Prediction of rainfall through a ML/DL approach over major metro cities of Uttar Pradesh, India

Abstract

 Weather forecasting is an important and challenging attribute to predict because most atmospheric and agricultural fields depend on day-to-day weather fluctuations. Rainfall is one of the most important parameters dependent on various climatic conditions. We used the Random Forest (RF) and Long Short-Term Memory (LSTM) Neural Network models to predict rainfall for the major cities of Uttar Pradesh in the months of June, July, August, and September (JJAS) from 1901 to 2020. We used various statistical indices like the correlation coefficient (CC), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE) to assess the quality of the forecast. Several climate indices were used as predictors to forecast rainfall as mentioned above. These indices include the North Atlantic Sea Surface Temperature, Nino 3.4, Equatorial south-east Indian Sea, and rainfall at a lag value of 12. The prediction of rainfall utilizes a predictive neural network model, and the output is compared to real-time observed rainfall data for the forecasted period. The investigation revealed that the LSTM generally performed better compared to the RF. The generated output is promising and can be widely extended in this type of application.