Satellite radar observations for hydrologic and hydrodynamic modelling

Abstract

As the availability and global coverage of in-situ observations of surface water dynamics is decreasing, remote sensing is playing an increasingly important role in hydrological monitoring. Satellite observations are often used to bridge the gap between data requirements and availability for water resource management tools, including hydrological-hydrodynamic models. In recent years, new radar missions have offered a myriad of new possibilities, as vast amounts of new data have become publicly available. Combining these remote observation technologies with reliable computational simulations can serve as valuable support for decision-making and management particularly in regions with weak in-situ monitoring networks. The aim of this PhD project was to explore the value of new radar missions for hydrologic monitoring and modelling. The research was centred on developing a full workflow, from rainfall-runoff model to hydrodynamic simulations of channel water level, for poorly instrumented catchments, using satellite radar observations to inform the models at different stages. The main contributions of the research were: 1) creating a robust and flexible hydrologic modelling framework with a calibration and parametrization strategy for nested, poorly gauged catchments, 2) developing basin-scale extraction of all water surface elevation observations from the operational Sentinel-3 satellites and CryoSat-2 and 3) a new hydraulic model parameter calibration approach using CryoSat-2 WSE. In data scarce regions, modelling is constrained by data availability; therefore, methods to fully exploit any available records are highly valuable. By informing the rainfall-runoff parameter regionalization with catchment characteristics from publicly available data sources (digital elevation model and land cover maps), it was possible to include valuable observations otherwise unsuitable for direct integration in the rainfall-runoff model. Furthermore, a better representation of the runoff-generation processes within the catchments was obtained by transferring information from gauged to ungauged sub-catchments. In poorly instrumented rivers, satellite observations are often a key source of information about hydrological variables. Recently, progress in instrumentation and public availability has created unique opportunities, particularly in these regions. Valuable information can be extracted from radar observations, e.g. dynamic water masks from Sentinel-1 SAR imagery, dense WSE monitoring networks at catchment scale from the Sentinel-3 SAR altimeters and spatially distributed characterization of river elevation from CryoSat-2. Sentinel-3 is the first altimetry mission to operate in SAR mode and open-loop tracking mode globally. The two-satellite configuration creates a uniquely dense WSE (Water Surface Elevation) monitoring network of 175 virtual stations in the Zambezi alone. In the Zambezi, Sentinel-3 outperformed past altimetry missions at in-situ stations with root mean square deviations as low as 3 cm and consistently lower than 32 cm. CryoSat-2 was originally designed for monitoring the cryosphere and was the first mission to operate in the high resolution modes SAR/SARIn over selected areas of interest. The geodetic orbit has interesting implications in hydrology as well, particularly to calibrate hydraulic parameters, although the high spatial resolution is challenging to integrate with traditional hydrological tools designed for time series at point locations. By using simulations of runoff to select valid observations of WSE, outlier filtering was possible even in ungauged catchments. Adequate hydraulic representation of rivers is necessary to produce useful flood simulations to informing water management decisions, and flood forecasting. In data-scarce regions, satellite WSE have very high value. Using a steady-state solver of the Saint-Venant equations and CryoSat-2 WSE observations, the channel roughness and bed elevation can be calibrated and used to parametrize a Lisflood-FP model for several reaches of the Zambezi River, while greatly reducing the resource requirements for the calibration. The model performance is comparable to, or better than, past calibration attempts, with RMSD down to 0.53 m when validated against Sentinel-3 WSE. A challenge remains in fully integrating the calibration with a 2D hydrodynamic model to benefit flood forecasting and water management efforts. The results from this study highlight the value of radar observations in hydrological monitoring and modelling applications, particularly in poorly gauged catchments where altimetry might provide the only water level information. By developing tools specifically adapted to the varying sampling patterns of altimetry missions, predictions of hydrological states can be improved. Currently operational missions and future missions all create new possibilities, which will require flexible approaches to improve the uptake of remote sensing data in the hydrologic community.