Local policymakers require spatially detailed climate projections to plan heat adaptation strategies, yet regional climate models (RCMs) at 5–12 km resolution cannot resolve the microclimatic heterogeneity that governs urban heat exposure. We present a statistical-dynamical downscaling method that combines bias-corrected EURO-CORDEX ensemble output with the mesoscale urban climate model FITNAH-3D to produce 50 m resolution maps of heat-related characteristic days across the German federal state of Baden-Württemberg (~36,000 km2, 24.7 million grid cells). The method transfers the spatial temperature pattern simulated by FITNAH-3D under idealized autochthonous conditions to the full temporal variability of the RCM ensemble through a validated empirical correction factor. Validation against 66 stations over the reference period (1971–2000) yields mean absolute errors (MAE) of 6 days/year for summer days (≥25 °C) and 2 days/year for hot days (≥30 °C). The downscaled ensemble mean projects, under +2 K summer warming (approximately RCP 4.5 mid-century), an average of 55 summer days and 17 hot days per year, with intra-domain ranges of 13–97 and 2–42, respectively. Under +3 K warming (approximately RCP 8.5 mid-century), tropical nights increase to a domain-averaged 3 per year, with maxima of 51 in densely built urban cores. A sensitivity analysis shows that the correction factor is robust: ±10% perturbation changes MAE by less than 2 days/year. We discuss the effective resolution after post-processing, the physical basis for the spatial transferability of the correction factor, and the implications of ensemble spread for adaptation planning. The methodology has been operationalized in Baden-Württemberg’s Climate Atlas.