Small bias, large errors: a dual-perspective evaluation of numerical model near-surface wind direction over beijing megacity

Abstract

Accurate surface wind fields are essential for understanding moisture transport and heavy rainfall forecasts, yet the ability of kilometer-scale numerical models to capture wind direction over heterogeneous surfaces remains poorly quantified. Using the EARS1km reanalysis for the summers of 2022–2023 and hourly observations from 134 surface stations across Beijing and its surroundings, we diagnose wind-direction error from two perspectives: weak-wind amplification and land-surface heterogeneity. Overall, EARS1km reproduces temperature, humidity, and wind speed with high fidelity, whereas wind direction exhibits a distinctive “Small Bias, Large Errors” pattern in which opposing deviations largely cancel in the mean while MAE and RMSE remain substantial. Weak-wind conditions define the primary high-sensitivity regime for wind-direction error, while land-surface heterogeneity at the 5–10 km scale further modulates its spatial distribution. The 3 × 3 classification also shows that the heterogeneity association varies across weak-wind-frequency categories. These findings provide a reference for assessing high-resolution reanalysis products and refining land-surface parameterizations in numerical models.