Urban Waterlogging Susceptibility Modeling and Predictive Contribution Analysis of 2D Urbanization and 3D Building Morphology in the Guangdong–Hong Kong–Macao Greater Bay Area

Urban waterlogging susceptibility in coastal urban agglomerations is shaped by horizontal urbanization and vertical densification, yet the incremental explanatory contribution of three-dimensional (3D) building morphology beyond natural environmental and two-dimensional (2D) urbanization factors remains unclear at the urban-agglomeration scale. Moreover, data-driven waterlogging susceptibility modeling is challenged by pseudo-absence uncertainty, spatial redundancy, and class imbalance in historical waterlogging inventories. This study examined the Guangdong–Hong Kong–Macao Greater Bay Area using 2014–2020 waterlogging records and multisource spatial data at a 300 m grid scale. KDE-based inverse-density downsampling and CTGAN augmentation strategy were integrated with an XGBoost-SHAP framework, and four ablation models were used to quantify the relative and incremental contributions of 2D urbanization and 3D building morphology. Under the stratified random-split evaluation, the sample-optimized model achieved ROC-AUC and PR-AUC values of 0.9681 and 0.8851. Relatively high- and high-susceptibility zones occupied only 11.85% of the study area but contained 95.04% of historical waterlogging records, forming a pronounced multicore pattern. Ablation analysis showed that 2D urbanization factors provided greater independent predictive benefit than 3D building morphology, while 3D morphology still yielded modest additional gains beyond natural environmental and 2D urbanization factors. SHAP identified impervious surface proportion and road density as showing the highest model-based predictive contributions, followed by building volume and building height, with 3D building morphology factors exhibiting stronger stage-dependent nonlinear responses. These findings indicate that horizontal urbanization provides the greatest predictive contribution to regional susceptibility assessment, whereas 3D building morphology plays a complementary rather than substitutive role by refining susceptibility differences within intensively developed urban areas.