Remote Sensing of Environment 255 (2021) 112302

Contents lists available at ScienceDirect

Remote Sensing of Environment

journal homepage: www.elsevier.com/locate/rse

Multi-angular reflectance spectra of small single trees

Petri R. Forsström a, *, Aarne Hovi a, Giulia Ghielmetti b, Michael E. Schaepman b,
Miina Rautiainen a, c
a
Department of Built Environment, Aalto University, P.O. Box 14100, 00076 Aalto, Finland
b
Department of Geography, Remote Sensing Laboratories, University of Zürich, Winterthurerstrasse 190, 8057 Zurich, Switzerland
c
Department of Electronics and Nanoengineering, Aalto University, P.O. Box 15500, 00076 Aalto, Finland

A R T I C L E I N F O A B S T R A C T

Keywords: Understanding the reflectance anisotropy of forests and the underlying scattering mechanisms is needed to
Goniometer improve the accuracy of retrievals of fundamental forest characteristics from optical remote sensing data. In this
Scattering paper, we developed a laboratory measurement set-up for a large goniometer (LAGOS) and measured multi-
Anisotropy
angular spectra (350–2500 nm) of 18 small trees, composed of three common European tree species: Scots
Pine
pine (Pinus sylvestris L.), Norway spruce (Picea abies (L.) H. Karst), sessile oak (Quercus petraea (Matt.) Liebl.). For
Spruce
Oak all trees, we measured tree spectra in 47 view angles in the upper hemisphere. To our knowledge, this is the first
Tree spectrum study reporting multi-angular reflectance spectra of single trees. We also measured the reflectance and trans­
mittance spectra of needles and leaves, as well as reflectance spectra of bark of the sample trees. We analyzed the
spectro-directional characteristics of the trees, and the inter- and intraspecific variations of these characteristics.
The anisotropy of trees was shown to be strongly asymmetrical and characteristic to species: while pine and
spruce exhibited strong hotspot effects, oak showed a strong specular component. Our results indicate that
simultaneous measurements of both spectral and directional characteristics of trees may enhance the discrimi­
nation of species and thus, support the retrieval of information of their biophysical properties.

1. Introduction monitoring of vegetation (e.g., Asner et al., 1998; Knyazikhin et al.,

1998; Chen et al., 2012; Pisek et al., 2016), because the reflectance
Currently, a wealth of satellite data is available for environmental anisotropy of vegetation carries in it fundamental information on
monitoring applications. However, so far there have been only a few structural characteristics of canopies. Especially forest canopies have
satellite instruments providing data on the directional scattering prop­ strongly anisotropic scattering patterns. The patterns have first been
erties of land areas. The most well-known instruments include NASA’s analyzed through radiative transfer modeling (e.g., Li and Strahler,
Multi-angle Imaging SpectroRadiometer (MISR) (Diner et al., 1998), 1992; Roujean et al., 1992; Jacquemoud, 1993; Gerard and North, 1997;
ESA’s Compact High Resolution Imaging Spectrometer (CHRIS) Chen and Leblanc, 1997; Rautiainen et al., 2004; Kobayashi and Iwa­
(Barnsley et al., 2004; Barnsley et al., 2000; Verrelst et al., 2010), and buchi, 2008; Kuusk et al., 2014) and later on using also multi-angular
CNES’s POLarization and Directionality of the Earth’s Reflectances in­ remote sensing data (e.g., Lacaze and Roujean, 2001; Canisius and
strument (POLDER) (Deschamps et al., 1994). In addition, spectro- Chen, 2007; Rautiainen et al., 2008; Verrelst et al., 2010). In general,
directional characteristics have been retrieved globally from e.g., these studies have demonstrated that the geometric structure of can­
MODerate resolution Imaging Spectroradiometer (MODIS) (Justice opies has a notable impact on the reflectance anisotropy of forests. More
et al., 2002) data of large areas, and in smaller areas from multi-angular specifically, strong reflectance anisotropy has been observed in conif­
airborne (e.g., Bréon et al., 1997; Sandmeier and Deering, 1999; Lobell erous forests (e.g., Deering et al., 1999). In forest canopies, the angular
et al., 2002; Korpela et al., 2011; Markiet et al., 2017) and unmanned effects substantially impact retrievals of biophysical variables using
aerial vehicle (UAV) data (Roosjen et al., 2018). indices (Verrelst et al., 2008) and are slightly larger in the visible (VIS)
Multi-angular remote sensing data have versatile uses in e.g., global than in the near-infrared (NIR) spectral region (e.g., Rautiainen et al.,

* Corresponding author.
E-mail addresses: [email protected] (P.R. Forsström), [email protected] (A. Hovi), [email protected] (G. Ghielmetti), michael.schaepman@
geo.uzh.ch (M.E. Schaepman), [email protected] (M. Rautiainen).

https://doi.org/10.1016/j.rse.2021.112302
Received 9 September 2020; Received in revised form 17 December 2020; Accepted 9 January 2021
Available online 21 January 2021
0034-4257/© 2021 The Author(s). Published by Elsevier Inc. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
P.R. Forsström et al. Remote Sensing of Environment 255 (2021) 112302

2008). The hot and dark spots are the largest in the principal plane and At the moment, the only operational goniometer which is large enough
decrease when the observation plane moves away from the principal for measuring small trees in both laboratory and field settings (Dangel
plane (e.g., Canisius and Chen, 2007). The angular characteristics at et al., 2005) is the LAGOS set-up at the University of Zürich (Schopfer
forest stand or tree level may also be studied in the future in more detail et al., 2008; Sandmeier and Itten, 1999). LAGOS is suitable for
using UAV-based remote sensing platforms. However, planning data measuring small trees and larger single trees can be measured in nature
collection by either air- or spaceborne platform relies on the availability using directional drones or cranes.
of high-quality reference data, typically collected at ground level or In this paper, we present the first empirical data on the spectro-
modeled (Schneider et al., 2014). directional characteristics of small single trees. We hypothesized that
Trees are typically the largest individual contributors to the spectra since tree spectra are strongly influenced by multiple scattering events
of a dense forest although the contribution of understory and soil layers within a crown, the species-specific distribution and orientation of
(i.e., forest background) should not be ignored in sparser or seasonally scattering elements in tree crowns (together with leaf and bark optical
dynamic forests (Rautiainen et al., 2011; Rautiainen and Lukeš, 2015). properties) introduce species-specific spectro-directional characteristics
Although forest spectra are continuously measured by satellite in­ for trees. To obtain the data, we developed a laboratory measurement
struments, reference data on the directional scattering properties of set-up for a large goniometer and measured multi-angular spectra
single trees are not available to date. Such multi-angular reference data, (350–2500 nm) of 18 small trees belonging to three common European
whether measured in a laboratory or in-situ, can be collected by utilizing tree species. Additionally, we measured the reflectance and trans­
a goniometer design that allows recording of radiation scattered by the mittance spectra of needles and leaves, and reflectance spectra of bark of
sample in different view angles. However, goniometers are, in general, the sample trees. We used the data to address the following research
laborious and slow to use, and so far, spectro-directional characteristics questions: 1) what are the spectro-directional characteristics of small
have been measured only for very limited vegetation types, e.g., ling­ single trees? and 2) how large are the inter- and intraspecific variations
onberry and blueberry shrubs (Forsström et al., 2019), Scots pine shoots of these characteristics? The measured data are openly available (Hovi
(Mõttus et al., 2012; Rautiainen et al., 2012), mosses, lichens and dwarf et al., 2021 (submitted)).
shrubs (Peltoniemi et al., 2005; Kuusinen et al., 2020), tree bark (Juola
et al., 2020), and grass and crop species (e.g., Lunagaria and Patel, 2017; 2. Materials and methods
Roosjen et al., 2012; Roosjen et al., 2017). Yet due to the physical
constrains of a ground-based instrument, sensor, and illumination, 2.1. Samples
measuring single trees is even more tedious compared to the previously
measured species that grow relatively close to the ground. Thus, labo­ We measured spectra of evergreen conifers Scots pine (Pinus sylvestris
ratory goniometers capable of incorporating entire trees are very rare. L.) and Norway spruce (Picea abies (L.) H. Karst), and deciduous

Fig. 1. Silhouettes of sample trees in comparable scale. The trees were 38–70 cm in height. See Table 1 for details on the tree species and structural characteristics.

2
P.R. Forsström et al. Remote Sensing of Environment 255 (2021) 112302

broadleaved sessile oak (Quercus petraea (Matt.) Liebl.) (Fig. 1). The 2.2.1. Multi-angular measurements of tree spectra
trees were up to 4 years old and 0.38 to 0.70 m in height, i.e., young The goniometer measurements of tree spectra were always the first in
trees. The trees were obtained from an outdoor tree nursery in Zürich the daily measurement routine and were made directly after bringing
between DOYs 240 and 263 in 2018. At the nursery, the trees had been the sample tree in the laboratory. Measuring one tree took approxi­
grown in pots with regular watering and were exposed to direct sunlight. mately 1.5 h. The utilized system represents a traditional goniometer
The study trees had healthy green leaves or needles, and intra- and design in which the sensor optics of the spectrometer (ASD FieldSpec3,
interspecific variation in structure, i.e., trees were of different height, serial number 16006) can be pointed at the center of the goniometer
and had different crown level clumping and leaf area. (see Table 1 and from any view angle in the upper hemisphere, and the amount of scat­
Fig. 1). Variation in the structural properties of the study trees was tered radiative energy recorded (Fig. 2). The measurement geometry
considered an important source of natural variation of spectra. We did was biconical, meaning both the incident radiation and the sensor field
not select trees with heavily irregular crown shapes or strongly bent of view (FOV) had conical angular characteristics (Schaepman-Strub
stems. The leaves of the sample trees were noted to have minor white et al., 2006).
residues of most likely calcium from the irrigation water applied at the We measured spectra of trees in view zenith angles (θ) 0◦ , ±21.2◦ ,
nursery. ±48.6◦ , ±76.2◦ , while having the goniometer half-arc turned in relative
The sample trees were stored at the university outdoor garden in azimuth angles (φ) 0◦ , 15◦ , 45◦ , 75◦ , 90◦ , 105◦ , 135◦ , 165◦ for each
watering beds with sprinklers on top to provide automatically controlled measurement (i.e., nadir was measured eight times). The azimuth 0◦ was
irrigation (scheduled daily at 6 am and 6 pm). We chose the watering in the principal plane, and positive view zenith angles indicate the
beds without a protective cover to keep the environmental conditions as backward viewing angles (towards illumination) and negative the for­
similar as possible to those at the nursery. The growing conditions, as ward viewing angles (away from illumination). The view zenith angles
well as the health of the sample trees were monitored visually every day. were chosen so that cos(θ) corresponds to nodes of Gauss-Legendre
Trees that showed visual symptoms of stress, e.g., yellowing or drooping integration, which enables, when needed, a reasonably accurate
of leaves or stem were disregarded from the measurements. Bamboo approximation of the spherical integral (Atkinson, 1982), and therefore
stakes were used to support the saplings and were removed only just also hemispherically reflected radiation, a pre-requisite for the work
before the spectral measurements (i.e., tree, and leaf or needle spectra), performed by Hovi et al. (2020a). To summarize, there were 47 different
which took approximately 6 h to complete for each tree. view angles, and because nadir was measured eight times, there were in
total 54 measurements per tree.
2.2. Measurements We used a bare fiber optic bundle as sensor optics for the spec­
trometer with an opening angle of 25◦ . With the fiber head at a 1.94 m
The measurements were made in a dark room at the University of distance from the goniometer center, a sample tree fit always fully in the
Zürich. The main instrument was the LAGOS goniometer in tandem with FOV of the sensor. Before and after measuring all spectra of a tree, we
a spectrometer. The interior of the goniometer laboratory has been took three averaged spectral reference readings from a diffuse (20 × 20
treated with special black paint to minimize reflections from its walls cm) Labsphere Zenith Lite™ 95% reflectance panel, placed in the center
and ceiling during measurements. Also, the instrument structures, such of the goniometer. The reference readings were used to convert the
as the stand holding the housing for the lamp, were covered with black spectral digital number (DN) data to a meaningful reflectance quantity,
canvases. With the aim to produce an extensive spectroscopic and as well as to validate the temporal stability of the light source during the
structural representation of the study trees, we made a series of mea­ measurements. In all view angles, we saved one averaged spectrum per
surements: (i) multi-angular spectra of entire trees using the goniometer, 10 measured spectra, using a 2.18 s integration time. The spectrometer
(ii) reflectance and transmittance spectra of leaves and needles, and and the light source lamp were allowed to warm-up at least 0.5 h before
reflectance spectra of bark, using an integrating sphere, and (iii) crown the measurements. A laptop computer with ASD RS3 software was used
level clumping using silhouette photographs, leaf surface areas, and leaf to save the spectra.
mass. Each measurement is described in detail in the following sub- As a light source, we utilized a broadband 1000 W tungsten-halogen
sections. lamp with a Thermo-Oriel housing and stabilized power supply. A
Köhler illuminator with an aspherical reflector and a condenser were
used to remove the effect of the lamp filament and to increase the
spectral and spatial homogeneity at the illuminated spot. The light
Table 1
source was fixed to a sturdy metal stand with vertical, horizontal, and
Structural parameters of sample trees. Tree species are Scots pine (Pinus sylvestris
L.), Norway spruce (Picea abies (L.) H. Karst), and sessile oak (Quercus petraea
tilt adjustments. During the measurements, the light source was inside
(Matt.) Liebl.). The silhouette to total area ratio (STAR) describes crown level the goniometer half-arc structure and as close to the outer edge as
clumping. possible in order to maximize the width of the light beam at the distance
of a tree sample. All sample trees were fully illuminated during the
Sample tree Height [cm] Green biomass [g] STAR
measurements from a zenith angle of θ = +40◦ , which is typical in
Pine 1 50 37.3 0.170 summer time acquisitions of satellite data in the midlatitudes, with the
Pine 2 62 96.8 0.138
Pine 3 69 87.2 0.153
maximum tree height still fitting inside the beam being around 70 cm.
Pine 4 70 116.6 0.145 The opening angle of the light beam was 22◦ and the lamp filament was
Pine 5 38 66.0 0.114 located 1.75 m from the goniometer center (Fig. 2).
Pine 6 50 59.1 0.135 To facilitate the measurements of tree spectra in the goniometer, we
Spruce 1 44 21.4 0.178
needed a non-reflective background in order to minimize the effect of
Spruce 2 41 38.0 0.122
Spruce 3 48 62.4 0.118 stray light in the recorded signal. By stray light we refer to any signal
Spruce 4 53 60.4 0.139 source other than the signal originating from the tree. In our measure­
Spruce 5 68 63.5 0.143 ments, most of the stray light was assumed to originate from the illu­
Spruce 6 70 36.1 0.183 minated fraction of the background. We constructed a 1.5 × 1.3 m
Oak 1 45 4.5 0.209
Oak 2 48 20.0 0.173
wooden frame, to support a spectrally black canvas (Fig. 2). The frame
Oak 3 46 9.6 0.195 had adjustable legs to allow correct vertical alignment for trees of
Oak 4 58 21.3 0.193 different size. We used four different frame heights in our measure­
Oak 5 58 23.0 0.209 ments: 60.0, 64.6, 69.2, and 73.8 cm. The height for each measured tree
Oak 6 39 14.0 0.204
was selected based on tree height, so that the tree crown was fully

3
P.R. Forsström et al. Remote Sensing of Environment 255 (2021) 112302

Fig. 2. Goniometer measurement set-up. The light source was fixed at +40◦ zenith angle (θ). During the measurements, the sensor view zenith angle (θ = 0◦ , ±21.2,
±48.6◦ , ±76.2◦ ) and sensor view azimuth angle (φ = 0◦ , 15◦ , 45◦ , 75◦ , 90◦ , 105◦ , 135◦ , 165◦ ) were altered.

illuminated, but the tree pot was covered by the canvas (Fig. 2). The were used (i) to normalize the measured spectra to the amount of ra­
canvas (Sunbrella® Solid VV M100) is composed of acryl fiber, and we diation intercepted by the tree crown, and (ii) to calculate a structural
selected it from several tested options, because it was the darkest ma­ parameter that quantifies crown level clumping, i.e., the silhouette to
terial available at a reasonably affordable price. The canvas was large total area ratio (STAR) (Stenberg et al., 2014). We explain the calcula­
enough to cover the horizontal projection of the light beam at the tion of silhouette areas in Section 2.3.1, the processing of tree spectra in
measurement spot. Thus, the light beam formed an ellipse-shaped illu­ Section 2.3.2, and the calculation of STAR in Section 2.3.4. STAR for
minated area on the background. The directional-hemispherical reflec­ each tree is reported in Table 1.
tance factor of the canvas was 0.013–0.02 over the whole spectral range We used a digital camera fixed next to the spectrometer sensor optics
350–2500 nm (Fig. A1). In preliminary tests outdoors, under clear-sky in the goniometer half-arc and pointed it towards the center of the
conditions and using a spectrometer with conical view geometry, we goniometer. The camera had an adjustable zoom lens which was fixed at
noticed some specular behaviour, i.e., signal from the canvas increased 45 mm focal length. The camera shutter was triggered remotely from the
towards the extreme forward scattering angles. However, the amount of ground. We first took photos of a checkerboard pattern at the center of
stray light depended not only on the reflectance of the background, but the goniometer and used Matlab Computer Vision Toolbox™ to solve the
also on how large fraction of the illuminated background was in the intrinsic and extrinsic camera parameters (i.e., focal length, lens dis­
spectrometer’s FOV (see Fig. 2). The stray light signal was measured tortions, and exact position and orientation of the camera at each
from the canvas at all view angles, separately for all four frame heights. viewpoint). During the silhouette area photography, a white canvas was
The relative fraction of stray light (compared to the signal from the tree) placed behind the tree to increase contrast between the tree and the
peaked at regions where the tree’s reflectance was low, being on average image background. The canvas was illuminated from the sides to mini­
61% at 400 nm, 39% at 660 nm, and 52% at 1930 nm. Stray light was mize shadows. This was important since the silhouette areas were
removed during data processing steps (Section 2.3.2.3). extracted from the photos using an automated binary thresholding
A tree was positioned in the goniometer through a slit cut in the black method (see Section 2.3.1). When taking a photo in the direction of
canvas and aligned to the measurement spot at the goniometer center illumination, the lamp was moved downwards so that it did not obstruct
using a projected laser dot directly in nadir. The middle point between the FOV of the camera.
trunk base and crown tip was aligned with the base level of the goni­
ometer (Fig. 2). Before the measurements, the slit in the canvas was 2.2.3. Structural characteristics of trees and optical properties of leaf and
sealed to hide the pot and the supporting structure. bark
Leaf optical properties (i.e., reflectance and transmittance) were
2.2.2. Silhouette area photography measured for each sample tree. We used an ASD RTS-3ZC integrating
After the tree spectral measurements, the sample trees were photo­ sphere with an ASD FieldSpec3 spectrometer (serial number 16007) to
graphed in all 47 view angles of the spectral measurements, and also in record the spectra (350–2500 nm) (Fig. 4). The methods were destruc­
the direction of the illumination. Projected silhouette areas of trees in tive and thus, the leaf level measurements were made after the mea­
the aforementioned view angles were calculated from the photos, and surements of tree spectra and silhouette photography. We applied the

4
P.R. Forsström et al. Remote Sensing of Environment 255 (2021) 112302

same measurement method and protocol for both leaves and needles as (Hovi et al., 2021 (submitted)) by multiplying DSC with π. In addition,
described for a single integrating sphere by Hovi et al. (2020b), except ratio of DSC to that of an ideal diffuse (spherically scattering) object can
that we used slightly thicker needle carriers (0.8 vs 0.3 mm). We aimed be obtained as DSC / (1 / 4π) = (4π × DSC). The processing is explained
for random leaf sampling, and always picked the leaf and needle samples below. For derivation of equations, details of stray light correction, and
from different heights and sides of the tree. We measured three samples discussion on measurement uncertainties, the reader is referred to Hovi
of leaves or needles for each tree. By a needle sample we refer to a et al. (2020a). DSC is wavelength-dependent, but the wavelength
measurement of several pine or spruce needles in the needle carrier, discriminator is omitted in the following formulae for the sake of clarity.
while for oak, one spectral sample means a measurement from a single, DSC for a given Ω was computed as
spatially homogeneous spot on the surface of a leaf. The spectra of
DNtree (Ω) SWR cos40◦ RWR cos0◦ fWR
needles were measured from the needle center for pines and closer to the DSCtree (Ω) = × × × , (1)
DNWR Stree (Ωi ) π ftree (Ω)
needle tips for spruce. This was because spruce needles needed to be
placed in the carrier in two rows, because they were shorter than the
where DNtree(Ω) and DNWR are the measured signals from the tree and
diameter of the sample port of the integrating sphere (15 mm). To ensure
white reference, Stree(Ωi) is the silhouette area of the tree in the direction
comparability of data between the leaf and needle measurements, we
of illumination [m2], SWR is the area of the white reference panel [m2],
used the same needle carriers for both. Measurements of leaf optical
and RWR is the reflectance factor of the white reference panel. Eq. 1 was
properties took approximately 1 h for a tree.
derived from the measurement equations that describe mathematically
One sample tree per species was also measured for bark reflectance
the signals observed when measuring the white reference panel and a
using the same measurement method and protocol as for leaves. Bark
tree, respectively. All DN values are assumed free from stray light. Stray
was peeled from the tree trunk (three samples per tree) using a sharp
light correction is explained in Section 2.3.2.3. In order to take into
knife and placed flat in the needle carrier, with the outer bark surface
account the different amount of radiation intercepted by the tree and the
towards the light. Bark samples were measured only for reflectance of
white reference panel, Eq. 1 computes the ratio of signals from the tree
outer surface since the transmittance of woody parts of a tree can be
and the white reference panel and normalizes the result with the ratio of
expected to be very close to zero.
their silhouette areas in the direction of illumination. The result is
We made direct measurements of tree height using a rigid mea­
brought into a physical scale by multiplying it with the DSC of the white
surement tape, and recorded fresh leaf mass for each tree (Table 1).
reference panel, which for a Lambertian surface is RWRcosθ / π. Finally,
Additionally, we calculated the leaf mass to leaf area conversion factors.
the result is multiplied with the correction factor for the point-spread-
These parameters were used in estimating the total leaf area for each
function (PSF) of the detector (fWR / ftree(Ω)), derivation of which is
tree and for calculating STAR (Table 1).
explained in detail in Section 2.3.2.2. Note that Eq. 1 is for direct illu­
For determining leaf mass, we picked all needles and leaves, and
mination only, i.e., the silhouette areas of the white reference panel and
weighed them using a laboratory grade scale. A smaller set of foliage
the tree (second term on the right-hand side), and thus the interception
from each tree (1 g and 10 g, i.e., approximately 150 needles, for pine
of incoming radiation, are computed using fixed illumination angle
and spruce needles, respectively, and 5 g for oak leaves) was picked to
(here zenith angle of 40◦ ). The equation could be adapted to outdoor
solve the conversion factor of leaf mass to projected area. The leaves or
measurements as well, by taking into account that some fraction of
needles of this subset were weighed, and the projected area was deter­
incoming radiation is diffuse. This requires that multiangular tree sil­
mined by scanning the leaves and needles in a digital film scanner
houettes are available (as in our study), thus enabling to compute the
(Epson Perfection V550) and by applying a binary thresholding to the
diffuse interception of the tree, and that the ratio of diffuse to total ra­
scanned silhouette images. For conifers, an additional small subset (10
diation and angular distribution of diffuse radiation are known.
needles) was taken for determining the projected area to total surface
area conversion factor. In this subset, measurements of projected area
2.3.2.2. Point-spread-functions of the spectrometer. The FOV of the
using the film scanner, and outer dimensions (i.e., length, breadth,
spectrometer had a nominal opening angle of 25◦ . In reality, the sensi­
thickness) of the needles using a digital caliper were made.
tivity of the spectrometer’s detector decreases gradually towards the
edges of the FOV, which means that the signal from a target is dependent
on the location of the target inside the FOV. For example, for a large tree
2.3. Data processing
(e.g., 70 cm in height) viewed at a zenith angle of 76.2◦ , the signal
originating from the treetop was already within the low sensitivity area.
2.3.1. Silhouette areas of trees
Thus, the signal was lower than what would be observed by a detector
The photos of sample trees (Section 2.2.2) were thresholded by
that has equal sensitivity across entire FOV (i.e., an isotropic detector).
applying an automatic method (Otsu, 1979) to the blue channel to yield
The correction factor fWR / ftree(Ω) was introduced to account for this.
binary black-and-white images. Before thresholding, manual selection
The terms fWR and ftree(Ω) are factors calculated for the white reference
was made by drawing a polygon around the tree to delineate areas in the
panel and for the tree, respectively. They describe the fraction that the
photos that contained only tree and white background. The silhouette
recorded signal represents, compared to a signal observed from the same
area of a tree was calculated from the obtained images by multiplying
target by an isotropic detector, and were obtained by weighting the
the number of black (tree) pixels with the area of a single pixel projected
silhouette images of the tree and white reference panel, respectively,
at the distance of the tree crown center.
with the PSF of the detector. The PSF was obtained from measurements
of a small Spectralon® panel, taken so that the panel was placed at
2.3.2. Estimates of directional scattering coefficients of trees
different locations inside the FOV, and by fitting an asymmetric 2D
Gaussian function in the measurements. This was done separately for
2.3.2.1. Equation for directional scattering coefficient. The data from the
each of the three detectors of the spectrometer, i.e. VNIR (350–1000
goniometer measurements (Section 2.2.1) and silhouette images (Sec­
nm), SWIR1 (1001–1800 nm), and SWIR2 (1801–2500 nm). Applying
tion 2.2.2) were processed into estimates of directional scattering co­
the correction factors fWR / ftree(Ω) slightly reduced the jumps in tree
efficients (DSC, [sr− 1]). We define DSC as the fraction of intercepted
spectra observed between different detectors of the spectrometer. These
radiation scattered into a unit solid angle around view direction vector
jumps, measured in relative terms [%] (i.e., (DSC(band 2) - DSC(band
Ω. The DSC was selected over the more common bidirectional reflec­
1)) / DSC(band 1) * 100%), were reduced from − 9.5 ± 9 (mean ±
tance factor (BRF), because a tree is not a surface, but rather scatters
standard deviation) to − 7 ± 7.9 at 1001 vs. 1000 nm, and from − 3 ± 3.9
spherically in all directions. If direct comparison with remote sensing
to 0.8 ± 3.5 at 1801 vs. 1800 nm. The remaining jumps indicate that the
data is desired, an estimate of a tree’s BRF can be obtained from the data

5
P.R. Forsström et al. Remote Sensing of Environment 255 (2021) 112302

correction was not perfect. This is most probably because the correction 2.3.3. Leaf and bark spectra
assumes that the tree is equally bright in all parts. In reality, the side of The directional-hemispherical reflectance and transmittance factors
the tree that was closer to the lamp received the largest amount of ra­ (referred as simply reflectance and transmittance) of leaves and needles,
diation and was therefore the brightest. as well as reflectance of bark were computed with commonly used for­
mulas for single integrating sphere (Eq. 37 and Eq. 38 in Hovi et al.,
2.3.2.3. Removal of stray light. Because the stray light fraction was 2020b). Processing of needle spectra required the retrieval of the gap
known from measurements (Section 2.2.1), stray light [DN] could be fractions of the needle sample within the collimated light beam. These
calculated for any view angle based on a measurement of the white were obtained by scanning the needle carrier with needles in it, using a
reference panel. The challenge was that the tree (or white reference digital film scanner, and by applying a threshold to the obtained 8-bit
panel) and its shadow covered partly the illuminated background and grayscale images within the area of the light beam to separate the
thus, obscured some fraction of the stray light. Therefore, for an accurate needles from the background. For needles of pine and spruce, the
stray light removal, we used the formula threshold value (202 for pine, 187 for spruce) was selected so that, when
the resulting gap fraction was applied in data processing (Eq. 38 in Hovi
DN = DNtotal − bDNstray , (2)
et al., 2020b), the mean needle transmittance at 410–420 nm was 0.021
for pine, and 0.039 for spruce. These values were obtained in a separate
where DN is the signal free from stay light, DNtotal is the original DN
measurement campaign in 2019, for the same species but growing in
value measured, DNstray is the stray light signal without the presence of
Finland. In that campaign, the gap fractions of the needle samples were
the tree or white reference panel, and b is the fraction of stray light not
obtained directly through measurements in the integrating sphere, by
obscured by the tree or white reference panel. Calculation of b was done
painting the illuminated side of the needles black, thus ensuring that the
for each of the detectors of the spectrometer separately. For each tree
measured transmittance signal was only due to the transmission through
and view angle, b was calculated by utilizing the silhouette images taken
the gaps between needles (Daughtry et al., 1989). An accurate estimate
from the view angle and the direction of illumination (Section 2.2.2), as
of needle transmittance could then be derived from measurements made
well as modeled PSF of the spectrometer and irradiance distribution of
before painting, because the gap fraction was known. In addition, we
the light beam. The latter was obtained using the red channel of RGB
applied an empirical transmittance correction that adjusted all trans­
photographs of the light beam on the black background. The image data
mittance spectra 5.5% downwards. It was taken from the measurements
were linearized (i.e., gamma correction removed) before using them in
made against a trusted reference method in Hovi et al. (2020b). The
modeling the irradiance distribution. First, the irradiance distribution of
correction ensured that leaf and needle albedos did not exceed unity.
the light beam and PSF of the spectrometer were projected on the image
Leaf and needle albedos were computed as the sum of reflectance and
taken from the view angle (Fig. 3a–b). Second, a hypothetical stray light
transmittance, and bark albedo was assumed equal to bark reflectance.
signal without a tree was computed by multiplying the PSF with the
The processed leaf, needle, and bark albedos are shown in Fig. 4 and
irradiance distribution of the light beam. The result is shown as the red-
used in the interpretation of our results.
yellow area in Fig. 3c. Third, the tree and its shadow were obtained from
the silhouette images and were then projected on the same image
2.3.4. Silhouette to total area ratio (STAR) of trees
(Fig. 3a–c) to compute the fraction of the stray light that was not
For each tree, leaf and needle mass were converted to total leaf (or
obscured by the tree and its shadow. The calculation of b was performed
needle) area, using two linear conversion factors: mass to projected area
similarly for the white reference panel, except for that the automatic
and projected to total surface area. The former was determined from the
image thresholding was not applicable, and silhouette and shadow of the
subset of leaves (needles) that had been scanned and weighed. The latter
panel were manually measured from images of the white reference panel
for coniferous needles was determined from the subset of needles that
at nadir.
had been scanned and measured for needle dimensions. In order to
In wavelength regions where the tree was dark and the ratio of stray
compute the total needle area from the measurements of needle di­
light (bDNstray) to signal from the tree (DN) was high (e.g., on average
mensions, the shape of pine needles was assumed as semi-fusiform (Eq. 7
61% at 400 nm, 39% at 660 nm, and 52% at 1930 nm), the stray light
in Flower-Ellis and Olsson, 1993), and that of spruce needles was
correction using Eq. 2 prevented negative DSC(Ω) values and resulted in
assumed as parallelepiped (Eq. 9 in Sellin, 2000). For broadleaved trees,
an average increase of DSC(Ω) by 59% at 400 nm, 30% at 660 nm, and
the total leaf area was obtained simply by multiplying the projected area
68% at 1930 nm, compared to a simple stray light correction, i.e. b set to
by a factor of two. Finally, STAR was computed for each tree as the ratio
1.
of the spherically averaged tree silhouette area to total leaf (or needle)
area.

Fig. 3. Illustration of the computation of fraction of stray light not obscured by the tree and its shadow in the goniometer measurements for one view angle (view
azimuth angle 165◦ , view zenith angle 21.2◦ , light originates from west-northwest direction). All images show silhouette and shadow of the tree projected onto a
plane perpendicular to the given viewing direction (black areas). Sub-figure (a) illustrates irradiance distribution of the light beam, (b) the point-spread-function
(PSF) of the spectrometer, and (c) the product of irradiance distribution and PSF. For details of stray light correction, the reader is referred to Hovi et al. (2020a).

6
P.R. Forsström et al. Remote Sensing of Environment 255 (2021) 112302

Fig. 4. Needle, leaf, and bark mean reflectance spectra (a), needle and leaf mean transmittance spectra (b), and corresponding albedo spectra (c). Coefficients of
variation (d) are for albedos, and are calculated using the mean spectrum for each individual tree, i.e. they measure the intraspecific variation between trees. Leaf and
needle reflectance and transmittance are means from both sides of the leaves, and bark reflectance is of the external surface. NDVI was 0.76 for pine needles, 0.68 for
spruce needles, and 0.69 for oak leaves, based on reflectance spectra (a) in red (665 ± 5 nm) and NIR (865 ± 5 nm) wavelengths. The spectra were smoothed to
remove spectral noise using a second order Savitzky-Golay filter with 31 nm (350–1680 nm) and 81 nm (1681–2400 nm) frame lengths. The noisiest wavelengths
below 400 nm and above 2400 nm are excluded. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of
this article.)

3. Results needles of the sample trees (Fig. 4).

In general, spruce and oak scattering was notably stronger compared
We present the spectra of the sample trees and analyze inter- and to pine in nadir (Fig. 5a): oak exhibited the highest DSC in blue (~450
intraspecific variations in their spectro-directional characteristics. We nm) and green spectral regions (~560 nm), while spruce DSC was the
show data in full spectral resolution (400–2400 nm) in the conventional highest in the red region (~660 nm). Pine was notably darker than oak
nadir, or near-nadir view angle of most satellite sensors, and visualize and spruce in the red region (34% and 24%, respectively). In a narrow
the multi-angular scattering patterns of each species in six wavelength green spectral region (~550 nm) pine was slightly brighter than spruce
regions commonly used to study vegetation characteristics from multi- (1%), while still being considerably darker than oak (15%). Overall,
spectral satellite data. Furthermore, we examine the spectral charac­ spruce and oak spectra were fairly similar in VIS and NIR (DSCs within
teristics and anisotropy of trees in the principal plane and cross-plane, 5% of each other), while less so in SWIR where oak was notably (39%)
sometimes referred to as the orthogonal plane, and report the normal­ brighter than spruce. Pine, on the other hand, scattered less than spruce
ized difference vegetation index (NDVI) (Rouse et al., 1974) at leaf and and oak in NIR (23% for both), as well as in SWIR (30% and 49%,
tree level. respectively). In the red-edge, moderate interspecific variation was
In nadir, all species exhibited spectral characteristics expected for observed between the broadleaved and coniferous species: oak was
healthy green vegetation (e.g., Gates et al., 1965; Knipling, 1970): brighter than pine and spruce (21% and 16%, respectively) between 710
relatively low scattering in VIS (400–700 nm), high scattering in NIR and 720 nm. Additionally, in the longer wavelengths of the red-edge,
(750–1300 nm), and decreasing scattering in the shortwave infrared where the spectral curve transitions into the NIR plateau, the transi­
region (SWIR) (>1300 nm) (Fig. 5a). These same basic spectral char­ tion was sharper for oak and pine than for spruce.
acteristics were observed in the spectral albedos of individual leaves and In nadir, the spectral absorption effect by water contained in the

7
P.R. Forsström et al. Remote Sensing of Environment 255 (2021) 112302

Fig. 5. Mean directional scattering coefficient (DSC) (a) and corresponding coefficient of variation (CV) (b) per tree species in nadir. CV was calculated using mean
nadir spectrum for each individual tree, i.e., it measures the intraspecific variation between trees. NDVI was 0.78 for pine, 0.76 spruce, and 0.78 for oak, based on
mean DSCs (a) in red (665 ± 5 nm nm) and NIR (865 ± 5 nm) wavelengths. The noisiest wavelengths below 400 nm and above 2400 nm are excluded.

trees was evident: clear absorption peaks were noted in water absorption specular component (DSC increased from nadir), pine scattering
regions in NIR and SWIR for all species (Fig. 5a). To analyze the inter­ changed notably less, and spruce exhibited no such forward enhance­
specific differences in water sensitivity, we compared the estimated ment being increasingly darker as the view zenith angle increased (DSC
mean DSC of each species at a wavelength of high reflectance (865 nm) decreased from nadir) (Fig. 6, Fig. 7 left column). In the cross-plane,
(i.e., low sensitivity to spectral effects of water) to those at the water each tree species showed a clear species-specific behaviour of scat­
absorption peaks (i.e., 970 nm and 1020 nm in NIR, and 1450 nm and tering (Fig. 6, Fig. 7 right column): although spruce scattered strongly
1930 nm in SWIR). In NIR, these ratios were similar for all species (i.e., upwards in most of the measured wavelengths (more than pine, less than
the ratios varied from 1.00 to 1.06). In SWIR, however, there were oak), its scattering decreased as the view angle was altered from nadir,
considerable differences between the species: the ratio at 1450 nm was as in the forward view angles. Pine scattering, on the other hand, always
5.1 for pine, 4.3 for spruce, and 2.7 for oak, and at 1930 nm it was 16.9 increased towards larger view zenith angles but was typically less than
for pine, 12.4 for spruce, and 9.8 for oak, i.e., pine and spruce exhibited that of spruce and oak. The strongest cross-plane scattering was
larger water absorption features than oak. observed for oak: DSC first increased when moving away from nadir and
The intraspecific variations in nadir DSCs, as presented by co­ then decreased at the maximum view zenith angle (θ = +76.2◦ ).
efficients of variation (CV), showed strong spectral dependence and Although the highest DSC was always observed close to the hot spot
were different for each species (Fig. 5b). For the spectral range, where for all species, the level of DSCs in other view angles varied (Fig. 6 and
the noise was low (400–2400 nm), the mean CV was the smallest for Fig. 7). To further compare the differences in reflectance anisotropy
spruce (7.7%) and somewhat larger for pine and oak (12.1% for both). between tree species, we computed backward to forward scattering
Intraspecific variations were the largest in SWIR for spruce and oak, and ratio. It was computed as mean of all backward DSCs divided by mean of
in VIS for pine. Smallest intraspecific variations were in NIR for all all forward DSCs (excluding DSCs in view angles in the exact cross-plane
species. (Φ = 90◦ ) and the principal plane (Φ = 0◦ ). Spruce and pine exhibited
The multi-angular scattering patterns (Fig. 6), and the mean spectra stronger backward scattering in VIS with ratio between 1.6 and 1.7 than
of trees in the principal and cross-plane (Fig. 7, Fig. A2) revealed a oak with ratio 1.1. While in longer wavelengths, the ratio decreased for
strong dependence of scattering on the view angle. Nevertheless, the spruce and pine to 1.3 in NIR, for oak it slightly increased in NIR to 1.2.
basic characteristics of vegetation spectra, recognized also at leaf level
and in nadir spectra (Fig. 4 and Fig. 5) were always present. The multi- 4. Discussion
angular scattering patterns were strongly asymmetrical along the prin­
cipal plane between backward and forward viewing angles (Fig. 6 and 4.1. Relationships between tree structure and spectra
Fig. 7 left column) and symmetrical along the cross-plane between the
left and right sides (Fig. 6, Fig. 7 right column). The small asymmetry Tree spectra (DSC) depended strongly on view angle and wave­
noted in the cross-plane anisotropy patterns for oak (Fig. 6, Fig. 7 right length. The spectro-directional characteristics were different for the
column) can be a coincidence due to the relatively small number of three species, thus implying that pine, spruce, and oak could be
sample trees (6), or it could be due to the measurement set-up (e.g. slight discriminated from each other in the forward viewing angles and in the
asymmetry in the spectrometer’s PSF), rather than any systematic cross-plane.
scattering characteristics of the trees. In the principal plane, all tree While all three species scattered the incident light strongly back­
species were observed to scatter more backwards towards the light wards, pine and oak showed also scattering peaks in the forward and
source (θ = +48.6◦ , Φ = ±15◦ ) than forwards away from the light cross-plane viewing angles. The spectral anisotropy of trees was always
source. Pine and spruce had stronger hot spot effects than oak, i.e. larger the largest in the principal plane and in the visible spectral region. Oak
increase of DSC from nadir towards the hotspot (Fig. 6, Fig. 7 left col­ had the most notable specular reflectance component while spruce was
umn). In forward viewing angles, however, while oak showed a strong the darkest in the forward viewing angles. Next, we will discuss the

8
P.R. Forsström et al. Remote Sensing of Environment 255 (2021) 112302

Fig. 6. Mean multi-angular scattering patterns of pine,

spruce, and oak (columns) in six wavelengths commonly used
in satellite sensors (±5 nm bands) (six top rows) and as NDVI
(bottom row). The reflectance quantity is directional scat­
tering coefficient (DSC) [sr− 1]. The colour scale covers the
entire data range in each wavelength and NDVI. Nadir is at
the center of each sub-figure. Off-nadir view zenith angles
(±76.2◦ , ±48.6◦ , ±21.2) are indicated by the data points
radiating outwards from the center points, while the view
azimuths angles (0◦ , 15◦ , 45◦ , 75◦ , 90◦ , 105◦ , 135◦ , 165◦ ) are
distributed radially around the center. The black asterisk
symbol represents the light source at a view zenith angle
+40◦ .

9
P.R. Forsström et al. Remote Sensing of Environment 255 (2021) 112302

Fig. 7. The angular distribution of mean directional scattering coefficient (DSC) [sr− 1] for pine, spruce and oak in the principal plane (left column) and in the cross-
plane (right column) at six wavelengths (±5 nm bands) (six top rows) and as NDVI (bottom row). Error bars correspond to standard deviations. Scaling of y-axes
differ between sub-figures. The black asterisk symbol represents the light source at view zenith angle +40◦ in the principal plane.

10
P.R. Forsström et al. Remote Sensing of Environment 255 (2021) 112302

reasons for these observed differences between the study species. reflectance component of oak depended on the wavelength so that in
The leaf level albedos (Fig. 4) could be used to explain the inter­ VIS, the effect was larger than in NIR and SWIR. Firstly, this was most
specific differences in nadir tree spectra only partially (Fig. 5). While likely due to the sparse distribution of horizontally oriented leaves in the
pine trees were darker than spruce trees in almost all wavelengths in oak crown. Secondly, it could be explained by the protective wax layer
nadir, pine needles scattered notably more than spruce needles on oak leaves: in VIS, while most of the incident energy penetrates the
throughout the NIR region and in green. Elsewhere in the spectrum, the leaf surface and gets absorbed by leaf pigments in energy conversion,
order of conifer trees followed that observed at needle level. Similarly, some fraction is reflected specularly by the protective wax layer residing
oak leaf and tree level spectra exhibited differences in VIS: oak leaves on the leaf surface (Bousquet et al., 2005). Since the scattering proper­
scattered relatively less in red and more in blue compared to conifer ties of wax itself are independent of wavelength (Bousquet et al., 2005),
needles, but the opposite behaviour was observed at tree level. In VIS, the spectral differences in the forward scattering component of a leaf
oak leaves were more similar to pine needles than to spruce needles. depend mainly on the optical interactions inside the leaf: unlike energy
Previous studies of leaf and needle albedos have reported similar spec­ in the visible wavelengths, a leaf efficiently diffuses NIR radiation in its
tral characteristics for leaves and needles of different tree species (e.g., cell structure, decreasing the contribution of specular scattering from
Hovi et al., 2017a) but, to-date, there is no systematic comparison of tree the surface on the overall spectra (Bousquet et al., 2005). Thus, although
level spectra of different species. Comparison of NDVIs, based on leaf the optical properties of leaves and needles are quite similar (e.g., Hovi
reflectance and tree DSC in nadir, revealed that while NDVI was always et al., 2017a), the specular effect is stronger in broadleaved species
large (from 0.68 to 0.78), it was larger at tree level for all species, and which have horizontally oriented leaves with large surface areas. Due to
more similar between species at tree level than at leaf level, mostly due the relatively large specular component and lower amount of backward
to the pine needles having higher NDVI at leaf level than spruce and oak. scattering, oak had the smallest backward to forward anisotropy of all
In addition to leaf level spectral properties, we examine species- three species.
specific differences in tree structure (Table 1). We will start by looking Spectral anisotropy measurements of single trees, as presented in this
at the two coniferous species. Even though spruce and pine had similar paper, pave the way for a more comprehensive understanding of how
mean values of crown level clumping (STARspruce = 0.147, STARpine = forest reflectance is formed. Similar measurements have not been pre­
0.143), spruce was brighter than pine in almost all wavelengths in nadir. viously made due to technical challenges related to e.g., developing
We speculate that the relatively open structure of pine (i.e., sparse suitable measurement (goniometer) set-ups. Thus, both the results and
branching pattern) resulted in more light entering deeper into the crown the measurement method presented in this paper are novel. For
before interacting with either needles or woody parts, and more of the measuring samples with height, such as trees, the instrument design
intercepted light escaping in forward and side viewing angles, and less should allow enough distance between the sensor optics and the sample
towards nadir. Based on measurements and visual observations of the so that the tree is within the FOV of the sensor in all view angles. This is a
sample trees, pine needles were long and distributed in a fewer number major challenge since facilitating a large goniometer indoors requires a
of shoots, whereas spruce needles were small, and more tightly aggre­ lot of resources. Moving the sensor away from the sample reduces the
gated around a larger number of shoots and branches. We also speculate parallax error, arising from the physical size of the sample, and increases
that spruce exhibits a higher amount of self-shadowing from its struc­ the accuracy of directional measurements by making the sensor aperture
tural parts, effectively blocking photons from escaping the crown. appear smaller to the sample. However, increasing the distance between
Next, we will compare results of broadleaved and coniferous species. the sensor and the sample decreases signal quality as the level of noise
Oak had clearly less clumping than the coniferous species (STARoak = increases. Additionally, if a sample does not fully cover the FOV of the
0.197). This was in line with oak being always brighter than pine, sensor, as was the case with our sample trees, correction for the back­
somewhat brighter than spruce in VIS and NIR, and notably brighter ground signal, i.e., stray light, which varies with the view angle, must be
than spruce in SWIR in nadir. We suggest that the spectral differences executed with care. Furthermore, defining the correct reflectance
between the broadleaved oak and the two coniferous species rise from quantity for vertical samples may not be as straightforward as for
the following factors: (i) the flat and more frequently horizontally ori­ samples with less height. Thus, even though goniometers offer currently
ented (planophile) oak leaves (Farque et al., 2001; Chianucci et al., the only possibility for obtaining spectro-directional data of vegetation
2018) scatter upwards more effectively than needles arranged on shoots, in a controlled environment, goniometer designs are always also a
(ii) the structure (as depicted by crown level STAR) of oak trees induces compromise between feasibility and accuracy.
a smaller amount of multiple scattering compared to the conifers, and
(iii) there was less water in oak leaves compared to conifer needles. It 4.3. Links to remote sensing of forests
should also be noted that the small amount of multiple scattering within
oak crowns resulted in more similar leaf and tree level spectra than in Overall, examination of leaf and tree spectra revealed some sur­
spruce and pine. prising inter- and intraspecific similarities between the two levels and
The interspecific variation in the transition from the red-edge spec­ can be compared with multi-angular characteristics of forest reflectance
tral region to the NIR could be related to species-specific contribution of in previous studies. However, due to the diversely different structural
woody parts on the overall spectra. The spectral measurements of bark and compositional properties, comparison of data between the two
reflectance (Fig. 4) were in line with our visual observations: while oak scales is not straightforward. While some similarities and differences can
had a notable absorption peak in red and brightening in green wave­ be identified, it should be noted that our results are for small trees, and
lengths, spruce and pine bark reflectance increased more linearly to­ further research would be needed on the characteristics of full-sized
wards longer wavelengths, due to a smaller influence of bark trees for a more comprehensive comparison. Experimentally, isolating
chlorophyll. Although all measured bark spectra were similar in the large trees comparable to our small size tree laboratory experiment will
transition between the red-edge and NIR regions (725–800 nm), spruce be one of the key challenges to transfer or even scale information to
bark reflectance was still the smallest of the three species. This might at larger areas.
least partly explain the differences observed in the transition region. The similarities in spruce and oak tree spectra observed in this study
(Figs. 5, 6, 7) are surprising since forest reflectance in satellite and
4.2. Measurements of spectral anisotropy of trees airborne images has been shown in general to be higher in NIR for de­
ciduous forests (in summer) compared to coniferous forests (e.g., Kimes
Our broadleaved study species, oak, displayed large anisotropy of et al., 1986; Ranson et al., 1994; Bréon et al., 1997; Eklundh et al., 2003;
reflectance in the principal plane with profound scattering peaks, not Canisius and Chen, 2007; Rautiainen et al., 2008; Heiskanen et al., 2013;
only backwards, but also forwards. The strength of the specular Rautiainen and Lukeš, 2015; Hadi et al., 2016). This can be explained by

11
P.R. Forsström et al. Remote Sensing of Environment 255 (2021) 112302

factors related to the absence of the natural scattering environment of and are available through Mendeley open access repository.
the trees, i.e., the surrounding forest: (i) the overall canopy structure (e.
g., spatial arrangement of trees) and thus multiple scattering as well as
Declaration of Competing Interest
mutual shadowing between trees, influence forest reflectance (e.g., Li
and Strahler, 1992), and make it substantially different from that of an
The authors declare that they have no known competing financial
individual tree crown, and (ii) in addition to structural and spectral
interests or personal relationships that could have appeared to influence
properties of trees, the reflectance of a forest also depends strongly on
the work reported in this paper.
the visibility of the understory vegetation and soil to the sensor (e.g.,
Rautiainen and Lukeš, 2015) which is typically largest in view angles
Acknowledgements
close to nadir (e.g., Korhonen et al., 2011; Hovi et al., 2017b; Kuusk
et al., 2018).
This study has received funding from the European Research Council
Our results can also be compared to multi-angular data from decid­
(ERC) under the European Union’s Horizon 2020 Research and Inno­
uous and coniferous forests at crown- (e.g., Korpela et al., 2011; Korpela
vation Programme (grant agreement No 771049). The text reflects only
et al., 2014), stand- (e.g., Kimes et al., 1986; Ranson et al., 1994; Bréon
the authors’ view and the Agency is not responsible for any use that may
et al., 1997; Sandmeier and Deering, 1999; Canisius and Chen, 2007;
be made of the information it contains. This study was also partly funded
Rautiainen et al., 2008) or landscape-level (e.g., Bicheron and Leroy,
by the Academy of Finland project BOREALITY (grant nr 286390). The
2000). In general, our study corroborated findings of previous airborne-
contribution of M.E.S. is supported by the University of Zurich.
and satellite measurements: there is a strong hot spot effect in the
principal plane for single trees and entire forests (e.g., Bréon et al., 1997;
Appendix A. Supplementary data
Sandmeier and Deering, 1999; Rautiainen et al., 2008). Similarly, in the
forward view angles, deciduous canopies (e.g., Kimes et al., 1986; Bréon
Supplementary data to this article can be found online at https://doi.
et al., 1997; Canisius and Chen, 2007) and single oak trees (Fig. 6, Fig. 7,
org/10.1016/j.rse.2021.112302.
left column) have now been shown to exhibit a clear scattering peak
(most prominent in VIS), and coniferous canopies (e.g., Sandmeier and
Deering, 1999; Canisius and Chen, 2007) along with pine and spruce References
trees (Fig. 6, Fig. 7 left column) have been shown to be dark at larger
Asner, G.P., Braswell, B.H., Schimel, D.S., Wessman, C.A., 1998. Ecological research
view zenith angles. Furthermore, the angular effects noted in the prin­ needs from multiangle remote sensing data. Remote Sens. Environ. 63 (2), 155–165.
cipal plane seem to decrease as a function of increasing wavelength for https://doi.org/10.1016/S0034-4257(97)00139-9.
both forests (e.g., Sandmeier and Deering, 1999; Rautiainen et al., 2008) Atkinson, K., 1982. Numerical integration on the sphere. Journal of the Australian
Mathematical Society (Series B) 23, 332–347.
as well as single tree crowns (Fig. 6, Fig. 7). Our study also revealed an Barnsley, M.J., Lewis, P., O’Dwyer, S., Disney, M.I., Hobson, P., Cutter, M., Lobb, D.,
interesting new finding: while NIR reflectance of deciduous stands have 2000. On the potential of CHRIS/PROBA for estimating vegetation canopy properties
previously been reported to be distinguishably stronger than that of from space. Remote Sens. Rev. 19, 171–189. https://doi.org/10.1080/
02757250009532417.
coniferous stands in nadir and backwards along the principal plane (e.g., Barnsley, M.J., Settle, J.J., Cutter, M.A., Lobb, D.R., Teston, F., 2004. The PROBA/CHRIS
Canisius and Chen, 2007; Rautiainen et al., 2008), in our study, how­ mission: a low-cost smallsat for hyperspectral multiangle observations of the earth
ever, single oak, pine, and spruce trees were equally bright in these view surface and atmosphere. IEEE Trans. Geosci. Remote Sens. 42 (7), 1512–1520.
https://doi.org/10.1109/TGRS.2004.827260.
angles in all wavelength regions except SWIR (Figs. 5, 6, 7). Bicheron, P., Leroy, M., 2000. Bidirectional reflectance distribution function signatures
While, in general, strong anisotropy patterns of spectral reflectance of major biomes observed from space. J. Geophys. Res. 105 (D21) https://doi.org/
have already earlier been measured for vegetation canopies using both 10.1029/2000JD900380.
Bousquet, L., Lachérade, S., Jacquemoud, S., Moya, I., 2005. Leaf BRDF measurements
ground-based methods (e.g., Kimes, 1983; Sandmeier and Itten, 1999;
and model for specular and diffuse components differentiation. Remote Sens.
Sandmeier et al., 1999; Peltoniemi et al., 2005) and air- and spaceborne Environ. 98 (2–3), 201–211. https://doi.org/10.1016/j.rse.2005.07.005.
sensors (e.g., Ranson et al., 1994; Sandmeier et al., 1999; Rautiainen Bréon, F.-M., Vanderbilt, V., Leroy, M., Bicheron, P., Walthall, C.L., Kalshoven, J.E.,
1997. Evidence of hot spot directional signature from airborne POLDER
et al., 2008), in practice, the use of multi-angular reflectance data in
measurements. IEEE Trans. Geosci. Remote Sens. 35 (2), 479–484. https://doi.org/
monitoring vegetation canopies is not yet common. Empirical evidence, 10.1109/36.563289.
such as that of spectro-directional characteristics of single trees, Canisius, F., Chen, J.M., 2007. Retrieving forest background reflectance in a boreal
collected at ground level both outdoors and in laboratory, is essential in region from Multi-angle Imaging SpectroRadiometer (MISR) data. Remote Sens.
Environ. 107 (1–2), 312–321. https://doi.org/10.1016/j.rse.2006.07.023.
understanding the scattering hierarchy within coniferous, broadleaved, Chen, J.M., Leblanc, S.G., 1997. A four-scale bidirectional reflectance model based on
and mixed forest canopies, and may improve the reliability of analysis of canopy architecture. IEEE Trans. Geosci. Remote Sens. 35 (5), 1316–1337. https://
remote sensing data. Such data are also fundamental in developing and doi.org/10.1109/36.628798.
Chen, J.M., Mo, G., Pisek, J., Liu, J., Deng, F., Ishizawa, M., Chan, D., 2012. Effects of
testing physically-based reflectance models. foliage clumping on the estimation of global terrestrial gross primary productivity.
Glob. Biogeochem. Cycles 26, GB1019. https://doi.org/10.1029/2010GB003996.
5. Conclusions Chianucci, F., Pisek, J., Raabe, K., Marchino, L., Ferrara, C., Corona, P., 2018. A dataset
of leaf inclination angles for temperate and boreal broadleaf woody species. Ann.
Forest Sci. 75, 50 https://doi.org/10.1007/s13595-018-0730-x.
Although empirical data on the spectra of different forest compo­ Dangel, S., Verstraete, M.M., Schopfer, J., Kneubühler, M., Schaepman, M., Itten, K.I.,
nents, such as tree leaves and needles, and of some understory species 2005. Toward a direct comparison of field and laboratory goniometer
measurements. IEEE Trans. Geosci. Remote Sens. 43 (11), 2666–2675. https://doi.
are nowadays available, the spectro-directional characteristics of indi­ org/10.1109/TGRS.2005.857324.
vidual trees had not been measured before this study. Our results of Daughtry, C.S.T., Biehl, L.L., Ranson, K.J., 1989. A new technique to measure the
small trees indicate that simultaneous measurements of both, spectral spectral properties of conifer needles. Remote Sens. Environ. 27 (1), 81–91. https://
doi.org/10.1016/0034-4257(89)90039-4.
and directional characteristics of may enhance the discrimination of tree
Deering, D.W., Eck, T.F., Banerjee, B., 1999. Characterization of the reflectance
species, and thus may aid in retrieval of information on their biophysical anisotropy of three boreal forest canopies in spring–summer. Remote Sens. Environ.
properties. We also described a tree crown level measurement set-up for 67 (2), 205–229. https://doi.org/10.1016/S0034-4257(98)00087-X.
a goniometer that can, in the future, be applied to collect multi-angular Deschamps, P.-Y., Bréon, F.-M., Leroy, M., Podaire, A., Bricaud, A., Buriez, J.-C., Sèze, G.,
1994. The POLDER mission: instrument characteristics and scientific objectives.
spectra of variety of individual plant species, not only trees. IEEE Trans. Geosci. Remote Sens. 32 (3), 598–599. https://doi.org/10.1109/
36.297978.
Data availability Diner, D.J., Beckert, J.C., Reilly, T.H., Bruegge, C.J., Conel, J.E., Kahn, R.A.,
Martonchik, J.V., Ackerman, T.P., Davies, R., Gerstl, S.A.W., Gordon, H.R.,
Muller, J.-P., Myneni, R.B., Sellers, P.J., Pinty, B., Verstraete, M.M., 1998. Multi-
The measurement data are presented in Hovi et al., 2021 (submitted) angle Imaging SpectroRadiometer (MISR) instrument description and experiment

12
P.R. Forsström et al. Remote Sensing of Environment 255 (2021) 112302

overview. IEEE Trans. Geosci. Remote Sens. 36 (4), 1072–1087. https://doi.org/ Kuusk, A., Pisek, J., Lang, M., Märdla, S., 2018. Estimation of gap fraction and foliage
10.1109/36.700992. clumping in forest canopies. Remote Sens. 10 (7), 1153. https://doi.org/10.3390/
Eklundh, L., Hall, K., Eriksson, H., Ardö, J., Pilesjö, P., 2003. Investigating the use of rs10071153.
Landsat thematic mapper data for estimation of forest leaf area index in southern Lacaze, R., Roujean, J.-L., 2001. G-function and HOt SpoT (GHOST) reflectance model:
Sweden. Can. J. Remote. Sens. 29 (3), 349–362. https://doi.org/10.5589/m03-004. application to multi-scale airborne POLDER measurements. Remote Sens. Environ.
Farque, L., Sinoquet, H., Colin, F., 2001. Canopy structure and light interception in 76 (1), 67–80. https://doi.org/10.1016/S0034-4257(00)00193-0.
Quercus petraea seedlings in relation to light regime and plant density. Tree Physiol. Li, X., Strahler, A.H., 1992. Geometric-optical bidirectional reflectance modeling of the
21 (17), 1257–1267. https://doi.org/10.1093/treephys/21.17.1257. discrete crown vegetation canopy: effect of crown shape and mutual shadowing.
Flower-Ellis, J.G.K., Olsson, L., 1993. Estimation of volume, total and projected area of IEEE Trans. Geosci. Remote Sens. 30 (2), 276–292. https://doi.org/10.1109/
Scots pine needles from their regression on length. In: Studia Forestalia Suecica, 190. 36.134078.
Forsström, P., Peltoniemi, J., Rautiainen, M., 2019. Seasonal dynamics of lingonberry Lobell, D.B., Asner, G.P., Law, B.E., Treuhaft, R.N., 2002. View angle effects on canopy
and blueberry. Silva Fennica 53 (2), 10150. https://doi.org/10.14214/sf.10150. reflectance and spectral mixture analysis of coniferous forests using AVIRIS. Int. J.
Gates, D.M., Keegan, H.J., Schleter, J.C., Weidner, V.R., 1965. Spectral properties of Remote Sens. 23 (11), 2247–2262. https://doi.org/10.1080/01431160110075613.
plants. Appl. Opt. 4 (1), 11–20. https://doi.org/10.1364/AO.4.000011. Lunagaria, M.M., Patel, H.R., 2017. Changes in reflectance anisotropy of wheat crop
Gerard, F.F., North, P.R.J., 1997. Analyzing the effect of structural variability and canopy during different phenophases. Int. Agrophys. 31 (2), 203–218. https://doi.org/
gaps on forest BRDF using a geometric-optical model. Remote Sens. Environ. 62 (1), 10.1515/intag-2016-0045.
46–62. https://doi.org/10.1016/S0034-4257(97)00070-9. Markiet, V., Hernández-Clemente, R., Mõttus, M., 2017. Spectral similarity and PRI
Hadi, Korhonen, L., Hovi, A., Rönnholm, P., Rautiainen, M., 2016. The accuracy of large- variations for a boreal forest stand using multi-angular airborne imagery. Remote
area forest canopy cover estimation using Landsat in boreal region. Int. J. Appl. Sens. 9 (10), 1005. https://doi.org/10.3390/rs9101005.
Earth Obs. Geoinf. 53, 118–127. https://doi.org/10.1016/j.jag.2016.08.009. Mõttus, M., Rautiainen, M., Schaepman, M.E., 2012. Shoot scattering phase function for
Heiskanen, J., Rautiainen, M., Stenberg, P., Mõttus, M., Vesanto, V.-H., 2013. Sensitivity Scots pine and its effect on canopy reflectance. Agric. For. Meteorol. 154–155,
of narrowband vegetation indices to boreal forest LAI, reflectance seasonality and 67–74. https://doi.org/10.1016/j.agrformet.2011.10.012.
species composition. ISPRS J. Photogramm. Remote Sens. 78, 1–14. https://doi.org/ Otsu, N., 1979. A threshold selection method from gray-level histograms. IEEE Trans.
10.1016/j.isprsjprs.2013.01.001. Syst. Man Cybernet. 9 (1), 62–66. https://doi.org/10.1109/TSMC.1979.4310076.
Hovi, A., Raitio, P., Rautiainen, M., 2017a. A spectral analysis of 25 boreal tree species. Peltoniemi, J.I., Kaasalainen, S., Näränen, J., Rautiainen, M., Stenberg, P.,
Silva Fennica 51 (4), 7753. https://doi.org/10.14214/sf.7753. Smolander, H., Smolander, S., Voipio, P., 2005. BRDF measurement of understory
Hovi, A., Lukeš, P., Rautiainen, M., 2017b. Seasonality of albedo and FAPAR in a boreal vegetation in pine forests: dwarf shrubs, lichen, and moss. Remote Sens. Environ. 94
forest. Agric. For. Meteorol. 247, 331–342. https://doi.org/10.1016/j. (3), 343–354. https://doi.org/10.1016/j.rse.2004.10.009.
agrformet.2017.08.021. Pisek, J., Chen, J.M., Kobayashi, H., Rautiainen, M., Schaepman, M.E., Karnieli, A.,
Hovi, A., Forsström, P., Ghielmetti, G., Schaepman, M.E., Rautiainen, M., 2020a. Sprinstin, M., Ryu, Y., Nikopensius, M., Raabe, K., 2016. Retrieval of seasonal
Empirical validation of photon recollision probability in single crowns of tree dynamics of forest understory reflectance from semiarid to boreal forests using
seedlings. ISPRS J. Photogramm. Remote Sens. 169, 57–72. https://doi.org/ MODIS BRDF data. J. Geophys. Res. 121, 855–863. https://doi.org/10.1002/
10.1016/j.isprsjprs.2020.08.027. 2016JG003322.
Hovi, A., Mõttus, M., Juola, J., Manoocheri, F., Ikonen, E., Rautiainen, M., 2020b. Ranson, K.J., Irons, J.R., Williams, D.L., 1994. Multispectral bidirectional reflectance of
Evaluating the performance of a double integrating sphere in measurement of northern forest canopies with the Advanced Solid-State Array Spectroradiometer
reflectance, transmittance, and albedo of coniferous needles. Silva Fennica 54 (2), (ASAS). Remote Sens. Environ. 47, 276–289. https://doi.org/10.1016/0034-4257
10270. https://doi.org/10.14214/sf.10270. (94)90161-9.
Hovi, A., Forsström, P.R., Ghielmetti, G., Schaepman, M.E., Rautiainen, M., 2021. Rautiainen, M., Lukeš, P., 2015. Spectral contribution of understory to forest reflectance
A Dataset Composed of Multiangular Spectral Libraries and Auxiliary Data at Tree, in a boreal site: an analysis of EO-1 Hyperion data. Remote Sens. Environ. 171,
Leaf, Needle, and Bark Level for Three Common European Tree Species (Submitted 98–104. https://doi.org/10.1016/j.rse.2015.10.009.
to Data in Brief). Rautiainen, M., Stenberg, P., Nilson, T., Kuusk, A., 2004. The effect of crown shape on
Jacquemoud, S., 1993. Inversion of the PROSPECT + SAIL canopy reflectance model the reflectance of coniferous stands. Remote Sens. Environ. 89 (1), 41–52. https://
from AVIRIS equivalent spectra: theoretical study. Remote Sens. Environ. 44 (2–3), doi.org/10.1016/j.rse.2003.10.001.
281–292. https://doi.org/10.1016/0034-4257(93)90022-P. Rautiainen, M., Lang, M., Mõttus, M., Kuusk, A., Nilson, T., Kuusk, J., Lükk, T., 2008.
Juola, J., Hovi, A., Rautiainen, M., 2020. Multiangular spectra of tree bark for common Multi-angular reflectance properties of a hemiboreal forest: an analysis using CHRIS
boreal tree species in Europe. Silva Fennica 54 (4), 10331. https://doi.org/ PROBA data. Remote Sens. Environ. 112 (5), 2627–2642. https://doi.org/10.1016/j.
10.14214/sf.10331. rse.2007.12.005.
Justice, C.O., Townshend, J.R.G., Vermote, E.F., Masuoka, E., Wolfe, R.E., Saleous, N., Rautiainen, M., Mõttus, M., Heiskanen, J., Akujärvi, A., Majasalmi, T., Stenberg, P.,
Roy, D.P., Morisette, J.T., 2002. An overview of MODIS land data processing and 2011. Seasonal reflectance dynamics of common understory types in a northern
product status. Remote Sens. Environ. 83, 3–15. https://doi.org/10.1016/S0034- European boreal forest. Remote Sens. Environ. 115 (12), 3020–3028. https://doi.
4257(02)00084-6. org/10.1016/j.rse.2011.06.005.
Kimes, D.S., 1983. Dynamics of directional reflectance factor distributions for vegetation Rautiainen, M., Mõttus, M., Yáñez-Rausell, L., Homolová, L., Malenovský, Z.,
canopies. Appl. Opt. 22 (9), 1364–1372. https://doi.org/10.1364/AO.22.001364. Schaepman, M.E., 2012. A note on upscaling coniferous needle spectra to shoot
Kimes, D.S., Newcomb, W.W., Nelson, R.F., Schutt, J.B., 1986. Directional reflectance spectral albedo. Remote Sens. Environ. 117, 469–474. https://doi.org/10.1016/j.
distributions of a hardwood and pine forest canopy. IEEE Trans. Geosci. Remote rse.2011.10.019.
Sens. GE-24 (2), 281–293. https://doi.org/10.1109/TGRS.1986.289647. Roosjen, P.P.J., Clevers, J.G.P.W., Bartholomeus, H.M., Schaepman, M.E., Schaepman-
Knipling, E.B., 1970. Physical and physiological basis for the reflectance of visible and Strub, G., Jalink, H., Van der Schoor, R., De Jong, A., 2012. A laboratory goniometer
near-infrared radiation from vegetation. Remote Sens. Environ. 1 (3), 155–159. system for measuring reflectance and emittance anisotropy. Sensors 12,
https://doi.org/10.1016/S0034-4257(70)80021-9. 17358–17371. https://doi.org/10.3390/s121217358.
Knyazikhin, Y., Martonchik, J.V., Diner, D.J., Myneni, R.B., Verstraete, M., Pinty, B., Roosjen, P.P.J., Suomalainen, J.M., Bartholomeus, H.M., Kooistra, L., Clevers, J.G.P.W.,
Gobron, N., 1998. Estimation of vegetation canopy leaf area index and fraction of 2017. Mapping reflectance anisotropy of a potato canopy using aerial images
absorbed photosynthetically active radiation from atmosphere-corrected MISR data. acquired with an unmanned aerial vehicle. Remote Sens. 9 (5), 417. https://doi.org/
J. Geophys. Res. 103 (D24), 32239–32256. https://doi.org/10.1029/98JD02461. 10.3390/rs9050417.
Kobayashi, H., Iwabuchi, H., 2008. A coupled 1-D atmosphere and 3-D canopy radiative Roosjen, P.P.J., Brede, B., Suomalainen, J.M., Bartholomeus, H.M., Kooistra, L.,
transfer model for canopy reflectance, light environment, and photosynthesis Clevers, J.G.P.W., 2018. Improved estimation of leaf area index and leaf chlorophyll
simulation in a heterogeneous landscape. Remote Sens. Environ. 112 (1), 173–185. content of a potato crop using multi-angle spectral data – potential of unmanned
https://doi.org/10.1016/j.rse.2007.04.010. aerial vehicle imagery. Int. J. Appl. Earth Obs. Geoinf. 66, 14–26. https://doi.org/
Korhonen, L., Korpela, I., Heiskanen, J., Maltamo, M., 2011. Airborne discrete-return 10.1016/j.jag.2017.10.012.
LIDAR data in the estimation of vertical canopy cover, angular canopy closure and Roujean, J.-L., Leroy, M., Deschamps, P.-Y., 1992. A bidirectional reflectance model of
leaf area index. Remote Sens. Environ. 115 (4), 1065–1080. https://doi.org/ the Earth’s surface for the correction of remote sensing data. J. Geophys. Res. 97
10.1016/j.rse.2010.12.011. (D18), 20455–20468. https://doi.org/10.1029/92JD01411.
Korpela, I., Heikkinen, V., Honkavaara, E., Rohrbach, F., Tokola, T., 2011. Variation and Rouse Jr., J.W., Haas, R.H., Schell, J.A., Deering, D.W., 1974. Monitoring vegetation
directional anisotropy of reflectance at the crown scale – implications for tree species systems in the Great Plains with ERTS. NASA, Goddard Space Flight Center Third
classification in digital aerial images. Remote Sens. Environ. 115, 2062–2074. ERTS-1 Symposium 1 (A), 309–317.
https://doi.org/10.1016/j.rse.2011.04.008. Sandmeier, St., Deering, D.W., 1999. Structure analysis and classification of boreal
Korpela, I., Mehtätalo, L., Markelin, L., Seppänen, A., Kangas, A., 2014. Tree species forests using airborne hyperspectral BRDF data from ASAS. Remote Sens. Environ.
identification in aerial image data using directional reflectance signatures. Silva 69 (3), 281–295. https://doi.org/10.1016/S0034-4257(99)00032-2.
Fennica 48 (3), 1087. https://doi.org/10.14214/sf.1087. Sandmeier, S.R., Itten, K.I., 1999. A field goniometer system (FIGOS) for acquisition of
Kuusinen, N., Juola, J., Karki, B., Stenroos, S., Rautiainen, M., 2020. A spectral analysis hyperspectral BRDF data. IEEE Trans. Geosci. Remote Sens. 37 (2), 978–986.
of common boreal ground lichen species. Remote Sens. Environ. 247, 111955. https://doi.org/10.1109/36.752216.
https://doi.org/10.1016/j.rse.2020.111955. Sandmeier, S.R., Middleton, E.M., Deering, D.W., Qin, W., 1999. The potential of
Kuusk, A., Kuusk, J., Lang, M., 2014. Modeling directional forest reflectance with the hyperspectral bidirectional reflectance distribution function data for grass canopy
hybrid type forest reflectance model FRT. Remote Sens. Environ. 149, 196–204. characterization. J. Geophys. Res. 104 (D8), 9547–9560. https://doi.org/10.1029/
https://doi.org/10.1016/j.rse.2014.03.035. 1999JD900094.

13
P.R. Forsström et al. Remote Sensing of Environment 255 (2021) 112302

Schaepman-Strub, G., Schaepman, M.E., Painter, T.H., Dangel, S., Martonchik, J.V., Sellin, A., 2000. Estimating the needle area from geometric measurements: application of
2006. Reflectance quantities in optical remote sensing – definitions and case studies. different calculation methods to Norway spruce. Trees 14, 215–222. https://doi.org/
Remote Sens. Environ. 103 (1), 27–42. https://doi.org/10.1016/j.rse.2006.03.002. 10.1007/PL00009765.
Schneider, F.D., Leiterer, R., Morsdorf, F., Gastellu-Etchegorry, J.P., Lauret, N., Stenberg, P., Mõttus, M., Rautiainen, M., Sievänen, R., 2014. Quantitative
Pfeifer, N., Schaepman, M.E., 2014. Simulating imaging spectrometer data: 3D forest characterization of clumping in Scots pine crowns. Ann. Bot. 114 (4), 689–694.
modeling based on LiDAR and in situ data. Remote Sens. Environ. 152, 235–250. https://doi.org/10.1093/aob/mct310.
https://doi.org/10.1016/j.rse.2014.06.015. Verrelst, J., Schaepman, M.E., Koetz, B., Kneubühler, M., 2008. Angular sensitivity
Schopfer, J., Dangel, S., Kneubühler, M., Itten, K.I., 2008. The improved dual-view field analysis of vegetation indices derived from CHRIS/PROBA data. Remote Sens.
goniometer system FIGOS. Sensors 8 (8), 5120–5140. https://doi.org/10.3390/ Environ. 112, 2341–2353. https://doi.org/10.1016/j.rse.2007.11.001.
s8085120. Verrelst, J., Schaepman, M.E., Clevers, J.G.P.W., 2010. Spectrodirectional Minnaert-k
retrieval using CHRIS-PROBA data. Can. J. Remote. Sens. 36 (6), 631–644. https://
doi.org/10.5589/m11-012.

14