Spatiotemporal quantification of inland waters resources and flooding is essential for research on hydrological and biogeochemical cycles, and greenhouse gas emissions, with satellite remote sensing offering the only means of monitoring and quantifying this distribution across vast areas and over decades. This is especially important in drylands, which cover 1/3 of the globe and have high hydroclimatic variability. Initially, surface water extent products were derived from MODIS-type sensors of coarse spatial resolution but daily temporal cadence. Recent advancements include regional to global scale products based on the entire Landsat archive. Despite the unprecedented spatial detail and the ability to map and quantify surface water dynamics and flooding from season-to-season and year-to-year, given the 16-day repeat frequency of Landsat, several ephemeral floods, very common in dynamic dryland systems, could be missed.

With the launch of NASA’s Landsat-8 (L8) and two Sentinel-2 (S2) satellites by the European Space Agency, it is now feasible to acquire optical observations at 10-60 m once every 5 days. The main goal of this research was to use these latest generation optical data streams, to map surface water dynamics in a dryland system, Australia’s Murray-Darling Basin (MDB), and quantify the importance of improved temporal cadence of HLS combined compared to each L8 and S2 individually in detecting surface water and floods.

We used NASA’s Harmonized Landsat/Sentinel-2 (HLS, https://hls.gsfc.nasa.gov/) Surface Reflectance Product, a seamless data set combining L8 and S2 observations, as input to machine learning-based mapping of surface water and flooding.
The increased cadence of HLS data resulted in more small and typically short-lived flooding events being detected compared to what we detected using L8 or S2 alone, with 1/3 more floods being captured in the more ephemeral north-west of MDB as compared to 1/5 more in the less dynamic parts of the MDB. The dense optical time series offered by the HLS data set is thus critical for capturing dynamic phenomena, such as ephemeral floods in dryland systems. Ongoing work quantifies the benefit of the improved spectral resolution of the HLS data in capturing ephemeral floods compared L8 or S2 alone.

Global surface water dynamics and its long-term changes have been documented at 30m spatial resolution using the entire multi-temporal orthorectified Landsat 5, 7 and 8 archive for the years 1984 to 2015. This validated dataset recorded the months and years when water was present, where occurrence changed and what form changes took (in terms of seasonality), documents inter-annual variability, and multi-annual trends. This information is freely available from the global surface water explorer https://global-surface-water.appspot.com. Here we extend this work by combining post 2015 Landsat 7 and 8 data with imagery from the Copernicus program's Sentinel satellites. Using these data in combination improves the spatial and temporal resolution. The improved geographic and temporal completeness of the combined Landsat / Sentinel dataset also offers new opportunities for the identification and characterization of seasonally occurring waterbodies. These improvements are also being examined in the light of reporting progress against Agenda 2030's Sustainable Development Goal 6, especially the indicator used to measure 'change in the extent of water-related ecosystems over time'.

Given current population growth rates and the increasingly visible effects of climate change on the environment and human activities, continuous monitoring of the spatio-temporal distribution of surface water bodies is becoming ever more important. This becomes particularly prominent in emergency response applications for which near-real time information about flood water extent and duration are crucial components to target often limited resources and prioritize response actions. To assure that the delivered information products have the highest possible spatial, temporal and thematic resolutions, it is critical to simultaneously harvest data from a large variety of satellite sensors.

In this contribution, we present an automated processing chain that covers all modules required for operational large-scale surface water monitoring at different spatial and temporal resolutions. This includes data ingestion and preparation, cloud / cloud shadow masking, extraction of water bodies and the preparation of thematic information products. All modules are specifically designed to be compatible with a large range of available optical satellite sensors. Cloud / shadow masking is performed with a globally trained convolutional neural network. Water bodies are mapped with a semi-supervised stochastic gradient descent classifier using training seeds which are automatically identified for each image scene by sampling a Normalized Difference Water Index (NDWI) image.

Compared to previous work in this direction, our method is purely data-driven and parameters are dynamically learned from the image to adapt to the sensor specific feature domain. Cloud / shadow masking and water mapping modules aim at being robust to radiometric, atmospheric and geometric variations across scenes and sensors. Hence, we evaluate the performance of the water mapping method against a globally distributed test dataset derived from Landsat TM, ETM+ and OLI, Sentinel-2 and RapidEye images, and compare it to a widely used water index thresholding method.

Furthermore, we present the application of our processing chain to surface water monitoring in the North-Eastern Indian state of Bihar, which is seasonally affected by flooding due to monsoon rain. In this context, we introduce a novel reference water layer that is dynamically derived for any given area and time period of interest. Combining the reference water layer with water masks allows distinguishing permanent water bodies from temporarily flooded areas. The results of our processing chain feed into a larger automatic flood monitoring system that jointly uses Synthetic Aperture Radar (SAR) and optical products to provide valuable information for emergency response and for an index-based flood insurance that aims at strengthening the resilience of vulnerable people in the region.

Lakes are common features of the Arctic tundra zone. In some Arctic areas, up to 40% of the land surface is covered by lakes. Arctic lakes are interesting study objects, because many of them can change fast. Changes occur in the form of drainage or expansion of lakes, as well as newly forming lakes, which can be observed in different areas of the Arctic. During recent years, Arctic lakes and ponds have become more and more important to study because of their role in the global carbon cycle. Many Arctic lakes are considered to be large sources of methane (CH₄). If the lake sediments are unfrozen, methane can be released during summertime and wintertime due to microbial activity. In addition to that, geologic methane can be released from the bottom of lakes in certain areas. Emissions from geological seeps can be substantially larger than that of biogenic sources, but their overall occurrence and their contribution to the total Arctic CH₄ emissions is uncertain.

Spaceborne Synthetic Aperture Radar (SAR) data has great value for the monitoring of Arctic lakes during wintertime due to the ability of SAR instruments to observe earth’s surface through persistent cloud cover and the polar night. A specific application of SAR data is the classification of shallow lake areas with ground-fast ice (ice that is connected to the lake sediments) and deeper areas with floating lake ice (ice with liquid water beneath) during winter. Areas where water remains unfrozen during winter under a layer of ice might release methane throughout the whole year. Different physical properties of the water, the ice and the frozen sediments lead to high differences in radar backscatter compared between ground-fast and floating lake ice. Typically, low backscatter from areas of ground-fast lake ice and high backscatter from areas of floating lake ice can be observed.

The Sentinel-1 mission of the European Union’s Copernicus Program has great potential for monitoring these two types of lake ice. The mission comprises two satellites carrying a C-band SAR sensor. The first satellite, Sentinel-1A was launched in April 2014. The second satellite, Sentinel-1B was launched in April 2016. With their high spatial and temporal resolutions lake ice can be mapped in unprecedented detail. A recent study focusing on mapping ground-fast and floating lake ice from Sentinel-1 data over the Yamal Peninsula in Northwestern Siberia also identified patterns of low backscatter that seem to be coming from areas of floating lake ice. These patterns vary strongly in shape and size for different lakes and their origin is unclear yet. For some lakes, they are mostly of circular shape, clearly stand out against surrounding areas exhibiting high backscatter, their positions vary during different years and their size and number tends to increase from mid to late winter.

The patterns in lake ice were also identified in optical Sentinel-2 imagery from late winter with the lake ice being partially and fully free of snow cover in different years. Possible explanations for the phenomenon are the formation of eddies under the ice layer, variations of snow properties in space and time or large ebullitions of methane.

We offer new insights into the phenomenon by analyzing data from multiple space-borne sensors over the Yamal Peninsula, including additional observations from Sentinel-1 and Sentinel-2, observations from ALOS PALSAR (L-band SAR) and Landsat 8’s Thermal Infrared Sensor (TIRS) during recent winters. For the Sentinel-2 data, also data acquired when the lake ice was fully covered by snow and data acquired during the break-up of the ice are analyzed. Additionally, an object-based image classification approach is proposed to map the locations of patterns from Sentinel-1 data and hence to be able to trace the evolution of the patterns in more detail. Moreover, a first basic evaluation of the ability to identify or exclude such lakes as an active source of methane based on data from Sentinel-5p, the first satellite from the Copernicus program dedicated to atmospheric monitoring, is anticipated. At the time of writing, the Sentinel-5p CH₄ product has not been released to the public yet, but is planned for December 2018.

Acknowledgements:
This research was supported by the Austrian Science Fund through Grant I 1401 (Joint Russian-Austrian project COLD-Yamal), by the CNES TOSCA "LakeIce", Toulouse Arctic Initiative and FBR-RGO 17-05-41043-RGO-а.

Understanding how lake ecosystems respond to anthropogenic disturbance and climate change is crucial in order to take suitable adaptation and mitigation measures for their sustainable management and protection. Superimposed on to these external pressures, lakes also naturally change with age, which means their resilience to change and sensitivity to various types of pressure will also change over time, often in different ways and at variable rates. The rate and character of change will depend upon each lake’s inherent characteristics and ecological behaviour. Importantly, lakes sit within a wider landscape, catchment and hydrological network, which means one needs to look beyond just the lake itself to determine the drivers of changing lake water status and behaviour. All of this suggests that a classification scheme that can assign levels of resilience (or sensitivity to change) to lakes is required, and that it should be able to monitor the trajectories of lake change over time. However, for most lakes around the world there is a lack of data (whether observational, sampled in situ or remotely sensed) to allow lakes to be grouped within complex classification and typology schemes, which has a profound implication on the ability to map, monitor and understand regional and global lake behaviour in response to environmental disturbance and change.

In this study, we adopted a lake geoecosystem approach originally proposed by Bajkiewicz-Grabowska (2008) at national level and applied it globally to more than 700 lakes across the globe. Using remotely-sensed, modelled and other freely available global datasets from the lake-catchment systems, we developed a universal typology to classify lakes around the world based on (i) their sensitivity to natural eutrophication due to ageing, and, by extension, (ii) their resilience to external pressures.

Knowing the rate at which each lake is expected to naturally age can provide a frame of reference in the interpretation of the effect on lake systems of human pressures and climate change, both now and in the future. In addition, by applying the proposed framework as a decision tool for designing field campaigns, targeted monitoring of more vulnerable geoecosystems can ensure that early warning signs will not be missed, and tipping points can, therefore, be avoided in order to better safeguard, manage and protect freshwater resources for future societal well-being and sustainable economic growth.

Freshwater wetlands ensure fundamental functions such as flood mitigation, groundwater recharge, water purification, and nutrient and sediment retention, as well as supporting high levels of biodiversity. Amazonian floodplains support one of Earth´s largest reservoirs of biodiversity, yet are increasingly threatened by land cover and land use changes induced by large-scale agriculture expansion, waterway network development, and hydroelectric dam construction. These pressures conjointly with climate change may have dramatic impacts on floodplain biodiversity and endemic plant and animal species. Because flooding dynamics is an important driver of floodplain biodiversity and productivity, characterizing and monitoring floodplain hydrology are important to supporting biodiversity conservation. Several global- or regional-scale wetland or flood extent maps have been produced. Until recently, these maps were of coarse spatial resolution, inadequate to support wetlands biodiversity conservation. Recently, based on Landsat imagery, Pekel et al. (2016) produced a global surface water (GSW) map at 30 m and analyzed changes in minimum/maximum flood extent and flood duration over the past three decades. However, in tropical regions, clouds and vegetation may significantly impact the accuracy of surface water mapping based on optical data. On the other hand, SAR sensors acquire data regardless of weather conditions and SAR imagery has been widely used over the past decades to monitor and map wetland inundation and vegetation worldwide, including in the Amazon region.
In this study, we use Sentinel-1 Synthetic Aperture Radar (SAR) time series (12-day repeat cycle at this latitude, 10 m resolution) to monitor the flood dynamics of a segment of the Solimões/Amazon river encompassing the Curuai floodplain (eastward) and the Janauacá floodplain (westward) for the year 2017 (covering 6 S1 tiles). The Curuai floodplain (4000 km2, including the local watershed) forms a vast complex system of temporally connected lakes, flooded forest and fringing wetlands along the Amazon river right margin. Several perennial or intermittent channels of various size link the floodplain lakes system with the Amazon River. The Janauacá floodplain is a medium size system (786 km2, including the local watershed), composed by a lake and associated flooded forest and other wetlands, linked to the Solimões River by a single channel.
Images from the S1 time series were stacked (around 30 images per tile), a mean image was calculated and a thresholding classification was applied on the basis of optical data (Sentinel 2). In this mean classification, areas always flooded will appear in black, areas never flooded in light grey and areas occasionally flooded in shades of grey on flood duration. For each date of the stack, the same thresholding classification is performed and compared to the mean classification in order to produce 4 land cover classes: open water, potentially flooded vegetation, low vegetation and forest. Classification results are then refined applying post-treatments: we used the HAND index combined with spatial and temporal rules to avoid overestimation of water in areas that are not compatible with the hydrodynamics of the area. We compare our results with the GSW products in terms of maximum water extent and inundation duration in order to assess the reliability of GSW for large Amazonian floodplains.

Maximum open water extent
Both studies are in good agreement, with an estimated open water maximal extent of 27 000 km2 (our study) and 30 000 km2 (Pekel et al., 2016). Most of the discrepancies are observed along floodplain and mainstream margins, and differences are greater at Curuai than at Janauacá. The lengths of the time series used to construct our product and GSW products are very different. Pekel et al. (2016) used a Landsat chronology over the 32 last years, while we used only the year 2017. Consequently, the maximum water level recorded at Óbidos gauge for our study was 760 cm while it was 860 cm over the last 32 years included in Pekel et al.’s study. As reported in Sippel et al. (1998), flood extent and main stream water level are directly related. According to their relationship between water level and flood extent, a 1 m water level variation induces an increase of roughly 11% of the flood extent, comparable with the expected flood extent increase for the same water level variation in the Curuai floodplain (Bonnet et al, 2008). Applying this percentage to our results leads to a maximal flood extent of circa 30 000 km2, similar to the extension found by Pekel et al. (2016).

Flood duration
Discrepancies between the GSW and Sentinel-based products are larger in the case of flood duration. These differences are not related to upstream or downstream position but to lateral flow propagation across the floodplain. Throughout the entire study area, strong heterogeneities are observed with variations between both results of several months.
At the level of the floodplains, we evidence smaller water residence duration in the main lake of the Janauacá floodplain (between 0 and 2 months). In Curuai, we observe longer water residence duration throughout the floodplain (up to 8 months). Part of the discrepancies might be explained by water level differences between the time series used to build the product (2014-2015 for GSW vs 2017 in this study). Thanks to the repetitivity of cloud-free Sentinel 1, we provide a finer quantification of the temporal dynamics of floods in the floodplains and explain the over-estimation and under-estimation of flood duration in Janauacá and Curuai respectively by the GSW product. Therefore, compared with the GSW product, the flood duration dynamics of our product correspond more closely with hydrodynamic modelling results obtained by Bonnet et al. (2017) for the Janauacá floodplain and Rudorff et al. (2014) and Bonnet et al. (2008) for the Curuai floodplain.

We conclude that GSW provides realistic maximum open water extents even at local scale and with accuracies suitable for supporting hydrologic applications, for example model calibration or validation. On the other hand, GSW should be used cautiously when looking at flood duration and subsequent hydrological connectivity analysis, which are fundamental properties to support biodiversity conservation. The Sentinel 1 and Sentinel 2 constellation should provide improved mapping of flood duration at global scale.