gerjoii

ground penetrating radar and electrical resistivity joint inversion (gerjoii)

by diego domenzain

https://diegozain.github.io/gerjoii

---

eager to start?

• field/tinyradar-sy/ for a synthetic example that can run in a laptop.
• field/example-sy/ for a synthetic example that should be run in a cluster.
• scroll down for a more detailed example (under construction, sorry)

---

Publications

• Joint inversion of full-waveform GPR and ER data. Part 2: enhancing low frequencies with the envelope transform and cross-gradients, Diego Domenzain, John Bradford, and Jodi Mead. Geophysics, 85(6). 2020.

• Joint inversion of full-waveform GPR and ER data. Part 1, Diego Domenzain, John Bradford, and Jodi Mead. Geophysics, 85(6). 2020.

• Inversion of 2.5D electrical resistivity data using the discrete adjoint method, Diego Domenzain, John Bradford, and Jodi Mead. SEG Technical Program Expanded Abstracts 2020.

• Joint inversion of full-waveform GPR and ER data enhanced by the envelope transform and cross-gradients, Diego Domenzain, John Bradford, and Jodi Mead. 18th International Conference on Ground Penetrating Radar, Golden, Colorado 2020.

• Joint inversion of GPR and ER data, Diego Domenzain, John Bradford, and Jodi Mead. SEG Technical Program Expanded Abstracts 2018.

---

gerjoii does, for both radar and electrical resistivity:

• forward models
• inversion routines (fwi for radar, adjoint method for both)
• inversions in parallel (inversion code parallelized)
• parallel inversions (many inversions at once, only with Slurm)

Forward modeling

• 2D and 2.5D
• linear & isotropic
• radar (or acoustic)
  • finite difference time domain
  • PML boundaries
  • topography
  • arbitrary position of sources and receivers
• electrical resistivity
  • finite volume
  • Robin & Neumann boundaries
  • flat topography
  • surface sources & receivers

Inversion routines

• 2D and 2.5D
• gradient descent
• full-waveform inversion for radar
  • instantaneous phase
  • envelope
• adjoint method for electrical resistivity
• complete access to all sensitivities and adjoint fields
• joint inversions
  • variable weights on sensitivities
  • cross-gradients
• support for field and synthetic data
• custom Slurm scripts ready to dominate the cluster

Signal processing

• waveform filtering
• image filtering
• velocity semblance
• beamforming
• radar source estimation
• common-offset gather processing for radar

Useful stuff

• 2D velocity generator (.png to .mat file, any .png will do)

• electrical resistivity sequence generator ready for Syscal

  • dipole-dipole
  • Wenner
  • Schlumberger

• Matlab & Python data visualization

Seismic interferometry

• cross-correlation
• multi-dimensional deconvolution

---

docs

Detailed documentation of the code lives in the directory docs/manuals/.

---

Examples

Synthetic data

For a quick example go to field/example-sy/.

For a longer recipe read these steps:

1. In local, go to field/shells_ and run cloner.sh to create a new synthetic toy example.
2. In the new folder field/toy-sy/ go to image2mat/mat-file/ to see the true parameters.
3. In image2mat/initial-guess/ see the initial parameters.
4. You can change both true and initial parameters as you wish. Read build_model.pdf.
5. Go to toy-sy/base/scripts/ and make sure [wdc,w,dc]_begin_.m loads the initial parameters you want.
6. Go to toy-sy/inv-param/ and edit whichever P_inv_[dc,w, ].txt you are going to use. This file controls the inversion parameters.
7. Upload gerjoii to remote. You can use rsync_gerjoii.sh.
8. Go to toy-sy/slurm/ and read slurm_howto.pdf and gerjoii2slurm.pdf.
9. Run your code.
10. See results of files in remote while in local with utils in field/server-see/.
11. Download with with download_[ ,obse,pinv,reco,u].sh.
12. See in local with utils in toy-sy/see/.

Field data

1. In data/raw/ make directory project and inside that make these:

   dc-data/syscal.txt
   dc-data/data-mat/
   dc-data/data-mat-raw/
   
   w-data/*[.HD, DT1]
   w-data/data-mat/
   w-data/data-mat-raw/
   w-data/data-mat-fwi/
   

2. Under construction, sorry.

---

Author

Diego Domenzain [email protected]

---