Plant traits are useful for predicting how species may respond to environmental change and/or influence ecosystem properties. Understanding the extent to which traits vary within species and across climatic gradients is particularly important for understanding how species may respond to climate change. We explored whether climate drives spatial patterns of intraspecific trait variation for three traits (specific leaf area (SLA), plant height, and leaf nitrogen content (Nmass)) across 122 grass species (family: Poaceae) with a combined distribution across six continents. We tested the hypothesis that the sensitivity (i.e., slope) of intraspecific trait responses to climate across space would be related to the species’ typical form and function (e.g., leaf economics, stature, and lifespan). We observed both positive and negative intraspecific trait responses to climate with the distribution of slope coefficients across species straddling zero for precipitation, temperature, and climate seasonality. As hypothesized, variation in slope coefficients across species was partially explained by leaf economics and lifespan. For example, acquisitive species with nitrogen-rich leaves grew taller and produced leaves with higher SLA in warmer regions compared to species with low N_mass. Compared to perennials, annual grasses invested in leaves with higher SLA yet decreased height and N_mass in regions with high precipitation seasonality. Thus, while the influence of climate on trait expression may at first appear idiosyncratic, variation in trait-climate slope coefficients is at least partially explained by the species’ typical form and function. Overall, our results suggest that a species’ mean location along one axis of trait variation (e.g., leaf economics) could influence how traits along a separate axis of variation (e.g., plant size) respond to spatial variation in climate.

We measured traits of grasses across the Bay Area of California and obtained additional records from published papers (Appendix 1) and trait databases, including TRY (Kattge et al. 2011), TTT (Bjorkman et al. 2018), BIEN (Maitner et al. 2018), BROT2 (Tavşanoğlu and Pausas 2018), and AusTraits (Falster et al. 2021) (see published paper for literature cited). For this analysis, we focus on three traits of interest: specific leaf area (SLA), plant height, and mass-based leaf nitrogen content (N_mass). All trait measurements were georeferenced with latitude and longitude coordinates. For each trait measurement, we extracted and paired the following high-resolution (30 arc sec, ~1km) climate statistics from CHELSA V2.1 (Karger et al. 2017; https://chelsa-climate.org/): mean annual precipitation (MAP), mean annual temperature (MAT), precipitation seasonality (PS; the standard deviation of the monthly precipitation estimates expressed as a percentage of the mean of those estimates (i.e. the annual mean)), temperature seasonality (TS; standard deviation of the monthly mean temperatures), mean monthly precipitation of the warmest and coldest quarter of the year (Pwarm and Pcold, respectively), and mean monthly temperature of the warmest and coldest quarter of the year (Twarm and Tcold, respectively).

To avoid over-weighting regions that were heavily sampled (i.e., parts of Europe and North America, see Fig. 2 in paper), we aggregated our trait-climate dataset by rounding latitude-longitude coordinates to the nearest first decimal point and averaging climate and trait values for a species within that binned coordinate. We then subset this binned dataset to include only species for which we had at least 10 records spanning a MAP gradient of 100 mm, a MAT gradient of 2℃, and a correlation between MAP and MAT of no more than 0.8. This was done to prevent fitting models when MAP and MAT were highly collinear or when all measurements were made over a narrow range of climate values which might result in extreme slope coefficients (Sandel et al. 2021). Based on these criteria, our final global grass trait-climate dataset spanned six continents, covered all of Earth’s major terrestrial biomes, and included 2,648 measurements of SLA (n= 109 species), 1,359 measurements of N_mass (n = 61 species), and 1,439 measurements of plant height (n = 66 species) (Fig. 2).

NOTE: THESE RAW TRAIT DATA ARE NOT SHARED HERE DUE TO RESTRICTIONS FROM DATABASES MENTIONED ABOVE. THE TRAIT-CLIMATE SLOPE COEFFICIENTS DESCRIBED IN THE NEXT PARAGRAPH ARE SHARED.

Trait data were log-transformed prior to analyses to meet assumptions of normality. For each species and trait, we ran two separate simultaneous autoregressive (SAR) models predicting trait values from climate variables using the errorsarlm() function in the spdep package (Bivand et al. 2015), with the neighborhood of a point being defined as the three nearest points. The first model included mean climate characteristics (MAP, MAT, PS, TS) while the second included mean monthly climate of the warmest and coldest quarters of the year (Pwarm, Pcold, Twarm, and Tcold). The use of SAR models accounts for spatial autocorrelation in our data, a common phenomenon in ecology where nearby observations are more similar than would be expected by chance (Legendre 1993). These models each produced 12 trait-climate slope coefficients (3 traits, 4 climate variables). Using Student t-tests, we assessed whether the mean of each SAR slope coefficients across species was significantly different from zero (Bonferroni-adjusted p-values for 12 independent tests; alpha = 0.004).

To test whether variation in intraspecific sensitivity of traits to climate could be explained by mean species traits, we ran phylogenetic generalized least squares (PGLS) regression models with the intraspecific trait-climate slope coefficients (e.g., MAP vs. SLA slope) as the dependent variable and the following mean species traits as independent variables: SLA, Nmass, height, lifespan (i.e., perennial or annual), and photosynthetic pathway (i.e., C₄ or C₃ photosynthesis). We used PGLS models to account for the possibility that more closely related species have more similar responses to climate than would be expected by chance (see Fig. S1 for complete phylogenetic tree). In our models, we log-transformed continuous mean traits (SLA, N_mass, and Height). To simplify these global models and determine which mean traits were most important for understanding variability in the trait-climate slope coefficients, we performed an automated model selection using the dredge() function in the MuMln package (Barton and Barton 2015). We then performed model averaging on those models with a delta AIC_c of <2, and produced partial residual plots to visualize the effects of individual significant predictor variables on variation in the slope coefficients while also considering other components of the final model.

Finally, we explored the correlation between mean continuous traits for all species used in this study (n=122) using standard major axis (SMA) regression (sma function in the smatr package; Warton et al. 2018) and tested for trait differences based on photosynthetic pathway and lifespan using two-sample t-tests. This was done to confirm that the mean traits for our species were similarly coordinated according to global spectrum of plant form and function (Diaz et al. 2016). All statistical analyses and data visualization were performed in R (version 4.2.2).

All statistical analyses were run in R (version 4.2.2).