TransMod Unifies Urban Mobility Forecasting Across Modes

Urban transportation systems are complex, featuring multiple coexisting mobility modes with intricate interdependencies that lead to correlated demand dynamics. However, forecasting demand across these diverse modes jointly is challenging due to significant spatial heterogeneity and often limited historical data for newer or emerging modes. Existing forecasting methods typically focus on individual modes and assume compatible spatial structures, which restricts their utility in multi-modal environments. To overcome these obstacles, a new framework called TransMod has been developed for unified urban mobility demand forecasting. TransMod's core innovation lies in constructing a shared, zone-level spatial representation. This representation aligns mobility systems with different spatial granularities into a common space, effectively reducing structural mismatches and distributional shifts between modes. Building upon this unified representation, TransMod then learns transferable spatio-temporal patterns from data-rich source modes and adapts them to data-scarce target modes. This mechanism significantly reduces the reliance on extensive historical data for newer modes. Extensive experiments using real-world datasets demonstrate that TransMod consistently outperforms existing approaches and maintains robust forecasting performance even when target data is limited.