Key points are not available for this paper at this time.
Agriculture is the primary profession in India which relies on several climatic factors such as rainfall, temperature, humidity, etc., for the successful growth of crops. Weather and drought forecasting may help to take preventive measures in an unusual situation due to crop failure. Most of the existing work attempts to forecast the weather or analyse the reason for the occurrence of drought and its effects in the past. This paper adopts the machine learning model called the long short-term memory (LSTM) neural network to forecast the long-term rainfall and standardised precipitation index for drought estimation. The predicted annual rainfall from LSTM is taken as an input to forecast the drought conditions of India for the upcoming years, which is a unique approach and objective, and based on this work appropriate decisions can be made for future actions. The past 117 years of rainfall and drought conditions are compared with the recent 50 years by analysing the data in several ways in order to predict the future scenario. The forecasted results are compared with actual observations to demonstrate the effectiveness of the LSTM model to produce adequate results. The error and network loss of the model is 0.059 and 0.0036, which is minimal, and the forecasted rainfall level is almost equal to actual level specifically accuracy is 99.46% for the previous year, 2021. It was found that there is a rainfall decline of 0.04% every year. Apart from the prediction for the country, a clear picture of the region regarding drought forecasts is presented in this work. The real-time drought level is mild and moderate for most of the regions in the country which matches with the drought level determined using forecasted rainfall.
Poornima et al. (Thu,) studied this question.