GOLD STANDARD · REMOTE SENSING · HOW X WORKS
Find the decision hidden inside the model
Surface reflectance is directional. BRDF normalisation asks what the same surface would look like under a chosen viewing geometry, helping separate real land change from changes in view angle and illumination.
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 directional reflectance
Follow the kernel calculation
Audit quality and limits
Transfer across sensors
Use evidence and next routes
CHAPTER 1 OF 12 · ORIENT THE GEOMETRY
1. The same surface can return different reflectance
Leaves, soil grains, shadows and canopy gaps scatter sunlight unevenly by direction. A pixel observed from two view angles can therefore change even when the surface itself does not. Solar zenith, view zenith and relative azimuth jointly define the observation geometry.
This directional dependence is described by the bidirectional reflectance distribution function, or BRDF. It relates illumination and viewing directions to reflected radiance. In satellite products, a fitted BRDF model is often used to compare observations under a common geometry.
CHAPTER 2 OF 12 · ORIENT THE GEOMETRY
2. Normalisation changes geometry, not history
BRDF normalisation predicts reflectance for a reference view and illumination convention. Nadir BRDF-adjusted reflectance, or NBAR, commonly uses a nadir view while retaining the illumination geometry associated with the product date.
That distinction matters. NBAR is not ordinary measured nadir reflectance, and it is not the same as white-sky or black-sky albedo. Each quantity integrates or fixes geometry differently and answers a different physical question.
CHAPTER 3 OF 12 · NORMALISE THE SIGNAL
3. Kernel models separate three directional components
The MODIS BRDF system uses a semi-empirical kernel-driven model. Reflectance is represented as fiso + fvolKvol + fgeoKgeo. The isotropic coefficient supplies a baseline; the volumetric and geometric coefficients weight kernels for volume scattering and shadow-casting structure.
The coefficients are estimated from multiple atmospherically corrected, cloud-screened observations over a moving period. Once estimated, the kernels can be evaluated at the observed geometry and at a target geometry.
CHAPTER 4 OF 12 · NORMALISE THE SIGNAL
4. Work a geometry adjustment explicitly
Take fiso = 0.20, fvol = 0.04 and fgeo = 0.02. At one oblique observation, suppose Kvol = 0.50 and Kgeo = −0.20. Modelled reflectance is 0.20 + 0.04×0.50 + 0.02×(−0.20) = 0.216.
At the target nadir geometry, suppose Kvol = 0.10 and Kgeo = −0.10. Adjusted reflectance is 0.20 + 0.004 − 0.002 = 0.202. The 0.014 difference is attributed to geometry within this fitted model, not automatically to land-cover change.
| Evaluation | Kvol | Kgeo | Modelled reflectance |
|---|---|---|---|
| Observed oblique geometry | 0.50 | −0.20 | 0.216 |
| Target nadir geometry | 0.10 | −0.10 | 0.202 |
| Geometry difference | — | — | 0.014 |
CHAPTER 5 OF 12 · NORMALISE THE SIGNAL
5. Build a comparable time series in the right order
Start from surface-reflectance observations with cloud, aerosol and snow handling appropriate to the product. Retrieve or fit BRDF parameters, evaluate them at the reference geometry, and then compare normalised values. Preserve acquisition geometry and quality flags alongside the result.
Do not treat normalisation as a substitute for atmospheric correction or co-registration. A geometry-consistent signal can still contain aerosol residuals, adjacency effects, mixed pixels or spatial misalignment.
CHAPTER 6 OF 12 · AUDIT THE RETRIEVAL
6. Quality flags tell you how much angular evidence exists
A full BRDF inversion supported by adequate, well-distributed angular samples is different from a magnitude inversion that relies more heavily on prior shape information. Product quality layers identify retrieval status and should be carried into analysis.
Sparse clear-sky observations, persistent cloud, snow transitions and extreme solar angles can weaken the estimate. Filter or stratify by quality rather than averaging high- and low-confidence retrievals into one smooth-looking series.
CHAPTER 7 OF 12 · AUDIT THE RETRIEVAL
7. Separate geometry correction from real uncertainty
The adjusted value inherits uncertainty from input reflectance, angular sampling, cloud screening and model mismatch. A small difference after normalisation is not automatically meaningful; compare it with retrieval quality and expected error.
Inspect observed-minus-model residuals by angle. Systematic residuals at particular view directions can reveal that the kernel model does not capture the scene, especially over rugged terrain, row crops, snow or strongly anisotropic canopies.
| Diagnostic | What it tests | Warning sign |
|---|---|---|
| Angular coverage | Evidence for BRDF fit | Angles clustered in one direction |
| Retrieval quality flag | Full versus constrained inversion | Low-quality magnitude retrieval |
| Residual by geometry | Kernel adequacy | Directional residual pattern |
| Normalised difference versus uncertainty | Change significance | Difference smaller than error |
CHAPTER 8 OF 12 · TRANSFER AND COMPARE
8. Transfer from vegetation monitoring to urban surfaces
Move from a forest canopy to roofs and street canyons. The stable problem is directional scattering and shadowing; the same view-angle change can alter apparent reflectance without material change.
The context changes the scattering geometry. A kernel model developed for land surfaces may be less adequate for vertical urban structure at fine resolution. Transfer succeeds by checking residuals, scale and angular support rather than assuming the same coefficients remain meaningful.
CHAPTER 9 OF 12 · TRANSFER AND COMPARE
9. Compare sensors only after harmonising definitions
Cross-sensor work must align spectral response, spatial support, atmospheric processing, reference geometry and temporal window. Two products called surface reflectance may still use different BRDF assumptions or quality rules.
Use overlapping acquisitions to estimate remaining angle-dependent bias after normalisation. Stratify by biome, season and view geometry. A single global regression can hide large errors in a smaller but important class.
CHAPTER 10 OF 12 · TRANSFER AND COMPARE
10. Know the limits of a nadir-adjusted product
Nadir adjustment reduces a major source of directional inconsistency; it does not reconstruct an angle-free intrinsic colour. The product still refers to a specified illumination, spectral band, spatial scale and model.
Rapid land change inside the multi-day retrieval window can also violate the assumption of a stable surface. High solar zenith, topographic shadow, snow and poor atmospheric inputs may dominate. Preserve the raw observations when the event timing matters.
CHAPTER 11 OF 12 · CHECK AND CONTINUE
11. Questions readers often ask
Is NBAR the reflectance a satellite actually measured at nadir?
Not necessarily. It is a modelled reflectance evaluated at nadir view using retrieved BRDF parameters and the product’s illumination convention. Read the product definition before mixing it with direct observations.
Can BRDF normalisation remove every seasonal pattern?
No. Phenology, moisture, snow and illumination all change seasonally. Normalisation targets directional geometry; it should not erase genuine surface dynamics.
CHAPTER 12 OF 12 · CHECK AND CONTINUE
12. Evidence, connected routes and the next check
NASA’s MODIS BRDF/Albedo product overview defines the parameter, albedo and NBAR product family. The NASA MCD43A1 Version 6.1 dataset record documents the isotropic, volumetric and geometric parameters and urges use of band-specific quality information.
Continue through atmospheric correction and adjacency effects, stored values and reflectance ratios, and incidence angle and false SAR change. The next practical check is to plot the corrected time series against view zenith angle; remaining slope is evidence the normalisation needs review.
Contents · Previous chapter · Return to the Remote Sensing root