PM2.5 extended-range forecast based on MJO and S2S using LightGBM

Abstract

We developed an extended-range fine particulate matter (PM2.5) prediction model in Shanghai using the light gradient-boosting machine (LightGBM) algorithm based on PM2.5 historical data, meteorological observational data, Subseasonal-to-Seasonal Prediction Project (S2S) forecasts and Madden-Julian Oscillation (MJO) monitoring data. The analysis and prediction results demonstrated that the MJO improved the predictive skill of the extended-range PM2.5 forecast. The MJO indexes, namely, real-time multivariate MJO series 1 (RMM1) and real-time multivariate MJO series 2 (RMM2), ranked the first, and seventh, respectively, in terms of the predictive contribution of all meteorological predictors. When the MJO was not introduced, the correlation coefficients for the forecasts on lead times of 11–40 days ranged from 0.27 to 0.55, and the root mean square errors (RMSEs) ranged from 23.4 to 31.8 μg/m3. After the MJO was introduced, the correlation coefficients for the 11–40 day forecast ranged from 0.31 to 0.56, among which those for the 16–40 day forecast substantially improved, and the RMSEs ranged from 23.2 to 28.7 μg/m3. When comparing the prediction scores, such as percent correct (PC), critical success index (CSI), and equitable threat score (ETS), the forecast model was more accurate when it introduced the MJO. A novel aspect of this study is to investigate the effects of the MJO mechanism on the meteorological conditions of air pollution in eastern China through advanced regression analysis. The MJO indexes RMM1 and RMM2 considerably impacted the geopotential height field of 28°–40° at 300–250 hPa 45 days in advance. When RMM1 increased and RMM2 decreased 45 days in advance, the 500 hPa geopotential height field weakened accordingly, and the bottom of the 500 hPa trough moved southward; thus cold air was more easily transported southward and the upstream air pollutants were transported to eastern China. With a weak ground pressure field and dry air at low altitudes, the westerly wind component increased, which led to the easier formation of a weather configuration favorable for the accumulation and transport of air pollution, thus resulting in an increase in PM2.5 concentration in the region. These findings can guide forecasters regarding the utility of MJO and S2S for subseasonal air pollution outlooks.