Problem
Leaves do not reflect light equally in all directions. Their anatomy, pigments and microscopic surface roughness determine how light is scattered and therefore how it is distributed inside a canopy, yet most canopy models treat leaves as Lambertian reflectors. Measuring directional reflectance leaf by leaf is slow, so the question is whether traits that are easy to measure can predict it.
Approach
- Measure directional reflectance. A custom-built Directional Spectrum Detection Instrument (DSDI), with a xenon source, fibre spectrometer and mechanically controlled illumination and viewing angles, measured both leaf surfaces at 400–1000 nm, calibrated against a Lambertian white reference.
- Fit a physical model. A Cook–Torrance BRDF represents diffuse and specular reflection with three wavelength-dependent parameters, fitted by adaptive grid search and least squares.
- Predict optics from traits. A stacking ensemble (SVR, random forest and gradient boosting, with a linear meta-learner) maps leaf thickness, specific leaf weight, pigments, microscopy-derived roughness and wavelength to the BRDF parameters.
- Quantify canopy effects. Predicted parameters were fed into rice-canopy ray tracing with fastTracer.
| Parameter | Physical interpretation | Related leaf properties |
|---|---|---|
| σ(λ) | Microfacet roughness | Epidermal texture and surface irregularity |
| k(λ) | Diffuse reflection coefficient | Internal scattering |
| n(λ) | Refractive index | Refraction and interface reflection |
Move the sliders to see how each parameter shapes the reflectance pattern: lower roughness concentrates light in a sharp lobe around the mirror direction, while the diffuse coefficient lifts the whole curve evenly.
- Peak
- 0.281 sr⁻¹ at 31°
- Diffuse level k/π
- 0.095 sr⁻¹
- Specular share at the mirror angle
- 66%
Figure 2. Interactive Leaf BRDF in the principal plane
f = F(n, θh)·D(α, σ)·G(L, N, V, H) / 2π²(L·N)(N·V) + k/π, computed in your browser as in the study's fitting code. The total is the diffuse semicircle plus the glossy lobe around the mirror direction. Slider ranges are the fitting bounds. The starting roughness σ = 0.2 is chosen to make the lobe visible; k and n are the code's initial guesses. None of these is a measured leaf; fitted values for each species are in the paper.
Source: doi:10.1016/j.plaphe.2025.100135 · Code
Results
- Material: maize, rice, cotton and poplar leaves from upper and lower canopy positions, both leaf surfaces.
- Model fit: BRDF fitting R² > 0.95.
- Prediction: ensemble prediction R² = 0.83–0.99, depending on the parameter.
Main findings:
- A physically based BRDF model represents directional leaf reflectance accurately.
- Structural and biochemical leaf traits carry the information needed to predict BRDF parameters.
- Differences in leaf optics substantially change the simulated light field inside a canopy, so leaf optics should be represented explicitly rather than assumed uniform.
Together these results connect leaf-scale phenotyping to radiative-transfer and canopy-photosynthesis models.
Limitations
The model was built from 270 data entries covering four species, two canopy positions and both leaf surfaces, and applies within the measured trait and wavelength ranges (400–1000 nm); new species and conditions require new measurements to extend it. The canopy simulations quantify changes in light distribution; linking them to field yield is a subsequent step. Extending the dataset to more genotypes, environments, developmental stages and water status is the next step of this work.
Code and data
- BRDF fitting scripts and Roughness Calculator
- fastTracer canopy ray-tracing software
- The study data are available from the corresponding author upon reasonable request, as stated in the published article.
