Abstract
The accuracy of hydrological model predictions is limited by
uncertainties in model structure and parameterization, and observations
used for calibration, validation and model forcing. While calibration is
usually performed with discharge estimates, the internal model processes
might be misrepresented, and the model might be getting the “right
results for the wrong reasons”, thus compromising model reliability. An
alternative is to calibrate model parameters with remote sensing (RS)
observations of the water cycle. Previous studies highlighted the
potential of RS-based calibration to improve discharge estimates,
focusing less on other variables of the water cycle. In this study, we
analyzed in detail the contribution of five RS-based variables (water
level (h), flood extent (A), terrestrial water storage (TWS),
evapotranspiration (ET) and soil moisture (W)) to calibrate a coupled
hydrologic-hydrodynamic model for a large Amazon sub-basin with
extensive floodplains. Single-variable calibration experiments with all
variables were able to improve discharge KGE from around 6.1% to 52.9%
when compared to a priori
parameter sets. Water cycle representation was improved with
multi-variable calibration: KGE for all variables were improved in the
evaluation period. By analyzing different calibration setups, a
consistent selection of complementary variables for model calibration
resulted in a better performance than incorporating all RS variables
into the calibration. By looking at multiple RS observations of the
water cycle, inconsistencies in model structure and parameterization
were found, which would remain unknown if only discharge observations
were considered.