A Deep Learning Ensemble Model for Short-Term Rainfall Prediction

Abstract

Rainfall Prediction is an integral part of research these days with its applications ranging from Disaster Management to Agricultural Technologies. Coastal cities like Chennai are extremely prone to irregular and incessant bursts of rainfall. Prior knowledge of such events is necessary to avoid wastage of resources and reduce damage to livelihood. In this paper, we have investigated several state-of-art algorithms such as SARIMA, LSTM, BiLSTM, RNN, RNN-LSTM that are used for rainfall forecasting. The investigations show that the state-of-art algorithms have reduced error rates in predictions, however fail to handle extreme rainfall events. To overcome this, an Ensemble Model of CNN, RNN-LSTM, and Bidirectional LSTM are proposed to forecast the daily rainfall statistics of Chennai. The proposed model is compared with the baseline models to analyze its performance. The features considered for model implementation are Rainfall, Relative Humidity, and Temperature that is collected on a daily scale. The evaluation result shows that the proposed model provides improved prediction results for Rainfall when compared to the baseline approaches.