Accurate predictions for wind and particle dispersion must be obtained quickly to inform emergency response to atmospheric releases of hazardous material in urban areas. In this study, we demonstrate and validate a newly implemented resolved building capability added to the exascale Energy Research and Forecasting (ERF) model. To represent buildings, an immersed forcing method is employed, which enforces near-zero velocities within buildings. We modify the immersed forcing method to implement a wall-model where we enforce a calculated velocity for building walls determined using the log law. The immersed forcing method is evaluated by simulating cases from the Joint Urban 2003 Oklahoma City tracer release experiment, analyzing cases with both stable and unstable thermodynamic conditions. For the stable case, we use adaptive mesh refinement for regions above a certain scalar concentration. Additionally, we conduct two sensitivity studies: one on building wall roughness length, finding that 0.1 m results in the most accurate scalar concentration predictions, and another on model configurations, where computational cost decreases by an order of magnitude when either not refining the grid or when using anelastic mode compared to fully compressible mode, with model skill only decreasing slightly. For the convective case, model skill is not as high as the stable case largely due to underprediction of scalar concentration at stations to the west and upwind of the release site. Ultimately, model skill for both cases are still among the highest in the published literature, demonstrating the efficacy of our method for accelerated high-fidelity urban atmospheric modeling applications.