Valuing scarce observation of rainfall variability with flexible semi-distributed hydrological modelling -...

To represent spatial and temporal variability in rainfall adequately, rainfall-runoff models must compromise among modelling objectives, data availability, conceptualization options, and the actual variability in rainfall. This is of utmost importance for challenges of integrated water management in the rapidly changing Mediterranean context. We evaluated the sensitivity of the SWAT model to combinations of spatial rainfall variability and catchment subdivision in a data-scarce mesoscale mountainous Mediterranean context. The case study focused on the Joumine catchment, in northern Tunisia, which is emblematic of agro-hydro-chemical changes and challenges. The double-mass curve method was used to verify the consistency of rainfall time series from 1991 to 2003, indicating proportionality between annual rainfall at the reference gauge and that of the nearest gauge. The rainfall lapse rate at the Joumine catchment was 69.9 mm per 100 m of altitude. Seven sets of rain gauges and five subdivision configurations of the catchment were simulated. Differences between measured and predicted streamflow at the outlet were assessed using three indices of model fit. Predicted streamflow was extremely sensitive to spatial rainfall variability but relatively insensitive to catchment subdivision. Daily predictions were most accurate for the wettest year (2002-2003) and least accurate for the driest year (1993-1994).