Domeier et al.: Tracking Thunnus thynnus onentalis with the aid of an automated algorithm 



295 



(2001) that light-based longitude estimates have a year 

 round constant error of ±0.32 degrees. 



Satellite imagery, temperature sensors, and land mask 



The PSAT Tracker code provides an interface to auto- 

 matically download, georeference, and display SST imag- 

 ery. As many as three different types of imagery can be 

 layered and prioritized to produce a collage of imagery 

 for processing and display. Higher priority layers are 

 searched first for SST matches before "drilling down" 

 to lower layers. The sources and types of available SST 

 data are numerous and have varied over the time frame 

 of this study; different sensors and algorithms produced 

 data of differing spatial and temporal resolution or accu- 

 racy (Table 1). To maximize the quality of the latitude 

 estimates produced by the PSAT Tracker algorithm, we 

 substituted better SST data as it became available. For 

 this study SST imagery was prioritized as follows: 1) 

 advanced very high resolution radiometer (AVHRR) or 

 moderate resolution spectroradiometer (MODIS) daily 

 data, 2) AVHRR or MODIS weekly data, and 3) multi- 

 channel sea surface temperature algorithm (MCSST) 

 weekly data. The MCSST algorithm is a weekly (or 8- 

 day) composite that is most helpful in analyzing regions 

 of frequent cloud cover; this algorithm was applied by the 

 University of Miami (Miami) from 1981 through Febru- 

 ary 2000 and has been applied by the Naval Oceano- 

 graphic Office (NAVOCEANO) since September 2001. 

 The MCSST algorithm provides a near complete picture 

 of SST data for the study area; although AVHRR and 

 MODIS data are higher resolution and more accurate. 



The difference in the resolution and accuracy of tem- 

 perature sensors on the tags verses those on the satel- 

 lites (Table 1) are worth mentioning. The accuracy of 

 the satellite SST data, particularly for MCSST/NAV- 

 OCEANO. is the limiting factor when attempting to 

 match tag data to satellite data. The degree to which 

 the satellite data and tag data must match can be set 

 by the user in PSAT Tracker; for this study it was set 



between the limit of MODIS and NAVOCEANO resolu- 

 tion (0.4°C). 



There is a fourth layer that is superimposed upon the 

 imagery. This is a land mask that is used to eliminate 

 placing a tag on land and to insure that tags move 

 around land barriers rather than across them. 



Computation of the track 



A detailed mathematical description of the computation 

 for the best track would take more space than is avail- 

 able. Instead, we present a more general description of 

 the algorithm and its logic, consisting of the following 

 five steps that are summarized below and then subse- 

 quently described in detail. 



1 Define the daily search area found within satellite 

 SST imagery. 



2 Define appropriate tag data (termed selection set) to 

 match to satellite SST values found within the daily 

 search area. 



3 Select candidate points within each daily search 

 area that provide the best match to the temperatures 

 found in the selection set. The cost of each candidate 

 point is largely determined by the difference between 

 the tag and satellite SST values. 



4 Calculate the cost for all possible steps, called arcs, 

 between pairs of candidate points of adjacent daily 

 search areas. The cost of each step is a function of 

 the length of the arc that connects adjacent candidate 

 points (the greater the distance, the greater the cost) 

 and the cost of each individual candidate point (see 

 step 3). 



5 Sum the costs of all tracks and identifying the track 

 with the lowest cost. 



Step I: Defining the daily search area A daily search 

 area is defined by the tag manufacturers' light-based 

 solution for longitude, a user defined bracket for lati- 

 tude and the value entered for maximum swimming 



