Influences of Leaf-Specular Reflection on Canopy

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 55, NO.

2, FEBRUARY 2017 619

Influences of Leaf-Specular Reflection on Canopy


BRF Characteristics: A Case Study of Real Maize
Canopies With a 3-D Scene BRDF Model
Donghui Xie, Wenhan Qin, Member, IEEE, Peijuan Wang, Yanmin Shuai, Yuyu Zhou, and Qijiang Zhu

Abstract—The diffuse and specular components of leaf reflec- I. I NTRODUCTION


tion are both important to determine the leaf optical properties as
well as to describe the leaf bidirectional reflectance distribution
function (BRDF). However, the specular component is usually
ignored in practice in numerous canopy reflectance models that
N UMEROUS physically based canopy reflectance mod-
els have been developed to describe the bidirectional
reflectance of the vegetation canopy and to understand the
describe the interaction between solar light and vegetated scene
components. To evaluate the impact of leaf-specular reflection on interaction of solar light with the ground scene components,
canopy bidirectional reflectance factor (BRF) characteristics, we such as the vegetation canopy, the intrinsic vegetation proper-
introduce a leaf BRDF model into the radiosity-graphics combined ties, and the underlying background (understory and/or soil).
model (RGM; a 3-D scene model) to calculate canopy BRFs with These models can be generally classified into three categories,
nondiffuse leaves. The modified RGM is validated by comparing i.e., radiative transfer model, geometric optical model, and
simulated BRFs against in situ measured BRFs over real maize
canopies. The results show that ignorance of leaf-specular reflec- computer simulation model [17]. All of these models rely on
tion can result in up to 50% of relative error in the blue band three sets of variables and parameters: angular distributions of
(435.8 nm). A series of maize canopies with different leaf angle radiance that characterize the light from the sun and sky, canopy
distributions (LADs) is reconstructed to investigate the effect of structures, and optical properties of phytoelements and the
five major biophysical/geometrical parameters such as leaf area
understory or soil background [5]. Certain compromises have
index, LAD, leaf surface property, view direction, and solar zenith
angle on leaf-specular reflection contributions to the canopy BRF. been made in the early development phase of these models to
It is demonstrated that increasing the incident solar zenith angle simplify the process and reduce the number of model variables.
and decreasing the mean leaf angle impact the angular distrib- However, these simplifications can cause the model accuracy
ution of the canopy BRF more significantly than other factors. shifted from an algorithmic nature toward user knowledge of
The cumulative hemispherical relative and absolute errors of
canopy BRF caused by the leaf-specular reflection are often too
the modeled system [25]. With a better understanding of light–
large to be ignored, even for canopies with rough surface leaves. matter interaction in vegetation canopies and an improved repre-
Moreover, the relative error of BRF in visible waveband shows sentation of the “ground truth,” many types of models have been
that, in general, leaf-specular reflection has a large impact than improved to increase accuracy. However, majorities of the veg-
that in near-infrared waveband. However, such impact can be suf- etation radiative transfer models widely used in remote sensing
ficiently accounted for by even just consideration of the first-order
leaf-specular reflection in canopy reflectance calculation, leading still treat individual leaves as perfect Lambertian scatters and
to a substantial improvement in simulation accuracy for most only use their spectral directional-hemispherical reflectance and
vegetation canopies. transmittance to specify leaf optical properties [26].
Index Terms—Bidirectional reflectance distribution func- In the past several decades, however, numerous studies have
tion (BRDF), canopy BRDF modeling, leaf-specular reflection, revealed that leaf is neither an ideal diffuser nor an ideal
radiosity-graphics combined model (RGM). specular reflector. Laboratory measurements have shown that
leaf transmittance has a near-Lambertian angular distribution,
Manuscript received June 24, 2014; revised January 6, 2015 and May 27, but leaf reflectance exhibits an obvious angular dependence on
2016; accepted July 20, 2016. Date of publication November 25, 2016; date of
current version December 29, 2016. This work is supported by the Chinese 973
the incident-outgoing geometry [4], [14]. The diffuse character
Program (Grant No. 2013CB733402) and the National Natural Science Founda- of leaf reflection emanates primarily from the interior of the
tion of China (Grant No. 41571341, 41331171 and 41371410). (Corresponding leaf via multiple scattering. The specular character of leaf
author: Donghui Xie.)
D. Xie and Q. Zhu are with the State Key Laboratory of Remote Sensing reflection arises at the leaf surface [14]. According to our field
Science, School of Geography, Beijing Normal University, Beijing 100875, observations, the amount of sunlight that is specularly reflected
China (e-mail: xiedonghui@bnu.edu.cn). by leaves is often sufficiently large enough to display some bright
W. Qin is with Science Systems and Applications, Inc., Lanham, MD
20706 USA. spots on their surfaces. Other studies show that leaf-specular
P. Wang is with State Key Laboratory of Severe Weather, Chinese Academy reflectance varies from 10% (soybean) to greater than 50%
of Meteorological Sciences, Beijing 100081, China. (corn and sorghum) in the visible wavelengths [10].
Y. Shuai is with the School for the Environment, University of Massachusetts
Boston, Boston, MA 02125 USA. Leaf is the principal physiological organ of vegetation
Y. Zhou is with the Department of Geological and Atmospheric Sciences, canopies, so its optical properties certainly affect the light
Iowa State University, Ames, IA 50011 USA. distribution within the canopy as well as canopy reflectance.
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. Andrieu et al. [1] indicated that the hypothesis of isotropic leaf
Digital Object Identifier 10.1109/TGRS.2016.2598442 reflectance is a crude approximation of the actual bidirectional
0196-2892 © 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
620 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 55, NO. 2, FEBRUARY 2017

reflectance pattern of leaves. A loss of information will occur transmitting surfaces. So far, most radiosity-based models used
if leaves are treated as Lambertian scatters in remote sensing in remote sensing continue to employ this assumption. Some
studies [18]. Monteith [19] suggested that the leaf-specular researchers extended several types of radiosity methods to
reflectance of the corn canopy should be properly considered simple scenes with non-Lambertian surfaces, such as the two-
because of its shiny leaves when calculating the canopy re- pass method that combines ray-tracing and radiosity methods
flected radiance. A limitation of the Lambertian assumption is together [23], [27] and Shao’s method with the definition of
also encountered when simulating a corn canopy bidirectional new form factors for a general environment [24], among others.
reflectance factor (BRF) using the radiosity-graphics combined These extended methods are useful for scenarios with only
model (RGM), a radiosity computer graphics combined model mirror/mirror-like reflecting surfaces or a small number of non-
[31]. The question now is not if but how and to what extent the Lambertian surfaces. Nevertheless, it is difficult to directly
leaf-specular reflection affects the light spatial/angular distrib- apply them to vegetated scenes with millions of specularly
ution and directional reflectance of the canopy. reflecting leaves. Therefore, in this paper, an improved radiosity
In fact, previous radiative transfer studies of vegetation method effectively working on vegetation canopies with spec-
canopy have shown that the bidirectional scattering characteris- ularly reflecting leaves is proposed based on RGM with an
tics of leaves are worth to consider in order to improve the quan- appropriate balance of accuracy and computability of radiosity.
titative interpretation of remote sensing data [1], [8], [18], [25]. The purpose of this study is to extend the radiosity method by
Some of the one- and three-dimensional (3-D) turbid medium introducing leaf BRDF to RGM and then to utilize the modified
models for vegetation are first explored and modified to con RGM to investigate the effect of leaf-specular reflection on the
sider the leaf-specular reflection process by using a phase canopy BRF. In Section II, we first introduce the methodology
function combining the diffuse and specular components or to extend the radiosity method. In Section III, a group of simu-
by adding a first-order specular scattering term directly to the lated BRFs based on a realistic 3-D maize canopy is validated
canopy BRF [1], [8], [18]. Such approaches may produce good with some field measured data. In Section IV, typical impact
results in some directions, such as the direction opposite to the factors on leaf-specular reflection contributions to canopy BRF
hot spot but underestimation in other directions, such as the hot- (i.e., leaf angle distribution (LAD), leaf area index (LAI), leaf
spot or nonspecular directions, especially for those canopies surface properties, and view and incident angles) are discussed
that have more horizontally oriented leaves [22]. Additionally, using the results from a series of computer-rendered 3-D maize
for the sake of simplification in the process of multiple scatter- canopies with different LADs and LAIs, and the conclusion is
ing within the canopy, the leaf directional- or hemispherical– presented in Section V.
hemispherical reflectance is commonly used instead of its real
bidirectional reflectance distribution function (BRDF), which II. M ETHODOLOGY
may result in an overestimation of the canopy reflectance,
especially in the visible to the short-wave infrared band [25]. A. Leaf BRDF Model
At the same time, some computer simulation models have As the primary physiological organ, leaves play an im-
been developed to simulate the canopy reflectance of a 3-D portant role in plant growth through photosynthesis that, in
scene based on the Lambertian leaf assumption. Compared with part, depends on light distribution inside the canopy. Canopy
turbid medium models, computer simulation models, such as absorption and scattering of incident light are mainly controlled
Raytran [13], PARCINOPY [7], DIANA [11], [12], and RGM by leaf optical properties (reflectance and transmittance), which
[21], are built either based on theories of ray-tracing or radios- determine the light scattering characteristics of leaves, and
ity, by taking advantage of computer graphics algorithms and are crucial for quantitative study of radiative transfer in the
removing many hypotheses and statistical rules used in most vegetation canopy.
1-D or 2-D models to improve the simulation accuracy. With the Leaf optical properties are described by the BSDF, including
development of computer technology, Monte Carlo ray-tracing the BRDF and bidirectional transmittance distribution func-
models have been modified to consider leaf-specular reflection tion (BTDF) [4]. Some laboratory measurements find that leaf
by including the leaf bidirectional scattering distribution func- transmittance has a near-Lambertian distribution, while leaf
tion (BSDF) into canopy radiative transfer modeling [5], [25]. reflectance is neither perfectly diffuse nor perfectly specular,
The classic radiosity method originates in the field of thermal exhibiting an apparent angular dependence on the incident-
engineering to study the state of energy equilibrium within an outgoing geometry [14]. In this paper, the leaf transmittance
enclosed environment. Borel et al. [2], [3] and Gerstl et al. [9] is assumed isotropic based on the current experimental knowl-
investigated the interaction between the canopy and the incident edge; however, the leaf-specular reflectance will be accurately
solar light using the radiosity method to calculate the BRF. simulated in the following section.
Thereafter, Goel et al. [11], [12] and Qin et al. [21] combined The sum of specular and diffuse components is often used to
the radiosity method with computer graphics based on virtual describe a leaf surface BRDF model [4], which can capture the
reality theory and developed DIANA and RGM, extending the directional variation of the optical properties adequately
application of radiosity theory into structure-realistic heteroge-
neous vegetated scenes. ρbd (λ, Ψs , Ψo ) = ρd (λ) + ρs (λ, Ψs Ψo ) (1)
Although the radiosity method for 3-D heterogeneous scenes
is powerful, the Lambertian assumption for leaves is still where ρbd is the leaf BRDF, λ is the wavelength, Ψo and Ψs are
used, which limits its application for specularly reflecting and the unit vectors of light outgoing and incident directions, and
XIE et al.: INFLUENCES OF LEAF-SPECULAR REFLECTION ON CANOPY BRF CHARACTERISTICS 621

ρd and ρs are the respective diffuse and nondiffuse (also called in the scene, and the integration is carried out over the area of
as directional or specular) reflectance. A physically based leaf each facet.
BRDF model proposed by Bousquet et al. [4] is used in this The first term at the right side of (5), the emittance, can
study (Appendix). be expressed as the single-scattering process, which originates
from the incident solar and sky diffuse light in the VIS–NIR
region. The second and third terms at the right side of (5) can
B. Estimation of Leaf Diffuse Reflectance (ρd ) be considered as multiple scattering processes, including high-
According to the definition of directional-hemispherical re- order (second and up) reflection and transmission.
flectance function (DHRF) [20], the DHRF of the leaf surface For an ideal Lambertian facet, a constant relationship be-
can be derived by integrating over the view hemisphere [4] tween the radiance I and the radiosity B (the energy flux
density) of a facet exists such as B = πI [2]. ρbd = ρdh /π,
ρdh (λ, Ψs ) = πρd (λ) + ρs−dh (λ, Ψs ) (2) and τ bd = τ dh /π. Therefore, the radiosity equation can be
simplified to
where the superscript dh denotes the directional-hemispherical
reflectance of leaf and s−dh denotes the directional-

N 
N
hemispherical reflectance contributed only by the specular com- Bi = Ei + ρdh Bj Fij + τidh Bk Fik (6)
i
ponent (DHRFspec ). j=1 k=1
Since ρd is assumed isotropic, it can be estimated at Ψs = 0,
i.e.,
where Ei is the flux density (radiosity) emitted from the facet i
 dh 
d ρ (λ, 0o ) − ρs−dh (λ, 0o ) and Fij is the form factor between facets j and i that specifies
ρ (λ) = . (3) the fraction of energy flux leaving facet j and reaching facet
π
i. Equation (6) is the standard radiosity equation for a scene
ρdh (λ, 0o ) is the normal-hemispherical reflectance (NHR) consisting of Lambertian facets.
and can be calculated by the PROSPECT model or measured
in the laboratory via spectrophotometers equipped with an in-
tegrating sphere over leaves [4]. ρs−dh (λ, 0o ) can be evaluated D. Extend Classic Radiosity Method to Consider
approximately by numerical integration of ρs (Appendix). Leaf-Specular Reflection
For an ideal Lambertian surface, ρs = 0; therefore, ρbd =
ρ = ρdh /π. Assuming that the transmittance is also indepen-
d To modify the classic radiosity method to consider the
dent of the incident direction, thus, τ bd = τ d = τ dh /π. For nondiffuse reflection of leaves, it is necessary to balance the
clarity, we omit the wavelength subscription in the following applicability and accuracy of the radiosity method in vegetation
sections. remote sensing. Naturally, the first-order specular reflection,
which accounts for more than 95% of total specular reflection
contributions (see Fig. 4), is taken into account directly and
C. Radiosity Equation for Leaf Canopies precisely. Then, the high-order directionally reflected light from
For a 3-D canopy scene consisting of many facets, light is leaves can be approximately considered as diffuse reflection and
assumed to be emitted, reflected, absorbed, or transmitted only added into the diffuse reflectance components for the canopy.
at facets. The corresponding radiance on facet i in the given In the following, radiance from a leaf facet can be calculated
outgoing direction Ψo can be expressed as in four steps. First, for leaves that directly receive the direct so-
lar light, the surface radiance within a given sun-view geometry
Ii (Ψo ) = Ie,i (Ψo )+Ir,i (Ψo )+It,i (Ψo ) i = 1, 2, . . . , 2N can be accurately calculated using the leaf BRDF model as (1).
(4) Second, the scattering process from nonleaf foliage components
where the subscripts e, r, and t denote the emitted, reflected, and those leaves that are not directly exposed to the direct solar
and transmitted components of the outgoing radiance, respec- light is approximated as from Lambertian scatterers. Third, the
tively. N is the number of polygons (with the front and back multiscattering among canopy components is processed based
faces) in the scene. on the ideal Lambertian hypothesis, i.e., the classic radiosity
Following a general energy equation, (4) is expanded as: method can still be applied to obtain the flux density of each
N 
facet. Finally, the directional radiance on each facet and the
 canopy BRF can be calculated by combining results from the
Ii (Ψo ) = Ie,i (Ψo ) + ρbd
i (Ψs , Ψo )Ij (Ψs )Hji cos θs dωs
j=1A
aforementioned three steps. These four steps are detailed in
j
the following sections.
N 
 Step 1—Singly Scattered Radiance: Any facet will reflect or
+ τibd (Ψs , Ψo )Ik (Ψs )Hki cos θs dωs transmit the light when it is lit by direct solar and diffuse sky
k=1A light. The total reflected or transmitted radiance of any facet
k

(5) could be written as the sum of two components: one from direct
sun sky
where ρbd and τibd are the BRDF and BTDF of the facet i and solar light Ie,i (Ψs , Ψo ) and the other from skylight Ie,i
i
Hji is the visibility function between the facet j and the facet i.
sky sun
In (5), the summation is carried out over all number of facets Ie,i (Ψs , Ψo ) = Ie,i + Ie,i (Ψs , Ψo ). (7)
622 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 55, NO. 2, FEBRUARY 2017

Let rs be the direct solar flux density incident on the canopy simplification are neglectable as long as the contribution from
crossing a horizontal surface and rd be the diffuse skylight flux first-order specular reflection is accurately calculated.
density. Assuming that the diffuse skylight is isotropic and the The leaf facet lit by direct solar light is treated as the first-
hemisphere space is divided into a finite number of directions order specular light source in the canopy, and the source can
Ns /2 with equal solid angles, the diffuse skylight intercepted subsequently illuminate other facets. For example, the second-
by facet i from all solid angles is given as order incident flux density of facet i from other facet j can be
Ns
 calculated as
2 
 rd  4π
Fd (i) = |n̂i · ŝk |ai (ŝk ) (8) N 

π Ns
k=1 Fs(2) (i) = s
Ie,j (Ψs , Ψo )Hij cos θi dω (12)
j=1A
where ŝk is the incident direction for the kth solid angle j

segment and ai (ŝk ) is the fraction of the area of facet i lit by


where Ψo is the direction from the center of facet i to facet j.
the ŝk direct beam.
If the area of facet j is sufficiently small (by subdivision of the
The diffusely reflected or transmitted radiance from skylight
facet) such that the first-order specularly reflected radiance can
is calculated as
be treated as a constant within the differential solid angle based
sky
Ie,i = Fd (i)ξi (9) on the definition of the view factor, (12) can be rewritten as

where ξi is equal to either the NHR or transmittance (NHT) of 


N

facet i depending on the incident direction facing the front or Fs(2) (i) = s
Ie,j (Ψs , Ψo )Fij . (13)
j=1
back side of facet i
ρdh  
i
π if ( n i · n j ) ≤ 0 Next, the second- and high-order specular reflection between
ξi = τidh   leaves is approximated similar to the diffuse scattering process
π if ( n i · nj) > 0.
s(2)
Direct solar light is the most important source for directional Ei = πFs(2) (i)ξi . (14)
reflection from a facet. It can be expressed as the sum of the dif-
d−sun The classic radiosity equation is then rewritten as
fusely reflected/transmitted radiance Ie,i and the specularly
s
reflected radiance Ie,i (Ψo )  
N 
N
s(2)
sun d−sun s
Bi = Eid +Ei +ρdh
i Bj Fij +τidh Bk Fik . (15)
Ie,i (Ψs , Ψo ) = Ie,i + Ie,i (Ψs , Ψo ). (10) j=1 k=1

First, the average flux density on facet i from a direct solar Thus, the matrix equation AB = E can be altered accord-
beam can be evaluated as ingly as
rs ⎛ ⎞
Fs (i) = |n̂i · ŝd |ai (ŝd ) (11) 1 −πξ1 F12 ··· −πξ1 F1N
cos θs ⎜ .. .. .. ⎟
A =⎝ .. ⎠
. . . .
where θs is the solar zenith angle and ŝd is the solar direction. −πξN FN 1 −πξN FN 2 ··· 1
The diffusely reflected or transmitted radiance of facet i due ⎛ ⎞
d−sun ⎛ ⎞ s(2)
to the direct solar light can be calculated by Ie,i = ρdi Fs (i) B1 E1d + E1
d−sun ⎜B2 ⎟ ⎜ s(2) ⎟
or Ie,i d
= τi Fs (i). The specularly reflected radiance from ⎜ ⎟ ⎜E2d + E2 ⎟
B = ⎜. ⎟ and E = ⎜. ⎜ ⎟.
a direct solar beam is given by Ie,i s
(Ψs , Ψo ) = ρsi (Ψs , Ψo ) · ⎝.. ⎠ ⎟
Fs (i). ⎝.. ⎠
BN E +E d s(2)
Therefore, the single scattering radiance can be divided into a N N
s
specular component Ie,i (Ψs , Ψo ) and a diffuse component, in-
cluding the diffusely reflected/transmitted radiance from direct For this linear matrix equation, the classic iteration method is
d
solar and diffuse sky light, i.e., Ie,i sky
= Ie,i d−sun
+ Ie,i . also valid to obtain the flux density (B) of each facet. Although
Step 2—Multiply Scattered Radiance: The classic radiosity the symbol B is still used to denote the radiosity of each facet,
method has the advantage of describing the multiple scattering its meaning and value in this context are not the same as in
by solving the radiosity equation with an iteration method. In the classic radiosity treatment, which represents the multiple
order to continue using the classic radiosity theory in which scattering in the canopy only with Lambertian facets.
the specularly reflected radiance is not considered, a tradeoff Step 3—Directional Exiting Radiance: The directional exiting
is made here. The diffusely scattered radiance, assuming each radiance of each facet can be calculated as the sum of the singly
facet is a Lambert, is independent of the view direction, so and multiply scattered radiances. Accordingly, the outgoing
it can be described by the classic radiosity equation similar radiance of a facet can be expressed as
to (6); high-order specular reflection from leaves (second and  τidh 
N N
up) is treated the same way as that for diffuse reflection. Ii (Ψo ) = Ie,i (Ψo )+ ρbd
i (Ψs , Ψo )Bj Fij + Bk Fik .
Because of the very small contribution (less than 2%) of high- j=1
π
k=1
order specular reflection to canopy BRF, errors caused by this (16)
XIE et al.: INFLUENCES OF LEAF-SPECULAR REFLECTION ON CANOPY BRF CHARACTERISTICS 623

Substituting the equations derived in step 1 into (16), then Step 4—Canopy BRF Calculation: The canopy BRF in any
the exiting radiance can be rewritten as the sum of the diffuse view direction Ψo can be calculated by
scattering component and the specular reflection component 
⎡ ⎛ ⎞ Ii (Ψo )|n̂i · ŝo |area(i)ai (Ψo )
i
1 ⎣ dh ⎝ N BRF(Ψo ) =  1 . (20)
Ii (Ψo ) = ρi Fd (i)+Fs (i)+ Bj Fij ⎠ π |n̂i · ŝo |area(i)ai (Ψo )
π i
⎤
j=1

N Substituting (18) into (20), then the canopy BRF can be
+ τidh Fd (i+N )+Fs(i+N )+ Bk Fik ⎦ divided into three components
⎛ ⎞ k=1

ρs−dh 
N BRF(Ψo ) = BRFdiff + BRFspec1 + BRFspec2 (21)
− i ⎝Fs (i) + Bj Fij ⎠ + ρsi (Ψs , Ψo )Fs (i)
π j=1 where the first one, BRFdiff , is the diffuse and multiple scat-

N tering reflectance among the canopy; the second, BRFspec1 , is
+ ρsi (Ψs , Ψo )Bj Fij . (17) contributed by the first-order specular reflectance of the leaf
j=1 facets from the solar direct light; and the last, BRFspec2 , is
contributed by the multiple specular reflectance of leaves within
The exiting radiance of a facet can be divided into three
the canopy.
components, the ideal diffuse scattering (single + multiple), the
Once the structure of the canopy is known, the first-order
first-order specular, and the multiple-order specular reflection
specular component will only depend on the following: 1) the
radiance
ratio of direct to total incident light in the radiance flux Fs
Ii (Ψo ) = Iidiff + Iispec1 (Ψo ) + Iispec2 (Ψo ) and 2) the specular reflectance of leaf facet ρsi , which is only
related to the leaf surface characteristics and independent of the
Where wavelength. Therefore, if we further assume that the ratio of
⎡ ⎛ ⎞ direct light is 1.0, the first-order specular BRFspec1 will become
1 ⎣ dh ⎝ N
Iidiff = ρi Fd (i) + Fs (i) + Bj Fij ⎠ a function of the canopy structure (i.e., LAI and LAD) and the
π j=1
leaf surface parameters (refraction index and roughness).
 ⎤

N
III. VALIDATION
+ τidh Fd (i + N ) + Fs (i + N ) + Bk Fik ⎦
k=1 The summer maize is an important food and forage crop
 widely planted in Northern China. Several complete sets of
ρs−dh in situ measurements for summer maize are selected to analyze
Iispec1 (Ψo ) = ρsi (Ψs , Ψo ) − i Fs (i)
π and validate its directional reflectance characteristics in this
N  study. All data are from a “Spectral Knowledge Library of Typ-
spec2 ρs−dh ical Land Surface Objects in China” [28], which has collected a
Ii (Ψo ) = ρsi (Ψs , Ψo ) − i Bj Fij . (18)
j=1
π lot of spectral data measured in the field.

According to (6), the classic radiosity of facet i can be A. Experiment


written as
⎛ ⎞ The site of the summer maize experiments is located at

N
Luancheng Agricultural Ecosystems Experimental Station,
Bi = ρdhi
⎝Fd (i) + Fs (i) + Bj Fij ⎠
Chinese Academy of Sciences, Shijiazhuang city, Hebei
j=1
  province, China (37◦ 52 N, 114◦ 39 E), with an average eleva-

N
tion of 50.5 m. BRFs of maize canopy together with the spectral
+ τidh Fd (i + Np ) + Fs (i + Np ) + Bk Fik . (19)
properties of the canopy components and the canopy structural
k=1
data were measured in July of 2002.
Then, the diffuse scattering component can be rewritten as 1) 3-D Maize Canopy: The canopy structural data such as the
Iidiff = Bi /π. average height of the canopy; the height of each leaf base; the
The integral of (18) within the hemispherical directions can length, width, shape, and orientation of leaves; and the radii of
produce the following: stems were measured by photographic method [29]. Empirical
  models proposed by España et al. [7] are improved and fitted
Iispec1 (Ψo )dΩo = Iispec2 (Ψo )dΩo = 0. with the measured structural data to generate a time series
2π 2π
of realistic 3-D maize canopies [29]. All components (leaves,
 stems, and soil) are discretized into many small facets for use
Then, we obtain 2π Ii (Ψo )dΩo = Bi , which satisfies the en- in the RGM [21].
ergy conservation laws. It should be noted that the radiosity B In this paper, a 3-D maize canopy similar to the real one
in (19) does not have the same value calculated by the classic on July 23, 2002, is rendered based on the field measurement
radiosity method. (Fig. 1). Its average row-width and plant-to-plant interval are
624 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 55, NO. 2, FEBRUARY 2017

Fig. 1. 3-D maize canopy (LAI = 2.23).

60 and 25 cm, respectively. The average height of the canopy is


about 1.09 m. The LAI of the maize canopy is 2.23 (2.24 in real
canopy).
2) Spectral Properties of Canopy Components: In the 3-D Fig. 2. Measured spectral curves of leaf, stem, and soil.
maize canopy, there are three categories of components,
namely, leaves, stems, and soil background. Their spectral TABLE I
properties were measured with a spectrophotometer (SE590, O PTICAL PARAMETERS OF C OMPONENTS IN THE S CENE
Spectron Engineering).
First, NHR and NHT spectral data from ten pieces of summer
maize leaves, corresponding to the tender and mature leaves
from top to down in one maize plant, were measured on July 21,
2002, by a laboratory spectrophotometer equipped with an
integrating sphere. According to their spectral similarity, all
maize leave reflectance spectra can be divided into three groups,
i.e., new leaf, first spread leaf, and all other leaves. All leaf
transmittance spectra are similar. Therefore, the NHR of the
leaves (except for the new leaf and the first spread leaf) and
NHT of all leaves are averaged as representatives.
The stems were pressed into slabs, and its NHR spectral data
were measured like a leaf in the laboratory.
Third, we measured bare soil BRF before maize was planted
in the field and found that soil BRF in the study area was almost TABLE II
L EAF S URFACE PARAMETERS
isotropic. Therefore, a Lambertian soil assumption was used,
and the nadir reflectance spectral properties of soil background
were measured in the field by the spectrophotometer right after
the maize canopy BRFs were measured.
All components of measured spectral data are shown
in Fig. 2.
Four representative wavebands, NIR (the central wavelength
is 804.4 nm, and the bandwidth is ∼15 nm), red (699.6 nm),
green (547.5 nm), and blue (435.8 nm), are adopted to simulate According to the measured time and the location, the solar
the canopy BRF, and the optical parameters are listed in Table I. incident direction can be determined with the zenith angle of
A group of maize leaf surface parameters (Table II) retrieved 38.14◦ and the azimuth of 259.18◦.
based on a set of polarized BRF measured in the laboratory [30] In addition, the ratios of the direct to the total incident
is adopted here. Then, ρs−dh (λ, 0o ) is calculated with numeri- light were measured when measuring canopy BRF. They were
cal integration of specular reflectance model (Appendix). 88.2% (blue), 91.7% (green), 94.1% (red), and 91.7% (NIR),
3) Canopy BRF Measurements: A multiangle observation de- respectively.
vice equipped with the SE590 spectrophotometer was used to
measure maize canopy BRFs. The detector of the spectropho-
B. BRF Simulations
tometer was mounted at the end of the arm of the bracket
device, with the height of 1.88 m, which can ensure the rep- BRFs are first simulated for the maize canopy by the classic
resentation of the canopy scene in the field of view. RGM with Lambertian leaves and then by the modified RGM
A group of BRFs measured in near principal plane at local with nondiffuse leaves. The simulated BRFs are compared with
time 16:00, July 23, 2002, is chosen for comparison (Fig. 3). in situ measured data (Fig. 3).
XIE et al.: INFLUENCES OF LEAF-SPECULAR REFLECTION ON CANOPY BRF CHARACTERISTICS 625

Fig. 3. Comparison of the simulated with the measured BRFs.

which makes up no more than 1.25% in VIS and 2.0% in NIR of


the canopy BRF. That means BRFspec2 generally can be ignored
for computational efficiency, which can be further confirmed in
the next section.
By contrast, BRFspec1 is independent of wavelength and
obviously has directional characteristics, such as the specular
peak in the forward directions with its location related to the
canopy structure. The ratio of BRFspec1 to the canopy BRF can
reach up to 52.7% in blue in specular direction, much bigger
than BRFspec2 .

IV. S ENSITIVITY A NALYSIS


To further evaluate the influence of some major impact
Fig. 4. Comparison of the canopy BRFspec1 and BRFspec2 . factors, such as canopy structures and leaf surface condition,
on the specular reflection distribution of the canopies, several
As shown in Fig. 3, the BRFs simulated with nondiffuse maize canopies with different LADs are created by changing
leaves are fitted better with the measured data in forward direc- each leaf shape (Fig. 5). The mean leaf angle (MLA) of maize
tions than those with Lambertian leaves. Moreover, both sets canopy is used as an indicator of the canopy type. When MLA
of BRFs in other directions are similar, following the measured increases, the canopy LAD is closing to erectophile. On the
data well. other hand, it closes to planophile as MLA decreases.
There is a drop of BRF in NIR band at about 25◦ of view The mature leaf, stem, and soil optical parameters in Ta-
direction. That may be caused by the shadow of the detector or ble I are still used here. The three sets of leaf surface parameters
the shelf of the device. selected from Bousquet et al. [4] are retrieved based on mea-
BRFspec1 and BRFspec2 , contributed by the first- and high- sured BRFs from laurel (Prunus laurocesarus), European beech
order specular reflectance of leaves in the canopy, are calculated (Fagus silvatica), and hazel (Corylus avellana) leaves, respec-
separately and shown in Fig. 4. tively. These surface parameters represent three kinds of leaves
BRFspec2 is almost independent of view directions but de- (smooth, intermediate, and rough; Table II) and can be used to
pendent on wavelength. BRFspec2 overall is very small (less analyze the impact of different leaf surface characteristics on
than 4.0 × 10−4 in the VIS domain and 6.8 × 10−3 in NIR), the canopy BRF.
626 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 55, NO. 2, FEBRUARY 2017

hemisphere is also defined to evaluate the total impacts of leaf


directional reflectance on the canopy BRF

1 
Nv
AE = |ΔBRFvi | (22)
Nv vi=1

1 
Nv
|ΔBRFvi |
RE = (23)
Nv vi=1 BRFnew
vi

where N v is the number of view directions in the upper


hemisphere.
Five factors are considered in the following section, includ-
ing leaf surface properties, view direction, LAD, LAI, and
incident solar zenith angle, to analyze their impact on leaf-
specular contribution to canopy BRFs over maize canopies
(Figs. 1 and 5). Several abstract canopies are also generated to
specially analyze the impact of LAI.
1) Leaf Surface Properties: Three types of leaf surface prop-
erties (Table II) represent the smooth (case 1), intermedi-
ate (case 2), and rough (case 3) leaf surfaces, respectively.
Fig. 6(a) shows typical canopy reflectance in the opposite solar
direction for case 1, whereas for case 2 and case 3, no obvious
canopy specular reflectance is observed even for those canopies
with planophile LADs (small MLA) [Fig. 6(b) and (c)].
Fig. 5. 3-D maize canopies with different MLAs. (a) MLA = 79.65.
(b) MLA = 49.27. (c) MLA = 33.18. (d) MLA = 23.74. (e) MLA = 17.65.
However, canopies with more horizontal rough-surface
leaves will cause additional difference in the backward scat-
tering directions (Fig. 7). It shows that, with the Lambertian
To emphasize the impact of leaf-specular reflection on the assumption, canopy reflectance is overestimated (ΔBRF < 0)
canopy BRF, the diffuse skylight is ignored so that the solar around the hot-spot direction but may be under- or overes-
direct light ratio is assumed to be 1.0 in all wavebands. timated around the “cold spot” direction depending on leaf
surface properties and the canopy MLA. In other words, even
A. First- and High-Order Specular Reflectance of Canopies for the rough-surface leaf, the leaf directional reflection still
BRFspec1 and BRFspec2 of the maize canopies with different greatly affects light distribution in the canopy. For example, RE
MLAs are calculated, respectively (Fig. 6). in case 3 is larger than those of the other two cases [Fig. 10(b)].
The angular distribution of BRFspec1 is significantly affected 2) View Direction: The leaf directional reflectance will af-
by the canopy MLA. BRFspec1 generally increases in forward fect the light distribution within the canopy not only at the
scattering directions with the decrease of MLA, and an obvious specular direction. For example, as shown in Fig. 7, there is
specular peak is often presented when leaf surface is smooth an overestimation in most backward scattering directions and
[Fig. 6(a)]. BRFspec2 is related to wavelength and increases an underestimation in some forward scattering directions. For
with leaf directional-hemispherical reflectance. According to case 1 of leaf surface, the relative difference near the specular
our simulated data, BRFspec2 in the NIR band is ten times direction increases with the decrease of MLA, and the maxi-
larger than that in the VIS domain, such as red band. However, mum relative difference appears in the specular direction (ap-
compared to BRFspec1 , BRFspec2 in any wavelength is much proximately 40% at the red band), but the relative differences
smaller. For example, based on the simulated data of these in backward directions are quite small [Fig. 8(a)]. However, for
maize canopies, the average of BRFspec2 with different leaf case 2 [Fig. 8(b)] and case 3 [Fig. 8(c)] of leaf surfaces, there
surface types is limited in the range of 10−4 ∼ 10−5 . Therefore, are larger relative differences at higher view angles. Moreover,
it is acceptable in most canopies to retain BRFspec1 and ignore the smallest relative differences appear at the forward scattering
BRFspec2 in modeling. directions and move from small zenith angles to big ones as
MLA increases.
3) LAD: LAD has a big influence on canopy BRF with
B. Impact Factors on Leaf-Specular Contribution
non-Lambertian leaves. BRFspec1 of the canopy with more
The absolute BRF difference (ΔBRFvi = BRFnew vi − horizontally orientated leaves will increase around specularly
BRFold
vi ) and relative BRF difference (|ΔBRF vi |% = |ΔBRF vi |/ reflected directions [Fig. 6(a)]. Additionally, increasing canopy
BRFnew
vi × 100%) in a given view direction between the reflectance can be observed in the opposite solar direction only
traditional RGM (old) and the modified RGM (new) are used from canopies with small MLA [Fig. 8(a)]. It is understandable
to quantitatively estimate canopy BRF difference over the that, the smoother the leaves that are oriented horizontally, the
same scene between diffuse and nondiffuse leaves. The sum more significant the canopy specular reflectance is. On the other
of absolute (AD) and relative differences (RD) in the upper side, there is no obvious specular reflectance of the canopy with
XIE et al.: INFLUENCES OF LEAF-SPECULAR REFLECTION ON CANOPY BRF CHARACTERISTICS 627

Fig. 6. Comparison of BRFspec1 (a)–(c) and the average of BRFspec2 (d) in principal plane for maize canopies in Fig. 5 (incident zenith angle = 40◦ ).

spherical or uniform LAD because the leaf-specular reflection specular reflectance [4], which changes the angular distribution
contribution to the canopy BRF spreads into a large range of of canopy reflectance.
outgoing angles, rather than one particular direction. Therefore, Canopy BRFs at different incident zenith angles are simu-
comparing to canopies with a horizontal LAD, it is difficult to lated (Fig. 1). RE and AE of the canopy BRF are calculated
measure specular reflectance above the canopy with a spherical (Fig. 10). Obviously, both RE and AE remain small when the
or uniform LAD in the field. However, it does not mean that incident zenith angle is smaller than 40◦ but increase quickly
leaf-specular reflection can be ignored in these canopies. after it is above 40◦ . In other words, leaf directional reflection
4) LAI: To control the canopy LAI, some abstract canopies has to be considered when modeling canopy reflectance at high
are generated, which are composed of leaves, distributed incident zenith angle (exceeding 40◦ ).
randomly on the soil background (length: 10 m, width: 10 m, Although the canopy specular reflectance is much larger with
and height: 2 m). All leaves consist of simple square facets with smooth leaf surface than that with rough leaf surface in some
four sides of 0.03 m each. The canopy LAI varies from 0.5 to view directions (such as the specular direction), the trend of AE
4.0. A uniform LAD function is used to produce leaf normal and RE with different leaf surface conditions is the opposite.
distribution [22]. That is, the rougher the leaf surface is, the greater AE and RE
AE and RE are calculated based on the simulated BRF from are (see Figs. 9 and 10).
each abstract canopy with different LAIs (Fig. 9). AE and RE
increase with LAI for all three leaf surface conditions. For
V. C ONCLUSION
example, the maximum RE could be greater than 50% (case 3)
when LAI exceeds 4.0. However, RE and AE will reach to the In this paper, we have developed a modified RGM coupled
saturation point if LAI continues to increase. It is well known with a leaf BRDF model to improve the simulation accuracy
that canopy reflectance will peak off as LAI increases, as well of the canopy BRF. The new model takes advantage of the
as the contribution of leaf directional reflection to the canopy radiosity method, which is superior for calculating the multiple
reflectance, because it mostly originates from the upper layer diffuse scattering components over other canopy models. Com-
leaves. pared with other radiosity-based models that consider specular
5) Solar Zenith Angle: Generally, canopy directional re- surfaces, the modified RGM simplifies the calculation process
flectance depends on the sun-view geometry, so does the leaf- of radiative transfer between leaves by separating the first- and
628 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 55, NO. 2, FEBRUARY 2017

Fig. 7. Comparison of absolute difference of the canopy BRF in NIR waveband in principal plane over maize canopies in Fig. 5 (incident zenith angle = 40◦ ).

high-order leaf-specular reflection process. The simulated tain canopies with more vertically oriented leaves, the under-
BRFs are validated using a group of in situ measured BRFs estimation occurs in most directions, except for large view
over the maize canopy. zenith angles in the forward scattering direction. Based on our
Several 3-D maize canopies with different LADs are used to simulation results, REs increase with the increment of LAI.
analyze the impact of leaf directional reflection on the canopy However, RE and AE will reach maxima if LAI exceeds 3.0
BRF. The influence of some key factors on the canopy BRF for a uniform canopy (Fig. 9). Specifically, AE will saturate
with diffuse or nondiffuse leaves is evaluated quantitatively, more quickly than RE. This observation explains the finding
i.e., leaf surface properties, view direction, LAI, LAD, and by Combes et al. [6] that the Lambertian approximation is
incident solar zenith angle. It is found that the simulated canopy less suitable for sparse canopies than dense canopies because
BRF with the Lambertian leaf hypothesis is often overestimated AE becomes relatively large when canopy LAI is small. Con-
in the backward scattering directions but underestimated in sidering that RE is wavelength dependent and larger in the
the forward scattering directions (Fig. 7). However, for cer- visible bands than that in NIR bands, it is necessary to consider
XIE et al.: INFLUENCES OF LEAF-SPECULAR REFLECTION ON CANOPY BRF CHARACTERISTICS 629

Fig. 9. (a) AE and (b) RE of BRFs in red waveband from abstract canopies
with different LAIs (the incident zenith angle is 40◦ ).

high-order leaf-specular reflection is approximated using the


approach similar to that for diffuse reflection.
Although our study is performed over crop canopy (maize),
the method and findings are applicable to other canopies as
well, such as broad leaf forest and bushes. However, a detail
investigation is still needed for other canopies. In this paper, we
have focused on the leaf-specular reflection to understand its
impact on the canopy BRF, so a Lambertian bare soil is assumed
and observed in the field in our study. However, in general,
the soil background is non-Lambertian; its BRDF effect could
be comparable or even larger than that from leaf BRDF. This
can be accounted for by turning on the non-Lambertian soil
reflectance module built in the RGM using Jacquemoud model
[16], a bare soil BRDF model developed based on Hapke
model [15] after determining the five input parameters in the
Fig. 8. Comparison of the relative difference of BRFs in red waveband from model through fitting the measured soil BRF data.
canopies with different MLAs (Fig. 5) and leaf surface parameters. (a) Case 1. In the future, we also expect to simulate multispectral and
(b) Case 2. (c) Case 3.
high-resolution images of canopies with nondiffuse leaves to
analyze the impact of canopy structures (such as LAD) on the
leaf-specular reflection when simulating canopy BRF at the
polarization property of canopy scattering, which is closely
visible domain, which accounts for greater than 50% of RE.
related with leaf-specular reflection.
By separating BRFspec1 and BRFspec2 , we have achieved an
optimal balance between simulation accuracy and computation
A PPENDIX
complexity because first-order leaf-specular reflection accounts
S PECULAR R EFLECTANCE M ODEL OF L EAF S URFACE
for up to 98% of the total leaf-specular reflection contribution
to the canopy BRF. Therefore, in the modified RGM, the first- According to the leaf BRDF model developed by
order leaf-specular reflection is accurately simulated, but the Bousquet et al. [4], ρd is assumed to be independent of the
630 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 55, NO. 2, FEBRUARY 2017

According to the definition of DHRF, DHRFspec of the leaf


surface can be derived by integration over the view hemisphere,
which can be discretized as needed in a numerical integration
π

2 

s−dh
ρ (λ, θi , φi ) = ρs (λ, θi , φi , θo , φo , n, σ)
θo =0 φo =0

× cos θo sin θo Δθo Δφo . (A4)

R EFERENCES
[1] B. Andrieu, F. Baret, S. Jacquemoud, T. Malthus, and M. Steven, “Evalu-
ation of an improved version of SAIL model for simulating bidirectional
reflectance of sugar beet canopies,” Remote Sens. Environ., vol. 60, no. 3,
pp. 247–257, 1997.
[2] C. C. Borel, S. A. W. Gerstl, and B. J. Powers, “The radiosity method in
optical remote sensing of structure 3-D surfaces,” Remote Sens. Environ.,
vol. 36, pp. 13–44, 1991.
[3] C. C. Borel and S. A. W. Gerstl, “Nonlinear spectral mixing models for
vegetative and soil surfaces,” Remote Sens. Environ., vol. 47, pp. 403–416,
1994.
[4] L. Bousquet, S. Lachérade, S. Jacquemoud, and I. Moya, “Leaf BRDF
measurements and model for specular and diffuse components differenti-
ation,” Remote Sens. Environ., vol. 98, pp. 201–211, 2005.
[5] M. Chelle, “Could plant leaves be treated as Lambertian surfaces in
dense crop canopies to estimate light absorption?” Ecol. Model., vol. 198,
pp. 219–228, 2006.
[6] D. Combes, L. Bousquet, S. Jacquemoud, H. Sinoquet, C. Varlet-Granche,
and I. Moya, “A new spectrogoniophotometer to measure leaf spec-
tral and directional optical properties,” Remote Sens. Environ., vol. 109,
pp. 107–117, 2007.
[7] M. L. España, F. Baret, F. Aries, M. Chelle, B. Andrieu, and L. Prévot,
“Modeling maize canopy 3D architecture application to reflectance simu-
lation,” Ecol. Model., vol. 122, pp. 25–43, 1999.
[8] J. P. Gastellu-Etchegorry, V. Demarez, V. Pinel, and F. Zagolski, “Model-
Fig. 10. (a) AE and (b) RE of BRF in red waveband at different incident zenith ing radiative transfer in heterogeneous 3-D vegetation canopies,” Remote
angles (for maize canopy of Fig. 1). Sens. Environ., vol. 58, pp. 131–156, 1996.
[9] S. A. W. Gerstl and C. C. Borel, “Principles of the radiosity method versus
radiative transfer for canopy reflectance modeling,” IEEE Trans. Geosci.
illumination-view geometry and only dependent on the wave- Remote Sens., vol. 30, no. 2, pp. 271–275, Mar. 1992.
[10] N. S. Goel, “Models of vegetation canopy reflectance and their use in
length. ρs relates to the roughness σ and refractive index n of estimation of biophysical parameters from reflectance data,” Remote Sens.
the leaf surface Rev., vol. 4, pp. 1–212, 1988.
[11] N. S. Goel, I. Rozehnal, and R. L. Thompson, “A computer graphics based
F (n, θa )G(θi , θo , φo ) model for scattering from objects of arbitrary shapes in the optical region,”
ρs (λ, θi , φi , θo , φo , n, σ) = Remote Sens. Environ., vol. 36, pp. 73–104, 1991.
4π cos θi cos θo
 [12] N. S. Goel and W. Qin, “Influences of canopy architecture on relation-
2 ships between various vegetation indices and LAI and FPAR: A computer
exp − tan
σ2
α
simulation,” Remote Sens. Rev., vol. 10, pp. 309–347, 1994.
× (A1) [13] Y. M. Govaerts and M. M. Verstraete, “Raytran: A Monte Carlo ray-
σ 2 cos4 α tracing model to compute light scattering in three-dimensional hetero-
where θi and φi are the illumination zenith and azimuth angles; geneous media,” IEEE Trans. Geosci. Remote Sens., vol. 36, no. 2,
pp. 493–505, Mar. 1998.
θo and φo are the viewing zenith and azimuth angles (in [14] L. Grant, “Diffuse and specular characteristics of leaf reflectance,”
spherical coordinates) Remote Sens. Environ., vol. 22, pp. 209–322, 1987.
 [15] B. W. Hapke, “Bidirectional reflectance spectroscopy 1. Theory,” J.
2 cos α cos θo 2 cos α cos θi Geophys. Res., vol. 86, pp. 3039–3054, 1981.
G(θi , θo , φo ) = min 1, , [16] S. Jacquemoud, E. Baret, and J. F. Hanocq, “Modeling spectral and bidi-
cos θa cos θa rectional soil reflectance,” Remote Sens. Environ., vol. 41, pp. 123–132,
(A2) 1992.
where cos α =(cos θi +cos θo )/2 cosθa , cos2θa = cosθi cosθo + [17] S. Liang, Quantitative Remote Sensing of Land Surfaces. Hoboken, NJ,
USA: Wiley, 2004.
sin θi sin θo cos φo . [18] R. B. Myneni, I. Impens, and G. Asrar, “Simulation of space measure-
The Fresnel-reflectance term in the model is modified to ments of vegetation canopy bidirectional reflectance factors,” Remote
consider the polarized reflectance measured from the leaf sur- Sens. Rev., vol. 7, no. 1, pp. 19–41, 1993.
[19] J. L. Monteith, Vegetation and the Atmosphere (Vol. 1: Principles).
face [30] London, U.K.: Academic, 1975.
2 2 [20] F. E. Nicodemus, J. C. Richmond, J. J. Hsia, I. W. Ginsberg, and
n·cos θa −cos θt cos θa −n·cos θt T. Limperis, “Geometrical Considerations and Nomenclature for Re-
n·cos θa +cos θt + cos θa +n·cos θt flectance,” Nat. Bureau Std., US Dept. Commerce, Washington, DC,
F (n, θa ) = . (A3) USA, NBS MN-160, p. 52, Oct. 1977.
2 [21] W. Qin and S. A. W. Gerstl, “3-D scene modeling of semidesert vege-
tation cover and its radiation regime,” Remote Sens. Environ., vol. 74,
Based on Snell’s law, sin θt = sin θa /n. pp. 145–162, 2000.
XIE et al.: INFLUENCES OF LEAF-SPECULAR REFLECTION ON CANOPY BRF CHARACTERISTICS 631

[22] J. Ross and A. Marshak, “The influence of leaf orientation and the specu- Peijuan Wang received the Ph.D. degree in ge-
lar component of leaf reflectance on the canopy bidirectional reflectance,” ography and geographic information systems from
Remote Sens. Environ., vol. 27, pp. 251–260, 1989. Beijing Normal University, Beijing, China, in 2006.
[23] H. E. Rushmeier and K. E. Torrance, “Extending the radiosity method From 2006 to 2008, she was a Research As-
to include specularly reflecting and translucent materials,” ACM Trans. sistant with the Chinese Academy of Meteorologi-
Graph., vol. 9, pp. 1–27, 1990. cal Sciences, China Meteorological Administration,
[24] M. Z. Shao, Q. S. Peng, and Y. D. Liang, “A new radiosity approach by Beijing. From 2008 to 2016, she was a Research
procedural refinements for realistic image synthesis,” ACM SIGGRAPH Associate Professor with the Chinese Academy of
Comput. Graph., vol. 22, no. 4, pp. 93–101, 1988. Meteorological Sciences. Since April 2016, she has
[25] J. Stuckens, B. Somers, S. Delalieux, W. W. Verstraeten, and P. Coppin, been a Research Professor. Her research interests
“The impact of common assumptions on canopy radiative transfer simu- include agricultural remote sensing modeling and the
lations: A case study in citrus orchards,” J. Quantitative Spectr. Radiative effects of agro-meteorological disasters on agriculture.
Transfer, vol. 110, no. 1/2, pp. 1–21, 2009.
[26] W. Verhoef, “Light scattering by leaf layers with application to canopy
reflectance modeling: The SAIL model,” Remote Sens. Environ., vol. 16, Yanmin Shuai received the Ph.D. degree in geog-
pp. 125–141, 1984. raphy and in remote sensing from Beijing Normal
[27] J. R. Wallace, M. F. Cohen, and D. P. Greenberg, “A two-pass solution to University, Beijing, China, and Boston University,
the rendering equation: A synthesis of ray tracing and radiosity methods,” Boston, MA, USA, respectively, in 2008 and 2009,
Comput. Graph., vol. 21, no. 4, pp. 311–320, 1987. with support from the joint doctoral program
[28] J. D. Wang et al., Typical Objects Spectral-Knowledge Base of China between Beijing Normal University and Boston
(in Chinese). Edinburgh, NSW, Australia: Science, 2009. University.
[29] M. X. Wu et al., “Approach for computation of structural parameters and After graduation, she became a Research Scientist
visual research of canopy in summer corn (in Chinese),” ACTA Agronom- with Earth Resources Technology, Inc., working on
ica Sinica, vol. 28, no. 6, pp. 721–726, 2002. NASA and USGS projects at Goddard Space Flight
[30] D. H. Xie, P. J. Wang, Q. J. Zhu, and H. M. Zhou, “Modeling polarimetric Center and University of Massachusetts Boston,
BRDF of leaves surfaces (in Chinese),” Spectr. Spectral Anal., vol. 30, Boston. Her research interests are focused on the development of routine
no. 12, pp. 3324–3328, 2010. direct broadcast anisotropic and radiative products for MODIS as well as
[31] D. H. Xie, R. Sun, Q. J. Zhu, J. D. Wang, M. X. Wu, and W. H. Qin, “Re- approaches for Landsat albedo, the detection and monitoring of phenological
flectance distribution of corn canopy simulated with radiosity-graphics events over agriculture and forested regions, and the radiative evolution of
combined model (in Chinese),” ACTA Agronomica Sinica, vol. 32, terrestrial ecosystems disturbed by fires, harvesting, and insect epidemics.
no. 3, pp. 317–323, 2006.

Yuyu Zhou received the B.S. degree in geography


Donghui Xie received the Ph.D. degree in remote and the M.S. degree in remote sensing from Beijing
sensing and geographic information systems from Normal University, Beijing, China, and the Ph.D.
Beijing Normal University, Beijing, China, in 2005. degree in environmental sciences from the University
From 2005 to 2007, she was a Postdoctoral Re- of Rhode Island, Kingston, RI, USA.
search Associate with Beijing Normal University. He is a Physical Geographer with the Department
She is currently with the State Key Laboratory of Re- of Geological and Atmospheric Sciences, Iowa State
mote Sensing Science, School of Geography, Beijing University, Ames, IA, USA. His research focuses
Normal University. Her research interests include on applications of remote sensing, GIS, integrated
canopy radiative transfer modeling and biophysical assessment modeling, and spatial analysis to under-
parameter retrieval of vegetation. stand the problems of environmental change and
their potential solutions.

Wenhan Qin (M’96) received the B.S. degree in Qijiang Zhu received the B.S. degree in geography
meteorology from Nanjing Institute of Meteorology from Peking University in 1960. He is a professor
and Graduate School, Nanjing, China, in 1983, the with School of Geography, Beijing Normal Univer-
M.S. degree in meteorology from the University of sity, China. His research focuses on environmental
Science and Technology of China, Hefei, China, in remote sensing and ecological modeling.
1986, and the Ph.D. degree in geoscience and remote
sensing from the Graduate School, University of
Science and Technology of China, in 1992.
He was a Postdoctoral Fellow with the Depart-
ment of Computer Science, Wayne State University,
Detroit, MI, USA, from 1993 to 1995. He is currently
a Senior Staff Scientist with Science Systems and Applications, Inc., Lanham,
MD, USA. He is the author of more than 80 scientific articles, including over
40 papers in refereed scientific journals, and has contributed to several books.
His early research interest focused on 3-D scene BRDF model development
and applications in remote sensing and ecological studies. His recent interests
include satellite remote sensing of atmospheric properties and the effects of
non-Lambertian underlying surface on the retrieval of trace gases, aerosol, and
cloud properties for satellite UV/VIS algorithms, as well as satellite sensor
calibration and validation.
Dr. Qin is one of the group recipients of several NASA awards such as
the 2013 NASA Group Achievement Award for the Suomi NPP Development
Team, 2009 NASA Group Achievement Award for the OMI Instrument Team,
and 2006 Pecora Group Achievement Award for the Total Ozone Mapping
Spectrometer (TOMS) Team in recognition of creating 25-year-long TOMS
data sets of unprecedented accuracy, significantly advancing the study of global
ozone and aerosols.

You might also like