Deep learning balloon anomaly detection

Suite of deep learning solutions for balloon pressure data denoising and anomaly detection. This is work in progress by Quentin Brissaud at NORSAR.

Data availability

• Balloon data from the 2019 Ridgecrest campaign can be found here: https://doi.org/10.6084/m9.figshare.14374067
• Balloon data from the the 15 January 2022 Hunga Eruption, Tonga can be found here: https://data.ipsl.fr/catalog/strateole2/eng/catalog.search#/search?from=1%26to%3D30
• Balloon data from Large Earthquakes can be found here: https://doi.org/10.5281/zenodo.6344454
• Balloon data from the MINI-BOOSTER campaign can be found here: https://snd.gu.se/en/catalogue/dataset/2021-257-1
• Balloon data from a Buried Chemical Explosion can be found here: https://ds.iris.edu/mda/21-021/

Dataset construction and synthetics

• .h5 datasets can be built using the notebook "generate_h5_files.ipynb"
• Extraction of Rayleigh waveforms to build synthetics can be done using "get_rayleigh_waves.ipynb"

Models

• Deep learning anomaly detection: "anomaly_detection.ipynb"

Extra

• The Fastmap algorithm for clustering was tested in routine "fastmap_test.ipynb"