Models have limited capacity in simulating coupled surface-subsurface urban hydrological processes at regional or larger scales due to two major challenges. On one hand, current urban flood models are highly distributed, data-hungry, and computationally demanding. On the other hand, urban infrastructure data are rarely available to the public at regional or larger scales. To tackle the first challenge, we propose a semi-distributed, physically based model called Model for Scale Adaptive River Transport – urban module, MOSART-Urban, that is applicable across the watershed, regional, and larger scales. MOSART-urban simulates surface runoff from impervious urban areas into below-ground urban stormwater networks (BUSNs). BUSNs and street networks are coupled through street inlets, i.e., water can flow from streets to BUSNs and vice versa (overflow). To tackle the second challenge, we develop a novel algorithm for estimating BUSNs at the watershed or larger scales based on ubiquitous street network, topographic, and land use/land cover datasets and drawing on graph theory concepts. We validate the derived BUSNs in the Baltimore, Houston, Los Angeles, and Seattle metropolitan areas in the US, where partial real BUSN data is available. By using a spatial proximity metric, we show that the framework can capture 59-76% of the real BUSN pipes in terms of the total length. We use the derived BUSNs to parameterize subsurface hydrologic processes in MOSART-Urban. Finally, we successfully validate MOSART-urban against the observed daily streamflow data at representative urban watersheds with different levels of urbanization, obtaining satisfactory performance with reasonable computing efficiency. MOSART-urban is thus suitable for representing coupled surface-subsurface urban hydrological processes in land surface or Earth system models at regional or larger scales.