Fish fanning in coastal marine environments results in the deposition of waste products on the seafloor and can cause changes in sediment chemistry and community structure. Rapid growth in the industry and an increased awareness and sensitivity to the environmental impacts over the last decade have resulted in more stringent compliance requirements for aquaculture producers. However, the selection of monitoring parameters is surrounded by uncertainty, which impedes the development of cost-effective monitoring procedures. Studies examining benthic impacts of fish farms have come to different conclusions concerning the severity of impacts. Previous attempts to quantitatively review these studies have had only limited success due to the uncertainty in the data originating from discrepancies in sampling and analytical protocols as well as different temporal and spatial scales covered in the studies. The main objective of this study was to review publications on fish farm benthic impacts and to develop a Bayesian network (BN) for the quantitative assessment of the relationships between impact parameters and site and farm characteristics. A BN was constructed based on parameters relationships obtained from 64 studies. It showed that benthic impact was a function of fish density, farm volume, food conversion ratio, water depth, current strength and sediment mud content. To examine the sensitivity of benthic impact variables to changes in these parameters, the BN was used to calculate three scenarios representing low, moderate and high impact. Porewater sulphide, acid volatile sulphide (AVS-S), water content, redox potential, sediment oxygen consumption, sediment NH4 release and macrofauna diversity showed the most certain and sensitive responses in these scenarios but methodological limitations have to be taken into consideration before characterising them as reliable monitoring parameters for a specific application. An examination of spatial trends in benthic impact parameters suggested that fish farm impacts were confined to a radius of about 40 to 70 m around the farms studied. The inability to satisfactorily model parameters as a function of distance from farms demonstrated the complexity of their spatial distribution and highlighted the need to improve our understanding of farm footprints to avoid detrimental environmental effects as a consequence of culture intensification. The BN approach successfully identified reliable monitoring parameters based on reviewed impact studies and has potential to support cost-effective monitoring. For use at specific farms, site-specific data should be integrated into the BN and additional variables could easily be added. BNs enable the synthesis of scientific and industry research data, practical experience and stakeholders' perspectives, which are important in the development of monitoring guidelines that meet the needs of all parties. (c) 2007 Elsevier B.V. All rights reserved.
