Tracking of climatic niche boundaries under recent climate change


  • Frank A. La Sorte,

    Corresponding author
      Correspondence author. E-mail:
    Search for more papers by this author
    • Present address: Cornell Lab of Ornithology, 159 Sapsucker Woods Rd., Ithaca, NY 14850 USA.

  • Walter Jetz

    1. Department of Ecology and Evolutionary Biology, Yale University, 165 Prospect St., New Haven, CT 06520-8106, USA
    Search for more papers by this author

Correspondence author. E-mail:


1. Global climate has changed significantly during the past 30 years and especially in northern temperate regions which have experienced poleward shifts in temperature regimes. While there is evidence that some species have responded by moving their distributions to higher latitudes, the efficiency of this response in tracking species’ climatic niche boundaries over time has yet to be addressed.

2. Here, we provide a continental assessment of the temporal structure of species responses to recent spatial shifts in climatic conditions. We examined geographic associations with minimum winter temperature for 59 species of winter avifauna at 476 Christmas Bird Count circles in North America from 1975 to 2009 under three sampling schemes that account for spatial and temporal sampling effects.

3. Minimum winter temperature associated with species occurrences showed an overall increase with a weakening trend after 1998. Species displayed highly variable responses that, on average and across sampling schemes, contained a strong lag effect that weakened in strength over time. In general, the conservation of minimum winter temperature was relevant when all species were considered together but only after an initial lag period (c. 35 years) was overcome. The delayed niche tracking observed at the combined species level was likely supported by the post1998 lull in the warming trend.

4. There are limited geographic and ecological explanations for the observed variability, suggesting that the efficiency of species’ responses under climate change is likely to be highly idiosyncratic and difficult to predict. This outcome is likely to be even more pronounced and time lags more persistent for less vagile taxa, particularly during the periods of consistent or accelerating warming. Current modelling efforts and conservation strategies need to better appreciate the variation, strength and duration of lag effects and their association with climatic variability. Conservation strategies in particular will benefit through identifying and maintaining dispersal corridors that accommodate diverging dispersal strategies and timetables.


Global climatic conditions have shown significant changes over the past 50 years and further dramatic and spatially heterogeneous changes are projected for this century (Easterling et al. 2000; Karl & Trenberth 2003; IPCC 2007; Loarie et al. 2009). Assuming animal populations are adapted to local environmental conditions and climatic changes exceed population-level mean plasticity (Phillimore et al. 2010), changes in climatic conditions will require either an evolutionary or a distributional response. Species’ potential to successfully adapt or shift distributions in response to climate change depends on a host of factors, such as the speed and variability of changing conditions, species’ dispersal abilities, characteristics of a species’ climatic niche and species interactions. Under rapid climate change, where adaptive capacities in some cases are likely to be overwhelmed resulting in increased extinction risk (Chevin, Lande & Mace 2010; Maclean & Wilson 2011), a key consideration is the degree of niche conservatism displayed by a species − the more constrained a species’ fundamental climatic niche, the greater the pressure to geographically track and maintain those conditions (Wiens & Graham 2005; Pearman et al. 2008; Soberón & Nakamura 2009; Wiens et al. 2010). With accelerating carbon dioxide emission rates (Raupach et al. 2007), the likely irreversibility of climate change (Solomon et al. 2009) and certain ecological consequences (Jetz, Wilcove & Dobson 2007; Sinervo et al. 2010), determining how effectively species are tracking their climatic niche is critical for improving the predictive quality of modelling efforts and the conservation value of mitigation strategies.

Indirect evidence to date suggests that species’ responses to climate change are currently ‘lagging’ behind climatic trends (Menéndez et al. 2006; Foden et al. 2007; Devictor et al. 2008; Bertrand et al. 2011). With the limited pressure exerted on species’ climatic niches during the initial period of climate change, lag effects in species’ responses are expected (Mustin et al. 2009; Schippers et al. 2011). As climate change progresses and populations more consistently occur outside of their climatic norms, responses are likely to accelerate as population dynamics (Mustin et al. 2009; Schippers et al. 2011) and evolutionary processes react (Travis & Dytham 2002; Simmons & Thomas 2004; Burton, Phillips & Travis 2010; Phillimore et al. 2010). Additionally, responses in some cases could develop more rapidly through the influence of compounding synergistic factors (Paine, Tegner & Johnson 1998; Denny et al. 2009; Harley & Paine 2009) or extreme events (Gaines & Denny 1993; Jentsch & Beierkuhnlein 2008). Similarly, responses for certain species could accelerate in an invasive fashion as they react opportunistically to changing environmental conditions (Hellmann et al. 2008).

Examples from the recent past and the geological record indicate that, in the majority of cases, species have responded geographically to climate change in an idiosyncratic fashion with persistent time lags and non-equilibrium associations with the environment (Davis 1986; Webb 1986; Graham & Grimm 1990; Graham et al. 1996; MacDonald et al. 2008). Select longitudinal studies suggest climatic niches are conserved over geological time periods (Martínez-Meyer, Peterson & Hargrove 2004). However, attempts to quantify the temporal structure of past responses, including the time required to reassert niche conservatism, have been limited. Evidence from lake sediment fossils and pollen suggests that communities have responded to rapid climate change with relatively limited lag effects (Webb 1986; Birks & Ammann 2000; Shuman et al. 2002; Williams et al. 2002; Yu 2003; Birks & Birks 2008) that in some cases are estimated at <100 years (Williams et al. 2002). However, some records suggest geographic responses for certain species can contain lag effects that exceed 400 years (Birks & Birks 2008). In summary, our current understanding of the trajectory of species’ responses during rapid climate change remains highly inadequate, especially regarding the tempo and degree of climatic niche conservatism.

We present a simple framework for identifying distributional responses species may display under rapid climate change for a particular climate variable. We distinguish five core type of responses ranging from no-response (scenario a), where a species does not track changing climatic conditions to full response (scenario e) where a species tracks changing climatic conditions (Fig. 1). The no-response scenario (scenario a) follows changing climatic conditions over time and, for simplicity, will be shown as following a linear trajectory. The primary assumption of these scenarios, with the possible exception of the no-response scenario (scenario a), is that the climate variable is a critical component of a species’ climatic niche that is conserved over time. Between the two bounding scenarios (scenarios a and e), a spectrum of trajectories is possible (Fig. 1), with three of particular relevance for short- to mid-term length trajectories: (b) species could show a limited response whose strength of niche tracking remains consistent over time; (c) species could show an intermediate response that eventually settles on a new climatic association which is then tracked; and (d) species could show a true lagged response where the response strengthens over time until the historic association is reasserted, suggesting delayed niche tracking. Two additional trajectories that we do not illustrate are unlikely in this particular setting but worth listing. First, species may over-track changing climatic conditions and, second, species that are successfully tracking their climatic niche could have a response that weakens and falls behind the climatic trend.

Figure 1.

 Schematics of five response scenarios displaying changes in a species’ geographic associations with a climatic factor between two time periods (t1 and t2) shown (a) along a spatial climatic gradient and (b) over time. At t1 a species has a presumed observed (historical) association with a climatic factor (x1) as indicated by its spatial location (circle) along the spatial climatic gradient. By t2 the climatic factor has shifted spatially (note change in colour gradient). Five response scenarios may arise: (a) the species’ association with the climatic factor may change from t1 to t2 in the same way as the shifting climatic gradient, thus having no geographic response; (e) alternatively, the species may retain its original association (x1) with the climatic factor over time tracking changes in the climatic factor geographically, thus having no lag effect and a full response. Three possible responses can occur between these two bounding scenarios: (b) a limited response with only minor divergence from the observed trend may occur; (c) the strength of the lag effect may diminish over time and an apparent new association (x2) arises; (d) finally, a weak lag effect may occur with diminishing strength over time with the species eventually returning to its historic association (x1).

The mechanisms that underlie these scenarios are likely quite varied, reflecting the ecological and evolutionary dynamics operating across levels of biological organization within species’ geographic ranges. For simplicity, these mechanisms can be summarized based on the characteristics of species’ climatic niches and their dispersal abilities. Species could fail to track over time climatic conditions (scenario a) that indeed represent a critical niche component because of ecological or geographic dispersal limitations or a large disjunction between the location of realized and fundamental niche boundaries. Alternatively, the particular climatic variable might not be relevant for this particular species. At the other extreme, a lack of dispersal constraints and a strong congruence between the location of realized and fundamental niche boundaries could promote full niche tracking (scenario e). Between these two extremes, responses are determined by how these factors are defined and function in combination over time under changing conditions. With a lagged response (scenario d), niche tracking accelerates as the fundamental niche boundaries are increasingly breached and dispersal opportunities become more viable. Over time, lag effects could diminish when the fundamental niche boundary is reached and then tracked (scenario e) or when the species returns to its original climatic association (scenario d).

Using this framework, we provide direct and broad-scale empirical evidence of species-specific lag effects and their temporal structure under recent climate change for North American wintering birds. Birds are highly mobile, and in a recent historical resurvey, the breeding ranges of 53 species were found over a 100-year period to reasonably track their geographic climatic niche centroids (Tingley et al. 2009). Occupancy patterns within geographic ranges for birds that winter in North America tend to be linked less to maintaining territories and more towards acquiring nutritional resources (Newton 2003), suggesting a greater capacity to track changing environmental conditions. Moreover, minimum winter temperature has been found to be associated with northern range boundaries for these taxa (Root 1988) and strong temporal-turnover dynamics at northern range boundaries have been shown to facilitate poleward range shifts (La Sorte & Thompson 2007). Temporal changes in the geographic associations with minimum winter temperature at the northern edge of the range for North American winter avifauna thus represents an ideal test case for examining response structures and associated lag effects. With its unique close to continental spatial and >30 year temporal extent, the Christmas Bird Count survey (National Audubon Society 2010) of the North American winter avifauna offers an exceptional opportunity to assess species’ broad-scale response to climate change (La Sorte & Jetz 2010a). We present an assessment of the temporal trajectories and lag effects observed for 59 species from 1975 to 2009. Our broad-scale assessment allows us to generalize species’ trajectories across multiple interacting biological and environmental factors in a robust fashion. We expect that this evaluation will provide critical baseline knowledge about species’ responses to a warming world that will guide future modelling and mitigation work that considers biological dynamics at broad spatial and temporal scales.

Materials and methods

We used the North American Christmas Bird Count (CBC) data base in our assessment (National Audubon Society 2010). Christmas Bird Count surveys are conducted by a team of observers within 12-km radius circles for a period of 24 h between 14 December and 5 January. The CBC represents a comprehensive census of early-winter avian assemblages. We only considered the presence/absence of individual species at CBC circles because of the lack of a standardized assessment of species’ abundance. We selected 476 CBC circles for analysis that were sampled every year from 1975 to 2009 that occurred between 25° and 49° N latitude (Fig. 2a). Survey effort was highly variable within and among the 476 CBC circles but did not change in a systematic fashion from 1975 to 2009 (Fig. S1, Supporting Information). We selected 59 terrestrial, native diurnal species for analysis that occurred every year from 1975 to 2009 and occurred at >9 CBC circles in 1975 and whose maximum occurrence in 1975 was ≤43° N latitude (Table 1). We combined subspecies into single species and combined species that had been split during the last 50 years. Five species that had experienced extreme colonization events during this time period (La Sorte & Thompson 2007) were not included in the list of 59 species. Any positive relationship for these species between survey effort and detection probability is likely to be encompassed in a nonbiased fashion in our assessment because of the strong and temporally consistent heterogeneity in survey effort from 1975 to 2009 (Fig. S1, Supporting Information). The most northernmost occurrence for the 59 species in 1975 was c. 667 km south from the most northernmost sampled CBC circles. Based on the observed distributional trends for these species’ northern range boundaries (1·36 km year−1; La Sorte & Thompson 2007), we would expect an average shift from 1975 to 2009 of c. 48 km (95% CI: 19–76 km), a trajectory well within the 667-km buffer.

Figure 2.

 (a) The location of 476 Christmas Bird Count (CBC) circles and the difference (anomaly) between the 1975–79 and 2005–09 average minimum winter temperature (red increase, blue decrease). (b) Minimum winter temperature averaged across 476 Christmas Bird Count (CBC) circles from 1975 to 2009. The trend line was estimated using a linear mixed model with CBC circle as a random effect. The dotted lines indicate the 95% confidence band.

Table 1.   The 59 Christmas Bird Count (CBC) species considered in the study with significant regression coefficients indicated for two ordinary least-squares regression models that estimated species’ associations with minimum winter temperature from 1975 to 2009
  1. The first model contained an intercept and coefficient for year1) and the second model contained these terms plus a second-order polynomial coefficient for year2). The models tested if species’ responses differed from two response scenarios: (a) regression coefficients whose 90% confidence intervals did not contain the coefficient from the no-response scenario a (Fig. 1); (e) regression coefficients whose 90% confidence intervals did not contain zero and differed from the no lag effect or full-niche-tracking scenario e (Fig. 1). The models were applied to three sampling schemes: Annual/Full that used all available data (Full), Annual/Stratification that used annual data that were spatially stratified (Spatial), and 5-year/Full that used the full sample summarized by 5-year time intervals (Time; see Materials and methods for details). The dashes indicate species that were not included in the spatial stratification sampling scheme (n = 45). Regressions for each species were based on 35 points for the Full and Spatial sampling schemes and seven points for the Time sampling scheme.

FamilyScientific nameCommon nameFullSpatialTime
OdontophoridaeCallipepla squamataScaled Quail  e   
OdontophoridaeCallipepla gambeliiGambel’s Quail      
CathartidaeCoragyps atratusBlack Vulturea     
CathartidaeCathartes auraTurkey Vulture aa   
AccipitridaeButeo regalisFerruginous Hawk aaa  
GruidaeGrus canadensisSandhill Crane  a, e   
ColumbidaeZenaida asiaticaWhite-winged Dovea, eaa, e e 
ColumbidaeColumbina incaInca Dovea, eaa, ea  
ColumbidaeColumbina passerinaCommon Ground-Dove a    
CuculidaeGeococcyx californianusGreater Roadrunner a a  
ApodidaeAeronautes saxatalisWhite-throated Swift      
TrochilidaeArchilochus colubrisRuby-throated Hummingbirda, e   
TrochilidaeCalypte costaeCosta’s Hummingbirda, e   
PicidaeMelanerpes aurifronsGolden-fronted Woodpecker      
PicidaePicoides scalarisLadder-backed Woodpeckera, e ea  
PicidaePicoides nuttalliiNuttall’s Woodpecker     a
PicidaePicoides borealisRed-cockaded Woodpeckere   
TyrannidaeSayornis phoebeEastern Phoebee a   
TyrannidaeSayornis sayaSay’s Phoebe      
TyrannidaePyrocephalus rubinusVermilion Flycatcher      
VireonidaeVireo griseusWhite-eyed Vireo  a   
VireonidaeVireo solitariusBlue-headed Vireo      
CorvidaeCorvus ossifragusFish Crow  a  a
HirundinidaeTachycineta bicolorTree Swallow a    
ParidaePoecile carolinensisCarolina Chickadeee e   
ParidaeParus inornatusPlain Titmouse  a   
RemizidaeAuriparus flavicepsVerdina a, ea  
SittidaeSitta pusillaBrown-headed Nuthatche e   
TroglodytidaeCampylorhynchus brunneicapillusCactus Wrena, e a   
TroglodytidaeSalpinctes obsoletusRock Wrena     
TroglodytidaeCistothorus platensisSedge Wrene e   
PolioptilidaePolioptila caeruleaBlue-grey Gnatcatcher      
TurdidaeSialia sialisEastern Bluebirda a, e   
MimidaeToxostoma curvirostreCurve-billed Thrasher      
MimidaeToxostoma redivivumCalifornia Thrashere   
MotacillidaeAnthus spragueiiSprague’s Pipita, e     
PtilogonatidaePhainopepla nitensPhainopepla  a, ea  
ParulidaeOreothlypis ruficapillaNashville Warbler    
ParulidaeParula americanaNorthern Parulaaa  
ParulidaeDendroica dominicaYellow-throated Warbler  e a 
ParulidaeDendroica pinusPine Warbler  a   
ParulidaeDendroica discolorPrairie Warbler    
ParulidaeDendroica palmarumPalm Warblerea    
ParulidaeMniotilta variaBlack-and-white Warblere     
ParulidaeSeiurus aurocapillaOvenbird    
ParulidaeIcteria virensYellow-breasted Chata   
EmberizidaeAimophila ruficepsRufous-crowned Sparrowa, e a, eaee
EmberizidaePeucaea cassiniiCassin’s Sparrow  e 
EmberizidaeSpizella pallidaClay-coloured Sparrow    
EmberizidaeSpizella breweriBrewer’s Sparrowe a, e  e
EmberizidaeChondestes grammacusLark Sparrow      
EmberizidaeAmphispiza bilineataBlack-throated Sparrowa aa  
EmberizidaeAmphispiza belliSage Sparrowa, e    a
EmberizidaeAmmodramus leconteiiLe Conte’s Sparrowe a   
EmberizidaeAmmodramus maritimusSeaside Sparrowe   
CardinalidaeCardinalis sinuatusPyrrhuloxia      
CardinalidaePasserina cyaneaIndigo Bunting  a 
IcteridaeAgelaius tricolorTricoloured Blackbirda   
IcteridaeQuiscalus mexicanusGreat-tailed Gracklea, e a, e   

To link CBC occurrences to climatic conditions representative of the time of observation, we compiled climate data for the period 1975–2009 from the 1-km resolution PRISM Climate Group data set (Oregon State University; We extracted minimum winter temperature for each CBC circle and year based on the monthly average daily minimum temperature averaged within the 12-km circle over the 2 months that encompassed the CBC sampling events (December–January).

To determine how the spatial and temporal sampling structure of our data affected our results, we repeated our analyses under three sampling schemes. The first used all the available data for a given year (= 35 years) and the full sample of available spatial data (Annual/Full). To assess how the unique spatial distribution of the 476 CBC circles (Fig. 2a) affected our results, the second sampling scheme used the data from single years contained within a spatially stratified subset of the CBC circles (Annual/Stratification). We selected 96 CBC circles randomly from 24 Bird Conservation Regions (BCR; North American Bird Conservation Initiative) where each BCR contained four circles. We used BCR’s because they represent ecologically distinct regions with similar bird communities and similar environmental and climatic conditions. BCR’s are also relatively large and are likely to encompass local to regional spatial autocorrelation that might exist among the CBC circles. A total of 45 species that occurred every year were considered, and were observed at >4 CBC circles in 1975 and whose maximum latitude in 1975 was ≤43° N. To determine how the use of annual associations affected our results, in the third sampling scheme, we used data summarized within seven 5-year time intervals (1975–79, 1980–84,…, 2005–09) for all 59 species and the full sample of 476 CBC circles (5-year/Full). Five years was selected because it encompassed the climatic variability in minimum winter temperature observed during this time period (Fig. 2b) and associated variability in species’ geographic responses.

Many bird species in temperate regions migrate south during winter months (Alerstam 1991) with winter climatic conditions playing a role defining population survival and resulting distributional boundaries as species track latitudinal temperature gradients (Mehlman 1997; Link & Sauer 2007; Robinson, Baillie & Crick 2007). Identifying which climatic factors are relevant, however, can be challenging (Kearney & Porter 2009). Given metabolic constraints on avian body temperature regulation (Jetz, Freckleton & McKechnie 2008) and seasonal variation in ectotherm prey availability (Bale 2002) and overall energy availability (Hurlbert & Haskell 2003), an obvious factor that is likely relevant for many bird species that winter in temperate regions of North American is minimum winter temperature (Root 1988; Zuckerberg et al. 2011). The climatic variable we used in our analysis was the lowest of the minimum winter temperature values measured at the CBC circles where a species occurred in a given year. This approach does not summarize minimum temperatures across a species’ geographic range. Rather, it estimates the lowest temperature each species encounters annually, a condition likely to occur in close proximity to the species’ north range limit (Root 1988). Thus, each species has one minimum winter temperature value for each year and, because we only considered species that were observed every year, each species has a total of 35 values.

We assessed geographic associations with minimum winter temperature in two fashions for the Annual/Full and the Annual/Stratification sampling schemes, first based on 1975 occurrences at CBC circles and then 1975 to 2009 occurrences at CBC circles (Fig. 3a,b for an example). For the first approach, we used the CBC circles where the species occurred in 1975 and extracted the lowest minimum winter temperature values estimated from those same circles for each subsequent year from 1976 to 2009. This approach generated the null expectation given no-response (scenario a; Fig. 1); that is, how minimum winter temperature would change over time if the species retained the same CBC occurrences from 1975 to 2009. The second approach extracted the lowest minimum winter temperature values from the CBC circles where the species actually occurred each year from 1975 to 2009. This approach documented the observed response and how each species’ geographic association with minimum winter temperature changed over time. For the 5-year/Full sampling scheme, we used the minimum winter temperature estimated at each CBC circle averaged over seven 5-year time periods. The first time period (1975–79) was used to generate the null expectation given no-response and the remaining 5-year periods was used to document the observed response.

Figure 3.

 The minimum winter temperature at Christmas Bird Count (CBC) circles where the Black vulture (Coragyps atratus) occurred in (a) 1975 and (b) in 2009. The open circles represent CBC circles occupied in the opposing year. (c) Minimum winter temperature averaged across CBC circles for the Black vulture based on 1975 CBC occurrences (blue) and 1975 to 2009 CBC occurrences (red) based on the Annual/Full sampling scheme. The red line is the observed response, the blue line is the expectation under no-response scenario a, and the dashed line is the expectation under no lag effect or full-niche-tracking scenario e (Fig. 1). Trend lines were estimated using ordinary least-squares regression.

To determine how the presence of outlying minimum winter temperature values might affect our analysis, we reran our analysis using the Annual/Full sampling scheme after identifying and removing outliers. Minimum winter temperature distributions were treated as normally distributed for each species and year, and outliers were detected based on the number of expected observations within the configuration of the model distribution (van der Loo 2010). Once outliers were removed, the lowest minimum winter temperatures values were re-identified for analysis.

To determine how species’ association with minimum winter temperature changed over time, we conducted a two-step analysis using a longitudinal study design (Fitzmaurice, Laird & Ware 2004). First, we used ordinary least-squares (OLS) regression to summarize the mean response trajectory over time for each individual species. Second, we used linear mixed models to summarize the mean response trajectory for the entire collection of species while accounting for the repeated measurements taken over time for each species. More specifically, we used OLS regression to estimate temporal associations with minimum winter temperature for each of the 59 species (Fig. 3c for an example) and linear mixed models to estimate temporal associations with minimum winter temperature across the 59 species. Both analyses were conducted separately using species’1975 CBC occurrences and species’ 1975 to 2009 CBC occurrence. With the linear mixed models, the intercept and slope in the annual trends were allowed to vary randomly among species. We also used linear mixed models to estimate the temporal trend in minimum winter temperature across CBC circles where the intercept and slope in the annual trends were allowed to vary randomly among circles.

We used a top-down model building strategy (Verbeke & Molenberghs 2000; West, Welch & Gałecki 2007) to select fixed and random effects for each linear mixed model with α set at 0·10. This setting allowed for a more inclusive model building strategy where < 0·10 is interpreted as evidence of a biological important association worthy of consideration. Model building was implemented with data from the Annual/Full sampling scheme. The resulting models were then applied to data sets derived from the Annual/Stratification and 5-year/Full sampling schemes. The modelling strategy started with a full model containing linear and second-order polynomial fixed effects and random effects for year, which defined the basis for the selection of random effects using restricted maximum likelihood-based likelihood ratio tests of nested paired models. The full model for one species or one CBC circle and 1 year has the form:


where Y is minimum winter temperature, β are the fixed effects, b are the random effects for species or CBC circle and e is the residual error. Depending on the analysis, the random effects were grouped by species or CBC circle. Once the random effects were chosen, the fixed effects were selected using maximum likelihood-based likelihood ratio tests of nested paired models. The outcomes of the likelihood ratio tests were used to make inferences on how minimum winter temperature changed over time across (fixed effects) and among (random effects) species and CBC circles. The variable year was centred for analysis by subtracting 1975 from the observed values placing the intercept at 1975. Diagnostic plots were examined for each linear mixed model and the distributional assumptions for the model residuals and random effects were found to be met in each case. All analysis was conducted in r version 2·13·0 (R Development Core Team 2011). The r libraries lme4 and nlme were used to implement the linear mixed models (Pinheiro & Bates 2000) and the library extremevalues were used to detect outliers (van der Loo 2010).


Minimum winter temperatures at CBC circles increased from 1975 to 2009 with the strongest gains occurring at higher latitudes (Fig. 2a) with evidence for a weakening trend after c. 1998 (χ2 = 693·8, df = 1, < 0·001; Fig. 2b, Table 2). There was additional evidence that trends differed among CBC circles (χ2 = 67·7, df = 2, < 0·001) and the shape of the trends differed among CBC circles (χ2 = 35·3, df = 3, < 0·001; Table 2). Strong interannual variability was present with several years standing out as outliers; in particular, 1976 which had the lowest average minimum winter temperature from 1975 to 2009 (Fig. 2b).

Table 2.   Linear mixed models summarizing the estimated trends in minimum winter temperature from 1975 to 2009 at 476 christmas bird count (CBC) circles and for 59 bird species based on 1975 CBC occurrences and 1975 to 2009 CBC occurrences
  1. Fixed effects and random effects for each model were selected based on a top-down model selection strategy, see Materials and methods for details.

Circle: 1975–2009
  Fixed effects
   Interceptβ1−6·297−21·2016 182<0·001
   Yearβ20·25340·8816 182<0·001
   Year2β3−0·005−31·9716 182<0·001
  Random effects
Species: 1975
  Fixed effects
  Random effects
Species: 1975–2009
  Fixed effects
  Random effects

The observed response for the 59 bird species from 1975 to 2009 was highly variable for the Annual/Full analysis (Fig. 4a; Table S1, Supporting Information). Looking at models with only linear terms, 18 species (32%) had responses that differed from the no-response scenario a and 21 species (36%) had responses that differed from the full-niche-tracking scenario e (Fig. 4a; Table 1). With the addition of second-order polynomial terms, 19 species had significant variation in the shape of the response and nine species had responses whose shape differed from the shape of the no-response scenario a (Fig. 4a; Table 1). Two species, Turkey vulture (Cathartes aura) and the Ruby-throated humming bird (Archilochus colubris), had trajectories that substantially over-tracked trends in minimum winter temperature (Fig. 4a; Table S1, Supporting Information). Thus, for the Annual/Full analysis, more species presented linear responses that followed the no-response scenario than the full-niche-tracking scenario. With the addition of nonlinear components, species’ responses showed evidence of converging on the full-niche-tracking scenario in a manner reflecting the delayed-niche-tracking scenario d (Fig. 1). The removal of minimum winter temperature outliers for individual species did little to change these overall patterns (Fig. S2a, Supporting Information).

Figure 4.

 (a-c) Fits of ordinary least-squares regression of minimum winter temperature by year for individual species based on their occupancy of Christmas Bird Count (CBC) circles from 1975 to 2009 under three sampling schemes: using the full annual and spatial data set (Annual/Full; n = 59), data that was spatially stratified (Annual/Stratification; n = 45) and data that was summarized into 5-year intervals (5-year/Full; n = 59; see Materials and methods for details). Bold lines identify species with significant second-order polynomial coefficients and, if not significant, species with significant linear coefficients. All fits were scaled to start at zero and the red dashed line is the expectation under niche tracking and no lag effect scenario e (Fig. 1). (d-f) Minimum winter temperature averaged across CBC circles based on 1975 CBC occurrences (blue) and 1975 to 2009 CBC occurrences (red) for bird species under three sampling schemes. The red line is the observed response and the blue line is the expectation under no-response scenario a (Fig. 1) summarized across species. The dashed line is the expectation under no lag effect or full-niche-tracking scenario e. Trend lines were estimated using linear mixed models with species as a random effect.

The Annual/Stratification analysis for 45 species accentuated the variability observed in the Annual/Full analysis (Fig. 4b; Table S2, Supporting Information). The majority of the 45 species presented associations that differed between the two analyses (Table 1). Looking at models with only linear terms, 19 species (42%) had responses that differed (< 0·10) from the no-response scenario a and 15 species (33%) had responses that differed from the full-niche-tracking scenario e (Fig. 4b; Table 1). With the addition of second-order polynomial terms, 13 species had significant variation in the shape of the response and eight species had responses whose shape differed from the shape of the no-response scenario a (Fig. 4b; Table 1). Under this setting, the white-winged Dove (Zenaida asiatica) had a linear trajectory that substantially over-tracked changing climatic conditions (Fig. 4b; Table S2, Supporting Information). In contrast to the Annual/Full analysis, a greater proportion of species presented linear responses that differed from the no-response scenario; eight of these species were shared across the two analyses. Similar to the Annual/Full analysis, with the addition of nonlinear components species’ responses converged on the full-niche-tracking scenario.

The 5-year/Full analysis presented a strong contrast to the previous two analyses with a substantial reduction in the variability in species’ trajectories and the number of trajectories that differed from scenarios a and e (Fig. 4c; Table 1; Table S3, Supporting Information). Looking at models with only linear terms, two species had responses that differed from the no-response scenario a and three species had responses that differed from the full-niche-tracking scenario e (Fig. 4c; Table 1). With the addition of second-order polynomial terms, two species had significant (< 0·10) variation in the shape of the response and three species had responses whose shape differed from the shape of the no-response scenario a (Fig. 4c; Table 1). The 5-year temporal summary therefore resulted in scenarios a and e becoming more similar, with the majority of species presenting trajectories that did not differ from either scenario. Nevertheless, the overall pattern still captured the signal evident in the previous two analyses with the majority of species presenting positive trajectories (64%) with some evidence for convergence on the full-niche-tracking scenario.

Using mixed models to summarize patterns across species, minimum winter temperature based on 1975 CBC occurrences for the no-response scenario increased from 1975 to 2009 with a weakening trend over time (χ2 = 23·0, df = 1, < 0·001; Fig. 4d, Table 2). Both trends (χ2 = 10·1, df = 2, = 0·007) and shape differed among species (χ2 = 41·32, df = 3, < 0·001; Table 2). For the observed response, minimum winter temperature increased from 1975 to 2009 in a more limited fashion with a similar weakening trend over time (χ2 = 26·40, df = 1, < 0·001; Fig. 4d, Table 2). While responses differed among species (χ2 = 35·2, df = 2, < 0·001), the shape of the response did not (χ2 = 0·0, df = 3, P = 1; Table 2). When combined into a single model, the linear components differed (χ2 = 27·5, df = 1, < 0·001), while the shape of the response did not differ (χ2 = 0·2, df = 1, P = 0·653) between the observed response and no-response scenarios. These results did not differ qualitatively when the two species that over-tracking climatic trends were excluded. In addition, the presence of a colder than average winter in 1976 did not appear to affect subsequent associations, which would be expected to have a higher minimum winter temperatures than those observed for 1975 CBC occurrences if populations were extirpated from the coldest portions of species’ ranges. Lastly, these results were robust to the removal of minimum winter temperature outliers for individual species (Fig. S2b, Supporting Information). In total, the trajectory for the Annual/Full analysis most closely followed the delayed niche-tracking scenario d (Fig. 1).

Similar associations were found when these results are contrasted with those generated using the spatial stratification sampling scheme (Annual/Stratification; Fig. 4e). Here, the delayed niche-tracking scenario d was the most relevant with evidence for over-tracking near the end of the survey period as the warming trend weakened. The analysis at 5-year instead of annual time intervals (5-year/Full; Fig. 4f) also emphasized the delayed niche-tracking scenario with evidence for a more persistent lag effect towards the end of the survey period. This outcome suggests that minimizing temporal variability and the influence of annual climatic outliers resulted in a more uniform response and a stronger lag effect.


Across CBC circles, minimum winter temperature increased with a pronounced weakening trend after 1998. The same general trend was mirrored in species’ climatic associations based on 1975 CBC occurrences. Based on 1975 to 2009 CBC occurrences and in agreement with past investigations (La Sorte & Thompson 2007), species responded to warming temperatures by tracking minimum winter temperature. Consistent with our predictions, species responded in a highly variable and idiosyncratic fashion. When summarized across species, a lag effect persisted for approximately 35 years and weakened in direct association with a ‘lull’ in the warming trend. These findings were generally consistent after accounting for the spatial and temporal sampling structure of the data. In total, winter avifauna in North America displayed highly variable and idiosyncratic responses that covered a broad array of possible scenarios. All species in the analysis considered in combination tracked the boundaries of their climatic niche, but did so in a delayed fashion suggesting that the conservation of geographically defined minimum winter temperature was broadly relevant during this time period.

Evidence for a broadly defined lag effect in a taxon as highly vagile as birds suggests these effects are likely to be even more pronounced for species with more limited dispersal abilities and stronger regional habitat associations. Moreover, evidence that the weakening in the warming trend provided an opportunity for niche tracking to accelerate suggests lag effects are likely to be even more substantial during periods defined by less variable or accelerating warming. There is evidence that the ‘lull’ in the warming trend after 1998 is related to short-lived sulphur emissions partially compensating for increasing greenhouse gas concentrations (Kaufmann et al. 2011). Based on our findings and the likelihood that the current ‘lull’ is temporary, it can be conjectured that modern climate change has the potential to leave many species with a diminishing ability to maintain geographic associations with suitable climates. This is particularly true in some tropical (Colwell et al. 2008) and montane regions (La Sorte & Jetz 2010b) where the impact of warming on species metabolism is likely to be greater (Dillon, Wang & Huey 2010) and dispersal opportunities and dispersal abilities are likely to be more limited.

The high variability in species responses documented in this study can be attributed to a variety of potential sources. First, stronger geographic responses to climate change have been noted for the taxa of winter avifauna whose ranges have more northerly locations in North America (La Sorte & Thompson 2007). Second, species and community level traits and attributes that indicate broader niche breadth have been associated with stronger responses (La Sorte et al. 2009; Angert et al. 2011). Both of these factors are likely responsible for some aspects of the observed variation, however, they do not represent strong correlates. In fact, there is little evidence to date for a close association between species’ level traits and the strength of species’ responses to recent climate change (Angert et al. 2011). The high variability observed in this study, therefore, contains evidence of an organized response defined by key geographic and biological features that in the end contains limited explanatory power. This suggests that communities are being reassembled under climate change in a nonuniform fashion.

When considering species’ responses to rapid climate change, several factors are important in determining how responses are likely to be structured over time. A primary consideration with geographically defined climatic associations is the broad-scale and dynamic interplay of populations and metapopulations with the environment (Holt & Keitt 2000, 2005). As has already been documented with winter avifauna, greater temporal dynamism at the leading edge of the range has been linked to distributional trends under recent climate change (La Sorte & Thompson 2007). However, an important consideration is that range boundaries do not function in isolation but are linked to core populations at the range centre, a region that is geographically more stable and resistant to environmental change because of larger population sizes (Brown 1984). The strength of the response at the boundary is thus determined by the degree of dispersal (and gene flow) across the range and the ecological and evolutionary balance that allows boundary populations to both persist and exploit changing environmental conditions (Gaston 2009). These dynamics will also depend on how climatic gradients are defined across the range; climate is likely to play a much more limited role delineating species range boundaries where latitudinal or altitudinal climatic gradients are weak or do not exist. Therefore, the structure and dynamics of species’ geographic ranges suggests tracking of climatic niche boundaries for a niche component that is conserved is likely to contain a lagged temporal response whose structure will be strongly species and region specific.

Beyond the spatial characteristics of species’ distributions, there are additional factors that affect how species’ responses to climate change are likely to develop over time. Evolutionarily, taxa and species vary in the conservatism of their broad-scale environmental climatic niches (Wiens et al. 2010; Cooper, Jetz & Freckleton 2011), and an interesting next research step will be whether such phylogenetic predictions are borne out in recent survey data. On an environmental level, the ability of species to cope with increasing climatic variation, as currently predicted under climate change projections, is likely to be critical for successful dispersal and persistence (Easterling et al. 2000; Boyce et al. 2006; Jongejans et al. 2010; Early & Sax 2011). This is likely to be especially important with migratory birds that will be affected by both spatial and temporal variation in climate change (Jones & Cresswell 2010). However, there is evidence that migratory behaviour for some species can evolve quite rapidly (Berthold et al. 1992). On an ecological level, the degree of overlap between species’ realized and fundamental niches (Soberón & Nakamura 2009; Dormann et al. 2010), and the plasticity or adaptive capacity of a species to environmental change are all likely to play critical roles (Skelly et al. 2007; Gienapp et al. 2008). Other important ecological factors include the presence of competitors (Case & Taper 2000; Ahola et al. 2007; Burton, Phillips & Travis 2010), parasites (Brooks & Hoberg 2007), mutualists or other ecological associates within and across trophic levels whose distributional response to climate change are likely to be even more strongly delayed under climate change (Harrington, Wolwod & Sparks 1999; Berg et al. 2010).

In total, species’ responses to climate change are being determined by an array of interacting biological and environmental factors whose outcome is likely to be complex, idiosyncratic and difficult to predict (Pimm 2009; Walther 2010). Moreover, these factors are more likely to hinder than to promote responses, particularly at climatic niche boundaries where physiologically appropriate climatic conditions could develop in the absence of other critical ecological or environmental features. Additional work is needed to further elucidate the dynamics or mechanisms underlying the responses observed in this study. For example, the assessment of standardized abundance data, not available as part of the CBC data set, would allow for the evaluation of additional key hypotheses regarding interspecific differences in observed responses.


As we enter this period of environmental change, effort is needed to understand the broad-scale biological consequences in a manner that will best inform mitigation strategies whilst limiting the number of ecological ‘surprises’ (Williams & Jackson 2007). Empirical work that adds realism to modelling efforts has the potential to narrow the range of likely scenarios, thus allowing investigators to further refine their predictions and helping policy makers to more efficiently allocate limited conservation resources. Warming trends during this century are likely to contain some level of interannual variability because of natural and anthropogenic factors (Kaufmann et al. 2011). As seen in this study, this variability can provide climatic ‘lulls’ and opportunities for species to reassert their climatic niche, a factor that is likely to be even more relevant for species with limited dispersal abilities. Modelling efforts that project species distributions under future climate change typically assume that species responses are uniform and niche-tracking is perfect (Pearson & Dawson 2003; Elith & Leathwick 2009). These assumptions were not supported by our findings where the level of variability in species’ responses covered a broad spectrum of possible response scenarios. In addition, considering the duration of lag effects documented in this study, forecasts made over the next 100 years where warming trends are projected to increase in strength (IPCC 2007) are likely to result in highly unrealistic predictions of future community composition. Modelling efforts that consider the distribution, strength and duration of lag effects under different climate change trajectories are likely to be more robust.

Lastly, our findings suggest that the development and application of conservation strategies designed to mediate the impacts of climate change for biological diversity (Hannah 2010) would benefit by the long-term conservation of high-quality dispersal corridors that accommodate diverging dispersal strategies and timetables under varying climatic trends. Conservation research and planning directed towards identifying critical dispersal corridors (Williams et al. 2005; Hannah et al. 2007; Vos et al. 2008) that are especially effective in promoting the dispersal of habitat specialists (Gillies & St. Clair 2008) would give a greater number of species the opportunity to independently disperse and persist, thus limiting anthropogenic biases in community reassembly. Once identified, the protection of the long-term natural integrity of these corridors will be necessary if species are to gain the greatest possible benefits.


We thank J.Belmaker, D. Fink, T.M. Lee, M. Tingley, J. Williams, and H. Wilman for valuable discussions and helpful comments on the manuscript. This article benefited by thoughtful comments by A. Phillimore and two anonymous reviewers. We also acknowledge the contributions of the thousands of Christmas Bird Count volunteers, K. Dale, G. LeBaron, and the National Audubon Society. The study was supported by National Science Foundation awards DEB 1026764 and DBI 0960550.