Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
4579323 | Journal of Hydrology | 2008 | 15 Pages |
SummaryThe difficulty of predicting rainfall–runoff responses in arid and semi-arid catchments using typically available data sets is well known, hence the need to carefully evaluate the suitability of alternative modelling approaches for a given problem and data set; and to identify causes of uncertainty in order to prioritise research and data. In this paper, we evaluate the distributed model, Kineros2, in application to an arid catchment in Oman, using rainfall–runoff data from 27 storm events. The analysis looks at model sensitivities, uncertainty and performance, based on uniform random sampling of the model parameter space and predictions of features of the observed hydrograph at the catchment outlet. A series of three experiments used different calibration strategies (an 11-parameter calibration, a 5-parameter calibration, and a 3-parameter calibration allowing some spatial variability of the saturated hydraulic conductivity). The parameters most significantly affecting flow peak and volume performance are those controlling infiltration rates on hillslopes. The model output was also generally sensitive to a parameter within the rainfall interpolation model. Relatively little sensitivity to initial catchment wetness was observed. Prediction performance was generally poor, for all events and for all the tested calibration and prediction strategies; and the uncertainty, estimated using model ensembles, was very high. A 2-parameter regression model used in previous work was found to perform better for predicting flow peaks. Literature review shows our results are consistent with experience of other modellers of arid and semi-arid climate hydrology. In order to realise the potential value of distributed, physically based models, for application to arid and semi-arid regions, significant data collection and further research is required, in particular regarding spatial rainfall observation and modelling.