Department of Mathematics
Environmental Modeling - Unveiling the Truth beneath the Fantasy
John Doherty
Watermark Numerical Computing, Brisbane, Australia
Dr. John Doherty, author of PEST, will be visiting us on Thursday, 4th October. John lives in Brisbane, Australia, where he runs his own consulting company, Watermark Numerical Computing. However for the last few months he has been working as a visiting professor with the University of Idaho in Idaho Falls. Much of his time there has been devoted to the development of parameter estimation and predictive analysis tools for use in surface water and ground water quantity and quality modeling.
Alongside the growing use of models in environmental management, is a growing skepticism of the usefulness of these models. While many outside the modeling profession still cling to the idea that "if it comes from a computer it must be right", there are a growing number of cases where the use of models in environmental management has been disappointing at best, and misleading at worst. In fact, there are signs of a crisis within the modeling profession. On the one hand there is general recognition that an attempt to simulate environment processes numerically can provide a sounder basis for the making of important decisions. On the other hand, many modelers are loathe to raise the expectations of their clients, or stakeholder groups, too high with regard to the usefulness of their models in the management process.
So in this time of re-assessment, just how high should expectations be raised? And where exactly should modeling fit into the environmental management process? And should a modeler suffer a severe identity crisis if his/her model cannot provide the impossible "answer at the back of the book" that many are seeking from it?
The talk will attempt to address these questions by first demonstrating that predictions made by an environmental model will, by their very nature, be accompanied by a (sometimes frighteningly large) margin of uncertainty. It is demonstrated that the higher the level of "system detail" that a model attempts to simulate (eg, contaminant movement in areas of high geological heterogeneity, the response of a catchment to extreme climatic events, nuances of groundwater-surface water interaction, etc), the greater will be the uncertainty with which such predictions are made. Methodologies whereby model predictive uncertainty can be quantified will be discussed. Finally, a rationale will be presented for determining the "point of diminishing return" in the model construction process - this being the point where the data at hand, and our understanding of environmental processes on a field scale, precludes the devoting of any further resources to the building of a model.
All interested persons are welcome.
The talk will be accessible to upper class students.