Conference Agenda

Overview and details of the sessions of this conference. Please select a date or location to show only sessions at that day or location. Please select a single session for detailed view (with abstracts and downloads if available).

 
Session Overview
Session
2.2: Large-scale Mapping of Specific LC
Time:
Wednesday, 15/Mar/2017:
10:50am - 12:30pm

Session Chair: Matthew C. Hansen, University of Maryland
Session Chair: Frédéric Achard, Joint Research Centre - European Commission
Mtg. Room: Big Hall
Bldg 14

Show help for 'Increase or decrease the abstract text size'
Presentations
10:50am - 11:10am

Global Mapping of Human settlement with Sentinel-1 and Sentinel-2 data: Recent developments in the GHSL

Christina Corbane, Martino Pesaresi, Vasileios Syrris, Thomas Kemper, Panagiotis Politis, Pierre Soille, Aneta J. Florczyk, Filip Sabo, Dario Rodriguez, Luca Maffenini, Stefano Ferri

Joint Research Centre, Italy

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.


11:10am - 11:30am

Mapping urban areas globally by jointly exploiting optical and radar imagery – the GUF+ layer

Mattia Marconcini, Soner Üreyen, Thomas Esch, Annekatrin Metz, Julian Zeidler

German Aerospace Center - DLR, Germany

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.


11:30am - 11:50am

Envisat ASAR and Sentinel-1: a decade of observations exploited to map inland water bodies

Maurizio Santoro1, Oliver Cartus1, Urs Wegmüller1, Andreas Wiesmann1, Penelope Kourkouli1, Celine Lamarche2, Sophie Bontemps2, Pierre Defourny2, Fabrizio Ramoino3, Olivier Arino3

1GAMMA Remote Sensing, Switzerland; 2Université catholique de Louvain, Belgium; 3ESA/ESRIN, Italy

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.


11:50am - 12:10pm

Global scale mapping of the when and where of inland and coastal waters over 32 years at 30m resolution.

Jean-Francois Pekel1, Andrew Cottam1, Noel Gorelick2, Alan Belward1

1European Commission - Joint Research Centre; 2Google Earth Outreach

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.


12:10pm - 12:30pm

Large Scale decametric Cropland Mapping from Sentinel-2 and Validation: Lessons Learned from 2016 nationwide Demonstration for different Countries

Pierre Defourny1, Sophie Bontemps1, Bellemans Nicolas1, Matton Nicolas1, Cara Cosmin2, Dedieu Gerard3, Hagolle Olivier3, Inglada Jordi3, Guzzonato Eric4, Savinaud Michael4, Udroiu Cosmin2, Grosu Alex2, Rabaute Thierry4, Nicola Laurentiu2, Koetz Benjamin5

1UCLouvain-Geomatics, Belgium; 2CS-Romania, Romania; 3CESBIO, France; 4CS-France; 5ESA-ESRIN

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.



 
Contact and Legal Notice · Contact Address:
Conference: WorldCover 2017
Conference Software - ConfTool Pro 2.6.113
© 2001 - 2017 by H. Weinreich, Hamburg, Germany