Land Use Regression Models: How Effective are they in Predicting Personal Exposure to Air Pollution?

Abstract

Poor air quality is known to be detrimental to human health and studies have shown that data from fixed air quality monitoring sites (FMS) are inadequate in estimating individual exposure to air pollutants. This is primarily because in developed countries, daily exposure to air pollution is largely determined by time spent in the commuter micro-environment which is not well represented by FMS. Pollutant concentrations in urban areas are highly variable in space and time with strong gradients in concentrations observed within, and with distance away from, transport corridors. Land use regression (LUR) models have been shown to be effective tools in interpolating air quality in urban areas and offer the possibility of higher spatial resolution estimates of air quality within transport corridors. Newer generation models are becoming increasingly effective at modelling air quality at local to micro-scales and a range of temporal scales. This study aims to evaluate how effective two of these new generation LUR models, a city scale and a local scale developed for Auckland, are at determining personal exposure to air pollution at a variety of temporal and spatial scales. A set of real-time measurements of personal exposure to nitrogen dioxide (NO2) as well as carbon monoxide (CO) and ultra-fine particles (UFP) were used within this environment to evaluate the given models and compare them to one another. It was found that there was not much difference in ability to predict NO2 concentrations, 73.8% city and 68.8% local predicted correctly, but when using the models as a proxy for UFP and CO, the local model performed significantly better with fewer areas under predicted. Spatial variation of NO2 at the microscale is complex, but the models chosen were able to predict NO2 concentrations within a reasonable margin of error most of the time, with a few exceptions. Being aware of the limitations of these models and LUR in general allows us to come up with new ways of sampling, designing and constructing new models that better represent the micro-environment of interest.