DL-HRGNSS

DL-HRGNSS

GNSS_Magnitude_V1

This is our preliminary DL model based on a convolutional neural network for magnitude estimation from HR-GNSS data (1 Hz sampling rate).

We have trained the model for three cases:

• Magnitude estimation from 3 stations, 181 seconds, 3 components.
• Magnitude estimation from 7 stations, 181 seconds, 3 components.
• Magnitude estimation from 7 stations, 501 seconds, 3 components.

Related work:

Claudia Quinteros-Cartaya, Jonas Köhler, Wei Li, Johannes Faber, Nishtha Srivastava, Exploring a CNN model for earthquake magnitude estimation using HR-GNSS data, Journal of South American Earth Sciences, 2024, 104815, ISSN 0895-9811, https://doi.org/10.1016/j.jsames.2024.104815.

Getting Started

Clone the repository

git clone https://github.com/srivastavaresearchgroup/DL-HRGNSS

Install dependencies (with python 3.8)

(virtualenv is recommended)

pip install -r requirements.txt

Data

The database for each configuration/case is in ./data. For example the data folder: ./data/GNSS_M3S_181 contains the data for 3 stations, 181 seconds.

The data is in numpy format, previously selected from the open-access database published by Lin et al., 2020 (https://doi.org/10.5281/zenodo.4008690).

You can find the information related to the data (ID, Hypocenter, Magnitude) in the dataframes info_data.csv located in the same folder as the respective data.

You can use Data_plot.ipynb to plot the waveforms.

Refer to Quinteros et al., 2024 (https://doi.org/10.1016/j.jsames.2024.104815) for more details about the data configuration.

Pretrained models and results

The pre-trained models are located in ./trained_models. These models are described in Quinteros et al., 2024 (https://doi.org/10.1016/j.jsames.2024.104815). You can find the respective results in ./predictions.

Processing and output

If you want to change the default configuration, you can edit the variables in main.py.

Configuration parameters you should change, depending on your choice:

• Number of stations (nst)
• Time window length (nt)
• Paths/folder names

To split, train and test the data, run:

python main.py

The outputs will be located in the path that you set.

By default the outputs will be saved in ./tests, and in the subfolders:

• ./data_inf: you can find the numpy files with the data index for train, validation, and test dataset, associated with the initial xdata.npy file.
• ./models: trained models are saved in this folder
• ./predictions: the predicted magnitude and error values are saved in text files in this folder.
• ./out_log: the output from logging is saved in this folder.