Optimization of parameters in the surface-layer scheme of the Weather Research and Forecasting (WRF) model for near-surface wind and temperature

Accurate simulations of surface wind and temperature are essential for boundary-layer, air-quality, and hazard-dispersion applications, yet the Weather Research and Forecasting (WRF) model exhibits persistent biases that are difficult to eliminate through model development alone. Here we optimize parameters in the surface-layer scheme of WRF to improve surface wind and temperature over Korea. We conduct a perturbed-parameter ensemble (PPE; 128 members) generated by Latin Hypercube Sampling of eight parameters. The PPE reveals large spreads in both magnitude and diurnal evolution of surface wind and temperature, confirming sensitivity to the selected parameters. However, the PPE shows that improvement is not guaranteed by single-objective optimization, as wind and temperature root-mean-square error (RMSE) are largely decoupled, RMSE and correlation coefficient (CORR) show weak correspondence, and daytime and nighttime performance can diverge. Using a generalized linear model within a Bayesian optimization framework, we derive optimal parameter sets for different targets: wind-only, temperature-only, and joint optimization using RMSE or CORR, evaluated separately for daytime, nighttime, and 24-h time periods. The optimized parameter sets improve the targeted metrics for targeted variables and time periods. The most robust overall result is achieved by the optimization targeting multi-variable, CORR, and 24-h time windows. The dominant controls are roughness-length scaling and stability-related parameters, which modify bulk transfer coefficients in a stability-dependent manner. Our results emphasize the need for multi-objective, regime-aware calibration of the parameters and motivate future development of dynamic (condition-dependent) parameters in surface-layer schemes.