Key concepts and a world-wide look at plant recruitment networks

Abstract

Plant-plant interactions are major determinants of the dynamics of terrestrial ecosystems. There is a long tradition in the study of these interactions, their mechanisms and their consequences using experimental, observational and theoretical approaches. Empirical studies overwhelmingly focus at the level of species pairs or small sets of species. Although empirical data on these interactions at the community level are scarce, such studies have gained pace in the last decade. Studying plant-plant interactions at the community level requires knowledge of which species interact with which others, so an ecological networks approach must be incorporated into the basic toolbox of plant community ecology. The concept of recruitment networks (RNs) provides an integrative framework and new insights for many topics in the field of plant community ecology. RNs synthesise the set of canopy-recruit interactions in a local plant assemblage. Canopy-recruit interactions describe which ("canopy") species allow the recruitment of other species in their vicinity and how. Here we critically review basic concepts of ecological network theory as they apply to RNs. We use RecruitNet, a recently published worldwide data set of canopy-recruit interactions, to describe RN patterns emerging at the interaction, species, and community levels, and relate them to different abiotic gradients. Our results show that RNs can be sampled with high accuracy. The studies included in RecruitNet show a very high mean network completeness (95%), indicating that undetected canopy-recruit pairs must be few and occur very infrequently. Across 351,064 canopy-recruit pairs analysed, the effect of the interaction on recruitment was neutral in an average of 69% of the interactions per community, but the remaining interactions were positive (i.e. facilitative) five times more often than negative (i.e. competitive), and positive interactions had twice the strength of negative ones. Moreover, the frequency and strength of facilitation increases along a climatic aridity gradient worldwide, so the demography of plant communities is increasingly strongly dependent on facilitation as aridity increases. At network level, species can be ascribed to four functional types depending on their position in the network: core, satellite, strict transients and disturbance-dependent transients. This functional structure can allow a rough estimation of which species are more likely to persist. In RecruitNet communities, this functional structure most often departs from random null model expectation and could allow on average the persistence of 77% of the species in a local community. The functional structure of RNs also varies along the aridity gradient, but differently in shrubland than in forest communities. This variation suggests an increase in the probability of species persistence with aridity in forests, while such probability remains roughly constant along the gradient in shrublands. The different functional structure of RNs between forests and shrublands could contribute to explaining their co-occurrence as alternative stable states of the vegetation under the same climatic conditions. This review is not exhaustive of all the topics that can be addressed using the framework of RNs, but instead aims to present some of the interesting insights that it can bring to the field of plant community ecology.

Keywords: canopy service; ecological networks; facilitation; interaction strength; plant–plant interactions; recruitment niche; replacement networks; sapling bank; stress gradient hypothesis; strongly connected components.

Fig. 1

Example of basic analysis of the recruitment network of a mountain semi‐desert community (Escara, Bolivia). The recruitment matrix contains the frequency of canopy–recruit interactions, with the canopy species represented in columns and the recruit species in rows. The first row (grey) and column (yellow) correspond to the node Open that represents recruitment far from other plants. The main diagonal (blue) contains the frequency of intraspecific recruitment. The graph representation of the recruitment has one node for each species (plus Open) and the arrows point from the canopy to the recruit species (species names are indicated with the first three letters of the genus and species). Node colours indicate their functional roles (as explained in Section V.3): yellow for the Open node, green for core species, blue for satellite species and orange for transient species. Values resulting from aggregating cell values along rows or columns inform, respectively, on properties of the “recruitment bank” and “canopy service” of each species. For example, the counts of non‐zero cells along rows or columns indicate the in‐ and out‐degree of each species, forming the width of the recruitment niche and canopy service. Similarly, the sums along rows and columns provide the species abundance in, and the canopy contribution to, the recruitment bank. Finally, Shannon diversity (H′) and the effective number of partners (e^H′) can be measured from row and column entries. Data from RecruitNet (Verdú et al., 2023).

Fig. 2

Relationships between aridity (measured as 1 – Aridity Index) and the mean strength of positive, negative and neutral canopy–recruit interactions measured as the Neighbour suitability index (Ns) in each local community. Background shading indicates the climate of the community: humid (<0.35), dry subhumid (0.35–0.50), semiarid (0.50–0.80), arid (0.80–0.95) and desert (>0.95). Aridity Index values were obtained from Zomer et al. (2022).

Fig. 3

Cumulative frequency of connectance (C) in recruitment networks (RNs) of the RecruitNet database. The dashed lines indicate the 22%, 50% and 90% percentiles. Observed C values range from a minimum of 0.014 to a maximum of 0.583.

Fig. 4

Fit of the in‐ and out‐degree distributions of the largest network in the RecruitNet database (Wanang, from Papua New Guinea, with 557 species) to the exponential, power law and truncated power law distributions (the exponential distribution could not be fitted to the out‐degrees). Statistical tests of fit are provided in Appendix S4. According to Akaike Information Criteria, the best fit was achieved with the truncated power law, but note that, nevertheless, this distribution overestimates the frequency of species with very large degrees.

Fig. 5

Functional topology of recruitment networks (RNs). Nodes represent species (except the Open node) and arrows point from the canopy to the recruit species. The toy example (inset panel) illustrates the types of functional roles of the nodes defined by the relationships between the strongly connected components (SCCs). In this toy example there are nine SCCs; only one of them (the core) is non‐trivial since it is formed by four nodes, and the rest are trivial (i.e. formed by a single node each). There are two trivial SCCs corresponding to disturbance‐dependent transients (they only receive arrows from Open node), two trivial SCCs corresponding to strict transient species (they do not receive any arrows) and three trivial SCCs corresponding to satellite species (they do not send any arrows). The examples of real RNs correspond to: SCBI which represents the typical RN of forests in humid regions (eastern mixed deciduous forest for North America), Cruz de Chimba which represents the RN of dryland forests (mediterranean pine‐oak forest), Negro II representing a shrubland from humid regions (Andean alpine dwarf shrubland) and Uquía, which is a shrubland from drylands (Andean mountain desert).

Fig. 6

Variation of the components of the functional structure with habitat (forest versus shrubland) and aridity. Lines are univariate fits to the data, with shadings indicating their 95% confidence interval. Functional roles are indicated as Core, Sat (satellite), ddTrans (disturbance‐dependent transients) and strTrans (strictly transient). Functional roles whose relative frequency varied significantly with aridity are indicated with an asterisk.