Authors: Qiang Pu*, University at Buffalo, SUNY, Eun-Hye Yoo, University at Buffalo, SUNY
Topics: Environment, Geographic Information Science and Systems, China
Keywords: Spatio-temporal Modeling, PM2.5, Satellite AOD
Session Type: Paper
Start / End Time: 8:00 AM / 9:40 AM
Room: Roosevelt 1, Marriott, Exhibition Level
Presentation File: No File Uploaded
Satellite retrieved aerosol optical depth (AOD) has been widely used as a proxy to estimate spatiotemporally resolved PM2.5 concentrations. However, due to cloud/snow cover and high surface reflectance, satellite AOD often suffers from missing data issue which makes the satellite-based PM2.5 estimates challenging. Using the capital region of China as an example, we developed a rigorous Two-Stage spatial temporal PM2.5 estimation model to estimate daily PM2.5 concentrations with full spatial coverage at 3 km resolution. A first stage linear mixed effect model was built to model the space-time relationships between PM2.5 and AOD and predicted PM2.5 at AOD-retrieved areas while inverse probability weighting was applied to account for possible sampling bias caused by non-random missingness in AOD. A second stage of Stochastic Partial Differential Equations approach using integrated nested Laplace approximation were applied using predicted PM2.5 from first stage and to fill the gap in daily PM2.5 predictions when AOD is not available. Results show that the PM2.5-AOD relationships were well modeled with cross validated R2 of 0.88 for year of 2016. In addition, the estimated PM2.5 with full coverage from our model yielded a very high prediction accuracy with out-of-sample validation R2 of 0.93 and RMSE of 9.64 μg/m3. Comparing with previous studies conducted in same region, our proposed two-stage spatio-temporal model performed better with higher accuracy, higher coverage and capability to quantify prediction uncertainty. This method can provide reliable regional PM2.5 predictions which can be further utilized in short-term and long-term exposures and health effects assessment of PM2.5.