Multi-sensor remote sensing with Bayesian uncertainty for urban heat vulnerability mapping in Dhaka City

Urban heat vulnerability is intensifying globally, yet long-term, spatially explicit assessments remain limited for rapidly growing megacities in the Global South. This study develops a decadal (2015–2024), uncertainty-aware Heat Vulnerability Index (HVI) for Dhaka, Bangladesh, addressing limitations of static and deterministic assessments, by integrating multi-sensor Earth observation indicators: surface urban heat island anomalies derived from Landsat thermal data, vegetation scarcity from NDVI, night-time lights from VIIRS, and built-up probability from Sentinel-2 Dynamic World. Indicators were robustly normalized (2–98%) and aggregated into a weighted composite index (0–1). Uncertainty was quantified using a Bayesian bootstrap framework, generating 95% credible intervals to ensure statistical robustness. Mean HVI increased from 0.25 (2015) to 0.43 (2017), a 72% rise, before stabilizing around 0.35–0.37. In contrast, the area exceeding the high-vulnerability threshold (HVI ≥0.7) expanded from <3 km² (2016) to 123.8 km² (2024), revealing spatial diffusion of extreme vulnerability despite stable citywide averages. Regression analysis shows SUHI explains 62% of inter-annual variability (p < 0.01). Bootstrap standard deviations remained below 0.06, confirming statistical stability. These findings indicate a shift from early intensification to outward diffusion driven by impervious expansion and vegetation loss. An interactive Google Earth Engine dashboard operationalizes the framework, supporting weight sensitivity, threshold analysis, and uncertainty visualization for climate adaptation planning in tropical megacities.