A risk-based approach for the enhanced understanding and management of denitrification in urban stormwater treatment wetlands

This thesis describes the development of a Bayesian network (BN) model for the enhanced understanding of denitrification in urban stormwater treatment wetlands. The motivation for the model was to maximise the potential for wetlands to control the loads of nitrate in runoff from urban centres to downstream bays and estuaries, which are at risk of eutrophication, e.g. Port Phillip Bay, Victoria. High rates of nitrate removal through denitrification are critical because of the relative abundance of N03" in stormwater and the ease by which it is transported in aquatic systems. Since the inception of urban stormwater treatment wetlands in the 1980s, continued research has led to gradual improvements in their design and management. However, significant uncertainty of important wetland treatment processes still exists due to the high degree of complexity. These uncertainties have impeded the ability to improve and predict wetland treatment performance, especially with respect to the removal of dissolved N species Given these existing knowledge gaps, a BN model was developed to assist with decision making in stormwater management and wetland design, with the aims of stimulating and predicting wetland denitrification efficiency. The model is a graphical representation of cause and effect relationships and incorporates all of the variables (nodes) influencing denitrification in urban stormwater wetlands. The relationships between nodes are depicted qualitatively by unidirectional linkages, and are quantified probabilistically, with information contained within the Conditional Probability Tables that underpin each node. The BN model developed represents current understanding of denitrification in urban stormwater treatment wetlands and can be readily modified in light of updated information. Sensitivity analysis of the developed BN model has allowed each of the system variables to be ranked in importance. Output from sensitivity analysis has identified that the variables hydraulic retention time, nitrate formation (conversion of other N species to N03"), organic carbon, and wetland design are the key drivers of denitrification in urban stormwater wetlands. Ensuring that these factors are optimised will increase the likelihood of achieving and maintaining high rates of wetland denitrification. The BN model was also used to test alternative options (scenario analysis) in wetland design and stormwater management. Scenario analysis helped explain the often poor treatment performance of some of Melbourne’s older style wetlands. Scenario analysis was also used to predict improvements in denitrification efficiency that might be achieved if best practice technologies and innovative wetland design features were implemented.