Constraining Air Pollutant Emissions Using Surface Observations

Surface monitoring observations reflect the combined effects of emissions, atmospheric transport, physicochemical transformations, and removal processes. Inferring emissions from these measurements therefore requires a framework that connects source contributions to observations. We address this inverse problem using a Bayesian inverse modeling approach that combines surface measurements, prior emission information, and source-tagged Community Multiscale Air Quality (CMAQ) simulations. Source-tagged concentration fields form a source–receptor matrix linking sectoral emissions to measurements at specific locations and times. For emission scaling, a log-space formulation maintains positive scaling factors and represents proportional adjustments and multiplicative model–observation errors. The Bayesian cost function balances departures from prior estimates against concentration mismatches, incorporating both prior and mismatch error covariances. Minimization yields the maximum a posteriori estimate, while local sensitivities support a Gaussian approximation of posterior uncertainty. Applications to elemental carbon (EC), airborne metals, and polycyclic aromatic hydrocarbons (PAHs) in California illustrate how the framework can refine emission estimates and source-specific chemical profiles while evaluating the constraints provided by sparse monitoring networks. Beyond parameter estimation, averaging kernels and degrees of freedom for signal characterize which emission factors, or combinations of factors, the observations can constrain. Recalculating these diagnostics for candidate monitoring locations provides a quantitative basis for selecting additional sites according to their expected information gain. The methodology connects emission improvement, uncertainty quantification, source attribution, and monitoring network design within a common inference framework.