This paper describes the use of citizen science-derived data for the creation of a land-use regression (LUR) model for particulate matter (PM2.5 and PMcoarse) for a vulnerable community in Imperial County, California (CA), near the United States (US)/Mexico border. Data from the Imperial County Community Air Monitoring Network community monitors were calibrated and added to a LUR, along with meteorology and land use. PM2.5 and PMcoarse were predicted across the county at the monthly timescale. Model types were compared by cross-validated (CV) R2 and root-mean-square error (RMSE). The Bayesian additive regression trees model (BART) performed the best for both PM2.5 (CV R2 = 0.47, RMSE = 1.5 µg/m3) and PMcoarse (CV R2 = 0.65, RMSE = 8.07 µg/m3). Model predictions were also compared to measurements from the regulatory monitors. RMSE for the monthly models was 3.6 µg/m3 for PM2.5 and 17.7 µg/m3 for PMcoarse. Variable importance measures pointed to seasonality and length of roads as drivers of PM2.5, and seasonality, type of farmland, and length of roads as drivers of PMcoarse. Predicted PM2.5 was elevated near the US/Mexico border and predicted PMcoarse was elevated in the center of Imperial Valley. Both sizes of PM were high near the western edge of the Salton Sea. This analysis provides some of the initial evidence for the utility of citizen science-derived pollution measurements to develop spatial and temporal models which can make estimates of pollution levels throughout vulnerable communities.
