AN ANALYSIS OF PREDICTIVE MODELS FOR STRE AM SALMONID POPULATIONS

A total of 73 multivariate (empirical) models predicting salmonid popu lations from biological and environmental variables were assessed. In conjunction with this, the published results of 15 tests of models are briefly commented on. The relationships between the sources of the si gnificant variables and the performance of the models are discussed in relation to the development of HABSCORE - a management tool for salmo nid fisheries which is based on empirical predictive models. An assess ment of the predictive capabilities of the models implied that a combi nation of variables (those which relate to the large-scale features of the catchment and those which describe the instream conditions) were more useful in predicting trout stocks than either source of variables alone. Whilst raw data gathered from relatively 'narrow' ecological r anges have been used to formulate the majority of models detailed in t he literature, such models often have high predictive power only withi n the same ecological range, and are consequently restricted in their applicability elsewhere. In order to develop models which can be used as fisheries management tools it is desirable to base the model develo pment on data from a wide geographical base. Although specific definit ions of many parameters may differ between fishery workers, there is g eneral agreement regarding the nature of those parameters perceived to be useful or important for model development. Given the production of a series of rigorous definitions for these parameters it should be po ssible to propose a system of habitat description that would be both w idely applicable and would give rise to reproducible results.