Sensitivity analyses are valuable tools for identifying important model parameters, testing the model conceptualization, and improving the model structure. They help to apply the model efficiently and to enable a focussed planning of future research and field measurement. Two different methods were used for sensitivity analyses of the complex process-oriented model TAC D (tracer aided catchment model, distributed) that was applied to the meso-scale Brugga basin (40 km 2) and the sub-basin St Wilhelmer Talbach (15.2 km 2). Five simulations periods were investigated: two summer events, two snow melt induced events and one summer low flow period. The model was applied using 400 different parameter sets, which were generated by Monte Carlo simulations using latin hypercube sampling. The regional sensitivity analysis (RSA) allowed determining the most significant parameters for the complete simulation periods using a graphical method. The results of the regression-based sensitivity analysis were more detailed and complex. The temporal variability of the simulation sensitivity could be observed continuously and the significance of the parameters could be determined in a quantitative way. A dependency of the simulation sensitivity on initial- and boundary conditions and the temporal and spatial variability of the sensitivity to some model parameters was revealed by the regression-based sensitivity analysis. Thus, the difficulty of transferring the results to different time periods or model applications in other catchments became obvious. The analysis of the temporal course of the simulation sensitivity to parameter values in conjunction with simulated and measured additional data sets (precipitation, temperature, reservoir volumes etc.) gave further insight into the internal model behaviour and demonstrated the plausibility of the model structure and process conceptionalizations.
Read full abstract