Composite Heat Wave Risk (CHWR) framework using Principal Component Analysis (PCA) and machine learning for assessing agro community resilience

Authors: Dolgobinda Pal, Saon Banerjee, Nidhi Nagabhatla, Sarath Chandran M. A., Amitava Panja, Sarathi Saha, Manish Kumar Naskar and Subhadeep Sarkar.

Scientific literature highlights that rising mean temperatures, combined with increasing frequency and duration of extreme heat events, pose risks to food security, particularly in climate sensitive regions of South Asia. These impacts are especially critical in West Bengal, India, where diverse agroecological condition and climate-sensitive farming systems create uneven socio-climatic stress. Despite growing concern, district-level heat wave risk assessment for agricultural communities in West Bengal remains limited, particularly in frameworks that integrate climatic, agricultural, and socio-economic dimensions with predictive modelling. In this milieu, the present research makes a significant contribution to climate resilience building by advancing a composite, spatially explicit heat wave risk assessment grounded in the IPCC Sixth Assessment Report (AR6) risk framework. By integrating climatic hazards with exposure and socio-economic vulnerability and using Principal Component Analysis (PCA) Composite heat wave risk index (CHWRI) were derived. Further to capture nonlinear spatio-temporal dynamics, Long Short-Term Memory (LSTM) model optimized using the Hippopotamus Optimization Algorithm (HOA) was employed to predict district-level heat wave risk classes. The results revealed pronounced spatial heterogeneity, with the red and lateritic, coastal saline, and old alluvial zones exhibiting compounded risk. The observed and predicted district-level CHWRI classes showed strong agreement, with 16 of 22 districts correctly classified (72.73%). Spatial statistics confirmed significant clustering (Global Moran's I = 0.47), while Purulia emerged as a major hotspot in Getis-Ord Gi∗ analysis. The proposed PCA–LSTM-HOA framework offers a novel, interpretable, and policy relevant tool for hotspot prioritization, early warning, and climate resilient agricultural planning —an urgent priority in the face of accelerating global warming.