Inferring ecosystem networks as information flows

Abstract The detection of causal interactions is great importance when inferring complex ecosystem functional and structural networks for basic applied research. Convergent cross mapping (CCM) based on nonlinear state-space reconstruction made substantial progress about network inference by measuring how well historical values one variable can reliably estimate states other variables. Here we investigate the ability a developed optimal information flow (OIF) model to infer bidirectional causality compare that CCM. Results from synthetic datasets generated simple predator-prey model, data real-world sardine-anchovy-temperature system multispecies fish highlight proposed OIF performs better than CCM predict population community patterns. Specifically, provides larger gradient inferred interactions, higher point-value accuracy smaller fluctuations $$\alpha$$ ? -diversity including their characteristic time delays. We propose an threshold maximize in predicting effective -diversity, defined as count model-inferred interacting species. Overall outperforms all models assessing predictive (also terms computational complexity) due explicit consideration synchronization, divergence diversity events define sensitivity, uncertainty complexity. Thus, offers broad ecological extracting ecosystems time-series space-time continuum. accurate species at any biological scale organization highly valuable because it allows biodiversity changes, instance function climate anthropogenic stressors. This has practical implications defining management design, such stock prioritization delineation marine protected areas derived collective assembly. be used evaluation design where should considered non-linear predictability diverse populations or communities.