Ground-level ozone estimation based on geo-intelligent machine learning by fusing in-situ observations, remote sensing data, and model simulation data

In recent years, near-surface ozone (O-3) pollution has been increasing, seriously endangering both the ecological environment and human health. Accurately monitoring spatially continuous surface O-3 is still difficult with only remote sensing observations. In this paper, to address this issue, we propose a method for estimating surface O-3 by fusing multi-source data, including in-situ observations, O-3 precursors obtained by remote sensing, and model simulation data, including O-3 profile data and reanalysis products of meteorological and radiative elements. The estimation method is geo-intelligent light gradient boosting (Geoi-LGB) which takes into account both the spatial and temporal geographical correlation based on the standard LGB model. The spatio-temporal autocorrelation factors of the site observations are also constructed and added into the input variables. In a case study of China, centered on North China in 2019, the Geoi-LGB method obtained a root-mean-square error of 10.25 mu g/m(3), a mean absolute error of 7.30 mu g/m(3), and a coefficient of determination of 0.912 under the site-based cross-validation strategy. The proposed method has the advantages of being able to obtain a higher accuracy than some of the popular O-3 estimation models. Furthermore, the excellent spatial mapping ability of the Geoi-LGB method was demonstrated, in that about 85 % of the sites had an annual average absolute error of less than 10 mu g/m(3). We believe that this study could provide some important reference information for the accurate estimation of ground-level O-3.