Sheridan et al. (1981). The conclusion made in the 1981 paper was that shrimp 

 production would not be affected by decreasing the discard rate. This conclusion 

 was based only on relative rates of nitrogen remineralization from different 

 sources, and food flows to shrimp from different sources. In the simulations, 

 the response of phytoplankton production to changes in water nitrogen 

 concentration was dependent upon the Michaelis-Menten half-saturation constant, 

 relative to the concentration of nitrogen in the water, rather than on the 

 relative rates of release of nitrogen from the various compartments. The 

 simulation exercise demonstrated that it is possible for a change in a nitrogen 

 inflow that is very small relative to other flows to affect model results. The 

 model, however, does not include nitrogen fixation and denitrif ication. These 

 flows are difficult to quantify. If they are large in this system relative to 

 the nitrogen flows included in the model, they may influence the way that 

 variations in other nitrogen flows affect the system. 



Model results were sensitive to the weighting factor determining the intensity 

 of bottomfish predation on shrimp. Model results appeared somewhat sensitive 

 to the apportionment of nitrogen between the excrement and fecal material of 

 bottomfish and possibly also of zooplankton and other animals. Results are 

 undoubtedly also sensitive to the Michaelis-Menten half -saturation constant 

 that relates phytoplankton production to the concentration of nitrogen in the 

 water. Although values for these parameters that were used in the model had 

 some basis in the literature, none were specific to the north-central Gulf of 

 Mexico. No specific sensitivity tests have yet been run on these parameters. 

 The impact of variations in these values on model results and conclusions has 

 not been fully explored. 



Do model results apply to the real-world system? In the development of an 

 ecosystem model, there are at least four levels of procedure at which variations 

 can affect model results: (1) assumptions of the conceptual model, (2) 

 translation of the conceptual model into mathematical equations, (3) model 

 quantification, and (4) model computerization. Ecosystem models are difficult 

 to validate, but there are a number of things that can be done to improve 

 confidence in an ecosystem model and its results. One is to test the validity 

 of the assumptions independently of the model and to assess the dependence of 

 the model on these assumptions. A second is to test the sensitivity of the 

 model to variation in initial values, rate-coefficients, and other constants, 

 such as those singled out above (bottomfish selectivity weighting factor for 

 shrimp relative to alternative prey, apportionment of waste nitrogen between 

 feces and excrement, and Michaelis-Menten half-saturation constant relating 

 gross primary productivity to water nitrogen concentration). Quantification of 

 a model requires many rough estimations. It is important to know whether any 

 of these have major effects on model results, not only to evaluate the present 

 reliability of the model but also to determine what further work is needed to 

 improve the model's reliability. Sensitivity testing is done by varying values 

 one at a time and comparing simulation results. Sensitivity test results can 

 lead to a more intensive search of existing literature, which may yield improved 

 estimates, or to field or laboratory studies designed to obtain the critical 

 information that is needed to adequately answer the question. 



Sensitivity testing is needed not only for model input values but also for 

 model structure. The model is, of necessity, a simplification. Does the 

 addition of structural detail significantly alter system results? If so, this 

 needs to be known. Finally, errors in program coding can cause responses that 



219 



