Frontiers in Environmental Science (Oct 2022)
Spatial downscaling of surface ozone concentration calculation from remotely sensed data based on mutual information
Abstract
Accurate near surface ozone concentration calculation with high spatial resolution data is very important to solve the problem of serious ozone pollution and health impact assessment. However, the existing remotely sensed ozone products cannot meet the requirements of high spatial resolution monitoring. In this study, surface O3 concentration (at 30 km spatial resolution) was extracted from the daily TROPOMI O3 profile products. Meanwhile, this study improved the downscaling algorithm based on the mutual information and applied it to the mapping of surface O3 concentration in China. Combined with the surface O3 concentration data (with 5 km spatial resolution) obtained by using the Light Gradient Boosting Machine (LightGBM) algorithm and AOD data (at 1 km resolution) from MODIS, the downscaling of TROPOMI ground O3 concentration data from 30 km to 1 km has been achieved in this study. The downscaled ground O3 concentration data were subsequently validated using an independent ground O3 concentration dataset. The main conclusion of this study is that the mutual information entropy between the bottom layer data of the TROPOMI ozone profile (at 30 km resolution), LightGBM surface O3 concentration data (at 5 km resolution), and MCD19A2 AOD data (at 1 km resolution) can accurately reduce the spatial resolution of ozone concentration in the ground layer. The downscaling procedure not only resulted in increase of the spatial resolution over the whole area but also significant improvements in precision with coefficient of determination (R2) increased from 0.733 to 0.823, mean biased error decreased from 7.905 μg/m3 to 3.887 μg/m3, and root-mean-square error decreased from 14.395 μg/m3 to 8.920 μg/m3 for ground O3 concentration.
Keywords
