Effects of the Choice of Data on Runoff Prediction with the Xinanjiang Model

The Xinanjiang model was in this study applied to the upper reach of the Huai River (10,190 km(2)), which is situated in a region with very erratic annual rainfall amounts. For the 22 years that were analysed, annual runoff coefficients varied from 13 to 53 %. Since the Xinanjiang model was developed for humid regions, it may not perform well when calibrated for dry years. Comparing the performance of the model when calibrated for two very dry years and two very wet years showed that the model indeed performs better when calibrated for wet years. However, when a relatively long dry period of 5 years was compared with another 5 - year relatively wet period, the difference was not significant. This result confirms the recommendation of the model developers to calibrate for periods of at least four years. It was furthermore found that dividing the basin into three sub - catchments did not lead to significantly worse results than when divided into 17 sub - areas, but application as a lumped model did lead to a serious degradation of its performance. Linking the Xinanjiang model to the automatic optimisation algorithm GLOBE largely solved the problems with the calibration of the 15 parameters.