Estimation of Ground-Level Fine Particulate Matter Using Remotely Sensed Data Augmenting Air Quality Monitoring Network Data in the Western United States

Recently the western United States has been experiencing an increase in particulate pollution associated with an increase in dust storms and wildfires. Particulate matter is traditionally monitored by ground-based air quality monitoring networks and can provide real time data at different temporal resolutions. However, measurements from these networks possess sparse spatial coverage and uneven distribution of the monitoring sites making the development of continuous spatial estimations of particulate matter concentrations at the regional scale very difficult. Moreover, these ground monitors are expensive and require regular maintenance. Due to their continuous, reliable, comprehensive spatial and temporal coverage, satellite remote sensing can supplement ground-level particulate measurements, though satellite data generally reflect columnar particular matter values. Therefore, the assimilation of satellite measurements of aerosol optical properties with large spatial coverage at high spatial and temporal resolution and multiple ground-based data records could be an ideal approach towards estimating the ground-level matrix of particulate concentration.

Given this hypothesis, this study develops a robust model to estimate ground-level fine particulate matter concentrations by employing ground-based particulate matter point measurements from 18 IMPROVE sites, satellite AOD retrieval from MODIS (Aqua and Terra) and MISR Level 2 products, meteorological data, and land cover change information. Meteorological data includes boundary layer height, air temperature, wind speed, relative humidity, and surface pressure from the Goddard Earth Observing System Forward Processing (GOES FP) product at 0.3125^o longitude by 0.25^o latitude spatial resolution. Normalized difference vegetation index (NDVI) is used as proxy of land cover change with 0.05^o x 0.05^o spatial resolution. We employ the recently emerged artificial neural network (ANN) and geographically weighted regression (GWR) statistical models to merge and assimilate these data sets for improved spatial coverage of particulate matter concentration across the West.