PM2.5 poses a serious threat to public health, yet it is poorly managed due to low spatial coverage of regulatory air quality monitoring (AQM) stations. This motivates novel low-cost methods to infer ground-level PM2.5 at a finer spatial resolution to better quantify PM2.5 exposure in epidemiological research. Most existing modeling approaches exploit aerosol optical depth (AOD) satellite information that is subject to large uncertainties, require extensive land-use variables as inputs, and predict PM2.5 at a coarse 1 km resolution. These limitations have motivated us to devise an algorithm for inferring ground-level PM2.5 by directly processing global-coverage, daily, real-time, 3 m resolution, 3-band micro-satellite imagery of sizes smaller than 1 km×1 km. In this study, we used a deep convolutional neural network (CNN) to process the imagery by extracting image features that characterize the dynamic changes in the built environment and the image colors reflecting AOD, and a random forest (RF) to predict PM2.5 from the extracted image features along with meteorological conditions. We validated our model on 35 AQM stations in Beijing from 2017 to 2019. We trained our CNN-RF model on 10,400 daily images of the stations labeled with the ground-truth PM2.5 and evaluated the model on 2,622 holdout images. Our model predicts PM2.5 accurately at a 200 m resolution with a mean absolute error (MAE) and a root mean squared error (RMSE) as low as 10.1 μg m-3 (23.7% error) and 17.6 μg m-3(41.2% error), respectively, and an R2 up to 0.83. Compared to current state-of-the-art approaches, our model’s errors are comparable at a much finer spatial resolution and are much simpler to implement as it only requires satellite imagery and meteorological inputs. Our results highlight the potential of augmenting limited spatial predictors of PM2.5 with satellite imagery to enhance spatial resolution that is useful for a wide range of applications.

Estimating ground-level PM2.5 using micro-satellite images by a convolutional neural network and random forest approach

Macromolecular Chemistry: The Second Century

[Author Benefits](https://scimeetings.acs.org/?utm_source=pubs_content_marketing&utm_medium=website&utm_campaign=0320_MCF_NPI_Launch_Spring_Homepage&ref=pubs_content_marketing)

[How to use SciMeetings](https://storage.googleapis.com/dl.morressier.com/acs/SciMeetings-ACSSpring2020-HowToGuide_M3020-Final.pdf)

[Frequently Asked Questions](https://storage.googleapis.com/dl.morressier.com/acs/SciMtng_ACSNMSp2020_FAQ_M3120_Final.pdf)

ACS Spring 2020 National Meeting & Expo

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.