Applicability of machine learning models for drought prediction using SPI in Kalahandi, Odisha
Year: 2025
Type: article
Abstract: This study assesses the performance of auto-regressive integrated moving average (ARIMA), artificial neural network (ANN), support vector machine (SVM) and extreme learning machine (ELM), in predicting meteorological drought with Standardized Precipitation Index (SPI-6 and SPI-12) for Kalahandi district, Odisha. Mann-Kendall tests showed no significant trend in SPI value for both shorter and longer scales. Model performance was evaluated using correlation coefficient (CC), root mean square error (RMSE), Nash-Sutcliffe efficiency (NSE), and mean absolute error (MAE) during the training as well as testing phases. For SPI-6, ARIMA performed well during training (NSE = 0.66, RMSE = 0.60) but showed a decline in testing (NSE = 0.25). Machine learning models, including ELM, SVM and ANN exhibited better consistency, with NSE values ranging from 0.45 to 0.47. For SPI-12, ANN delivered the highest accuracy with NSE values of 0.91 and 0.89 and RMSE values of 0.31 and 0.29 in training and testing, respectively. Graphical analysis further demonstrated that ANN and SVM outperformed ARIMA by effectively capturing nonlinear trends and extreme fluctuations. Overall, machine learning models, particularly ANN and SVM, proved to be superior for predicting both long-term (SPI-12) and short-term (SPI-6) precipitation indices, highlighting their effectiveness for accurate drought forecasting.
Source: Journal of Agrometeorology
Authors: AMIT PRASAD, R.K. SINGH, K V RAMANA RAO, C. K. SAXENA
Language: English
Subfield: Global and Planetary Change
Field: Environmental Science
Domain: Physical Sciences
SDGs: Climate action, Zero hunger
Open Access status: diamond
OpenAlex ID: W4410990821