Spatiotemporal Impacts of Vertical Urban Development Patterns on Carbon Emission Intensity: A GeoAI Framework Using Multisource Remote Sensing Data

Urban carbon performance emerges from complex interactions among built form, functional mix, population concentration, and ecological conditions, making it difficult to interpret how vertical urban development shapes carbon emission intensity (CEI) across space and time. Existing approaches often rely on single-form indicators or context-insensitive models that overlook typological differentiation, spatial heterogeneity, and nonlinear effects. This study develops a typology-aware GeoAI framework to examine the association between vertical-development-oriented urban morphology and CEI across five major Chinese cities in 2005, 2010, 2015, and 2020 using multisource geospatial data, including ODIAC emissions, gridded GDP, building information, OSM-based POIs, WorldPop, and NDVI. Multiple machine-learning models were benchmarked; MGWR was used as a spatial explanatory reference, and SHAP was applied to recover nonlinear associations. XGBoost achieved the best cross-validation performance (R2 = 0.862, RMSE = 0.077, MAE = 0.004). Office-building POI density, population density, and functional diversity were consistently associated with CEI variation, whereas the dominant relationships were typology-specific and time-sensitive. Building height showed a nonlinear positive association in compact high-density cores, while the building-related greening proxy and office-building POI density exhibited recurring nonlinear patterns in mixed-use and expansion zones. Overall, the framework identifies planning-relevant association ranges rather than deterministic policy cutoffs.