Multiple regression analysis of winter visibility，PM_2.5 concentration and humidity in Beijing-Tianjin-Hebei and its surrounding regions

Since 2013, low visibility events have been repeatedly observed in Beijing-Tianjin-Hebei and its surrounding regions. PM_2.5 concentration and humidity are considered to be key factors leading to low visibility. Using surface meteorological data from MICAPS and PM_2.5 concentration observation data from the China Environmental Monitoring Center, the influences of PM_2.5 and humidity on visibility under different relative humidity (RH) and pollution levels are investigated. According to the differences in geographical environment and pollution degree, the study region was divided into Beijing-Tianjin and Hebei-Shandong regions. The multiple regression equations of visibility, PM_2.5 concentration, temperature and dew point temperature are established based on data of January 2017, and these equations are tested using the data of January 2015, 2016, 2018 and 2019. Results show that when RH＜70% and PM_2.5 concentration＜75 μg/m³, the visibility in Beijing-Tianjin region and Hebei-Shandong region is usually higher than 10 km. The increase in PM_2.5 concentration is the dominant factor for the rapid decrease in visibility. The combination of increase in RH (70%—85%) and increase in PM_2.5 concentration (75—200 μg/m³) can result in further decrease of visibility (10—5 km). The decrease in visibility (5—2 km) is mostly depended on further increase in RH (85%—95%), while the correlation between PM_2.5 concentration and visibility becomes weaker in this situation. The decrease in visibility to 2 km or even lower is mainly due to the extinction of droplets under the near saturation of water vapor (RH＞95%), and has little relation with changes in PM_2.5 concentration. Compared with establishing the visibility fitting equation directly without grouping, establishing the visibility fitting equation according to the RH above or below 85% respectively can greatly optimize the multivariate regression models. The RMSEs for visibility fittings with RH＞85% decreases from 9.2 and 5.2 km to 0.5 and 0.7 km. The visibility in January of 2015, 2016, 2018 and 2019 are well reproduced by these fitting models. Correlation coefficients between the observed visibility and the calculated visibility all are higher than 0.91. This study provides a new visibility parameterization for the haze-fog numerical prediction system.