GOLD STANDARD · REMOTE SENSING · HOW X WORKS
Find the decision hidden inside the model
A satellite measures radiance after sunlight has travelled through the atmosphere and interacted with a neighbourhood, not a surface reflectance value in isolation. Atmospheric correction estimates the surface contribution, while adjacency effects explain why light scattered from nearby bright or dark terrain can still bias a pixel.
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.
Separate measured and intended quantities
Build a radiance-to-reflectance ledger
Diagnose adjacency effects
Transfer and validate
Use sources and connected routes
CHAPTER 1 OF 12 · DEFINE THE SIGNAL
1. Top-of-atmosphere radiance is not surface reflectance
A passive optical sensor records spectral radiance at the instrument. That signal can include sunlight reflected by the target, absorption and scattering along the path, and light scattered into the line of sight by the atmosphere.
Surface reflectance is the fraction and directional distribution of incident radiation reflected by the surface under stated geometry and spectral conventions. Atmospheric correction is an inverse problem: estimate that surface quantity from a mixed, noisy top-of-atmosphere observation.
CHAPTER 2 OF 12 · DEFINE THE SIGNAL
2. Name the product level before comparing pixels
USGS states that Landsat Level-1 products are not atmospherically corrected, whereas Level-2 science products include surface-reflectance processing. Digital numbers also require product-specific scale factors and offsets.
Comparing a Level-1 radiance value with a Level-2 reflectance value as if they were the same measurement creates false change. Preserve collection, processing level, algorithm version, band, acquisition geometry and quality masks.
CHAPTER 3 OF 12 · CORRECT THE PATH
3. Build a simplified path-radiance model
For teaching, write measured radiance as LTOA = Tdown Tup ρ Es/π + Lpath, where ρ is an idealised surface reflectance, Es is solar irradiance on the surface, T terms are transmissions and Lpath is path radiance. A practical algorithm uses fuller radiative-transfer physics and band responses.
The equation reveals the correction task: subtract path radiance, then divide by the illumination and transmission terms. Every estimated atmospheric input contributes uncertainty to the retrieved ρ.
CHAPTER 4 OF 12 · CORRECT THE PATH
4. Work one controlled correction
Take normalised values LTOA = 0.24, Lpath = 0.04 and combined illumination-transmission factor K = Tdown Tup Es/π = 0.80. The estimated surface reflectance is (0.24−0.04)/0.80 = 0.25.
If path radiance were underestimated as 0.02, the result would be 0.275. A two-hundredths radiance error creates a 0.025 reflectance difference in this constructed example. It is not a universal Landsat conversion.
| Assumed Lpath | LTOA − Lpath | K | Retrieved ρ |
|---|---|---|---|
| 0.04 | 0.20 | 0.80 | 0.250 |
| 0.02 | 0.22 | 0.80 | 0.275 |
| 0.06 | 0.18 | 0.80 | 0.225 |
CHAPTER 5 OF 12 · CORRECT THE PATH
5. Aerosol, water vapour and geometry all matter
Molecular scattering, aerosols and gaseous absorption affect bands differently. Solar zenith angle changes path length and illumination. Terrain slope, view geometry and bidirectional reflectance can add structure that an atmospheric-only correction does not remove.
Use the product’s ancillary layers and quality flags. A valid surface-reflectance product is not automatically cloud-free, shadow-free, terrain-normalised or comparable across every sensor and season.
CHAPTER 6 OF 12 · CHECK ADJACENCY
6. Adjacency mixes the neighbourhood into the pixel
Atmospheric scattering can carry photons reflected by surrounding terrain into the sensor’s line of sight. A small dark water body beside bright sand may therefore appear brighter than a model treating each pixel independently would predict.
The effect depends on aerosol conditions, wavelength, target size, point-spread behaviour and the reflectance contrast around the target. It is not the same as ordinary pixel mixing at the ground, though both can occur together.
CHAPTER 7 OF 12 · CHECK ADJACENCY
7. Work a bright-neighbour sensitivity check
Suppose the independently corrected target would have reflectance 0.08. A simplified adjacency contribution adds α(ρN−ρT), with neighbourhood reflectance 0.40 and α = 0.05. The added term is 0.016, giving an apparent 0.096.
If α falls to 0.01 under clearer conditions, the contribution is 0.0032 and apparent reflectance is 0.0832. These invented values show the direction and scale logic; they do not replace a three-dimensional radiative-transfer model.
| Target ρT | Neighbour ρN | α | Adjacency term | Apparent ρ |
|---|---|---|---|---|
| 0.08 | 0.40 | 0.05 | 0.0160 | 0.0960 |
| 0.08 | 0.40 | 0.01 | 0.0032 | 0.0832 |
| 0.40 | 0.08 | 0.05 | −0.0160 | 0.3840 |
CHAPTER 8 OF 12 · TRANSFER AND VALIDATE
8. Changed-context transfer: a coastal water boundary
Move from a large inland field to a narrow coastal inlet. The atmospheric path remains, but adjacency contrast becomes stronger because bright land surrounds dark water and the target may occupy only a few pixels.
Transfer succeeds when target size, neighbourhood, aerosol conditions and sensor spatial response are rebuilt. Copying the field correction without a coastal adjacency check can turn land brightness into a false water-quality signal.
CHAPTER 9 OF 12 · TRANSFER AND VALIDATE
9. Validate with more than a smooth map
Compare corrected values with field or airborne reference measurements whose time, spectral response and viewing geometry are documented. Use stable pseudo-invariant sites and cross-sensor checks, while recognising that each reference has its own uncertainty.
Inspect residuals against aerosol optical thickness, target-neighbour contrast, distance from boundaries and solar geometry. A correction that removes scene-wide haze but leaves boundary-dependent bias has solved only part of the problem.
CHAPTER 10 OF 12 · TRANSFER AND VALIDATE
10. Limits and common misconceptions
Surface reflectance is an estimated geophysical product, not a direct sensor reading. Atmospheric correction does not restore information completely hidden by thick cloud, saturated signals or an unsuitable observation geometry.
Adjacency correction is not generic sharpening. It needs a physical account of scattered neighbourhood light and sensor response. Aggressive correction can invent contrast when its aerosol or neighbourhood assumptions are wrong.
CHAPTER 11 OF 12 · EVIDENCE AND ROUTES
11. Questions readers often ask
A darker Level-2 pixel than Level-1 display does not by itself prove overcorrection; the products use different physical quantities, scaling and visual stretches. Compare like with like in physical units.
Is top-of-atmosphere reflectance useless?
No. It is useful for calibration-aware comparisons and algorithms designed for it. It simply retains atmospheric effects and should not be labelled surface reflectance.
When are adjacency effects most important?
They deserve attention for small or high-contrast targets, hazy conditions, short wavelengths and boundaries such as coasts, snow edges, bright urban surfaces or clouds beside dark terrain.
CHAPTER 12 OF 12 · EVIDENCE AND ROUTES
12. Evidence, connected routes and the next check
USGS Landsat Surface Reflectance guidance explains the purpose and product constraints. NASA’s Harmonized Landsat Sentinel-2 algorithms describe operational atmospheric correction, and the NASA adjacency-effect study documents scattered light from neighbouring fields.
Continue through stored values, offsets and reflectance ratios, co-registration and false boundary change, and emissivity and brightness-temperature bias. The next evidence check is to stratify residual reflectance error by aerosol state, boundary distance and neighbourhood contrast.
Contents · Previous chapter · Return to the Remote Sensing root