In climatology
statistical downscaling techniques have been used for predicting local rainfall from GCM (Global Circulation
Model) output. Since the characteristics of GCM output are nonlinear and do not follow any standard statistical distribution, the use of parametric technique will not be appropriate. Projection pursuit
regression (PPR) is one of nonparametric methods which can be used to model the data that have such characteristics. The result of analysis shows that PPR performs better than the common parametric method, i.e. principal component regression (PCR). In respective of the length of data, the correlations of the predicted values of the PPR model with the observed data are much higher (between 0.71 and 0.84) than those of PCR model (between 0.60 and 0.66).