418 John Woods
detritus and faecal pellets. As usual, the biological sophistication of models
designed for LE integration lags several years behind current practice in Eulerian
modelling. Here we give an indication of contemporary research aimed at improving both the specification and number of functional groups in LE modelling.
Carlotti and Wolf (1998) replaced the simple parametrization of herbivorous
zooplankton used in the WB model (based on body weight) by one based on structured growth. (For a general discussion ofthis approach see Tuljapurkar & Caswell
1997.) Broekhuizen (1999) developed an LE model with two functional groups of
phytoplankton based on diatoms and dinoflagellates respectively. Partridge (2000)
developed a more ambitious LE food web model with two groups each of phytoplankton and herbivorous zooplankton. It is too early to report on the multi-year
stability of virtual plankton ecosystems based on these models, but experience
gained with the WB model shows how to proceed.
18.16.4 Conclusion
Although the long term goal is to simulate individual species explicitly in plankton ecosystem models, that lies well into the future. That is for two reasons: (1) we
have biological primitive equations for only a handful of species, and (2) integrating models with thousands of species will require petaflops computers (which can
do one billion sums per microsecond). Both these barriers to explicit simulation of
biodiversity can be overcome, and probably will be overcome before your career
ends. Meanwhile, we must be pragmatic and base our models on functional groups
each of which represents many species. In this section I have discussed the scientific basis for those functional groups. The stability of VPEs based on the LE
method and their ability to simulate competition provide a sound basis for parametrizing functional groups in terms of biologic al primitive equations and ambient
climate.
18.17 The Ecological Turing Test
Testing the credibility of computer simulation goes back to the earliest days of
computing. Turing (1950) famously proposed a test to discover whether one was
communicating with a computer or a person. This "Turing test" involved asking a
series of questions devised to discriminate between man and machine. The
machine was deemed to have a mind if no question could be devised that showed
that it was not a person. We can extend that idea to testing an ecosystem model by
asking a succession of questions designed to discover whether a particular data set
originated in observations or in a computer simulation. I call this the "Ecological
Turing Test" (ETT).
18.17.1 Verification
The ETT is related to the classical procedure of verifying primitive equation
models by comparing selected features described by the simulation and by obser-
detritus and faecal pellets. As usual, the biological sophistication of models
designed for LE integration lags several years behind current practice in Eulerian
modelling. Here we give an indication of contemporary research aimed at improving both the specification and number of functional groups in LE modelling.
Carlotti and Wolf (1998) replaced the simple parametrization of herbivorous
zooplankton used in the WB model (based on body weight) by one based on structured growth. (For a general discussion ofthis approach see Tuljapurkar & Caswell
1997.) Broekhuizen (1999) developed an LE model with two functional groups of
phytoplankton based on diatoms and dinoflagellates respectively. Partridge (2000)
developed a more ambitious LE food web model with two groups each of phytoplankton and herbivorous zooplankton. It is too early to report on the multi-year
stability of virtual plankton ecosystems based on these models, but experience
gained with the WB model shows how to proceed.
18.16.4 Conclusion
Although the long term goal is to simulate individual species explicitly in plankton ecosystem models, that lies well into the future. That is for two reasons: (1) we
have biological primitive equations for only a handful of species, and (2) integrating models with thousands of species will require petaflops computers (which can
do one billion sums per microsecond). Both these barriers to explicit simulation of
biodiversity can be overcome, and probably will be overcome before your career
ends. Meanwhile, we must be pragmatic and base our models on functional groups
each of which represents many species. In this section I have discussed the scientific basis for those functional groups. The stability of VPEs based on the LE
method and their ability to simulate competition provide a sound basis for parametrizing functional groups in terms of biologic al primitive equations and ambient
climate.
18.17 The Ecological Turing Test
Testing the credibility of computer simulation goes back to the earliest days of
computing. Turing (1950) famously proposed a test to discover whether one was
communicating with a computer or a person. This "Turing test" involved asking a
series of questions devised to discriminate between man and machine. The
machine was deemed to have a mind if no question could be devised that showed
that it was not a person. We can extend that idea to testing an ecosystem model by
asking a succession of questions designed to discover whether a particular data set
originated in observations or in a computer simulation. I call this the "Ecological
Turing Test" (ETT).
18.17.1 Verification
The ETT is related to the classical procedure of verifying primitive equation
models by comparing selected features described by the simulation and by obser-