High-resolution numerical forecast simulations of two snow events and one flash flood event over the Black Hills of South Dakota were made with a coupled atmospheric-hydrologic numerical modeling system. The Coupled Atmospheric Hydrologic Modeling System (CAHMS) consists of the ARPS atmospheric model, the NCAR-LSM land surface model, overland flow and stream routing based on the CASC2D model, and sub-surface flow using the MODFLOW model. Simulations were run in a nested configuration, with the inner atmospheric model covering most of western SD and neighboring states centered on the Black Hills with a horizontal grid interval of 4 km.

Simulations of a moderate snowfall case and a heavy (blizzard) snow case revealed that the model was capable of making reasonable predictions of precipitation, location, amounts and type. Critical elements included: 1. Prediction of precipitation type, snow or rain, was greatly improved through the use of the precipitation type from the explicit microphysical scheme, rather than use of a surface temperature-based parameterization as is common in operational models. 2. Use of a constant generic snow density value caused large errors in snow water equivalent-snow depth conversions. 3. Precipitation location and amounts were better predicted when a detailed land surface parameterization (LSM) was used, than when a simple force-restore parameterization was used. 4. Use of the detailed land surface parameterization permitted a reasonable simulation of the pattern of snow melting, and perhaps of snow melting rates.

Simulation of the Rapid City flash flood event of 1972 revealed that: 1. Simulation of the flood-producing storms over the Black Hills was dependent on initialization, including a low level frontal (wind shift) boundary along the Hills. 2. Simulated peak discharge in narrow stream basins in complex terrain is very sensitive to spatial errors in the simulated precipitation. Shifts of one-to-two grid points easily changed maximum stream discharge amounts by factors of two or more. 3. The hydrologic simulation was very sensitive to several parameters in the land surface model in partitioning between infiltration and runoff.