The detection of causal interactions between variables is of great importance when inferring complex networks but also remains challenging due to the high-dimensionality and nonlinearity of observational time series with a limited sample size. Convergent cross mapping (CCM) based on nonlinear state space reconstruction, as a network inference model, made substantial progress by measuring how well historical values of one variable can reliably estimate states of other variables. Here, we investigate the ability of an Optimal Information Flow (OIF) model to infer bidirectional causality and validate the model on a mathematically simulated predator-prey model, a real-world sardine-anchovy-temperature system and a multispecies fish ecosystem by comparing that to CCM. The validation work demonstrates that the proposed OIF model performs better than CCM since it provides a larger gradient of inferred interactions, higher point-value accuracy with smaller fluctuations and no requirement of convergence. Besides, OIF offers broad ecological information by extracting predictive causal networks of complex ecosystems from time-series data in the space-time continuum. Therefore, OIF is a robust model in estimating predictive causality (also in terms of computational complexity) due to the explicit consideration of synchronization, divergence and diversity of events that define model sensitivity, uncertainty and complexity. The accurate inference of species interactions allows to predict biodiversity changes as a function of climate and other anthropogenic changes. This has practical implications for defining optimal ecosystem management such as fish stock prioritization and marine protected area delineation based on the derived collective multispecies assembly.
