Project title: Knowledge integration and Management Strategy Evaluation modelling
Program: Kimberley Marine Research Program

Modelling the future of the Kimberley region

The Models

In this project, the Management Strategy Evaluation will be carried out via two computer models: ALCES and Ecopath with Ecosim (EwE). The purpose of using these models is to integrate existing and new knowledge about the Kimberley system and to provide an estimation of the likely impacts of different stressors on the land (ALCES) and marine (EwE) environments.

Before the models can start answering management questions, the models need to be set up for the specific needs of the Management Strategy Evaluation project. This involves mostly three interrelated steps:

1. the model domain need to be chosen (this defines what is ‘internal’ vs ‘external’ to our system)
2. the models’ structure has to be defined (this defines what components of the system we include) and
3. the models need to be parameterised (we need to provide the model information about the system).

These steps require expertise and technical knowledge but also experience and good judgement. They define what is often called ‘the art of modelling’.

A model domain represents the geographical area covered by the model. The borders of the domain thus discriminate between what spatial features are ‘internal’ or ‘external’ to the model. The choice of the model domain thus reflects what we treat as internal and external to the Kimberly system. This choice is problem-specific. It depends on type of questions we need to answer and what processes are relevant to those questions. It also depends on technical issues related to the characteristics of the models we use, the computational effort they require and the level of model complexity we are willing to accept.
In addition, workflow constraints are critical. While ideally we would like to be able to change the model domain according to how the project develops and what questions become relevant when more information is made available, in practise this is rarely possible, since all model implementation steps depend on the model domain. As a result this choice has to be finalised early in a modelling project. Similar considerations apply to the models’ structure.

Steps 1 and 2 have been carried out in two phases. First, in October 2013 both Ewe and Alces teams met in Perth for a three-day workshop, during which it was decided i) that the two model domains would match spatially, ii) what the spatial resolution of the models should be, iii) what processes each model should include and iv) that Alces visualisation capabilities will be used to deliver the model outcomes. Second, via a number of video-conferences, the modelling team agreed on how the two models should communicate. Communication between the two models is needed to account for the interaction between terrestrial and marine environments and to fully integrate of the ecological, social and economic data. This integration will be carried out by i) defining a number of futures to simulate (see section ‘What we mean by Futures’) ii) running these future in Alces, to determine the land impact and stressors on the marine environment and iii) use Alces output to characterise EwE runs in order to model the marine impacts.

Step 3 (Models parameterisation) has been carried out independently by the Alces and EwE modelling teams. A model parameterisation takes currently available information about the state of the Kimberley system and codifies it in a way that can be used by a computer model. The role of this step is at times misunderstood and undervalued.  Three concepts are particularly important not only for this modelling project that for the overall Kimberly research effort. First, the outcome of a model parameterisation is not just a model input but also a snapshot of what we know about the system. The actual extent of this knowledge is often laid bare by the very process of model parameterisation. The need to provide a model with a numerical value for (say) turtle biomass and age structure forces the modellers to look for and certify this information if available and to extrapolate credible ranges from other sources if not available. Often it is not until this process needs to be carried out that the extent of our (lack of) knowledge is made clear. In Donald Rumsfel’s parlance, this process turns some data gaps from ‘unknown unknowns’ into known ones  Second, since data collection always needs to be prioritised, model sensitivity analysis may provide information on what type of information could provide the largest impact on management-relevant knowledge. Third, the existence of large data gaps at times leads critics to question the value and reliability of a modelling effort. This overlooks the fact that ultimately decisions will have to be made even in the presence of data gaps. Rather than being invalidated by large data gaps, modelling provides a way to recognise, formalise and evaluate the relative impacts of these data gaps.