Time series modeling of PM2.5 concentrations with residual variance constraint in eastern mainland China during 2013–2017

Abstract

Satellite-based mapping has been proven to be an effective method to reveal the spatiotemporal variations of PM2.5 distributions. However, most satellite AOD (aerosol optical depth) statistical models suffer from unstable accuracy over long time spans. This study thus aims to propose an accurate and stable method for PM2.5 concentration estimations in time series. Specifically, a three-step residual variance constraint method (RVCM) is developed to simulate PM2.5 concentrations from January 2013 to December 2017 with the aid of AODs and other auxiliary data. Results show that the five-year fitting R2 and cross-validation R2 of RVCMs improved from 0.77 to 0.88 and 0.71 to 0.84, respectively, compared to those models without residual variance constraint (WO-RVCM). Additionally, RVCM demonstrated more stable performance on time series simulation of PM2.5 concentrations than WO-RVCM, with the yearly fitting R2 of 0.89, 0.88, 0.85, 0.87 and 0.88, and corresponding cross validation R2 of 0.85, 0.84, 0.80, 0.82 and 0.83, respectively. Furthermore, accuracy verification of removed outliers in residual variance constraint modeling confirmed the credibility of RVCM in outliers' simulation compared to WO-RVCM models. Finally, RVCM-aided estimations of time series PM2.5 concentrations and associated premature deaths in the study area (east and southeast mainland China) revealed their total decrease rates were 35.21% and 21.57%, and excellent air quality days increased from 7% to 35%. These findings suggest that residual variance constraint is effective and could be a reliable solution to providing time series AOD-PM2.5 modeling with stable accuracy over long time spans.