Microclimate moderates plant responses tomacroclimate warmingPieter De Frennea,b,1, Francisco Rodríguez-Sánchezb, David Anthony Coomesb, Lander Baetena,c, Gorik Verstraetena,Mark Vellendd, Markus Bernhardt-Römermanne,2, Carissa D. Brownd,f, Jörg Brunetg, Johnny Cornelish,Guillaume M. Decocqi, Hartmut Dierschkej, Ove Erikssonk, Frank S. Gilliaml, Radim Hédlm, Thilo Heinkenn,Martin Hermyo, Patrick Hommelp, Michael A. Jenkinsq, Daniel L. Kellyr, Keith J. Kirbys, Fraser J. G. Mitchellr, Tobias Naaft,Miles Newmanr, George Peterkenu, Petr Pet�ríkv, Jan Schultzw, Grégory Sonnierx, Hans Van Calstery, Donald M. Wallerx,Gian-Reto Waltherz, Peter S. Whiteaa, Kerry D. Woodsbb, Monika Wulft, Bente Jessen Graaecc, and Kris Verheyena

aForest and Nature Lab, Ghent University, BE-9090 Gontrode-Melle, Belgium; bForest Ecology and Conservation Group, Department of Plant Sciences,University of Cambridge, Cambridge CB2 3EA, United Kingdom; cTerrestrial Ecology Unit, Department of Biology, Ghent University, BE-9000 Ghent, Belgium;dDépartement de biologie, Université de Sherbrooke, Sherbrooke, QC, Canada J1K 2R1; eInstitute of Botany, University of Regensburg, DE-93053 Regensburg,Germany; fDepartment of Geography, Memorial University, St. John’s, NL, Canada A1B 3X9; gSouthern Swedish Forest Research Centre, Swedish University ofAgricultural Sciences, SE-230 53 Alnarp, Sweden; hAgency for Nature and Forests, BE-1000 Brussels, Belgium; iEdysan (FRE 3498), Centre National de laRecherche Scientifique/Université de Picardie Jules Verne, FR-80037 Amiens Cedex, France; jDepartment of Vegetation and Phytodiversity Analysis, Albrecht-von-Haller-Institute for Plant Sciences, Georg-August-Universität Göttingen, DE-37073 Göttingen, Germany; kDepartment of Ecology, Environment and PlantSciences, Stockholm University, SE-106 91 Stockholm, Sweden; lDepartment of Biological Sciences, Marshall University, Huntington, WV 25701; mDepartmentof Vegetation Ecology, Institute of Botany of the Academy of Sciences of the Czech Republic, CZ-65720 Brno, Czech Republic; nDepartment of BiodiversityResearch/Systematic Botany, Institute of Biochemistry and Biology, University of Potsdam, DE-14469 Potsdam, Germany; oDepartment of Earth andEnvironmental Sciences, Division of Forest, Nature and Landscape, Katholieke Universiteit Leuven, BE-3001 Leuven, Belgium; pAlterra Research Institute,Wageningen UR, 6700 AA Wageningen, The Netherlands; qDepartment of Forestry and Natural Resources, Purdue University, West Lafayette, IN 47907;rBotany Department and Trinity Centre for Biodiversity Research, School of Natural Sciences, Trinity College Dublin, Dublin 2, Ireland; sDepartment of PlantSciences, University of Oxford, Oxford OX1 3RB, United Kingdom; tInstitute of Land Use Systems, Leibniz-ZALF, DE-15374 Müncheberg, Germany; uBeechwoodHouse, St. Briavels Common, Lydney GL15 6SL, United Kingdom; vDepartment of Geographic Information Systems and Remote Sensing, Institute of Botany,Academy of Sciences of the Czech Republic, CZ-25243 Pr�uhonice, Czech Republic; wUS Forest Service, Milwaukee, WI 53203; xDepartment of Botany, Universityof Wisconsin–Madison, Madison, WI 53706; yResearch Institute for Nature and Forest, BE-1070 Brussels, Belgium; zSpecies, Ecosystems, Landscapes Division,Federal Office for the Environment FOEN, CH-3003 Bern, Switzerland; aaDepartment of Biology, University of North Carolina at Chapel Hill, Chapel Hill, NC27599; bbProgram in Natural Sciences, Bennington College, Bennington, VT 05201; and ccDepartment of Biology, Norwegian University of Science andTechnology, NO-7491 Trondheim, Norway

Edited by Harold A. Mooney, Stanford University, Stanford, CA, and approved September 24, 2013 (received for review June 13, 2013)

Recent global warming is acting across marine, freshwater, andterrestrial ecosystems to favor species adapted to warmer con-ditions and/or reduce the abundance of cold-adapted organisms(i.e., “thermophilization” of communities). Lack of community re-sponses to increased temperature, however, has also been re-ported for several taxa and regions, suggesting that “climaticlags” may be frequent. Here we show that microclimatic effectsbrought about by forest canopy closure can buffer biotic re-sponses to macroclimate warming, thus explaining an apparentclimatic lag. Using data from 1,409 vegetation plots in Europeanand North American temperate forests, each surveyed at leasttwice over an interval of 12–67 y, we document significant ther-mophilization of ground-layer plant communities. These changesreflect concurrent declines in species adapted to cooler conditionsand increases in species adapted to warmer conditions. However,thermophilization, particularly the increase of warm-adapted spe-cies, is attenuated in forests whose canopies have become denser,probably reflecting cooler growing-season ground temperaturesvia increased shading. As standing stocks of trees have increasedin many temperate forests in recent decades, local microclimaticeffects may commonly be moderating the impacts of macroclimatewarming on forest understories. Conversely, increases in harvestingwoody biomass—e.g., for bioenergy—may open forest canopiesand accelerate thermophilization of temperate forest biodiversity.

climate change | forest management | understory | climatic debt | range shifts

Biological signals of recent global warming are increasinglyevident across a wide array of ecosystems (1–7). However,

the temperature experienced by organisms at ground level (mi-croclimate) can substantially differ from the atmospheric tem-perature due to local land cover and terrain variation in terms ofvegetation structure, shading, topography, or slope orientation(8–15). The daytime or nighttime surface temperature in roughmountain terrain, for instance, can deviate by up to 9 °C from theair temperature (10). Likewise, forest structure creates substantial

temperature heterogeneity, with the interior daytime temperature indense forests being commonly several degrees cooler than in moreopen habitats during the growing season (12–15). Spatial microcli-matic temperature variation can thus be substantial relative toprojected changes in average temperature over time, and biotic

Significance

Around the globe, climate warming is increasing the domi-nance of warm-adapted species—a process described as“thermophilization.” However, thermophilization often lagsbehind warming of the climate itself, with some recent studiesshowing no response at all. Using a unique database of morethan 1,400 resurveyed vegetation plots in forests across Europeand North America, we document significant thermophilizationof understory vegetation. However, the response to macro-climate warming was attenuated in forests whose canopieshave become denser. This microclimatic effect likely reflectscooler forest-floor temperatures via increased shading duringthe growing season in denser forests. Because standing stocksof trees have increased in many temperate forests in recentdecades, microclimate may commonly buffer understory plantresponses to macroclimate warming.

Author contributions: P.D.F., F.R.-S., D.A.C., G.M.D., B.J.G., and K.V. designed research;P.D.F., F.R.-S., L.B., G.V., M.V., M.B.-R., C.D.B., J.B., J.C., G.M.D., H.D., O.E., F.S.G., R.H., T.H.,M.H., P.H., M.A.J., D.L.K., K.J.K., F.J.G.M., T.N., M.N., G.P., P.P., J.S., G.S., H.V.C., D.M.W.,G.-R.W., P.S.W., K.D.W., M.W., and K.V. performed research; P.D.F., F.R.-S., L.B., G.V., and K.V.analyzed data; and P.D.F., F.R.-S., D.A.C., L.B., M.V., M.B.-R., C.D.B., J.B., J.C., G.M.D., O.E., F.S.G.,R.H., T.H., M.H., M.A.J., D.L.K., K.J.K., M.N., G.P., P.P., G.S., H.V.C., D.M.W., P.S.W., K.D.W., B.J.G.,and K.V. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.1To whom correspondence should be addressed. E-mail: pieter.defrenne@ugent.be.2Present address: Instituteof Ecology, Friedrich-Schiller-University Jena,DE-07743 Jena,Germany.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1311190110/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1311190110 PNAS Early Edition | 1 of 5

ECOLO

GY

responses to macroclimate warming may be buffered by micro-climatic heterogeneity (9–11). It seems likely that microclimatescan modulate the large-scale, multispecies, and long-term re-sponse of biota to macroclimate warming, but this is currentlyunverified. Testing this idea will permit incorporation of fine-grained thermal variability into bioclimatic modeling of futurespecies distributions (11, 16), and is particularly relevant to forestunderstories, which play a key role in vital ecosystem services offorests such as tree regeneration, nutrient cycling, and pollina-tion (17, 18). Additionally, microclimatic buffering might help toexplain the lag of community responses to increased temperaturethat has been reported for several taxa and regions (3, 5, 7).Temperate forests comprise 16% (5.3 million km2) of the

world’s forests (19), and understory plants represent on averagemore than 80% of temperate forest plant diversity (17). Tem-perate forests have recently experienced pronounced climatewarming, but have also been heavily influenced by other envi-ronmental changes. Changing forest management regimes dueto altered socioeconomic conditions, but also eutrophication, cli-mate warming, and fire suppression, have resulted in increasedtree growth, standing stocks, and densities in many temperateforests of the northern hemisphere (20–25). In Western Europe,for instance, logging and natural losses of tree biomass have beenconsistently lower than annual growth increments, resulting in analmost doubling of standing stocks of trees per hectare between1950 and 2000 (21). Hence, forests’ powerful influence on ter-

restrial microclimates raises the intriguing possibility that non-climatic drivers of global change, such as forest canopy closure,might have lowered ground-level temperatures via increasedshading, thereby counteracting the effects of macroclimate warmingon the forest understory.Here we compiled plant occurrence data (1,032 species in

total) from 1,409 resurveyed vegetation plots in temperate de-ciduous forests. The plots were distributed across 29 regions oftemperate Europe and North America (Fig. 1 A and B) with anaverage interval of 34.5 y (range: 12–67 y) between the originaland repeated vegetation surveys (Table S1). From these plots, wetested for plant community responses to recent macroclimatewarming and assessed the potential role of changes in forestcanopy cover in modulating such responses. To quantify possiblethermophilization of communities, we inferred the temperaturepreferences of species from distribution data by means of eco-logical niche modeling (16) (Fig. 1 A–C). This method buildson previous use of species’ temperature preferences to assesscommunity-level climate-change impacts (3–6). We then calcu-lated the floristic temperature for each plot by sampling from thetemperature preference distributions of all species that werepresent at the time of the surveys. The probability of samplinga temperature for a particular species was determined by theshape of its thermal response curve as estimated by niche mod-eling. We repeated this resampling procedure 500 times to ac-count for the variability and uncertainty in species’ temperature

A

24

Temperature (°C)

Pro

babi

lity

ofoc

curr

ence

C

B

Anemone nemorosa

esnedanacmumehtnaiaMasoromenenomenA

Maianthemum

canadense

D

Species 1

Species 2

Species 3

Community time 1

Community

time 2

12 34

5

67

9

8

10

11

19

171622

1514

121318

2320

21

25

26

272829

1.0

0.8

0.6

0.4

0.2

0.0

0 10 20 30

Temperature (°C)0 10 20 30

Fig. 1. Estimation of plant thermal response curves and temporal community-level responses to warming. (A–C) Thermal response curves were estimatedfrom species’ current distribution ranges (green areas in maps). The two most frequent understory species in the database are shown as an example:Anemone nemorosa from Europe (A) and Maianthemum canadense from North America (B) (Tables S2 and S3). The red dots indicate the locations of the 29study regions and numbers refer to site descriptions in Table S1. (D) Hypothetical community response to warming: the community consists of species 1 and 2at time 1 and species 2 and 3 at time 2, which results in thermophilization between time 1 and 2 because the more cold-adapted species 1 is lost from thecommunity, and the more warm-adapted species 3 is gained. The green arrows represent the temporal shift of the mean, left tail (fifth percentile) and righttail (95th percentile) of the distribution of floristic temperatures and reflect the degree to which the mean thermophilization increases, cold-adapted speciesdecrease, and warm-adapted species increase, respectively.

2 of 5 | www.pnas.org/cgi/doi/10.1073/pnas.1311190110 De Frenne et al.

preferences (26). Communities with many cold-adapted spe-cies will thus have a lower floristic temperature, and vice versa.To assess thermophilization over time, we compared the mean,fifth, and 95th percentiles of the temperature distribution forevery plot at the old and recent survey, respectively (Fig. 1D).The shift of the mean of the distribution of floristic temperatures(in degrees Celsius per decade) then reflects the mean thermo-philization. In contrast, shifts in the tails of the distribution ofplot-level floristic temperatures (fifth and 95th percentiles) reflectchanges in the occurrence of cold and warm-adapted species,respectively (Fig. 1D).

Results and DiscussionSignificant community turnover took place over time in thetemperate forests we sampled: on average, one-third of thespecies present in the old surveys has been replaced by otherspecies today; the mean Lennon dissimilarity index (SI Materialsand Methods) across all plots was 0.69 (95% bootstrappingconfidence interval: [0.68, 0.70]), both in Europe (dissimilaritywas 0.70 [0.69, 0.71]) and North America (0.65 [0.62, 0.68]). Thisfloristic turnover partly arose from the nonrandom replacementof species in terms of their temperature preferences, illustratedby significant thermophilization both in European and easternNorth American forests (Fig. 2A). On average, the estimatedthermophilization rate was 0.041 °C·decade−1 (the range across10 different modeling methods was 0.027–0.056 °C·decade−1;Table S4). Significant interregional variation was present, withthermophilization rates ranging from +0.83 °C·decade−1 (GreatSmoky Mountains) to –0.64 °C·decade−1 (Ireland). Thermo-philization was significantly positive in 20 of 29 regions, sig-

nificantly negative in eight study regions, and unchanged in oneregion (Fig. 2B).The overall thermophilization of understory plant communi-

ties has been driven by concurrent gains of relatively warm-adapted species and loss of cold-adapted taxa, as revealed by theshifts in the cold (fifth percentile) and warm (95th percentile)ends of the floristic temperature distribution (Fig. 2C). In theeastern North American forest plots, however, both warm-adapted and cold-tolerant species have increased (Fig. 2C) dueto continuous immigration of new species (i.e., overall increasein species richness), which does not occur in the European plots(SI Results). The mean thermophilization of understory plantcommunities that we observe across temperate deciduous forestsin two continents expands on earlier findings that mountainvegetation communities are showing increases of lower-altitudespecies at higher altitudes, leading to novel species assemblages(3, 4, 27). The thermophilization of vegetation is consistent with thewarming climate observed across the regions: the mean rise in April-to-September temperatures between the old and recent survey was0.28 °C·decade−1 (Table S1). We found a positive relationshipbetween the thermophilization and the region-specific April-to-September temperature change, indicating higher thermo-philization in areas with higher rates of warming (mean slope0.07, P < 0.001; SI Results). European and North Americantemperate deciduous forest vegetation is thus changing as ex-pected by macroclimate warming, but thermophilization lags behindrising temperatures.We found that local changes in forest canopy cover modu-

late the thermophilization of vegetation; thermophilization waslowest in forests that became denser, and highest in forests that

Gaume, BE (1)

Tullgarn, SE (10)

Elbe-Weser, DE (11)

Devín, CZ (12)Milovice Wood, CZ (13)

Rychlebské Mts., CZ (14)

Wytham Woods, UK (15)

Göttingen, DE (16)

Milícovský, CZ (17)

Switzerland, CH (18)

Hirson/Saint-Michel, Fr (19)

Flanders, BE (2)Andigny, Fr (20)

Speulderbos, NL (21)

Lady Park Wood, UK (22)

Münich, DE (23)

County Kerry, IRL (24)

Zoerselbos, BE (3)

Herenbossen, BE (4)

Vorte Bossen, BE (5)

Meerdaalwoud, BE (6)

Florenne, BE (7)

Tournibus, BE (8)

Dalby, SE (9)

Smoky Mountains, TN, USA (25)

Fernow, WV, USA (26)

Mont Mégantic, QC, CAN (27)Dukes, MI, USA (28)

WI, USA (29)

-0.80 -0.40 0.00 0.40 0.80Thermophilization (°C.decade-1)

Europe

North-America

0.00

0.04

0.08

0.12

Ther

mop

hiliz

atio

n (°

C.d

ecad

e-1)

C

BA

Warming

Cooling

WarmingCooling

All d

ata

Euro

peNo

rth-A

mer

ica

-0.20

-0.10

0.00

0.10

0.20

Ther

mop

hiliz

atio

n (°

C.d

ecad

e-1)

Shift cold speciesShift warm species

Warming

Cooling

Fig. 2. Thermophilization of temperate forest understories across Europe and North America. (A and B) Mean thermophilization (positive values denoteincreases over time) for all data and in European and American forests (A) and for the individual regions (B). The numbers between brackets refer to the sitesin Fig. 1. (C) Mean shifts in relatively cold-adapted (blue) and warm-adapted species (red) for all plots, and in Europe and North America. Positive valuesreflect positive shifts of the left and right tail, i.e., decreases of cold-adapted and increases of warm-adapted taxa, respectively. Error bars denote the 95%confidence intervals based on 500 resampled species’ temperature preferences.

De Frenne et al. PNAS Early Edition | 3 of 5

ECOLO

GY

became more open over time (Fig. 3A). The relationship be-tween forest canopy cover changes and the mean thermophi-lization was significantly negative (mean slope = –0.0073, P <0.001, range of slopes across 10 different modeling methods–0.0392 to –0.0015; Table S5). Moreover, the increase ofwarm-adapted species was consistently lower in plots thatincreased in canopy cover compared with plots that becamemore open over time, which experienced stronger thermo-philization (mean slope = –0.0170, P < 0.001, range across 10different modeling methods –0.0463 to –0.0099). For cold-adapted species, the effects of canopy cover changes werelower and more variable (mean slope = –0.0071, P < 0.05,range across 10 different modeling methods –0.0584 to +0.0083). Thus, cold-adapted taxa responded to a lesser extent tochanges in forest canopy cover (Fig. 3B). Taken together, theseresults suggest that recent forest canopy closure in northern-hemi-spheric temperate forests has buffered the impacts of macroclimatewarming on ground-layer plant communities, thus slowing changesin community composition.Forest canopy closure modulates macroclimatic trends through

the effects on local microclimates. Dense tree canopies not onlylower ground-layer temperatures but also increase relative airhumidity and shade in the understory (12–15). Hence, the re-ported decrease in light-demanding understory plants in Europe(28) is also congruent with the local environmental effects causedby forest canopy closure. Higher relative humidity in dense forestscan also protect forest herbs and tree seedlings from summerdrought, decreasing mortality and thus buffering the impacts oflarge-scale climate change (15, 29). Furthermore, many forestherbs are known to be slow-colonizing species (30). Given thehigh degree of habitat fragmentation in contemporary landscapes,microclimatic buffering in dense forests may be a critical mechanismto ensure the future conservation of temperate forest plant diversity.If forest canopy closure attenuates warming in the understory,

atmospheric temperatures provide an unrealistic benchmarkagainst which to compare floristic temperatures. Hence, land-usechanges such as forest canopy closure could partially explainthe lag observed between, for instance, lowland forest plantcommunity composition (3) and temperature trends as measured

in weather stations (i.e., above dwarf vegetation in open areas).Instead of accumulating climatic lags, these understory communi-ties could be mostly exploiting the buffering microclimatic effectsbrought about by canopy closure. Therefore, measuring climatechange in the field and identifying the actual climatic lags of biotais crucial to further our understanding of community reorderingand future biodiversity conservation in the face of climate change.In sum, we observed increasing dominance of warm-adapted

understory plants across more than 1,400 plots and 29 regions inEuropean and North American temperate deciduous forests.Additionally, our most striking finding is the temporal bufferingof the continent-wide response of understory vegetation to mac-roclimate warming by forest canopy closure. The importance ofincreased canopy cover in influencing understory biodiversityis particularly relevant in an era when forest managementworldwide is confronted with increasing demands for woodybiomass, not least as an alternative source of renewable energy(31, 32). In addition, current conservation actions in Europeanforests are regularly directed toward restoring traditional man-agement (e.g., coppicing in ancient forests), resulting in canopyopening. Such actions could not only result in soil nutrient de-pletion, lower biomass pools, and enhanced soil nitrogen release(28, 31), but also, depending on the sylvicultural system applied,increase temperatures at the forest floor (12–15). Large-scalereopening of the canopy for woody biomass harvesting may thushasten thermophilization of understory plant communities oftemperate forests.

Materials and MethodsUnderstory Resurveys. We compiled complete species lists of 1,409 rigorouslyselected (SI Materials and Methods) resurveyed vegetation plots in Europeanand North American ancient deciduous forests, and determined forest can-opy cover changes (sum of the tree and shrub species’ canopy cover) for 854of the plots (19 regions; Table S1 and Fig. S1). Plots were either permanentlymarked or semipermanent (i.e., with known coordinates; SI Materials andMethods), and plot sizes ranged between 1 and 1,000 m2 (Table S1). Plot-levelchanges in canopy cover between the old and recent surveys were quantifiedas response ratios log(coverrecent/coverold).

Mean Shift cold speciesShift warm species

Denser More open DenserMore open-0.05

0.00

0.05

0.10

0.15

0.20

-0.05

0.00

0.05

0.10

0.15

0.20

Forest canopy cover change (log(cover /cover ))recent old

-4 -2 0 2 4-4 -2 0 2 4

Ther

mop

hiliz

atio

n(°

C.d

ecad

e)

-1

A B

Fig. 3. Forest canopy closure modulates understory thermophilization. (A) Relationship between forest canopy cover change and mean thermophili-zation of understory plant communities in temperate European and North American forests. (B) Relationship between forest canopy cover change anddecreases of cold-adapted species (expressed by the shift of the left tail of the plot-level distribution of floristic temperatures, blue) and increases ofwarm-adapted species (expressed by the shift of the right tail of the plot-level distribution of floristic temperatures, red). Relationships result from mixed-effect models for each of 500 samples; shaded areas denote 95% confidence intervals based on those samples. These results are mainly based on theEuropean data (Table S1).

4 of 5 | www.pnas.org/cgi/doi/10.1073/pnas.1311190110 De Frenne et al.

Calculation of Thermophilization.We calculated the thermophilization for eachplot by sampling from the inferred temperature preference distributions of thespecies present (Fig. 1). The long-term mean temperature and precipitation inthe growing season (April to September; Fig. S2) were used to estimate spe-cies’ thermal response curves by means of ecological niche modeling (16). Toaccount for variability and uncertainty in species’ thermal preferences andniche widths (26), the distribution of plot-level floristic temperatures at eachsurvey was constructed by resampling 500 times from species’ thermal re-sponse curves. The mean thermophilization per plot was quantified as thedifference between the mean floristic temperature (in degrees Celsius) be-tween the recent and original survey, divided by the time interval (indecades) between the two surveys. In addition, we determined thecontribution of the loss of cold-adapted and the gain of warm-adaptedspecies to the thermophilization patterns by quantifying the shifts in theleft and right tails (fifth and 95th percentiles, respectively) of the plot-leveldistribution of floristic temperatures (Fig. 1D and Figs. S3 and S4).

Forest Cover and Temperature Change vs. Thermophilization. The relationshipsbetween forest canopy cover and temperature changes on the one hand,

and thermophilization on the other hand (shifts in themean, fifth, and 95thpercentiles of the distribution of floristic temperatures over time) were

assessed usingmixed-effect models with “study region” as a random-effect term

for each of the 500 resampled species’ temperature preferences. Sensitivity

analyses revealed that excluding precipitation, applying various climatic

periods, study area extents, and modeling approaches, and randomly re-

moving subsets of species resulted in consistent results (see SI Materials and

Methods for a detailed account of the methods and SI Results for sup-

porting results).

ACKNOWLEDGMENTS. We thank J. Kartesz and M. Nishino for the Americanspecies distribution maps, two anonymous reviewers for valuable comments,and the Research Foundation–Flanders (FWO) for funding the scientific re-search network FLEUR. Support for this work was provided by FWO Postdoc-toral Fellowships (to P.D.F. and L.B.), European Union Seventh FrameworkProgramme FP7/2007-2013 Grant 275094 (to F.R.-S.), the Natural Sciences andEngineering Research Council of Canada (M.V. and C.D.B.), and long-term re-search development project RVO 67985939 (to R.H. and P.P.).

1. Walther GR, et al. (2002) Ecological responses to recent climate change. Nature416(6879):389–395.

2. Parmesan C (2006) Ecological and evolutionary responses to recent climate change.Annu Rev Ecol Evol Syst 37:637–669.

3. Bertrand R, et al. (2011) Changes in plant community composition lag behind climatewarming in lowland forests. Nature 479(7374):517–520.

4. Gottfried M, et al. (2012) Continent-wide response of mountain vegetation to climatechange. Nature Clim Change 2(2):111–115.

5. Devictor V, et al. (2012) Differences in the climatic debts of birds and butterflies ata continental scale. Nature Clim Change 2(2):121–124.

6. Cheung WWL, Watson R, Pauly D (2013) Signature of ocean warming in global fish-eries catch. Nature 497(7449):365–368.

7. Dullinger S, et al. (2012) Extinction debt of high-mountain plants under twenty-first-century climate change. Nature Clim Change 2(8):619–622.

8. Kraus G (1911) Soil and Climate in a Small Space (Fischer, Jena, East Germany), German.9. Dobrowski SZ (2011) A climatic basis for microrefugia: The influence of terrain on

climate. Glob Change Biol 17(2):1022–1035.10. Scherrer D, Körner C (2010) Infra-red thermometry of alpine landscapes challenges

climatic warming projections. Glob Change Biol 16(9):2602–2613.11. Lenoir J, et al. (2013) Local temperatures inferred from plant communities suggest

strong spatial buffering of climate warming across Northern Europe. Glob ChangeBiol 19(5):1470–1481.

12. Geiger R, Aron RH, Todhunter P (2009) The Climate Near the Ground (Rowman &Littlefield, Plymouth, UK), 7th Ed.

13. Chen J, et al. (1999) Microclimate in forest ecosystem and landscape ecology. Bio-science 49(4):288–297.

14. Norris C, Hobson P, Ibisch PL (2012) Microclimate and vegetation function as in-dicators of forest thermodynamic efficiency. J Appl Ecol 49(3):562–570.

15. von Arx G, Graf Pannatier E, Thimonier A, Rebetez M (2013) Microclimate in forestswith varying leaf area index and soil moisture: Potential implications for seedlingestablishment in a changing climate. J Ecol 101(5):1201–1213.

16. Peterson AT, et al. (2011) Ecological Niches and Geographic Distributions (PrincetonUniv Press, Princeton, NJ).

17. Gilliam FS (2007) The ecological significance of the herbaceous layer in temperateforest ecosystems. Bioscience 57(10):845–858.

18. Nilsson MC, Wardle DA (2005) Understory vegetation as a forest ecosystem driver:Evidence from the northern Swedish boreal forest. Front Ecol Environ 3(8):421–428.

19. Hansen MC, Stehman SV, Potapov PV (2010) Quantification of global gross forestcover loss. Proc Natl Acad Sci USA 107(19):8650–8655.

20. United Nations Economic Commission for Europe/Food and Agricultural Organization(2000) Forest Resources of Europe, CIS, North America, Australia, Japan and NewZealand (Industrialized Temperate/Boreal Countries). Geneva Timber and ForestStudy Papers (United Nations, New York).

21. Gold S, Korotkov AV, Sasse V (2006) The development of European forest resources,1950 to 2000. For Policy Econ 8(2):183–192.

22. Rautiainen A, Wernick I, Waggoner PE, Ausubel JH, Kauppi PE (2011) A national andinternational analysis of changing forest density. PLoS ONE 6(5):e19577.

23. Luyssaert S, et al. (2010) The European carbon balance. Part 3: Forests. Glob ChangeBiol 16(5):1429–1450.

24. Thomas RQ, Canham CD, Weathers KC, Goodale CL (2010) Increased tree carbonstorage in response to nitrogen deposition in the US. Nat Geosci 3(1):13–17.

25. McMahon SM, Parker GG, Miller DR (2010) Evidence for a recent increase in forestgrowth. Proc Natl Acad Sci USA 107(8):3611–3615.

26. Rodríguez-Sánchez F, De Frenne P, Hampe A (2012) Uncertainty in thermal tolerancesand climatic debt. Nature Clim Change 2(9):636–637.

27. Lenoir J, Gégout JC, Marquet PA, de Ruffray P, Brisse H (2008) A significant upwardshift in plant species optimum elevation during the 20th century. Science 320(5884):1768–1771.

28. Verheyen K, et al. (2012) Driving factors behind the eutrophication signal in under-storey plant communities of deciduous temperate forests. J Ecol 100(2):352–365.

29. Lendzion J, Leuschner C (2009) Temperate forest herbs are adapted to high air hu-midity—evidence from climate chamber and humidity manipulation experiments inthe field. Can J Res 39(12):2332–2342.

30. De Frenne P, et al. (2011) Interregional variation in the floristic recovery of post-agricultural forests. J Ecol 99(2):600–609.

31. Schulze ED, Körner C, Law BE, Haberl H, Luyssaert S (2012) Large-scale bioenergy fromadditional harvest of forest biomass is neither sustainable nor greenhouse gas neu-tral. GCB Bioenergy 4(6):611–616.

32. Tilman D, et al. (2009) Energy. Beneficial biofuels—the food, energy, and environmenttrilemma. Science 325(5938):270–271.

De Frenne et al. PNAS Early Edition | 5 of 5

ECOLO

GY

Supporting InformationDe Frenne et al. 10.1073/pnas.1311190110SI Materials and MethodsUnderstory Resurveys.We compiled as many data sets as possible inwhich vascular plants in the herb layer (<1 m height, includingferns and tree seedlings) of lowland and lower mountainous(<800 m above sea level) temperate deciduous forests weresurveyed at least twice over a period of at least 10 y. The re-sulting database has a total of 1,032 plant species (640 Europeanand 392 American species) and 1,409 plots, i.e., 1,220 plots werelocated in 24 European regions and 189 plots in five regions ineastern North America (Fig. 1 and Table S1). All plots werepermanent (i.e., plots were permanently marked in the fieldat the time of the first survey) or semipermanent (i.e., non-permanently marked in the field, but with known coordinates,often supplemented with field descriptions) and could thus becompared on a pairwise basis. Semipermanent plots were re-located based on some combination of detailed maps (mostlyscaled 1:25,000 or better), field descriptions of plot positionsrelative to reference points such as tracks and large, old trees,location sketches, slope and aspect of the site, and global posi-tioning system coordinates. More information on the relocationaccuracy of the semipermanent plots is available in refs. 1, 2, and3. Comparable abundance data to use abundance weighting werenot available for all regions and, therefore, all vegetation datawere converted into presence/absence data throughout (ran-domly removing subsets of species did not affect the conclusions,which suggests that this was of minor influence; SI Results). Wespecifically only focus on “ancient” deciduous forests (i.e., thoseon lands never converted to agriculture since the oldest land-usemaps) (4) because these are of important conservation concern,and because recent forests (often established on former agri-cultural land) remain floristically impoverished for centuries tomillennia and are likely to show more successional dynamicscompared with communities of ancient forests (4–7). Addi-tionally, we focused on plots where no whole-stand-replacingmanagement actions have taken place during the period ofobservation; heavily disturbed plots such as those situated onharvester tracks created by logging activities were deliberatelyexcluded. Permanent and semipermanent plots in ancient tem-perate forests provide an invaluable opportunity to test thecontribution of changes in microclimate to plant communityreordering over time. The European resurveys spanned a pe-riod of 17–67 y, and the American data spanned a period of12–55 y (mean: 34.5 y; Table S1). The 50 most frequent Eu-ropean and American species and their life form and change infrequency between the old and recent survey are displayed inTables S2 and S3.

Inference of Species’ Temperature Preferences.We inferred species’temperature preferences from their current distribution rangesby means of ecological niche modeling (8, 9).Species distribution data.Distribution maps were obtained from theatlases of Hultén and Fries (10) and Meusel and Jäger (11) forthe European species, and Kartesz (12) for the American spe-cies. Maps were available for 92.6% of the European species and100% of the American species. Taxa that were only identified tothe genus level were excluded (e.g., some Carex sp., Dryopterissp., Poa sp., Rosa sp., Rubus sp., and Viola sp.; 29 taxa in theEuropean and 25 in the American surveys). The European mapswere georeferenced using Quantum GIS v.1.7.4 (http://qgis.org/)with a thin-plate transformation and nearest-neighbor resam-pling methods. All other analyses were performed in R 2.15.2

(13). The American maps were directly available in digital for-mat (12).Ecological niche models. The long-term mean temperature andprecipitation in the growing season (April to September) andwhole year (January to December) were extracted from World-Clim (14). To estimate the climatic niche of each species, we setthe study area extent to all land comprised between 15°W (At-lantic Ocean) and 60°E (Ural Mountains), and 25°N (Sahara)and 90°N (North Pole) for Europe; and 180°W (Pacific Ocean)and 51°W (Atlantic Ocean), and 10°N (Panama Canal) and 90°N(North Pole) for North America. Within these geographicallimits, the amount of background data (pseudoabsences) was∼10,000 (Europe: 8,556 grid cells, America: 10,602) (15).Changing the study area extent to a buffer of 3,000 km aroundthe study regions did not affect the thermophilization patterns(SI Results). The species’ temperature preferences were inferredfrom the distribution maps by using a wide array of comple-mentary methods: (i) assuming Gaussian response curves vs.more flexible curve shapes, and (ii) with or without precipitationincluded in the models (because both temperature and pre-cipitation are primary determinants of species distributions)(16). It may be important to also consider precipitation eventhough we are only interested in inferring species’ temperaturepreferences, because failure to account for both climatic factorsmay result in biased inferences of thermal response curves (17).When precipitation was included in the models, species’ thermalresponse curves were defined by setting precipitation to themedian precipitation amount across each species’ range. Notethat the long-term mean growing season and mean annualtemperature (Europe: r = 0.965, n = 8,556 grid cells, P < 0.001;America: r = 0.981, n = 10,602, P < 0.001) were significantlycorrelated.Two statistical modeling methods were used to build the

ecological niche models based on the distribution range of eachspecies. In the first method, we assumed a Gaussian relationshipbetween the environment (temperature T and precipitation P)and species j’s probability of occurrence at every grid cell i, Gij:

Gij = aj exp−�

Ti − bjcj

�2

; [S1]

with Ti the temperature value at grid cell i, and species-specificparameters aj the maximum probability of occurrence, bj theoptimum temperature of species j, and cj the rate at which prob-ability of occurrence decays away from the optimum bj (18). TheGaussian model including precipitation P then becomes

Gij = aj exp−

��Ti − bj

cj

�2

+

�Pi − dj

ej

�2�; [S2]

with Pi the precipitation value at grid cell i, and dj and ej theoptimum precipitation and the rate at which the probability ofoccurrence declines away from the optimum dj for species j,respectively. Parameter estimation was performed using non-linear generalized least-squares (gnls function in nlme packagein R) (19).The second approach to build the ecological niche models

consisted of generalized additive models (GAMs; gam functionwith binomial errors in gam package in R) (20) and thus allowedfor more flexible shapes of the thermal response curves.

De Frenne et al. www.pnas.org/cgi/content/short/1311190110 1 of 13

Both modeling methods (Gaussian vs. GAMs) including orexcluding precipitation resulted in robust patterns of thermo-philization and relationships with forest canopy cover (SI Results).We used the Gaussian thermal response curves derived from thecombined growing-season temperature and growing-season pre-cipitation throughout the main text.

Forest Canopy Cover and Temperature Change.We quantified changesin canopy cover as a proxy for changes in microclimate at theforest floor, with higher canopy cover associated with deepershade and reduced growing season daytime temperatures (21–23). Information on the forest canopy cover (sum of tree andshrub species’ canopy cover in percentage per species) for boththe old and recent survey was available for 854 of the 1,409 plots.The median, fifth, and 95th percentile of cover values were116%, 64%, and 255%, respectively, in the old surveys, and119%, 53%, and 268% in the recent surveys. Note that this estimateof forest cover is commonly larger than 100% because tree andshrub canopies tend to significantly overlap. We expressed changesin canopy cover between the old and recent survey as natural log-arithm response ratios log(coverrecent/coverold). Positive values thusrefer to plots where the canopy cover has increased, negative valuesto forest plots where the canopy cover has decreased, and zerovalues to no change. Dividing the forest cover change responseratios by the time interval between the two survey periods (t, indecades) using log(coverrecent/coverold)·t

−1 resulted in similar results(SI Results). We here primarily focus on forest cover changes be-cause these have been identified as the most important driver offloristic turnover in understory plant communities of Europeantemperate forests in recent decades (24). The forest cover responseratios range between −3.8 and +3.3. Despite this significant varia-tion in forest cover changes between the old and recent surveys,canopy cover increased on average by 5.9% (Fig. S1), a patternrepresentative of most northern hemisphere temperate forests (seemain text).To quantify the rate of temperature change and its effect on the

thermophilization patterns, we obtained temperature data overland for both the growing season (April to September) and wholeyear (January to December) for each of the 29 study regions fromthe Goddard Institute for Space Studies (GISS) database at 1°resolution and 250 km smoothing (25). For each region, wecalculated the mean rate of temperature change between the oldand recent surveys as the difference between the average tem-perature during the 5 y preceding the old and recent survey di-vided by the time interval (again expressed in degrees Celsius perdecade) (26). The mean temperature change ranged between +0.02 and +0.67 °C·decade−1 from January to December (overallmean +0.27 °C·decade−1) and between +0.02 and +0.69 °C·de-cade−1 in the growing season (overall mean +0.28 °C·decade−1;Table S1). The temperature change calculated for the whole yearvs. the growing season only were significantly correlated (r =0.867, n = 29, P < 0.0001; Fig. S2) and therefore, because of thehigher relevance of growing-season temperatures for plants(26–28), we used the growing-season temperature change forfurther analyses.

Analyses of Community-Level Changes over Time. To determine thefloristic turnover in the forest understories, we calculated theLennon dissimilarity index (29) between the old and recentsurvey for each plot j as min(bj, cj)/[min(bj,cj) + aj] using thedesigndist function from the vegan package (30), with aj repre-senting the number of species shared by both surveys, bj thenumber of species unique to the first (old) survey, and cj the numberof species unique to the second (recent) survey. The Lennonindex quantifies actual floristic turnover and not nestedness de-rived from differences in species richness between time periods(24). Due to the bounded nature (between 0 and 1) of the Lennoncoefficients, the 95% confidence intervals (CIs) were based on

999 bootstrap samples using the boot function from the bootpackage in R (31).Second, the thermophilization was calculated per plot. A

flowchart of the whole analysis is given in Fig. S3. We charac-terized species’ relationship with temperature by their wholethermal response curve (as estimated by niche modeling), be-cause simply using species’ temperature optima (bj in Eq. S1) orthe mean temperature across species’ distribution ranges (32)fails to account for variability and uncertainty in species’ thermaltolerances and can provide biased estimates of communitythermophilization (17). Therefore, we used a bootstrap approachto account for uncertainty and variability in species’ thermaltolerances. A temperature value was sampled independently 500times for each species present in a plot with the probability de-termined by the shape of its thermal response curve (Fig. 1 andFig. S4). Thus, species mainly distributed across cold areas (e.g.,higher latitudes or altitudes) will contribute relatively low tem-peratures across most samples, whereas species distributedacross warmer areas will generally contribute relatively hightemperature values.Subsequently, for each of the 500 samples, species’ tempera-

ture values were then averaged across all present plant speciesper plot and survey period into the “floristic temperature” [sensuBertrand et al. (33)] at the community level. Next, the thermo-philization Thj of plot j was quantified as the difference betweenthe mean floristic temperature in the recent survey (FTrecent,j)and the mean floristic temperature in the original survey (FTold,j)divided by the time interval between the two survey periods (tj, indecades):

Thj =�FTrecent;j −FTold;j

�:t−1j : [S3]

A positive value of Th thus quantifies an overall temporal in-crease of the floristic temperature in degrees Celsius per decade.This calculation was thus repeated 500 times for every plot.Next, to determine the contribution of decreases of cold-

adapted and increases of warm-adapted species to the thermo-philization, we also quantified shifts in the left (Cj) and right tails(Wj) of the plot-level distribution of floristic temperatures, re-spectively (Fig. 1D and Figs. S3 and S4; see refs. 34–36 foranalogous applications). The tails were characterized as the fifthand 95th percentiles (to minimize the impact of outliers) of theplot-level distribution of floristic temperatures per survey period.The shifts in the cold (Cj) and warm (Wj) ends of the distributionof floristic temperature between the recent and old survey perdecade were then calculated as

Cj =�LTrecent;j −LTold;j

�:t−1j [S4]

Wj =�RTrecent;j −RTold;j

�:t−1j ; [S5]

with LTold,j and LTrecent,j the left tails (fifth percentile) and RTold,jand RTrecent,j the right tails (95th percentile) of the distributionof floristic temperature in plot j in the first (old) and second(recent) survey, respectively. Thus, positive values of Cj reflectthermophilization due to a decrease of relatively cold-adaptedspecies, and positive values of Wj reflect thermophilization dueto an increase of the frequency of more warm-adapted species(Fig. S4).To estimate a summary Th, C, and W across all studies, ac-

counting for the different numbers of plots between studies, weused a mixed-effect model assuming study region-level values tocome from a common distribution ∼N(μ,τ) with μ the grandmean and τ the superpopulation SD for each of the 500 samples.The parameter μ is thus the summary metric and τ expressesthe heterogeneity across studies; this was calculated using an

De Frenne et al. www.pnas.org/cgi/content/short/1311190110 2 of 13

intercept-only mixed-effect model with study region as a ran-dom term using the lme function in the nlme package (19, 37)for all data, and for Europe and North America separately.Finally, the relationships between forest cover changes and

thermophilization, referred to as shifts in the mean, left, and righttails of the distribution of floristic temperatures, were assessedthrough mixed-effect modeling with study region as a randomeffect term using the lme function in the nlme package (19, 37).For every bootstrap sample (thus, n = 500), we calculated theslope and intercept of the relationship between plot-level forestcanopy cover changes and thermophilization (the mean, fifth, or95th percentile shifts). We tested random intercept models aswell as random intercept and slope models (37): the difference inAkaike Information Criterion (AIC) between the more complexmodels (random intercept and slope) and the random interceptmodels of the mean thermophilization was ΔAIC = −1.5, andthus we used the simpler random intercept model for all analy-ses. We performed similar analyses using mixed-effect models toassess the relationship between thermophilization and the re-gion-specific rate of temperature change. The significance of thethermophilization and relationships with forest canopy cover andtemperature changes was tested by Student t tests on the boot-strapped values (500 samples) (34).

SI ResultsSensitivity Analyses. We ran a wide array of sensitivity analysesboth on the thermophilization patterns as well as on the re-lationship with forest canopy cover and modeling approaches:applying various climatic periods (annual vs. growing season) andclimatic variables (excluding or including precipitation), studyarea extents (including all grid cells of the whole continent or onlyusing a buffer of 3,000 km around the study areas as environ-mental background data), and shapes of the thermal responsecurves (Gaussian response curves vs. GAMs) resulted in con-sistent thermophilization patterns and relationships with forestcanopy cover changes. Additionally, repeating the resamplingprocedure after a new subset of 20% of all species in each con-tinent had been randomly removed from the dataset in each ofthe 500 bootstrap samples again resulted in robust thermophili-zation patterns and relationships with forest canopy cover changes(Tables S4 and S5). Dividing the forest cover change responseratios by the time interval between the two survey periods (SIMaterials and Methods) also resulted in similar results (Table S5),

which clearly indicates that the thermophilization patterns are notdetermined by the trends of only few species and that the emergingpatterns are robust to different methods of inferring species’ tem-perature preferences and expressing forest cover change.

Thermophilization vs. Rate of Temperature Change. We found apositive relationship between the mean thermophilization andthe region-specific growing season temperature change (meanslope of the mixed-effect models with random term study regionacross 500 samples: 0.07, 95% CIs of the slopes: [0.05, 0.09], P <0.001). Similarly, the decrease of cold-adapted species and in-crease of warm-adapted species was higher in areas with higherrates of warming (same method but with the shift of the fifth and95th percentiles of the plot-level distribution of floristic tem-peratures: mean slope: 0.27 [CI: 0.23, 0.31], P < 0.001, and 0.05[0.03, 0.07], P < 0.001, respectively). Additionally, there was norelationship between the rate of warming and forest canopychanges (mixed-effect model between forest cover change andgrowing-season temperature change, t = 0.25, n = 854 plots, P =0.805), which indicates that forest canopy closure was not asso-ciated to areas with higher rates of recent climate warming inour dataset.

Increase of Cold-Adapted American Species (Fig. 2C). Surprisingly,though there has been a significant increase in the dominance ofwarm-adapted species across all plots, in eastern North Americanforests, both warm-adapted and cold-adapted species have in-creased as indicated by the negative shift of the left tail andpositive shift of the right tail of the distribution of floristictemperatures, respectively (Fig. 2C). Thus, in the North Amer-ican plots, the distribution of floristic temperatures has becomewider over time, and might be related to continuous immigrationof new species (i.e., overall increase in species richness), whichdoes not occur in Europe. There was no systematic change inspecies richness over time in Europe (24), but an increase inherb-layer plant diversity in North American forests: the meanspecies number per European plot amounted to 17.2 in old and17.6 in recent surveys, whereas species richness per plot in-creased from 15.1 species in the old surveys to 16.5 species in therecent North American surveys. Accordingly, 60% of the 50 mostfrequent American species in our database increased in fre-quency, whereas 40% decreased between the two surveys (seeTables S2 and S3 for species information).

1. Baeten L, Hermy M, Van Daele S, Verheyen K (2010) Unexpected understoreycommunity development after 30 years in ancient and post-agricultural forests. J Ecol98(6):1447–1453.

2. Baeten L, et al. (2009) Herb layer changes (1954–2000) related to the conversion ofcoppice-with-standards forest and soil acidification. Appl Veg Sci 12(2):187–197.

3. Hédl R, Kopecky M, Komarek J (2010) Half a century of succession in a temperateoakwood: From species-rich community to mesic forest. Divers Distrib 16(2):267–276.

4. Flinn KM, Vellend M (2005) Recovery of forest plant communities in post-agriculturallandscapes. Front Ecol Environ 3(5):243–250.

5. Dupouey JL, Dambrine E, Laffite JD, Moares C (2002) Irreversible impact of past landuse on forest soils and biodiversity. Ecology 83(11):2978–2984.

6. Verheyen K, Honnay O, Motzkin G, Hermy M, Foster DR (2003) Response of forestplant species to land-use change: A life-history trait-based approach. J Ecol 91(4):563–577.

7. De Frenne P, et al. (2011) Interregional variation in the floristic recovery of post-agricultural forests. J Ecol 99(2):600–609.

8. Franklin J (2009) Mapping Species Distributions: Spatial Inference and Prediction(Cambridge Univ Press, Cambridge, UK).

9. Peterson AT, et al. (2011) Ecological Niches and Geographic Distributions (PrincetonUniv Press, Princeton, NJ).

10. Hultén E, Fries M (1986) Atlas of North European Vascular Plants: North of the Tropicof Cancer (Koeltz Scientific, Königstein, Germany).

11. Meusel H, Jäger E (2011) Comparative Chorology of the Central European Flora(Fischer, Jena, Germany).

12. Kartesz JT; The Biota of North America Program (BONAP) (2013) North AmericanPlant Atlas. Floristic Synthesis of North America, Version 1.0 (BONAP, Chapel Hill, NC).

13. R Core Team (2012) R: A Language and Environment for Statistical Computing (RFoundation for Statistical Computing, Vienna).

14. Hijmans RJ, Cameron SE, Parra JL, Jones PG, Jarvis A (2005) Very high resolutioninterpolated climate surfaces for global land areas. Int J Climatol 25(15):1965–1978.

15. Barbet-Massin M, Jiguet F, Albert CH, Thuiller W (2012) Selecting pseudo-absences forspecies distribution models: How, where and how many? Methods Ecol Evol 3(2):327–338.

16. Woodward FI (1987) Climate and Plant Distribution (Cambridge Univ Press,Cambridge, UK).

17. Rodríguez-Sánchez F, De Frenne P, Hampe A (2012) Uncertainty in thermal tolerancesand climatic debt. Nature Clim Change 2(9):636–637.

18. McInerny GJ, Purves DW (2011) Fine-scale environmental variation in speciesdistribution modelling: Regression dilution, latent variables and neighbourly advice.Methods Ecol Evol 2(3):248–257.

19. Pinheiro J, et al. (2012) nlme: Linear and Nonlinear Mixed Effects Models. R packageversion 3.1-104 (R Foundation for Statistical Computing, Vienna).

20. Hastie T (2011) gam: Generalized Additive Models. R package version 1.06.2 (RFoundation for Statistical Computing, Vienna).

21. Geiger R, Aron RH, Todhunter P (2009) The Climate Near the Ground (Rowman &Littlefield, Plymouth, UK), 7th Ed.

22. Chen J, et al. (1999) Microclimate in forest ecosystem and landscape ecology.Bioscience 49(4):288–297.

23. Norris C, Hobson P, Ibisch PL (2012) Microclimate and vegetation function asindicators of forest thermodynamic efficiency. J Appl Ecol 49(3):562–570.

24. Verheyen K, et al. (2012) Driving factors behind the eutrophication signal inunderstorey plant communities of deciduous temperate forests. J Ecol 100(2):352–365.

25. Hansen J, Ruedy R, Sato M, Lo K (2010) Global surface temperature change. RevGeophys 48:RG4004.

De Frenne et al. www.pnas.org/cgi/content/short/1311190110 3 of 13

26. Gottfried M, et al. (2012) Continent-wide response of mountain vegetation to climatechange. Nature Clim Change 2(2):111–115.

27. Graae BJ, et al. (2012) On the use of weather data in ecological studies alongaltitudinal and latitudinal gradients. Oikos 121(1):3–19.

28. Lenoir J, et al. (2013) Local temperatures inferred from plant communities suggeststrong spatial buffering of climate warming across Northern Europe. Glob ChangeBiol 19(5):1470–1481.

29. Lennon JJ, Koleff P, Greenwood JJD, Gaston KJ (2001) The geographical structure ofBritish bird distributions: Diversity, spatial turnover and scale. J Anim Ecol 70(6):966–979.

30. Oksanen J, et al. (2012) vegan: Community Ecology Package. R package version 2.0-5(R Foundation for Statistical Computing, Vienna).

31. Canty A, Ripley B (2012) boot: Bootstrap R (S-Plus) Functions. R package version 1.3-6(R Foundation for Statistical Computing, Vienna).

32. Devictor V, et al. (2012) Differences in the climatic debts of birds and butterflies ata continental scale. Nature Clim Change 2(2):121–124.

33. Bertrand R, et al. (2011) Changes in plant community composition lag behind climatewarming in lowland forests. Nature 479(7374):517–520.

34. Maggini R, et al. (2011) Are Swiss birds tracking climate change? Detectingelevational shifts using response curve shapes. Ecol Modell 222(1):21–32.

35. Feeley KJ (2012) Distributional migrations, expansions, and contractions of tropicalplant species as revealed in dated herbarium records. Glob Change Biol 18(4):1335–1341.

36. Zhu K, Woodall CW, Clark JS (2012) Failure to migrate: Lack of tree range expansionin response to climate change. Glob Change Biol 18(3):1042–1052.

37. Zuur AF, Ieno EN, Walker NJ, Saveliev AA, Smith GM (2009) Mixed Effects Models andExtensions in Ecology with R (Springer, Berlin).

0.4

0.6

0.8

1

-0.2

0

0.2

-0.8

-0.6

-0.4

-10 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29

Study region

Fore

stca

nopy

cove

r cha

nge

(log(

cove

r/c

over

))re

cent

old

More open

More dense

Fig. S1. Forest cover change in each of the 19 regions (854 plots) with canopy information. The dashed line indicates no change (response ratio = 0); negativeresponse ratios indicate forest plots that became more open over time; and positive response ratios indicate plots where the forest became denser over time.Study region numbers refer to sites in Figs. 1 and 2 and Table S1. Error bars denote 95% confidence intervals.

De Frenne et al. www.pnas.org/cgi/content/short/1311190110 4 of 13

lllllllllllllllllllllllllllllllllllllllllll

lllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllll

llllllllllllllllllllllllllllllllllllllllllllllllll

llllllllllllllllllllllllllllllllllllllllllllllllll

llllllllllllllllllllllllllllllllllllllllllllll

lllllllllllllllllllll

lllllllllllllllllllllllllllllllllllllllllllllllll

llllllllllllllllllllllllllllllllllllllllll

lllllllllllllllllll

lllllllllllllllllllllllllllllllllllll

llllllllllllllllllllll

lllllllllllllllllllllllllllllllllllllllllllllll

lllllllllllllllllll

lllllllllllllllllllllllllll

llllllllllllllllllllllllllllllllll

lllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllll

llllllllllllllll

llllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllll

llllllllllllllllllllllllll

lllllllllllllllllllll

llllllllllllllllllllllllllllllllllllllllllllllllllllllllll

lllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllll

llllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllll

llllllllllllll

lllllllllllllllll

llllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllll

llllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllll

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

Growing−season temperature change (°C.decade−1)

Ann

ual t

empe

ratu

re c

hang

e (°

C.d

ecad

e−1)

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Fig. S2. Relationship between the rate of growing season (April to September) and annual (January to December) temperature change between the 5 ypreceding the old and 5 y preceding the recent survey per decade. The gray dashed line depicts the fitted linear regression.

De Frenne et al. www.pnas.org/cgi/content/short/1311190110 5 of 13

Species’ distribution ranges

Species’ temperature

Ecological niche modelling

p ppreferences

Sample one temperature value per species

Distribution of

Species lists per plot and survey

period

temperatures per plot and survey period

Calculate

Mean floristic temperature (FTj), left tail (LTj) and

right tail (RTj) per plot and survey period

Recent minus old surveydivided by time

Thermophilisation Th, decrease of cold-adapted

i fspecies C, increase of warm-adapted species W

Mixed-effect model withrandom term study region

Data on changes in forest canopy cover

Effect of forest canopy cover changes

Fig. S3. Flowchart of the thermophilization analyses. The whole procedure indicated by orange arrows was bootstrapped 500 times to account for variabilityand uncertainty in species’ temperature preferences.

De Frenne et al. www.pnas.org/cgi/content/short/1311190110 6 of 13

Temperature (°C)

A

Species 1

Species 2

Species 3

0 10 20 30

1.0

0.8

0.6

0.4

0.2

0.0

Pro

babi

lity

ofpr

esen

ce

C

0 10 20 30

1.0

0.8

0.6

0.4

0.2

0.0

Pro

babi

lity

D

0 10 20 30

1.0

0.8

0.6

0.4

0.2

0.0

+ warm-adapted sp.

- cold-adapted sp.

B

+ warm-adapted sp.

- cold-adapted sp.

Pro

babi

lity

Large shiftLT

Small shiftLT

Large shiftRT

Large shiftRT

Mean shift

Mean shift

Temperature (°C)0 10 20 30

1.0

0.8

0.6

0.4

0.2

0.0

Pro

babi

lity

Community time 1Community time 2

Mean shift

Small shift RTLarge shiftLT

Fig. S4. Three hypothetical thermophilization trajectories between time 1 and time 2. (A) Three hypothetical species are described: species 1 is relatively morecold-adapted, species 2 is intermediate, and species 3 is relatively more warm-adapted, as indicated by the relative position of the thermal response curves. (B)Increase of warm-adapted species. The hypothetical community consists of species 1 at time 1 and species 1 and 2 at time 2; this results in thermophilizationbetween time 1 and 2 because the more warm-adapted species 2 is gained, and the resulting shift of the right tail (RT; 95th percentile) is much larger than theshift of the left tail (LT; fifth percentile). (C) Concurrent decline of cold-adapted species and increase of warm-adapted species. The hypothetical communityconsists of species 1 and 2 at time 1 and species 2 and 3 at time 2; this results in thermophilization between time 1 and 2 because the more cold-adapted species1 is lost from the community, whereas the more warm-adapted species 3 is gained. Both the LT and RT shift considerably to the right (this option is displayed inFig. 1D). (D) Decline of cold-adapted species. The hypothetical community consists of species 1 and 2 at time 1 and species 2 at time 2; this results in ther-mophilization because the more cold-adapted species 1 is lost from the community, and thus the LT of the distribution of floristic temperatures shifts con-siderably to the right, whereas the shift of the RT is much smaller. The green arrows represent the shift of the LT, mean, and RT.

De Frenne et al. www.pnas.org/cgi/content/short/1311190110 7 of 13

Table

S1.

Characteristicsofthe29

studyregions

Number

Studyregion,

country

Latitude,

°NLo

ngitude,

°EPlotsize

,m

2No.ofplots

PorSP

†

Yea

r(s)

of

old

survey

Yea

r(s)

of

recentsurvey

Forest

canopy

cove

rdata

Temperature

chan

ge,

°C·decad

e−1

Ref.

1Gau

me,

BE

49.6

5.5

50–40

043

SP19

50–19

5520

08Tree

andshrub

cove

r0.10

1

2Cen

tral

Flan

ders,BE

51.2

3.3

100–

200

47SP

1977

–19

8020

09Tree

andshrub

cove

r0.55

2

3Zo

erselbos,BE

51.2

4.7

100

17SP

1982

2008

Tree

andshrub

cove

r0.49

*

4Heren

bossen

,BE

51.1

4.8

196

111

SP19

8020

04Notav

ailable

0.56

35

VorteBossen

,BE

51.1

3.4

150

26SP

1977

–19

8019

98Notav

ailable

0.30

46

Mee

rdaa

lwoud,BE

50.8

4.7

125–

225

21SP

1954

2000

Tree

andshrub

cove

r0.12

5

7Florenne,

BE

50.2

4.7

100

58SP

1957

2005

Notav

ailable

0.29

68

Tournibus,BE

50.3

4.6

100

139

SP19

6720

05Tree

andshrub

cove

r0.26

6

9Dalby,

SE55

.713

.31–

2(16forcanopy)

74SP

1935

2002

Tree

andshrub

cove

r0.06

7

10Tu

llgarn,SE

58.1

17.1

100

127

SP19

7120

03Notav

ailable

0.27

*11

Elbe-W

eser,DE

53.6

910

0–40

050

SP19

86–19

8920

08Tree

andshrub

cove

r0.51

8

12D� ev

ín,CZ

48.9

16.6

100–

1,00

050

SP19

53–19

6320

02–20

03Tree

andshrub

cove

r0.10

9

13Milo

vice

Wood,CZ

48.8

16.7

500

46SP

1953

–19

5420

06Tree

andshrub

cove

r0.11

10

14RychlebskéMountains,CZ

50.3

17.1

315

21SP

1941

–19

4319

98–19

99Tree

andshrub

cove

r0.12

11

15W

ytham

Woods,UK

51.8

−1.3

100

49P

1974

1999

Tree

andshrub

cove

r0.35

12

16Göttingen

,DE

51.5

10.1

100–

400

42P

1980

2001

Tree

andshrub

cove

r0.69

13

17Milí� co

vW

ood,CZ

5014

.550

–62

519

SP19

8620

08–20

09Tree

andshrub

cove

r0.44

14

18Sw

itze

rlan

d,CH

47.3

7.8

100–

400

37SP

1940

–19

6519

98Tree

andshrub

cove

r0.11

15

19Hirson/Saint-Michel,FR

49.9

4.1

500–

800

22SP

1956

–19

6519

96–19

98Tree

andshrub

cove

r0.26

*

20Andigny,

FR50

3.6

500–

800

19SP

1957

–19

6319

95–19

96Tree

andshrub

cove

r0.18

*

21Sp

eulderbos,NL

52.3

5.7

100–

250

27SP

1957

–19

5919

87–19

88Tree

andshrub

cove

r0.18

*

22La

dyPa

rkW

ood,UK

51.7

−2.7

3235

P19

7920

09Notav

ailable

0.32

1623

Munich,DE

48.3

11.7

100

125

P19

8620

03Tree

andshrub

cove

r0.58

17

24County

Kerry,IE

52.2

−9.5

1616

P19

9120

11Notav

ailable

0.13

*25

Smoky

Mountains,TN

,US

35.7

−83

.51,00

018

P19

77–19

7819

95Notav

ailable

0.24

1826

Fernow,W

V,US

39.1

−79

.810

×1m

214

P19

9120

03Notav

ailable

0.02

19

De Frenne et al. www.pnas.org/cgi/content/short/1311190110 8 of 13

Table

S1.

Cont.

Number

Studyregion,

country

Latitude,

°NLo

ngitude,

°EPlotsize

,m

2No.ofplots

PorSP

†

Yea

r(s)

of

old

survey

Yea

r(s)

of

recentsurvey

Forest

canopy

cove

rdata

Temperature

chan

ge,

°C·decad

e−1

Ref.

27MontMég

antic,

QC,CA

45.5

−71

.280

017

SP19

70–19

7120

11Tree

andshrub

cove

r0.25

*

28Duke

s,MI,US

46.0

−87

.21

74P

1978

–19

8020

02–20

07Notav

ailable

0.36

2029

NorthernW

I,US

46.8

−91

20×1m

266

SP19

49–19

5020

00–20

04Notav

ailable

0.11

21

Theregionnumbersco

rrespondto

those

inFigs.1an

d2.

Thetemperature

chan

gebetwee

nsurvey

sisthedifference

inmea

ngrowing-sea

sontemperature

(ApriltoSe

ptember)betwee

nthe5yprecedingthe

old

and5yprecedingtherecentsurvey

sex

pressed

per

decad

e.Countryan

dstateco

des:BE,

Belgium;CA,Can

ada;

CH,Sw

itze

rlan

d;CZ,

Cze

chRep

ublic;DE,

German

y;FR

,Fran

ce;IE,Irelan

d;MI,Michigan

;QC,

Queb

ec;SE

,Sw

eden

;TN

,Te

nnessee;

UK,United

Kingdom;US,

United

States;W

I,W

isco

nsin;W

V,W

estVirginia.

*Unpublished

.†P,

perman

ent;

SP,semiperman

ent.

1.Verstraeten

G,et

al.(201

3)Te

mporalch

anges

inforest

plantco

mmunitiesat

differentsite

types.ApplVeg

Sci16

(2):23

7–24

7.2.

Bae

tenL,

HermyM,Van

Dae

leS,

Verhey

enK(201

0)Unex

pectedunderstoreyco

mmunitydev

elopmen

tafter30

yearsin

ancien

tan

dpost-agricu

lturalforests.JEcol98

(6):14

47–14

53.

3.CornelisJ,Rombouts

L,HermyM

(200

7)Chan

ges

intheve

getationofthe‘Heren

bossen

’in

Hulshoutsince

1980

.Are

effectsofdesiccation,acidificationan

deu

trophicationnoticeab

le?Natuur.Fo

cus6(1):4–10

.4.

Lameire

S,HermyM,Honnay

O(200

0)Tw

odecad

esofch

angein

thegroundve

getationofamixed

deciduousforest

inan

agricu

lturallandscap

e.JVeg

Sci11

(5):69

5–70

4.5.

Bae

tenL,

etal.(200

9)Herblaye

rch

anges

(195

4–20

00)relatedto

theco

nve

rsionofco

ppice-with-standardsforest

andsoilacidification.ApplVeg

Sci12

(2):18

7–19

7.6.

Van

Calster

H,et

al.(200

8)Divergingeffectsofove

rstoreyco

nve

rsionscen

ariosontheunderstoreyve

getationin

aform

erco

ppice-with-standardsforest.Fo

rEcolMan

age25

6(4):519

–52

8.7.

VonOheimbG,Brunet

J(200

7)DalbySö

derskogrevisited:Lo

ng-term

vegetationch

anges

inasouth

Swed

ishdeciduousforest.ActaOecol31

(2):22

9–24

2.8.

Naa

fT,

WulfM

(201

0)Hab

itat

specialists

andgen

eralists

drive

homogen

izationan

ddifferentiationoftemperateforest

plantco

mmunitiesat

theregional

scale.

BiolConserv

143(4):848

–85

5.9.

Kopecky

M,Héd

lR,Sz

aboP(201

3)Non-ran

dom

extinctionsdominateplantco

mmunitych

anges

inab

andoned

coppices.JApplEcol50

(1):79

–87

.10

.Héd

lR,Kopecky

M,Komarek

J(201

0)Halfacentury

ofsuccessionin

atemperateoak

wood:From

species-rich

communityto

mesic

forest.DiversDistrib

16(2):26

7–27

6.11

.Héd

lR(200

4)Veg

etationofbee

chforestsin

theRychlebskeMountains,Cze

chRep

ublic:Re-inspectedafter60

yearswithassessmen

tofen

vironmen

talch

anges.Plan

tEcol17

0(2):243

–26

5.12

.KirbyKJ,Morecroft

MD

(201

0)Th

eFlowersoftheFo

rest.W

ytham

Woods:Oxford’sEcological

Laboratory,ed

sSa

villPS

,Pe

rrinsCM,KirbyKJ,Fisher

N(O

xford

Univ

Press,Lo

ndon),pp75

–89

.13

.Diersch

keH

(200

9)Veg

etationdyn

amicsofafencednea

r-naturalcalcareo

usbee

chforest—co

mparisonofve

getationrelevé

es19

80an

d20

01.Fo

rstarchiv

80:143

–15

0.14

.Pe

t� rík

P(200

9)Chan

ges

inflora

innaturalheritag

eMilí� co

vský

forest

andpondsin

Prag

ueaftermore

than

20ye

ars.Muze

um

aSo

u� ca

snost

Roztoky

uPrah

y24

:133

–15

2.15

.W

alther

GR,Grundman

nA

(200

1)Tren

dsofve

getationch

angein

collinean

dsubmontaneclim

axforestsin

Switze

rlan

d.BullGeo

botInst

ETH

67:3–12

.16

.Bae

tenL,

etal.(201

2)Distinguishingbetwee

nturnove

ran

dnestednessin

thequan

tificationofbiotichomogen

ization.BiodiversConserv

21(6):13

99–14

09.

17.Bernhardt-Römerman

nM,et

al.(200

7)Lo

ng-term

effectsofnitrogen

-dep

ositiononve

getationin

adeciduousforest

nea

rMunich,German

y.ApplVeg

Sci10

(3):39

9–40

6.18

.JenkinsMA,Jose

S,W

hitePS

(200

7)Im

pacts

ofan

exoticdisea

sean

dve

getationch

angeonfolia

rcalcium

cyclingin

Appalachianforests.EcolAppl17

(3):86

9–88

1.19

.Gilliam

FS,Hocken

berry

AW

,Adam

sMB(200

6)Effectsofatmospheric

nitrogen

dep

ositionontheherbaceo

uslaye

rofacentral

Appalachianhardwoodforest.JTo

rrey

BotSo

c13

3(2):240

–25

4.20

.W

oodsKD,Hicks

DJ,SchultzJ(201

2)Lo

sses

inunderstory

diversity

ove

rthreedecad

esin

anold-growth

cool-temperateforest

inMichigan

,USA

.Can

JRes

42(3):53

2–54

9.21

.Rooney

TP,W

iegman

nSM

,RogersDA,W

allerDM

(200

4)Bioticim

pove

rishmen

tan

dhomogen

izationin

unfrag

men

tedforest

understory

communities.Conserv

Biol18

(3):78

7–79

8.

De Frenne et al. www.pnas.org/cgi/content/short/1311190110 9 of 13

Table S2. Overview of the 50 most frequent European understory species with their life form, average frequency (SD betweenbrackets) in the recent and old surveys, and the change in frequency compared with the old surveys (red, decreases; blue, increases)

No. Species Life form*Average frequencyrecent survey, %

Average frequencyold survey, % Change in frequency, %

13 Acer pseudoplatanus 0.9)6.1(6.41)2.2(6.32ST38 Aegopodium podagraria 0.0)1.2(1.41)0.2(1.41oF40 Ajuga reptans Fo 13.9 (1.6) 15.5 (1.1) −1.61 Anemone nemorosa Fo 45.2 (2.8) 46.3 (2.9) −1.119 Arum maculatum 5.2)1.2(7.81)2.2(1.12oF39 Asarum europaeum Fo 14.0 (1.6) 16.1 (1.8) −2.122 Athyrium filix-femina 9.2)2.1(5.61)2.1(3.91F17 Brachypodium sylvaticum 0.5)6.1(3.71)0.2(3.22G18 Carex sylvatica 6.1)0.1(4.02)2.1(0.22G34 Carpinus betulus 8.4)0.1(0.01)6.1(8.41ST33 Circaea lutetiana 6.3)0.1(5.21)9.0(1.61oF12 Convallaria majalis Fo 23.7 (2.1) 25.7 (2.1) −2.041 Corylus avellana 4.7)7.0(6.6)1.1(9.31ST47 Crataegus monogyna 4.3)2.1(1.8)8.1(5.11ST9 Deschampsia cespitosa 5.0)3.1(5.52)4.1(0.62G21 Dryopteris carthusiana 8.9)6.0(9.9)2.1(7.91F24 Dryopteris dilatata 5.41)5.0(2.4)3.1(7.81F14 Dryopteris filix-mas 7.0)7.1(5.22)9.1(2.32F49 Euonymus europaeus 4.6)7.0(8.4)5.1(2.11ST26 Fagus sylvatica 7.4)9.0(3.31)1.1(0.81ST6 Fraxinus excelsior 8.41)1.1(3.41)1.2(2.92ST45 Galeopsis tetrahit 9.7)4.0(6.4)1.1(5.21oF29 Galium odoratum 1.1)0.1(7.51)2.1(7.61oF5 Geum urbanum 6.4)8.1(9.62)1.2(5.13oF36 Glechoma hederacea 8.2)7.0(6.11)9.0(4.41oF8 Hedera helix 3.9)0.1(8.61)3.1(1.62oF25 Hyacinthoides nonscripta Fo 18.3 (1.9) 18.7 (2.0) −0.42 Lamium galeobdolon Fo 38.0 (2.3) 38.8 (2.3) −0.727 Lonicera periclymenum Fo 17.2 (1.4) 19.9 (1.8) −2.731 Maianthemum bifolium Fo 16.6 (1.5) 18.4 (1.4) −1.846 Melica uniflora 0.1)9.0(8.01)9.0(8.11G11 Mercurialis perennis Fo 23.9 (1.7) 30.2 (2.3) −6.310 Milium effusum 1.2)4.1(6.22)8.1(8.42G44 Moehringia trinervia 8.2)6.0(7.01)9.0(5.31oF7 Oxalis acetosella 4.3)6.1(2.42)7.1(5.72oF20 Paris quadrifolia Fo 20.2 (2.1) 22.0 (1.9) −1.823 Poa nemoralis G 18.9 (2.0) 22.6 (2.0) −3.843 Poa trivialis 3.4)8.0(3.9)0.1(6.31G4 Polygonatum multiflorum Fo 34.6 (2.3) 35.9 (2.3) −1.332 Pteridium aquilinum F 16.6 (1.3) 17.2 (1.4) −0.730 Quercus robur 7.6)9.0(0.01)4.1(7.61ST42 Ranunculus auricomus agg. Fo 13.8 (1.7) 13.9 (1.3) −0.128 Ranunculus ficaria Fo 16.8 (1.2) 18.2 (1.3) −1.43 Rubus fruticosus agg. 6.3)6.1(4.33)9.1(0.73oF37 Sorbus aucuparia 3.4)0.1(9.9)4.1(2.41ST15 Stachys sylvatica 4.8)0.1(2.41)2.2(5.22oF50 Stellaria holostea 1.1)7.0(1.01)8.0(1.11oF48 Ulmus glabra 8.4)7.0(6.6)7.1(5.11ST16 Urtica dioica 3.4)0.1(0.81)3.1(4.22oF35 Viola reichenbachiana/riviniana Fo 14.8 (1.1) 15.2 (1.0) −0.4

No. is the ranking (1–50) according to the average frequency in the recent survey (high to low).*TS, tree and shrub species that occurred in the understory; Fo, forbs; G, graminoids; F, ferns.

De Frenne et al. www.pnas.org/cgi/content/short/1311190110 10 of 13

Table S3. Overview of the 50 most frequent North American understory species with their life form, average frequency (SD betweenbrackets) in the recent and old surveys, and the change in frequency compared with the old surveys (red, decreases; blue, increases)

m*rofefiLseicepS.oNAverage frequencyrecent survey, %

Average frequencyold survey, % Change in frequency, %

23 Acer pensylvanicum 5.8)2.3(5.8)6.3(9.61ST24 Acer rubrum 5.9)6.2(3.6)7.3(9.51ST20 Acer saccharum 9.7)4.2(5.9)3.3(5.71ST40 Adiantum pedatum 5.0)2.3(1.01)4.3(6.01F48 Amphicarpaea bracteata Fo 9.0 (3.0) 9.5 (2.7) −0.513 Anemone quinquefolia 7.3)2.9(6.02)9.01(3.42oF9 Aralia nudicaulis Fo 28.6 (7.7) 33.3 (12.4) −4.816 Arisaema triphyllum 4.7)8.3(3.51)5.5(8.22oF6 Athyrium angustum 8.31)7.4(5.71)7.8(2.13F2 Carex 5.9)1.01(2.13)4.41(7.04G.ps29 Caulophyllum thalictroides 9.6)8.1(3.6)4.3(2.31oF49 Circaea canadensis 1.2)1.3(9.6)5.3(0.9oF14 Claytonia caroliniana Fo 24.3 (8.7) 30.7 (12.3) −6.318 Clintonia borealis Fo 20.6 (5.5) 25.4 (7.3) −4.821 Dendrolycopodium obscurum C 17.5 (6.7) 19 (7.7) −1.650 Dicentra canadensis Fo 9.0 (4.0) 21.7 (9.7) −12.712 Dryopteris carthusiana 6.01)4.6(3.41)5.01(9.42F7 Dryopteris intermedia 3.5)1.7(4.52)7.8(7.03F3 Erythronium americanum 6.2)2.31(8.23)3.21(4.53oF19 Eurybia macrophylla Fo 18.5 (8.3) 31.7 (11.1) −13.215 Galium triflorum Fo 24.3 (8.3) 29.1 (9.4) −4.842 Gaultheria procumbens TS 10.1 (4.5) 15.3 (6.9) −5.325 Gymnocarpium dryopteris 3.6)4.2(5.9)8.3(9.51F11 Huperzia lucidula C 25.4 (6.5) 36.5 (9.1) −11.136 Laportea canadensis 6.2)4.2(9)2.2(6.11oF43 Liriodendron tulipifera 7.3)6.2(3.6)0.3(1.01ST1 Maianthemum canadense 6.2)7.31(7.93)4.31(3.24oF8 Maianthemum racemosum ssp. racemosum 5.8)6(7.12)8.7(2.03oF30 Mitchella repens TS 13.2 (3.4) 23.3 (6.4) −10.110 Oryzopsis asperifolia 3.5)2.9(6.02)3.11(9.52G31 Osmorhiza claytonii Fo 13.2 (4.2) 19 (6) −5.837 Oxalis montana Fo 11.1 (3.8) 12.7 (3.2) −1.644 Parathelypteris noveboracensis F 10.1 (2.3) 12.2 (3.1) −2.133 Phegopteris connectilis 2.3)2.2(9)3.3(2.21F4 Polygonatum pubescens 3.5)5.8(2.03)1.9(4.53oF34 Polystichum acrostichoides 3.5)2(9.6)5.2(2.21F38 Prunus serotina 2.4)1.3(9.6)1.3(1.11ST22 Pteridium aquilinum F 17.5 (7.2) 21.2 (8.9) −3.745 Pyrola elliptica Fo 10.1 (4.5) 15.9 (6.8) −5.846 Quercus rubra 5.9)2.0(5.0)2.3(1.01ST41 Sanguinaria canadensis 1.2)8.2(5.8)4.3(6.01oF32 Schizachne purpurascens 4.7)6.2(8.5)9.5(2.31G35 Smilax rotundifolia 6.2)8.2(5.9)6.3(2.21oF17 Streptopus lanceolatus var. lanceolatus Fo 21.7 (6) 37 (10.8) −15.347 Taraxacum officinale 5.9)0(0)3.4(5.9oF5 Trientalis borealis 8.4)8.9(5.72)9.11(3.23oF27 Trillium grandiflorum Fo 14.3 (6.1) 17.5 (6.7) −3.239 Uvularia sessilifolia Fo 11.1 (4.1) 19.6 (7.3) −8.528 Viola blanda Fo 14.3 (6.1) 15.3 (6) −1.126 Viola pubescens 5.0)3.5(8.41)4.5(3.51oF

No. is the ranking (1–50) according to the average frequency in the recent survey (high to low).*TS, tree and shrub species that occurred in the understory; Fo, forbs; G, graminoids; F, ferns; C, club mosses.

De Frenne et al. www.pnas.org/cgi/content/short/1311190110 11 of 13

Table S4. Sensitivity analyses of the mean thermophilization (Th), the shift of cold-adaptedspecies (fifth percentiles, C), and warm-adapted species (95th percentiles, W)

Temperature onlyTemperature and

precipitation

Mean 95% CI Mean 95% CI

(A) Gauss response curves Th 0.033 0.030 0.035 0.041 0.038 0.044C 0.027 0.023 0.031 0.017 0.011 0.023W 0.035 0.032 0.038 0.060 0.057 0.064

(B) GAM response curves Th 0.027 0.025 0.030 0.043 0.039 0.048C −0.002 −0.006 0.003 −0.009 −0.016 −0.003W 0.040 0.037 0.043 0.071 0.066 0.075

(C) Annual climatic variables Th 0.036 0.033 0.039 0.050 0.046 0.053C 0.014 0.008 0.020 0.003 −0.004 0.010W 0.044 0.040 0.048 0.062 0.058 0.067

(D) 3,000-km buffer Th 0.029 0.027 0.031 0.056 0.053 0.060C 0.013 0.008 0.018 0.029 0.024 0.034W 0.037 0.034 0.040 0.078 0.074 0.082

(E) 20% sp. randomly removed Th 0.032 0.030 0.035 0.042 0.039 0.046C 0.010 0.005 0.015 −0.006 −0.013 0.001W 0.046 0.042 0.049 0.070 0.065 0.075

Results from 500 resampled species’ temperature preferences; 95% confidence intervals are based on thosesamples. The data within the black box were used in the main text. A wide array of modeling techniques to inferspecies’ temperature preferences resulted in consistent thermophilization patterns by applying varying climaticvariables (left column: temperature only, right column: including both temperature as well as precipitation),modeling approaches (A, Gauss response curve; B, flexible-shape response curves using GAMs), climatic periods(whole year instead of growing season, C), and study area extents (3,000-km buffer around study regions instead ofwhole continent, D). (E) Randomly excluding 20% of the species in each of the 500 samples also resulted in robustthermophilization patterns.

De Frenne et al. www.pnas.org/cgi/content/short/1311190110 12 of 13

Table S5. Sensitivity analyses of the effects of forest canopy cover changes on the mean thermophilization (Th), and on the shift ofcold-adapted species (fifth percentiles, C) and warm-adapted species (95th percentiles, W)

noitatipicerpdnaerutarepmeTylnoerutarepmeT

Slope 95% conf. interval Slope 95% conf. interval

(A) Gauss response curves Th −0.0027 −0.0052 −0.0001 * −0.0073 −0.0108 −0.0037 ***C 0.0042 −0.0007 0.0091 ns −0.0071 −0.0136 −0.0006 *W −0.0117 −0.0159 −0.0076 *** −0.0170 −0.0233 −0.0108 ***

(B) GAM response curves Th −0.0064 −0.0093 −0.0034 *** −0.0392 −0.0436 −0.0348 ***C −0.0125 −0.0183 −0.0066 *** −0.0584 −0.0674 −0.0493 ***W −0.0127 −0.0171 −0.0083 *** −0.0463 −0.0552 −0.0375 ***

(C) Annual climatic variables Th −0.0039 −0.0076 −0.0002 * −0.0092 −0.0130 −0.0053 ***C 0.0007 −0.0063 0.0076 ns 0.0083 0.0006 0.0159 *W −0.0099 −0.0154 −0.0044 *** −0.0250 −0.0309 −0.0190 ***

(D) 3,000-km buffer Th −0.0015 −0.0042 0.0012 ns −0.0046 −0.0083 −0.0010 *C 0.0031 −0.0023 0.0084 ns 0.0021 −0.0047 0.0088 nsW −0.0142 −0.0184 −0.0100 *** −0.0237 −0.0297 −0.0176 ***

(E) 20% sp. randomly removed Th −0.0032 −0.0062 −0.0003 * −0.0039 −0.0081 0.0004 (*)C 0.0016 −0.0037 0.0068 ns 0.0022 −0.0055 0.0099 nsW −0.0129 −0.0176 −0.0081 *** −0.0161 −0.0224 −0.0098 ***

(F) Forest cover change per decade Th −0.019 −0.028 −0.009 *** −0.035 −0.048 −0.022 ***C −0.005 −0.025 0.014 ns −0.045 −0.070 −0.019 ***W −0.030 −0.045 −0.014 *** −0.051 −0.074 −0.028 ***

A wide array of techniques to infer species’ temperature preferences resulted in consistent effects of forest canopy cover changes on the thermophilizationby applying varying climatic variables (left column: temperature only, right column: including both temperature as well as precipitation), modeling approaches(A, Gauss response curve; B, more flexible-shape response curves using GAMs), climatic periods (whole year instead of growing season, C), and study areaextents (3000 km buffer around study regions instead of whole continent, D). Randomly excluding 20% of the species (E) and dividing the forest cover responseratios by the time interval between the two survey periods (F) also resulted in robust patterns. Relationships result from mixed-effect models with random termstudy region for each of 500 samples; 95% confidence intervals are based on those samples. The data within the black box were used in the main text.Significances of the relationships with forest canopy cover were tested by Student t tests on the bootstrapped values (500 samples). ns: P > 0.10, (*): P < 0.10, *:P < 0.05, and ***: P < 0.001.

De Frenne et al. www.pnas.org/cgi/content/short/1311190110 13 of 13