Geospatial machine learning maps urban heat inequality for climate resilient planning in Bologna

Sustainable, climate-resilient urban development requires evidence linking thermal exposure, social vulnerability, and the built environment. This study develops a geospatial machine-learning framework for Bologna, Italy, to identify areas where urban heat, social vulnerability, and green-infrastructure deficits overlap. The framework integrates land surface temperature (LST) with data on 51,976 building footprints, urban morphology, infrastructure, vegetation and land cover, meteorological conditions, and demographic and socioeconomic characteristics. A Random Forest (RF) model predicted building-level LST during summer 2023, while a Long Short-Term Memory (LSTM) model characterized hourly air-temperature dynamics using meteorological records from 2019–2023. With spatially tuned hyperparameters, the RF model achieved R 2 = 0.749, MAE = 1.16 °C, and RMSE = 1.50 °C under a conventional random split. Under five-fold spatially blocked cross-validation, pooled performance was R 2 = 0.579, MAE = 1.58 °C, and RMSE = 1.97 °C, indicating optimistic random validation while retaining predictive capability in spatially separated areas of Bologna. Local relief, built-up cover, tree cover, sky-view factor, shadow conditions, industrial and commercial land, and distance to parks were the most influential predictors. The tuned LSTM model captured one-hour-ahead air-temperature dynamics in 2023 with R 2 = 0.995, MAE = 0.43 °C, and RMSE = 0.59 °C, outperforming the persistence baseline RMSE of 1.07 °C. At the planning scale, a Heat Vulnerability Index (HVI) was developed for 90 statistical areas by integrating heat exposure with socioeconomic and demographic vulnerability. Sensitivity analysis showed that HVI rankings were robust to variations in component weights, with a median Spearman rank correlation of 0.992 and agreement in nine of ten areas between recalculated and baseline top ten rankings. CAAB, Michelino, Ducati–Villaggio INA, San Donnino, and Pilastro emerged as the highest-priority areas for targeted heat-mitigation and climate-adaptation interventions. By linking fine-scale heat patterns with local adaptive capacity, the framework supports prioritizing tree-canopy expansion, shading, cool surfaces, and improved access to green spaces. It contributes to the Sustainable Development Goals (SDGs) and can be adapted to other compact European cities after local calibration.