An Improved Geographically and Temporally Weighted Regression for Surface Ozone Estimation From Satellite-Based Precursor Data

It is very essential to resolve the issues of atmospheric ozone pollution and health impact evaluation about high spatial resolution and accurate near-surface ozone concentration. Nevertheless, the existing remotely-sensed ozone products could not meet the demands of high spatial resolution monitoring. For this purpose, this study using surface ozone precursor (the surface nitrogen dioxide concentration and formaldehyde concentration) data developed an improved geographically and temporally weighted regression (IGTWR) method to estimate the surface ozone concentration. This method calculated a generalized distance between sample points in that multidimensional space constructed using the longitude, latitude, day, and normalized difference vegetation index (NDVI). Next, the surface ozone precursor data were as independent variables to retrieve the daily ozone concentrations. The contribution of the proposed model is that the NDVI data was introduced as the underlaying factor to explain the heterogeneity of underlaying conditions and indicated ozone concentration more accurately to improve estimation accuracy. And then the ground station observations were used to validate the estimated ground-level ozone concentration results. Based on the cross-validation results of all test data, the model estimated the root mean squared error (RMSE) and the correlation coefficient (R ² ) of surface ozone are 9.456 µg/m ³ and 0.983, respectively. The results demonstrate that it is feasible to estimate surface ozone concentrations using data from the TROPOMI sensor and an improved geographically weighted regression model.