Water surface elevation (WSE) is an essential quantity for water resource monitoring and hydrodynamic modelling. We investigate the use of satellite altimetry from recent satellites to provide such for medium to large lakes in remote regions like on the Himalayan Plateau where we specifically focus on the medium sized lake Lake Namco which has an extend of 100 80 km east-west by 30 40 km north-south and it situated at 4 km altitude around 90E and 31N.
For a lake like Namco satellite altimetry from exact repeat can readily be used to study the temporal variations in the WSE by referencing these to a mean track. Geodetic mission satellites like Cryosat-2, SARAL/AltiKa (since July 2016) and Jason-2 (since July 2017) offers valuable lake height information, however these can not be referenced to a mean track. One of the problem with data from a satellite like Cryosat-2 has a near annual repeat period which is unsuitable for WSE estimation, however Cryosat-2 crosses at least once per month but at different locations. The same can be said about SARAL and Jason-2.
The global geoid EGM2008 is normally used to reference the lake height. However, local geoid errors not accounted for by the EGM2008 will consequently contaminate the WSE when tracks from different locations in the lake are used to compute the WSE along with possible mean lake topography.
With multiple years of satellite altimetry from multiple satellite we devise a way to correct for this ”geoid” residual geoid error and demonstrate that the local WSE are improved when such is applied. We assume that the lake level follows the local gravity field at the altitude of the lake and that the shorter spatial wavelength is related to possible errors in the the geoid model (at zero altitude).
We suggest initially deriving a high resolution spatial Mean lake level (MLL) model using geodetic mission altimetry from the CryoSat-2 and SARAL/AltiKa and to remove this correction from the observations prior to estimating the WSE.
A validation of use of this high resolution MLL model on WSE is performed with data from in-situ observations, a laser altimetry satellite (ICESat), a Ka-band radar altimetry satellite (SARAL) and a SAR radar altimetry satellite (Sentinel-3).

The monitoring of water resources at the global scale is already a major challenge whose importance will increase in the next years. Continental waters are essential components of the water, energy and carbon cycles. The rivers and lakes water levels are identified as essential climate variables by the GCOS. Traditionally, this information was provided by gage data. However, the availability of these in situ measurements is very heterogeneous and is declining (GDRC).
For the last twenty years, numerous research teams have demonstrated that measurements from space altimeters could be used to mitigate this lack of in situ measurements. However, it is well known that the altimeters, being primarily designed and optimized to observe the oceans, can have some difficulties to track the surface of the inland water bodies especially in cases where the surrounding topography is rugged. In the past, this issue has limited the number of available measurements with respect to what could have been potentially acquired.
However, lots of progresses have been made on this subject in the last few years with the now operational use of a new tracking mode on recent radar altimeters (on the Jason3 and Sentinel3 satellites). These altimeters can be operated in two tracking modes: the classic Close Loop (CL) mode and the Open Loop (OL) mode. In OL mode the position of the tracking window is no more defined autonomously by the altimeter but is imposed by internal tables that are built from a database of (mainly) hydrology targets. This database provides position and approximate width and elevation of the water bodies, the so called potential virtual stations (PVS). The OL tracking mode improves the number of available measurements for hydrology and their accuracy.
Recently, the methods used for the definition of PVS for inland waterbodies have been refined and extended at LEGOS, so that a large number of new PVS have now been defined: 33 000 for Sentinel-3A (S3A) and 32 500 for Sentinel-3B (S3B) on its operational orbit. This database was delivered to CNES for conversion in tables used by the onboard software of the altimeter. They will be activated by the end of November 2018. In this presentation we will highlight the main characteristics of the database for S3A and S3B, present a summary of the results of the validation over inland waterbodies on a few orbit cycles and demonstrate the benefits that the new mode will bring for hydrology users using illustrative examples (for example some rivers of middle range width were observed for the first time when the OL tracking mode was activated). And finally we conclude with the current plans for extending and densifying the current database for hydrology in the next few years in preparation for the future missions Sentinel6 and SWOT.

The water resources management at the global scale is already a major challenge whose importance will increase in the next years. Continental waters are essential components of the water, energy and carbon cycles. The rivers and lakes water levels are identified as essential climate variables by the GCOS. Traditionally, this information was provided by gage data. However, the availability of in situ measurements is very heterogeneous and is declining (GDRC).
For the last twenty years, numerous research teams have demonstrated that measurements from space altimeters could be used to mitigate this lack of in situ measurements on a global scale even if these instruments where primarily designed for measurements over the oceans. However, careful post-processing and data editing is necessary to obtain accurate water levels over continental surfaces. This consumes a lot of scientists and engineers time because the number of geophysical situations seems to be almost infinite and there seems to be a huge number of particular cases to be considered. Therefore, automated algorithms are needed to globally extract valuable information from altimetry data over inland water, especially in a context where the number of available measurements is expected to increase a lot in the next years (operational use of on-board DEM mode for the satellites Jason-3 and Sentinel-3).
The expertise used for post-processing and editing altimetry data for inland waters appears to be mainly visual. In view of the recent deep learning achievements, this work investigates how to use deep neural networks to increase the number of available inland water levels derived from satellite altimetry data for the hydrology community. Especially, convolutional neural networks (CNN) are designed to extract features from images. The ResNet architecture (He et al., 2015) has been adapted and implemented in this study, as well as various deep learning techniques such as dropout or adaptive learning rate methods.
Since a sequence of consecutive altimeter waveforms (a radargram) can be considered as an image, our work consists in implementing CNN to automate water surface heights calculation from radargrams. The work has been done on Jason-3 Low Resolution Mode (LRM) data and Sentinel-3 A/B Synthetic Aperture Radar (SAR) mode data. Later, the developed algorithms could be used to process past and future satellite radar altimetry missions.

First, neural networks have been trained on Jason-3 simulated waveforms with supervised learning algorithms. These simulations were generated using the LEGOS nadir altimetry waveforms simulator. The performances of the neural networks were characterized by comparing their percentage of correct predictions of water heights within a +/- 20 cm margin. The best network achieved 99% success rate on the validation set of examples generated by simulation. Then, using transfer learning, this network performed 72% of right predictions on real Jason-3 measurements.
Improving the results on real measurements implies the neural networks to be trained on a better database with more realistic radargrams from simulations. Because the Jason-3 altimeter is optimized for sea surface height measurements, approximatively 85% of the radargrams over rivers contain at least one saturated waveform, when none of the simulated waveforms was saturated. When ignoring the saturated radargrams, the neural network obtains a success rate of almost 91% (20 cm). Hence, ongoing work focuses on training a neural network on an improved database with more realistic simulated radargrams including saturated measures. Although it takes a few days to train a neural network, a trained network needs only 2 milliseconds to compute the water surface height from a radargram.
Also, we are now working on applying these algorithms to Sentinel-3 SAR waveforms given that more than 65000 potential virtual stations are to be monitored by the constellation by the end of November 2018.

The SHAPE 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.

This Research and Development intends to make the best use of SAR (delay-Doppler) altimetry data for applications in hydrology. The study focuses on three main variables of interest: river water level (RWL), river discharge (RWD) and lake water level (LWL), RWD and LWL being part of the Terrestrial Essential Climate Variables (TECV) defined by GCOS.

The project embraces data processing from L1A altimetry products up to L2 (geophysical products), L3 (water level time series) and L4 (River discharge). It started with CryoSat-2 data (before the launch of Sentinel-3A) and is integrating Sentinel-3A as another input for the SHAPE processor.

The project has developed its own modular and configurable altimetric processor comprising a delay-Doppler processor (from L1A to L1b), a L2 processor including state-of-the-art geophysical corrections and new SARM retrackers. On top of this, the SHAPE project also implements its own L3 processor (from L2 to RWL and LWL) and L4 processors (from RWL to RWD) and a validation and verification framework. With the confidence brought by the validation and verification steps, the project implements hydrological dynamic and semi-distributed models of river catchments able to assimilate RWL measurements in order to estimate RWD.

The high level of configuration of the processor allows to work in parallel on two different baselines. The first one is dedicated to mimic as much as possible the real Sentinel-3 baseline and the second a baseline optimised for hydrology, at all processing levels.

The project focuses on 3 rivers (Amazon, Danube and Brahmaputra) and 2 lakes (Vänern and Titicaca). Sentinel-3A L1A data is considered to be used on the Brahmaputra river while CryoSat-2 L1A data is used on the other water bodies.

In this communication, we report both on the achievements made by the project as well as providing results, we also report about its status and planning.

GNSS-Reflectometry is a remote sensing technique mainly used for the estimation of wind speed over the ocean. The retrieval of wind speeds in the inner core of tropical cyclones represents the main focus of the NASA CYGNSS Mission [1], a constellation of 8 microsatellites, launched in December 2016, which collects GPS reflections over both oceans and land, and within a latitudinal range of approximately +-38°.
CYGNSS data have recently proven particularly useful and informative over land, and the successful use of these data for soil moisture retrieval [2] and flood detection [3], has been recently highlighted.
Here, we explore the possibility to estimate river discharge using CYGNSS data over land, exploiting the strong coherent and localized response of GPS-reflections from bodies of water over land, compared to the diffuse and incoherent scattering provided by the surroundings.
River discharge data from the US Geological Survey (USGS) have been analysed for three sites: one for the Mississippi river in Vicksburg, MI; one for the Gila River, near Dome, AZ, and one for the Colorado River at Yuma, AZ. A correlation analysis has been carried out between the daily discharge data from these three sites and the CYGNSS Signal-to-Noise Ratio (SNR). The CYGNSS SNR has been computed in different ways, namely:
a) uncalibrated SNR from the Delay/Doppler Maps of scattered power;
b) calibrated SNR, where the calibration removes the effect of the receive antenna, of the attenuation due to the path followed by the signal, and the different transmit power of GPS satellites.
In addition to that, the SNRs collocated with daily in-situ discharge measurements have been computed as both the daily SNRs of the specular point closest to the station site, and as the daily average of the CYGNSS SNRs from the specular points within a 0.1° latitude/longitude box, centered at the station site. Time series of the CYGNSS SNRs are compared to those of the discharge data, and similar patterns and characteristics can be identified. The correlation between the two is also found to be significant, although it appears stronger for the Gila River data, compared to the other two cases.
A discussion on the validity of the current calibration of CYGNSS data over land is presented, as well as on the possibility to implement alternative calibration approaches such as in [4]. Potential approaches for future estimations of river discharge from CYGNSS data are discussed.

[1] Ruf, C., R. Atlas, P. Chang, M. Clarizia, J. Garrison, S. Gleason, S. Katzberg, Z. Jelenak, J. Johnson, S. Majumdar, A. O’Brien, D. Posselt, A. Ridley, R. Rose, V. Zavorotny (2015), “New Ocean Winds Satellite Mission to Probe Hurricanes and Tropical Convection,” Bulletin of American Meteorological Society, doi:10.1175/BAMS-D-14- 00218.1.
[2] Chew, C. and E. Small, “Soil Moisture Sensing Using Spaceborne GNSS Reflections: Comparison of CYGNSS Reflectivity to SMAP Soil Moisture”, Geophysical Research Letters, Vol. 45, Issue 9, May 2018,
[3] Chew, C., J. Reager and E. Small, “CYGNSS data map flood inundation during the 2017 Atlantic hurricane season”, Scientific Reports, volume 8, Article number: 9336 (2018).
[4] Tarpanelli, A; Brocca, L; Lacava, T; Melone, F; Moramarco, T; Faruolo, M; Pergola, N; Tramutoli, V (2013). Toward the estimation of river discharge variations using MODIS data in ungauged basins. REMOTE SENSING OF ENVIRONMENT, 136, 47-55.

RIDESAT - RIver flow monitoring and Discharge Estimation by integrating multiple SATellite data, is an ESA project investigating the capacity of altimeter, optical and thermal satellite sensors to observe river discharge. The purpose of the project is to develop and validate a novel methodology for the estimation of river discharge in ungauged basins based on the combination of multi-sensors. The reason behind the project is related to the limits of each single sensor: cloudy conditions for optical sensors hamper the land observation; the low revisit time of the radar altimetry often is not compatible with the variations of the flow in the many channels. With the combination of multi-sensor measurements significant advantages over single sensors can be achieved also contributing to improve the quality of the final products. Indeed, the integrated product guarantee a continuous monitoring of the inland water and, hence, of the river discharge.
RIDESAT has three primary technical objectives that will be addressed to: i) validate altimetry water level product over rivers, especially the SAR data of Sentinel-3; 2) test and/or confirm the potential of optical and thermal sensors to obtain a product that is able to represent the hydraulic variables; 3) implement a technique for combining the multi-sensor products, also collocated in the same platform, to obtain reliable estimates of river discharge.
To achieve the scientific objectives, first the single-instrument products must be processed independently to generate a dataset of proxies of hydraulic variables strongly linked with river discharge (e.g. water level, flow velocity, width). Successively, these proxies are implemented as integrated techniques for the final estimation of the river discharge.
To test the ability of the approach to retrieve river discharge at global scale, 10 pilot sites are selected all over the world, based on the availability of in-situ measurements of hydraulic and morphological variables: water level, cross section, width, surface and bottom slope, flow velocity and discharge. The availability of large datasets of in situ measurements will be used for: 1) a first validation of single-instrument products; 2) an extensive validation of the river discharge product; 3) the implementation of the technique for integrating the products; 4) the evaluation of the uncertainty attributed to the combination process; 5) the evaluation of the limitation of the procedure.