A CNN Bidirectional LSTM framework for predicting monsoon rainfall in India

Abstract

Rainfall prediction has evolved as a paramount research significance in recent times due to its complexities and ongoing demand such as water resource planning and management. Agriculture is a major source of employment in India, as well as a substantial contributor to gross domestic product, and crop output is dependent on the monsoon season. Rainfall prediction is useful to authorities for water storage and timely release to increase crop productivity. The current study proposes a Deep Neural Network (DNN) based hybrid model using a combination of convolutional neural network bi-directional long short-term memory (CNN BiLSTM) to predict monthly rain fall during monsoon seasons. The DNN models were used to analyze the average monthly rainfall data collected across the country from 1871 to 2019 during the monsoon seasons. Furthermore, the hybrid model's results were compared to the Bidirectional LSTM (BiLSTM) architecture. In predicting rainfall in India, the proposed hybrid model framework has been found to be more accurate than the BiLSTM. The findings of the study suggest that a DNN frame work can be successfully adopted for time series analysis in water resource management and related domains to reduce the associated risks.