The Bracciano Lake is a volcanic caldera lake located in the Latium, Roma region, Italy connected to the aquifer. Its surface is around 57 km² and its depth around 160 m. The area is exposed to continuous stresses half from climate changes and half from several public and private pumping wells tapping the groundwater aquifer. Over the last thirty years, the withdrawals from the aquifer have increased. Another stress on the system is climate change leading to changes in precipitation (50% of rainfall deficit in 2017 compared with the 1971-2000 average) and temperature conditions (+1.3°C compared with the 1971-2000 average), which in turn affect aquifer recharge and the lake water budget. The Bracciano Lake water crisis started in 2015 up to 2017. In 2017, the lake experienced a decrease in level around 0.7 m in six months, from January to the end of July, that caused a strong decrease of the lake water surface storage. In November 2017, the lake level reached a historic low of −1.98 m with respect to the hydrological zero (corresponding to a 13.5% reduction in the area of the lake).
In this study, we first use remotely sensed observations, mostly from the recently launched Sentinel missions, to estimate changes in the lake area from Sentinel-1 and 2 SAR and multi-spectral images, stage derived from Sentinel-3A radar altimetry acquisitions, and hence in volume combining both source of information. Our estimates are validated through comparison against level measurements made by the Bracciano smart lake organization and using a bathymetry of the lake.
Then, the water balance of the lake is estimated using in situ rainfall data, temperature-based modeling of the evaporation to infer the contribution from or to the groundwater. A similar estimate will be performed on the Martignano Lake, located two kilometer away from the Bracciano Lake, and under an ERS/ENVIAT/SARAL altimeter track will be also performed. From these different sources of information, solving the water balance equation, we should be able to determine if the reason of this steep decrease in the water level of Bracciano Lake is the decrease of the difference between the rainfall and the evapotranspiration or the increase of groundwater seepage, and hence, if it is it climate-related, caused by intensive extraction, or both.

In this paper, we present a methodology to map water bodies in tropical areas under dense vegetation at high spatial and temporal resolution using multi-source data. The water surface product (SWAF-HR) at high spatial resolution (30 arcsec) and high temporal resolution (3 days) over the Amazon basin for recent years (2010-2016) is presented. Three main datasets are used : (1) water surface fraction at coarse spatial resolution from L-band microwave sensor (Soil Moisture and Ocean Salinity - SMOS) (2) Global Surface Water Occurrence from Landsat sensor (GSWO) and (3) the new Digital Elevation Model (DEM) Multi-Error-Removed-Improved-Terrain (MERIT) based on the Shuttle Radar Topography Mission (SRTM). The product is compared to the GSWO dataset, the inundated area extent estimated by the SWAF-HR product is higher. This is due to the high capability of passive L-band microwave to sense open water compared to optical sensors due to the all-weather capabilities and sensing underneath the vegetation. We flooded area of the Branco floodplain varies from 2.7 104 and 0.2 104 km2 whereas the extent of the Mamore floodplain inundation ranges between 0.8 104 and 8.1 104 km2. The flooded area over the Branco floodplain gradually decreased from 2010 to 2015 but in 2016, the flooded area has increased during the rainy season. During 2010-2016, the minimum of the inundated surface extent has been reached during the winter 2015-2016 leading to a drought event related to ENSO. The SWAF-HR data is also compared to ESA CCI and GBP land cover, in situ discharge at outlet and main tributaries, surface height derived from altimetry and precipitation data. The results of the comparison showed a correct spatial and dynamic behavior of these new products.

Measuring and monitoring river reach area is important for river discharge estimation, hydrological model calibration, water management and hazard monitoring. However, our knowledge of the spatio-temporal variations of hydraulic river parameters is surprisingly poor. In situ gauge stations are limited in spatial and temporal coverage, and the number of available stations has been decreasing during the past decades. On the other hand, remote sensing techniques have proven their ability to measure different geometrical aspects of rivers. Satellite imagery, for instance, can provide river reach area variations with a fine spatial and temporal resolution.

Rivers show diverse behaviour along their path due to the contribution of different parameters like the gradient of elevation, river slope, tributaries and morphology of the river bed. In the sub-tropical regions, some rivers completely disappear in the dry session and overflow during the wet season. The pixel value of the water surface in a satellite image depends on the quality of water, the roughness of its surface, its chemical properties, load and type of sediments and the depth of the water column. Also, in river borders, we usually deal with shallow water covered by vegetation and complicated combinations of wet and dry areas within a pixel. Therefore, most of the pixel-based image segmentation techniques fail to determine accurate river masks. Like other natural phenomena, water bodies have strong spatial and temporal correlation in satellite images. Therefore including contextual and temporal information as additional constraints and using region-based image segmentation algorithms improves the accuracy of the derived water masks significantly.

In this study, we develop an energy function based on all available information in the images (pixel intensities, spatial and temporal constraints). Then, to minimize the energy function, we introduce a graph with two terminals. Accordingly this optimization problem will be solved by using the graph cuts technique. The maximum-flow solution in the graph is equivalent to the best labelling structure for the image. Apart from binary river masks, we generate probabilistic river masks by calculating the marginal probabilities for the final residual graphs.

Since direct validation of derived river masks is not possible, we will validate our results indirectly by comparing with measured river discharge. To this end we select two river reaches of Niger River near Lokoja and Koulikoro stations where in situ river discharge is available during the MODIS period (2000-2006). Then effective river widths are calculated by dividing the river reach area by the reach length. Finally, the river discharge estimation models are developed by using the simultaneous river width and discharge measurements. These models allow us to compare the measured and estimated river discharge.

Monitoring the ice cover on lakes is important for safety as well as economic reasons. In Norway, some tens of victims fall through the ice every year, leading to several fatalities. Furthermore, lake ice has implications for the water flow, and monitoring lake ice is an important part of flood prediction. Floods might cause enormous amounts of damage.

The Sentinel-3 sensors provide excellent opportunities for an operational lake ice monitoring service. The OLCI sensor on board the Sentinel-3A satellite alone provides daily coverage for northern latitudes at 300 m spatial resolution, allowing monitoring of a large amount of lakes.

In Norway, The Norwegian Water Resources and Energy Directorate (NVE) monitors lake ice and issues flood warnings. We have developed an automatic algorithm for monitoring lake ice cover, to be used operationally as a part of this service. The lake ice cover retrieval algorithm is based on linear spectral unmixing, where the spectrum is used to retrieve the abundance of each possible endmember. The OLCI sensor provides reflectance data in 21 different bands, and is thus well suited to this technique. Three endmembers are used: snow, ice and open water. The former two of these are combined to give the fractional ice coverage within a grid cell.

For cases of low solar elevation, the reflectance spectrum tends to be a bit different due to longer path length for light through the atmosphere. This causes the spectrum to shift slightly towards blue wavelengths, which would produce a bias in the spectral unmixing. Therefore, a simplified method for atmospheric correction (SMAC) is applied on the OLCI data to correct for this effect before retrieving the fractional ice cover.

As the OLCI sensor does not have any thermal bands, data from the SLSTR sensor (also on Sentinel-3) are used to perform cloud masking. Clouds are detected in the SLSTR data, and the cloud mask is resampled to the same grid as the ice cover product.

An initial evaluation of the lake ice products has shown them to be in agreement with observations.

We present the algorithm and processing chain used to retrieve the lake ice cover. Some example products are shown. The results from an evaluation of the products are presented and discussed.

Over hydrological areas as well as coastal areas, the complementarity between satellite data and external and independent in-situ datasets has been used for the multiple altimeter space missions for the last 25 years, developed by the European Space Agency (ESA) and the Centre National d’Etudes Spatiales (CNES). In addition, new technologies are emerging, giving access to further possibilities in terms of environmental studies.

In this context, the innovative HyDrones light altimeter is a complementary space-derived remote sensing solution for bridging data gaps and comparing data time series with other instruments. Indeed, HyDrones (https://hydrones.cls.fr) relies on the complementarity between flying UAVs and an autonomous light altimeter payload, aiming to monitor continental water bodies, estuaries and coastal zones. Taking the advantage of UAV versatility and fast deployment capabilities, the solution is based on an innovative instrument capable of performing high resolution and real-time measurements of hydrological areas. Moreover, the HyDrones solution was designed to provide a cost-effective solution that can be easily and rapidly deployed on the end-users’ areas of interest. Thus, on-demand fast data acquisition is possible, as well as offline long term monitoring. The first version of the HyDrones instrument (#MK1) is a lightweight altimeter providing centimeter accuracy water surface height estimate.

In this study, the HyDrones #MK1 solution is considered to validate recent improvements performed on SAR processing techniques for the Sentinel-3 space mission over hydrological surfaces (Fully-Focused SAR), allowing to exploit the whole phase history of the scatterers. Assessments have been carried out by comparing water surface height estimates derived from FFSAR processing technique with the HyDrones innovative LiDAR instrument embedded on a flying UAV over hydrological systems. Intercomparisons between instruments thus provide a great opportunity to assess the performance of the different techniques, and reveals that the HyDrones #MK1 light altimeter is very valuable for satellite activities and climate related experience, providing centimeter accuracy value-added products for in-situ Calibration/Validation activities of space altimeter missions over inland water bodies.

Monitoring of inundation regimes for wide areas and frequent time intervals is of great value for the fulfilment of the 2020 Aichi Biodiversity Targets and United Nations Sustainable Development Goals (SDGs) and targets, such as Aichi Targets 7, 11, 12, 14, 15 & SDG targets 6.6, 13.1, 15.1, 15.5. Multispectral satellite data are suitable for the estimation of the inundation extent under cloud-free weather conditions achieving a high degree of accuracy. Radar data on the contrary, especially Sentinel-1 ones, face limitations and challenges in reaching credible results. As such may be referred the land cover synthesis of the scene, as, for example, shallow waters or waters with emergent vegetation or flat sandy areas may produce a confusing signal with either non-inundated areas in the first two cases or inundated ones in the latter. Exploitation of high accuracy Sentinel-2 inundation derived maps to train Sentinel-1 varying pixel values in time leads to credible results up to a time interval and coverage percentage from the date that training occurred. The aim of this study is to unravel the conditions, especially the time interval from available accurate inundation maps that the sole use of Sentinel-1 data, in view of lack of available multispectral data for the desired date, may lead to credible results.
The introduced methodology generates a Random Forest (RF) classification model per 3x3 pixel window. It relies on local training sets of the Sentinel-1 polarization bands and their algebraic combinations. Reference is provided by accurate inundation maps, usually generated by Sentinel-2 data. For this process Sentinel-1 and Sentinel-2 acquisition dates shall be identical or pretty close to each other. The RF model estimated per window is then applied to the target Sentinel-1 image, when non-favorable atmospheric conditions prevent multispectral data acquisition, to derive the desired inundation map.
This study examines two approaches, each one including multiple test cases, for the formation of the local training sets. In the first approach, abbreviated as TIM, aim is to utilize the minimum possible dataset (thus, saving resources) and examine its efficacy. Four (4) pairs of Sentinel-2-based inundation maps are examined to train a Sentinel-1 based RF model per pair and finally retrieve the time break point, exceeding which this approach may result to non-credible results. Each pair is further distant in time from the Sentinel-1 target date for inundation map generation than the previous examined one. One Sentinel-2 map shall precede the target date, and the other one shall follow this date. In the second approach, abbreviated as GRP, aim is to examine the accuracy influence, if all available datasets may be utilized. Four (4) groups of even number (ranging from two to eight) Sentinel-2 based inundation maps are used as reference to train a Sentinel-1 based RF model per group and finally retrieve the time break point, exceeding which this approach may result to non-credible results. Half of the maps of the group shall timely precede the target date and the rest of them shall follow this date. The evaluation of the TIM and GRP experiments performed for 8 target dates confirm that a Sentinel-1 inundation map with a kappa coefficient 0.66 and 0.75 on average for TIM and GRP, respectively, can be generated, when mean days interval between the target Sentinel-1 image and the Sentinel-2 inundation maps is less than 30 days. GRP approach allows the estimation of inundation maps with higher accuracy compared to the maps of the TIM approach, where average kappa for GRP approach is about 0.2 higher than TIM for all time intervals examined.
This methodology is developed, supported and funded by the European Union’s Horizon 2020 research and innovation program ECOPOTENTIAL. Focus is on Doñana marshlands in Spain, a breeding and wintering habitat for many water birds, where inundation maps are required to balance bird nesting and cattle breeding ecosystem services, as expressed via the biomass production due to water presence. Generated inundation maps can be provided via the ECOPOTENTIAL Virtual Laboratory, fully integrated in GEOSS and in support of GEO ECO initiative’s objectives.