An innovative ensemble approach of deep learning models with soft computing techniques for GIS-based drought-zonation mapping in Rarh Region, West Bengal.

Abstract

Drought is a complex natural calamity that has serious consequences for ecosystems and society, demanding its identification for effective mitigation. This study analyzed drought scenarios in West Bengal's Rarh Region at 3-, 6-, and 12-month intervals, as the Birbhum and Purba Bardhhaman districts are experiencing decreasing rainfall trends. Purba Bardhhaman, noted for its rice production, is undergoing severe drought, affecting agriculture and food security. The current study analyzed 27 drought assessment factors from meteorological, agricultural, hydrological, and socioeconomic perspectives. A Multi-Layer Perceptron Neural Network (MLP NN) was used as the benchmark, followed by a DenseNet neural network. A Hybrid Deep Learning Ensemble model was built to provide a precise drought-prone map. The results showed that 26.66% of the region is very highly drought-prone at a 3-month interval, 20% at 6 months, and 25% at 12 months. The Hybrid Deep Learning Ensemble model had the highest accuracy, with ROC-AUC values of 94.2%, 94.3%, and 95.3% at 3, 6, and 12-month intervals, respectively. The study provides crucial insights for West Bengal policymakers to handle rising drought risks, underlining the importance of implementing appropriate drought management techniques. This study emphasizes the importance of the spatial scope and underlying causes of drought sensitivity to specific mitigation strategies that ensure sustainable development.