Biodiversity facets, canopy structure and surface temperature of grassland communities

Canopy structure is an important driver of the energy budget of the grassland ecosystem and is, at the same time, altered by plant diversity. Diverse plant communities typically have taller and more densely packed canopies than less diverse communities. With this, they absorb more radiation, have a higher transpiring leaf surface, and are better coupled to the atmosphere which leads to cooler canopy surfaces. However, whether plant diversity generally translates into a cooling potential remains unclear and lacks empirical evidence. Here, we assessed how functional identity, functional diversity, and species richness of grassland communities in the Jena Experiment predict the mean and variation of plant surface temperature mediated via effects of canopy structure. Using terrestrial laser scanning, we estimated canopy structure describing metrics of vertical structure (mean height, LAI), the distribution (evenness), and the highest allocation (center of gravity) of biomass along height strata. As metrics of horizontal structure, we considered community stands gaps, canopy surface variation, and emergent flowers. We measured surface temperature with a thermal camera. We used SEM models to predict biodiversity effects on the surface temperature during two seasonal peaks of biomass. Before the first cut in May, herb-dominated communities directly promoted lower leaf surface temperatures. However, communities with a lower center of gravity (mostly herb-dominated) also increased canopy surface temperatures compared with grass-dominated communities with higher biomass stored in the top canopy. Grass-dominated communities showed a smaller variation of surface temperatures, which was also positively affected by species richness via an increase in mean height. In August, mean surface temperature decreased with increasing community clumpiness and LAI. The variation of surface temperature was greater in herb-dominated than in grass-dominated communities and increased with plant species richness (direct effects). Synthesis: The mean and variation of canopy surface temperature were driven by differences in functional group composition (herbs- vs. grass dominance) and to a lesser extent by plant diversity. These effects were partly mediated by the metrics of canopy structure but also by direct effects unrelated to the structural metrics considered.There are no missing values. It follows a README *.txt file with further documentation. Funding provided by: Deutsche ForschungsgemeinschaftCrossref Funder Registry ID: http://dx.doi.org/10.13039/501100001659Award Number: FOR 1451Funding provided by: Universität LeipzigCrossref Funder Registry ID: http://dx.doi.org/10.13039/501100008678Award Number: Funding provided by: German Centre for Integrative Biodiversity Research*Crossref Funder Registry ID: Award Number: FZT 118Funding provided by: German Centre for Integrative Biodiversity ResearchCrossref Funder Registry ID: Award Number: FZT 118Our field data was collected within the Trait-Based Biodiversity Experiment in 2014 (TBE; Ebeling et al., 2014) at the Jena Experiment site (Thuringia, Germany; 50°55´ N, 11°35`E, 130 m above sea level (Roscher et al., 2004; Weisser et al., 2017). We conducted this study in 92 plots of the Trait-Based experiment (3.5 m x 3.5 m size), comprehending the two species pools, with a gradient of plant species richness of 1 to 8 species. We performed a non-destructive measurement of plant community canopy structure at high resolution, we used a terrestrial laser scanner (TLS) Faro Focus 3D X330 (FARO Technologies Inc., 2011). We scanned 92 plots on April 31st (the first peak of biomass) and August 20th, 2014 (the second biomass peak). The TLS was mounted upside-down on a tripod that was elevated 3.35 m above ground level. The legs of the tripod were centered on permanent survey markers to guarantee identical scanning areas on both dates. We extracted an area of 3.75 m² (1.5 m x 2.5 m) in each plot below the scanner to reduce the effect of shadows within scans. The point clouds of the 92 plots were filtered using statistical outlier removal (SOR) and noise filter. We used the 3D point clouds from terrestrial laser scanning to calculate metrics characterizing vertical and horizontal dimensions of the community canopy structure. We produced height-based metrics from the point cloud of each community. We used mean height as the first vertical dimension metric. To characterize vertical space-filling properties, we calculated the evenness and the center of gravity of the point cloud. Evenness reports the homogeneity of the point cloud density in their vertical distribution, while the center of gravity identifies the height stratum (definition see below) with the highest density of points (Spehn et al., 2000; Barry et al., 2020). As a baseline for only these two vertical metrics (evenness and center of gravity), we calculated voxel grids from the 3D point cloud for each plot. For each scan, a voxel grid of 5 cm was created containing at least one laser return, and the volume was then calculated as the product of the cell area and the attributed height. We used the function 'vox' from the R package VoxR (Lecigne et al., 2014). We used the voxel grids to define five different strata of height (0.3 - 20 cm, 20 - 40 cm, 40 - 60 cm, 60 – 80 cm, and 80 - 100 cm). For every stratum, we applied the method 'Sum of Voxel,' which calculated the sum of all voxels separately for each of the five strata. As a result, we obtained volumetric data based on 3D point clouds for five different strata and the community canopy height. Based on this information, the evenness metric represents the mean proportion of filled voxels across strata of vegetation height, calculated as the sum of all five voxel strata volumes divided by 5. The center of gravity, in turn, used the volume of voxel grids per height strata to identify the location with the highest density of points. This location was measured in terms of the height-weighted average volume allocation of the community. We then calculated the center of gravity by multiplying each stratum's volume with the mean height of the strata and dividing by the total community volume. Center of gravity range from 1 to 5, in which 1 is the bottom layer (0-20 cm) and five the top canopy (80-100 cm). Further, the leaf area index was also measured at the same time in all 92 plots using the LAI-2000 plant canopy analyzer (LI-Cor, Inc, 2013). Ten random measurements were averaged to a mean of LAI value per plot. Hence, we used LAI as an additional vertical dimensional metric to characterize plant ground area covered by the plant community. To assess the horizontal heterogeneity of the plant community for each plot, we also calculated two horizontal metrics describing the canopy surface variation and clumpiness. We used the surface reconstruction method, which fits a mesh on the 3D point cloud density of each plot (the filtered point clouds and not voxel grids) (Attene & Spagnuolo, 2000). We applied the Poisson Surface Reconstruction method, which fits a mesh on all oriented points (perpendicular vectors to the tangential plane to the surface at that point) (Kazhdan & Hoppe, 2019). After producing the surface mesh for all plots, a surface area of the mesh in square meters was calculated and divided by the area of the plot (3.75 m²). The variation metric is a dimensionless ratio between the mesh surface area and the ground area. For clumpiness, we evaluated the size and distribution of clusters in the spatial arrangement of the point cloud into two dimensions based on the rasterized 3D point clouds. For this, we computed Geary´s index, an identifier of cluster points with similar attributes, assessed by the pixel spatial autocorrelation. We used the function Geary from the R package "raster". The two response variables, the mean and the coefficient of variation (hereafter CV) of community surface temperature were obtained using the Testo 882 Thermal Imaging Camera, which also recorded RGB images of all 92 plots (for example, Figure 3). We obtained the thermal data and terrestrial laser scans within two days. All thermal measurements were carried out around noon (12:30 – 13:30), at 150 cm height, and facing north. The thermal camera settings controlled the canopy's emissivity as 0.95 with reflectance temperature at 20°C. The sensor detects long-wave infrared radiation in the spectral range from 7.5 to 14 μm and has a thermal sensitivity of 50 mK at +30°C and accuracy of ±2.0°C. With the thermal matrix (registered pixel temperature with an original resolution of 640 x 480), we computed the mean and the coefficient of variation of surface temperature for each plot. As flower heads are often 10 K warmer than the surrounding leaves of herbs and grasses, we calculated the Normalized Green-Red Difference Index (NGRDI, the difference between the green and red bands divided by their sum (Pérez et al., 2000). Further, we also calculated the Normalized Green-Blue Difference Index (NGBDI, the difference between the green and the blue bands divided by their sum (Wang Xiaoqin et al., 2015) better to distinguish the hot spots of inflorescences and green vegetation. We expected that biodiversity effects on mean canopy surface temperature are indirectly mediated by predictors related to vertical structure metrics, while biodiversity effects on temperature CV are mediated by the horizontal structure. To test these assumptions further, we constructed a more detailed formal hypothesis using linear mixed-effects models within a PiecewiseSEM (Lefcheck, 2016). we ran the initial SEM model as a list of causal relationships between canopy structure and biodiversity facets. The last linear model inside the SEM was between the mean and CV of surface temperature and biodiversity facets (functional identity, dispersion, and species richness) and all canopy structure metrics to test the fit of the model to the data. Second, we inspected this initial SEM model results for goodness-of-fit tests for both the full and causal relationships, we then added the predictors that significantly improved the model fit with P values higher than 0.05