CNN-Based Forecasting of Intraseasonal Mean and Active/Break Spells for Indian Summer Monsoon

Abstract

Indian summer monsoon is highly significant for the economic growth of the country. The rainfall during the monsoon period varies over temporal and spatial scales. Existing works mostly focus on predicting the seasonal mean rainfall and fail to provide insights for any discrepancies, such as persistent dry periods or heavy rainfall within a season. This paper provides a prediction of the intraseasonal mean and active/break spells of Indian summer monsoon with five and ten days lead. The prediction model learns the spatio-temporal relationship of the climatic variables using a Convolutional Neural Network (CNN). The CNN-based model predicts the mean rainfall with a Pearson correlation coefficient of 0.63. We accurately predict the rainfall as active and break spells with an Area Under Curve score of 0.81 and 0.84, respectively. We evaluate the performance of our model against the state of the art model showing significant skill improvement of 36.4% and 29.2% in precision and recall.