Combined uncertainty of hydrological model complexity and satellite-based forcing data evaluated in two data-scarce semi-arid catchments in Ethiopia

摘要

In water resources modeling, meteorological data scarcity can be compensated by various global data sets, but those data sets can differ tremendously. In the literature, hydrological models of differing complexity are proposed for estimating the water resources of semi-arid catchments, and also to evaluate rainfall data sets. The goal of this paper is to provide a joint analysis of modeling uncertainty due to different input data and increasing model complexity. Impacts of mutually concealed uncertainties on model performance and model outputs are exemplified in two data sparse semi-arid catchments in Ethiopia. We applied a semi-distributed and a fully distributed hydrological model, having different levels of complexity. Three different satellite-based rainfall data sets and two temperature products were used as model inputs. The semi-distributed model demonstrated good validation performances, while the fully distributed model was more sensitive to data uncertainties. The application of TRMM version 6 completely failed and the high-resolution CMORPH precipitation estimate outperformed TRMM version 7. In contrast, the use of high-resolution temperature data did not improve the model results. The large differences in remotely sensed input data were buffered inside the hydrological models. Therefore, data set evaluations regarding only the simulated hydrographs were less meaningful. In contrast, the investigation of parameter evolution and distributed outputs' variability appeared to be a valuable tool to uncover the interdependencies of data and model uncertainties. We suggest this procedure to be applied by default in water resources estimations that are affected by data scarcity, but especially when data sets are evaluated using hydrological models. Our case study demonstrates that estimations of groundwater recharge and actual evapotranspiration vary largely, depending on the applied data sets and models. The joint analysis reveals large interdependencies between data and model evaluations. It shows that traditional studies focusing only on one part of uncertainty, either the input uncertainty or the uncertainty arising from the choice of model structure are limited in their explanatory power of the modeling performance, particularly in poorly gauged regions. (C) 2014 Elsevier B.V. All rights reserved.