Estimating Daily PM2.5 Concentrations in New York City at 100-m Resolution Using MAIAC AOD: Lessons Learned on Integrating Nonregulatory Measurements in Exposure Models

Previous studies on the health impact of fine particulate matter (PM_2.5) have mainly relied on the data from high-quality but relatively sparse regulatory monitors to assess human exposure levels. In recent years, satellite remote sensing data such as MAIAC AOD calibrated by ground observations have been increasingly used to estimate PM_2.5 concentrations in many regions of the world. However, the spatial resolution of satellite model predictions (e.g., 1 km ~ 10 km) is often insufficient to support intra-urban PM_2.5 exposure assessment. The introduction of various low-cost PM_2.5 sensors presents both opportunities and challenges to public health researchers and air quality managers. In this study, we evaluated the impact of integrating a dense non-regulatory PM_2.5 sampling network into a satellite-based daily PM_2.5 exposure model at 100 m resolution in New York City (NYC) in 2015. Two separate machine learning models were developed, one with only the PM_2.5 data from the US EPA, and the other with measurements from both EPA and the New York City Community Air Survey (NYCCAS). The EPA-only model obtained a cross-validation (CV) R² of 0.85, with a root mean square error (RMSE) of 1.98 µg/m3. The EPA+NYCCAS model obtained a CV R² of 0.73, with a RMSE of 2.35 µg/m3. The annual mean PM_2.5 predictions in NYC from EPA model and EPA+NYCCAS model were 8.39 µg/m3 and 8.81 µg/m3, respectively. The EPA+NYCCAS model predicted distinctly different spatial patterns of PM_2.5 concentrations compared with EPA model with more pollution hot spots identified (Figure below). Along the major roads and in densely populated areas, the EPA+NYCCAS model predictions were more than 15% higher than the EPA model, whereas they were comparable in suburban areas, parks and forested areas. Our results showed that satellite AOD data together with non-regulatory low-cost PM_2.5 samplers can be fused together to estimate intra-urban PM_2.5 levels at high spatial-temporal resolution.