Content uploaded by Matthias Forkel
Author content
All content in this area was uploaded by Matthias Forkel on Apr 18, 2016
Content may be subject to copyright.
Codominant water control on global interannual
variability and trends in land surface phenology and
greenness
MATTHIAS FORKEL
1
, MIRCO MIGLIAVACCA
1
, KIRSTEN THONICKE
2
,MARKUS
REICHSTEIN
1
, SIBYLL SCHAPHOFF
2
, ULRICH WEBER
1
and NUNO CARVALHAIS
1,3
1
Department for Biogeochemical Integration, Max-Planck-Institute for Biogeochemistry, Hans-Kn
€
oll-Str. 10, 07745 Jena,
Germany,
2
Potsdam Institute for Climate Impact Research, Earth System Analysis, Telegraphenberg A31, 14473 Potsdam,
Germany,
3
Faculdade de Ci
^
encias e Tecnologia, Universidade Nova de Lisboa, 2829-516 Caparica, Portugal
Abstract
Identifying the relative importance of climatic and other environmental controls on the interannual variability and
trends in global land surface phenology and greenness is challenging. Firstly, quantifications of land surface phenol-
ogy and greenness dynamics are impaired by differences between satellite data sets and phenology detection meth-
ods. Secondly, dynamic global vegetation models (DGVMs) that can be used to diagnose controls still reveal
structural limitations and contrasting sensitivities to environmental drivers. Thus, we assessed the performance of a
new developed phenology module within the LPJmL (Lund–Potsdam–Jena managed Lands) DGVM with a compre-
hensive ensemble of three satellite data sets of vegetation greenness and ten phenology detection methods, thereby
thoroughly accounting for observational uncertainties. The improved and tested model allows us quantifying the rel-
ative importance of environmental controls on interannual variability and trends of land surface phenology and
greenness at regional and global scales. We found that start of growing season interannual variability and trends are
in addition to cold temperature mainly controlled by incoming radiation and water availability in temperate and bor-
eal forests. Warming-induced prolongations of the growing season in high latitudes are dampened by a limited avail-
ability of light. For peak greenness, interannual variability and trends are dominantly controlled by water availability
and land-use and land-cover change (LULCC) in all regions. Stronger greening trends in boreal forests of Siberia than
in North America are associated with a stronger increase in water availability from melting permafrost soils. Our
findings emphasize that in addition to cold temperatures, water availability is a codominant control for start of grow-
ing season and peak greenness trends at the global scale.
Keywords: browning, dynamic global vegetation model, greening, land-use and land-cover change, model evaluation, phenol-
ogy, remote sensing, vegetation dynamics
Received 16 December 2014 and accepted 2 March 2015
Introduction
Satellite observations demonstrated globally significant
interannual variability and trends of phenology and
greenness in the last three decades (Myneni et al., 1997;
Tucker et al., 2001; Xu et al., 2013; Zeng et al., 2013). The
use of satellite-derived time series of vegetation indices,
such as the normalized difference vegetation index
(NDVI) (Tucker, 1979), to study the timing of changes
in vegetation greenness is usually referred as land sur-
face phenology (De Beurs & Henebry, 2004). Variability
and trends in land surface phenology and greenness
have been associated with regionally different climatic
and environmental controlling factors: Positive trends
in land surface greenness (‘greening’) (Myneni et al.,
1997; Goetz et al., 2005) and phenological changes such
as an earlier start and a lengthening of the growing sea-
son (Tucker et al., 2001; Tateishi & Ebata, 2004; Julien &
Sobrino, 2009) in high-latitude regions have been con-
cordantly associated with warming climate (Lucht
et al., 2002; Menzel et al., 2006; Xu et al., 2013; Keenan
et al., 2014). Nevertheless, also increasing atmospheric
CO
2
and nitrogen deposition can potentially fertilize
vegetation and thus contribute to global greening
trends (Piao et al., 2006; Mao et al., 2012). The CO
2
fer-
tilization effect is supposed to be mainly important in
dry lands (Donohue et al., 2013). On the other hand,
multiple controls have been identified for negative
trends in vegetation greenness (‘browning’) in boreal
forests of North America, and in some temperate and
subtropical grasslands (Goetz et al., 2005; Bi et al., 2013;
De Jong et al., 2013a) such as fire regimes (Goetz et al.,
2005), heat stress (Bunn et al., 2007), forest type (Beck &
Correspondence: Matthias Forkel, tel. +49 3641 576283,
fax +49 3641 577200, e-mail: mforkel@bgc-jena.mpg.de
3414 © 2015 John Wiley & Sons Ltd
Global Change Biology (2015) 21, 3414–3435, doi: 10.1111/gcb.12950
Goetz, 2011), cooling spring temperatures (Wang et al.,
2011), reduced soil moisture, and possibly permafrost
(Barichivich et al., 2014). In the Sahel, greening trends
are discussed in the face of opposing effects of increas-
ing precipitation and increasing land degradation
(Fensholt et al., 2013; Dardel et al., 2014). Land manage-
ment might be also important for end of growing sea-
son in temperate regions (Garonna et al., 2014). For
tropical forests, it has been intensively discussed
whether vegetation index time series have seasonal
dynamics and whether these are driven by variations in
light or water availability (Huete et al., 2006; Samanta
et al., 2010; Morton et al., 2014). Seasonal-to-decadal
dynamics of land surface greenness affect ecosystem
structure (Wolkovich & Cleland, 2010; Fridley, 2012)
and the climate system through changes in albedo, sur-
face roughness, and through exchange of energy, water
and carbon (Bonan, 2008; Richardson et al., 2013). Con-
sequently, it is important to understand the relative
importance of regional competitive explanations and
controlling factors on average spatial patterns, interan-
nual variability and trends in land surface phenology
and greenness. An observation-based identification of
this relative importance is difficult because of several
challenges that limit the use of explanatory data analy-
sis approaches because controlling factors (1) act on dif-
ferent spatial scales (e.g. uniform atmospheric CO
2
increase vs. regional fire events), (2) are temporally cor-
related (e.g. CO
2
and air temperature increase), (3) exhi-
bit nonlinear dynamic interactions (e.g. drought, fire,
land-cover change and succession), or (4) are not read-
ily available from observations (e.g. spatial distributed
observation of permafrost dynamics). On the other
hand, a consistent framework such as dynamic global
vegetation model (DGVM) is not limited by these chal-
lenges if such processes are accurately represented in
the model. DGVMs can be applied to identify the rela-
tive importance of controlling factors such as tempera-
ture, prescription, fire disturbance, CO
2
fertilization,
permafrost dynamics and soil moisture, and land-use
and land-cover change (LULCC) (Piao et al., 2011).
DGVMs were previously applied to identify control-
ling factors for land surface phenology and greenness
(Lucht et al., 2002; Piao et al., 2006; Mao et al., 2012,
2013) but need to be critically evaluated with respect to
the model performance in reproducing observations.
Especially, DGVMs cannot well reproduce observed
phenology (Richardson et al., 2012) and seasonal-to-
decadal dynamics of land surface greenness (Anav
et al., 2013; Murray-Tortarolo et al., 2013). On the other
hand, new phenology models have been recently devel-
oped and parameterized using satellite-derived green-
ness observations (Knorr et al., 2010; St
€
ockli et al., 2011;
Caldararu et al., 2014; Forkel et al., 2014). This new gen-
eration of phenology models describes the temporal
development of canopy greenness and thus follows a
different paradigm than traditional phenology models
that usually simulate specific events of leaf develop-
ment such as budburst or leaf senescence (Richardson
et al., 2012). Forkel et al. (2014) developed and parame-
terized a new phenology model within the LPJmL
(Lund–Potsdam–Jena managed Lands) DGVM that bet-
ter reproduces satellite observations of seasonal-to-dec-
adal dynamics of the fraction of absorbed
photosynthetic active radiation (FAPAR) by consider-
ing effects of temperature, light and water availability.
Using LPJmL with this improved phenology scheme
and additional model developments (Bondeau et al.,
2007; Thonicke et al., 2010; Schaphoff et al., 2013) allows
going beyond an earlier LPJ-based analysis of climate
effects on peak greenness (Lucht et al., 2002) to addi-
tionally quantify the effects of water availability, light,
fire, permafrost, and land-cover dynamics on land sur-
face phenology and greenness.
The estimation of land surface phenology and green-
ness metrics (PGMs) and their interannual and decadal
dynamics from satellite-derived time series is challeng-
ing for several reasons: (1) PGMs such as the start and
end of the growing season (SOS, EOS) are known to con-
trol the annual CO
2
uptake, but they are limited descrip-
tors of site-specific seasonal changes in canopy structure
and plant physiology (Schwartz & Reed, 1999; Studer
et al.,2007;Xiaoet al., 2009; Liang et al., 2011; Fu et al.,
2014b). (2) Satellite data sets differ remarkably in spatial
patterns and temporal dynamics due to different sensor
properties (sun–sensor geometry, spectral and spatial
resolution), observational distortions (cloud and snow
cover, aerosols) and processing algorithms (McCallum
et al., 2010; Fensholt & Proud, 2012; Wang et al., 2012; Ji-
ang et al., 2013; D’Odorico et al., 2014; Guay et al.,2014;
Scheftic et al., 2014). Additionally, the temporal resolu-
tion of satellite data sets and thus of greenness time ser-
ies affects the timing of phenological or trend changes,
but it has been shown that temporal resolution is of
minor importance for the variability in timing in com-
parison with the statistical time series analysis method
(Zhang et al., 2009; Forkel et al., 2013; White et al., 2014).
Thus, (3) the estimation of PGMs from vegetation index
time series is highly sensitive to the chosen analysis
methods. Satellite-derived greenness time series require
usually smoothing and interpolation to exclude short-
term variability and to estimate daily phenology events
from less frequent observations. Therefore, different
smoothing, interpolation, curve-fitting and de tection
methods were developed (hereinafter simply called
‘phenology methods’) (De Beurs & Henebry, 2010).
These methods can result in remarkable differences in
estimated patterns and dynamics of land surface phenol-
© 2015 John Wiley & Sons Ltd, Global Change Biology, 21, 3414–3435
CONTROLS ON PHENOLOGY AND GREENNESS 3415
ogy and greenness (Hird & McDermid, 2009; Musial
et al.,2011;Forkelet al., 2013; Kandasamy et al.,2013;
Geng et al., 2014; White et al., 2014). Thus, it is necessary
to consider the uncertainty from different data sets and
phenology methods for a robust quantification and for
model evaluation of recent patterns and temporal
dynamics of land surface phenology and greenness.
Here, we aimed to (1) to quantify the relative impor-
tance of climatic and ecosystem controlling factors on
average patterns, interannual variability and trends in
land surface phenology and greenness at the regional
and global scale, for which we need to (2) quantify glo-
bal patterns and temporal dynamics of land surface
phenology and greenness by considering uncertainties
from different satellite data sets and phenology meth-
ods, and to (3) evaluate the performance of the LPJmL
DGVM with an improved phenology scheme (Forkel
et al., 2014) in reproducing the observed patterns and
dynamics when considering these uncertainties.
Material and methods
FAPAR data sets
FAPAR time series used here originated from three satel-
lite-based data sets and from LPJmL with two different
phenology schemes (Fig. 1). The GIMMS3g (Global Inven-
tory Modeling and Mapping Studies, 3rd generation) FA-
PAR data set was derived from a harmonized NDVI data
set (Zhu et al., 2013; Pinzon & Tucker, 2014) and covers
fully the years 1982–2011. MODIS FAPAR (Moderate-Reso-
lution Imaging Spectroradiometer) was taken from the
MOD15A2 product (Knyazikhin et al., 1999; USGS, 2001)
and covers the period since February 2000. GL2-VGT2 FA-
PAR (Geoland2 BioPar GEOV1, Vegetation 2) was derived
from SPOT (Satellite Pour l’Observation de la Terre) obser-
vations (Baret et al., 2013). The data set covers originally the
period 1999–2012 based on VGT1 and VGT2 observations.
However, we only use the data set from 2003 onwards
(only VGT2 observations) because the combined data set
has discontinuities at the sensor shift between VGT1 and
VGT2 (Forkel et al., 2014; Horion et al., 2014). FAPAR from
GIMMS3g, MODIS and GL2-VGT2 is defined as black-sky
green canopy instantaneous FAPAR at 10:35, 10:35 and
10:15 solar time, respectively. Instantaneous FAPAR obser-
vations during this time are close approximations of the
daily integrated FAPAR (Baret et al., 2007) and thus compa-
rable with LPJmL model simulations. All satellite-based
data sets were aggregated from their original resolution to
0.5° spatial resolution and to monthly time steps to be com-
parable with LPJmL model simulations. Submonthly FA-
PAR values were aggregated to monthly values using the
maximum value composite approach (Holben, 1986) to
exclude potential remaining low-biased FAPAR values.
Data comparison and model
evaluation using Kling-Gupta
efficiency (KGE)
FAPAR datasets
Phenology methods to extract PGMs
(phenology and greenness metrics)
Satellite-based FAPAR
GIMMS3g (1982-2011)
MODIS (2001-2011)
GL2-VGT2 (2003-2011)
LPJmL FAPAR (1982-2011)
LPJmL-OP
(original phenology)
LPJmL-GSI
(improved phenology)
Factorial model experiments
with LPJmL-GSI
LPJmL
-
noCold
LPJmL-noLight
LPJmL-
noWater
LPJmL-
noLULCC
LPJmL-noPermafrost
LPJmL-
noFire
LPJmL-noCO2
Filling of permanent (winter) gaps
Temporal smoothing and interpolation using
5 methods: LIN, SPL, SSA, DL1, DL2
Calculation of PGMs using 2 approaches:
Trs, Deriv
KGE (data ~ data)
KGE (LPJmL ~ data)
LPJmL model performance within
agreement of datasets?
PGM: SOS, EOS, LOS, POP, PEAK, MGS, MSP, MAU
Wilcoxon rank-rum test,
paired by methods
x 10 methods
x 10 methods
Quantification of effects of a controlling factor on phenology and greenness
Total effect = sqrt(effects on mean, variance, correlation and trend)
eTotal (LPJmL-factor ~ LPJmL-GSI)
Monthly FAPAR time series
aggregated to 0.5° spatial resolution
Dataset-method ensemble of PGMs
10 PGM estimates per dataset
L
I
N
.
T
r
s
L
I
N
.
D
e
r
i
v
S
P
L
.
T
r
s
S
P
L
.
D
e
r
i
v
S
S
A
.
Tr
s
S
S
A
.
D
e
r
i
v
D
L
1
.
T
r
s
D
L
1
.
D
e
r
iv
D
L
2
.
Tr
s
D
L
2
.
D
e
r
iv
Med, IQR
Te
m
p
o
r
a
l
a
g
g
r
e
g
a
t
i
o
n
E
n
s
e
m
b
l
e
s
t
a
t
s
Mean, Sd
Mean, Sd
Mean, Sd
Mean, Sd
Mean, Sd
Med, IQR
Med, IQR
Med, IQR
Med, IQR
Mean, Sd
Med, IQR
x 10 methods
Ensemble median + IQR
x 10 methods
H0: KGE(LPJmL~data) – KGE(data~data) = 0
H1: KGE(LPJmL~data) – KGE(data~data) < 0
–
–
–
–
–
–
Fig. 1 Flow chart of the described data sets and methods. Abbreviations of methods are explained in the Material and Methods section
and in Table S1.
© 2015 John Wiley & Sons Ltd, Global Change Biology, 21, 3414–3435
3416 M. FORKEL et al.
FAPAR and phenology in LPJmL
The LPJmL DGVM simulates carbon and water fluxes and
stocks as a result of natural vegetation dynamics (Sitch
et al., 2003) and has extensions for human land use and
agriculture (Bondeau et al., 2007), fire (Thonicke et al.,
2010), surface albedo and snow cover (Strengers et al.,
2010), and soil hydrology and permafrost dynamics (Schap-
hoff et al., 2013). Additionally, a new phenology scheme for
natural vegetation plant functional types (PFTs) has been
implemented in LPJmL, and FAPAR-related model parame-
ters were optimized against GIMMS3g FAPAR, albedo and
gross primary production data (Forkel et al., 2014). FAPAR
of a PFT in LPJmL depends mainly on annual changing
foliar projective cover (FPC) and the daily phenology status
(PHEN) (Forkel et al., 2014).
We applied two phenology schemes in LPJmL to simulate
PHEN. LPJmL-OP is the original LPJ phenology scheme (Sitch
et al., 2003) and is based on accumulated temperature condi-
tions (i.e. growing degree-days) for summergreen PFTs.
LPJmL-GSI (growing season index) is an alternative and
improved phenology scheme and simulates seasonal leaf
development in response to cold temperature, light, water
availability and heat stress controlling functions:
PHEN
PFT
¼ f
cold;PFT
ðTÞf
light;PFT
ðSWdownÞf
water;PFT
ðWÞ
f
heat;PFT
ðTÞ
ð1Þ
where T is the average daily air temperature, SWdown the
daily short-wave radiation, and W is the percentage water
availability (Forkel et al., 2014).
LPJmL model set-up and factorial model experiments
LPJmL was driven by monthly time series of air temperature
and precipitation from the CRU TS3.1 data set (Harris et al.,
2014), and by monthly short-wave downward and long-wave
net radiation time series from the ERA-Interim reanalysis data
set (Dee et al., 2011). Monthly observed burnt area time series
were prescribed to the fire module as described in Forkel et al.
(2014) to constrain fire simulation by observations. For this,
we used burnt area estimates from the Global Fire Emissions
Database (GFED4) (Giglio et al., 2010) for the period 1996–
2011, from the Alaskan Large Fire Database (Kasischke et al.,
2002; Frames, 2012), and from the Canadian National Fire
Database (Stocks et al., 2002; CFS, 2010) for North America for
the period 1979–1996. To assess LULCC effects, we prescribed
to LPJmL a data set of recent and historic (1700–2005) crop-
land distributions (Fader et al., 2010). All model simulations
and the required model spinup were performed according to
the standard LPJmL modelling protocol (Thonicke et al., 2010;
Schaphoff et al., 2013). Transient model runs were analysed
for the period 1982–2011 for which FAPAR satellite observa-
tions are available.
We performed several experiments with LPJmL-GSI to
quantify the effects of different controlling factors on interan-
nual variability and trends in land surface phenology and
greenness (Table 1). Specifically, we investigated the effects of
seasonal climatic controls (cold temperature, light and water
availability), fire, land-use and land-cover change (LULCC),
permafrost dynamics and CO
2
fertilization by running the
LPJmL-GSI phenology model. In a second step, we ran a series
of model experiments using the same set-up and drivers as
the standard model run but with one factor fixed at a time
Table 1 Overview of factorial model experiments and corresponding effects on FAPAR in LPJmL
LPJmL model
experiment Factor
Effect on FAPAR
in LPJmL
Factorial changes to LPJmL model components
Phenology Permafrost
Agriculture
(land use)
Natural
vegetation Burnt area CO2
LPJmL-GSI Standard – GSI Yes Yes Dynamic Observed Growing
LPJmL-OP –– OP Yes Yes Dynamic Observed Growing
LPJmL-noCold Cold Direct effects
on phenology
status PHEN (daily)
GSI, but
f
cold
= 1
Yes Yes Dynamic Observed Growing
LPJmL-noLight Light GSI, but
f
light
= 1
Yes Yes Dynamic Observed Growing
LPJmL-noWater Water GSI, but
f
water
= 1
Yes Yes Dynamic Observed Growing
LPJmL-noLULCC LULCC Direct effect on
FPC (annual)
GSI Yes Land-use
fractions
fixed to
1982
Maximum
FPC
fixed to
1982
Observed Growing
LPJmL-noPf Permafrost
(Pf)
Indirect effects on
FPC (annual)
GSI No Yes Dynamic Observed Growing
LPJmL-noFire Fire GSI Yes Yes Dynamic No fire Growing
LPJmL-noCO2 CO2 GSI Yes Yes Dynamic Observed Constant
after 1982
© 2015 John Wiley & Sons Ltd, Global Change Biology, 21, 3414–3435
CONTROLS ON PHENOLOGY AND GREENNESS 3417
(Table 1). We fixed the cold temperature, light and water
availability controlling functions of the LPJmL-GSI phenology
scheme to unity to assess the effects of temperature, light and
water availability, respectively. This implies that temperature,
light or water availability does not affect FAPAR phenology in
LPJmL-GSI but can still affect productivity and thus FAPAR
through annual changes in FPC. All other model experiments
directly affected FAPAR dynamics through annual changes in
FPC. We disabled the simulation of soil thermal dynamics to
assess a possible effect of permafrost. This experiment implies
that seasonal soil freezing and thawing does not affect soil
moisture and rooting depth and thus generally increases pro-
ductivity and thus FPC. We made another model run without
simulating any fire activity to assess the possible effect of fire
disturbance. To assess the importance of LULCC, we fixed
land-use and land-cover fractions. For this, cropland fractions
and the maximum FPC of natural vegetation in a grid cell
were kept constant after 1982. This approach implies that the
FPC of PFTs can still change because of competition or mortal-
ity, but the area extent cannot exceed the coverage conditions
of 1982. To assess the effect of CO
2
fertilization, atmospheric
CO
2
fractions were kept constant at 341.22 ppm after 1982 and
thus did not further fertilize photosynthesis and increase FPC.
Finally, we performed one model experiment with the original
LPJmL phenology model (LPJmL-OP) (Sitch et al., 2003) to use
a classical growing degree-day-based phenology model as an
alternative and benchmark in model evaluation.
Phenology methods and trend analysis
Different phenology methods were used to estimate the uncer-
tainty in PGMs that is caused by different smoothing, interpo-
lation and analysis methods (Fig. 1). Although FAPAR time
series from LPJmL model simulations are gap-free and with-
out observational distortions such as in satellite data sets, we
applied all phenology methods also to modelled FAPAR time
series to ensure comparability with estimated PGMs from
satellite data sets. All applied methods are freely available in
the
R software package ‘greenbrown’ (http://greenbrown.
r-forge.r-project.org/).
In our approach, all phenology methods consist of three
steps (Fig. 1): (1) permanent gaps (i.e. usually winter months
in northern regions) were filled in each time series; (2) the time
series were smoothed and interpolated to daily time steps
using five different methods; and (3) PGMs were calculated
from smoothed and daily interpolated time series using two
different approaches. All methods are described with more
details within the Data S1.
In the first step, we filled months with permanent gaps ser-
ies (i.e. gaps that occur in at least 20% of all years during the
same season) with the minimum FAPAR value. This approach
was already used by Beck et al. (2006) to fill missing winter
observation in NDVI time series.
In the second step, we used five different methods for tem-
poral smoothing and for interpolation to daily values (Data
S1). These methods use linear interpolation, spline smoothing
and interpolation (Migliavacca et al., 2011; Musial et al., 2011),
singular spectrum analysis (Golyandina et al., 2001; Mahecha
et al., 2010), or two curve-fitting approaches with double-logis-
tic functions (Beck et al., 2006; Elmore et al., 2012) to derive
daily interpolated and smoothed FAPAR time series.
In the third step, we used the smoothed and daily interpo-
lated time series to estimate start of growing season (SOS) and
end of growing season (EOS) by either using 50% thresholds
on the seasonal greenness curve (approach Trs) (White et al.,
1997) or the derivative of the seasonal curve (approach Deriv)
(Tateishi & Ebata, 2004) (Data S1). Both approaches are based
on the definition of SOS and EOS as the mid-points of spring
greenup and autumn senescence, respectively. We followed
this definition of SOS and EOS (White et al., 1997; Tateishi &
Ebata, 2004), although lower thresholds or extreme values of
the second derivative of the seasonal greenness curve better
agree with phenology transitions observed at the surface
(White et al., 2014). Nevertheless, SOS and EOS definitions
that are based on lower values are more strongly affected from
nonvegetation changes as snow cover or cloud contaminations
and thus less reliable. All other PGMs were derived after-
wards: the length of the growing season (LOS) is the differ-
ence between EOS and SOS. Mean growing season FAPAR
(MGS) is the average FAPAR value from all days between
SOS and EOS. Mean spring (MSP) and mean autumn (MAU)
FAPAR are the average FAPAR values from a period of
20 days around SOS and EOS, respectively. Peak FAPAR
(PEAK) is the maximum FAPAR value of the year from the
smoothed and interpolated curve. The position of the peak
(POP) is the day of the year when PEAK is reached. In sum-
mary, we used ten phenology methods (five smoothing and
interpolation methods in step #2 times two detection
approaches in step #3). Thus, we derive for each data set and
each PGM an ensemble of ten annual time series of land sur-
face phenology and greenness metrics.
Trends in all annual PGM time series were computed based
on linear least square regression with break point detection
(Bai & Perron, 2003; Zeileis et al., 2003; Forkel et al., 2013). The
significance of the trend was estimated using the Mann–Ken-
dall trend test (Mann, 1945; Kendall, 1975).
Data comparison, model evaluation and quantification of
factorial effects
We compared the time series of PGM derived from all satel-
lite data sets to assess their agreement and their uncertainty
with the aim of evaluating the performance and usability of
LPJmL. Specifically, we assessed the agreement of PGM time
series regarding the mean, two measures of interannual vari-
ability (standard deviation and correlation) and overall agree-
ment. For this, we computed the Kling–Gupta efficiency
(KGE) with its components that account for bias, difference in
standard deviation and correlation (Gupta et al., 2009). KGE
ranges between negative infinity (worst agreement) and 1
(perfect agreement) and is defined based on the Euclidean
distance in a 3-dimensional coordinate system of agreement
measures:
KGE ¼ 1
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ða 1Þ
2
þðb 1Þ
2
þðc 1Þ
2
q
ð2Þ
© 2015 John Wiley & Sons Ltd, Global Change Biology, 21, 3414–3435
3418 M. FORKEL et al.
where c is the Pearson correlation coefficient between two
time series. a and b account for the difference in standard
deviation r and in the mean values l between a times series x
and a reference time series r, respectively:
a ¼
r
x
r
r
ð3Þ
b ¼
l
x
l
r
ð4Þ
We computed KGE between two data sets for the same
phenology method. This resulted for each satellite data set
in an ensemble of 20 KGEs (e.g. GIMMS3g compared
against MODIS and GL2-VGT2 using 10 methods each) and
for each LPJmL model experiment in an ensemble of 30
KGEs (i.e. LPJmL compared against three satellite data sets
using 10 methods each). We tested whether the performance
of LPJmL is within the agreement of the data sets by testing
whether the differences between KGE from LPJmL against
satellite data sets and KGE from the cross-comparison of
satellite data sets are significantly < 0 using the Wilcoxon
rank-sum test (paired along phenology methods, Fig. 1). We
also applied the KGE metric to quantify the effect of a fac-
tor on PGM time series in the factorial model experiment.
For this, we computed KGE between a PGM time series
from a factorial model run and the reference LPJmL-GSI
model run for each phenology method. To additionally
quantify the effect on the trend, we extended the KGE met-
ric by a fourth metric d which accounts for the differences
in linear trend slopes s between a times series from a facto-
rial model experiment x and a time series r from the refer-
ence model run:
d ¼
s
x
s
r
ð5Þ
Thus, we are defining the total effect (eTotal), the effect on
the mean (eMean), the effect on the variance of annual values
(eVar), the effect on correlation or interannual dynamic (eCor),
and the effect on the trend (eTrend) of a factor accordingly to
the Kling–Gupta efficiency as the Euclidean distance in a 4-
dimensional space:
eTotal ¼
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
eVar þ eMean þ eCor þ eTrend
p
ð6Þ
where eVar, eMean, eCor and eTrend are defined as the
squared differences between 1 and a, b, c and d, respectively.
To consider the uncertainty of phenology methods, we are
reporting results as the median and interquartile range of an
effect over all phenology methods.
Results
Comparison of land surface phenology and greenness
metrics from data sets and models
Spatial patterns of phenology and greenness metrics
differed remarkably between data sets (Fig. 2). In most
biomes, the mean annual SOS and EOS dates were
detected earlier in the GIMMS3g data set and later in
the GL2-VGT2 data set (Fig. 2, Table 2). Mean annual
SOS differed between satellite data sets by up to
50 days in Savannas and in boreal needle-leaved
summergreen forests. Mean annual SOS and EOS dates
simulated by LPJmL-OP were out of phase in temperate
grasslands and in Savannas. All satellite data sets and
LPJmL-GSI agreed regarding the global patterns of
mean annual peak FAPAR, but LPJmL-OP overesti-
mated peak FAPAR in temperate to arctic regions
(Fig. 2c). These differences between LPJmL-OP
and LPJmL-GSI are related to structural limitations
hampering the optimization of LPJmL-OP (Forkel et al.,
2014).
Estimated PGMs had large differences between phe-
nology methods (Fig. 2, Table 2). For example, mean
annual SOS ranged over almost 80 days for the GL2-
VGT2 data set in boreal needle-leaved summergreen
forests. These differences cannot be solely associated
with distortions of optical remote sensing observations
but with weaknesses on the phenology methods as
well, as similar differences were found for the LPJmL
SOS (DOY)
0 100 200 300
0 100 200 300
EOS (DOY)
–40 –20 0 20 40 60–40 –20 0 20 40 60–40 –20 0 20 40 60
0.0
0.2 0.4 0.6 0.8 1.0
Latitude (°N)Latitude (°N) Latitude (°N)
PEAK (–)
GIMMS3g
MODIS
GL2−VGT2
LPJmL−OP
LPJmL−GSI
(a) (b)
(c)
Fig. 2 Latitudinal gradients of mean annual (a) SOS, (b) EOS and (c) peak FAPAR from data sets and LPJmL. For each data set or
model, the median and the interquartile range of the method ensemble are shown.
© 2015 John Wiley & Sons Ltd, Global Change Biology, 21, 3414–3435
CONTROLS ON PHENOLOGY AND GREENNESS 3419
Table 2 Biome-averaged mean annual start of season (SOS) and end of season (EOS) (in days of year, DOY) from satellite-based data sets (GIMMS3g, MODIS, GL2-VGT2) and
LPJmL phenology models
Biome PGM GIMMS3g MODIS GL2-VGT2 LPJmL-OP LPJmL-GSI
Savannas (N-Hemisphere) SOS 137 8 [129, 153]*** 146 10 [137, 165]** 147 8 [137, 161]** 133 20 [105, 157] 138 5 [130, 146]
EOS 287 9 [270, 301] 286 12 [270, 306] 291 14 [269, 318]* 272 20 [253, 298]*** 280 7 [267, 290]***
Savannas (S-Hemisphere) SOS 261 13 [233, 275]*** 277 9 [261, 292]** 275 12 [253, 293] 291 8 [277, 305]*** 265 10 [246, 273]*
EOS 152 6 [143, 161]*** 161 8 [151, 176] 170 9 [158, 185]*** 139 18 [119, 167]*** 162 14 [146, 183]
Temperate grasslands SOS 113 6 [106, 123]*** 120 6 [112, 132] 139 9 [124, 150]*** 102 5 [98, 112]*** 129 7 [119, 138]
EOS 266 12 [247, 290]** 272 15 [253, 302] 271 12 [252, 290] 289 9 [280, 306]*** 258 14 [244, 281]***
Temperate broadleaved summergreen forests SOS 98 4 [92, 104]*** 104 5 [98, 111] 118 6 [111, 127]*** 88 4 [84, 97]*** 87 5 [77, 95]***
EOS 288 13 [274, 305]*** 294
13 [281, 312] 300 12 [284, 320]*** 280 14 [264, 301]*** 292 17 [275, 316]
Boreal needle-leaved evergreen forests SOS 100 5 [94, 106]*** 113 9 [102, 131] 140 6 [129, 146]*** 118 8 [108, 132] 105 9 [92, 120]**
EOS 286 14 [274, 313]*** 295 15 [280, 325] 306 16 [262, 320]* 292 11 [277, 306] 268 10 [251, 279]***
Boreal needle-leaved summergreen forests SOS 107 16 [85, 129]*** 121 11 [101, 132] 149 22 [120, 199]*** 136 24 [102, 180]* 116 16 [94, 137]
EOS 257 10 [237, 268]** 272 11 [255, 284] 270 27 [212, 301] 275 9 [258, 287]** 260 12 [239, 276]
Tundra SOS 120 21 [90, 147]*** 138 14 [112, 160] 160 8 [149, 178]*** 142 21 [113, 175] 123 16 [99, 146]***
EOS 263 10 [244, 272]*** 276 12 [259, 287] 291 24 [228, 317]** 282 10 [263, 296] 269 12 [247, 282]
Numbers are the mean, standard deviation and the minimum and maximum values (in square brackets) of SOS (or EOS) from 10 phenology methods. Star symbols indicate the
P-value of a two-sided Wilcoxon rank-sum test (paired by phenology method) if the multimethod ensemble of SOS (or EOS) estimates of a data set or model eq
uals the ensemble
of all other satellite-based data sets (null hypothesis) or if it is outside the data set ensemble (alternative hypothesis). P-values are as follows: ***P ≤ 0.001, **0.001 < P ≤ 0.01,
*0.01 < P ≤ 0.05, no symbol for P > 0.05. All biome-averaged values were derived for the Northern Hemisphere (except Savannas). Results in bold font highlight data sets or
models without a significant difference to the other data sets. See Fig. S1a for definitions of biomes.
© 2015 John Wiley & Sons Ltd, Global Change Biology, 21, 3414–3435
3420 M. FORKEL et al.
model. These wide ranges cannot be related to single
methods because the estimated PGMs of a method
depend also on biome and data set, that is a method
might result in an extreme SOS in one biome or data
set, whereas it might result in an average SOS in
another biome or data set (Fig. S5). Thus, the use of a
single phenology method can result in wrong conclu-
sions about land surface phenology and greenness
dynamics in some biomes.
Temporal dynamics and trends in PGMs agreed only
partly between data sets (Fig. 3, Table 3). Peak FAPAR
had significant greening in tundra, boreal forests of
Siberia, and temperate forests of Europe in GIMMS3g
and LPJmL-GSI in 1982–2011. In boreal forests of North
America, peak FAPAR had significant greening in
LPJmL-GSI and negative but nonsignificant trends in the
satellite-based data sets. Nevertheless, in the overlap-
ping period of all data sets (2003–2011), GIMMS3g had
positive peak FAPAR trends, whereas MODIS and GL2-
VGT2 had negative trends, reflecting also a disparity
between data sets for this region. In the Sahel, most
data sets had greening trends (but only for GIMMS3g
significant), which were reproduced by LPJmL-GSI. In
the Amazon, only the GIMMS3g data set had signifi-
cant greening, whereas MODIS, GL2-VGT2 and
LPJmL-GSI had no trends. In the Congo basin, only
MODIS had significant browning, whereas the other
data sets had positive trends (GL2-VGT2) or no trends
(GIMMS3g, LPJmL-GSI). All data sets had (partly sig-
nificant) trends towards earlier SOS in boreal and tem-
perate forests and in the tundra (Fig. 3h–i). Negative
SOS trends in temperate forests of Europe were in
agreement with positive trends in LOS (Fig. S3). We
found no trends in LOS in boreal forests of Siberia
because trends towards earlier SOS were compensated
by trends towards an earlier EOS (Fig. S3). These paral-
lel changes in spring and autumn phenology have also
been observed in some temperate tree species (Fu et al.,
2014a). LPJmL-GSI reproduced the observed SOS,
EOS and LOS trends in these regions. We found no
1985 1990 1995 2000 2005 2010
0.35 0.45 0.55
PEAK (–)
*
***
.
1985 1990 1995 2000 2005 2010
0.65 0.75 0.85
PEAK (–)
***
*
**
1985 1990 1995 2000 2005 2010
0.65 0.70 0.75
PEAK (–)
*
.
.
.
1985 1990 1995 2000 2005 2010
0.72 0.76 0.80
PEAK (–)
**
***
*
*
1985 1990 1995 2000 2005 2010
0.25 0.35 0.45
PEAK (–)
*
1985 1990 1995 2000 2005 2010
0.86 0.90
PEAK (–)
**
1985 1990 1995 2000 2005 2010
0.84 0.88 0.92
PEAK (–)
***
*
.
*
*
1985 1990 1995 2000 2005 2010
50 70 90 110
**
1985 1990 1995 2000 2005 2010
90 110 130 150
SOS (DOY)
SOS (DOY)
**
*
.
*
LPJmL−GSI
GIMMS3g
MODIS
GL2−VGT2
(a) (b) (c)
(d) (e)
(f)
(g) (h) (i)
Fig. 3 Regional-averaged time series and linear trends of (a–g) peak FAPAR and (h–i) SOS. Time series are the multimethod median
and interquartile range. Dashed lines indicate the linear trend for the overlapping period of all data sets (2003–2011). The significance
of the trend in each segment is indicated by star symbols. See Fig. S1b for definitions of regions.
© 2015 John Wiley & Sons Ltd, Global Change Biology, 21, 3414–3435
CONTROLS ON PHENOLOGY AND GREENNESS 3421
significant trends in SOS, EOS or LOS in other regions
(Fig. S3). Nevertheless, SOS trends for one data set had
partly large uncertainties and even opposite trend
directions depending on the phenology method
(Table 3). These results demonstrate the need to use
multiple phenology methods and satellite data sets in
order to robustly provide confidence for the application
of LPJmL to identify controlling factors for land surface
phenology and greenness.
The evaluation of the performance of LPJmL-OP and
LPJmL-GSI with respect to the uncertainty of satellite
data sets and phenology methods showed that LPJmL-
GSI agreed for SOS and EOS in temperate, boreal and
arctic regions better with the satellite-based data sets
than the data sets with each other (Fig. 4, Fig. 5i–k).
Especially in boreal forests and tundra, the KGE of
LPJmL-GSI for SOS, EOS and LOS was in more than
96% of all grid cells within the KGE ensemble of the
satellite data sets. In regions south of 40°N, the KGE
ensemble of LPJmL was usually significantly lower than
the KGE ensemble of each data set. Nevertheless, it is
possible to find always a combination of a satellite data
set and phenology method that attests the model either
a modest or almost perfect performance for SOS (Fig.
S7). For peak FAPAR, LPJmL-GSI had lower KGEs than
the satellite data ensemble in most regions except the
arctic tundra and tropical evergreen forests (Fig. 4)
where satellite data sets had generally the lowest agree-
ments (Fig. 5a, a–h). LPJmL-GSI had the weakest perfor-
mance for peak FAPAR in comparison with the satellite
data sets in the Sahel where the KGE was only in 6% of
all grid cells within the KGE ensemble of the data sets.
In all other regions, the performance of LPJmL-GSI in
reproducing peak FAPAR was in at least 29% of all grid
cells within the agreement of the data sets. Globally, the
null hypothesis that the KGE ensemble for peak FAPAR
and SOS from LPJmL-GSI equals the KGE ensemble of
data set cross-comparisons was accepted in 42% and
74% of all land grid cells, respectively. The KGE of
LPJmL-GSI was for most PGMs and grid cells higher
than the KGE of LPJmL-OP. Low KGEs between satel-
lite data sets and between LPJmL-GSI and satellite data
sets were usually related to weak correlations and sec-
ondly to differences in standard deviation (Fig. 5). In
summary, we demonstrate large uncertainties in the
timing of land surface phenology from different meth-
ods and satellite-based data sets but also highlight the
improved usability of LPJmL-GSI over LPJmL-OP to
explain phenology and greenness dynamics.
Controls on phenology and greenness dynamics
We identified seasonal water limitation and land-
use-driven changes in vegetation composition as the
Table 3 Biome-averaged trends in SOS (days/year)
Biome GIMMS3g MODIS GL2-VGT2 LPJmL-OP LPJmL-GSI
Time period 1982–2011 2001–2011 2003–2011 1982–2011 1982–2011
Savannas (N-Hemisphere) 0.04 0.16 [0.24,
0.26] (3*)
0.35 0.33 [0.96,
0.15] (1*)
0.36 0.4 [0.03, 1.07] (0*) 0.13 0.14 [0.09, 0.41] (5*) 0.09 0.04 [0.01,
0.14] (1*)
Savannas (S-Hemisphere) 0.03 0.1 [0.18,
0.12] (2*)
0.19 0.54 [0.67,
0.99] (1*)
0.64 0.6 [0.33, 1.56] (1*) 0.21 0.09 [0.08, 0.39] (7*) 0.08 0.19 [0.21,
0.39] (4*)
Temperate grasslands 0.07 0.04 [0.13,
0] (2*)
0.39 0.22 [0.1,
0.72] (0*)
0.5 0.16 [0.19, 0.76] (0*) 0.1 0.07 [0.19, 0.01] (6*)
0.1 0.07 [0.22,
0.02] (2*)
Temperate broadleaved
summergreen forests
0.12 0.49 [1.22,
1.31] (7*)
0 0.13 [0.22,
0.18] (0*)
0.32 0.27 [0.94, 0.16] (1*) 0.09 0.04 [0.12, 0.01] (0*) 0.09 0.05 [0.16,
0.01] (1*)
Boreal needle-leaved
evergreen forests
0.11 0.05 [0.23,
0.06] (4*)
0.29 0.26 [0.88,
0.14] (0*)
0.2 0.29 [0.48, 0.31] (0*) 0.12 0.05 [0.2, 0.01] (5*) 0.11 0.04 [0.18,
0.05] (5*)
Boreal needle-leaved
summergreen forests
0.03 0.05 [0.11,
0.04] (1*
)
0.94 0.25 [1.33,
0.49] (7*)
0.09 1.63 [1.31, 3.74] (1*) 0.14 0.11 [0.29, 0.08] (4*) 0.23 0.13 [0.53,
0.09] (5*)
Tundra 0.01 0.05 [0.09,
0.07] (0*)
0.27 0.17 [0.66,
0.02] (3*)
0.1 0.18 [0.33, 0.19] (0*) 0.17 0.08 [0.26, 0.03] (7*) 0.18 0.08 [0.3,
0.07] (8*)
Numbers are the mean, standard deviation and the minimum and maximum values (in square brackets) of SOS trend slopes from 10 phenology detection methods. Numbers in
round brackets indicate the number of methods that resulted in significant SOS trends (* indicates P ≤ 0.05, Mann–Kendall trend test). Results in bold font highlight trends for which
most methods agree in trend direction and for which at least one method indicated a significant trend. See Fig. S1a for definitions of biomes.
© 2015 John Wiley & Sons Ltd, Global Change Biology, 21, 3414–3435
3422 M. FORKEL et al.
globally dominant controls on annual changes in vege-
tation phenology and greenness (Fig. 6, Fig. S6). Water
availability had the highest total effects on peak FAPAR
in latitudes south of 40°N (Fig. 7). Seasonal cold tem-
perature and light limitation determined total effects on
peak FAPAR in regions polewards of 30°N and 30°S,
respectively. A multitude of factors had high total
effects on peak FAPAR in temperate, boreal and arctic
regions north of 40°N, especially through seasonal cold
temperature, seasonal light limitation, LULCC and per-
mafrost. CO
2
fertilization had an effect on peak FAPAR
in all regions but was of minor importance in compari-
son with all other factors. Fire had small total effects on
peak FAPAR at the global scale but dominated region-
ally in fire-prone ecosystems such as Savannas and
parts of boreal forests.
The interannual variability of land surface phenology
was not only affected by seasonal effects of cold tem-
perature, light and water availability, but also by
annual effects such as LULCC and fire (Fig. 6b–c). In
subtropical regions (South America, Sahel, India,
Northern Australia), water availability together with
LULCC and fire had high total effects on SOS, whereas
EOS was more influenced by water availability. In tem-
perate regions (eastern USA, Europe), SOS and EOS
were mostly affected by seasonal cold temperature, sea-
sonal light limitation and LULCC. In agricultural
regions in the central USA, Ukraine and southern Rus-
sia, LULCC was the dominant controlling factor for
SOS and EOS. In boreal and arctic regions, a mixture of
seasonal cold temperature, light, water availability and
permafrost dynamics had high total effects on SOS and
EOS. These results demonstrate the importance of
water availability, LULCC and permafrost dynamics
for annual dynamics in vegetation phenology.
Environmental factors had regionally different effects
on mean, variance, correlation and trends of vegetation
phenology and greenness (Fig. 8). In the tundra, the
variance and correlation of annual peak FAPAR were
mainly affected by seasonal cold temperature, perma-
frost and LULCC (Fig. 8a). Light limitation strongly
influenced the trend in peak FAPAR because LPJmL-
GSI resulted in even stronger peak FAPAR greening
trends if the length of the growing season in tundra
regions would not be limited by radiation. In boreal for-
ests of Siberia, permafrost determined the variance in
annual peak FAPAR, whereas both LULCC and perma-
frost had high effects on trends in peak FAPAR
(Fig. 8b). On the other hand, permafrost dynamics were
less important in boreal forests of North America than
in Siberia: the variance of annual peak FAPAR was
equally affected by a mixture of seasonal cold, light and
water limitation, whereas LULCC had the highest effect
on correlation and trends (Fig. 8c). In temperate forests
–1.5 –1.0 –0.5 0.0 0.5 1.0
Latitude )
SOS.KGE (–)
–1.5 –1.0 –0.5 0.0 0.5 1.0
Latitude ( )
EOS.KGE (–)
–40 –20 0 20 40 60–40 –20 0 20 40 60 –40 –20 0 20 40 60
–1.5 –1.0 –0.5 0.0 0.5 1.0
Latitude ( )
PEAK.KGE (–)
0 0.1 0.3 0.5 0.7 0.9 1
LPJmL−OP
LPJmL−GSI
GIMMS3g
MODIS
GL2−VGT2
(a) (b) (c)
(d) (e) (f)
Fig. 4 Kling–Gupta efficiency for SOS, EOS and peak FAPAR for data sets and LPJmL phenology models. (a–c) Latitudinal gradients
of KGE. For each data set, the median KGE and the range between the 1st quartile and the maximum of the method ensemble are
shown. As KGE can potentially reach until negative infinity, the figures are cut at 1.5. (d–f) P-value maps for the test if the perfor-
mance of LPJmL-GSI is within the agreement of data sets.
© 2015 John Wiley & Sons Ltd, Global Change Biology, 21, 3414–3435
CONTROLS ON PHENOLOGY AND GREENNESS 3423
–1.0 –0.5 0.0 0.5 1.0
KGE PEAK
KGE (LPJmL−GSI ~ data)
KGE (data ~ data)
H0: 42%
KGE (LPJmL−GSI ~ data)
KGE (data ~ data)
H0: 79%
KGE (LPJmL−GSI ~ data)
KGE (data ~ data)
H0: 29%
KGE (LPJmL−GSI ~ data)
KGE (data ~ data)
H0: 41%
–1.0 –0.5 0.0 0.5 1.0
KGE PEAK
KGE (LPJmL−GSI ~ data)
KGE (data ~ data)
H0: 64%
KGE (LPJmL−GSI ~ data)
KGE (data ~ data)
H0: 6%
KGE (LPJmL−GSI ~ data)
KGE (data ~ data)
H0: 59%
KGE (LPJmL−GSI ~ data)
KGE (data ~ data)
H0: 88%
–1.0 –0.5 0.0 0.5 1.0
KGE SOS
KGE (LPJmL−GSI ~ data)
KGE (data ~ data)
H0: 74%
KGE (LPJmL−GSI ~ data)
KGE (data ~ data)
H0: 99%
KGE (LPJmL−GSI ~ data)
KGE (data ~ data)
H0: 75%
KGE components
β
(bias)
α
(standard deviation)
γ
(correlation)
(a)
(b) (c) (d)
(e)
(f) (g)
(h)
(i) (j) (k)
Fig. 5 Kling–Gupta efficiency (KGE) for satellite data comparison and LPJmL-GSI model evaluation for peak FAPAR (a–h) and SOS (i–
k) globally and for different regions. Boxplots are regional distributions of KGE from all grid cells of a region and from the full data-
method ensemble. Plots are limited to 1. Barplots show the regionally averaged (median) components of KGE and are plotted as the
difference between the regional median KGE and a perfect KGE at 1. H0 indicates the percentage of grid cells in a region for which the
null hypothesis, H0: KGE (LPJmL-GSI ~ data) – KGE (data ~ data) = 0, was accepted (one-sided Wilcoxon rank-sum test, paired by
phenology methods, P ≤ 0.05).
© 2015 John Wiley & Sons Ltd, Global Change Biology, 21, 3414–3435
3424 M. FORKEL et al.
of Europe, peak FAPAR was mostly affected by LUL-
CC, which mainly affected the variance, correlation and
trend of annual dynamics (Fig. 8d). In the Sahel,
seasonal water availability and LULCC had both high
effects on the mean, variance, correlation and trends in
peak FAPAR (Fig. 8e). In tropical forest regions, the
R: Water or Pf
G: LULCC, Fire or CO2
B: Cold or Light
R
G
B
(a)
(b)
(c)
Fig. 6 Total effect of different factors on (a) peak FAPAR, (b) SOS and (c) EOS. The maps are red-green-blue composites of the maximum
eTotal of grouped factors. Red colours indicate a high total effect of seasonal water limitation or permafrost, green colours of LULCC, fire
or CO2, and blue colours of seasonal cold or light limitation. In case of black or white colours, all factors have a low or high total effect,
respectively. Shown are the eTotal results from the median of the phenology method ensemble. All eTotal values were mean-centred and
scaled, and histogram stretching was applied to enhance plotting. Grey areas are without vegetation or seasonality was not detected.
© 2015 John Wiley & Sons Ltd, Global Change Biology, 21, 3414–3435
CONTROLS ON PHENOLOGY AND GREENNESS 3425
trends in peak FAPAR were dominantly affected by
water availability (Fig. 8f,g). Although the total effect of
CO
2
fertilization was small in all regions, CO
2
affected
mainly trends in peak FAPAR.
In temperate forests of Europe, the SOS was mainly
affected by seasonal cold temperature and light limita-
tion (Fig. 8h). In boreal forests of Siberia, SOS trends
were additionally affected by permafrost (Fig. 8i) point-
ing to a combined effect of increasing temperatures,
melting permafrost and thus higher water availability.
Light had an additional strong effect on SOS trends
because these trends would be even stronger if light is
not ultimately limiting the growing season in these
high-latitude regions.
On the global scale, two satellite data sets and the
LPJmL-GSI standard simulation had greening trends
and trends towards earlier start of season (but not sig-
nificant) (Fig. 9a–b). These positive trends in peak FA-
PAR and negative trends in SOS are well correlated
with global trends in air temperature and water avail-
ability (Fig. 9c–d). Positive trends in water availability
from LPJmL-GSI are related to positive trends in global
precipitation (Fig. S8). Interestingly, peak FAPAR from
LPJmL-GSI and GL2-VGT2, and SOS from LPJmL-GSI,
GIMMS3g and MODIS had globally stronger correla-
tions with water availability than with temperature
(Fig. 9b). Specifically, the trends and variance of global
peak FAPAR were mostly affected by seasonal cold
temperature and water limitation and LULCC in the
factorial model experiment (Fig. 9e). Correlation and
trends in SOS were in the global average dominated by
seasonal cold temperature and water limitation (Fig. 9f).
These results suggest that increasing water availability
is a codominating control for the global greening and
start of season trends, along with air temperature.
Discussion
Uncertainties in detection and explanation of land surface
phenology and greenness
Our results demonstrated the need to use multiple data
sets and phenology methods for detection and model
evaluation of land surface phenology and greenness
metrics. Moreover, the detection and explanation of
land surface phenology and greenness dynamics is
affected from several sources of uncertainties such as
(1) the used observational response variable (here FA-
PAR) (Walker et al., 2014; White et al., 2014; Wu et al.,
2014), (2) the definition of phenology events, (3) the
phenology method applied, (4) the used observational
data sets with associated errors and differences in pro-
cessing algorithms, and (5) the explanatory model.
We did not account for the uncertainty sources #1
and #2 in this study to ensure the comparability of
LPJmL model results with satellite-based data sets. A
major source of uncertainty or mismatches between dif-
ferent PGM estimates is the definition of phenology
events (especially SOS and EOS) (De Beurs & Henebry,
2010). We followed the definition of SOS and EOS as
the days of half spring greenup and half autumn senes-
cence (White et al., 1997, 2014; Tateishi & Ebata, 2004;
Karlsen et al., 2006) because such observations are less
affected by soil reflectance and snow cover or are less
proved to noise than observations at the spring onset or
autumn offset (Huete et al., 1992; Beck et al., 2006; Del-
bart et al., 2006). Therefore, we avoid drawing conclu-
sions about dynamics or model performance that were
likely affected by such observational limitations.
The use of different phenology methods (uncertainty
#3) is one of the major sources of uncertainty (De Beurs
& Henebry, 2010). Smoothing and interpolation meth-
ods have been developed to exclude short-term vari-
ability and nonvegetation changes from phenology
detection (Jonsson & Eklundh, 2002; Beck et al., 2006;
Delbart et al., 2006). The accuracy of these methods
often depends on the number of missing observations,
land-cover type, data set properties and the magnitude
of short-term and interannual variability (Verbesselt
et al., 2010; Musial et al., 2011; Forkel et al., 2013; Kan-
dasamy et al., 2013; Geng et al., 2014). Additionally, the
approach to detect phenology dates (i.e. thresholds,
extreme values of the derivative of the smoothed time
series, or parameters of fitting functions) introduces
Fig. 7 Latitudinal gradients of the total effect of different fac-
tors on peak FAPAR. Shown is the median and interquartile
range of the phenology method ensemble. eTotal values were
averaged (median value) over 1° bands, and a 7° running med-
ian filter was applied.
© 2015 John Wiley & Sons Ltd, Global Change Biology, 21, 3414–3435
3426 M. FORKEL et al.
further differences (De Beurs & Henebry, 2010; White
et al., 2014). It is possible to identify an optimal phenol-
ogy method for an ecosystem if ground observations
are available (White et al., 2014). Nevertheless, for con-
tinental to global applications an identification of an
optimal phenology method is not straightforward. Phe-
nology methods tuned in a particular area with ground
observation can be too specific and not easily generaliz-
able and usable in another area. This implies that a phe-
nology method might result in a good agreement
between data sets whereas another phenology method
might result in a worse agreement (Fig. S4e). Thus, the
use of a single method may result in erroneous diag-
nostics on phenology dynamics and may undermine
the robustness in analysing data set agreement or
model performance. Unless a single robust method can
be identified from ground observations, the use of
ensembles of state-of-the-art methods for the detection
of land surface phenology and greenness provides an
empirical probabilistic approach to detect PGMs.
The fourth source of uncertainty arises from the vari-
ety of data sets with differences in sensor properties,
observational errors and processing algorithms. Our
results confirm previous findings of negative SOS
trends in Europe, North America and Asia (Julien &
Sobrino, 2009; Hamunyela et al., 2013; Zhang et al.,
2014) and of greening trends from the GIMMS3g data
set (Fensholt & Proud, 2012; Bi et al., 2013; De Jong
et al., 2013a; Xu et al., 2013). The weak agreement
between data sets was more related to low correlations
and variance than to biases (Fig. 5). Low correlations
between PGMs were found previously in Europe
eTotal PEAK
eTotal PEAK
eTotal PEAK
eTotal PEAK
eTotal PEAK
eTotal PEAKeTotal PEAK
Cold
Light
Water
Fire
Pf
LULCC
CO2
1.12
0.57
0.34
0.27
0.65
0.57
0.3
0.0 0.4 0.8
Cold
Light
Water
Fire
Pf
LULCC
CO2
0.54
0.27
0.27
0.15
0.75
1.06
0.12
0.0 0.4 0.8
Cold
Light
Water
Fire
Pf
LULCC
CO2
0.69
0.29
0.24
0.23
0.46
1.24
0.12
0.0 0.2 0.4 0.6 0.8 1.0
Cold
Light
Water
Fir
e
P
f
LULCC
CO2
0.89
0.37
0.42
0.02
0.02
3
0.13
0.0 0.4 0.8
Cold
Light
Water
Fire
LULCC
CO
2
0.02
0
0.54
0.21
0.6
0.08
0.0 0.1 0.2 0.3 0.4 0.5
Cold
Light
Water
Fire
LULCC
CO2
2.52
0
5.49
0.73
51.24
2.35
0246
C
old
Light
Water
Fire
LULCC
CO2
3.58
0
5.32
2.3
100.9
2.56
0246810
eTotal SOS
eTotal SOS
Cold
Light
Water
Fire
Pf
LULCC
CO2
0.75
0.9
0.12
0.01
0.03
0.14
0.05
0.0 0.2 0.4 0.6 0.8
Cold
Light
Wate
r
Fire
Pf
L
ULCC
CO2
0.36
0.25
0.43
0.05
0.39
0.08
0.04
0.0 0.1 0.2 0.3 0.4 0.5
Regional
distribution
of eTotal
Regional
mean
eTotal
Regional averaged
IQR of method
ensemble of eTotal
eTrend
eCor
eVar
eMean
Regional averaged
components of eTotal:
Tundra
Boreal forest Siberia Boreal forest North America
Temperate forest Europe Sahel Amazon
Congo Temperate forest Europe (SOS) Boreal forest Siberia (SOS)
(a) (b) (c)
(d) (e) (f)
(g) (h) (i)
Fig. 8 Regional distributions and averaged total effects with components on (a–g) peak FAPAR and (h–i) SOS.
© 2015 John Wiley & Sons Ltd, Global Change Biology, 21, 3414–3435
CONTROLS ON PHENOLOGY AND GREENNESS 3427
(Atzberger et al., 2013), Mongolia (Miao et al., 2013),
tundra, temperate arid regions and tropical forests
(Fensholt & Proud, 2012), and in parts of the USA and
Mexico (Scheftic et al., 2014). Given these differences
between data sets, the use of a single data set for model
evaluation might result in biased conclusions about
model performance (Anav et al., 2013; Murray-Tortarol-
o et al., 2013). Thus, the use of multiple data sets (Guay
et al., 2014) and their uncertainty is required for a
robust analysis of land surface phenology and green-
ness dynamics.
Finally, the structure, parameters and forcing data of
the used model to explain land surface phenology and
greenness is another uncertainty source (Raupach et al.,
2005; Migliavacca et al., 2012). Differences between
forcing data sets such as precipitation introduce a high
model uncertainty in the simulation of FAPAR interan-
nual variability especially in regions with sparse cover-
age of weather stations (Traore et al., 2014). Thus, the
low performance of LPJmL in the Sahel is probably to a
large extent caused by the uncertainty in the precipita-
tion and land-use data sets. To assess structural uncer-
(a) (b)
(c) (d)
(e) (f)
Fig. 9 Global temporal dynamic and controlling factors for land surface phenology and greenness. Globally averaged time series with
trends in (a) peak FAPAR and (b) SOS anomalies, respectively. SOS anomalies are differences to the mean SOS of each grid cell. SOS
time series from MODIS and GL2-VGT2 were shifted by +4 days to improve the readability. Scatterplots with linear regression and cor-
relation coefficients between (c) peak FAPAR and (d) SOS, and global averaged anomalies (relative to 1982) of air temperature (CRU
TS3.2) and water availability (LPJmL) for satellite data sets and LPJmL-GSI. Global distribution and global averaged total effects with
components on (e) peak FAPAR and (f) SOS. See Fig. 8 for an explanation of (e) and (f).
© 2015 John Wiley & Sons Ltd, Global Change Biology, 21, 3414–3435
3428 M. FORKEL et al.
tainty, we compared results of two phenology modules,
LPJmL-OP and LPJmL-GSI, and demonstrated an
improved ability of the latter for reproducing observed
dynamics. Generally, different model structures from
empirical models (e.g. De Jong et al., 2013b; Barichivich
et al., 2014; Broich et al., 2014), model-data integration
approaches (e.g. St
€
ockli et al., 2011; Caldararu et al.,
2014) to process-oriented models (e.g. Lucht et al., 2002;
Piao et al., 2006; Mao et al., 2013) were used to explain
observed phenology and greenness dynamics. It has
been shown that structural uncertainty is larger than
parameter uncertainty in phenology modelling (Mi-
gliavacca et al., 2012). We reduced the uncertainty of
FAPAR and phenology-related parameters by optimiz-
ing LPJmL-GSI against observations (Forkel et al.,
2014). Consequently, modelled SOS and EOS were
within the uncertainty of data sets in many regions
(Fig. 5). On the other hand, the model performance for
peak FAPAR was usually lower than the data agree-
ment. Peak FAPAR is related to FPC in LPJmL and thus
depends on productivity, allocation, mortality, estab-
lishment, fire and land-use change. These model rou-
tines were not yet improved through parameter
optimization and thus provide potential for further
model development. Other controls on greenness are
not considered in LPJmL such as nutrient availability
(Fisher et al., 2012) or disturbances as insect infestations
and storms (Eklundh et al., 2009; Bright et al., 2013),
changes in grazing (Hilker et al., 2014) or in topography
and soil conditions (Walker et al., 2009; Frost et al.,
2014). However, we assume that these potential drivers,
as for the case of fires, can be important regionally but
are not of global importance for dynamics of land sur-
face greenness and phenology. Increasing nutrient
availability such as increasing nitrogen deposition
might be the reason why LPJmL underestimated peak
FAPAR trends in comparison with GIMMS3g in tem-
perate forests of Europe (De Vries et al., 2006). Never-
theless, although nitrogen deposition changes globally
(Vitousek et al., 1997), nitrogen deposition was only
identified of regional importance for trends in LAI in
another modelling study (Mao et al., 2013). In sum-
mary, our approach considers uncertainties from differ-
ent data sets and phenology methods and allowed us
for the first time to quantify the relative importance of
light and water availability, observed fire activity, per-
mafrost dynamics and LULCC on annual dynamics in
land surface phenology and greenness.
Importance of light, water availability and LULCC for
phenology dynamics
Besides the known effect of seasonal cold temperature
on SOS in temperate and boreal forests (Menzel et al.,
2006; Zhang et al., 2007; Jeong et al., 2011; Wolkovich
et al., 2012; Keenan et al., 2014), we demonstrated addi-
tionally the importance of seasonal light and water
availability for interannual variability and trends in
phenology. The availability and intensity of light affects
spring leaf development especially in understory or
late-successional species (Maeno & Hiura, 2000; Rich-
ardson et al., 2009; K
€
orner & Basler, 2010; Caffarra &
Donnelly, 2011). Moreover, our results suggest that the
warming-induced advancing SOS trend is dampened
by a limited availability of light in boreal forests. Thus,
further warming in temperate and boreal regions might
not necessarily result in a longer growing season
(K
€
orner & Basler, 2010; Richardson et al., 2013).
We found a strong effect of water availability and per-
mafrost on SOS in boreal forests (Fig. 8i). The absence of
permafrost in LPJmL increased water availability which
resulted in stronger advancing SOS trends. Permafrost
soils and the seasonal dynamic of the active layer are
regulating the plant available water in many tundra and
boreal forest regions and thus are affecting species com-
position and productivity (Benninghoff, 1952; Sugimoto
et al., 2002; Schuur et al., 2007) and possibly phenology
(Molau, 1997; Natali et al., 2012). Moreover, it has been
shown that winter precipitation determines spring gree-
nup in northern latitudes (Fu et al., 2014c). The equally
important roles of seasonal light and water availability
on trends and variance in SOS suggest that the tempera-
ture sensitivity of phenology might be overestimated if
these factors are neglected in experimental studies
(Wolkovich et al., 2012).
It is generally known that phenology depends on
land-cover type or species (Cleland et al., 2007; K
€
orner
& Basler, 2010; Richardson et al., 2013). However, the
role of LULCC on interannual dynamics of phenology
was only little studied (Bradley & Mustard, 2008; Davi-
son et al., 2011). Our findings emphasize, to our knowl-
edge for the first time, the control of LULCC on SOS in
temperate regions (Fig. 6b). Therefore, we conclude
that besides climatic drivers, anthropogenic factors
need to be carefully considered to explain the interan-
nual variability and trends of land surface phenology.
Regional controls for greening and browning trends
Seasonal water availability and LULCC had dominant
effects on variance, correlation and trends in peak
greenness in all regions (Fig. 8). In boreal forests and
tundra, permafrost and land-cover change had the
highest effects on variance and trends in peak FAPAR
(Fig. 8a–c), despite temperature. Conversely to previ-
ous studies explaining greening trends in arctic and
boreal regions with warming temperatures (Lucht et al.,
2002; Jia et al., 2009; Bhatt et al., 2013; Xu et al., 2013;
© 2015 John Wiley & Sons Ltd, Global Change Biology, 21, 3414–3435
CONTROLS ON PHENOLOGY AND GREENNESS 3429
Barichivich et al., 2014), our results indicate that this
warming is mediated into greening through cold tem-
perature effects on phenology, and through changes in
permafrost-regulated water availability and increasing
vegetation coverage. The arctic and boreal greening is
associated with an increase and intensification of shrub
cover (Forbes et al., 2010; Myers-Smith et al., 2011; Bern-
er et al., 2013). Warming increases the risk of perma-
frost degradation (Shur & Jorgenson, 2007; Jorgenson
et al., 2010), and the associated increases in water avail-
ability in lowlands and nutrient availability contribute
to greening trends (Walker et al., 2009; Berner et al.,
2013; Raynolds et al., 2013; Frost et al. , 2014). The
importance of these effects could be a potential expla-
nation for the weakening relationship between temper-
ature and greenness dynamics (Piao et al., 2014). The
influence of permafrost dynamics on greening trends
contributes to the divergent continental patterns of less
greening/more browning in boreal forests of North
America than in Eurasia (Bi et al., 2013) because boreal
forests of Eurasia are to larger extent underlain by per-
mafrost.
In temperate forest of Europe, interannual variability
and trends in peak FAPAR were mostly related to
LULCC (Fig. 8d). Indeed, large areas experienced forest
regrowth in the last decades (Fuchs et al., 2013). In cen-
tral and eastern Europe, this was induced through
socio-economic transitions in the former socialist states
from the 1980 to the 1990s: croplands were reforested
(Kozak, 2003; Kuemmerle et al., 2008; Alcantara et al.,
2013) and forests experienced a recovery after damages
by air pollution in the 1980s (Schulze, 1989; Kub
ıkov
a,
1991; Pol
ak et al., 2006; Main-Knorn et al., 2009). Addi-
tionally, disturbance-induced land-cover changes such
as storms and insect infestations damaged especially
the dominant spruce forests (Wermelinger, 2004;
J
€
onsson et al., 2009). This effect can be also seen in satel-
lite-derived vegetation indices (Eklundh et al., 2009).
Although not all of these disturbances that affect land
cover are represented in LPJmL, our results demon-
strate that fixed land-cover conditions of 1982 cannot
explain the observed greening trends in temperate
forests of Europe.
In the Sahel, peak FAPAR was equally affected by
seasonal water availability and LULCC (Fig. 8e). This is
in agreement with previous findings of regreening in
1982–2011 because of increasing precipitation trends
after a period of severe droughts (Eklundh & Olsson,
2003; Herrmann et al., 2005; Hickler et al., 2005; Olsson
et al., 2005). Opposing to these precipitation-induced
greening trends, increased land degradation occurs
(Dardel et al., 2014). Increasing trends in precipitation
and land degradation are accepted as the main controls
for the variability and trends in greenness in the Sahel,
but the relative contribution of these opposing factors is
debated (Fensholt et al., 2013; Brandt et al., 2014a; Dar-
del et al., 2014). Some studies associate the greening
with a recovery of trees (Brandt et al., 2014b), but it has
been shown that changes in tree cover depend highly
on the used satellite data set with diverging results
(Horion et al., 2014). Our results demonstrate that
changes in vegetated area are more important for the
trend in peak FAPAR than seasonal changes in water
availability.
In both tropical forest regions (Amazon and Congo),
seasonal water availability and LULCC were the domi-
nant controls for variance and trends in peak FAPAR
(Fig. 8f –g). According to our results, CO
2
fertilization
was of minor importance although it has been shown
that CO
2
fertilization has an high effect on photosynthe-
sis in tropical forests (Schimel et al., 2014). However,
FAPAR of tropical rainforests is likely to be insensitive
to the long-term CO
2
-driven growth trend because it
already reaches values close to saturation. Although
some data sets have significant trends in vegetation
greenness over the Amazon and the Congo basin
(Fig. 8), the satellite data sets have a poor agreement in
temporal dynamics. Greenness observations over tropi-
cal forest regions are affected by several observational
limitations (Samanta et al., 2012), and seasonal changes
have been identified as artefacts from sun–sensor geom-
etries (Morton et al., 2014). Consequently, previous
hypotheses about a greening of the Amazon during
drought periods because of increased light availability
(Huete et al. , 2006; Myneni et al., 2007; Saleska et al.,
2007) have been falsified (Morton et al., 2014). More-
over, decreased productivity and increased tree mortal-
ity were reported during drought periods (Nepstad
et al., 2007; Phillips et al., 2009; Zhou et al., 2014). Our
results support these findings that water availability
rather than light affects seasonal-to-decadal variations
in peak greenness in topical forests.
Controls for global greening trends and consequences for
prognostic modelling
Our results demonstrate that global greening and phe-
nology trends are mostly associated with (1) annual
changes in land cover and land use, and (2) seasonal
effects of cold temperature and water availability on
phenology (Fig. 9e–f). This finding partly agrees with
previous studies about warming-induced greening
trends in high latitudes conducted at regional scale
(Lucht et al., 2002; Xu et al., 2013). However, we addi-
tionally emphasize the strong contributions of LULCC
and water availability on interannual and decadal vari-
ability of vegetation greenness globally. Thus, greening
trends are linked to an intensification of the global
© 2015 John Wiley & Sons Ltd, Global Change Biology, 21, 3414–3435
3430 M. FORKEL et al.
water cycle (Huntington, 2006) that leads to regional
increasing precipitation (New et al., 2001), soil moisture
(Sheffield & Wood, 2008) or earlier soil thaw dates
(Smith et al., 2004). Land-use changes have global con-
sequences for ecosystem structure, functions and ser-
vices (Foley et al., 2005). Important changes in forest
cover have been reported for the last decade (Hansen
et al., 2013) that should clearly be represented in the in-
terannual variability of land surface phenology and
greenness. Additionally, changes in anthropogenic land
use affected the interannual variability and trends in
land surface phenology and greenness especially in
agricultural regions of Europe, the USA, South-East
Asia, sub-Sahelian Africa and South America (Mueller
et al., 2014; Wang et al., 2014). Although CO
2
fertiliza-
tion clearly affected the trend in global greenness in
1982–2011, the effect of CO
2
fertilization was small in
comparison with seasonal temperature and water
effects, and land-use and land-cover change.
In conclusion, our study reveals spatially and tem-
porally distinct drivers of land surface phenology and
greenness dynamics. The importance of a driver varies
according to temporal scale, region and metric of inter-
est (start of season, end of season, peak greenness).
Thus, future studies should clearly state the investi-
gated aspect of phenology. In temperate and boreal
forests, seasonal light and water availability, in addi-
tion to seasonal cold temperature, mainly control inter-
annual variability and trends of the start of growing
season. In all regions, interannual variability and
trends of peak greenness are driven by seasonal water
limitation and land-use and land-cover change (LUL-
CC). The large importance of LULCC and water avail-
ability on phenology and greenness dynamics requires
a better observation-based quantification and under-
standing of land transitions, natural vegetation dynam-
ics, and water-vegetation couplings to improve
prognostic land surface models. Current prognostic
land surface models have large uncertainties about the
future development of the terrestrial carbon cycle
(Friedlingstein et al., 2006, 2014). These uncertainties
are partly related to an underestimation of the role of
water in comparison with temperature in ecosystem
carbon turnover (Carvalhais et al., 2014) and different
representations of vegetation transitions (Friend et al.,
2014). Our findings emphasize that in addition to limi-
tation by low temperatures, water availability is glob-
ally a codominant control for start of growing season
and greening trends. These results point towards the
reformulation of phenology models contributing to a
better prognostic description of vegetation controls on
the global carbon cycle. The model introduced and
tested here overcomes some of these limitations and
missing drivers.
Acknowledgements
We are grateful to Ranga Myneni and colleagues for providing
the GIMMS3g FAPAR data set. We thank Trevor F. Keenan for
his comments on phenology methods. We thank Werner von
Bloh for help in setting up LPJmL model simulations. MF was
funded from the EU FP7 Carbones project (242316) and con-
ducted this work under the International Max Planck Research
School for Global Biogeochemical Cycles.
References
Alcantara C, Kuemmerle T, Baumann M et al. (2013) Mapping the extent of aban-
doned farmland in Central and Eastern Europe using MODIS time series satellite
data. Environmental Research Letters, 8, 035035.
Anav A, Murray-Tortarolo G, Friedlingstein P, Sitch S, Piao S, Zhu Z (2013) Evalua-
tion of land surface models in reproducing satellite derived leaf area index over
the high-latitude Northern hemisphere. Part II: earth system models. Remote Sens-
ing, 5, 3637–3661.
Atzberger C, Klisch A, Mattiuzzi M, Vuolo F (2013) Phenological metrics derived over
the European continent from NDVI3 g data and MODIS time series. Remote Sens-
ing, 6, 257–284.
Bai J, Perron P (2003) Computation and analysis of multiple structural change mod-
els. Journal of Applied Econometrics, 18,1–22.
Baret F, Hagolle O, Geiger B et al. (2007) LAI, fAPAR and fCover CYCLOPES global
products derived from VEGETATION. Remote Sensing of Environment, 110, 275–
286.
Baret F, Weiss M, Lacaze R, Camacho F, Makhmara H, Pacholcyzk P, Smets B (2013)
GEOV1: LAI and FAPAR essential climate variables and FCOVER global time ser-
ies capitalizing over existing products. Part1: principles of development and pro-
duction. Remote Sensing of Environment, 137, 299–309.
Barichivich J, Briffa KR, Myneni R et al. (2014) Temperature and snow-mediated
moisture controls of summer photosynthetic activity in Northern terrestrial eco-
systems between 1982 and 2011. Remote Sensing, 6, 1390–1431.
Beck PSA, Goetz SJ (2011) Satellite observations of high northern latitude vegetation
productivity changes between 1982 and 2008: ecological variability and regional
differences. Environmental Research Letters, 6, 045501.
Beck PSA, Atzberger C, Høgda KA, Johansen B, Skidmore AK (2006) Improved moni-
toring of vegetation dynamics at very high latitudes: a new method using MODIS
NDVI. Remote Sensing of Environment, 100, 321–334.
Benninghoff WS (1952) Interaction of vegetation and soil frost phenomena. Arctic, 5,
34–44.
Berner LT, Beck PSA, Bunn AG, Goetz SJ (2013) Plant response to climate change
along the forest-tundra ecotone in northeastern Siberia. Global Change Biology, 19,
3449–3462.
Bhatt US, Walker DA, Raynolds MK et al. (2013) Recent declines in warming and
vegetation greening trends over pan-arctic tundra. Remote Sensing, 5
, 4229–
4254.
Bi J, Xu L, Samanta A, Zhu Z, Myneni R (2013) Divergent arctic-boreal vegetation
changes between North America and Eurasia over the Past 30 years. Remote Sens-
ing, 5, 2093–2112.
Bonan GB (2008) Forests and climate change: forcings, feedbacks, and the climate
benefits of forests. Science, 320, 1444–1449.
Bondeau A, Smith PC, Zaehle S et al. (2007) Modelling the role of agriculture for
the 20th century global terrestrial carbon balance. Global Change Biology, 13, 679–
706.
Bradley BA, Mustard JF (2008) Comparison of phenology trends by land cover class:
a case study in the Great Basin, USA. Global Change Biology, 14, 334–346.
Brandt M, Romankiewicz C, Spiekermann R, Samimi C (2014a) Environmental
change in time series – An interdisciplinary study in the Sahel of Mali and Senegal.
Journal of Arid Environments, 105,52–63.
Brandt M, Verger A, Diouf AA, Baret F, Samimi C (2014b) Local vegetation trends in
the sahel of mali and senegal using long time series FAPAR satellite products and
field measurement (1982–2010). Remote Sensing, 6, 2408–2434.
Bright BC, Hicke JA, Meddens AJH (2013) Effects of bark beetle-caused tree mortality
on biogeochemical and biogeophysical MODIS products. Journal of Geophysical
Research: Biogeosciences, 118, 974–982.
Broich M, Huete A, Tulbure MG et al. (2014) Land surface phenological response to
decadal climate variability across Australia using satellite remote sensing. Biogeo-
sciences, 11, 5181–5198.
© 2015 John Wiley & Sons Ltd, Global Change Biology, 21, 3414–3435
CONTROLS ON PHENOLOGY AND GREENNESS 3431
Bunn AG, Goetz SJ, Kimball JS, Zhang K (2007) Northern high-latitude ecosystems
respond to climate change. EOS, Transactions American Geophysical Union, 88, 333.
Caffarra A, Donnelly A (2011) The ecological significance of phenology in four differ-
ent tree species: effects of light and temperature on bud burst. International Journal
of Biometeorology, 55, 711–721.
Caldararu S, Purves DW, Palmer PI (2014) Phenology as a strategy for carbon opti-
mality: a global model. Biogeosciences, 11, 763–778.
Carvalhais N, Forkel M, Khomik M et al. (2014) Global covariation of carbon turnover
times with climate in terrestrial ecosystems. Nature, 514, 213–217.
CFS (2010) Canadian large fire database. Natural Resources Canada, Canadian Forest
Service (ed Service CF).
Cleland EE, Chuine I, Menzel A, Mooney HA, Schwartz MD (2007) Shifting plant
phenology in response to global change. Trends in Ecology & Evolution, 22, 357–365.
Dardel C, Kergoat L, Hiernaux P, Grippa M, Mougin E, Ciais P, Nguyen C-C (2014)
Rain-use-efficiency: what it tells us about the conflicting Sahel greening and Sah-
elian paradox. Remote Sensing, 6, 3446–3474.
Davison JE, Breshears DD, van Leeuwen WJD, Casady GM (2011) Remotely sensed
vegetation phenology and productivity along a climatic gradient: on the value of
incorporating the dimension of woody plant cover. Global Ecology and Biogeogra-
phy, 20, 101–113.
De Beurs KM, Henebry GM (2004) Land surface phenology, climatic variation, and
institutional change: analyzing agricultural land cover change in Kazakhstan.
Remote Sensing of Environment, 89, 497–509.
De Beurs KM, Henebry GM (2010) Spatio-temporal statistical methods for modelling
land surface phenology. In: Phenological Research (eds Hudson IL, Keatley MR), pp.
177–208. Springer, Netherlands.
De Jong R, Verbesselt J, Zeileis A, Schaepman M (2013a) Shifts in global vegetation
activity trends. Remote Sensing, 5, 1117–1133.
De Jong R, Schaepman ME, Furrer R, de Bruin S, Verburg PH (2013b) Spatial relation-
ship between climatologies and changes in global vegetation activity. Global
Change Biology, 19, 1953–1964.
De Vries W, Reinds GJ, Gundersen P, Sterba H (2006) The impact of nitrogen deposi-
tion on carbon sequestration in European forests and forest soils. Global Change
Biology, 12, 1151–
1173.
Dee DP, Uppala SM, Simmons AJ et al. (2011) The ERA-Interim reanalysis: configura-
tion and performance of the data assimilation system. Quarterly Journal of the Royal
Meteorological Society, 137, 553–597.
Delbart N, Le Toan T, Kergoat L, Fedotova V (2006) Remote sensing of spring phenol-
ogy in boreal regions: a free of snow-effect method using NOAA-AVHRR and
SPOT-VGT data (1982–2004). Remote Sensing of Environment, 101,52–62.
D’Odorico P, Gonsamo A, Pinty B, Gobron N, Coops N, Mendez E, Schaepman ME
(2014) Intercomparison of fraction of absorbed photosynthetically active radiation
products derived from satellite data over Europe. Remote Sensing of Environment,
142, 141–154.
Donohue RJ, Roderick ML, McVicar TR, Farquhar GD (2013) Impact of CO2 fertiliza-
tion on maximum foliage cover across the globe’s warm, arid environments. Geo-
physical Research Letters, 40, 3031–3035.
Eklundh L, Olsson L (2003) Vegetation index trends for the African Sahel 1982–1999.
Geophysical Research Letters, 30,1–4.
Eklundh L, Johansson T, Solberg S (2009) Mapping insect defoliation in Scots pine
with MODIS time-series data. Remote Sensing of Environment, 113, 1566–1573.
Elmore AJ, Guinn SM, Minsley BJ, Richardson AD (2012) Landscape controls on the
timing of spring, autumn, and growing season length in mid-Atlantic forests. Glo-
bal Change Biology, 18, 656–674.
Fader M, Rost S, M
€
uller C, Bondeau A, Gerten D (2010) Virtual water content of tem-
perate cereals and maize: present and potential future patterns. Journal of Hydrol-
ogy, 384, 218–231.
Fensholt R, Proud SR (2012) Evaluation of Earth Observation based global long term
vegetation trends — Comparing GIMMS and MODIS global NDVI time series.
Remote Sensing of Environment, 119, 131– 147.
Fensholt R, Rasmussen K, Kaspersen P, Huber S, Horion S, Swinnen E (2013) Assess-
ing land degradation/recovery in the African Sahel from long-term earth observa-
tion based primary productivity and precipitation relationships. Remote Sensing, 5,
664–686.
Fisher JB, Badgley G, Blyth E (2012) Global nutrient limitation in terrestrial vegeta-
tion. Global Biogeochemical Cycles,
26, GB3007.
Foley JA, DeFries R, Asner GP et al. (2005) Global consequences of land use. Science,
309, 570–574.
Forbes BC, Macias Fauria M, Zetterberg P (2010) Russian Arctic warming and
‘greening’ are closely tracked by tundra shrub willows. Global Change Biology, 16,
1542–1554.
Forkel M, Carvalhais N, Verbesselt J, Mahecha M, Neigh C, Reichstein M (2013)
Trend change detection in NDVI time series: effects of inter-annual variability and
methodology. Remote Sensing, 5, 2113–2144.
Forkel M, Carvalhais N, Schaphoff S, von Bloh W, Migliavacca M, Thurner M, Tho-
nicke K (2014) Identifying environmental controls on vegetation greenness phenol-
ogy through model–data integration. Biogeosciences, 11, 7025–7050.
Frames (2012) Alaska large fire database.
Fridley JD (2012) Extended leaf phenology and the autumn niche in deciduous forest
invasions. Nature, 485, 359–362.
Friedlingstein P, Cox P, Betts R et al. (2006) Climate-carbon cycle feedback
analysis: results from the C 4 MIP model intercomparison. Journal of Climate, 19,
3337–3353.
Friedlingstein P, Meinshausen M, Arora VK, Jones CD, Anav A, Liddicoat SK, Knutti
R (2014) Uncertainties in CMIP5 climate projections due to carbon cycle feedbacks.
Journal of Climate, 27, 511–526.
Friend AD, Lucht W, Rademacher TT et al. (2014) Carbon residence time dominates
uncertainty in terrestrial vegetation responses to future climate and atmospheric
CO2. Proceedings of the National Academy of Sciences, 111, 3280–3285.
Frost GV, Epstein HE, Walker DA (2014) Regional and landscape-scale variability of
Landsat-observed vegetation dynamics in northwest Siberian tundra. Environmen-
tal Research Letters, 9, 025004.
Fu YSH, Campioli M, Vitasse Y et al. (2014a) Variation in leaf flushing date influences
autumnal senescence and next year’s flushing date in two temperate tree species.
Proceedings of the National Academy of Sciences, 111, 7355–7360.
Fu YH, Piao S, Op de Beeck M et al. (2014b) Recent spring phenology shifts in western
Central Europe based on multiscale observations. Global Ecology and Biogeography,
23, 1255–
1263.
Fu YH, Piao S, Zhao H et al. (2014c) Unexpected role of winter precipitation in deter-
mining heat requirement for spring vegetation green-up at northern middle and
high latitudes. Global Change Biology, 20, 3743–3755.
Fuchs R, Herold M, Verburg PH, Clevers JGPW (2013) A high-resolution and harmo-
nized model approach for reconstructing and analysing historic land changes in
Europe. Biogeosciences, 10, 1543–1559.
Garonna I, de Jong R, de Wit AJW, M
€
ucher CA, Schmid B, Schaepman ME (2014)
Strong contribution of autumn phenology to changes in satellite-derived growing
season length estimates across Europe (1982–2011). Global Change Biology, 20,
3457–3470.
Geng L, Ma M, Wang X, Yu W, Jia S, Wang H (2014) Comparison of eight techniques
for reconstructing multi-satellite sensor time-series NDVI data sets in the Heihe
river Basin, China. Remote Sensing, 6, 2024–2049.
Giglio L, Randerson JT, van der Werf GR, Kasibhatla PS, Collatz GJ, Morton DC, De-
Fries RS (2010) Assessing variability and long-term trends in burned area by merg-
ing multiple satellite fire products. Biogeosciences, 7, 1171–1186.
Goetz SJ, Bunn AG, Fiske GJ, Houghton RA (2005) Satellite-observed photosynthetic
trends across boreal North America associated with climate and fire disturbance.
Proceedings of the National Academy of Sciences of the United States of America, 102,
13521–13525.
Golyandina N, Nekrutkin VV, Zhigljavskiy AA (2001) Analysis of Time Series
Structure: SSA and Related Techniques. Chapman & Hall/CRC, Boca Raton, Fla,
xii, pp. 305.
Guay KC, Beck PSA, Berner LT, Goetz SJ, Baccini A, Buermann W (2014) Vegetation
productivity patterns at high northern latitudes: a multi-sensor satellite data
assessment. Global Change Biology, 20, 3147–3158.
Gupta HV, Kling H, Yilmaz KK, Martinez GF (2009) Decomposition of the mean
squared error and NSE performance criteria: implications for improving hydrolog-
ical modelling. Journal of Hydrology, 377,80–91.
Hamunyela E, Verbesselt J, Roerink G, Herold M (2013) Trends in spring phenology
of western European deciduous forests. Remote Sensing, 5, 6159–6179.
Hansen MC, Potapov PV, Moore R et al. (2013) High-resolution global maps of 21st-
century forest cover change. Science, 342, 850–853.
Harris I, Jones PD, Osborn TJ, Lister DH (2014) Updated high-resolution grids of
monthly climatic observations
– the CRU TS3.10 Dataset. International Journal of
Climatology, 34, 623–642.
Herrmann SM, Anyamba A, Tucker CJ (2005) Recent trends in vegetation dynamics
in the African Sahel and their relationship to climate. Global Environmental Change,
15, 394–404.
Hickler T, Eklundh L, Seaquist JW et al. (2005) Precipitation controls Sahel greening
trend. Geophysical Research Letters, 32, L21415.
Hilker T, Natsagdorj E, Waring RH, Lyapustin A, Wang Y (2014) Satellite observed
widespread decline in Mongolian grasslands largely due to overgrazing. Global
Change Biology, 20, 418–428.
© 2015 John Wiley & Sons Ltd, Global Change Biology, 21, 3414–3435
3432 M. FORKEL et al.
Hird JN, McDermid GJ (2009) Noise reduction of NDVI time series: an empirical com-
parison of selected techniques. Remote Sensing of Environment, 113, 248–258.
Holben BN (1986) Characteristics of maximum-value composite images from tempo-
ral AVHRR data. International Journal of Remote Sensing, 7, 1417–1434.
Horion S, Fensholt R, Tagesson T, Ehammer A (2014) Using earth observation-based
dry season NDVI trends for assessment of changes in tree cover in the Sahel. Inter-
national Journal of Remote Sensing, 35, 2493–2515.
Huete AR, Hua G, Qi J, Chehbouni A, van Leeuwen WJD (1992) Normalization of
multidirectional red and NIR reflectances with the SAVI. Remote Sensing of Envi-
ronment, 41, 143–154.
Huete AR, Didan K, Shimabukuro YE et al. (2006) Amazon rainforests green-up with
sunlight in dry season. Geophysical Research Letters, 33, L06405.
Huntington TG (2006) Evidence for intensification of the global water cycle: review
and synthesis. Journal of Hydrology, 319,83– 95.
Jeong S-J, Ho C-H, Gim H-J, Brown ME (2011) Phenology shifts at start vs. end of
growing season in temperate vegetation over the Northern Hemisphere for the
period 1982–2008. Global Change Biology, 17, 2385–2399.
Jia GJ, Epstein HE, Walker DA (2009) Vegetation greening in the canadian arctic
related to decadal warming. Journal of Environmental Monitoring, 11, 2231–2238.
Jiang N, Zhu W, Zheng Z, Chen G, Fan D (2013) A comparative analysis between
GIMSS NDVIg and NDVI3 g for monitoring vegetation activity change in the
Northern hemisphere during 1982–2008. Remote Sensing, 5, 4031–4044.
Jonsson P, Eklundh L (2002) Seasonality extraction by function fitting to time-series of
satellite sensor data. IEEE Transactions on Geoscience and Remote Sensing, 40,1824–1832.
J
€
onsson AM, Appelberg G, Harding S, B
€
arring L (2009) Spatio-temporal impact of cli-
mate change on the activity and voltinism of the spruce bark beetle, Ips typogra-
phus. Global Change Biology, 15, 486– 499.
Jorgenson MT, Romanovsky VE, Harden JW et al.
(2010) Resilience and vulnerability
of permafrost to climate change. Canadian Journal of Forest Research - Revue Canadi-
enne de Recherche Foresti{
e}re, 40, 1219–1236.
Julien Y, Sobrino JA (2009) Global land surface phenology trends from GIMMS data-
base. International Journal of Remote Sensing, 30, 3495–3513.
Kandasamy S, Baret F, Verger A, Neveux P, Weiss M (2013) A comparison of meth-
ods for smoothing and gap filling time series of remote sensing observations –
application to MODIS LAI products. Biogeosciences, 10, 4055–4071.
Karlsen SR, Elvebakk A, Høgda KA, Johansen B (2006) Satellite-based mapping of the
growing season and bioclimatic zones in Fennoscandia. Global Ecology and Biogeog-
raphy, 15, 416–430.
Kasischke ES, Williams D, Barry D (2002) Analysis of the patterns of large fires in the
boreal forest region of Alaska. International Journal Of Wildland Fire, 11, 131–144.
Keenan TF, Gray J, Friedl MA et al. (2014) Net carbon uptake has increased through
warming-induced changes in temperate forest phenology. Nature Climate Change,
4, 598–604.
Kendall MG (1975) Rank Correlation Methods. Griffin, London.
Knorr W, Kaminski T, Scholze M, Gobron N, Pinty B, Giering R, Mathieu P-P (2010)
Carbon cycle data assimilation with a generic phenology model. Journal of Geophys-
ical Research: Biogeosciences, 115, G04017.
Knyazikhin Y, Glassy J, Privette JL et al. (1999) MODIS Leaf Area Index (LAI) and
Fraction of Photosynthetically Active Radiation Absorbed by Vegetation (FPAR)
Product (MOD15) algorithm theoretical basis document.
K
€
orner C, Basler D (2010) Phenology under global warming. Science, 327, 1461–1462.
Kozak J (2003) Forest cover change in the Western Carpathians in the past 180 years.
Mountain Research and Development, 23, 369–375.
Kub
ıkov
a J (1991) Forest dieback in Czechoslovakia. Vegetatio, 93 , 101–108.
Kuemmerle T, Hostert P, Radeloff VC, van der Linden S, Perzanowski K, Kruhlov I
(2008) Cross-border comparison of post-socialist farmland abandonment in the
carpathians. Ecosystems, 11, 614–628.
Liang L, Schwartz MD, Fei S (2011) Validating satellite phenology through intensive
ground observation and landscape scaling in a mixed seasonal forest. Remote Sens-
ing of Environment, 115, 143–157.
Lucht W, Prentice IC, Myneni RB et al. (2002) Climatic control of the high-latitude
vegetation greening trend and pinatubo effect. Science, 296, 1687–1689.
Maeno H, Hiura T (2000) The effect of leaf phenology of overstory trees on the repro-
ductive success of an understory shrub, Staphylea bumalda DC. Canadian Journal
of Botany, 78, 781–785.
Mahecha MD, F
€
urst LM, Gobron N, Lange H (2010) Identifying multiple spatiotem-
poral patterns: a refined view on terrestrial photosynthetic activity. Pattern Recog-
nition Letters, 31, 2309–2317.
Main-Knorn M, Hostert P, Kozak J, Kuemmerle T (2009) How pollution legacies and
land use histories shape post-communist forest cover trends in the Western Carpa-
thians. Forest Ecology and Management, 258,60–70.
Mann HB (1945) Nonparametric tests against trend. Econometrica, 13, 245–259.
Mao J, Shi X, Thornton PE, Piao S, Wang X (2012) Causes of spring vegetation growth
trends in the northern mid–high latitudes from 1982 to 2004. Environmental
Research Letters, 7, 014010.
Mao J, Shi X, Thornton PE, Hoffman FM, Zhu Z, Myneni RB (2013) Global latitudinal-
asymmetric vegetation growth trends and their driving mechanisms: 1982–2009.
Remote Sensing, 5, 1484–1497.
McCallum I, Wagner W, Schmullius C, Shvidenko A, Obersteiner M, Fritz S, Nilsson
S (2010) Comparison of four global FAPAR datasets over Northern Eurasia for the
year 2000. Remote Sensing of Environment, 114, 941–949.
Menzel A, Sparks TH, Estrella N et al. (2006) European phenological response to cli-
mate change matches the warming pattern. Global Change Biology, 12, 1969–1976.
Miao L, Luan Y, Luo X
et al. (2013) Analysis of the phenology in the Mongolian pla-
teau by inter-comparison of global vegetation datasets. Remote Sensing, 5, 5193–
5208.
Migliavacca M, Galvagno M, Cremonese E et al. (2011) Using digital repeat photogra-
phy and eddy covariance data to model grassland phenology and photosynthetic
CO2 uptake. Agricultural and Forest Meteorology, 151, 1325–1337.
Migliavacca M, Sonnentag O, Keenan TF, Cescatti A, O’Keefe J, Richardson AD
(2012) On the uncertainty of phenological responses to climate change, and impli-
cations for a terrestrial biosphere model. Biogeosciences, 9, 2063–2083.
Molau U (1997) Responses to natural climatic variation and experimental warming in
two tundra plant species with contrasting life forms: cassiope tetragona and
Ranunculus nivalis. Global Change Biology, 3,97–107.
Morton DC, Nagol J, Carabajal CC et al. (2014) Amazon forests maintain consistent
canopy structure and greenness during the dry season. Nature, 506, 221–224.
Mueller T, Dressler G, Tucker CJ et al. (2014) Human land-use practices lead to global
long-term increases in photosynthetic capacity. Remote Sensing, 6, 5717–5731.
Murray-Tortarolo G, Anav A, Friedlingstein P et al. (2013) Evaluation of land surface
models in reproducing satellite-derived LAI over the high-latitude Northern hemi-
sphere. Part I: uncoupled DGVMs. Remote Sensing, 5, 4819–4838.
Musial JP, Verstraete MM, Gobron N (2011) Technical Note: comparing the effective-
ness of recent algorithms to fill and smooth incomplete and noisy time series.
Atmospheric Chemistry and Physics, 11, 7905–7923.
Myers-Smith IH, Forbes BC, Wilmking M et al. (2011) Shrub expansion in tundra eco-
systems: Dynamics, impacts and research priorities. Environmental Research Letters,
6, 045509.
Myneni RB, Keeling CD, Tucker CJ, Asrar G, Nemani RR (1997) Increased
plant growth in the northern high latitudes from 1981 to 1991. Nature, 386,
698–702.
Myneni RB, Yang W, Nemani RR et al. (2007) Large seasonal swings in leaf area of
Amazon rainforests. Proceedings of the National Academy of Sciences, 104, 4820–4823.
Natali SM, Schuur EAG, Rubin RL (2012) Increased plant productivity in Alaskan
tundra as a result of experimental warming of soil and permafrost. Journal of Ecol-
ogy, 100, 488–498.
Nepstad DC, Tohver IM, Ray D, Moutinho P, Cardinot G (2007) Mortality of large
trees and lianas following experimental drought in an Amazon forest. Ecology, 88,
2259–2269.
New M, Todd M, Hulme M, Jones P (2001) Precipitation measurements and trends in
the twentieth century. International Journal of Climatology, 21, 1889–1922.
Olsson L, Eklundh L, Ard
€
o J (2005) A recent greening of the Sahel—trends, patterns
and potential causes. Journal of Arid Environments, 63, 556–566.
Phillips OL, Arag
~
ao LEOC, Lewis SL et al. (2009) Drought sensitivity of the amazon
rainforest. Science, 323, 1344–1347.
Piao S, Friedlingstein P, Ciais P, Zhou L, Chen A (2006) Effect of climate and CO2
changes on the greening of the Northern Hemisphere over the past two decades.
Geophysical Research Letters, 33, L23402.
Piao S, Wang X, Ciais P, Zhu B, Wang T, Liu J (2011) Changes in satellite-derived veg-
etation growth trend in temperate and boreal Eurasia from 1982 to 2006. Global
Change Biology, 17, 3228–3239.
Piao S, Nan H, Huntingford C et al. (2014) Evidence for a weakening relationship
between interannual temperature variability and northern vegetation activity. Nat-
ure Communications, 5.
Pinzon JE, Tucker CJ (2014) A non-stationary 1981–2012 AVHRR NDVI3 g time ser-
ies. Remote Sensing, 6, 6929–6960.
Pol
ak T, Albrechtov
a J, Rock BN (2006) Bud development types as a new macroscopic
marker of Norway spruce decline and recovery processes along a mountainous
pollution gradient. Forestry, 79, 425–437.
Raupach MR, Rayner PJ, Barrett DJ et al. (2005) Model-data synthesis in terrestrial
carbon observation: methods, data requirements and data uncertainty specifica-
tions. Global Change Biology,
11, 378–397.
© 2015 John Wiley & Sons Ltd, Global Change Biology, 21, 3414–3435
CONTROLS ON PHENOLOGY AND GREENNESS 3433
Raynolds MK, Walker DA, Verbyla D, Munger CA (2013) Patterns of change within a
tundra landscape: 22-year landsat NDVI trends in an area of the Northern foothills
of the Brooks range, Alaska. Arctic, antarctic, and alpine research, 45, 249–260.
Richardson AD, Braswell BH, Hollinger DY, Jenkins JP, Ollinger SV (2009) Near-sur-
face remote sensing of spatial and temporal variation in canopy phenology. Ecolog-
ical Applications, 19, 1417–1428.
Richardson AD, Anderson RS, Arain MA et al. (2012) Terrestrial biosphere models
need better representation of vegetation phenology: results from the North Ameri-
can Carbon Program Site Synthesis. Global Change Biology, 18, 566–584.
Richardson AD, Keenan TF, Migliavacca M, Ryu Y, Sonnentag O, Toomey M (2013)
Climate change, phenology, and phenological control of vegetation feedbacks to
the climate system. Agricultural and Forest Meteorology, 169, 156–173.
Saleska SR, Didan K, Huete AR, da Rocha HR (2007) Amazon forests green-up during
2005 drought. Science, 318, 612.
Samanta A, Ganguly S, Hashimoto H et al. (2010) Amazon forests did not green-up
during the 2005 drought. Geophysical Research Letters, 37, L05401.
Samanta A, Ganguly S, Vermote E, Nemani RR, Myneni RB (2012) Why is remote
sensing of amazon forest greenness so challenging? Earth Interactions, 16,1–14.
Schaphoff S, Heyder U, Ostberg S, Gerten D, Heinke J, Lucht W (2013) Contribution
of permafrost soils to the global carbon budget. Environmental Research Letters, 8,
014026.
Scheftic W, Zeng X, Broxton P, Brunke M (2014) Intercomparison of seven NDVI
products over the United States and Mexico. Remote Sensing, 6, 1057–1084.
Schimel D, Stephens BB, Fisher JB (2014) Effect of increasing CO2 on the terrestrial
carbon cycle. Proceedings of the National Academy of Sciences, 112, 436–441.
Schulze E-D (1989) Air pollution and forest decline in a spruce (Picea abies) forest.
Science, 244, 776–783.
Schuur EAG, Crummer KG, Vogel JG, Mack MC (2007) Plant species composition
and productivity following permafrost thaw and thermokarst in Alaskan tundra.
Ecosystems, 10, 280–292.
Schwartz MD, Reed BC (1999) Surface phenology and satellite sensor-derived onset
of greenness: an initial comparison. International Journal of Remote Sensing, 20
,
3451–3457.
Sheffield J, Wood EF (2008) Global trends and variability in soil moisture and drought
characteristics, 1950–2000, from observation-driven simulations of the terrestrial
hydrologic cycle. Journal of Climate, 21, 432–458.
Shur YL, Jorgenson MT (2007) Patterns of permafrost formation and degradation in
relation to climate and ecosystems. Permafrost and Periglacial Processes, 18,7–19.
Sitch S, Smith B, Prentice IC et al. (2003) Evaluation of ecosystem dynamics, plant
geography and terrestrial carbon cycling in the LPJ dynamic global vegetation
model. Global Change Biology, 9, 161–185.
Smith NV, Saatchi SS, Randerson JT (2004) Trends in high northern latitude soil
freeze and thaw cycles from 1988 to 2002. Journal of Geophysical Research: Atmo-
spheres, 109, D12101.
St
€
ockli R, Rutishauser T, Baker I, Liniger MA, Denning AS (2011) A global reanalysis
of vegetation phenology. Journal of Geophysical Research, 116, G03020.
Stocks BJ, Mason JA, Todd JB et al. (2002) Large forest fires in Canada, 1959–1997.
Journal of Geophysical Research, 107, 8149.
Strengers BJ, M
€
uller C, Schaeffer M et al. (2010) Assessing 20th century climate–vege-
tation feedbacks of land-use change and natural vegetation dynamics in a fully
coupled vegetation–climate model. International Journal of Climatology, 30, 2055–
2065.
Studer S, St
€
ockli R, Appenzeller C, Vidale PL (2007) A comparative study of satellite
and ground-based phenology. International Journal of Biometeorology, 51, 405–414.
Sugimoto A, Yanagisawa N, Naito D, Fujita N, Maximov TC (2002) Importance of
permafrost as a source of water for plants in east Siberian taiga. Ecological Research,
17, 493–503.
Tateishi R, Ebata M (2004) Analysis of phenological change patterns using 1982–2000
Advanced Very High Resolution Radiometer (AVHRR) data. International Journal
of Remote Sensing, 25, 2287–
2300.
Thonicke K, Spessa A, Prentice IC, Harrison SP, Dong L, Carmona-Moreno C (2010)
The influence of vegetation, fire spread and fire behaviour on biomass burning
and trace gas emissions: results from a process-based model. Biogeosciences, 7,
1991–2011.
Traore AK, Ciais P, Vuichard N et al. (2014) Evaluation of the ORCHIDEE ecosystem
model over Africa against 25 years of satellite-based water and carbon measure-
ments. Journal of Geophysical Research: Biogeosciences, 119, 2014JG002638.
Tucker CJ (1979) Red and photographic infrared linear combinations for monitoring
vegetation. Remote Sensing of Environment, 150, 127–150.
Tucker CJ, Slayback DA, Pinzon JE, Los SO, Myneni RB, Taylor MG (2001) Higher
northern latitude normalized difference vegetation index and growing season
trends from 1982 to 1999. International Journal of Biometeorology, 45, 184–190.
USGS (2001) NASA Land Processes Distributed Active Archive Center (LP DAAC).
MOD15A2 Leaf Area Index - Fraction of Photosynthetically Active Radiation 8-
Day L4 Global 1 km. USGS/Earth Resources Observation and Science (EROS)
Center, Sioux Falls, South Dakota.
Verbesselt J, Hyndman R, Newnham G, Culvenor D (2010) Detecting trend and sea-
sonal changes in satellite image time series. Remote Sensing of Environment, 114,
106–115.
Vitousek PM, Aber JD, Howarth RW et al. (1997) Human alteration of the global
nitrogen cycle: sources and consequences. Ecological Applications, 7, 737–750.
Walker DA, Leibman MO, Epstein HE et al. (2009) Spatial and temporal patterns of
greenness on the Yamal Peninsula, Russia: interactions of ecological and social fac-
tors affecting the Arctic normalized difference vegetation index. Environmental
Research Letters, 4, 045004.
Walker JJ, de Beurs KM, Wynne RH (2014) Dryland vegetation phenology across an
elevation gradient in Arizona, USA, investigated with fused MODIS and Landsat
data. Remote Sensing of Environment, 144,85–97.
Wang X, Piao S, Ciais P, Li J, Friedlingstein P, Koven C, Chen A (2011) Spring temper-
ature change and its implication in the change of vegetation growth in North
America from 1982 to 2006. Proceedings of the National Academy of Sciences of the Uni-
ted States of America, 108, 1240–1245.
Wang D, Morton D, Masek J et al. (2012) Impact of sensor degradation on the MODIS
NDVI time series. Remote Sensing of Environment, 119,55–61.
Wang Y, Roderick ML, Shen Y, Sun F (2014) Attribution of satellite-observed vegeta-
tion trends in a hyper-arid region of the Heihe River basin, Western China. Hydrol-
ogy and Earth System Sciences, 18, 3499–
3509.
Wermelinger B (2004) Ecology and management of the spruce bark beetle Ips ty-
pographus—a review of recent research. Forest Ecology and Management, 202,67–
82.
White MA, Thornton PE, Running W (1997) A continental phenology model for mon-
itoring vegetation responses to interannual climatic variability. Global Biogeochemi-
cal Cycles, 11, 217–234.
White K, Pontius J, Schaberg P (2014) Remote sensing of spring phenology in north-
eastern forests: a comparison of methods, field metrics and sources of uncertainty.
Remote Sensing of Environment, 148,97–107.
Wolkovich EM, Cleland EE (2010) The phenology of plant invasions: a community
ecology perspective. Frontiers in Ecology and the Environment, 9, 287–294.
Wolkovich EM, Cook BI, Allen JM et al. (2012) Warming experiments underpredict
plant phenological responses to climate change. Nature, 485, 494–497.
Wu C, Gonsamo A, Gough CM, Chen JM, Xu S (2014) Modeling growing season phe-
nology in North American forests using seasonal mean vegetation indices from
MODIS. Remote Sensing of Environment, 147,79–88.
Xiao X, Zhang J, Yan H, Wu W, Biradar C (2009) Land surface phenology. In: Phenol-
ogy of Ecosystem Processes (ed. Noormets A), pp. 247–270. Springer, New York.
Xu L, Myneni RB, Chapin Iii FS et al. (2013) Temperature and vegetation seasonality
diminishment over northern lands. Nature Climate Change, 3, 581–586.
Zeileis A, Kleiber C, Kr
€
amer W, Hornik K (2003) Testing and dating of structural
changes in practice. Computational Statistics & Data Analysis, 44, 109–123.
Zeng F-W, Collatz G, Pinzon J, Ivanoff A (2013) Evaluating and quantifying the cli-
mate-driven interannual variability in Global Inventory Modeling and Mapping
Studies (GIMMS) Normalized Difference Vegetation Index (NDVI3 g) at global
scales. Remote Sensing, 5, 3918–3950.
Zhang X, Tarpley D, Sullivan JT (2007) Diverse responses of vegetation phenology to
a warming climate. Geophysical Research Letters, 34, L19405.
Zhang X, Friedl MA, Schaaf CB (2009) Sensitivity of vegetation phenology detection
to the temporal resolution of satellite data. International Journal of Remote Sensing,
30, 2061–
2074.
Zhang X, Tan B, Yu Y (2014) Interannual variations and trends in global land surface
phenology derived from enhanced vegetation index during 1982–2010. Interna-
tional Journal of Biometeorology, 58, 547–564.
Zhou L, Tian Y, Myneni RB et al. (2014) Widespread decline of Congo rainforest
greenness in the past decade. Nature, 509,86–90.
Zhu Z, Bi J, Pan Y et al. (2013) Global data sets of vegetation Leaf Area Index (LAI)3 g
and Fraction of Photosynthetically Active Radiation (FPAR)3 g Derived from Glo-
bal Inventory Modeling and Mapping Studies (GIMMS) Normalized Difference
Vegetation Index (NDVI3 g) for the period 1981 to 2011. Remote Sensing, 5, 927–
948.
© 2015 John Wiley & Sons Ltd, Global Change Biology, 21, 3414–3435
3434 M. FORKEL et al.
Supporting Information
Additional Supporting Information may be found in the online version of this article:
Data S1. Description of phenology methods.
Table S1. Explanation of abbreviations.
Figure S1. Used masks for biome-aggregated, region-aggregated and global-aggregated results in tables and figures of the main
text.
Figure S2. Latitudinal gradients of all PGMs.
Figure S3. Regional-averaged time series and trends in SOS, EOS and LOS.
Figure S4. Latitudinal gradients of median annual SOS as estimated from 10 phenology methods for GIMMS3g, MODIS and GL2-
VGT2.
Figure S5. RGB composite maps of all effects on all PGMs.
Figure S6. KGE between SOS from LPJmL-GSI and the data-method ensemble.
Figure S7. Globally spatial averaged time series and trends of peak FAPAR, water availability anomalies from LPJmL, anomalies in
total tree foliar projective cover from LPJmL and air temperature and precipitation anomalies.
© 2015 John Wiley & Sons Ltd, Global Change Biology, 21, 3414–3435
CONTROLS ON PHENOLOGY AND GREENNESS 3435
