Capability and Limitation of a Semi-distributed Water Quality Model to Represent Nutrient Concentrations in a Mesoscale Catchment in Central Germany

Abstract:

Semi-distributed hydrological and water quality models are broadly used to assess nonpoint source pollutant inputs to receiving water bodies, investigate their sources and predict the impacts of climate and land-use changes on water quality. However, spatiotemporal testing of these models is necessary. This spatially-distributed validation of such models at internal stations is rare. In this study, the semi-distributed model HYPE (Hydrological Predictions for the Environment) was tested for nitrate-N (NO₃-N) and total phosphorus (TP) concentration simulations at spatially distributed and non-calibrated internal gauging stations. First, the HYPE model was applied at the mesoscale nested catchment Selke (463 km²) in central Germany to simulate discharge and nutrient concentrations at three gauging stations in the main stem of the river. These gauging stations were selected in purpose, representing the different geographical features and climatic variabilities of the catchment from upstream forest-dominant to downstream agricultural-dominant land use areas. The DiffeRential Evolution Adaptive Metropolis (DREAM) tool was used for automatic calibration and uncertainty analysis of the HYPE model using a multisite and multi-objective approach. Second, the model performance was assessed at eight internal stations that were not used in the calibration process.

Results showed the capability of the HYPE model to represent well discharge for calibration (1994-1998) and validation (1999-2014) periods with lowest Nash-Sutcliffe Efficiency (NSE) value of 0.75 and percentage bias (PBIAS) of less than 18% with low predictive uncertainty. The model performance declined substantially when only the outlet gauging station was considered, reflecting the importance of multisite calibration. The HYPE model, also, showed well the dynamics of NO₃-N and TP load simulations, represented by the lowest PBIAS values of -16% and -20% for NO₃-N and TP loads, respectively. Results confirmed that changing seasonal pattern of NO₃-N concentrations were controlled by combined effects of both hydrological and biogeochemical processes. TP concentration simulations were strongly impacted by the availability of accurate point source data. Results also confirmed the capability of HYPE model to simulate the spatiotemporal dynamics of NO₃-N and TP concentrations at eight internal validation stations with PBIAS values varies in the range of -9% to 14% and -25% to 34% for NO₃-N and TP concentrations, respectively. Overall results revealed that training the semi-distributed water quality model with enough observed data at key gauging stations representing the different hydro-meteorological and geographical conditions of the whole catchment can ensure better the spatiotemporal capability of the model.