Documentation for computing seismic velocity variations (doublets) and invert them

This documentation explains how we performed the computation of seismic velocity variations and how we invert them.

Here is the gitlab repository associated to the DOUBLETS (Whisper and F-IMAGE) project.

1. Download data for 2 seismic traces

$ iget -rv /mantis/home/lecoinal/DATA_TESTS_DOUBLETS
D- ./DATA_TESTS_DOUBLETS :
D- ./DATA_TESTS_DOUBLETS/oar.prepSHUJUAN.76736803 :
   CI.BRE.__.Z.trace.int8.h5    2614.394 MB | 6.518 sec | 16 thr | 401.090 MB/s
   CI.DLA.__.Z.trace.int8.h5    2612.931 MB | 5.788 sec | 16 thr | 451.461 MB/s
   CI.LAF.__.Z.trace.int8.h5    2674.234 MB | 6.300 sec | 16 thr | 424.484 MB/s
   CI.LGB.__.Z.trace.int8.h5    2636.603 MB | 7.389 sec | 16 thr | 356.832 MB/s
   CI.LLS.__.Z.trace.int8.h5    2672.918 MB | 8.210 sec | 16 thr | 325.570 MB/s
   namelist_doublets_iworms.       0.005 MB | 0.327 sec | 0 thr |  0.016 MB/s
   preprocess_doublets_iworm       0.005 MB | 0.063 sec | 0 thr |  0.083 MB/s
   readme                          0.003 MB | 0.043 sec | 0 thr |  0.064 MB/s
   small.CI.BRE.__.Z.trace.i     213.430 MB | 1.228 sec | 16 thr | 173.753 MB/s
   small.CI.LLS.__.Z.trace.i     240.661 MB | 0.850 sec | 16 thr | 283.275 MB/s
   submit_preprocess_doublet       0.001 MB | 0.064 sec | 0 thr |  0.020 MB/s

$ h5ls -r small.CI.BRE.__.Z.trace.int8.h5 |head -5
/                        Group
/2000                    Group
/2000/122                Dataset {1728000}
/2000/123                Dataset {1728000}
/2000/124                Dataset {1728000}
[...]
/2001                    Group
/2001/001                Dataset {1728000}
/2001/002                Dataset {1728000}
/2001/003                Dataset {1728000}
/2001/004                Dataset {1728000}
/2001/005                Dataset {1728000}
[...]
/2001/362                Dataset {1728000}
/2001/363                Dataset {1728000}
/2001/364                Dataset {1728000}
/2001/365                Dataset {1728000}

2. Get doublets code, dependancies installation, compilation and linking

2.a. Get doublets code

> git clone https://gricad-gitlab.univ-grenoble-alpes.fr/lecoinal/doublets.git

2.b. Install dependancies

For our code to work, one should install the following dependencies:

  • gfortran
  • hdf5
  • lapack
  • blas
  • fftw

2.c. Code compilation and linking

Now, you can compile the code using the Makefile_local: 

$ make -f Makefile_local 
$ make -f Makefile_local 
h5fc -cpp -Dlk_1bit  -O2  -c   -o constants.o constants.f90 
h5fc -cpp -Dlk_1bit  -O2  -c   -o corrmodule.o corrmodule.f90 -I/usr/include -lfftw3  
h5fc -cpp -Dlk_1bit  -O2  -c   -o doubletmodule.o doubletmodule.f90 
h5fc -cpp -Dlk_1bit  -O2  -c   -o inversionmodule.o inversionmodule.f90 
h5fc -cpp -Dlk_1bit  -O2  -o corr_doublet_inversion_bypair constants.o corrmodule.o doubletmodule.o inversionmodule.o corr_doublet_inversion_bypair.f90 -I/usr/include -lfftw3  -llapack -lblas 
h5fc -cpp -Dlk_1bit  -O2  -o compute_Cminv corrmodule.o compute_Cminv.f90 -I/usr/include -lfftw3  -llapack -lblas

It creates 2 binaries :

  • compute_Cminv: This program computes the inverse matrix Cminv as Cminv(i,j) = exp( -(dist(i,j)) / (2*Lcorr) ) with Lcorr in days. Cm and Cminv are symetric band Topelitz matrix. This program stores Cminv in packed storage and normal format. Cminv [N,N] is a covariance matrix on model params built from distance between model params. It represents the temporal smoothing of the final dt/t curve.

  • corr_doublet_inversion_bypair: This program takes 2 input files (2 seismic traces) and computes correlations, computes doublets for each pair of dates, and then invert doublets to get one value per date

Remarque: méthode iworms aussi pour l'"inversion" : code dvov_iworms.py

2.d. For GriCAD users (Université Grenoble Alpes)

An (old) installation procedure with using MODULE environment on ISTerre (ist-oar) and on GriCAD clusters is detailed here:

Dependencies list, and installation/compilation instructions on ISTerre/GriCAD clusters (module environment).

An installation procedure with using Guix environment on GriCAD clusters (only luke and dahu) is detailed here:

Dependencies list, and installation/compilation instructions on GriCAD clusters (guix environment).

3. Prepare metadata

You need the inter-station distance (in km).

Example here:

  • CI.BRE.00.Z CI.DLA.00.Z 11.550312518734689

  • CI.BRE.00.Z CI.LLS.00.Z 14.12678310723991

  • CI.LAF.00.Z CI.LGB.00.Z 20.57029577662673

4. Edit the namelist

todo: detail namelist entries...

Example namelist

5. Run the code

Once you have completed the namelist, you can use the wrapper submit_doublets_iworms.sh to run the MFP code. Using this submission script, you can directly change the name of the 2 input files and the corresponding interstation distance without changing the namelist (you can use namelist as a template that is automatically filled by submission script). 

> ./submit_doublets_iworms.sh CI.BRE.__.Z.trace.int8 CI.LLS.__.Z.trace.int8 14.12678310723991
[...]