WP8: SEG-Y and LAS adapters, the iwt command, and the critical assessment by ingehap · Pull Request #9 · ingehap/ES_PHYS_P011

ingehap · 2026-10-15T06:31:03Z

Summary

Work package 8 of docs/IMPLEMENTATION_PLAN.md, the last one. It adds three things:

file adapters, so recorded data can replace the synthetic set;
the iwt command, which runs the ZO-VSP → model → tie chain on files;
the README's consolidated assessment of the article.

New code

iwt/io.py:
- SEG-Y in numpy, revisions 1 and 2 with fixed-length traces: IBM, IEEE and integer samples, both byte orders, EBCDIC or ASCII text, and scaled elevation and coordinate words. The core needs no new dependency;
- ZO-VSP and CMP-gather files, with a configurable receiver-depth word, sign and datum shift;
- LAS through lasio: common mnemonics; us/ft, us/m, g/cm³ and kg/m³; depth from a TVDSS curve or MD − KB; missing samples dropped; optional resampling;
- CSV logs, layer models (with Q and labels) and time-depth tables.
iwt/vsp/processing.py: the WP3 ZO-VSP chain in one call (picks, separation, 30 Hz times, interval velocities, t*, interval Q, corridor stack). The stratigraphic synthetic is driven by the recorded first arrival.
iwt/cli.py: iwt demo | vsp | model | tie, plus python -m iwt and a console entry point.
tests/test_io_cli.py (14 tests):
- IBM floats against the textbook value −118.625 = 0xC276A000;
- the byte layout against fixed byte positions;
- segyio in both directions (segyio added to the dev extras only; wheels exist for cp39 and cp313);
- LAS unit and depth handling;
- the processing chain on a model with known Q;
- the command line end to end.
README: a command-line section, WP8 results and a real-data guide. It also adds the article's claims tested on the synthetic well, the consolidated critical assessment and the errata (one new: the article does not give the inputs of its sonic-to-seismic correction).
Plan: WP8 marked done, with as-built notes. Version 0.8.0.

Findings

Intrinsic Q depends on the wavelet driving the stratigraphic synthetic:
- with the recorded first arrival: 45.5/59.0/120/72 (true 45.5/58.8/80/70);
- with a Ricker at the first arrivals' peak frequency (47 Hz, raised by the ghost): the middle block is 20 % high.
The recorded arrival beats even the true source wavelet.
iwt model: the model's times are within 0.83 ms RMS of the VSP. What remains is the bias of the reference pick plus the ray-theory against effective-medium scale effect.
iwt tie:
- corridor stack 0.944;
- seismic trace 0.950 with Q in the synthetic, and 0.909 with a 16° phase rotation without it;
- an estimated wavelet scores 0.986, but that tie is in-sample and is documented as a diagnostic only.

Not built (documented)

DLIS, deviation surveys, 3C rotation, walkaway picking, and command-line access to walkaway inversion and CMP migration (library only). Whether the Hanssen data are released was not checked.

Test plan

pytest -q -W error::RuntimeWarning -W error::DeprecationWarning (Python 3.13): 168 passed
pytest -q (Python 3.9, without lasio and segyio): 161 passed, 4 skipped
pyflakes iwt tests examples
pytest -W error tests/test_io_cli.py: clean
README quick start runs; pip install -e . gives a working iwt command (--version, every --help)
Demo chain iwt demo → vsp → model → tie reproduces the README numbers

🤖 Generated with Claude Code

https://claude.ai/code/session_019vaUDUPicoFeQaUhfhnoYq

Generated by Claude Code

…ment - iwt.io: numpy SEG-Y reader/writer (rev 1/2, fixed-length traces, IBM/IEEE/int samples, both byte orders, scaled header words), ZO-VSP and gather files, LAS via lasio (mnemonics, us/ft-us/m, g/cc-kg/m3, TVDSS or MD - KB), CSV logs, layer models and time-depth tables - iwt.vsp.processing: the WP3 ZO-VSP chain in one call; the stratigraphic t* is driven by the recorded first arrival (a peak-frequency Ricker biased Q by 20 %) - iwt.cli: iwt demo / vsp / model / tie, python -m iwt, console entry point - tests/test_io_cli.py (14 tests): IBM floats, byte layout, segyio cross-check both ways, LAS units and depths, processing chain, CLI end to end - README: command line, WP8 results, using real data, the article's claims tested, critical assessment, errata; plan marked done; version 0.8.0 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019vaUDUPicoFeQaUhfhnoYq

Copilot

🟡 Changes recommended

Recording-delay, geometry-unit and Q/datum handling can corrupt results, while accepted input cases can cause command failures.

11 open findings

What changed in this PR

Completes WP8 by connecting file-based data to the existing well-tie workflow and documenting the synthetic results and limitations.

Changes:

Adds SEG-Y, LAS and CSV adapters.
Introduces the ZO-VSP processing chain and iwt commands.
Adds interoperability tests and consolidates the article assessment.

File	Description
`tests/test_io_cli.py`	Adapter, processing and CLI tests.
`README.md`	Usage, results and critical assessment.
`pyproject.toml`	CLI entry point, test dependency and version.
`iwt/vsp/processing.py`	Integrated ZO-VSP processing chain.
`iwt/vsp/__init__.py`	Documents the processing module.
`iwt/io.py`	SEG-Y, LAS and CSV adapters.
`iwt/cli.py`	Demo, VSP, model and tie commands.
`iwt/__main__.py`	Enables `python -m iwt`.
`iwt/__init__.py`	Exports adapters and updates version.
`docs/IMPLEMENTATION_PLAN.md`	Records WP8 completion and limitations.

🧠 Review effort: Balanced

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

+    depths = depth_sign * sg.headers[depth_field] + depth_shift
+    order = np.argsort(depths, kind="stable")
+    zs = float(np.median(sg.headers["source_depth"])) if source_depth is None else source_depth
+    xs = float(np.median(np.abs(sg.headers["offset"]))) if source_offset is None else source_offset


+    off = np.abs(sg.headers["offset"]).astype(float)
+    order = np.argsort(off, kind="stable")
+    return Gather(sg.t, off[order], sg.data[order], float(np.median(sg.headers["source_depth"])),
+                  float(np.median(-sg.headers["receiver_elevation"])))


+        syn = Z.synthetic_zovsp(strat_model, vsp.depths, vsp.source_depth, dt, t.size,
+                                source_wavelet(t, down, pk), f_max=f_max, attenuation=False,
+                                source="downgoing", workers=workers)
+        cs_strat = _chain(syn.t, syn.data, f_seismic, band)[4]


+    if args.vsp:
+        vsp = json.loads(Path(args.vsp).read_text())
+        if qp is None:
+            q = vsp["q"]


+            qv = np.where(inv > 0, 1.0 / np.where(inv > 0, inv, 1.0), np.inf)
+            qp = _q_curve(z, q["top"], q["bottom"], qv)
+            note.append(f"{q['kind']} Q from {args.vsp}")
+        dz, tz = np.asarray(vsp["depths"], float), np.asarray(vsp["vertical_time"], float)


+    sg = IO.read_segy(args.seismic)
+    obs = sg.data[args.trace]
+    t, dt = sg.t, sg.dt
+    i0 = int(round(sg.t0 / dt))


+    else:
+        spike = (np.array([0.0, dt]), np.array([1.0, 0.0]))
+        refl = normal_incidence(model, dt, nt, spike, **kw)[1][i0:]
+        tw, w = estimate_wavelet(obs[win], refl[win], dt, args.half_length, args.damping)


+    samples = np.ascontiguousarray(raw[:, 240:]).view(order + kind).reshape(n_traces, ns)
+    data = ibm_to_ieee(samples) if fmt == 1 else samples.astype(np.float64)
+    dt_us = binary["interval"] & 0xFFFF or int(headers["sample_interval"][0])
+    t0 = float(headers["delay"][0]) * 1e-3 if n_traces else 0.0


+    write_segy(path, vsp.data, vsp.dt, dict(
+        receiver_elevation=-vsp.depths, source_depth=np.full(n, vsp.source_depth),
+        offset=np.full(n, vsp.source_offset), field_record=np.ones(n),
+        trace_number=np.arange(1, n + 1)), text=text or "ZERO-OFFSET VSP WRITTEN BY IWT", fmt=fmt)


+    q = A.interval_q(vsp.depths, tv, dts, q_blocks)
+    corridor = None
+    if corridor_peak is not None:
+        dec = Z.deconvolve(t, up, down, tv, ricker(corridor_peak, vsp.dt))


ingehap marked this pull request as ready for review October 15, 2026 06:38

Copilot AI balanced review requested due to automatic review settings October 15, 2026 06:38

ingehap merged commit 5255fbd 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 07:40

View session

Navigation Menu

Repository navigation

WP8: SEG-Y and LAS adapters, the iwt command, and the critical assessment - #9

WP8: SEG-Y and LAS adapters, the iwt command, and the critical assessment#9
ingehap merged 1 commit into
mainfrom
claude/blissful-gauss-3y1u5w

ingehap commented Oct 15, 2026

Uh oh!

Uh oh!

Copilot AI left a comment

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

ingehap commented Oct 15, 2026

Summary

New code

Findings

Not built (documented)

Test plan

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟡 Changes recommended

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants