A crucial aspect of the bioregional assessment (BA) methodology is the open and transparent accounting for uncertainty in the predictions. It is this quantification of uncertainty that will feed directly into the risk assessment.
Uncertainty analysis has always been an essential part of any scientific research. In recent decades, however, uncertainty analysis has received increased attention in all disciplines of environmental research. It is not hard to imagine that various environmental science disciplines have adopted and adapted different uncertainty analysis techniques in order to address the key issues relevant to their domain. For example, in geology and hydrogeology emphasis has traditionally been on characterising the spatial heterogeneity of model parameters, while the focus in hydrology is more on characterising the uncertainty in time series of inputs and observations (Gupta et al., 2012).
There are very few studies published in which uncertainty analysis is carried out through a chain of models from different environmental science disciplines at a scale that is comparable to what needs to be achieved in the BA. The 2008 performance assessment of the high-level radioactive waste repository at Yucca Mountain, Nevada, United States is a prime example of a similar uncertainty analysis exercise. The performance assessment is the culmination of over 30 years of multidisciplinary research (Helton et al., 2014) and assesses the dose of radioactivity a reasonably maximally exposed individual would be exposed to after 106 years of nuclear waste disposal at the Yucca Mountain site. The performance assessment incorporates a complex model chain in which the uncertainty in 392 physical variables is propagated over a range of scenarios by running the model chain several hundreds of times for each scenario. The model runs provide a probability distribution of the dose of radioactivity after 106 years, while post-processing of these enables the identification of the most influential variables. An example for the baseline scenario can be found in Hansen et al. (2014).
Bastin et al. (2013) provide a more general overview of managing uncertainty in integrated environmental models. They argue for probabilistic characterisation of uncertainty as the most quantitative and objective approach, recognising that this will always entail a level of qualitative treatment of model structural uncertainty, mostly based on expert judgement. In order for a framework to be able to manage uncertainty through a chain of interlinked models or components, Bastin et al. (2013) highlight the need for a well-defined protocol for communication of probabilistic uncertainty between model components, an appropriate mechanism to propagate uncertainty, which ideally should require minimal change to the component’s model code, and the ability to accommodate different levels of spatial and temporal support among the different components. To make such a framework workable, tools are needed that can aid in expert elicitation (the formal process of capturing expert knowledge) or statistical inference based on observation data of input uncertainties, formal sensitivity and uncertainty analysis techniques, visualisation methods of uncertainty in space and time and finally, tools to validate model chain outputs against observation data.
The methodology for the propagation of uncertainty through models in the bioregional assessment presented in this document will take on board several of these concepts and tailor these to the specific scope of and resources available for bioregional assessments. This submethodology focuses on the propagation of uncertainty in individual, bioregion-specific physical models. The companion submethodology M10 (as listed in Table 1) for identifying and analysing risk outlines and discusses the overarching risks and sources of uncertainty that are shared among bioregions and are inherent to the BA methodology. The companion submethodology M08 (as listed in Table1) about receptor impact modelling also discusses the propagation of uncertainty from the physical models into the receptor impact models.
Chapter 3 provides an overview of the requirements that the methodology will need to fulfil to achieve the goals set out by the BA and, equally important, will establish a number of key assumptions that are needed to have a methodology that is applicable in practice. Chapter 4 discusses the methodology in greater detail. Chapter 5 outlines the outputs of the analysis and where they are used. The last Chapter reiterates and summarises the key points of the methodology.
