STUDY OF NEW RAINFALL PREDICTION METHOD BASED ON DEEP LEARNING

Abstract

Heavy rainfall has been causing many dreadful disasters, but its prediction is quite difficult in current weather forecast models. Today deep learning breakthrough in various field. We attempted to apply a deep learning algorithm, or stacked LSTMs, to rainfall prediction. Our LSTM weather prediction model was established with micro climate data in Kyusyu from 1990 to 2015 and validated for 2016-2018 rainfall events. The model was able to predict even weak rainfall but showed disadvantage for heavy rain fall events as well as rainfall onset prediction. We also suggested possible solutions for increasing reliability of LSTM rainfall models.