Spatially Explicit Socio-Environmental Vulnerability to Urbanisation and Climate Extremes in Bihar Using Multi-Source Earth Observation, Spatial Statistics and Machine Learning

Abstract

Bihar, one of India's most populous and disaster-prone states, is experiencing rapid, largely unplanned urbanisation superimposed on escalating climate extremes — recurrent floods, extreme rainfall, droughts and heatwaves. Conventional assessments overlay land use/land cover (LULC) change on socio-economic layers but rarely ask where, and for whom, urbanisation and climate extremes combine into vulnerability. This study develops a spatially explicit socio-environmental vulnerability assessment that integrates multi-source Earth observation (EO), spatial statistics and machine learning (ML) to ask how rapid urbanisation alters exposure to climate extremes, and how unevenly these pressures are distributed across socially vulnerable populations. We develop a Socio-Environmental Exposure and Vulnerability Index, SEVI = f(H, E, S, A), integrating hazard intensity, population exposure, social sensitivity and adaptive capacity, and a Socio-Environmental Inequality Gradient, SEIG = z(EEI) − z(SAC), which contrasts environmental exposure with social adaptive capacity and sorts places into four inequality quadrants. The framework is applied to all 38 districts and to a 500 m grid across 12 rapidly transforming cities, using multi-source EO (Landsat 5/7/8/9, Sentinel-1 SAR, Sentinel-2 and MODIS, 2000–2024), gridded climate records, Census and facility data, and spatial statistics (Global and Local Moran's I, Getis-Ord Gi*, spatial lag/error models, multiscale geographically weighted regression, MGWR) and ML (random forest, XGBoost with SHAP interpretation, K-means clustering). In the present simulated realisation of the pipeline, built-up area expanded by 135 per cent, heatwave days rose by 2.4 days per decade, newly urbanised cells warmed 2.8 °C more than stable non-urban cells, and SEVI was strongly clustered (Moran's I = 0.71) with a hot spot across the Kosi–Mithila floodplain; SEVI tracked observed hazard impacts (ρ = 0.80). Seven districts (20 per cent of the population) and 28 per cent of the urban population fall in the 'double-burden' quadrant of high pressure and low capacity. Inequality is hazard-specific: flood burden is concentrated among low-capacity populations (concentration index −0.17 for districts, −0.13 within cities), whereas heat burden is concentrated in denser urban cores. SHAP shows that realised vulnerability is driven primarily by flood exposure, amplified by low literacy, livelihood precarity and thin health services. The resulting vulnerability atlas supports climate-resilient urban planning and inequality-aware adaptation in Bihar.