Traits help explain species' performance away from their climate niche centre

Climate change impacts on biota are variable across sites, among species and throughout individual species' ranges. Niche theory predicts that population performance should decline as site climate becomes increasingly different from the species' climate niche centre, though studies find significant variation from these predictions. Here, we propose that predictions about climate responses can be improved by incorporating species' trait information.


| INTRODUC TI ON
Climate change is reshaping plant communities through increasing extinction risk Urban, 2015), introducing novel colonists through species range expansions (Savage & Vellend, 2015;Steinbauer et al., 2018) and causing shifts in abundance within communities . Climate change experiments often find that moist temperate grassland plant communities shift towards taller, more productive species as climates warm (Elmendorf, Henry, Hollister, Björk, Bjorkman, et al., 2012;Fridley et al., 2016). However, considerable variation among species and communities in their responses suggests that the effects of climate change can be highly context-dependent (Dunne et al., 2004;Vandvik et al., 2020).
Some of the variation in species responses to climate change may be explained by macroecological context: a population's relative environmental position compared with average conditions encountered across the species' range or niche (Lynn, Klanderud, et al., 2021;Reich et al., 2015). Drawing from classical niche theory (Blonder, Moulton, et al., 2017;Hutchinson, 1957), macroecological context predicts that local populations will perform better with climate change if local temperature or precipitation shifts towards conditions more similar to the species' range/niche centres but perform worse if climate shifts away from range centre conditions. This can be assessed by looking at climate differences: the difference between the average climate a species occupies across its geographic distribution (its climate niche centre) and the climate experienced by the local population (Lynn, Klanderud, et al., 2021). For example, in a turf transplant experiment where whole grassland communities were moved to warmer and/or wetter climates, species with climate niches warmer than a given transplant site were less likely to go extinct, more likely to colonize, and increased in abundance compared with species with niches increasingly cooler than their transplant site (Lynn, Klanderud, et al., 2021). Similarly, a single-site warming and drought experiment found that species with warmer summer temperature niches increased in abundance under warming while species with drier spring precipitation niches increased in abundance under drought (D. Liu et al., 2018). Macroecological context thus provides a universalizable framework to assess and predict both population-and community-level variation, that is, differences in response to climate change among species in a given site and/or across sites within a species. Nevertheless, a considerable amount of interspecific variation in species climate responses remains unexplained after accounting for macroecological context.

This additional interspecific variation in macroecological studies
of species responses to climate change likely arises from aspects of species life history strategies that are not closely tied to responses to climate. That is, even if species have similar climatic optima, they likely vary in life history and investment strategies for light capture, tissue quality, reproduction and more (Díaz et al., 2016;Fridley, 2017;Grime, 2006) that are not causally linked to climate but may dictate their performance under different macroecological contexts. For instance, in tree species across the Pacific west of the United States, taller tree species tended to occupy wetter habitats than expected based on models of optimal soil water balance for survival (Bohner & Diez, 2020). This suggests that traits (i.e. height) related to life history strategy outside of those that governed survival responses to climate explained variation in habitat occupancy. Therefore, variation in the magnitude and direction of climate responses across species may partially depend on variation in other traits that relate to their ability to maximize fitness in environments, independent of that site's climate.
Here, we develop a novel modelling framework to test how species' responses to climate variation around their niche centre is associated with plant functional traits. We hypothesize that variation in trait strategies indicative of stress tolerance may in part explain why some species perform better away from their climate niche centre (Dallas et al., 2017;Sporbert et al., 2020). This approach extends the macroecological context framework (see below) to fulfil one of the key aims of functional ecology by explaining how species respond to environmental variation based on their trait strategies (Grime, 2006;Reich, 2014). Our framework is generalizable to other systems and forms of environmental change and, thus, may have broad utility in predicting species vulnerability to various global change drivers.

| The framework
We first define macroecological context in Box 1, then provide hypotheses for how traits relate species' climate sensitivity in Table 1. A schematic diagram of the framework is provided in Figure 1.
We start with the general assumption that a species' niche centre represents its climate optimum, such that increasing differences in either direction represents sub-optimal conditions that reduce performance relative to the niche centre (Brown, 1995). That is, we expect unimodal climate difference sensitivities (CDS; Box 1).
However, in empirical cases where observed climate differences are mostly positive or mostly negative, interpretation is greatly simplified and enables linear investigations on one side of a species' optimum. For example, in our study system (grasslands in southwestern Norway), conditions are generally colder and wetter than the estimated niche centres of the species, suggesting we should generally find negative temperature and positive precipitation CDS (Figure 1b,c). Assuming performance is maximal at the niche centre and given our distribution of niche differences in Figure 1b, a negative temperature CDS means that the species performs worse as site temperature is increasingly cooler than a species' niche centre.

K E Y W O R D S
abundance, biodiversity, body size, climate change, extinction and colonization, functional traits, niche theory, plants A positive precipitation CDS means that the species performs better as site precipitation becomes more similar to the species' niche centre.
We assume that species' performances are optimal at their climate niche centres, which are derived from their observed distributions. However, this simplifying empirical assumption requires further scrutiny given dispersal (e.g. Goel & Keitt, 2022), biotic interactions (e.g. Lynn, Miller, & Rudgers, 2021) and other aspects of the abiotic environment (e.g. Moutouama & Gaoue, 2023) will constrain the observed distribution to a subset of possible climates. Ultimately, climate niche centres should be derived from physiological response curves (Michaletz et al., 2015;Perez et al., 2023), but such physiological data are lacking for the vast majority of species. Therefore, using species distributions to estimate climate niche centres is a pragmatic solution that allows testing hypotheses across large numbers of species, given predictions of climate suitability are generally well related to physiological climate performance (VanDerWal et al., 2009).
As outlined above, species vary in the amount of decline or even whether or not performance declines at all away from the niche centre and towards presumably marginal climates (Pironon et al., 2017;Sporbert et al., 2020; but see Britnell et al., 2023;Fristoe et al., 2023). We propose that functional traits may help explain species' ability to increase performance under non-optimal climate conditions. Specifically, resource acquisitive traits are advantageous in environments that are relatively less stressful (i.e. under nonproduction limiting conditions) compared with their niche centre, while conservative trait strategies are more advantageous in environments that are relatively more stressful than their niche centre (Grime, 2006). Given these interpretations, we expected species with positive temperature CDSs to have more resource-conservative traits (lower trait values are more conservative) than species with zero or negative CDSs (Figure 1e; Table 1), reflecting a capacity to maintain performance in environments that are increasingly cooler than their niche centre. In contrast, we expected species with positive precipitation sensitivities to have more resource-acquisitive traits (higher trait values are more acquisitive) than the zero and negative precipitation sensitivity species (Figure 1f; Table 1), reflecting the improved performance of species as the precipitation environment becomes relatively drier and hence more like their niche centre. This hypothesis may seem unintuitive given water is often a limiting resource, but this system receives extremely high precipitation annually (sites receiving >3000 mm annual precipitation) to the point where cloud cover and lack of sunlight are a greater impediment to productivity than water availability (Seddon et al., 2016).
For instance, low-precipitation communities (~800 mm) in western Norway tend to have more resource-acquisitive leaves (i.e. higher specific leaf area) than high-precipitation communities (Guittar et al., 2016). We focused on traits from the leaf-height-seed spectrum, which are hypothesized to capture much of the variation in plant life history strategies from fast-growing resource acquisitive, to slow-growing stress-tolerant and small-seeded ruderal strategies (Laughlin et al., 2010;Moles, 2018;Westoby, 1998). Past observational and experimental work supports these a priori predictions and is summarized in Table 1, along with descriptions of the specific traits we test.

| Study design
We used a 12-site turf-transplant experiment that began in 2009 in southwestern Norway for our observations of species abundance over time (Vandvik et al., 2020). The sites are regionally distributed into a factorial grid with three summer-temperature levels (~6, 8 and 10°C) and four annual precipitation levels (~700, 1200, 1900 and 2900 mm) based on downscaled normal period climate data (1961-1990; met.no; coordinates in Table S1.1 in Appendix S1). We transplanted intact grassland plant communities or 'turfs' as treatments either one step 'warmer' (~2°C; eight site-to-site transplants), 'wetter' (~700 mm, nine site-to-site transplants) or 'warmer/wetter' (six site-to-site transplants) to factorially isolate the effects of these predicted climate change drivers on plant communities (Hanssen-Bauer et al., 2017). Every site contained local control plots. Over the 10 years since transplanting, the transplanted communities

BOX 1 Definitions and interpretation.
Macroecological context is the relation of a site and plant community characteristics to the geographic range and niche characteristics of the species making up the community.
Climate difference is the mean climate variable of a species (niche centre) minus the climate of the site where an individual of the species was observed (i.e. Δclimate = niche climate -site climate). In this study, we focused on mean annual temperature (MAT) and annual precipitation (AP).
Negative values mean the site where a species was observed is warmer or wetter than the species' niche centre while positive values indicate the site is cooler or drier than the species' niche centre.
Climate difference sensitivity (CDS) represents the strength and direction of response metric change as climate differences change from negative to positive ( Figure 1c).
Mathematically, this is the slope from a linear model of climate differences predicting a given response metric.
We focus on the direction (positive, zero/unrelated or negative) of CDS. For a given species, positive sensitivity indicates a given response metric increases as climate differences increase, zero sensitivity indicates no relationship between a response metric and climate differences, and negative sensitivity indicates a response metric decreases as climate differences increase from negative to positive.
TA B L E 1 Specific hypotheses for how the traits we studied are related to species macroecological context relationships for colonization and abundance change. Resource-conservative traits: • Leaf dry matter content (LDMC): Higher LDMC leaves have more dense tissue (Kichenin et al., 2013;Venn et al., 2014;Wilcox et al., 2021) • Leaf thickness (L th ): Higher L th leaves are physically stronger (Blonder, Salinas, et al., 2017;Soudzilovskaia et al., 2013;Wilcox et al., 2021) • Leaf carbon content (leaf C): Leaves with more carbon have higher structural support (e.g. cell walls) (Peng et al., 2020;Zhao et al., 2014) • Leaf C:N: Leaves with higher C:N have more structural support relative to photosynthetic capacity on a mass basis (Zhang, He, et al., 2020; • Seed mass (SM): Species with higher SM have higher seedling survival and establishment probability but are generally made in lower numbers (Chelli et  Traits are divided into those for which higher values are considered more resource-acquisitive and associated with higher maximum potential growth rate and those for which higher values are considered more resource-conservative and typically associated with lower maximum potential growth rate. The climate difference sensitivity (CDS) groups are indicated with +, 0, and −, representing positive, zero/null and negative linear relationships with performance, respectively. Our hypotheses for how traits differ between CDS groups are indicated by ↑, =, ↓, signs, which represent whether a group is likely to have higher (↑) or lower (↓) trait values compared with the zero group (=). Support for our trait interpretations and predictions come from the observational and experimental work cited in the 'Traits and interpretation' column.
have shifted towards the species composition of their 'destination' sites, albeit with extinction lags (Vandvik et al., 2020). All sites were located on south-facing, shallow slopes with calcareous substrate and grazed, historically (Klanderud et al., 2015). Turfs were 29 × 29 × 10 cm in length, width and depth, and we analysed the inner 25 × 25 cm for treatment effects, leaving the outer edge as a buffer.
Further experimental details can be found in Vandvik et al. (2020).

| Colonization, extinction and abundance change
We

F I G U R E 1
Conceptual diagram to demonstrate how we analysed climate difference sensitivity (CDS; Box 1) in relation to species traits.
(a) Two hypothetical species with different distributions that both occur in our study region in southwestern Norway. Macroecological context contrasts the characteristics of a species niche with the characteristics of the site where a species is observed as shown in (b), which are the real distributions (histograms) of climate differences of individual species from plots in our data. The black coloured bars are the data from a particular species, Agrostis capillaris. We focus on contrasts of species mean annual temperature (MAT) and annual precipitation (AP) niche centres with site MAT and AP as metrics of climate differences (ΔMAT = niche − site temperature; ΔAP = niche − site annual precipitation). (c) The slope of the relationship between a given performance metric and climate difference for each species is their climate difference sensitivity (CDS). We hypothesize that species performance should decrease away from the centre of their climate niche, and so CDS (slope) should be positive when sites are warmer or wetter than the range centre, and negative when sites are cooler or drier, given the data in (b). Thus, in our case, we predict negative CDS for temperature and positive CDS for precipitation, as indicated by solid versus dashed lines. We propose that traits may be key in explaining why some species do not follow this expected pattern. (d) We then classified the distribution of species CDSs as either negative, zero or positive, based on their slopes and the uncertainty around that estimate (see Figure 2 for real data; this figure is hypothetical). This classification simplifies species CDS into species that do better, worse or are unresponsive as site climates become cooler or more similar in precipitation to their niche centres. The asymmetric distribution in (d) reflects our hypotheses as explained in panel (c). (e and f) Finally, we assessed our hypotheses (Table 1) for what groups of species CDSs would present which trait distributions. Whether we expected a given trait to present either (e) or (f) is dependent on whether the higher trait values are indicative of resource-acquisitive or conservative strategies, whether the CDS is for ΔMAT or ΔAP, and which performance metric (colonization, extinction and abundance change) the CDS represents (Table 1). Note, (c) and the panels following are all hypothetical data and colour labels for each panel are inherited from the legend at the bottom of the figure.
Colonizations were counted as species that were absent in a turf in the first survey (2009)  With Δc s as cover change for species, s, in a turf calculated as the natural log (plus one to avoid taking the natural log of zero) of the species final cover, f, minus its initial cover, i. A table of species information found in the survey can be found in Table S1.2 in Appendix S1. and annual precipitation (AP in m/year) because of their widespread availability and strong correlation in our system with other climate variables such as potential evapotranspiration. We then took the mean of MAT and AP across the species' occurrences to estimate a species' 'climate niche centre'. Mean and median niche centres were highly correlated (r > .9 for both MAT and AP). We also used WORLDCLIM data to extract MAT and AP data for the experimental destination sites to stay consistent between site and species' climate data sources. Finally, we calculated 'climate differences' or ΔMAT and ΔAP as a species' niche centre minus the site climate (Box 1). cm), specific leaf area (SLA; cm 2 /g), leaf dry matter content (LDMC; g/g), leaf thickness (L th in mm), leaf nitrogen (leaf N; %mass), leaf carbon (leaf C; %mass) and leaf C:N ratio (%/%) and seed mass (SM in g/1000 seeds without dispersal tissues from Royal Botanic Gardens Kew, 2021) can be found in the Appendix S1. Our analyses aimed at assessing species-level trait variation, but if there is more variation among individuals of a species than between, interpretations of mean traits would be questionable. Variance partitioning analyses suggest this was not the case (Appendix S1).

| Statistical analyses
Testing our hypotheses required two steps: (1) fit the 'speciesspecific' macroecological context models, and (2) investigate how traits are associated with the direction of macroecological predictions. We only included species with trait data (n = 85). We used three response metrics: colonization, extinction and abundance change (Δc). Colonization and extinction probability were Bernoulli distributed, and Δc was normally distributed.
For step 1, we fit models with species-specific slope and intercept parameters relating climate differences to a response metric (cf. Figure 1c). Each data point represents one species per turf. This species-specific analysis is data demanding. We, therefore, dropped species that did not occur at multiple sites (<2 sites; three species) and/or had too little data (<10 observations; seven species), leaving 75 species for analysis. Additionally, transplant treatment effects were not included in the model because it further thinned power across treatments and because the main pattern of interest is how species respond to climate differences, regardless of the source of these differences. As mentioned above, a key effect of transplantation is expanding the set of temperature and precipitation environments a species experiences (Lynn, Klanderud, et al., 2021). The model was specified as: where μ is one of the three response metrics linearly predicted by species i ([species i ]) with an intercept (α) and slope (β, or climate difference sensitivity/CDS) describing how species performance relates to a climate difference metric (either ΔMAT or ΔAP). The model included site and block as group-intercept variance terms (σ 2 ).
For each species, we took the mean slope (β or CDS) posteriors generated from step 1 and classified them by direction (positive or negative) and uncertainty (degree of posterior overlap with zero). The uncertainty was classified based on the inner 68% or ±1 SD of the posterior slope estimate. If the 68% interval included zero, the sensitivity was classified as zero/unrelated. If the 68% interval did not include zero, it was further classified by its direction as either positive or negative. We simplified CDS into a directional classification rather than analysing raw slope values because (1) raw slope values as measures of strength are contingent on historical and environmental factors unassociated with climate differences (e.g. priority effects), and (2) the magnitude of the slopes may be dependent on where in the climate difference spectrum the species' were sampled and, therefore, may be unreliable as a precise measure of CDS but useful for general (1) Δ c s = ln c s,f + 1 − ln c s,i + 1 (2) = species i + species i * ΔMAT + N 0, 2 site + N 0, 2 block (step 1) characterizations of directional responses. We focused on slopes and not intercepts from the model because the slope represents how a species is predicted to respond to climate change away from its niche centre, while the intercepts represent a species' base performance when a site's climate is equal to the species' niche centre. Slopes and intercepts were also highly correlated (all r ~ .7 or greater). Additionally, some species had parameters that failed to converge (failed models checks; see below) in step 1 for colonization and extinction and were dropped for analyses in step two (final counts per analysis in Figure 2).
In step 2, we tested if the direction (negative, zero/unrelated, positive) of species climate difference sensitives (CDS) for each response metric were associated with different trait strategies (cf., where values of a trait, m, were analysed as a function of species' CDS direction (factor with levels negative, zero/unrelated, positive) with intercepts, α m . We investigated these relationships for all eight traits in All models were fit with Bayesian methods (Hobbs & Hooten, 2015) in JAGS (Plummer, 2003) in the 'R2jags' package (Su & Yajima, 2015) in R 4.0.3 (R Core Team, 2020). Model specifications and checks can be found in Appendix S1. Bayesian analyses produce posterior probability distributions of parameters allowing for probabilistic statements on the strength and direction of a term (e.g. probability a slope term, β, is greater than zero given as P(β > 0) = 0.99).
We additionally calculated the joint probability distribution (degree of overlap) between posteriors of negative and positive CDSs to (3) trait m = m * CDS (step 2) F I G U R E 2 Distributions of species climate sensitivities (step 1) used in the functional trait analyses. Histograms show the mean slope divided by their standard deviation to reflect if the posterior distribution overlaps 0 at the 68% credibility interval. CDSs were classified as 'Negative' if the CDS was less than −1, 'Positive' if the CDS was greater than 1 and 'Zero' if the CDS was between −1 and 1. These boundaries are marked by vertical lines and CDS colours for all panels are in the legend of (a). Panels are ordered by performance metric colonization (a, b), extinction (c, d) and abundance change or 'ΔCover' (e, f) modelled by climate differences (ΔMAT (a, c, e) and ΔAP (b, d, f)). Numbers above the histogram bars label the number of species that fall into a given CDS classification.
assess their degree of difference (in percentages) when one or the other had a large amount of support for being different from the zero CDS.

| RE SULTS
Species climate difference sensitivities spanned the full range of negative, zero and positive for each of the different performance metrics and climate variables (Figure 2), but the distribution of CDSs did not meet our expectations. Temperature CDSs suggested most species performed better (positive colonization and cover, negative extinction CDS) in sites cooler than their niche centre, while precipitation CDSs were more evenly distributed (Figure 2). While not consistent with our expectation from niche theory, this broad distribution of CDSs increased the power for testing associations between traits and CDSs (Figure 2 for trait sample sizes).

| Temperature climate difference sensitivity and traits
Results for all three-response metrics (i.e. colonization, extinction, abundance change) were consistent with our predictions for height, but opposite of expectations for LDMC and with little relationship of other traits to temperature difference sensitivities (Table 1). Species with higher colonization probability at sites cooler than their niche centre (positive CDSs) were shorter and had higher LDMC compared with species that decreased in colonization probability in sites cooler than their niche centre (negative CDSs). Specifically, species for which colonization probability increased as sites became cooler than their niche centres (i.e. with increasing ΔMAT), were, on average, ~40% shorter (P(β < 0) ~ 1) and had ~20% lower LDMC (P(β < 0) ~ 0.98) than species with zero CDS (Figure 3a; Table 2). This trait difference was especially substantial for height which overlapped by only 9% between positive and negative temperature CDS species. In contrast, there was high overlap between negative and positive CDS species for LDMC (~50%) even though the LDMC of negative CDS species was not different from the zero CDS species (P(β < 0) ~ 0.74). Differences in other traits among CDS groups were more variable, but species with positive CDS had lower leaf C:N than the zero CDS species (P(β < 0) ~ 0.92).
Additionally, species with positive extinction CDS tended to be taller but with higher uncertainty (P(β > 0) ~ 0.85).
Again, supporting height but not leaf hypotheses (Table 1), shorter species and species with more resource-acquisitive leaf strategies increased in abundance as sites became cooler than species' niche centres. Species that increased in cover as temperature differences increased were 73% shorter (P(β < 0) ~ 0.99) and had 26% lower LDMC (P(β < 0) ~ 1; Figure 3c; Table 2). Species with positive temperature CDS for cover were different from negative CDS species for height and LDMC (both <8% overlap).

| Precipitation climate difference sensitivity and traits
Species colonization-precipitation CDSs were weakly related to traits at best, which was counter to our hypotheses (Table 1) and consistent across response metrics. With height, species that decreased in colonization ability as sites became drier and more similar to their niche centre (negative CDSs) were, on average, 24% shorter than (P(β < 0) ~ 0.90) and species with positive CDSs were 29% taller than (P(β > 0) ~ 0.90) species with zero CDS (Figure 4a; Table 2), which generally supported predictions ( Table 1). The probability that positive and negative precipitation CDS species overlapped in height was only ~10%. Together, this suggests that taller species had increasing colonization probability while shorter species had decreasing colonization probability as site precipitation became more like their niche centre precipitation.
In contrast to our predictions (Table 1), species with lower seed mass had an increased chance of going extinct as site precipitation moved from wetter to more like species' niche centres.
Species with higher extinction probability as precipitation differences increased had 175% lower seed mass than species with zero extinction CDS for precipitation (P(β < 0) ~ 0.99; Figure 4b; Table 2). However, there was ~25% overlap in seed mass for species with positive and negative extinction CDS for precipitation, even though negative CDS was not very different from zero CDS species for seed mass (P(β < 0) ~ 0.80). Species with positive extinction CDS for precipitation also had higher leaf C:N (P(β > 0) ~ 0.89) than zero CDS species, though this pattern had high uncertainty.
Species that increased abundance as site precipitation became more similar to their niche centre were taller (P(β > 0) ~ 0.95), had higher LDMC (P(β > 0) ~ 0.97), higher leaf C:N (P(β > 0) ~ 0.93) and higher seed mass (P(β > 0) ~ 0.97; Figure 4b; Table 2). However, all of these traits except for height overlapped greatly with traits of the species negative precipitation CDS for cover (lowest overlap among them was leaf C:N at ~46%) and the direction of these trait differences from the zero CDS species was the same for positive and negative CDS groups (Figure 4c). Together, this suggests the only consistent pattern emerging for precipitation CDS was species that increased abundance as sites went from wetter to more like species' niche centres were taller than other species, supporting our hypothesis (Table 1).

| DISCUSS ION
Relating functional traits to climate difference sensitivities yielded several novel insights: (1) shorter species and species with more resource-acquisitive leaves displayed higher performance in sites increasingly cooler than their niche centre, (2) temperature sensitivity was more related to trait strategies than precipitation sensitivity, and (3) plant vegetative height (body size) was the most consistent trait for understanding species climate difference sensitivities. Generally, these results supported our a priori hypotheses for height (Table 1), but not for leaf or seed traits, as we discuss below.

F I G U R E 3
Mean trait values for species with positive, zero or negative (a) colonization, (b) extinction and (c) abundance change or 'ΔCover' climate difference sensitivities (CDSs) for temperature. Violins represent the density of the 95% posterior estimates for a given trait with the midline representing the median posterior value. Panel titles label the traits with units of height in cm, leaf dry matter content (LDMC) in g/g, leaf carbon to nitrogen ratio (leaf C:N) is a unitless ratio (%/%) and seed mass in g/1000 seeds. Asterisks (*) indicate a given CDS is different from the zero group at the 95% credibility level (see Table 2). Numbers in each panel are Bayesian R 2 values.

| Does species' performance peak away from their temperature niche centre?
A surprising and novel result from our analyses in step 1 is that species performance generally peaked in sites cooler than their niche centre rather than near their niche centre (Figure 2a,c,e). This aligns, through new methodology, with a growing body of evidence that species do not perform best towards either their geographic or climatic niche centre (Dallas et al., 2017;Pironon et al., 2017;Sporbert et al., 2020). We offer three non-mutually exclusive hypotheses for this result that deserve more attention in future work. First, species' climate optima are likely cooler than can be estimated from current distributions because (a) species current ranges are lagging behind the pace at which the climate has already changed (Alexander et al., 2018;Svenning & Sandel, 2013), and (b) many species have not yet recolonized the total habitable climate space that has been left open since the last glacial maximum (Knight et al., 2020;Normand et al., 2011). Both mechanisms suggest an underfilling of species' potential cold range edge in their global distribution that would result in the 'optimum cooler than centre' pattern. Second, intensified biotic interactions may limit the performance of a species near the centre of their fundamental niche (Louthan et al., 2018;Lynn, Miller, & Rudgers, 2021), representing a truncation of the fundamental niche to the realized niche (Hutchinson, 1957). Finally, species performance in a given temperature environment is dependent on other niche dimensions like nutrient availability (Chapin & Shaver, 1985). Therefore, both characterizations of species temperature niche and observations within our sites may be confounded by failing to consider other niche dimensions. The multidimensional and interacting niche requirements and histories of species will inevitably disrupt attempts to generalize by simplifying investigations down to a few aspects of their niche (e.g. temperature). Nevertheless, the level of predictive success that such simplified models can achieve should TA B L E 2 Summary statistics for analyses of how climate difference sensitivity (CDS) groups are related to traits. Note: Estimates are sorted by performance metrics, climate difference variables and traits. The estimates are presented by how much trait values for the 'negative' and 'positive' CDS groups differ from the 'zero' CDS, indicated by the sign. The parentheses in the negative and positive CDS columns refer to the probability that the posterior distribution of the estimate is greater than 0 (P(β > 0)). Estimates that have 95% credibility intervals that do not overlap 0 and are different from the 'zero' CDS group are bolded. Finally, we present a Bayesian R 2 for each model (see Section 2). All trait estimates were log transformed. See Figures 3 and 4 for predicted trait means from these analyses.
inspire confidence that our base assumptions are on track and general prediction is obtainable.
One possible caveat to our interpretation is that the range of temperatures and environments investigated outside the species' niche centres was narrow relative to the breadth of environments a species occupies across their entire range. Expanding the climate range of performance observations is a key next step in assessing this framework. Furthermore, progress in predicting species responses to warming using their temperature niche will come from increased measurements of physiological performance over temperature gradients (Michaletz et al., 2015;Perez et al., 2023;Reich et al., 2015;Yamori et al., 2014) which will better characterize a species niche.
Our work presented here can serve as the prior information to inform such physiologically based investigations and assess/resolve the issues of using observed distributions to infer a species' niche characteristics.

F I G U R E 4
Mean trait values for species with positive, zero or negative (a) colonization, (b) extinction and (c) abundance change or 'ΔCover' CDSs for precipitation. Violins represent the density of the 95% posterior estimates for a given trait with the midline representing the median posterior value. Panel titles label the traits with units of height in cm, leaf dry matter content in g/g, leaf carbon to nitrogen ratio (leaf C:N) is a unitless ratio (%/%) and seed mass in g/1000 seeds. Asterisks (*) indicate a given CDS is different from the zero group at the 95% credibility level (see Table 2). Numbers in each panel are Bayesian R 2 values.

| Plant height, body size and prediction under climate change
The consistent relationship of plant vegetative height with climate difference sensitivities adds to a large and growing body of evidence that organismal body size is a key determinant of species' climate distributions and responses. Taller and larger sized plants tend to have greater competitive effect within a community (Grime, 1973), suggesting that the lower performance of taller species in sites cooler than their niche centre in our system may reflect a stress tolerance-competitive ability tradeoff and their relative competitive dominance in the communities under optimal abiotic conditions. The strong pattern that shorter species performed better in sites increasingly cooler than their niche centre suggests that shorter stature is an advantage in cooler, more stressful temperature environments (Körner, 2016). This and past work supports this hypothesis (Table 1), given both experimental (Guittar et al., 2016) and observed warming  drive height increases via species turnover within communities. Body sizes' utility in predicting species responses to climate change holds across biomes and taxonomic groups (Fuller et al., 2021;Gardner et al., 2014;Ma et al., 2021;Pardee et al., 2022;Sheridan & Bickford, 2011). These firm correlational bedrocks of plant height/body size and climate change responses should move into developing and testing more mechanistic theoretical models of how climate change will drive shifts in communities and ecosystems (Choi et al., 2016;Schramski et al., 2015).

| Lower leaf investment favoured in relatively colder environments?
We hypothesized that species with more resource-conservative leaves (e.g. high LDMC) would perform better as sites became increasingly cooler than their niche centre (Table 1). Conservative, high LDMC leaves generally relate to higher cold tolerance (Ladwig et al., 2018) while species with resource-acquisitive leaves have higher photosynthesis and growth rates (Kazakou et al., 2006;Wright et al., 2004) and are better poised to take advantage of a favourable climate. Our unexpected finding of a relative advantage of low LDMC species in cooler than niche centre climates may be the result of short growing seasons and potentially high nutrient environments selecting for cheap, fast, deciduous leaves. Plants occupying higher elevations with shorter growing seasons can have lower LDMC leaves than plants in lower elevations (Venn et al., 2014) and lower LDMC species senesce earlier than high LDMC leaves (Bucher & Römermann, 2021). High-nutrient environments tend to select for lower LDMC leaves because the advantage of investing carbon to protect mineral nutrients in leaves is low if replacing the nutrients is easy (Daou et al., 2021). Given long dormant periods and high nutrient availability, the optimal strategy for species in environments colder than their niche centre may be low-LDMC, resourceacquisitive leaves. We further hypothesize that these physiological advantages of low LDMC leaves in sites cooler than their range centre may confer a competitive advantage over other species at the sites with more conservative leaves (Dong et al., 2020;Kunstler et al., 2016). Testing this hypothesis would require cross-range transplant experiments paired with nutrient additions and models of competitive hierarchies, where a given species' performance is modelled by the difference between its trait value and its competitors (Kunstler et al., 2012).

| Seed mass was not strongly associated with species climate difference sensitivities
Seed mass, as an estimate of maternal resource investment without dispersal, was generally unrelated to climate difference sensitivities, counter to our hypotheses (Table 1). Globally, seed mass across species tends to increase with temperature and precipitation (Moles et al., 2014), suggesting that greater seed mass can lead to greater establishment and survival probability in environments with high competition for light (DeMalach et al., 2019;Westoby, 1998).
Instead, we found weak evidence that smaller seeded species had greater extinction probability as site precipitation became more like their niche centre. This result was contrary to our main hypothesis, though evidence from the literature is mixed on how seed mass relates to a species climate response (Table 1). Past work has shown that species vary considerably in their optimal temperature and precipitation for germination (Gya et al., 2023;Vandvik et al., 2017), and other work suggests that optimal germination temperatures do not necessarily relate to a species' distribution (Marques et al., 2014).
Seed mass' lack of relation to climate difference sensitivities, even in models of colonization, may stem from disparities between the climate where a species occurs and the narrower weather window when a species germinates. Additionally, species in this system exhibit a range of clonal reproduction strategies that may be more important for recruitment dynamics than seed strategies (Guittar et al., 2016), further explaining the lack of differences among species in seed mass with CDS.

| Traits are unrelated to species precipitation difference sensitivity
We found less support that traits were related to precipitation difference sensitivities-a simple vote count of how many relationships had 95% CIs that did not include zero across all traits and models (24 total) for precipitation versus temperature was one versus 10, respectively. Almost all the species (88%) had ranges that were drier than our sites, suggesting that precipitation variation in this oceanic region (~800-4400 precipitation mm/year) is a relatively minor driver of vegetation dynamics (reviewed in Wilcox et al., 2017). Nevertheless, we found weak evidence that taller species increased in abundance and had higher colonization probability as site precipitation became more like their niche centre precipitation. These results are in slight contrast with evidence that higher precipitation favours taller, larger bodied plants (Liu et al., 2019), but support our hypothesis that taller stature confers a competitive advantage at a species niche centre (Table 1). This further suggests that the extremely high precipitation found in our study system may actually act as a stressor that shortens growing season length via snowpack (Rixen et al., 2022) and may restrict photosynthetically active radiation via clouds (Nijp et al., 2015), rather than leading to higher productivity Seddon et al., 2016) and more competition for light. Therefore, given the extremity of this system, the precipitation results should not be extrapolated globally and sensitivity to precipitation deserves more study in other systems. Additionally, root traits may be better positioned to determine plant water use strategies than the aboveground traits we assessed (Li et al., 2017;Zwicke et al., 2015).

| Limitations
There are several caveats to consider when interpreting our results. First, because of data limitations, we assumed that the effects of climate transplant treatments were contained in climate difference estimates. This assumption is false in absolute terms, given that transplants also change the plant, herbivore, pollinator, etc. community in which a turf is embedded. Nevertheless, prior work suggests that this simplifying assumption is reasonable (Lynn, Klanderud, et al., 2021;Vandvik et al., 2020 Finally, while a strength of our study is that it is based on trait data collected from within the study system, our trait data were obtained from individuals in the natural communities outside of the observed turfs (except for seed data), which could have two consequences. Given the prominence of intraspecific trait variation within and across communities along environmental gradients (Siefert et al., 2015), trait values measured in one site may not reflect the trait expression of the individuals in permanent resampling plots in another site. We recognize that finer-grained trait data (e.g. site specific) might increase predictive power, but such data are difficult and time-consuming to obtain and thus would not be useful in making broad-scale predictions of responses to climate change. Additionally, intraspecific variation in our system was less than interspecific variations (Appendix S1), suggesting our trait sampling protocols provided useful data for prediction, as we found. Alternatively, past work suggests that traits may be most useful when viewed as syndromes of species measured in a common environment that reflect differences across species (Fridley et al., 2016;Grime et al., 1997;Reich, 2014;Shipley et al., 2016). However, this trait screening method would have been logistically and financially difficult in our system given the number of potential species (> 140).

| CON CLUS IONS
Generally, shorter plant species with more resource-acquisitive leaves had higher performance as sites became cooler than their niche centres. We found little evidence that seed mass was related to species climate difference sensitivities. Additionally, trait relationships to precipitation difference sensitivities were generally weaker than to temperature differences. Importantly, the results support the foundations of functional trait ecology (Shipley et al., 2016) and suggest the functional approach can be valuable for predicting future community and ecosystem responses to global change. The results provide a novel, mechanistic, trait-based explanation for the recent and puzzling findings that species do not perform best in their expected climatic optima. Furthermore, we suggest that this analysis framework is adaptable to any dataset with spatial/environmentally distributed observations of species performance over time that can be paired with traits. Expanding this framework to other systems is crucial to fully assess its potential as a tool for conservation.

ACK N O WLE D G E M ENTS
We thank the editor Prof. J.S. Cabral and two anonymous reviewers for their thoughtful comments that improved the manuscript.
We thank Linn C. Krüger, Siri L. Olsen and numerous interns and students for their assistance in data collection. Funding came from the Norwegian Research Council (NORKLIMA grants #184912 and #244525) to V.V. J.S.L. was supported by a UiB Postdoctoral Fellowship in Community Ecology.

CO N FLI C T O F I NTE R E S T S TATE M E NT
The authors declare that they have no conflict of interest.

PEER R E V I E W
The peer review history for this article is available at https:// www.webof scien ce.com/api/gatew ay/wos/peer-revie w/10.1111/ ddi.13718.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data are available at Dryad (https://doi.org/10.5061/dryad.63xsj 3v69) and code for all analyses can be found on GitHub (https:// github.com/jslyn n33/macro conte xt-and-traits).