Satellite-based estimation of full-coverage ozone (O3) concentration and health effect assessment across Hainan Island

Abstract Ambient ozone (O3) poses great damage to human health, biodiversity, and climate change. Although the health risk due to O3 pollution has been widely evaluated, the spatiotemporal variation of O3 level and its potential health impact in a remote tropical island was less concerned. In the present study, we firstly employed an efficient machine learning model named random forest (RF) to fill the missing O3 column amount over Hainan Island of China based on the meteorological factors and other geographical covariates. The result suggested that the RF model achieved the best performance (R2 = 0.87, RMSE = 6.36 DU) among all of the filling methods. We further used an extreme gradient boosting (XGBoost) algorithm to estimate the surface O3 level around Hainan Island on the basis of the full-coverage O3 column amount and other predictors. The XGBoost method showed the better predictive performance (R2 = 0.59, RMSE = 24.14 μg/m3) compared with other estimation models. The predicted O3 concentration exhibited remarkably spatial difference with the highest value observed in Dongfang (77.61 ± 5.25 μg/m3), followed by that in Changjiang (72.51 ± 5.22 μg/m3), and the lowest one in Lingshui (64.16 ± 2.71 μg/m3). Based on the estimated O3 level, we found that the annually mean nonattainment days in Dongfang reached 29.26 ± 35.20 days and about one third of days in the winter fell into the severe O3 pollution over Hainan Island. Besides, the mean all-cause mortalities owing to severe O3 pollution over Hainan Island were calculated to be 391 (95%CI: (212–570)) cases each year, accounting for 0.004% of all the population in Hainan Island. The statistical simulation of O3 level at a remote island updated the traditional knowledge and aroused the public concern on island air pollution.