Quality layers and masking¶

Every DISP granule carries quality layers next to displacement. This example looks at them, compares the "recommended" and "strict" mask presets, and summarizes how often each pixel passes over time.

In [1]:

import tempfile
from pathlib import Path

import matplotlib.pyplot as plt

import disp_xr as dxr
from disp_xr.tutorial import make_disp_stack

folder = Path(tempfile.mkdtemp()) / "granules"
make_disp_stack(folder, "two_ministacks", shape=(128, 128))
ds = dxr.open_disp(folder)
print([name for name in ds.data_vars if name != "displacement"])

import tempfile from pathlib import Path import matplotlib.pyplot as plt import disp_xr as dxr from disp_xr.tutorial import make_disp_stack folder = Path(tempfile.mkdtemp()) / "granules" make_disp_stack(folder, "two_ministacks", shape=(128, 128)) ds = dxr.open_disp(folder) print([name for name in ds.data_vars if name != "displacement"])

['short_wavelength_displacement', 'estimated_phase_quality', 'shp_counts', 'timeseries_inversion_residuals', 'temporal_coherence', 'phase_similarity', 'water_mask', 'connected_component_labels', 'persistent_scatterer_mask', 'recommended_mask']

The layers of one epoch¶

recommended_mask is 1 where the processor recommends the pixel; water, low temporal coherence and unreliable unwrapping regions are 0.

In [2]:

epoch = ds.isel(time=-1)
fig, axes = plt.subplots(1, 3, figsize=(12, 3.6), layout="constrained")
for ax, name in zip(axes, ["recommended_mask", "temporal_coherence", "water_mask"]):
    epoch[name].plot.imshow(ax=ax, add_labels=False)
    ax.set_title(name, loc="left", fontsize=10)
    ax.set_aspect("equal")
plt.show()

epoch = ds.isel(time=-1) fig, axes = plt.subplots(1, 3, figsize=(12, 3.6), layout="constrained") for ax, name in zip(axes, ["recommended_mask", "temporal_coherence", "water_mask"]): epoch[name].plot.imshow(ax=ax, add_labels=False) ax.set_title(name, loc="left", fontsize=10) ax.set_aspect("equal") plt.show()

Mask presets¶

mask sets displacement to NaN where a pixel fails. "strict" adds the connected-component, water and temporal-coherence criteria to the recommended mask. Criteria can also be given one by one, for example dxr.mask(ds, None, temporal_coherence=0.5). In this synthetic frame the recommended mask already excludes everything "strict" adds. A coherence threshold alone keeps more: it drops the low-coherence corner (0.3) but not the water columns, which the recommended mask removes.

In [3]:

for label, masked in [
    ("recommended", dxr.mask(ds, "recommended")),
    ("strict", dxr.mask(ds, "strict")),
    ("coherence > 0.5", dxr.mask(ds, None, temporal_coherence=0.5)),
]:
    kept = masked.displacement.notnull().mean().compute()
    print(f"{label:>17}: {float(kept):.1%} of pixel-epochs kept")

for label, masked in [ ("recommended", dxr.mask(ds, "recommended")), ("strict", dxr.mask(ds, "strict")), ("coherence > 0.5", dxr.mask(ds, None, temporal_coherence=0.5)), ]: kept = masked.displacement.notnull().mean().compute() print(f"{label:>17}: {float(kept):.1%} of pixel-epochs kept")

      recommended: 96.4% of pixel-epochs kept
           strict: 96.4% of pixel-epochs kept
  coherence > 0.5: 99.6% of pixel-epochs kept

Over time¶

quality_summary reduces over time: the share of epochs each pixel was masked, the harmonic mean of temporal coherence (which weights bad periods more than a plain mean), and whether the pixel is reliable (recommended in at least 90% of epochs).

In [4]:

summary = dxr.quality_summary(ds).compute()
fig, axes = plt.subplots(1, 3, figsize=(12, 3.6), layout="constrained")
for ax, name in zip(axes, ["pct_masked", "temporal_coherence_hmean", "reliable"]):
    summary[name].plot.imshow(ax=ax, add_labels=False)
    ax.set_title(name, loc="left", fontsize=10)
    ax.set_aspect("equal")
plt.show()
print(f"reliable pixels: {float(summary.reliable.mean()):.1%}")

summary = dxr.quality_summary(ds).compute() fig, axes = plt.subplots(1, 3, figsize=(12, 3.6), layout="constrained") for ax, name in zip(axes, ["pct_masked", "temporal_coherence_hmean", "reliable"]): summary[name].plot.imshow(ax=ax, add_labels=False) ax.set_title(name, loc="left", fontsize=10) ax.set_aspect("equal") plt.show() print(f"reliable pixels: {float(summary.reliable.mean()):.1%}")

reliable pixels: 96.4%