Choosing a model

Introduction

Before chosing a model for a particular purpose, it is useful to understand what the models studied during ECASA are, and what they are not. They are not miniaturized versions of reality. Instead they are sets of mathematical equations that are intended to represent certain features of the behaviour of real-fish-farming systems and coastal ecosystems. Usually, these equations are solved, using computer programmes, for specified conditions: the solutions require information about the conditions. A model can give incorrect results: if it is used to simulate features or predict things for which it has not been designed; if its mathematical formulation is incorrect; and if the computer programme gives erroneous or inexact results. Here are some questions to ask when chosing a model:

• What are you planning to use the model to tell you? See 'Relevance';
• How reliable is the model for this purpose? See 'Reliability';
• How costly will it be to use the model for this purpose? See 'Resources'.

Model relevance

The models studied during ECASA have been categorized on the basis of:

• type of farmed organism - especially, bivalve shellfish versus finfish, and in some cases containing an organism model that can be adapted to a particular species;
• type of environment for which the model has been designed and in which it has been tested; where relevant, we list the ECASA sites for which model test reports have been provided; some models have been previously tested prior to ECASA, and this will be documented in the model descriptors that are obtainable from each model's page in this toolbox;
• the scale for which they have been designed;

There are several other matters that influence the relevance, reliability and cost of using models.

• grain: is the part of space that is simulated by a model, modelled as a single box, several boxes, or a network of points? More finely grained models can be more accurate, but they are also more expensive to use;
• dynamical vs steady-state/worst case models: dynamical models predict rates of change and hence time-series of modelled state variables; steady-state or worse case models can provide simpler or cheaper answers than dynamical models, but cannot, for example, simulate the course of events during a year;
  
• Parameters: these are the 'constants' in model equations; the more equations and parameters, the more work needs to be done in getting parameter values and, in some caes, adjusting them for the site being studied;
  
• boundary conditions ...: most models must be supplied with information about conditions outside the region that is simulated; this task is more demanding for models with many state variables and fine grain.

Model reliability

Because biological organisms and natural ecosystems are very complex, it is rare for model predictions to be completely correct. Nevertheless, a model will be useful if it can explain, or predict, a significant part of change in variables of interest.

During ECASA we applied a standard test of model reliability whenever possible. The test was based on the extent to which values calculated by the computer solutions of the mathematical equations for specified conditions, agreed with observations of the values of the model's state variables for the same conditions. See the model Reliability page for details of this test. Reports on reliability, using the standard test where appropriate, are included on most of the individual model pages. Some of the models considered within ECASA were not tested during the project. References for tests carried out in other contexts can be found in the model descriptors that can be downloaded from pages dealing with individual models.

The models in this toolbox can be used in several ways. The simplest is that of providing advice to managers in the form of answers to 'what-if' questions. For example, 'what would be the effect on the seabed near my farm if I double my production of sea-bass?' This question is best answered by running the model, first, for the existing production, and comparing the results with the observed sea-bed effect. If there are discrepancies, it may be possible to adjust some of the model's 'parameters' to get better agreement. Then run the model for the proposed higher level of production. This method will reduce uncertainity in prediction.

The second type of use, relates to assimilative or carrying capacities. Some models can be used to estimate these capacities, following the theory described in Management for Sustainability. Other models make use of the ideas of assimilative or carrying capacities by estimating the maximum stock or production that can be sustained within the appropriate capacity, as this is defined by a national quality standard (an EQS or EcoQO). In principle, the confidence limits of such estimates can be estimated from the statistics obtained during the reliability test of the model, especially when it has been tested over several sites. In practice, more work needs to be done to provide a robust estimate of confidence that will be good for all relevant sites. Our present advice is therefore this: limits estimated from models should be approached in steps, with appropriate monitoring at each step. See, in particular, the page on Monitoring for Public Environment Managers.

Resources needed to use a model

Developing and documenting a model is expensive. ECASA has done this so ... Nevertheless, there are costs (in euros or in skilled peoples' time) in using even a well developed model. The following costs should be considered when planning to use a model for the purposes described in this toolbox:

1. Licensing costs - for computer program used to run model. In some cases this program is proprietary to ECASA partner institutions. In other cases the program might be 'open source' but needs a proprietary software, such as Matlab (link) to run. Some of the ECASA models can be assessed through web sites, but a password may be needed.
2. costs of running the program and interpreting the results: computing sosts negligible in most cases, but users may need to develope some skills in using the program and model;
3. costs of obtaining and preparing the information on the 'specified conditions' relevant to the site or water body; detailed information on sea-bed topography (needed for hydrodynamical models), and boundary condition data, are often difficult or expensive to acquire.

Some of ECASA's models use standard software such as a spreadsheet or web browser, so minimizing costs 1 and 2; however there always remains the cost 3 of getting the information needed.