When indicator species identified from a network structure predict biodiversity

Comprehensive biodiversity surveys are difficult to sustain across large and environmentally heterogeneous regions, creating a need for approaches that can recover community composition from a limited number of observed species. We ask whether a small set of indicator species selected from species co-occurrence networks can be used to reconstruct the occurrence of the remaining assemblage. We develop an integrated framework combining probabilistic co-occurrence networks, network-based indicator selection, and species distribution modeling, and evaluate it across contrasting bird and beetle assemblages prediction. Network-guided indicator selection supported species-level reconstruction across both taxa and was competitive with, and in several assemblages superior to, conventional selection strategies. Meaningful predictive performance was also retained for several low-prevalence species. However, indicator-based models generally matched or underperformed models based directly on climate, suggesting that part of the information carried by co-occurrence networks reflects environmental gradients already captured by measured climatic variables. The clearest gains beyond climate occurred in an assemblage characterized by a larger number of significant associations, broader spatial extent, and greater ecological heterogeneity, suggesting that the value of network-guided indicators may increase when richer association structure is available. Together, these results show both the potential and the limits of co-occurrence networks for identifying compact sets of species that can support species-level biodiversity reconstruction.