Formulation and Implementation of a Dynamic Roughness Length for Improved Surface Wind Speeds in WRF

Abstract

Accurate near-surface winds from the Weather Research and Forecasting (WRF) model are essential for wind-energy applications, yet persistent biases remain that depend on terrain, season, and land cover. This study develops and evaluates an observation-informed, time-varying aerodynamic roughness length for WRF. The roughness length is inferred from the neutral log-law using measured 10-m winds and friction velocity from control WRF runs. A random-forest model, trained on local meteorology is then used to produce a continuous 3-hourly dynamic roughness length series. The dynamic roughness length is implemented in WRF by overwriting its default values each cycle while retaining the local land-surface properties. The framework is tested at three Texas environments: complex topography, homogeneous cropland, and a coastal site during January and July 2021. Across sites, the inferred roughness length frequently exceeds WRF's default bounds, especially in summer, indicating under-representation of effective surface drag in default tables. With the dynamic roughness length implemented into WRF, model performance improves across sites: mean-bias magnitude decreases, point-by-point scatter tightens, explained variance increases, and high-wind events are captured with smaller errors. These gains appear alongside frequent exceedance of the static WRF default bounds, indicating that observation-informed roughness more faithfully represents effective surface drag. Overall, allowing roughness length to vary in time, anchored by local measurements—provides a practical and physically motivated lever to reduce near-surface wind errors in WRF and highlights limitations of fixed, class-based roughness values.