A Bayesian network model linking nutrient management actions in the Tully catchment (northern Queensland) with Great Barrier Reef condition Will Shenton A , Barry T. Hart A,C and Jon Brodie B A Water Studies Centre, Monash University, Melbourne, Vic. 3800, Australia. B Australian Centre for Tropical Freshwater Research, James Cook University, Townsville, Qld 4811, Australia. C Corresponding author. Email: barry.hart@waterscience.com.au Abstract. Correlating catchment management actions with improvements in the ecological condition of downstream coastal regions is challenging. We describe a Bayesian network (BN) model that predicts the effects of nitrogen-fertiliser management strategies in the Tully River catchment (northern Queensland) on the condition of inshore reefs of the Great Barrier Reef (GBR). The model consists of three linked submodels that relate sugarcane nitrogen management with runoff into the Tully River and nitrate concentration in the GBR lagoon, predicts phytoplankton biomass in the GBR lagoon from the nitrate inputs, and links the phytoplankton biomass with three marine influences to predict the probability of the reefs being dominated by coral (good) or macro-algae (bad). Four scenarios were modelled – current and the ‘six easy steps’ nitrogen management, and active and depleted algal grazing (herbivory) of the reef. The model predicts an increased probability of the reef being coral-dominated with current fertiliser practice and with active reef herbivory, with increased algal-dominance if reef herbivory is decreased. Introduction of a better nitrogen-fertiliser management with active herbivory resulted in an increased probability of coral dominance. This comparative-scenario analysis highlights the importance of both agricultural nutrient management practices and marine processes in predicting reef condition. Additional keywords: decision support, fertiliser, nitrogen. Introduction The Great Barrier Reef (GBR) in north-eastern Australia is an iconic ecosystem of global significance that is threatened by climate change, the impacts of fisheries, and contaminants from the catchment (Hughes et al. 2007). The main water-quality issues are elevated concentrations of suspended sediment, nutrients and pesticides from diffuse agricultural sources, including rangeland grazing and intensive agricultural cropping systems (Baker 2003; Brodie et al. 2008). The Reef Water Quality Partnership was a collaborative structure formed in response to uncertainty in the science supporting the reef plan and regional water-quality management plans (including water-quality improvement plans (WQIPs)), and the need for institutional collaboration across Queensland and Australian Governments and the regional natural resource management (NRM) bodies of the GBR to address these needs (Reef Plan 2003). The Reef Water Quality Partnership provided a linkage between the GBR and catchment-scale policy and planning processes, and focussed on improved target-setting, monitoring and reporting at both these scales. A good example of these WQIPs is provided by that devel- oped by the Mackay–Whitsunday Natural Resources Manage- ment Group (Drewry et al. 2008) and the Tully WQIP (Kroon 2008). These WQIPs are all intended to be adaptive manage- ment plans, with the capacity to change with time as more information becomes available. Eberhard et al. (2009) devel- oped a protocol to guide the practical application of an adaptive approach in the coastal catchments of the Great Barrier Reef, and tested this protocol using the Tully WQIP. A major problem faced by each of these WQIPs is the difficulty in relating catchment management actions with poten- tial improvement in the condition of the GBR. Currently, several models are available to estimate the loads of suspended sedi- ment and nutrients exported from each of the catchments (Drewry et al. 2007; Armour et al. 2009). Models have been developed to predict what happens to these contaminants once they enter the GBR lagoon, and their effects on the biota (e.g. coral reefs, seagrasses, phytoplankton, fish) (e.g. Wolanski et al. 2004); however, these models do not include a robust link to catchment land-use practices. Within WQIPs, targets for land- use management linked to marine ecosystem targets have been set using a variety of models strung together from the paddock to the reef (Kroon 2008, 2009). Problems can arise when the outputs of one model (e.g. at a paddock scale) are used as an input into the next model downstream (e.g. a catchment model) (Brodie et al. 2009). CSIRO PUBLISHING www.publish.csiro.au/journals/mfr Marine and Freshwater Research, 2010, 61, 587–595 Ó CSIRO 2010 10.1071/MF09093 1323-1650/10/050587