Interpretable multimodal generative design of three-dimensional urban hub morphology for resilient and sustainable planning

Rapid urbanization is intensifying the planning challenge of high-density urban hubs, where transport demand, land-market pressures, and population dynamics interact nonlinearly. This study proposes an interpretable, end-to-end generative framework that fuses multimodal urban signals—transportation accessibility (ACC), housing prices (HP), and population heat (POPH)—to synthesize block-scale three-dimensional (3D) hub morphologies and support prospective scenario testing. Guangzhou (China) is used as a testbed with six representative 600 × 600 m hub types (innovation, MICE, tourism, financial, transportation, and industrial). Two conditional GAN surrogates (Pix2pix and CP-GAN) are benchmarked, and CP-GAN is selected for its higher fidelity in reproducing urban texture, land-use/land-cover patterns, and functional distributions. To move beyond “black-box” generation, we conduct controlled univariate perturbations of ACC/HP/POPH (−30% to +30%) and quantify type-specific responses in urban morphology indicators, LUCC, and POI composition, complemented by 2D-to-3D massing reconstruction. Results reveal distinct development regimes: accessibility induces near-linear intensification in industrial hubs but saturating or U-shaped trajectories in mature hubs; Land-value perturbations indicate an intermediate range in which redevelopment more strongly promotes diversification (e.g., industrial and MICE hubs), whereas outside this range the system tends toward intensified specialization (e.g., financial and tourism hubs); and population pressure yields sustained densification in industrial/innovation hubs but threshold effects in tourism hubs. The framework provides an actionable pathway to generate, compare, and interpret resilient and sustainable hub design alternatives before implementation.