Frank Köster

Combining Methodical Procedures from Knowledge Discovery in Databases and Individual-Oriented Simulation

Systems analysis and the exploration of data are important tasks during the course of environmental epidemiological studies. We examine knowledge discovery in databases combined with individual-oriented modeling and simulation to support the detection of hypotheses about cause-and-effect relationships within environmental systems. An individual-oriented model and detected hypotheses can uncover possible explanations for the current state of health of a study population. Such a model can support future planning and decision making for healthcare management. The main goal of attempting to use these methods in epidemiological research is to reduce expenditures in costs and time for a study as well as to improve the analysis and interpretation of available data.

Scenarios and Experiments

You will spend most of your time working with META-X specifying scenarios and experiments, i.e. with translating your real-world or theoretical problems into parameter sets for the generic metapopulation model which is implemented in META-X. Most elements of how to specify scenarios and experiments are described in the Guided Tour and will not be repeated here in detail; in particular, the Screenshots are not repeated. Instead, this chapter gives an overview of: n n nThe structure and elements of the Experiment Wizard. n n nThe structure and elements of the Scenario Wizard. n n nThe specification and purpose of ‘homogeneous’ parameters. n n nThe hierarchy of model parameters in META-X which allows you to create ‘user-defined’ scenarios.

The Landscape Editor

The right-hand part of the META-X window contains the Landscape Editor which is designed to visualize the landscape of a scenario, to modify existing scenarios via a graphical interface, and to create new scenarios with this graphical interface. Once a scenario has been selected in the Project Tree by a double-click or by choosing EditorpLoad Scenario in the menu, the Landscape Editor will indicate: n n1. n nWhether the scenario is completely specified or some parameters regarding patches, dispersal range or correlation length are still missing. n n n n n2. n nThe position of the patches and if a patch is occupied by a subpopulation. n n n n n3. n nThe scales of the scenario, i.e. the basic scale of the map, dispersal range and correlation length. n n n n n4. n nIf a so-called ‘Local Aspect’ is chosen, different aspects of the patches, i.e. the mean time to extinction, number of emigrants produced or number of immigrants needed to establish a new subpopulations (with 50% probability). n n n n n5. n nThe connections between patches, i.e. all pairs of patches which can recolonize each other via dispersal.

The Project Tree

The Project Tree in the left half of the META-X screen is designed to organize your projects: n n nIt hierarchically lists the experiments of a project, the scenarios of an experiment n n nand the parameters and evaluations of experiments and scenarios. n n nThe most important procedures of META-X (scenario, experiment and simulation n n nwizard) can be started directly from the Project Tree. n n nScenarios may be selected to be displayed in the Landscape Editor. n n nYou can copy (or cut) and paste existing experiments and scenarios. n n nExperiments and scenarios can be deleted or renamed. n n nYou can modify the sequence of scenarios in the experiment.

Parameterizing META-X

One main task when working with META-X is to translate questions regarding hypothetical or real metapopulations into parameterizations of the META-X model. To be able to do this, you need to know the model and exactly what its parameters mean (Chap. 14). The next step is to compile all the relevant empirical information available and to extract the model parameters. To give you an idea of how to do this, we briefly describe some general concepts of parameterizing META-X (or any other PVA model) in this chapter. However, this is not the place for a complete introduction into the problem of parameterizing (meta-)population models. There is a whole body of literature on, for example, obtaining demographic or dispersal data from mark/recapture studies (Moilanen et al. 1998, Henle et al 1999, Moilanen 1999, Hanski et al. 2000). In general, if you want to learn how to parameterize PVA models, you will need to scan the PVA literature (see suggested readings at the end of Chap. 13) and assemble your own tool chest.

Import, Export and Report

META-X has four ways to communicate with ‘the rest of the (computer) world’: n n nImport of model parameters. This is useful if you want to import model parameters from other programs, for example programs which calculate ‘Patch characteristics’ (mean time to extinction, etc.). The format of the import files is thus an interface between all kinds of sources which produce model parameters for all kinds of species and landscapes, and META-X. n n nExport of simulation results. Raw data from the simulations may be exported and analyzed with other programs. n n nExport of graphs. Some of the diagrams of META-X and the entire window of the Landscape Editor can be exported to other programs via the clipboard, or sent directly to the printer. n n nReports. You can generate reports of scenarios and whole experiments which contain all the information and parameters specified by you, and all evaluations. Reports are HTML format and can thus easily be imported into wordprocessing programs, printed and modified.

Goals, Methods, and Concepts of PVA

META-X has been developed as a tool for education and for practical use by specialists such as field ecologists, conservation biologists, managers of natural resources, and environmental decision-makers. Working with META-X requires no programming and — at least at the level of the metapopulation — no modelling, and is therefore easy to use by non-experts. However, as with any tool, even nonexperts have to know the purpose of the tool and its basic function if they want to use it appropriately. Moreover, META-X is not a fully ‘canned’ PVA software because external sub-models, for example of local population dynamics, are needed to parameterize META-X for actual species and landscapes. Therefore, to work with META-X you need to be familiar with the basics of PVA. In this chapter we briefly introduce the goals, methods and concepts of population viability analysis (PVA). In particular, we explain in detail how to quantify the persistence and viability of populations (and metapopulations), since this is usually not explained in PVA literature. For those needing a full introduction into PVA, we strongly recommend consulting the literature listed at the end of this chapter.

Simulation and Evaluation

As soon as you have specified scenarios and experiments, the time has come for the computer to do its job: to simulate metapopulation dynamics for a certain number of years, e.g. 300 years. ‘Simulation’ means starting with a certain state of the model metapopulation (i.e. a certain pattern of patch occupancy) and then using the model parameters to calculate the state of the metapopulation after one time step. This new state is then used to calculate the state after the next time step, and so on.

The META-X Model in Detail

A generic metapopulation model which is to be used for many different species, landscapes and problems has to be general but still structurally realistic. Structural realism is crucial because users cannot change the model’s overall structure, only tailor it to their problems by specific parameterizations. Structural realism means including the key structures and processes of metapopulation dynamics. In META-X, structural realism is achieved by focusing the main model on key structures and processes at the metapopulation level.

Modeling a library with THORNs

Thorns combine the widely used object-oriented programming language C++ with various features of Petri nets for modeling concurrency and time. In this way complex distributed systems can be modeled in a detailed manner. Thorns can be transformed to C++ code and executed sequentially or concurrently by simulators for validation and experiments. This paper shows both, features of Thorns and their modeling approach by an example.