(dominant) component, the mean response, which can often be different from the 

 central tendency response, may better represent the total functional group. A 

 simple deterministic solution will mimic the typical response only. This 

 particular type of analysis addresses both the uncertainty introduced by model 

 aggregation as well as that introduced by imprecise measurements of kinetic 

 coefficients. Methods other than Monte Carlo can be used to introduce 

 s tochas ticity in ways which are based on different ecological assumptions. For 

 example, where Monte Carlo assumes parameters are uncertain but constant 

 throughout the simulation, 1st Order Variance Propagation Analysis and other 

 methods based on random differential equations assume instantaneous probability 

 distributions for parameters during each simulation. These methods can help 

 place in perspective the potential errors (both real and artificial) introduced 

 by measurement uncertainty, conceptual aggregations, etc. 



Including potential variability in ambient properties (e.g., currents, 

 light) and in aggregation effects (e.g., within functional group variability) 

 in a Monte Carlo analysis or other method, for example, is one way to produce 

 distributional output that can be summarized as percent occurrence and duration 

 of certain desirable or undesirable events. This becomes more usable information 

 relating to problems like that of dissolved oxygen as mentioned above. 



Although methods of uncertainty analysis are well documented and have had 

 extensive application in physical sciences, they have had only limited applica- 

 tions to environmental/ecological problems. More work needs to be done to 

 assess the relative utility of various approaches to uncertainty analysis in 

 ecosystem models. 



Thus, each approach in the range of strategies included in the general 

 term "ecosystem modeling" has identifiable strengths and weaknesses. None is 

 "best" or "worst," but each may contribute useful input to the management 

 process given different questions being asked, different available data, 

 different personnel, and different constraints in time and money. Generally, 

 the simple models will be easiest to understand by the manager inexperienced in 

 modeling, and may well be the most effective in terms of cost and time. However, 

 their appeal is deceptive; if the proposed perturbations are beyond the prior 

 range experienced by the system, predictions will be very tentative at best. 

 In these cases, especially, detailed mechanistic models may provide the only 

 numerical tool for evaluating various scenarios. 



Given the spectrum of modeling strategies and their respective assumptions, 

 it is clear that it is desirable to apply more than one modeling approach. 

 When the predictions of various models spanning this range of resolution 

 converge, managers and scientists may have more confidence in the results of 

 the modeling experiment. 



Another viable though expensive alternative is what has been called "living 

 models." Large scale macrocosms have recently reached a level of sophistication 

 to be useful in management. A specific example is the Marine Ecosystem Research 

 Center (MERC) in Rhode Island. This facility has met two critical criteria: 

 the experimental systems can be shown to 1) replicate each other, and 2) mimic 

 changes in the natural system within acceptable limits. Real experiments can 

 be designed on these living models which have a capability to respond in ways 

 no numerical model can ever approach. 



247 



