Seasonal variation in the canopy color of temperate evergreen conifer forests

Abstract

Evergreen conifer forests are the most prevalent land cover type in North America. Seasonal changes in the color of evergreen forest canopies have been documented with near‐surface remote sensing, but the physiological mechanisms underlying these changes, and the implications for photosynthetic uptake, have not been fully elucidated.
Here, we integrate on‐the‐ground phenological observations, leaf‐level physiological measurements, near surface hyperspectral remote sensing and digital camera imagery, tower‐based CO2 flux measurements, and a predictive model to simulate seasonal canopy color dynamics.
We show that seasonal changes in canopy color occur independently of new leaf production, but track changes in chlorophyll fluorescence, the photochemical reflectance index, and leaf pigmentation. We demonstrate that at winter‐dormant sites, seasonal changes in canopy color can be used to predict the onset of canopy‐level photosynthesis in spring, and its cessation in autumn. Finally, we parameterize a simple temperature‐based model to predict the seasonal cycle of canopy greenness, and we show that the model successfully simulates interannual variation in the timing of changes in canopy color.
These results provide mechanistic insight into the factors driving seasonal changes in evergreen canopy color and provide opportunities to monitor and model seasonal variation in photosynthetic activity using color‐based vegetation indices.

Full text

Seasonal variation in the canopy color of temperate evergreen
conifer forests
Bijan Seyednasrollah
1,2
, David R. Bowling
3
, Rui Cheng
4,5
, Barry A. Logan
6
, Troy S. Magney
7
,
Christian Frankenberg
4,5
, Julia C. Yang
3
, Adam M. Young
1,2
, Koen Hufkens
8,9
, M. Altaf Arain
10
,
T. Andrew Black
11
, Peter D. Blanken
12
, Rosvel Bracho
13
, Rachhpal Jassal
11
, David Y. Hollinger
14
,
Beverly E. Law
15
, Zoran Nesic
11
and Andrew D. Richardson
1,2
1
School of Informatics, Computing & Cyber Systems, Northern Arizona University, Flagstaff, AZ 86011, USA;
2
Center for Ecosystem Science and Society, Northern Arizona University,
Flagstaff, AZ 86011, USA;
3
School of Biological Sciences, University of Utah, Salt Lake City, UT 84112, USA;
4
Division of Geological and Planetary Sciences, California Institute of
Technology, Pasadena, CA 91125, USA;
5
Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA 91125, USA;
6
Department of Biology, Bowdoin College, Brunswick, ME
04011, USA;
7
Department of Plant Sciences, University of California Davis, Davis, CA 95616, USA;
8
Computational & Applied Vegetation Ecology Lab, Ghent University, Ghent 9000,
Belgium;
9
INRA, UMR ISPA, Villenave d’Ornon 75011, France;
10
School of Earth, Environment and Society and McMaster Center for Climate Change, McMaster University, Hamilton, ON
L8S 4K1, Canada;
11
Faculty of Land and Food Systems, University of British Columbia, Vancouver, BC V6T 1Z4, Canada;
12
Department of Geography, University of Colorado, Boulder, CO
80309, USA;
13
School of Forest Resources and Conservation, University of Florida, Gainesville, FL 32611, USA;
14
Northern Research Station, USDA Forest Service, Durham, NH 03824,
USA;
15
College of Forestry, Oregon State University, Corvallis, OR 97330, USA
Author for correspondence:
Bijan Seyednasrollah
Email: bijan.s.nasr@gmail.com
Received: 23 July 2020
Accepted: 13 October 2020
New Phytologist (2021) 229: 2586–2600
doi: 10.1111/nph.17046
Key words: AmeriFlux, evergreen conifer,
PhenoCam, phenology, PRI, seasonality,
xanthophyll.
Summary
Evergreen conifer forests are the most prevalent land cover type in North America. Seasonal
changes in the color of evergreen forest canopies have been documented with near-surface
remote sensing, but the physiological mechanisms underlying these changes, and the implica-
tions for photosynthetic uptake, have not been fully elucidated.
Here, we integrate on-the-ground phenological observations, leaf-level physiological mea-
surements, near surface hyperspectral remote sensing and digital camera imagery, tower-
based CO
2
flux measurements, and a predictive model to simulate seasonal canopy color
dynamics.
We show that seasonal changes in canopy color occur independently of new leaf produc-
tion, but track changes in chlorophyll fluorescence, the photochemical reflectance index, and
leaf pigmentation. We demonstrate that at winter-dormant sites, seasonal changes in canopy
color can be used to predict the onset of canopy-level photosynthesis in spring, and its cessa-
tion in autumn. Finally, we parameterize a simple temperature-based model to predict the
seasonal cycle of canopy greenness, and we show that the model successfully simulates inter-
annual variation in the timing of changes in canopy color.
These results provide mechanistic insight into the factors driving seasonal changes in ever-
green canopy color and provide opportunities to monitor and model seasonal variation in pho-
tosynthetic activity using color-based vegetation indices.
Introduction
Evergreen conifer forests are the dominant land cover type
across much of the mid-latitudes of North America (Pielou,
2011). These forests account for a substantial portion of con-
tinental-scale CO
2
uptake and carbon sequestration, influence
water cycling and the ecohydrology of some of the continent’s
largest watersheds, and have major effects on land–atmosphere
interactions that are relevant in the context of the global
Earth system (Bonan, 2008). Therefore, they play a critical
role in the response of the terrestrial biosphere to environ-
mental change.
In cold climates, conifer forests are photosynthetically dormant
during the winter months, despite maintaining foliage year-
round (H¨
anninen, 2016). Downregulation of photosynthetic
capacity occurs in response to shorter and cooler days in autumn
(Hollinger et al., 1999; Way & Montgomery, 2015; Gamon
et al., 2016; Bowling et al., 2018). Likewise, upregulation of pho-
tosynthetic capacity is observed to occur in spring in response to
increasing temperature (Tanja et al., 2003; Richardson et al.,
2009b). However, incident solar radiation at temperate latitudes
remains relatively high during winter, and thus the photosyn-
thetic apparatus must be protected from excess absorbed energy
to prevent damage (Demmig-Adams & Adams, 1996). Conifers
2586 New Phytologist (2021) 229: 2586–2600 Ó2020 The Authors
New Phytologist Ó2020 New Phytologist Foundation
www.newphytologist.com
This is an open access article under the terms of the Creative Commons Attribution License, which permits use,
distribution and reproduction in any medium, provided the original work is properly cited.
Research

achieve winter photoprotection through thermal dissipation of
excess absorbed excitation energy (Mı´guez et al., 2015). As
reviewed by Verhoeven (2014) and Ensminger et al. (2006), this
appears to involve both rearrangement of the photosynthetic
apparatus—with the light harvesting centers essentially being
converted into ‘energy dissipating centers’—and upregulation of
the xanthophyll cycle.
Conventional remote sensing indices, which are based on the
red-edge transition between low visible-wavelength reflectance
and high near infrared-wavelength reflectance (e.g. normalized
difference vegetation index or NDVI; Tucker, 1979) typically fail
to robustly detect seasonal changes in the carotenoid pigment
pool (Magney et al., 2019) or photosynthetic activity (Balzarolo
et al., 2016; Gamon et al., 2016) in evergreen conifer forests.
Intriguingly, visible-wavelength indices may offer improved
potential for remote sensing of the seasonality of evergreen
conifer photosynthesis (Wong & Gamon, 2015a). Gamon et al.
(2016) presented a ‘chlorophyll/carotenoid index’ (CCI), calcu-
lated from bands 11 (531 nm) and 1 (645 nm) on NASA’s
MODerate resolution Imaging Spectroradiometer (MODIS)
satellite sensors. CCI performed better than NDVI at characteriz-
ing the seasonality of daily gross photosynthesis (or gross primary
productivity, GPP, estimated from eddy covariance CO
2
flux
measurements) across three evergreen sites.
Other studies have shown that visible-wavelength indices
derived from digital camera (or ‘phenocam’; Richardson, 2019)
imagery offer similar potential (Toomey et al., 2015), although it
is uncertain whether any single color-based index can robustly
track seasonal changes in photosynthetic activity across different
evergreen species mixtures, ecosystem types, or climate regimes.
Richardson et al. (2009a) reported that a ‘green excess’ index
mimicked the seasonal pattern of GPP at the conifer-dominated
Howland Forest. Results from the Niwot Ridge subalpine forest
(Colorado, USA) suggested that seasonal variation in the light-
saturated rate of canopy photosynthesis, a proxy for photosyn-
thetic capacity, was better correlated (r=0.92 vs r=0.76,
respectively) with an index known as the green chromatic coordi-
nate (G
cc
) than with the green–red vegetation index (GRVI), an
approximation to CCI (Bowling et al., 2018). Most recently, Liu
et al. (2020), reported that start-of season (SOS) and end-of-sea-
son (EOS) transition dates derived from the red chromatic coor-
dinate (R
cc
) better predicted seasonal transitions in evergreen
conifer photosynthesis than transition dates derived from G
cc
.
However, the seasonal trajectory of R
cc
did not itself align with
the seasonal trajectory of GPP, and day-to-day variation in R
cc
was much higher than G
cc
. On top of uncertainties about which
is the best index to use, mechanistic studies to understand the
underlying physiological basis for the observed seasonal variation
in canopy color have been lacking.
Upregulation of photosynthetic capacity in spring has been
modeled as a function of air temperature (Tanja et al., 2003;
Richardson et al., 2009b, 2019), and observational studies point
to temperature as the key factor driving concurrent seasonal
changes in evergreen conifer canopy color and photosynthesis
(e.g. Bowling et al., 2018). Experimental warming has been
shown to advance spring ‘green-up’ transition dates, and delay
autumn ‘green-down’ transition dates (both derived from G
cc
)
for mature black spruce trees (Richardson et al., 2018b). To date,
the potential to model the seasonal trajectory of evergreen conifer
canopy color as a function of temperature has not yet been inves-
tigated.
Here, using data from evergreen conifer sites across the
PhenoCam Network, we explore the relationships between sea-
sonal variation in canopy color and photosynthetic activity. We
begin by using leaf-level field data from three winter-dormant
sites (Howland Forest, Harvard Forest, and Niwot Ridge) to
investigate the physiological mechanisms associated with seasonal
changes in canopy color. Next, using data from 11 sites in the
AmeriFlux and FLUXNET-Canada networks, we conduct an
analysis that links seasonal variation in G
cc
and GRVI to the sea-
sonality of GPP at the seven sites that are winter-dormant.
Finally, using a multi-year time series of G
cc
from 26 evergreen
PhenoCam sites, we parameterize and test a temperature-based
model that simulates the seasonal trajectory of G
cc
, from winter
dormancy to the summertime peak in activity and back to winter
dormancy.
Materials and Methods
PhenoCam data
The PhenoCam Network uses digital repeat photography to track
seasonal changes in canopy color at over 600 sites across North
America and around the world (Richardson, 2019). Over 100
sites in the network include evergreen conifer vegetation, and the
most recent curated PhenoCam data release (v.2; Seyednasrollah
et al., 2019b) includes 265 site-years of data for evergreen conifer
sites. The PhenoCam processing workflow is fully described in
previous publications (Richardson et al., 2018a; Seyednasrollah
et al., 2019a); we summarize the key steps here.
Imagery used in the current study was obtained from cam-
eras (NetCam SC IR; StarDot Technologies, Buena Park,
CA, USA) deployed (see http://phenocam.sr.unh.edu/) and
configured (https://khufkens.github.io/phenocam-installation-
tool/) using a common protocol. Images were recorded from
04:00 to 22:00 (local time; images captured when the sun
was less than 5°above the horizon were excluded), uploaded
to the PhenoCam server, and processed to yield quantitative
information about canopy color for a predefined region of
interest (ROI) corresponding to the vegetation under study.
We used only ROIs specific to evergreen needleleaf (EN) veg-
etation, while excluding exposed ground and other vegetation
types, in this analysis.
For each image, the mean intensity (across the ROI) of the red,
green, and blue (RGB) color channels was determined, and from
these 8-bit digital numbers (R
DN
,G
DN
,B
DN
) we calculated the
G
cc
and GRVI (Richardson et al., 2013b; Bowling et al., 2018)
as:
Gcc ¼GDN
RDN þGDN þBDN
Eqn 1
Ó2020 The Authors
New Phytologist Ó2020 New Phytologist Foundation
New Phytologist (2021) 229: 2586–2600
www.newphytologist.com
New
Phytologist Research 2587

GRVI ¼GDN RDN
GDN þRDN
Eqn 2
Following Richardson et al. (2018b), we visually inspected the
imagery from each camera, and identified days on which the
canopy was at least partially obscured by snow at midday. These
days were excluded from analysis. To further minimize the effects
of variation in lighting conditions (e.g. clouds, aerosols, and illu-
mination geometry), we aggregated data from multiple images
recorded each day to composite products at a 1- and 3-d time
step using the quantile-based approach described by Sonnentag
et al. (2012). We then used a LOESS-based method to smooth
the G
cc
and GRVI time series, identify outliers, and extracted sea-
sonal transition dates based on predefined thresholds (10%, 25%,
and 50%) of the seasonal amplitude (Richardson et al., 2018a;
Seyednasrollah et al., 2019a).
All PhenoCam data used here (Fig. 1) are publicly available
from the PhenoCam dataset v.2.0 (Seyednasrollah et al., 2019a)
which is available for download from the ORNL DAAC (Seyed-
nasrollah et al., 2019b).
Leaf-level observations and measurements
Howland Forest The Howland Forest site (US-Ho1, 45.2°N,
68.7°W, elevation 60 m above sea level (asl) is located in a boreal-
northern hardwood transition forest about 50 km north of Ban-
gor, ME, USA. Forest composition is dominated by the evergreen
conifers, red spruce (Picea rubens Sarg., 44% of basal area) and
eastern hemlock (Tsuga canadensis (L.) Carri`
ere, 26% of basal
area). Mean annual temperature (MAT) is 5.3°C and mean annual
precipitation (MAP) is 1070 mm. Phenological observations (bud-
burst) have been conducted at Howland Forest since 1990 and are
ongoing. The mean aggregate date of budburst by the dominant
coniferous species was estimated from visual observations made
during weekly or twice-weekly site visits (Richardson et al.,
2009b). A phenocam was installed at Howland in 2007.
Harvard Forest The Harvard Forest (US-Ha1, 42.5°N,
72.2°W, elevation 340 m asl) is a mixed temperate forest about
110 km west of Boston, MA, USA. Forest composition is domi-
nated by the deciduous species red oak (Quercus rubra L., 36% of
basal area) and red maple (Acer rubrum L., 22% of basal area),
although mixed and pure stands of evergreen conifer species are
found throughout the forest, for example eastern hemlock (13%
of basal area) and white pine (Pinus strobus L., 6% of basal area).
MAT is 6.6°C and MAP is 1070 mm. Field observations of phe-
nology have been conducted at Harvard Forest since 1990
(Richardson & O’Keefe, 2009) and are ongoing, although obser-
vations of budburst for both white pine and eastern hemlock
were discontinued in 2001. Phenocams were installed on the
Hemlock tower (forest dominated by eastern hemlock) in 2010
and the Barn tower (mixed forest dominated by red oak and
white pine) in 2011.
Leaf-level physiological studies of white pine and eastern hem-
lock were initiated at Harvard Forest in mid-winter 2015 and
continued through late-winter 2016. Needle samples were col-
lected from three trees of each species in the vicinity of the Hem-
lock and Barn towers, monthly during the dormant season and
every 2 weeks during spring and autumn. The chlorophyll fluo-
rescence ratio F
v
/F
m
was measured using a hand-held Opti-
sciences OS-30p fluorometer (Opti-Sciences, Hudson, NH,
USA) on needles (n=20) that had been dark acclimated for a
minimum of 15 min. Spectral reflectance (350–2500 nm) was
measured using a FieldSpec 3 portable spectrometer (Analytical
Spectral Devices, Boulder, CO, USA). The spectrometer was
connected to a needle leaf clip (PP Systems, Amesbury, MA,
USA), with illumination provided by a quartz tungsten halogen
light source (Thorlabs SLS201L), referenced to a white Spec-
tralon (Labsphere, North Sutton, NH, USA) standard. We calcu-
lated the photochemical reflectance index (PRI; Gamon et al.,
1992, 1997) using reflectance at 531 and 570 nm (ρ
531
and ρ
570
,
respectively) as:
PRI ¼ρ531 ρ570
ρ531 þρ570
Eqn 3
Niwot Ridge The subalpine Niwot Ridge site (US-NR1,
40.0°N, 105.6°W, elevation 3050 m asl) is located about 25 km
west of Boulder, CO, USA. Forest composition is dominated by
the evergreen conifer species: subalpine fir (Abies
lasiocarpa (Hook.) Nutt., 46% stem density), Engelmann spruce
(Picea engelmannii Parry ex Engelm., 28% stem density), and
lodgepole pine (Pinus contorta Dougl. ex. Loud., 26% stem den-
sity). MAT is 1.5°C and MAP is 800 mm, of which 65% typi-
cally falls as snow. A phenocam was first installed at Niwot Ridge
in 2009.
We used leaf pigment data from Magney et al. (2019) to inves-
tigate relationships between changes in canopy color and leaf-
level pigment pools and ratios. Approximately monthly from
June 2017 to June 2018, samples were collected from two lodge-
pole pine and three Engelmann spruce trees growing in close
proximity to the tower, and flash-frozen in liquid nitrogen. Pig-
ments were extracted in acetone and analyzed by high-perfor-
mance liquid chromatography (HPLC; Bowling et al., 2018).
Pigments measured included Chlaand Chlb, violaxanthin (V),
antheraxanthin (A), zeaxanthin (Z), neoxanthin, lutein, and α-
and β-carotene.
Three of the trees sampled for pigments (two lodgepole
pines and one Engelmann spruce) were in the field of view
of the niwot5 phenocam. We developed custom ROI masks
and extracted canopy color information for each tree on the
day on which pigment sampling occurred. We used these data
to assess the linear correlations between color indices (G
cc
and
GRVI) and pigment contents (total chlorophyll, total
carotenoids, and total xanthophyll cycle pigments), their ratios
(chl : car, chl : xan), and xanthophyll cycle epoxidation state
([Z +A]/[V +Z+A]).
For further insight into relationships between pigments and
seasonal changes in canopy color, we used hyperspectral
reflectance measured from a two-dimensional (2D) scanning
New Phytologist (2021) 229: 2586–2600 Ó2020 The Authors
New Phytologist Ó2020 New Phytologist Foundation
www.newphytologist.com
Research
New
Phytologist
2588

telescope, PhotoSpec (Grossmann et al., 2018), mounted on
top of the Niwot Ridge AmeriFlux tower (Cheng et al.,
2020). PhotoSpec measures canopy reflectance (ρ
λ
) from 400
nm to 900 nm (1.2 nm full-width-half-maximum) with a
Flame-S spectrometer (Ocean Optics Inc., Dunedin, FL,
USA). The field of view of PhotoSpec is 0.7°. The narrow
field of view allowed us to point the instrument at the indi-
vidual trees that were sampled for pigment analysis. From the
PhotoSpec data, we calculated a full suite of normalized dif-
ference reflectance indices (NDIs), using all possible combina-
tions for ρ
λ1
and ρ
λ2
, as in Eqn 4.
NDI λ1,λ2
ðÞ¼
ρλ1ρλ2
ρλ1þρλ2
Eqn 4
Eddy covariance measurements of forest–atmosphere
exchange
To place seasonal changes in canopy color in the context of the
seasonality of canopy photosynthesis, we used tower-based eddy
covariance measurements of forest–atmosphere CO
2
fluxes (Bal-
docchi, 2008). We identified 11 evergreen sites in North America
that had multiple years of flux data with overlapping PhenoCam
data (Supporting Information Table S1). The eddy covariance
measurements (net ecosystem exchange (NEE) of CO
2
,inµmol
m
−2
s
−1
) were obtained from the AmeriFlux (sites: US-Ho1, US-
NR1, US-Me2, US-Me6, CA-TP1, CA-TP3, CA-TP4, and CA-
Obs) and FLUXNET2015 (CA-Qfo) archives, or directly from
site PI (Ca-Obs (most recent data), Ca-Ca3, and US-SP1).
Our workflow for processing the CO
2
flux measurements is
illustrated in Fig. S1. The flux data were evaluated to determine
an appropriate friction velocity threshold (u*) to remove periods
of low turbulence, and flux and meteorological data were then
gap-filled using the REDDYPROC package in R (Wutzler et al.,
2018).
Gross primary productivity (GPP) was derived following
Hollinger et al. (2004) by first estimating ecosystem respiration
(R
eco
) from the average nighttime NEE over a moving 5-d win-
dow, and then subtracting R
eco
from NEE. The light response of
GPP was characterized using a Michaelis–Menten approach (Eqn
5):
GPP ¼Amax PPFD=KmþPPFDðÞ Eqn 5
where PPFD is photosynthetic photon flux density (in µmol
m
−2
s
−1
), A
max
is the theoretical maximum rate of photosynthe-
sis, and K
m
is the half-saturation constant; A
max
and K
m
were fit
parameters. The canopy-level photosynthetic capacity index
(Bowling et al., 2018), GPP
sat
, was calculated by evaluating the
fit to Eqn 5 within each moving window, assuming full-sun con-
ditions (PPFD =2000 µmol m
−2
s
−1
).
Fig. 1 Distribution of the study sites used in the present analysis. PhenoCam site-year counts, used in the modeling analysis, are aggregated to 1°×1°
grids.
Ó2020 The Authors
New Phytologist Ó2020 New Phytologist Foundation
New Phytologist (2021) 229: 2586–2600
www.newphytologist.com
New
Phytologist Research 2589

The GPP
sat
time series were used to determine the start and
end of the photosynthetically active season (SOS and EOS,
respectively, see Fig. S1). For each year, the time series of daily
GPP
sat
was smoothed with a LOESS filter, from which the
annual maximum GPP
sat
was determined. At winter dormant
sites, the annual minimum of GPP
sat
was zero (Fig. S1). Similar
to how transition dates were determined from PhenoCam data,
the GPP-based SOS transition dates were determined as the day
that the smoothed GPP
sat
crossed the thresholds of 10, 25 and
50% (SOS10, SOS25, and SOS50) of the annual maximum
GPP
sat
. EOS transition dates were similarly determined during
the decrease of GPP
sat
in autumn. Uncertainties on these transi-
tion dates were estimated by bootstrapping.
We evaluated the agreement between PhenoCam-based transi-
tion dates and GPP-based transition dates using Type II regres-
sion and correlation analysis. Because there are various ways to
statistically aggregate (e.g. mean, median, 75
th
and 90
th
quantile)
the G
cc
and GRVI data from multiple images to a single 3-d
index value, and then to extract SOS and EOS transition dates
from 3-d data using different thresholds (10%, 25%, and 50%)
of the seasonal amplitude (Richardson et al., 2018a), we evalu-
ated all 12 possible combinations of these approaches for both
G
cc
and GRVI (Table S3, see later). We report results here for
only one method each for G
cc
and GRVI. We selected the ‘best’
method by identifying, for each index, the method that had the
highest aggregate correlation (sum of SOS and EOS
PhenoCam–GPP correlation coefficients) across all site-years of
data. For G
cc
, 3-d values were calculated as the 90
th
quantile of
the index, calculated across all daytime images, and transition
dates were determined based on the 25% threshold of the sea-
sonal amplitude. For GRVI, 3-d values were calculated as the
mean value of the index, across all daytime images, and transition
dates were determined based on the 50% threshold of the sea-
sonal amplitude.
Modeling the seasonal trajectory of canopy color
We developed a temperature-driven model to simulate the sea-
sonal trajectory of canopy greenness (cf. directly modeling SOS
and EOS transition dates, e.g. Melaas et al., 2016; Richardson
et al., 2019b), whereby day-to-day changes in greenness are
driven by the preceding day’s maximum (spring) or minimum
(autumn) air temperature. The model was inspired by formula-
tions previously presented to simulate the seasonality of develop-
mental processes (H¨
anninen & Kramer, 2007; H¨
anninen, 2016),
and seasonal upregulation and downregulation of photosynthetic
capacity (Richardson et al., 2010b) of winter-dormant conifers.
Our Bayesian framework (Eqns 6–8) accommodates missing data
and uncertainty in the observations, which makes it powerful for
statistical inference from phenological time series.
Gccs,t∼Nðγs,t,σ2ÞEqn 6
γs,t¼min max γs,t1þΔγs,t,γs,min

,γs,max

Eqn 7
Δγs,t¼max Tmax,s,tθs,1 ,0ðÞρs,1t<D
min Tmin,s,tθs,2 ,0ðÞρs,2t≥D

Eqn 8
Briefly, canopy greenness, G
cc
(Eqn 6), for site sat time tis
assumed to be distributed by a Gaussian distribution (N) with
mean γs,tand variance σ2:The state variable, γs,t(Eqn 7), tracks
daily changes in G
cc
and is constrained to fall within the range
defined by γ
s,min
and γ
s,max
, which are estimated parameters. The
daily change in that state variable, Δγs,t(Eqn 8), is prescribed as a
function of air temperature. In spring, the increase in G
cc
occurs as
the product of the daily T
max
above the threshold temperature
θs,1, multiplied by the temperature sensitivity factor, ρs,1.In
autumn, the decrease in G
cc
occurs as the product of the daily
T
min
below the threshold temperature θs,2, multiplied by the tem-
perature sensitivity factor, ρs,2. The timing of the transition from
the spring phenology phase to the autumn phenology phase is
determined by the parameter D, which is specified as a day of year.
The posterior distributions of the model parameters (θs,1 ,θs,2 ,
ρs,1,ρs,2 and D) were fitted on a site-by-site basis for 26 evergreen
conifer PhenoCam sites in North America (Table S2). We
selected evergreen conifer sites with more than 4 yr of data in the
PhenoCam data set v.2.0 (Seyednasrollah et al., 2019a) and
where the seasonal signal-to-noise ratio was at least 20 : 1. We
quantified ‘signal’ as the seasonal amplitude of G
cc
, and ‘noise’ as
the mean absolute difference in G
cc
across successive days (over
which time phenological change is expected to be small). Daily
minimum and maximum temperature data were obtained from
Daymet (Thornton et al., 2016). A Gibbs sampling method
(‘RJAGS’ package in R 4.0; Plummer, 2013) was used to fit the
posterior distributions. For cross-validation of an n-year time
series, the first n–1 year were used for model calibration param-
eters (see Fig. S2), and the final year of data was used for out-of-
sample validation. To evaluate the ability of the model to capture
interannual variation in the timing of phenological shifts, SOS
and EOS transition dates were estimated from the model as 50%
of the seasonal amplitude of simulated G
cc
and compared against
the 50% seasonal amplitude transition dates described in section
entitled ‘PhenoCam data’.
Results
Leaf-level observations
Howland Forest At Howland Forest, the time series of G
cc
and
GRVI showed strong seasonality in canopy color, with both color
indices following approximately sinusoidal patterns, reaching a
minimum in winter and a maximum in summer (Fig. 2). On
days with snow on the canopy (open symbols), G
cc
was shifted
downward by a relatively small amount relative to G
cc
measured
on days without snow on the canopy (blue symbols). By compar-
ison, when there was snow on the canopy, GRVI was shifted
upward to levels that were more similar to summer values. Filter-
ing data for snow on canopy is therefore important for G
cc
but
critical for GRVI. Even when there was no snow on the canopy,
New Phytologist (2021) 229: 2586–2600 Ó2020 The Authors
New Phytologist Ó2020 New Phytologist Foundation
www.newphytologist.com
Research
New
Phytologist
2590

GRVI appeared to be more variable, day-to-day, than G
cc
, with
some obvious outliers. There was less year-to-year variation in
the winter minimum of G
cc
than GRVI, but overall the two time
series were well correlated (r=0.89).
The data indicate that the onset of increases in G
cc
and GRVI
occurred independently of the production of new foliage, as
determined by on-the-ground phenological surveys (Fig. 2).
Notably the date of budburst by the dominant conifer species,
red spruce and eastern hemlock, occurs after the point when
more than half of the annual seasonal amplitude of these indices
has already been reached.
Harvard Forest At Harvard Forest, leaf-level measurements on
white pine and eastern hemlock showed strong seasonal patterns
in two key physiological indices, chlorophyll fluorescence, F
v
/F
m
,
and the PRI (Fig. 3). The value of F
v
/F
m
indicates that leaf-level
changes in the quantum efficiency of photosystem II (PS II;
Bolh`
ar-Nordenkampf et al., 1989) occurred in parallel with
changes in G
cc
over the course of the year. PRI suggests that the
changes in G
cc
were associated with variation in the chl : car ratio
(Garbulsky et al., 2011). As was the case at Howland Forest (Fig.
2), increases in G
cc
(normalized in Fig. 3c to account for the dif-
ferences in winter baseline and summer maxima between the two
species) began many weeks before the production of new foliage
occurred in June.
Niwot Ridge At Niwot Ridge, our analysis of the temporal pat-
terns in leaf pigments highlights the strong relationship between
canopy color and leaf-level pigment pools and ratios. First, the
seasonal patterns in both color indices (G
cc
and GRVI) and pig-
ment data were highly consistent across the three trees (Fig. 4;
two lodgepole pines: P1 and P2, and one Engelmann spruce: S1)
in the field of view of the niwot5 phenocam, with one exception:
chlorophyll content did not vary across individuals, or over time.
This is consistent with results presented by Bowling et al. (2018)
using data from a previous year.
Second, the seasonal patterns of variation in G
cc
correlated best
with those for the chl : xan ratio, while GRVI correlated best with
the chl : car ratio (Fig. 5). These results show clear covariation
on seasonal timescales between leaf-level pigments and canopy
color as quantified from phenocam imagery.
We complement the earlier analysis of broad-band changes in
canopy color with a more detailed analysis using hyperspectral
vegetation reflectance data (Fig. 6). Across all wavelength com-
binations, we found λ
1
~530 nm (green) and λ
2
~690 nm
(red) resulted in high correlation (r>0.95) between a two-
wavelength NDI and the chl : car ratio, highlighting the impor-
tance of seasonal changes in leaf pigment content as a likely
mechanism driving changes in spectral reflectance, and the
potential to detect these changes with color-based indices
(specifically GRVI, given the optimal bands for λ
1
and λ
2
)
retrieved from camera imagery. There were a number of nar-
rowly-defined spectral ranges (e.g. for any λ
2
,500<λ
1
<550
nm and 675 <λ
1
<700 nm, and also 700 <λ
2
<725 nm
when 525 <λ
1
<650) where correlations with chl : car were
high. We further note that for any value of λ
2
when λ
1
<500
nm, correlation of NDI with chl : car was not as strong as with
the green–red band combinations.
0.34
0.36
0.38
0.40
Gcc
Jan 2012 Jan 2013 Jan 2014 Jan 2015 Jan 2016 Jan 2017
-0.04
-0.02
0.00
0.02
0.04
0.06
GRVI
Fig. 2 Seasonality of canopy color, as
characterized by the green chromatic
coordinate (G
cc
, upper panel) and green–red
vegetation index (GRVI, lower panel), from
PhenoCam imagery for Howland Forest.
Filled blue symbols indicate 3-d composite
values from high-frequency imagery; hollow
symbols are data that have been screened
because of snow on the evergreen canopy.
Shaded blue bands indicate LOESS
smoothing spline fit to the snow-filtered
data, 95% confidence interval. Vertical
green bars indicate the period between
conifer budburst and completion of leaf
development, based on on-the-ground
surveys conducted each spring since 1990.
Ó2020 The Authors
New Phytologist Ó2020 New Phytologist Foundation
New Phytologist (2021) 229: 2586–2600
www.newphytologist.com
New
Phytologist Research 2591

Links between seasonal changes in canopy color and
photosynthetic capacity
SOS and EOS transition dates derived from G
cc
and GRVI were
generally aligned with the corresponding dates indicating the
start and end of photosynthetic activity, as estimated from GPP
(Fig. 7). But there were obvious outliers to these relationships at
sites which were not fully winter dormant; although photosyn-
thetic capacity does decline in winter at the warm sites, it does
not shut down completely, and thus the concept of SOS and
EOS dates is not applicable (even though seasonal variation in
canopy color was observed even at the warmest site, US-SP1).
Our analysis is therefore restricted to the remaining seven winter-
dormant sites, which ranged across 14°in latitude, 37°in
longitude, and more than 8°C in MAT. SOS and EOS transition
dates varied by more than 2 months across site-years (n=38 and
n=34, respectively).
As described in the Methods section, we found that many
different calculations could provide PhenoCam-based transi-
tion dates that were well-correlated with GPP-based transition
dates, although some methods were superior (Table S3). The
best methods resulted in transition dates for both G
cc
and
GRVI that were well-correlated with GPP-based dates for
both SOS (r=0.70 and r=0.69, respectively) and EOS
(r=0.73 and r=0.74) (Fig. 7). However, the root-mean-
square error (RMSE) for G
cc
SOS (6 d) was substantially
smaller than for G
cc
EOS (11 d), and for either GRVI SOS
(10 d) or EOS (11 d).
0.20
0.40
0.60
0.80
Fv/Fm
White pine
Eastern hemlock
–0.10
–0.05
0.00
0.05
0.10
PRI
Jan 2015 Apr 2015 Jul 2015 Oct 2015 Jan 2016 Apr 2015
0.00
0.25
0.50
0.75
1.00
Month and Year
Normalized Gcc
Budburst,1990-2001
White pine
Eastern hemlock
(a)
(b)
(c)
Fig. 3 Seasonal variation in leaf-level
physiological measurements for white pine
and eastern hemlock at Harvard Forest.
(a) Chlorophyll fluorescence F
v
/F
m
; (b) the
photochemical reflectance index, PRI; (c) the
green chromatic coordinate (G
cc
), normalized
by observed seasonal maxima and minima to
fall within the range 0–1. In (c), the
horizontal bars indicate the mean dates of
budburst (production of new foliage) for
each species, over the period 1990–2001
(ground observations were discontinued
before the start of the present study). The
errors bars in (a) and (b) show the SD of the
observations.
New Phytologist (2021) 229: 2586–2600 Ó2020 The Authors
New Phytologist Ó2020 New Phytologist Foundation
www.newphytologist.com
Research
New
Phytologist
2592

Temperature-based modeling of evergreen phenology
Our temperature-based phenology model performed well, in that
it was able to reproduce both the seasonal trajectory of canopy
color and transition dates for in-sample data, as well as out-of-
sample data (Fig. 8).
Despite the large climatological range, and differences in
species composition across sites, the model consistently replicated
the seasonal patterns of G
cc
and successfully predicted the year
held out for cross-validation at each site. For example, US-Mpj is
a semi-arid pinyon-juniper forest in New Mexico, USA while
CA-Obs is a boreal spruce forest in Saskatchewan, Canada. The
two contrasting sites exhibited distinct trends in their seasonal
change of canopy color (Fig. S2). The seasonal amplitude of
greenness at CA-Obs was more than twice as much as that at US-
Mpj, and green-up and green-down occurred at a faster rate at
CA-Obs than at US-Mpj. The model performed almost equally
well at both sites (Fig. S2).
Our temperature-based model closely tracked the timing and
the rate of changes in greenness for all site-years. The values of R
2
for calibration data (82% of the data) and validation data (18%
of the data) were 0.94 and 0.89, respectively (Fig. 8). The spring
and autumn transition dates extracted from the modeled time
series also matched well with the transition dates derived from
the G
cc
time series. The median absolute error was 6 d for both
spring and autumn transition dates when we used the calibration
data. For the validation data, the median absolute errors were 6
and 9 d for spring and autumn transition dates, respectively.
Model parameters were generally similar across different species
and climate regimes (see Table S4). More specifically, G
min
,
G
max
and Dshowed small variability across sites. The spring
threshold temperature (θ
1
) expressed higher variability than the
autumn threshold temperature (θ
2
), but the variability of temper-
ature sensitivity coefficient of spring (ρ
1
) was larger than that of
autumn (ρ
2
). Overall, the model parameters did not vary in rela-
tion to climate (MAT or MAP) or site location (latitude,
0.38 0.40 0.42
Gcc
Tree ID: P1
0.00 0.02 0.04 0.06 0.08
GRVI
Tree ID: P1
0.0 0.5 1.0 1.5
chl (µmol g–1)chl (µmol g–1)chl (µmol g–1)
Tree ID: P1
2.0 2.5 3.0 3.5 4.0
chl:car (mol mol–1)chl:car (mol mol–1)chl:car (mol mol–1)
Tree ID: P1
5 10152025
chl:xan (mol mol–1)chl:xan (mol mol–1)chl:xan (mol mol–1)
Tree ID: P1
0.39 0.40 0.41 0.42
Gcc
Tree ID: P2
-0.01 0.01 0.03 0.05
GRVI
Tree ID: P2
0.0 0.5 1.0 1.5
Tree ID: P2
2.4 2.8 3.2 3.6
Tree ID: P2
10 15 20 25
Tree ID: P2
Jul Sep Nov Jan Mar May
0.37 0.38 0.39 0.40 0.41
Gcc
Tree ID: S1
Jul Sep Nov Jan Mar May
0.00 0.02 0.04 0.06
GRVI
Tree ID: S1
Jul Sep Nov Jan Mar May
0.0 0.5 1.0 1.5
Tree ID: S1
Jul Sep Nov Jan Mar May
2.5 3.0 3.5 4.0
Tree ID: S1
Jul Sep Nov Jan Mar May
510203040
Tree ID: S1
Fig. 4 Seasonal patterns in phenocam-derived canopy color indices (G
cc
and GRVI), and pigment contents and ratios, for three trees (two lodgepole pine:
P1 and P2, and one Engelmann spruce: S1) in the field of view of the niwot5 phenocam. For additional pigment content (total carotenoids and total
xanthophylls) data, see Supporting Information Fig. S3.
Ó2020 The Authors
New Phytologist Ó2020 New Phytologist Foundation
New Phytologist (2021) 229: 2586–2600
www.newphytologist.com
New
Phytologist Research 2593

longitude, elevation) factors, but as the PhenoCam data archive
grows, it may be possible to identify biogeographic patterns in
model parameters.
Discussion
Our study shows that changes in evergreen conifer canopy color
that occur on seasonal timescales are associated with variation in
leaf pigment ratios. Furthermore, our results show that changes
in canopy color can be detected from digital camera imagery,
and SOS and EOS transition dates derived from canopy color
metrics align with corresponding dates representing the start
and end of photosynthetic activity in winter-dormant ecosys-
tems (Gamon et al., 2016; Wong et al., 2020). Finally, the sea-
sonal trajectory of changes in canopy color can be simulated
with a temperature-based model which likewise successfully
reproduces the timing of SOS and EOS transition dates across a
wide range of evergreen ecosystems. Our work builds on previ-
ous studies (Richardson et al., 2009a; Toomey et al., 2015;
Bowling et al., 2018; Liu et al., 2020) by providing new insight
into the underlying mechanisms and drivers of the observed
relationship between canopy color and photosynthesis, and
highlighting the broader generality of these results across North
America.
The seasonality of canopy color variation is directly linked
to photochemical processes
We found that two visible-wavelength spectral indices, G
cc
and
GRVI, reveal strong seasonal variation in color of evergreen
conifer canopies. This variation occurred independently of new
leaf development in spring (Figs 2, 3c; see also Zhang et al.,
2020). Seasonal variation in canopy color occurred in tandem
with shifts in the quantum efficiency of PS II photochemistry as
measured by F
v
/F
m
(Fig. 3a), as well as PRI (Fig. 3b) and leaf
pigment pools and their ratios (Fig. 5). The leaf-level processes
associated with seasonal upregulation and downregulation of the
photosynthetic machinery, and related mechanisms of photopro-
tection, in evergreen conifer forests have been well-documented
(Busch et al., 2009; Fr´
echette et al., 2015, 2016; Wong et al.,
2019; Grebe et al., 2020). Downregulation of photosynthesis
involves inactivation of PS II reaction centers and re-organization
of light harvesting complexes to function for energy dissipation
(Ensminger et al., 2006). Ensminger et al. (2004) reported that
the springtime increase in NEE of a Scots pine forest, which
occurred as temperatures rose above 0°C, is correlated with
detectable changes in leaf pigments, F
v
/F
m
, and chloroplast pro-
tein synthesis. This previous work provides context for the results
and support for the conclusions drawn here.
The PRI has been shown to be a reliable index of photosyn-
thetic radiation use efficiency in evergreen forests (Garbulsky
et al., 2011; Soudani et al., 2014). Although originally conceived
to detect changes in xanthophyll cycle epoxidation state occurring
on minutes to diurnal timescales (Gamon et al., 1992), on sea-
sonal timescales PRI is more indicative of shifts in the chl : car
ratio that occur independently of the xanthophyll cycle (Wong &
Gamon, 2015a). Upregulation and downregulation of photosyn-
thesis occurs as a complex process involving a number of discrete
components that are coordinated but operate on different
timescales (Wong & Gamon, 2015b; see also Busch et al., 2009;
Fr´
echette et al., 2016). Nevertheless, on seasonal timescales PRI
and CCI are reliable indicators of potential photosynthetic activ-
ity because of their sensitivity to the chl : car ratio (Wong &
Gamon, 2015b; Gamon et al., 2016; Wong et al., 2020). Like-
wise, the indices used here to track canopy color, G
cc
and GRVI,
mimic the seasonality of PRI and F
v
/F
m
(Fig. 3) and are corre-
lated with the chl : xan and chl : car ratios (Fig. 5), thereby
establishing a plausible link between canopy color and photo-
chemical processes.
Changes in pigment pools and ratios are detected from
canopy-scale near-surface remote sensing
Near-surface remote sensing at the canopy level presents new
challenges that are not typically an issue with leaf-level spectral
0.00 0.02 0.04 0.06 0.08
GRVI
R = 0.78
2.0 2.5 3.0 3.5 4.0
R = 0.7
Chl : car
0.2 0.4 0.6 0.8 1.0
5 10152025303540
R = 0.8
Chl : xan
Normalized Gcc
R = 0.78
0.00 0.02 0.04 0.06 0.08
R = 0.72
GRVI
P1
P2
S1
Fig. 5 Correlation of canopy color indices and pigment ratio data for
Niwot Ridge. Data are plotted using different symbols for each of the
three trees (two lodgepole pine: P1 and P2, and one Engelmann spruce:
S1) within the niwot5 phenocam field of view. Rather than presenting the
full correlation matrix (see Supporting Information Fig. S3), here we show
only the strongest correlations. To account for the different seasonal
maxima and minima of green chromatic coordinate (G
cc
) across different
trees, the values were normalized.
New Phytologist (2021) 229: 2586–2600 Ó2020 The Authors
New Phytologist Ó2020 New Phytologist Foundation
www.newphytologist.com
Research
New
Phytologist
2594

measurements. These include the effects of canopy structure and
nonphotosynthetic biomass, as well as changing illumination
geometry and weather. Our study builds on previous work that
has linked leaf- and canopy-level spectral measurements to sea-
sonal co-variation in pigments and photosynthesis (Wong et al.,
2019, 2020). In addition to showing that vegetation indices cal-
culated from hyperspectral PhotoSpec data correlated well with
pigment pools and ratios, we found that color indices derived
from broadband phenocam imagery performed almost equally
well (Fig. 5). Phenocam indices have been shown to have a non-
linear relationship with pigment concentrations in senescing
deciduous foliage (Junker & Ensminger, 2016; Liu et al., 2018).
Thus, changes in pigment pools and ratios are associated with
changes in foliage color that can be detected using broad-band
visible-wavelength color indices.
Snow is problematic for remote sensing of evergreen forests
(J¨
onsson et al., 2010); it obscures the vegetation of interest and
confounds the spectral signature. This is particularly an issue with
GRVI (Fig. 3). An advantage of the phenocam approach is that
images can be visually inspected to identify days when there is
snow on the canopy, and the affected images removed from the
analysis (Richardson et al., 2018b). Automated methods have also
been developed to identify snowy images (Kosmala et al., 2018).
By comparison, snow cover data products from satellite remote
sensing (e.g. MODIS MOD10A1 and MYD10A1) may fail to
detect discontinuous snow cover (Kosmala et al., 2018), which
could still adversely impact spectral indices. Future studies that
attempt to link data from orbiting sensors to tower-based carbon
and water flux measurements could benefit from making use of
the imagery from collocated phenocams to identify snowy days.
Canopy color indices derived from digital camera imaging
can be used to characterize photosynthetic phenology
Although the term ‘phenology’ is typically applied to visually
observable developmental events, such as budburst and flowering,
(a)
(b)
Fig. 6 (a) Heatmap plot showing Pearson (r)
correlation of normalized difference indices
(Eqn 3), λ
1
and λ
2
wavelengths, with leaf-
level measurements of chl : car pigment ratio
conducted over the course of the year. The
white contour line indicates |r|= 0.95. (b)
Quantum efficiency of PhenoCam digital
cameras for each color channel. The
hyperspectral reflectance data were obtained
from a 2D scanning telescope, PhotoSpec.
Ó2020 The Authors
New Phytologist Ó2020 New Phytologist Foundation
New Phytologist (2021) 229: 2586–2600
www.newphytologist.com
New
Phytologist Research 2595

it has also been used to describe the seasonality of physiological
processes at leaf-to-ecosystem scales, e.g. ‘photosynthetic phenol-
ogy’ (Gu et al., 2009; Richardson et al., 2013a; Gamon et al.,
2016). The timing of the start and end of the photosynthetically
active season plays a large role in controlling annual ecosystem
productivity, and how productivity varies in time and space
(Richardson et al., 2010a). Low-cost sensors that could be used
to characterize photosynthetic phenology would be of value for
scaling photosynthetic uptake from a small set of intensively-
monitored sites with eddy covariance instrumentation to a much
larger, more spatially extensive observation network.
Previous studies have highlighted the ability of CCI and PRI,
measured at both the leaf- (Eitel et al., 2019; Wong et al., 2019)
and canopy-level (Gamon et al., 2016), to provide robust mea-
sures of the SOS and EOS dates for photosynthetic phenology.
The results presented here show that G
cc
and GRVI offer similar
potential. Our analysis includes sites across a wide geographic
range in North America, with SOS and EOS dates both varying
by up to 70 d across sites and years (Fig. 7). We acknowledge
that relationships between our phenocam indices and GPP-
derived transition dates broke down at warm sites where photo-
synthesis was maintained year-round (albeit at a reduced rate in
winter), even though G
cc
and GRVI indicated variation in
canopy color. One hypothesis to explain this pattern is that the
photoprotective mechanisms at these warmer sites are not upreg-
ulated (or downregulated) to the same degree that they are in
colder sites, and this drives differences in the relationship between
reflectance characteristics and photochemistry. Nevertheless, our
results support the idea that phenocams are low-cost but sensitive
instruments that have a wide applicability in environmental mon-
itoring (Richardson, 2019). Here, we have specifically shown the
potential value of phenocams for identifying seasonal patterns in
evergreen conifer forests.
A temperature-based model can simulate the trajectory of
seasonal changes in canopy color
We presented a dynamic modeling framework for simulating
canopy color, and by proxy photosynthetic activity, that can be
flexibly adapted to larger scales and applications. The structure of
our model has roots in previous representations of evergreen phe-
nology (H¨
anninen & Kramer, 2007; H¨
anninen, 2016). The
strong sensitivity of increases in canopy greenness to warm tem-
perature in spring, and decreases in canopy greenness in autumn
that are driven by cold temperature, is consistent with our under-
standing of the role of temperature and photoperiod in driving
evergreen seasonality (Tanja et al., 2003; Ensminger et al., 2004;
Fr´
echette et al., 2016). The results are also consistent with previ-
ous empirical analyses of phenocam data, which found that inter-
annual variation in evergreen conifer SOS and EOS dates was
mostly explained by air temperature during a narrow window
preceding each transition (Richardson et al., 2019). But, similar
to Tanja et al. (2003), we found that it was necessary to parame-
terize our model separately at each site. And, although we had
anticipated that unusually cold temperature in spring (or warm
temperatures in autumn) might cause temporary reversal of the
overall upward (downward) trajectory, we were unable to identify
the associated temperature thresholds, perhaps because at the site
level, such events are relatively rare. While further work is there-
fore needed to develop an improved model that generalizes well
across sites, this framework may ultimately have application in
global land surface models in which the seasonality of evergreen
50 100 150 200
50 100 150 200
GPP SOS 25% (d)
PhenoCam SOS 25% (d)
(a)
50 100 150 200
50 100 150 200
GPP SOS 50% (d)
PhenoCam SOS 50% (d)
(c)
250 300 350
250 300 350
GPP EOS 25% (d)
PhenoCam EOS 25% (d)
(b)
250 300 350
250 300 350
GPP EOS 50% (d)
PhenoCam EOS 50% (d)
(d)
CA-Obs
CA-Qfo
CA-TP1
CA-TP3
CA-TP4
US-Ho1
US-NR1
CA-Ca3
US-Me2
US-Me6
US-SP1
Fig. 7 Start-of-season (SOS) and end-of-
season (EOS) derived from PhenoCam
reflectance indices and from gross primary
productivity (GPP) were highly correlated.
Vertical axes show PhenoCam-based
transition dates. Horizontal axes show GPP-
based transition dates. Upper panels (a, b)
are based on the 90
th
percentile of the green
chromatic coordinate (G
cc
) time series. Lower
panels (c, d) are based on the mean
green–red vegetation index (GRVI) time
series. Individual years for each site are
shown separately. Uncertainties are shown
as error bars. Warm sites, where
photosynthesis occurs year-round, are shown
with hollow symbols.
New Phytologist (2021) 229: 2586–2600 Ó2020 The Authors
New Phytologist Ó2020 New Phytologist Foundation
www.newphytologist.com
Research
New
Phytologist
2596

conifer forests is poorly represented (Lawrence et al., 2011;
Richardson et al., 2012; Peaucelle et al., 2019) and in incorporat-
ing potential effects from other factors such as daylength (Way &
Montgomery, 2015).
Conclusion
The novel contribution of our study is the finding that seasonal
pigment changes in evergreen conifers are detectable at the
canopy level using imagery from low-cost, commercially available
digital cameras. Given the low cost of digital cameras, our work
highlights the potential value in co-locating phenocams with
more advanced instruments such as solar induced fluorescence
(SIF). SIF has been shown to track changes in canopy-level pho-
tosynthetic capacity across a range of timescales in evergreen
ecosystems (Zuromski et al., 2018; Magney et al., 2019). We
note that many leaf-level studies have found it advantageous to
simultaneously measure both F
v
/F
m
and PRI to fully understand
the factors that are regulating photosynthesis at different times of
the year (Busch et al., 2009; Fr´
echette et al., 2016; Wong et al.,
2019). Likewise, installing phenocams at sites where SIF instru-
mentation is already installed may have great value for offsetting
the specific limitations of each method. These measurements
would be complementary and not redundant and would allow
(c)
0.30 0.35 0.40 0.45
0.30
0.35
0.40
0.45
0.30
0.35
0.40
0.45
Observed greenness
Modeled greenness
MAE = 6 d
MAE = 6 d
(b)
MAE = 6 d
MAE = 9 d
(d)
100 200 300
100
200
300
100
200
300
Observed transition day
Modeled transition day
(a)
² = 0.94
R
² = 0.89
R
Fig. 8 Validations of the phenology model for (a, b) in-sample and (c, d) out-of-sample data. The plots on the left (a, c) are modeled vs observed
greenness values. The plots on the right (b, d) are modeled vs observed transition dates values. MAE stands for median absolute error. Green circles and
orange triangles on the right-hand-side panels indicate spring and autumn transition dates, respectively. Horizontal features in (a) and (c) are due to the
minimum and maximum greenness values in the model parameters.
Ó2020 The Authors
New Phytologist Ó2020 New Phytologist Foundation
New Phytologist (2021) 229: 2586–2600
www.newphytologist.com
New
Phytologist Research 2597

independent quantification of the temporal kinetics of both
chlorophyll fluorescence and leaf pigmentation.
Acknowledgements
This is a product of the IMGG workshop held in Flagstaff, AZ,
in May 2019. ADR and DRB thank J. Page, J. White, and D.
Evans for inspiration. ADR acknowledges support from the
National Science Foundation, through the Macrosystems Biology
(awards EF-1065029 and EF-1702697) and Long-Term Ecologi-
cal Research (award DEB-1832210) programs. DRB acknowl-
edges support from the NASA Carbon Monitoring Systems
(awards NNX16AP33G and 80NSSC20K0010) and ABoVE
(80NSSC19M0130) programs, and the National Science Foun-
dation, through the Macrosystems Biology and NEON-Enabled
Science program (DEB-1926090). Thanks to flux tower scientists
for hosting the PhenoCams and providing flux and met data, and
to many agencies for funding the flux towers. PhenoCam
acknowledgments for individual sites are presented in Supporting
Information. Needle pigments were quantified by Sophia Lopez,
who held a Langbein Research Fellowship (Bowdoin College),
with Jaret Reblin’s support. The authors declare no conflict of
interest.
Author contributions
ADR and DRB planned the IMGG workshop and designed the
research. BS, DRB, RC, BAL, TSM, CF, JCY, AMY and KH
conducted research. MAA, TAB, PDB, RB, RJ, DYH, BEL and
ZN contributed tower measurements. BS, RC, DRB and ADR
analyzed data. BS, DRB and ADR drafted the manuscript with
input from RC, BAL, TSM and KH. All authors provided feed-
back on manuscript drafts and approved the manuscript for sub-
mission.
ORCID
M. Altaf Arain https://orcid.org/0000-0002-1433-5173
T. Andrew Black https://orcid.org/0000-0002-7494-9767
Peter D. Blanken https://orcid.org/0000-0002-7405-2220
David R. Bowling https://orcid.org/0000-0002-3864-4042
Rosvel Bracho https://orcid.org/0000-0002-8015-9796
Rui Cheng https://orcid.org/0000-0002-3003-8339
Christian Frankenberg https://orcid.org/0000-0002-0546-
5857
Koen Hufkens https://orcid.org/0000-0002-5070-8109
David Y. Hollinger https://orcid.org/0000-0002-4284-1575
Rachhpal Jassal https://orcid.org/0000-0002-6727-5215
Beverly E. Law https://orcid.org/0000-0002-1605-1203
Barry A. Logan https://orcid.org/0000-0003-3920-2139
Troy S. Magney https://orcid.org/0000-0002-9033-0024
Zoran Nesic https://orcid.org/0000-0001-5728-513X
Andrew D. Richardson https://orcid.org/0000-0002-0148-
6714
Bijan Seyednasrollah https://orcid.org/0000-0002-5195-2074
Julia C. Yang https://orcid.org/0000-0001-9698-9033
Adam M. Young https://orcid.org/0000-0003-2668-2794
Data availability
Freely available code and data can be accessed from https://
github.com/bnasr/ENPhenology.
References
Baldocchi D. 2008. Turner review no. 15. ‘Breathing’ of the terrestrial biosphere:
Lessons learned from a global network of carbon dioxide flux measurement
systems. Australian Journal of Botany 56:1–26.
Balzarolo M, Vicca S, Nguy-Robertson AL, Bonal D, Elbers JA, Fu YH,
Gr¨
unwald T, Horemans JA, Papale D, Pe ˜
nuelas J et al. 2016. Matching the
phenology of Net Ecosystem Exchange and vegetation indices estimated with
MODIS and FLUXNET in-situ observations. Remote Sensing of Environment
174: 290–300.
Bolh`
ar-Nordenkampf HR, Long SP, Baker NR, Oquist G, Schreiber U,
Lechner EG. 1989. Chlorophyll fluorescence as a probe of the photosynthetic
competence of leaves in the field: a review of current instrumentation.
Functional Ecology 3: 497.
Bonan GB. 2008. Forests and climate change: forcings, feedbacks, and the
climate benefits of forests. Science 320: 1444–1449.
Bowling DR, Logan BA, Hufkens K, Aubrecht DM, Richardson AD, Burns SP,
Anderegg WRL, Blanken PD, Eiriksson DP. 2018. Limitations to winter and
spring photosynthesis of a Rocky Mountain subalpine forest. Agricultural and
Forest Meteorology 252: 241–255.
Busch F, H¨
uner NPA, Ensminger I. 2009. Biochemical constrains limit the
potential of the photochemical reflectance index as a predictor of effective
quantum efficiency of photosynthesis during the winter spring transition in
Jack pine seedlings. Functional Plant Biology 36: 1016–1026.
Cheng R, Magney TS, Dutta D, Bowling DR, Logan BA, Burns SP,
Blanken PD, Grossmann K, Lopez S, Richardson AD et al. 2020.
Decomposing reflectance spectra to track gross primary production
in a subalpine evergreen forest. Biogeosciences Discussions 17:
4523–4544.
Demmig-Adams B, Adams WW. 1996. The role of xanthophyll cycle
carotenoids in the protection of photosynthesis. Trends in Plant Science 1:
21–26.
Eitel JUH, Maguire AJ, Boelman N, Vierling LA, Griffin KL, Jensen J, Magney
TS, Mahoney PJ, Meddens AJH, Silva C et al. 2019. Proximal remote sensing
of tree physiology at northern treeline: do late-season changes in the
photochemical reflectance index (PRI) respond to climate or photoperiod?
Remote Sensing of Environment 221: 340–350.
Ensminger I, Busch F, Huner NPA. 2006. Photostasis and cold acclimation:
sensing low temperature through photosynthesis. Physiologia Plantarum 126:
28–44.
Ensminger I, Sveshnikov D, Campbell DA, Funk C, Jansson S, Lloyd J,
Shibistova O,
¨
Oquist G. 2004. Intermittent low temperatures constrain spring
recovery of photosynthesis in boreal Scots pine forests. Global Change Biology
10: 995–1008.
Fr´
echette E, Chang CY-Y, Ensminger I. 2016. Photoperiod and temperature
constraints on the relationship between the photochemical reflectance index
and the light use efficiency of photosynthesis in Pinus strobus.Tree Physiology
36: 311–324.
Fr´
echette E, Wong CYS, Junker LV, Yao C, Chang Y, Ensminger I. 2015.
Zeaxanthin-independent energy quenching and alternative electron sinks cause
a decoupling of the relationship between the photochemical reflectance index
(PRI) and photosynthesis in an evergreen conifer during spring. Journal of
Experimental Botany 66: 7309–7323.
Gamon JA, Huemmrich KF, Wong CYS, Ensminger I, Garrity S, Hollinger
DY, Noormets A, Pe˜
nuelas J. 2016. A remotely sensed pigment index reveals
photosynthetic phenology in evergreen conifers. Proceedings of the National
Academy of Sciences, USA 113: 13087–13092.
New Phytologist (2021) 229: 2586–2600 Ó2020 The Authors
New Phytologist Ó2020 New Phytologist Foundation
www.newphytologist.com
Research
New
Phytologist
2598

Gamon JA, Pe˜
nuelas J, Field CB. 1992. A narrow-waveband spectral index that
tracks diurnal changes in photosynthetic efficiency. Remote Sensing of
Environment 41:35–44.
Gamon JA, Serrano L, Surfus JS. 1997. The photochemical reflectance index: an
optical indicator of photosynthetic radiation use efficiency across species,
functional types, and nutrient levels. Oecologia 112: 492–501.
Garbulsky MF, Pe˜
nuelas J, Gamon J, Inoue Y, Filella I. 2011. The
photochemical reflectance index (PRI) and the remote sensing of leaf, canopy
and ecosystem radiation use efficiencies. A review and meta-analysis. Remote
Sensing of Environment 115: 281–297.
Grebe S, Trotta A, Bajwa AA, Mancini I, Bag P, Jansson S, Tikkanen M, Aro
EM. 2020. Specific thylakoid protein phosphorylations are prerequisites for
overwintering of Norway spruce (Picea abies) photosynthesis. Proceedings of the
National Academy of Sciences, USA 117: 17499–17509.
Grossmann K, Frankenberg C, Magney TS, Hurlock SC, Seibt U, Stutz J. 2018.
PhotoSpec: a new instrument to measure spatially distributed red and far-red
solar-induced chlorophyll fluorescence. Remote Sensing of Environment 216:
311–327.
Gu L, Post WM, Baldocchi DD, Black TA, Suyker AE, Verma SB, Vesala T,
Wofsy SC. 2009. Characterizing the seasonal dynamics of plant community
photosynthesis across a range of vegetation types. In: Noormets A, ed.
Phenology of ecosystem processes. New York, NY, USA: Springer, 35–58.
H¨
anninen H. 2016. Boreal and temperate trees in a changing climate. Dordrecht,
the Netherlands: Springer.
H¨
anninen H, Kramer K. 2007. A framework for modelling the annual cycle of
trees in boreal and temperate regions. Silva Fennica 41: 167–205.
Hollinger DY, Aber J, Dail B, Davidson EA, Goltz SM, Hughes H, Leclerc MY,
Lee JT, Richardson AD, Rodrigues C et al. 2004. Spatial and temporal
variability in forest-atmosphere CO
2
exchange. Global Change Biology 10:
1689–1706.
Hollinger DY, Goltz SM, Davidson EA, Lee JT, Tu K, Valentine HT. 1999.
Seasonal patterns and environmental control of carbon dioxide and water
vapour exchange in an ecotonal boreal forest. Global Change Biology 5:
891–902.
J¨
onsson AM, Eklundh L, Hellstr¨
om M, B¨
arring L, J¨
onsson P. 2010. Annual
changes in MODIS vegetation indices of Swedish coniferous forests in relation
to snow dynamics and tree phenology. Remote Sensing of Environment 114:
2719–2730.
Junker LV, Ensminger I. 2016. Relationship between leaf optical properties,
chlorophyll fluorescence and pigment changes in senescing Acer saccharum
leaves. Tree Physiology 36: 694–711.
Kosmala M, Hufkens K, Richardson AD. 2018. Integrating camera imagery,
crowdsourcing, and deep learning to improve high-frequency automated
monitoring of snow at continental-to-global scales. PLoS ONE 13: e0209649.
Lawrence DM, Oleson KW, Flanner MG, Thornton PE, Swenson SC,
Lawrence PJ, Zeng X, Yang Z-L, Levis S, Sakaguchi K et al. 2011.
Parameterization improvements and functional and structural advances in
version 4 of the Community Land Model. Journal of Advances in Modeling
Earth Systems 3: art. M03001.
Liu Y, Wu C, Sonnentag O, Desai AR, Wang J. 2020. Using the red chromatic
coordinate to characterize the phenology of forest canopy photosynthesis.
Agricultural and Forest Meteorology 285–286: 107910.
Liu Z, An S, Lu X, Hu H, Tang J. 2018. Using canopy greenness index to
identify leaf ecophysiological traits during the foliar senescence in an oak forest.
Ecosphere 9: e02337.
Magney TS, Bowling DR, Logan BA, Grossmann K, Stutz J, Blanken PD,
Burns SP, Cheng R, Garcia MA, K¨
ohler P et al. 2019. Mechanistic evidence
for tracking the seasonality of photosynthesis with solar-induced fluorescence.
Proceedings of the National Academy of Sciences, USA 116: 11640–11645.
Melaas EK, Friedl MA, Richardson AD. 2016. Multiscale modeling of spring
phenology across deciduous forests in the eastern United States. Global Change
Biology 22: 792–805.
Mı´guez F, Fern´
andez-Marı´n B, Becerril JM, Garcı
´a-Plazaola JI. 2015.
Activation of photoprotective winter photoinhibition in plants from different
environments: a literature compilation and meta-analysis. Physiologia
Plantarum 155: 414–423.
Peaucelle M, Ciais P, Maignan F, Nicolas M, Cecchini S, Viovy N. 2019.
Representing explicit budburst and senescence processes for evergreen conifers
in global models. Agricultural and Forest Meteorology 266–267:97–108.
Pielou EC. 2011. The World of Northern evergreens. Ithaca, NY, USA: Cornell
University Press.
Plummer M. 2013. rjags: Bayesian graphical models using MCMC,v.4-10.The
Comprehensive R Archive Network [WWW document] URL
https://cran.r-project.org/package=rjags.
Richardson AD. 2019. Tracking seasonal rhythms of plants in diverse ecosystems
with digital camera imagery. New Phytologist 222: 1742–1750.
Richardson AD, Anderson RS, Arain MA, Barr AG, Bohrer G, Chen G, Chen
JM, Ciais P, Davis KJ, Desai AR et al. 2012. Terrestrial biosphere models
need better representation of vegetation phenology: results from the North
American carbon program site synthesis. Global Change Biology 18: 566–584.
Richardson AD, Andy Black T, Ciais P, Delbart N, Friedl MA, Gobron N,
Hollinger DY, Kutsch WL, Longdoz B, Luyssaert S et al. 2010a. Influence of
spring and autumn phenological transitions on forest ecosystem productivity.
Philosophical Transactions of the Royal Society B: Biological Sciences 365:
3227–3246.
Richardson AD, Braswell BH, Hollinger DY, Jenkins JP, Ollinger SV. 2009a.
Near-surface remote sensing of spatial and temporal variation in canopy
phenology. Ecological Applications 19: 1417–1428.
Richardson AD, Hollinger DY, Dail DB, Lee JT, Munger JW, Okeefe J. 2009b.
Influence of spring phenology on seasonal and annual carbon balance in two
contrasting New England forests. Tree Physiology 29: 321–331.
Richardson AD, Hufkens K, Li X, Ault TR. 2019. Testing Hopkins’ Bioclimatic
Law with PhenoCam data. Applications in Plant Sciences 7: e01228.
Richardson AD, Hufkens K, Milliman T, Aubrecht DM, Chen M, Gray JM,
Johnston MR, Keenan TF, Klosterman ST, Kosmala M et al. 2018a.
Tracking vegetation phenology across diverse North American biomes using
PhenoCam imagery. Scientific Data 5.
Richardson AD, Hufkens K, Milliman T, Aubrecht DM, Furze ME,
Seyednasrollah B, Krassovski MB, Latimer JM, Nettles WR, Heiderman RR
et al. 2018b. Ecosystem warming extends vegetation activity but heightens
vulnerability to cold temperatures. Nature 560: 368–371.
Richardson AD, Keenan TF, Migliavacca M, Ryu Y, Sonnentag O, Toomey M.
2013a. Climate change, phenology, and phenological control of vegetation
feedbacks to the climate system. Agricultural and Forest Meteorology 169:
156–173.
Richardson AD, Klosterman S, Toomey M. 2013b. Near-surface sensor-derived
phenology. In: Schwartz MD, ed. Phenology: an Integrative environmental
science. Dordrecht, the Netherlands: Springer, 413–430.
Richardson AD, O’Keefe J. 2009. Phenological differences between understory
and overstory. In: Noormets A, ed. Phenology of ecosystem processes. New York,
NY, USA: Springer, 87–117.
Richardson AD, Williams M, Hollinger DY, Moore DJP, Dail DB, Davidson
EA, Scott NA, Evans RS, Hughes H, Lee JT et al. 2010b. Estimating
parameters of a forest ecosystem C model with measurements of stocks and
fluxes as joint constraints. Oecologia 164:25–40.
Seyednasrollah B, Young AM, Hufkens K, Milliman T, Friedl MA, Frolking S,
Richardson AD. 2019a. Tracking vegetation phenology across diverse biomes
using version 2.0 of the PhenoCam dataset. Scientific Data 6: 222.
Seyednasrollah B, Young AM, Hufkens K, Milliman T, Friedl MA, Frolking S,
Richardson AD, Abraha M, Allen DW, Apple M et al. 2019b. PhenoCam
dataset v2.0: vegetation phenology from digital camera imagery, 2000–2018. Oak
Ridge, TN, USA: ORNL DAAC. doi: 10.3334/ornldaac/1674.
Sonnentag O, Hufkens K, Teshera-Sterne C, Young AM, Friedl M, Braswell
BH, Milliman T, O’Keefe J, Richardson AD. 2012. Digital repeat
photography for phenological research in forest ecosystems. Agricultural and
Forest Meteorology 152: 159–177.
Soudani K, Hmimina G, Dufrˆ
ene E, Berveiller D, Delpierre N, Ourcival JM,
Rambal S, Joffre R. 2014. Relationships between photochemical reflectance
index and light-use efficiency in deciduous and evergreen broadleaf forests.
Remote Sensing of Environment 144:73–84.
Tanja S, Berninger F, Vesala T, Markkanen T, Hari P, M¨
akel¨
a A, Ilvesniemi H,
H¨
anninen H, Nikinmaa E, Huttula T et al. 2003. Air temperature triggers the
Ó2020 The Authors
New Phytologist Ó2020 New Phytologist Foundation
New Phytologist (2021) 229: 2586–2600
www.newphytologist.com
New
Phytologist Research 2599

recovery of evergreen boreal forest photosynthesis in spring. Global Change
Biology 9: 1410–1426.
Thornton PE, Thornton MM, Mayer BW, Wei Y, Devarakonda R, Vose RS,
Cook RB. 2016. Daymet: daily surface weather data on a 1-km grid for North
America, Version 3. Oak Ridge, TN, USA: ORNL DAAC.
Toomey M, Friedl MA, Frolking S, Hufkens K, Klosterman S, Sonnentag O,
Baldocchi DD, Bernacchi CJ, Biraud SC, Bohrer G et al. 2015.
Greenness indices from digital cameras predict the timing and seasonal
dynamics of canopy-scale photosynthesis. Ecological Applications
25:99–115.
Tucker CJ. 1979. Red and photographic infrared linear combinations for
monitoring vegetation. Remote Sensing of Environment 8: 127–150.
Verhoeven A. 2014. Sustained energy dissipation in winter evergreens. New
Phytologist 201:57–65.
Way DA, Montgomery RA. 2015. Photoperiod constraints on tree phenology,
performance and migration in a warming world. Plant, Cell & Environment 38:
1725–1736.
Wong CYS, D’Odorico P, Arain MA, Ensminger I. 2020. Tracking the
phenology of photosynthesis using carotenoid-sensitive and near-infrared
reflectance vegetation indices in a temperate evergreen and mixed deciduous
forest. New Phytologist 226: 1682–1695.
Wong CYS, D’Odorico P, Bhathena Y, Arain MA, Ensminger I. 2019.
Carotenoid based vegetation indices for accurate monitoring of the phenology
of photosynthesis at the leaf-scale in deciduous and evergreen trees. Remote
Sensing of Environment 233: 111407.
Wong CYS, Gamon JA. 2015a. Three causes of variation in the photochemical
reflectance index (PRI) in evergreen conifers. New Phytologist 206: 187–195.
Wong CYS, Gamon JA. 2015b. The photochemical reflectance index provides an
optical indicator of spring photosynthetic activation in evergreen conifers. New
Phytologist 206: 196–208.
Wutzler T, Lucas-Moffat A, Migliavacca M, Knauer J, Sickel K,
ˇ
Sigut L,
Menzer O, Reichstein M. 2018. Basic and extensible post-processing of eddy
covariance flux data with REddyProc. Biogeosciences 15: 5015–5030.
Zhang S, Butt`
o V, Khare S, Deslauriers A, Morin H, Huang J-G, Ren H, Rossi
S. 2020. Calibrating PhenoCam data with phenological observations of a black
spruce stand. Canadian Journal of Remote Sensing 46: 154–165.
Zuromski LM, Bowling DR, K¨
ohler P, Frankenberg C, Goulden ML, Blanken
PD, Lin JC. 2018. Solar-induced fluorescence detects interannual variation in
gross primary production of coniferous forests in the western United States.
Geophysical Research Letters 45: 7184–7193.
Supporting Information
Additional Supporting Information may be found online in the
Supporting Information section at the end of the article.
Fig. S1 Illustration of workflow for processing tower-measured
fluxes of net ecosystem exchange (NEE) of CO
2
to extract season-
ality of photosynthetic capacity and associated transition dates.
Fig. S2 Temperature-based phenology model captures the sea-
sonal trajectory of changes in canopy color for two sites with
strong climatological and species composition differences.
Fig. S3 Seasonal patterns in phenocam-derived canopy color
indices (G
cc
and GRVI), and pigment contents and ratios, for
three trees (two lodgepole pine: P1 and P2, and one Engelmann
spruce: S1) in the field of view of the niwot5 phenocam.
Fig. S4 Heatmaps show correlation values between color- and
pigment-based indices.
Table S1 Metadata for eddy covariance study sites.
Table S2 Metadata for PhenoCam study sites.
Table S3 Evaluation of correlation of start-of-season (SOS) and
end-of-season (EOS) transition dates, derived from PhenoCam
imagery, with corresponding dates derived from tower-based esti-
mates of gross primary production (GPP).
Table S4 List of the fitted model parameters.
Please note: Wiley Blackwell are not responsible for the content
or functionality of any Supporting Information supplied by the
authors. Any queries (other than missing material) should be
directed to the New Phytologist Central Office.
New Phytologist (2021) 229: 2586–2600 Ó2020 The Authors
New Phytologist Ó2020 New Phytologist Foundation
www.newphytologist.com
Research
New
Phytologist
2600