sse-denoising-gnss

Source code for the paper "Denoising of Geodetic Time Series Using Spatiotemporal Graph Neural Networks: Application to Slow Slip Event Extraction". Please cite the original paper when using this code:

Costantino, G., Giffard-Roisin, S., Mura, M. D., & Socquet, A. (2024). Denoising of Geodetic Time Series Using Spatiotemporal Graph Neural Networks: Application to Slow Slip Event Extraction. 10.1109/JSTARS.2024.3465270

Installation

Use the file requirements.txt to install the correct dependencies for the project.

How the code works

The code follows a sequence of steps:

1. Generation of the synthetic database: generate_synthetic_data.py
2. Training of the model: train.py
3. Inference
   • inference on the synthetic database: inference_synthetic_data.py
   • inference on real GNSS data: inference_running_window.py
4. Test: tables and figures
   • test on the synthetic database: test_synthetic_data.py
   • test on real GNSS data: test_real_data.py

Additional files:

• ablation_study.py: reproduces the figures and results of the ablation study (Table II and Fig. 4 of the paper)
• adj_matrix_plots.py: reproduces the figures on the optimal graph connections (Fig. 5 of the paper)

Note well: in order to reproduce the results presented in the paper (i.e., steps 1 to 3 should be skipped), the synthetic database as well as the prediction files can be downloaded at: 10.5281/zenodo.11283069

After downloading, the file placement should look like the following:

denois_synth_ts_cascadia_realgaps_extended_v5_200stations_6_7_depth_20_40.data
predictions
   ablation
      pred_denoising_test_data_ablation_notransf
      pred_denoising_test_data_ablation_spatial_att_only
      pred_denoising_test_data_ablation_temp_att_only
   pred_denoising_test_data
   pred_denoising_test_data_1d_xue_freymueller
   pred_denoising_test_data_2d_unet