REVIEW SUMMARY◥

CLIMATE CHANGE

Improving the forecast forbiodiversity under climate changeM. C. Urban,* G. Bocedi, A. P. Hendry, J.-B. Mihoub, G. Pe’er, A. Singer, J. R. Bridle,L. G. Crozier, L. De Meester, W. Godsoe, A. Gonzalez, J. J. Hellmann, R. D. Holt,A. Huth, K. Johst, C. B. Krug, P. W. Leadley, S. C. F. Palmer, J. H. Pantel, A. Schmitz,P. A. Zollner, J. M. J. Travis

BACKGROUND: As global climate change ac-celerates, one of the most urgent tasks for thecoming decades is to develop accurate pre-dictions about biological responses to guide theeffective protection of biodiversity. Predictivemodels in biology provide ameans for scientiststo project changes to species and ecosystemsin response to disturbances such as climatechange. Most current predictive models, how-ever, exclude important biological mechanismssuch as demography, dispersal, evolution, andspecies interactions. These biological mech-anisms have been shown to be important inmediating past and present responses to cli-mate change. Thus, current modeling effortsdo not provide sufficiently accurate predic-

tions. Despite the many complexities involved,biologists are rapidly developing tools thatinclude the key biological processes neededto improve predictive accuracy. The biggestobstacle to applying these more realistic mod-els is that the data needed to inform themare almost always missing. We suggest waysto fill this growing gap betweenmodel sophis-tication and information to predict and preventthe most damaging aspects of climate changefor life on Earth.

ADVANCES: On the basis of empirical andtheoretical evidence, we identify six biologicalmechanisms that commonly shape responsesto climate change yet are too often missing

from current predictive models: physiology;demography, life history, andphenology; speciesinteractions; evolutionary potential and popula-tion differentiation; dispersal, colonization, andrangedynamics; and responses to environmentalvariation. We prioritize the types of informationneeded to inform each of these mechanisms andsuggest proxies for data that are missing ordifficult to collect. We show that even for well-studied species,weoften lack critical informationthatwould be necessary to applymore realistic,mechanistic models. Consequently, data limi-tations likely override the potential gains inaccuracy of more realistic models. Given the

enormous challenge ofcollecting this detailedinformation on millionsof species around theworld, we highlight prac-tical methods that pro-mote the greatest gains

in predictive accuracy. Trait-based approachesleverage sparse data to make more generalinferences about unstudied species. Target-ing species with high climate sensitivity anddisproportionate ecological impact can yieldimportant insights about future ecosystemchange. Adaptive modeling schemes provideameans to target themost important datawhilesimultaneously improving predictive accuracy.

OUTLOOK: Strategic collections of essentialbiological information will allow us to buildgeneralizable insights that inform our broaderability to anticipate species’ responses to climatechange and other human-caused disturbances.By increasing accuracy andmakinguncertaintiesexplicit, scientists can deliver improved pro-jections for biodiversity under climate changetogether with characterizations of uncertaintyto support more informed decisions by policy-makers and land managers. Toward this end,a globally coordinated effort to fill data gapsin advance of the growing climate-fueled bio-diversity crisis offers substantial advantagesin efficiency, coverage, and accuracy. Biologistscan take advantage of the lessons learnedfrom the Intergovernmental Panel onClimateChange’s development, coordination, and inte-gration of climate change projections. Climateandweather projectionswere greatly improvedby incorporating important mechanisms andtesting predictions against global weather sta-tion data. Biology can do the same.Weneed toadopt this meteorological approach to predict-ing biological responses to climate change toenhance our ability to mitigate future changesto global biodiversity and the services it pro-vides to humans.▪

RESEARCH

SCIENCE sciencemag.org 9 SEPTEMBER 2016 • VOL 353 ISSUE 6304 1113

Environment

Evolution

PhysiologyDemography

Dispersal

Species interactions

Energy and mass balance to predict physiological responses

Predicting land-use changes at relevant scales

Quantitative genetic or genetically explicit models to predict adaptive responses

Interaction matrices to predict novel communities

Climate-dependent dispersal behavior to predict spatial responses

Climate-dependent demography to predict population dynamics

Cane toad

Simulated land use

Dengue mosquito

Duskywingskipper & oaks

Meadow brown

Emperor penguin

Emerging models are beginning to incorporate six key biological mechanisms that canimprove predictions of biological responses to climate change. Models that include biologicalmechanisms have been used to project (clockwise from top) the evolution of disease-harboringmosquitoes, future environments and land use, physiological responses of invasive species such ascane toads, demographic responses of penguins to future climates, climate-dependent dispersalbehavior in butterflies, and mismatched interactions between butterflies and their host plants. Despitethesemodelingadvances,weseldomhave thedetaileddataneeded tobuild thesemodels, necessitatingnew efforts to collect the relevant data to parameterize more biologically realistic predictive models.

The list of author affiliations is available in the full article online.*Corresponding author. Email: mark.urban@uconn.eduCite this article as M. C. Urban et al., Science 353, aad8466(2016). DOI: 10.1126/science.aad8466

ON OUR WEBSITE◥

Read the full articleat http://dx.doi.org/10.1126/science.aad8466..................................................

on

Sept

embe

r 9,

201

6ht

tp://

scie

nce.

scie

ncem

ag.o

rg/

Dow

nloa

ded

from

REVIEW◥

CLIMATE CHANGE

Improving the forecast forbiodiversity under climate changeM. C. Urban,1* G. Bocedi,2 A. P. Hendry,3 J.-B. Mihoub,4,5 G. Pe’er,5,6 A. Singer,6,7,8

J. R. Bridle,9 L. G. Crozier,10 L. De Meester,11 W. Godsoe,12 A. Gonzalez,13

J. J. Hellmann,14 R. D. Holt,15 A. Huth,7,6,16 K. Johst,7 C. B. Krug,17,18 P. W. Leadley,17,18

S. C. F. Palmer,2 J. H. Pantel,19 A. Schmitz,5 P. A. Zollner,20 J. M. J. Travis2

Newbiologicalmodels are incorporating the realistic processes underlying biological responsesto climate change and other human-caused disturbances. However, these more realisticmodels require detailed information,which is lacking formost species onEarth.Currentmonitoringefforts mainly document changes in biodiversity, rather than collecting the mechanistic dataneeded to predict future changes.We describe and prioritize the biological information needed toinformmore realistic projections of species’ responses to climate change.We also highlight howtrait-based approaches and adaptive modeling can leverage sparse data to make broaderpredictions.Weoutlineaglobal effort to collect thedatanecessary tobetter understand, anticipate,and reduce the damaging effects of climate change on biodiversity.

We need to predict how climate changewillalter biodiversity if we are to preventserious damage to the biosphere (1). Biol-ogists develop predictivemodels to antic-ipate how environmental changes might

affect the future properties of species and eco-systems (2, 3). Manymodels have been developedto understand climate change impacts (fig. S1) (4),but biological responses remaindifficult to predict(5, 6). One reason is that most models forecastingbiodiversity change ignoreunderlyingmechanisms,such as demographic shifts, species interactions,

and evolution, and instead extrapolate correlationsbetween current species’ ranges and climate (Fig. 1)(4). These omissions are troubling because weknow that these missing biological mechanismsplayed key roles in mediating past and presentbiotic responses to climate change (7–9).Moreover,models ignoring biological mechanisms oftenbecome unreliable when extrapolated to novelconditions (10–13). As climates and ecologicalcommunities without historical precedent becomemore common and correlations between currentspecies distributions and climate become uncoupled(10, 14, 15), we cannot rely on tools based onstatistical descriptions of the past. Given the essen-tial role of biological processes inmediating species’responses to climate change, accurate forecastsof future biodiversity likely will require morerealistic models.Emerging models incorporate fundamental

biologicalmechanisms rather than relying solely onstatistical correlations (16–19). Unlike correlativeapproaches, mechanistic models do not assumethat a species’ range reflects its niche perfectly,has reached equilibrium with the environment,or is independent of species interactions—allcommonly violated assumptions (7, 13, 20, 21).Mechanistic models can also integrate multiple,interacting biological processes, as well as non-linear and stochastic dynamics (Fig. 2) (17, 18, 22),and canbetter characterize uncertainty by directlymodeling error sources (2, 21, 23).By incorporating realistic features such as

demography and dispersal, mechanistic modelscommonly outperform correlative approaches inprojecting climate change responses (13, 20). Forexample,mechanisticmodels consistently predictedsimulated species’ range dynamics over a periodof 75 years, whereas correlative models becameincreasingly inaccurate over this same time frame(20). Mechanistic models improve predictive accu-

racy, especially when species face strong bioticinteractions, experience novel climates, or cannotdisperse far (13, 20, 24). Moreover, mechanisticmodels can informpredictive efforts by indicatingprocesses (e.g., biotic limits on ranges) hidden bycurrent associations between environments andspecies distributions (24). Althoughmorework isneeded to craft more sophisticated and accuratemechanisticmodels that are customizable for indi-vidual species andecosystems, the tools are alreadymature enough to improve projections (2, 16, 19).Mechanistic models, however, require high-

quality data about how a species’ unique biologygoverns its responses to climate. Parametersprovidethis information. For example, a parameter such aspopulation growth rate determines how popula-tion abundances change through time. In contrast,a model variable such as population abundancedescribes emergent properties. Differentiating be-tween parameters and variables is importantgiven the recent focus on harmonizing effortsto collect variables that monitor the state of globalbiodiversity (25). We believe that such endeavorsshould focus not solely on collecting variablesthat indicate the state of biodiversity, but alsoon measuring mechanistic parameters criticalfor predicting future responses.Here, we identify the mechanistic data needed

tomake substantial gains in predictivemodeling.Rather than focusing on one particular mecha-nism (15, 18, 22, 26, 27), we take a comprehensiveapproach, assess data availability for each mech-anism, prioritize data needs, demonstrate how toleverage sparse data tomake general predictions,and suggest how global coordination could facili-tate these efforts. By synthesizing this informationin one framework, we aim to inspire the futureresearch agenda needed to develop the full pre-dictive potential ofmechanisticmodels. Consistentwith the Intergovernmental Panel on Climate

RESEARCH

SCIENCE sciencemag.org 9 SEPTEMBER 2016 • VOL 353 ISSUE 6304 aad8466-1

1Institute of Biological Risk, Ecology and EvolutionaryBiology, University of Connecticut, Storrs, CT, USA. 2Instituteof Biological and Environmental Sciences, University ofAberdeen, Aberdeen, UK. 3Redpath Museum, Department ofBiology, McGill University, Montreal, Canada. 4SorbonneUniversités, UPMC Université Paris 06, Muséum Nationald’Histoire Naturelle, CNRS, CESCO, UMR 7204, Paris,France. 5Conservation Biology, UFZ-Helmholtz Centre forEnvironmental Research, Leipzig, Germany. 6German Centrefor Integrative Biodiversity Research (iDiv) Halle-Jena-Leipzig, Leipzig, Germany. 7Ecological Modelling, UFZ-Helmholtz Centre for Environmental Research, Leipzig,Germany. 8Swedish University of Agricultural Sciences,Swedish Species Information Centre, Uppsala, Sweden.9School of Biological Sciences, University of Bristol, Bristol,UK. 10NOAA Fisheries Northwest Fisheries Science Center,Seattle, WA, USA. 11Laboratory of Aquatic Ecology, Evolutionand Conservation, KU Leuven, Leuven, Belgium. 12Bio-Protection Research Centre, Lincoln University, Lincoln, NewZealand. 13Biology, McGill University, Montreal, Canada.14Institute on the Environment; Ecology, Evolution, andBehavior, University of Minnesota, St. Paul, MN, USA.15Biology, University of Florida, Gainesville, FL, USA.16Institute for Environmental Systems Research, Departmentof Mathematics/Computer Science, University of Osnabrück,Osnabrück, Germany. 17Ecologie Systématique Evolution,University Paris-Sud, CNRS, AgroParisTech, Université Paris-Saclay, Orsay, France. 18DIVERSITAS, Paris, France. 19Centred’Ecologie fonctionnelle et Evolutive, UMR 5175 CNRS-Université de Montpellier-EPHE, Montpellier Cedex, France.20Forestry and Natural Resources, Purdue University, WestLafayette, IN, USA.*Corresponding author. Email: mark.urban@uconn.edu

No mechanism

1 or more mechanisms

Dispersal

Demography

Physiology

Species interactions

Population interactions

Adaptive potential

0 10 20 30 40 50 60 70 80

Percent of total studies

Fig. 1. Most models of biological responses to cli-mate change omit important biological mecha-nisms.Only 23% of reviewed studies (4) includeda biological mechanism. Models that included onemechanism usually incorporated others, but nomodel included all six mechanisms. All models in-cluded environmental variation, generally via cor-relations, but usually did not explicitly incorporatespecies’ sensitivities to environmental variation atrelevant spatiotemporal scales.

on

Sept

embe

r 9,

201

6ht

tp://

scie

nce.

scie

ncem

ag.o

rg/

Dow

nloa

ded

from

Change (IPCC), we use “projection” to define alldescriptions of the future and reserve “forecast”for the most likely projections.

Crucial biological information

In Table 1, we identify six mechanisms thatdetermine biological responses to climate change.On the basis of these mechanisms, we assess dataavailability for four well-studied species (Fig. 3).We find that although information on the six keymechanisms partly exists for species with higheconomic value, it is incomplete for even the best-studied species and absent for the vast majorityof Earth’s species. Consequently, themost realisticmodels usually rely on sparse data or data ex-trapolated from nonrepresentative populations,environments, or species.We next describe each mechanism in further

detail, highlighting key parameters and discussingchallenges of measurement, uncertainty, and sen-

sitivity. Here, uncertainty encompasses bothlimited knowledge and random outcomes; sen-sitivity denotes how changes in a parameter valueinfluencemodel outcomes. After describing thesemechanisms, we recommend how to collect dataefficiently and leverage imperfect data.

Physiology

Physiologymediates how climate conditions suchas temperature, growing degree-days, water avail-ability, and potential evapotranspiration influencesurvival, growth, development, movement, andreproduction (18, 28–30). Physiological parametersinclude critical thermal minima or maxima (thelow and high temperatures at which organismscease organized movement), evaporative waterloss, photosynthetic rate, andmetabolic rate. Theseindividual physiological responses often are usedto informhigher-level processes such as populationpersistence and range shifts (29, 31). For example,

knowledge of the proportion of time a lizardremains active outside its burrow, where it isthermally neutral, can help in predicting its ex-tinction risk under future climates (32).Physiologistsmeasure parameters fromnatural

observations or experiments in climate-controlledchambers (28). However, using natural observa-tions risks confounding responses to climatewithother environmental factors (28). High-prioritytraits include responses to extremeheat or dryness,where survival often declines steeply. Uncertaintyabout physiological responses increases when welack information on habitat heterogeneity, localadaptation, and physiological impacts on overallfitness.

Demography, life history, and phenology

Demographic (birth, death, migration), life his-tory (schedule of life cycle events), and phenological(timing of life history events) traits play critical

aad8466-2 9 SEPTEMBER 2016 • VOL 353 ISSUE 6304 sciencemag.org SCIENCE

Cold ColdHot

Evolution

Dispersal Dispersal

Cold

Hot

Cold ColdHot

Species interactionsEvolution

EnvironmentD

ispe

rsal

Demography Physiology

Additive genetic variation + gene flow

Spatial climate gradient

+ climate change

Temperature-dependentpopulation growth

Logistic populationgrowth

Gaussian dispersalfunction

Competition

A B

C D

Model components Model equations

Change in population (N)

Populationdynamics

Dispersal

Fitness

max exp

Temperature-dependent growth Competition

Change in trait (z)

Directional selection

Gene flow

Model spatiotemporal dynamics Model results

Tim

e

Space

Spec

ies

abun

danc

es

Space

Fig. 2. A generic model integrates six biological mechanisms to predictclimate change responses. (A to C) The six mechanisms (A) are matchedby color to their representation in equations (B) simplified from (11) (see tableS1 for symbol descriptions). Results suggest how dispersal (blue-purple),adaptive evolution (yellow), and their combination (red-orange) determine thematch between community-wide thermal traits and changing local temper-

atures (C). Temperatures increase before stabilizing at the white dashed line.Black indicates no trait change. In cold regions,warm-adapted species disperseinto newly suitable, warmer habitats. In warm regions, evolution dominatesbecause no species with higher thermal tolerances exist. (D) Equilibriumabundances of five hypothetical species (each indicated by differently coloredlines) after climate change.

RESEARCH | REVIEW

on

Sept

embe

r 9,

201

6ht

tp://

scie

nce.

scie

ncem

ag.o

rg/

Dow

nloa

ded

from

roles in climate change responses (29, 33, 34).Important parameters include birth and deathrates, age at maturity, development rate, and re-productive investment. Parameters are best collectedonmarked individuals across representative popu-lations spanning different densities and climates.

However, these efforts require long-term, costlycommitments. Changes in population abundancesfrom short-term weather variation can provideproxies, but these become unreliable over time.Long-term vegetation plots can provide detaileddemographic information for plants. Citizen scien-

tists can collect data over large regions, on traitssuch as flowering time or breeding date, butconcerns about data quality likely limit their use-fulness for less easily measured traits such asgenetic variation.Certain demographic parameters are especially

important. For example, adult survival often affectspopulation growth rate more than fecunditydoes in long-lived species (35). Density depen-dence and generation length also strongly affectextinction risk from climate change (22). Addi-tional uncertainty stems from local adaptation,responses to novel environments, mismatchedphenology, community shifts, and interactionswith nonclimate stressors (15, 36, 37).

Evolutionary potential andlocal adaptation

Assaying genetic variation is crucial for predictingfuture responses (27, 38) because it could allowpopulations to adapt to climate change in situ.Unfortunately, scientists seldom know if, or howquickly, populations can evolve climate-sensitivetraits (36). Moreover, species usually comprisemany locally adapted populations that each re-spond differently to climate change (39). Speciesmight not shift their ranges with climate changeif locally adapted populations become isolatedand cannot colonize new habitats (39). Alterna-tively, individuals dispersing from locally adaptedpopulations might track optimal climates acrosslandscapes, and thus might not need to adaptlocally (Fig. 2) (11).The breeder’s and Price equations can be used

to predict responses to natural selection basedon selection strength and genetic (co)variances(40). Genetic (co)variances are commonly mea-sured through controlled breeding experimentsor pedigrees. However, these estimates can be-come unreliable over long time scales or in novelenvironments if selection regimes or adaptivepotential change (41). Also, genetic (co)variancesoften vary among populations and environments,thus requiring broad sampling and careful sen-sitivity analyses.Other approaches involve trackingevolution using long-term observations, recon-structing evolution from layered propagule banks,or applying experimental evolution (42, 43). Forinstance, comparingBrassica rapa plants grownfrom seeds collected before and after a droughtrevealed rapid evolution of flowering time (43).Past local adaptations to spatial climatic gra-dients are easier to assess. However, these pat-terns suggest past adaptive potential, not futureevolutionary rates (36). By scanning entire ge-nomes, next-generation sequencing offers a pro-mising tool to uncover fine-scale evolutionarydiversification (44), and declining cost for genomicscould rapidly expand our limited knowledge ofadaptive potential. Other frequently applied ap-proaches include common garden experiments,natural transplants, and observations of pheno-typic variation (Table 1).Adaptive potential and population differen-

tiation represent high-priority parameters becauseignoring them contributes high levels of uncer-tainty (12, 27, 36, 43). For example, the Quino

SCIENCE sciencemag.org 9 SEPTEMBER 2016 • VOL 353 ISSUE 6304 aad8466-3

Fence lizard (Sceloporus undulatus) Sockeye salmon (Oncorhynchus nerka)

Speckled wood butterfly (Pararge aegeria)

High

Medium

Low

PhysiologyDemography& life history

Disp

ersa

l

EvolutionSp. interactions

Envi

ronm

ent

European beech (Fagus sylvatica)

A B

C D

Data quality

Fig. 3. Data gaps exist even for well-studied species. We rated data quality for some of the best-studied species in climate change research: (A) fence lizard, (B) sockeye salmon, (C) speckled woodbutterfly, and (D) European beech. Data quality: high = near-complete information; medium = informationavailable but missing critical components; low = information mostly absent.We evaluated data availabilityby examining models of climate responses, reviewing species-specific literature, and contacting experts.

RESEARCH | REVIEW

on

Sept

embe

r 9,

201

6ht

tp://

scie

nce.

scie

ncem

ag.o

rg/

Dow

nloa

ded

from

checkerspot butterfly was expected to becomeextinct from climate change, but it persists afteradapting to live on a new host plant (45). Givenlimited genetic and evolutionary information,we will often need to generalize adaptive ratesacross species based on characteristics such asgeneration time, genetic isolation, phenotypicvariation, and phylogenetic position. Fortunately,even coarse estimates of maximum adaptiverate compared to climate change suggest tippingpoints, where minor changes in climate initiatemajor biological disruptions and thus representtargets for facilitating adaptation in threatenedpopulations (46).

Species interactions

Species interactions often underlie unexpectedresponses to climate change (10, 15), and mostextinctions attributed to climate change to datehave involved altered species interactions (47).Surprises occur when specialist interactions suchasmutualism constrain species’ responses (48),phenologicalmismatches alter species interactions(37), or top consumers propagate climate changeeffects throughout food webs (8). For instance,high temperatures along the Pacific Coast exa-cerbated predation by sea stars onmussels, whichcaused local extirpations (49). Yet few modelsaccount for species interactions explicitly, insteadassuming that each species responds independentlyto climate change (6, 15) (Fig. 1).High-quality information on species interactions

requires well-resolved information about inter-acting species, interaction types and strengths,spatiotemporal variation, and phenology. Un-fortunately, such detailed information is usuallymissing. One approach to overcome this deficitis to analyze important subsets of strongly inter-acting species (15). Less robust alternatives in-clude estimating trophic position using isotopes,understanding competition via diet breadth orspecies co-occurrence patterns, extrapolating fromcorrelations between body size and trophic level,or discerning species co-occurrence patterns frommetagenomics. High-priority parameters includethose characterizing specialist interactions, top-down foodweb interactions, and timingmismatchesamong interacting species.High uncertainty arisesfrom changes in species interactions themselves(e.g., shifts from competition to facilitation) andcomplex indirect effects that propagate throughfoodwebs (9). Additional uncertainties arise fromspecies’differential abilities to track climate changein space, creating previously unseen communitiesas coevolved interactions disappear and novelinteractions form (10).

Dispersal, colonization,and range dynamics

To persist, species often must track suitable cli-mates into new regions through dispersal, colo-nization, and subsequent range shifts (50, 51).Mostmodels unrealistically assume that all orga-nisms disperse comparably and across any land-scape (Fig. 1) (26). In reality, dispersal dependson the interplay among individual behavior, fit-ness, habitat quality, and landscape configuration

(52). Range shifts are particularly sensitive todynamics at range boundaries where low abun-dances challenge accurate estimation (53).Global positioning system units can record

fine-scaled individual movement but are costlyand unsuitable formany small organisms. Passiveintegrated transponders, acoustic tags, and telem-etry devices track smaller individuals at lowercost, but these require strategically placed recorders.

Neutral genetic variation across landscapes canindicate movement patterns, but demographichistory can confound these estimates. Citizen sciencesometimes enables cost-effective, coordinated, andlarge-scale data collection, assuming adequatequality control. Dispersal distances also can beinferred from proxies [e.g., body size–dispersalrelationships in animals (51) and growth form,seedmass, and vegetation type in plants (54)] until

aad8466-4 9 SEPTEMBER 2016 • VOL 353 ISSUE 6304 sciencemag.org SCIENCE

Revise model to includenew parameters or different relationships

Improve estimatesfor sensitive parameters

Target sensitive parameters

Define model

Popu

latio

n

Estimate parameter sensitivities

Use available parameters

Obtainprojections frompreliminarymodel

Projections improve across iterations

0 100

0 20

05

10

Temperature

Gro

wth

rate

1

2

dN

dt= rN N 2

Validate with monitoring data

Define model

0 100

010

YearYear

20

010

20

Fig. 4. Biological models improve iteratively through time by applying an adaptive modelingscheme. Steps include parameterizing models using available data, estimating parameter sensitivities,targeting better measurements for sensitive parameters, validating projections with observations, anditeratively refining and updating the model to improve predictive accuracy and precision through time.

RESEARCH | REVIEW

on

Sept

embe

r 9,

201

6ht

tp://

scie

nce.

scie

ncem

ag.o

rg/

Dow

nloa

ded

from

SCIENCE sciencemag.org 9 SEPTEMBER 2016 • VOL 353 ISSUE 6304 aad8466-5

Table

1.Biological

param

eters,

colle

ctionmethods,

proxies

,priorities

,andke

yunce

rtainties

.We

listsixclas

sesof

biologica

lmec

hanism

s,ex

ample

param

eters,

metho

dsto

colle

ctthem

,pos

sible

proxy

relation

shipsthat

could

fillin

gapsforpoo

rlystud

ied

taxa

,priorityparam

eters,

and

key

remaining

unce

rtainties.

Welistmetho

dsin

anillus

trativedes

cend

ingorder

ofdataac

curacy

.The

orderingof

colle

ctionmetho

dsisillus

trativeon

lyan

dwillclea

rlych

ange

dep

endingon

theparticu

lar

attributes

ofsp

eciesan

dsystem

s.The

bes

tmetho

ds,

howev

er,m

ight

notbeea

sily

implemen

tedfor

sometaxa

,ne

cessitating

morepractical

metho

dsfollo

wed

byse

nsitivityan

alys

is.The

ywill

also

chan

gethroug

htime—

forex

ample,as

emerging

metho

dsbec

omeless

costly.In

reality,

theidea

l

approac

hforco

llectingdataon

ake

yproce

sswill

invo

lvejointus

eof

more

than

onemethod

.For

exam

ple,for

dispersa

lwemight

curren

tlywan

tto

colle

cthigh

-qua

litytelemetry

dataforthemov

e-men

tof

arelative

lysm

allnu

mber

ofdisperse

rsbec

ause

ofco

stco

nstraintswhile

also

obtaining

population

-lev

eles

timates

ofdispersa

lthrough

either

land

scap

ege

netics

ormark-releas

e-

reca

pture

method

s(orboth).Ween

courag

eread

ersto

tailo

rco

stsan

dben

efitsof

thealternative

and

complemen

tary

approac

hes

totheirow

nsy

stem

and

adjust

dec

isions

forinve

stmen

tof

reso

urces

appropriately.

Note

that

each

ofthes

emec

hanism

slik

ely

interacts

with

othe

r

mec

han

isms.

Biologica

l

mec

hanism

s

Exam

ple

parameters

Alterna

tive

and

complem

entary

metho

dsProxy

relation

ships

Priority

parameters

Key

unce

rtainties

Phy

siolog

yThe

rmal,d

esicca

tion

,

andch

emical

toleranc

es;

environm

ent-dep

enden

t

perform

ance

and

metab

olic

rate;

pho

tosy

nthe

sis

•Exp

erim

entalu

nderstan

ding

ofph

ysiologica

l

resp

onse

sto

environm

entalc

onditio

ns

inna

ture

orthelabo

ratory

•Obs

ervedco

rrelations

betw

eenph

ysiologica

l

resp

onse

san

den

vironm

entalc

onditio

nsin

timeor

spac

e

•Trait-ba

sedprox

ies(e.g.,bo

dymass

formetab

olism)

•Bod

ymas

sco

rrelates

strong

lywithen

ergy

requiremen

ts

•Water

andlig

ht

requiremen

tsin

vege

tation

mod

els

Phy

siolog

ical

resp

onse

s

inex

trem

een

vironmen

ts

(e.g.,perform

ance

under

hotor

dry

conditions)

•How

does

beh

avior

modify

phys

iology

?

•To

what

deg

reedoorgan

isms

evolvedifferen

tphys

iologica

l

resp

onse

sac

ross

arange

?

•How

dophys

iologica

l

sensitivities

ofdifferen

t

perform

ance

traits

scale

towhole-organ

ism

fitnes

s?............................................................................................................................................................................................................................................................................................................................................................................................................................................

Dem

ography

,

lifehistory,

andphe

nology

Birth

anddea

thrates,

includ

ingag

eor

stag

e

structure,

ageof

maturity,

dev

elop

men

t

andgrow

thrates,

environm

ental

dep

enden

ce,tim

ing,

andindividua

lvariability

•Lo

ng-term

markreca

pture

paren

tage

stud

iesor

long

-term

dem

ographicdata

from

vege

tation

plots

•Exp

erim

entals

tudiesof

environmen

t-dep

enden

tbirth

anddea

th

ratesin

nature

(bes

t)or

thelaboratory

•Pop

ulationgrow

thratesfrom

obse

rved

abun

dan

cedata

Dem

ographicparam

eters

correlatewithlife

historytraits

(e.g.,slow

-fas

tco

ntinuu

m)

andnich

esp

ecialization

Vital

ratesthat

mos

tinflu

ence

pop

ulationgrow

thrates

(e.g.,ad

ultsu

rvival

for

long

-lived

organ

isms,

gene

rationlength,

mismatch

esin

timing

oflifehistory

even

ts)

•To

wha

tde

gree

doorganism

sevolve

diffe

rent

lifehistoriesacross

arang

e?

•Doe

srapidad

aptatio

nto

clim

atech

ange

play

arole?

•Whe

ndo

esph

enolog

yde

pend

onclim

ateversus

nonc

limate

trigge

rs(e.g.,da

yleng

th)?

•How

doothe

ren

vironm

ental

chan

ges(e.g.,ha

bitatde

grad

ation)

interact

with

clim

ateresp

onses?

............................................................................................................................................................................................................................................................................................................................................................................................................................................

Evolutiona

ry

poten

tial

andloca

l

adap

tation

Additivege

netictrait

(co)va

rian

ce/h

eritab

ility

andad

ditivege

netic

cova

rian

cebetwee

n

traits

andfitne

ss

•Qua

ntitativege

neticva

riationin

keytraits

estimated

from

controlle

dbreed

ingdes

igns

,

from

pop

ulations

withped

igrees

,orfrom

individualsraised

under

common

cond

itions

•Exp

erim

entalo

rco

rrelationa

lestim

ationof

selectiongrad

ients

•Gen

eex

press

ionpatternsforun

derstan

ding

func

tiona

ltraitva

riationun

der

differen

t

environmen

talc

onditions

•Phe

notypic

variationwithinpop

ulations

•Ev

olutiona

ryrates

correlatene

gative

ly

withge

neration

leng

th

•Gen

etic

variation

withinpop

ulations

pos

itivelyco

rrelates

withpop

ulationsize

•Spac

e-for-timesu

bstitutions

Adap

tive

potential,

loca

ladap

tationof

clim

ate-se

nsitive

param

etersac

ross

spec

iesrange

•To

what

deg

reeis

traitch

ange

determined

byge

netics

versusen

vironmen

t?

•How

welld

osh

ort-term

mea

suremen

tsof

adap

tive

mec

han

isms

perform

inthelongrun?

•How

does

loca

ladap

tation

within

arange

altersp

ecies-leve

l

resp

onse

sto

clim

atech

ange

?............................................................................................................................................................................................................................................................................................................................................................................................................................................

Fitnes

sdifferen

ces

amon

gpop

ulations

anden

vironm

ents;

gene

ticva

riationam

ong

pop

ulations

;phe

notypic

variation,

includ

ing

plastic

ity,a

mon

gpo

pulatio

ns

•Rec

iproca

ltrans

plant

andco

mmon

garden

experim

ents

that

reve

alfitne

ss

andtraitdifferen

cesam

ong

pop

ulations

inresp

onse

toreleva

nt

environm

entalg

radients

•Statistical

search

forva

riationin

loci

under

selection

•Gen

etic

variationam

ong

pop

ulations

pos

itively

correlates

withrang

esize

............................................................................................................................................................................................................................................................................................................................................................................................................................................

Con

tinu

edon

thene

xtpag

e

RESEARCH | REVIEW

on

Sept

embe

r 9,

201

6ht

tp://

scie

nce.

scie

ncem

ag.o

rg/

Dow

nloa

ded

from

aad8466-6 9 SEPTEMBER 2016 • VOL 353 ISSUE 6304 sciencemag.org SCIENCE

Biologica

l

mec

hanism

s

Exam

ple

parameters

Alterna

tive

and

complem

entary

metho

dsProxy

relation

ships

Priority

parameters

Key

unce

rtainties

............................................................................................................................................................................................................................................................................................................................................................................................................................................

•Gen

eex

press

ionpatterns

forun

derstan

dingfunc

tion

altrait

variationun

der

differen

ten

vironm

ental

cond

itions

•Pop

ulationge

netics

withne

utralloc

ito

understan

dpop

ulationdifferen

tiation

throug

hbarriersto

gene

flow

•Obse

rvationof

phe

notypic

variationwithin

andam

ongpop

ulations

............................................................................................................................................................................................................................................................................................................................................................................................................................................

Spe

cies

interactions

Interactionweb

s

withsp

atiotemporal

variationan

dphe

nology

,

interactiontypes

and

streng

ths,

commun

ity

mod

ule,

dietor

reso

urce

overlap,

trop

hic

pos

ition

•Exp

erim

entale

valuationof

spec

ies

interactionstreng

than

ddirec

tion

inna

ture

(bes

t)or

thelaboratory

•Natural

historyob

servations

ofinteractions

•Isotop

ean

alys

isto

reve

altrop

hic

leve

lsan

dfood

web

links

•Statistical

co-occ

urrenc

e

patterns(e.g.,ch

ecke

rboa

rd

patternsforco

mpetition)

•Trop

hicleve

linc

reas

es

withbod

ysize

•Sim

ilartrop

hicleve

ls

aresh

ared

by

phy

loge

netica

lly

simila

rsp

ecies

Spec

ialistinteractions,

sens

itivityoftop

cons

umers,

phe

nologica

l

mismatch

esbetwee

n

interactingsp

ecies

•What

hap

pen

sas

coev

olved

interactionsdisap

pea

ran

dnew

spec

iesinteractionsform

?

•How

sensitive

arefoodweb

sto

top-dow

nve

rsusbottom-up

clim

atedisturban

ces?

•To

what

deg

reeca

nsp

ecies

adap

tto

nov

elsp

ecies

interactions?

............................................................................................................................................................................................................................................................................................................................................................................................................................................

Dispe

rsal,

colonizatio

n,

andrang

e

dyna

mics

Dispersa

lbeh

aviors;

mov

emen

tan

d

settlemen

trules;

interind

ividua

l

variab

ility;e

nviron

men

t-,

den

sity-,an

dco

ndition-

dep

enden

tdispersa

l;

land

scap

epermea

bility

(e.g.,leas

t-co

stpath

analys

is)

•Satellitetelemetry

ofmov

ingorga

nism

s

toreve

alland

scap

emov

emen

ttrac

ks

•Mark-reca

pture

and

reloca

tion

sto

evalua

teab

solute

mov

emen

t

•Exp

erim

ents

(e.g.,lin

ked

mes

ocos

ms)

toun

derstan

dmov

emen

t

•La

ndscap

ege

netic

sto

reveal

land

scap

eco

nnec

tivity

amon

gpo

pulatio

ns

•Historica

lrec

onstructionof

movem

ent

patterns

during

expa

nsion

•Incide

ncefunc

tions

inmetap

opulations

tode

term

inepo

pulatio

nco

nnec

tivity

•Citizenscienc

eto

trac

k

orga

nism

s(e.g.,tagg

edbirds)

•La

rger-bod

iedan

imals

disperse

farthe

r

•Smallerse

edstrav

elfarthe

r

•Animal-disperse

dse

eds

trav

elfarthe

r

•La

rger-w

inge

dorga

nism

s

disperse

farthe

r

•Pelag

ican

imalsdisperse

farthe

rthan

ben

thic

ones

Long

-distance

dispersa

l,fitnes

sat

rang

eboundaries

•How

importan

tis

long-range

dispersa

lforrange

dyn

amics?

•How

does

fitnes

sva

ry

across

arange

?

............................................................................................................................................................................................................................................................................................................................................................................................................................................

Respo

nses

to

environm

ental

varia

tion

Func

tion

alrelation

ships

betwee

ntraits

and

environm

ents;

iden

tific

ationan

d

qua

ntifica

tion

ofke

y

environm

entalg

radients

across

spec

ies-releva

nt

scales

ofsp

ace

andtime

•Exp

erim

entalm

anipulationof

keyen

vironm

ents

toun

derstan

dfunc

tion

alresp

onse

s

•Statistical

analys

isof

environm

entalg

radients

andresp

onse

s

•Cha

racterizationof

environm

entalg

radientsat

biologica

llyreleva

ntsc

ales

♦Surve

ysof

environm

entalp

aram

eters

cond

uctedat

releva

ntsp

atial

andtemporal

scales

♦Groun

d-truthed

map

sto

beus

edin

environm

entalg

radient

analys

es

♦Statistical

interpolationof

coarse

map

data

•Determiningne

tworks

ofco

-acting

environm

ental

variab

les

•Correlating

easily

colle

cted

remotely

sens

eddatato

othe

rfactors

such

asreso

urce

s

Iden

tifyingke

ygrad

ients,

spatials

cale

dep

enden

ce

ofen

vironmen

tal

resp

onse

s,dyn

amic

chan

gein

grad

ients

•Are

therege

neral

way

sto

predict

thereleva

ntsc

ales

atwhich

differen

tsp

ecieswill

resp

ond

toen

vironmen

talv

ariation?

•What

biologica

lparam

etersare

linke

dwiththeen

vironmen

tal

factors

andhow?

•How

areim

portan

ten

vironmen

tal

grad

ients

chan

gingthrough

time?

............................................................................................................................................................................................................................................................................................................................................................................................................................................

RESEARCH | REVIEW

on

Sept

embe

r 9,

201

6ht

tp://

scie

nce.

scie

ncem

ag.o

rg/

Dow

nloa

ded

from

better estimates become available. Long-distancedispersal and fitness at range edges are high-priority parameters because they introduce highuncertainty in model outcomes (26), yet are dif-ficult to measure.

Responses to environmental variation

Responses to climate change depend on species-specific sensitivities and exposures to climate andhabitat variation at relevant spatiotemporal scales.For instance, birds respond idiosyncratically todifferent climate variables, depending on theirindividual sensitivities to temperature andprecipi-tation change (55). Researchers must carefullyidentify which specific climate components actu-ally affect species. Many organisms respond notto average annual temperature or precipitation,but rather to temperature thresholds, season length,humidity, potential evapotranspiration, or extremeevents such as droughts. Species also differ in therelevant spatiotemporal scales of environmentalvariation. Researchers should evaluate the envi-ronment through the eyes of the organism. Thescales relevant to focal organisms often aremetersand minutes rather than the measurements inkilometers andmonths typically available. Despitethe increasing availability of fine-scaled informa-tion, most predictions are still made at coarsescales, which can substantially reduce predictiveaccuracy (56). Hierarchical sampling canmaximizeinformation content by combining large-scalesampling with targeted fine-scale measurementsthat capture relevant gradients. Species charac-teristics such as body size or generation lengthalso can provide proxies for missing data on spe-cies’ environmental responses.In addition, we need to integrate predictions

of climate change with other human disturbances,including land use, pollution, invasive species, andharvesting, to gauge the full extent of future envi-ronmental change. Improving predictions of thesedisturbances [e.g., (57)] and downscaling data torelevant ecological resolutions are critical forreducing future uncertainty.

Interacting mechanisms

Eachmechanismpotentially interacts withmanyothers. Specifically, climate responses dependprox-imately on dispersal and demography; demog-raphy in turn depends on physiology, speciesinteractions, and environments; and each traitcan evolve. For example, great tit birds in theNeth-erlands do not lay eggs earlier in warmer springs(involving demography, phenology, and environ-mental responses), whereas their caterpillar prey(species interaction) emerge earlier. This pheno-logical mismatch between birds and their preydecreases nestling fitness (demography) (37). Yetgreat tits from the United Kingdom do breedearlier in warmer springs, suggesting populationgenetic differentiation (58). A challenge is to inte-gratemultiple interactingmechanismswithout un-necessarily increasing model complexity (Fig. 2).

A practical way forward

We recognize that the complexity of naturalsystems will add uncertainty to even the best-

parameterized and most realistic models (59).Collecting the relevant information and devel-oping realistic biological models will requiresubstantial investment in time and resources.Despite these challenges, we believe that collect-ing mechanistic data will both enhance ourfundamental understanding of the biological pro-cesses that underlie climate responses and con-tribute tomore accurate, longer-term projectionsthat facilitate more effective conservation. Mech-anistic models might not make accurate pre-dictions initially, but learning from those failuresprovides the insights that ultimately improveprojections. Predictive science advances mostquicklyvia iterativeprediction-failure-improvementcycles, and mechanistically grounded models oftenquicken the pace of these advances (2, 3, 19). Evensmall gains in understanding can improve futuremodels by indicating criticalmissing information,highlighting key uncertainties, suggesting generaltrait-based predictions for nonmodeled organisms,and delimiting the best options for retaining bio-diversity under a range of future policy scenarios.Given limited time and resources, however, we

need to develop strategies that leverage existingdata and target essential information. Towardthis end, we advocate for an adaptive modelingscheme that facilitates cost-effective model devel-opment anddata collection (Fig. 4). The process ofmodel testing and revision—steps rarely taken to-day, but facilitatedbyamore systematic approach—can reveal data of particular importance for im-provingpredictions.Researchers first parameterizemodelswith available data. In Table 1, we demon-strate how to tailor data collection efforts to system-specific constraints by listing ideal methods alongwith more easily collected proxies. Researchersthen use independently collected variables frommonitoring efforts to test outcomes and fit un-certain relationships. Sensitivity analyses identifythemost important parameters to collect, ensuringthat resources go toward producing the greatestgains in accuracy. On the basis of these analyses,researchers can collect improved or new param-eter estimates and revise the model throughsuccessive iterations of the approach. Crucially,results frommultiple independentmodels shouldbe combined because ensemble forecasts oftenprove more accurate (3, 60). Researchers alsoneed to articulate clearly how uncertainty in pa-rameter estimates and model choice propagatesat each modeling step. We recommend adoptingthe IPCC’s standards (1) for classifyingmodel con-fidence and probabilistic uncertainty.Several approaches are available to extend pro-

jections from a few carefully studied species tomany unstudied ones.We often possess extensiveinformation that is spread across many speciesbut is incomplete foranyparticularspecies.Emergingphylogenetic and trait-based approaches couldfill these data gaps. Trait-based approaches usetrait correlations (e.g., between adult survival andfecundity) to predict missing parameters forspecies (50). Researchers also can simulate theclimate responses of virtual species with realisticcombinations of traits. For example, this virtualapproach predicted that 30% of terrestrial mam-

mals might not keep pace with climate change(61). Minimally, these efforts provide qualitativeinsights about which types of species are mostvulnerable to climate change and therefore shouldbe targeted for future, in-depth study (22). Anothercost-effective strategy is to prioritize research onspecies with both high climate sensitivity and dis-proportionately large impacts on ecosystems. Theseso-called biotic multipliers—often, top predatorsandother keystone species—amplify small changesin climate to produce large ecological effects (8)such that their future dynamics drive overall eco-system changes (9).Conservation sometimes focuses on overall bio-

diversity rather than focal species. Estimates fromsubsets of species might be cautiously extrapo-lated to overall biodiversity, assuming suitablerepresentation across taxonomic and phylogeneticdiversity. However, trait-based approaches mightmore efficiently suggest species that have vulner-able trait combinations or amplify community-wide impacts of climate change. For example,focusing on top consumers and other keystonespecies can indicate how their responses rever-berate through entire food webs (8), thus furtherextending the value of single-species forecasts.Lastly, hybrid correlative-mechanistic approaches

offer a pragmatic initial approach to improvingpredictions by adding key mechanisms to simplemodels. For example, adjusting predicted rangesfrom correlativemodels with species-specific dis-persal abilities (62) or interacting species’ ranges(48) can add realism and improve predictions.Given the simplicity of most current approaches(Fig. 1), even minimally more realistic modelsmight improve projections untilmore complicatedmodels can be developed (13, 19).

Global coordination

Global coordination will be critical at all stages,including defining projection goals, developingbettermodels, collating and incorporating existingdata, determining which additional data mightimprove forecasts, collecting newdata,monitoringbiodiversity changes, and organizing and main-taining data. Researchers and policymakers firstmust agree on the nature of the projection itself,including the accuracy, coverage, and timehorizonof forecasts. A global clearinghouse would beuseful to organize trait data, standardize termi-nology (e.g., dispersal versus migration), andmonitor climate responses.It would also be useful to form regionalworking

groupswith local experts. Regionalworking groupswould define representative ecosystems and cli-matic and environmental gradients in their region,while taking advantage of existing data and long-termmonitoring sites. Groupswould select speciesrepresenting abroad rangeof regional trait diversityand build initial models with available data toestimate parameter sensitivity. To address imme-diate extinction threats, regional working groupsmight also characterize the climate change riskfor threatened species on the International UnionforConservationofNatureRedList. Groups shouldthendevelop plans to refine sensitive parametersthrough targeted fundingopportunities and citizen

SCIENCE sciencemag.org 9 SEPTEMBER 2016 • VOL 353 ISSUE 6304 aad8466-7

RESEARCH | REVIEW

on

Sept

embe

r 9,

201

6ht

tp://

scie

nce.

scie

ncem

ag.o

rg/

Dow

nloa

ded

from

science. Collected biological informationmust beaccessible, quality-checked, standardized, andmain-tained in databases such as Encyclopedia of Life’sTraitBank (traits) and Global Biodiversity Infor-mation Facility (species occurrences).The IPCC’s development of climate change

predictions provides a template for how to achievecomparable progress in biodiversity projections.The IPCC’s biodiversity analog, the Intergovern-mental Platform on Biodiversity and EcosystemServices, can also help to coordinate this effort.Already, the Group on Earth Observations–Biodiversity Observation Network is developinga list of essential biodiversity variables (EBVs) formonitoring global biodiversity (25) and isworkingto address monitoring gaps (19). Despite someoverlap between the modeling parameters out-lined here and EBVs, the two collection schemeshave divergent objectives. The EBVs monitorchanges in biodiversity and provide variables forinitializing and testing mechanistic predictions.Mechanisticmodels, however, also require param-eters governing key processes, which often man-date more detailed observations or experimentsthan monitoring programs currently entail.

Combining predictive modeling withrobust scenario analysis

Collecting the data necessary to informmechanisticbiologicalmodels presents an enormous challengegiven the vast diversity of life, its complexity, andour inadequate knowledge about it. This inherentcomplexity and stochasticity limits the accuracy ofbiological predictions for policy and management(59, 63), especially over long forecast horizons (3).Wemust accept that even thebest-informedpredic-tions could fail for a variety ofunanticipated reasons.An alternative approach to planning for climate

change develops conservation strategies robust toa broad range of future scenarios (64), thus in-suring against inevitable surprises. For example,applying this “robust scenario” approach mightinclude maintaining dispersal corridors, pre-serving existing natural habitat and geneticdiversity, and facilitating monitoring and flex-ible, adaptive management (59, 65). This strat-egy broadly protects biodiversity and dependsless on accurate predictions. However, practicalconsiderations will often limit the number of op-tions that are feasible, especially whenmanage-ment options for one species trade off againstanother.The two approaches are notmutually exclusive,

and we believe that they work best in tandem.Mechanistic approaches likely will improve pre-dictions at intermediate time horizons (e.g., 25 to50 years), when current environmental correla-tions break down and correlative approachesbecome less accurate (3). Beyond this time frame,even the best mechanistic models become un-certain as keyparameters can shift anduncertaintypropagates. Yet predictive models are still neededto delimit plausible expectations, place boundson uncertainty, and direct limited resources towardstrategies that target themost threatened regionsand species (23, 59). Hence, a tandem approachbuilds general insights from key representative

species while preserving flexible options thatwork when models fail.

Conclusions

Climate scientists in 1975 acknowledged theirinability to predict climate accurately and high-lighted the many challenges to reaching thisobjective (66). Despite these challenges, they out-lined an ambitious long-term research programaimedatunderstandingkeymechanismsgoverningclimate change and collecting key pieces of missinginformation. This program ultimately produced theimprovements in forecasting weather and climatechange that society benefits from today. We believethat biology can and must do the same.We advocate for a renewed global focus on

targeting the natural history information neededto predict the future of biodiversity. Such effortswould more than compensate for their cost byimproving our ability to understand, anticipate,and thereby prevent biodiversity loss and dam-age to ecosystems from climate change as well asother disturbances. Ultimately, understanding hownature works will provide innumerable benefits forlong-term sustainability and human well-being.

REFERENCES AND NOTES

1. J. Settele et al., in Climate Change 2014: Impacts, Adaptation,and Vulnerability. Fifth Assessment Report of theIntergovernmental Panel on Climate Change, C. B. Field et al.,Eds. (Cambridge Univ. Press, 2014), pp. 1–153.

2. N. Mouquet et al., Predictive ecology in a changing world.J. Appl. Ecol. 52, 1293–1310 (2015). doi: 10.1111/1365-2664.12482

3. O. L. Petchey et al., The ecological forecast horizon, andexamples of its uses and determinants. Ecol. Lett. 18, 597–611(2015). doi: 10.1111/ele.12443; pmid: 25960188

4. M. C. Urban, Accelerating extinction risk from climate change.Science 348, 571–573 (2015). doi: 10.1126/science.aaa4984;pmid: 25931559

5. A. L. Angert et al., Do species’ traits predict recent shifts atexpanding range edges? Ecol. Lett. 14, 677–689 (2011).doi: 10.1111/j.1461-0248.2011.01620.x; pmid: 21535340

6. A. J. Davis, L. S. Jenkinson, J. H. Lawton, B. Shorrocks,S. Wood, Making mistakes when predicting shifts in speciesrange in response to global warming. Nature 391, 783–786(1998). doi: 10.1038/35842; pmid: 9486646

7. S. D. Veloz et al., No-analog climates and shifting realizedniches during the late quaternary: Implications for 21st-century predictions by species distribution models. GlobalChange Biol. 18, 1698–1713 (2012). doi: 10.1111/j.1365-2486.2011.02635.x

8. P. L. Zarnetske, D. K. Skelly, M. C. Urban, Biotic multipliers ofclimate change. Science 336, 1516–1518 (2012). doi: 10.1126/science.1222732; pmid: 22723403

9. E. Post, Ecology of Climate Change: The Importance of BioticInteractions (Princeton Univ. Press, 2013).

10. M. C. Urban, J. J. Tewksbury, K. S. Sheldon, On a collisioncourse: Competition and dispersal differences create no-analogue communities and cause extinctions during climatechange. Proc. R. Soc. B 279, 2072–2080 (2012). doi: 10.1098/rspb.2011.2367; pmid: 22217718

11. J. Norberg, M. C. Urban, M. Vellend, C. A. Klausmeier,N. Loeuille, Eco-evolutionary responses of biodiversity toclimate change. Nat. Clim. Change 2, 747–751 (2012).doi: 10.1038/nclimate1588

12. G. Bocedi et al., Effects of local adaptation and interspecificcompetition on species’ responses to climate change. Ann. N.Y.Acad. Sci. 1297, 83–97 (2013). pmid: 23905876

13. D. Zurell et al., Benchmarking novel approaches for modellingspecies range dynamics. Global Change Biol. 22, 2651–2664(2016). doi: 10.1111/gcb.13251; pmid: 26872305

14. J. W. Williams, S. T. Jackson, J. E. Kutzbach, Projecteddistributions of novel and disappearing climates by 2100 AD.Proc. Natl. Acad. Sci. U.S.A. 104, 5738–5742 (2007).doi: 10.1073/pnas.0606292104; pmid: 17389402

15. S. E. Gilman, M. C. Urban, J. Tewksbury, G. W. Gilchrist,R. D. Holt, A framework for community interactions underclimate change. Trends Ecol. Evol. 25, 325–331 (2010).doi: 10.1016/j.tree.2010.03.002; pmid: 20392517

16. D. Purves et al., Ecosystems: Time to model all life on Earth.Nature 493, 295–297 (2013). pmid: 23325192

17. G. Bocedi et al., RangeShifter: A platform for modelling spatialeco-evolutionary dynamics and species’ responses toenvironmental changes. Methods Ecol. Evol. 5, 388–396(2014). doi: 10.1111/2041-210X.12162

18. M. Kearney, W. Porter, Mechanistic niche modelling: Combiningphysiological and spatial data to predict species’ ranges.Ecol. Lett. 12, 334–350 (2009). doi: 10.1111/j.1461-0248.2008.01277.x; pmid: 19292794

19. S. M. McMahon et al., Improving assessment and modelling ofclimate change impacts on global terrestrial biodiversity.Trends Ecol. Evol. 26, 249–259 (2011). doi: 10.1016/j.tree.2011.02.012; pmid: 21474198

20. J. Pagel, F. M. Schurr, Forecasting species ranges by statisticalestimation of ecological niches and spatial populationdynamics. Glob. Ecol. Biogeogr. 21, 293–304 (2012).doi: 10.1111/j.1466-8238.2011.00663.x

21. H. R. Pulliam, On the relationship between niche anddistribution. Ecol. Lett. 3, 349–361 (2000). doi: 10.1046/j.1461-0248.2000.00143.x

22. R. G. Pearson et al., Life history and spatial traits predictextinction risk due to climate change. Nat. Clim. Change 4,217–221 (2014). doi: 10.1038/nclimate2113

23. A. Singer et al., Community dynamics under environmentalchange: How can next generation mechanistic models improveprojections of species distributions? Ecol. Model. 326, 63–74(2016). doi: 10.1016/j.ecolmodel.2015.11.007

24. L. B. Buckley et al., Can mechanism inform species’distribution models? Ecol. Lett. 13, 1041–1054 (2010).pmid: 20482574

25. H. M. Pereira et al., Essential biodiversity variables. Science339, 277–278 (2013). doi: 10.1126/science.1229931;pmid: 23329036

26. M. C. Urban, P. L. Zarnetske, D. K. Skelly, Moving forward:Dispersal and species interactions determine biotic responsesto climate change. Ann. N.Y. Acad. Sci. 1297, 44–60 (2013).pmid: 23819864

27. A. A. Hoffmann, C. M. Sgrò, Climate change and evolutionaryadaptation. Nature 470, 479–485 (2011). doi: 10.1038/nature09670; pmid: 21350480

28. M. J. Angilletta, Thermal Adaptation: A Theoretical andEmpirical Synthesis (Oxford Univ. Press, Oxford, 2009).

29. L. Crozier, G. Dwyer, Combining population-dynamic andecophysiological models to predict climate-induced insectrange shifts. Am. Nat. 167, 853–866 (2006). doi: 10.1086/504848; pmid: 16685639

30. M. Kearney et al., Modelling species distributions without usingspecies distributions: The cane toad in Australia under currentand future climates. Ecography 31, 423–434 (2008).doi: 10.1111/j.0906-7590.2008.05457.x

31. M. Kearney, W. P. Porter, C. Williams, S. Ritchie,A. A. Hoffmann, Integrating biophysical models andevolutionary theory to predict climatic impacts on species’ranges: The dengue mosquito Aedes aegypti in Australia. Funct.Ecol. 23, 528–538 (2009). doi: 10.1111/j.1365-2435.2008.01538.x

32. B. Sinervo et al., Erosion of lizard diversity by climate changeand altered thermal niches. Science 328, 894–899 (2010).doi: 10.1126/science.1184695

33. D. A. Keith et al., Predicting extinction risks under climatechange: Coupling stochastic population models with dynamicbioclimatic habitat models. Biol. Lett. 4, 560–563 (2008).doi: 10.1098/rsbl.2008.0049; pmid: 18664424

34. S. Jenouvrier et al., Demographic models and IPCC climateprojections predict the decline of an emperor penguinpopulation. Proc. Natl. Acad. Sci. U.S.A. 106, 1844–1847(2009). doi: 10.1073/pnas.0806638106;pmid: 19171908

35. B.-E. Sæther, Ø. Bakke, Avian life history variation andcontribution of demographic traits to the population growthrate. Ecology 81, 642 (2000). doi: 10.2307/177366

36. J. Merilä, A. P. Hendry, Climate change, adaptation, andphenotypic plasticity: The problem and the evidence. Evol.Appl. 7, 1–14 (2014). doi: 10.1111/eva.12137

37. M. E. Visser, A. J. van Noordwijk, J. M. Tinbergen, C. M. Lessells,Warmer springs lead to mistimed reproduction in great tits(Parus major). Proc. R. Soc. B 265, 1867–1870 (1998).doi: 10.1098/rspb.1998.0514

aad8466-8 9 SEPTEMBER 2016 • VOL 353 ISSUE 6304 sciencemag.org SCIENCE

RESEARCH | REVIEW

on

Sept

embe

r 9,

201

6ht

tp://

scie

nce.

scie

ncem

ag.o

rg/

Dow

nloa

ded

from

38. S. P. Carroll et al., Applying evolutionary biology to addressglobal challenges. Science 346, 1245993 (2014). doi: 10.1126/science.1245993; pmid: 25213376

39. S. L. Pelini, J. A. Keppel, A. E. Kelley, J. J. Hellmann, Adaptationto host plants may prevent rapid insect responses to climatechange. Global Change Biol. 16, 2923 (2010). doi: 10.1111/j.1365-2486.2010.02177.x

40. M. B. Morrissey et al., The prediction of adaptive evolution:Empirical application of the secondary theorem of selectionand comparison to the breeder’s equation. Evolution 66,2399–2410 (2012). doi: 10.1111/j.1558-5646.2012.01632.x;pmid: 22834740

41. J. P. Reeve, Predicting long-term response to selection. Genet.Res. 75, 83–94 (2000). doi: 10.1017/S0016672399004140;pmid: 10740924

42. W. E. Bradshaw, C. M. Holzapfel, Evolutionary response to rapidclimate change. Science 312, 1477–1478 (2006). doi: 10.1126/science.1127000; pmid: 16763134

43. S. J. Franks, S. Sim, A. E. Weis, Rapid evolution of floweringtime by an annual plant in response to a climate fluctuation.Proc. Natl. Acad. Sci. U.S.A. 104, 1278–1282 (2007).doi: 10.1073/pnas.0608379104; pmid: 17220273

44. J. Buckley, R. K. Butlin, J. R. Bridle, Evidence for evolutionarychange associated with the recent range expansion of theBritish butterfly, Aricia agestis, in response to climate change.Mol. Ecol. 21, 267–280 (2012). doi: 10.1111/j.1365-294X.2011.05388.x; pmid: 22118243

45. C. Parmesan, A. Williams-Anderson, M. Moskwik, A. S. Mikheyev,M. C. Singer, Endangered Quino checkerspot butterfly and climatechange: Short-term success but long-term vulnerability? J. InsectConserv. 19, 185–204 (2015). doi: 10.1007/s10841-014-9743-4

46. C. A. Botero, F. J. Weissing, J. Wright, D. R. Rubenstein,Evolutionary tipping points in the capacity to adapt toenvironmental change. Proc. Natl. Acad. Sci. U.S.A. 112,184–189 (2015). doi: 10.1073/pnas.1408589111;pmid: 25422451

47. A. E. Cahill et al., How does climate change cause extinction?Proc. R. Soc. B 280, 20121890 (2013). doi: 10.1098/rspb.2012.1890; pmid: 23075836

48. O. Schweiger, J. Settele, O. Kudrna, S. Klotz, I. Kühn, Climatechange can cause spatial mismatch of trophically interactingspecies. Ecology 89, 3472–3479 (2008). doi: 10.1890/07-1748.1; pmid: 19137952

49. C. D. G. Harley, Climate change, keystone predation, andbiodiversity loss. Science 334, 1124–1127 (2011). doi: 10.1126/science.1210199; pmid: 22116885

50. C. A. Schloss, T. A. Nuñez, J. J. Lawler, Dispersal will limitability of mammals to track climate change in the WesternHemisphere. Proc. Natl. Acad. Sci. U.S.A. 109, 8606–8611(2012). doi: 10.1073/pnas.1116791109; pmid: 22586104

51. B. J. Anderson et al., Dynamics of range margins formetapopulations under climate change. Proc. R. Soc. B 276,1415–1420 (2009). doi: 10.1098/rspb.2008.1681;pmid: 19324811

52. T. Delattre et al., Interactive effects of landscape and weatheron dispersal. Oikos 122, 1576–1585 (2013). doi: 10.1111/j.1600-0706.2013.00123.x; pmid: 19324811

53. J. P. Sexton, P. J. McIntyre, A. L. Angert, K. J. Rice, Evolutionand ecology of species range limits. Annu. Rev. Ecol. Syst. 40,415–436 (2009). doi: 10.1146/annurev.ecolsys.110308.120317

54. F. J. Thomson et al., Chasing the unknown: Predicting seeddispersal mechanisms from plant traits. J. Ecol. 98, 1310–1318(2010). doi: 10.1111/j.1365-2745.2010.01724.x

55. M. W. Tingley, M. S. Koo, C. Moritz, A. C. Rush, S. R. Beissinger,The push and pull of climate change causes heterogeneousshifts in avian elevational ranges. Global Change Biol. 18,3279–3290 (2012). doi: 10.1111/j.1365-2486.2012.02784.x

56. R. Early, D. F. Sax, Analysis of climate paths reveals potentiallimitations on species range shifts. Ecol. Lett. 14, 1125–1133(2011). doi: 10.1111/j.1461-0248.2011.01681.x; pmid: 21955643

57. D. Murray-Rust et al., Combining agent functional types,capitals and services to model land use dynamics. Environ.Model. Softw. 59, 187–201 (2014). doi: 10.1016/j.envsoft.2014.05.019

58. A. Charmantier et al., Adaptive phenotypic plasticity inresponse to climate change in a wild bird population. Science320, 800–803 (2008). doi: 10.1126/science.1157174;pmid: 18467590

59. D. E. Schindler, R. Hilborn, Prediction, precaution, and policyunder global change. Science 347, 953–954 (2015).doi: 10.1126/science.1261824; pmid: 25722401

60. M. B. Araújo, M. New, Ensemble forecasting of speciesdistributions. Trends Ecol. Evol. 22, 42–47 (2007).doi: 10.1016/j.tree.2006.09.010; pmid: 17011070

61. L. Santini et al., A trait-based approach for predicting speciesresponses to environmental change from sparse data: How

well might terrestrial mammals track climate change? GlobalChange Biol. 22, 2415–2424 (2016). doi: 10.1111/gcb.13271;pmid: 27073017

62. S. Dullinger et al., Extinction debt of high-mountain plantsunder twenty-first-century climate change. Nat. Clim. Change2, 619–622 (2012). doi: 10.1038/nclimate1514

63. B. Beckage, L. J. Gross, S. Kauffman, The limits to prediction inecological systems. Ecosphere 2, 125 (2011). doi: 10.1890/ES11-00211.1

64. R. J. Lempert, M. E. Schlesinger, Robust strategies for abatingclimate change. Clim. Change 45, 387–401 (2000).doi: 10.1023/A:1005698407365

65. B. Rayfield, D. Pelletier, M. Dumitru, J. A. Cardille, A. Gonzalez,Multipurpose habitat networks for short-range and long-rangeconnectivity: A new method combining graph and circuitconnectivity. Methods Ecol. Evol. 7, 222–231 (2016).doi: 10.1111/2041-210X.12470

66. U.S. Committee for the Global Atmospheric Research Program,National Research Council, Understanding Climatic Change(National Academy of Sciences, Washington, DC, 1975).

ACKNOWLEDGMENTS

This paper originates from the “Ecological Interactions and RangeEvolution Under Environmental Change” and “RangeShifter” workinggroups, supported by the Synthesis Centre of the German Centre forIntegrative Biodiversity Research (DFG-FZT-118), DIVERSITAS, and itscore projects bioDISCOVERY and bioGENESIS. Supported by theCanada Research Chair, Natural Sciences and Engineering ResearchCouncil of Canada, and Quebec Centre for Biodiversity Science (A.G.);the University of Florida Foundation (R.D.H.); KU Leuven ResearchFund grant PF/2010/07, ERA-Net BiodivERsA TIPPINGPOND, andBelspo IAP SPEEDY (L.D.M.); European Union Biodiversity ObservationNetwork grant EU-BON-FP7-308454 (J.-B.M. and G.P.); KU LeuvenResearch Fund (J.P.); and NSF grants DEB-1119877 and PLR-1417754and the McDonnell Foundation (M.C.U.).

SUPPLEMENTARY MATERIALS

www.sciencemag.org/content/353/6304/aad8466/suppl/DC1Fig. S1Table S1

10.1126/science.aad8466

SCIENCE sciencemag.org 9 SEPTEMBER 2016 • VOL 353 ISSUE 6304 aad8466-9

RESEARCH | REVIEW

on

Sept

embe

r 9,

201

6ht

tp://

scie

nce.

scie

ncem

ag.o

rg/

Dow

nloa

ded

from

(6304), . [doi: 10.1126/science.aad8466]353Science Zollner and J. M. J. Travis (September 8, 2016) P. W. Leadley, S. C. F. Palmer, J. H. Pantel, A. Schmitz, P. A. Gonzalez, J. J. Hellmann, R. D. Holt, A. Huth, K. Johst, C. B. Krug,Singer, J. R. Bridle, L. G. Crozier, L. De Meester, W. Godsoe, A. M. C. Urban, G. Bocedi, A. P. Hendry, J.-B. Mihoub, G. Pe'er, A.Improving the forecast for biodiversity under climate change

 Editor's Summary

   

This copy is for your personal, non-commercial use only.

Article Tools

http://science.sciencemag.org/content/353/6304/aad8466article tools: Visit the online version of this article to access the personalization and

Permissionshttp://www.sciencemag.org/about/permissions.dtlObtain information about reproducing this article:

is a registered trademark of AAAS. ScienceAdvancement of Science; all rights reserved. The title Avenue NW, Washington, DC 20005. Copyright 2016 by the American Association for thein December, by the American Association for the Advancement of Science, 1200 New York

(print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last weekScience

on

Sept

embe

r 9,

201

6ht

tp://

scie

nce.

scie

ncem

ag.o

rg/

Dow

nloa

ded

from