DROUGHT PREDICTION BASED ON SPI WITH VARYING TIMESCALES USING MACHINE LEARNING AND DEEP LEARNING MODELS

Abstract

Drought modeling is an important issue because it is required for curbing or mitigating its effects, alerting the people to its consequences and water resources planning. The moni toring and forecasting of droughts plays a key role in the assessment of ecosystem health and mitigating the impact of extreme weather events on human society. This study in vestigates the capability of a deep learning method, Long Short-Term Memory (LSTM), in forecasting drought calculated from monthly rainfall data of Palakkad, Kasaragod and Punalur in Kerala. Due to the complexity of the drought phenomena and the requirements for its assessment, several indices have been developed and used in assessing drought events. Among these indices, SPI (Standardized Precipitation Index) is recommended by the WMO (World Meteorological Organization). Root Mean Square Error (RMSE), Mean Absolute Er ror (MAE) , Mean Square Error (MSE), Coefficient of Determination (R²) , Radar Plot,Box Plot and Violin Plot of SPI-1, SPI-3, SPI-6 and SPI-12 for different models like Random Forest, Support Vector Regression (SVR) and LSTM are compared with each other in order to find the best model. The overall results showed that the LSTM method performed su perior to the Random Forest and SVR in forecasting drought based on SPI-1, SPI-3, SPI-6 and SPI-12. From the study it is proven that SPI-12 shows better performance in LSTM time series prediction