Mondaic
This tutorial is presented as Python code running inside a Jupyter Notebook, the recommended way to use Salvus. To run it yourself you can copy/type each individual cell or directly download the full notebook, including all required files.
This micro-tutorial compares different parameterizations of acoustic media and demonstrates how to run VTI acoustic simulations with SalvusProject. The concept is shown for a simple 2-D layered medium, but it works just the same in 3-D by just adding an extra dimension.
import numpy as np
import os
import salvus.namespace as sn
# Run simulations on this site.
SALVUS_FLOW_SITE_NAME = os.environ.get("SITE_NAME", "local")# Either load an existing project or create a new one.
p = sn.Project.from_domain(
path="project",
# Should match the volume model.
domain=sn.domain.dim2.BoxDomain(x0=0, x1=3000, y0=-1000, y1=0),
load_if_exists=True,
)
We start with an isotropic acoustic model with 3 layers using a custom
bm model file.%%writefile layered_model.bm
NAME radial_model
ACOUSTIC true
UNITS m
COLUMNS depth rho vp
0.0 1000.0 1500.0
200.0 1000.0 1500.0
200.0 1300.0 1900.0
500.0 1300.0 1900.0
500.0 2200.0 3200.0
4000.0 2200.0 3200.0Writing layered_model.bm
Similarly, we introduce a VTI layered model, where the second and third layer have a higher wave speed in horizontal direction.
%%writefile layered_model_ani.bm
NAME radial_model
ACOUSTIC true
ANISOTROPIC true
UNITS m
COLUMNS depth rho vpv vph
0.0 1000.0 1500.0 1500.0
200.0 1000.0 1500.0 1500.0
200.0 1300.0 1900.0 2100.0
500.0 1300.0 1900.0 2100.0
500.0 2200.0 3200.0 3500.0
4000.0 2200.0 3200.0 3500.0Writing layered_model_ani.bm
The anisotropic parameters for vertical and horizontal P-wave velocities cam also
represent an isotropic medium equivalent to the first example when using the same value for
VPV and VPH.%%writefile layered_model_fake_ani.bm
NAME radial_model
ACOUSTIC true
ANISOTROPIC true
UNITS m
COLUMNS depth rho vpv vph
0.0 1000.0 1500.0 1500.0
200.0 1000.0 1500.0 1500.0
200.0 1300.0 1900.0 1900.0
500.0 1300.0 1900.0 1900.0
500.0 2200.0 3200.0 3200.0
4000.0 2200.0 3200.0 3200.0Writing layered_model_fake_ani.bm
Now, we create simulation configurations for all three 1D models. Obviously, the fake anisotropic model will result in the same waveforms as the isotropic model.
for name, bm in zip(
["isotropic", "anisotropic", "fake-anisotropic"],
["layered_model.bm", "layered_model_ani.bm", "layered_model_fake_ani.bm"],
):
p.add_to_project(
sn.SimulationConfiguration(
name=name,
max_frequency_in_hertz=20.0,
elements_per_wavelength=1.5,
model_configuration=sn.ModelConfiguration(
background_model=sn.model.background.one_dimensional.FromBm(
filename=bm, reference_datum=0.0
),
),
event_configuration=sn.EventConfiguration(
wavelet=sn.simple_config.stf.Ricker(center_frequency=10.0),
waveform_simulation_configuration=sn.WaveformSimulationConfiguration(
end_time_in_seconds=1.5,
),
),
absorbing_boundaries=sn.AbsorbingBoundaryParameters(
reference_velocity=2200.0,
reference_frequency=10.0,
number_of_wavelengths=7,
),
),
)
p.viz.nb.simulation_setup("anisotropic")[2024-03-15 09:19:44,934] INFO: Creating mesh. Hang on.
<salvus.mesh.unstructured_mesh.UnstructuredMesh at 0x7f1840e10710>
Finally, we highlight two more model parameterizations for anisotropic acoustic material using the Thomsen parameters and a second-order tensor, which can be seen as a reduced form of the 4th-order elastic tensor.
We add two more meshes representing the equivalent isotropic and anisotropic media in this parameterization as
UnstructuredMeshSimulationConfiguration.def convert_thomsen_params_to_vti_tensor(
dim: int,
rho: np.ndarray,
vp: np.ndarray,
delta: np.ndarray,
epsilon: np.ndarray,
):
if dim == 2:
model = {
"M0": 1 / (vp**2 * rho),
"D11": (1 + 2 * epsilon) / rho,
"D12": (delta - epsilon) / rho,
"D22": 1 / rho,
}
elif dim == 3:
model = {
"M0": 1 / (vp**2 * rho),
"D11": (1 + 2 * epsilon) / rho,
"D12": np.zeros_like(rho),
"D13": (delta - epsilon) / rho,
"D22": (1 + 2 * epsilon) / rho,
"D23": (delta - epsilon) / rho,
"D33": 1 / rho,
}
return model
def interpolate_vti_model_onto_mesh(mesh: sn.UnstructuredMesh):
if "VPV" in mesh.elemental_fields.keys():
vti_model = convert_thomsen_params_to_vti_tensor(
dim=2,
rho=mesh.elemental_fields["RHO"],
vp=mesh.elemental_fields["VPV"],
delta=0.5
* (
mesh.elemental_fields["VPH"] ** 2
/ mesh.elemental_fields["VPV"] ** 2
- 1.0
),
epsilon=0.5
* (
mesh.elemental_fields["VPH"] ** 2
/ mesh.elemental_fields["VPV"] ** 2
- 1.0
),
)
for k, v in vti_model.items():
mesh.elemental_fields[k] = v
del mesh.elemental_fields["VPV"]
del mesh.elemental_fields["VPH"]
del mesh.elemental_fields["RHO"]
else:
vti_model = convert_thomsen_params_to_vti_tensor(
dim=2,
rho=mesh.elemental_fields["RHO"],
vp=mesh.elemental_fields["VP"],
delta=np.zeros_like(mesh.elemental_fields["VP"]),
epsilon=np.zeros_like(mesh.elemental_fields["VP"]),
)
for k, v in vti_model.items():
mesh.elemental_fields[k] = v
del mesh.elemental_fields["VP"]
del mesh.elemental_fields["RHO"]
return mesh
mesh = p.simulations.get_mesh("isotropic")
mesh = interpolate_vti_model_onto_mesh(mesh)
p.add_to_project(
sn.UnstructuredMeshSimulationConfiguration(
unstructured_mesh=mesh,
name="thomsen-isotropic",
event_configuration=p.entities.get(
"simulation_configuration", "anisotropic"
).event_configuration,
)
)
mesh = p.simulations.get_mesh("anisotropic")
mesh = interpolate_vti_model_onto_mesh(mesh)
p.add_to_project(
sn.UnstructuredMeshSimulationConfiguration(
unstructured_mesh=mesh,
name="thomsen-anisotropic",
event_configuration=p.entities.get(
"simulation_configuration", "anisotropic"
).event_configuration,
)
)[2024-03-15 09:19:46,051] INFO: Creating mesh. Hang on.
At the surface of the domain, we consider a single point source at the center of the horizontal extent and model a linear array of receivers with a spacing of 20 m.
src = sn.simple_config.source.cartesian.ScalarPoint2D(x=1500, y=0, f=1.0)
recs = sn.simple_config.receiver.cartesian.collections.ArrayPoint2D(
x=np.linspace(100, 2900, 141), y=np.zeros(141), fields=["phi"]
)
p += sn.Event(event_name="event", sources=src, receivers=recs)
p.viz.nb.domain()
It is time to actually run the simulations for all 5 scenarios.
for name in [
"isotropic",
"anisotropic",
"fake-anisotropic",
"thomsen-anisotropic",
"thomsen-isotropic",
]:
p.simulations.launch(
simulation_configuration=name,
events=p.events.list(),
site_name=SALVUS_FLOW_SITE_NAME,
ranks_per_job=1,
)
p.simulations.query(block=True)[2024-03-15 09:19:46,485] INFO: Submitting job ... Uploading 1 files... 🚀 Submitted job_2403150919699652_f6fffc4328@local
[2024-03-15 09:19:50,785] INFO: Submitting job ... Uploading 1 files... 🚀 Submitted job_2403150919796562_6aead41959@local
[2024-03-15 09:19:55,986] INFO: Creating mesh. Hang on. [2024-03-15 09:19:56,164] INFO: Submitting job ... Uploading 1 files... 🚀 Submitted job_2403150919180867_2457e5b66a@local
[2024-03-15 09:20:01,953] INFO: Submitting job ... Uploading 1 files... 🚀 Submitted job_2403150920966237_d683b560b1@local
[2024-03-15 09:20:06,958] INFO: Submitting job ... Uploading 1 files... 🚀 Submitted job_2403150920998630_19a8dd8d1e@local
Of course, regardless of the parameterization of the medium, the waveforms of all isotropic simulations coincide. The same is true for both anisotropic parameterizations. Note, however, the difference between isotropic and anisotropic material.
p.viz.nb.waveforms(
data=["isotropic", "fake-anisotropic", "thomsen-isotropic"],
receiver_field="phi",
)
p.viz.nb.waveforms(
data=["anisotropic", "thomsen-anisotropic"],
receiver_field="phi",
)
p.viz.nb.waveforms(
data=["isotropic", "anisotropic"],
receiver_field="phi",
)
p.waveforms.get(data_name="isotropic", events=p.events.list())[0].plot(
component="A",
receiver_field="phi",
plot_types=["shotgather"],
)
p.waveforms.get(data_name="anisotropic", events=p.events.list())[0].plot(
component="A",
receiver_field="phi",
plot_types=["shotgather"],
)
PAGE CONTENTS