Assessing the Impact of Incorporating Bias-Adjusted Satellite Observations into Global Data Fusion for Ground-level Health-Related Ozone Metrics

Abstract

Understanding global ground-level ozone is hindered by significant monitoring gaps. To address this, we introduce a Bayesian Maximum Entropy data fusion framework that includes satellite ozone observations, generating a global 0.1° surface ozone dataset (2005–2022). Satellite observations (IASI+GOME2, OMI/MLS), adjusted to ground-level estimates, were analyzed using Flexible Regionalized Air Quality Model Performance to correct biases and estimate variance. Ground observations (hard data) and satellite-derived estimates (soft data) were then integrated, using a composite of 10 atmospheric chemistry models and 3 reanalyses as the global spatial-temporal trend. Cross-validations demonstrated high overall accuracy (R 2 = 0.83; RMSE = 3.73 ppb). Although integrating satellite-derived soft data yields only marginal performance gains in densely monitored areas, it substantially reduces model uncertainties and improves model accuracy across large spatial gaps. Notably, for the furthest unmonitored gap group at a 20° checkerboard, including satellite data increased R 2 by 25.9% and decreased RMSE by 21.1% after the integration. Applying this optimized data fusion configuration from 2005 to 2022, global population-weighted ozone is estimated to have increased by 0.21 ppb/decade. Consequently, the global population exposed to ozone levels exceeding 50 ppb more than doubled over this 18-year period, rising from 17.2% to 41.0%.