Full-coverage mapping and spatiotemporal variations of ground-level ozone (O3) pollution from 2013 to 2020 across China

Association of short‐term exposure to ambient air pollution with mortality from ischemic and hemorrhagic stroke

“…Daily gridded data (spatial resolution: 10 km × 10 km) on PM 2.5 , PM 10 , SO 2 , NO 2 , CO, and O 3 concentrations were collected from the ChinaHighAirPollutants (CHAP) dataset, which has been known as a long-term, full-coverage, high-resolution, and highquality near-surface air pollutants dataset in China. This dataset was generated from our proposed Space-Time Extra-Trees (STET) model combining with big data including ground measurements, satellite remote sensing products, atmospheric reanalysis, and model simulations [30][31][32][33]. The cross-validation coefficient of determination (R 2 ) for PM 2.5 , PM 10 , SO 2 , NO 2 , CO, and O 3 was 0.91, 0.88, 0.84, 0.84, 0.80, and 0.87, respectively.…”

Cluster‐Enhanced Ensemble Learning for Mapping Global Monthly Surface Ozone From 2003 to 2019

“…Second, we include 89 predictors, which can better handle the complex interactions in surface ozone prediction. For example, previous ML‐based modeling studies mainly focused on remotely sensed total ozone column, emission and meteorology (Liu et al., 2020; Ma et al., 2021; Wei et al., 2022), which may not be sufficient to model surface ozone, due to complex and nonlinear processes in ozone formation. Our model incorporates additional predictors such as satellite tropospheric ozone, atmospheric chemistry reanalysis products, and land cover to capture known processes that generate surface ozone variability.…”

“…The importance of variables also confirms our selection that previously unaccounted variables are of high rank (e.g., CAMS ozone concentration). Third, gap‐free extrapolation of ozone by machine learning is usually applied in a specific region, such as the United States (Requia et al., 2020) and China (Liu et al., 2020; Ma et al., 2021; Wei et al., 2022). Other regions suffering from high ozone pollution are less frequently modeled, such as India and Europe.…”

“…Generally, products that use chemical transport models suffer from a high computational burden, unresolved or oversimplified physical and chemical mechanisms, and uncertainty in emission inventories. The advantages of statistical models and geostatistical data fusion are their rapidity and interpretability, but their performance is not comparable to machine learning methods (Wei et al., 2022). In recent years, ML has gained popularity as it can capture complex and nonlinear relationships among environmental features and satellite data (Yuan et al., 2020).…”

“…Multiple methods can be used to estimate gridded surface ozone pollution but trade‐offs are inherent in each. There are primary four methods in the literature: (a) regional and global chemical transport models (e.g., Astitha et al., 2017; Guo et al., 2016; Liu et al., 2021, and others), (b) statistical models (e.g., Eeftens et al., 2012; Kerckhoffs et al., 2015, and others), (c) geostatistical data fusion (e.g., Chen et al., 2021; DeLang et al., 2021; Xu et al., 2016, and others), and (d) machine learning models (e.g., Liu et al., 2020; Ma et al., 2021; Requia et al., 2020; Wei et al., 2022, and others). Generally, products that use chemical transport models suffer from a high computational burden, unresolved or oversimplified physical and chemical mechanisms, and uncertainty in emission inventories.…”