Mapping the distribution of subsea permafrost is a key step for understanding its impact on global warming. Although conventional seismic techniques have been used to determine the lateral extent of subsea permafrost, they are limited for evaluating its vertical variation in regional-scale mapping. However, the work of Bustamante et al. 2023 presented a deep learning approach that enebled the generation of more reliable and accurate velocity and attenuation models from seismic data using a multi-input multi-ouput NN and a transfer learning technique.
This repository follows the code and results in "Mapping subsea permafrost distribution in the Beaufort Sea with marine seismic and deep learning" submitted to JGR: Solid Earth. This work extends the work of Bustamante et al. 2023 by evaluating the NN in 15 seismic lines from the ARA04C and ARA05C surveys in the Beaufort Sea (Kang et al. 2023). The distribution of the seismic lines is shown in the following figure:
In addition, following predefined thresholds, this repository calculate the permafrost distribution in all the seismic lines as described in the paper.
The repository contains 4 folders aiming to store the required information (CheckpointsTL, DataPreprocessed, InvertedModels and SSPInterpretation) and 1 code folder. The data folders are as follows:
Checkpoints: Store the last checkpoint on the TL methodologyDatapreprocessed: Store de seismic lines aranged in CMPs.
- The subfolder
CMPs_coordssaves the location files of the lines and the subfolderMultiInputsave the seismic line data transformed in the 4 input domains
InvertedModels: Store the inverted models after evaluating the 64 NN in the subfolder TL and the average of the Vp model in the subfolder VP_modelsSSPInterpretation: Store the inverted velocity models after applying the defined thresholds
The coding folder contain the scrpts necessary for obtaining the results from the TL methodology applied to the seismic data. There are two main files to consider:
Evaluate.py: Read the information on the Datapreprocessed folder and generate the MultiInput file if it does not exist. It also generates the output file in InvertedModels/TL subfolderSections_100mIsobath.py: Generate the interpeted sections of permafrost distribution in the sesimic lines and store them in the folder SSPInterpretation. In addition, it saves the Average Vp velocity models in the subfolder InvertedModels/VP_models.
The requiements for running the scripts are summarized in the file requirements.txt. Note that the package GeoFlow is available in https://github.com/gfabieno/GeoFlow
In addition to the code and data, the WellLogs folder contains the crystal cable logs in wells Irkaluk B-35, Kopanoar
M-13, and Nektoralik K-59 used in the paper.
After locating the pre-processed seismic lines in the folder Datapreprocessed, the NN can be evaluated line by line as
follows:
cd code
python Evaluate.py -ln 05-06Please change the line number option -ln to the desired seismic line. Choose between ['04-01', '04-02', '04-08', '04-09', '04-10', '04-11', '05-01', '05-03', '05-05', '05-06', '05-07', '05-08', '05-11', '05-12', '05-14', '05-15', '05-16', '05-17'].
Note that the input follows the (ARAC survey number - line number) format.
Evaluate.py reads the information from the folder Datapreprocessed and generates the MultiInput file if it does not
exist. It also generates the output file in InvertedModels/TL subfolder. The output file contains the average of the
inverted velocities and attenuations of the seismic line. Note that Evaluate.py can be run in parallel for all the
seismic lines in the ARA04C and ARA05C surveys.
Sections_100mIsobath.py generates the interpeted sections of permafrost distribution in the 100m isobath for the
evaluated sesimic lines and store them in the folder SSPInterpretation. In addition, it saves the Average Vp velocity
models in the subfolder InvertedModels/VP_models. The script can be run as follows:
cd code
python Sections_100mIsobath.py -ln 05-06Again, please change the line number option -ln to the desired seismic line.
Figures 5-7 in the paper can be reproduced by running the following scripts:
- Figure 5:
python Figure_Parallel_lines_1.py - Figure 6:
python Figure_Parallel_lines_2.py - Figure 7:
python Figure_Orthogonal_lines.py
These figures show the inverted velocity and attenuation models evaluated in the seismic lines parallel (Figures 5-6) and orthogonal (Figure 7) to the shelf.
The prediction at the intersection between different seismic lines is shown in Figure 8. To reproduce this figure, run
In addition, Figure 9 compared the inverted P-wave velocity models with crystal cable logs collected in wells close to the seismic lines. To reproduce this figure, run
The code also generate the corresponding .sgy file for each seismic line. Te fence diagrams displayed on the paper
can be easily visualized using 3D seismic visualization software such as OpendTect
(https://www.opendtect.org/osr/Main/HomePage).
Jefferson Bustamante Restrepo (Polytechnique Montreal, Geological Survey of Canada),
Gabriel Fabien-Ouellet (Polytechnique Montreal),
Mathieu Duchesne (Geological Survey of Canada),
Amr Ibrahim (Polytechnique Montreal),
[1] Bustamante, J., Fabien-Ouellet, G., Duchesne, M.J. and Ibrahim, A., 2023. Deeplearning viscoelastic seismic inversion for mapping subsea permafrost. Geophysics, Submitted
[2] Kang Seung-Goo, Young Keun Jin, Jongkuk Hong, 2023. Geophysical data (multi channel seismic data) collected during the 2013 ARA04C and 2014 ARA05C expeditions on the Beaufort sea. Available at: https://doi.org/10.22663/kopri-kpdc-00002217.1.





