Urban policy transfer has the potential to enhance urban systems by applying successful strategies from one city to another. However, existing models for predicting policy impacts are often context-specific and lack generalisability across different urban settings. This study introduces TransGM, a novel framework that enables the transfer of spatial interaction models between cities via adaptive transfer learning. The framework employs: (1) spatial Kullback–Leibler divergence to quantify structural differences across urban contexts; (2) production-constrained gravity models for flow prediction, accounting for residues in urban features; and (3) feature-specific regularisation weights that adapt parameters based on the degree of spatial pattern similarity between source and target cities. TransGM is demonstrated through a case study of workplace attractiveness and its impact on commuting flows transferred from Birmingham to Coventry. The adapted transfer model replicates Birmingham's urban configuration in Coventry rather than learning Coventry's own priority structure. Transfer process is jointly governed by spatial divergence between the two cities and the data sufficiency: where spatial patterns align, local data dominate parameter estimation, while in data-sparse contexts, source-city regularisation guides model behaviour. This balance between place-specific urban structure and cross-city mobility and amenity transferability positions adaptive transfer learning as an effective tool for evaluating the feasibility of replicating one city's development model in another, offering simulation-based evidence to inform urban planning and policy decisions. • Novel TransGM framework enables adaptive inter-city transport policy transfer. • A modular gravity model framework with adaptive regularisation. • Effective adaptation of pre-trained models to new cities with limited data. • A case study of transferring the mixed-use strategy between Birmingham and Coventry. • Adaptive digital urban and transport models applicable to various cities.
Enaya et al. (Tue,) studied this question.