Lake Monitoring from a Combination of Multi Copernicus Missions: Sentinel-1 A and B and Sentinel-3A
Monitoring water volume changes of a given lake needs precise estimating of water level and water surface variations. The water level of the lake might be erroneous due to multi-reflection from the illuminated areas inside the radar footprint. Despite improvements in the radar systems, the footprint is still quite large, especially in the cross track direction that leads to contaminated waveforms, called multi-peak waveforms. Bathymetry data are also not available everywhere to measure the absolute water volume storage.
To derive optimized ranges and consequently more precise water level, corrupted waveforms need to be analyzed. We developed a new approach to select an optimal peak in a given waveform to be retracked with the threshold retracker. We selected a peak, which provides a water level closest to the in situ gauge. In another scenario, we involved all meaningful peaks in a given waveform and considered the average of retracking corrections obtained from all sub-waveforms.
The water surface of the lake was estimated from analyzing SAR images. To distinguish water from non-water surfaces, the threshold algorithm based on the histogram has been used. The surface time series was validated against external data. Finally, relative water volume changes were estimated from the water level and surface variations according to the Heron method.
In this study, we used L2 and L1b data of Sentinel-3 A SRAL and SAR images from Sentinel-1 A and B from June 2016 to May 2018 to monitor Lake Vanern in Sweden. Our analysis in water level determination shows an improvement of 50% for our novel optimized peak selection compared to L2 data in front of in situ gauge measurements. We also found that for more than 90% of the waveforms, the first peak, called the first sub-waveform, leads to a better result. The second scenario, i.e. involving all meaningful peaks, called mean-all sub-waveforms, has almost the same performance as the optimized sub-waveform, which shows the effectiveness of this scenario for water level monitoring.
We found a correlation of 97% and 71% for the water level with respect to the water volume and surface variations, respectively. A 78% correlation was achieved between water surface-volume variations. There is also a correlation of 83% and 88% for our water surface and volume variations with respect to that of the Hydroweb database, respectively. An RMSE of 5 cm in water level variation is a significant achievement for the Sentinel-3 SAR altimeters over the inland waters.