Processed air pollution data (NO2, PM10, PM2.5) collected from Antwerp in April, 2019
This repository contains the Antwerp's air pollution data collected in April, 2019. Three pollutants are considered including NO2, PM10 and PM2.5. The measurements provided by this dataset are the concentration of the considered pollutants in terms of parts per billion (ppb) (NO2) or microgram per cubic meter (μgm-3) (PM10 and PM2.5).
There are four folders, namely SSTR, HST, HTR and HSTR. Each folder has the data corresponding to a specific setting as follow.
- SSTR: Standard spatio-temporal resolution
- HSR: High spatial resolution
- HTR: High temporal resolution
- HSTR: High spatio-temporal resolution
Each folder has air quality measurements in sparse matrix format (e.g., no2_measurements.npz).
A row in the air quality measurement matrix corresponds to a location. A column in the air quality measurement matrix corresponds to a time instance (e.g., 60 minutes, 30 minutes).
In addition, it also contains the geo-coordinates of all the considered locations (nodes.txt).
A row in the file nodes.txt has the following format:
node_id longitude latitude
The underlying graph created by the road network of Antwerp is also provided via the corresponding adjacency matrix (adj.npz).
Furthermore, the training and testing masks are provided for result reproduction.
The following table presents the result for air quality inference achieved by using variational graph autoencoder (AVGAE) and multi-input variational graph autoencoder (MAVGAE).
| NO2 | PM2.5 | PM10 | |||||
|---|---|---|---|---|---|---|---|
| Dataset | Method | MAE | RMSE | MAE | RMSE | MAE | RMSE |
| SSTR | Krigging exponential | 7.21 | 10.73 | 3.26 | 4.92 | 6.34 | 8.57 |
| AVGAE | 6.27 | 9.25 | 2.55 | 3.65 | 5.04 | 6.87 | |
| MAVGAE | 6.03 | 8.86 | 2.45 | 3.61 | 4.3 | 5.7 | |
| HSR | Krigging exponential | 6.80 | 10.15 | 3.08 | 4.76 | 5.08 | 7.13 |
| AVGAE | 6.25 | 9.26 | 2.62 | 4.08 | 4.59 | 6.08 | |
| MAVGAE | 5.97 | 8.92 | 2.51 | 3.95 | 4.29 | 5.66 | |
| HTR | Krigging exponential | 7.04 | 10.35 | 3.23 | 4.92 | 7.11 | 9.70 |
| AVGAE | 6.46 | 9.37 | 2.79 | 4.03 | 5.55 | 7.34 | |
| MAVGAE | 6.11 | 9.01 | 2.59 | 3.83 | 4.62 | 5.98 | |
| HSTR | Krigging exponential | 6.95 | 10.35 | 3.13 | 4.86 | 5.04 | 7.12 |
| AVGAE | 6.72 | 10.07 | 2.74 | 4.12 | 4.59 | 6.06 | |
| MAVGAE | 6.32 | 9.48 | 2.61 | 3.98 | 4.34 | 5.66 | |
If you find the dataset useful, please consider citing the following works:
@article{do2020graph,
title={Graph-Deep-Learning-Based Inference of Fine-Grained Air Quality from Mobile IoT Sensors},
author={Do, Tien Huu and Tsiligianni, Evaggelia and Qin, Xuening and Hofman, Jelle and La Manna, Valerio Panzica and Philips, Wilfried and Deligiannis, Nikos},
journal={IEEE Internet of Things Journal},
year={2020},
publisher={IEEE}
}
@inproceedings{do2019matrix,
title={Matrix completion with variational graph autoencoders: Application in hyperlocal air quality inference},
author={Do, Tien Huu and Nguyen, Duc Minh and Tsiligianni, Evaggelia and Aguirre, Angel Lopez and La Manna, Valerio Panzica and Pasveer, Frank and Philips, Wilfried and Deligiannis, Nikos},
booktitle={ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
pages={7535--7539},
year={2019},
organization={IEEE}
}