Predicting Indian Monsoonal Rainfall Using a Sequential Neural Network on Principal Components of Atmospheric Data

Thomas Favata

Abstract

Monsoonal variation effects many people across India, and a better way to predict the extent of this phenomena would greatly benefit the Indian people. Surface temperature has been used in most literature to create statistically significant forecasts of the Indian Monsoon. Empirical Orthogonal Functions (EOFs) and their Principal Components (PCs) were calculated to find the spatial and time series components respectively. This research uses a sequential neural network to incorporate the PCs of the heights of pressure layers in addition with the temperature data to create a more comprehensive prediction method of Indian monsoons.