Dataset and R code for "Relationship between wind speed and plant hydraulics at the global scale"

Data collection

      Plant hydraulic traits and height data were obtained from three sources: (1) field measurements of plant hydraulics for 210 forest species in China; (2) the TRY Plant Traits Database (https://www.try-db.org/TryWeb/Home.php; Kattge et al., 2020); and (3) published literature. For the latter we conducted searches on Web of Science, Google Scholar, and China National Knowledge Infrastructure (http://www.cnki.net) using keywords such as “hydraulic traits,” “xylem hydraulic conductivity,” “xylem vulnerability,” “water potential at 50% loss of hydraulic conductivity,” “xylem embolism resistance,” and “plant water conductivity.” A substantial portion of data in our study were obtained from published literature (Choat et al., 2012; Gleason et al., 2016) and the Xylem Functional Traits Database (XFT; https://xylemfunctionaltraits.org).

To minimize ontogenetic and methodological variation, we only included data that met the following criteria: (a) plants were grown in natural ecosystems, excluding greenhouse and common garden experiments; (b) measurements were made on adult plants and not on seedlings; (c) hydraulic traits were measured on terminal stem or branch segments in the sapwood at the crown; (d) trait data were calculated as the mean value for each species at the same site when data were from multiple sources; and (e) data values > 3 SD (standard deviation) were removed to reduce the effect of outliers (Carmona et al., 2021); (f) height data were reported at the same site where plant hydraulic traits were measured. 

Climate data were obtained either from the original reports or from WorldClim version 2 (http://worldclim.org/version2; Fick & Hijmans, 2017; Table 1) if the original data were not available. The following variables measured at ~1 km2 scale were extracted from WorldClim: mean annual wind speed (μ), mean annual precipitation, mean annual temperature, precipitation seasonality, temperature seasonality, wind seasonality (μS; coefficient of variation across monthly measurements × 100), precipitation of driest month, and minimum temperature of coldest month. The VPD data were extracted from the TerraClimate dataset (http://www.climatologylab.org/terraclimate.html; Abatzoglou et al., 2018). Annual PET (potential evapotranspiration) data were extracted from the CGIAR-CSI consortium (http://www.cgiar-csi.org/data; Zomer et al., 2008). Moisture index (MI), which is the ratio of precipitation to PET. 

Data analysis

Trait and environment data were log10-transformed to achieve approximate normality, except for P50 and temperature data. We first calculated correlations among all climatic variables and for subsequent analyses retained only those variables with correlation coefficients lower than |0.7| (Dormann et al., 2013). We then ran independent multiple linear models for each trait of interest using the retained climatic variables. Model selection based on a corrected Akaike information criterion and using the R package glmulti (Calcagno & de Mazancourt, 2010), identified the best linear model for each trait. The R package ‘visreg’ (Breheny & Burchett, 2017) was used to visualize the partial relationships between wind speed and hydraulic traits. Two-dimensional contour plots were then used to explore and visualise how plant hydraulic traits varied simultaneously with wind speed and moisture index.

To quantify the strength of wind effects on plant hydraulics, models with wind parameters μ and μS included were compared to those without these wind parameters. 

To test for differences in the relationship between hydraulic traits and wind speed among species grouped into different climatic regions (i.e., dry vs. wet sites, and tropical vs. temperate regions), we used standardized major axis (SMA) analyses using the R package ‘smatr’ (Warton et al., 2012). A grouping factor was added in each SMA to test whether species groups share a common slope, with p > 0.05 indicating species groups share a common slope.

Variance partitioning analysis was performed using the ‘rdacca. hp’ R package to quantify the degree to which the effect of wind speed was independent from other climatic variables (Lai et al., 2022). The individual contribution of each predictor was estimated in this analysis. This analysis also helped to illustrate the significant values of climatic variables on plant hydraulics.

A Random Forest machine-learning algorithm (implemented using the R package ‘randomForest’) was utilized to further assess the relative importance of environmental variables for each plant hydraulic trait (Breiman, 2001). To avoid multicollinearity, this analysis only included variables with correlation coefficients lower than |0.7|. A higher value of the mean decrease in accuracy (%IncMSE) indicates the increased importance of a variable (e.g., a %IncMSE value of 50 indicates that the overall mean square error would increase by 50% if that variable were to be excluded from the analysis). This provides a measure of a variable's importance in estimating the value of the target variable across the trees in the forest.