Predicting regional space-time variation of PM2.5 with land-use regression model and MODIS data

Existing land-use regression (LUR) models use land use/cover, population, and traffic information to predict long-term intra-urban variation of air pollution. These models are limited to explaining spatial variation of air pollutants, and few of them are capable of addressing temporal variability. This article proposes a space-time LUR model at a regional scale by incorporating aerosol optical depth (AOD) data from the Moderate Resolution Imaging Spectroradiometer (MODIS).

A multivariate regression model was established to predict the distribution of particle matters less than 2.5 μm in aerodynamic diameter (PM(2.5)) in Florida, USA. Monthly PM(2.5) averages at 34 monitoring sites in the year 2005 were used as the dependent variable, while independent variables include land-use patterns, population, traffic, and topographic characteristics. In addition, a monthly AOD variable derived from the MODIS data was integrated into the regression as a space-time predictor. Cross-validation procedures were conducted to validate this AOD-enhanced LUR model.

The final regression model yields a coefficient of determination (R (2)) of 0.63, which is comparable to other studies that employ aerodynamic/meteorological models. The cross validation indicated a good agreement between the observed and predicted PM(2.5) with a mean residual of 0.02 μg/m(3). The distance to heavy-traffic roads is negatively associated with the concentrations of PM(2.5), while agricultural land use is positively correlated. PM(2.5) tends to concentrate in high-latitude areas of Florida and during summer/fall seasons. The monthly AOD has a significant contribution to explaining the variation of PM(2.5) and remarkably enhances the model performance.

This research is the first attempt to improve current LUR models by integrating remote sensing technologies. The integrative model approach offers an effective means to estimate air pollution over time and space, and could be an alternative to the classic meteorological approach. The model results would provide adequate measurements for epidemiological studies, particularly for chronic health effects in large populations.