Spatio-Temporal Bayesian Quantile Regression for High Air-Pollution Concentrations

This short paper summarises ongoing work on quantile-based exposure mapping for urban air pollution using Bayesian quantile regression for daily concentrations in the Rome municipality (Italy) over 2011–2022. We analysed a traffic-related pollutant (nitrogen dioxide, NO2), targeting conditional quantiles on the right tail of the distribution (τ = 0.90) to characterise high-pollution exposures, beyond mean-based approaches. Temporal variability is represented through time-varying covariates and seasonal terms, including meteorological predictors derived from ERA5 reanalysis (ECMWF), while persistent spatial heterogeneity is captured by a Gaussian process providing local adjustments to the global intercept. We perform a validation strategy combining leave-one-year-out and leave-one-site-out schemes. The fitted models are used to generate daily, quantile-specific exposure surfaces on a 1 km grid across the whole municipality: such tools can be useful to support exposure assessment, especially in more polluted areas. This framework provides a practical tool for exposure mapping of air-pollution quantiles from routine monitoring data, with assessment of predictive performance.