Influence of the uncertainties of monitoring data on model calibration and evaluation

The model-based prediction of the impact of different land management on nutrient loading requires measured nutrient flux data. Thereby the accurate calibration and evaluation of the models need an adequate data base in form of monitoring data. Uncertainties in the monitoring data influence the calibration and thus the parameter settings which affect the modelling results. Hence, we compared three different time-based sampling strategies and four different load estimation methods for model calibration and compared the results. For our study we used the river basin model SWAT (Soil and Water Assessment Tool). Study area is the intensively used loess-dominated Parthe watershed (315 km²) in Central Germany. Nitrate-N load estimation results differ considerable depending on sampling strategy, used load estimation method and period of interest. For study period the load estimation results for the daily composite data set have the lowest ranges (14% and 2% maximum deviation related to the mean value of all applied methods). In contrast estimation results for the sub-monthly and the monthly data set vary in greater ranges (between 25% and 52%). To show differences between sampling strategies we calculated the percentage deviation of mean load estimations of sub-monthly and monthly data sets related to the mean estimation value of composite data set. The maximum deviation of 82% occurs for the sub-monthly data set in 2000. This affects the model and leads to different parameter settings in model calibration and evaluation. Therefore we recommend both the implementation of optimised monitoring programs and the use of more than one load estimation method to describe the water quality situation in a better way and to establish a good calibration base for simulation models.