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

Abstract

Increased nitrogen loading of estuaries and coastal systems can lead to eutrophication and toxic effects. Considering the extent of urban development in coastal catchments, concerns about high concentrations of N, and in particular nitrate-N, in urban stormwater runoff are well justified. Denitrification represents the major pathway of NO3- removal from aquatic systems. Wetlands provide an ideal environment for denitrification from surface runoff and are widely used to improve the quality of stormwater. However, the treatment capacity of wetlands remains largely unknown, and wetland treatment of stormwater is still considered to be an emerging technology. Climatic and hydrological conditions and the geographical and biogeochemical variations between wetlands and associated catchments make it difficult to apply generic concepts of design and management. Wetlands are highly complex systems and are characterised by extreme variability, which leads to unpredictable outcomes and makes it difficult to translate results from one wetland to another. In recent years, there has been a large research effort aimed at better understanding denitrification, at both the microbiological scale and the larger catchment or wetland scale. We are applying risk assessment techniques to make practical use of the existing and cumulative knowledge to further our understanding of ways to stimulate and maintain high rates of denitrification in urban wetlands. The application will assist with improving stormwater and urban wetland management. The risk assessment methodology is based on determining the level of risk to denitrification posed by stressors within urban stormwater and wetland systems through consideration of the multiple factors in operation and their various interactions. A Bayesian Network (BN) is being used as the modelling environment. This paper describes the iterative development of the BN, and provides examples of sources of data and information and the methods by which they have been incorporated into the model. Information has been obtained from multiple sources and at various scales, including expert literature, monitoring data, and domain experts. It will be demonstrated that one of the key advantages of using BNs as the modelling framework is that it readily allows information from a range of scales and sources to be incorporated. During ongoing model development, sensitivity analysis (SA) has been used as an important model validation and assessment tool and has allowed structural and probabilistic errors to be identified and corrected. Through ranking the relative importance of network variables on the output, SA has enabled the identification of the key drivers of the system. The modelled variables found to be exerting the greatest influence over variations in the output (the removal of stormwater NO3- by constructed urban treatment wetlands or "Denitrification efficiency") are hydraulic retention time (time taken for the input stream to pass through the wetland), the input NO3- load, available organic carbon, and toxic inhibition by contaminants sequestered within wetland substrates (eg heavy metals). Identifying the primary driving factors within a system can assist with the prioritisation of management actions and research resources, thus fulfilling the intended use of the BN as a Decision Support Tool (DST). Uncertainty in many of the processes being modelled is high but the BN represents our current knowledge of a highly complex system within an accessible framework, and where uncertainty is demonstrated explicitly. With continued research under an adaptive management framework, additional information will become available and the model can be further developed and updated, thus further satisfying the fundamental requirements of risk assessment.