K2-RHEM, a recently integrated event-based rangeland watershed model, represents one of the few process-based models available for rangeland applications. However, to gain wider acceptance and demonstrate its reliability, comprehensive evaluation results are essential. In this study, K2-RHEM was evaluated in five small semi-arid watersheds within the USDA-ARS Walnut Gulch Experimental Watershed. Using extensive runoff and sediment data, along with field surveys on channel heads, soil textures, and channel cross-sections, the model showed strong performance in predicting hydrology metrics without calibration: NS ranged from 0.53 to 0.87 and KGE from 0.54 to 0.88 for runoff; NS from 0.59 to 0.85 and KGE from 0.69 to 0.90 for runoff peak; and NS from 0.98 to 0.99 and KGE from 0.94 to 0.98 for time to peak. Sediment yield predictions were particularly accurate in watersheds with significant channel incisions, with NS of 0.65 and KGE of 0.79. Good sediment yield calibration and validation results were achieved in three watersheds, and reasonable results achieved in the smallest watershed. Sediment yield and runoff peak were found to be sensitive to level of watershed discretization. Improved model performance was seen with additional rain gauges even in small watersheds. These findings demonstrate the potential of K2-RHEM as a reliable tool for the prediction of hydrology and erosion for small-scale rangeland watershed management and highlight the importance of both proper watershed discretization and rainfall data resolution in model applications. (c) 2025 International Research and Training Center on Erosion and Sedimentation, China Water & Power Press, and China Institute of Water Resources and Hydropower Research. Publishing services by Elsevier B.V. on behalf of KeAi Communications Co. Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).