Heterogeneous urban decarbonization pathways revealed by typological and machine learning frameworks: A global analysis across Annex I cities

Cities are central to global climate mitigation, yet existing projections often rely on coarse-grained assessment models that overlook the interplay between urban typology and socioeconomic dynamics. Here, we harness a context-aware deep learning framework combined with typological clustering to project CO₂ emissions for 46,833 cities in Annex I countries through 2050 under Shared Socioeconomic Pathways. Specifically, we classify cities by multidimensional urban morphology and couple the resulting typologies with FiLMSeq2Seq prediction model we proposed to generate city-level emission projections. We identify five distinct urban typologies characterized by differing patterns of horizontal expansion, economic intensification, and vertical structure. We find that emission trajectories diverge significantly across these typologies: Under the Business-As-Usual scenario, ‘Sprawling’ cities have the highest proportion of emission increases, with 31.5% of the 1335 cities showing growth. Conversely, ‘Moderate vertical’ cities will dominate future mitigation with a total emission reduction of 798.2 Mt. Our results suggest that uniform national policies are unsuitable for diverse urban typologies, underscoring the need for type-specific strategies to realize the full decarbonization potential of cities.