Modeling ecosystem-wide responses to environmental stressors: A multi-trophic hierarchical Bayesian network approach

Effective management of aquatic ecosystems requires models that capture complex interactions among multiple trophic levels and environmental stressors. However, many existing data-driven ecological models are limited in scope, often focusing on a narrow subset of taxa or employing simplified representations of community structure, making them inadequate for ecosystem-wide assessments. Hierarchical Bayesian networks (HBNs) address these limitations by incorporating latent variables that capture correlated ecological relationships, such as trophic interactions and shared environmental responses. This structure reduces the dense connectivity often seen in conventional Bayesian networks, allowing for more concise and interpretable representations of complex interactions. This study developed an HBN to predict the responses of aquatic communities-including phytoplankton, zooplankton, benthic macroinvertebrates, and fish-to a wide range of environmental drivers, represented by meteorological, water quality, hydrological, and riverbed variables, across the four major river basins in South Korea. The network structure was informed by KF-METAWEB, a comprehensive trophic interaction database for Korean freshwater ecosystems, ensuring that the modeled relationships reflect ecologically validated interactions across trophic levels. The hierarchical design of the HBN enabled the model to capture cascading effects across biological communities and environmental gradients. Compared to conventional Bayesian networks-both knowledge-based (mean accuracy = 0.733; AUC = 0.648) and data-driven (mean accuracy = 0.745; AUC = 0.681)-the HBN achieved superior predictive performance (mean accuracy = 0.787; AUC = 0.705). Sensitivity and scenario analyses identified water quality parameters and substrate composition as critical factors of the community structure of benthic macroinvertebrates and fish. This study presents the first application of an HBN to predict multi-trophic dynamics in riverine ecosystems and demonstrates its potential as a transparent, data-informed tool for ecological assessment and adaptive river basin management.