Generalized Additive Model With Dynamic Coefficients for Spatiotemporal Ozone Predictions

Accurate prediction of surface‐level ozone concentrations is critical for air quality management and public health protection. This study develops a flexible spatiotemporal statistical modeling framework to predict daily mean O3 concentrations across Italy by integrating satellite‐derived ozone estimates with ground‐based observations and high‐resolution environmental predictors. The proposed model is based on a linear regression with dynamic intercept and slope that relate in situ O3 measurements to satellite data, explicitly addressing additive (systematic shifts) and multiplicative (scaling) biases in satellite‐derived ozone estimates. These spatiotemporally varying coefficients are modeled through a generalized additive model framework, allowing the capture of complex and potentially nonlinear relationships between ozone levels and environmental covariates. This unified and interpretable approach enables a detailed understanding of bias patterns in satellite data. Model diagnostics and crossvalidation demonstrate superior explanatory power and predictive performance compared to simpler models. The interpretability of the model is illustrated by revealing the influence of elevation, nitrogen dioxide concentrations, and seasonal variation on bias structures. Furthermore, the model's downscaling capability is demonstrated by producing fine‐scale ozone concentration predictions over Italy and its surrounding regions. The proposed modeling framework offers an accurate, scalable, and interpretable tool for mapping surface‐level ozone, supporting improved environmental monitoring and informing policy decisions.