Abstract Evaluating kilometer‐scale atmospheric models in data‐sparse mountains is challenging because in situ meteorological observations are scarce and remote‐sensing products are uncertain. Using hydrological models to link atmospheric model‐simulated precipitation to streamflow is equally problematic, because those models carry substantial structural and parameter uncertainty in mountain terrain. We therefore propose a simple, low‐uncertainty alternative: route atmospheric model runoff through a network routing model calibrated against observed streamflow. We apply this approach to assess thirteen 3‐km Weather Research and Forecasting (WRF) experiments over the Yarlung Zangbo River basin on the Tibetan Plateau, with systematically varied parameterizations for radiation, microphysics, planetary boundary layer, and orographic drag. In mountainous basins, routing can be simplified using the Muskingum model. This parsimonious approach proves effective and robust: hourly streamflow simulations achieve a median Pearson correlation of 0.75 across all six examined gauging stations, and calibrated wave celerity shows little sensitivity to the driving WRF experiment. Statistical analysis confirms that routing model uncertainty is small enough to distinguish performance differences among the WRF configurations. Compared with precipitation‐based evaluation, the routing‐based approach provides complementary and more hydrologically relevant measures of atmospheric model performance, especially when basin‐mean precipitation is already well captured. The method offers a discriminative tool that leverages the superior spatial representativeness of streamflow observations to evaluate atmospheric models in data‐sparse mountainous regions.
Yang et al. (Sun,) studied this question.