban density, natural habitat fragmentation, soil sealing ...), especially for spatial statistics and key environmental indicators.

The cloud-free composite over Southern Africa, at 10m spatial resolution, has been produced using data acquired from Sentinel-2A, and has been processed to demonstrate and promote the spatial coverage and resolution. The selected dataset, covering South of Africa, is part of Sentinel-2A overpasses from the 1st of January 2016 to 5th of December 2016. The Sentinel-2A composite have been generated applying 2 steps algorithm based on three-monthly darkest pixel selection to remove clouds and successively yearly composite based on max-NDVI method to remove cloud shadows. It covers an area of about 7430000 km² and has been constructed from 740 Sentinel-2A L1C tiles (110km x 110km, at 10m spatial resolution).
The Sentinel-2A cloud-free composite at 10m spatial resolution over Southern Africa has been processed using CalESA system with SNAP 4.0 processing software.
The yearly composites are based on Sentinel-2A L1c tiling grid (110km x 110km) and then mosaicked using the geo-location information.
• Preliminary S2A data screening based on cloud coverage information reported in the metadata (less than 5% or when not possible, the three less cloudy images). More than 25000 Sentinel-2A data have been processed.
• S2A data re-projection from UTM to Lat/Lon (WGS-84).
• STEP-1: For each quarter (January-March, April-June, July-September, October-December) select the darkest pixel from the stack of Sentinel-2A images identified as input.
• STEP-2: From the quarterly composites choose the best pixel applying the “max-NDVI” method replacing the remaining cloud shadow pixels with clear ones, if present.
• Using GDAL utilities and Python routines to generate the RGB mosaic over Southern Africa at 10m spatial resolution.
• Using the ‘S2 visualisation tool’, developed in house and available at 2016africacomposite10m.esrin.esa.int, the user can visualise the S2 cloud-free composite and navigate through it till full resolution.

Temporal consistency of land cover time series and detection of major land cover change are key requirements from the users’ community to describe terrestrial surface over time. Land cover was listed as an Essential Climate Variables (ECVs) by the Global Climate Observing System (GCOS) as critical information to further understand the climate system and support climate feedback modelling. However, only global land cover maps from the same instruments have been produced till now, keeping the time series rather short. In the framework of the Climate Change Initiative (CCI) supported by the European Space Agency (ESA) climate modelling and remote sensing teams joined forces to design, implement and deliver global datasets matching the climate science needs for long-term global products.

Building on the ESA-GlobCover experiences, this research first revisited the land cover definition into a stable component and seasonal component and designed a new approach to produce consistent land cover time series decoupling consistent mapping and change detection. Then, the archives of several satellite missions including ENVISAT Meris FR and RR, SPOT-Vegetation, the more recent PROBA-V, and the 90s’ archive of 1 km AVHRR, were reprocessed using state-of-the art methods to produce weekly surface reflectance composite and quality flags throughout the years. According to a stratification splitting the world into 22 equal-reasoning areas from ecological and remote sensing points of view, different seasonal composites were compiled to enhance the land cover discrimination. A typology of 22 land cover classes was defined based on the UN Land Cover Classification System and its classifiers to support the further conversion into Plant Functional Types distribution required by the Earth System Models (ESM). The classification processes combined first machine learning and unsupervised algorithms at 300 m resolution from the whole MERIS FR archives using most of the MERIS bands to establish the land cover baseline. Based on similar algorithms annual global land cover maps from 1 km AVHRR HRPT, 1 km SPOT-Vegetation data sets and from 300 m PROBA-V time series were then produced to serve as input in the land cover change detection algorithm. Systematic analysis of the temporal trajectory of each pixel allows depicting the main change for a simplified land cover typology matching the IPCC classes. This new land cover change detection method was found quite reliable for SPOT-Vegetation and PROBA-V thanks to their excellent temporal co-registration. In contrast, the poorer radiometric and geometric quality of AVHRR HRPT time series only provided major change in contrasted landscapes. The change detected at 1 km was then disaggregated at 300 m whenever higher resolution imagery was available. Finally, these products were validated from an independent reference dataset built by a network of international experts.

All land cover maps can be visualized from an interactive web interface and downloaded along with an aggregation tool, enabling re-projection and re-sampling as well as the translation from LC classes into Plant Functional Types for different climate models. Three major ESM successfully completed several joint experiments based on the CCI land cover products.

The Copernicus Global Land Service is the component of the European Copernicus service which ensures a global systematic monitoring of the Earth’s land surface. It provides bio-geophysical variables in near real time describing the daily state, and changes in state, of vegetation, land surface processes and is currently preparing the release of a Moderate resolution Dynamic Global Land Cover layer.

In this presentation the methodology and rationale behind the Moderate resolution Dynamic Global Land cover layer is explained. This layer complements several global land cover ‘epoch’ datasets which have been created at medium (and high) spatial resolution during the last decade by providing a yearly dynamic land cover layer at 100m resolution. We will present the sub-product covering continental Africa for the year 2015. To build this global land cover layer, 100m spatial resolution PROBA-V data are used as primary EO data. Data fusion techniques are applied for areas with insufficient 5-daily PROBA-V100 m data and daily 300 m datasets are fused in. Next, time series metrics together with ancillary data sets (e.g. other Copernicus global land service biophysical products) are used in a supervised classification approach. Finally, at a third level, we build upon the success of previous global mapping efforts and focus on the improvement in areas where the thematic accuracy of the respective maps was insufficient to perform the final classification of each pixel. The map uses a hierarchical legend based on the United Nations Land Cover Classification System (LCCS). Compatibility with existing global land cover products is hereby taken into account, and extended by providing several cover layers. Training data have been collected from multiple sources, among others by using existing reference datasets (e.g. GOFC-GOLD) and by collecting reference data through Geo-Wiki (http://geo-wiki.org/). The product has been validated by local experts. The validation sample design has been random stratified where each sample site has been classified by visual interpratation of a high resolution imagery (Google and Bing).

GlobeLand30 is an open-access 30-m resolution global land cover (GLC) data product with 10 major classes for years 2000 and 2010. Since its first release on the 22 Sept, 2014, it has been utilized by users from about 120 countries and found applications in many Societal Benefit Areas. At the same time, the users have put forward new demands, such as providing more land cover classes, up-to-dateness and time-series. This has led to an international validation and the preparation of the updating of GlobeLand30.

The validation of 30-m GLC data products is facing several critical challenges related to the high spatial heterogeneity of land cover in the entire earth land surface, and the lacking of standardized approaches and efficient on-line tools to support collaborative practices. With the support of GEO and UN-GGIM, a technical specification of validation has been formulated and a web-based validation system has been developed. About 40 GEO-UN_GGIM members have participated in the joint validation of GlobeLand30.

The updating of 30-m GlobeLand30 is different than its original creation, and aims to produce a 2015’s version product. From the technical point of view, change detection with remote sensed imagery is the major approach and the rapidly increasing crowdsourcing information provides another valuable resource. Due to the extreme complexity of spectral heterogeneity of land cover classes, no one change detection algorithm could be universally applicable to all kinds of imagery and geographic regions. In order to support efficient updating with the consideration of the existing land cover data sets, a specific on-line system was developed to facilitate the design and execution of suitable change detection workflow with the help of a domain knowledge-based service relation model and dynamic service composition.

In the context of the Climate Change Initiative supported by ESA, the Land Cover team aims to map the whole Africa based on the entire archive of Sentinel-2 mission. To address the requirement of a high spatial resolution LC map expressed by the climate science community, the research team will generate a prototype map at 10 m resolution over the whole Africa with a consistent legend of 10 classes. This pioneer experiment faces several challenges including the big data management issues, the preprocessing chain development for improved and cloud screened Sentinel-2 surface reflectance, the precise definition of a scalable land cover typology, the land cover processing chain development and the design of reference database collection for validation.
The increase in spatial resolution requires indeed significant methodological adjustments and innovations to the processing chains developed for medium spatial resolution imagery at global scale. For the pre-processing, for example, the topography and adjacency effects have to be taken into account in the atmospheric correction. In addition, due to the lower revisiting capacity of high spatial resolution sensors such as Sentinel-2, the spatial consistency of surface reflectance between few images becomes a critical aspect in the production of high spatial resolution composites.

Sentinel-2 imagery requires several quality control procedures in order to be processed by large scale processing facilities. From a data management point of view, 72 TB of Sentinel-2 data have been downloaded and preprocessed into surface reflectance L3. A surface reflectance product covering one month of acquisition corresponds to 10 TB and continental surface reflectance mosaic reaches 2 TB of data.

For the LC classification, challenges are of a different nature. The decametric resolution captures the landscape elements diversity and distinct evolution through time due to slightly different seasonality and ecological gradients. Specific effort is made to ensure the consistency between this decametric map and the medium resolution global LC maps already developed within the Climate Change Initiative. While the rich literature on LC mapping at high resolution supports the processing chain development, the data flow provided by Sentinel-2 forces to revisit the classification strategy to map the LC consistently over space and time.

A review of various regional and global mapping efforts completed at different scales from 30 to 300 m resolution, a land cover typology is proposed and currently tested over different regions. The compilation and harmonization of all available land cover maps have been completed in order to support the processing strategy. A set of 10 test sites widely distributed in Africa and representative of different EO conditions and ecoregions allows benchmarking several automated methods. These benchmarking results support the processing chain development in order to optimize the performances of the each step. Finally, results over the whole Africa as well as regional results are presented, highlighting the potential of Sentinel-2 for global land cover mapping and the challenges ahead.

The new global policy framework for the sustainable development of urban areas calls for timely, consistent and accurate information on human settlements. Free and open earth observation data (e.g. Landsat, Sentinel) offer a great potential for large area mapping of human settlements. The Global Human Settlement Layer (GHSL) is the first open and free information layer describing the spatial evolution of human settlements in the past 40 years. It has been produced from Landsat image collections (1975, 1990, 2000 and 2014) and publically released on the JRC open data portal. The recent availability of Sentinel-1 and Sentinel-2 data is expected to bring land cover mapping and monitoring to an unprecedented level. With the great advantage of being free and immediately available for the users, Sentinel data can provide up-to-date global information on the status and evolution of human settlements. With the shift to Sentinel imagery, regular updates and incremental improvements of the GHSL will become more feasible and reliable. This study presents the recent developments in global mapping of human settlements with Sentinel-1 data. Taking advantage of the capabilities offered by the Symbolic Machine learning approach and the functionalities of JRC Big Data infrastructure, the challenges posed by the processing and analytics of the Sentinel-1 global coverage were effectively addressed. In view of the future deployment of the GHSL framework on Sentinel-2 data, a benchmark experiment over selected European cities has been performed in order to assess the added-value of Sentinel-1 and Sentinel-2 with respect to Landsat for improving global high-resolution human settlement mapping. The results show that noticeable improvement could be gained from the increased spatial detail and from the thematic contents of Sentinel-2 compared to the Landsat derived product as well as from the complementarity between Sentinel-1 and Sentinel-2 images.

From the beginning of the years 2000, more than half of the global population is living in urban environments and the dynamic trend of urbanization is growing at an unprecedented speed. Accordingly, an effective monitoring of urbanization represents a key issue to analyze and understand the complexity of human settlements and ensure their sustainable development.

To this purpose, starting from the last decade different global maps outlining urban areas have started being produced. In this framework, the two currently most largely employed are JRC’s Global Human Settlement Layer (GHSL) derived at 38m spatial resolution from Landsat data and, especially, DLR’s Global Urban footprint (GUF) derived at 12m spatial resolution from TanDEM-X/TerraSAR-X data. However, it is worth noting that, despite generally accurate, these layers still exhibit both some over- and underestimation issues. Specifically, this is mostly due to the fact that they have been generated by means of: i) single-date scenes (which can be strongly affected by the specific acquisition conditions) and ii) solely using either optical or radar data, which are sensible to different structures on the ground (e.g., with optical imagery bare soil and sand generally tend to be misclassified as urban, while this does not occur with radar data; on the contrary, with radar imagery complex topography areas or forested regions can be wrongly categorized as urban areas, whereas this does generally not happen if optical data are employed).

In order to overcome these limitations, in the framework of the ESA SAR4Urban project, we have developed a novel methodology that jointly exploits multitemporal optical and radar data for automatically outlining urban areas. In particular, the basic assumption of the intended approach is that the temporal dynamics of urban settlements over time are sensibly different than those of all other non-urban classes. Hence, given all the multitemporal images available over the region of interest in the selected time interval we first extract key temporal statistics (i.e., temporal mean, minimum, maximum, etc.) of: i) the original backscattering value in the case of radar data; and ii) different spectral indices (e.g., vegetation index, built-up index, etc.) derived after performing cloud masking in the case of optical imagery. Then, different classification schemes based on Support Vector Machines are separately applied to the optical and radar temporal features, respectively, and, finally, the two outputs are properly combined together.

At present, the technique is being employed for generating the so-called GUF+ 2015, a global map of urban areas at 10m spatial resolution derived jointly using the TimeScan-Landsat 2015 product (i.e., a dataset including temporal statistics for several spectral indexes derived from ~420,000 Landsat-7/8 scenes produced within the ESA Urban-TEP platform) and temporal statistics of Sentinel-1 IW GRDH data computed globally using Google Earth Engine. The whole classification activities are supported by the Urban-TEP infrastructure and the GUF+ is expected to be completed by March 2017.

Experimental results are extremely promising and confirm the great potential of combining optical and radar imagery and the higher accuracy of the GUF+ compared to the other existing layers.

Ten years of operations of the Envisat ASAR instrument have generated an invaluable archive of repeated observations of the SAR backscatter over land masses. The potential of such data is still being unravelled in applications despite the Envisat mission ended in 2012. The CCI Water Bodies dataset is one example of a global thematic dataset encompassing the ASAR data archive. More in general, thematic applications over land have been possible in spite of an uncoordinated acquisition strategy. Only at coarse resolution (1,000 m), the complement of all ASAR acquisition modes yielded wall-to-wall repeated coverage throughout the Envisat mission. Multi-temporal metrics of the SAR backscatter were found to overcome typical confusion between water and land occurring in single images under windy or frozen conditions. Nonetheless, such observables where not totally unique over water since specific land surface types such as glaciers and sand dunes were characterized by similar values under specific imaging conditions.

With the Sentinel-1 mission, it is envisaged that several caveats identified when mapping inland water bodies with ASAR are somewhat overcome. This appears to be the case after two years of operations and the operations of two satellites. Repeated acquisitions are planned according to a predefined strategy aiming at maximizing the information content in the scene observed. Stacks of multi-temporal observations, often in dual-polarization, are being created. The 6-12 days acquisition rate over Europe and other intensively observed regions even opens possibility to track water seasonality, which was only possible with the ASAR mission locally or for a short time period.

Here, two global scale applications of the Envisat ASAR data archive to map water bodies and some examples of the contribution of Sentinel-1 to map water bodies are presented.

The SAR Water Body Indicator derived to support the CCI Water Body Dataset is briefly reviewed. We then present the contribution of ASAR multi-year observations (2005-2012) to capture inland water dynamics at 1,000 m with a weekly time step. A novel approach based on the functional relationship between ASAR backscatter and local incidence angle is applied. The water seasonality appears to be well identified in the northern hemisphere thanks to the very dense ASAR observations in time. In regions characterized by small water bodies and dynamics, or when the data sampling was irregular, the dynamics appear to be underrepresented.

While the detection of water bodies with ASAR had to rely on a sophisticated construct and required multi-temporal observations, the availability of cross-polarized backscatter from the Sentinel-1 satellites relaxes the constraints on the input data source and allows for improved thematic accuracy. In boreal landscapes, the detection of water bodies using a simple threshold-based approach on a summer mean of cross-polarized backscatter images performed at 20 m with over 90% accuracy when compared to samples interpreted in high-resolution images. We are currently extending our investigations to other landscapes in Europe and Africa, here with a focus to complement the land cover mapping activities based on Sentinel-2 within the CCI Land Cover Project. Results will be presented at the conference.

The location and persistence of surface water is both affected by climate and human activity and affects climate, biological diversity and human wellbeing.
Global datasets documenting surface water location and seasonality have been produced, but measuring long-term changes at high resolution remains a challenge.
To address the dynamic nature of water, the European Commission’s Joint Research Centre (JRC), working with the Google Earth Engine (GEE) team, has processed each single pixel acquired by Landsat 5, 7, and 8 between 16th March 1984 and 10th October 2015 (> 3,000,000 Landsat scenes, representing > 1,823 Terabytes of data).
The produced dataset records the months and years when water was present across 32 years, where occurrence changed and what form changes took in terms of seasonality and persistence, and documents intra-annual persistence, inter-annual variability, and trends.
This validated dataset shows that impacts of climate change and climate oscillations on surface water occurrence can be measured and that evidence can be gathered showing how surface water is altered by human activities.
Freely available, we anticipate that this dataset will provide valuable information to those working in areas linked to security of water supply for agriculture, industry and human consumption, for assessing water-related disaster reduction and recovery, and for the study of waterborne pollution and disease spread. The maps will also improve surface boundary condition setting in climate and weather models, improve carbon emissions estimates, inform regional climate change impact studies, delimit wetlands for biodiversity and determine desertification trends. Issues such as dam building (and less widespread dam removal), disappearing rivers, the geopolitics of water distribution and coastal erosion are also addressed.

Amongst all land use change processes, the agricultural area change due to spatial expansion or cultivated lands abandonment is one of the most dynamic land cover change. Furthermore, cropland or agriculture areas corresponds to very diverse land features which vary over time as the interaction result of crop management practices and seasonal weather conditions. In order to capture the cropland evolution, the JECAM network has adopted a restrictive definition of cropland which corresponds to annually cultivated lands; in spite of its remote sensing perspective this definition still makes the cropland and crop type mapping particularly challenging.

Early 2017, the Sentinel-2 (S2) mission will reach the optimal capacity for cropland mapping and agriculture monitoring in terms of resolution (10-20 m), revisit frequency (5 days with two satellites) and systematic coverage (global). In order to exploit these new capabilities, specific methods for dynamic cropland mapping and main crop type classification have been developed in the framework of the Sentinel-2 for Agriculture project funded by ESA.

Dynamic Cropland masks correspond to a set of successive masks to depict annually cultivated areas. The production can rely on two alternative approaches depending on the availability or not of in-situ data. Both methods are based on a random forest classifier, trained in the first case with in-situ data and in the second case, with samples collected from an existing reference land cover map. The Crop Type map classifies the main crop groups, i.e. irrigated versus rainfed and summer versus winter crops. The map is produced at the half and at the end of the season using a random forest classifier over a combination of S2 and L8 time series.

During the 2016 growing season, these methods were applied at national scale over Ukraine, Mali and South Africa covering more than 500 000 km² in each country, where fast track nationwide field campaigns were organized by national partners. These demonstration cases delivered a new type of land mask delineating the cropland at 10 m resolution over the entire country for a given season and delivered less than few weeks after the last observation.

The mapping results obtained over Ukraine for 2016 have been thoroughly validated through an independent accuracy assessment. National workshops are also organized with various key users in order to discuss the timeliness and relevance of the products accuracy for different operational applications. A similar approach is going on for Mali and South Africa. The accuracy assessment results are very high for Ukraine and will be available for Mali and South Africa.

Beyond these very encouraging results of the Sen2Agri system, such an automatic production of high resolution land map obtained from freely and continuously available time series changes completely the classical remote sensing approach. Indeed this system is designed to deliver products on a yearly basis and to run over very large areas providing a new capability for regional to continental mapping. Based on the lessons learnt from the Sen2Agri system demonstration, challenges ahead are discussed towards a more general land cover mapping system.

Rice is the most important food crop in Asia and the mapping and monitoring of paddy rice fields is an important task in the context of food security, food trade policy, water management and greenhouse gas emissions modelling. Asia’s biggest rice producers are facing increasing pressure in terms of food security due to population and economic growth while agricultural areas are confronted with urban encroachment and the limits of yield increase. At the same time demand for rice imports is increasing, spurred by global population growth.

Despite the importance of knowledge about rice production the countries official land cover products and rice production statistics are of varying quality and sometimes even contradict each other. Available remote sensing studies focused either on time-series analysis from optical sensors or from Synthetic Aperture Radar (SAR) sensors. We try to address the sensor specific limitations by proposing a paddy rice mapping approach that combines medium spatial resolution, temporally dense time-series from the optical MODIS sensors and high spatial resolution time-series from the Sentinel-1 A/B SAR sensors.

We developed a method to use MODIS time-series and a one-class classifier to create medium resolution rice maps [1]. In a next step we used these medium resolution rice maps to mask Sentinel-1 Interferometric Wide Swath images, which limits the amount of data to process and allows efficient rice mapping over larger areas. The high resolution rice masks are then created by segmentation of multi-temporal SAR images into objects, from which backscatter time-series are derived and classified. We created 10m resolution rice-maps that also allow seasonality extraction, given enough Sentinel-1 acquisitions. This method allows concurrent, accurate and high resolution mapping of paddy rice areas from freely available data. Results of our paddy rice classification will be presented for selected study sites in Asia.

1. Clauss, K.; Yan, H.; Kuenzer, C. Mapping Paddy Rice in China in 2002, 2005, 2010 and 2014 with MODIS Time Series. Remote Sensing. 2016, 8, 434.

The need for accurate information on the state and evolution of tropical forest types at regional and continental scales is widely recognized, particularly to analyze the forest diversity and dynamics, to assess degradation and deforestation processes and to better manage these natural resources (Achard et al. 2014).

A few global and continental land cover or forest cover products have been derived from Landsat satellite imagery at 30m resolution: they either contain detailed thematic information without temporal dynamics (Chen et al. 2015, Giri and Long 2010) or contain information on forest-cover changes over long time periods (10 to 30 years) without thematic classes such as the discrimination of evergreen forests (Kim et al. 2014, Hansen et al. 2013, Potapov et al. 2015).

The objective of this study is to map undisturbed evergreen and semi-deciduous forests at 30m resolution over the full tropical humid domain and to better characterize the changes and disturbances which occurred during the last 33 years in these forests. Therefore we exploited the full archive of Landsat imagery between 1984 and 2016 and developed a pixel-based automatic methodology which includes four steps: (i) pre-processing of the Landsat time series with cloud masking and filtering of sensor artefacts, (ii) single-date image classification (driven by a large spectral library) into three basic classes (evergreen forest, vegetative non-forest cover and poorly/non vegetated cover), (iii) creating forest/non-forest maps for three epochs based on the occurrence of non-forest classes, and (iv) production of a final map of detailed forest types based on the temporal succession of observed basic classes from 1984 to 2016.

The resulting map includes six classes: undisturbed forest cover, old and young vegetation regrowths, old or recent deforested areas (during last 10 years), recently disturbed areas (during last 3 years) and other land cover. This map at 30 m resolution allows the identification of small linear features such as gallery forests and of small disturbance events such as skid trails and logging decks. The use of a 33-year Landsat time series allows (i) to identify most deforestation and degradation events (when > 0.1 ha) that occurred during this period, (ii) to provide the dates of the forest disturbances, and (iii) to considerably reduce the confusion that usually occurs with small scale agricultural fields (shifting cultivation, tree plantation and irrigated crops…). Finally we characterize the deforested and disturbed classes by providing their timing and occurrence (date of first and last events, number of events).

The accuracy of the forest map was assessed over Africa from an independent sample of reference data (3830 plots) created through visual expert interpretation of Landsat imagery at several dates and finer resolution satellite imagery with an overall agreement of 90%. The pan-tropical map and the accuracy assessment results over Africa will be presented at the conference.

It is intended in the future to adapt and apply the methodology on Sentinel-2 data for a better characterization of forest-cover disturbances.

Remote sensing is an important tool for understanding the spatial and temporal dynamics of forest disturbances over large areas. The most recent developments in disturbance detection algorithms utilize dense time series information that enables the detailed characterization of abrupt and gradual disturbance events. These advances allow the mapping of a wide variety of disturbance agents, including harvest, fire, blowdown, and insect attacks. However, current algorithms have primarily been developed and tested in the forest ecosystems of North America, which are characterized by relatively homogeneous coniferous forests, little to none management, and medium to large-scale disturbance patches. Those algorithms have not been validated for mapping forest disturbances in Europe yet, where forest ecosystems are much more variable in terms of species composition, landscape structure, and forest management. Consequently, our aim was to evaluate current disturbance detection algorithms for mapping forest disturbance in Europe by i) comparing spectral-temporal characteristics of forest disturbances across a sample of five protected forest areas in the temperature forests of central Europe; and by ii) comparing disturbance characteristics inside the protected forests to the surrounding unprotected forests in order to understand the compound effect of natural and management disturbances on spectral-temporal characteristics. We utilized dense Landsat time series from the USGS and ESA archives covering the time period 1985 to 2016. We mapped forest disturbance characteristics (i.e., magnitude and duration) using state-of-the-art time series tools in conjunction with random forests classification and a set of photo-interpreted reference plots. Primary results show that disturbances were detected with high accuracies (>90% overall accuracy) across all sites. Spectral-temporal characteristics varied substantially within and outside protected forests. In particular, unmanaged areas showed more long-term disturbances – likely related to blowdown and following bark beetle mortality – whereas disturbances in managed forests were mainly short and of high magnitude. However, temporal patterns of disturbances were similar inside and outside the protected areas, suggesting that spectral disturbance patters outside protected forests are superimposed by a strong management signal. Our analysis improves the understanding of forest disturbances – and how to map them using remote sensing – for European temperate forests, a forest ecosystem of high economic and ecological value. As such, this study paves the way for using long and dense time series of optical satellite data, i.e. Landsat and Sentinel-2, for mapping and understanding forest dynamics across Europe.

Wetlands are amongst the planet‘s most productive ecosystems providing a wealth of ecosystem services, e.g., nutrition, flood control and protection, or support of biodiversity, to name a few. Nevertheless, wetlands are exposed to multiple threats due to climate change, agricultural pressure, hydrological modifications, fragmentations, etc. Thus, a consistent mapping and monitoring of global wetland ecosystems is very important to track changes and trends aiming at supporting wetland conservation and sustainable management. Although EO data are ideal for large scale inventorying of wetlands, the large diversity of them makes remote detection particular challenging. This diversity and resulting challenge has been tackled by many researchers applying different sensors (optical and radar) and mapping techniques to delineate wetland from non-wetland areas. A global and homogenous inventory, however, is still not available and scope of ongoing research.

Herein, we present an innovative and operational water and wetness product building on data from Sentinel-1 SAR and Sentinel-2 MSI complemented with historical data from the Landsat missions. Rather than trying to detect wetlands in the ecological sense, we derive wetlands in the physical meaning identifying the wetness of the underlying land surface.

Using a hybrid sensor approach, i.e., combining optical and radar observations, provides a more robust wetland delineation with optical imagery being more sensitive to the vegetation cover and radar imagery to soil moisture content. Additionally, the higher frequency of observations stemming from the combined data streams contributes to a better characterization of seasonal dynamics which is important so that seasonal and temporary changes do not lead to false conclusions of the overall long-term trend in wetland extent. Within the domain of optical remote sensing, the identification of wetlands is based on the enhancement of the spectral signature using bio-physical indices sensitive to water and wetness and subsequent derivation of a water and wetness probability index. The radar-based algorithm builds on geophysical parameters, surface soil moisture dynamics and water bodies, derived from historical Envisat ASAR and Sentinel-1 backscatter time series to identify permanent/temporary wet and flooded areas. In addition, it is possible to identify flooded vegetation according to the double-bounce scattering principle in densely vegetated wetlands. The non-flood prone areas are masked using the Height Above Nearest Drainage (HAND) index. After the separate processing of the optical and radar imagery, the data is fused into a combined water and wetness product. With our methods we aim at detecting the current status of wetland areas, but also to capture the historic evolution taking into account the past 25 years in a fully automated manner.

The above described methods are currently applied for several large regional sites throughout Africa within the GlobWetland Africa project and for the Pan-European production of the “water and wetness” High Resolution Layer of the Copernicus Land Monitoring Service. We will further present a thorough validation of the product for different wetland ecosystems and discuss remaining issues, mainly due to global data availability and coverage.

Land-cover and Land Use information are important parameters in most of the studies related to natural environment, ecosystem services and many other important disciplines. However despite its importance and the many efforts toward data harmonization (especially for LC) do not exist an accepted model on how to link and functionally correlate those two information. On the contrary there is often a contamination of LCLU terms in many LC nomenclatures (Anderson, Corine, etc) and surprising also in some LU classifications (UNFCC, NLUD, etc). Even when the two information are keep correctly clearly distinct (E.U. Inspire spatial data infrastructure) no effort is made to model/describe their functional relationship. FAO (and UNEP) have a long prominent role on the efforts to develop standardized LCLU classification and data harmonization. Especially in LC the development of the LCCS parametric method and subsequently the “object oriented” approach underlining the LCML (Land Cover Meta-Language) model (ISO Standard 19144-2) has open a new path forward for the representation/harmonization of LC information. Based on this experience and using as base part of the original LCML UML schema, a new model is under development, the LCHML (Land Characterization Meta-Language). LCHML not only propose a revised LC and a new LU model but also try to create a comprehensive standardized framework were is possible to create an exhaustive and functional correlation of both biophysical and human related activities. LCHML therefore try to integrate in a unique model both LC and LU. The objective is to create a standardized framework were it is possible to describe any geographic area from different perspectives: pure LC, pure LU, or a functional combination of the two called (tentatively) LCH (Land Characterization).

The Copernicus Programme, headed by the European Commission (EC) in partnership with the European Space Agency (ESA), offers Earth observation-based services for six core thematic areas: Land, Atmosphere, Oceans, Climate Change, Emergency and Security. Among these services – mainly based on Earth Observation (EO) data provided by ESA through the Copernicus Space Component – the Copernicus Land Monitoring Service delivers products on local, continental and global levels. As part of the pan-European Copernicus Land Service, coordinated by the European Environment Agency (EEA), the High Resolution Layers (HRLs) map multi-temporal land cover characteristics for five thematic areas (Imperviousness, Forest, Grassland, Water/Wetness, Small Woody Features) in 20 meters spatial resolution and in a consistent manner for 39 European countries. All thematic HRLs contain specific information on current environmental conditions and temporal variance of major land cover types with thematic accuracies exceeding 80–90% (depending on the product). The HRL products are tailored towards a multi-user community and are freely provided for download on the Copernicus website.

With the current production for the 2015 reference year, the HRLs are entering the era of big data multi-temporal image processing, incorporating large data volumes from different sensors in a decentralized processing framework in a network of industrial service providers. Our contribution will describe the framework, methodology and first results of the current HRL 2015 production comprising the update of the existing (2012) pan-European HRLs Imperviousness and Forest, including 2012–2015 change products, as well as 2015 mapping of other newly defined HRLs (Grassland, Water/Wetness, Small Woody Features).

Primary information source are multi-temporal, high-resolution satellite images from Sentinel-1 and -2, as well as data from SPOT, Resourcesat and Landsat contributing missions. Whereas the 2015 Forest and Imperviousness HRLs will be produced based on optical time series imagery, the newly defined Grassland and Water/Wetness products will benefit from innovative approaches on the basis of a fusion of optical and synthetic aperture radar (SAR) time series data. The novel HRL on Small Woody Features is the HRL using very high resolution (VHR) data as primary input. The VHR data sets will also be used for reference data collection and validation alongside national and pan-European in-situ data sets.

The full chain of image and in-situ data acquisition, pre-processing, generation of biophysical variables, multi-temporal image classification and validation will be demonstrated, and first results will be presented for all five HRLs. Semi-automatic classification techniques based on multi-temporal pixel-based as well as segment-based approaches are applied specifically tailored towards each HRL, resulting in raster products in full 20m resolution, as well as vector products on 1:5000 scale for the Small Woody Features.

The Copernicus HRLs are designed for a broad user community as basis for environmental and regional geo-spatial analyses as well as for supporting political decision-making. With future updates, the HRLs will significantly benefit from ESA’s growing Sentinel-1/-2 archive, further improving the products’ consistency, timeliness and accuracy. An outlook concludes on the potential usability of the presented methods and products for future European to global LC/LU applications on a HR scale.

The National Land Cover Database (NLCD) is an example of a national land cover change monitoring system that incorporates user requirements, scientific advances and the results from rigorous accuracy assessments to provide accurate and current data products that are useful to land managers and the public. It is managed by a consortium of United States governmental agencies, the Multi-Resolution Land Characteristics (MRLC) consortium, that require land cover information to assess environmental quality and promote the sustainable use of natural resources. This consortium is a collaborative forum, where members share research, methodological approaches, and data to establish protocols promoting the development and use of integrated land cover data products. The NLCD began as a one-time land cover thematic mapping effort of the conterminous US in 1992 and now encompasses four epochs (1992, 2001, 2006 and 2011) of thematic land cover data, as well as continuous field datasets such as percent impervious surface, required for water quality assessments and percent canopy cover, required for biodiversity, biomass and carbon sequestration assessments. All datasets are derived from Landsat imagery and so have a spatial resolution of 30 metres by 30 metres. Monitoring land cover change is accomplished by integrating a remote sensing image change analysis with a knowledge-based system on land change. This system consists of rules and/or attributes derived from the spectral, spatial, and temporal characteristics of remote sensing data and historical knowledge on land cover change and its trajectories. All of the datasets have undergone rigorous accuracy assessments that have been used to guide continuous improvements in the data, as well as advancing remote sensing science. The consortium has stressed advancing the technological aspects of land cover data production, including data preprocessing, classification methodologies and advancing an integrated database paradigm that enables change monitoring over time. The result has been a dramatic decrease in the amount of time required to create a dataset and its cost, as well as the identification of new compatible datasets such as the newly formulated percent bareground and shrub cover. The results of the consortium’s efforts have been a suite of land cover data products that provide valuable, tangible societal benefits. These benefits include: water quality monitoring and assessment; identifying potential relationships between land cover patterns and human well-being; assessing the impacts of natural disasters on ecosystem services; and influencing federal energy policies on land use change and potential environmental degradation that may arise from land use change.

A component of the EU-funded ECOPOTENTIAL project funded under the Horizon 2020 Program (Reference 641762) has been the development of the EO Data for Ecosystem Monitoring (EODESM) system. This system allows classification of land covers according to the Food and Agricultural Organisation (FAO) Land Cover Classification System (LCCS-2) taxonomy from EO-derived biophysical and thematic information. The presentation will a) demonstrate the EODESM system, giving examples from protected areas in Europe, Africa and the Middle East, and its wider application (including at the global level), and b) convey the uncertainty associated with including layers generated at different scales within the classification.

The EODESM system has been developed such that LCCS classifications can be generated from pre-prepared (including global) data layers as well as satellite data (e.g., as acquired by the Sentinel’s and Landsat). These layers relate to the broader LCCS Level 3 categories (natural/semi-natural and cultivated/managed landscapes, natural and artificial bare areas, and water) and include additional information (including lifeform, structure, phenology, hydroperiod, and urban density). Inputs can also include existing thematic information (e.g., cadastral and infrastructure maps), modeled outputs (e.g., hydrological) or knowledge. These layers are used to generate the component LCCS codes (e.g., A3 or A4 for trees or shrubs, D1 or D2 for broadleaved or evergreen), which are then combined. The system has also been adapted to consider changes in these codes but also biophysical and other attributes (including those unrelated to the classification; e.g., above ground biomass), and can attribute changes to specific causes and indicate or suggest consequences and the impacts on, for example, protected areas. An advantage is that it can be applied at any scale (local to global)and also integrate data from sensors operating in different modes (lidar, radar, optical) and spatial and temporal resolutions. Furthermore, consistent and highly detailed land cover classifications are provided together with a diverse range of information describing landscapes (e.g., biomass, hydrology, Leaf Area Index, soil moisture). The system is also well suited to follow the LCCS-3 or Land Cover Macro Language (LCML) and is being modified accordingly. In the context of ECOPOTENTIAL, the EODESM approach has been applied to a wide range of protected areas and their surrounds with the intention of providing timely and historical information to ensure protection and facilitate restoration of ecosystem services. The EODESM system has potential for global application and this will be conveyed within the presentation with reference to other land cover classification schemes. The EODESM system will be made available through a Virtual Laboratory being prepared by the ECOPOTENTIAL project and integrated within GEOSS. Land cover maps and change maps will also be accompanied by comprehensive metadata including data quality documentation.

Traditionally remote sensing of land cover has focused on mapping land cover types, with monitoring being devoted to finding the conversions between land cover types. As time series of observations have become available, it is now possible to detect more subtle characteristics about landscapes, including trends in ecosystem health and condition. Using many observations it is possible to separate effects related to seasonality from those providing longer term indications of the condition of ecosystems. For example, it is possible to observe trends indicative of growth in forests, decline related to pests or degradation, and interannual variability related to climate. One particularly interesting case is to monitor the recovery of ecosystems following disturbance. Additionally, it is possible to begin to identify events that influence ecosystems, such as extreme weather. The net effect is the ability to derive more subtle information about land cover that will prove helpful in both ecosystem modeling efforts and land management.

In the framework of the ESA Climate Change Initiative (CCI), the Land Cover (LC) team delivered a new generation of satellite-derived series of 300 m global annual LC products spanning a period from 1992 to 2015. These maps, consistent in space and time, were obtained from AVHRR (1992 - 1999), SPOT-Vegetation (1999 - 2012), MERIS (2003 - 2012) and PROBA-V (2012 - 2015) time series. The typology was defined using the UN Land Cover Classification System (LCCS) and counts 22 classes compatible with the GLC2000, GlobCover 2005 and 2009 products.

A critical step in the acceptance of these products by users is providing confidence in the products’ quality and in their uncertainties through validation against independent data. Building on the GlobCover experience, a validation strategy was designed to globally validate the maps series and change at four points in time: 2000, 2005, 2010 and 2015.

In order to optimize data and resources availability, 2600 Primary Sample Units (PSUs), defined as a 20 × 20 km box, were selected based on a stratified random sampling. Over each PSU, five Secondary Sample Units (SSUs) were defined, located at the center of each 20 × 20 km box. This cluster sampling strategy increased the number of sample sites and thus lowered standard error of accuracy estimates.

An international network of land cover specialists with regional expertise was in charge of interpreting high spatial resolution images over SSUs to build the reference database that eventually allowed assessing the accuracy of the CCI global LC map series.

Over each SSU, the visual interpretation of very high resolution imagery close to 2010 allowed labeling each object derived by segmentation according to the CCI LC legend. Change between 2010, 2005 and 2000 was then systematically evaluated on Landsat TM or ETM+ scenes acquired from the Global Land Surveys (GLS) for the respective years. Annual NDVI profiles derived from the SPOT-Vegetation time series facilitated image interpretation by providing seasonal variations of vegetation greenness. This reference validation database was then complemented for 2015 thanks to the large FAO data set obtained using the Collect Earth tool.

Overall, users and producers accuracies of the CCI LC map series are then derived. In addition, various quality indices, related to specific use in the climate models (carbon content, net primary productivity, methane emissions, etc.), will be constructed by taking into account the semantic distance between LC classes.

Such validation scheme was made possible thanks to a tailored user-friendly validation web interface which integrated large amount of very high spatial resolution imagery, pixel-based NDVI profiles from various sensors, as well as GIS interactive tools to facilitate LC and change interpretation. This efficient web platform was already selected by several other international initiatives to collect reference data sets as this unique object-based approach provides reference database which can be used to validate land cover maps at different resolution.

The validation of datasets such as the Copernicus Pan-European Tree Cover Density (TCD) high resolution layer (TCD) and UMD Global Forest Change (GFC) Tree percent requires considerable effort to provide validation results at Pan-European level over nearly 6 million km². A stratified systematic sampling approach was developed based on the LUCAS sampling frame. A two-stage stratified sample of 17,296 1ha square primary sampling units (PSU) was selected over EEA39 based on countries or groups of countries which area was greater than 90,000km² and a series of omission and commission strata. In each PSU, a grid of 5 x 5 Secondary Sample units (SSUs) with a 20 m step was applied. These points were photo-interpreted on orthophotos with a resolution better than 2.5m.

The UMD GFC data was processed to provide a 2012 Tree Percent layer comparable to the Copernicus High Resolution Layer by including Tree losses and gains over the selected period. An appropriate interpolation procedure was then applied to both the UMD GFC and Copernicus HRL TCD to provide a precise match with the PSU validation data and account for potential geographic differences between validation and map data and SSU sampling errors.

Initial results based on the binary conversion of the HRL TCD data by applying a 10, 15 and 30% thresholds indicate a level of omission errors in line with the required maximum level of 15% and exceeds the target level of 10% for commission errors set in the product specifications. However, disaggregated results were also computed at country / group of countries level as well as for biogeographical regions and showed considerable geographical variability. Best results were obtained in countries or biogeographical regions with high tree cover (e.g. Continental and Boreal regions) and the worst results in countries or biogeographical regions with low tree cover (e.g. Anatolian, Arctic). There is less variability between production lots and the outcome of the analysis of the scatterplots show that there is a strong relationship between validation and map data with the best results in the countries and biogeographical regions mentioned previously. However, there seems to be a general trend to slightly underestimate TCD. Results for the UMD GFC dataset are currently in progress but will be made available in time for the conference. Results provided at HRL TCD production lot, bio-geographical regions and country/group of country level, should provide a sound basis for targeting further improvement to the products.

The stratification procedure was based on a combination of the HRL TCD Layer and CORINE Land Cover (CLC) was effective for commission errors, but less so for omission errors. Therefore, the stratification should be simplified to only include a commission (tree cover mask 1-100) and an omission stratum (rest of the area) or an alternative stratification to CLC should be applied to better target omission. Finally, the approach developped could be rolled out at global scale for the complete validation of global datasets.

The main objective of the Copernicus Global Land – Hot Spot Monitoring Service is to provide detailed land information on specific areas of interest, including protected areas or hot spots for biodiversity and land degradation. For such areas of interest, land cover and land cover change products which are mainly derived from medium (Landsat, Sentinel2) and high resolution satellite data are made available to global users. The service directly supports field projects and policies developed by the European Union (EU) in the framework of EU’s international policy interests. It is coordinated by the Joint Research Center of the European Commission and answers ad-hoc requests and focus mainly within the domain of the sustainable management of natural resources.

Comprehensive, independent validation and accuracy assessment of all thematic map products is performed before providing the land cover and land cover change products to the global users. The following rigorous methodology is conducted:

Spatial, temporal and logical consistency is assessed by determination of the positional accuracy, the assessment of the validity of data with respect to time, and the logical consistency of the data e.g. topology, attribution and logical relationships.

A Qualitative-systematic accuracy assessment is performed wall-to-wall by a systematic visual examination of the land cover and land cover change maps within a geographic information system and their accuracies are documented in terms of type of errors.

For quantitative accuracy assessment, a stratified random sampling approach is implemented which is based on inclusion probabilities. A web-based interpretation tool based on PostgreSQL is implemented which provides high resolution time series imagery e.g. from Sentinel 2, derived temporal trajectories of reflection, and ancillary information in addition to the very high resolution imagery. In order to quantify the uncertainty of the derived accuracy measures, confidence intervals are derived by analytic formulas as well as by applying bootstrapping.

Area Estimation is performed according to the requirements for international reporting. In general, the errors of omission and errors of commission are not equal and such bias shows up in the course of the quantitative accuracy assessment. As the implemented approach applies probability sampling and the error matrix is based on inclusion probabilities, area estimates are directly derived from the error matrix. The area estimates are complemented by confidence intervals.

The approach and methodology will be discussed in detail on the basis of a systematic accuracy assessment and area estimation results for sites in Africa which are the focus in the first year of implementation of the Copernicus Global Land Hot Spot Monitoring Service.

REDD+, Reducing emissions from deforestation and forest degradation, is a mechanism developed by Parties to the United Nations Framework Convention on Climate Change (UNFCCC). It creates a financial value for the carbon stored in forests by offering incentives for developing countries to reduce emissions from forested lands and invest in low-carbon paths to sustainable development. Developing countries would receive results-based payments for results-based actions. Forest monitoring is always remained challenging to asses with the least error. So far, a lot of countries started to report on the deforestation on a national scale, but the technical issue to assess the degradation in-country is still a major point of research. Remote sensing options are actually being explored in order to map degradation.

The ForMosa project aim was to develop a sound methodology for forest degradation monitoring for REDD+ using medium coarse satellite data sources such as Landsat-8, Sentinel-2, and SPOT 5 in addition to Rapid Eye imagery. The project is carried out by three partners: Planet, the Wageningen University (Service Providers) and the Forest department of FAO, which carried out the accuracy assessment of the project products. Initially three pilot study sites, Kafa Tura in Ethiopia, Madre de Dios in Peru and Bac Kan in Vietnam were selected. The initial product developed at 10 m resolution with five classes, representing different levels of forest degradation with increasing intensity.

The Forest department of FAO used the in-house built open source tools for the accuracy assessment. The process consists of four steps, (i) map data, (ii) sample design, (iii) response design and (iv) analysis. A stratified random sampling approach was used to access the product by using high-resolution Google earth and Bing map imagery. In this paper, the methodology of the work and its results will be presented.

Classification techniques are dependent on the quality and quantity of reference data, where the data should represent different land cover types, varying landscapes and be of a high overall quality. In general there is currently a lack of reference data for large scale land cover mapping. In particular, reference data are needed for a new dynamic 100m land cover product that will be added to the Copernicus Global Land services portfolio in the near future. These reference data must correspond to the 100m Proba-V data spatially and temporally. Therefore the main objectives of this study are as follows: to develop algorithms and tools for collecting high quality reference data; to collect reference data that correspond to Proba-V data spatially and temporally; to develop a strategy for validating the new dynamic land cover product and to collect validation data.

To aid in the collection of reference data for the development of the new dynamic land cover layer, the Geo-Wiki Engagement Platform (http://www.geo-wiki.org/) was used to develop tools and to provide user-friendly interface to collect high quality reference data. Experts, trained by staff at IIASA (International Institute for Applied Systems Analysis) interpreted various land cover types of a network of reference locations via the Geo-Wiki platform. At each reference location, experts were presented with 100m x 100m area, subdivided into 10m x 10m cells superimposed on high-resolution Google Earth or Microsoft Bing imagery. Using visual cues, acquired from multiple training sessions, the experts interpreted each cell into land cover classes which include trees, shrubs, water, arable land, burnt areas, etc. This information is then translated into different legends using the UN LCCS (United Nations Land Cover Classification System) as a basis. The distribution of sample sites is systematic, with the same distance between sample sites. However, land cover data are not collected at every sample site as the frequency depends on the heterogeneity of land cover types by region. Data quality is controlled by ensuring that multiple classifications are collected at the same sample site by different experts. In addition, there are control points that have been validated by IIASA staff for use as additional quality control measures.

As a result, we have collected the reference data (at approximately 20000 sites) for Africa, which has been chosen due to its high density in complex landscapes and areas covered by vegetation mosaics. Current efforts are underway to expand the reference data collection to a global scale.

For validation of the new global land cover product, a separate Geo-Wiki branch was developed and accessed by local experts who are not participated in the training data collection. A stratified random sampling procedure is used due to its flexibility and statistical rigorousness. For the stratification, a global stratification based on climate zones and human density by Olofsson et al (2012) was chosen owing to its independence from existing land cover maps and thus its suitability for future map evolutions.

The reference datasets will be freely available for use by the scientific community.

Long-term data records of Earth Observation data are a key input for climate change analysis and climate models. The goal of this research is to create cloud free surface reflectance composites over Africa using Sentinel-2 L1C TOA products, to extent end enrich a time series from multiple sensors (MERIS, SPOT VGT, Proba-V and AVHRR). While the focus of previous work was to merge the best available missions, providing near weekly optical surface reflectance data at global scale, to produce the most complete and consistent possible long-term data record, Sentinel-2 data will be used to map a prototype African land cover at 10-20 meters. To achieve this goal the following processing methodology was developed for Sentinel-2: Pixel identification, atmospheric correction and compositing. The term “Pixel identification” – IdePix – refers to a classification of a measurement made by a space borne radiometer, for the purpose of identifying properties of the measurement which are influencing further algorithmic processing steps. Most importantly is the classification of a measurement as being made over cloud and cloud shadow, a clear sky land surface or a clear sky ocean surface. This step was followed by atmospheric correction including aerosol retrieval to compute surface directional reflectance. The atmospheric correction includes the correction for the absorbing and scattering effects of atmospheric gases, in particular ozone, and water vapour, of the scattering of air molecules (Rayleigh scattering) and the correction of absorption and scattering due to aerosol particles. All components except aerosols can be rather easily corrected because they can be taken from external sources or can be retrieved from the measurements itself. Aerosols are spatially and temporally highly variable and the aerosol correction is the largest error contributor of the atmospheric correction. The atmospheric correction particularly in case of high-resolution data like Sentinel 2 data has to take into account the effects of the adjacent topography or terrain. Furthermore, the final step of the atmospheric correction should be an approximate correction of the adjacency effect, which is caused by atmospheric scattering over adjacent areas of different surface reflectance, and is required for high spatial resolution satellite sensors. The source of uncertainty associated with the atmospheric correction are observation and viewing geometry angles, aerosol optical thickness and aerosol type, digital elevation model, accuracy of ortho-rectification, pixel identification, atmospheric parameter (e.g. water vapour column), and accuracy of spectral / radiometric calibration. All sources of errors are taken into account for the uncertainty calculation, with only one exception, which corresponds to the pixel identification. In case of the uncertainty estimation for the Sentinel 2 data, the Monte Carlo simulation, a mostly used modelling approach, will be applied. Afterwards the data were binned to 10-day cloud free surface reflectance composites including uncertainty information on a specified grid. The used compositing technique includes multi-temporal cloud and cloud shadow detection, to reduce their influence. The results will be validated against CEOS LANDNET and RadCalNet sites measurements. This very large scale feasibility study should pave the way for regular global high resolution land cover mapping.

Mapping land cover signatures with satellite SAR sensors has in the past been significantly constrained by topographic effects on both the geometry and radiometry of the backscatter products used. To avoid the significant distortions introduced by strong topography to radiometric signatures, many established methods rely on single track exact-repeat evaluations, at the cost of not integrating the information from revisits from other tracks.

Modern SAR sensors offer wide swaths, enabling shorter revisit intervals than previously possible. The open data policy of Sentinel-1 enables the development of higher level products, built on a foundation of level 1 SAR imagery that meets a high standard of geometric and radiometric calibration. We systematically process slant or ground range Sentinel-1 data to terrain-flattened gamma nought backscatter. After terrain-geocoding, multiple observations are then integrated into a single composite in map geometry.

Although composite products are ubiquitous in the optical remote sensing community (e.g. MODIS), no composite SAR backscatter products have yet seen similar widespread use. In the same way that optical composites are useful to avoid single-scene obstructions such as cloud cover, composite SAR products can help to avoid terrain-induced local resolution variations, providing full coverage backscatter information that can help expedite multitemporal analysis across wide regions. The composite products we propose exhibit improved spatial resolution (in comparison to any single acquisition-based product), as well as lower noise. Backscatter variability measures can easily be added as auxiliary channels.

We present and demonstrate methods that can be applied to strongly reduce the effects of topography, allowing nearly full seamless coverage even in Alpine terrain, with only minimal residual effects from fore- vs. backslopes.

We use data from the Sentinel-1A (S1A), Sentinel-1B (S1B), and Radarsat-2 (RS2) satellites, demonstrating the generation of hybrid backscatter products based on multiple sources. Unlike some other processing schemes, here data combinations are not restricted to single modes or tracks. We define temporal windows that support ascending/descending combinations given data revisit rates seen in archival data. Next, that temporal window is cycled forward in time merging all available acquisitions from the set of satellites chosen into a time series of composite backscatter images that seamlessly cover the region under study. We demonstrate such processing over the entirety of the Alps, as well as coastal British Columbia, and northern Nunavut, Canada. With S1A/S1B combinations, we demonstrate full coverage over the Alps with time windows of 6 days. Results generated at medium resolution (~90m) are presented together with higher resolution samples at 10m.

The radar composites demonstrated offer a potential level 3 product that simplify analysis of wide area multi-temporal land cover signatures, just as e.g. 16-day MODIS composite products have in the optical domain.

Use of the Radarsat-2 data was made possible through the SOAR-EU programme, and an initiative of the WMO’s Polar Space Task Group SAR Coordination Working Group (SARCWG). This work was supported by a subcontract from ESA Contract No. VEGA/AG/15/01757.

Time series remote sensing data propose important features for land cover and cover change mapping and monitoring due to its capability in capturing intra and inter-annual variation in land reflectance. Higher spatial and temporal resolution time series data are particularly useful for mapping land cover types in areas with heterogeneous landscapes and highly fluctuating vegetation dynamics. Although, for large area land monitoring, satellite data such as PROBA-V that provides five-daily time series at 100 m spatial resolution, improves spatial detail and resilience against high cloud cover, it also creates challenges in handling increased data volume. Cloud-based processing platforms namely ESA (European Space Agency) Cloud Toolbox infrastructure can leverage large scale time series monitoring of land cover and its change.

We demonstrate current activities of Wageningen University and Research in time series based land cover mapping, change monitoring and map updating based on PROBA-V 100 m time series data. Using Proba-V based temporal metrics and cloud filtering in combination with machine learning algorithms, our approach resulted in improved land and forest cover maps for a large study area in West Africa. We further introduce an open source package for Proba-V data processing.

Aiming to address varied map user’s requirements, different machine learning algorithms are tested to map cover percentages of land cover types in a Boreal region. Our study also extends to automatic updating of land cover maps based on observed land cover changes using Proba-V full time series.

Cloud-based “big-data” driven land cover and change monitoring approaches showed clear advantages in large area monitoring. The advent of cloud-based platforms (e.g., PROBA-V mission exploitation platform), will not only revolutionize the way we deal with satellite data, but also enable the capacity to create multiple land cover maps for different end-users using various input data.

In the last decades, satellite optical imagery has proved to be one of the most effective means for supporting land-cover classification; in this framework, the availability of data has been lately growing as never before mostly due to the launch of new missions as Landsat-8 and Sentinel-2. Accordingly, methodologies capable of properly handling huge amount of information are becoming more and more important.

So far most of the techniques proposed in the literature made use of single-date acquisitions. However, such an approach might often result in poor or sub-optimal performances, for instance, due to specific acquisition conditions or, above all, the presence of clouds preventing to sense what lies underneath. Moreover, the problem becomes even more critical when investigating large areas which cannot be covered by a single scene, as in the case of national, continental or global analyses. In such circumstances products are derived from data necessarily acquired at different times for different locations, thus generally resulting not spatially consistent.

In order to overcome these limitations we propose a novel paradigm for the exploitation of optical data based on the use of multitemporal imagery which can be effectively applied from local to global scale. First, for the given study area and the time frame of interest all the available scenes acquired from the chosen sensor are taken into consideration and pre-processed if necessary (e.g., radiometric calibration, orthorectification, spatial registration). Afterwards, cloud masking and, optionally, atmospheric correction are performed. Next, a series of features suitable for addressing the specific investigated application are derived for all scenes as, for instance, spectral indexes [e.g., the normalized different vegetation index (NDVI), the atmospherically resistant vegetation index (ARVI), the normalized difference water index (NDWI), etc.] or texture features (e.g., occurrence textures, co-occurrence texture, local coefficient of variation, etc.). The core idea is then to compute per each pixel key temporal statistics for all the extracted features, like temporal maximum, minimum, mean, variance, median, etc. Indeed, this allows compressing all the information contained in the different multi-temporal acquisitions, but at the same time to easily and effectively characterize the underlying dynamics.

In our experiments, we focused the attention on Landsat data. Specifically, we generated the so-called TimeScan-Landsat 2015 global product derived from almost 420,000 Landsat-7/8 scenes collected at 30m spatial resolution between 2013 and 2015 (for a total of ~500 terabytes of input data and more than 1.5 petabyte of intermediate products). So far, the dataset is being employed for supporting the detection of urban areas globally and estimating the corresponding built-up density. Additionally, it has also been tested for deriving a land-cover classification map of Germany. In the latter case, an ensemble of Support Vector Machines (SVM) classifiers trained by using labelled samples derived from the CORINE land-cover inventory was used (according to a novel strategy which properly takes into account its lower spatial resolution). Preliminary results are very promising and assess the great potential of the proposed approach which is planned to be applied at larger continental scale in the next months.

Mapping and monitoring of global land cover and land use is a challenge, as each theme requires different inputs for accurate characterization.
This talk presents results on global tree cover, bare ground, surface water and crop type, with the goal of realizing a generic approach. The ultimate goal is to map land themes and their change over time, using the map directly to estimate areas. However, much work needs to be done in demonstrating that maps, particularly of land change, may be used in area estimation. Good practices require the use of probability-based samples in providing unbiased area estimates of land cover extent and change. All of the aforementioned themes will have sample-based reference data presented in explaining the challenges of generic mapping and monitoring at the global scale.

For more than 20 years, “Earth Observation” (EO) satellites developed or operated by ESA have provided a wealth of data. In the coming years, the Sentinel missions, along with the Copernicus Contributing Missions as well as Earth Explorers and other, Third Party missions will provide routine monitoring of our environment at the global scale, thereby delivering an unprecedented amount of data.

As for global land monitoring and mapping the fleet of heritage and operational missions allow to analyse, extract, condense and derive relevant information on the status and change of our land cover, heading towards the development of a sustainable operational system for land cover classifications to meet the various user’s needs for land monitoring purposes.

ESA as either owner or operator has been and is handling a variety of heritage and operational missions such as Sentinel-2, Sentinel-3 on behalf of the European Commission, the latter based on 10 years MERIS heritage. Furthermore Proba-V is delivering since more than 3 years data to a growing land user base, again following on 15 years SPOT VGT heritage as operated by CNES and with data dissemination via Belgium including VITO as the current archive manager. Through missions such as Landsat, Earthnet builds on a heritage on land monitoring since >35 years.

While the availability of the growing volume of environmental data from space represents a unique opportunity for science and applications, it also poses a major challenge to achieve its full potential in terms of data exploitation.

In this context ESA has started in 2014 the EO Exploitation Platforms (EPs) initiative, a set of R&D activities that in the first phase (up to 2017) aims to create an ecosystem of interconnected Thematic Exploitation Platforms (TEPs) on European footing, addressing a variety of thematic areas.

The PROBA-V Mission Exploitation Platform (MEP), complements the PROBA-V user segment by offering an operational Exploitation Platform on the PROBA-V data, correlative data and derived products. The MEP PROBA-V addresses a broad vegetation user community with the final aim to ease and increase the use of PROBA-V data by any user. The data offering consists of the complete archive from SPOT-VEGETATION, PROBA-V , as well as selected high-resolution data/products in the land domain.

Together with the European Comission, ESA is furthermore preparing the way for a new era in Earth Observation, with a concept to bring users to the data, under the ‘EO Innovation Europe’ responding to paradigm shift in the exploitation of Earth Observation data.

The Copernicus Data Information and Access Service (DIAS) will focus on appropriate tools, concepts and processes that allow combining the Copernicus data and information with other, non-Earth Observation data sources to derive novel applications and services. It is foreseen that the data distribution and access initiatives will support, enable and complement the overall user and market uptake strategy for Copernicus.

The presentation will debrief on the current status and planning with regard to exploitation platforms in operations and planned with ESA involvement, as relevant for Land monitoring and land cover classification.

Sentinel-2 data are being distributed for more than a year now. However, they are still not as widely used as they should be based on their usefulness. The reason probably lies in technical complexity of using S-2 data, especially if one wants to use full potential of multi-temporal and multi-spectral imaging. Vast volume of data to download, store and process is technically too challenging.
We will present a Copernicus Award [1] winning service for archiving, processing and distribution of Sentinel data, Sentinel Hub [2]. It makes it easy for anyone to tap into global Sentinel archive and exploit its rich multi-sensor data to observe changes in the land. We will demonstrate, how one is able not just observing imagery all over the world but also creating its own statistical analysis in a matter of seconds, performing comparison of different sensors through various time segments. The result can be immediately observed in any GIS tool or exported as a raster file for post-processing. All of these actions can be performed on a full, worldwide, S-2 archive (multi-temporal and multi-spectral). To demonstrate the technology, we created a simple web application, called "Sentinel Playground" [3], which makes it possible to query Sentinel-2 data anywhere in the world.
Sentinel-2 data are only as useful as the applications built on top of it. We would like people to not bother too much with basic processing and storing of data but rather to focus on value added services. This is why we are looking for partners, who would bring their remote sensing expertise and create interesting new services.
[1] http://www.copernicus-masters.com/index.php?anzeige=press-2016-03.html
[2] http://www.sentinel-hub.com
[3] http://apps.sentinel-hub.com/sentinel-playground/

Accurate, detailed and continuously updated information on land cover is fundamental to fulfil the information requirements of new environmental legislation, directives and reporting obligations, to address sustainable development and land resource management and for supporting climate change impact and mitigation studies. Each of these applications has different data requirements in terms of spatial detail, thematic content, topicality, accuracy, and frequency of updates. To date, such demands were largely covered through bespoken services based on a variety of relevant EO satellite sensors and customized, semi-automated processing steps.

To address public and industry multi-user requirements, we present a system for retrieval of high resolution global land cover monitoring information, designed along Space 4.0 standards. The highly innovative framework provides flexible options to automatically retrieve land cover based on multi-temporal data streams from the Sentinel 1, Sentinel 2 as well as third party missions. Users can specify desired land cover data for any place on the globe for any given time period since the operational start of Sentinel-2, and receives a quality controlled output within hours or days (depending on product level).

The core of the operational mapping system is a modular chain consisting of sequential components for operational data access, pre-processing, time-series image analysis and classification, pre-acquired in-situ data supported calibration and validation, and service components related to product ordering and delivery. Based on the user’s selection for a target area, date/period, and the type of the requested product, the system modules are automatically configured into a processing chain tailored to sector-specific information needs.

The data access component retrieves all necessary data by connecting the processing system to Sentinel data archives (e.g. the Austrian EODC) as well as other online image and in-situ databases. After pre-processing, all satellite data streams are converged into data cubes hosting the time-series data in a scalable, tile-based system. Targeted land cover information is extracted in a class specific manner in the thematic image analysis, classification and monitoring module, representing the core of the processing engine. Key to the retrieval of thematic land cover data is an automated, iterative training sample extraction approach based on data from existing regional and global land cover products and in-situ data bases. The system is self-learning and -improving and thereby continuously building a global database of spatially and temporally consistent training samples for calibration and validation. Finally, the class specific land cover maps are assembled into a coherent land cover database according to the user’s specifications.

The developed system is currently tested in various R&D as well as operational customer projects. The aim is to solidify the performance of the various modules with a multi-staged opening of the system portal, starting with selected industry customers along B2B service models.

We will demonstrate the service capacity for a number of use cases, which are already applied for the current production of the High Resolution Layers within the Copernicus Land Monitoring Service, GlobalWetland-Africa, the Land Information System Austria (CadastrENV) and related mapping services for the international development sectors.

One of the main interests in exploiting Earth Observation (EO) data and collateral information in the Space and Security domain is related to the detection of changes on the Earth Surface; to this aim, having an accurate and precise information on Land Cover is essential. Thus it is crucial to improve the capability to access and analyse the growing amount of data produced with high velocity by a variety of EO and other sources as the current scenario presents an invaluable occasion to have a constant monitoring of Land Cover changes.

The increasing amount of heterogeneous data imposes different approaches and solutions to exploit such huge and complex datasets; the new paradigms are changing the traditional approach where the data are downloaded to users’ machines, and the key role of technologies such as Big Data and Cloud Computing are emerging as important enablers for productivity and better services, where the “processes are brought to the data”.

The European Union Satellite Centre (SatCen) is currently outlining a system using Big Data and Cloud Computing solutions, built on the results of two Horizon 2020 projects: BigDataEurope (Integrating Big Data, software & communities for addressing Europe’s Societal Challenges) and EVER-EST (European Virtual Environment for Research – Earth Science Themes). Main aims are: to simplify the access to satellite data (e.g. Sentinel missions); to increase the efficiency of the processing using distributed computing (e.g. by open source toolboxes); to detect and visualise changes potentially related to Land Cover variations; to integrate the final output with collateral information.

Through a web-based Graphical User Interface the user can define an Area of Interest (AoI) and a specific time range for the analysis. The system is directly connected to relevant catalogues (e.g. the Sentinels Data Hub) and the data (e.g. Sentinel-1) can be accessed and selected for the processing. Several SNAP operators (e.g. subset, calibration and terrain correction) have been chained, so the user can automatically trigger the pre-processing chain and the successive change detection algorithm (based on an in-house tool). The output is then mapped as clustered changes on the specific AoI.

The integration of the detected changes with Land Cover information and collateral data (e.g. from social media and news) allows to characterize and validate changes in order to provide decision-makers with clear and useful information.