Improving process-consistency of an ecohydrological model through inclusion of spatial patterns of satellite-derived land surface temperature

Distributed hydrological models can correctly simulate spatial patterns if the model parameters adequately represent the spatial heterogeneity of the basin. However, determining realistic values of these parameters is often difficult. Considering remote sensing-derived land-surface states and fluxes in combination with streamflow observations is a suitable strategy to better constrain parameters and improve process-consistency. For ecohydrological modelling, an accurate representation of vegetation characteristics is important to capture the vegetation response to varying moisture availability. Land surface temperature (T_s) may serve as a valuable diagnostic because it is pivotal to the surface energy and water balance, and conveys information about ecosystem stress and water use. This study aims at assessing the benefits of integrating spatial patterns of Landsat-derived T_s into calibration of a process-based ecohydrological model to improve process representation of catchment-scale energy fluxes and vegetation response to moisture deficits. We explicitly analyze the trade-off between streamflow and T_s performance, and explore the value of adding an increasing number of T_s images in the calibration process. The study is performed in a mixed land cover catchment in Germany using the ecohydrological model EcH₂O. Our results demonstrate the value of satellite-derived T_s data for reducing uncertainties of energy-balance related vegetation parameters, which are hardly constrained in streamflow-only calibration. Including satellite-derived T_s for model calibration reduced the mean absolute error of spatial anomalies in simulated T_s patterns by 15 % with negligible deterioration of streamflow performance. Inclusion of spatial T_s patterns improved capturing the differences in T_s between vegetation types, and affected simulated evapotranspiration fluxes. Improvements in simulated T_s could already be achieved by including only few (4–5) images of satellite-derived T_s in calibration. Based on our results, we advocate a wider use of satellite-based T_s data for multivariate calibration to improve model parameter identifiability and process consistency of highly parameterised, process-based distributed ecohydrological models, especially when capturing hydrology-vegetation interactions is crucial.