Analysis of shift in Indian monsoon and prediction of accumulated cyclone energy in Indian subcontinent using deep learning

Abstract

Every year India faces many cyclones and erratic monsoon seasons are common in recent times. Cyclones destroy the infrastructure and lead to loss of life and damage property in coastal areas. The agriculture sector is also affected by random and unexpected rainfall. In recent years, India gets rainfall during the harvest season which leads to financial loss. Also, the number of drought events is on the rise in the Indian subcontinent as the rainwater is not managed properly. Farmers need to know whether the monsoon rainfall pattern has been shifted or not and need to shift their agricultural activity accordingly to handle the impacts of climate change. From the rainfall and accumulated cyclone energy (ACE) data analysis, it is found that monsoon seasons in India are not shifted, but, rainfall is intense during the initial months of each monsoon season. ACE values are predicted using techniques such as ARIMA, LSTM, Prophet, and stacked ensemble with multi-layer perceptron. Based on the experimental results, the proposed stacked ensemble model achieves 91% accuracy.

Keywords

• ARIMA
• prophet
• LSTM
• rainfall shift analysis
• accumulated cyclone energy
• ensemble