Itoh et al.: Migration patterns of Thunnus orientalis determined with archival tags 



517 



-1 1 n 



10Nov94 18Feb95 29 May 95 6 Sep 95 15Dec95 



_ 40 -, 



10 Nov 94 18 Feb 95 29 May 95 6 Sep 95 15 Dec 95 



Date 



Figure 1 



Seasonal change of errors in location estimate. Upper and lower panels 

 show longitudinal and latitudinal errors, respectively. Triangles show 

 equinoxes. Data are from an archival tag implanted in a fish held in a 

 pen for 453 days. 



formation about latitude. The tag did not provide a latitude 

 estimate for 18 days around and at the vernal (autumn) 

 equinox and had large errors for one month before (or after) 

 as well as 10 days after (or before) that period, respectively. 

 The same pattern was observed in the test with archival 

 tags that were left in air. In addition, the latitude estimates 

 were biased toward south in summer and toward north in 

 winter, that is, toward erroneously short day lengths. 



Occasional large deviations were observed in both lati- 

 tude and longitude estimates. These were easily identified 

 as outliers in our analyses of data obtained from wild fish 

 by comparing them with estimated locations for adjacent 

 days. When evaluating the accuracy of location estimates 

 for practical use in analyses of wild fish movements, we 

 excluded longitude or latitude estimates that differed more 

 than 10° from the real location and the latitude estimates 

 not provided by the tag near the equinoxes. These account- 

 ed for 2.8% of longitude estimates and 8.9% of latitude 

 estimates obtained in the tests in air, as well as 4.8% of 

 longitude data and 47.5% of latitude data obtained in the 

 tests of pen-held fish. 



Figure 2 shows the position estimates and error bars 

 corresponding to one standard deviation for 1 17 tags tested 

 in air — most of them for a .5-day period, five for 82 days, 

 and twelve others for various intermediate durations. The 

 aggregate of all observations in air yields an error estimate 

 (mean ±standard deviation) of -0.54° ±0.75° for longitude, 

 and -0.12° ±3.06° for latitude. 



When individual tags tested in air were examined 

 separately, 96%- of tags (112/117) showed average posi- 

 tion errors within a range of ±1.5° in longitude. Among 

 these 112 tags with small longitude errors, 95 had been 

 manufactured within the last half year and had an av- 

 erage and standard deviation of position error equal to 

 -0.50° ±0.19°, and the other 17 tags were more than 

 one year old and had an average position error of -0.51 

 ±0.75°. The average is not significantly different (ANOVA 

 F=0.01,P>0.05) and the younger tags had a smaller stan- 

 dard deviation (F=412, P<0.01). No significant difference 

 of accuracy was observed among the 17 older tags that 

 could be related to their history, i.e. among four tags kept 

 in air without release and 13 tags released with fish and 

 recovered (F=1.01 for average and F=2.81 for standard 

 deviation, both P>0.05). 



For the two tags attached to fish in pens, one tag mea- 

 sured only five positions with a resulting error estimate of 

 -2.38 ±0.39 for longitude, and -1.82° ±1.58° for latitude. 

 The other measured 432 positions, with a resulting error 

 estimate of -0.53° ±2.46° for longitude and 1.26° ±5.33° 

 for latitude. This is the data series presented earlier in 

 Figure 1. The large standard deviation in longitude error — 

 much larger than that obtained in other tests — initially 

 raised questions regarding the effect of water on the posi- 

 tioning techniques. However as mentioned earlier, the low 

 light sensitivity of both tags used in captive fish was identi- 

 fied as the likely cause of these large errors. 



