Sensitivity Experiments of Meteorological Parameterization Schemes for WRF Model during a Heavy Air Pollution Episode in Beijing

Meteorological forecasting is an important factor affecting the accuracy of atmospheric heavy pollution prediction. In response to a heavy pollution event in Beijing during 16−21 December 2016, this paper carried out a sensitivity test for the parameterization scheme of a mesoscale meteorological model Weather Research and Forecasting (WRF). Combining microphysical, long-wave radiation, short-wave radiation, land surface, boundary layer, near-surface, and cumulus convective parameterization processes, a total of 51 sets of parameterization schemes were designed to analyze the simulation accuracy and sensitivity of the temperature, relative humidity, and 10-m height wind speed of eight meteorological stations in Beijing under different simulation schemes. The temperature simulation is the most sensitive to a long-wave process parameterization scheme, the set dispersion is 2.4–7.4°C, followed by the short-wave process parameterization scheme. Additionally, the relative humidity simulation is the most sensitive to the long-wave process parameterization scheme, followed by the land surface process and the wind speed simulation had little difference in sensitivity to different process parameterization schemes. In the comparison of the statistical results of the simulation results with observations, we prefer the combination of the smallest simulation error: Lin microphysical, RRTMG long-wave, RRTMG short-wave, Tiedtke cumulus convection, Noah land surface, MYNN 3rd boundary layer and MYNN near-surface scheme, and compared the best scheme to the ensemble mean and baseline scheme. For the ensemble mean, the correlation coefficient between the temperature simulation and observation was 0.69, which is greater than the baseline scheme. The simulated deviation and root-mean-square error were 25% and 11% less than the baseline scheme and the ensemble mean relative humidity and wind speed simulation were less variable than the baseline scheme. Compared with the ensemble mean, the best scheme can simultaneously improve the temperature, relative humidity, and wind speed simulation, such that the temperature simulation deviation and root-mean-square error decreases by 35% and 17% compared with the baseline scheme, the relative humidity simulation deviation and root-mean-square error decreases by 43% and 13%, and the wind speed simulation deviation and root-mean-square error decreases by 33% and 24%. The above results show that the sensitivity test and optimization of the parameterization scheme can significantly reduce the simulation error of meteorological elements during heavy pollution. The improvement of heavy pollution prediction needs to focus on the uncertainty of the parametric scheme simulation. Additionally, the MYNN 3rd boundary layer scheme has good performance in the simulation of meteorological elements in this heavy pollution process, which can provide reference for future improvements of heavy pollution forecasting.