Global navigation satellite system precipitable water vapour combined with other atmospheric factors to predict the short-term change of PM2.5 mass concentration

Abstract

With the rapid development of the global navigation satellite system (GNSS), GNSS-derived precipitable water vapour (PWV) is of great significanc in weather forecasting. The present study reports the effect of monitoring haze (mainly PM2.5) using GNSS PWV data from the Beijing Fangshan station (BJFS, International GNSS Service). However, the correlation between the PWV time series based on data from the BJFS and corresponding PM2.5 mass concentration series data from the Temple of Heaven station in Beijing is low over long periods (for DOYs 1-60, 2016 and DOYs 275- 366 in 2016). In addition, the correlation between the two was low in summer when considering the DOYs 183-244, 2016. We suspect the low correlation is due to the inf uence of meteorological factors, such as the relative humidity, wind speed, pressure, and heavy rainfall. To accurately analyze the relationship between the two, we select a short period with relatively stable weather conditions (e.g., no heavy rainfall and relatively continuous data) during haze periods., which improves the correlation coeff cient between the GNSS PWV and PM2.5 time series to a value of 0.66. To further improve the correlation between PWV and the PM2.5 time series, two datasets were reconstructed by use of the fourth-layer low-frequency coeff cient (a4) with the standard orthogonal wavelet Daubechies method (db5). By removing the inf uence of high-frequency noise with this method, the correlation is improved, with the value of the maximum correlation coeff cient reaching 0.73. Although the real change in PM2.5 mass concentration is affected by other factors (e.g., NO2, SO2, relative humidity), the reconstructed NO2, relative humidity, average wind speed, PWV, O-3, CO and SO2 time series serve as the independent variables, with reconstructed PM2.5 as the dependent variable, to establish a multiple-regression model for the prediction of the change in the PM2.5 mass concentration. This approach is successful in predicting a short-term change in PM2.5 mass concentration based on four statistical values of the regression model. Therefore, the GNSS PWV data combined with other parameters can be used to forecast haze weather with a linear-multiple regression method after wavelet decomposition.