Key Uncertainties in the PATH Decision Analyses


PATH conducted extensive sensitivity analyses to determine which uncertainties were most influential in determining model outcomes. This involved comparing outcomes for all of the runs (i.e., all combinations of hypotheses) with the outcomes under a subset of runs associated with a particular hypothesis, or a particular combination of hypotheses. These comparisons looked at both the ability to meet survival and recovery standards and the ranking of decisions (i.e., which action had the higher probability of meeting a given standard). We considered a range of differences in these probabilities (i.e., 0.02, 0.06, and 0.1) to assess how robust our conclusions were. To independently check the inferences we drew from these methods, we applied Categorical Regression Tree analysis to the complete data set of decision analysis outcomes. The ‘CART’ trees clearly showed the relative importance of each action and hypothesis to the computed probabilities of survival and recovery. Sensitivity analyses for spring/summer chinook are described in Marmorek et al. (1998b, c, d) and Peters and Marmorek (in press). Sensitivity analyses for fall chinook are described in Peters et al. (1999, in press).

Using this approach, PATH scientists identified two key uncertainties that have the strongest effects on survival and recovery of Snake River spring/summer and fall chinook: extra mortality of nontransported fish and the relative post-Bonneville survival of transported fish compared to post-Bonneville survival of nontransported fish. Bonneville Dam is the last of eight dams that smolts pass on their way to the ocean (“BON” in Fig. 1).

1. Extra mortality of nontransported fish

Extra mortality is defined as any mortality occurring outside the juvenile migration corridor that is not accounted for by the other terms in the life cycle model used for retrospective and prospective modeling (i.e., terms for stock productivity and carrying capacity, mortality in dams and reservoirs, and estuarine/ocean mortality affecting all salmonid populations). Because many of the changes that may account for historical patterns in extra mortality all happened around the same time (e.g., Fig. 2) there is uncertainty about which of these factors (or mix of factors) influences extra mortality. Therefore, PATH formulated three alternative hypotheses about the source of this extra mortality:

   a. Hydro – extra mortality is related to the experience of smolts that pass through the hydropower system (e.g., delayed effects of stress).

   b. Regime shift – extra mortality follows a 60-yr cycle that is related to long-term cycles in ocean conditions. There are no actions that can be taken to reduce extra mortality, but extra mortality will eventually go down when ocean conditions improve.

   c. Stock viability (here to stay) – extra mortality is due to some phenomenon that will not be affected by any hydrosystem action or regime shift (i.e., interaction with hatchery fish, presence of diseases such as Bacterial Kidney Disease, or reduction in nutrients associated with historical declines in spawning stock).

Extra mortality can only be inferred from other measured quantities; it cannot be directly measured. This makes it difficult to monitor changes in extra mortality resulting from an experimental action, and thus to test alternative hypotheses. Nevertheless, extra mortality is still an important construct because (a) it helps to design experimental management actions that address its potential causes; and (b) it is needed to simulate the range of effects of alternative experimental actions to assess their relative risks and benefits.

2. The relative post-Bonneville survival of transported fish compared to post-Bonneville survival of nontransported fish

In the PATH modeling framework, the ratio of these two values is known as “D”. Like extra mortality, D cannot be directly measured, but must be inferred from other measured quantities (e.g., Transport:Control ratios and in-river survival estimates from transportation studies for spring/summer chinook). Differences in the assumptions used to estimate D led to alternative hypotheses about both historical and future D values, for both spring/summer and fall chinook.


In general, the ability of transportation to recover stocks depends directly on D (i.e., more likely to recover stocks when D is high, less likely when D is low). Drawdown actions were forecast to recover stocks over a wider range of D values (Marmorek et al. 1998b, Peters et al. 1999, in press, Peters and Marmorek in press). The ability of both drawdown and transportation to recover stocks also depends on the extra mortality hypothesis – both actions were more likely to recover stocks with the hydro hypothesis than with the regime shift or stock viability hypotheses.

Reducing these key uncertainties can help to identify the long-term management action that is best able to recover the stocks. There is an interaction between extra mortality hypotheses and the D value: forecasts of recovery are generally more sensitive to the extra mortality hypotheses if D is a high value. If D is high, fewer transported fish die below Bonneville Dam. Other factors causing extra mortality of all fish are then required to explain historical declines in overall survival. If D is low, post-Bonneville mortality of transported fish is sufficient to explain most of the observed historical declines in overall survival rates, and extra mortality factors affecting all fish become less important. This suggests that we should not measure D without also narrowing down the extra mortality hypotheses, and vice versa.