APPLYING DEEP LEARNING WITH CMIP5 DATASET FOR MONTHLY RAINFALL PREDICTION IN THAILAND

Abstract

 The accuracy of climate models decreases drastically when predicting rainfall for more than two weeks, so it is expected that deep learning models can be applied. In this study, deep learning is applied to the Chao Phraya River basin in Thailand to predict rainfall for one to three months during the rainy season. We examined whether the lack of observational data can be solved by using the CMIP5 dataset, which is a non-observational dataset. The results show that CMIP5 can be adapted to the training data. Furthermore, we found that deep learning methods tend to outperform climate models when the prediction lead time is long.