Rock Island Sockeye Travel Time

The Data

The fish were tagged and released at Rock Island and observed at McNary Dam. 37 cohorts were analyzed over the 5 year period 1992-1996. The migration period was mid April through early May.

Models

We applied three nested models of increasing complexity to the data:

• Model 1 uses a constant migration rate.
• Model 2 incorporates a linear flow relationship.
• Model 3 adds a seasonal/flow interaction - fish use more of the flow for migration as the season progresses.
• Model 4 adds an experience factor - flow independent migration speed increases the longer the fish have been in the river. However, since these fish only passed through one reach, (RIS to MCN), this model is inappropriate.

Model details are provided by Zabel, et al. (1996).

Graphical Results

The plots below show observed average travel times versus modeled average travel times to each of the three observation sites for the three models.

corresponds to observations at McNary Dam.

Tabular Results

Table 1 . Travel-time parameters for sockeye from RIS releases (1992-1996). *(Vvar is fixed at 200 to overcome the problems of small samples, long reach, and single observation site.)

model	parameter estimates (std. error)							resid. ss	mult. R2
	MIN	MAX	FLOW		TINFL	2	2*
1	13.8064 (3.6336)	-	-	-	-	-	200*	346.545	-
2	-2.0506 (12.769)	-	1.6402 (1.2784)	-	-	-	200*	145.087	.5813
3	-1.8461 (2.4843)	-	0.9069 (0.1945)	0.0438 (0.0301)	80.8237 (35.418)	-	200*	130.663	.6230

Discussion

The small cohort size creates some problems in the analysis of the data. As the fish spread through the river, their arrival times at McNary are very spread out. As a result. The VVar term becomes large and distorts the predicted average travel time. For the RIS releases we have held Vvar constant at a relatively high 200. This allows us to approach the minimum residual sum of squares (within 1 to 3 points) without getting the travel time distortions.

Model 3 gives a slightly better fit than model 2 and greatly reduces the standard error around the parameter estimates.