A Comparative Study of SARIMAX and Artificial Neural Networks for Drought Forecasting in North Karnataka

Abstract

Drought forecasting is crucial for managing its effects on crop production and water supply. This study evaluates the effectiveness of the Seasonal Auto-Regressive Integrated Moving Average model with external predictors (SARIMAX) and Artificial Neural Networks (ANN) in predicting the Standardized Precipitation Index (SPI), a crucial indicator of drought conditions in Karnataka. Using data from 2000 to 2023, both models incorporate variables such as rainfall, temperature, NDWI (Normalized Difference Water Index), and NDVI (Normalized Difference Vegetation Index). The results indicate that the SARIMAX model significantly outperforms the ANN model, with a lower RMSE of 0.2699 and MAE of 0.2114, highlighting its superior accuracy and predictive reliability. The SARIMAX model also demonstrates minimal bias (ME = 5.89e-14) and uncorrelated residuals (ACF1=-0.0750), confirming its robustness in capturing the underlying trends. In contrast, the ANN model exhibited higher errors and lower predictive performance in extreme drought conditions. Based on these findings, the SARIMAX model is recommended as the more effective tool for SPI forecasting in North Karnataka, offering a reliable approach to enhancing agricultural resilience in drought-prone areas.