Effect of watershed delineation and areal rainfall distribution on runoff prediction using the SWAT model

The effect of the division in number of subcatchments and the spatial distribution of areal rainfall on the prediction of streamflow was evaluated using the SWAT model and data from the Grote Nete River catchment (Flanders, Belgium). A multi-automatic calibration scheme (MACS), using the Shuffled Complex Evolution (SCE) optimization algorithm, was applied. A total of 6 delineations were examined. The performance of each model set-up was assessed with respect to the outlet measured daily total, quick and slow flow component. The highest Nash-Sutcliffe Efficiency (EF) value for daily total flow was obtained by a delineation of 21 subcatchments. The EF of daily slow flows are high (>0.7) and comparatively stable for all analyzed delineations. Although quick flows are systematically underestimated, for larger number of subcatchments a relative good agreement exists between observed and simulated extreme flows with medium to high return period, expect for the 65 subdivision. The effect of the spatial density of rainfall input was evaluated running the model with uniform and non-uniform areal distribution of rainfall. A modified definition of Nash-Sutcliffe Efficiency (NSEref) was introduced to measure the performance of the simulated runoff versus the reference flow, derived with the Thiessen-based areal rainfall as input. The analysis revealed that: (a) the NSEref decreases with the number of subcatchments in which the basin is divided, and (b) simulations using a uniform rainfall distribution equal to the rainfall recorded in a rainfall station situated centrally in the catchment underperform as input.