Design and Analysis of Samonid Tagging Studies in the Columbia Basin. Volume III, Experiment Designs and Statistical Models to Estimate the Effect of Transportation on Survival of Columbia River System Salmonids.

Experiment designs to estimate the effect of transportation on survival and return rates of Columbia River system salmonids are discussed along with statistical modeling techniques. Besides transportation, river flow and dam spill are necessary components in the design and analysis, otherwise questions as to the effects of reservoir drawdowns and increased dam spill may never be satisfactorily answered. Four criteria for comparing different experiment designs are: (1) feasibility; (2) clarity of results; (3) scope of inference; and (4) time to learn. A controlled experiment with treatments that are a combination of transport status (transported or left in-river), river flow level, and dam spill level should provide the clearest results of transport effect. The potential for bias due to interactions between year effects and the treatments is minimized by running as many treatments as possible within a single outmigration year. Relatedly, the most rapid learning will occur if several different treatments are implemented at randomly chosen time periods within thesame outmigration season. If the range of flow and dam manipulation includes scenarios of interest to managers, the scope of inference should be satisfactory. On the other hand these designs may be the least feasible; trying to manage the river system under a sequence of deliberately chosen flow regimes within a single season, for example, may be quite impractical. At the other end of the spectrum are designs that simply have two treatment combinations, transportation and being left in-river, and the influence of flow and spill are controlled for, if possible, in after-the-fact statistical analysis. Because of possible confounding influences of flow and spill on the transportation effect, these designs could yield the most ambiguous results and require the most years of experimentation to learn. If flows and spill are not manipulated in a planned, well defined, and impartial manner the scope and quality of inference may not be satisfactory. On the other hand, these designs are the simplest to implement. Implementation issues are: (1) The nature of flow and spill level manipulations will need clear definition, either in absolute terms, cfs, or relative terms, such as spilling 10% of the water. (2) Relatedly, system wide implementation of flow and spill levels will provide simpler interpretation of results than will mixing spill rates, for instance, between dams. Transporting fish from just one location will also simplify interpretation. (3) Tagging of experimental fish should be done well upstream of the dams with random assignment to transport or in-river groups done later, near the dams, to minimize biases from delayed tagging mortality. (4) Tagging with PIT tags and CWTs in combination will provide evidence of any potential homing problems. (5) High PIT tag retention rates are important to minimizing potential analysis problems (thus on-going research to improve retention is vital). (6) Approximate sample sizes to achieve a desired level of precision can be calculated fairly easily using formulas provided in the report