In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
648 
AN OVERVIEW OF TWO DECADES OF SYSTEMATIC 
EVALUATIONS OF CANOPY RADIATIVE TRANSFER MODELS 
J-L. Widlowski 
Global Environment Monitoring Unit, Institute for Environment and Sustainability, DG Joint Research Centre of the 
European Commission, Via E. Fermi 1, Ispra 21020 (VA), Italy - Jean-Luc.Widlowski@jrc.ec.europa.eu 
KEY WORDS: Model Intercomparison, Canopy Radiative Transfer, Quality Assurance, Optical Remote Sensing 
ABSTRACT: 
Space borne observations constitute a highly appropriate source of information to quantify and monitor Earth surface processes. The 
reliability that may be associated with the outcome of interpretation and assimilation efforts of these data, however, relies heavily on 
the actual performance of the available modeling tools. Scientists, space agencies and policy makers that want to make use or support 
the derivation of quantitative information from space observations must therefore have access to indicators describing the quality of 
the models and algorithms that are used in retrievals. As a formalization of earlier model verification efforts the RAdiation transfer 
Model Intercomparison (RAMI) initiative was launched in 1999 in an attempt to shed light on the reliability and accuracy of 
physically-based canopy radiative transfer models simulating the interactions between sunlight and vegetation. This contribution 
documents the evolution and achievements of RAMI and provides an outlook of challenges and opportunities that still lie ahead. 
1. INTRODUCTION 
1.1 Purpose of Canopy Radiative Transfer Models 
The exploitation of global Earth Observation data hinges more 
and more on physically-based radiative transfer (RT) models. 
These models simulate the interactions of solar radiation within 
a given medium (e.g., clouds, plant canopies) and are used to 
generate look-up-tables, to train neural networks or to develop 
parametric formulations that are then embedded in quantitative 
retrieval algorithms such as those delivering the operational 
surface products for MODIS, MISR and MERIS, for example. 
Assessing the quality of RT models is thus essential if accurate 
and reliable information is to be derived from them. Biases and 
errors in RT models may also affect the outcome of new 
mission concept studies, as well as our capacity to quantify 
Earth surface processes and the reliability of downstream 
applications that assimilate such remotely sensed data streams. 
The focus of this contribution lies with the quality of 
physically-based models that deal with the representation of 
radiative processes in vegetated environments within the optical 
domain of the solar spectrum. 
Most land surfaces are strongly anisotropic reflectors when 
observed from optical to thermal-infrared wavelengths. The 
angular dependence of their reflectance function (termed the 
bidirectional reflectance distribution function or BRDF) results 
from 1) the three-dimensional (3-D) nature of terrestrial targets, 
i.e., the size, shape and spacings of trees in a forest or crops in a 
field, which produces distinct patterns of shadows that change 
with the direction of view (and illumination), and 2) the 
scattering behaviour of individual foliage, wood and soil 
elements together with their density and orientation with respect 
to the illumination and viewing directions (Ross, 1981). 
Physically-based canopy RT models, when used in forward 
mode, are capable to predict the BRDF of a vegetation target on 
the basis of architectural, spectral and illumination related 
descriptions. Conversely, when used in inverse mode, 
physically-based canopy RT models may, in principle, retrieve 
the structural and spectral canopy properties that gave rise to the 
directionally (and spectrally) varying reflectance observations. 
Knowledge of the BRDF of terrestrial surfaces is necessary for 
1) the accurate retrieval of surface albedo (via a hemispherical 
integral of the BRDF), 2) the specification of the lower 
boundary condition in atmospheric corrections and for the 
estimation of cloud and aerosol properties (e.g., Hu et al., 
1999), and 3) the retrieval of sub-pixel surface characteristics 
(e.g., Widlowski et al, 2004). 
1.2 Types of Canopy RT Models 
A large variety of physically-based canopy RT models have 
been developed over the past five decades or so. According to 
Qin and Liang (2000) the purpose of modelling the radiation 
distribution in the 1960s was primarily to estimate the canopy 
photosynthetic rate; in the 1970s the focus was on the 
calculation of surface albedo and net radiation for energy- 
balance and micro-meteorological research; in the 1980s and 
90s canopy RT modelling was driven by the need to accurately 
describe the angular distribution of the reflected radiation. In 
the last decade or so the emphasis was placed on efficient 
representations of radiative processes in increasingly complex 
3-D canopy architectures and the retrieval of sub-pixel surface 
structure information. In a landmark paper, Goel (1987) 
grouped canopy RT models into 4 different categories: 
1. Geometrical models, that assume the canopy to be 
made up of a ground surface (of known reflective 
properties) with geometrical objects of prescribed 
shapes and dimensions (e.g., spheres, ellipsoids, 
cones) and optical properties (reflectance, 
transmittance and absorption) placed on it in a 
defined manner (random or clustered) to represent the 
spatial distribution of tree crowns. The canopy 
reflectance is the weighted sum of four components: 
sunlit and shaded crown, and sunlit and shaded 
background. These models are best suited for small 
view and sun zenith angles and for sparse canopies 
with high leaf densities, where the effects of mutual