In this study, we explore the potential to improve the capability of an existing global flood model in monitoring and forecasting flash floods which usually occur in upstream headwater catchments at spatial scales not necessarily commensurate with the resolution of global models. Specifically, based on the widely used GFMS and the Dominant river tracing-Routing Integrated with VIC Environment (DRIVE) model (Wu et al., 2014, WRR), we recently further proposed a Glocal (global to local) Hydrometeorological Solution to Floods (GHS-F, Wu et al., 2020, AAS) which is able to provide both large scale flood detection and detailed flood delineation at local scales with sufficient levels of accuracy,for better preparedness, mitigation, and management of significant precipitation-caused different types of flooding. A Time-Space varying Distributed Unit Hydrograph (TS-DUH) based on publicly-available-only data is proposed for efficient flash flood forecasting within the GHS-F framework. As in the traditional spatially distributed unit hydrograph (SDUH) method, TS-DUH initially estimates the runoff travel time (and flow velocity) from each location within a catchment to the outlet based on topographic and hydroclimate characteristics. The delineation of runoff-drainage process is further adjusted by considering the heterogeneous and dynamic runoff contribution caused by rainfall and soil moisture variations. The excess rainfall is estimated by the Soil Conservation Service's (SCS) curve number method. An alternative excess rainfall input is taken from the GFMS which provides long-term (2000-present) well-archived and real-time operative global runoff datasets from the state-of-the-art DRIVE model. The performance of the TS-GUH method is evaluated using 5,517 flash flood events of 243 small-to-medium-sized catchments in the CONUS, with 1458 events used for calibration. The validation results show that using DRIVE-Runoff is better than SCS-CN, 98% and 72% of events have KGE values greater than 0 and 0.5, respectively, with a median KGE value of 0.6 and the probability of detection (POD) of flood events is 0.9. More importantly, using near real-time satellite rainfall-driven DRIVE-Runoff, long-term flow simulation (2003-2020) without calibration at 803 gauges show better performance of TS-DUH than the original GFMS, with a median KGE improvement of 0.14. This combined UH and numerical hydrological model approach showed great potential for flash flood monitoring and forecasting at regional or global scales.

