SAR Interferogram and Coherence

Purpose

SAR Interferogram and Coherence computes spatial and temporal coherence from SAR image pairs, quantifying phase stability for interferometric analysis. High coherence indicates reliable phase information; low coherence suggests volume scattering or temporal decorrelation.

Typical Questions This Tool Helps Answer

• Which areas of the scene have stable coherence suitable for surface displacement mapping or subsidence monitoring?
• Where is temporal or volumetric decorrelation too severe to produce reliable interferometric phase, and should those zones be masked?
• Are the coherence values consistent with the expected land-cover behavior — strong in urban areas, lower in vegetated and agricultural zones?

When To Use This Workflow

Use SAR Interferogram and Coherence when:

• You have a registered (or registerable) SAR pair and need coherence and/or interferogram products
• You need embedded acceptance gates in your deliverable for governance or QA purposes
• You want a compact inline pipeline that handles coregistration and coherence in a single step

For best results, run sar_analysis_readiness on the pair before this tool to confirm scene compatibility.

Background

Coherence quantifies phase stability between two SAR acquisitions:

$$\gamma = \frac{\left|E[s_1 s_2^*]\right|}{\sqrt{E[|s_1|^2]E[|s_2|^2]}}$$

With $\gamma\in[0,1]$, high values indicate stable scattering and reliable interferometric support, while low values suggest decorrelation or poor registration/support.

Physical Decorrelation Components

Observed coherence is often interpreted as the product of multiple effects:

$$\gamma_{obs} \approx \gamma_{geom},\gamma_{temp},\gamma_{vol},\gamma_{snr},\gamma_{proc}$$

Where geometric baseline, temporal change, volumetric scattering, SNR, and processing quality each contribute to coherence loss.

Estimation Trade-Offs

Coherence is estimated over finite windows (or multilook neighborhoods). Larger windows reduce variance but blur fine features; smaller windows preserve detail but increase estimator noise. Production choices should match the scale of the decision task.

Methodological Considerations

• Enforce strong coregistration before coherence estimation to avoid artificial decorrelation.
• Use consistent windowing/multilook settings when comparing scenarios or dates.
• Interpret coherence jointly with land-cover and geometry context; low coherence is not a single-cause diagnosis.

Practical Interpretation Pitfalls

Common errors include using fixed coherence thresholds across heterogeneous terrain, confusing low SNR effects with true temporal change, and over-interpreting isolated high-coherence patches without neighborhood continuity.

Inputs

Parameters

Outputs

Runtime output keys (returned in ToolRunResult.outputs):

Summary JSON Schema (summary)

Top-level keys:

• workflow: "sar_interferogram_coherence"
• schema_version: "1.0.0"
• acceptance_gates:
  • valid_fraction, valid_fraction_threshold, valid_fraction_pass
  • mean_coherence, mean_coherence_threshold, mean_coherence_pass
  • coreg_residual_gate_applicable, coreg_residual_mae_px, coreg_residual_mae_threshold_px, coreg_residual_mae_pass
  • overall_pass (bool)
• summary: rows, cols, valid_fraction, mean_coherence
• coregistration: performed, residual_mae_px, search_best_score
• parameters: echo of all runtime parameter values

Example

import whitebox_workflows as wbw

env = wbw.WbEnvironment()
env.working_directory = "/data/sar_project"

# Pre-aligned pair
result = env.earth_observation_sar.sar_interferogram_coherence(
    reference_sar="s1_20260401_vv.tif",
    moving_sar="outputs/s1_aligned.tif",
    assume_prealigned_pair=True,
    coherence_window=7,
  output_prefix="outputs/coherence_run"
)

import json
s = json.loads(open(result.outputs["summary"]).read())
print(s["acceptance_gates"]["overall_pass"])
print(s["summary"]["mean_coherence"])

Inline Coregistration Example

result = env.earth_observation_sar.sar_interferogram_coherence(
    reference_sar="s1_20260401_vv.tif",
    moving_sar="s1_20260413_vv.tif",
    auto_coregister_pair=True,
    coherence_window=9,
  output_prefix="outputs/coherence_run"
)

References

• Rosen, P. A., et al. (2000). Synthetic aperture radar interferometry. Proc. IEEE, 88(3), 333–382.
• Bamler, R., & Hartl, P. (1998). Synthetic aperture radar interferometry. Inverse Problems, 14(4), R1–R54.
• Touzi, R., et al. (1999). Coherence estimation for SAR imagery. IEEE TGRS, 37(1), 135–149.

Parameter Interaction Notes

• auto_coregister_pair=true and assume_prealigned_pair=true are mutually exclusive — setting both is an error.
• coherence_window and coherence_decimation: large window + decimation > 1 saves runtime at the cost of local spatial detail.
• acceptance_min_mean_coherence should be adjusted for land cover context: vegetated or water-heavy scenes will naturally report lower values.
• write_interferogram=false suppresses the interferogram output — use when only coherence is needed.

QA and Acceptance Criteria

1. Input readiness: use sar_analysis_readiness to confirm scene compatibility first.
2. Acceptance gate: summary["acceptance_gates"]["overall_pass"] must be true before delivering coherence products.
3. Mean coherence plausibility: values < 0.2 are expected over vegetation/water; > 0.7 expected over stable urban surfaces.
4. Valid fraction: if valid_fraction < 0.25, check terrain masking extent and DEM/scene overlap.
5. Visual review: coherence map should show spatial gradients aligned with land cover; uniform blobs indicate artifacts.

Advanced Operational Guidance

• For deformation monitoring, use complex component mode for true interferometric phase.
• Archive summary.json per pair as the canonical processing record.
• Fix coherence_window and output_compression across all pairs in a multi-date stack to ensure comparability.

Implementation Patterns

• InSAR prep: sar_analysis_readiness → sar_coregistration → sar_interferogram_coherence (with assume_prealigned_pair=true)
• Inline pipeline: sar_interferogram_coherence with auto_coregister_pair=true
• Coherence-only: write_interferogram=false to skip interferogram output
• COG delivery: output_layout=cog, output_compression=deflate

Related Tools

• sar_coregistration — dedicated coregistration step for full per-burst QA record
• sar_analysis_readiness — scene pair compatibility validation
• time_series_change_intelligence — multi-date coherence-stack change analysis

Results Delivery Checklist

• acceptance_gates.overall_pass=true confirmed
• coherence raster delivered and visually inspected
• summary.json archived per pair as processing record
• valid_fraction reviewed to confirm masking is not excessive
• HTML report reviewed and attached to project record

Common Questions

Q: What does overall_pass: false mean for my deliverable? A: One or more acceptance gates failed. Check valid_fraction_pass (excessive masking?), mean_coherence_pass (expected for vegetation/water?), and coreg_residual_mae_pass (alignment quality). Each has its own threshold value in the summary for diagnosis.

Q: Mean coherence is 0.12 — is the processing wrong? A: Not necessarily. Values below 0.2 are normal over dense vegetation, agricultural land, or scenes spanning seasonal change. Check valid_fraction to confirm enough valid samples are present, then consider whether low coherence is physically meaningful given the land cover.

Q: Should I run sar_coregistration first, or use auto_coregister_pair=true? A: Use the standalone sar_coregistration tool when you need the per-burst diagnostic transform JSON as a separate QA record. Use auto_coregister_pair=true for a compact pipeline when only the coherence product and summary JSON are needed.

Q: What does coherence_window=7 mean for my spatial resolution? A: Coherence is estimated over a 7×7 pixel neighborhood. Larger windows produce smoother maps at the cost of spatial detail. For Sentinel-1 IW (10 m pixels), the effective resolution of the coherence estimate is ~70 m.