Bayesian Network and System Thinking modelling to manage water quality related health risks from extreme events

Abstract

The occurrence of extreme events can challenge the capacity of water utilities to deliver potable water of sufficient quality with respect to minimising health risks to consumers. As a consequence, proactive risk-assessment and decision support tools are necessary to assist in managing and mitigating such critical events effectively. However, the utility of these tools can be limited due to the lack of comprehensive data and a high degree of epistemic and stochastic uncertainty. We use a combination of Bayesian Network (BN), System Dynamics (SD) and participatory modelling to develop a risk assessment tool for managing water-related health risks associated with extreme events. The combination of BN and SD modelling offers a number of advantages over other environmental modelling techniques; the capacity for dealing with a high degree of uncertainty, the use of feedback loops (SD only) and the ability to elicit and integrate quantitative and qualitative data (including expert opinion). The risk assessment tool developed is applied to the raw water delivery system supplying Prospect water filtration plant system (Sydney, Australia), which is the main source of potable water for the Sydney metropolitan region. Key-stakeholders were engaged in developing and populating the conceptual models that form the basis of developing the BN and SD models. Conceptual models were developed by the stakeholders around the key indicator parameters of turbidity, water colour and Cryptosporidium sp. levels. These three conceptual models were combined into a single risk model and used for developing separate BN and SD models. Additional stakeholder workshops were conducted to refine the models (structure and parameter values) and to provide validation of the model outputs. Here we present the development of a BN model designed to understand the risk of extreme events on the ability to provide potable water of a specified quality. The model has undergone development and preliminary parameterization via two participatory workshops. However, its development is an ongoing process with the next stage involving supplementing the ‘expert opinion’ used to parameterize the model so far with ‘hard’ data. The completed models will quantify the sensitivity of the Prospect raw water delivery system to different types and combinations of extreme events (both natural and anthropogenic). The BN model will provide a risk management tool for estimating the probability of (top-down modelling), and requirements for (bottomup modelling), meeting water quality guidelines. The SD model will provide a means of testing the implementation of different management scenarios and the impact that this has on water quality for different time horizons. Overall, these complementary modelling methodologies will assist water treatment operators, water managers and other stakeholders in developing evidence-based mitigation strategies leading an to enhanced resilience of the system.