Herbarium specimens reveal substantial

and unexpected variation in phenological

rstb.royalsocietypublishing.org sensitivity across the eastern United States
Daniel S. Park1,†, Ian Breckheimer1,†, Alex C. Williams2, Edith Law2,
Aaron M. Ellison3 and Charles C. Davis1
Research 1
Department of Organismic and Evolutionary Biology and Harvard University Herbaria, Harvard University,
Cite this article: Park DS, Breckheimer I, Cambridge, MA 02138, USA
2
Williams AC, Law E, Ellison AM, Davis CC. 2018 David R. Cheriton School of Computer Science, University of Waterloo, Waterloo, Ontario, Canada N2L 3G1
3
Harvard Forest, Harvard University, Petersham, MA 01366, USA
Herbarium specimens reveal substantial and
unexpected variation in phenological sensitivity CCD, 0000-0001-8747-1101
Downloaded from https://royalsocietypublishing.org/ on 08 November 2024

across the eastern United States. Phil.

Phenology is a key biological trait that can determine an organism’s survival
Trans. R. Soc. B 374: 20170394. and provides one of the clearest indicators of the effects of recent climatic
http://dx.doi.org/10.1098/rstb.2017.0394 change. Long time-series observations of plant phenology collected at conti-
nental scales could clarify latitudinal and regional patterns of plant
Accepted: 15 October 2018 responses and illuminate drivers of that variation, but few such datasets
exist. Here, we use the web tool CrowdCurio to crowdsource phenological
data from over 7000 herbarium specimens representing 30 diverse flowering
One contribution of 16 to a theme issue plant species distributed across the eastern United States. Our results, span-
‘Biological collections for understanding ning 120 years and generated from over 2000 crowdsourcers, illustrate
biodiversity in the Anthropocene’. numerous aspects of continental-scale plant reproductive phenology. First,
they support prior studies that found plant reproductive phenology signifi-
cantly advances in response to warming, especially for early-flowering
Subject Areas: species. Second, they reveal that fruiting in populations from warmer,
ecology, evolution, taxonomy and systematics lower latitudes is significantly more phenologically sensitive to temperature
than that for populations from colder, higher-latitude regions. Last, we
Keywords: found that variation in phenological sensitivities to climate within species
between regions was of similar magnitude to variation between species.
citizen science, digitization, geographical
Overall, our results suggest that phenological responses to anthropogenic cli-
range, herbarium specimens, phenology, mate change will be heterogeneous within communities and across regions,
phenological sensitivity with large amounts of regional variability driven by local adaptation, pheno-
typic plasticity and differences in species assemblages. As millions of
Authors for correspondence: imaged herbarium specimens become available online, they will play an
increasingly critical role in revealing large-scale patterns within assemblages
Daniel S. Park
and across continents that ultimately can improve forecasts of the impacts of
e-mail: [email protected] climatic change on the structure and function of ecosystems.
Charles C. Davis This article is part of the theme issue ‘Biological collections for
e-mail: [email protected] understanding biodiversity in the Anthropocene’.

1. Introduction
Ecosystems on every continent have been affected by local, regional and global
changes in climate, especially increases in temperature [1]. Changes in phenol-
ogy—the timing of life-history events—are among the most conspicuous and
well-documented species responses to climatic change, especially for plants
[2– 7]. Phenological disruption has already impacted species’ local persistence
†
These authors contributed equally to this and community diversity [8– 10], which may have widespread consequences
for critical ecosystem processes, including carbon sequestration [11–13],
work.
ecosystem –atmosphere interactions [14] and trophic interactions [15 –28].
Despite these trends, our knowledge of plant phenological responses to cli-
Electronic supplementary material is available matic change remains inadequate. In particular, although phenological
online at https://dx.doi.org/10.6084/m9. responses may differ among species with different functional or life-history
figshare.c.4274660. traits and biogeographical origins [29 –32], long-term observational datasets

& 2018 The Author(s) Published by the Royal Society. All rights reserved.
 to assess such trends are limited in geographical, temporal temperature across 238 of latitude in the eastern United 2
and taxonomic scope [33]. Many of these data track woody States. We examined both native and introduced plant

rstb.royalsocietypublishing.org
plant species of the Northern Hemisphere (most commonly, species from northern coniferous forests, eastern deciduous
abundant tree species), and only for the last approximately forests, subtropical evergreen forests, grasslands, wetlands,
40 years (but see [34,35]). These biases limit our understand- alpine meadows and aquatic plant communities. Environ-
ing of variation in phenological responses across species and mental conditions in this region vary considerably across
biomes. Furthermore, although population-level variation in species’ ranges, and populations may exhibit substantial
phenology has been demonstrated for a few species [36–38], variation in phenological response across this latitudinal
there are very few studies that quantify both inter- and gradient. Our overall goals with this study were: (i) to
intraspecific variation in phenological response [32,33]. demonstrate the power of characterizing phenology from
Variability in species’ phenology is particularly relevant herbarium data using an efficient and rapid workflow that
because climatic change is not geographically uniform. For leverages a nearly fully mobilized online flora of the eastern

Phil. Trans. R. Soc. B 374: 20170394

example, high-latitude regions are warming faster than sub- United States [56,57]; (ii) to greatly increase the taxonomic
tropical and tropical ones [1,39]. Short growing seasons also and ecological diversity of species sampled for this purpose
may cause high-latitude ecosystems to be especially sensitive (from woody perennials to herbaceous annuals across a
to temperature, leading to stronger selective pressures for range of biomes); and (iii) to sample species with broad lati-
populations to initiate growth as soon as conditions become tudinal ranges to assess regional and inter- and intra-species
Downloaded from https://royalsocietypublishing.org/ on 08 November 2024

favourable in early spring. Additionally, plants adapted to variation in phenological responses.

highly variable climates may exhibit higher phenological
thresholds to temperature, as it provides a less reliable
signal [40]. Thus, the effects of climatic change on species’
phenology may differ across their ranges depending on vari- 2. Methods
ation in phenological sensitivity. Variability in phenological
responses to climatic change within species may also alter
(a) Specimen data collection
We examined phenological responses of species using digitized
patterns of gene flow, which could either promote or counter- specimens from two of the most comprehensive digitized regional
act adaptive evolution via the sharing of locally (mal)adapted floras in the world, the Consortium of Northeastern Herbaria
alleles [41–44]. Recent studies have shown that plant phenol- (CNH; http://portal.neherbaria.org/portal/) [56] and Southeast
ogy may be more responsive in more northern-ranging Regional Network of Expertise and Collections (SERNEC;
populations where there are more variable and extreme cli- http://sernecportal.org/portal/index.php) [57]. These two
mates, especially during the early part of the growing online portals include more than six million digitized herbarium
season [44,45]. Although these studies are suggestive, they records, including specimen images. Our criteria for selecting
are restricted in spatial scale and taxonomic scope, and angiosperm species for analysis were that specimens: (i) included
at least county-level location data; (ii) included at least 50 unique
broad regional patterns of phenological response to climate
collections across space and time; (iii) were of species with rela-
may differ from patterns at smaller, local scales.
tively easily identifiable and quantifiable reproductive structures;
Herbarium specimens represent snapshots of phenology and (iv) were from species with broad latitudinal ranges sufficient
(i.e. flowering and fruiting) at a specific place and time, and to enable quantification of population-level variation.
have shown tremendous promise to increase the spatial, tem- Applying these criteria yielded 30 species with varying life-
poral and taxonomic resolution of phenological data [46– 48]. history traits, growth forms, native status and general reproduc-
They provide rich historical depth, wide geographical scope tive seasonality (e.g. early- versus late-spring flowering). We
and taxonomic diversity, all of which allow researchers to downloaded over 10 000 digital herbarium specimen images of
track long-term changes of vast numbers of species and com- these species from CNH and SERNEC, removed duplicate, mis-
munities through space and time [4,46– 49]. Despite their identified or sterile specimens, and those with notable insect
representation of phenological responses [48,50], herbarium damage on reproductive structures, extensive physical damage
or poor preservation. We also removed all 30 specimens from
specimens have been used less frequently than other data
Florida, which were geographical and climatic outliers. Our
sources, such as field observations, to address phenological
final dataset comprised 7722 specimens and spanned 120 years
change, in part because they have been inaccessible to across 512 United States counties (table 1). Species’ life-history
many researchers [46,51]. However, the widespread digitiz- (annual versus perennial), growth form (woody versus herbac-
ation of herbarium collections [52] combined with new eous) and native status (native versus introduced) were
approaches to collecting [46,53] and analysing [54,55] pheno- inferred from the United States Department of Agriculture
logical data derived from herbarium specimens has the PLANTS Database (https://plants.usda.gov/). For individual
power to transform our understanding of plant responses specimen metadata, see electronic supplementary material,
to global climatic change. table S1.
Here, we applied a newly developed web-enabled crowd-
sourcing platform, CrowdCurio: Thoreau’s Field Notes (https://
www.crowdcurio.com/) [46], to examine more than 7000 (b) Crowdsourcing phenological data collection
specimens of 30 phylogenetically diverse flowering plant The phenological state ( phenophase) of a plant can be inferred
from the presence and quantity of relevant structures, such as
species. The Thoreau’s Field Notes module facilitates the
leaves, flowers or fruits [46,48]. Past researchers generally have
rapid quantification of phenological traits via image annota-
focused on the presence or the absence of a single structure or
tion and has been demonstrated to yield reliable data trait (e.g. [58]) or applied majority estimates for scoring a
regardless of the level of expertise among crowdsourcers single phenophase (e.g. [49]). Here, we quantified data for two
(i.e. expert versus non-expert scoring) [46]. We used these reproductive phenophases, flowering and fruiting. Specimens
crowdsourced data to infer the magnitude, direction, and were scored as flowering if open flowers represented greater
variability in reproductive phenological responses to spring than or equal to 50% of the total reproductive structure count
 Table 1. The number of specimens and different categorical traits of examined species. FL and FR refer to the number of specimens classified as flowering and 3
fruiting, respectively.

rstb.royalsocietypublishing.org
time span
family species (years) FL FR lifespan growth form status

Ranunculaceae Anemone canadensis L. 116 41 59 perennial herbaceous native

Ranunculaceae Anemone hepatica L. 95 40 60 perennial herbaceous native
Ranunculaceae Aquilegia canadensis L. 119 251 238 perennial herbaceous native
Asteraceae Bidens vulgata Greene 119 83 136 annual herbaceous native
Celastraceae Celastrus orbiculatus Thunb. 103 151 220 perennial herbaceous introduced

Phil. Trans. R. Soc. B 374: 20170394

Asteraceae Centaurea stoebe Tausch 111 93 161 perennial herbaceous introduced
Asteraceae Cirsium arvense (L.) Scop. 118 186 171 perennial herbaceous introduced
Asteraceae Cirsium discolor (Muhl. ex Willd.) 117 46 93 perennial herbaceous native
Spreng.
Downloaded from https://royalsocietypublishing.org/ on 08 November 2024

Geraniaceae Geranium maculatum L. 119 489 513 perennial herbaceous native

Geraniaceae Geranium robertianum L. 119 48 307 perennial herbaceous native
Xanthorrhoeaceae Hemerocallis fulva (L.) L. 115 144 45 perennial herbaceous introduced
Malvaceae Hibiscus moscheutos L. 119 105 100 perennial herbaceous native
Balsaminaceae Impatiens capensis Meerb. 120 153 501 annual herbaceous native
Iridaceae Iris pseudacorus L. 117 90 66 perennial herbaceous introduced
Iridaceae Iris versicolor L. 119 344 185 perennial herbaceous native
Liliaceae Lilium canadense L. 117 139 27 perennial herbaceous native
Caprifoliaceae Lonicerabella Zab. 107 37 62 perennial woody introduced
Caprifoliaceae Lonicera canadensis Bartram 120 194 201 perennial woody native
ex Marshall
Caprifoliaceae Lonicera japonica Thunb. 116 329 115 perennial woody introduced
Rosaceae Malus pumila Mill. 118 74 40 perennial woody introduced
Malvaceae Malva neglecta Wallr. 116 25 140 perennial herbaceous introduced
Onagraceae Oenothera perennis L. 120 194 214 perennial herbaceous native
Orobanchaceae Orobanche uniflora L. 118 213 105 annual herbaceous native
Rosaceae Rosa gallica L. 108 45 17 perennial woody introduced
Rosaceae Rubus odoratus L. 120 176 318 perennial woody native
Sarraceniaceae Sarracenia purpurea L. 119 234 75 perennial herbaceous native
Iridaceae Sisyrinchium mucronatum Michx. 117 86 157 perennial herbaceous native
Solanaceae Solanum rostratum Dunal 115 21 85 annual herbaceous native
Melanthiaceae Trillium grandiflorum (Michx.) Salisb. 119 129 40 perennial herbaceous native
Melanthiaceae Trillium undulatum Willd. 120 402 156 perennial herbaceous native

and scored as fruiting if they had less than 50% flowers and buds difference in organ counts for each phenophase by the total
and at least one fruit present. We used the Thoreau’s Field Notes count of that specimen across the two duplicate specimens and
instance of CrowdCurio to crowdsource phenological data from subtracting this value from 1 (1 – (jcount1 – count2j/(count1 þ
digitized herbarium specimens [46]. Citizen scientists hired count2)) [55]. Reliability scores range from zero (unreliable/
through Amazon’s Mechanical Turk service (MTurk; https:// inconsistent) to one (reliable/consistent). Participants who
www.mturk.com/) counted the number of buds, flowers and reported no organs on one sheet and a non-zero number of the
fruits observed for a set of 10 specimen images. Participants same organ on the duplicate sheet were assigned a reliability
first watched a short (1 min) instructional video on how to score of zero for that organ (i.e. the lowest reliability score).
score phenological traits using CrowdCurio and then were We conservatively selected the lowest reliability score among
provided with three tutorial images of each reproductive struc- the three calculated for each organ per participant and
ture for every species. The 2364 anonymous participants were assigned it to each participant as their final score. That is, if a
compensated at a rate of $0.12 per image. participant got a reliability score equal to zero on one organ,
To provide an estimate of measurement error, each 10-image they would be assigned a reliability score of zero for all
set scored by a single crowdsourcer included nine unique images organs. Specimen observations scored by crowdsourcers with a
and a single duplicate image randomly selected from the other reliability score of zero were excluded from the analysis. We
nine [55]. We estimated the reliability score for each participant also spot checked for suspicious outliers manually and removed
based on the data for each 10-image set by dividing the absolute such data.
 (c) Historical climate data The linear predictor m[i] was estimated as a function of covariates, 4
We used estimates of historic (1895–2016) average monthly temp- including mean spring (March– May) air temperatures
(SpringT[i]) and the average elevation of the county in which

rstb.royalsocietypublishing.org
erature and precipitation data at 4 km resolution from PRISM
(product AN81 m; http://prism.oregonstate.edu/), which provide the specimen was recorded (Elev[c]). Additional intercept terms
high-resolution time-series estimates of climatic elements for the (a1– a5) were added for each species (s), region (r), species 
contiguous United States. As accurate locality data are not available region combination (sr), county (c) and observer (o). The full
for the majority of historic specimen records [59], we used county as expression for estimating m[i] was
our geographical unit of analysis. The vast majority (79%) of speci- m[i] ¼ a1[s] þ a2[r] þ a3[sr] þ a4[c] þ a5[o] þ b1[s]  SpringT[i]
mens used in this study were collected before 1980, and while 72%
of the specimens used in this study had associated coordinate data, þ b2[r]  SpringT[i] þ b3[sr]  SpringT[i] þ b4[s]  Elev[c] :
at least 91% of those coordinates had been georeferenced post hoc ð2:2Þ
(e.g. assigned county or township centroid coordinates), and thus
Species-specific slope and intercept terms (a1[s], b1[s] and

Phil. Trans. R. Soc. B 374: 20170394

may not represent precise sampling locales. For each county and
year, we estimated the mean monthly temperature, precipitation b4[s] in equation (2.2)) were drawn from normal distribu-
and elevation, and assigned these values to each specimen [59]. tions, with species assemblage means ma1, mb1 and mb4, and
Although counties can vary in size and climate, counties in states hypervariances sa1, sb1 and sb4.
along the Atlantic coast of the United States are generally small in
a1[s]  N(ma1 , sa1 ), ð2:3Þ
size and geographically homogeneous. We estimated within-
Downloaded from https://royalsocietypublishing.org/ on 08 November 2024

county climatic heterogeneity as the standard deviation of b1[s]  N(mb1 , sb1 ) ð2:4Þ
estimated monthly climatic values across each county and year and b4[s]  N(mb4 , sb4 ): ð2:5Þ
and included it in our initial analyses, but coefficients had Bayesian
credibility intervals that were not credibly different from zero, so we Region and species  region slopes (b2[r], b3[sr]), and region,
dropped these terms from our final models. species  region, county and observer intercepts (a2[r], a3[sr],
a4[c], a5[o]) in equation (2.2) were drawn from zero-centred
normal distributions with hypervariances sb2, sb3, sa2, sa3, sa4
and sa5, respectively. The species-specific sampling variation
(d) Statistical analyses (s[s]) terms in equation (2.1) were estimated independently to
Phenological sensitivity to spring temperature—defined as the
account for differences in the duration of flowering and fruiting
mean of March, April and May temperatures [46]—was defined
phases between species.
as the slope of the linear relationship between the day of year
The three different groups of slopes estimated for spring
(DOY) of a phenophase and the spring temperature of the corre-
temperature decomposed variation in phenological sensitivity
sponding location and year (shifts in days per degree Celsius
into components representing between-species variability (b1[s]),
change: days/8C) [44,46]. These months have been used to
between-region variability (b2[r]) and within-species variability
define spring across the east coast of the United States [46,60].
across regions (b3[sr]). The accompanying hypervariances (sb1,
To calculate phenological sensitivity, we binned our specimen
sb2, sb3) directly represented these different sources of variability;
data into both broad climatic zones [61,62] and finer-scale
comparing their relative magnitudes quantified the contributions
National Ecological Observatory Network (NEON) domains.
of each source of variation to overall variation in phenological sen-
Our data comprised two climatic zones (cold/very cold; mixed-
sitivity. This model structure also provided estimates of the
humid/hot-humid—hereafter referred to as cold and mixed-
contributions of species turnover to differences in phenological
warm) and five NEON ecoclimatic domains (NE, northeast;
sensitivity between regions. We estimated these contributions
MA, mid-Atlantic; AP, Appalachians & Cumberland Plateau;
by analysing the output of Model 1, computing phenological sen-
OZ, Ozarks Complex; SE, southeast; electronic supplementary
sitivities for each observation for each iteration of our model. We
material, table S2). We also estimated phenological sensitivity to
then used the mean and standard deviation of these estimates for
elevation as the slope of the linear relationship between the
each region to create region-specific estimates of mean phenologi-
DOY of a phenophase and metres above sea level (m a.s.l.).
cal sensitivities and their variability. We assessed the contribution
We estimated the mean timing of flowering and fruiting phe-
of community turnover by comparing estimates that included all
nophases, and environmental influences on them, using Bayesian
three climate sensitivity terms (b1[s] þ b2[r] þ b3[sr]) with esti-
hierarchical linear regression models [63]. In our models, species,
mates that included only the terms that represent species-level
region, county and observer were considered random effects,
variability in climate sensitivity (b1[s]). This strategy allowed us
while spring temperature and county elevation were covariates.
to infer what the mean phenological sensitivities would be
The hierarchical nature of the model, in which the phenological
across regions in the hypothetical case that they differed only in
responses of individual species were assumed to be drawn
species composition, and species responded identically to climate
from statistical distributions instead of fixed estimates [64],
across their ranges.
allowed us to better estimate their climatic sensitivities. These
Model 2 differed from Model 1 in treating the region term
models also more accurately quantified uncertainty in our esti-
(b2[r]) as a two-level fixed effect representing the climatic
mates and partitioned the variance in phenological timing and
region from which the specimen was drawn.
phenological sensitivity within and between species and regions.
We fitted two models for each phenological phase. ‘Model 1’ m[i] ¼ a2[r] þ a3[sr] þ a4[c] þ a5[o] þ b2[r]  SpringT[i]
estimated species-specific phenological sensitivities and parti- ð2:6Þ
tioned their variances. ‘Model 2’ provided a more powerful þ b3[sr]  SpringT[i] þ b4[s]  Elev[c] :
comparison between phenological sensitivities found in the
warmer, lower latitudes of our study area and those in Model 2 maximized statistical power to compare overall phe-
the cooler, higher-latitude regions (figure 1). nological sensitivities of species between warmer, more southerly
In Model 1, the dependent variable was the DOY for which a parts of our study area and cooler, more northerly areas. Instead
given phenological phase (flowering or fruiting) was recorded of treating region-specific slopes and intercepts as normally dis-
for the ith specimen. DOY[i] was assumed to be normally tributed random grouping factors, we represented species 
distributed, with mean m[i] and species-specific variance s[s]. region, county and observer terms as zero-centred and normally
distributed. This structure allowed more direct inference about
DOY[i]  N(m[i] , s[s] ) : ð2:1Þ overall differences in phenological sensitivity between cool and
 (a) (c) 5

rstb.royalsocietypublishing.org
45 45

40
40
latitude

NEON domain
Appalachians (AP) 35
35 mid-Atlantic (MA)
northeast (NE)

Phil. Trans. R. Soc. B 374: 20170394

Ozarks Complex (OC)
30 southeast (SE) 30
–85 –80 –75 –70 0 5 10 15 20
longitude spring temperature (ºC)
Downloaded from https://royalsocietypublishing.org/ on 08 November 2024

(b) (d )

45 45

40
latitude

40

35 climate zone 35
cold
mixed-warm

30 30
–85 –80 –75 –70 0 300 600 900
longitude elevation a.s.l. (m)

Figure 1. Distribution of herbarium specimens across geographical and environmental space, colour-coded by NEON domain (a,c) or climate region (b,d ). Panels (a)
and (c) show specimens in flower, and panels (b) and (d) show specimens in fruit. Specimens are referenced to county centroids, and jitter has been added to
coordinates to reduce over-plotting. Average spring (March – May) air temperatures are strongly negatively correlated with latitude (Pearson correlation r ¼ 20.92,
panel c), but elevation and latitude are largely independent (panel d ).

mixed-warm climate regions (differences between estimates of transformed to the original data scale to facilitate illustration
b2[r] for the two regions). Unlike Model 1, however, Model 2 and interpretation.
lacks species-specific parameters (a1[s] and b1[s]) and did not Code and data for reproducing these analyses are archived
provide species-specific estimates of phenological sensitivity or by Harvard Forest [71].
permit comparison of variability in phenological sensitivity
within and between species.
We estimated all parameters of the two models using
Hamiltonian Monte Carlo (HMC) [65] implemented in Stan 3. Results
(v. 2.17.3) [66] called from the rstan interface [67] in R [68].
Our focal species spanned wide geographical and climatic
HMC is a form of Markov chain Monte Carlo (MCMC) that effi-
ciently estimates hierarchical Bayesian regression models for space (figure 1). They demonstrated diverse patterns of
larger datasets like ours [69]. We used relatively uninformative phenology and significant variation in responses to climate
prior distributions: zero-centred normal priors for slopes and across species and geographies. Using Model 1, estimated
intercepts, and truncated normal distributions for variances mean (non-leap-year) flower timing at 7.48C and 216 m a.s.l.
and hypervariances. To account for sampling behaviour and sim- (mean collection conditions for the specimens) varied from
plify prior choices, we scaled and centred the response variable 10 May (Day 130, Anemone hepatica) to 10 September (Day
DOY[i] and all continuous predictors by subtracting the mean 253, Bidens vulgata) for flowering and 22 May (Day 142, A.
and dividing by the standard deviation of each variable. For hepatica) to 14 September (Day 257, B. vulgata) for fruiting
each model run for each phenophase, we estimated parameters (figure 2). The average lag time between flowering and fruit-
using four MCMC chains of 4000 iterations each and discarded
ing across all species was approximately 20 days. Most
the first 2000 iterations of each chain (as burn-in). We assessed
species flowered and fruited earlier with warmer spring temp-
convergence of each model both visually and with the
Gelman– Rubin statistic (^r , 1:1 for all parameters). We also eratures (assemblage mean 22.56 days/8C, 95% CI 23.64 to
assessed good model fit using visual posterior predictive 21.48, figure 2), and these responses were credibly different
checks implemented in the bayesplot R package [70]. All from zero ( posterior probability . 0.95) for 21 out of 30
parameter estimates were based on at least 1000 effective pos- species for flowering and 15 out of 30 species for fruiting
terior samples. Estimates reported in the results were back- (electronic supplementary material, tables S3 and S4).
 flowering fruiting 6
(a) (b) species

rstb.royalsocietypublishing.org
1 1 Anemone canadensis 16 Lilium canadense
phenological sensitivity
0 4
13
22 13
8
4 2 Anemone hepatica 17 Lonicera bella
−1 25
6
8
Aquilegia canadensis Lonicera canadensis
(days/°C)

25 3 18
19 10
−2 30 5 26 2
26
15
28 19
2918 21 28 12 22 16 24
10
17 7 17 21 12 5 4 Bidens vulgata 19 Lonicera japonica
−3 2 3
209
1
24
16
6 18
3 11
30
7
27 15 29 1
−4 27 5 Celastrus orbiculatus 20 Malus pumila
11 9 14
23 14
23
−5 20 6 Centaurea stoebe 21 Malva neglecta
−6 7 Cirsium arvense 22 Oenothera perennis
−7 8 Cirsium discolor 23 Orobanche uniflora
9 Geranium maculatum 24 Rosa gallica

Phil. Trans. R. Soc. B 374: 20170394

2 (c) (d ) 10 Geranium robertianum 25 Rubus odoratus
phenological sensitivity

1 1 24 11 Hemerocallis fulva 26 Sarracenia purpurea

(days/100 m a.s.l.)

9 7 8 8
27
0 29
145 6
4 29
3016
28
4 12 Hibiscus moscheutos 27 Sisyrinchium mucronatum
2018 23 11 13 19
26 13 3 20 14
−1 3
27 22 21 12 9 22
21
6 12
5
13 Impatiens capensis 28 Solanum rostratum
15 10 2615
2 17 25 2 23 17 1 24
−2 30 28 25
Downloaded from https://royalsocietypublishing.org/ on 08 November 2024

11 18 7 14 Iris pseudacorus 29 Trillium grandiflorum

19
−3 16 15 Iris versicolor 30 Trillium undulatum
−4
10
−5 community means
−6 50% CI

80% CI
100 150 200 250 100 150 200 250
phase mean at 7.4°C and 216 m a.s.l. (DOY)

Figure 2. Mean flowering and fruiting time compared with estimated phenological sensitivities to spring (March – May) temperatures (a,b), and collection elevation
(c,d ) of 30 species estimated from herbarium specimens using a Bayesian hierarchical model (Model 1). Coloured and enumerated dots indicate species-specific
estimates, and white circles at panel margins represent estimated community means for each quantity. Thick and thin bars represent 50 and 80% posterior credible
intervals, respectively. Thick black lines represent credible linear relationships between quantities on x- and y-axes ( posterior slope estimate different from zero
with greater than 90% probability). Dotted lines represent non-credible relationships ( posterior slope estimate not different from zero with greater than
90% probability).

For both flowering and fruiting, species with earlier (figure 3b). All differences between cold and mixed-warm
reproductive phenologies were substantially more sensitive climatic regions had a posterior probability greater than
to spring temperature than species that flowered and fruited 0.95 except for mean differences in flowering, where differ-
later in the season. This sensitivity manifested in a strong ences had a posterior probability of 0.87. These qualitative
positive correlation between mean flowering and fruiting patterns were robust to an alternative spatial binning strategy
date and spring temperature sensitivity, with a slope of that used only latitude, and not climate, to differentiate more
0.018 days/8C per day for flowering and 0.023 days/8C per northerly and southerly regions (electronic supplementary
day for fruiting (figure 2a,b). These slopes were different material, figure S3).
from zero with greater than 99% posterior probability. We Overall differences between cool, northern and warm,
also found that, all other conditions being equal, flowering southerly parts of the study area were accompanied by
and fruiting came earlier at higher elevations (community large amounts of regional variation not explained by climate
means 21.71 to 20.11 days earlier per 100 m greater or latitude (figure 4). For example, using Model 1 we esti-
elevation for flowering, and 22.07 to 0.14 days earlier for mated that plants in the Ozark Complex and mid-Atlantic
fruiting, figure 2c,d). These effects were credibly different NEON domains were substantially more phenologically
from zero for 7 of 30 species for flowering and 8 of 30 species sensitive than those in the northeast and Appalachians
for fruiting, but elevation influenced early-flowering and (figure 4a), while the northeast and Ozark Complex had
late-flowering species approximately equally. plant assemblages with greater variability in phenological
Species in the warm and mixed-temperate climatic sensitivity (figure 4c). Differences in phenological sensitivity
regions showed greater mean sensitivities to spring tempera- between regions were not adequately explained by differ-
ture and also greater variability (standard deviation) in ences in species composition, as per-sample weighted
climate sensitivity between species than in the cool-temperate means computed using only species effects (b1[s] in equation
northeast and Appalachians (figure 3; electronic supplemen- (2.2)), did not show strong regional differences (figure 4b,d).
tary material, figures S1 and S2). Using Model 2, we Our hierarchical approach allowed us to compare within-
estimated that mean sensitivities in the mixed-warm region species, between-species and between-region sources of
(figure 1b) were 22.96 days/8C (95% CI 3.69 to 22.25) for variability for both mean flowering time and sensitivity to
flowering and 23.37 days/8C (95% CI 24.12 to 22.60) for spring temperature (figure 5). This analysis shows that
fruiting, but were substantially closer to zero in the cool- between-species variation dominated variability in mean
temperate region (22.51 days/8C, 95% CI 22.86 to 22.19 flowering time (figure 5a), but there was a similar amount
for flowering, and 22.09 days/8C, 95% CI 22.63 to 21.57 of variation in phenological sensitivity within species
for fruiting, figure 3a). The mixed-warm climatic region also between regions to that seen between species for both
had greater assemblage variability in phenological sensitivity flowering and fruiting (figure 5b).
 flowering fruiting growing seasons, which exert pressure for species to initiate 7
growth as soon as conditions become favourable [44,76 –78].

rstb.royalsocietypublishing.org
(a) 0 Despite these general trends, we observed significant
variation among species in their responses to warming. In
** general, early-flowering and early-fruiting species were
phenological sensitivity

assemblage mean
−1 more sensitive to spring temperatures than late-flowering/
fruiting species (figure 2), a pattern also observed at smaller
(days/°C)

scales [75,79]. Warming-induced leaf budburst advancement

−2
has been suggested to be less prominent in late-flushing
species compared with early-flushing ones owing to their
−3 greater chilling requirements [80]. Similar mechanisms may
affect flowering and fruiting, where advances in the flowering

Phil. Trans. R. Soc. B 374: 20170394

date of late-flowering species caused by spring warming
−4 would be smaller than those of early-flowering species,
(b) which would manifest as weaker phenological responses to
3 ** temperature in late-flowering species. Further, as flowering
phenological sensitivity

and fruiting events later in the year are more separated from
Downloaded from https://royalsocietypublishing.org/ on 08 November 2024

spring climatic conditions, there is an extended window of

assemblage s.d.
time in which other factors could affect or modify reproduc-
(days/°C)

2
tive timing. For instance, late-flowering species may be more
sensitive to photoperiod or precipitation.
A large amount of variability in phenological sensitivity
1 across species suggests that phenological responses to cli-
*
matic change will be heterogeneous within communities.
This could cause temporal reorganization of the structure
0 and composition of plant communities, potentially altering
cool warm-mixed direct and indirect interactions among plant species and
climate region between plant and animal species, and ecosystem services
[24,34,81–83].
Figure 3. Estimated community phenological sensitivities to spring (March –
May) air temperature in cold-temperate versus mixed-warm temperate climate
zones (depicted in figure 1) using a Bayesian hierarchical model (Model 2). (b) Phenological sensitivity to spring temperature tends
Panel (a) depicts species assemblage means, and panel (b) depicts assemblage
standard deviations. Black and grey represent flowering and fruiting stages, and
to decrease with latitude
The consequences of phenological shifts can be further com-
thick and thin bars represent 50 and 80% posterior credible intervals, respect-
plicated by intraspecific variation in phenological sensitivity
ively. Comparisons with a posterior probability greater than or equal to 0.8 and
to environmental cues [33,38]. For instance, using 20 years
less than 0.95 are depicted with a ‘ ’, comparisons with probabilities greater
B
of observational data, Prevéy et al. [44] found that the pheno-
than or equal to 0.95 and less than 0.99 are depicted with ‘*’, and comparisons
logical sensitivity to temperature of tundra plants at colder,
with probabilities greater than or equal to 0.99 are depicted with ‘**’.
higher latitudes was greater than at warmer, lower latitudes.
However, contrary to such studies, we found that plants in
4. Discussion warmer, lower-latitude regions tended to be more pheno-
Our analyses revealed that (i) plant species from the eastern logically sensitive to temperature, especially for fruiting
United States exhibit advanced timing of flowering and fruit- (figure 3). We hypothesize that this is due to the lower and
ing in response to warmer spring temperatures, (ii) the less predictable winter and spring climates of the north-
magnitude of these responses varies significantly between eastern United States. In such environments, dynamic
and within species across their latitudinal ranges and (iii) phenological tracking of spring temperatures (i.e. high phe-
that phenological sensitivity to temperature tends to be nological sensitivity to temperature) presents high risks to
higher in the warmer, more stable climates of lower-latitude reproductive success, because warm periods may often be fol-
regions. lowed by periods of dramatic chilling [40]. At lower latitudes,
the advent and progression of spring is less variable and
average temperatures are higher; thus phenological tracking
of temperature is less risky (electronic supplementary
(a) Differential responses to spring warming material, figures S1 and S2). Indeed, species exhibited a
across species larger amount of variability in their responses to temperature
Consistent with previous field observations of community in the warmer, lower-latitudinal parts of their ranges.
phenology, we found that reproductive phenology of flower- Climate and phenology might play different roles in filter-
ing plants accelerated with warming spring temperatures ing species assemblages in regions with longer growing
(e.g. [46,72,73]; but see [63]). The average number of days of seasons than in regions where the growing season is short
phenological advancement per degree increase in temperature and reproductive phenologies are strongly constrained by
(22.56 days/8C) that we observed also fell within previous shorter freeze-free periods [35]. Indeed, studies synthesizing
estimates [46,74,75]. All else being equal, flowering and fruit- plot-level observational data have suggested phenological
ing tended to occur earlier at higher elevations. Higher sensitivity of plant communities to warming may be
elevations tend to be relatively colder and have shorter positively correlated with mean annual temperature, but
 turnover + species 8
local adaptation/plasticity turnover only

rstb.royalsocietypublishing.org
phenological sensitivity (days/°C)
(a) (c)
0

assemblage mean
2.0

–4.0

Phil. Trans. R. Soc. B 374: 20170394

phase
flowering
phenological sensitivity (days/°C)

(b) (d ) fruiting
2.5
Downloaded from https://royalsocietypublishing.org/ on 08 November 2024

assemblage s.d.
2.0

1.5

1.0

SE OC AP MA NE SE OC AP MA NE
NEON domain NEON domain

Figure 4. Differences in sensitivity to spring temperatures between NEON domains are driven by regional variation due to local adaptation or plasticity, not com-
munity turnover between regions. This applies to both differences in species assemblage mean phenological sensitivities (a,b) and assemblage standard deviations
(c,d ). Panels (a) and (b) represent best estimates of regional variations in phenological sensitivity incorporating species identity, NEON Domain and NEON Domain 
species identity as random effects (b1[s], b2[r], b3[sr]) and panels (c) and (d ) represent estimates incorporating only species-level effects (b1[s]). This means that in
panels (c) and (d ), phenological sensitivity is assumed to be constant within a species across domains. Estimates for flowering are represented in black and
estimates for fruiting are represented in grey. Thick and thin bars represent 50 and 80% posterior credible intervals, respectively, from a hierarchical Bayesian
model (Model 1).

negatively correlated with seasonal temperature range (i.e. phenological sensitivity that was not clearly linked to climate
variability) in Europe [84] and China [85]. Alternatively, it or latitude. These regional differences were not explained
is possible that plants in warmer climates exist closer to by species turnover, but rather suggest the presence of
their response thresholds in terms of phenology, and thus inter-population variation driven by local adaptation or
react more dramatically to small changes in temperature. phenological plasticity (figure 4). Further, within-species
However, Körner & Basler [86] noted that cherry cultivars variation in phenological sensitivity between regions was of
from regions with less variable spring temperatures flowered similar magnitude to differences in sensitivity between species
earlier in common gardens, suggesting phenological sensi- (figure 5). Other studies examining leaf out and senescence in
tivity does vary with climate. Plants in regions with high trees also have shown that individuals from geographically
spring temperature variability also tend to be less pheno- and climatically separated populations differ in their
logically sensitive in terms of leaf out and bud break to phenology even when grown in common gardens [40,87,88].
temperature than those in less variable climates [40]. Our Because the eastern United States is experiencing geo-
results demonstrating that phenological sensitivity to temp- graphically variable climatic change, the heterogeneity in
erature is higher in areas with low standard deviation of phenological responses to warming that we observed within
intra-annual temperature and inter-annual variation in and among species may have important consequences for
spring temperature and high mean annual temperature plant communities in the near future. Colder, climatically
support these findings. variable high-latitude regions are experiencing dispropor-
tionate warming and climatic homogenization (i.e. reduced
standard variation of intra-annual temperature), while
(c) Consequences of variation in phenological responses warmer, climatically less variable more southerly regions are
across species ranges experiencing increases in intra-annual temperature variability
Our results imply that with equal warming, individuals in (electronic supplementary material, figure S1). These climatic
lower-latitude populations will advance their reproductive changes could alter patterns of overlap in reproductive timing
phenology more dramatically than those at higher latitudes. among species in a community and across their individual
This observed variation in phenological response may reflect ranges. Changes in phenological overlap across ranges could
adaptation to local climatic conditions, especially in annual have direct consequences for adaptive evolution and species
species. We found a large amount of regional variation in resilience to current and impending climatic changes, as
 flowering fruiting biases; some regional differences could be due to differing 9
(a) patterns of collection across space and time. Including

rstb.royalsocietypublishing.org
40 county-level random effects as we have done here minimizes
the impacts of these biases but does not eliminate them
altogether. It is possible that different patterns of pheno-

intercept (DOY)
30
variance (s)

logical sensitivity may be observed across species ranges

depending on how climatic and/or geographical regions
20 are delimited, though testing an alternative threshold yielded
similar results, suggesting that the patterns we observe are
10 robust to spatial binning choices. Additionally, we addressed
crowdsourcing bias by including crowdsourcer random
effects and removing observations for crowdsourcers with

Phil. Trans. R. Soc. B 374: 20170394

0
low reliability scores even though phenological data

phenological sensitivity (days/°C)

(b) collected by citizen scientists do not differ significantly in
quality from those collected by experts [46,55]. Lastly,
3 although spring temperature is a critical driver of flowering
variance (s)

phenology in temperate climates, we cannot fully exclude

Downloaded from https://royalsocietypublishing.org/ on 08 November 2024

the possibility that other variables correlated with latitude

2
or mean spring temperature may determine observed
variance in phenological sensitivity [46,73,75,94– 97]. For
1 example, spring temperature tends to be highly correlated
with mean annual and mean monthly temperatures in east-
ern North America (electronic supplementary material,
0
between between within figure S4). Photoperiod or snow melt may further influence
domain species species and alter species phenological responses [98 –103]. Future
(sb2) (sb1) (sb3) research into how these environmental cues interact to trigger
phenological events is necessary and will greatly improve our
variance component
understanding of plant phenology.
Figure 5. Variance components from Model 1, allowing comparisons of varia-
bility between NEON domains, between species and within species. Variance in
the intercept (a) represents variability in flowering (black) or fruiting times (grey)
at 7.48C spring temperature and 216 m elevation above sea level, median con- 5. Conclusion
ditions for the specimens. Variance in phenological sensitivity to spring air Building on previous phenological research by scoring mul-
temperature (days/8C) represents variability in the slope of the linear relationship tiple phenological traits across over 8000 herbarium
between spring temperature and flowering or fruiting times. specimens spanning 120 years, we have demonstrated that
phenological sensitivity can vary greatly across species’
ranges. This variance may be attributed to adaptation or
gene flow may increase among previously (temporally) iso- acclimation to local climates. The large amount of within-
lated populations of some species and decrease among species variation in phenological sensitivity that we observed
others [41–43,89–91]. Decreased phenological overlap could underlines the complex and contingent nature of phenological
genetically isolate fringe populations, potentially leading to sensitivities. Phenological responses of individual species to
local extirpation and species range contractions. Moreover, climate are not stable phenotypic traits, but instead emerge
phenologically sensitive plants at lower latitudes may from a multitude of potentially reciprocal interactions between
especially be at risk due to increasingly variable temperatures organisms and their environment. Populations in different
and increased probabilities of phenological mismatch with regions could have differences in frequencies of genes that
mutualists [92]. However, we cannot fully rule out the control how climate affects the timing of development or
possibility that individuals of long-lived perennial species differences in microhabitat distributions between regions that
may also be able to acclimate their phenological sensitivity alter how populations experience local climate. The regions
to changing climatic conditions over longer periods of time. themselves may have differences in unmeasured environ-
Some of the variation we observe in phenological sensi- mental factors that interact with responses to temperature or
tivity to temperature across and within species may be due differences in species interactions that may alter phenological
to differences in microclimate. However, the lack of accurate signalling. The circumstances and extent to which these or
location data for most historic specimens limited our ability other factors explain regional variation in species responses
to infer fine-scale climate, necessitating coarser, county-level to climate is currently unknown. To untangle the roles of
analyses. Also, we cannot ignore the possibility that the phe- ecological and evolutionary processes governing the hetero-
nological trends we observed are unique to the species that geneous phenological responses of plant species to warming,
we studied and/or reflect biases in herbarium collections researchers will have to take advantage of new techniques
[93]. To minimize effects of spatial bias and uncertainty, we and datasets. In addition to continued field observations and
studied common, well-collected species and accounted for laboratory analysis of mechanisms responsible for flowering
climatic heterogeneity present in each sampling locale in and fruiting, herbarium specimens can provide a comprehen-
our models. However, these taxa are not necessarily represen- sive, nuanced picture of phenological responses to ongoing
tative of species assemblages across regions, and our analyses climatic change across many species. Our study further
do not account explicitly for spatial and temporal sampling demonstrates that we can now harness the treasure trove of
 information in herbaria across the world to examine hundreds, A.M.E. and I.B. analysed the data; D.S.P. drafted the first version of 10
if not thousands of species across myriad plant lineages, habi- the manuscript; C.C.D. and D.S.P. made first substantial revisions
to this draft, and all authors contributed to subsequent revisions.

rstb.royalsocietypublishing.org
tats and regions. Such efforts will be critical to enhance our
Competing interests. We declare we have no competing interests.
ability to forecast future changes in plant assemblages across
Funding. This study was funded as part of the New England Vascular
space and time in an era of accelerating climate change
Plant Project to C.C.D. (NSF-DBI: EF1208835), NSF-DEB 1754584 to
[104,105]. C.C.D. and A.M.E., and an NSERC Discovery Grant to E.L.
(RGPIN-2015-04543). A.M.E.’s participation in this project was sup-
Ethics. The use and collection of data by citizen scientists were
ported by Harvard Forest. D.S.P.’s contribution was supported by
approved by an ethics review committee at the University of
the Harvard University Herbaria and NSF-DEB 1754584. I.B.’s contri-
Waterloo (ORE no. 21647).
bution was supported by the National Science Foundation’s
Data accessibility. Data and R code are available from the Harvard Forest Postdoctoral Research Fellowship in Biology (NSF-DBI-1711936).
Data Archive (http://harvardforest.fas.harvard.edu/dataarchive,
Acknowledgements. The authors thank X. Feng, S. Worthington, and
dataset HF309).

Phil. Trans. R. Soc. B 374: 20170394

members of the Davis lab for their invaluable insights and comments
Authors’ contributions. C.C.D. conceived the study; D.S.P., C.C.D., A.C.W. on the project and manuscript.
and E.L. designed the study; D.S.P. and A.C.W. collected data; D.S.P.,

References
Downloaded from https://royalsocietypublishing.org/ on 08 November 2024

1. Field CB et al. 2014 Climate change 2014: impacts, with abiotic conditions—implications for flowering Mismatch between birth date and vegetation
adaptation, and vulnerability. Part A: global and in Arctic plants. Ecology 96, 775– 787. (doi:10. phenology slows the demography of roe deer.
sectoral aspects. In Contribution of Working Group II 1890/14-0338.1) PLoS Biol. 12, e1001828. (doi:10.1371/journal.pbio.
to the Fifth Assessment Report of the 11. White MA, Nemani RR. 2003 Canopy duration has 1001828)
Intergovernmental Panel on Climate Change. Geneva, little influence on annual carbon storage in the 21. Kharouba HM, Ehrlén J, Gelman A, Bolmgren K,
Switzerland: IPCC. deciduous broad leaf forest. Glob. Change Biol. 9, Allen JM, Travers SE, Wolkovich EM. 2018 Global
2. Root TL, Price JT, Hall KR, Schneider SH, Rosenzweig 967 –972. (doi:10.1046/j.1365-2486.2003.00585.x) shifts in the phenological synchrony of species
C, Pounds JA. 2003 Fingerprints of global warming 12. Richardson AD, Hollinger DY, Dail DB, Lee JT, interactions over recent decades. Proc. Natl Acad.
on wild animals and plants. Nature 421, 57 –60. Munger JW, O’keefe J. 2009 Influence of spring Sci. USA 115, 5211–5216. (doi:10.1073/pnas.
(doi:10.1038/nature01333) phenology on seasonal and annual carbon balance 1714511115)
3. Parmesan C, Yohe G. 2003 A globally coherent in two contrasting New England forests. Tree 22. Navarro-Cano JA, Karlsson B, Posledovich D,
fingerprint of climate change impacts across natural Physiol. 29, 321 –331. (doi:10.1093/treephys/ Toftegaard T, Wiklund C, Ehrlén J, Gotthard K. 2015
systems. Nature 421, 37 –42. (doi:10.1038/ tpn040) Climate change, phenology, and butterfly host plant
nature01286) 13. Barford CC et al. 2001 Factors controlling long- and utilization. Ambio 44, 78 –88. (doi:10.1007/s13280-
4. Miller-Rushing AJ, Primack RB, Primack D, Mukunda short-term sequestration of atmospheric CO2 in a 014-0602-z)
S. 2006 Photographs and herbarium specimens as mid-latitude forest. Science 294, 1688–1691. 23. Posledovich D, Toftegaard T, Wiklund C, Ehrlén J,
tools to document phenological changes in (doi:10.1126/science.1062962) Gotthard K. 2018 Phenological synchrony
response to global warming. Am. J. Bot. 93, 14. Fitzjarrald DR, Acevedo OC, Moore KE. 2001 Climatic between a butterfly and its host plants:
1667–1674. (doi:10.3732/ajb.93.11.1667) consequences of leaf presence in the eastern United experimental test of effects of spring temperature.
5. Menzel A, Sparks TH, Estrella N, Roy DB. 2006 States. J. Clim. 14, 598–614. (doi:10.1175/1520- J. Anim. Ecol. 87, 150–161. (doi:10.1111/1365-
Altered geographic and temporal variability in 0442(2001)014,0598:CCOLPI.2.0.CO;2) 2656.12770)
phenology in response to climate change. Glob. 15. Inouye DW, Barr B, Armitage KB, Inouye BD. 2000 24. Burkle LA, Marlin JC, Knight TM. 2013 Plant-
Ecol. Biogeogr. 15, 498–504. (doi:10.1111/ Climate change is affecting altitudinal migrants and pollinator interactions over 120 years: loss of
j.1466-822X.2006.00247.x) hibernating species. Proc. Natl Acad. Sci. USA 97, species, co-occurrence, and function. Science 339,
6. Chambers LE et al. 2013 Phenological changes in 1630 –1633. (doi:10.1073/pnas.97.4.1630) 1611– 1615. (doi:10.1126/science.1232728)
the southern hemisphere. PLoS ONE 8, e75514. 16. Visser ME, Both C. 2005 Shifts in phenology due to 25. Kudo G, Ida TY. 2013 Early onset of spring increases
(doi:10.1371/journal.pone.0075514) global climate change: the need for a yardstick. the phenological mismatch between plants and
7. Ovaskainen O, Skorokhodova S, Yakovleva M, Proc. R. Soc. B 272, 2561–2569. (doi:10.1098/rspb. pollinators. Ecology 94, 2311–2320. (doi:10.1890/
Sukhov A, Kutenkov A, Kutenkova N, Shcherbakov A, 2005.3356) 12-2003.1)
Meyke E, del Delgado M. 2013 Community-level 17. Doiron M, Gauthier G, Lévesque E. 2015 Trophic 26. Iler AM, Inouye DW, Høye TT, Miller-Rushing AJ,
phenological response to climate change. Proc. Natl mismatch and its effects on the growth of young in Burkle LA, Johnston EB. 2013 Maintenance of
Acad. Sci. USA 110, 13 434 –13 439. (doi:10.1073/ an Arctic herbivore. Glob. Change Biol. 21, temporal synchrony between syrphid flies and floral
pnas.1305533110) 4364 –4376. (doi:10.1111/gcb.13057) resources despite differential phenological responses
8. Willis CG, Ruhfel B, Primack RB, Miller-Rushing AJ, 18. Post E, Forchhammer MC. 2008 Climate change to climate. Glob. Change Biol. 19, 2348–2359.
Davis CC. 2008 Phylogenetic patterns of species loss reduces reproductive success of an Arctic herbivore (doi:10.1111/gcb.12246)
in Thoreau’s woods are driven by climate change. through trophic mismatch. Phil. Trans. R. Soc. B 27. Visser ME, van Noordwijk AJ, Tinbergen JM, Lessells
Proc. Natl Acad. Sci. USA 105, 17 029 –17 033. 363, 2367–2373. (doi:10.1098/rstb.2007.2207) CM. 1998 Warmer springs lead to mistimed
(doi:10.1073/pnas.0806446105) 19. Post E, Pedersen C, Wilmers CC, Forchhammer MC. reproduction in great tits (Parus major). Proc. R. Soc.
9. Inouye D. 2008 Effects of climate change on 2008 Warming, plant phenology and the spatial Lond. B 265, 1867– 1870. (doi:10.1098/rspb.1998.
phenology, frost damage, and floral abundance of dimension of trophic mismatch for large herbivores. 0514)
montane wildflowers. Ecology 89, 353–362. Proc. R. Soc. B 275, 2005–2013. (doi:10.1098/rspb. 28. Mayor SJ et al. 2017 Increasing phenological
(doi:10.1890/06-2128.1) 2008.0463) asynchrony between spring green-up and arrival of
10. Wheeler HC, Høye TT, Schmidt NM, Svenning JC, 20. Plard F, Gaillard J-M, Coulson T, Hewison AJM, migratory birds. Sci. Rep. 7, 1902. (doi:10.1038/
Forchhammer MC. 2015 Phenological mismatch Delorme D, Warnant C, Bonenfant C. 2014 s41598-017-02045-z)
 29. Wolkovich EM, Cleland EE. 2011 The phenology of 42. Alleaume-Benharira M, Pen IR, Ronce O. 2006 55. Williams AC, Goh J, Willis CG, Ellison AM, Brusuelas 11
plant invasions: a community ecology perspective. Geographical patterns of adaptation within a JH, Davis CC, Law E. 2017 Déja vu: characterizing

rstb.royalsocietypublishing.org
Front. Ecol. Environ. 9, 287–294. (doi:10.1890/ species’ range: interactions between drift and gene worker reliability using task consistency. In Proc. 5th
100033) flow. J. Evol. Biol. 19, 203 –215. (doi:10.1111/ AAAI Conf. Human Computation and Crowdsourcing,
30. Craine JM, Wolkovich EM, Towne EG, Kembel SW. j.1420-9101.2005.00976.x) HCOMP 2017, 23–26 October 2017, Québec City,
2012 Flowering phenology as a functional trait in a 43. Sexton JP, Strauss SY, Rice KJ. 2011 Gene flow Québec, Canada (eds S Dow, AT Kalai),
tallgrass prairie. New Phytol. 193, 673–682. increases fitness at the warm edge of a species’ pp. 197–205. Palo Alto, CA: AAAI Press.
(doi:10.1111/j.1469-8137.2011.03953.x) range. Proc. Natl Acad. Sci. USA 108, 56. Schorn C, Weber E, Bernardos R, Hopkins C, Davis C.
31. Colautti RI, Ågren J, Anderson JT, Colautti RI, Jon A, 11 704– 11 709. (doi:10.1073/pnas.1100404108) 2016 The New England Vascular Plants Project:
Colautti RI. 2017 Phenological shifts of native and 44. Prevéy J et al. 2017 Greater temperature sensitivity 295 000 specimens and counting. Rhodora 118,
invasive species under climate change: insights from of plant phenology at colder sites: implications for 324. (doi:10.3119/15-34)
the Boechera –Lythrum model. Phil. Trans. R. Soc. B convergence across northern latitudes. Glob. Change 57. SouthEast Regional Network of Expertise and

Phil. Trans. R. Soc. B 374: 20170394

372, 20160032. (doi:10.1098/rstb.2016.0032) Biol. 23, 2660–2671. (doi:10.1111/gcb.13619) Collections. 2017 SERNEC Data Portal. See http://
32. Pau S, Wolkovich EM, Cook BI, Davies TJ, Kraft NJB, 45. Riihimäki M, Savolainen O. 2004 Environmental and sernecportal.org/portal/index.php (accessed 28 June
Bolmgren K, Betancourt JL, Cleland EE. 2011 genetic effects on flowering differences between 2018).
Predicting phenology by integrating ecology, northern and southern populations of Arabidopsis 58. Hart R, Salick J, Ranjitkar S, Xu J. 2014 Herbarium
evolution and climate science. Glob. Change Biol. lyrata (Brassicaceae). Am. J. Bot. 91, 1036–1045. specimens show contrasting phenological responses
Downloaded from https://royalsocietypublishing.org/ on 08 November 2024

17, 3633 –3643. (doi:10.1111/j.1365-2486.2011. (doi:10.3732/ajb.91.7.1036) to Himalayan climate. Proc. Natl Acad. Sci. USA 111,
02515.x) 46. Willis CG et al. 2017 CrowdCurio: an online 10 615 –10 619. (doi:10.1073/pnas.1403376111)
33. Wolkovich EM, Cook BI, Davies TJ. 2014 Progress crowdsourcing platform to facilitate climate change 59. Park DS, Davis CC. 2017 Implications and
towards an interdisciplinary science of plant studies using herbarium specimens. New Phytol. alternatives of assigning climate data to
phenology: building predictions across space, time 215, 479 –488. (doi:10.1111/nph.14535) geographical centroids. J. Biogeogr. 44,
and species diversity. New Phytol. 201, 1156 –1162. 47. Willis CG et al. 2017 Old plants, new tricks: 2188– 2198. (doi:10.1111/jbi.13029)
(doi:10.1111/nph.12599) phenological research using herbarium specimens. 60. PeñA JE, Duncan R, Browning H. 1996 Seasonal
34. CaraDonna PJ, Iler AM, Inouye DW. 2014 Shifts in Trends Ecol. Evol. 32, 531–546. (doi:10.1016/j.tree. abundance of Phyllocnistis citrella (Lepidoptera:
flowering phenology reshape a subalpine plant 2017.03.015) Gracillariidae) and its parasitoids South Florida
community. Proc. Natl Acad. Sci. USA 111, 48. Davis CC, Willis CG, Connolly B, Kelly C, Ellison AM. citrus. Environ. Entomol. 25, 698–702. (doi:10.
4916–4921. (doi:10.1073/pnas.1323073111) 2015 Herbarium records are reliable sources of 1093/ee/25.3.698)
35. Diez JM, Ibáñez I, Miller-Rushing AJ, Mazer SJ, phenological change driven by climate and provide 61. Baechler MC, Gilbride TL, Cole PC, Hefty MG, Ruiz K.
Crimmins TM, Crimmins MA, Bertelsen CD, Inouye novel insights into species’ phenological cueing 2010 Guide to determining climate regions by
DW. 2012 Forecasting phenology: from species mechanisms. Am. J. Bot. 102, 1599–1609. (doi:10. county. Build. Am. Best Pract. Ser. 7, 1–45.
variability to community patterns. Ecol. Lett. 15, 3732/ajb.1500237) (doi:10.2172/969181)
545–553. (doi:10.1111/j.1461-0248.2012.01765.x) 49. Primack D, Imbres C, Primack RB, Miller-Rushing AJ, 62. Zhang Y, Bielory L, Georgopoulos PG. 2014 Climate
36. Banta JA, Ehrenreich IM, Gerard S, Chou L, Wilczek Del Tredici P. 2004 Herbarium specimens change effect on Betula (birch) and Quercus (oak)
A, Schmitt J, Kover PX, Purugganan MD. 2012 demonstrate earlier flowering times in response to pollen seasons in the United States.
Climate envelope modelling reveals intraspecific warming in Boston. Am. J. Bot. 91, 1260–1264. Int. J. Biometeorol. 58, 909–919. (doi:10.1007/
relationships among flowering phenology, niche (doi:10.3732/ajb.91.8.1260) s00484-013-0674-7)
breadth and potential range size in Arabidopsis 50. Robbirt KM, Davy AJ, Hutchings MJ, Roberts DL. 63. Hobbs NT, Hooten MB. 2015 Bayesian models: a
thaliana. Ecol. Lett. 15, 769– 777. (doi:10.1111/ 2011 Validation of biological collections as a source statistical primer for ecologists. Princeton, NJ:
j.1461-0248.2012.01796.x) of phenological data for use in climate change Princeton University Press.
37. Wilczek AM et al. 2009 Effects of genetic studies: a case study with the orchid Ophrys 64. Gelman A, Hill J. 2006 Data analysis using regression
perturbation on seasonal life history plasticity. sphegodes. J. Ecol. 99, 235– 241. (doi:10.1111/ and multilevel/hierarchical models. Cambridge, UK:
Science 323, 930–934. (doi:10.1126/science. j.1365-2745.2010.01727.x) Cambridge University Press.
1165826) 51. Vellend M, Brown CD, Kharouba HM, McCune JL, 65. Hoffman MD, Gelman A. 2014 The No-U-turn
38. Matthews ER, Mazer SJ. 2016 Historical changes in Myers-Smith IH. 2013 Historical ecology: using sampler: adaptively setting path lengths in
flowering phenology are governed by unconventional data sources to test for effects of Hamiltonian Monte Carlo. J. Mach. Learn. Res. 15,
temperature  precipitation interactions in a global environmental change. Am. J. Bot. 100, 1593– 1623.
widespread perennial herb in western North 1294 –1305. (doi:10.3732/ajb.1200503) 66. Carpenter B et al. 2017 Stan: a probabilistic
America. New Phytol. 210, 157–167. (doi:10.1111/ 52. Sweeney PW, Starly B, Morris PJ, Xu Y, Jones A, programming language. J. Stat. Softw. 76, 1 –32.
nph.13751) Radhakrishnan S, Grassa CJ, Davis CC. 2018 (doi:10.18637/jss.v076.i01)
39. Barros VR et al. 2014 Climate change 2014: impacts, Large-scale digitization of herbarium specimens: 67. Stan Development Team. 2018 RStan: the R
adaptation, and vulnerability. Part B: regional development and usage of an automated, high- interface to Stan. R package version 2.18.1. See
aspects.. In Contribution of Working Group II to the throughput conveyor system. Taxon 67, 165–178. http://mc-stan.org/.
Fifth Assessment Report of the Intergovernmental (doi:10.12705/671.9) 68. R Core Team. 2018 R: a language and environment
Panel on Climate Change. Geneva, Switzerland: IPCC. 53. Yost JM et al. 2018 Digitization protocol for scoring for statistical computing. Vienna, Austria: R
40. Zohner CM, Renner SS. 2014 Common garden reproductive phenology from herbarium specimens Foundation for Statistical Computing. See http://
comparison of the leaf-out phenology of woody of seed plants. Appl. Plant Sci. 6, e1022. (doi:10. www.R-project.org.
species from different native climates, combined 1002/aps3.1022) 69. Betancourt M, Girolami M. 2015 Hamiltonian Monte
with herbarium records, forecasts long-term change. 54. Pearse WD, Davis CC, Inouye DW, Primack RB, Carlo for hierarchical models. Curr. Trends Bayesian
Ecol. Lett. 17, 1016 –1025. (doi:10.1111/ele.12308) Davies TJ. 2017 A statistical estimator for Methodol. Appl. 79, 30. (doi:10.1201/b18502-5)
41. Lenormand T. 2002 Gene flow and the limits to determining the limits of contemporary and historic 70. Gabry J, Simpson D, Vehtari A, Betancourt M,
natural selection. Trends Ecol. Evol. 17, 183 –189. phenology. Nat. Ecol. Evol. 1, 1876. (doi:10.1038/ Gelman A. 2017 Visualization in Bayesian workflow.
(doi:10.1016/S0169-5347(02)02497-7) s41559-017-0350-0) See http://arxiv.org/abs/1709.01449.
 71. Park DS, Breckheimer I, Williams AC, Law E, Ellison closely related wildflower species. Am. J. Bot. 96, forest canopy. Glob. Change Biol. 12, 1174–1188. 12
AM, Davis CC. 2018 Assessing plant phenological 1821 –1829. (doi:10.3732/ajb.0800411) (doi:10.1111/j.1365-2486.2006.01164.x)

rstb.royalsocietypublishing.org
patterns in the eastern United States over the last 83. Memmott J, Craze PG, Waser NM, Price MV. 95. Ellwood ER, Temple SA, Primack RB, Bradley NL,
120 years. Harvard Forest Data Archive, HF309. 2007 Global warming and the disruption of Davis CC. 2013 Record-breaking early flowering in
(doi:10.6073/pasta/ plant – pollinator interactions. Ecol. Lett. the eastern United States. PLoS ONE 8, 1–9.
f6afa728bb5edfd79f458d7d5e23f559) 10, 710– 717. (doi:10.1111/j.1461-0248. (doi:10.1371/journal.pone.0053788)
72. Roberts AMI, Tansey C, Smithers RJ, Phillimore AB. 2007.01061.x) 96. Augspurger CK. 2013 Reconstructing patterns of
2015 Predicting a change in the order of spring 84. Lapenis A, Henry H, Vuille M, Mower J. 2014 temperature, phenology, and frost damage over 124
phenology in temperate forests. Glob. Change Biol. Climatic factors controlling plant sensitivity to years: spring damage risk is increasing. Ecology 94,
21, 2603 –2611. (doi:10.1111/gcb.12896) warming. Clim. Change 122, 723–734. (doi:10. 41– 50. (doi:10.1890/12-0200.1)
73. Yu H, Luedeling E, Xu J. 2010 Winter and 1007/s10584-013-1010-2) 97. Güsewell S, Furrer R, Gehrig R, Pietragalla B. 2017
spring warming result in delayed spring phenology 85. Zhang H, Yuan W, Liu S, Dong W, Fu Y. 2015 Changes in temperature sensitivity of spring phenology

Phil. Trans. R. Soc. B 374: 20170394

on the Tibetan Plateau. Proc. Natl Acad. Sci. USA Sensitivity of flowering phenology to changing with recent climate warming in Switzerland are related
107, 22 151–22 156. (doi:10.1073/pnas. temperature in China. J. Geophys. Res. Biogeosci. to shifts of the preseason. Glob. Change Biol. 23,
1012490107) 120, 1658–1665. (doi:10.1002/2015JG003112) 5189–5202. (doi:10.1111/gcb.13781)
74. Wolfe DW, Schwartz MD, Lakso AN, Otsuki Y, 86. Körner C, Basler D. 2010 Phenology under global 98. Bjorkman AD, Elmendorf SC, Beamish AL, Vellend
Pool RM, Shaulis NJ. 2005 Climate change and warming. Science 327, 1461–1462. (doi:10.1126/ M, Henry GHR. 2015 Contrasting effects of warming
Downloaded from https://royalsocietypublishing.org/ on 08 November 2024

shifts in spring phenology of three horticultural science.1186473) and increased snowfall on Arctic tundra plant
woody perennials in northeastern USA. 87. Zohner CM, Benito BM, Fridley JD, Svenning J, phenology over the past two decades. Glob. Change
Int. J. Biometeorol. 49, 303–309. (doi:10.1007/ Renner SS. 2017 Spring predictability explains Biol. 21, 4651– 4661. (doi:10.1111/gcb.13051)
s00484-004-0248-9) different leaf-out strategies in the woody floras of 99. Way DA, Montgomery RA. 2015 Photoperiod
75. Miller-Rushing AJ, Primack RB. 2008 Global North America, Europe and East Asia. Ecol. Lett. 20, constraints on tree phenology, performance and
warming and flowering times in Thoreau’s Concord: 452 –460. (doi:10.1111/ele.12746) migration in a warming world. Plant. Cell Environ.
a community perspective. Ecology 89, 332–341. 88. Zohner CM, Renner SS. 2017 Innately shorter 38, 1725–1736. (doi:10.1111/pce.12431)
(doi:10.1890/07-0068.1) vegetation periods in North American species 100. Nakagawa H, Yamagishi J, Miyamoto N, Motoyama
76. Bliss LC. 1962 Adaptations of arctic and alpine explain native–non-native phenological M, Yano M, Nemoto K. 2005 Flowering response of
plants to environmental conditions. Arctic 15, asymmetries. Nat. Ecol. Evol. 1, 1655. (doi:10.1038/ rice to photoperiod and temperature: a QTL analysis
117–144. (doi:10.14430/arctic3564) s41559-017-0307-3) using a phenological model. Theor. Appl. Genet.
77. Billings WD, Mooney HA. 1968 The ecology of arctic 89. Savolainen O, Pyhäjärvi T, Knürr T. 2007 Gene flow 110, 778–786. (doi:10.1007/s00122-004-1905-4)
and alpine plants. Biol. Rev. 43, 481–529. (doi:10. and local adaptation in trees. Annu. Rev. Ecol. Evol. 101. Dunne JA, Harte J, Taylor KJ. 2003 Subalpine
1111/j.1469-185X.1968.tb00968.x) Syst. 38, 595–619. (doi:10.1146/annurev.ecolsys. meadow flowering phenology responses to climate
78. Oberbauer SF et al. 2013 Phenological response of 38.091206.095646) change: integrating experimental and gradient
tundra plants to background climate variation 90. Nagy ES, Rice KJ. 1997 Local adaptation in two methods. Ecol. Monogr. 73, 69 –86. (doi:10.1890/
tested using the International Tundra Experiment. subspecies of an annual plant: implications for 0012-9615(2003)073[0069:SMFPRT]2.0.CO;2)
Phil. Trans. R. Soc. B 368, 20120481. (doi:10.1098/ migration and gene flow. Evolution 51, 102. Wolkovich EM, Davies TJ, Schaefer H, Cleland EE,
rstb.2012.0481) 1079 –1089. (doi:10.1111/j.1558-5646.1997. Cook BI, Travers SE, Willis CG, Davis CC. 2013
79. Fitter AH, Fitter RSR. 2002 Rapid changes in tb03955.x) Temperature-dependent shifts in phenology
flowering time in British plants. Science 296, 91. Fox GA. 2003 Assortative mating and plant contribute to the success of exotic species with
1689–1691. (doi:10.1126/science.1071617) phenology: evolutionary and practical consequences. climate change. Am. J. Bot. 100, 1407–1421.
80. Murray MB, Cannell MGR, Smith RI. 1989 Date of Evol. Ecol. Res. 5, 1– 18. (doi:10.3732/ajb.1200478)
budburst of fifteen tree species in Britain following 92. Thomson JD. 2010 Flowering phenology, fruiting 103. Park IW, Mazer SJ. In press. Overlooked climate
climatic warming. J. Appl. Ecol. 26, 693–700. success and progressive deterioration of pollination parameters best predict flowering onset: assessing
(doi:10.2307/2404093) in an early-flowering geophyte. Phil. Trans. R. Soc. B phenological models using the elastic net. Glob.
81. Theobald EJ, Breckheimer I, HilleRisLambers J. 2017 365, 3187–3199. (doi:10.1098/rstb.2010.0115) Change Biol. (doi:10.1111/gcb.14447)
Climate drives phenological reassembly of a 93. Daru BH et al. 2018 Widespread sampling biases 104. Chuine I, Beaubien EG. 2001 Phenology is a major
mountain wildflower meadow community. Ecology in herbaria revealed from large-scale digitization. determinant of tree species range. Ecol. Lett. 4,
98, 2799 –2812. (doi:10.1002/ecy.1996) New Phytol. 217, 939 –955. (doi:10.1111/nph. 500–510. (doi:10.1046/j.1461-0248.2001.00261.x)
82. Miller-Rushing AJ, Inouye DW. 2009 Variation in the 14855) 105. Chuine I. 2010 Why does phenology drive species
impact of climate change on flowering phenology 94. Richardson AD, Bailey AS, Denny EG, Martin CW, distribution? Phil. Trans. R. Soc. B 365, 3149–3160.
and abundance: an examination of two pairs of O’Keefe J. 2006 Phenology of a northern hardwood (doi:10.1098/rstb.2010.0142)