Thermokarst depressions, called alasses, are an important feature of the periglacial landscape of the Central Siberia. They are considered as a good indicator of cryogenic history of the terrain and of changes in modern hydro-climate conditions. Alasses play a crucial role in the life of the local people providing a habitat for villagers, areas for grazing and for hay-harvest. Alas lake ice is a main source of the potable water during the winter.

Our study is based on synergy of different remote sensing techniques, including optical, SAR and altimetric instruments, and in situ observations. We used Landsat imagery, ground trust data and cartographic sources to understand alass lakes and alass grasslands spatial distribution. Then, by retrieving the series of water masks from 30 TarraSAR-X images, we investigated differences in seasonal dynamic of thermokarst lakes in two sub-regions with different rate of permafrost degradation. With help of altimetric radar missions (ENVISAT and Jason-2) provided an information on long-term water level variability in thermokarst lakes and on relative dynamic of their area, the changes of the hydrological conditions during the last 15 years in large 57 000 km2 domain were assessed. Finally, combination of the Landsat derived alass grasland masks with in situ observations on grassland productivity during last 30 years, was used for assessment of the potential ecosystem services provided by the alasses for local agriculture.

This work was supported by the CNES TOSCA "Permasensing", SWOT "Wetlands", ERA.NET RUS Plus S&T #226 “ERALECC”, Toulouse Arctic Initiative and University of Toulouse IDEX “InHERA” projects.

Area B - Board 181

One key type of chemical constituent that must be considered in water quality assessments in inland waters (e.g. lakes, rivers, and streams) is Dissolved Organic Matter (DOM). It will be considered in this study as the main substance that can be quantified to assess water quality by understanding of how carbon fluxes move from the terrestrial ecosystem to the aquatic ecosystem. Coloured Dissolved Organic Matter (CDOM) is a readily measurable component of Dissolved Organic Matter (DOM) that can contribute to understanding the movement of carbon in terrestrial and aquatic ecosystems. In this study, Remote Sensing techniques provides a very powerful tool for understanding the roles of landscape pattern and the topographic drainage structure of watersheds. Firstly, the Sentinel-2 Multispectral Imager (MSI) was used to monitor a key water quality parameter, Coloured Dissolved Organic Matter (CDOM), in 230 northern lakes. Secondly, digital elevation models (DEMs) produced using data from the Shuttle Radar Topography Mission (STRM) were used to model landscape elevation with a horizontal resolution of approximately 90x90 meters. The DEMs are applied in this study to automatically map the drainage networks and watersheds in 230 northern lakes. The topographic characteristics of the upstream drainage areas of each of the 230 observations of CDOM are created to describe the physical characteristics and internal drainage structure through digital terrain analysis techniques (e.g. flow direction and flow accumulation). To perform the conventional approach to acquire the drainage structure for these 230 locations, an eight flow direction matrix (D8) is used which is derived from the digital elevation models (DEMs). This study provides potential insights into the roles of landscape pattern and various topographic descriptors of the watershed, and the connectivity of aquatic DOM to its terrestrial sources. Thus, Digital Elevation Models (DEMs) can be used to acquire information about the morphology of a land surface and how carbon moves from that land surface to adjoining waters.
Keywords: Inland waters, terrestrial ecosystem, water quality, Sentinel-2 Multispectral Imager (MSI), Shuttle Radar Topography Mission (STRM), northern lakes, watershed, eight flow direction matrix (D8), digital elevation model (DEM), Dissolved organic carbon (DOC), coloured dissolved organic matter (CDOM).

Area B - Board 182

We investigate the applicability of Sentinel-2 for ‚lake colour‘ applications by providing a first field-based & satellite-based inventory of optical lake types in the permafrost landscapes of the Lena River Delta (Arctic Siberia) and central Yamal (Western Arctic Siberia) using field investigations and multi-sensor satellite data. We also include TerraSAR-X derived water body extent to support the understanding of the seasonal lake dynamics, namely of the season of snow melt and spring flood in both lake systems.

Within the thematic network between our groups we seek to investigate:
• the water-leaving spectral signal of the lake water bodies: ‘the lake colour’
• how to link ‘lake colour’ to SAR-derived information
• geochemical and hydrodynamical in-situ investigations measurements (cDOM, DOC, SPM, bathymmetry)

The Lena River Delta is the biggest delta in the Arctic. The delta consists of young river floodplains with shallow floodplain lakes and of different morphological terraces with differing lake types. The youngest polygonal-ice rich river terrace is characterised by a wide range of lake types: shallow polygonal ponds and lakes up to 2 m depth that regularly downfreeze during autumn to early summer and the type of thermokarst lake deeper then 2 m depth. The highest terrace in the delta is no river terrace but lifted up tectonically and composed of Pleistocene ice-rich Yedoma deposits with an old thermokarst landscape characterised by deep thermokarst basins often filled with thermokarst lakes reaching down to a depth of around 5 meters. The Russian-German long-term monitoring site Samoylov is located on an island including the first terrace and a young floodplain with good access to the Yedoma deposits. The Alfred Wegener Institute AWI in Germany works in long-term cooperation with the St.Petersburg-State University and the Arctic and Antarctic Research Institute AARI from St. Petersburg, Russia, sampling since 2009 in the central Lena River delta a range of inland water bodies, the Lena river water and coastal waters. The sampling that is most dense and continous includes shallow and deep thermokarst lakes on the Samoylov holocene terrace and the deep thermikarst lakes on the Yedoma as well as Lena river water from early summer to late summer.
In contrast, the Central Yamal permafrost landscape in Western Siberia has widely distributed tabular ground ice. Tabular ground ice degradation results in the formation of much deeper lake basins up to 10 m depth and deeper. There is a wide range of different lake types, small water bodies on river floodplains and the large thermokarst lakes on high old morphological terraces. The landscape is strongly dissected by river network and land slides. The Earth Cryopshere Instiute (ECI) established a long-term monitoring site, Vaskiny Dachi (VD) located in central Yamal, close to the Bovanenkova gas field. The Circumpolar Active Layer Monitoring CALM site VD was established by Leibman (ECI) in 1993 and is part of the Global Terrestrial Network for Permafrost, GTN-P. ECI has ongoing cooperations with US, Finnish, Austrian and German institutions (e.g., US/RU LCLUC Yamal, AT/RU COLD-YAMAL, DE/RU DAAD/OSL POLYAR 1 and 2 POLYAR goes LENA). First water sampling at VD started in 2011, then continued with including a large number of different lake types in the river floodplaines and on the different high terraces.
In this study, we give an overview on different optical types of lakes in the thermokarst landscapes of central Yamal and the Lena River Delta. For several of the lakes in Yamal and for some lakes in the central Lena River Delta we were able to sample for Dissolved Organic Carbon, DOC, and coloured dissolved organic matter, cDOM (the absorbing fraction of the DOC pool) and SPM. The cDOM regimes influence the transparency of the different lake types. We show simulations of cDOM influence on lake colour using the ‘Water Colour Simulator’ WASI©. However, turbidity seems to play the dominant role in providing the water-leaving colour of thermokarst lake types.

Area B - Board 183

Cyanobacteria blooms adversely impact water quality with detrimental environmental impacts and health risks for consumption and recreational use of inland waters. Detection of cyanobacterial blooms has been demonstrated from orbital sensors with a band in the orange region centered near 620 nm, the absorption peak of the diagnostic pigment phycocyanin. The Ocean and Land Colour Instrument (OLCI) onboard Sentinel-3 is an example of the current sensors designed for water colour remote sensing to offer such a band. However, the spatial resolution of those sensors (~300 m) is too coarse to resolve small inland waters. Recent studies report that with the nominal spatial resolution of 300 m it is possible to survey only about ~1% of the total number of lakes with a surface area larger than 2000 m². Higher resolution sensors (≤ 30 m) originally designed for land remote sensing comparatively can resolve about 80% of the total number of lakes, but lack a band centered in the orange region - with the exception of the commercial missions WorldView-2 and 3. Recently, we demonstrated a method to extract radiometric information centered at ~620 nm by contrasting the panchromatic and red + green bands of the Operational Land Imager (OLI) onboard Landsat 8, a 30 m resolution global coverage mission with open data policy. It was further demonstrated that this contra-band can be retrieved with an average uncertainty of ~5% and that the orange line height calculated with this band is proportional to the in situ phycocyanin concentration. However, possible errors due to atmospheric correction were not addressed. Additionally, estimation of phycocyanin from OLI will require a correct magnitude of the orange contra-band and therefore a correct calibration of the Panchromatic band. In this study we employ several methods to evaluate the accuracy of the estimated orange signal. First the accuracy of the OLI retrieval is estimated using matchups with in situ data and with OLCI over large lakes. Observed bias and uncertainty above the algorithm uncertainty are evaluated in terms of atmospheric correction errors and OLI Panchromatic band calibration. Atmospheric correction errors are evaluated by using 6SV to propagate in situ spectra to the top of atmosphere with a range of atmospheric properties and aerosol models and applying atmospheric correction to calculate uncertainty. Finally, the panchromatic band calibration is evaluated against in situ spectra and at top of atmosphere against interpolated OLCI data and the hyperspectral sensor Hyperion.

Area B - Board 184

Remote sensing is one of the most dynamically developing fields of science, and due to its versatility, it can be applicable in many different fields of interest, i.e., biomedical science, forestry, water monitoring, agriculture, urban planning, hydrology, etc. Recent advances in remote sensor technology and remote techniques make the use of remote sensing to study aquatic ecosystems more common.
Monitoring of water environment and ecosystem, detecting water contaminants and understanding water quality parameters are the essential tasks in water management and protection of the whole aquatic environment. Among many techniques available nowadays, i.e. biomonitoring, in- situ measurements, remote sensing methods not only offer possibilities for the understanding of a significant number of changes in nature but also enable the monitoring of the state of water reservoirs, assessment several water quality parameters (e.g. chlorophyll and phycocyanin concentration, etc.) and level of physical, chemical or biological pollution. Detection of biological contaminants plays a significant role in preserving human health and water management. To obtain accurate and precise results of determination of the level of biological contamination and to distinguish its type it is necessary to determine precisely spectral reflectance coefficients of several water biological pollutants.
Monitoring of water bodies with remote sensing techniques, first of all, is based on the definition of the type of pollutant that should be detected or monitored and next on the defying the useful range of electromagnetic spectrum which will be used for the quality control of the investigated water body. With established useful range of the spectrum, it is possible to compare the spectral characteristics of water with the spectral response of a contaminant’s sample, i.e., blue-green radiation penetrates through the water layer, wherein the penetration depth is dependent on the chemical composition and physical properties of water, and infrared radiation is completely absorbed in surface water layer.
The accurate and precise determination of algae and cyanobacteria species will allow to determine their qualitative and quantitative status in the analyzed inland waters, based on the analysis of image data from the Sentinel-2 missions, as well as image data acquired using multispectral and hyperspectral sensors mounted on platforms aeronautical or UAV. The analysis of imagery data could be conducted using empirical indices based on band ratios, and derivative analysis, and their dependencies with their concentration.
The presented research aimed to obtain spectral reflectance characteristics in laboratory conditions using a hyperspectral camera, of algae and cyanobacteria occurring in Poland's inland waters. For the study one VNIR hyperspectral camera was used, produced by Headwall. The Headwall MicroHyperspec A-series VNIR is a push- broom sensor, that allows for continuous recording of the image line by line. This hyperspectral scanner allows for registration of images with 327 spectral channels with a 1.9 nm spectral bandwidth in the 380- 1000 nm range.
Spectral reflectance characteristics have been obtained for several species of biological contaminants, i.e. BA-1- Oocystis submarina; BA-2- Chlorella vulgaris; BA-5- Monoraphidium contortum; BA-15- Nodularia spumigena; BA-81- Nostoc sp.; BA-120- Synechococcus sp.; FA-7- Leptolyngbya sp.; MA-4- Porphyridium purpureum - Figure 1 (please find the attached image) at various concentrations. Such an approach would provide for the possibility to determine the relationship between the reflectance values at specified wavelengths (corresponding to the spectral bands of sensors from mission Sentinel- 2) and the species and the concentration of the investigated algae and cyanobacteria.
The laboratory tests carried out using hyperspectral data allowed to determine empirical formulas that would allow distinguishing selected species of cyanobacteria and algae, as well as determining their concentration on satellite images. However, the studies carried out have shown that some species cannot be distinguished, and thus the determination of their concentration is burdened with a high error- even when using hyperspectral data.
Next, the developed formulas were validated on the available image data from the Sentinel-2 missions for the one selected in Poland, for several dates, for which laboratory tests were also carried out.
The obtained results were characterized by moderate accuracy of species determination and concentration determination, however, not exceeding 80%. This was mainly due to the high mixing of the species in the tested tank. In addition, the spatial resolution of image data from Sentinel-2 affected the low spectral unmixing accuracy, and thus translated into the determination of species found in the tested reservoir.
It is anticipated that the use of commercial aerial or UAV platforms with hyperspectral cameras would improve the accuracy of the obtained results, however, due to the generally available data of the Sentinel-2, it can be concluded that the obtained results are satisfactory.

Area B - Board 185

Over the last few decades inland waters have suffered from heavy anthropogenic pressures enhancing eutrophication processes and causing harmful algal blooms. Current observation system of these waters is based on laborious, low frequency, point in situ measurements that allow to evaluate the ecological status of water bodies that is required for the implementation of the European Union Water Framework Directive (WFD, 2000/60/EC). The pigment chlorophyll-a concentration is a good indicator of phytoplankton biomass and is often used for assessment of ecological status of a water body. However, 4-6 measurements per year carried out by the Environmental Protection Agency (EPA) provide little information on chlorophyll-a variation over the whole season, including short-time variability of phytoplankton biomass, and phytoplankton blooms. A more timely and cost-efficient system needs development. The recently launched Sentinel-2 satellites with high spatial (10-20 m) and temporal (3-5 days) resolution provide continuous data for inland water monitoring.
The aim of this study was to investigate seasonal variation of chlorophyll-a concentration in 12 eutrophic and hypertrophic lakes in Lithuania that have high recreational potential and are affected by human activities.
We used in situ chlorophyll-a concentration measured by the Lithuanian EPA and satellite derived chlorophyll-a concentration for the year 2016. We used our derived empirical equation (R² = 0.85, RMSE = 7.4 μg/L) to obtain chlorophyll-a concentrations from the Sentinel-2 imagery. The empirical equation was based on reflectances in bands B4 and B5 (the central wavelengths at the 665 nm and at the 705 nm, respectively). We analysed seasonal pattern of chlorophyll-a concentration in 12 shallow eutrophic and hypertrophic lakes (average depth 1.5-4.2 m, size 0.74-7.9 km²) in Lithuania.
The use of satellite data increased the number of observations in Lithuania over a lake to 11-16 observations per season (May-October) when cloudless conditions were present. In addition, we found that in some cases, there was no masking of cloud shadows by the Sen2Cor scene classification and, therefore, results of those pixels were not reliable. We got large differences of chlorophyll-a concentration (~100 µg/L) for the same lake, however, lakes are relatively small and, thus, such large differences may result from method uncertainties.
The peak of chlorophyll-a concentration in four lakes (Jieznas, Kairių, Krokų lanka, Siesikų) occurred in July; in four other lakes (Alovės, Rimietis, Pravalas, Spenglas) occurred in August; in the remaining four lakes (Niedulis, Lielukas, Kampuolis, Širvys) it occurred in the autumn months.
For increased accuracy of chlorophyll-a concentrations, better masking of dark objects, such as, cloud shadows is needed, since the Sen2Cor scene classification fails to detect them in some cases. Using satellite data combined with in situ measurements provides more information on chlorophyll-a concentration variability in eutrophic and hypertrophic lakes. It is possible to observe chlorophyll-a concentration increase in summer months and in some cases to detect the peak time when using combined satellite-derived and in situ measurement database.

Area B - Board 186

The main objective of DCS4COP project (EU-H2020, 2018-2020) is to set-up, demonstrate and establish a novel DataCube service for the value-adding Earth Observation industry at highly competitive costs. Today, ocean color observations represent one of the most used tools in water quality monitoring relying on satellite products at high spatial and temporal scales.
In the present study, we attempt to validate key ocean color products mainly Chlorophyll-a (Chl-a), suspended particulate matter (SPM) and water Turbidity in the framework of DCS4COP. The test site is a highly complex and sensitive coastal lagoon (Étang de Berre) impacted by human-induced pollution. This lagoon is located in the vicinity of industrial zones of Marseille (south-east of France). It receives massive river inputs (1.2 billion m3 of freshwater per year) controlled by Electricité de France in charge of a hydropower plant just north of the lagon which is also connected to the Mediterranean Sea by a canal. The water quality of this lagoon is regularly monitored by monthly field measurements and analyzes since 1994. The objective here is to complement these field observations using high spatial ocean color products generated from Landsat-8-OLI, Sentinel-2-MSI and Sentinel-3-OLCI satellite data.
Preliminary results show the failure of standard ocean color products concentrations (Chl-a nd SPM concentrations, water turbidity). Therefore, regular field bio-optical measurements are used to successively assess the validity of several atmospheric correction algorithms (e.g., ACOLITE, POLYMER) and develop regional inversion relationships. A second set of field measurements is used to validate the best ocean color satellite products for delivery to users with associated uncertainties.

Area B - Board 187

In any satellite mission the validation of derived products is essential to assess and ensure high-quality geophysical data products. In the frame of HYPERMAQ and HYPERNETS projects a new network of automated hyperspectral radiometers is being developed. One of the planned test sites is an eutrophic, turbid, permanently mixed shallow water body called Chascomús lake located in the Pampa Plain in the Buenos Aires Province (Argentina). This lake has been intensively studied in the last 20 years and new field campaigns have been recently performed in which radiometric measurements using TriOS/RAMSES radiometer were made for the first time. Measured hyperspectral spectra have shown a dip at 620 nm, the characteristic absorption feature of the photopigment phycocyanin (PC), a marker pigment for Cyanobacteria which presence has been confirmed by microscopy. High spectral resolution information allows for the full potential of the reflectance signal from lakes to be explored to improve the retrieval accuracy of existing products, like chlorophyll-a concentration, but also to derive new products and validate any optical mission, like S2, S3, PROBA-V, MODIS, VIIRS, L8, Pléiades, ENMAP, PRISMA, SABIA/MAR, etc.
The MultiSpectral Instrument (MSI) on board of Sentinel-2 mission provides high spatial resolution data which can be relevant for monitoring coastal and in-land waters because of the high spatial resolution (10 to 20 m) and spectral bands which allows the estimation of total suspended matter and chlorophyll concentration, even though they were originally designed for land applications. In turn, OLCI on board Sentinel-3, a mission designed to monitor the ocean color, has more bands which might be capable of capturing changes in the reflectance at specific wavelengths of biological interest, like the 620 nm and close bands, but with a coarser spatial resolution (300 m) which might prevent its use in small water bodies. Even though the few existing measurements are not enough to perform a strict match-up analysis, a first evaluation of Sentinel-2 and Sentinel-3 potential use to identify cyanobacteria-dominated waters is here analyzed. Different processors, like ACOLITE and C2RCC were tested. Reflectance retrieved using ACOLITE showed better performance for bands between 510 and 680 nm, overestimating at higher and underestimating at lower wavelengths, but generally reproducing the shape of the spectra and the dip at 620 band. In turn, C2RCC systematically underestimated in situ values with better results in the blue bands compared to ACOLITE, but the dip at 620 nm and characteristic peak at 708 nm was not reproduced. For Sentinel-2 ACOLITE retrieved values tend to overestimate while C2RCC tended to underestimate field reflectance measurements, but both reproducing in general the spectral shape.

Area B - Board 188

In the frame of climate change, the knowledge of the continental water cycle plays an important key role. For such investigations, hydrologists mainly require information about lake volume changes and river discharge which can be achieved nowadays by using remote sensing data. On the one hand, water level time series can be already derived successfully from satellite altimetry, on the other hand, surface area information can be derived from optical images.
In this contribution, the “Automated Water Area Extraction Tool” (AWAX) for inland waters is presented. AWAX provides monthly water surface area time series for users independently of their remote sensing knowledge and experience.
The methodology of AWAX is based on preprocessed Landsat images (1984-today), of which the best possible cloud-free monthly composites are used for the processing. Additionally, Sentinel-2 images (2015-today) are used in combination in order to achieve high-resolution (10m) land-water masks. Five different indexes - Modified Normalized Difference Water Index (MNDWI), New Water Index (NWI), Automated Water Extraction Index for Non-Shadow Areas (AWEInsh), Automated Water Extraction Index for Shadow Areas (AWEIsh) and Tasseled Cap for Wetness (TCwet) - are applied to the composite and adaptive thresholds are used to classify water. Furthermore, data gaps caused by clouds, snow or missing data are filled by using statistical information of all existing combined images. This leads to reliable water area time series without gaps. AWAX provides a long-term inundation probability mask, monthly maps and a water area time series for the investigated inland waters. The extracted water area time series are validated by a comparison to in-situ data of water level time series for different inland waters.

Area B - Board 189

Generation of a New Global Water Bodies Layer from TanDEM-X Data
--------------------------------------------------------
In this paper we propose a new method for extracting water bodies from TanDEM-X bistatic SAR images, aimed at generating a new global water bodies layer.
A reliable detection of water bodies from SAR Remote Sensing images is useful for many scientific purposes, such as water extent monitoring, flood monitoring, digital elevation models editing, flow modeling, runoff estimation, coastal and shoreline monitoring, and agriculture sustainability assessment.
SAR images can provide reliable information on all-weather and all-day conditions and the global coverage of TanDEM-X represents a high valuable data set for the generation of large-scale geomaps.
The iterative algorithm that we developed is based upon the segmentation algorithm called “Watershed”, taking as input the interferometric coherence derived from a single bistatic TanDEM-X acquisition.
The “Watershed” algorithm begins from an arbitrary group of well-known seeds called “markers”, and from these seeds it iteratively segments the entire image into different connected and homogeneous group of pixels.
In order to successfully exploit such an algorithm, the choice of the “markers” is crucial; this operation has been performed through an ad-hoc training process of the input data.
The algorithm allows for the preservation of water bodies borders even if they have complex shapes, leading to the generation of highly precise water masks.
Then, once a fully segmented image has been returned, the analysis continues with different refinement steps in order to mitigate the effects of geometric distortions, such as Shadow and Layover, caused by the side-looking nature of the SAR system, and with a threshold-based final classification.
In the final paper we will present an analysis of the algorithm’s performance using full-resolution bistatic TanDEM-X data acquired over Pennsylvania, USA, and over the Berchtesgaden Alps, Germany, together with the first results from a global processing of TanDEM-X quicklook data, at 50 meters resolution, for the generation of a new global water bodies binary layer.

Area B - Board 190

Mapping water occurrence on the terrestrial surface remains to be one of the most fundamental tasks in remote sensing. It supports water resources assessment, land-cover and land-use mapping, and studies of climate change, public health, disaster adaptation and so on. Accurate water mapping depends largely on the water index method that produces high-resolution water maps with high accuracy of classification. However, previously widely-adopted water index methods are developed from 30-m resolution Landsat imagery, with low-albedo commission error (e.g., shadow misclassified as water) identified as a primary issue. Besides, high-albedo omission error from sunglint phenomenon becomes more apparent with the increased resolution. In light of Sentinel-2 that has the highest 10-m native resolution, this study aims to develop a water index originated from Sentinel-2 that improves resolution and accuracy at the same time. The support vector machine (SVM) is used to exploit the 10-m spectral bands among Sentinel-2 bands of three resolutions (10-m, 20-m, 60-m). The new Multi-Spectral Water Index (MuWI), consisting of the complete version and the revised version (MuWI-C and MuWI-R), is designed as the combination of normalized differences for threshold stability. The proposed method is assessed on coincident Sentinel-2 and sub-meter images over four locations in Venice, Northern India, Gulf of Mexico and Colombia. In comparisons with Normalized Difference Water Index (NDWI), Modified Normalized Difference Water Index (MNDWI) and Automated Water Extraction Index (AWEI), results show that both versions of MuWI enable to produce 10-m resolution water maps with higher overall classification accuracies and kappa coefficients (p-value<0.01). Commission and omission errors are also significantly reduced particularly in terms of shadow and sunglint. Consistent accuracy over complex water mapping scenarios is obtained by MuWI due to high threshold stability. MuWI-C achieves slightly higher overall accuracy than MuWI-R (96.4% vs. 95.9%), but MuWI-R has simpler form owing to fewer terms (4 vs. 15) and integer coefficients (vs. decimal coefficients). Overall, the proposed MuWI method is applicable to accurate water mapping with improved spatial resolution and accuracy, which facilitates water mapping and its related studies and applications on growing Sentinel-2 images.

Area B - Board 191

Continental surface waters only cover a few percent of the Earth surface, but they are essential for life and human environment, and play a major role in the global water cycle and climate variability. As a consequence, there is a widespread demand for accurate and long‐term quantitative observations of their distribution and dynamics over the whole globe.

An initial methodology was developed to estimate the extent and dynamics of surface waters, at global scale, from multiple satellite observations (passive and active microwave observations and visible and near infrared). These satellite data have different sensitivities to the surface properties (e. g., vegetation, topography, soil properties), and their combined use makes it possible to disentangle the contributions of the various factors in the observations. The technique is globally applicable without tuning for specific environments. A monthly estimate of surface waters was produced for 1993-2007, with a 0.25° spatial resolution: the Global Inundation Extent from Multi- Satellites (GIEMS) (Prigent et al., 2001, 2007, 2012; Papa et al., 2010). This dataset has been extensively used by the community, for methane emission estimation or for hydrological modeling for example. The data set has also been downscaled up to 90 m, for local applications (Aires et al., 2017; 2018).
This initial GIEMS methodology requires key ancillary data, and to obtain a climate quality dataset, these data must be of similar provenance and of constant quality over time. However, some data used in this initial method were not available after 2007. An alternative methodology has been developed that requires less ancillary data, to overcome the problems of discontinuity of data sources over long time record. Overestimation issues that were noticed in GIEMS in specific environments in South East Asia have also been solved.
The new method has been applied to the successive SSM/I and SSMIS instruments, from 1992 to 2016, using the carefully inter-calibrated satellite observations from the Satellite Application Facility on Climate Monitoring (CM SAF) under EUMETSAT support. The new data set (GIEMS-2) has been carefully analyzed, at the global scale and for the different environment types. It has been compared to the former GIEMS version and to other estimates of surface water extent (e.g., Pekel et al., 2016). The GIEMS-2 time series are seamless over the full record. As expected, the surface water extents show a significant annual cycle and a strong interannual variability. Comparisons with river height and precipitation are also very convincing.



References:
Aires, F., L. Miolane, C. Prigent, E. Fluet-Chouinard, B. Lehner, and F. Papa, A global, long-term and high spatial resolution inundation extent database, J. Hydrometeorology, http://dx.doi.org/10.1175/JHM-D-16-0155.1, 2017.
Aires, F., C. Prigent, F. Papa, E. Fluet-Chuinard, B. Lehner, and D. Yamazaki, Comparison of visible (G3WBM and GSWO) and multi-satellite (GIEMS-D3) global inundation datasets at high-spatial resolution, RSE, 216, pp. 427-441, DOI: 10.1016/j.rse.2018.06.015, 2018.
Papa, F., C. Prigent, C. Jimenez, F. Aires, W.B. Rossow, and E. Matthews, Interannual variability of surface water extent at global scale, 1993-2004, Journal of Geophysical Research, 115, D12111, doi:10.1029/2009JD012674., 2010.
Pekel, J.-F., A. Cottam, N. Gorelick, and A. S. Belward, High- resolution mapping of global surface water and its long-term changes, Nature, 540(7):418–422, 2016.
Prigent, C., F. Papa, F. Aires, C. Jimenez, W.B. Rossow, E. Matthews, Changes in land surface water dynamics since the 1990s and relation to population pressure, Geophys. Res. Lett., 39, L08403, 5 pp., DOI: 10.1029/2012GL051276, 2012.
Prigent, C., F. Papa, F. Aires, and W.B. Rossow Global inundation dynamics inferred from multiple satellite observations, Journal of Geophysical Research, 112, D12107, doi: 10.1029/2006JD007847, 2007.
Prigent, C., E. Matthews, F. Aires, and W. B. Rossow, Remote sensing of global wetland dynamics with multiple satellite data sets, Geophysical Research Letter, Vol. 28 , No. 24 , pp. 4,631-4,634, 2001.

Area B - Board 192

ABSTRACT
The objective of this paper is to present the experience acquired by CS on the development of algorithms and tools related to the SWOT radar altimetry mission in support of the CNES (the French Space Agency), dedicated to the study of inland waters. This work is in line with the definition of future missions (Sentinel 3Top) or complementarity with other missions (Sentinel 1).

Index Terms— Radar altimetry, inland waters, water detection, hydrological products, SWOT, Sentinel 1, Sentinel 3 Top.

1. INTRODUCTION
CNES, NASA, CSA and UKSA have invested in the SWOT (Surface Water & Ocean Topography) wide-swath altimetry mission, whose launch is planned for 2021. Its interferometric synthetic aperture radar (InSAR) has the potential to improve ocean observation and access global measurement of water levels in rivers, lakes and flooded areas. This partnership is based on 25 years of cooperation between CNES and NASA in the field of ocean altimetry with the Topex-Poseidon and Jason satellites.
Compared to conventional altimetry, which provides point data along profiles spaced several tenths or hundreds of km apart, wide-swath InSAR altimetry provides a two-dimensional image with a horizontal resolution of about 500 m over ocean and around 20 m average posting over continental surfaces. The SWOT mission represents a major conceptual breakthrough in space altimetry, and is a precursor to the definition of future operational swath altimetry missions for measuring the topography of oceans and inland waters.

2. ALGORITHMS AND HYDROLOGICAL PRODUCTS

3.1. Water detection
Water detection in SWOT images is based on the fact that water will, most of the time, have a much higher radar reflectivity than land. The current baseline method for water detection in the operational SWOT processing chains is an improved MRF classification method [Lobry, 2017]. A dedicated method for narrow rivers has also been proposed [Lobry, 2017]. Fuzzy and majority vote fusion schemes have been implemented in order to combine both of them and possibly other detection or classification methods.
Water and land radar reflectivities, needed to perform water detection, can be predicted from knowledge of the radar system and hypotheses about water and land normalized backscattering coefficients. The possibility to estimate water and land properties using a prior water mask such as that [Pekel et al.] has also been implemented. This and other water masks are also used to define zones to always include in the so-called pixel cloud (beyond what is detected as water), and possibly also to define the areas to always exclude (areas where the presence of water is impossible).
Another contribution is the conception and implementation of tools to assess water detection performance, both based on classic confusion matrix metrics, and with respect to the specific criteria of the SWOT science requirements at the water body level.

3.2. Hydrological products and processing
The pixel cloud will be the rawest level-2 SWOT product. It is a point cloud consisting of geolocated pixels with, for each pixel, information of height, correction… The three hydrological products are computed from the pixel cloud, i.e. shapefile river product (each object consists of a river section about 10km long), lake products (products per pass and cycle averages are foreseen), and NetCDF raster file.

Lake products
The LOCNES processing chain produces lakes products from the pixel cloud. One of the key points in the processing is the computation of the lake boundaries. Several methods have been implemented and tested:
• simple concave hull,
• Delaunay triangulation,
• vectorization in radar geometry.
Each observed water body is linked, if possible, to a reference lake, stored in an a priori lake database. Assessment of candidate databases has been conducted.

Raster products
The raster product displays the pixel cloud in grid format. It will be generated systematically at fixed resolution, but possibly also in a more flexible way upon user request via the data distribution system.

River products
The SWOT Science Team works actively on the definition of algorithms to compute globally discharge at the reach level. Our involvement here is to provide the joint ICUBE - IMFT - IMT - IMT - LEGOS laboratories team (under the direction of P.A. Garambois and J.Monnier) with CS' know-how to enrich and validate flow algorithms, in particular by:
• prospecting and studying auxiliary data to better constrain algorithms (topography/bathymetry, ground data, remote sensing data),
• the implementation of operational test cases using the SWOT data simulator.
The river product will contain flow information calculated from the heights measured by SWOT. Five flow calculation algorithms are being developed and compared in a dedicated SWOT working group ("PEPSI challenge").
The ICUBE-IMFT-IMT research team has several tools that are part of the flow algorithm approach. These can be organized into three categories:
• FirstGuess": tools for generating drafts (or "a-priori") on the required variables (flow, bathymetry and friction),
• LowQ": low complexity calculation algorithms based on original direct and inverse relationships ,
• "HydrauVDA": algorithms with fine representation of hydraulics and Variational Data Assimilation (VDA).
The team has been working for several months to integrate the lot within a single algorithm, called "HiVDI" (Hierarchical Variational Discharge Inference).

3. MULTI-TEMPORAL AND MULTI-SENSOR METHODS
Algorithmic development work will be extended in 2019 to a multi-temporal and multi-sensor approach as part of a PhD conducted in collaboration between CNES, Telecom ParisTECH and CS SI under the direction of Prof. Florence Tupin. These algorithms are not intended for operational SWOT processing, but can be seen as risk-mitigation actions or preparation of future missions.
The SWOT mission will be able to deliver long time series, and in parallel other sources of information are available ( Sentinel-1). SWOT multi-temporal data can be used to improve measurements in the event of stable signals or, on the contrary, to highlight the evolution of the hydrographic network over time, both in terms of shape on the amplitude image and in terms of height from the interferometric phase.
The second part of the thesis will focus on the use of Sentinel-1 data to help extract water networks from SWOT HR images. Rapid processing of this data will be developed to exploit multi-temporal information in the context of water surface detection, using CS experience in Sentinel toolboxes (SNAP).

4. CONCLUSION
C-S provides technical support to the French Space Agency (CNES) to develop and/or assess algorithms and prior data for the processing of hydrological products in the framework of the SWOT mission. The experience gained also enables us to take part in simulation and processing activities to prepare future missions. We are also involved in studies that consists in performing multitemporal analysis or combining data from several missions (SAR, InSAR, optical images etc.), such as water detection based on multitemporel SWOT and Sentinel-1 data.

5. ACKNOWLEDGMENT
We thank the CNES and in particular Claire Pottier and Roger Fjørtoft for the fruitful cooperation in the framework of the SWOT mission.

6. REFERENCES
• Brisset, P., Garambois, P.-A., Monnier, J., & Roux, H. (2018), On the assimilation of altimetric data in 1D Saint–Venant river flow models. Advances in Water Resources, Volume 119, pp 41-59
• Durand, M., C-J, G., Garambois, P.-A., Bjerklie, D., Smith, L.-C., Roux, H., et al. (2016). An intercomparison of remote sensing river discharge 1 estimation algorithms from measurements of river height, width, and slope. Water Resources Research, 52(6), 4527-4549.
• Durand, M., Gleason, C.-J., Bjerklie, D.-M., Garambois, P.-A., Smith, L.-C., Roux, H., et al. (n.d.). Including stage-dependent roughness coefficient in algorithms to estimate river discharge from remotely sensed water elevation, width, and slope. Abstract H21F-1487, presented at 2016 Fall Meeting, AGU, San Francisco, Calif., 12-16 Dec.
• Larnier, K., Monnier, J., & Garambois, P.-A. (n.d.). River discharge and bathymetry estimations from swot altimetry measurements. Submitted.
• Lobry S., Tupin F., Fjørtoft R., Unsupervised detection of thin water surfaces in SWOT images based on segment detection and connection. International Geoscience and Remote Sensing Symposium (IGARSS), Fort Worth, USA, July 2017.
• Lobry S., Denis L., Tupin F., Fjørtoft R., Double MRF for Water Classification in SAR images by joint detection and reflectivity estimation », submitted to IGARSS 2017.
• Lobry S., Markovian models for SAR images: Application to water detection in SWOT satellite images and multi-temporal analysis of urban areas, November 2017, PhD.
• Lobry S., Tupin F., Fjørtoft R., Non-Uniform Markov Random Fields for Classification of SAR Images. European Conference on Synthetic Aperture Radar (EUSAR), Hamburg, Germany, June 2016.
• Pekel J.F. et al., « High Resolution Mapping of Global Surface Water and its Long-Term Changes”. Nature 540 (7633), 2016.
• Tuozzolo S, Lind G., Overstreet B, Mangano J., Fonstad M., Hagemann M., Frasson R., Larnier K., Garambois P.A.,Monnier J.,Durand M. (n.d.). Estimating river discharge with swath altimetry: A proof of concept using AirSWOT observations, Geophysical Research Letters, Submitted.

Area B - Board 193

The ACA-DDP project is part of SEOM, Scientific Exploitation of Operational Missions, an ESA program element which aims at expanding the international research community, strengthening the leadership of the European EO research community and addressing new scientific researches.

In this Project, isardSAT and Brockmann Consult, together with a group of scientists in a larger consortium, have developed “DeDop” (for delay-Doppler), an open source tool to allow the processing of delay-Doppler altimetry data.

Being open source and distributed online, the tool allows the user to select input data, choose and run processing options, and immediately query and view the results. This introduces a new paradigm in the processing of altimetry data, allowing a much more direct and immediate interaction with the processor in comparison to the use of Agency’s or specialist’s products, with periodic updates implemented over longer-time scales.

In this communication, we present results from a case study on inland water (a task that is part of the ACA-DDP project) in which the DeDop processor has been implemented on Sentinel-3A data over the Amazon river. The processor has been run using a custom configuration specifically designed for inland surfaces and improving the overall quality of the river water level measurements. (Actually, rivers are one of the most challenging water surface type to be monitored by altimetry, could it be in low resolution (LRM) or in SAR (SARM) modes.)

Area B - Board 194

Accurate water level time series of rivers, lakes, reservoirs and wetlands contain important information about regional hydrological regimes, water resources and their dynamics. Even the Earth’s hydrosphere on a global scale might be monitored, supposing that a comprehensive world-wide data set of water levels is available.
Satellite altimetry (SA) may complement observational data from in-situ gauging stations, as this technique is proven to yield reliable estimates of absolute water level heights not only for open ocean but also for inland water bodies. Due to its nature as satellite remote sensing technique SA is capable of monitoring targets even in remote areas without infrastructure. Observations from SA allow not only the determination of seasonal fluctuations or short-term variations of water storage and discharge, but also long-term trends and the occurrence of extreme events. Furthermore, they serve as important input to hydrological models, accounting for realistic assessment of potential risks like flooding or droughts when combined with climate predictions.

One key factor controlling the quality of inland water level time series obtained by satellite altimetry is a careful data preprocessing. The exclusion of unusable measurements, i.e. reflections contaminated by land, is crucial here. This is especially true in case of small or narrow water bodies, e.g. small lakes and rivers.
In the past, mainly two different strategies have been utilized to address this issue. Usually, existing static land-water masks are used to discriminate between radar returns from water and land. A different approach is based on the analysis of the returned radar echo (called waveform) itself. Here, waveforms are classified as land or water returns based on one or more parameters that are analyzed in order to characterize the echo type.
However, both methods exhibit shortcomings: In the first one, temporal variations of water outlines are not taken into account. These variations can be significant, especially in case of meandering rivers and wetlands. In the second one, misclassification rates are difficult to determine.

Within this study, a methodology is developed in which time-dependent water masks are incorporated, which contrasts the approaches mentioned before. These masks are derived from remote sensing images taken as simultaneously as possible with the altimetry measurements. The impact of using temporarily changing water masks on resulting water level time series is investigated. Therefore, time series are computed for specific sample regions exhibiting different characteristics, such as varying water coverage or changing channel locations.

Area B - Board 195