Mapping yearly fine resolution global surface ozone through Regionalized Air Quality Model Performance corrections and Bayesian Maximum Entropy data fusion of observations and model output for 1990–2017

Abstract:

Bayesian Maximum Entropy (BME) framework to integrate observations from 7,269 globally distributed surface monitoring sites from the Tropospheric Ozone Assessment Report and 1,565 sites from the Chinese National Environmental Monitoring Center Network, with nine global atmospheric chemistry models. Using the M³Fusion method, a multi-model composite is created by weighting models based on their ability to predict observations in each region and year. We then further refine the multi-model composite using a non-homogenous, non-linear, non-homoscedastic Regionalized Air Quality Model Performance (RAMP) bias correction. RAMP allows us to evaluate and correct the model composite regionally by basing corrections on the nearest observations and to capture local trends by not assuming model performance is homogenous by region. BME estimates match observations at each monitoring site with the observational influence decreasing across space and time until the output matches the multi-model or RAMP composite. The method successfully incorporates observations, as quantified by an R² increase from 0.28 when using only the M³Fusion composite to 0.81 after BME data fusion, and 0.83 when using the RAMP corrected composite as the global background. After estimating at 0.5° resolution using BME, we add fine spatial detail from an additional model at 0.125° resolution. Our final product estimates annual global surface ozone for each year between 1990 and 2017 at 0.1° resolution, and results not using the RAMP correction were delivered to support the GBD 2019 study. Global ozone exposure is estimated to be increasing over this period, driven by highly populated and polluted regions of Asia and Africa, despite decreases in the United States and Russia.