Estimation of Ground-Level NO<sub>2</sub> Concentrations Over Megacities Using Sentinel-5P and Machine Learning Models: A Case Study of Istanbul
Air pollution is a serious issue in terms of public and environmental health. In this regard, it is important to determine its compliance with the standards by continuous monitoring. For this purpose, air quality ground monitoring stations have been established as part of monitoring systems. While t...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Copernicus Publications
2025-05-01
|
| Series: | The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences |
| Online Access: | https://isprs-archives.copernicus.org/articles/XLVIII-M-6-2025/303/2025/isprs-archives-XLVIII-M-6-2025-303-2025.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Air pollution is a serious issue in terms of public and environmental health. In this regard, it is important to determine its compliance with the standards by continuous monitoring. For this purpose, air quality ground monitoring stations have been established as part of monitoring systems. While these stations provide highly accurate data, they are point-based and costly. Satellite data, which provides a wide coverage area, enables local and global analysis while providing data with low spatial resolution. Integration of ground and satellite data using machine learning (ML) algorithms enables more accurate regional analysis. For this purpose, estimation analysis of the NO<sub>2</sub> parameter, which is the most measured parameter at ground monitoring stations and has a major impact on its formation by human activities, was conducted for the Istanbul megacity using freely accessible Sentinel-5P satellite data. The performance of three ML algorithms, namely multi-layer perceptron (MLP), support vector regression (SVR), and XGBoost regression (XGB), in estimating the ground level-NO<sub>2</sub> parameter was evaluated both quantitatively using RMSE and MAE accuracy metrics and qualitatively by visual analysis. The model was trained with data covering the years 2019–2022, validated with data for 2023, and tested with data for 2024. According to the results obtained, while the three models gave similar results with RMSE values of 19.59, 19.65, and 20.03 μg/m<sup>3</sup> and MAE values of 15.00, 14.34, and 15.90 μg/m<sup>3</sup> in the test data, SVR and MLP models provided higher accuracy in the seasonal assessment. In the visual assessment, the SVR model results provide a more accurate approach. |
|---|---|
| ISSN: | 1682-1750 2194-9034 |