Loopstructural

Constrained implicit geological models

Use LoopStructural when observations constrain continuous geological scalar fields. A scalar value labels an interface; it is not automatically an age, measured depth or elevation. Record the coordinate CRS, shared XYZ units, positive-Z convention, feature order and meaning of each scalar level.

A constrained planar model

This synthetic example uses metre coordinates with a nonzero origin and an asymmetric box. Its scalar field is Z - 3030 + 0.1*(X - 1050) - 0.05*(Y - 2100) in metres. Scalar observations and gradient constraints anchor both level and orientation; a single point alone cannot define a geological surface.

pythonCopy code
from itertools import product
from LoopStructural import GeologicalModel
import numpy as np
import pandas as pd

origin = np.array([1000., 2000., 3000.])
maximum = np.array([1100., 2200., 3060.])
xyz = np.array(list(product([1010., 1050., 1090.], [2020., 2180.], [3010., 3050.])))
data = pd.DataFrame(xyz, columns=['X', 'Y', 'Z'])
data['feature_name'] = 'strat'
data['val'] = xyz[:, 2] - 3030 + .1 * (xyz[:, 0] - 1050) - .05 * (xyz[:, 1] - 2100)
data[['gx', 'gy', 'gz']] = [.1, -.05, 1.]
model = GeologicalModel(origin, maximum)
model.data = data
model.create_and_add_foliation('strat', interpolatortype='FDI', nelements=1000)
model.update()
query = np.array([[1020., 2040., 3020.], [1080., 2160., 3040.], [1050., 2100., 3030.]])
values = model.evaluate_feature_value('strat', query)

The expected values are approximately −10, +10 and 0. Pass world coordinates to model.evaluate_feature_value(..., scale=True) (the default). Direct feature evaluation has a different local-coordinate contract. GeologicalModel 1.8 accepts the two bounds positionally; do not assume origin= and maximum= constructor keywords from older examples still work.

Evaluate a VTK grid

Continue with the model above. regular_grid returns an N×3 array; disable shuffling and use Fortran ordering for VTK's x-fastest points. Write actual scalar values into point data before saving.

pythonCopy code
import pyvista as pv

shape = (9, 7, 5)
points = model.regular_grid(nsteps=shape, shuffle=False, order='F')
grid = pv.StructuredGrid()
grid.points = points
grid.dimensions = shape
grid.point_data['strat'] = model.evaluate_feature_value('strat', points)
surface = grid.contour([0.], scalars='strat')
if not np.isfinite(grid['strat']).all() or surface.n_cells == 0:
    raise ValueError('Model evaluation or requested isosurface is invalid')

Use grid.save('model.vtk') or surface.save('strat.vtk') when export is requested. A rendered surface is a discretized level set; check constraint residuals and mesh convergence before interpreting small structures.

CSV helper and conditional models

The model builder supports constrained FDI/PLI foliations and grid/isosurface export. It validates finite coordinates, complete constraints and positive 3D extents, and reports failed or empty exports with a nonzero exit status. Provide explicit bounds for planar or zero-span input geometry rather than inventing an extent.

bashCopy code
python scripts/build_model.py constraints.csv --grid --output model.vtk \
  --origin 1000 2000 3000 --maximum 1100 2200 3060 --nsteps 9 7 5

Resolve the command relative to this installed skill directory. The helper requires X,Y,Z,feature_name, plus val and/or complete gx,gy,gz or nx,ny,nz constraints; each feature needs a scalar level anchor. Missing vector rows are all NaN, not zero vectors.

Read interpolation and validation for numerical choices, and fault/fold requirements only when those structures are needed. Faults, fold frames and uncertainty ensembles need explicit kinematics/data and a separate configured model; they are not created by a nominal CSV column or by this foliation helper.

Verification scope

LoopStructural 1.8.0 with loop-interpolation 0.0.2 was exercised with FDI and PLI planar models, world-coordinate evaluation, asymmetric grid ordering, isosurface and VTK readback, plus malformed-input/CLI failures. This validates the entrypoint and export path, not fault-network inversion, folded geology, uncertainty estimates or the geological adequacy of sparse field constraints.

Official model source, checked 2026-09-14 against the installed 1.8.0 API.