ArticlePDF Available

Abstract and Figures

Land Surface Phenology (LSP) and Leaf Area Index (LAI) are important variables that describe the photosynthetically active phase and capacity of vegetation. Both are derived on the global scale from optical satellite sensors and require robust validation based on in situ sensors at high temporal resolution. This study assesses the PAI Autonomous System from Transmittance Sensors at 57 • (PASTiS-57) instrument as a low-cost transmittance sensor for simultaneous monitoring of LSP and LAI in forest ecosystems. In a field experiment, spring leaf flush and autumn senescence in a Dutch beech forest were observed with PASTiS-57 and illumination independent, multi-temporal Terrestrial Laser Scanning (TLS) measurements in five plots. Both time series agreed to less than a day in Start Of Season (SOS) and End Of Season (EOS). LAI magnitude was strongly correlated with a Pearson correlation coefficient of 0.98. PASTiS-57 summer and winter LAI were on average 0.41 m 2 m −2 and 1.43 m 2 m −2 lower than TLS. This can be explained by previously reported overestimation of TLS. Additionally, PASTiS-57 was implemented in the Discrete Anisotropic Radiative Transfer (DART) Radiative Transfer Model (RTM) model for sensitivity analysis. This confirmed the robustness of the retrieval with respect to non-structural canopy properties and illumination conditions. Generally, PASTiS-57 fulfilled the CEOS LPV requirement of 20% accuracy in LAI for a wide range of biochemical and illumination conditions for turbid medium canopies. However, canopy non-randomness in discrete tree models led to strong biases. Overall, PASTiS-57 demonstrated the potential of autonomous devices for monitoring of phenology and LAI at daily temporal resolution as required for validation of satellite products that can be derived from ESA Copernicus' optical missions, Sentinel-2 and-3.
Content may be subject to copyright.
remote sensing
Article
Monitoring Forest Phenology and Leaf Area Index
with the Autonomous, Low-Cost Transmittance
Sensor PASTiS-57
Benjamin Brede 1,*, Jean-Philippe Gastellu-Etchegorry 2, Nicolas Lauret 2, Frederic Baret 3,
Jan G. P. W. Clevers 1, Jan Verbesselt 1and Martin Herold 1
1Laboratory of Geo-Information Science and Remote Sensing, Wageningen University & Research,
Droevendaalsesteeg 3, 6708 PB Wageningen, The Netherlands; jan.clevers@wur.nl (J.G.P.W.C.);
jan.verbesselt@wur.nl (J.V.); martin.herold@wur.nl (M.H.)
2Centre d’Etudes Spatiales de la BIOsphere, Toulouse University, CNES, CNRS, IRD, UPS, (Toulouse),
31401 Toulouse, France; gastellu@cesbio.cnes.fr (J.-P.G.-E.); nicolas.lauret@cesbio.cnes.fr (N.L.)
3Institut National de la Recherche Agronomique–Université d’Avignon et des Pays du Vaucluse
(INRA-UAPV), 228 Route de l’Aérodrome, 84914 Avignon, France; baret@avignon.inra.fr
*Correspondence: benjamin.brede@wur.nl
Received: 15 May 2018; Accepted: 27 June 2018; Published: 30 June 2018


Abstract:
Land Surface Phenology (
LSP
) and Leaf Area Index (
LAI
) are important variables that
describe the photosynthetically active phase and capacity of vegetation. Both are derived on the
global scale from optical satellite sensors and require robust validation based on in situ sensors at high
temporal resolution. This study assesses the PAI Autonomous System from Transmittance Sensors at
57
(
PASTiS-57
) instrument as a low-cost transmittance sensor for simultaneous monitoring of
LSP
and
LAI
in forest ecosystems. In a field experiment, spring leaf flush and autumn senescence in a
Dutch beech forest were observed with
PASTiS-57
and illumination independent, multi-temporal
Terrestrial Laser Scanning (
TLS
) measurements in five plots. Both time series agreed to less than
a day in Start Of Season (
SOS
) and End Of Season (
EOS
).
LAI
magnitude was strongly correlated
with a Pearson correlation coefficient of 0.98.
PASTiS-57
summer and winter
LAI
were on average
0.41 m2m2
and
1.43 m2m2
lower than
TLS
. This can be explained by previously reported
overestimation of
TLS
. Additionally,
PASTiS-57
was implemented in the Discrete Anisotropic
Radiative Transfer (
DART
) Radiative Transfer Model (
RTM
) model for sensitivity analysis. This
confirmed the robustness of the retrieval with respect to non-structural canopy properties and
illumination conditions. Generally,
PASTiS-57
fulfilled the CEOS LPV requirement of 20% accuracy
in
LAI
for a wide range of biochemical and illumination conditions for turbid medium canopies.
However, canopy non-randomness in discrete tree models led to strong biases. Overall,
PASTiS-57
demonstrated the potential of autonomous devices for monitoring of phenology and
LAI
at daily
temporal resolution as required for validation of satellite products that can be derived from ESA
Copernicus’ optical missions, Sentinel-2 and -3.
Keywords:
Land Surface Phenology; Leaf Area Index; ground-based; forest; validation;
radiative transfer model; DART model
1. Introduction
Vegetation phenology describes the “timing of seasonal developmental stages in plant life cycles
including bud burst, canopy growth, flowering, and senescence [...]” [
1
]. This includes the start, end and
length of the photosynthetically active phase in the year. A mechanistic understanding of controls on these
events is still lacking for most biomes [
2
]. This leads to a general misrepresentation of vegetation temporal
Remote Sens. 2018,10, 1032; doi:10.3390/rs10071032 www.mdpi.com/journal/remotesensing
Remote Sens. 2018,10, 1032 2 of 19
behaviour in global circulation models and uncertainty in vegetation-climate feedbacks. Standardised,
wide-spread observations are paramount for the quantification of phenology and climate feedbacks.
Proximal sensing techniques such as phenocams or webcams allow high revisit frequency and
objective analysis techniques [
3
5
]. They are based on the principle that changes in canopy biophysical
and -chemical composition, which go along with leaf development and senescence, alter its radiative
regime. Most often exploited is the decrease of reflectance in the visible wavelengths due to absorption
for photosynthesis and the increase in the Near-Infrared (
NIR
) due to reflecting properties of leaves.
The latter effect can be exploited when sensors also record
NIR
[
6
,
7
]. This can be done for example
with tower-based proximal sensing. However, these sensors are often not standardised in terms of
measurement protocols [8].
Apart from this, the change in reflective behaviour is also utilised to detect phenological events
over large areas with satellite-borne sensors [
9
,
10
]. The spatially aggregated, temporal behaviour of
plants over larger areas that are observed from space is referred to as Land Surface Phenology (
LSP
) [
1
].
In contrast to ground-based systems, space-borne missions have the advantage to use only single or
few sensors, which makes it easier to derive comparable products.
The case is similar for another quantitative vegetation property, the Leaf Area Index (
LAI
). It is
defined as the one-sided leaf area per unit of ground surface area [
11
]. Hence, it quantifies the amount
of leaves that are available for photosynthesis during the photoperiod. Similar to
LSP
,
LAI
can be
inferred from ground-based and space-borne instruments, but also from air-borne sensors [
12
14
] and
Unmanned Aerial Vehicless (
UAV
s) [
15
,
16
]. Ground-based
LAI
observations are typically performed
with consumer-grade cameras equipped with fish-eye lenses, referred to as Digital Hemispherical
Photography (
DHP
) [
17
]. This method relies on gap fraction theory to infer
LAI
[
18
]. Most often,
observations with a viewing zenith angle of
57.5
, the so-called hinge angle, are analysed, where
impact of the Leaf Angle Distribution (
LAD
) on gap fraction is minimal. Other hand-held instruments
are also available. However, in general, ground-based instruments for
LAI
are used manually and in
campaigns that cover larger areas to capture the extent of satellite scenes [
13
]. Regular re-sampling of
the same locations is possible, but labour-intensive [19,20].
In the context of Earth observation satellite missions and programmes, such as NASA’s
Moderate-Resolution Imaging Spectroradiometer (
MODIS
) or ESA’s Sentinel-3 mission [
21
,
22
], both
LSP
and
LAI
spatial products require robust quality ground-based validation. This demands
monitoring devices that match land product’s temporal resolution, potentially able to record
LAI
at
high resolution so that
LSP
can be inferred from the time series. Low-cost devices would be preferred
to allow deployment over larger areas and at many sites.
The PAI Autonomous System from Transmittance Sensors at
57
(
PASTiS-57
) is a candidate to fulfil
these requirements. It was developed by Institut national de la recherche agronomique (INRA)-Hiphen
(Avignon, France) during the FP7 ImagineS project (http://fp7-imagines.eu). Its main application
was to support multi-day calibration and validation field campaigns for retrieval of
LAI
and Fraction
of Absorbed Photosynthetically-Active Radiation (
FAPAR
) with hectometric resolution space-borne
sensors [
23
26
]. Recently, it was compared with seasonal measurements of LAI-2200 and
DHP
in
agricultural fields [
27
]. Its measurement principle is based on gap fraction theory. Similar to
DHP
,
it exploits the
LAD
quasi-invariance at the hinge angle. However, a more detailed assessment of its
performance characteristics especially with respect to changing illumination conditions and plant
biochemical properties has not been presented yet.
In this context, vegetation Radiative Transfer Models (
RTM
s) can support sensitivity analysis,
the definition of new sensors and the development of inversion procedures to translate the radiative
signal into canopy variables [
28
]. Generally, vegetation
RTM
s model the interaction of sun radiation
with canopy elements based on the canopy’s biophysical and -chemical properties. In this way, they
can be exploited to assess the sensors sensitivity to canopy parameters in idealised conditions, i.e.,
without measurement noise, and for a wide range of possible canopy conditions. For example, the
widely used PROSAIL model, a combination of the Scattering by Arbitrarily Inclined Leaves (SAIL)
Remote Sens. 2018,10, 1032 3 of 19
canopy and PROSPECT leaf radiative models, has been exploited to design vegetation indices [
29
] and
for biophysical parameter retrieval via inversion [
30
32
]. However, SAIL represents the canopy as a
homogeneous layer of scatterers. This assumption does not hold for clumped canopies such as forests.
In these cases, a heterogeneous representation that can take into account canopy clumping and
non-random structure is more appropriate. The Discrete Anisotropic Radiative Transfer (
DART
)
model implements this paradigm [
33
].
DART
applications include canopy biophysical and -chemical
parameter retrieval [
34
,
35
], surface energy budget studies [
36
] and recently chlorophyll fluorescence
modelling [
37
]. Additionally,
DART
incorporates the option to implement in-scene sensors such as
hemispherical or pinhole cameras [37]. This provides the option to test arbitrary sensor designs.
The aim of this study has been twofold: (i) testing the ground-based transmittance sensor
PASTiS-57
for monitoring
LSP
and
LAI
in a forest stand with daily frequency; and (ii) assessing
the sensor’s sensitivity to canopy properties other than
LAI
and their interactions by means of
RTM
experiments, thereby testing the robustness of the measurement principle to different canopy
conditions. This study is structured as follows: The
PASTiS-57
sensor is presented in detail in
Section 2.1. Section 2.2 describes the field data collection and analysis. Section 2.3 elaborates on
implementation of the sensor in
DART
, the set up of the synthetic canopy and how sensitivity analysis
was conducted. Results are presented in Section 3and discussed in Section 4. Section 5summarises
the results and lists implications for future sensor design.
2. Materials and Methods
2.1. PASTiS-57 Instrument
The
PASTiS-57
consists of a weather proof, battery powered datalogger with 6 photodiode-based
sensors (Figure 1). The sensors are fixed to a viewing zenith angle of
57.5
, sensitive in the blue spectral
region to minimise canopy multiple scattering and have different lengths of wire for sensor distribution
around the data-logger. The logger is battery powered and can autonomously collect data at
1 min
interval for up to four months. Intervals of
2 min
and
5 min
are also possible. Radiation is recorded with
uncalibrated Digital Number (
DN
) in the interval 0–4000, whereas larger
DN
s are treated as unreliable
due to saturation effects. The signal can be calibrated with dedicated Photosynthetically-Active
Radiation (
PAR
) sensors [
25
]. However, a more practical approach is to utilise the relative signal by
installing one device above the canopy, which serves as reference for incoming radiation, and another
device below the canopy, which represents the observations. In this way, many plots can be served by
one reference sensor, as long as it is within a distance where illumination conditions can be assumed
comparable. The observed signal is the spectral directional transmittance τfor each sensor:
τ=DNbelow c ano py
DNabove cano py
(1)
2.2. Field Experiment
2.2.1. Study Area and Field Data Collection
PASTiS-57
sensors have been installed at the Speulderbos Fiducial Reference site in the Veluwe
forest area (N
5215.150
E
542.000
), The Netherlands [
38
] (www.wur.eu/fbprv). This site represents a
maturing stand of mixed European beech (Fagus sylvatica), pedunculate oak (Quercus robur) and sessile
oak (Quercus petraea) with few understorey. The trees were initially planted in 1835. Nowadays, the
stand has a density of around 200 trees/ha.
Remote Sens. 2018,10, 1032 4 of 19
Figure 1.
The
PASTiS-57
instrument installed in the Speulderbos site: (
A
) With the opened data-logger
box, the six sensors can be seen on top, a spare cable is curled up at the back of the data-logger, and
two D-cell batteries for power supply in the box; and (
B
) two
PASTiS-57
installed at the centre of Plot C
(centre marker not visible).
PASTiS-57
instruments were installed in the centres of five plots. For redundancy, each plot was
equipped with two devices (Figure 2). Contrary to previous studies where sensors were put onto the
ground, the individual sensors in this study have been mounted on top of the data logger on a plastic
board to face the NE, E, SE, SW, W and NW directions. Each
PASTiS-57
unit was fixed at
1.30 m
above
ground at an iron rod, aligned to north with a magnetic compass and levelled with a bubble level.
Another two devices have been mounted at the top of a
42 m
high scaffold tower approximately
550 m
west of Plot A to record above canopy reference downward radiation. Campaigns were conducted in
spring 2016 during leaf flush, autumn 2016 during leaf senescence, and summer and autumn 2017.
Campaigns were programmed with 2min interval in 2016 and 1 min interval in 2017.
A
B
E
C
D
5°42'10"E5°42'0"E5°41'50"E5°41'40"E5°41'30"E5°41'20"E
52°15'10"N52°15'5"N52°15'0"N
0 300150 m
Scaffold Tower
¯
Figure 2.
Map of the study site with the scaffold tower where reference instruments were placed and
the five sampling plots. Background is an airborne false-colour composite of 2013. The location of the
study site within The Netherlands is marked on the inset.
Remote Sens. 2018,10, 1032 5 of 19
2.2.2. Plant Area Index (PAI) Retrieval
It should be noted that many proximal sensing techniques cannot distinguish between foliage and
woody canopy elements. The
PAI
includes both classes, while
LAI
refers only to photosynthetically
active plant tissue [
39
,
40
]. In the following,
PAI
refers to observed plant area, while
LAI
refers to the
actual quantity of green leaf area. Typically,
LAI
retrieval from below canopy sensors such as
DHP
uses
gap fraction theory [
18
,
41
]. For this, the canopy is assumed as a uniform cloud of randomly oriented,
black facets (
ρ=
0,
τ=
0). In this case, the gap fraction is related to
LAI
based on Beer–Lambert’s law:
P(θ) = eG(θ)(θ)L/ cos θ(2)
where
θ
is the viewing zenith angle,
P(θ)
the canopy gap fraction in direction
θ
,
G(θ)
the projection of
unit foliage in the
θ
direction, which characterizes the foliage angular distribution,
(θ)
the clumping
index that describes the non-randomness of the canopy and
L
the
LAI
.
P(θ)
has been variably
interpreted as gap fraction in the case of
DHP
[
18
], hit probability in the case of lidar sensors [
42
,
43
] or
transmittance in the case of PASTiS-57 [24,26].
For canopy clumping estimation, the method of Lang and Xiang (
LX
) [
44
] and the six different
viewing directions of each
PASTiS-57
instrument were exploited. The
LX
method assumes that within
a segment the foliage is random and it contains gaps. In the case of
PASTiS-57
the instantaneous field
of view of a single sensor can be interpreted as a segment. In that case, canopy clumping can be
described as:
(θ) = ln P(θ)
ln P(θ)(3)
where
P(θ)
is the mean gap fraction of all segments and
ln P(θ)
the logarithm of the mean gap
fraction of all segments. “The reasoning behind this technique is that since the [
LAI
] is related to the
natural logarithm of the gap fraction, the average LAI should follow the logarithm average of the
gap
fraction” [41]
. In this sense, the denominator in Equation
(3)
normalises the gap fraction by the
logarithm. Using this definition for clumping, replacing
P(θ)
with
PASTiS-57
measured transmittance
τPAST IS
and exploiting the near constant value of the
G
-function at
θ=
57.5
for many
LAD
reduces
Equation (2) to
L=1.075 ln(τPAST IS)
(4)
The goal of this study was to produce daily observations of
LAI
, from which phenological
parameters can be derived. Earlier studies in forests showed that sub-daily retrievals with
PASTiS-57
based on Equation
(4)
produce results with strong, high frequency noise with
±0.5 LAI
amplitude [
45
].
Therefore, raw readings require appropriate quality filters to be applied. Here, these filters were
based on experience gathered while investigating the raw time series. The primary result was that
high frequency noise stems from changing illumination conditions that violate the assumptions for
the retrieval.
Firstly, broken cloud cover can result in different sky illumination conditions at the location of the
reference and the observation sensors. Here, this was counteracted by aggregating via averaging of the
1 min
transmission readings to a daily time series. Secondly, strong cloudiness reduces the radiation
reaching the forest floor, especially at the north-facing sensors, resulting in below canopy
DN
of 0,
which are interpreted as infinite
PAI
(Equation
(4)
) and do not match the assumptions behind the
clumping appraisal. Therefore,
DN
readings of 0 for the below canopy sensors were removed from the
time series. Thirdly, large canopy gaps result in direct illumination on the below canopy sensor, which
violates the assumption of diffuse illumination and produces high
DN
readings. These conditions
are only of short duration when the sun moves over the specific gap. Even the NW and NE sensors
experience these conditions when canopy elements are overly strong illuminated through canopy gaps
and result in recorded high transmission. To counteract this effect, all
DN
exceeding the 95th daily
percentile were removed.
Remote Sens. 2018,10, 1032 6 of 19
2.2.3. Reference Datasets
Next to
PASTiS-57
records, a multi-temporal campaign with a RIEGL VZ-400 Terrestrial Laser
Scanning (
TLS
) (RIEGL LMS GmbH, Horn, Austria) was conducted at the Speulderbos site. This
scanner has shown good results for monitoring phenology [
39
]. The main advantage of
TLS
as a
gap fraction sensor is its independence from illumination conditions [
42
,
46
,
47
]. This results in high
precision time series, i.e., with low noise in the temporal domain. On the other side, partial hits lead
to underestimation of gap fraction by these kind of sensors [
46
,
48
]. Partial hits result from objects
that only partially cover the laser instantaneous field of view, but are registered as full interceptions
by the waveform analysis methods of commercial suppliers to maximise point cloud density. In this
way, gap fraction is underestimated and, consequently,
PAI
is overestimated. In addition, wet canopy
conditions have to be avoided for the sampling, because water droplets on canopy elements absorb
the laser beam, thereby apparently increasing gap fraction.
In total 45 sampling events were conducted. The sampling strategy was to focus efforts during
change periods, i.e., Start Of Season (
SOS
) and End Of Season (
EOS
), and to avoid rain conditions.
For each sampling event, the scanner was mounted on a surveying tripod in each centre of the five
plots, at a maximum distance of
3 m
from the respective
PASTiS-57
devices.
P(θ)
was derived from the
hemispherical scans by taking into account the multi-return capability of the scanner [
43
]. The hinge
angle was approximated with the 55to 60region, which is a typical strategy [39,42,43].
Apart from the ground-based
TLS
time series,
MODIS
Collection 6 MCD15A3H
LAI
products
were retrieved [
49
,
50
]. MCD15A3H is a four-day composite product based on inversion of a vegetation
RTM
. Its eight-day companion product was validated to Stage 2 according to the Land Product
Validation (
LPV
) subgroup (https://lpvs.gsfc.nasa.gov/). However, the four-day product was
preferred over the eight-day product as the goal here included estimating temporal metrics, thus
denser samples were important.
The samples were retrieved from the Application for Extracting and Exploring Analysis Ready
Samples (
APPEARS
) service of the US Geological Survey (https://lpdaacsvc.cr.usgs.gov/appeears/)
for the period 1 January 2016 until 28 February 2018 and for the
500 m
pixel centred at the Speulderbos
site. This means that the MCD15A3H samples also included forest patches other than beech, e.g.,
some mixed species stand included in the Speulderbos site. After downloading, the time series was
filtered with the accompanying quality flags to allow only good quality
LAI
retrievals. After visual
inspection it was clear that some outliers in summer with
LAI
below
3 m2m2
occurred, which have
been excluded as well.
2.2.4. Phenological Model Fitting
LSP
can be modelled with logistic functions [
1
,
9
]. These mathematically simple models are
fitted piecewise to time series of vegetation indices or
LAI
to describe the spring growth and autumn
senescence periods [
39
]. From these models
LSP
indicators like
SOS
and
EOS
can be derived. Here,
a logistic model was used [9,39]:
PAI(t) = UL
1+ek(ttm)+L(5)
where
t
is time (expressed as Day Of Year (
DOY
)),
U
the upper asymptote (
m2m2
),
L
the lower
asymptote (
m2m2
),
k
the growth rate (
d1
) and
tm
the inflection point, where
k
is maximal (
DOY
).
The model was fitted with a non-linear least squares routine implemented in the
nls
function of the R
stats
package [
51
,
52
]. Separate sigmoids were estimated for the spring 2016 and autumn 2017 periods.
In addition to the model parameters’ best estimates, non-linear least squares estimate also produces
prediction intervals. Finally,
SOS
and
EOS
were estimated as the time of the year when the fitted
model reached the 95% upper and lower prediction interval for the
L
and
U
parameter, respectively.
Remote Sens. 2018,10, 1032 7 of 19
This fitting strategy was applied separately to the
PASTiS-57
,
TLS
and
MODIS
time series, and results
were compared.
2.3. Radiative Transfer Model Experiments
In-scene sensors in
DART
can be implemented as frame cameras with arbitrary viewing direction
and properties (see Figure 3for an example), and record at-sensor spectral radiance for any number
of spectral bands with any bandwidths (
W m2sr1µm1
). For this study, the frame camera
characteristic was not exploited, but only the integrated radiance over the whole sensor FOV was
regarded similar to the
PASTiS-57
photo-diodes. As in the field set up (Section 2.2), in this simulation
the 6
PASTiS-57
sensors were directed in the NE, E, SE, SW, W and NW directions with a viewing zenith
angle of
57.5
. In the
DART
scene, one device consisting of six sensors was positioned below, another
above the simulated canopy, so that canopy transmittance could be calculated in the same way as in
the field experiment (Equation
(1)
). The
PASTiS-57
spectral response curve is not exactly known, so a
blue waveband centred at
490 nm
with
20 nm
bandwidth (FWHM) was chosen. Additionally, bands
centred in the green (
560 nm
), red (
665 nm
) and
NIR
(
865 nm
) were tested following the specifications
of the Sentinel-2 Multi Spectral Instrument (
MSI
) [
53
]. These additional bands allowed judging the
instrument performance in the case photo-diodes would have been chosen that are sensitive in another
spectral region.
Figure 3. DART
sample scene: (
A
) True colour image sample for below-canopy sensor in
DART
.
The viewing zenith angle is
57.5
, but the field of view is extended compared to the sensors used in the
modelling to give an overview of the scene. (B) Top view of the created mock-up with 50 trees.
Two scenarios were set up to test different canopy parameters. In both, only the sensitivity of
PASTiS-57
to change in
LAI
was investigated, but not to the temporal evolution. This is justified with
the direct dependence of temporal sensitivity on the sensitivity to
LAI
. The first scenario modelled
the canopy as a turbid medium, which is in accordance to gap fraction theory that is underlying the
LAI
retrieval (Section 2.2.2). This scenario was intended to test the retrieval robustness to different
illumination conditions as well as variation in biophysical and -chemical canopy composition. Table 1
summarises the parameters and their chosen values. For each case, one typical and two extreme cases
were chosen. All parameters were varied in a full grid approach, resulting in a total of 450 simulations.
For this experiment, clumping was not investigated (
=
1) because the canopy was homogeneous in
all directions. The solar azimuth angle was kept constant at
180
. The simulated sensors were analysed
with respect to their prediction performance of the true
LAI
. The results were compared with the
Global Climate Observing System (
GCOS
) requirement for
LAI
retrieval accuracy, which is 20% as
Remote Sens. 2018,10, 1032 8 of 19
well as the accuracy goal for agricultural meteorology applications identified by the WMO , which is
5% [54]. The Relative Error (RE) was chosen as accuracy metric and calculated as:
RE =PAIsi mul ated PAST IS LAIDART
LAIDART
(6)
where
LAIDART
is the
DART
input
LAI
and
PAIsimulated PAST I S
the
PAI
derived with the simulated
PASTiS-57. Using this formulation, positive REs meant overestimation of the PASTiS-57-derived PAI.
During analysis of the results of these
RTM
simulations, a systematic bias in
PAI
estimation has
been identified. This could be linearly modelled with the form
PAI =aLA I +b+e
independently
for each
LAD
(
p<
0.01). For assessment of this error’s impact on
LSP
metrics estimation,
PAI
in
Equation
(5)
was replaced with the linear bias model and solved for
t
. The comparison of this with
the unbiased estimation of
t
gave the expected error in
LSP
metric. Since the analytic solution was
complex, the impact of the relative error was assessed numerically by testing a range of values for
U
,
L
and
k
. In the case of
U
and
L
extreme combinations were tested, i.e.,
L=
0 and
U∈ {
1, 2, ...10
}
.
In the case of
k
, estimates from the field derived models were used (Section 2.2.4), i.e.,
k∈ {
0.5,
0.08
}
.
Values for
tm
were not necessary, because it cancels out when only considering the difference between
two estimates.
Table 1.
Biophysical, biochemical and illumination parameters and values used for turbid DART
experiments.
Parameter Values Unit
Leaf Area Index (LAI) 1, 2, ..., 10 m2m2
Leaf Angle Distribution (LAD) spherical, erectrophile, planophile, extremophile, plagiophile -
Chlorophyll a and b (Cab) 20, 50, 80 µg cm2
Solar Zenith Angle (SZA) 0, 57.5, 80
The second scenario was intended to test the retrieval performance with respect to canopy
non-randomness. For this, discrete trees were modelled with ellipsoid crowns with
10 m
diameter
and
5 m
height, and trunks with
40 cm
diameter. The number of trees per scene was varied between
50 and 400 trees on a scene of
80 m×80 m
. Illumination and canopy biochemical parameters were held
constant with a spherical
LAD
,
50 µg cm2Cab
and
SZA
of
57
to extract the effect of clumping alone.
3. Results
3.1. Field Experiment
Figure 4shows the raw
DN
recordings of two sampling days, one before
SOS
, one after leaf flush
in summer. For both days, averages were clearly lower than reference readings above the canopy,
resulting in average canopy transmittance of 29.1% and 0.9% before and after the start of season,
respectively. Another feature was the high number of
0 DN
readings at early and late hours of the
day after leaf flush, which made up 25.7% of all observations on that day. At these times, the
SZA
is
typically large, so that the direct path through the canopy is long and not sufficient radiation reaches
the below canopy sensors. In contrast to this, the SW sensor on the reference device experienced
saturation in the afternoon, probably due to direct illumination. Overall, the two days showed high
agreement in temporal evolution, indicating similar impact of changes in illumination. These stem
from the course of the sun, resulting in the rise and fall of readings over the course of the day, and
from clouds and stems, causing high frequency changes.
Remote Sens. 2018,10, 1032 9 of 19
Avg = 162.08 N = 2247
Avg = 556.62 N = 2247
Avg = 10.27 N = 1112
Avg = 1119.41 N = 1112
2016−04−02
2016−05−29
Obs
Ref
06:00 09:00 12:00 15:00 18:00 06:00 09:00 12:00 15:00 18:00
10
1000
10
1000
DN
Sensor
E
NE
NW
SE
SW
W
Figure 4.
Recordings of two sampling days for one device in Plot A. Upper panels are the observations
below the canopy, lower panels are reference readings from above the canopy.
DN
axis is on log-scale.
Dotted horizontal line is saturation point at
4000 DN
. Discontinued lines on lower panel reach
saturation. Only pairs for which the observation did not reach 0 DigitalNumber were considered.
A full
PAI
time series for all three campaigns derived from a device on Plot C can be seen in
Figure 5. While the difference between the single sensors was only marginal, the impact of the filtering
was clearly visible. The naive retrieval resulted in strong, high frequency noise with positive spikes
and a Lag 1 Auto-Correlation Coefficient (
ACC1
) of 0.93. The noise after filtering was modest and
evolved equally around a mean course of
PAI
with
ACC1
of 0.97. The former resulted from situations
under full canopy in summer, when the below-canopy sensors had
0 DN
readings. This results in
theoretically infinite
PAI
according to Equation
(4)
. When comparing the two years, an earlier decrease
in
PAI
could be observed in 2017. This can be explained with the natural variability of
EOS
. This results
from different wind loads, which is the main force to defoliate the trees once the leaves have died.
Spring 2016
Autumn 2016
Summer 2017
Apr May Jun Sep Oct Nov Aug Sep Oct Nov Dec
2.5
5.0
7.5
10.0
PAI [m2m2]
Method
Naive
Filtered
Figure 5.
All campaigns of one instrument in Plot C before (Naive) and after application of filtering
(Filtered).
Remote Sens. 2018,10, 1032 10 of 19
In Figure 6
PASTiS-57
and
TLS
estimated
PAI
are plotted together for the dynamic phases of
the yearly phenology, which are spring leaf flush and autumn senescence. Overall,
PASTiS-57
and
TLS
showed high agreement in temporal development. Both sensors’ time series reflect the fast leaf
development during spring and the longer senescence period in autumn. The
TLS
sampling intervals
were not sufficient to record the fast changes in spring, especially between the sampling events of
4 May 2016
and 12 May 2016 when
PAI
increased by
1.86 m2m2
on average within eight days.
The PASTiS-57 with their daily interval could closely follow the development.
Considering the magnitude, both sensors agreed strongly with a Pearson correlation coefficient of
0.98. However,
PASTiS-57
retrievals were lower in winter by
1.44 m2m2
compared to
TLS
. This was
likely due to the different sensing techniques. In case of
PASTiS-57
, backward scattering from woody
elements increases recorded radiation at the below canopy sensors, thus decreases
PAI
estimates.
In the case of
TLS
, the partial hits are mainly responsible for the sensitivity to the recording of canopy
elements. For both instruments,
PAI >
1.5 in winter pointed to the large influence of woody material on
the retrievals. In contrast to this,
PASTiS-57
average and
TLS
agreed in summer to within
0.74 m2m2
.
Parameter estimates for the fitted phenological models are summarised in Tables 2and 3. A total
of 21 and 40 samples for each plot were used for
TLS
and
MODIS
, respectively. Concerning the upper
and lower asymptotes
U
and
L
,
PASTiS-57
showed significantly lower estimates compared to
TLS
in
all plots. For both spring and autumn campaigns,
PASTiS-57 U
and
L
were on average
0.41 m2m2
and
1.43 m2m2
lower than
TLS
, reflecting the difference in acquisition mechanism. Compared to
MODIS
,
PASTiS-57 U
and
L
were
0.19 m2m2
lower and
0.97 m2m2
higher, respectively. Again,
this reflects the different nature of the retrieval algorithms.
MODIS LAI
makes use of top of canopy
reflectance and is stronger utilising the
NIR
signal. This makes it less sensitive to woody material in
the canopy, thus MODIS LAI showed generally lower values than PASTiS-57.
Summer 2017
A
Summer 2017
B
Summer 2017
C
Summer 2017
D
Summer 2017
E
Spring 2016
A
Spring 2016
B
Spring 2016
C
Spring 2016
D
Spring 2016
E
Aug Sep Oct Nov Dec Jan Feb Aug Sep Oct Nov Dec Jan Feb Aug Sep Oct Nov Dec Jan Feb Aug Sep Oct Nov Dec Jan Feb Aug Sep Oct Nov Dec Jan Feb
Apr May Jun Apr May Jun Apr May Jun Apr May Jun Apr May Jun
0.0
2.5
5.0
7.5
0.0
2.5
5.0
7.5
PAI (LAI) [m2m2]
Sensor
E
NE
NW
SE
SW
W
LX
MODIS
TLS
Figure 6.
Comparison of
PASTiS-57
and
TLS
derived
PAI
for single sensors (coloured) and Land and
Xiang clumping correction (LX), and
MODIS LAI
for five plots during the spring 2016 and summer
2017 campaigns.
Remote Sens. 2018,10, 1032 11 of 19
Table 2.
Phenological model fitting results for the spring 2016 campaign with parameter mean estimates
and their 95% standard error.
U
is the upper asymptote (
m2m2
),
L
the lower asymptote (
m2m2
),
k
the growth rate (
d1
),
tm
the inflection point (
DOY
) and
SOS
the Start Of Season (
DOY
).
MODIS
results refer to all plots and represent LAI in case of Uand L.
Parameter A B C D E
UPASTIS 5.31 (±0.04) 5.65 (±0.04) 5.46 (±0.05) 5.09 (±0.05) 5.67 (±0.05)
UTLS 5.82 (±0.03) 6.10 (±0.03) 5.72 (±0.04) 5.62 (±0.04) 5.99 (±0.03)
UMOD IS 5.63 (±0.15) – – – –
LPASTIS 1.63 (±0.03) 1.94 (±0.03) 1.85 (±0.04) 1.57 (±0.03) 1.75 (±0.04)
LTLS 3.10 (±0.02) 3.51 (±0.03) 3.02 (±0.03) 3.03 (±0.03) 3.32 (±0.02)
LMOD IS 0.80 (±0.09) – – – –
kPASTIS 0.43 (±0.03) 0.49 (±0.04) 0.54 (±0.06) 0.48 (±0.04) 0.44 (±0.03)
kTLS 0.41 (±0.03) 0.52 (±0.05) 0.47 (±0.04) 0.43 (±0.04) 0.31 (±0.02)
kMOD IS 0.77 (±0.26) – – – –
tm,PASTIS 129.5 (±0.2) 129.3 (±0.2) 129.3 (±0.2) 129.2 (±0.2) 129.1 (±0.2)
tm,TLS 130.0 (±0.3) 128.6 (±0.4) 129.3 (±0.4) 129.5 (±0.4) 130.9 (±0.3)
tm,MOD IS 126.0 (±0.5) – – – –
SOSPAST I S 117.9 119.5 121.0 119.6 118.4
SOST LS 118.2 119.9 119.7 119.2 115.8
SOSMO DIS 120.9 – – – –
Table 3. Same as Table 2, but for the autumn 2017 campaign referring to the EOS (DOY).
Parameter A B C D E
UPASTIS 5.69 (±0.04) 5.72 (±0.03) 5.64 (±0.05) 5.49 (±0.05) 5.79 (±0.04)
UTLS 6.06 (±0.08) 6.07 (±0.08) 6.00 (±0.13) 6.05 (±0.12) 6.15 (±0.10)
UMOD IS 5.06 (±0.11) – – – –
LPASTIS 1.59 (±0.04) 1.80 (±0.04) 1.69 (±0.05) 1.55 (±0.05) 1.64 (±0.04)
LTLS 3.05 (±0.10) 3.43 (±0.14) 2.74 (±0.17) 2.78 (±0.18) 3.34 (±0.16)
LMOD IS 0.65 (±0.09) – – – –
kPASTIS 0.08 (±0.00) 0.08 (±0.00) 0.07 (±0.00) 0.08 (±0.01) 0.09 (±0.01)
kTLS 0.08 (±0.01) 0.12 (±0.02) 0.07 (±0.01) 0.10 (±0.02) 0.09 (±0.02)
kMOD IS 0.16 (±0.02) – – – –
tm,PASTIS 307.6 (±0.7) 307.8 (±0.7) 307.5 (±1.0) 309.8 (±1.0) 311.7 (±0.7)
tm,TLS 307.2 (±1.9) 317.4 (±2.0) 307.4 (±3.0) 314.4 (±2.6) 316.7 (±2.6)
tm,MOD IS 292.5 (±1.7) – – – –
EOSPAST I S 250.0 250.4 243.2 256.2 262.8
EOST LS 258.8 288.4 263.7 280.8 281.4
EOSMO DIS 269.0 – – – –
Furthermore,
PASTiS-57
agreed very well with
TLS
in terms of
SOS
with an average difference of
less than a day.
EOS
was estimated on average
22 days
later by
TLS
. However, agreement among
TLS
plots was low with a range of
29 days
in
EOS
. More samples during winter would have been necessary
to decrease the estimation error. Moreover,
PASTiS-57
achieved the lowest estimation standard error on
the sigmoid inflection points
tm
. This was made possible by the high temporal density of the
PASTiS-57
time series. Additionally,
PASTiS-57
agreed well with
MODIS SOS
to within
2 days
. As in the case of
TLS
,
EOS
estimation of
MODIS
was impaired. Only for
MODIS
, persistent cloud cover— which is
common in autumn in The Netherlands—prevented frequent observations.
Remote Sens. 2018,10, 1032 12 of 19
3.2. Radiative Transfer Model Experiments
The
DART RTM
experiments permitted to have control over all canopy and illumination
parameters, and to model abstract canopies. Figure 7summarises the results for the turbid medium
canopy case. Most influential was the choice of the spectral band. For instance,
NIR
retrievals were
generally more than 75% lower than true
LAI
. This strong misinterpretation stems from the retrieval
assumption of black leaves, which is not fulfilled in the
NIR
. In fact, leaves typically transmit around
45% of incoming radiation in this band. This leads to higher recorded radiation below canopy and
underestimation of
LAI
. Additionally, the
RE
was larger at small
SZA
. This could be explained by a
smaller optical path through the canopy at small
SZA
, which leads to increased below-canopy recorded
radiation compared to what would be expected for black leaves. These effects could also be observed
to some degree in the green spectral band, where leaves typically transmit >10%.
Blue
Green
Red
NIR
0.0 2.5 5.0 7.5 10.0 0.0 2.5 5.0 7.5 10.0 0.0 2.5 5.0 7.5 10.0 0.0 2.5 5.0 7.5 10.0
−200
−150
−100
−50
0
−40
−30
−20
−10
0
10
−40
−30
−20
−10
0
10
−40
−30
−20
−10
0
10
DART LAI [m2m2]
Relative error [%]
LAD
Spherical
Erectophile
Planophile
Extremophile
Plagiophile
SZA
0
57.5
80
Figure 7. DART
model results for turbid canopy representation for west facing sensors. Positive errors
mean over-estimation by the retrieval. Light grey and darker grey areas are the 20% and 5% accuracy
requirement of GCOS [54].
In contrast, the blue and red bands were less compromised. They both underestimated true LAI
by maximum 40.0% and on average by 14.2% and 14.5%, respectively. Thus, the average accuracy was
within the
GCOS
threshold accuracy of 20%. Leaf absorption is strongest at these wavelengths due to
absorption by chlorophyll, so that the canopy comes close to the approximation of black facets. Typical
transmittance for these spectral regions is
<
2% and reflectance
<
4%. This is why blue channels of
digital cameras are recommended for
LAI
retrieval [
20
,
55
]. Nonetheless, even these low values in
ρ
and
τ
led to higher detected radiation at the sensor compared to what would be expected with black
leaves, so that canopy transmittance was overestimated, which leads to underestimation of PAI.
Within the blue and red spectral bands most variation was across the different
LAD
s. While
the spherical
LAD
underestimated true
LAI
by an average of 2.8% and a maximum of 11.0% in the
blue, simulated canopies with planophile
LAD
s resulted in average and maximum underestimation of
38.2% and 40.0%, respectively. In the latter case, the deviation of the
G
-function value from 0.5 and
transmitting properties of the leaves probably interacted to increase the deviation from true LAI.
Apart from this, underestimation was generally increasing with true
LAI
. This means that
at higher
LAI
more radiation was reaching the sensor than expected by the model, i.e., canopy
transmittance is larger than expected. This effect could be created by multiple scattering in the canopy.
According to the model assumption there are no scattering processes within the canopy. Radiation is
only absorbed or transmitted without interaction. However, the
DART
simulated leaves had
τ>
0,
which allows radiation to go through leaves. The more leaves there are, the stronger the mismatch
between gap fraction and DART model.
Remote Sens. 2018,10, 1032 13 of 19
Concerning
Cab
, pairwise Student’s
t
-tests between any of the
Cab
levels showed no significant
differences in
LAI
estimation (
p>
0.95) and differences were below 0.1%. This showed that the direct
influence of Cab on LAI estimation was very low.
Finally,
SZA
had minor overall impact on the retrievals. This was on average 0.9% in the blue
band between
SZA 0
and
57.5
. However, the difference was larger for spherical and erectophile
LAD
s. The extreme case was at
LAI
10, where the relative error for
0
and
57.5
differed by
5.1
and
9.7for spherical and erectophile LADs, respectively.
When translating the impact of the bias in
LAI
retrieval on
LSP
metrics, erectophile
LAD
s
delivered the largest error with
1.9 days
later estimation of
EOS
. Spherical, planophile, extremophile
and plagiophile
LAD
s resulted in
1.5 days
,
0.8 days
,
1.0 days
and
1.1 days
later
EOS
, respectively.
SOS
estimation showed lower errors with on average
0.1 days
. This means
EOS
as the generally slower
process experiences larger errors in
LSP
metric retrieval based on the
LAI
bias error. It should be noted
that this difference was based on the particular phenological model used here (Section 2.2.4), but models
based on sigmoid functions in general should experience errors on the same order of magnitude.
RTM
results for the heterogeneous scenario are presented in Figure 8. Tree density was
significantly altering retrieval performance in scenarios with
<
200 trees and
LAI
>
5 m2m2
. This led
to underestimation of up to 69.4% at true
LAI 10 m2m2
and a scene with 50 trees. For these scenarios
the present leaf mass was concentrated in few crowns, so that the assumption of a homogeneous
canopy did not hold and
LAI
was underestimated. The clumping correction after Lang and Xiang [
44
]
could account for some of these effects, but could only reduce the underestimation to 55.1% in the case
of 50 trees. In those cases, the assumption of a random foliage distribution within the sensor FOV
was violated. Actually, the horizontal FOV of the
PASTiS-57
is large compared to the solid angles that
camera pixels represent. The clumping correction after Lang and Xiang [
44
] corresponds rather to the
small FOVs represented by camera pixels.
Trees = 50
Trees = 100
Trees = 200
Trees = 300
Trees = 400
0.0 2.5 5.0 7.5 10.0 0.0 2.5 5.0 7.5 10.0 0.0 2.5 5.0 7.5 10.0 0.0 2.5 5.0 7.5 10.0 0.0 2.5 5.0 7.5 10.0
0
3
6
9
DART LAI [m2m2]
Retrieved PAI [m2m2]
Clumping
correction
LX
None
Figure 8. DART
model results of discrete canopy representation for five different tree densities
(horizontal panels in number of trees). Retrieval without (None) and with clumping correction after
Lang and Xiang (LX) [44].
4. Discussion
Ground-validation of
LSP
and
LAI
require high temporal density canopy observations. This
study explored the
PASTiS-57
instrument for autonomous monitoring of phenology and
PAI
in a
Dutch beech forest.
DART RTM
experiments helped to evaluate sensing mechanism of the
PASTiS-57
in relation to changes in canopy biochemical and structural properties other than LAI.
The field experiments showed very good temporal agreement with illumination independent
TLS
and
MODIS LAI
products when temporal density of these reference products was high. Biases in
PAI magnitude were attributed to differences in sensing mechanism. The field observations required
filtering and aggregating the readings to daily time series to reduce high frequency noise, especially
during full canopy coverage in summer. This noise can be partially tracked back to the sensor’s
radiometric resolution of
4000 DN
. Considering Equation
(4)
, the change in
PAI
per
DN
, which is
the sensitivity to signal digitisation, is inversely proportional to the
DN
. This is because the first
derivative of Equation
(4)
with respect to
τ
is proportional to the inverse of
τ
:
L01
τ
. This can result
Remote Sens. 2018,10, 1032 14 of 19
in differences as large as
0.75 m2m2
between
DN
observation readings of 1 and 2 when the reference
sensor is close to saturation (Figure 9). Radiometric sensitivity also impacts the maximum
PAI
that
can be recorded. In the case of
PASTiS-57
, it lies at
8.91 m2m2
with a single measurement. Modern
digital cameras typically offer digitisation up to
14 bit
for raw images, resulting in
16 384
grey levels,
so that theoretically
10.43 m2m2
can be retrieved. Therefore, a higher signal bit depth improves
the sensitivity to high
LAI
as well as maximum retrievable
LAI
. In the case of the field experiment,
the maximum summer PAI was 6 m2m2(Figure 5), which was within the theoretical range.
DN Reference = 2000
DN Reference = 4000
0 10 20 30 40 50 0 10 20 30 40 50
4
5
6
7
8
9
DN Observation
Retrieved PAI [m2m2]
Figure 9.
Sensitivity of
PASTiS-57 PAI
due to digitisation at low observation readings (
DN
) for two
levels of reference readings.
The
RTM
experiments confirmed the principles underlying the retrievals. In particular, below
canopy transmittance measurements in blue and red spectral bands have a high sensitivity to canopy
structure and are robust against variation in biochemical composition and illumination conditions. This
is not the case for top of canopy reflectance measuring sensors, e.g., tower based or satellite sensors,
which often exploit the NIR. These are much more dependent on Cab and illumination angle [28].
However, heterogeneous scenarios confirmed the strong effect of canopy non-randomness on
LAI
estimation. In particular in case of low tree density scenarios, which violate the homogeneous
canopy assumption more than dense canopies with closed cover,
LAI
was strongly underestimated.
Clumping correction after Lang and Xiang [
44
] counteracted this effect somewhat. Other clumping
correction strategies exist, but these usually require estimation of gap size distribution [
41
,
56
]. This is
possible with
DHP
, but not with pointing devices, such as
PASTiS-57
. Therefore, a strategy for field
measurements would be to employ multiple
PASTiS-57
instruments per plot. Alternatively, a new
sensor design based on low-cost micro-computers equipped with fish-eye cameras could be tested.
Such an imaging sensor could also retrieve LAD concurrently with LAI [18].
Another disadvantage of the single-band, pointing device design of the
PASTiS-57
is the lack of
options to distinguish woody and foliage canopy elements. Gower et al. [
57
] list ranges of 7% to 34% of
wood area index contribution to
PAI
based on a literature review. Previous studies proposed solutions
to this problem with multi-band imaging sensors, including NIR [
58
] or imaging sensors combined
with radiative transfer modelling [
40
]. Another way is multi-temporal estimation by using the winter
measured PAI as branch area index and subtract it from the summer measured PAI. However, this
neglects radiative interaction processes when both elements are present in the canopy (e.g., occlusion
of leaves by branches) and is not agreed on [
59
]. The lack of consolidated correction methods has
also led to a prevalent neglect of correction [
40
,
59
]. This topic needs to be addressed with dedicated
devices, e.g., dual-wavelength lidar [60].
In the context of sensor simulation,
DART
proved to be a versatile tool. Especially the option to
simulate arbitrary sensors allowed the implementation of the
PASTiS-57
sensor in this study. Although
Remote Sens. 2018,10, 1032 15 of 19
sensor simulation with
RTM
s is not new, below canopy sensor simulations have been restricted to
DHP
[
56
] or
TLS
[
61
,
62
]. Another advantage of
DART
was the option to simulate heterogeneous
canopies, which is crucial for forest radiative transfer modelling.
Next to considerations concerning the retrieval principle, thoughts should be given to practical
instrument design choices. For instance, the power supply with batteries is a good choice for remote
sites and proved to provide electricity for ~1 year. However, close to field stations constant power
could be supplied via the electricity grid or from centrally organised solar cells to prevent power loss
and missing observations. A permanent data-link to the logger and upload to cloud servers could help
to identify sensor problems and monitor results in real-time. Furthermore, the contamination of the
sensors with water, falling leaves or needles, or with insects should be taken care of in a long-term
deployment. In sites with substantial understorey, sensors could be deployed at different heights and
below understorey plants to sample the vertical profile. In addition, sensors at larger heights might be
able to focus on the foliage and prevent large stems to be in the FOV.
In the context of a set-up in larger, permanent sample sites, the representative area of
PASTiS-57
should be considered to determine the number of required devices. In this respect,
PASTiS-57
is
comparable to other below-canopy sensors such as
DHP
and Licor LAI-2000 that measure
τ
at the
hinge angle. Therefore, the diameter of the measurement area is 2
×
canopy height
/ tan(
57.5
)
. This
results in a diameter of
32 m
for a
25 m
high canopy, as is the case for Speulderbos. Considering
geo-location error of 1 pixel [
53
] this would be representative for Sentinel-2
10 m
resolution bands.
However, replicates need to be installed per plot to improve precision in the case of
LAI
validation [
54
].
In the case of Sentinel-3—when considering geo-location error—a footprint of
1000 m
would need to
be covered. Locations should be sampled to account for the site heterogeneity, i.e., number of species,
differences in canopy structure and presence/variability of understorey.
Furthermore, low-cost, passive sensors such as
PASTiS-57
can be combined with
light-independent monitoring. For instance, Culvenor et al. [
63
] presented a monitoring lidar system
that samples the hinge angle, similar to the
TLS
used in this study. These systems are more cost
and maintenance intense, but offer opportunities for inter-comparison and benchmarking, also with
traditional manual sampling methods. Such a combination of sensors would offer the option of high
precision light-independent sensors for site central areas and low-cost sensors for covering larger areas.
This would provide the instrument infrastructure necessary for continuous validation of
LSP
and
LAI
products, as required by validation Stage 4 of the GCOS LPV group [54].
5. Conclusions
Robust tracking of the phenological cycle and thereby connected canopy biophysical conditions
requires sampling techniques with high temporal resolution. This study assessed the ground-based,
autonomous, cost-efficient
PASTiS-57
instrument in both field and
RTM
experiments for its
performance in forest
SOS
,
EOS
and
LAI
estimation. The instrument design supported acquisition
of yearly time series at up to
1 min
raw data resolution with low maintenance effort. The choice of
the blue spectral region and a viewing angle of
57.5
was found to be robust for a range of canopy
biochemical and illumination conditions, thereby focussing on changes in canopy structure, mainly
LAI
. However, clumping assessment in irregular canopies was limited by the low number of sensors
per instrument and the sensors’ pointing measurements. In addition, the restriction to a viewing
angle of
57.5
alone does not allow retrieval of
LAD
together with
LAI
, as is possible with
DHP
.
Future studies should compare
PASTiS-57
with other phenology monitoring devices and develop
combinations of instruments as site concepts. Other sensor designs could be tested, e.g., based on
imaging sensors.
Author Contributions:
B.B. designed and conducted the experiments, and wrote the paper. J.-P.G.-E. and N.L.
supported the
DART
experiments. N.L. implemented the sensor option in
DART
. F.B. designed and built the
PASTiS-57 instrument. J.G.P.W.C., J.V. and M.H. scientifically supported and reviewed the paper.
Funding: This work was carried out as part of the IDEAS+ contract funded by ESA-ESRIN.
Remote Sens. 2018,10, 1032 16 of 19
Acknowledgments:
Data analysis was supported by the Research and User Support (RUS) Service. The RUS
Service is funded by the European Commission, managed by the European Space Agency, and operated by CSSI
and its partners. The authors thank the Dutch Forestry Service (Staatsbosbeheer) for granting access to the site,
and Murat Üçer and Christiaan van der Tol for access to the scaffold tower. The fieldwork design of this study
was inspired by the M.Sc. thesis of Tom Schenkels (http://edepot.wur.nl/333531). Further thanks go to Marcello
Novani, Alvaro Lau Sarmiento and Mathieu Decuyper for help during the fieldwork.
Conflicts of Interest: The authors declare no conflict of interest.
References
1.
Kimball, J. Vegetation Phenology. In Encyclopedia of Remote Sensing; Njoku, E.G., Ed.; Encyclopedia of Earth
Sciences Series; Springer: New York, NY, USA, 2014; pp. 886–890.
2.
Richardson, A.D.; Keenan, T.F.; Migliavacca, M.; Ryu, Y.; Sonnentag, O.; Toomey, M. Climate change,
phenology, and phenological control of vegetation feedbacks to the climate system. Agric. For. Meteorol.
2013,169, 156–173.
3.
Browning, D.; Karl, J.; Morin, D.; Richardson, A.; Tweedie, C. Phenocams Bridge the Gap between Field and
Satellite Observations in an Arid Grassland Ecosystem. Remote Sens. 2017,9, 1071.
4.
Keenan, T.F.; Darby, B.; Felts, E.; Sonnentag, O.; Friedl, M.A.; Hufkens, K.; O’Keefe, J.; Klosterman, S.;
Munger, J.W.; Toomey, M.; et al. Tracking forest phenology and seasonal physiology using digital repeat
photography: A critical assessment. Ecol. Appl. 2014,24, 1478–1489.
5.
Richardson, A.D.; Braswell, B.H.; Hollinger, D.Y.; Jenkins, J.P.; Ollinger, S.V. Near-surface remote sensing of
spatial and temporal variation in canopy phenology. Ecol. Appl. 2009,19, 1417–1428.
6.
Petach, A.R.; Toomey, M.; Aubrecht, D.M.; Richardson, A.D. Monitoring vegetation phenology using an
infrared-enabled security camera. Agric. For. Meteorol. 2014,195–196, 143–151.
7.
Soudani, K.; Hmimina, G.; Delpierre, N.; Pontailler, J.Y.; Aubinet, M.; Bonal, D.; Caquet, B.; de Grandcourt, A.;
Burban, B.; Flechard, C.; et al. Ground-based Network of NDVI measurements for tracking temporal
dynamics of canopy structure and vegetation phenology in different biomes. Remote Sens. Environ.
2012,123, 234–245.
8.
Balzarolo, M.; Anderson, K.; Nichol, C.; Rossini, M.; Vescovo, L.; Arriga, N.; Wohlfahrt, G.; Calvet, J.C.;
Carrara, A.; Cerasoli, S.; et al. Ground-based optical measurements at European flux sites: A review of
methods, instruments and current controversies. Sensors 2011,11, 7954–7981.
9.
Zhang, X.; Friedl, M.A.; Schaaf, C.B.; Strahler, A.H.; Hodges, J.C.F.; Gao, F.; Reed, B.C.; Huete, A. Monitoring
vegetation phenology using MODIS. Remote Sens. Environ. 2003,84, 471–475.
10.
Che, M.; Chen, B.; Zhang, H.; Fang, S.; Xu, G.; Lin, X.; Wang, Y. A New Equation for Deriving Vegetation
Phenophase from Time Series of Leaf Area Index (LAI) Data. Remote Sens. 2014,6, 5650–5670.
11. Chen, J.; Black, T.A. Defining leaf area index for non-flat leaves. Plant Cell Environ. 1992,15, 421–429.
12.
Koetz, B.; Baret, F.; Poilvé, H.; Hill, J. Use of coupled canopy structure dynamic and radiative transfer models
to estimate biophysical canopy characteristics. Remote Sens. Environ. 2005,95, 115–124.
13.
Frampton, W.J.; Dash, J.; Watmough, G.; Milton, E.J. Evaluating the capabilities of Sentinel-2 for quantitative
estimation of biophysical variables in vegetation. ISPRS J. Photogramm. Remote Sens. 2013,82, 83–92.
14.
Casas, A.; Riaño, D.; Ustin, S.L.; Dennison, P.; Salas, J. Estimation of water-related biochemical and
biophysical vegetation properties using multitemporal airborne hyperspectral data and its comparison to
MODIS spectral response. Remote Sens. Environ. 2014,148, 28–41.
15.
Chianucci, F.; Disperati, L.; Guzzi, D.; Bianchini, D.; Nardino, V.; Lastri, C.; Rindinella, A.; Corona, P.
Estimation of canopy attributes in beech forests using true colour digital images from a small fixed-wing
UAV. Int. J. Appl. Earth Obs. Geoinf. 2016,47, 60–68.
16.
Roosjen, P.P.; Brede, B.; Suomalainen, J.M.; Bartholomeus, H.M.; Kooistra, L.; Clevers, J.G. Improved
estimation of leaf area index and leaf chlorophyll content of a potato crop using multi-angle spectral
data–potential of unmanned aerial vehicle imagery. Int. J. Appl. Earth Obs. Geoinf. 2018,66, 14–26.
17.
Jonckheere, I.; Fleck, S.; Nackaerts, K.; Muys, B.; Coppin, P.; Weiss, M.; Baret, F. Review of methods for in situ
leaf area index determination Part I. Theories, sensors and hemispherical photography. Agric. For. Meteorol.
2004,121, 19–35.
18.
Weiss, M.; Baret, F.; Smith, G.J.; Jonckheere, I.; Coppin, P. Review of methods for in situ leaf area index (LAI)
determination Part II. Estimation of LAI, errors and sampling. Agric. For. Meteorol. 2004,121, 37–53.
Remote Sens. 2018,10, 1032 17 of 19
19.
Lukasová, V.; Lang, M.; Škvarenina, J. Seasonal Changes in NDVI in Relation to Phenological Phases, LAI
and PAI of Beech Forests. Balt. For. 2014,20, 248–262.
20.
Lang, M.; Nilson, T.; Kuusk, A.; Pisek, J.; Korhonen, L.; Uri, V. Digital photography for tracking the
phenology of an evergreen conifer stand. Agric. For. Meteorol. 2017,246, 15–21.
21.
Berger, M.; Moreno, J.; Johannessen, J.A.; Levelt, P.F.; Hanssen, R.F. ESA’s sentinel missions in support of
Earth system science. Remote Sens. Environ. 2012,120, 84–90.
22.
Donlon, C.; Berruti, B.; Buongiorno, A.; Ferreira, M.H.; Féménias, P.; Frerick, J.; Goryl, P.; Klein, U.; Laur, H.;
Mavrocordatos, C.; et al. The Global Monitoring for Environment and Security (GMES) Sentinel-3 mission.
Remote Sens. Environ. 2012,120, 37–57.
23.
Simic, A.; Baret, F.; Weiss, M.; Lecerf, R.; Alessandrini, A.; Hanocq, J.F.; Marloie, O. Production of the high
resolution maps of biophysical variables based on SPOT imagery and in-situ measurements generated by
PASTIS 57 for Hyytiala, Finland. In Proceedings of the IEEE International Geoscience and Remote Sensing
Symposium, Munich, Germany, 22–27 July 2012; pp. 7655–7658.
24.
Raymaekers, D.; Garcia, A.; Di Bella, C.; Beget, M.; Llavallol, C.; Oricchio, P.; Straschnoy, J.; Weiss, M.;
Baret, F. SPOT-VEGETATION GEOV1 biophysical parameters in semi-arid agro-ecosystems. Int. J. Remote
Sens. 2014,35, 2534–2547.
25.
Latorre, C.; Camacho, F.; De la Cruz, F.; Lacaze, R.; Weiss, M.; Baret, F. Seasonal monitoring of FAPAR
over the Barrax cropland site in Spain, in support of the validation of PROBA-V products at 333 m.
In Proceedings of the 4th International Symposium on Recent Advances in Quantitative Remote Sensing,
Torrent, Spain, 22–26 September 2014; pp. 1–6.
26.
Nestola, E.; Sánchez-Zapero, J.; Latorre, C.; Mazzenga, F.; Matteucci, G.; Calfapietra, C.; Camacho, F.
Validation of PROBA-V GEOV1 and MODIS C5 & C6 fAPAR products in a deciduous beech forest site in
Italy. Remote Sens. 2017,9, 126.
27.
Fang, H.; Ye, Y.; Liu, W.; Wei, S.; Ma, L. Continuous estimation of canopy leaf area index (LAI) and clumping
index over broadleaf crop fields: An investigation of the PASTIS-57 instrument and smartphone applications.
Agric. For. Meteorol. 2018,253–254, 48–61.
28.
Jacquemoud, S.; Verhoef, W.; Baret, F.; Bacour, C.; Zarco-Tejada, P.J.; Asner, G.P.; François, C.; Ustin, S.L.
PROSPECT + SAIL models: A review of use for vegetation characterization. Remote Sens. Environ.
2009,113, S56–S66.
29.
Clevers, J.G.P.W.; Verhoef, W. LAI estimation by means of the WDVI: A sensitivity analysis with a combined
PROSPECT-SAIL model. Remote Sens. Rev. 1993,7, 43–64.
30.
Atzberger, C.; Richter, K. Spatially constrained inversion of radiative transfer models for improved LAI
mapping from future Sentinel-2 imagery. Remote Sens. Environ. 2012,120, 208–218.
31.
Rivera, J.P.; Verrelst, J.; Leonenko, G.; Moreno, J. Multiple cost functions and regularization options for
improved retrieval of leaf chlorophyll content and LAI through inversion of the PROSAIL model. Remote Sens.
2013,5, 3280–3304.
32.
Campos-Taberner, M.; García-Haro, F.J.; Camps-Valls, G.; Grau-Muedra, G.; Nutini, F.; Crema, A.;
Boschetti, M. Multitemporal and multiresolution leaf area index retrieval for operational local rice crop
monitoring. Remote Sens. Environ. 2016,187, 102–118.
33.
Gastellu-Etchegorry, J.P.; Demarez, V.; Pinel, V.; Zagolski, F. Modeling radiative transfer in heterogeneous
3-D vegetation canopies. Remote Sens. Environ. 1996,58, 131–156.
34.
Demarez, V.; Gastellu-Etchegorry, J. A Modeling Approach for Studying Forest Chlorophyll Content.
Remote Sens. Environ. 2000,71, 226–238.
35.
Banskota, A.; Serbin, S.P.; Wynne, R.H.; Thomas, V.A.; Falkowski, M.J.; Kayastha, N.; Gastellu-Etchegorry, J.P.;
Townsend, P.A. An LUT-Based Inversion of DART Model to Estimate Forest LAI from Hyperspectral Data.
IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2015,8, 3147–3160.
36.
Gastellu-Etchegorry, J.P.; Martin, E.; Gascon, F. DART: a 3D model for simulating satellite images and
studying surface radiation budget. Int. J. Remote Sens. 2004,25, 73–96.
37.
Gastellu-Etchegorry, J.P.; Lauret, N.; Yin, T.; Landier, L.; Kallel, A.; Malenovsky, Z.; Bitar, A.A.; Aval, J.;
Benhmida, S.; Qi, J.; et al. DART: Recent Advances in Remote Sensing Data Modeling With Atmosphere,
Polarization, and Chlorophyll Fluorescence. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens.
2017
,10,
2640–2649.
Remote Sens. 2018,10, 1032 18 of 19
38.
Brede, B.; Bartholomeus, H.; Suomalainen, J.; Clevers, J.; Verbesselt, J.; Herold, M.; Culvenor, D.; Gascon, F.
The Speulderbos Fiducial Reference Site for Continuous Monitoring of Forest Biophysical Variables.
In Proceedings of the Living Planet Symposium 2016, Prague, Czech Republic, 9–13 May 2016; p. 5.
39.
Calders, K.; Schenkels, T.; Bartholomeus, H.; Armston, J.; Verbesselt, J.; Herold, M. Monitoring spring
phenology with high temporal resolution terrestrial LiDAR measurements. Agric. For. Meteorol.
2015,203, 158–168.
40.
Woodgate, W.; Armston, J.D.; Disney, M.; Jones, S.D.; Suarez, L.; Hill, M.J.; Wilkes, P.; Soto-Berelov, M.
Quantifying the impact of woody material on leaf area index estimation from hemispherical photography
using 3D canopy simulations. Agric. For. Meteorol. 2016,226–227, 1–12.
41.
Leblanc, S.G.; Chen, J.M.; Fernandes, R.; Deering, D.W.; Conley, A. Methodology comparison for canopy
structure parameters extraction from digital hemispherical photography in boreal forests. Agric. For. Meteorol.
2005,129, 187–207.
42.
Jupp, D.L.B.; Culvenor, D.S.; Lovell, J.L.; Newnham, G.J.; Strahler, A.H.; Woodcock, C.E. Estimating
forest LAI profiles and structural parameters using a ground-based laser called ‘Echidna’. Tree Physiol.
2009,29, 171–81.
43.
Calders, K.; Armston, J.; Newnham, G.; Herold, M.; Goodwin, N. Implications of sensor configuration and
topography on vertical plant profiles derived from terrestrial LiDAR. Agric. For. Meteorol.
2014
,194, 104–117.
44.
Lang, A.; Xiang, Y. Estimation of leaf area index from transmission of direct sunlight in discontinuous
canopies. Agric. For. Meteorol. 1986,37, 229–243.
45.
Lecerf, R.; Baret, F.; Hanocq, J.; Marloie, O.; Rautiainen, M.; Mottus, M.; Heiskanen, J.; Stenberg, P.
PASTIS 57: Autonomous Light Sensors for PAI Continuous Monitoring. Principles, Calibration and Application to
Vegetation Phenology; AGU Fall Meeting Abstracts 12/2010; 2010.
46.
Hancock, S.; Essery, R.; Reid, T.; Carle, J.; Baxter, R.; Rutter, N.; Huntley, B. Characterising forest gap
fraction with terrestrial lidar and photography: An examination of relative limitations. Agric. For. Meteorol.
2014,189–190, 105–114.
47.
Calders, K.; Origo, N.; Disney, M.; Nightingale, J.; Woodgate, W.; Armston, J.; Lewis, P. Variability
and bias in active and passive ground-based measurements of effective plant, wood and leaf area index.
Agric. For. Meteorol. 2018,252, 231–240.
48.
Vaccari, S.; van Leeuwen, M.; Calders, K.; Coops, N.C.; Herold, M. Bias in lidar-based canopy gap fraction
estimates. Remote Sens. Lett. 2013,4, 391–399.
49.
Myneni, R.; Knyazikhin, Y.; Park, T. MCD15A3H MODIS/Terra+Aqua Leaf Area Index/FPAR 4-Day L4 Global 500
m SIN Grid V006 [Data Set]; NASA EOSDIS Land Processes DAAC; dx.doi.org/10.5067/MODIS/MCD15A3H.
006, 2015.
50.
Myneni, R.; Knyazikhin, Y.; Shabanov, N. Leaf Area Index and Fraction of Absorbed PAR Products from
Terra and Aqua MODIS Sensors: Analysis, Validation, and Refinement. In Land Remote Sensing and Global
Environmental Change–NASA’s Earth Observing System and the Science of ASTER and MODIS; Ramachandran, B.,
Justice, C.O., Abrams, M.J., Eds.; Springer: New York, NY, USA; Dordrecht, The Netherlands;
Heidelberg, Germany; London, UK; 2011; Chapter 27, pp. 603–633.
51.
Bates, D.M.; Watts, D.G. Nonlinear Regression Analysis and Its Applications; John Wiley & Sons:
New York, NY, USA, 1988; p. 365.
52.
R Core Team. R: A Language and Environment for Statistical Computing; R Foundation for Statistical Computing:
Vienna, Austria, 2014.
53.
Drusch, M.; Del Bello, U.; Carlier, S.; Colin, O.; Fernandez, V.; Gascon, F.; Hoersch, B.; Isola, C.; Laberinti, P.;
Martimort, P.; et al. Sentinel-2: ESA’s Optical High-Resolution Mission for GMES Operational Services.
Remote Sens. Environ. 2012,120, 25–36.
54.
Fernandes, R.; Plummer, S.; Nightingale, J.; Baret, F.; Camacho, F.; Fang, H.; Garrigues, S.; Gobron, N.;
Lang, M.; Lacaze, R.; et al. Global Leaf Area Index Product Validation Good Practices. In Best Practice
for Satellite-Derived Land Product Validation, 2.0.1 ed.; Schaepman-Strub, G., Román, M., Nickeson, J., Eds.;
Land Product Validation Subgroup (WGCV/CEOS): 2014; p. 76.
55.
Jonckheere, I.; Nackaerts, K.; Muys, B.; Coppin, P. Assessment of automatic gap fraction estimation of forests
from digital hemispherical photography. Agric. For. Meteorol. 2005,132, 96–114.
56.
Gonsamo, A.; Pellikka, P. The computation of foliage clumping index using hemispherical photography.
Agric. For. Meteorol. 2009,149, 1781–1787.
Remote Sens. 2018,10, 1032 19 of 19
57.
Gower, S.T.; Kucharik, C.J.; Norman, J.M. Direct and indirect estimation of leaf area index, f(APAR), and net
primary production of terrestrial ecosystems. Remote Sens. Environ. 1999,70, 29–51.
58.
Kucharik, C.J.; Norman, J.M.; Murdock, L.M.; Gower, S.T. Characterizing canopy nonrandomness with a
multiband vegetation imager (MVI). J. Geophys. Res. 1997,102, 29455.
59.
Bréda, N.J.J. Ground-based measurements of leaf area index: A review of methods, instruments and current
controversies. J. Exp. Bot. 2003,54, 2403–2417.
60.
Howe, G.A.; Hewawasam, K.; Douglas, E.S.; Martel, J.; Li, Z.; Strahler, A.; Schaaf, C.; Cook, T.A.;
Chakrabarti, S. Capabilities and performance of dual-wavelength Echidna
R
lidar. J. Appl. Remote Sens.
2015,9, 095979.
61.
North, P. Three-dimensional forest light interaction model using a Monte Carlo method. IEEE Trans. Geosci.
Remote Sens. 1996,34, 946–956.
62.
Grau, E.; Durrieu, S.; Fournier, R.; Gastellu-Etchegorry, J.P.; Yin, T. Estimation of 3D vegetation density
with Terrestrial Laser Scanning data using voxels. A sensitivity analysis of influencing parameters.
Remote Sens. Environ. 2017,191, 373–388.
63.
Culvenor, D.; Newnham, G.; Mellor, A.; Sims, N.; Haywood, A. Automated In-Situ Laser Scanner for
Monitoring Forest Leaf Area Index. Sensors 2014,14, 14994–15008.
c
2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
... to reduce sensitivity to canopy leaf angle distribution. The PASTIS-57 instrument has been used to derive time-series of PAI over cropland and forest environments (Brede et al., 2018;Fang et al., 2018). Although the systems based on radiometric sensors have proven useful for low-cost automated monitoring of vegetation biophysical variables, they are subject to some limitations. ...
... Firstly, careful calibration of above-and below-canopy sensors is required to ensure accurate estimates of transmittance are acquired, and regular recalibration may be necessary due to sensor degradation, which could occur at different rates (particularly since the above-canopy sensors are subject to increased solar radiation). Additionally, variable and/or direct illumination can lead to substantial artefacts in the derived time-series (Brede et al., 2018;Fang et al., 2018). Finally, with the exception of the PASTIS-57 system, for deriving LAI/PAI, there is again a need to specify the leaf angle distribution of the canopy (Section 1.5.2.2.1). ...
... This enables a robust transfer function to be established between the in situ measurements and a single high spatial resolution image obtained at the time of the campaign. In contrast, the automated in situ measurement techniques described in Section 1.5.6.2 provide dense characterisation of the temporal dynamics of vegetation, but only cover a small number of locations within a site (Brede et al., 2018;Culvenor et al., 2014;Qu et al., 2014a;Ryu et al., 2012). ...
Thesis
Accurate and timely information on vegetation status, in the form of biophysical and biochemical variables, is key to the effective management of vegetated environments. Using optical instruments capable of resolving the spectral characteristics of vegetation, satellite-derived vegetation products now provide users routine estimates of these variables at the regional and global scale. However, to ensure their fitness-for-purpose, quality assessment is required. Unfortunately, progress in the validation of these products has been restricted by temporally limited reference data, with periodic field campaigns providing few in situ reference measurements throughout the growing season or over multiple years. Information on how product performance varies over time is, therefore, scarce (resulting in uncertainty in models of crop yield, carbon exchange, and the weather and climate systems, for which vegetation seasonality is an important driver). In recent years, several techniques have emerged with the potential to provide temporally continuous in situ reference data, automating the data collection process and overcoming the logistical issues associated with periodic field campaigns. This thesis focuses on addressing challenges associated with these techniques, including those related to data processing methods, measurement assumptions, spatial representativeness, and upscaling approaches (which are a necessity for validating moderate spatial resolution products). Of various emerging techniques, above-canopy digital repeat photography was identified as being of particular interest due to its maturity and degree of spatial integration. However, a critical appraisal of the approach revealed that due to non-linear and seasonal effects, the resulting timeseries of colour indices were prone to asymptotic saturation and could not be easily linked to any one biophysical property. To overcome these limitations, a new technique based on automated below-canopy digital hemispherical photography was proposed and evaluated. Benchmarking against manually collected data provided confidence that the approach could deliver leaf area index measurements of comparable quality to traditional in situ measurement techniques (but with substantially improved temporal characterisation). Upscaling methods were then investigated, as existing approaches are not well-suited to dense temporal characterisation of a limited number of locations. It was concluded that radiative transfer model-based approaches, which incorporate physical knowledge and enable seasonal variations in sun-sensor geometry to be accounted for, were more robust than vegetation index-based multitemporal transfer functions. Overall, the thesis provides a framework for routine quality assessment of satellitederived vegetation products, from a cost-effective automated in situ measurement technique, through to an upscaling approach capable of deriving time-series of high spatial resolution reference maps suitable for product validation. By facilitating a temporally explicit quantification of product performance in future work, the framework will enable targeted product improvements to be made, ultimately reducing uncertainties in downstream applications.
... For example, LAI has been continuously observed using radiometric sensors (e.g., PASTIS-57) [21]. However, these instruments cannot carry out clumping correction strategies because they involve single measurements [22]. In addition, downward-looking or upward-pointing digital cameras have been used to obtain the LAI based on gap fraction theory. ...
... Extensive research has been conducted on the relationship between LAI e and LAI [23,25]. The focus of this study was on the measurement of LAI e , which can be converted into LAI if the clumping index is known [6,22]. ...
Article
Full-text available
The leaf area index (LAI) is an important structural parameter of plant canopies used in terrestrial biosphere models. Optical methods are commonly used for measuring LAI due to their non-destructive nature, convenience, and rapidity. In the present study, a novel instrument, named the Automated Hemispherical Scanner (AHS), was developed to measure plant area index (PAI) for monitoring daily changes in LAI in forest ecosystems. In the AHS, an optical sensor driven by a pair of servomotors is used to observe hemispherical light transmission continuously at adjustable intervals, and a blue filter is used to reduce the multiple scattering effect of light on the measured transmission. A set of algorithms was developed to screen the direct radiation transmitted through the canopy and to compute the transmissions from the diffuse radiation at seven zenith (0–60) and seven azimuth (0–150) angles for calculating PAI. Field experiments were conducted to verify the reliability of the AHS in three forests of Northeast China against an existing instrument named the LAI-2200 Plant Canopy Analyzer. The PAI values obtained using the AHS agreed well (R2 = 0.927, root mean square error = 0.41) with those from the LAI-2200. Since both instruments use the same gap fraction theory for calculating the PAI from diffuse radiation transmissions obtained from multiple angles, the agreement of these two instruments means that the AHS can reliably measure the transmittance of diffuse radiation and the theory has been implemented correctly. Compared with LAI-2200, the AHS has the advantage of automated and continuous measurements, and therefore it is suitable for monitoring variations in PAI over extended periods, such as the whole growing season. Compared with widely used digital photographic techniques, the AHS also avoids the requirement of determining a suitable photographic exposure, which is often problematic in the field with variable sky conditions. With these advantages, the AHS could be deployed in forest growth monitoring networks.
... The described sensor device alone provides an indicator of the plant's health. By combining this device with further sensors measuring the quality of light reaching the crown [4] or the leaf (LAI; leaf area index [6]), one can additionally obtain a continuous measure of the plant's wellbeing as well as a forecast for its development. ...
Article
Measuring chlorophyll fluorescence is an important tool in plant research, since it is a reliable non-invasive method for capturing photosynthetic efficiency of a plant and, hence, an indicator of plant stress/health. The principle of chlorophyll fluorometry is based on the optical illumination of a plant’s leaf at a certain wavelength, while simultaneously measuring the emitted fluorescence light intensity at a different optical wavelength. By relating the fluorescence light energy at small and large excitation power, conclusions on the efficiency of the photosystem and, therefore, on the plant’s photosynthesis capability can be drawn. Current mobile chlorophyll fluorometers are either (i) compact and energy efficient but limited in functionality and accuracy by omitting modulated measurement signals or (ii) sophisticated and precise with respect to the measurement, but with the drawback of extended weight, size, energy consumption and cost. This contribution presents a smaller, lighter and cheaper sensor device that can be built with sufficiently low energy consumption to be powered by energy harvesting while being light enough to be attached nearly anywhere such as tree branches. With a device cost below 250 €, the performance of the developed device is similar to more expensive commercial devices considering measurements of the relative variable fluorescence. Moreover, the sensor device provides a wireless interface in the European 868 MHz SRD band with up to 10 km of range in free space while just consuming 150 µW in receiving mode due to a custom duty cycling technique.
... Traditionally, satellite images have been used for this process and these are still the method of choice for large-scale analyses (Voigt et al. 2018). However, new sensors offer comparable data for phenology (Brede et al. 2018). At a smaller scale, drones can offer advantages over satellite images, such as flexible data collection, capturing imagery below cloud cover and providing much higher resolution imagery than commercially available satellites (Rodríguez et al. 2012;Wich and Koh 2018). ...
Article
Full-text available
Observing and quantifying primate behavior in the wild is challenging. Human presence affects primate behavior and habituation of new, especially terrestrial, individuals is a time-intensive process that carries with it ethical and health concerns, especially during the recent pandemic when primates are at even greater risk than usual. As a result, wildlife researchers, including primatologists, have increasingly turned to new technologies to answer questions and provide important data related to primate conservation. Tools and methods should be chosen carefully to maximize and improve the data that will be used to answer the research questions. We review here the role of four indirect methods—camera traps, acoustic monitoring, drones, and portable field labs—and improvements in machine learning that offer rapid, reliable means of combing through large datasets that these methods generate. We describe key applications and limitations of each tool in primate conservation, and where we anticipate primate conservation technology moving forward in the coming years.
... Using the consolidated data, it should be possible to identify the most common and substantial uncertainty contributors and inform future in situ sampling and measurement protocols accordingly. In addition to further field campaigns, detailed consideration will be given to the use of permanent instrumentation [72][73][74][75][76][77][78][79]. This will include a review of site deployment considerations and an initial plan for the establishment of permanent ESAsupported FRM4VEG 'supersites'. ...
Article
Full-text available
With a wide range of satellite-derived vegetation bio-geophysical products now available to users, validation efforts are required to assess their accuracy and fitness for purpose. Substantial progress in the validation of such products has been made over the last two decades, but quantification of the uncertainties associated with in situ reference measurements is rarely performed, and the incorporation of uncertainties within upscaling procedures is cursory at best. Since current validation practices assume that reference data represent the truth, our ability to reliably demonstrate compliance with product uncertainty requirements through conformity testing is limited. The Fiducial Reference Measurements for Vegetation (FRM4VEG) project, initiated by the European Space Agency, is aiming to address this challenge by applying metrological principles to vegetation and surface reflectance product validation. Following FRM principles, and in accordance with the International Standards Organisation’s (ISO) Guide to the Expression of Uncertainty in Measurement (GUM), for the first time, we describe an end-to-end uncertainty evaluation framework for reference data of two key vegetation bio-geophysical variables: the fraction of absorbed photosynthetically active radiation (FAPAR) and canopy chlorophyll content (CCC). The process involves quantifying the uncertainties associated with individual in situ reference measurements and incorporating these uncertainties within the upscaling procedure (as well as those associated with the high-spatial-resolution imagery used for upscaling). The framework was demonstrated in two field campaigns covering agricultural crops (Las Tiesas–Barrax, Spain) and deciduous broadleaf forest (Wytham Woods, UK). Providing high-spatial-resolution reference maps with per-pixel uncertainty estimates, the framework is applicable to a range of other bio-geophysical variables including leaf area index (LAI), the fraction of vegetation cover (FCOVER), and canopy water content (CWC). The proposed procedures will facilitate conformity testing of moderate spatial resolution vegetation bio-geophysical products in future validation exercises.
... Except for the traditional direct and indirect methods, the use of several automatic instruments has increased rapidly with technological progress. For example, LAI was continuously observed using radiometric sensors (e.g. the LAINet and the PASTIS-57) (Qu et al., 2014;Brede et al., 2018;Fang et al., 2018) and digital cover photography (Ryu et al., 2012Toda and Richardson, 2018). Despite the potential of these automatic instruments, there are still some unsolved problems. ...
Article
Automatic and accurate measurements of the leaf area index (LAI) can provide information about key parameters of vegetation structure in an ecosystem. Little attention, however, has been paid to LAI dynamics for shrub canopies in both arid and semiarid regions compared to forests and crops. One of the main sources of error when measuring shrub LAI from the indirect method is the woody part. The influence of woody parts on LAI estimations can be expressed as the woody-to-total area ratio, α; there are also relatively few studies on α value of shrubs, especially the dynamic changes of α. This study used a new cost-effective tool, the fisheye webcam, to derive daily LAI and α in the fields of Caragana korshinskii and Salix psammophila on the Loess Plateau in China. The LAI data were compared with measurements acquired by an LAI-2200 Plant Canopy Analyzer and direct methods. The LAI-2200 data were consistent with the effective plant area index (PAIeff) estimated using upward-pointing webcams (UPWs) (R² = 0.89, root mean square error (RMSE) = 0.21) but not downward-pointing webcams (DPWs) (R² = 0.52, RMSE = 0.68). Shrub PAIeff derived from the DPWs was much smaller than PAIeff derived from the UPWs over the season (y = 0.48x–0.03, R² = 0.79, RMSE = 0.85). LAI derived from the direct methods was strongly linearly correlated with the PAI minus WAI obtained from upward digital hemispherical photography (R² = 0.94, RMSE = 0.62 for C. korshinskii and R² = 0.99, RMSE = 0.31 for S. psammophila). This result indicates that the upward digital hemispherical photography for deriving PAI and WAI performs good, which can be utilized to calculate LAI accurately. The woody-to-total area ratio (α) of deciduous shrub canopies was a dynamic parameter that varied throughout the season, decreasing as the canopy leaves developed and increasing when they fell. α = 1 for the leaf-off season and < 0.1 when the foliage peaked. α differed little between the two shrub canopies (R² = 0.95, RMSE = 0.07), and it could be used to convert PAI to LAI for C. korshinskii and S. psammophila. The UPWs demonstrated excellent potential for continuous LAI measurement for the two shrub canopies, but the DPWs should be further tested before use. The fisheye-webcam is inexpensive and can easily monitor LAI at field scale and can be used to evaluate LAI products derived from remote sensing.
Article
The definition of LAI (Leaf Area Index) is important when deriving it from reflectance observation for model application and validation. Canopy reflectance and the corresponding quantities of LAI, PAI (Plant Area Index), GAI (Green Area Index) and effective GAI (GAIeff) are first calculated using a 3D radiative transfer model (RTM) applied to 3D wheat and maize architecture models. A range of phenological stages, leaf optical properties, soil reflectance, canopy structure and sun directions is considered. Several retrieval methods are compared, including vegetation indices (VIs) combined with a semi-empirical model, and 1D or 3D RTM combined with a machine learning inversion approach. Results show that GAIeff is best estimated from remote sensing observations. The RTM inversion using a 3D model provides more accurate GAIeff estimates compared with VIs and the 1D PROSAIL model with RMSE = 0.33 for wheat and RMSE= 0.43 for maize. GAIeff offers the advantage to be easily accessible from ground measurements at the decametric resolution. It was therefore concluded that the most efficient retrieval approach would be to use machine learning algorithms trained over paired GAIeff and the corresponding canopy reflectance derived either from realistic 3D canopy models or from experimental measurements.
Article
Full-text available
To further progress the validation of global leaf area index (LAI) products, temporally continuous reference data is a key requirement, as periodic field campaigns fail to adequately characterise temporal dynamics. Progress in cost-effective automated measurement techniques has been made in recent years, but appropriate upscaling methodologies are less mature. Recently, the use of multitemporal transfer functions has been proposed as a potential solution. Using data collected during an independent field campaign, we evaluated the performance of both vegetation index-based multitemporal transfer functions and a radiative transfer model (RTM)-based upscaling approach. Whether assessed using cross validation or data from the independent field campaign, the RTM-based approach provided the best performance ( ${r^{2}\ge 0.88}$ , RMSE ≤ 0.41, NRMSE < 13%). For upscaling temporally continuous in situ data, the ability of RTM-based approaches to account for seasonal changes in sun-sensor geometry is a key advantage over vegetation index-based multitemporal transfer functions.
Article
Full-text available
In situ leaf area index (LAI) measurements are essential to validate widely-used large-area or global LAI products derived, indirectly, from satellite observations. Here, we compare three common and emerging ground-based sensors for rapid LAI characterisation of large areas, namely digital hemispherical photography (DHP), two versions of a widely-used commercial LAI sensor (LiCOR LAI-2000 and 2200), and terrestrial laser scanning (TLS). The comparison is conducted during leaf-on and leaf-off conditions at an unprecedented sample size in a deciduous woodland canopy. The deviation between estimates of these three ground-based instruments yields differences greater than the 5% threshold goal set by the World Meteorological Organization. The variance at sample level is reduced when aggregated to plot scale (1 ha) or site scale (6 ha). TLS shows the lowest relative standard deviation in both leaf-on (11.78%) and leaf-off (13.02%) conditions. Whereas the relative standard deviation of effective plant area index (ePAI) derived from DHP relates closely to TLS in leaf-on conditions, it is as large as 28.14–29.74% for effective wood area index (eWAI) values in leaf-off conditions depending on the thresholding technique that was used. ePAI values of TLS and LAI-2x00 agree best in leaf-on conditions with a concordance correlation coefficient (CCC) of 0.796. In leaf-off conditions, eWAI values derived from DHP with Ridler and Calvard thresholding agrees best with TLS. Sample size analysis using Monte Carlo bootstrapping shows that TLS requires the fewest samples to achieve a precision better than 5% for the mean ± standard deviation. We therefore support earlier studies that suggest that TLS measurements are preferential to measurements from instruments that are dependent on specific illumination conditions. A key issue with validation of indirect estimates of LAI is that the true values are not known. Since we cannot know the true values of LAI, we cannot quantify the accuracy of the measurements. Our radiative transfer simulations show that ePAI estimates are, on average, 27% higher than eLAI estimates. Linear regression indicated a linear relationship between eLAI and ePAI–eWAI (R² = 0.87), with an intercept of 0.552 and suggests that caution is required when using LAI estimates.
Article
Full-text available
In addition to single-angle reflectance data, multi-angular observations can be used as an additional information source for the retrieval of properties of an observed target surface. In this paper, we studied the potential of multi-angular reflectance data for the improvement of leaf area index (LAI) and leaf chlorophyll content (LCC) estimation by numerical inversion of the PROSAIL model. The potential for improvement of LAI and LCC was evaluated for both measured data and simulated data. The measured data was collected on 19 July 2016 by a frame-camera mounted on an unmanned aerial vehicle (UAV) over a potato field, where eight experimental plots of 30 × 30 m were designed with different fertilization levels. Dozens of viewing angles, covering the hemisphere up to around 30° from nadir, were obtained by a large forward and sideways overlap of collected images. Simultaneously to the UAV flight, in situ measurements of LAI and LCC were performed. Inversion of the PROSAIL model was done based on nadir data and based on multi-angular data collected by the UAV. Inversion based on the multi-angular data performed slightly better than inversion based on nadir data, indicated by the decrease in RMSE from 0.70 to 0.65 m 2 /m 2 for the estimation of LAI, and from 17.35 to 17.29 μg/cm 2 for the estimation of LCC, when nadir data were used and when multi-angular data were used, respectively. In addition to inversions based on measured data, we simulated several datasets at different multi-angular configurations and compared the accuracy of the inversions of these datasets with the inversion based on data simulated at nadir position. In general, the results based on simulated (synthetic) data indicated that when more viewing angles, more well distributed viewing angles, and viewing angles up to larger zenith angles were available for inversion, the most accurate estimations were obtained. Interestingly, when using spectra simulated at multi-angular sampling configurations as were captured by the UAV platform (view zenith angles up to 30°), already a huge improvement could be obtained when compared to solely using spectra simulated at nadir position. The results of this study show that the estimation of LAI and LCC by numerical inversion of the PROSAIL model can be improved when multi-angular observations are introduced. However, for the potato crop, PROSAIL inversion for measured data only showed moderate accuracy and slight improvements.
Article
Full-text available
Near surface (i.e., camera) and satellite remote sensing metrics have become widely used indicators of plant growing seasons. While robust linkages have been established between field metrics and ecosystem exchange in many land cover types, assessment of how well remotely-derived season start and end dates depict field conditions in arid ecosystems remain unknown. We evaluated the correspondence between field measures of start (SOS; leaves unfolded and canopy greenness >0) and end of season (EOS) and canopy greenness for two widespread species in southwestern U.S. ecosystems with those metrics estimated from near-surface cameras and MODIS NDVI for five years (2012–2016). Using Timesat software to estimate SOS and EOS from the phenocam green chromatic coordinate (GCC) greenness index resulted in good agreement with ground observations for honey mesquite but not black grama. Despite differences in the detectability of SOS and EOS for the two species, GCC was significantly correlated with field estimates of canopy greenness for both species throughout the growing season. MODIS NDVI for this arid grassland site was driven by the black grama signal although a mesquite signal was discernable in average rainfall years. Our findings suggest phenocams could help meet myriad needs in natural resource management.
Article
Automatic leaf area index (LAI) measurements are important for obtaining sufficient amounts of field data over an extended period of time. A seasonal field campaign was carried out to obtain continuous LAI measurements over maize, soybean, and sorghum fields in northeast China in 2016. Field LAI measurements were acquired with the automatic PASTIS-57 (PAI Autonomous System from Transmittance Instantaneous Sensed from 57°) instrument and two smartphone applications, PocketLAI and LAISmart. These measurements were compared with data obtained using the LAI-2200 Plant Canopy Analyzer, digital hemispherical photography (DHP), and destructive sampling measurements. The effective plant area index (PAIeff) estimates from LAI-2200 and DHP are consistent over the season, with the overall relative errors (RE) of less than 5%. The PASTIS-57 data exhibit a small underestimation of the LAI-2200 and DHP values (RE < 20%). The relative errors for the LAISmart data are between −20% and −30%. PocketLAI significantly underestimates the LAI-2200 values (RE > 40%) and saturates at around PAIeff = 3.5. The canopy clumping index (CI) exhibits an S-shaped seasonal variation that decreases with the increase of PAIeff during the vegetative growth stage but increases after this stage. PASTIS-57 shows great potential for obtaining continuous LAI measurements in agricultural crop fields, but the smartphone applications should be further examined before they can be used for research purposes. The data collected in this study are valuable for the validation of remote sensing products.
Article
Gap fraction measurements with digital a hemispherical camera were carried out in 2007–2016 in a Scots pine (Pinus sylvestris L.) stand growing on transitional Sphagnum bog in Järvselja, Estonia. The measurements were done at the time when (1) the pines had maximum foliage, or (2) after autumn needle fall or before bud burst. Data for gap fraction calculation were extracted from camera raw files. For data recording, the sensor output signal maximum was kept according to the image brightness histogram within the interquartile range of the sensor dynamic range. A linear conversion method (LinearRatio) was used for image processing. The relative needle loss in autumn (20–30%) increased gap fraction by 0.02–0.05 at view zenith angles from 10° to 60° and decreased the mean plant area index from 1.86 to 1.68 after autumn litter fall. The measurement results were in good agreement with a theoretical gap fraction model simulation and with gap fraction estimates made from cover photos. The measurement and data processing protocol is appropriate for long-term monitoring in permanent sample plots.
Article
The availability of new fAPAR satellite products requires simultaneous efforts in validation to provide users with a better comprehension of product performance and evaluation of uncertainties. This study aimed to validate three fAPAR satellite products, GEOV1, MODIS C5, and MODIS C6, against ground references to determine to what extent the GCOS requirements on accuracy (maximum 10% or 5%) can be met in a deciduous beech forest site in a gently and variably sloped mountain site. Three ground reference fAPAR, differing for temporal (continuous or campaign mode) and spatial sampling (single points or Elementary Sampling Units-ESUs), were collected using different devices: (1) Apogee (defined as benchmark in this study); (2) PASTIS; and (3) Digital cameras for collecting hemispherical photographs (DHP). A bottom-up approach for the upscaling process was used in the present study. Radiometric values of decametric images (Landsat-8) were extracted over the ESUs and used to develop empirical transfer functions for upscaling the ground measurements. The resulting high-resolution ground-based maps were aggregated to the spatial resolution of the satellite product to be validated considering the equivalent point spread function of the satellite sensors, and a correlation analysis was performed to accomplish the accuracy assessment. PASTIS sensors showed good performance as fAPARPASTIS appropriately followed the seasonal trends depicted by fAPARAPOGEE (benchmark) (R² = 0.84; RMSE = 0.01). Despite small dissimilarities, mainly attributed to different sampling schemes and errors in DHP classification process, the agreement between fAPARPASTIS and fAPARDHP was noticeable considering all the differences between both approaches. The temporal courses of the three satellite products were found to be consistent with both Apogee and PASTIS, except at the end of the summer season when ground data were more affected by senescent leaves, with both MODIS C5 and C6 displaying larger short-term variability due to their shorter temporal composite period. MODIS C5 and C6 retrievals were obtained with the backup algorithm in most cases. The three green fAPAR satellite products under study showed good agreement with ground-based maps of canopy fAPAR at 10 h, with RMSE values lower than 0.06, very low systematic differences, and more than 85% of the pixels within GCOS requirements. Among them, GEOV1 fAPAR showed up to 98% of the points lying within the GCOS requirements, and slightly lower values (mean bias = -0.02) as compared with the ground canopy fAPAR, which is expected to be only slightly higher than green fAPAR in the peak season.
Article
The 3D distribution of plant material is a key parameter to describe vegetation structure, which influences several processes such as radiation interception and ecosystem functioning. Vegetation covers are often described using Leaf Area Index (LAI) or Plant Area Index (PAI) for monitoring or modeling purposes. Characterizing vegetation 3D structure at fine scale is increasingly required, notably in order to be able to apply radiative transfer simulations at scales consistent with the spatial resolution of recent remote sensing sensors. To assess 3D PAI of a vegetation plot, this paper evaluates the potential of a voxelization method using Terrestrial Laser Scanning (TLS) data, based on the Beer-Lambert transmittance computation law. The theoretical validation was performed using a simulation framework based on a radiative transfer model (DART). The framework allowed simulating TLS acquisition on a theoretical distribution of leaves and a realistic representation (single tree), for which all characteristics are well known. Hence, a sensitivity analysis was performed to study the influence of instrument parameters (i.e. single- or multi-echo, beam divergence), scanning configuration (scan angle step), vegetation characteristics (leaf size and density, leaf angle distribution), and voxel parameters (cubic versus spherical geometry, at different resolutions, with and without occlusion) on the estimation of PAI. For a theoretical distribution of leaves, results showed good accuracy of the voxelization method (R² = 0.91 and RMSE = 20% for a mean case, at voxel level) with a high resolution multi-echo TLS scan, cubic voxels over 0.5-m resolution, low inter-voxel occlusion, small leaves, and up to a surface density of 2 m².m −3. Error increased with a larger scan angle resolution, single echo TLS systems, and vegetation density. Also, without clumping, error increased with smaller voxels or larger leaves. Best results were obtained with multi-echo TLS scans (angular resolution of 0.05°), cubic voxels at 1-m resolution when occlusion is low (voxel sampling higher than 50% of maximum sampling at 15 m) and small leaves (e.g. 10 cm²), which provided very good agreement (RMSD = 7.6%, R² = 0.98, p = 0.99). On a realistic isolated tree, PAI was correctly assessed with cubic voxels at 0.25 m resolution. A method to merge voxelized scans was proposed to deal with inter-voxel occlusion effects.
Article
This paper presents an operational chain for high-resolution leaf area index (LAI) retrieval from multiresolution satellite data specifically developed for Mediterranean rice areas. The proposed methodology is based on the inversion of the PROSAIL radiative transfer model through the state-of-the-art nonlinear Gaussian process regression (GPR) method. Landsat and SPOT5 data were used for multitemporal LAI retrievals at high-resolution. LAI estimates were validated using time series of in situ LAI measurements collected during the rice season in Spain and Italy. Ground LAI data were collected with smartphones using PocketLAI, a specific phone application for LAI estimation. Temporal evolution of the LAI estimates using Landsat and SPOT5 data followed consistently the temporal evolution of the in situ LAI measurements acquired on several Mediterranean rice varieties. The estimates had a root-mean-square-error (RMSE) of 0.39 and 0.51 m2/m2 in Spain and 0.38 and 0.47 m2/m2 in Italy for Landsat and SPOT5 respectively, with a strong correlation (R2 > 0.92) for both cases. Spatial-temporal assessment of the estimated LAI from Landsat and SPOT5 data confirmed the robustness and consistency of the retrieval chain. This paper demonstrates the importance of an adequate characterization of the underlying rice background in order to address changes in background condition related to water management. Results highlight the potential of the proposed chain for deriving multitemporal near real-time decametric LAI maps fundamental for operational rice crop monitoring, and demonstrate the readiness of the proposed method for the processing of data such as the recently launched Sentinel-2.