Challenges in Designing Management Experiments to Resolve Critical Uncertainties

Experimental management requires spatial/temporal contrasts in treatments. In our initial scoping of experimental management actions (Marmorek et al. 1999), we determined that for many actions affecting the mainstem migration corridor (e.g., changes in flow, methods or amount of smolt transportation, magnitude of hatchery releases, harvest rates), one can only create temporal contrasts, not true spatial contrasts (i.e., an independent control system). Partial spatial contrasts are possible by comparing the performance of Snake River stocks with stocks in the Lower or Mid-Columbia region (Fig. 1), using the approach developed by Deriso et al. (1996). However some of the Snake River treatments could also affect Lower Columbia stocks, confounding the ‘control’ stocks. For example, upstream changes in flow, transportation, or hatchery releases could all affect conditions for Lower Columbia River stocks during the critical period of entry into the estuary and ocean. Because the four Snake River projects are not used as storage reservoirs, drawdown actions are less likely to have confounding impacts on downriver stocks. PIT-tags offer an opportunity to create control groups within the same year for different transportation methods, but the total number of tags required depends exactly on what one defines as a control (e.g., smolts passing down river and never undetected at any project, or simply smolts that are never put in a barge or truck), as well as the level of marine survival of both treatment and control groups.

In our more detailed analyses of the five priority experimental management actions, we used relatively simple models to simulate “true” future survival changes associated with the candidate actions. We then assessed the ability to learn by seeing how well the experiment and future monitoring could estimate the “true” survival change in a fluctuating environment (Peters et al. 2000; C. M. Paulsen and R. A. Hinrichsen, unpublished data). The main metrics of how much can be learned from an action were expressed in terms of the probability of estimating effects of an action over various time frames, or, conversely, how long it would take to estimate an effect with a certain level of confidence. Various criteria can be applied to determine how long an experiment needs to be run to estimate effect sizes that reflect the risk preferences of decision makers. The models also estimated various conservation metrics for each action, which were compared to the ability to learn to assess the trade-off between learning and conservation objectives. To avoid the confounding of gradual changes in climate or ocean conditions, we found that the most effective temporal contrast is switching an action on and off in alternate years. Sub-basin actions such as carcass fertilization could utilize more efficient experimental designs with both spatial and temporal contrasts, and hence obtain a faster rate of learning. However, their survival benefits are expected to be only modest, and insufficient on their own to recover the stocks.