Aim: Addressing how woody plant species are distributed in space can reveal inconspicuous drivers that structure plant communities. The spatial structure of conspecifics varies not only at local scales across co-existing plant species but also at larger biogeographical scales with climatic parameters and habitat properties. The possibility that biogeographical drivers shape the spatial structure of plants, however, has not received sufficient attention.

Location: Global synthesis.

Time period: 1997 - 2022.

Major taxa studied: Woody angiosperms and conifers.

Methods: We carried out a quantitative synthesis to capture the interplay between local scale and larger scale drivers. We modelled conspecific spatial aggregation as a binary response through logistic models and Ripley’s L statistics and the distance at which the point process was least random with mixed effects linear models. Our predictors covered a range of plant traits, climatic predictors and descriptors of the habitat.

Results: We hypothesized that plant traits, when summarized by local scale predictors, exceed in importance biogeographical drivers in determining the spatial structure of conspecifics across woody systems. This was only the case in relation to the frequency with which we observe aggregated distributions. The probability of observing spatial aggregation and the intensity of it was higher for plant species with large leaves but further depended on climatic parameters and mycorrhiza.

Main Conclusions: Compared to climatic variables, plant traits perform poorly in explaining the spatial structure of woody plant species, even though leaf area is a decisive plant trait that is related to whether we observe homogenous spatial aggregation and its intensity. Despite the limited variance explained by our models, we found that the spatial structure of woody plants is subject to consistent biogeographical constraints and that these exceed beyond descriptors of individual species, which we captured here through leaf area.

On the 8^th of September 2022 we carried out a search in the Web of Science with the search string “(Ripley's K function) AND (forest)”. The search yielded 356 hits. We screened those 356 studies for eligibility, first based on the suitability of their article titles and second based on their abstracts (Figure S1). The 240 eligible studies were subsequently screened manually upon reading the entire article based on the following inclusion criteria:

(1)           The study reported on univariate Ripley's K or L statistics or else it was possible to extract those from figures or maps.

(2)           The study had been carried out in a woody ecosystem or a rangeland.

(3)           The univariate Ripley’s K statistics described the distribution of individuals from a single plant species.

(4)           The authors named the plant species for which the univariate Ripley's K statistics had been described.

(5)           The landscape (for example a logging area) did not induce conspicuous point processes that could not be corrected within the analysis.

We manually processed the remaining 240 studies through reading the main text which reduced the final number of eligible studies to 69. A list of those data sources can be found in Appendix Three. From those studies we extracted the following moderators and we fitted them as predictors in subsequent models:

Mean annual temperature: continuous variable. When unreported, we extracted the variable based on coordinates from WorldClim (Fick & Hijmans, 2017).

Total annual precipitation: continuous variable. When unreported, we extracted the variable based on coordinates from WorldClim (Fick & Hijmans, 2017).

Latitude of the study location: continuous variable. When unreported, we extracted the information based on the closest location reported.

Longitude of the study location: continuous variable. When unreported, we extracted the information based on the closest location reported.

Site area: continuous variable. We extracted the site area from the studies and converted it into a unified unit, square meter.

Tree species: categorical variable.

Plant traits: we collected data on 7 traits: leaf area (i.e. the size of the leaves), seed mass, wood density, leaf mass per area, tree height, plant species biomass and stem specific density. We first gathered data on tree height, seed mass and leaf area from the subset of common species in TRY (Díaz et al., 2022). We subsequently searched for seed mass data the SID database (Royal Botanic Gardens Kew, 2023) and the ICRAF database for wood density data (Ketterings et al., 2001). In the cases we observed no records in those databases we checked the EOL database (http://eol.org.). For leaf area, leaf mass per area, tree height, plant species biomass and stem specific density, we extracted them from the EOL database (http://eol.org.). We opted with these traits to cover as many trait syndromes as possible but the main criterion which we used to decide on the traits was the feasibility of acquiring them for the plant species in our database.

Woody system age: categorical variable. We classified non woody habitats, plantations and systems that had recently experienced serious disturbances as “young” whereas natural forests or woody stands that had reached maturity as “old”.

Mycorrhiza type: categorical variable. We extracted mycorrhizal types for each species from Wang and Qiu (2006). In the cases that we could find no mycorrhizal classification information in the database at a species level we searched instead the database compiled by Delavaux et al. (2021) containing information at a genus level. We only extracted mycorrhizal classifications if these supported a single mycorrhizal type at a minimum probability of 85%. Otherwise, we left the plant species unclassified in relation to mycorrhiza.

Ripley's L effect size: continuous variable. We first calculated for all distances the ratio between the (1) difference between the Ripley's L statistic and the width of the 95% CI envelope divided by two and (2) the difference between the upper and lower points of the envelope divided by two. A large absolute value suggests a strong deviation from randomness whereas any value below 1 suggest a random process. We identified the location where the absolute value of this ratio was maximum.

Ripley's L statistic: continuous variable. We transformed Ripley´s K statistics (when they had not been transformed) into Ripley´s L statistics. We only used the value at the location where we observed the maximum in absolute value Ripley's K effect size.

Distance when Ripley's L peaked: continuous variable describing the distance at which we observed the maximum in absolute value Ripley´s L effect size.

Köppen climate zone: a categorical variable with 4 levels describing the main climatic zones based on the Köppen classification: A (tropical climates); B (arid climates); C (temperate climates); D (continental climates). We extracted those from the raster files published by Beck et al. (2018).

In the cases that we observed multiple values in databases (referring here mainly to plant trait values) per species, we used the median value. In the cases when we had to digitize plots to extract data, we did so with Plot Digitizer v2.6.8.