See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/312450378

Delayed response of spring phenology to global

warming in subtropics and tropics

Article in Agricultural and Forest Meteorology · March 2017

DOI: 10.1016/j.agrformet.2017.01.002

CITATIONS READS

0 36

3 authors, including:

Xiaoqiu Chen David Inouye

Peking University University of Maryland, College Park
40 PUBLICATIONS 610 CITATIONS 192 PUBLICATIONS 8,645 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Phenological change in a spring ephemeral: Implications for pollination and plant reproduction View
project

Long-term study of the phenology and abundance of flowering at the Rocky Mountain Biological
Laboratory, and how that affects pollinators and seed predators. View project

All content following this page was uploaded by Xiaoqiu Chen on 25 January 2017.

The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document
and are linked to publications on ResearchGate, letting you access and read them immediately.
 Agricultural and Forest Meteorology 234 (2017) 222–235

Contents lists available at ScienceDirect

Agricultural and Forest Meteorology

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

Delayed response of spring phenology to global warming in

subtropics and tropics
Xiaoqiu Chen a,∗ , Lingxiao Wang a , David Inouye b
a
College of Urban and Environmental Sciences, Laboratory for Earth Surface Processes of the Ministry of Education, Peking University, Beijing 100871, PR
China
b
Department of Biology, University of Maryland, College Park, MD 20742-4415, United States

a r t i c l e i n f o a b s t r a c t

Article history: Climate drivers of plant phenology in the subtropics and tropics are still unclear, which significantly
Received 13 June 2016 hinders accurate prediction of climate change impacts on vegetation growth and carbon balance in these
Received in revised form unique ecoregions. The basic hypothesis of process-based phenology models is that spring tree phenology
10 December 2016
is regulated by temperature and triggered by chilling temperatures during the dormancy period, followed
Accepted 3 January 2017
by forcing temperatures during the growth period. That is, trees require cool temperatures during a
chilling period to break endodormancy, and then enter the phase of ecodormancy during which the rate
Keywords:
of ontogenetic development increases with increasing air temperature. Therefore, insufficient chilling
Global warming
Leafing and flowering
requirements may slow bud growth and consequently delay budburst. Many studies have shown that
Process-based phenology model chilling requirement is at present sufficient to release bud dormancy fully in temperate regions, and thus
Chilling and forcing requirement forcing temperatures play a dominant role in triggering spring tree phenology. To identify differences
Spatiotemporal prediction in mechanisms of spring tree phenology responses to air temperature between the subtropics/tropics
and the temperate zone, and their possible effects on future phenological trends under global warming,
we used leaf unfolding and flowering data from a tree species of tropical origin, Melia azedarach, and
output daily mean air temperature data from a regional climate model (HadGEM3-RA) for the period
1981–2005 at 42 stations in southeastern China to fit unified forcing and chilling phenology models.
Then, we selected optimum models for each phenophase at each station. Moreover, we predicted leafing
and flowering dates across the research region over 2021–2100 under global climate warming scenarios.
The results show a previously unreported phenological phenomenon: chilling will often be insufficient
to break bud dormancy in the northern tropical zone and may become a crucial factor limiting leafing and
flowering responses to spring warming. However, chilling will still be sufficient to break bud dormancy
in the warm temperate zone, and thus may not limit leafing and flowering responses to spring warming
there. Consequently, predicted leafing and flowering dates both will be delayed in the northern tropical
zone but will advance in the warm temperate zone from 2021 to 2100. In the subtropical zone, the effect
of chilling temperature on spring phenology will be reduced gradually from earlier to later phenophases.
Thus, predicted leafing and flowering dates show decreased delaying trends and increased advancing
trends from earlier to later phenophases in the subtropical zone. Our findings suggest that insufficient
chilling temperature accumulation during the dormancy period may counteract the forcing temperature
accumulation during the growth period in the northern tropical zone and parts of the subtropical zone,
resulting in delayed leafing and flowering dates.
© 2017 Elsevier B.V. All rights reserved.

1. Introduction
The changing rate and direction of plant phenology in response
to climate change are diversified, depending highly on geograph-
∗ Corresponding author at: 5 Yiheyuan Road, Yifuerlou Room 3352, College of
Urban and Environmental Sciences, Peking University, Beijing 100871, PR China.
E-mail addresses: cxq@pku.edu.cn (X. Chen), wanglingxiao@pku.edu.cn
(L. Wang), Inouye@umd.edu (D. Inouye).
ical locations, time scales, and specific species (Menzel, 2003;
Matsumoto et al., 2003; Gordo and Sanz, 2010; Chen and Xu, 2012;
Chen et al., 2015). Therefore, revealing environmental and biolog-
ical mechanisms of the diversified phenological responses across
different climate zones is key to predicting accurately vegetation
phenology at regional and continental scales (Morin et al., 2009; Xu
and Chen, 2013). Process-based phenology models provide a reli-
able and effective tool to identify external and internal mechanisms
of phenological responses.

http://dx.doi.org/10.1016/j.agrformet.2017.01.002
0168-1923/© 2017 Elsevier B.V. All rights reserved.
 X. Chen et al. / Agricultural and Forest Meteorology 234 (2017) 222–235
Numerous studies have shown that spring phenology is highly
dependent on temperature during both the endodormancy phase
and the ecodormancy phase (Hänninen and Kramer, 2007; Chuine
et al., 2003). Photoperiod may also influence budburst but the mag-
nitude of the effect is generally minor (Bernier, 1988; Mouradov
et al., 2002; Laube et al., 2014). So far, no evidence has demon-
strated that photoperiod is more dominant than temperature when
predicting leafing and flowering, even in beech—one of the species
most sensitive to photoperiod (Kramer, 1994a; Vitasse et al., 2009;
Chuine et al., 2010). Thus, the basic hypothesis of process-based
phenology models is that spring tree phenology is controlled by
air temperature and triggered by a period with chilling temper-
atures followed by a period with forcing temperatures (Chuine,
2000). Because chilling temperatures may accelerate bud growth
from the state of quiescence to the state of budburst, in addition
to breaking dormancy (Nelson and Lavender, 1979; Cannell and
Smith, 1983; Murray et al., 1989; Hänninen et al., 1993), insuffi-
cient chilling due to rapid late winter-early spring air temperature
increase may slow bud growth and consequently delay budburst.
By contrast, forcing temperatures may also promote bud growth
after bud dormancy has been released in spring (Cannell and Smith,
1983; Hunter and Lechowicz, 1992; Kramer, 1994b; Chuine, 2000),
and early fulfilment of forcing requirements under rapid spring air
temperature increase may speed up bud growth and advance bud-
burst. Therefore, budburst timing might be attributed to the net
effect of temperature variations on the fulfilment of chilling and
forcing requirements.
Generally speaking, the less chilling temperatures are received,
the more forcing temperatures are subsequently needed to trigger
budburst (Chuine, 2000). To date, many process-based phenology
models have been established to simulate and predict spring phe-
nology of temperate-zone trees, and its responses to regional and
global climate change (Cannell and Smith, 1983; Murray et al.,
1989; Hänninen, 1990; Kramer, 1994a, 1994b; Chuine et al., 1998,
1999; Chuine, 2000; Linkosalo et al., 2008; Morin et al., 2009;
Richardson and O’Keefe, 2009; Luedeling et al., 2011; Fu et al., 2012;
Xu and Chen, 2013). Overall, results show that the fulfilment of forc-
ing requirements is not necessarily influenced by the fulfilment of
chilling requirements in triggering spring tree phenology. That is,
models excluding chilling requirements were best supported by
the data, while models including both forcing and chilling require-
ments were less well supported (Chuine, 2000; Linkosalo et al.,
2008; Morin et al., 2009; Richardson and O’Keefe, 2009; Parker
et al., 2011; Vitasse et al., 2011; Fu et al., 2012; Xu and Chen, 2013).
This implies that chilling requirements are currently sufficient to
release fully bud dormancy of trees in many temperate regions of
the world even though dramatic late winter-early spring temper-
ature increases have occurred. Therefore, forcing temperature is
currently the main trigger of spring phenology.
Observed spring phenological advancement and its response
to forcing temperature in the Northern Hemisphere over the past
decades (Fitter et al., 1995; Peñuelas and Filella, 2001; Fitter and
Fitter, 2002; Walther et al., 2002; Parmesan, 2007; Cook et al.,
2012) verify the above process-based modeling. Recently, a few
studies based on statistical models and experimental approaches
found that several temperate species are responsive to cold temper-
ature in the previous autumn and winter (Cook et al., 2012; Laube
et al., 2014; Roberts et al., 2015). Species that are highly sensitive to
chilling, such as hawthorn and birch, were predicted to be delayed
or advance less (Roberts et al., 2015). However, sensitive species
detected by statistical models and experimental approaches were
inconsistent (Laube et al., 2014; Roberts et al., 2015). Moreover,
responses to autumn and winter temperatures were strongest for
early-flowering species, and declined for species that normally
flower later in spring and summer (Cook et al., 2012; Roberts et al.,
2015).
223
By contrast, climate drivers of plant phenology in the sub-
tropics and tropics are still unclear (Hatta and Darnaedi, 2005;
Sanchez-Azofeifa et al., 2013), which significantly hinders accu-
rate prediction of climate change impacts on vegetation growth and
carbon balance in these unique ecoregions. A statistical analysis of
data from the northern tropical zone of China detected a dominant
delaying trend in budburst and flowering dates of tropical plants,
which are mainly influenced by increasing temperature. In contrast,
rainfall showed little effect on spring phenological variation (Zhao
et al., 2013). Nevertheless, whether the delayed spring phenology
in tropics of China is induced by insufficient chilling requirement
is unknown. Moreover, a process-based rough estimate of winter
chilling for temperate fruit and nut trees across the entire terrestrial
globe indicates that warm regions are likely to experience severe
reductions in available winter chilling, potentially threatening pro-
duction there (Luedeling et al., 2011). However, no robust surface
evidence has been detected so far for a warmer winter-induced
insufficient chilling requirement and delayed spring phenology in
the subtropics and tropics. To address this gap in our knowledge,
which hinders assessments of how spring phenology in the sub-
tropics and tropics is responding to climate warming, this study
presents the results of process-based modeling using first leaf
unfolding (FLU), 50% leaf unfolding (50% LU), first flowering (FF)
and 50% flowering (50% F) data from a tree species of tropical
origin, Melia azedarach, and daily mean air temperature data at
42 stations across southeastern China from 1981 to 2005 (Fig. 1
and Table 1). The scientific questions we intend to answer are:
1) Whether mechanisms of spring tree phenology response to air
temperature are different in different climate zones and different
seasonal stages (phenophases), and if so, why?; 2) Do occurrence
dates of spring tree phenology indicate a delayed trend in response
to global warming in the subtropics and tropics, and if so, why?; and
3) How do effects of chilling and forcing temperatures on spring tree
phenology interact?
2. Materials and methods
2.1. Study area and tree species
The study area is located in southeastern China from 19.0◦ N to
35.4◦ N and from 106.6◦ E to 120.5◦ E, covering approximately 1.58
million km2 , which accounts for 16.5% of China’s land area (Fig. 1,
Table 1). The climate is characterized by monsoonal winds and a
variety of climatic types. There are five climate zones from south to
north, namely, northern tropical, southern subtropical, middle sub-
tropical, northern subtropical and warm temperate zones (China
Meteorological Administration, 1979). The annual mean air tem-
perature decreases from 23.8 ◦ C in the south to 11.9 ◦ C in the north
and the annual mean total precipitation decreases from 1878.7 mm
in the southeast to 579.5 mm in the northwest. The topographic
structure in this area is composed of fluvial plains, low mountains,
and hills with an average elevation less than 500 m. The dominant
vegetation types from south to north include rain forest, evergreen
broadleaf forest, deciduous and evergreen broadleaf mixed forest,
and deciduous broadleaf forest on the low mountains and hills, as
well as cultivated vegetation on the plains.
To detect the diversified responses of spring tree phenol-
ogy to climate warming over the large geographical ranges, we
selected Melia azedarach (Meliaceae) as the indicator species. Melia
azedarach is a deciduous tree with odd-pinnate leaves, about 10 m
tall. The leaflets are opposite, blades ovate or elliptic. The thyrsus
is similar in length to leaves. Flowers are fragrant and lilac-colored.
It grows widely in mixed evergreen broad-leaved and decidu-
ous forests, sparse forests, field margins, roadsides at elevations
between 500 and 2100 m from warm temperate to northern trop-
224 X. Chen et al. / Agricultural and Forest Meteorology 234 (2017) 222–235
Fig. 1. Distribution of phenological and meteorological stations with grid’s center points of the regional climate model HadGEM3-RA in different climate zones.
ical zones of the eastern monsoonal region of China. This species
is cultivated and sometimes naturalized in many warm temperate
and tropical parts of the world, such as Bhutan, India, Indonesia,
Laos, Nepal, Papua New Guinea, Philippines, Sri Lanka, Thailand,
Vietnam; tropical Australia, Pacific islands (Editorial Committee of
Flora Repubulicae Popularis Sinicae, 1997).
2.2. Phenological data
Phenological data for Melia azedarach were provided by the
Meteorological Information Center of the China Meteorological
Administration (Chen, 2013). The phenological observations at this
network are carried out normally every other day by professionals
at weather stations according to uniform observation guidelines
(China Meteorological Administration, 1993). In order to reveal the
mechanisms of diversified phenological responses to seasonal cli-
mate warming across growth and reproductive stages, we selected
four spring phenophases, namely, first leaf unfolding, 50% leaf
unfolding, first flowering, and 50% flowering (Chen, 2013). The
first leaf unfolding is defined as when the first batch of leaves
is fully spread on a few twigs of the observed tree, whereas the
50% leaf unfolding is defined as when leaves are fully spread on
50% of twigs of the observed tree. Similarly, the first flowering is
identified as when the first batch of flower buds is fully opened
on the observed tree, while 50% flowering is identified as when
50% of flower buds are fully opened on the observed tree (China
Meteorological Administration, 1993). To ensure data quality, we
verified time series of the four Melia azedarach phenophases (day of
year, DOY) at each station by means of correlation analyses between
time series of first and 50% leaf unfolding and between time series
of first and 50% flowering (Chen and Xu, 2012). Then, we eliminated
the few anomalous data and obtained altogether 42 stations with
 X. Chen et al. / Agricultural and Forest Meteorology 234 (2017) 222–235 225

Table 1
Geographical coordinates, elevation, average thermal-moisture conditions, and total number of phenological samples at each station from 1981 to 2005.

Station number Station North East Elevation Annual mean Annual Total number of
name latitude longitude (m a.s.l.) temperature (◦ C) precipitation (mm) phenological samples

1 Qiongzhong 19◦ 02 109◦ 50 248.6 22.7 1774.5 84

2 Haikou 20◦ 00 110◦ 22 10.7 23.8 1666.4 92
3 Huazhou 21◦ 39 110◦ 37 31.8 22.7 1591.9 96
4 Qinzhou 21◦ 57 108◦ 37 5.5 22.6 1419.0 88
5 Lingshan 22◦ 25 109◦ 18 66.0 21.2 1460.6 96
6 Zhongshan 22◦ 32 113◦ 21 2.1 22.2 1790.6 91
7 Yulin 22◦ 39 110◦ 10 85.3 21.4 1593.1 91
8 Nanning 22◦ 49 108◦ 21 73.7 21.6 1270.4 92
9 Foshan 23◦ 01 113◦ 06 6.5 22.2 1702.7 80
10 Gaoyao 23◦ 03 112◦ 28 12.4 20.6 1774.5 90
11 Tiandeng 23◦ 05 107◦ 09 438.0 21.2 1144.3 99
12 Guiping 23◦ 24 110◦ 05 44.1 21.1 1550.2 80
13 Cangwu 23◦ 25 111◦ 14 41.4 21.2 1513.1 84
14 Chaozhou 23◦ 40 116◦ 38 11.3 21.5 1652.9 72
15 Heyuan 23◦ 44 114◦ 41 40.8 21.0 1878.7 89
16 Baise 23◦ 54 106◦ 36 175.8 22.1 1069.2 92
17 Du’an 23◦ 56 108◦ 06 172.2 20.5 1228.5 82
18 Mengshan 24◦ 12 110◦ 31 147.0 18.0 1573.4 96
19 Meixian 24◦ 16 116◦ 06 89.3 19.5 1620.2 96
20 Liuzhou 24◦ 28 109◦ 23 99.1 20.0 1358.0 100
21 Hechi 24◦ 42 108◦ 03 214.3 20.3 1147.0 84
22 Lianzhou 24◦ 47 112◦ 23 98.1 17.8 1492.6 96
23 Longnan 24◦ 55 114◦ 49 206.5 19.1 1564.8 100
24 Guilin 25◦ 05 110◦ 19 171.6 19.0 1481.9 88
25 Rong’an 25◦ 13 109◦ 24 122.1 18.7 1452.3 92
26 Lianhua 27◦ 08 113◦ 57 181.4 16.3 1537.7 96
27 Nanfeng 27◦ 13 116◦ 32 106.5 17.4 1715.3 100
28 Nanchang 28◦ 33 115◦ 57 28.7 17.8 1657.3 84
29 Wuyuan 29◦ 16 117◦ 51 82.0 15.7 1630.7 99
30 Fengdu 29◦ 52 107◦ 41 214.8 14.8 1165.5 80
31 Jiangxia 30◦ 21 114◦ 19 38.2 16.7 1256.5 72
32 Xuancheng 30◦ 56 118◦ 45 34 15.2 1226.0 96
33 Xinyang 32◦ 08 114◦ 03 115.1 15.3 986.7 92
34 Zhenjiang 32◦ 11 119◦ 28 28.8 14.7 1050.1 88
35 Chuzhou 32◦ 18 118◦ 18 32.9 15.0 1043.0 92
36 Fuyang 32◦ 55 115◦ 49 38.6 15.1 974.6 90
37 Zhumadian 33◦ 00 114◦ 01 83.3 14.8 860.7 98
38 Dafeng 33◦ 12 120◦ 29 7.3 14.2 1007.5 92
39 Suzhou 33◦ 38 116◦ 59 36.7 14.7 869.6 92
40 Ruzhou 34◦ 11 112◦ 50 214.2 11.9 691.5 98
41 Zhengzhou 34◦ 43 113◦ 39 111.3 13.9 579.5 94
42 Jining 35◦ 26 116◦ 35 45.2 13.9 635.0 100

a time series length of at least 18 years from 1981 to 2005 for the
four phenophases (Fig. 1). The detailed station names, geo-location
parameters (latitude, longitude and elevation), mean annual air
temperature, mean annual precipitation, and the total number of
phenological samples at each station are listed in Table 1. The aver-
age time series of each phenophase during 1981–2005 within each
climate zone are shown in Fig. A1.
2.3. Climate data
An apparent contradiction has arisen in spring phenological
simulation and prediction by means of process-based spring phe-
nology models: Model generation is based on daily mean air
temperature from meteorological stations, whereas the model pre-
diction is based on daily mean air temperature from outputs of a
regional climate model. Clearly, these two temperature datasets
are different, and thus it is not reasonable to make predictions
until both datasets are calibrated (Luo et al., 2014). However, cal-
ibrating air temperature under climate change scenarios on the
basis of historical differences between air temperature time series
from regional climate model outputs and meteorological stations
introduces great uncertainties. To avoid the inconsistency of air
temperature datasets in fitting and predicting occurrence dates of
the four Melia azedarach phenophases, we used the daily mean air
temperature data acquired from outputs of a Regional Climate Mod-
els (RCMs) under the project Coordinated Regional Downscaling
Experiment (CORDEX)-EAST ASIA (Giorgi et al., 2009) with a spa-
tial resolution of 0.44◦ as inputs of the process-based phenology
models. The CORDEX project was aimed at generating regional cli-
mate change projections for the major terrestrial regions of the
world using experiment results of the 5th Phase of Coupling of
Model Intercomparison Program (CMIP5) as input data. Currently,
projections of five Regional Climate Models under scenarios of
the Representative Concentration Pathway (RCP) 4.5 and 8.5 are
available. The RCP 4.5 was defined as a stabilization scenario, in
which total radiative forcing is stabilized before 2100 through the
employment of a range of technologies and strategies for reducing
greenhouse gas emissions (Thomson et al., 2011), while the RCP 8.5
scenario is characterized by increasing greenhouse gas emissions
over time and is thought as representative of the scenario with the
highest greenhouse gas concentration level in existing literature
(Riahi et al., 2011).
We used daily mean air temperature data acquired from out-
puts of the regional climate model HadGEM3-RA as inputs of the
process-based phenology models. The data selection was based on
a correlation analysis and bias estimate between historical annual
mean air temperatures from different RCM outputs and annual
mean air temperature from ground observations at 8 km grids from
1979 to 2005 (Luo et al., 2014). The temperature dataset contains
historical daily mean air temperatures from 1981 to 2005 and
226 X. Chen et al. / Agricultural and Forest Meteorology 234 (2017) 222–235
future daily mean air temperatures under climate change scenar-
ios RCP8.5 and RCP4.5 from 2021 to 2100. For each phenological
station, we selected the daily mean air temperature dataset at the
nearest grid’s center point of HadGEM3-RA (Fig. 1) to fit and predict
occurrence dates of Melia azedarach’s leaf unfolding and flowering.
The distances between phenological stations and corresponding
grid’s center points are between 2.8 km and 25.8 km, with an aver-
age distance of 12.1 km.
2.4. Phenology models and their evaluations
To assess whether warmer winters are slowing the fulfilment
of chilling requirements and inducing a later spring phenology of
Melia azedarach in different climate zones, we employed the Uni-
Forc and UniChill models (Chuine, 2000) to fit leaf unfolding and
flowering dates of Melia azedarach at each station from 1981 to
2005. Both output daily mean air temperature data of HadGEM3-RA
and observed phenological data are available for this period.
The UniForc model assumes that occurrence date of a spring
phenophase is triggered only by daily mean air temperature during
spring. When the state of forcing temperature, Sf , achieves a critical
value F∗ on the date of y (DOY), the phenophase happens [Eq. (1)]:

y
Sf = Rf (xt ) = F ∗
t1
The state of forcing temperature is an accumulation of the rate
of forcing temperature, Rf (xt ), which begins on 1 January (t1 , DOY)
and xt is the daily mean air temperature. The rate of forcing tem-
perature is defined as an exponential function of xt [Eq. (2)]:
1
Rf (xt ) =
1 + ed(xt −e)
The Uniforc model has three fitted parameters d, e, and F∗ , with
d < 0 and e > 0.
The UniChill model hypothesizes that occurrence date of a spring
phenophase is induced by both chilling temperatures during the
dormancy period and forcing temperatures during the growth
period. When the state of chilling temperature, Sc , achieves a criti-
cal value of C∗ on the date of t1 , the bud dormancy is released [Eq.
(3)]:

t1
Sc = Rc (xt ) = C ∗
t0
The state of chilling temperature is a sum of the rate of chilling
temperature, Rc (xt ), which begins on 1 September of the preceding
year (t0 , DOY) and xt is the daily mean air temperature. The rate of
chilling temperature is defined as an exponential function of xt [Eq.
(4)]:
1 on the lowest AIC value. The NSEs of all local optimum models
Rc (xt ) = 2
1 + ea(xt −c) +b(xt −c)
After the bud dormancy is released on the date of t1 , the forcing
temperature accumulation starts until it reaches a critical value F∗
[Eqs. (1) and (2)]. It should be noted that the time point t1 denotes
not only the end date of chilling temperature accumulation but also
the starting date of forcing temperature accumulation. The UniChill
model includes seven fitted parameters, in which a, b, c, C∗ belong to
the chilling temperature function and d, e, F∗ belong to the forcing
temperature function.
Using the simulated annealing algorithm of Metropolis (Chuine
et al., 1998), we fitted model parameters and then selected the
optimum models by means of the root mean square error (RMSE),
the Akaike Information Criterion (AIC) (Akaike, 1973), and the
Nash-Sutcliffe Efficiency (NSE) (Nash and Sutcliffe, 1970) for each
phenophase at each station. The model parameters were fitted
by the lowest value of the RMSE [Eq. (5)], while the optimum
model (UniForc or UniChill) for each phenophase at each station was
selected by the lowest value of the AIC [Eq. (6)]. In order to eval-
uate the effectiveness of the optimum models, we also calculated
the NSE [Eq. (7)] for comparing the model performance with the
null model (the mean occurrence date of a specific phenophase).

n

 (obsi − prei )
2

i=1
RMSE = (5)
n

n
2
(obsi − prei )
i=1
AIC = n × ln( ) + 2(k + 1) (6)
n

n
2
(obsi − prei )
i=1
NSE = 1 − n 
(7)
 2
obsi − obs
i=1
(1) where obsi denotes the observed date of a specific phenophase
in year i; prei denotes the simulated or predicted date of the
phenophase in year i; obs denotes the mean observed date of the
phenophase during the research period; n is the number of years;
k is the number of parameters.
2.5. Model prediction
(2)
To predict possible responses of Melia azedarach spring phenol-
ogy to global climate warming, we substituted output daily mean
air temperature data from the regional climate model (HadGEM3-
RA) under scenarios RCP8.5 and RCP4.5 from 2021 to 2100 into
the 168 local optimum models and obtained occurrence dates of
each phenophase at each station under both scenarios from 2021 to
2100 (336 time series). Then, we calculated anomalies of phenolog-
ical occurrence dates by taking the observed average phenological
occurrence date during 1981–2010 as the reference value.
(3) 3. Results
3.1. Local optimum models and their spatial and seasonal
dependence
We employed the UniForc and UniChill models to fit 168 datasets
(four phenophases at 42 stations) from 1981 to 2005, and obtained
an optimum model for each phenophase at each station based
(4) are larger than zero, indicating the effectiveness of the optimum
models. In addition, the coefficient of determination for regression
lines (R2 ) between observed and simulated occurrence dates are
all significant for the four phenophases at the 42 stations (P < 0.05).
The regression equations between observed and simulated occur-
rence dates for the four phenophases at the 42 stations explained
59–93% of the FLU variance (mean = 71%), 48–93% of the 50% LU
variance (mean = 70%), 60–93% of the FF variance (mean = 75%), and
58–93% of the 50% F variance (mean = 76%) from 1981 to 2005.
The root mean square errors (RMSEs) for differences between
observed and simulated occurrence dates are between 2.7 days
and 8.9 days (mean = 5.1 days) for the FLU, between 2.5 days and
8.5 days (mean = 4.6 days) for the 50% LU, between 3.1 days and
8.8 days (mean = 5.6 days) for the FF, and between 3.3 days and
8.7 days (mean = 5.5 days) for the 50% F, respectively.
 X. Chen et al. / Agricultural and Forest Meteorology 234 (2017) 222–235 227

Fig. 2. Spatial pattern of local optimum models (C: UniChill model; F: UniForc model) in different climate zones for (a) first leaf unfolding, (b) 50% leaf unfolding, (c) first
flowering, and (d) 50% flowering. The colored dots denote different climate zones.

The structures of the optimum models display an obvious
spatial and seasonal dependence. All 24 optimum models (4
phenophases × 6 stations) in the warm temperate zone belong to
the UniForc model, which assumes that occurrence date of a spring
phenophase is triggered only by forcing temperatures during the
growth period, whereas all 12 optimum models (4 phenophases × 3
stations) in the northern tropical zone pertain to the UniChill model,
which assumes that occurrence date of a spring phenophase is
induced by both chilling temperatures during the dormancy period
and forcing temperatures during the growth period. In the remain-
ing 132 optimum models (4 phenophases × 33 stations) of the
subtropical zone, the UniChill model accounts for 67%, while the
UniForc model accounts for only 33%. Specifically, the UniChill
model gives a better simulation than the UniForc model at 32 of
33 stations for the FLU (97%, Fig. 2a), at 29 of 33 stations for the 50%
LU (88%, Fig. 2b), at 25 of 33 stations for the FF (76%, Fig. 2c), and at 2
of 33 stations for the 50% F (6%, Fig. 2d), respectively. Thus, the effect
of chilling temperature on spring phenology of Melia azedarach
was gradually weakened but the effect of forcing temperature was
steadily enhanced along with the progression of seasons from ear-
lier phenophases (such as FLU and 50% LU) to later phenophases
(such as FF and 50% F).
228 X. Chen et al. / Agricultural and Forest Meteorology 234 (2017) 222–235

Table 2
Linear trends (days per decade) and their significance levels in mean anomalies of phenological occurrence dates (baseline: average phenological occurrence date during
1981–2010) in different climate zones from 2021 to 2100 predicted under global climate change scenarios RCP8.5 and RCP4.5.

Phenophase Scenario Warm temperate zone Subtropical zones Northern tropical zone

First leaf unfolding RCP8.5 −0.42*** 1.53* 1.78**

RCP4.5 −0.26 0.24 0.45
50% leaf unfolding RCP8.5 −0.59*** 1.46* 2.55**
RCP4.5 −0.33 0.38 0.81
First flowering RCP8.5 −0.58*** 1.14* 2.10*
RCP4.5 −0.21 0.06 0.62
50% flowering RCP8.5 −0.65*** −0.52* 2.89***
RCP4.5 −0.39 −0.22 0.91
*
P < 0.05.
**
P < 0.01.
***
P < 0.001.

Fig. 3. Percentage of stations with predicted advancing (a) and delaying (b) linear trends under scenarios RCP8.5 and RCP4.5 for each phenophase (FLU: first leaf unfolding;
50% LU: 50% leaf unfolding; FF: first flowering; 50% F: 50% flowering) in the northern, middle, and southern subtropical zones, respectively.

3.2. Spring phenology prediction under global warming
Using the 168 local optimum models and output daily mean
air temperature data from the regional climate model (HadGEM3-
RA) under scenarios RCP8.5 and RCP4.5 for the period 2021–2100,
we predicted the future change of leafing and flowering dates of
Melia azedarach at the 42 stations in different climate zones. These
predictions were based on the UniForc and UniChill models. For dif-
ferent climate zones, diversified responses of spring phenology to
global warming were predicted. A consistent advancing trend for
the four phenophases was detected under both scenarios in the
warm temperate zone (P < 0.001 under scenario RCP8.5; P > 0.05
under scenario RCP4.5), whereas a consistent delaying trend for the
four phenophases was found under both scenarios in the north-
ern tropical zone (P < 0.05–0.001 under scenario RCP8.5; P > 0.05
under scenario RCP4.5). Under both scenarios in the intermedi-
ate subtropical zone, a delaying trend for FLU, 50% LU and FF,
and an advancing trend for 50% F were identified (P < 0.05 for
the four phenophases under scenario RCP8.5; P > 0.05 for the four
phenophases under scenario RCP4.5) (Table 2). Further analyses
show that the effect of chilling temperature on spring phenology
in the subtropical zone will be reduced gradually from earlier to
later phenophases, resulting in an increased percentage of stations
with advancing trends and a decreased percentage of stations with
 X. Chen et al. / Agricultural and Forest Meteorology 234 (2017) 222–235
Table 3
Linear trends (days per decade) and their significance levels in mean anomalies of phenological occurrence dates from 2021 to 2100 under global climate change scenarios
RCP8.5 and RCP4.5 predicted by the UniForc model and UniChill model, respectively.
Model Scenario First leaf unfolding
UniForc RCP8.5 −0.4***
RCP4.5 −0.2***
UniChill RCP8.5 1.6**
RCP4.5 0.3
**
P < 0.01.
***
P < 0.001.
delaying trends from earlier to later phenophases in the northern,
middle, and southern subtropical zones, respectively (Fig. 3).
Considering the different effects of forcing and chilling tem-
peratures on predicted phenological trends, we calculated time
series of mean anomalies of phenological occurrence date under
both scenarios based on the UniForc and UniChill models, respec-
tively. Results show that future phenological trends depend highly
on model structures and have an obvious divergence in response
to global climate warming. For the spring phenology prediction
based on the UniForc model, mean anomalies of phenological occur-
rence dates indicate significant advancements (P < 0.01 ∼ 0.001)
for the four phenophases under both climate change scenarios.
We note that the absolute values of advancing trends (days per
decades) under scenario RCP8.5 are generally larger than those
under scenario RCP4.5. In addition, the absolute values of advancing
trends under scenario RCP8.5 represent a slight increasing ten-
dency from earlier to later phenophases, whereas the absolute
values of advancing trends under scenario RCP4.5 do not display
any tendency from earlier to later phenophases. With regard to
spring phenology prediction based on the UniChill model, how-
ever, a significant delay (P < 0.01–0.001) was detected for the four
phenophases under scenario RCP8.5, whereas a nonsignificant
delay (P > 0.05) was discovered for the four phenophases under
scenario RCP4.5 (Table 3).
4. Discussion
This study shows that strategies of spring tree phenology
response to air temperature are highly dependent on geographical
locations and phenological occurrence timings. Leafing and flow-
ering of Melia azedarach were triggered predominantly by forcing
temperature (UniForc model) at all stations of the warm temper-
ate zone, and by both chilling and forcing temperatures (UniChill
model) at all stations of the northern tropical zone. In the subtrop-
ical zone however, the effect of chilling plus forcing temperatures
on leafing and flowering of Melia azedarach was stronger for the
earlier phenophases than the later phenophases (Fig. 2). Thus,
chilling requirements are sufficient to break bud dormancy and
may not be an important driver influencing Melia azedarach phe-
nological responses to spring warming in the warm temperature
zone, which has been verified by numerous studies in process-
based spring phenology modeling in the temperature zone (Chuine,
2000; Linkosalo et al., 2008; Morin et al., 2009; Richardson and
O’Keefe, 2009; Parker et al., 2011; Vitasse et al., 2011; Fu et al.,
2012; Xu and Chen, 2013). In contrast, chilling requirements are
insufficient to release bud dormancy and become a crucial limiting
factor affecting Melia azedarach phenological responses to spring
warming at all stations of the northern tropical zone and most sta-
tions of the subtropical zone (except for 50% F; Fig. 2), although
winter temperature (from December to February) increased much
more slowly in the northern tropical zone and the subtropical zone
(0.08–0.23 ◦ C/decade, P > 0.05) than in the warm temperate zone
(0.31 ◦ C/decade, P < 0.05) over the 1981–2005 period. Namely, the
fulfilment of chilling requirements in the northern tropical zone
and the subtropical zone might be more sensitive to winter temper-
229
50% leaf unfolding First flowering 50% flowering
−0.5*** −0.5*** −0.6***
−0.3** −0.2** −0.3**
1.8** 1.7** 2.0***
0.5 0.2 0.6
ature increases than that in the warm temperate zone. Observations
indicate that Melia azedarach experienced a shorter over-wintering
period from 38 to 81 days (average number of days from the end
date of defoliation in the previous year to the start date of budburst
in the current year) under higher winter temperatures in the sub-
tropical and northern tropical zones but a longer over-wintering
period of 143 days under lower winter temperatures in the warm
temperate zone. Therefore, Melia azedarach growing in the subtrop-
ical and northern tropical zones may need less chilling temperature
accumulation for recovering than that living in the warm temperate
zone. That is, even slight winter temperature increases may still dis-
turb dormancy of Melia azedarach in the subtropical and northern
tropical zones. This implies that the delaying effect of insufficient
chilling temperature accumulation on leafing and flowering may
counteract or even exceed the advancing effect of forcing tempera-
ture accumulation, resulting in delayed Melia azedarach leafing and
flowering dates in the subtropical and northern tropical zones.
Further analysis indicates that spatial pattern and its seasonal
shift of local optimum models (Fig. 2) depend obviously on spatial
pattern of long-term mean winter (from December to February)
and spring (from March to May) temperatures. Insufficient chilling
requirements (Unichill model) appeared mainly in warmer areas
and the range of insufficient chilling requirements decreased from
earlier phenophases to later phenophases (Fig. 4). The weakened
effect of chilling temperatures on Melia azedarach spring phenology
from earlier phenophases to later phenophases might be attributed
to time intervals between budburst and the following phenophases.
As chilling temperatures may accelerate bud growth from the state
of quiescence to the state of budburst, and leafing happens soon
after budburst, chilling temperatures may subsequently influence
first leaf unfolding and 50% leaf unfolding dates at most stations
(Fig. 4a, b). By contrast, 50% flowering dates were much less influ-
enced by chilling temperatures because they occur much later than
budburst dates and in the warm season. As a result, the effect of
chilling temperature on 50% flowering dates appears only at the five
stations with the most sensitive chilling requirements and highest
winter temperatures (Fig. 4d).
The above-mentioned spatial pattern and its seasonal shift of
local optimum models have directly led to divergent linear trends
of predicted Melia azedarach spring phenology in different climate
zones. At stations in the warm temperate zone, spring phenol-
ogy affected by the forcing temperature (Uniforc model) shows a
consistent advancing trend of the four phenophases from 2021 to
2100 under both scenarios. As the mean increase rate of spring
temperature under scenario RCP8.5 (0.42 ◦ C/decade, P < 0.001) is
much more rapid than that under scenario RCP4.5 (0.14 ◦ C/decade,
P > 0.05), the significant advancing trends of the four phenophases
occur only under scenario RCP8.5 (Table 2). At stations in the
northern tropical zone, however, spring phenology affected by
the chilling plus forcing temperature (Unichill model) displays a
consistent delaying trend of the four phenophases from 2021 to
2100 under both scenarios, and the significant delaying trends
are detected also under scenario RCP8.5 (Table 2). Moreover, at
stations in the subtropical zone, the mixed chilling plus forcing
temperature (Unichill) and forcing temperature (Uniforc) models
predict a significant (under scenario RCP8.5) and a nonsignificant
(under scenario RCP4.5) delaying trend of FLU, 50% LU and FF, but
230 X. Chen et al. / Agricultural and Forest Meteorology 234 (2017) 222–235
Fig. 4. Temperature dependence of spatial pattern and its seasonal shift of local optimum models for (a) first leaf unfolding, (b) 50% leaf unfolding, (c) first flowering, and (d)
50% flowering at the 42 stations.
a significant (under scenario RCP8.5) and a nonsignificant (under
scenario RCP4.5) advancing trend of 50% F (Table 2), which might
be explained by the weakened effect of chilling temperatures and
intensified effect of forcing temperatures from earlier phenophases
to later phenophases. Overall, climate warming may significantly
decrease the spatial inhomogeneity of leafing and flowering dates
of Melia azedarach and result in a more homogenous spatial pattern
of spring phenology than at present. Our model predictions might
be also expected to occur in other subtropical and tropical areas of
the northern hemisphere.
When average linear trends of Melia azedarach spring phenology
were calculated based on the Uniforc and Unichill models sep-
arately, a clearly divergent trend change can also be identified.
For the spring phenology prediction based on the UniForc model,
the mean anomalies of phenological occurrence dates indicate a
consistently significant advancement (P < 0.01–0.001) for the four
phenophases under both climate change scenarios (Table 3). The
advanced response of Melia azedarach leafing and flowering dates
to global warming signifies that the advancing effect of forcing tem-
perature on leafing and flowering dates might be sped up in the
future at stations with UniForc as the optimum model (Fig. 2). By
contrast, for the spring phenology prediction based on the UniChill
model, a consistently significant delay (under scenario RCP8.5) and
a consistently nonsignificant delay (under scenario RCP4.5) are
detected for the four phenophases (Table 3). The delayed response
of Melia azedarach leafing and flowering dates to global warming
implies that the delaying effect of insufficient chilling temperature
on leafing and flowering dates may always exceed the advancing
effect of forcing temperature in the future at stations with Unichill
as the optimum model. Uncertainties about the spring phenology
prediction may exist in the adopted local optimum phenology mod-
els and regional climate change projections. Further studies should
therefore be carried out to assess suitability and effectiveness of the
local optimum phenology models fitted by historical daily mean air
temperature data from outputs of regional climate models in pre-
dicting phenological occurrence dates based on future daily mean
air temperature data under global climate change scenarios.
5. Conclusions
Global warming and its impacts on subtropical and tropical
ecosystems are a matter of global concern not only for scientists
but also for the general public. Yet, our knowledge of leafing and
flowering mechanisms in the subtropics and tropics is meager,
which significantly hinders accurate prediction of climate change
impacts on vegetation growth and reproduction, and carbon bal-
ance in these unique ecoregions. Detecting the interaction between
chilling and forcing requirements of plants in triggering leafing and
flowering phenology, and its spatial and seasonal differences are of
crucial importance for revealing mechanisms of leafing and flow-
ering response to climate change. These insights would allow us
to make accurate predictions of vegetation growth and reproduc-
tion, and carbon balance under global warming at regional and
continental scales.
Our research provides insights into the diversified mechanisms
of spring phenological responses to air temperature across different
climate zones, and we show that chilling requirements are suffi-
cient to break bud dormancy and may not be an important driver
influencing phenological responses of Melia azedarach to spring
warming in the warm temperature zone, but may not meet the
needs for releasing bud dormancy and become a crucial limiting
factor affecting phenological responses to spring warming in the
northern tropical zone. In the subtropical zone however, the effect
of chilling temperature on spring phenology was reduced gradually
from earlier to later phenophases. This implies that the fulfilment
of chilling requirements of Melia azedarach in the subtropical and
northern tropical zones might be more sensitive to winter temper-
ature increase than that in the warm temperate zone.
Prediction of future phenological trends depends greatly on
model structures, which can result in very divergent responses
 X. Chen et al. / Agricultural and Forest Meteorology 234 (2017) 222–235 231

to global warming. Predicted leafing and flowering dates dis-
play a coincident delaying trend in the northern tropical zone
(Unichill model) but a coincident advancing trend in the warm tem-
perate zone (Uniforc model) from 2021 to 2100 under scenarios
RCP8.5 and RCP4.5. Moreover, predicted leafing and flowering dates
show decreased delaying trends and increased advancing trends
from earlier (Unichill model dominance) to later (Uniforc model
dominance) phenophases in the subtropical zone. This might be
attributed to the net effect of temperature variations on the ful-
filment of chilling and forcing requirements. Namely, the delaying
effect of insufficient chilling temperatures on earlier phenophases
may counteract or exceed the advancing effect of forcing temper-
atures at most subtropical stations, leading to a predominantly
delaying trend, whereas the delaying effect of insufficient chilling
temperatures on later phenophases may not be strong enough to
counteract the advancing effect of forcing temperatures, resulting
in a predominantly advancing trend. As a result, shortened vege-
tation growth and reproduction periods might be expected in the
subtropics and tropics, which may lead to reduced regulatory func-
tion of vegetation to the seasonal carbon cycle, and decreased fruit
quality.
Acknowledgements
This work was funded by the National Natural Science Foun-
dation of China under grant no. 41471033. The authors thank the
Meteorological Information Center of the China Meteorological
Administration for providing phenological data. In addition, we
acknowledge the CORDEX-East Asia Databank, which is responsi-
ble for the CORDEX dataset, and we thank the National Institute of
Meteorological Research (NIMR), three universities in the Repub-
lic of Korea (Seoul National University, Yonsei University, Kongju
National University) and other cooperative research institutes in
East Asia region for producing and making available their model
output.
Appendix A.
See Table A1–A4.

Table A1
Parameters and performance of the optimum models for first leaf unfolding at each station (AIC = AICUniForc − AICUniChill ).

Station number a b c d e C* F* AIC

1 0.16 13.93 98.66 −83.85 15.97 119.48 2.02 2.31

2 1.12 −24.79 8.98 −43.76 12.08 132.99 10.52 4.60
3 0.42 85.32 25.87 −23.37 9.81 123.93 19.68 6.13
4 1.26 12.41 22.34 −95.98 3.71 22.90 98.97 15.49
5 0.20 −21.18 23.62 −52.42 3.28 13.83 162.00 2.40
6 1.07 83.96 10.89 −84.10 5.24 4.80 70.07 2.72
7 3.65 −81.90 14.35 −83.85 2.46 55.64 121.10 6.87
8 0.03 −81.04 −48.01 −9.07 13.62 171.80 1.34 5.64
9 0.08 −93.23 5.29 −29.27 0.52 126.12 40.81 5.65
10 4.74 75.60 30.24 −75.50 3.73 61.35 114.88 7.12
11 4.95 85.05 33.69 −69.85 4.21 39.78 140.09 5.83
12 1.15 −62.91 −27.37 −85.27 2.03 126.35 60.58 5.58
13 2.43 27.57 32.64 −26.56 0.53 31.18 137.57 4.24
14 9.06 80.48 15.66 −14.75 15.26 52.98 3.87 3.75
15 0.14 −12.85 −69.84 −28.93 16.25 146.26 2.24 2.75
16 9.19 −71.38 16.30 −81.24 4.05 35.94 113.87 8.81
17 0.14 −2.69 1.13 −10.27 9.80 100.40 19.10 4.43
18 3.38 92.56 24.60 −83.66 1.47 119.31 54.66 2.85
19 0.14 12.31 74.30 −28.47 3.72 108.74 59.16 8.42
20 0.61 43.56 27.35 −67.99 2.28 96.90 84.09 3.56
21 4.30 55.56 13.85 −74.56 12.37 98.99 0.39 2.70
22 2.55 −86.45 −9.52 −90.56 2.60 172.11 6.41 5.18
23 6.07 11.68 5.16 −81.43 10.35 4.66 18.82 3.38
24 0.75 −7.14 −0.76 −0.11 2.45 38.81 44.65 3.34
25 0.21 −40.54 1.07 −41.41 11.60 180.03 2.94 5.55
26 1.50 45.05 25.66 −74.59 0.85 195.03 1.69 3.48
27 0.94 −88.74 −72.76 −0.87 14.42 142.91 3.77 4.09
28 0.27 −17.95 −39.42 −57.46 5.77 192.35 9.13 5.06
29 0.57 80.66 26.69 −41.49 4.92 197.99 12.63 5.77
30 3.58 −52.33 −4.08 −16.07 11.25 113.27 2.66 3.45
31 0.15 −81.57 −21.90 −62.88 3.97 179.59 32.58 6.79
32 0.11 10.81 67.14 −0.16 13.69 206.88 2.96 3.92
33 0.53 −77.26 −18.66 −79.20 4.65 186.66 27.93 4.97
34 2.33 −95.53 −12.92 −66.99 3.62 181.46 40.67 5.43
35 1.45 82.73 46.14 −7.57 3.22 176.08 43.89 5.86
36 −0.01 11.60 50.86 −2.70
37 −0.01 6.97 51.90 −8.28
38 −0.04 0.09 63.16 −4.56
39 −0.01 11.56 52.67 −4.85
40 −0.01 15.64 52.60 −7.88
41 −0.01 1.89 50.67 −6.91
42 −0.01 5.28 54.48 −3.27
232 X. Chen et al. / Agricultural and Forest Meteorology 234 (2017) 222–235

Table A2
Parameters and performance of the optimum models for 50% leaf unfolding at each station (AIC = AICUniForc − AICUniChill ).

Station number a b c d e C* F* AIC

1 3.81 −70.69 10.25 −89.14 15.94 119.28 6.71 4.93

2 0.78 8.10 28.08 −67.22 6.55 62.40 47.90 4.86
3 2.56 50.04 22.38 −44.52 6.23 4.71 129.69 5.24
4 3.05 75.82 22.68 −16.36 7.24 112.46 16.53 3.99
5 8.14 46.64 29.12 −19.47 3.84 15.36 165.15 2.84
6 1.62 64.35 27.24 −99.00 9.26 138.21 28.77 4.26
7 5.71 95.86 29.99 −56.13 2.44 68.22 114.16 4.27
8 3.31 47.61 28.99 −6.42 4.52 58.76 119.65 2.57
9 1.07 11.94 11.13 −85.28 10.67 18.03 27.08 9.03
10 5.80 38.25 26.39 −32.85 2.15 0.39 187.76 4.54
11 2.67 −67.99 16.36 −93.73 4.21 42.05 146.02 4.97
12 7.66 92.53 32.73 −13.92 0.27 36.96 158.00 2.93
13 0.00 29.13 61.68 −84.15 12.43 180.19 1.88 4.63
14 1.21 95.75 10.17 −41.20 13.68 19.11 11.79 4.61
15 4.78 −70.09 4.99 −12.80 2.21 16.65 117.94 1.68
16 7.07 42.62 24.31 −22.32 4.98 18.91 134.85 6.17
17 9.53 −82.05 −0.86 −51.02 10.71 7.73 23.79 3.36
18 1.00 −74.59 9.76 −73.86 0.73 67.04 119.03 5.15
19 1.91 −87.95 −18.27 −38.26 3.40 131.25 48.66 5.92
20 0.44 −18.01 −13.74 −78.79 2.06 16.27 172.66 5.01
21 9.37 22.04 22.76 −34.88 0.09 4.10 168.26 1.12
22 2.73 97.49 19.79 −96.16 16.06 128.12 3.66 2.84
23 2.04 46.46 20.19 −43.55 10.07 151.97 8.01 1.68
24 −77.02 11.19 41.25 −2.27
25 −91.66 1.39 74.02 −3.44
26 0.85 −94.25 −85.38 −21.18 9.32 208.21 1.85 1.50
27 1.67 97.78 56.74 −50.62 7.33 175.03 18.34 9.29
28 0.17 −20.02 −88.56 −22.05 2.51 175.73 33.18 4.09
29 0.00 −0.24 −17.05 −20.00 6.67 136.90 41.56 6.48
30 3.13 51.46 13.85 −29.41 0.02 84.56 73.39 0.13
31 0.26 84.86 98.19 −91.21 3.91 178.31 39.82 5.65
32 0.51 −97.01 −26.24 −10.14 11.75 193.24 12.82 2.18
33 −0.03 13.03 53.47 −5.16
34 0.18 −72.60 −61.75 −0.07 16.98 169.19 24.65 5.69
35 0.54 −38.74 −15.69 −47.15 11.42 218.26 10.63 1.77
36 −0.01 0.48 53.66 −6.36
37 −0.02 11.28 56.21 −5.19
38 −0.04 6.26 58.15 −5.39
39 −0.01 19.56 52.67 −4.48
40 −0.02 19.11 54.94 −6.96
41 −0.02 8.13 52.78 −7.79
42 −0.02 15.78 55.29 −2.75

Table A3
Parameters and performance of the optimum models for first flowering at each station (AIC = AICUniForc − AICUniChill ).

Station number a b c d e C* F* AIC

1 1.00 69.71 79.73 −47.54 6.38 100.46 52.80 3.67

2 0.13 −38.95 −64.28 −57.75 12.04 136.73 24.70 6.14
3 0.04 −20.17 −74.38 −84.93 10.32 135.73 46.90 2.16
4 9.63 39.18 29.02 −31.52 2.80 9.29 90.95 5.25
5 1.16 −83.13 23.16 −24.79 2.69 16.49 178.85 4.05
6 9.35 57.60 10.86 −7.24 8.35 8.23 67.95 5.74
7 2.01 50.77 10.38 −83.43 5.47 10.87 87.00 6.94
8 9.15 −51.09 21.42 −36.08 3.72 8.18 184.87 3.75
9 6.72 −44.04 3.91 −7.11 6.50 8.24 57.53 4.61
10 6.14 64.80 33.93 −44.20 3.06 14.46 181.83 3.51
11 9.98 0.06 8.07 −74.38 9.19 0.95 53.91 0.32
12 1.71 −95.14 −28.62 −75.62 0.09 6.89 196.04 2.43
13 7.06 −86.49 23.13 −23.09 0.92 9.92 190.73 1.40
14 0.87 95.47 23.23 −79.14 6.96 109.73 54.73 5.70
15 1.30 −61.79 −32.54 −58.50 18.10 87.37 8.96 4.62
16 0.09 64.11 24.14 −95.88 6.49 115.24 50.13 4.39
17 0.00 62.02 12.47 −43.53 12.95 75.37 18.13 0.87
18 1.37 6.26 25.88 −43.36 0.54 8.39 199.85 1.14
19 9.13 98.26 30.55 −76.08 3.28 38.57 151.90 2.59
20 0.05 −31.11 21.39 −52.79 0.24 11.92 200.02 5.82
21 4.91 83.82 26.63 −85.76 1.06 66.74 132.83 2.10
22 2.28 66.54 24.41 −36.91 2.90 197.71 10.89 3.53
23 2.77 −99.90 −9.40 −23.26 4.06 169.81 39.53 4.37
24 −5.00 19.45 52.13 −5.79
25 0.04 −83.25 −66.21 −96.70 19.18 214.28 0.94 2.33
26 2.83 −93.09 −6.34 −60.19 8.49 198.21 27.91 3.86
27 0.06 −4.18 −44.82 −3.96 1.89 165.35 56.65 6.94
 X. Chen et al. / Agricultural and Forest Meteorology 234 (2017) 222–235 233

Table A3 (Continued)

Station number a b c d e C* F* AIC

28 0.07 46.06 45.65 −84.60 18.58 220.76 2.40 2.59

29 −0.01 2.41 58.12 −4.64
30 3.86 −72.43 −5.65 −45.41 2.21 12.11 156.56 2.41
31 −0.50 18.16 59.56 −3.69
32 −0.03 6.04 59.48 −3.82
33 −1.01 4.57 59.76 −6.86
34 −0.03 4.33 65.91 −0.37
35 −0.71 10.43 59.47 −3.84
36 −0.31 12.45 57.34 −2.85
37 −2.30 7.47 61.27 −7.62
38 −1.15 6.02 66.07 −4.70
39 −0.02 11.51 58.65 −5.01
40 −0.02 15.30 55.42 −3.82
41 −0.53 18.70 58.19 −5.44
42 −0.70 8.61 62.94 −3.01

Table A4
Parameters and performance of the optimum models for 50% flowering at each station (AIC = AICUniForc − AICUniChill ).

Station number a b c d e C* F* AIC

1 0.29 −6.47 46.30 −41.08 5.36 10.04 146.81 8.28

2 4.80 −78.70 10.98 −8.28 11.98 100.24 56.38 2.34
3 0.53 −14.73 0.87 −71.74 8.89 134.24 57.42 2.34
4 0.16 −61.29 5.34 −64.93 9.39 171.54 25.54 8.15
5 −0.40 3.86 81.77 −6.06
6 5.32 −23.07 7.28 −50.23 12.23 11.11 43.21 1.49
7 −54.00 2.58 86.65 −2.78
8 −2.93 3.14 85.94 −5.70
9 −62.65 1.93 82.56 −5.73
10 −13.84 4.09 86.14 −1.82
11 −9.39 18.22 9.89 −3.43
12 −2.14 0.21 96.79 −3.90
13 −1.94 0.77 89.99 −4.82
14 −0.09 18.43 31.06 −1.68
15 −97.26 10.97 39.35 −1.18
16 −0.77 5.92 60.58 −1.13
17 −26.24 11.43 33.53 −0.82
18 −0.01 5.64 51.70 −2.75
19 −1.42 1.65 87.99 −6.73
20 −0.05 11.65 51.96 −2.73
21 −6.07 0.02 92.78 −2.98
22 −0.01 17.97 49.17 −5.44
23 −0.02 8.11 51.82 −4.80
24 −0.03 17.17 53.29 −5.48
25 −0.01 11.48 54.23 −5.61
26 −0.01 14.79 55.45 −4.21
27 −0.05 8.48 55.56 −6.40
28 −0.02 0.07 65.16 −3.42
29 −0.01 3.48 63.04 −5.09
30 −1.63 6.02 107.00 −5.01
31 −0.01 8.51 62.52 −4.69
32 −0.03 7.41 60.56 −3.84
33 −0.01 5.30 62.23 −5.73
34 −0.04 3.84 69.87 −6.89
35 −0.03 3.46 65.85 −3.66
36 −0.04 8.00 60.80 −4.00
37 −0.02 11.26 63.43 −7.51
38 −0.03 3.60 68.12 −5.32
39 −0.02 4.14 64.60 −6.56
40 −0.01 4.47 66.18 −5.17
41 −0.03 2.80 61.10 −6.78
42 −0.05 9.15 64.56 −6.13
234 X. Chen et al. / Agricultural and Forest Meteorology 234 (2017) 222–235
Fig. A1. Average time series of each phenophase during 1981–2005 within each climate zone. (a) first leaf unfolding, (b) 50% leaf unfolding, (c) first flowering, and (d) 50%
flowering.
References
Akaike, H., 1973. Information theory and an extension of the maximum likelihood
principle. In: Petrov, B.N., Csaki, F. (Eds.), Proceedings of the Second
International Symposium on Information Theory. Akademiai Kiado, Budapest,
pp. 267–281.
Bernier, G., 1988. The control of floral evocation and morphogenesis. Ann. Review
Plant Physiol. 39, 175–219.
Cannell, M., Smith, R.I., 1983. Thermal time, chill days and prediction of budburst
in Picea sitchensis. J. Appl. Ecol. 20 (3), 951–963.
Chen, X.Q., Xu, L., 2012. Phenological responses of Ulmus pumila (Siberian Elm) to
climate change in the temperate zone of China. Int. J. Biometeorol. 56 (4),
695–706.
Chen, X.Q., Tian, Y., Xu, L., 2015. Temperature and geographic attribution of change
in the Taraxacum mongolicum growing season from 1990 to 2009 in eastern
China’s temperate zone. Int. J. Biometeorol., http://dx.doi.org/10.1007/s00484-
015-0955-4.
Chen, X.Q., 2013. East Asia. In: Schwartz, M.D. (Ed.), Phenology: An Integrative
Environmental Science. Springer Netherlands, Dordrecht, pp. 9–22.
China Meteorological Administration, 1979. Atlas of the Climate of China.
Sinomaps Press, Beijing, pp. 222–223 (in Chinese).
China Meteorological Administration, 1993. Observation Criterion of Agricultural
Meteorology. China Meteorological Press, Beijing, pp. 139–140 (in Chinese).
Chuine, I., Cour, P., Rousseau, D.D., 1998. Fitting models predicting dates of
flowering of temperate-zone trees using simulated annealing. Plant Cell
Environ. 21 (5), 455–466.
Chuine, I., Cour, P., Rousseau, D.D., 1999. Selecting models to predict the timing of
flowering of temperate trees: implications for tree phenology modelling. Plant
Cell Environ. 22 (1), 1–13.
Chuine, I., Kramer, K., Hänninen, H., 2003. Plant development models. In: Schwartz,
M.D. (Ed.), Phenology: An Integrative Environmental Science. Kluwer
Academic Publishers, Dordrecht, Boston, London, pp. 217–235.
Chuine, I., Morin, X., Bugmann, H., 2010. Warming, photoperiods, and tree
phenology. Science 329, 277–278, author reply 278.
Chuine, I., 2000. A unified model for budburst of trees. J. Theor. Biol. 207 (3),
337–347.
Cook, B.I., Wolkovich, E.M., Parmesan, C., 2012. Divergent responses to spring and
winter warming drive community level flowering trends. Proc. Natl. Acad. Sci.
U.S. A. 109, 9000–9005.
Editorial Committee of Flora Repubulicae Popularis Sinicae, 1997. Flora
Repubulicae Popularis Sinicae, Vol. 43. Science Press, Beijing, pp. 100 (3) (in
Chinese).
Fitter, A.H., Fitter, R.S.R., 2002. Rapid changes in flowering time in British plants.
Science 296, 1689–1691.
Fitter, A.H., Fitter, R.S.R., Harris, I.T.B., Williamson, M.H., 1995. Relationships
between first flowering date and temperature in the flora of a locality in
central England. Funct. Ecol. 9 (1), 55–60.
Fu, Y.H., Campioli, M., Van Oijen, M., Deckmyn, G., Janssens, I.A., 2012. Bayesian
comparison of six different temperature-based budburst models for four
temperate tree species. Ecol. Modell. 230, 92–100.
Giorgi, F., Jones, C., Asrar, G.R., 2009. Addressing climate information needs at the
regional level: the CORDEX framework. WMO Bull. 58, 175–183.
Gordo, O., Sanz, J.J., 2010. Impact of climate change on plant phenology in
Mediterranean ecosystems. Global Change Biol. 16 (3), 1082–1106.
Hänninen, H., Kramer, K., 2007. A framework for modelling the annual cycle of
trees in boreal and temperate regions. Silva Fenn. 41 (1), 167–205.
Hänninen, H., Kellomäki, S., Laitinen, K., Pajari, B., Repo, T., 1993. Effect of increased
winter temperature on the onset of height growth of Scots pine: a field test of a
phenological model. Silva Fenn. 27 (4), 251–257.
Hänninen, H., 1990. Modelling bud dormancy release in trees from cool and
temperate regions. Acta For. Fenn. 213, 1–47.
Hatta, H., Darnaedi, D., 2005. Phenology and Growth Habits of Tropical Trees.
National Science Museum, Tokyo, pp. 1–436.
Hunter, A.F., Lechowicz, M.J., 1992. Predicting the timing of budburst in temperate
trees. J. Appl. Ecol. 29 (3), 597–604.
Kramer, K., 1994a. Selecting a model to predict the onset of growth of Fagus
sylvatica. J. Appl. Ecol. 31 (1), 172–181.
Kramer, K., 1994b. A modelling analysis of the effects of climatic warming on the
probability of spring frost damage to tree species in The Netherlands and
Germany. Plant Cell Environ. 17 (4), 367–377.
Laube, J., Sparks, T.H., Estrella, N., Höfler, J., Ankerst, D.P., Menzel, A., 2014. Chilling
outweighs photoperiod in preventing precocious spring development. Global
Change Biol. 20, 170–182.
Linkosalo, T., Lappalainen, H.K., Hari, P., 2008. A comparison of phenological
models of leaf bud burst and flowering of boreal trees using independent
observations. Tree Physiol. 28 (12), 1873–1882.
Luedeling, E., Girvetz, E.H., Semenov, M.A., Brown, P.H., 2011. Climate change
affects winter chill for temperate fruit and nut trees. PLoS One 6 (5), 1–13.
Luo, X., Chen, X.Q., Wang, L., Xu, L., Tian, Y., 2014. Modeling and predicting spring
land surface phenology of the deciduous broadleaf forest in northern China.
Agric. For. Meteorol. 198, 33–41.
Matsumoto, K., Ohta, T., Irasawa, M., Nakamura, T., 2003. Climate change and
extension of the Ginkgo biloba L: growing season in Japan. Global Change Biol.
9 (11), 1634–1642.
Menzel, A., 2003. Plant phenological anomalies in Germany and their relation to air
temperature and NAO. Clim. Change 57 (3), 243–263.
Morin, X., Lechowicz, M.J., Augspurger, C., O’Keefe, J., Viner, D., Chuine, I., 2009.
Leaf phenology in 22 North American tree species during the 21 st century.
Global Change Biol. 15 (4), 961–975.
 X. Chen et al. / Agricultural and Forest Meteorology 234 (2017) 222–235
Mouradov, A., Cremer, F., Coupland, G., 2002. Control of flowering time: interacting
pathways as a basis for diversity. Plant Cell 14 (Suppl), S111–S130.
Murray, M.B., Cannell, M., Smith, R.I., 1989. Date of budburst of fifteen tree species
in Britain following climatic warming. J. Appl. Ecol. 26 (2), 693–700.
Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through conceptual models
part I—A discussion of principles. J. Hydrol. 10 (3), 282–290.
Nelson, E.A., Lavender, D.P., 1979. The chilling requirement of western hemlock
seedlings. For. Sci. 25, 485–490.
Parker, A.K., de Cortazár-Atauri, I.G., Van Leeuwen, C., Chuine, I., 2011. General
phenological model to characterise the timing of flowering and veraison of
Vitis vinifera L. Aust. J. Grape Wine Res. 17 (2), 206–216.
Parmesan, C., 2007. Influences of species, latitudes and methodologies on
estimates of phenological response to global warming. Global Change Biol. 13
(9), 1860–1872.
Peñuelas, J., Filella, I., 2001. Responses to a warming world. Science 294 (5543),
793–795.
Riahi, K., Rao, S., Krey, V., Cho, C., Chirkov, V., Fischer, G., Kindermann, G.,
Nakicenovic, N., Rafaj, P., 2011. RCP 8.5-A scenario of comparatively high
greenhouse gas emissions. Clim. Change 109, 33–57.
Richardson, A.D., O’Keefe, J., 2009. Phenological differences between understory
and overstory. In: Noormets, A. (Ed.), Phenology of Ecosystem Processes.
Springer, Dordrecht, Heidelberg, London, New York, pp. 87–117.
Roberts, A.M.I., Tansey, C., Smithers, R.J., Phillimore, A.B., 2015. Predicting a change
in the order of spring phenology in temperate forests. Global Change Biol. 21,
2603–2611.
View publication stats
235
Sanchez-Azofeifa, A., Kalacska, M.E., Quesada, M., Stoner, K.E., Lobo, J.A.,
Arroyo-Mora, P., 2013. Tropical dry climates. In: Schwartz, M.D. (Ed.),
Phenology: An Integrative Environmental Science. Springer Netherlands,
Dordrecht, pp. 157–171.
Thomson, A.M., Calvin, K.V., Smith, S.J., Kyle, G.P., Volke, A., Patel, P., Delgado-Arias,
S., Bond-Lamberty, B., Wise, M.A., Clarke, L.E., Edmonds, J.A., 2011. RCP4.5: A
pathway for stabilization of radiative forcing by 2100. Clim. Change 109, 77–94.
Vitasse, Y., Delzon, S., Dufrêne, E., Pontailler, J.-Y., Louvet, J.-M., Kremer, A.,
Michalet, R., 2009. Leaf phenology sensitivity to temperature in European
trees: do within-species populations exhibit similar responses? Agric. For.
Meteorol. 149, 735–744.
Vitasse, Y., François, C., Delpierre, N., Dufrêne, E., Kremer, A., Chuine, I., Delzon, S.,
2011. Assessing the effects of climate change on the phenology of European
temperate trees. Agric. For. Meteorol. 151 (7), 969–980.
Walther, G., Post, E., Convey, P., Menzel, A., Parmesan, C., Beebee, T.J.C., Fromentin,
J., Hoegh-Guldberg, O., Bairlein, F., 2002. Ecological responses to recent climate
change. Nature 416 (6879), 389–395.
Xu, L., Chen, X.Q., 2013. Regional unified model-based leaf unfolding prediction
from 1960 to 2009 across northern China. Global Change Biol. 19 (4),
1275–1284.
Zhao, J., Zhang, Y., Song, F., Xu, Z., Xiao, L., 2013. Phenological response of tropical
plants to regional climate change in Xishuangbanna, south-western China. J.
Trop. Ecol. 29, 161–172.