304 



Fishery Bulletin 103(2) 



Figure 10 



Track showing northward excursions of fish 159 (track extending to 38°) and 

 441 (track extending to 34.5°) between 1 April and 10 May 2003. Displayed SST 

 imagery is a composite for the month of April showing a 7°C temperature gradient. 



PSAT Tracker is similar to the Sibert et al. algorithm 

 in that it invokes a model of fish behavior; there is a 

 simple constraint on the maximum distance that a fish 

 can swim during a time step, and shorter tracks that 

 require lesser expenditures of energy by the fish are 

 favored. Like the Sibert et al. algorithm, the PSAT 

 Tracker also incorporates candidate points that are not 

 limited to the initial light-based estimate of longitude 

 but includes adjacent longitudes based upon the user's 

 assessment of the accuracy of the initial estimate. Fi- 

 nally, both the Sibert et al. and the PSAT Tracker al- 

 gorithms yield a solution that provides a best fit to the 

 time series of satellite (in the case of PSAT Tracker) 

 and tag measurements to the model of fish behavior. 



Unfortunately, it is difficult to make a general assess- 

 ment of the accuracy of either approach. In the case of 

 the PSAT Tracker algorithm, the accuracy of the track 

 will decrease in the absence of a north-south tempera- 

 ture gradient. We have not found a means of quantita- 

 tively determining the accuracy of PSAT Tracker cal- 

 culations. However, the quality of the fit between pixel 

 values of temperature from imagery and tag values for 

 positions along the track is calculated as a x 2 value. In 

 the case of the Sibert et al. algorithm, the accuracy of 

 the track will decrease during the period of the equinox 

 when the latitudinal errors of the light-based estimates 



are extremely large. Our data indicate that this period 

 can be as long as two months surrounding each equinox 

 (skewed towards winter). At such times the estimates of 

 position derived by the Sibert et al. algorithm depend 

 largely on the random walk model of fish movement, 

 which provides only a generic description of movement. 

 Although the algorithm provides values for the mean 

 square errors of bias and randomness for the tag es- 

 timates of latitude and longitude, these values are not 

 true values for error of predicting location; rather they 

 represent of the discrepancy between the estimates of 

 position by the random walk model, the formulation 

 of the latitude estimation error, and the tag measure- 

 ments. Additionally, the Sibert et al. algorithm does not 

 exclude the possibility of placing a fish on land. 



The PSAT Tracker worked well for this study because 

 of the strong north-to-south temperature gradient that is 

 presented in the northeastern Pacific. Studies conducted 

 in regions with poor temperature gradients will continue 

 to rely on light-based latitude estimates and approaches 

 like the Sibert et al. algorithm. Further development 

 of PSAT Tracker, or other SST-based geolocation al- 

 gorithms, should explore a means of using light-based 

 latitude positions in combination with SST matching 

 when light data are reliable, but excluding light-derived 

 latitude positions when they are unreliable. 



