## Data overview
To elucidate differences between patterns in δ15N across space versus one site through time, we used two complementary approaches. To build our spatial explanatory regressions, global foliar δ15N were obtained from the dataset available from (Craine et al. 2018, 2019), consisting of over 43,000 samples spanning all continents except Antarctica and acquired from 258 datasets over 128 unique years (between 1876-2017). The temporal component of this study was conducted at the Jornada Basin Long Term Ecological Research (LTER) site (32.56 latitude, -106.78 longitude; Las Cruces, NM, USA). Rainfall manipulation treatments commenced in multiple years and spanned a range of experimental durations at the time of field sampling, from 4 years to 14 years (Table S1). We present multiple years and time durations of rainfall manipulations within one site and categorize this portion of our analyses as "temporal" as a means to distinguish the nature of these data compared to the single-time point samples in the global, spatial dataset. We recognize that foliar δ15N provides limited interpretation compared to whole-plant δ15N values, given that differential fractionation may occur within plant stems and roots during assimilation. Nevertheless, the results presented here match global studies and therefore are generalizable.
## Synthesis of spatial relationships
All spatial isotope data were pooled, regardless of plant functional type. Craine et al. assigned mean annual precipitation (MAP) and mean annual temperature (MAT) to each data point based on geographic location from (New et al. 2002). In addition, we included foliar isotope data made available from the National Ecological Observatory Network (NEON; data product ID DP1.10026.001) (NEON [National Ecological Observatory Network] 2022) across North America over 6 years (2016–2021). Climate data were obtained from the NEON field site metadata (NEON [National Ecological Observatory Network] 2022), available on the Field Sites webpage. Isotopic data were then site-averaged according to the latitude and longitude rounded to the nearest tenth decimal place.
Site-averaged foliar δ15N for the global spatial dataset were then regressed against mean annual precipitation and mean annual temperature using multiple linear regression:
* δ15N ~ precipitation (eq. 1)
* δ15N ~ temperature (eq. 2)
* δ15N ~ precipitation + temperature (eq. 3)
* δ15N ~ precipitation + temperature + precipitation\*temperature (eq. 4)
Some continents exhibited mean annual precipitation values that were not normally distributed (i.e., Australia). Thus, the above regressions were also conducted using ln(precipitation):
* δ15N ~ precipitation (eq. 5)
* δ15N ~ ln(precipitation) + temperature (eq. 6)
* δ15N ~ ln(precipitation) + temperature + ln(precipitation)\*temperature (eq. 7)
The best global model was then selected using Akaike Information Criteria (Sakamoto et al. 1986); the model that met our criteria had the lowest AIC with ΔAIC > 2, (eq. 7). The selected model was then applied to all continental-scale analyses for uniformity and comparisons. All spatial statistical analyses were performed in R version 4.2.2 (R Team 2013). Regression assumptions were tested and met for all analyses.
## Study site description, experimental design, and stable isotope analyses
The Jornada LTER study site description, experimental design, and isotopic analyses for this temporal dataset are described in EDI package knb-lter-jrn.210586001.
## Statistical analyses of temporal data
To assess the response of δ15N to natural interannual variability in precipitation, within the Jornada Basin LTER data set, we analyzed site-level control plots across our study site that were not parsed temporally using a linear mixed effects model in the lme4 package (Bates et al. 2014):
* δ15N ~ precipitation + (1 | experiment / plot) + (1 | year sampled) (eq. 8)
For temporally parsed statistics with the Jornada Basin LTER dataset, which included rainfall treatments. We did not include temperature because it was a relatively constant during the study period and at our study site (Currier & Sala 2022). It is worthy to note that the random effect variances were estimated as zero for these analyses. Nonetheless, we felt that it was important to retain the random effect terms (year and plots within each experiment) to reflect the semi-repeated sampling and blocked nature of our experimental design. Bolker et al. (2009) affirm that in cases like this, the results remain unchanged and the random effect parameters may be retained.
Finally, rates of acclimation for changes in δ15N versus precipitation through time at the Jornada Basin LTER were calculated by estimating the time until each temporal slope would match the global and individual continental slopes of δ15N versus precipitation, where precipitation in this case was left untransformed for the spatial slopes:
* slope_spatial ~ slope_JRN\*time_years + intercept_JRN (eq. 9)
All temporal statistical analyses were performed in R version 4.2.2 (R Team 2013). Regression assumptions were tested for all analyses. In the case for the temporal soil analyses, the dependent variable (δ15N) exhibited some non-normality, which was not rectified through transformation. Nevertheless, the residuals of all soil models were tested for and exhibited normality.
## References
Bates, D., Maechler, M., Bolker, B. & Walker, S. (2014). lme4: Linear mixed-effects models using Eigen and S4. R package version, 1, 1–23.
Bolker, B.M., Brooks, M.E., Clark, C.J., Geange, S.W., Poulsen, J.R., Stevens, M.H.H., et al. (2009). Generalized linear mixed models: a practical guide for ecology and evolution. Trends in Ecology & Evolution, 24, 127–135.
Currier, C.M. & Sala, O.E. (2022). Precipitation versus temperature as phenology controls in drylands. Ecology, e3793.
NEON (National Ecological Observatory Network). (n.d.). Explore field sites. Metadata accessed from https://www.neonscience.org/field-sites/explore-field-sites on January 24, 2022.
NEON (National Ecological Observatory Network). (n.d.). Plant foliar traits (DP1.10026.001), RELEASE-2022. https://doi.org/10.48443/kmc7-8g05. Dataset accessed from https://data.neonscience.org on January 24, 2022.
New, M., Lister, D., Hulme, M. & Makin, I. (2002). A high-resolution data set of surface climate over global land areas. Clim. Res., 21, 1–25.
Sakamoto, Y., Ishiguro, M. & Kitagawa, G. (1986). Akaike information criterion statistics. Dordrecht, The Netherlands: D. Reidel, 81, 26853.
Team, R.C. (2013). R: A language and environment for statistical computing.
## Data provenance
