3.7. Exports And Reproducibility#

The desktop app produces several kinds of outputs: figures, re-exported EDI files, station metadata tables, pipeline configuration files, solver input folders, interpretation products, logs, and the desktop session state. Treat these as different artifacts. A figure explains a decision; an EDI folder stores a processed data product; a JSON configuration makes the workflow repeatable.

The guiding rule is simple: every exported result should carry enough context for another user to understand what data state it came from and which settings produced it.

3.7.1. Think In Data States#

Before exporting, name the data state you are working from. Most confusion in desktop projects comes from mixing raw, corrected, recomputed, and pipeline outputs in the same folder or figure caption.

Data State

Meaning

Export Habit

raw

Original files loaded from the field or source archive.

Keep immutable; do not overwrite or mix with processed EDIs.

qc_reviewed

Raw data with QC figures and notes, but no data values changed.

Save figures, station inventory, and review notes.

corrected

Active survey after a correction was committed to the desktop session.

Export corrected EDIs plus before/after figures and parameters.

recomputed

EDIs rewritten by the recompute tool after rotation, trimming, filling, or impedance-derived value updates.

Save the recompute manifest and reload the output before using it.

pipeline

Output from an automated processing chain.

Save pipeline JSON, logs, preview figures, and exported EDIs together.

inversion_input

Files prepared for an external or internal inversion run.

Save input files with the QC/correction evidence that defines them.

Use these state names in folder names and captions. A file named L30_corrected_static_shift_ama_rhoa_phase.png is easier to audit than profile_final.png.

3.7.2. Export Types#

Output

Desktop Surface

Typical Use

Figure files

More > Export Figure, panel export buttons

Save the current map, profile, QC, correction, forward, or inversion plot for reports and processing notes.

Batch figures

Tools > Batch Export Plots…

Save every open canvas figure into one folder.

Survey metadata

Tools > Format Converter…

Export station metadata to CSV or JSON.

Re-exported EDI files

Tools > Format Converter…, correction tools, pipeline export

Save a processed or converted survey as station EDI files.

Recomputed EDIs

Tools > Recompute EDIs…

Rotate tensors, trim frequency range, fill missing values, recompute resistivity/phase, and write an optional manifest CSV.

Pipeline configuration

Processing Pipeline > Save config

Preserve method and parameter choices for a repeatable processing chain.

Inversion input/output

Inversion Wizard

Store solver input files, logs, and results in a run-specific working directory.

Interpretation products

Interpretation Studio

Export Oasis Montaj XYZ, LAS, CSV tables, and VTK grids.

Desktop session

File > Save Session

Restore theme, recent files, selected station, window geometry, and preferences.

3.7.3. Figure Export#

Most desktop plots are matplotlib figures. Use More > Export Figure or a panel’s own Export button to save the active figure. The export dialog delegates to Figure.savefig and supports:

Format

Type

Best Use

PNG

Raster, lossless

Screenshots, documentation, quick sharing, and web pages.

TIFF

Raster, high resolution

Journal or report workflows that require high-DPI raster figures.

PDF

Vector

Reports, print, and figures that need sharp text and lines.

SVG

Vector

Web documentation or editing in vector graphics tools.

EPS

Vector, legacy

Older publishing pipelines that still request EPS.

The dialog defaults to 300 DPI, which is a good report setting for raster exports. Use 150 DPI for quick review images, 300 DPI for most reports, and 600 DPI only when a publisher or print workflow needs it. Vector formats are usually better for line plots, rose diagrams, and profile curves because labels remain sharp at any zoom level.

When exporting a figure for a processing decision, put the decision in the filename:

qc/L30_coverage_before_frequency_edit.png
qc/L30_snr_after_noise_filter.png
corrections/L30_static_shift_before_after.pdf
strike/L30_strike_stability_bands.svg

Avoid generic names such as figure1.png. They become untraceable as soon as several windows are open.

3.7.4. Figure Naming Pattern#

A useful figure name usually contains:

<line>_<data-state>_<view>_<component-or-band>_<purpose>.<ext>

Examples:

L30_raw_station_map_geometry_check.png
L30_qc_reviewed_snr_before_filtering.png
L30_corrected_static_shift_ama_before_after.pdf
L30_recomputed_rot30_profile_xy_1s.svg
L30_pipeline_qc_ss_rotate_coverage_after.png

Use short names, but include the part that would otherwise be forgotten: line, data state, view, component or band, and purpose.

3.7.5. Batch Plot Export#

Use Tools > Batch Export Plots… when several panel windows are open and you want to capture the current desktop evidence in one folder. The batch exporter collects visible matplotlib canvases from open panel windows and saves them as <label>.<format>.

Batch export supports PNG, PDF, SVG, EPS, and TIFF. It also lets you choose DPI and whether the background should be transparent. Transparent backgrounds are useful for slide decks, but they can make dark theme labels hard to read on a light page. For documentation and reports, prefer a normal opaque background unless you know the final page color.

Use batch export after a review pass:

  1. Open the map, profile, QC, and correction windows you want to document.

  2. Set each window to the exact plot and station/band selection you want.

  3. Run batch export to a dated folder.

  4. Rename or annotate key figures if the automatic label is not specific enough.

Batch export saves visual evidence. It does not save the processing settings that produced corrected data. Pair it with pipeline JSON, correction notes, or exported EDIs when reproducibility matters.

3.7.6. Minimum Evidence Set#

When a processing decision changes the data state, save at least:

Decision

Minimum Figures

Minimum Machine-Readable Files

Frequency or band selection

Coverage and SNR before/after.

Pipeline JSON or notes listing selected bands.

Static-shift correction

Before/after station curves and pseudosection.

Corrected EDIs or correction parameters.

Rotation

Strike/stability evidence and profile comparison.

Recomputed EDIs plus manifest or pipeline JSON.

Source-effect correction

Diagnostic figure and before/after response comparison.

Corrected EDIs in a clearly named output folder.

Inversion setup

QC summary, profile/map figure, and starting model figure.

Solver input files, run config, and log.

This minimum set is what lets another user understand both what changed and why it changed.

3.7.7. Survey Metadata Export#

Use Tools > Format Converter… to export station metadata from the loaded survey. The converter can write:

  • survey_stations.csv for spreadsheet review and quick station inventory;

  • survey_stations.json for structured downstream tools;

  • EDI re-exports when the loaded station objects expose an EDI writer.

The CSV and JSON metadata include station name, latitude, longitude, frequency count, minimum and maximum period, and whether impedance error data are present. Use these files as an inventory, not as the full MT/CSAMT data product.

Metadata export is useful before sharing a survey because it lets a reviewer check station count, coordinate coverage, and frequency availability without opening every EDI file.

3.7.8. Metadata Review Uses#

Station metadata exports are especially useful for:

  • checking whether all expected stations are present;

  • comparing field sheets with loaded station IDs;

  • reviewing coordinate or elevation problems in a spreadsheet;

  • documenting frequency coverage before processing;

  • sharing a lightweight inventory when the EDI files are too large or private;

  • giving inversion collaborators a quick summary of the profile.

Metadata are not a replacement for EDIs or QC figures. Treat them as the project index.

3.7.9. EDI Data Products#

EDI outputs are the data products that other tools can reload. The desktop can write EDI files from several places:

  • Format Converter can re-export the currently loaded survey when station objects support EDI writing.

  • Data Corrections can export Stratagem-corrected EDIs from the Stratagem correction path.

  • Processing Pipeline writes processed EDIs during its final Export step using pycsamt.emtools.export_edis.

  • Recompute EDIs writes a transformed EDI set with optional tensor rotation, frequency trimming, missing-value fill, and resistivity/phase recomputation.

Always keep raw EDIs separate from corrected or recomputed EDIs. A practical layout is:

raw_edi/
corrected_edi/static_shift_ama/
corrected_edi/source_effect_test/
recomputed_edi/rotated_to_strike/
pipeline/L30_qc_ss_rotate/exported_edi/

After exporting an EDI product, reload it in a fresh desktop session or clear mental state by reopening the folder. Check the station count, map geometry, profile curves, and QC coverage before using the exported product for inversion.

3.7.10. EDI Product Safety Rules#

Use these rules whenever the desktop writes EDI files:

  • never write corrected EDIs into the raw input folder;

  • keep one output folder per correction, recompute, or pipeline attempt;

  • include method names in folder names, not only dates;

  • keep failed or experimental outputs separate from accepted outputs;

  • reload exported EDIs before using them as inversion input;

  • keep the figures and parameter files beside the exported EDIs;

  • do not rename station files in a way that breaks station identity.

An EDI folder is a data product. Once another program consumes it, the folder name and nearby evidence become the processing history.

3.7.11. Recomputed EDIs#

The Recompute EDIs… tool is for controlled data-product generation. It can use the loaded stations or a selected EDI file/folder as input, then write a new EDI set with chosen transformations.

Important options:

  • Rotation angle – rotates impedance and/or tipper components when a strike decision has already been justified.

  • Frequency limits – trims the output to a trusted frequency band.

  • Fill missing values – fills missing data according to the selected strategy; document this choice carefully.

  • Recompute resistivity/phase – updates derived values from impedance.

  • Manifest CSV – writes a station-level record of the operation.

  • Filename template – controls output names, for example {source_stem}.edi.

Use recomputed EDIs when you need a clean, explicit handoff to external software. Do not use this tool as a hidden correction step; save the manifest and the diagnostic figures that justify every transformation.

3.7.12. Recompute Audit Checklist#

Before accepting a recomputed EDI folder:

  • confirm the output folder name states the operation, for example rotated_30deg_trim_1e-3_1e2hz;

  • save the manifest CSV when available;

  • reload the recomputed folder in the desktop;

  • compare station count with the source folder;

  • inspect one early, one middle, and one late station profile;

  • verify frequency limits, rotation, or filled values are visible as expected;

  • keep the source EDIs unchanged and nearby.

If the recomputed folder cannot pass this audit, keep it as an experiment and do not use it as the accepted inversion input.

3.7.13. Pipeline Configurations#

The processing pipeline can save and load its configuration as JSON. The JSON stores each step ID, step name, selected method, and method parameters. It does not replace the exported EDIs; it explains how those EDIs were generated.

Save the pipeline JSON beside the pipeline output:

pipeline/
  L30_qc_ss_rotate.json
  L30_qc_ss_rotate.log
  exported_edi/
  figures/

When you rerun a saved configuration, confirm that step 1 is using the intended input source. A valid JSON file can still produce the wrong output if it is run against a different active survey.

3.7.14. Pipeline Export Package#

A pipeline export is complete when it contains:

  • the pipeline JSON;

  • the run log;

  • exported EDIs, when the final step writes data;

  • preview or summary figures for important steps;

  • a short note naming the input data state;

  • a clear output folder name that includes line and major methods.

For example:

pipeline/
  L30_raw_to_qc_ss_rotate/
    L30_raw_to_qc_ss_rotate.json
    L30_raw_to_qc_ss_rotate.log
    figures/
    exported_edi/
    README.md

The README.md can be short. One or two sentences about input state, major choices, and accepted output are enough to make the folder easier to review later.

3.7.15. Inversion Exports#

The inversion wizard writes into the selected working directory. For external engines such as Occam2D, ModEM, and MARE2DEM, input-file generation is already an export worth preserving. Keep one folder per run attempt:

inversion/
  L30_occam2d_corrected_ss_v01/
    inputs/
    logs/
    results/
    figures/

Use run names that identify line, data state, engine, and attempt number. A folder such as modem_run is not enough once several corrected data states exist.

Before archiving an inversion folder, save:

  • the input files generated by the builder;

  • the solver log or desktop run log;

  • the starting model or forward-model parameters;

  • result plots;

  • the QC and correction figures that define the input data state.

3.7.16. Inversion Run Naming#

Use run names that can survive months of project history:

<line>_<engine>_<data-state>_<model-or-band>_<attempt>

Examples:

L30_occam2d_corrected_ss_v01
L30_mare2dem_recomputed_rot30_v02
L18_modem_pipeline_qc_ss_rotate_v01

Avoid names such as test, final, new_final, or run2. They only make sense on the day they are created.

3.7.17. Interpretation Exports#

The Interpretation Studio exports derived geological products. Available exports include:

  • XYZ for Oasis Montaj style workflows;

  • LAS for station or borehole-style logs;

  • CSV for flat interpretation tables;

  • VTK for gridded model visualization.

These files are downstream interpretation products. Save them under a folder that points back to the inversion run and corrected EDI set that produced them:

interpretation/
  from_L30_occam2d_corrected_ss_v01/
    L30_interpretation.csv
    L30_model.vtk
    station_logs/

Interpretation outputs should not be separated from their parent inversion context. A VTK grid or CSV classification is only meaningful if the reviewer can find the model, inversion run, corrected data state, and assumptions that produced it.

3.7.18. Session And Settings#

Use File > Save Session to persist the desktop session. The session is stored at ~/.pycsamt/session.json and includes theme, recent files, last data directory, selected station, frequency preferences, window geometry, solver paths, default inversion work directory, logging level, map tile provider, and API key setting.

Session state is useful for convenience, but it is not a project archive. It does not guarantee that another user can reconstruct a processing product. For project reproducibility, save exported data, figures, pipeline JSON, and inversion/interpretation folders in the project directory.

Settings profiles are separate JSON snapshots of application/API settings. Use them when you want to move a configuration between machines or preserve a known set of plotting and processing defaults.

3.7.19. What Not To Rely On#

Do not rely on these as the only record of a project:

  • screenshots copied from the screen;

  • the desktop session file alone;

  • recent-file history;

  • a solver result folder without input files;

  • corrected EDIs without before/after figures;

  • pipeline EDIs without the JSON configuration;

  • an interpretation export without its parent inversion run.

These items are useful, but alone they do not explain the processing chain. Pair visual, data, and configuration outputs whenever a result matters.

3.7.21. Export Checklist#

Before sharing or archiving a desktop result, check that you have:

  • raw input EDIs or a link to their immutable location;

  • exported corrected/recomputed EDIs if the active data were changed;

  • QC figures before and after major corrections;

  • strike, dimensionality, or topography figures when they affected modelling;

  • pipeline JSON and logs for automated workflows;

  • inversion input files, logs, and result figures;

  • interpretation exports with a clear parent inversion run;

  • a short note explaining the data state and major assumptions.

This is the difference between “I saved the plot” and “I can reproduce the result.” The second one is the one future you will thank present you for.