Find your next learning step
Wrapped InSAR phase records a value modulo one cycle. Recovering a continuous field is simple along a clean line but becomes path-dependent around noise, gaps and real discontinuities in a two-dimensional image.
Open the chapter contents · Remote Sensing root · Gold Standard of Education · How X Works
Chapter contents
Choose the route closest to the question, or read from Chapter 1 for the complete argument. The chapters remain open on the page; these grouped routes are an index.
Route 1 · Build the wrapped problem
Route 2 · Find and contain residues
Route 3 · Handle components and jumps
Route 4 · Validate the result
Route 5 · Check and continue
CHAPTER 1 OF 12 · BUILD THE WRAPPED PROBLEM
1. Wrapped phase hides whole cycles
An interferogram records phase modulo 2π. Values separated by any integer number of complete cycles are displayed as the same wrapped phase. Phase unwrapping tries to assign an integer cycle count to each valid pixel so neighbouring differences form a physically plausible continuous field.
The task is not equivalent to adding 2π whenever a displayed colour jumps. Noise, layover, shadow, water, vegetation change and large deformation gradients can make the local clues inconsistent. In two dimensions, different integration paths may imply different answers.
Residues mark local inconsistency in the wrapped phase. Connected-component labels mark regions the algorithm could unwrap internally but could not tie together through reliable data. Both are evidence about what the unwrapped product can and cannot support.
CHAPTER 2 OF 12 · BUILD THE WRAPPED PROBLEM
2. One-dimensional continuity is the simple case
Suppose adjacent wrapped phases are 170° and −170°. Direct subtraction gives −340°, but wrapping that difference into the interval (−180°,180°] gives +20°. If the true field is locally smooth, the second pixel is interpreted as 190°, not −170°.
Repeating this rule along one unbranched line constructs a continuous sequence, provided each true neighbouring change is below half a cycle and the data are reliable. A true step larger than 180° is then aliased to a smaller wrapped difference.
Two-dimensional images add loops. The unwrapped value at one pixel should not depend on whether it was reached from the north, south, east or west. Noise can violate that path-independence condition.
CHAPTER 3 OF 12 · FIND AND CONTAIN RESIDUES
3. A residue is a loop that fails to close
Take a small pixel loop and compute each wrapped edge difference in a consistent direction. Consider four increments of +100°, +100°, +100° and +60°. Their sum is +360°, one complete positive cycle. A genuinely single-valued continuous phase would return to its starting value with sum zero; the wrapped observations contain a positive residue inside the loop.
| Directed edge | Wrapped increment | Running sum |
|---|---|---|
| 1 | +100° | +100° |
| 2 | +100° | +200° |
| 3 | +100° | +300° |
| 4 | +60° | +360° |
A sum of −360° gives a negative residue. Residues often occur near low coherence or steep gradients, but they locate inconsistency; they do not name its physical cause.
CHAPTER 4 OF 12 · FIND AND CONTAIN RESIDUES
4. Branch cuts control which paths are allowed
Path-following algorithms may connect positive and negative residues with branch cuts or connect a residue to the data boundary. The unwrapping path is not allowed to cross those cuts. Within the remaining region, integration avoids loops whose wrapped sum is non-zero.
This does not erase the bad data. It makes a particular consistent solution possible by declaring where continuity will not be enforced. Different residue-pairing choices can produce different error propagation when the scene is noisy or residue charges are unbalanced.
Minimum-cost-flow and statistical-cost algorithms express the problem differently, but every method relies on assumptions about smoothness, reliability and allowable discontinuities. The algorithm name alone does not guarantee a correct absolute phase field.
CHAPTER 5 OF 12 · FIND AND CONTAIN RESIDUES
5. Connected components carry independent integer offsets
A coherence mask or data gap may split valid pixels into disconnected islands. An algorithm can unwrap each island internally, yet no reliable path crosses the gap to establish how many whole 2π cycles separate one island from another. Products often store a connected-components raster to expose these independently unwrapped regions.
| Product element | What it tells you | What it cannot settle alone |
|---|---|---|
| Unwrapped phase | Continuous solution within the chosen model | Absolute integer offset across disconnected regions |
| Coherence | Similarity and phase stability indicator | Correctness of every high-coherence pixel |
| Connected-component label | Pixels linked in one unwrapping solution | Which offset aligns separate components |
| Residue map | Local loop inconsistency | Whether the cause is noise or real discontinuity |
Subtracting values from different labels as if they shared one datum can create a false displacement step of whole cycles.
CHAPTER 6 OF 12 · HANDLE COMPONENTS AND JUMPS
6. Changed context: a real discontinuity can resemble an error
Across a surface rupture, landslide edge or sharp subsidence boundary, the physical phase may genuinely change too rapidly for neighbouring wrapped samples to reveal the correct cycle count. A smoothness-driven algorithm may bridge the discontinuity incorrectly, cut it, or divide the scene into components.
Conversely, a long artificial phase ramp can arise from an unwrapping error that propagated away from a noisy patch. Context decides: compare the suspected boundary with amplitude imagery, geology, coherence, acquisition geometry and independent displacement observations.
The transfer rule is important: “discontinuous” does not automatically mean wrong, and “visually smooth” does not automatically mean right. The measurement should preserve plausible physical breaks without turning noise into deformation.
CHAPTER 7 OF 12 · HANDLE COMPONENTS AND JUMPS
7. Coherence guides trust but is not a proof
Low coherence weakens the phase relationship between acquisitions, increasing residue density and unwrapping ambiguity. Masking those pixels can prevent errors from spreading. Yet a mask that is too aggressive fragments the scene, creating many components whose relative offsets are unknown.
High coherence is also not sufficient. A steep deformation gradient can be coherent but undersampled. Atmospheric delay can be spatially smooth and coherent. Coregistration errors or topographic residuals may remain systematic.
Choose thresholds with the decision scale in mind, document morphological cleaning or multilooking, and retain the component labels. A polished unwrapped map without its reliability layers removes the evidence needed for interpretation.
CHAPTER 8 OF 12 · HANDLE COMPONENTS AND JUMPS
8. Reference points fix one constant, not every component
Interferometric phase is relative. Analysts often choose a stable reference pixel or region and set its phase or displacement to zero. That fixes one additive constant within the connected component containing the reference.
It does not automatically anchor disconnected components. Each can carry an independent integer-cycle ambiguity and sometimes an independent offset introduced by processing. External control, overlap with another observation, or a defensible physical constraint may be needed to align them.
Reference selection also matters when the chosen point moves, lies in atmospheric structure or sits near an unwrapping boundary. Report the reference location, component and rationale rather than presenting displacement as absolute.
CHAPTER 9 OF 12 · VALIDATE THE RESULT
9. Validate with loops, paths and independent observations
Inspect residue density and whether residues cluster around masked areas or the signal of interest. Compare results from different unwrapping parameters or algorithms. Test whether alternate paths within a component give consistent phase change, and examine jumps at component boundaries.
Use temporal networks when available: closure phase around acquisition triangles can reveal inconsistencies that a single interferogram cannot. Compare line-of-sight displacement with GNSS, levelling, corner reflectors or physically constrained models, allowing for different spatial and temporal sampling.
Validation should target the claimed measurement. A regional mean may tolerate isolated defects; a fault-slip profile crossing a component boundary may not. Keep the wrapped interferogram and quality layers alongside the derived displacement.
CHAPTER 10 OF 12 · VALIDATE THE RESULT
10. Geometry converts a cycle error into a distance error
After unwrapping, line-of-sight displacement is commonly proportional to phase change, with sign depending on the processing convention. A one-cycle phase error therefore becomes a fixed displacement jump tied to radar wavelength. For a wavelength of 5.6 cm, one 2π cycle corresponds to 2.8 cm of line-of-sight path change under the usual two-way geometry.
That number can be operationally large. A component misaligned by three cycles would shift the interpreted displacement by 8.4 cm. The arithmetic is exact only after checking the product’s phase sign and displacement convention.
Do not convert to vertical motion without a geometry model. InSAR primarily measures change along the radar line of sight, and horizontal motion can contribute strongly.
What phase unwrapping cannot recover
If the true phase changes by more than half a cycle between reliable samples and no additional information exists, the wrapped data alone may not determine the missing integer. Complete decorrelation removes phase information rather than merely hiding it. Water and shadow gaps may permanently separate components.
Unwrapping cannot correct atmospheric delay, orbital ramps, topographic model error or deformation that occurred outside the acquisition interval. Nor does it distinguish a correct mathematical unwrap from the wrong physical attribution.
State excluded areas and unresolved components. A smaller defensible map is more useful than a seamless surface assembled from unsupported offsets.
CHAPTER 11 OF 12 · CHECK AND CONTINUE
11. Questions readers should ask
Is every 2π jump an unwrapping error?
No. Wrapped phase naturally cycles. An error is an incorrect integer assignment in the inferred continuous field.
Can two connected components be compared directly?
Not unless their relative offset has been established by overlap, external evidence or a justified constraint.
Do residues always occur in low coherence?
They are common there, but steep real gradients and sampling limits can also generate inconsistency.
Why retain the wrapped phase?
It lets reviewers inspect the original modulo-2π evidence instead of trusting only the derived surface.
CHAPTER 12 OF 12 · CHECK AND CONTINUE
12. Continue through the verified remote-sensing ecosystem
Begin with the Remote Sensing root, the Gold Standard of Education and the How X Works hub. Read Interferometric Coherence, Decorrelation and Phase Reliability before treating a phase value as stable, and use Co-Registration Offsets and False Change at a Boundary to separate alignment artefacts from real edges.
For measurement geometry and scaling, continue to Incidence Angle, SAR Backscatter and False Change Across Orbits and SAR Backscatter Decibels and False Change Claims. For missing observations, see Revisit Gaps, Cloud Cover and Missed Short Events.
Authoritative technical reading: the European Space Agency’s InSAR Principles processing and interpretation guide explains residues, branch cuts and unwrapping assumptions. The Alaska Satellite Facility’s burst InSAR product guide documents connected-component files as contiguous unwrapped regions. Use both layers when deciding where a displacement comparison is supported.
Top · Contents · Previous · How X Works