On the AOD contribution to ground PM prediction models

Abstract

Introduction. The use of satellite observations for estimating ground level particulate matter (PM) concentrations is growing. Specifically, numerous researchers use satellite-based Aerosol Optical Depth (AOD) as an explanatory variable in different statistical models for estimating PM concentrations. Whereas some models use only the AOD, others use additional variables, such as meteorological, environmental, and land use variables. AOD measures light extinction at a given wavelength due to scattering and absorption along the measured atmospheric column. The rationale behind using the AOD stems from the physical interaction of airborne PM with electromagnetic radiation, and since it has a much better spatial coverage than ground PM measurements. In many works AOD is the primary variable in spite of the coarse temporal resolution of the MODIS-based AOD (once per day) and the large number of missing data due to clouds. Methods. This work assesses the actual contribution of AOD to PM estimations by analyzing (a) a large database of fine and coarse PM concentrations from different geographic locations, (b) satellite-borne AOD obtained with different spatial grid by applying different algorithms on MODIS raw data, and (c) applying various statistical models to link these databases. Results. We found that the date, as a variable, has a crucial effect on the performance of all the models, including mixed effects models (MEM) that consistently result in better predictions. In fact, the seemingly higher explained spatiotemporal variability of MEMs results from the model fitting the temporal PM patterns, for which the date is a strong predictor. Surprisingly, the overall AOD contribution to the explained variability is little. Conclusion. While AOD has a lot of potential for PM estimation in locations where monitoring is very sparse or not existing at all, currently its actual contribution to predicted PM levels is marginal, and should be further studied.