Run this example

Generate synthetic data

Run 00_forward.sh to generate synthetic teleseismic waveforms using the target model. The FK plane-wave boundary conditions are applied at the mesh boundaries for each event defined in src_rec/sources_telecc.dat. The synthetic data is saved in the fwat_data folder.

00_forward.sh


#!/bin/bash
set -e
 
rm -rf OUTPUT_FILES
 
mkdir -p DATABASES_MPI
mkdir -p OUTPUT_FILES
 
NPROC=`grep ^NPROC DATA/Par_file | grep -v -E '^[[:space:]]*#' | cut -d = -f 2 | cut -d \# -f 1`
 
# Copy target model and run forward simulation
cp target_model.h5 DATA/tomo_files/tomography_model.h5
mpirun -np $NPROC ../../bin/xfwat_mesh_databases -s tele
mpirun -np $NPROC ../../bin/xfwat_fwd_measure_adj -m M00 -s tele -r 1

The -s tele flag activates the teleseismic mode in SpecFWAT, which reads FKmodel_* files from src_rec/ and injects plane-wave boundary conditions computed by the FK method. The source time function (STF) estimation is handled internally during the measurement step.

Conduct the inversion

Run 01_run_this_test.sh to perform the inversion process for 10 iterations. Each iteration consists of meshing, forward simulation with measurement and adjoint source computation, post-processing, and model update.

01_run_this_test.sh


#!/bin/bash
set -e
 
rm -rf OUTPUT_FILES
 
mkdir -p DATABASES_MPI
mkdir -p OUTPUT_FILES
 
NPROC=`grep ^NPROC DATA/Par_file | grep -v -E '^[[:space:]]*#' | cut -d = -f 2 | cut -d \# -f 1`
 
# Run inversion for 10 iterations
for it in `seq 0 9`; do
    model=`printf "M%02d" $it`
    if [ $it -eq 0 ]; then
        cp initial_model.h5 DATA/tomo_files/tomography_model.h5
    fi
    mpirun -np $NPROC ../../bin/xfwat_mesh_databases -s tele
    mpirun -np $NPROC ../../bin/xfwat_fwd_measure_adj -m $model -s tele -r 3
    mpirun -np $NPROC ../../bin/xfwat_post_proc -m $model
    mpirun -np $NPROC ../../bin/xfwat_optimize -m $model
done

The -r 3 flag enables forward simulation, measurement, and adjoint simulation in a single call. Kernels from all 8 events are summed during post-processing (xfwat_post_proc) before the model is updated by the L-BFGS optimizer (xfwat_optimize).