WRF-CMAQ-MOS studies based on extremely randomized trees

With the rapid development of urbanization and industrialization, air pollution has become one of the problems which are most concerned with. In this paper, the CMAQ (Community Multi-Scale Air Quality) model is combined with the WRF (Weather Research and Forecast Model) forecast of meteorological data, observed meteorological data and pollutants concentration data to build the WRF-CMAQ-MOS statistical correction model based on the method of Extremely Randomized Trees. The output of CMAQ model like AQI and pollutants concentrations of PM_2.5, PM₁₀, NO₂, SO₂, O₃ and CO etc. are optimized. In the experiment that combines WRF meteorological elements with the CMAQ-MOS model, it is found that the linear regression method cannot well reflect the effect of optimization. Thereby the Extremely Randomized Trees method and Gradient Boosted Regression Trees were selected to improve the model performance. The investigations show that the Extremely Randomized Trees method has greatly improved the performance of CMAQ-MOS model combined with WRF meteorological elements. Finally, the model and method are evaluated using the data of January, February and March 2016. The results show that the most obvious optimization effect can be found in two pollutants, i.e., NO₂ and O₃. The correlation coefficient of NO₂ in January, February and March 2016 is increased from 0.3542 to 0.6274, the correlation coefficient of O₃ is increased from 0.3883 to 0.7886, the root-mean-square error of NO₂ is decreased from 0.0346 to 0.0243 mg/m³, and the root-mean-square error of O₃ is decreased from 0.0447 to 0.0367 mg/m³. The WRF-CMAQ-MOS statistical correction model developed in this study can effectively improve the prediction accuracy of various pollutants, and thus has a great application prospect in operational air quality prediction.