ABSTRACT
The prediction of daily rainfall continues to be challenging task in the field of water resources management. The present study emphasizes the comparison of statistical downscaling techniques for the prediction of daily rainfall in a mountainous catchment. Different models were developed for the prediction of rainfall on monthly and daily timescales using the same atmospheric predictors as inputs. The monthly model developed using local polynomial regression was proved to be computationally simple, easy to implement as it captures the linear and non-linear features in the data set preserving the dynamics of the atmosphere and the statistical properties of the historical data set. The daily prediction of rainfall based on generalized linear models was found to be flexible to accommodate heterogeneity of the catchment with different covariates and easy to develop. The models were applied to the Idukki reservoir catchment in Kerala, India. The performance of the models was found to be acceptable satisfying the different performance criteria. The uncertainty involved in statistical downscaling models was also addressed to by comparing the models on the basis of correlation mapping.
