Achieving a predictive understanding of how ecological communities and processes respond to global environmental change is fundamental for biodiversity conservation. To this end, network or graph-theoretic analytical tools are frequently applied in community and landscape ecology to quantify interactions among species and movement of individuals across landscapes, respectively. Recent advances in multilayer networks (i.e. networks of networks), which explicitly account for multiple link types, permit the examination of feedbacks between local species interactions and regional spatial dynamics. However, such complex networks require immense amounts of data, which hinders their application to conservation decision-making. Here, we present a novel framework for assembling Predictive Multilayer Networks (PMNs) from readily available ecological and geospatial data. Then, we provide a proof-of-concept using a simulated plant-pollinator community, and compare land-cover scenarios to predict effects of forest loss and restoration on metacommunity-wide connectance.
We constructed an example network by first simulating a raster landscape consisting of four land cover classes. We then created hypothetical habitat preferences for pollinators and flowering plants to predict species occurrence across the landscape. Pollinator movement (spatial connectivity) was represented by intralayer links calculated from least-cost paths. Interlayer links (plant-pollinator interactions) were weighted by plant-pollinator co-occurrence probability and hypothetical pollinator foraging preferences.
Results/Conclusions
We were able to construct PMNs from 21 unique land-cover change scenarios representing a gradient from 100% deforestation to 100% forest restoration. Through calculation of the sum of all link weights relative to a null network (in which every species occurs universally across the landscape), we found that connectance ranged from 0.05 to 0.51, relative to a maximum of 1 in which all biologically possible interactions are realized. Connectance was highly correlated with forest cover (R2 = 0.996), indicating forest loss drove a metacommunity-wide decline in functional redundancy. Despite all species showing some relationship with forest cover, each species responded uniquely to land-cover change. Since PMNs incorporate species-specific habitat preferences, real-world networks would likely show more nuanced responses to land-cover change. We are developing additional multilayer adaptations of standard network metrics (robustness and centrality) to evaluate network stability and identify the most important species and landscape cells for maintaining network structure.
Despite frequent calls to integrate spatial and species interaction networks, collecting sufficient data is a major barrier. Instead, our approach leverages readily available data on species distributions, interaction preferences, and movement capabilities to construct PMNs, which can then be used to guide conservation decisions.