Improving accuracy of downscaling rainfall by combining predictions of different statistical downscale models

A flexible framework of multi-model of three statistical downscaling approaches was established in which predictions from these models were used as inputs to Artificial Neural Network (ANN). Traditional ANN, Simple Average Method (SAM), and combining models (SDSM, Multiple Linear Regressions (MLR), Generalized Linear Model (GLM)) were applied to a studied site in North-western England. Model performance criteria of each of the primary and combining models were evaluated. The obtained results indicate that different downscaling methods can gain diverse usefulness and weakness in simulating various rainfall characteristics under different circumstances. The combining ANN model showed more adaptability by acquiring better overall performance, while GLM, MLR and showed comparable results and the SDSM reveals relatively less accurate results in modelling most of the rainfall amount. Furthermore traditional ANN has been tested and showed poor performance in reproducing the observed rainfall compared with above methods. The results also show that the superiority of the combining approach model over the single models is promising to be implemented to improve downscaling rainfall at a single site.