Machine Learning Approach on Input Variable Selection for Improved Monthly Rainfall Prediction Considering Teleconnections

Abstract

Monthly rainfall prediction is a challenging task due to the complexity of rainfall and atmospheric system. In prediction models based on Artificial Neuron Network (ANN), meteorological parameters and their spatiotemporal variations should be carefully selected. In this study, a new input variable selection method called Zero Input Test (ZIT) is proposed. A comparative study between ZIT and correlation coefficient (CC) is carried out considering the global data of atmospheric variables from lags of 1 to 12 months. Selections from the two approaches are evaluated by building two ANN models based on CC and ZIT selections to predict monthly rainfall of Upper Chao Phraya River Basin (UCPRB), Thailand. The results of the numerical experiment confirm superiority of ZIT towards CC. However, the rainfall prediction results from both methods show a similar accuracy and prediction trends even though selections from CC and ZIT are different. This suggests that there might be multiple regions affecting UCPRB rainfall.