Biodivers Conserv
DOI 10.1007/s10531-016-1097-7
ORIGINAL PAPER
Shifting targets: spatial priorities for ex situ plant
conservation depend on interactions between current
threats, climate change, and uncertainty
Adam B. Smith1 • Quinn G. Long1 • Matthew A. Albrecht1
Received: 2 November 2015 / Revised: 4 March 2016 / Accepted: 2 April 2016
Ó Springer Science+Business Media Dordrecht 2016
Abstract Few strategies for conservation seed banking consider current and climate
threats simultaneously and few—if any—represent uncertainty inherent in the assessment
process. Here we evaluate the vulnerability of 5148 populations of 71 rare plant species in
the North American Central Highlands to current threat, threat from climate change, and
their combination. We calculated priorities based on current threat using existing conservation status and protection, and priorities based on climate threat using ecological
niche models and species-level traits related to reproduction and dispersal. Current- and
climate-based priorities were integrated using a weighted average of rank priority. We
managed uncertainty using either a precautionary strategy that avoids any extinctions or a
resource-conservative strategy that directs attention to species known to be vulnerable with
high certainty. Priorities based on current threats highlighted presently rare species while
priorities based on climate threat emphasized presently common species. The location of
geographic ‘‘hotspots’’ providing opportunities for efficient seed banking depended
strongly on the weight of the climate module relative to the current module, the strategy
used to handle uncertainty, and emissions scenario. Integrating threats highlighted some
hotspots that were not identified using just current or climate threat, indicating the
importance of considering current and climate threats simultaneously. Only the Central
Basin of Tennessee, a known center of endemism, was consistently emphasized. We urge
(1) integrating current and climate threats when designing seed-banking strategies; and (2)
reporting of uncertainty in a manner that allows decision-makers to choose actions based
on available resources and tolerable risk.
Communicated by Danna J. Leaman.
Electronic supplementary material The online version of this article (doi:10.1007/s10531-016-1097-7)
contains supplementary material, which is available to authorized users.
& Adam B. Smith
[email protected]
1
Center for Conservation & Sustainable Development, Missouri Botanical Garden,
PO Box 299, Saint Louis, MO 63166-0299, USA
123
Biodivers Conserv
Keywords Climate change vulnerability Conservation seed banking Endemism North American Central Highlands Species distribution model Uncertainty
Introduction
Seed banking has become a widely advocated ex situ conservation strategy for plant
species threatened by habitat loss, fragmentation, and degradation, and future climate
change (Maunder et al. 2004; Vitt et al. 2010; Guerrant et al. 2014). As a cost-effective
measure for capturing genetic variation in wild populations, ex situ seed banks serve as a
safety-net in case of extinction in the wild and provide source material for in situ conservation efforts, including restoration, reintroduction, and population management
(Guerrant et al. 2004; Cochrane et al. 2007). Indeed, the Global Strategy for Plant Conservation calls for preservation of at least 75 % of all plant species in ex situ facilities
(Wyse Jackson and Kennedy 2009). Vulnerability assessments have yet to be widely
applied in the context of conservation seed banking despite their potential for maximizing
return on effort. Seed banking strategies that do employ vulnerability assessments generally focus either on current threat (e.g., Farnsworth et al. 2006; Gauthier et al. 2010;
Kricsfalusy and Trevisan 2014) or threat from anticipated climate change (e.g., Wyse
Jackson 2008; Godefroid and Vanderborght 2010; Still et al. 2015), but very few integrate
current levels of endangerment and vulnerability to climate change (Vitt et al. 2010;
Havens et al. 2015).
The most straightforward method for prioritizing seed-banking effort is to evaluate taxa
based on current levels of threat and rarity (e.g., Farnsworth et al. 2006; Gauthier et al.
2010; Kricsfalusy and Trevisan 2014). In general we should expect that species that are
rare and endangered now will fare worse under climate change since their populations are
already in peril. Indeed, recent work suggests that traits contributing to current vulnerability like small population size, small range, and current population trend correlate with
vulnerability under climate change (Keith et al. 2014; Pearson et al. 2014; Stanton et al.
2015). If these results are general, then priorities established using current conservation
status should be robust even as impacts of climate change become more severe in the
future. However, it is possible that the impact of future climate change on species’ welfare
will be independent of current conservation status. If so, then priorities based on threat
from climate change (e.g., Wyse Jackson 2008; Godefroid and Vanderborght 2010;
Anacker et al. 2013) could be very different from those established using current threats.
Nonetheless, many current threats will likely persist largely unabated so must be considered when designing strategies that incorporate threats from anticipated climate change.
Hence, it is important to understand the similarities between priorities established using
current threats (including contemporary climate change) and anticipated climate change.
Uncertainty about species’ current conservation status and their responses to climate
change complicates efforts to characterize vulnerability (Anacker et al. 2013; Wright et al.
2015). In particular, factors expected to affect sensitivity (e.g., dependence on other species
for pollination) and adaptive capacity (e.g., ability to disperse) to climate change are less
well known for rare species. Another source of uncertainty arises from using species’
current distributions to estimate their climatic tolerances. If species’ ranges are restricted
by non-climatic factors like limited dispersal or the distribution of patchy edaphic habitats
on which they specialize, then the range of climatic conditions associated with their current
123
Biodivers Conserv
presences may be less than the range of the environments the species can actually withstand (Van der Veken et al. 2008). As a result, methods that rely on species’ current
distributions to measure climatic niche breadth may overestimate exposure to climate
change. Together these sources of uncertainty can propagate through a vulnerability
assessment to make the outcome itself uncertain. Few—if any—seed banking strategies
account for uncertainty in the assessment process.
Here assess the vulnerability of 5148 populations of 71 rare plant species to current
threats, threat from future climate change, and a combination of current and climate threat.
The system ranks species and populations within species. The assessment is comprised of
two primary modules, one for current threat and another for climate threat. Priorities can be
integrated across the two modules to explore their combination. Our primary objective was
to identify high-priority ‘‘hotspots’’ that could be targeted for seed banking in an efficient
manner. We asked three questions: (1) Where are hotspots for efficient seed banking of
highly vulnerable populations? (2) How does the location of hotspots change when priorities are established using just current threat, just climate threat, or both in combination?
and (3) Are hotspot locations robust to uncertainty?
Methods
Species and study region
Our analysis focuses on 71 rare taxa that are the subject of a long-term seed-banking
program at the Missouri Botanical Garden (Table A1). The core geographic ranges of these
species lie within the North American Central Highlands (hereafter ‘‘Highlands’’), which is
composed of two largely disjunct subregions (Fig. A1). The western subregion comprises
the Ozark and Ouachita Mountains, and the eastern subregion comprises the Tennessee
Central Basin and surrounding areas up to the western escarpment of the Southern
Appalachian Mountains (Mayden 1987). Floristic diversity and plant endemism in both
subregions is concentrated in patchy, isolated habitats including fens, remnant grasslands,
and especially glades (Baskin and Baskin 2003; Estill and Cruzan 2001; Noss 2013), which
are embedded in a forest/woodland matrix. Glades are rocky openings with thin soil
dominated by short-statured forbs and grasses (Ware 2002; Baskin and Baskin 2003).
Notably, 46 % (33 species) of the species in our study occur on glades while 35 % (25
species) are considered endemic or near-endemic to glades (Ware 2002; Baskin and Baskin
2003).
General approach
Our assessment system combines the predictive capacity of ecological niche models for
estimating exposure to climate change with the power of trait-based assessments of species’ sensitivity and adaptive capacity to adverse climate change (Pacifici et al. 2015;
Willis et al. 2015). Priorities are established using two primary modules, one reflecting
current conservation status and protection and one reflecting threat from future climate
change. Each module produces a ranked list of populations from 1 (highest priority) to
5148 (lowest priority—equal to the total number of populations across all species). We
integrate the rankings of the two modules using a weighted average:
123
Biodivers Conserv
(a) current module
current
conservaon
status (G-rank)
current
species
trend
seed bank
(SB) rank
G1
declining
SB0
stable
SB1
increasing
SB2
higher priority
SB3
(as above)
(apply
genec
diversity
module)
(as above)
G2
(as above)
(as above)
G3
(as above)
(as above)
(b) climate module
climate change
(CC) rank
habitat type
pollinaon
syndrome
CC0
patchy
matrix
change in
climac
suitability
by animal
not by
wind
Δ
suitability
by wind/
selfing
by wind
Δ
suitability
(as above)
Δ
suitability
(as above)
(as above)
Δ
suitability
CC1
(as above)
(as above)
(as above)
Δ
suitability
CC2
(as above)
(as above)
(as above)
Δ
suitability
CC3
(as above)
(as above)
(as above)
Δ
suitability
(c) genec diversity module: original ranks
(d) genec diversity module: new ranks
4
5
3
5
3
8
2
8
2
7
123
(apply
genec
diversity
module)
6
1
4
7
6
1
higher priority
dispersal
Biodivers Conserv
b Fig. 1 Conceptual diagrams of modules used to rank species and populations. Current vulnerability (a) and
climate change vulnerability (b) are calculated using hierarchical frameworks that account for dependencies
between factors affecting vulnerability. In the current and climate modules each step is applied sequentially
with populations ending in bins higher in the last column of a module receiving higher rank priority. For
example, under the current module (a) the most vulnerable species has a G1 rank, is declining, and has no
seed bank accessions (SB0). Under the climate module (b) the most vulnerable species has all known
populations negatively exposed to climate change (CC0), lives in patchy habitats, is pollinated by animals,
and is not dispersed by wind. The last step in the climate module ranks populations in each dispersal bin
based on change in suitability estimated from niche models. The genetic diversity module (c, d) is applied to
the rankings generated from the current or climate modules (shown as irregular polygons). Its effect on
rankings is illustrated for Amoprha ouachitensis which occurs in 3 ecoregions. For this illustration the
species is assumed to have the top 8 ranked populations (for visual clarity not all population ranks are
shown). Ranks generated by the current or climate modules (c) are swapped within the species to obtain a
more even distribution of ranks across ecoregions (d)
ð1 wÞrcurr þ wrclim
ð1Þ
where rcurr is a population’s rank from the current module and rclim the rank from the
climate module. When w = 0 the climate module has no effect and the ranking reflects
only threat in the present (here defined roughly as a period from 1981 to 2010). When
w = 1 priorities are assigned using only climate threat (here the period from 2041 to 2070).
When w = 0.5 both receive equal weight.
Both current and climate modules are structured in a hierarchical fashion to reflect
dependencies between factors that influence vulnerability (Fig. 1). Dependences are
reflected by the structure of the hierarchies: specifically, splits made higher in the hierarchies affect the outcomes of splits made at lower levels. Other commonly-employed
systems use score-based modules in which each factor is assigned a value depending on its
perceived importance to vulnerability (e.g., Gardali et al. 2012; Reece and Noss 2014;
Young et al. 2015). Scores are then added or otherwise aggregated within and across
modules. While conceptually simple, score-based systems do not account for important
dependencies between components of vulnerability. For example, the current module is
based on the presumption that the rarest of species should have priority over more common
species, even if the latter are declining. Likewise, in our climate module we placed dispersal lower in the hierarchy than pollination because successful pollination is a prerequisite for successful dispersal of propagules. Score-based systems cannot be guaranteed to
reflect these dependences.
Assessing current vulnerability and protection
The first step in the current module (leftmost step in Fig. 1a) sorts species by their current
conservation status (NatureServe 2014; Table A1). NatureServe classifies species by ‘‘Granks’’ ranging from G1 (critically imperiled) to G5 (secure). G-ranks are based on several
criteria including but not limited to total number of known populations, overall trend
(increasing/stable/declining), and levels of in situ protection. We only considered species
ranked G3 (vulnerable) or less. The second level sorts each species within a G-rank by its
overall species-level trend: increasing, stable, or declining (NatureServe 2014). NatureServe considers trend when designating G-ranks, but by sorting species within a G-rank
by trend ‘‘double-counting’’ of trend is avoided. The third level sorts each species with a
particular trend according to its existing representation in conservation seed banks. To
assess seed bank representation we used a system akin to population categories of G-ranks:
loosely, G1 species have from 1 to 5 populations, G2’s 6–20, G3’s 21–80, and G4’s C80
123
Biodivers Conserv
populations. We adopted these ranges to define ‘‘seed bank’’ rank (SB1 has from 1 to 5
seed bank accessions, SB2 6–20, etc.; cf. Farnsworth et al. 2006). Species with no
accessions were assigned SB0. SB rank was calculated using the most comprehensive
databases from the Center for Plant Conservation’s National Collection of Endangered
Plants and the US Department of Agriculture’s Germplasm Resources Information Network. In some cases there were more accessions than known populations which occurred if
now-extirpated populations had been collected or if populations had been collected more
than once. Thus, a species could be assigned a higher SB-rank than its current G-rank.
Current conservation status was assessed first by sorting species by G-rank, then trend,
then SB-rank (Fig. 1a). The final priority was assigned based on the position of the species
in the third level of bins in Fig. 1a. For example, the highest-priority species (a G1 species
with a declining trend and SB0) would fall into the topmost bin and species with lower
priority would fall into bins below this. Populations of each species were then assigned
ranks by bins into which their species were sorted, with average rank assigned to populations in the same bin.
Assessing vulnerability to future climate change
The first step in the climate module estimates exposure to climate change using ecological
niche models (Fig. 1b). Details of the modeling procedure are presented in the Online
Resource, so here we outline just the general approach. We adopted a strategy to address
the problems of (1) modeling species with few presences and (2) potential underestimation
of species’ climatic tolerances. The strategy employs ecological and geographic surrogates
to augment the number of presences of focal species available for modeling. Conservation
surrogates are used widely to reflect rare species’ distributions (Lewandowski et al. 2010),
habitat preferences (Carmel and Stoller-Cavari 2006), and response to threats (Pierson
et al. 2015). Specifically, we trained two or three ecological niche models per species: (1) a
species-level model (SLM) using the presence sites of each species (as is typical among
modeling studies); (2) a ‘‘narrow-proxy’’ model (NPM) using the focal species’ presences
plus presences of nearby, ecologically similar species; and (3) a ‘‘broad-proxy’’ model
(BPM) using the focal species’ presences plus presences of ecologically similar species
selected from a larger area to represent potentially broader climatic preferences of the
target species. SLMs were trained for all species with C25 geographically unique presences (51 of 71 species). For NPMs and BPMs we first identified species that had the same
lifespan/phenology (perennial, winter annual, or summer annual) and habitat type
(woodland, glade, riparian, wetland, and ‘‘open’’ habitats including rocky outcrops and
barrens) as the focal species. For NPMs we selected presences among these species from
the area circumscribed by the intersection of the US EPA Level IV ecoregion(s) (Omernik
1987) and US Forest Service Provisional Seed Transfer Zones climatic zone(s) (Bower
et al. 2014) in which the focal species’ presences occurred (Table A1). BPMs were trained
using all presences of all species with the same lifespan/phenology in the same region of
distribution (east or west of the Mississippi River or both). As a check on the validity of
this approach we also trained NPMs for the more common species ([25 presences) and
compared the output and performance to their SLMs.
All climate-based ecological-niche models were projected to a period centered on the
2050s under the RCP4.5 and 8.5 emissions scenarios based on ensemble predictions of 15
global circulation models (Wang et al. 2012). For each model type (SLM, NPM, or BPM)
we calculated species’ exposure to climate change (first level in the climate module;
Fig. 1b) by counting the number of populations that had a future suitability less than the
123
Biodivers Conserv
10th-percent quantile of suitability across all populations of that species in the present.
Each species was then assigned a ‘‘climate change’’ rank (CC-rank) analogous to G-rank
based on the number of populations that were not exposed. For example, Eriogonum
longifolium var. harperi (a G2 species) has 16 populations, 11 of which were classified as
exposed (using its NPM for RCP8.5), so it was assigned CC1 because it has between 1 and
5 unexposed populations. Species for which all populations were exposed were classified
as CC0.
The next three levels in the climate module assess species’ sensitivity (ability to
reproduce) and adaptive capacity (ability to disperse) in light of changing climate. Both
reproduction and dispersal are affected by the patchiness of habitat, where patchier habitats
pose greater barriers to success in both (Aizen and Feinsinger 1994; Wolf and Harrison
2001; Honnay 2007). Species were categorized based on whether they were associated
with matrix habitat (forests and woodlands—less vulnerable) or patchy habitats (nonmatrix habitats—more vulnerable; Table A1). This level was placed higher in the hierarchy
than the subsequent two levels which assessed pollination and dispersal since both of the
latter are affected by the nature of habitat connectivity (Aizen and Feinsinger 1994; Wolf
and Harrison 2001; Honnay 2007). Species were then categorized based on their reliance
on other species for pollination (insects or birds—more vulnerable) or not (selfing or via
wind—less vulnerable) based on their floral morphology and autecological descriptions.
Pollination was placed higher in the hierarchy than dispersal because successful pollination
is a prerequisite for successful dispersal. Finally, dispersal syndrome was coded into two
categories based on whether the species could disperse by wind (less vulnerable) or not
(more vulnerable). These aspects of sensitivity and adaptive capacity are not the only ones
that are relevant to climate vulnerability, but they do reflect key aspects of potential
responses to changing climate. Other factors such as genetic variation, mating system, seed
dormancy, and interactions with other species are likely also very relevant but much less
data exists for quantifying these aspects.
The final level of the climate module ranks populations based on the change in their
expected exposure, with populations experiencing greater proportional loss of suitability
assigned higher priority. Ties were assigned average ranks. Overall climate vulnerability
was calculated by sorting species sequentially by CC-rank, habitat type, pollination syndrome, and dispersal syndrome, then ranking populations in each pollination syndrome bin
by change in exposure. For example, the highest-priority species in the climate module
would have all of its populations exposed (CC0), live in patchy habitats, be pollinated by
animals, and have seeds not dispersed by wind (i.e., by animals or by gravity or ballistically). Within this species the highest-priority populations would have the greatest loss of
climatic suitability as estimated by ecological niche models.
Capturing genetic diversity
Rare alleles are more likely to be collected if sampling is spread among regions in which
populations have become locally adapted (vs. concentrating effort within a single region;
Hoban and Schlarbaum 2014). Hence, we applied a re-ranking procedure to rankings from
the current and climate modules (Fig. 1). This genetics module swaps ranks within species
such that if the species is distributed among n ecoregions the first n-ranked populations are
distributed among them (one per region). Otherwise populations’ original ranks are
maintained as much as possible. Following Johnson et al. (2010), we assumed local
adaptation occurs within US EPA Level IV ecoregions (Omernik 1987). For species
present in both eastern and western portions of the Highlands we instead used ‘‘east’’ or
123
Biodivers Conserv
‘‘west’’ as ecoregions. The procedure was not applied to species occurring in just one
ecoregion. When combining rankings of the current and climate module we first ranked
populations using each module then applied the genetics module to each of these rankings.
We then calculated a weighted rank according to Eq. 1.
Strategies to manage uncertainty
We accounted for uncertainty in our prioritization system by ranking species based on two
opposing uncertainty-management strategies: precautionary and resource-conservative.
The precautionary uncertainty-management strategy targets species that have high levels of
priority regardless of the uncertainty associated with their assessment. This strategy
assumes a ‘‘pessimistic’’ evaluation, meaning that when there was a range of values to
calculate vulnerability, we used values that maximized vulnerability. For example, if a
species had an uncertain G-rank (which occurred for 13 % of the species; Table A1), we
used the lowest one, assumed a declining trend if trend was unknown (44 % of species),
and used the lesser number of accessions among the two seed banks’ databases if both
banks had accessions (11 % of species), the greater value if one had no accessions (39 %
of species), and zero if neither had accessions (51 %). Pessimistic evaluation for the
climate module used predictions from SLMs for species with C25 presences (72 % of
species) and NPMs for species with \25 presences (28 %), and assumed only animalassisted pollination unless we knew otherwise. A resource-conservative uncertainty-management strategy assumed optimistic values, and prioritized species that are more certainly
vulnerable and therefore safer investments. Climate exposure for the resource-conservative
uncertainty-management strategy was calculated using BPMs.
We calculated 10 sets of rankings using three weights of current versus climate threat
(w = 0, 0.5, and 1; Eq. 1), two uncertainty-management strategies, and two RCPs.
Hereafter different combinations of w, uncertainty-management strategy, and RCPs are
referred to as ‘‘scenarios.’’ We compared rankings between scenarios using the Spearman
rank correlation coefficient.
Identifying hotspots for collection
For each scenario we identified hotspots for efficient collection using the top 10 %-ranked
populations. Here, hotspots represent areas with high densities of vulnerable populations in
which effort can be concentrated efficiently. Hotspots were located using the 25th- and
50th-percentile contour intervals of a kernel density estimator trained on these populations
(Wiegand and Moloney 2014). Other levels (e.g., top 5 or 2 % of populations or 75 or
100 % contours) circumscribed areas either similar to these or too spatially restricted or
generous to guide efficient collection.
Results
Population ranks varied most by module, then strategy for managing uncertainty, then by
emissions scenario. Correlations between scenarios using only the current module (w = 0)
and climate module alone (w = 1) were very weak (range rs = 0.05–0.18, depending on
uncertainty-management strategy and RCP). Correlations were moderate between integrated scenarios (w = 0.5) and their respective current-only scenarios (rs = 0.41–0.77)
123
Biodivers Conserv
and the integrated scenarios and climate-only scenarios (rs = 0.52–0.75). Correlations
between scenarios with the same module (and possibly RCP) but different uncertaintymanagement strategies were moderate to high, ranging from 0.58 to 0.81. Correlations
between scenarios with different RCPs were high, ranging from 0.82 to 0.94. The genetics
module barely changed rankings across species (rs for any scenario with and without
application of the genetics module was 0.99).
For each scenario we designated a species a ‘‘priority species’’ if at least one of its
populations fell within the top 10 % of ranks for that scenario. Increasing the importance of
climate in the assessment increased the mean G-rank of priority species, from 1.47 ± 0.10
(± SE) under the current-only scenario, to 1.66 ± 0.08 under integrated scenarios, to
2.22 ± 0.08 under the climate-only scenarios (averaged across uncertainty-management
strategies and RCPs). Each integrated scenario and its respective and current-only scenario
shared between 9 and 13 priority species (depending on uncertainty-management strategy
and RCP), and the integrated and climate scenarios shared between 10 and 12 species. In
contrast, current-only and climate-only scenarios shared just 6 to 7 priority species. Only
Astragalus bibullatus, a G1 species, was consistently designated a priority species across
all scenarios.
For species modeled with both NPMs and SLMs, predictions at presence sites were
strongly correlated in the current (mean r = 0.89 ± 0.02, P \ 0.001) and future periods
(r = 0.82 ± 0.04, P \ 0.001 for both RCPs). SLMs and NPMs also had similar performance (Online Resource). Hence, we felt confident using predictions from NPMs for
species with \25 presence sites. As expected, SLMs/NPMs which use the current distribution or assume narrow potential distributions predicted more exposure to climate change
than the BPMs which assume broader potential distributions (mean ± SE change in
suitability at presence sites for SLMs/NPMs across species and RCPs was -0.30 ± 0.01
for RCP4.5 and 0.37 ± 0.01 for RCP8.5; for BPMs -0.17 for RCP4.5 and -0.24 ± 0.00
for RCP8.5).
Across all 10 scenarios 11 hotspots were identified in total, varying from 2 to 6 per
scenario (Figs. 2, 3). Areas in the western region were rarely highlighted (western region:
Ouachita Mountains 2 times across scenarios, Western Ozarks 2, Tiak Glades 1; eastern
region: Central Basin 10, Eastern Highland Rim 8, Bluegrass Region: 5, Moulton Valley 4,
Big South Fork 4, all others 1 to 2 each). Only the Central Basin of Tennessee was
consistently highlighted across all scenarios. In contrast, the locations of other hotspots
depended on an interaction between the weight of the climate module relative to the
current module, the strategy used to handle uncertainty, and emissions scenario. For
example, the Eastern Highland Rim was highlighted in all scenarios except for the integrated and climate-only scenarios using the resource-conservative uncertainty-management strategy for RCP4.5 (Fig. 2e, f), and the Western Highland Rim was only highlighted
under the current module using the resource-conservative uncertainty-management strategy (Fig. 2d). Likewise, emissions pathway had an important effect. For example, the
integrated resource-conservative scenario for RCP4.5 emphasized both eastern and western
subregions, but the integrated resource-conservative scenario for RCP8.5 emphasized only
areas in the east (Figs. 2e, 3e).
Surprisingly, some hotspots were only highlighted under the integrated scenarios and
not their respective current- or climate-only scenarios, meaning they would have been
overlooked were the assessment conducted using only current or climate threat. For
example, the Ouachita Mountains, Tiak Glades, and Ketona Glades were highlighted under
the integrated resource-conservative scenario for RCP4.5 (Fig. 2e) but only weakly
highlighted or entirely absent from the current- or climate-only scenarios with the same
123
Biodivers Conserv
Fig. 2 Hotspots for efficient seed banking using the precautionary (a–c) or resource-conservative (d–
f) strategies for managing uncertainty for greenhouse gas emissions pathway RCP4.5. The precautionary
strategy directs attention toward species and populations that might be threatened, even if it is not known
with certainty that they are. The resource-conservative strategy directs attention toward species and
populations that are known to be vulnerable. To identify hotspots we first used a kernel density estimator
trained on the top 10 %-ranked populations in each scenario. We then located hotspots using polygons
drawn around the top 25- and 50 % of the density estimator’s values (i.e., the highest concentrations of topranked populations). The Central Basin of Tennessee was always highlighted as a hotspot. The locations of
other hotspots depended on the strategy used to manage uncertainty and the relative weight of climate versus
current threat. Hotspots in the integrated scenarios were not necessarily identified by their respective
current- or climate-only scenarios. All hotspots in this figure are based on rankings from the current and
climate modules subsequently processed by the genetics module (Fig. 1)
123
Biodivers Conserv
Fig. 3 Hotspots identified using the precautionary (a–c) or resource-conservative (d–f) strategies for
managing uncertainty for emissions pathway RCP8.5. Hotspots shown for the current module (a, d) are the
same as those in Fig. 2 but reproduced here to aid visual comparison. The Central Basin was consistently
identified as a hotspot in all scenarios, but the locations of other hotspots were less dependable. Increasing
the weight of climate under the precautionary strategy for managing uncertainty dispersed hotspots away
from the Central Basin (b, c), whereas increasing the weight of climate under the resource-conservative
strategy removed more hotspots (e, f). See Fig. 2 for more details
uncertainty-management strategy and RCP (Fig. 2d, f). Increasing the weight of climate in
the assessment had different effects depending on the uncertainty-management strategy
and RCP. For example, increasing the weight of climate under the precautionary uncertainty-management strategy for RCP8.5 shifted emphasis to areas more peripheral to the
Central Basin (Fig. 3a–c), whereas increasing climate’s weight under the resource-
123
Biodivers Conserv
conservative uncertainty-management strategy for the same RCP removed peripheral
hotspots (Fig. 3d–f). Application of the genetics module had almost no effect on the
locations of hotspots (Figs. A2–A5).
Discussion
The objectives of our assessment were to (1) identify hotspots for efficient seed banking of
highly vulnerable species; (2) compare spatial priorities based on current threat, climate
treat, and their integration; and (3) determine the robustness of hotspots to uncertainty. The
locations of hotspots varied according the particular combination of type of threat (current,
climate, integrated), strategy used to manage uncertainty, and emissions pathway (Figs. 2,
3). Only one hotspot (the Central Basin) appeared consistently in all scenarios, and only
one species (A. bibullatus) was consistently flagged for special attention by current, climate, and integrated scenarios. In fact, the current module (by design) prioritized critically
imperiled species for ex situ seed banking, while the climate module emphasized species
that are currently more secure in their conservation status. Some hotspots were only
highlighted under the integrated modules. These results underscore the risks of relying on
priorities based on just current or climate threats. Uncertainty-management strategy and
emissions pathways had as much of an effect on hotspots as the module(s) used to rank
populations (Figs. 2, 3). Thus, a full consideration of uncertainty is just as important as
integrating current and climate threats. In contrast, we found few differences in hotspots
when applying methods to enhance acquisition of genetic diversity within a species
(Figs. A2–A5). Hence, there seems to be little cost to pursuing a strategy that maximizes
the probability of capturing genetic diversity even though it is more likely to spread
priority among more distant populations.
Hotspots for efficient seed banking
The eastern Highlands had hotspots in every scenario, while areas in the western Highlands
were rarely highlighted (Figs. 2, 3). In part this may be explained by unequal frequencies
of rare species in each subregion: 33 species with 2812 populations in the east and 28
species with 1354 populations in the west (10 species with 982 populations occur in both
regions). All scenarios highlighted the Central Basin in the eastern Highlands. The Basin
maintains some of the highest levels of endemism and concentration of rare species in the
southeastern US due to its abundance of limestone glades (Estill and Cruzan 2001), which
are threatened by urban expansion of the greater Nashville metropolitan area (Estill and
Cruzan 2001; NatureServe 2014). Hence, the Basin is a robust target for seed banking
under any scenario. In contrast, the Ouachita Mountains comprise the other area of high
endemism within our study region (Zollner et al. 2005) with many of its species currently
threatened by forestry, recreation, and mining (NatureServe 2014). Nonetheless, the
Ouachitas were rarely highlighted. Although most hotspots circumscribed many species, a
few centered on highly-threatened populations of just one or two species. For example, the
Powell River Valley hotspot contains just Trifolium calcaricum (a G1 species). Nevertheless, most offer opportunities for efficient collection because they are fairly small in size
(\150 km in diameter) and contain many other populations that can be collected opportunistically. For example, *75 populations that fall within the top 10 % of rankings under
123
Biodivers Conserv
different scenarios are contained by the Moulton Valley hotspot, but *30 additional
populations occur in the same location, all together representing 15 species.
Integrating threats
In general priorities based on the integrated module were intermediate between those based
on current- or climate-only modules. The climate-only scenarios emphasized species that
are currently more secure (higher G-rank) in their conservation status than the current-only
scenarios, which by design ranked species that are presently rare and threatened higher. In
turn, the integrated module emphasized species that were on average not as endangered as
those under the current module but not as secure as those under the climate module.
Likewise, correlations between rankings of current- and climate-only modules were very
low (rs always B0.18), but moderate to high between integrated and current rankings and
between integrated and climate rankings (0.41–0.77). Although integrated rankings tended
to produce results in the ‘‘middle’’ of current- and climate-only scenarios, integration of
current and climate modules sometimes produced distinct spatial results. For example,
some hotspots identified by integrated rankings did not emerge strongly in their corresponding current- or climate-only scenarios (e.g., the Ouachita Mountains, Tiak Glades,
and Ketona Glades; Fig. 2a–c).
These results highlight two important generalizable lessons for vulnerability assessments. First, conservation priorities identified using just current or just climate threat alone
may misdirect attention from species and regions that might be most threatened by their
combination. Presumably, species that are currently rare are predisposed to harm from
climate change because of their precarious status (Pearson et al. 2014). Indeed, recent work
suggests that factors that are explicitly considered when assigning species’ G-ranks
(number of populations, range size, and overall trend) correlate positively with propensity
to go extinct under simulated climate change (Keith et al. 2014; Pearson et al. 2014;
Stanton et al. 2015). On this basis, we expected high correlations between current- and
climate-only ranks, but this is not what we found (rs between current and climate
ranks B0.18). Hence, we caution against assuming current priorities will align with future
priorities, even if current threats remain unabated. To date, published seed banking
assessments concentrate on just one type of threat (current or climate) but very few
consider them simultaneously (Vitt et al. 2010; Havens et al. 2015). Thus, we urge integration of current and climate threats when designing seed-banking strategies.
We acknowledge that integrating vulnerability from current and climate threats is difficult and encourage more research in this area. In particular, we expect that current and
climate threats can be more easily combined when they are measured with the same
‘‘currency’’ (e.g., ranks as used here). For example, the latest version of NatureServe’s
Climate Change Vulnerability Index (CCVI; Young et al. 2015; version 3.0), now widely
used to prioritize species-level conservation in North America, does not explicitly consider
population size, range size, or demographic parameters, even though these are key variables in calculating NatureServe’s G-ranks (Young et al. 2015). Hence, the 5 categories
into which the CCVI classifies species are not directly comparable to NatureServe’s five
G-rank classes, which complicates their joint interpretation. For example, is a G2 species
that falls into the CCVI’s ‘‘less vulnerable’’ class a higher priority than a G3 species that
falls into its ‘‘extremely vulnerable’’ class? Ideally, output in this case would provide
predictions of future G-rank.
A second, related lesson is that the time-evolving nature of climate change has
important implications for prioritization of seed-banking strategies. Though inexpensive
123
Biodivers Conserv
compared to other conservation actions, seed banking is not without cost, and adequate
collection will require years if not decades to complete (Guerrant et al. 2014). In our case
near-term collection should prioritize species identified where the climate module has little
weight (w *0), but as time passes priorities should shift to reflect greater climate threat
(w [0 but \1). An alternative approach is to extrapolate climatic conditions between the
current and future periods and model exposure in a series of time steps (e.g., Keith et al.
2014; Pearson et al. 2014; Stanton et al. 2015), but this still requires integrating currentand climate-based priorities.
Uncertainty
Uncertainty complicates conservation efforts but if appropriately incorporated into
assessments and communicated effectively, uncertainty can be strategically managed by
conservation practitioners in the context of their resources and tolerance of risk (Regan
et al. 2005; McCarthy 2014). In our assessment we used sensitivity analysis to reflect
uncertainty, which had as large of an effect as the relative weight of the current and climate
modules on hotspot location and identity of priority species for ex situ conservation
(Figs. 2, 3). For example, for RCP4.5 with the integrated scenario the precautionary
uncertainty-management strategy highlighted four hotspots (Fig. 2b), of which only two
were in common with the integrated resource-conservative scenario, which itself highlighted six hotspots (Fig. 2e). We encourage careful use of vulnerability assessments that
weigh priority scores by available information (e.g., Thomas et al. 2011), assume high-risk/
high-reward actions are always unfavorable (Shoo et al. 2013), or assume ‘‘neutral’’ scores
for situations where information is unknown (NatureServe’s CCVI; Young et al. 2015).
These systems mask uncertainty and so presume decision makers will always want to adopt
a given strategy to manage uncertainty. This is of special concern for assessment systems
designed to accommodate a wide variety of species and situations not necessarily known to
the system’s developers (e.g., Thomas et al. 2011; Shoo et al. 2013; Young et al. 2015).
Uncertainty can be incorporated and communicated alongside vulnerability assessments
using a separate uncertainty score for each component (e.g., Gardali et al. 2012; Moyle
et al. 2013), Monte Carlo randomization (Reece and Noss 2014), comparing vulnerability
across completely different assessment systems (Still et al. 2015), or sensitivity analysis
(e.g., Anacker et al. 2013; Foden et al. 2013; Wright et al. 2015; this study).
Dependencies between factors affecting vulnerability
Our results highlight the importance of considering not only the absolute level of vulnerability within each component of vulnerability but also the interactions between different components. Interactions are likely to be common in natural systems. For example,
the benefits of high genetic diversity to allow adaptation to climate change would be nearly
irrelevant if available habitat is inundated by sea level rise. Likewise, the vulnerability of
specialized predators, pathogens, and mutualists will depend in part on the vulnerability of
other species (Warren et al. 2010). Dependencies will also occur when new habitat is
unlikely to develop within the timeframe expected for species needing to migrate. For
example, the ability of many species to colonize areas above treeline will depend on
formation of adequate substrate.
Some vulnerability assessments use differential weighting to emphasize potentially
offsetting factors (e.g., Benscoter et al. 2013; Reece et al. 2013; Reece and Noss 2014).
However, careful attention should be given to the manner in which weighted scores are
123
Biodivers Conserv
aggregated since this could produce illogical outcomes, such as the sum of several smallweighted factors offsetting the effect of a ‘‘monolithic’’ threat like complete habitat
destruction (e.g., from sea level rise). A hierarchical system similar to the one presented
here disallows these kind of incongruities since it models the dependency of one factor on
another akin to the manner in which a classification and regression tree model represents
interactions.
The problem of rare species
Rare species are difficult to model (Lomba et al. 2010) and thus are often discarded in
vulnerability analyses, a practice which can dramatically alter priorities (Platts et al. 2014).
Moreover, their current distributions may underrepresent their true climatic tolerances if
their restricted distributions are the result of other, non-climatic related factors. For
example, historical records indicate that at least one of our species, Trifolium stoloniferum,
was once widespread, but declined likely because of overhunting of the American bison
which created microhabitats favored by the species (NatureServe 2014). Our proxy models
emulate the common practice in conservation of using surrogates to stand in for lesser
known, rare species (Carmel and Stoller-Cavari 2006; Lewandowski et al. 2010; Pierson
et al. 2015). The NPMs offer a way to sidestep the problem of modeling rare species by
generating presences that are presumably ecologically equivalent and spatially relevant to
those were the focal species more common in its current range. The BPMs offer a way to
estimate climatic envelopes for species whose ranges of extent have been diminished by
non-climatic threats or otherwise do not fully occupy their potential range limits. It is
possible that the climatic envelopes of edaphically-limited species are frequently underestimated given that plants that are not edaphically limited often do not fill their full
climatically-tolerable area of extent (Van der Veken et al. 2008). Since rare, edaphicallyrestricted plant species are of special conservation concern, these methods deserve more
attention.
Conclusions
In an era of rapid global change, conservation seed banking will always be chasing a
moving target. Although species that are currently rare might be expected to suffer more
under changing climate, this may not always be the case, meaning that priorities established using current conservation status could misdirect conservation efforts. Consequently, conservation practitioners should carefully consider the manner in which current
threats are integrated with future threats when establishing priorities. In particular, using
just current conservation status or just vulnerability to climate change can produce priorities that are distinctly different from an integrated assessment. We expect integration
will be facilitated in cases when vulnerability to different kinds of threats are measured
using the same currency (e.g., current and eventual levels of rarity). We also advise
explicitly considering the effect of uncertainty on conservation priorities. Uncertainty can
be managed if assumptions are clear and results presented in a manner that allows decisions to be made based on levels of tolerable risk and available resources.
Acknowledgments We thank Iván Jiménez and Sebastián Tello for helpful discussions and J. Leighton
Reid for helpful comments on the manuscript. This work was funded by an Institute of Museum and Library
Services National Leadership Grant (LG-25-10-0035-10). We thank NatureServe and the Tennessee
123
Biodivers Conserv
Department of Environment and Conservation, Georgia Department of Natural Resources, Missouri
Department of Conservation, Kentucky State Nature Preserves Commission, Alabama Natural Heritage
Program, Illinois Department of Natural Resources, and Arkansas Natural Heritage Commission. Two
anonymous reviewers generously helped to improve the manuscript.
References
Aizen MA, Feinsinger P (1994) Forest fragmentation, pollination, and plant reproduction in a Chaco dry
forest, Argentina. Ecology 75:330–351
Anacker BL, Gogol-Prokurat M, Leidholm K, Schoenig S (2013) Climate change vulnerability assessment
of rare plants in California. Madroño 60:193–210
Baskin JM, Baskin CC (2003) The vascular flora of cedar glades of the southeastern United States and its
phytogeographical relationships. J Torrey Bot Soc 130:101–118
Benscoter AM, Reece JS, Noss RF, Brandt LA, Mazzotti FJ, Romañach SS, Watling JI (2013) Threatened
and endangered subspecies with vulnerable ecological traits also have high susceptibility to sea level
rise and habitat fragmentation. PLoS One 8:e70647
Bower AD, Clair BS, Erickson V (2014) Generalized seed transfer zones for native plants. Ecol Appl 24:913–919
Carmel Y, Stoller-Cavari L (2006) Comparing environmental and biological surrogates for biodiversity at a
local scale. Isr J Ecol Evol 52:11–27
Cochrane JA, Crawford AD, Monks LT (2007) The significance of ex situ seed conservation to reintroduction of threatened plants. Aust J Bot 55:356–361
Estill JC, Cruzan MB (2001) Phytogeography of rare plant species endemic to the southeastern United
States. Castanea 66:3–23
Farnsworth EJ, Klionsky S, Brumback WE, Havens K (2006) A set of simple decision matrices for prioritizing collection of rare plant species for ex situ conservation. Biol Conserv 128:1–12
Foden WB, Butchart SHM, Stuart SN, Vié J-C, Akçakaya R, Angulo A, DeVaniter LD, Gutsche A, Turak E,
Cao L, Donner SD, Katariya V, Bernard R, Hollamd RA, Hughes AF, O’Hanlon SE, Garnett ST,
Şekercioğlu ÇH, Mace GM (2013) Identifying the world’s most climate change vulnerable species: a
systematic trait-based assessment of all birds, amphibians and corals. PLoS One 8:e65427
Gardali T, Seavy NE, DiGuadio RT, Comrack LA (2012) A climate change vulnerability assessment of
California’s at-risk birds. PLoS One 7:e29507
Gauthier R, Debussche M, Thompson JD (2010) Regional priority setting for rare species based on a method
combining three criteria. Biol Conserv 143:1501–1509
Godefroid S, Vanderborght T (2010) Seed banking of endangered plants: Are we conserving the right
species to address climate change? Biol Conserv 19:3049–3058
Guerrant EO Jr, Havens K, Maunder M (2004) Ex situ plant conservation: supporting species survival in the
wild. Island Press, Washington
Guerrant EO Jr, Havens K, Vitt P (2014) Sampling for effective ex situ plant conservation. Int J Plant Sci
175:11–20
Havens K, Vitt P, Still S, Kramer A, Fant J, Schatz K (2015) Seed sourcing for restoration in an era of
climate change. Nat Areas J 35:122–133
Hoban S, Schlarbaum S (2014) Optimal sampling of seeds from plant populations for ex situ conservation of
genetic biodiversity, considering realistic population structure. Biol Conserv 177:90–99
Honnay O, Jacquemyn H (2007) Susceptibility of common and rare plant species to the genetic consequences of habitat fragmentation. Conserv Biol 21:1–9
Johnson R, Stritch L, Olwell P, Lambert S, Horning ME, Cronn R (2010) What are the best seed sources for
ecosystem restoration on BLM and USFS lands? Nativ Plants J 11:117–131
Keith DA, Mahoney M, Hines H, Elith J, Regan TJ, Baumgartner JB, Hunter D, Heard GW, Mitchell NJ,
Parris KM, Penman T, Scheele B, Simpson CC, Tingley R, Tracy CR, West M, Akçakaya HR (2014)
Detecting extinction risk from climate change by IUCN Red List criteria. Conserv Biol 28:810–819
Kricsfalusy VV, Trevisan N (2014) Prioritizing regionally rare plant species for conservation using
herbarium data. Biodivers Conserv 23:39–61
Lewandowski AS, Noss RF, Parsons DR (2010) The effectiveness of surrogate taxa for the representation of
biodiversity. Conserv Biol 24:1367–1377
Lomba A, Pellissier L, Randin C, Vicente J, Horondo J, Guisan A (2010) Overcoming the rare species
modeling complex: a novel hierarchical framework applied to an Iberian endemic plant. Biol Conserv
143:2647–2657
123
Biodivers Conserv
Maunder M, Havens K, Guerrant EO Jr, Falk DA (2004) Ex situ methods: a vital but underused set of
conservation resources. In: Guerrant EO Jr, Havens K, Maunder M (eds) Ex situ plant conservation:
supporting species survival in the wild. Island Press, Washington, pp 3–20
Mayden RL (1987) Historical ecology and North American Highland fishes: a research program in community ecology. In: Matthews WJ, Heins DC (eds) Community and evolutionary ecology of North
American stream fishes. University of Oklahoma Press, Norman, pp 210–220
McCarthy MA (2014) Contending with uncertainty in conservation management decisions. Ann N Y Acad
Sci 1322:77–91
Moyle PB, Kiernan JD, Crain PK, Quiñones RM (2013) Climate change vulnerability of native and alien
freshwater fishes of California: a systematic assessment approach. PLoS One 8:e63883
NatureServe (2014) NatureServe explorer: an online encyclopedia of life. Version 7.1. NatureServe,
Arlington, Virginia. http://explorer.natureserve.org. Accessed 9 Mar 2014
Noss RF (2013) Forgotten grasslands of the south. Island Press, Washington
Omernik JM (1987) Ecoregions of the conterminous United States. Map (scale 1:7,500,000). Ann Assoc Am
Geogr 77:118–125
Pacifici M, Foden WB, Visconti P, Watson JEM, Butchart SHM, Kovacs KM, Scheffers BR, Hole DG,
Martin TG, Akçakaya HR, Corlett RT, Huntley B, Brickford D, Carr JA, Hoffmann AA, Midgley GF,
Pearce-Kelly P, Pearson RG, Williams SE, Willis SG, Yoing B, Rondinini C (2015) Assessing species
vulnerability to climate change. Nat Clim Change 5:215–225
Pearson RG, Stanton JC, Shoemaker KT, Aiello-Lammens ME, Ersts PJ, Horning N, Fordham DA, Raxworthy CJ, Ryu HY, McNees J, Akçakaya HR (2014) Life history and spatial traits predict extinction
risk due to climate change. Nat Clim Change 4:217–221
Pierson JC, Barton PS, Lane PW, Lindenmayer DB (2015) Can habitat surrogates predict the response of
target species to landscape change? Biol Conserv 184:1–10
Platts PJ, Garcia RA, Hof C, Foden W, Hansen LA, Rahbek C, Burgess ND (2014) Conservation implications of omitting narrow-ranging taxa from species distribution models, now and in the future. Divers
Distrib 20:1307–1320
Reece JS, Noss RF (2014) Prioritizing species by conservation value and vulnerability—a new index applied
to species threatened by sea-level rise and other risks in Florida. Nat Areas J 34:31–45
Reece JS, Noss RF, Oetting J, Hoctor T, Volk, M (2013) A vulnerability assessment of 300 species in
Florida: Threats from sea level rise, land use, and climate change. PloS one 8:e80658
Regan HM, Ben-Haim Y, Langford B, Wilson WG, Lundburg P, Andelman SJ, Burgman MA (2005) Robust
decision-making under severe uncertainty in conservation management. Ecol Appl 15:1471–1477
Shoo LP, Hoffmann AA, Garnett S, Pressey RL, Williams YM, Taylor M, Falconi L, Yates CJ, Scot JK,
Alagador D, Williams SE (2013) Making decisions to conserve species under climate change. Clim
Change 119:239–246
Stanton JC, Shoemaker KT, Pearson RG, Akçakaya HR (2015) Warning times for species extinctions due to
climate change. Glob Change Biol 21:1066–1077
Still SM, Frances AL, Treher AC, Oliver L (2015) Using two climate change vulnerability assessment
methods to prioritize and manage rare plants: a case study. Nat Areas J 35:106–121
Thomas CD, Hill JK, Anderson BJ, Bailey S, Beale CM, Bradbury RB, Bulman CR, Crick HQP, Eigenbrod
F, Griffiths HM, Kunin WE, Oliver TH, Walmsley CA, Watts K, Worsfold NT, Yardley T (2011) A
framework for assessing threats and benefits to species responding to climate change. Methods Ecol
Evol 2:125–142
Van der Veken S, Hermy M, Vellend M, Knapen A, Verheyen K (2008) Garden plants get a head start on
climate change. Front Ecol Environ 6:212–216
Vitt P, Havens K, Kramer AT, Sollenberger D, Yates E (2010) Assisted migration of plants: changes in
latitudes, changes in attitudes. Biol Conserv 143:18–27
Wang T, Hamann A, Spittlehouse DL, Murdock TQ (2012) ClimateWNA—High resolution spatial climate
data for Western North America. J Appl Meteorol Climatol 51:16–29
Ware S (2002) Rock outcrop plant communities (glades) in the Ozarks: a synthesis. Southwest Nat
47:585–597
Warren M, Robertson MP, Greeif JM (2010) A comparative approach to understanding factors limiting
abundance patterns and distributions in a fig tree-wasp mutualism. Ecography 33:148–158
Wiegand T, Moloney KA (2014) Handbook of spatial point-pattern analysis in ecology. Taylor & Francis,
Boca Raton
Willis SG, Foden W, Baker DJ, Belle E, Burgess ND, Carr JA, Doswald N, Garcia RA, Hartley A, Hof C,
Newbold T, Rahbek C, Smith CJ, Visconti P, Young BE, Butchart SHM (2015) Integrating climate
change vulnerability assessments from species distribution models and trait-based approaches. Biol
Conserv 190:167–178
123
Biodivers Conserv
Wolf AT, Harrison SP (2001) Effects of habitat size and patch isolation on reproductive success of the
serpentine morning glory. Conserv Biol 15:111–121
Wright AN, Hijmans RJ, Schwartz MW, Shaffer HB (2015) Multiple sources of uncertainty affect metrics
for ranking conservation risk under climate change. Divers Distrib 21:111–122
Wyse Jackson P (2008) The potential impact of climate change on native plant diversity in Ireland. National
Botanic Gardens of Ireland. www.botanicgardens.ie/news/(2008)0122.htm. Accessed 1 July 2015
Wyse Jackson P, Kennedy K (2009) The global strategy for plant conservation: a challenge and opportunity
for the international community. Trends Plant Sci 14:578–580
Young BE, Byers E, Hammerson G, Frances A, Oliver L, Treher A (2015) Guidelines for using the
natureserve climate change vulnerability index, Rlease 3.0. NatureServe, Arlington
Zollner D, MacRoberts MH, MacRoberts BR, Ladd D (2005) Endemic vascular plants of the Interior
Highlands, U.S.A. Sida 21:1781–1791
123