MetaWeave: a reproducible framework for the spatial inference of ecological networks

A regional species-interaction network, or metaweb, describes interactions that potentially occur among species within a region. MetaWeave is an R package that combines regional metawebs, local species assemblages, and alternative interaction models to reconstruct local networks. Its main feature is to separate the effects of interaction assumptions from changes in community composition. We illustrate the package with three applications: a synthetic benchmark comparing interaction models against a known generating process; spatial downscaling of a Neotropical palm–mammal metaweb; and a reanalysis of six Ecuador plant–frugivore networks. In the Ecuador example, an externally fitted regional stochastic block model is applied to both observed and richness-matched alternative assemblages. Holding the observed assemblages fixed, the block model increased simulated nestedness relative to neutral link allocation at the two lowest sites, while the composition contrasts were less consistent across sites. These comparisons are conditional on species recorded in the interaction data and use a model fitted to the pooled networks. MetaWeave 1.0.0 is available on CRAN. By retaining local interaction matrices and allowing assemblages and interaction rules to vary independently, MetaWeave supports explicit comparisons of ecological network hypotheses across space.