The statistical approach is the oldest, most common, and potentially the 

 simplest form of impact estimated. The basis of the approach is the derivation 

 of an empirical transfer function relating some ecosystem metric to impact 

 magnitude. Typical metric constructs include matrices, perhaps with weighting 

 functions (Cantor 1977), species diversity indices, and correlation procedures 

 (Pielou 1977). Field sampling and monitoring programs, coupled with a variety 

 of statistical methods, represent the standard format for statistical environ- 

 mental impact analyses. Major drawbacks of the approach arise from its sensi- 

 tivity to the metric selected, and the high degree of natural variability usually 

 reflected in the underlying data. This latter problem results in a common 

 inability to answer the question, "When is an impact (i.e., a change in the 

 system metric) significant?" In addition, questions of which variables to 

 measure, over what time span, and at what sampling intervals and locations are 

 non-trivial, but must be answered. Because of the high degree of agglomeration 

 inherent in statistical methods, the dynamic complexities of the "real" system 

 become obscured. Thus, little insight into the processes controlling system 

 response is gained, and the ability to address a variety of realistic management 

 options is usually lost. 



At the other end of the complexity spectrum lies the full ecosystem modeling 

 approach. In its purest form, this methodology incorporates all our knowledge 

 of environmental functioning at the process level. In practical applications 

 some agglomeration at the biomass level (Laevastu and Larkins 1981) or at the 

 level of numbers (Anderson and Ursin 1977; Reed and Balchen 1982) is necessary 

 to achieve completion of a project within prescribed temporal, economic and 

 computational constraints. In the past, this agglomeration commonly led to a 

 "box model" representation, in which species or species types and their inter- 

 actions were represented through a set of highly parameterized differential 

 equations (e.g., Chen 1975; Kelly 1975). Increasing accessibility of powerful 

 computational facilities has made feasible more detailed, pro cess -specific 

 representations. Such detail is attractive because of increased "realism". The 

 construction of such an ecosystem model is based on some form of conservation 

 laws, including appropriate source, sink, and interaction processes, to relate 

 ecologically critical components. 



The more detailed and "realistic" these models are, the more variables 

 and parameters they require. Such efforts are therefore sometimes characterized 

 as "data hungry", and may require relatively arbitrary assignment of values to 

 many parameters. Conclusions drawn from simulation outputs may therefore be 

 subject to considerable uncertainty. The data necessary to support this modeling 

 approach, in terms of input as well as validation, is rarely available at present. 



On the positive side, an ecosystem model with appropriate spatial, temporal, 

 and biological resolution makes better use of available data than simple 

 statistical analyses, in that information inherent in the data is conserved in 

 the model. Spatial species dynamics, for example, are well documented in most 

 fisheries data, but traditional fish population models (e.g., Beverton and Holt 

 1954) have essentially neglected this aspect. More recent ecosystems models 

 for fisheries management (Laevastu and Larkins 1981; Reed and Balchen 1982) 

 have demonstrated the importance of spatial dimensions in arriving at an under- 

 standing of the governing processes. 



104 



