Conclusion

We calibrated and evaluated a hydrological-hydrodynamic model with five different RS-based observations of the water cycle: water levels (Jason-2), flood extent (ALOS-PALSAR), TWS (GRACE), vegetation ET (MOD16), and soil moisture (SMOS), for a study basin in a tropical region with floodplains (Purus River Basin in the Amazon), and analyzed the redundancy and complementarity between different variables and processes.
Results showed that calibration with current RS observations was able to improve discharge estimates. For instance, in the uncalibrated setup (a priori parameter sets), average performances for discharge were around KGE = 0.30. By calibrating the model with ET from MOD16 (and evaluating for the same time period), discharge average performance was improved to KGE = 0.64, representing a Skill Score of S = 52.9%. Also in the calibration period, a joint scheme of calibration with water level + soil moisture led to discharge improvements of S = 59.9%. When evaluating for a different time period, discharge performance was improved by calibration with water level, TWS and a joint scheme of all RS-variables (S = 25.9%, S = 27.9% and S = 17.4%, respectively). We conclude that RS observations are useful to predict discharge estimates. However, the utility of each RS variable might depend on the study area characteristics and the time period considered.
Our results also showed that RS-based calibration led to an overall improvement of the water cycle representation. For instance, calibration with water level was able to improve estimates of water level itself, but also flood extent, TWS and ET; calibration with soil moisture was able to improve estimates of soil moisture itself, but also discharge, flood extent and TWS.
Moreover, calibration with multiple RS variables was able to highlight deficiencies that might be related to model structure, parameterization, and observations. In the context of model structure, for instance, calibration with ET highlighted the model inability to represent the root water intake in dry season in this region, thus compensating it by misrepresenting other variables. In the context of model parameterization, for instance, we found a wide range of different parameters by varying the calibration target variable.
Besides individual calibration with each RS variable, we conducted two multi-variable calibration experiments: calibration with all variables except discharge, and calibration with water level and soil moisture. Calibration with all variables was useful to some extent, but appropriately selecting complementary variables for model calibration may result in a better overall performance. Even though we used a lumped calibration approach, results highlighted the overall model capability to retrieve ET spatial pattern, but not for TWS and soil moisture.
The main conclusions presented here are of great interest for the hydrological community, and agree with previous works in that RS–based calibration is useful to improve the water cycle representation in hydrological models. To further investigate the potentiality of RS data, future developments should test the methodology presented here for multiple basins at contrasting hydro-climatic regions. Here, we assessed an Amazonian Equatorial basin, with particular climate and land cover characteristics and an overall spatial homogeneity of rainfall-runoff processes. Other basins with different hydroclimatic regimes could be also assessed, e.g., in arid basins subject to long dry periods, more erratic precipitation patterns, and different runoff generation mechanisms than the Amazon, which require different model structures.
Finally, here we used one state-of-the-art RS product for each variable, but future developments should explore to its potential other missions as SWOT for surface water observation (Biancamaria et al., 2016), as well as considering different products for representing each variable (e.g., ET could be estimated by GLEAM, MODIS, SSEBop, SEBS, ALEXI, METRIC, etc., besides MOD16).