Seed dispersal distance classes and dispersal modes for the European flora

Motivation: Although dispersal ability is one of the key features determining the spatial dynamics of plant populations and the structure of plant communities, it is also one of the traits for which we still lack data for most species. We compiled a compre - hensive dataset of seed dispersal distance classes and predominant dispersal modes for most European vascular plants. Our seed dispersal dataset can be used in functional biogeography, dynamic vegetation modelling and ecological studies at local to continental scales.

However, the morphological adaptations and sizes of seeds and other diaspores do not fully describe the dispersal potential of a given species, which limits the use of such data in ecological studies (e.g. Beckman et al., 2018;Bullock et al., 2017;Nogales et al., 2007;Tamme et al., 2014;Thompson et al., 2010Thompson et al., , 2011. Few studies have measured, estimated or mathematically modelled dispersal distances, which seem to be a more reliable indicator of dispersal ability. These were reviewed by Vittoz and Engler (2007) and Thompson et al. (2011). Unfortunately, such data are only available for a limited number of European plant species. Recently, Tamme et al. (2014) developed a statistical method to estimate dispersal distances based on several plant traits (dispersal mode, growth form, seed mass, seed release height and terminal velocity) using data on 576 plant species. Although estimating dispersal distances from plant traits is a promising approach, there are other important factors affecting dispersal distances, namely humanassisted dispersal (anthropochory) and species habitat preferences (Vittoz & Engler, 2007). Plant dispersal patterns have been significantly influenced by humans, who intentionally or unintentionally transport species at long distances, often outside their natural distribution range. Most databases only consider dispersal mode estimated from species morphological characteristics, such as dispersal by wind, water or animals (e.g. Royal Botanic Gardens Kew Seed Information Database, 2021; SID). Indeed, natural dispersal strategy is crucial for colonizing new areas at the landscape scale. However, rare events of human-assisted dispersal should also be considered.
Information on the dispersal by humans is directly available in some databases (e.g. LEDA, Kleyer et al., 2008), or it can be inferred for species with native ranges on other continents (e.g. GloNAF;van Kleunen et al., 2019). Another group of species that are commonly dispersed by humans are weeds of arable fields or other man-made habitats (compare the characteristic species combinations in the EUNIS Habitat Classification; Chytrý et al., 2020). Dispersal distance is also influenced by the interaction between the environment and plant traits (Vittoz & Engler, 2007 et al., 2020), and data on habitat preferences of individual plants Mucina et al., 2016) make it possible to combine this information to estimate the actual plant dispersal distances. Vittoz and Engler (2007) proposed to estimate plant dispersal distances using a clearly defined and methodologically simple approach. They assigned plant species to seven dispersal distance classes based on the morphology of dispersal units (diaspores), other species traits and habitat preferences. The original dataset of Vittoz and Engler (2007)  Here, we present a new dataset of seed dispersal distance classes and predominant dispersal types for 10,327 widely distributed European vascular plants. We use the approach of Vittoz and Engler (2007) with a slight modification of some classes. To evaluate our approach, we compared our seed dispersal distance classes with dispersal distances estimated using the trait-based method proposed by Tamme et al. (2014).

| Species selection
We prepared a list of frequent European vascular plant species based on species occurrence frequency in vegetation plots in the European Vegetation Archive (EVA, Chytrý et al., 2016), a centralized database of European vegetation plots. We standardized species taxonomy and nomenclature according to Euro+Med (http://europ lusmed.org).
Subspecies, varieties and forms were merged at the species level.
Species in taxonomically difficult groups were merged into aggregates following the EUNIS-ESy expert system for the EUNIS Habitat Classification . Hereafter, we use the term 'species' to refer to both species and aggregates.
As we aimed to include all frequent species and species that could attain high cover in plant communities, we selected those species that occurred in >1% of plots, had a cover of >25% in >0.1% of plots or had the highest cover of all vascular plant species in >0.1% of plots. This selection included 5569 frequent or locally dominant vascular plant species of the European flora. Hereafter, we refer to them as priority species.
For seed mass and plant height, we checked the frequency distribution across all species and excluded outlying, improbable values.
We then calculated the median value of each trait for each species across all databases.
On this basis, we estimated species affinity for forest or open habitats.
Another important aspect was whether or not the species is native to Europe, which indicates human-assisted long-distance dispersal. For the assignment of the alien or native status, we used several additional sources, particularly the list of European alien species in the FloraVeg.EU database (https://flora veg.eu).

| Indicators of human-assisted dispersal
We have considered several groups of alien species: (1) Species that were introduced to Europe unintentionally or as a curiosity rather than for economic profit and that are now freely spreading and have well-established populations in different habitats.
(2) Crops and weeds, that is, plants introduced with agriculture and restricted to anthropogenic habitats such as arable land and human settlements. They are dispersed mainly due to human activities, especially by speirochory (unintentional dispersal with seeds of cultivated plants) or ethelochory (intentional introduction by seeds or seedlings into new areas). To determine the most important weeds that are unintentionally dispersed by humans, we used lists of weeds for EUNIS arable habitats .
(3) Planted trees and shrubs. To distinguish frequently cultivated species from naturally occurring species, we used the list of non-native trees planted in Europe Wagner et al., 2017). We did not include cultivated species that do not spread spontaneously.
Some native species are also strongly supported by humans and are repeatedly planted or sown in the landscape, especially in the case of seed mixtures used for establishing new grasslands or increasing grassland productivity. We have considered these species to be both naturally dispersed and human-dispersed.

| Delimitation of dispersal distance classes
We adopted the delimitation of dispersal distance classes from Vittoz and Engler (2007). They compiled all available measurements of seed dispersal distances of Central European species and classified them into (1) mean, mode and median values, (2) maximum values (99th percentiles of distribution kernels) and (3) values for long-distance dispersal events, that is, extreme distances reached by only a very small fraction of seeds. They did not consider the last category in the delimitation of dispersal classes. They assigned each dispersal distance to a dispersal mode. Dispersal distance classes were then defined by grouping dispersal modes with similar dispersal distances and similar diaspore adaptations and traits. Finally, Vittoz and Engler (2007) estimated, for each class, the upper limits of the distances within which 50% and 99% of seeds would disperse. We adopted a similar delimitation of dispersal distance classes (Table 1).

| Assignment to dispersal distance classes
We summarized the plant characteristics obtained from various sources in a table (Supplement S1) that includes plant height, life form, predominant dispersal mode, seed mass and typical habitat.
We also included data on plant geographical origin and the use by humans to distinguish a group of plants intentionally dispersed by humans (Table 1). Plants were classified into seven dispersal distance classes based on the above-mentioned characteristics, as illustrated included human-dispersed species, although the natural dispersal strategy of these species was different and often less efficient than long-distance dispersal by humans. For example, Robinia pseudoacacia is a tree with dry fruits suitable for wind dispersal, which corresponds to dispersal distance class 4 in the scheme of Vittoz and Engler (2007). However, this species was introduced to Europe by humans and is frequently planted, which would lead to a classification in class 7. Nowadays, long-distance dispersal of this species by humans rarely happens, but once introduced to a new area, it can spread locally due to wind dispersal (Vítková et al., 2017).
For animal dispersal, we did not distinguish between dispersal by small and large animals as proposed by Vittoz and Engler (2007) because such information is missing for most plant species. However, we did distinguish between dispersal by ants and vertebrates. We also modified the delimitation of species with dusty seeds. Vittoz After standardizing nomenclature, the dataset contained 13,333 species. These species had at least partial information necessary for their assignment to the dispersal distance classes. We assigned the priority species with incomplete or no data to dispersal distance classes based on our expert judgement combined with the information available for other species in the same genus. For example, dispersal mode information was available for several species of the genus Abies but not for Abies borisii-regis and A. sibirica. We decided to classify these species in the same dispersal distance class as the other Abies species. However, we did this only for the genera with uniform dispersal traits, similar height and similar habitat preferences. We did not do this for the genera with diverse diaspores (e.g. Anemone, Medicago and Ranunculus) or for anemochorous species without information on habitat preferences or height. Finally, we were able to classify 10,327 species and aggregates of European vascular plants.

| Analyses
To evaluate our classification of species into dispersal classes, we related dispersal classes to estimated maximum dispersal We ran all models with the 'dispeRsal' function (Tamme et al., 2014), which estimates dispersal distances using simple linear regressions and mixed-effect models that account for taxonomic dependencies by using species taxonomy (either as family or order) as a random variable. We determined species families using the 'TPL' function of the R package Taxonstand (Cayuela et al., 2017). Finally, to estimate the dispersal distance for each species, we averaged dispersal distances for each model and structure type (i.e. simple regression, species' family or order as a random component).
To test the relationship between mean estimated dispersal distances and our dispersal distance classes, we fitted linear models with dispersal distances as dependent variables and dispersal distance classes as categorical predictors and visually inspected the distribution of dispersal distances within each class. Such analyses were conducted for comparisons within dispersal distance classes 1-6, considering natural dispersal modes only and within all seven classes.

| RE SULTS
Our dataset (Supplement S1) contains information on the dispersal distance class and the predominant dispersal mode for 10,327 species.
The dispersal distance classes are ordered from class 1, which contains species with the shortest dispersal distances, to class 7, which contains species with the longest dispersal distances (Figure 3 Tree, shrub Herb, dwarf shrub Seeds with wings or hem Seeds with trichomes animals, which sometimes hide them as stock), endozoochorous (i.e. dispersal in animal gastrointestinal tract) or epizoochorous (i.e. dispersal on animal fur). Finally, class 7 contains 1288 human-dispersed (anthropochorous) species. The species of the last class are also classified into one of the previous six classes based on their natural dispersal mode (see Supplement S1).
Our dispersal classes were positively related to dispersal distances estimated using the 'dispeRsal' function. When included as categorical predictors in a linear model, dispersal classes explained a significant portion of the variation in dispersal distances (R 2 = 0.75 for six naturally dispersed classes and R 2 = 0.70 for all seven classes; Figure 4).

| DISCUSS ION
We estimated dispersal distances for 10,327 European plant species using dispersal distance classes following the method proposed by Vittoz and Engler (2007). This semi-quantitative trait provides better information for ecological studies than categorical proxies of dispersal distance, such as dispersal modes and dispersal vectors.
Our dataset is based on the definition of the predominant dispersal modes, which are also given for all 10,327 species. We recognize that plants usually combine multiple dispersal vectors and that the efficiency of these vectors varies in space and time (Beckman et al., 2020;Bullock et al., 2017;Sádlo et al., 2018;Thompson et al., 2010). We have attempted to overcome this uncertainty by dividing species adapted to human dispersal into two groups reflecting less effective but more frequent natural dispersal and more effective but rare human-assisted dispersal. Consequently, researchers can decide whether to use only the first six dispersal classes (corresponding to natural dispersal) or all seven dispersal classes (also including rare events of human-assisted dispersal). Six classes are appropriate for landscape-scale studies, while seven classes may be more suitable for studies at the continental scale and over longer time periods. While data coverage is relatively high for Central and Western Europe, less information is generally available for plants from the Balkan Peninsula and Eastern Europe and for rare species. However, the dataset can be extended with new data using our decision tree.
We have shown that our classification is consistent with predictions of seed dispersal distances using the 'dispeRsal' function (Tamme et al., 2014). Relatively low dispersal distances estimated using this function for classes 5 and 7 (see Figure 4) are due to considering more factors in our approach. Tamme et al. (2014) did not consider the habitat preferences of anemochorous species, leading to F I G U R E 3 Numbers of species included in dispersal distance classes classified into (a) six classes, excluding anthropochorous dispersal and (b) seven classes, also including dispersal distance class 7 with anthropochorous dispersal. an underestimate of dispersal distances for class 5. They also did not account for human dispersal, which is partly based on human preferences for some species rather than on the biological traits of species, resulting in an underestimate of dispersal distances for class 7.
The seed dispersal dataset is important for several types of macroecological studies. Dispersal distance has significant implications for the distribution and coexistence of plant species (Beckman et al., 2020;Levin et al., 2003). For example, Poschlod et al. (1998) have shown the importance of dispersal ability for the diversity of calcareous grasslands, Pärtel and Zobel (2007) (Beckman et al., 2020) or estimating the potential spread of invasive species (Chapman et al., 2017).
Our dataset contains all the necessary information to assign species to dispersal distance classes, including relevant traits, predominant dispersal modes, species biogeographical origin (native vs. alien) and habitat preferences. Besides using the dispersal distance classes, it is also possible to work with partial data. For example, in landscapescale studies focused on a single habitat type, one might want to consider only dispersal traits without habitat preferences.
The seed dispersal dataset for the European flora is a resource for addressing a variety of ecological and evolutionary questions.
We hope this dataset and approach will stimulate research in ecology and vegetation modelling related to plant dispersal potential on a broad continental scale. It can be used to create more realistic models of plant species range shifts and vegetation changes in response to global environmental changes.

AUTH O R CO NTR I B UTI O N S
W.T. and M.C. conceived the idea. Z.L. classified the species. I.A.
prepared the trait data and carried out nomenclature unification.
G.M. conducted the statistical analyses. J.R., S.A., J.V.E. and P.V. provided background databases. Z.L. wrote the manuscript. All authors discussed and critically commented on the manuscript.

ACK N O WLE D G E M ENTS
We thank Ivana Mileová for drawing Figure 1.