Modeling 455 



real controls operating within the ecosystem to construct a whole-system 

 model that was predictive. Yet we were able to contruct a model that 

 simulated the seasonal cycles of biomass and production quite well. If we 

 had not known so much about this system, perhaps we would have believed 

 that the model was entirely adequate. Some of the aspects of the model 

 that made us distrust it were its sensitivity to extremely small changes in 

 some parameters, its dependence upon tuning of some parameters or 

 constants about which we really knew little, and our ability to make this 

 model simulate any conceivable seasonal cycle by making slight changes 

 (this is true of most nonlinear models). In addition, our development of 

 stochastic versions of the deterministic models showed us how the slight 

 variability found in values measured in nature often had a drastic effect on 

 the output from the model. In contrast, our experience with this ecosystem 

 indicated that this amount of variability had no profound effect on the 

 behavior of the natural system. We concluded that perhaps a stochastic 

 model is preferable to a deterministic one but that we could not construct a 

 realistic stochastic model because of lack of data on the natural 

 variability. 



There were, however, several positive aspects of our experience with 

 modeling. First, a number of scientists and students learned a great deal 

 about the good and bad aspects of ecosystem modeling. Education in 

 ecosystem research was one goal of IBP. Second, the modeling forced us 

 to think critically about processes, interrelationships, and controls within 

 our pond. We hope we would have done this anyway but it is all too easy to 

 become caught up in the problems of rather routine data gathering and 

 avoid the more difficult problems. Third, the modeling forced a whole- 

 system approach and focused our attention upon some aspects of the 

 ecosystem that had been little studied; for example, the feeding rates of 

 protozoans on bacteria. Conversely, we became aware that some areas like 

 primary productivity of planktonic algae had been over-studied and 

 further intensive study was unnecessary. Fourth, the modeling effort 

 quickly made us aware of how little we really knew about some of the 

 processes and relationships in the ponds. This realization caused us to 

 consolidate our efforts into one or two ponds. The natural tendency of 

 ecologists seems to be to sample as many ponds as possible and to pick a 

 different pond for the study of each process. Finally, a fifth positive aspect 

 was the tests we were able to run with the small models or with small parts 

 of the large models. These were tests of the importance of various factors 

 or of the possible range of some rates that could not be measured. One 

 example was the determination of the likely rates of mixing of benthic 

 algae into the sediment. 



Overall, the use of modeling as described here was probably 

 worthwhile and was probably an aid to the project. The models did not 

 turn out to be the predictive tools we had hoped for but the process of 

 modeling was extremely valuable for a number of reasons. 



