In order to join virtual sessions, you must be registered and logged-in(Were you registered for the in-person meeting in Denver? if yes, just log in.) 
Note: All session times are in Mountain Daylight Time.

A machine learning approach to model spatial variations of daily fine particulate matter (PM2.5) and nitrogen dioxide (NO2) in Shanghai, China

Authors: Chao Liu*, College of Architectrue and Urban Planning, Tongji University, Xinyi Song, Shanghai Jiaotong University
Topics: Urban and Regional Planning, Environmental Science, Quantitative Methods
Keywords: Land Use Regression (LUR), machine learning, Random Forest, Intra-urban air pollution, PM2.5, NO2, spatial-temporal
Session Type: Guided Poster
Presentation File: No File Uploaded

It is challenging to predict high-resolution spatial-temporal patterns of intra-urban air pollution and identify impacting factors in regional scale. Studies have attempted to capture features of air pollutants such as fine particulate matter (PM2.5) and nitrogen dioxide (NO2) using Land Use Regression (LUR) models but this method overlooks factor multicollinearity. While machine learning is a feasible approach to establish persuasive intra-urban air pollution daily-variation models. In this paper, random forest was exploited to establish intra-urban PM2.5 and NO2 spatial-temporal variation models and compared to traditional LUR method. Taking City of Shanghai, China as the case area, 36 station-measured daily records in one year of PM2.5 and NO2 concentration and over 80 different predictors associated with meteorological and geographical conditions, transportation, community population density, land use, and points of interest were used to construct LUR and random forest models. Results from the two methods were compared, and impacting factors were identified. Explained variance (R2) was used to quantify and compare model performance. The final LUR model explained 49.3% and 42.2% of the spatial variation in ambient PM2.5 and NO2, respectively, whereas the random forest model explained 78.1% and 60.5% of the variance. Regression mapping for unsampled sites on a grid pattern of 1km×1km was also implemented. The random forest models performed much better than the LUR model. In general, our findings suggest that the random forest approach offer a robust improvement in predicting performance compared to LUR model to estimate daily spatial variations in ambient PM2.5 and NO2.

Abstract Information

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

To access contact information login