Research On O₃ And NO₂ Prediction For Typical Cities In Sichuan-Tibet Region Based On Machine Learning And Mesoscale Mode

The Tibetan Plateau（TP）is usually regarded as a relatively primitive area for monitoring the atmospheric background.In recent years,with the development of Chinese economy and the acceleration of urbanization,some cites on TP are also experiencing urbanization.In addition,the atmospheric circulation carrying air pollutants from South and Central Asia to the plateau has a great impact on its ecological balance.Therefore,it is very important to study the prediction of air pollutant concentration in TP to prevent the air pollution events in the plateau.In this study,three typical cities（Chengdu,Lhasa and Qamdo）in TP and its surrounding areas were selected as the research objects,and the changes of each city at different time scales（year,month,day and week）were analyzed.The air pollutant concentration monitoring data,GFS and time characteristic quantity of each city from 2017 to 2021 were used as predictors.Then,the air pollutant prediction models based on SVM,Lasso,RF and XGBoost were established and evaluated.And the optimal model（XGBoost）is selected to predicted air pollutants in each city.Finally,the XGBoost（ML）and WRF-Chem models were combined（ML＿WRF-Chem）to predict O₃ and NO₂ in each city in the next three days.The main conclusions are as follows:（1）Based on four machine learning algorithms（SVM,Lasso,RF and XGBoost）,O₃ and NO₂ in TP and its surrounding areas were predicted,and it was found that the XGBoost model performed best.The order of the four machine learning algorithms for NO₂ and O₃simulation in each city is:XGBoost>RF>SVM>Lasso.（2）Based on the XGBoost model,the six-hour forecast of O₃ and NO₂ in some cities of the TP and its surrounding areas in the next three days is realized.Among them,the correlation and the measured between the predicted value of O₃ concentration and the measured value in Chengdu in the next three days reaches 0.9,and the correlation coefficient between the predicted value of O₃ in the first day of Qamdo and Lhasa reaches 0.8,which is better than the WRF-Chem model.In addition,the predicted values of NO₂ in each city based on the XGBoost model are closer to the measured values than the WRF-Chem prediction results.（3）The fusion model（ML＿WRF-Chem）was used to predict O₃ and NO₂ for the next three days.The correlation coefficients between the predicted and measured values of O₃ and NO₂ in Chengdu in the next three days by machine learning model（ML）and fusion model are 0.9 and 0.69,respectively.The forecast results of NO₂ concentration in Lhasa and Qamdo in the next three days by ML＿WRF-Chem model are closer to the measured values than ML model.