Abstract

Particulate matter (PM) is a major air pollutant of public health concern. It represents a complex mixture of chemically and physically diverse substances, from both primary, secondary, anthropogenic and natural sources. Identifying ambient PM sources is key for developing strategies to reduce PM through targeted actions. Broadly speaking, source apportionment (SA) refers to a class of methods aimed at partitioning pollution to the sources from which it was emitted. SA has been an active area of research in the last thirty years, but there is still no consensus on the best methodology.
In this work, we consider long-term time series of particle number size distribution (PNSD) from an urban background station in London, and we develop a novel Bayesian receptor model allowing for a data driven selection of the number of sources and explicitly accounting for nonnegativity constraints on the source contributions and source compositions, while including in the model an autoregressive component to account for temporal correlation in the data. Then we evaluate the effect of the estimated sources on respiratory hospital admissions in children, propagating the uncertainty, from the first stage (sources estimation) to the second stage (health effect model).