GOLD STANDARD · REMOTE SENSING · HOW X WORKS
Find the decision hidden inside the model
A cast shadow carries geometric information about the object that made it. With the solar direction, image scale and ground geometry declared, shadow displacement can estimate height from one image—but occlusion, slope and sensor perspective can turn simple trigonometry into a biased inversion.
This guide connects the Gold Standard of Remote Sensing, Gold Standard of Education, and How X Works hub.
Chapter contents and reading routes
Choose the route closest to your present task. Every chapter remains expanded, so the article also reads naturally from start to finish.
Orient the shadow geometry
Calculate a height estimate
Audit displacement and occlusion
Transfer and validate
Use evidence and next routes
CHAPTER 1 OF 12 · ORIENT THE GEOMETRY
1. A shadow is a projection, not a ruler laid on the image
The Sun defines a three-dimensional ray direction. An upright object blocks that ray, and the ground receives a displaced shadow. The image records a two-dimensional projection of the object, shadow and terrain through the sensor geometry.
Height can be inferred only after the measured image displacement is converted to ground distance along the solar-azimuth direction. Pixel count alone is not height, and the darkest region beside an object is not automatically its own complete cast shadow.
CHAPTER 2 OF 12 · ORIENT THE GEOMETRY
2. Start with the flat-ground, near-nadir case
For a vertical object of height H on level ground, a complete shadow of ground length L and solar elevation β obeys H=L tan β. Solar zenith angle is 90°−β, so mixing elevation and zenith reverses the trigonometric relationship.
This simple equation is a baseline. Sloping ground changes the receiving plane, off-nadir viewing displaces the visible roof and base, and the building itself may hide part of the shadow when Sun and sensor occupy similar directions.
CHAPTER 3 OF 12 · CALCULATE THE HEIGHT
3. Measure displacement along the Sun direction
Find the object base and the corresponding shadow tip, then project their ground coordinates onto the horizontal direction opposite the solar azimuth. A diagonal image distance must not be multiplied by one pixel size if map pixels are anisotropic or the image is not orthorectified.
Use acquisition-time solar metadata or an independently computed ephemeris. A one-degree azimuth error can pair the base with the wrong shadow edge in a dense scene. Record how the endpoints were chosen.
CHAPTER 4 OF 12 · CALCULATE THE HEIGHT
4. Work one building-height estimate
Suppose an orthorectified image has 0.5 m ground sampling, and the base-to-tip displacement along the shadow direction is 56 pixels. Then L=28 m. With solar elevation β=35°, H=28 tan35°≈19.6 m.
If endpoint placement is uncertain by two pixels in total, L carries about ±1 m, which alone gives height uncertainty of roughly ±0.70 m at the same solar angle. Solar-angle, terrain and registration errors must be added rather than hiding behind the sub-metre arithmetic.
| Input | Worked value | Role |
|---|---|---|
| Pixel displacement | 56 pixels | Measured along shadow direction |
| Ground sampling | 0.5 m/pixel | Converts to L=28 m |
| Solar elevation | 35° | Sets tan β≈0.700 |
| Estimated height | 19.6 m | Flat-ground baseline |
CHAPTER 5 OF 12 · CALCULATE THE HEIGHT
5. See how Sun angle controls sensitivity
At low solar elevation, shadows are long and endpoint errors are a smaller fraction of length, but shadows overlap more objects and terrain. At high elevation, shadows are short and may be hard to segment; one pixel produces a larger fractional height error.
For a fixed one-metre length error, height error is tanβ metres. It is about 0.36 m at 20°, 0.70 m at 35° and 1.73 m at 60°. Scene usability depends on both geometry and visibility, not one preferred angle.
CHAPTER 6 OF 12 · AUDIT THE INVERSION
6. Correct the roof–base–shadow correspondence
In off-nadir imagery, a tall building’s roof appears displaced relative to its ground footprint. Measuring from the visible roof edge to the shadow tip can omit or double-count part of the geometry. Sensor azimuth and elevation determine the correction.
Dense cities add self-occlusion and neighboring shadows. Reject or flag cases where the object’s own shadow cannot be separated. An automated segmenter should output a quality mask, not silently force every footprint to a height.
| Failure mode | Bias mechanism | Practical check |
|---|---|---|
| Roof used as base | Perspective displacement enters L | Use footprint/base geometry |
| Neighbor shadow overlap | Tip belongs to another object | Trace along solar azimuth |
| Sloping ground | Receiving plane is not horizontal | Use local terrain model |
| Dark material mistaken for shadow | Segmentation error | Check spectral and geometric consistency |
CHAPTER 7 OF 12 · AUDIT THE INVERSION
7. Separate image resolution from endpoint certainty
Ground sampling distance describes nominal pixel spacing, not guaranteed boundary accuracy. Blur, resampling, compression, mixed pixels and penumbra can make the usable shadow edge uncertain by several pixels.
Estimate endpoint uncertainty from repeated annotation, segmentation probability or edge spread. A reported height to centimetres is not defensible when the shadow tip is uncertain by a metre.
CHAPTER 8 OF 12 · TRANSFER AND VALIDATE
8. Transfer the geometry to tree height
A tree casts a shadow under the same solar geometry, but the crown is irregular, partly transparent and wind-sensitive. The visible tip may come from one branch, while the trunk base may be hidden.
The changed-context check replaces a planar roof with crown structure and asks which height is desired: maximum tree height, representative crown height or canopy surface. Validate against lidar or field measurements before applying a building rule.
CHAPTER 9 OF 12 · TRANSFER AND VALIDATE
9. Use multiple shadows when one image is ambiguous
Images at different Sun angles provide repeated equations and reveal endpoint mistakes. If estimated height changes systematically with azimuth, terrain slope, occlusion or segmentation is likely unresolved.
Multi-date use requires stable objects and accurate co-registration. Do not average incompatible shadows blindly; fit a geometry model and inspect residuals by date and direction.
CHAPTER 10 OF 12 · TRANSFER AND VALIDATE
10. Validate height, not just shadow segmentation
A high intersection-over-union score for shadow masks does not guarantee accurate height. Endpoint errors concentrated at the tip can matter greatly while changing little total area.
Compare inferred heights with lidar, stereo photogrammetry, surveyed structures or reliable building records. Report bias, root-mean-square error and failure rate by height, density, Sun angle and terrain class. Preserve excluded cases.
CHAPTER 11 OF 12 · CHECK AND CONTINUE
11. Questions readers often ask
Can one satellite image recover every building height?
No. The method needs a visible, attributable shadow, suitable geometry and a known receiving surface. Dense overlap, cloud, low contrast and off-nadir occlusion can make height unidentified.
Is the longest dark line always the correct shadow length?
No. Measure the displacement from the correct base to the correct tip along the solar direction, then account for ground slope and sensor projection. Darkness alone does not establish correspondence.
CHAPTER 12 OF 12 · CHECK AND CONTINUE
12. Evidence, connected routes and the next check
Raju, Chaudhary and Jha’s single-image shadow-height method and accuracy assessment documents rooftop, shadow and height-estimation stages. Recent high-resolution work on building footprints, occlusion and shadow-based height inversion shows why sensor geometry and rejected overlaps matter.
Continue through terrain shadows and horizon angles, DEM-driven shadow-boundary uncertainty, and co-registration offsets and false change. The next check is to repeat the height from a second Sun direction and inspect the residual.
Contents · Previous chapter · Return to the Remote Sensing root