This folder contains the implementation code of the paper "Back to the Future: GNN-based NO2 Forecasting via Future Covariates" (IGARSS 2024).

Authors: Antonio Giganti, Sara Mandelli, Paolo Bestagini, Umberto Giuriato, Alessandro D’Ausilio, Marco Marcon, Stefano Tubaro

The directory is structured as follows:

.
├── dataset/
├── lib/
├   ├── models.py
├   ├── utils.py
├   └── dataset/
├       ├── Madrid_2019.csv 
├       ├── Madrid_2019_tsl.pkl
├   	└── tsl_dataset_refactor.py
├── logger/
├── model/
├── conda_env.yaml
├── experiment.yaml
├── README.yaml
└── run_experiment.py

The datasets used in the experiments is available here. We provide a Torch Spatiotemporal (tsl)-ready version of the data, available at ./lib/dataset/Madrid_2019_tsl.pkl. This was created using the tsl_dataset_refactor.py and the Madrid_2019.csv data.

The experiment.yaml file stores all the parameter used to run the experiment. The results of the experiment are stored in the ./logs/ folder.

To solve all dependencies, we recommend using Anaconda and the provided environment configuration by running the command:

conda env create -f conda_env.yml
conda activate conda_env

The script used for the experiments in the paper is the run_experiment.py file. Change the settings in the experiment.yaml file according to the settings you want to use for the experiment, i.e., the model, the training parameters, etc. In addition, you have to set the ROOT_PATH in the ./lib/utils.py according to your configuration. After that, run the script with the command:

python run_experiment.py 

@inproceedings{giganti_magcrn_2024,
    author = {Giganti, Antonio and Mandelli, Sara, and Bestagini, Paolo and Giuriato, Umberto and D’Ausilio, Alessandro and Marcon, Marco and Tubaro, Stefano},
    title = {{Back to the Future: GNN-based NO2 Forecasting via Future Covariates}},
    year = {2024}
}

• Image and Sound Processing Lab (ISPL)

• Department of Electronics, Information and Bioengineering (DEIB)

• Politecnico di Milano

• Antonio Giganti, ResearchGate, LinkedIn

• Sara Mandelli, ResearchGate, LinkedIn

• Paolo Bestagini, LinkedIn

• Umberto Giuriato, ResearchGate, LinkedIn

• Alessandro D’Ausilio, ResearchGate, LinkedIn

• Marco Marcon, LinkedIn

• Stefano Tubaro, ResearchGate, LinkedIn

These works were supported by the Italian Ministry of University and Research MUR and the European Union (EU) under the PON/REACT project.