DEVELOPMENT OF A SEASONAL FORECAST MODEL OF PRECIPITATION USING GLOBAL SURFACE TEMPERATURE IMAGES BY DEEP LEARNING

Abstract

 In this study, we predicted monthly rainfall after two months in the Chao Phraya river in Thailand by using global surface temperature image and convolutional nueral network which is one of the Deep Learning. We evaluated accuracy of two prediction models which was different neuron number. As a result, The model in which the number of neurons was undulated was more higher accuracy prediction. The undulation like as increasing, decreasing and increasing. We divided all of prediction results into every month’s data and evaluated them using RMSE, MAE and RMSE/MAE. As a result, RMSE and MAE were the high values in November to April and the low values in May to October. RMSE/MAE values are higher than values follow normal distribution. Accuracy of this model which predicts monthly rainfall was higher in the rainy season than in the dry season but in the rainy season, many outliers were predicted.