Introducing a Nonlinear Model Based on Hybrid Machine Learning for Modeling and Prediction of Precipitation and Comparison with SDSM Method (Cases Studies: Shahrekord, Barez, and Yasuj)

In the present study, a nonlinear hybrid model, based on multivariate adaptive regression splines (MARS), artificial neural networks (ANN) and K-nearest neighbor (KNN) has been presented for downscaling the precipitation of Shahrekord, Barez, and Yasuj under climate change conditions. This model, similar to SDSM, is composed of two steps; classification and regression. The MARS model is employed for classification of precipitation occurrence and the ANN and KNN are employed for determination of the amount of precipitation. The results of MARS showed that the mentioned model is more accurate than the SDSM model. Comparing the results of downscaled precipitation showed that the ANN model is more accurate than the SDSM and KNN in prediction of average annual and monthly precipitation. So that the R value for ANN was 54% more than the one in SDSM model, in Shahrekord. Also, according to the highest accuracy, standard deviation and skewness coefficient, the ANN, KNN and SDSM model ranked first, second, and third, respectively, for prediction of monthly average precipitation in three investigated stations. Eventually, the precipitation changes in the near future (2020-2040) and far future (2070-2100) periods were investigated under the A2 and B2 scenarios of the HADCM3 model. Results revealed that the lowest precipitation reduction is corresponded to ANN (in Shahrekord) and A2 scenario in the near future period and the highest precipitation reduction is corresponded to SDSM (in Yasuj) and A2 scenario in the far future period. Finally, it can be concluded that the proposed model is more accurate than the SDSM model and can be used as an alternative to the SDSM model.