基于随机森林的川东北城市群臭氧污染预报研究

Ozone pollution forecasting in northeast Sichuan City Cluster based on random forest

为探究川东北城市群气象因子间的非线性交互作用对近地层臭氧(O_(3))污染的影响,并开展未来1~10天的O_(3)浓度预报,本研究基于2017-2021年川东北城市群O_(3)浓度监测数据、气象数据与欧洲中心细网格模式预报资料,分析近5年川东北O_(3)污染时空特征,并以此为基础选取预报因子,经随机森林(RF)模型筛选关键特征因子作为模型输入数据,构建本地化动态时效RF臭氧预报模型。结果表明:1)2017-2021年川东北O_(3)浓度年评价值呈现“M”型的变化趋势;各月O_(3)超标天数与O_(3)浓度月评价值变化均呈现“双峰型”趋势;O_(3)浓度日变化则呈现“单峰型”趋势。2)特征选择结果显示川东北城市群影响O_(3)浓度变化的关键特征因子以气温、云量、起报天O_(3)浓度、海平面气压等变量为主。3)以川东北城市群整体来看,RF臭氧预报模型误差项处于合理范围内,预报性能良好,对川东北城市群O_(3)浓度有较高的解释性。

To investigate the influence of the non-linear interaction between meteorological factors on near-surface ozone(O_(3))pollution in northeast Sichuan City Cluster,and to enable day-by-day O_(3) concentration forecasting for future 1-10 days.The spatial and temporal characteristics of O_(3) pollution in northeast Sichuan in the past five years were analyzed,based on the O_(3) concentration monitoring data,meteorological data,and Eurocentric fine-grid model forecasts data from 2017 to 2021.On this basis,the localized dynamic time-dependent RF ozone forecasting model was constructed by screening key characteristic factors as model input data through the random forest(RF)model.The results showed that:1)From 2017 to 2021,the annual assessed value of O_(3) concentration in northeast Sichuan showed an"M"trend;the number of exceedance days and the monthly assessed value of O_(3) concentration showed a"double-peak"trend;and the daily variation of O_(3) concentration showed a"single-peak"trend.2)The results of feature selection showed that the key feature factors affecting O_(3) concentration changes in the northeast Sichuan City Cluster were dominated by variables such as air temperature,cloudiness,O_(3) concentration on the starting day,and sea level pressure.3)Taking the northeast Sichuan City Cluster as a whole,the RF ozone forecasting model error term was within a reasonable range,with good forecast performance and high interpretability of the O_(3) concentration.