Loading...
|
Please use this identifier to cite or link to this item:
http://ir.nhri.org.tw/handle/3990099045/15821
|
| Title: | Geo-AI prediction model on estimating spatiotemporal variation of PM2.5 concentrations in morning and evening rush hours- a case study in taipei metropolitan, Taiwan |
| Authors: | Wong, PY;Su, HJ;Wu, CD |
| Contributors: | National Institute of Environmental Health Sciences |
| Abstract: | Urban air pollution has been a critical issue worldwide. It is indicated that PM2.5 poses many adverse health effects. Since the PM2.5 level during rush hours significantly contributes to overall exposure, it is crucial to investigate the effects during commutes. This study aimed to propose a geospatial artificial intelligence (Geo-AI) prediction model to estimate the spatiotemporal variation of PM2.5 concentrations during morning and evening rush hours in Taiwan. Hourly PM2.5 measurements from 2006 to 2020 were collected from 74 air quality monitoring stations established by the Taiwan Environmental Protection Administration. Total of 0.35 million observations were involved in analysis. We further aggregated hourly PM2.5 into morning (7:00-9:00) and evening (16:00-18:00) averages. Potential predictor variables included co-pollutants (NO2 and SO2), land-use/land cover information, landmarks, satellite images. Totally, around 400 potential variables were included in the modelling process. For feature selection, we adopted SHapley Additive exPlanations (SHAP) index as a reference to remove the redundant features. To develop the Geo-AI prediction model, Kriging interpolation, land-use regression, machine learning, and ensemble stacking approach were utilized. For model validation, 10-fold cross validation (CV) and temporal external data were included to test overfitting issue and the model extrapolation capability. The Geo-AI model captured 90% and 95% of PM2.5 variability, and the root mean square error (RMSE) was 4.85 and 3.75 μg/m3 for morning and evening periods, respectively. Similar results were obtained from 10-fold CV and external data validation with R2 of 0.90 and 0.83 for morning, 0.94 and 0.77 for evening periods. The selected variables showed that Kriged PM2.5, distance to industrial parks, and the density of roads explained most of PM2.5 variation for morning rush hour. It's also discovered that Kriged PM2.5, distance to industrial parks, and the density of temples affected PM2.5 variation in evening rush hour. The developed Geo-AI could estimate the spatiotemporal variation of PM2.5 concentrations with a high prediction accuracy for morning and evening rush hours. Important features were identified by the explainable SHAP index through machine learning process. The spatial distribution of estimated PM2.5 could provide information for the government to manage urban air pollution controlling strategies. |
| Date: | 2023-10-30- |
| Relation: | 2023 Asian Conference on Remote Sensing. 2023 Oct 30-Nov 03:Abstract number 198676. |
| Link to: | https://a-a-r-s.org/proceeding/ACRS2023/Remote%20Sensing%20Applications/ACRS2023145.pdf |
| Cited Times(Scopus): | https://www.scopus.com/inward/record.url?partnerID=HzOxMe3b&scp=85191231361 |
| Appears in Collections: | [其他] 期刊論文
|
Files in This Item:
| File |
Description |
Size | Format | |
|---|
| SCP85191231361.pdf | | 217Kb | Adobe PDF | 35 | View/Open |
|
All items in NHRI are protected by copyright, with all rights reserved.
|
