Skip to content

Workflows

The processing chain of moraine as five verified pipeline files (decisions 0009, 0033): each one is run end to end on the sample data set before it changes, and the numbers and figures in its guide come from that run. The sample data set is Campi Flegrei: Sentinel-1 descending track 22, 92 dates from 2019-01-02 to 2021-12-29, 981 x 4160 pixels at full resolution, the uplifting caldera west of Naples (data/CampiFlegrei/README.md in the repository says how the GAMMA results were made). The [vars] defaults of the files are its values; for other data give yours with --var. The files share one working directory and read each other's outputs, so run them in order:

moraine run examples/01_load.toml --workdir WORK --var gamma=/path/to/gamma --var reference=YYYYMMDD --var geo=YYYYMMDD
moraine run examples/02_ps.toml --workdir WORK --var shape=NLINES,WIDTH
moraine run examples/03_ds.toml --workdir WORK --var shape=NLINES,WIDTH
moraine run examples/04_refine.toml --workdir WORK
moraine run examples/05_unwrap.toml --workdir WORK --var shape=NLINES,WIDTH --var range_pixel_spacing=... --var azimuth_pixel_spacing=...
pipeline guide what it does outputs
examples/01_load.toml 01 Load GAMMA results GAMMA results to zarr, coordinates, web mercator WORK/raw/
examples/02_ps.toml 02 PS candidates amplitude dispersion + Noise2Fringe temporal coherence WORK/ps/
examples/03_ds.toml 03 DS processing SHP, DS candidates, coherence, phase linking, DS selection WORK/ds/
examples/04_refine.toml 04 Merge and refine PS + DS, Noise2Fringe Transformer, temporal coherence WORK/pc/
examples/05_unwrap.toml 05 Phase unwrapping minimum cost flow on a Delaunay network WORK/unw/

To change a parameter, copy the file into the working directory and edit the copy, or override a [vars] value with --var name=value. A rerun skips the steps whose arguments and inputs did not change and reruns everything downstream of a change; moraine run FILE --dry-run shows the plan, moraine status FILE the state. A failing step prints its error and log; fix the cause and run the file again. The file format is the pipeline files contract; moraine run FILE --json prints one JSON object with the plan, the step records and the summaries of the outputs (JSON output).

Each guide lists the steps, the parameters to tune, the expected results on the sample data set with the figures of that run, and what to do when they are off. The guides are the same files the AI agents read; the recorded session ran exactly these five files on the sample data set.