III.4.4 - Summary of validation with survival studies

Yearling chinook provide the largest number of studies for comparison, as well as broad geographic coverage. In general, model predictions of survival are in excellent agreement with the survival observations in three regions of the river system: the Snake River, mid-Columbia, and below Bonneville. All of the model-estimated survivals are plotted against observed survivals in Fig. 80; the fit appears excellent, and the regression, though not forced through the origin, is highly significant and close to a one-to-one line (modeled survival (%) = 5.05 + 0.94*(observed survival), r2 = 0.86).

The validation of the model with the survival studies from brand releases of fish between 1966 and 1983 for the Snake River and 1984 through 1986 in the Upper Columbia indicates that for this species the model fits the survival pattern in the years that the hydrosystem was being completed.

It appears that in many of the PIT tag studies that fish, following tagging, were sedentary for a time, ranging from a few days to a few weeks, during which they were presumably acclimating to the presence of a PIT tag. Observations suggest that tagged fish took a longer time to arrive at the first collection point but afterwards their migration rate increased. We found that delaying movement of yearling chinook for a time after actual release produced model predictions that were close to observations.

Fig. 80 Graph of all yearling chinook survival validation efforts. Solid line is a one-to-one relationship, the dotted line is a linear regression.

Fall chinook

Data to evaluate the ability of the model to predict fall chinook survival was limited, reliant on assumptions and constrained by uncertainties in the analysis. Even so, the model does provide a consistent representation of Snake River and mid-Columbia fall chinook during their in-river migration. The analysis also suggests that temperature is an important factor in fall chinook survival. Some of the inconsistencies between the modeled and observed survival may be resolved with better temperature modeling. Modeled and observed survivals are plotted in Fig. 81; the fit is not particularly good, and the regression line suggests a bias toward overestimating survival at the low end and underestimating at the high end {modeled survival (%) = 13.71 + 0.64*(observed survival), r2 = 0.67}. We suspect that making model parameters more stock-specific (i.e. distinguishing the different travel time properties of Snake and mid-Columbia stocks) will improve the fit.

The CRiSP model was not able to predict fall chinook survival between the time they were tagged in their juvenile habitat and arrived at Lower Granite dam. Preliminary work does suggest that the timing of arrival can be predicted from accumulative temperature but survival predictions remain problematic.

Fig. 81 Graph of all subyearling chinook survival validation efforts. Solid line is a one-to-one relationship, the dotted line is a linear regression.

Steelhead

Data for steelhead comparisons are neither as abundant as for yearling chinook, nor as sparse as for subyearling chinook. CRiSP comparisons, to 1994 PIT tagged fish from the NMFS survival study and mid-Columbia brand releases from 1984-1986, produce consistently good fits with regard to travel time, but equally consistently underestimate survival. Because travel time fits are adequate, it appears that the current calibration overestimates the rate of mortality suffered by steelhead as they travel through reaches (but we note that the 1994 NMFS survival study provides estimates of lower survival for steelhead than for yearling chinook, contrary to expectations; Muir et al. 1995). At the same time, fits to survival data from the 1970's in the Snake River provide excellent agreement to travel times and good fits to survivals. Future calibration work will take this into account, and model estimates of survival will be adjusted to provide agreement to all available data. Modeled and observed survivals for steelhead are plotted in Fig. 82; the tendency for slight overestimation of survival at the low end produces a moderately large intercept in the regression {modeled survival (%) = 4.74 + 0.80*(observed survival), r2 = 0.94}, but the attenuation of the data provide an excellent overall fit. Here, as with subyearling chinook, our analysis will benefit from distinguishing behavioral components that differ among the various steelhead stocks in the river system.

Steelhead, due to their size, are also more vulnerable to gas bubble-based mortality than are chinook salmon. This may be a source of overestimation of mortality, and will also be examined.

Fig. 82 Graph of all steelhead survival validation efforts. Solid line is a one-to-one relationship, the dotted line is a linear regression.

Summary of fits to survival estimates

The difference between the predicted and observed survival can be expressed in terms of a generalized fitting measure: the percent difference between the two survival estimates relative to the observed survival estimate. This is expressed as the FIT measure and is defined

(173)

Values of FIT greater than zero imply that observed survivals exceed model predicted survivals, while negative values imply the model predicted survivals higher than observed. The mean and standard deviation of the FIT measure is illustrated in Table 80.