Impact of Spatial Resolution and Forcing on the Simulation of Drought and Floods

Abstract:

A Variable Infiltration Capacity (VIC) model for the Thur basin (1703 km2, situated in Switzerland) was set-up and run with four different spatial resolutions (1x1 km, 5x5 km, 10x10 km, lumped) and evaluated at an hourly, daily, and monthly time step. We used the novel Hierarchical Latin Hypercube Sample (HLHS) methodology (Vorechovsky, 2014) to efficiently sample the most sensitive parameters of the VIC model. The model was run 3150 times with the HLHS and the best 10% of the runs based on the NSE(Q) and NSE(logQ) was selected as behavioral and used for further analysis. This whole exercise was conducted both with homogeneously applied forcing (equal precipitation over the whole catchment) and with distributed forcing (specific for each grid cell). We ran the model for the period August 2002 to August 2003. This period is characterised by three very high peaks (August, September 2002) and the 2003 heatwave and associated drought (June, July, August 2003). The 2002 peaks are the highest measured in the past 39 years (1974-2012) at the outlet of the Thur. The peaks were caused by a larger weather system that also caused the heavy floods in the Elbe and the Danube. In contrast, the 2003 summer was extremely warm and dry in Western and Central Europe, with Switzerland being among the hottest and driest regions. For each of the four spatial resolutions we selected the behavioral sets. A large overlap in behavioral sets implies that the model is not very sensitive to spatial resolution. The model was shown not to be very sensitive to spatial resolution if the forcing was applied homogeneously over the catchment, i.e. there was a large overlap in behavioral sets. The model became slightly more sensitive to spatial resolution if the forcing was applied in a distributed fashion. Both for the distributed and for the homogeneously applied forcing the model was shown to be highly sensitive to the temporal resolution, but the sensitivity for temporal resolution decreased with heterogeneous forcing. In this case study it was shown that the VIC model is more sensitive to temporal resolution than to spatial resolution. When the forcing is applied homogeneously over the catchment, this signal is strengthened (even more sensitive to temporal resolution, even less sensitive to spatial resolution), while distributed forcing slightly weakens this signal.