Find your next learning step
An InSAR map does not measure motion against an invisible absolute zero. Its values are tied to a chosen spatial and temporal reference, and that choice travels through every reported pixel.
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 · Set the displacement datum
Route 2 · Work the reference bias
Route 3 · Design a stable reference
Route 4 · Test limits and uncertainty
Route 5 · Check and continue
CHAPTER 1 OF 12 · SET THE DISPLACEMENT DATUM
1. Every InSAR displacement field has a datum
Interferometric synthetic aperture radar measures phase differences between acquisitions. After unwrapping and conversion, the displacement field is still relative: a spatial reference point or area and a temporal reference determine the zero. The map does not independently reveal absolute motion at every pixel.
If the reference area moves, its motion is subtracted from the entire scene. A stable-looking reference is therefore an assumption that must be tested, not a decorative processing choice.
Write the reported line-of-sight displacement at pixel x as drep(x,t)=dtrue(x,t)−dref(t), plus residual atmospheric, orbital, topographic and unwrapping errors. The sign convention must be stated because products differ on whether positive means motion toward or away from the satellite.
CHAPTER 2 OF 12 · SET THE DISPLACEMENT DATUM
2. A reference point and a reference area trade noise differently
A single coherent point offers a precise location but can inherit local motion or one bad pixel. An area reference averages several pixels and can reduce uncorrelated noise, yet it may span a deformation gradient or heterogeneous land cover.
The standard error does not necessarily fall as one over the square root of pixel count because atmospheric and processing errors are spatially correlated. Ten neighbouring pixels are not automatically ten independent observations.
| Reference choice | Main advantage | Main risk |
|---|---|---|
| Single persistent scatterer | Clear physical location | Local motion or point failure controls the scene |
| Compact stable area | Robust spatial average | Correlated noise limits averaging gain |
| Large broad area | More samples | May cross a deformation gradient or several regimes |
| External geodetic tie | Physical datum from another sensor | Frame, projection and time matching errors |
CHAPTER 3 OF 12 · WORK THE REFERENCE BIAS
3. Work a moving-reference example
Suppose a target truly moves −18 mm in the adopted line-of-sight sign convention during the interval. The chosen reference area moves −5 mm in the same convention. The reported target displacement is −18−(−5)=−13 mm.
The processing has not lost only five millimetres at the target. It has shifted every reported pixel by +5 mm relative to an absolute frame. A genuinely stable pixel with true displacement 0 would appear to move +5 mm.
If the analyst later switches to a verified zero-motion reference, the target becomes −18 mm and the whole map shifts by −5 mm. Spatial differences within the original map remain the same, but absolute interpretations and comparisons with external data change.
| Location | True displacement | Reference-subtracted result |
|---|---|---|
| Moving reference area | −5 mm | 0 mm by definition |
| Target | −18 mm | −13 mm |
| Stable validation point | 0 mm | +5 mm |
CHAPTER 4 OF 12 · WORK THE REFERENCE BIAS
4. Common-mode removal can remove real broad deformation
Subtracting a reference-area mean can suppress a scene-wide phase offset from orbit or atmosphere. It can also erase a real signal if the reference shares broad tectonic, hydrologic, volcanic or subsidence motion with the rest of the scene.
This is a gauge choice: adding the same time-dependent constant to all true pixels and subtracting it from the reference can leave the relative observations unchanged. Without an external constraint, the common component is not identifiable from the relative field alone.
The correct question is not whether a zero must be chosen. It must. The question is whether the chosen zero matches the claim being made.
CHAPTER 5 OF 12 · WORK THE REFERENCE BIAS
5. Changed context: two studies can disagree by a constant
Study A references a stable bedrock area. Study B references an urban district subsiding at 3 mm per year. Over four years, B’s velocities or cumulative displacements can differ from A’s by a scene-wide 12 mm offset even if both process the interferometric phase correctly.
Transfer check: compare the two maps after subtracting each map’s value at the same validation location. If the residual fields agree, the apparent dispute was largely a datum difference. If spatial gradients remain, look for atmospheric corrections, unwrapping, viewing geometry or time-span differences.
Never average two products until their sign convention, line-of-sight vector, temporal reference and spatial datum have been harmonised.
CHAPTER 6 OF 12 · DESIGN A STABLE REFERENCE
6. Choose a candidate area using physics and data quality
Screen for high coherence, low unwrapping risk, compactness and adequate distance from known deformation. Check geology, infrastructure, groundwater, slope processes and land-cover history. “Far from the target” is not equivalent to stable.
Inspect the candidate through time rather than on one interferogram. A point can be quiet during the calibration interval and drift later. Avoid areas at the edge of coverage or near strong atmospheric gradients where residual ramps can bias the mean.
Document the polygon or point coordinates, pixel-selection rule, weights, rejected dates and the exact operation used to set zero. Reproducibility requires more than naming a neighbourhood.
Use an area statistic that survives outliers
A simple mean is transparent but sensitive to unwrapping errors and isolated bad pixels. A median or robust weighted mean can reduce outlier influence. Coherence weighting may help precision but can bias the reference toward a particular surface type.
First remove or flag pixels that fail quality rules. Then calculate the reference statistic and its time series. Map residuals inside the reference area: a strong internal gradient warns that the polygon is not acting as one stable unit.
Repeat the solution with several reasonable polygons. The spread among them is a reference-choice sensitivity, not ordinary pixel noise, and should be carried into interpretation.
CHAPTER 7 OF 12 · DESIGN A STABLE REFERENCE
7. Tie the relative field to independent observations carefully
GNSS can anchor an InSAR time series, but the sensors observe different quantities. Project the GNSS three-dimensional displacement into the radar line of sight using the correct look vector. Match reference frame, epoch, time span and sign.
The GNSS monument and radar pixels may sample different materials or spatial scales. A roof, road, soil patch and deeply anchored monument can move differently even when close on a map.
Leveling, corner reflectors and stable geological benchmarks can provide additional checks. Use multiple anchors where possible; one external point cannot diagnose a residual spatial ramp.
CHAPTER 8 OF 12 · DESIGN A STABLE REFERENCE
8. Diagnose reference problems in the residual pattern
| Residual pattern after comparison | Plausible cause | Next check |
|---|---|---|
| Nearly constant offset over the scene | Different reference datum | Align at common stable control |
| Smooth long-wavelength ramp | Orbit or atmospheric residual | Fit only with independent justification |
| Offset changes by acquisition date | Moving temporal/spatial reference | Inspect reference-area time series |
| Local disagreement near reference | Local deformation or decorrelation | Move or shrink polygon and reprocess |
Do not automatically remove a ramp because it looks inconvenient. A broad geophysical signal can also be smooth. Use weather models, orbital information, neighbouring tracks and ground data to decide.
CHAPTER 9 OF 12 · TEST LIMITS AND UNCERTAINTY
9. What a stable-looking map cannot prove
Low variance within a reference area does not prove zero absolute motion; the whole area may move coherently. High coherence measures phase consistency, not geological stability. Agreement with one nearby pixel can reflect shared bias.
Reference subtraction does not correct phase unwrapping mistakes, topographic residuals or look-direction ambiguity. It can hide a common offset while leaving those errors intact.
In a time series, changing the reference polygon between dates can manufacture steps. Maintain one documented rule or explicitly transform all epochs into a common datum.
CHAPTER 10 OF 12 · TEST LIMITS AND UNCERTAINTY
10. A practical evidence checklist
Publish the spatial reference, temporal reference, sign convention and line-of-sight vector. Verify the reference area’s internal residuals and coherence through time. Recompute with at least two defensible alternative reference areas and report the change at key targets.
Compare with projected GNSS or another independent datum where available. Check whether disagreement is constant, ramp-like, date-specific or local. Preserve the unreferenced or conventionally referenced intermediate product so later analysts can transform it consistently.
Finally, phrase the result honestly: “13 mm relative to this reference area” is stronger science than “13 mm absolute motion” when no absolute tie exists.
CHAPTER 11 OF 12 · CHECK AND CONTINUE
11. Questions readers should ask
Why must a reference be chosen?
Interferometric phase constrains differences. A spatial and temporal datum fixes an otherwise undetermined common offset.
Does a larger reference area always improve precision?
No. Spatial correlation, deformation gradients and mixed surfaces can prevent the expected averaging gain or introduce bias.
Can I change the reference after processing?
Often yes, by subtracting the chosen new reference time series from every pixel, provided the saved products and conventions are compatible.
Is coherence evidence of stability?
It is evidence of phase consistency. A coherent surface can still move.
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. For viewing geometry, read Line-of-Sight Decomposition, Ill-Conditioned Geometry and North–South Blindness. For phase reliability, continue to Atmospheric Phase Screens, Spatial Correlation and False-Deformation Ramps, Interferometric Coherence, Decorrelation and Phase Reliability and InSAR Phase Unwrapping, Residues and Disconnected Components.
Authoritative reading: the peer-reviewed EZ-InSAR processing paper states that reference-point location strongly influences final InSAR results. NASA’s NISAR science overview explains how repeat-pass SAR reveals surface change. Use these principles with documented reference coordinates and independent ground checks.
Top · Contents · Previous · How X Works