Adaptive AOD Forecast Model Based on GNSS-Derived PWV and Meteorological Parameters

Abstract

Aerosol optical depth (AOD) is one of the basic parameters for determining the total aerosol content, and it exerts an important impact on regional environment pollution. To investigate the spatiotemporal variations of AOD, this study analyzes the relationship of AOD with precipitable water vapor (PWV) derived from a global navigation satellite system (GNSS) and meteorological parameters and proposes an adaptive AOD forecasting (AAF) model. In this model, the initial AOD value is determined using an empirical AOD model that considers annual periodicity, and the AOD difference is fitted using PWV, temperature ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$T$ </tex-math></inline-formula> ), and surface pressure ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$P$ </tex-math></inline-formula> ). In addition, this model also considers the time autocorrelation of the AOD difference; the model coefficients can be adaptively updated with training data. AOD data at 550 nm derived from the aerosol robotic network (AERONET), second modern-era retrospective analysis for research and applications (MERRA-2), and Copernicus atmosphere monitoring service (CAMS) for the Beijing–Tianjin–Hebei area are utilized to validate the proposed AAF model. Numerical results show that: 1) the accuracy of AOD derived from MERRA-2 is superior to that obtained from CAMS; 2) AOD is negatively correlated with <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$P$ </tex-math></inline-formula> , is positively correlated with PWV and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$T$ </tex-math></inline-formula> , and has a high time autocorrelation with the AOD difference at consecutive times; and 3) the proposed AAF model demonstrates better performance than the traditional multiple linear regression (MLR) model. The average root mean square error (RMSE), mean absolute error (MAE), and bias of the AAF model are 0.17, 0.14, and −0.04, respectively, and those of the MLR model are 0.31, 0.25, and 0.06, respectively. These results reveal that the proposed AAF model can estimate AOD with high precision and has considerable potential for application in AAF research.