Implementing Machine Learning Models for Drought Prediction Based on Metrological Drought Indices with Varying Time Scales: A Case of Latur Region

Abstract

Droughts have a huge socioeconomic impact, thus globally affecting the people, the environment, and the economy on a massive scale. Proper monitoring and accurate predictions can help mitigate the ill effects of such natural calamities to a considerable level. Providing warning signals and proper preplanning can provide further solutions. Drought occurrences highly depend on the environmental characteristics of the region. To identify drought and analyze its severity level, metrological drought indices like Standard Precipitation Index (SPI), Standardized Precipitation and Evapotranspiration Index (SPEI), and Reconnaissance Drought Index (RDI) are popularly used. The present work focuses on drought prediction using three drought indices SPI, SPEI, RDI. The case of Latur region of Maharashtra, India, has been taken for the research work. The prediction model is developed using support vector regression (SVR) and long short-term memory (LSTM) for varying timescales 1, 3, 6, 9 and 12 months, and the performance of the model is evaluated across measures like mean absolute error (MAE) and root mean squared error (RMSE).