Development of Bayesian Belief Networks for modelling the impacts of falling groundwater

Media type: 
Reports
Author/s: 
Peter Speldewinde
Institution/s: 
The University of Western Australia
State: 
Western Australia
Year: 
2013

Supporting Document 6 from the NCCARF project ‘Adapting to climate change: a risk assessment and decision making framework for managing groundwater dependent ecosystems with declining water levels'.

Executive summary

Bayesian Belief Networks (BBNs) are an excellent tool for assessing the impact of climate change on groundwater dependent ecosystems. Due to its visual nature BBNs present a tool for communicating the environmental issues and processes and also a means of gathering additional information to feed into models or develop new models.   BBNs are based on Bayesian probability which states that for any two events, A and B, the probability of event B occurring given that event A has happen (p(B│A) can be determined using the formula:  p(B│A)=p(A│B)×p(B)/p(A)  where p(A│B) is the probability of event A occurring given B, p(B) is the probability of event B and p(A) is the probability of event A. A BBN is composed of nodes (or variables) which have causal links where changes in the state of  one node may influence other nodes linked to it. The nature of these changes are defined by conditional probability tables which give the probability of an outcome given the change in the influencing nodes.

BBNs have a number of advantages (Jakeman, 2009 #6)-

  • Easily updated with new submodels and new information
  • Spatial and landscape components can be included as separate nodes
  • Easily used as a tool for communicating complex environmental problems among experts, managers and stakeholders
  • Can integrate models of different types
  • Can be used as a decision making tool
  • Transparent.

Bayesian belief networks (BBNs) were developed to model the potential impacts of climate change on groundwater dependent ecosystems. Three systems were chosen as case studies (Gnangara Mound, Blackwood River and Margaret River Caves).  Each system had varying degrees of data available, ranging from a data rich case study (Gnangara Mound (invertebrates and vegetation) through to a data poor case study (Margaret River Caves).

The development and testing of the BBNs followed the process of-

  • Developing a conceptual model for each of the systems: In this stage the identification of important system variables and links between variables were established.  In this case this was done at a workshop with experts defining variable and causal links.
  • Parameterisation of the models with data: In this stage states and probabilities for each variable were assigned, with each variable being discreet. As the three systems varied in the quality and quantity of data available, ranging from a completely data driven approach for the data rich Gnangara Mound wetlands (invertebrates and vegetation) through to an expert opinion approach for the Margaret River Caves and Gnangara Mound frogs. For all case studies NeticaTM v4 (www.norsys.com) was used for the construction of the BBNs (there is a range of BBN software, such as GenieTM (www.genie.sis.pitt.edu) and HuginTM (www.hugin.com), any of these could have been used).  
  • Evaluation of the models: Evaluation of models was undertaken in two forms, expert opinion and sensitivity analysis. Qualitative feedback was obtained through stakeholders and experts in workshops where the models were demonstrated. Sensitivity analysis identifies how sensitive a conclusion is to the evidence provided. Sensitivity analyses were conducted at different groundwater levels (node set to 100% for a particular groundwater level) to determine major driving nodes.
  • Analysis of the impacts of various climate change scenarios on the systems using the BBN: Analysis of the impacts of various climate change scenarios was conducted using GIS for the Gnangara Mound and Blackwood River study sites, where groundwater level projections under different climate change scenarios were modelled using the BBNs (see SD7 Neville 2013).

In the case of the Gnangara Mound wetland invertebrates and wetlands, which had an extensive data set, BBNs were constructed using only available data. In the case of the Blackwood River where data was less extensive a combination of data and expert opinion was utilised. In the case of the amphibians and Margaret River caves case studies, where there was not appropriate data, expert opinion was utilised. In all cases BBNs could be constructed and the networks were able to model the impacts on the systems examined due to changing groundwater levels.

The case studies demonstrate the use of BBN’s in modelling the impact of altered groundwater levels, due to climate change, on groundwater dependent ecosystems.  The case studies used a variety of information from extensive datasets (Gnangara mound invertebrates and vegetation) through to expert opinion (Gnangara mound frogs and Margaret River caves). The models provided a visual representation of the systems examined and allowed the manipulation of starting conditions for the models for the testing of different scenarios.

Please cite this report as:
Speldewinde, P 2013, Adapting to climate change: A risk assessment and decision making framework for managing groundwater dependent ecosystems with declining water levels. Supporting document 6: Development of Bayesian Belief Networks for modelling the impacts of falling groundwater due to climate change on groundwater dependent ecosystems, National Climate Change Adaptation Research Facility, Gold Coast, 35 pp.

View the main report from this suite
Visit the research project page 

Cover image: Quin Brook is located on the northern Gnangara Mound, Western Australia. This photo was taken in 2008 when there still used to be water in it © Dr Bea Sommer, Edith Cowan University, Centre for Ecosystem Management 

X