microstructure-tpw

Python tools for reading Rockland Scientific microprofiler data and computing turbulent dissipation rates from VMP (Vertical Microstructure Profiler) and MicroRider instruments.

Overview

microstructure-tpw provides a complete processing pipeline for ocean turbulence measurements from Rockland Scientific instruments equipped with shear probes and fast thermistors (FP07). The package reads proprietary .p binary data files, converts channels to physical units, detects profiles, and computes both the rate of dissipation of turbulent kinetic energy (epsilon) and the rate of dissipation of thermal variance (chi), following the methods described in the Rockland Scientific ODAS MATLAB Library and associated Technical Notes.

• Epsilon (TKE dissipation rate) from shear probe spectra (detailed mathematics), including Goodman coherent noise removal, Nasmyth spectrum fitting, and Macoun & Lueck wavenumber correction.

• Chi (thermal variance dissipation rate) from FP07 thermistor spectra (detailed mathematics), including Batchelor/Kraichnan spectrum models, FP07 transfer function correction, and spectral fitting (iterative Peterson & Fer 2014, the default, or MLE).

The package is organized into six subpackages under odas_tpw:

• rsi — Rockland Scientific instrument I/O, NetCDF conversion, profiles, epsilon/chi orchestration
• chi — Chi (thermal variance dissipation) calculation
• scor160 — ATOMIX shear-probe benchmark processing and shared physics modules
• perturb — Full campaign processing pipeline (trim, merge, calibrate, compute, bin)
• processing — Instrument-agnostic profile processing (top trim, bottom-crash, CT alignment, multi-probe epsilon combining)

Installation

pip install -e ".[dev]"    # editable install with dev dependencies
pip install .              # standard install

See docs/installation.md for more options.

Quick Start

Several CLIs ship with the package (rsi-tpw, perturb, perturb-plot, perturb-diag, scor160-tpw); the two primary ones, for different workflows:

• rsi-tpw — single-file or short batch processing on the command line.
• perturb — full campaign pipeline driven by a YAML config file (trim → merge → profiles → epsilon → chi → bin → combo), with parallel workers and stage-aware logging. Use this for cruise-scale runs.

The two CLIs use different spectral defaults (rsi-tpw: fft_length=1024 with a 4096-sample dissipation window; perturb: duration-based fft_sec=1.0 — one second, resolved per instrument sampling rate, = 512 samples on a 512-Hz VMP-250 — for both epsilon and chi), so their outputs differ in vertical resolution and noise behavior.

rsi-tpw (single files / short batches)

# Full pipeline: .p files → epsilon → chi
rsi-tpw pipeline VMP/*.p -o results/

# Or run individual stages
rsi-tpw eps VMP/*.p -o epsilon/                       # writes into epsilon/eps_00/
rsi-tpw chi VMP/*.p --epsilon-dir epsilon/ -o chi/    # finds eps_00/ automatically

# Pre-deployment bench test (dummy probes, instrument at rest): raw-count
# figures + spectra + an auto-evaluated Rockland checklist (see docs/rsi-tpw/bench.md)
rsi-tpw bench VMP/*.p -o bench/

Note: eps writes into a hash-tracked subdirectory (epsilon/eps_00/, see output directories). The chi command searches the given --epsilon-dir and its eps_* subdirectories (most recently modified first), and matches both single-profile ({stem}_eps.nc) and per-profile ({stem}_prof001_eps.nc, ...) file names, concatenating the latter along time. Only when no matching epsilon files exist anywhere does chi fall back to Method 2 for that file, with a console warning.

from odas_tpw.rsi.pipeline import run_pipeline
from pathlib import Path

# Full pipeline: .p → profiles → epsilon → chi → binning → combine
run_pipeline([Path("VMP/file.p")], Path("results/"))

# Or use the modular API
from odas_tpw.rsi.dissipation import compute_diss_file
from odas_tpw.rsi.chi_io import compute_chi_file

compute_diss_file("VMP/file.p", "epsilon/")
compute_chi_file("VMP/file.p", "chi/")

Note: get_diss() and get_chi() still work for backward compatibility but are deprecated in favor of run_pipeline() or the modular compute_diss_file() / compute_chi_file() functions.

perturb (campaign-scale batch pipeline)

# Write a template config you can edit (cruise paths, GPS file,
# instrument-specific overrides, parallelism, etc.).
perturb init my-cruise.yaml

# Run the full pipeline — outputs land in <output_root>/{profiles_NN,
# diss_NN, chi_NN, ctd_NN, *_binned_NN, combo}, with a per-run log under
# <output_root>/logs/.
perturb run -c my-cruise.yaml

# Or run individual stages against the same config:
perturb trim -c my-cruise.yaml
perturb profiles -c my-cruise.yaml
perturb diss     -c my-cruise.yaml
perturb chi      -c my-cruise.yaml
perturb bin      -c my-cruise.yaml
perturb combo    -c my-cruise.yaml

See docs/perturb/pipeline.md for the full stage list and docs/perturb/configuration.md for the config schema.

Documentation

rsi-tpw (science library)

perturb (batch pipeline)

scor160 (ATOMIX benchmark)

Shared

References

This package is a Python implementation derived from the Rockland Scientific ODAS MATLAB Library (v4.5.1) and associated Technical Notes (TN-028, TN-051, TN-061).

See docs/bibliography.md for the full list of references.

Testing

python -m pytest                          # run all tests
python -m pytest tests/test_epsilon.py    # epsilon pipeline tests only

Development

This project was developed in collaboration with Claude Code.

Validation and known limitations

Epsilon is validated against the ATOMIX shear-probe benchmark (Fer et al. 2024) — 1416 spectra from six benchmark records spanning 10⁻¹⁰ to 10⁻³ W/kg, with a log10 bias below 0.09, log10 RMSD of 0.008–0.096, and 100% of estimates within half a decade of the published reference.

Chi has no equivalent benchmark and its absolute level is not anchored: report it as a relative field with a systematic of order 1.5×. Note also that a chi fom near 1.0 does not mean the spectral fit is good — it is an integrated ratio, and cancellation across the band hides a factor-of-two structured residual.

Before trusting an absolute level on your own data, audit the calibration constants (rsi-tpw sensors --cal-dir, --diff-gain): they, not the estimator, are usually the dominant uncertainty.

Full detail, and suggested wording for a manuscript, in docs/validation_and_limitations.md.

Citation

If you use this software in published work, please cite it. Citation metadata lives in CITATION.cff — GitHub renders a "Cite this repository" button from it (BibTeX/APA export). Each tagged release is archived on Zenodo with a DOI.

Cite the version DOI of the release that produced your numbers — the concept DOI resolves to whatever release is newest, which is not what you ran. Version DOIs are listed in CITATION.cff; v0.3.0 is 10.5281/zenodo.21366143. The concept (all-versions) DOI 10.5281/zenodo.21366142 is the right one to cite when referring to the software in general rather than to a specific result.

License

GPLv3 — see LICENSE for details.