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

Abstract

Surface ozone is damaging to human health and crop yields. When evaluating global air pollution risk, gridded datasets with high accuracy are desired to reflect the local variations in air pollution concentrations. Here, a cluster‐enhanced ensemble machine learning method was used to develop a new 0.5‐degree monthly surface ozone data set during 2003–2019 by combining numerous informative variables. The overall accuracy of our data set is 91.5% (90.8% for space and 92.3% for time). Historically, populations in South Asia, North Africa and Middle‐East, and High‐income North America are exposed to the highest ozone concentrations. Globally, the population weighted ozone concentration in the peak season is 47.07 ppbv. Our results highlight that ozone pollution is intensifying in some regions, and implicate air quality management is crucial to secure human health from air pollution.

Future Local Ground‐Level Ozone in the European Area From Statistical Downscaling Projections Considering Climate and Emission Changes

“…According to Liu et al. (2022) methods to obtain O 3 estimates can be divided into regional and global chemical transport models, statistical models, geostatistical data fusion, and machine learning models. While dynamical climate chemistry models are a useful tool to assess future O 3 , they come with the drawback of being computationally expensive and thus the number of model runs is limited.…”

“…While dynamical climate chemistry models are a useful tool to assess future O 3 , they come with the drawback of being computationally expensive and thus the number of model runs is limited. Machine learning models have a very good performance, but they are generally characterized by low interpretability (Liu et al., 2022). Statistical models, which are very flexible, computationally inexpensive, and easy to interpret, have so far used meteorological information to assess future O 3 concentrations, neglecting information about the precursor emissions.…”

“…In this regard, knowledge about local future ground-level O 3 concentrations is of particular interest to protect human health and to design corresponding mitigation and adaptation strategies. According to Liu et al (2022) methods to obtain O 3 estimates can be divided into regional and global chemical transport models, statistical models, geostatistical data fusion, and machine learning models. While dynamical climate chemistry models are a useful tool to assess future O 3 , they come with the drawback of being computationally expensive and thus the number of model runs is limited.…”

“…The present study contributes to the health-relevant topic of O 3 air pollution under climate and emission changes by assessing local-scale daily MDA8/MDA1 changes using the latest data and statistical methodology. In this regard, state-of-the-art ESM fusion methods (e.g., Liu et al, 2022;Sun & Archibald, 2021) are promising approaches to further improve the methodology presented in this study. Besides, for health impact assessments and public health and policy interventions, it is important to provide further information about changes in health-relevant thresholds such as the MDA8 value of 100 μg/m 3 for short-term O 3 exposure and of 60 μg/m 3 in the peak season (highest average in six consecutive months) for long-term exposure (World Health Organization, 2021).…”

Estimating birthweight reduction attributable to maternal ozone exposure in low- and middle-income countries

“…We obtained O 3 data from a global dataset of monthly surface O 3 concentrations with a more complete spatiotemporal coverage ( 30 ); this O 3 dataset was constructed by applying ensemble machine learning techniques to satellite observations, chemical transport model outputs, atmospheric reanalysis and emission data, and ground-surface observations. The monthly mean daily maximum 8-hour average (MDA8) O 3 concentrations during 2003–2019 were calculated across a regular 0.5° × 0.5° (≈50 km × 50 km) grid.…”