ABSTRACT: Benthic gross primary productivity (GPP), net primary production (NP), and respiration (R) were measured seasonally in each of 12 major benthic habitats in 3 southeast Australian estuaries, along with a suite of biological, physical, and chemical parameters to construct a benthic carbon budget and to elucidate controls over benthic metabolism. We also tested the performance of an artificial neural network (ANN) model in predicting benthic metabolism from the suite of measured parameters, and compared model performance to traditional stepwise regression methods. Carbon budgets indicated that macrophyte communities made the greatest contribution to whole system benthic metabolism (51 to 79% of gross productivity and 38 to 74% of respiration), and net benthic metabolism of the 3 estuaries ranged from −25 to ~90 g C m⁻² yr⁻¹. Metabolism in non-macrophyte communities was tightly coupled to light, temperature, organic matter supply, and benthic algal biomass, and metabolism in macrophyte communities was coupled predominantly to temperature and light. ANN outperformed stepwise regression for all benthic metabolic parameters in both macrophyte and non-macrophyte habitats. Root mean square errors of ANN were up to 3-fold lower than stepwise regression models, indicating the potential use of ANN in modeling ecosystem-scale metabolism. We used ANN models to predict system-wide changes in benthic net production associated with an increase in temperature of 1 to 2°C. Model results indicate that system-wide net production increased with temperature, indicating that carbon burial in, and/or export from estuaries may increase as a result of increasing water temperature associated with climate change.