A Random Forest approach for predicting temporal-spatial exposures of fine particulate matter with calibrated low-cost sensor networks and GIS predictors.

Authors: Philip Orlando*, , Linda George, Portland State University, Vivek Shandas, Portland State University, Meenakshi Rao, Portland State University
Topics: Environmental Science, Spatial Analysis & Modeling, Geography and Urban Health
Keywords: urban air pollution low-cost sensor networks, random forest, fine particulate matter, land-use regression, urban canopy
Session Type: Poster
Day: 4/5/2019
Start / End Time: 8:00 AM / 9:40 AM
Room: Lincoln 2, Marriott, Exhibition Level
Presentation File: No File Uploaded

A low-cost air quality sensor network was deployed in Portland, Oregon to assess the temporal-spatial variability of fine particulate matter (PM2.5) between July 2017 and December 2018. Sensor-specific correction factors were determined from field evaluations comparing each node to a centrally located regulatory monitoring station maintained by the Oregon Department of Environmental Quality. Strong linear relationships (R2 > 0.90) were observed throughout the network during wildfire haze events each August due to the spatially homogeneous PM2.5 levels from this regional source. The temporal variation of these haze events also provided a wide enough range in PM2.5 concentrations (0-150 μgm-3) necessary to perform meaningful regressions against the reference instrument. Overestimations between 1.5 to 2.5 times were observed by the low-cost sensors before correction factors were applied. A Random Forest model was trained using these corrected PM2.5 data as a response matrix, in addition to over 200 land-use, meteorological, and geographical covariates. Relying on a 70/30 holdout method, an adjusted R2 of 0.88 was observed when comparing the validation dataset with model predictions. This model is currently being used to generate mean daily PM2.5 exposure surfaces at a 1 km2 resolution throughout the Portland metropolitan area. Ultimately, these PM2.5 surfaces, in addition to other environmental stressors including Nitrogen Dioxide (NO2) and urban heat (oC), will assess the cumulative effect of prenatal exposure on gestational and maternal health. This research also aims to explore the potential role the urban canopy may play in mitigating exposure to these environmental stressors.

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login