649 
In: Wagner W„ Szekely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
shading and multiple scattering are minimal (for a 
recent review see Chen et al., 2000), 
2. Turbid medium models assume the canopy to be 
horizontally uniform and to be composed of plane 
parallel layers that are filled with dimensionless 
scatterers randomly distributed throughout the 
available volume and oriented in accordance with a 
given leaf normal distribution function. These models 
are best used for dense canopies with small vegetation 
elements (e.g., mature agricultural crops) and 
relatively inappropriate for open forest canopies (for a 
recent review see Qin and Liang, 2000), 
3. Hybrid models represent vegetation canopies using 
both of the above approaches. Typically they assume 
that the interior of geometric objects (representing 
tree crowns) are uniformly filled with a ‘gas’ of point 
like oriented scatterers of specified orientations and 
spectral properties. These models can be used to 
represent both sparse and dense canopies. However, 
multiple scattering is not rigorously treated and the 
models are often limited by one single type of crown 
geometries (Goel and Thompson, 2000), 
4. Computer simulation models can represent arbitrarily 
complex canopy architectures using constructive solid 
geometry or similar computer graphics techniques. All 
facets of a geometric object (needle, trunk, leaf, twig, 
etc.) can be tagged with spectral and directional 
scattering properties. In ray-tracing models a Monte 
Carlo procedure is then used to determine the location 
and direction of incident light beams; the type of 
interaction, i.e., reflection, absorption or transmission, 
that such rays undergo when intersecting with an 
object, and in the case of a scattering event also the 
direction of further propagation. To compute the 
BRDF of a plant canopy one keeps shooting rays into 
the scene, follows them through their various 
interactions until they exit the scene and end up in 
certain small solid angles around predefined viewing 
directions. Due to the large number of photons needed 
for reliable statistics this type of RT model tends to be 
relative computer intensive (Disney et al., 2000). 
1.3 Early Canopy RT Model Validation Efforts 
With the availability of a large set of canopy RT models in the 
1980s the question arose as to how one could assess their 
quality and reliability. Of primary interest here was the 
validation of ‘simple’ canopy RT models that - because of fast 
execution times and small numbers of parameters - were likely 
to play a role in the operational retrieval of quantitative surface 
information from optical remote sensing data. So far the 
verification of canopy RT models in forward mode has always 
relied on comparison strategies with respect to one or more of 
the following types of reference data: 
1. Air or space borne observations, that were acquired 
over specific test sites and subsequently corrected on 
the basis of concurrently measured atmospheric 
properties, are used as a means to evaluate the 
simulations of canopy RT models based on structural, 
spectral and illumination related information 
pertaining to the same canopy target and (ideally also) 
the same time of acquisition as the space or air-borne 
observations, e.g., Schaaf et al., (1994), Soffer et al., 
(1995), 
2. In-situ or laboratory measurements, that were 
acquired with sensors - typically supported by a tram 
system or goniometer structure - looking down at the 
canopy target, are used as a means to evaluate the 
quality of RT model simulations based on the 
spectral, structural and illumination characteristics of 
the canopy target (ideally acquired at the same time as 
the BRDF measurements), e.g., Franklin and Duncan 
(1992), Strahler and Liang (1994), 
3. Canopy RT model simulations, that were (ideally) 
generated by sophisticated Monte Carlo RT models 
on the basis of detailed 3-D description of the canopy 
architecture, are used to assess the output of simpler 
canopy RT models making use of the same canopy- 
target characteristics (albeit adapted to their need for 
input parameter specifications), e.g., Goel and Kuusk, 
(1992), Liang and Strahler (1992). 
In some cases the quality of canopy RT models is also 
addressed in inverse mode, that is, by looking at how well a 
model allows to retrieve certain biophysical parameters on the 
basis of measured BRDF data. Such an approach is, however, 
more suited to comment on the numerical inversion procedure 
than the physical correctness of the canopy RT model. Pinty 
and Verstraete (1992) advocated the use of both forward and 
inverse modes. Their idea was to acquire detailed descriptions 
of the structural and spectral properties of a canopy target and 
to feed these into an RT model to simulate BRDF patterns of 
the target under a specific set of illumination and observation 
conditions. These forward simulations can then be compared to 
actual observations previously acquired over the target in 
question at the same viewing and illumination geometries. The 
RT model can then be inverted against the measured and/or 
simulated data sets and the output of this operation compared to 
the canopy characteristics that had been measured initially and 
used as input to the forward simulations. 
Unfortunately, the verification of canopy RT models on the 
basis of actual measurements has always been hampered by the 
lack of accurate, comprehensive and self-consistent field data 
sets, e.g., Strahler, (1997). This situation has not changed much 
since the 1990s and even the use of artificial targets in 
laboratory environments suffer from the same difficulties, that 
is, instruments and methods that allow for a highly precise 
characterisation of 1) the light environment surrounding the 
target, 2) the position, orientation, size and shape of all the 
physical components making up the target, 3) the magnitude, 
directionality and spatial variation of scattering properties of all 
canopy and background elements, and 4) the detector location, 
foreoptics (if present) and spectral response functions. None of 
these issues are present, however, when canopy RT models are 
compared using virtual plant environments. 
2. SYSTEMATIC RT MODEL EVALUATION 
2.1 Strategy 
The RAdiative transfer Model Intercomparison (RAMI) 
initiative was launched in the late 1990s to provide a platform 
for the systematic evaluation of physically-based canopy RT 
models (Pinty et al., 2001). Of primary relevance was the need 
to eliminate sources of uncertainty that affect the outcome of 
verification efforts but that do not pertain to the quality of the 
canopy RT models themselves. At the time, this strategy 
precluded the evaluation of RT model simulations on the basis