Mapping agricultural fragility in India through a yield gap vulnerability framework: a national-scale machine...

In India, regional yield gaps continue to widen despite increased agricultural productivity, owing to socioeconomic inequality and climate variability. This study develops a novel Yield Gap Vulnerability (YGV) framework to measure agricultural fragility at the district-level by integrating agricultural, hydrological, meteorological, and socioeconomic indicators to observed yield gaps for major cereals (rice and wheat) and nutri-crops (maize and millet). An integrated Machine Learning (ML) approach is used that can support sustainable agricultural management, spatial planning, and risk reduction. Results reveal pronounced spatial heterogeneity in crop performance across the nation. While modal yields of rice and wheat have increased, the share of high-vulnerability districts has also risen, particularly in resource-stressed regions such as the Indo-Gangetic Plains. In contrast, maize and millet exhibit emerging yield bimodality, indicating localized productivity gains from intensified management within predominantly rainfed systems. Model evaluation further shows that XGBoost consistently outperforms Random Forest and Artificial Neural Network models in both the hydro-meteorology-only baseline and the all-inclusive (hydro-meteorology and socioeconomic) variant. These findings provide robust statistical evidence that India's agricultural vulnerability is shaped not only by climatic stressors but also by socioeconomic and institutional capacity. Our proposed YGV framework provides a data-driven decision-support tool for adaptive agriculture management by combining biophysical and socioeconomic aspects. It strengthens long-term risk management methods by allowing policymakers to identify vulnerability hotspots and prioritize region-specific measures. Scalable and adaptable, this ML-based approach can be applied across various agricultural systems facing compounded socioeconomic and climate risks.