Trophic shifts have been observed to cause significant changes in the composition, structure and functioning in aquatic ecosystems. Complex spatial and temporal interactions with biotic and abiotic processes make quantifying trophic thresholds difficult, with traditional statistical models unable to account for numerous simultaneously occurring confounding biophysical and biogeochemical factors across multiple scales. The effective quantification of positive and negative feedbacks leading to trophic shifts is not well represented by traditional models, whereas Bayesian models can take into account the integration of subsystems within the whole system using both quantitative and qualitative data. Bayesian models are naturally suited to modelling and predicting outputs in complex ecosystems through the incorporation of a priori information, and the quantification of uncertainties and high spatial and temporal resolution. Bayesian belief networks (BBNs) were developed to derive nutrient water quality guidelines and identify trophic thresholds in unregulated coastal rivers in south-eastern Australia. Spatial and temporal heterogeneity of water column nutrients, and the identification of limiting nutrients and their thresholds, were quantified using diffusing substrata (NDS) in freshwater and estuarine reaches of five unregulated river catchments. The nutrient limitation data collected from seasonal nutrient diffusing strata experiments were used to link similar catchments and define the spatial and temporal scope for the models. The Bayesian models developed predict shifts in trophic state in freshwater and estuarine reaches of coastal rivers providing managers with spatially and temporally explicit low-risk thresholds for nutrient concentrations and ratios to assist in the management of trophic shifts.