Temporal instability of parameters in an event-based distributed hydrologic model applied to a small semiarid catchment

Event-based hydrologic modeling is common practice for semiarid basins where runoff is restricted to short periods after a storm. Such models are used to predict runoff production and soil erosion in agricultural areas as well as the effects of storms on sewer systems, all in areas with limited information. Sometimes, model parameterization is done through infiltration experiments to obtain a parametric infiltration curve or using standard values in lookup tables associated with land use as it is often the case for hydraulic roughness. The model may then be used to predict soil losses or runoff production in storms of different intensities. In the present study a distributed hydrologic model was calibrated to see if rainfall–runoff events of different intensities in a single semiarid basin may have different optimal calibrated sets of parameters. To achieve this, 17 sequential events were calibrated covering a wide range of conditions and storm types in the semiarid southwest of Spain. Two parameters related to roughness (Manning’s n and standard deviation of terrain micro-heights) and three related to infiltration (initial and final infiltration capacities and infiltration decay rate) were calibrated for each event. The results show that the calibrated set of parameters and their sensitivities change through time. The drift of the minima in the parameter space is partially explained by the type of storm. Hydraulic roughness and initial infiltration capacity showed the highest sensitivity to rainfall intensity, while steady state infiltration capacity showed sensitivity to information used as a proxy for the wetness state of the basin. The dynamics of the parameters and their relative sensitivities indicate that the model has to adjust itself to the different conditions of the basin so no single set of parameters characterizes the basin.