WP7: the integrated loop, discrepancy register, multiples and the Fig. 2/4/5 analogues by ingehap · Pull Request #8 · ingehap/ES_PHYS_P011

ingehap · 2026-10-14T21:16:43Z

Summary

Work package 7 of docs/IMPLEMENTATION_PLAN.md. This runs the article's integrated loop (Lewis et al., 2016, Fig. 1) end to end on the synthetic well, using estimated models only:

logs → ZO-VSP → walkaway → surface seismic;
each step is recorded in an auditable register of model versions, discrepancies and pre-fixed acceptance criteria.

New code

iwt/welltie.py:
- least-squares multi-trace wavelet estimation (Walden & White, 1998);
- constant phase rotation;
- tie metrics (White & Simm, 2003). The time shift and the phase come from the complex cross-correlation of the analytic signals, because a real cross-correlation mistakes one for the other.
iwt/workflow.py:
- model_from_logs: sonic and dipole moved to the seismic frequency with Q (Kjartansson, 1979); ε, δ, γ and Q curves; a gap layer above the logs, with its velocity from the VSP;
- Register: versions, discrepancies, criteria, Markdown output.
iwt/multiples.py: CMP gathers with the physics switched off step by step (full, primaries with conversions, no mode conversion, P primaries, Born).
iwt/avo.py: linear_resolution for the identifiability study.
iwt/synthetic.py: dipping_reservoir_models (wedge, dipping reservoir, flat OWC, all-brine twin).
iwt/modeling/reflection.py: exact_pp_slowness.
iwt/raytrace.py: an exact zero-offset fast path in reflection_table.
examples/reproduce_paper.py: the loop (about 6 min with 4 workers). It writes docs/discrepancy_register.md and docs/figures/wp7_{tie,ava,section}.png.
tests/test_workflow.py (11 tests), README and plan updates, version 0.7.0.

Results (synthetic well)

criterion (fixed before tuning)	value	threshold	passed
VSP first-break residual	0.22 ms	≤ 1 ms	yes
corridor stack vs synthetic	0.93	≥ 0.8	yes
reservoir event time shift	0.02 ms	≤ 2 ms	yes
Stø AVA gradient, contamination-free	0.40	≤ 0.10	no
Stø AVA gradient, as acquired	0.73	≤ 0.10	no

The time tie closes:

the ZO-VSP Q and drift calibration do most of the work;
filling the 52 m blind gap above the logs from the VSP lifts the corridor correlation from −0.06 to 0.93.

The AVA does not close, as in the article. After migration:

the full-physics contamination is 0.9–1.5 × Born, and it grows with angle;
without mode conversion, P-wave interbed multiples are small (0.14–0.23 × Born);
with mode conversion, the converted primaries and the multiples largely cancel, so an acoustic demultiple would not repair the AVA.

Synthetic-only attribution:

the true earth itself predicts 2–14 % too much amplitude. That is the Q model above the first VSP receiver; with the true Q the misfit falls from 0.088 to 0.040;
swapping one part of v2 into the truth is not additive. The 5 m-smoothed γ gives the isotropic stringers shale anisotropy.

Identifiability:

the reservoir AVA constrains ρ and the overburden Q, barely the Vs and not δ;
reservoir Vs and overburden Q correlate +0.59.

Fig. 5 analogue: VTI places the fluid signal inside the oil column, while isotropic imaging puts it 17 m deep. Q compensation barely sharpens the signal in 1D.

Deviations from the plan (documented in the plan's as-built notes)

Predictions use full reflectivity. A ray-Born prototype failed beyond about 27° with the thin stringers.
The AVA criterion uses 20–44°, not 30–50°, because the seabed reflection crosses the reservoir.
Not built: hold-out validation, Monte Carlo ensembles, free-surface multiples.

Test plan

pytest -q -W error::RuntimeWarning -W error::DeprecationWarning (Python 3.13 venv): 154 passed
pytest -q (Python 3.9 venv): 150 passed, 1 skipped (no matplotlib)
pyflakes iwt tests examples
python examples/reproduce_paper.py --workers 4 completes and regenerates the register and figures
README quick start runs

🤖 Generated with Claude Code

https://claude.ai/code/session_019vaUDUPicoFeQaUhfhnoYq

Generated by Claude Code

…. 2/4/5 analogues - iwt.welltie: least-squares wavelet estimation, phase rotation, tie metrics from the complex cross-correlation of analytic signals (shift and constant phase separated) - iwt.workflow: model_from_logs (dispersion with Q, anisotropy and Q curves, gap layer from the VSP), versioned models, discrepancy register, acceptance criteria, Markdown - iwt.multiples: CMP gathers with physics switched off step by step - iwt.avo.linear_resolution: identifiability (posterior std, data weight, correlation) - iwt.synthetic.dipping_reservoir_models: wedge, dipping reservoir, flat OWC - iwt.modeling.reflection.exact_pp_slowness; zero-offset fast path in reflection_table - examples/reproduce_paper.py runs logs -> ZO-VSP -> walkaway -> surface seismic on estimated models and writes docs/discrepancy_register.md and docs/figures/wp7_*.png - tests/test_workflow.py (11 tests); README, plan; version 0.7.0 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019vaUDUPicoFeQaUhfhnoYq

Copilot

🟡 Changes recommended

Numerical errors, ground-truth leakage, and inconsistent model history and scoring undermine the workflow’s reliability.

8 open findings

What changed in this PR

Implements WP7, connecting the existing modelling and calibration tools into an integrated synthetic-well workflow with an auditable discrepancy register.

Changes:

Adds well-tie metrics, model versioning, physics comparisons, and identifiability analysis.
Adds the reproduction script and dipping-reservoir experiment.
Documents results, adds regression tests, and bumps the package version.

File	Description
tests/test_workflow.py	Tests WP7 helpers and numerical behaviour.
README.md	Documents workflow, results, and limitations.
pyproject.toml	Bumps version to 0.7.0.
iwt/workflow.py	Adds log-derived models and discrepancy register.
iwt/welltie.py	Adds wavelet estimation and tie metrics.
iwt/synthetic.py	Adds dipping-reservoir model pairs.
iwt/raytrace.py	Optimizes zero-offset reflection tables.
iwt/multiples.py	Compares physics variants and relative errors.
iwt/modeling/reflection.py	Exposes PP reflection coefficients by slowness.
iwt/avo.py	Adds linear parameter-resolution analysis.
iwt/__init__.py	Exports new modules and updates version.
examples/reproduce_paper.py	Runs the integrated workflow and generates outputs.
docs/IMPLEMENTATION_PLAN.md	Records WP7 completion and deviations.
docs/discrepancy_register.md	Records generated discrepancies and acceptance results.

🧠 Review effort: Balanced

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

+    criteria: List[Criterion] = field(default_factory=list)
+
+    def new_version(self, name: str, model: ElasticModel1D, changes: str) -> ModelVersion:
+        v = ModelVersion(len(self.versions), name, model, changes)


+    reg.check("VSP first-break residual after calibration", 1e3 * np.sqrt(np.nanmean(d1 ** 2)), 1.0,
+              "<=", "ms")


+
+def anisotropy_step(model, logs, reg, zo):
+    z = logs.depth
+    g_sonic, g_up = wp5.sonic_gamma(model, logs)


+    a, b = np.asarray(a, dtype=float), np.asarray(b, dtype=float)
+    if mask is not None:
+        a, b = a[mask], b[mask]
+    return float(np.sqrt(np.nanmean((a - b) ** 2) / np.nanmean(b * b)))


+        units.sort(key=lambda u: u.top)
+        below = units[units.index(water) + 1:]
+        if below and below[0].top <= water.top:
+            raise ValueError("the dip moves the base of the reservoir above the contact")


+    L = int(round(max_shift / dt))
+    lags = np.arange(-L, L + 1)
+    # |analytic| norms are twice the real ones (equal energy in the quadrature part)
+    cc = np.array([np.vdot(np.roll(sa, k), oa) for k in lags]) / (2.0 * no * ns)


+        a, b, c = env[i - 1], env[i], env[i + 1]
+        den = a - 2 * b + c
+        off = 0.5 * (a - c) / den if den != 0 else 0.0
+    phi = float(-np.angle(cc[i]))


+    curves = {k: mid(_at(v, n)) for k, v in (("eps", eps), ("delta", delta), ("gamma", gamma),
+                                               ("qp", qp), ("qs", qs))}


ingehap marked this pull request as ready for review October 15, 2026 06:04

Copilot AI balanced review requested due to automatic review settings October 15, 2026 06:04

ingehap merged commit a7f5530 into main Oct 15, 2026
4 checks passed

Copilot AI reviewed Oct 15, 2026

View reviewed changes

Copilot started reviewing on behalf of ingehap October 15, 2026 07:27

View session

Navigation Menu

Repository navigation

WP7: the integrated loop, discrepancy register, multiples and the Fig. 2/4/5 analogues - #8

WP7: the integrated loop, discrepancy register, multiples and the Fig. 2/4/5 analogues#8
ingehap merged 1 commit into
mainfrom
claude/blissful-gauss-3y1u5w

ingehap commented Oct 14, 2026

Uh oh!

Uh oh!

Copilot AI left a comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

		reg.check("VSP first-break residual after calibration", 1e3 * np.sqrt(np.nanmean(d1 ** 2)), 1.0,
		"<=", "ms")

		curves = {k: mid(_at(v, n)) for k, v in (("eps", eps), ("delta", delta), ("gamma", gamma),
		("qp", qp), ("qs", qs))}

Conversation

ingehap commented Oct 14, 2026

Summary

New code

Results (synthetic well)

Deviations from the plan (documented in the plan's as-built notes)

Test plan

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟡 Changes recommended

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants