Skip to content

Checking results

Results are large; never load one to look at it. moraine gives three ways to check a result without a plot script, and the workflow guides say what to expect from each step.

Pyramids

A pyramid is a multi resolution copy of a raster (ras-pyramid) or of a point cloud rendered on a grid (pc-pyramid), stored next to the result (contract). The example pipelines build pyramids for the results worth checking, and moraine run saves a PNG of every pyramid a step makes in WORK/.moraine/<file name>/quicklook/.

moraine ras-pyramid --ras ps/ras_adi.zarr --out_dir ps/ras_adi_pyramid
moraine pc-pyramid --pc pc/pc_ph.zarr --out_dir pc/pc_ph_pyramid --x pc/pc_e.zarr --y pc/pc_n.zarr --ras_resolution 20

moraine info: shape, chunks and statistics

moraine info ps/ras_adi.zarr ps/ras_adi_pyramid

  ps/ras_adi.zarr: float32 (981, 4160) chunks (1000, 1000)
  ps/ras_adi_pyramid: float32 (981, 4160) raster pyramid, 10 levels
    stats_level=0, nan_fraction=0.40655, min=0.0335447, max=4.39345, mean=0.537522, std=0.0957743, p01=0.255305, p50=0.541975, p99=0.786118

For a plain array nothing is read but the metadata. For a pyramid the statistics come from a coarse level (nan_fraction, min, max, mean, std, p01, p50, p99; the amplitude for complex data, true_fraction for booleans) and warnings name anomalies: all values nan, infinite or constant values. With --json the same numbers are one JSON object (contract).

moraine quicklook: a picture

moraine quicklook ps/ras_temp_coh_pyramid -o tcoh.png
moraine quicklook raw/rslc_pyramid --show intf_seq --index 5 -o intf_5.png
moraine quicklook raw/rslc_pyramid --show intf_all --index 0 91 --extent 1400,0,2600,600 -o zoom.png

--show chooses what to draw of a stack: phase (default for complex data), intf_0 (interferogram with the first image), intf_seq (the I-th sequential interferogram), intf_all (any pair ref sec), coh, coh_abs. Phases use a cyclic colour map over (-π, π]; other values viridis over the 1-99 % range. --extent draws a part of the scene (west,south,east,north degrees on a map, range_min,azimuth_min,range_max,azimuth_max pixels on the radar grid): a smaller part is drawn from a finer level, down to single pixels or points; the title gives the extent and the level. Look at the whole scene first, then zoom into what looks wrong.

What to look for: fringes should be continuous, noise should sit where the coherence is low, selected points should cover the stable areas (towns, rock), unwrapped phase should have no isolated 2π jumps between neighbouring areas.

moraine view: interactive maps in a notebook

moraine view ps/ras_adi_pyramid pc/pc_ph_pyramid -o view.ipynb --show intf_seq --dates raw/meta.toml

writes a notebook with one map per pyramid: zoom and pan load details from the pyramid levels, sliders choose the image, a click plots the time series of a pixel or point, a double click makes it the reference, polygons drawn on the map are saved for polygon-mask. The notebook holds only paths and runs on the data machine, so it works over any Jupyter or VS Code connection without a server or port forwarding (decision 0018).

The same views are available in Python: mc.view(data) takes a pyramid, a raster array or point data (x=, y=); show= chooses what to show of a stack ('intf_seq', ... or a function lambda v, ref, sec: ...), a * b overlays views, a + b puts them side by side, repr(v) describes a view in text and v.png('out.png', ...) saves an image. See moraine.cli in Python.

Expected ranges

Each workflow guide has a table of the values of the sample data and the sane ranges, e.g. for the PS candidates: amplitude dispersion p50 roughly 0.3 to 1, temporal coherence between 0 and 1, a few percent to about 30 % of the pixels as candidates. Compare moraine info with the table after every run and look at the quicklooks before continuing; a value outside the range usually means a wrong parameter or input, not a different site.