POLACHECK: DISTRIBUTION OF SEARCHING EFFORT 



by a value of 50 for the clustering parameter and 

 from 196 to 691 miles for a value of 150 miles 

 (Table 1). The frequency distribution of these in- 

 tercluster distances (Fig. 7) is an indication of the 

 stability of the clusters to the value of the cluster- 

 ing parameter. Thus, for a value of 100, over 65% 

 of the clusters have an intercluster distance 

 exceeding 175 miles. This suggests that 65% of 

 clusters will be stable up to a value of 175 miles 

 for the clustering parameter. (Note, this is not 

 strictly true. If the set preceding the first mem- 

 ber of a cluster was less than 175 miles away and 

 this set was also less than 175 miles from the first 

 set of the next cluster, this set plus these two 

 clusters would be combined in a single cluster 

 for a value of the cluster parameter less than 

 175.) 



These statistics describing the characteristics 

 of the defined clusters suggest that the algorithm 

 used to create them successfully separates the ac- 

 tivities of a cruise into areas where sets are com- 

 mon and areas where they are infrequent. The 

 major differences in the clusters with different 

 values for the clustering parameters result from 

 the merging of two relatively close clusters or the 

 inclusion of an isolated set or chase near the 

 boundary of a cluster (e.g., for 80% of the cruises, 

 the actual number of clusters decreases or re- 

 mains the same over a range of 50-150 miles 



for the clustering parameter). However, the fact 

 that many of these descriptive statistics vary 

 continuously with the value of the clustering 

 parameter suggests that these defined clusters 

 do not represent distinct units, but areas of 

 high concentration in a continuously grading 

 system. 



Cruises vary greatly with respect to the 

 amount of variability they exhibit in response to 

 changes in the value of the clustering parameter 

 (Table 2). Such variability is to be expected since 

 no single searching strategy is used by all vessels 

 and vessels may change their strategy during the 

 course of a cruise. In addition, the spatial distri- 

 bution of potential sets probably also varies with 

 time and space. These sources of variability 

 among cruises, combined with the relatively 

 small sample sizes within a cruise, may be part of 

 the reason that the descriptive statistics charac- 

 terizing clusters vary continuously with the value 

 of the clustering parameter. 



The lack of any sharp demarcation in the clus- 

 ters as a function of the clustering parameter, 

 combined with the large amount of variability 

 exhibited among different cruises, creates a prob- 

 lem in presenting results based on the clustering 

 algorithm. Consequently, whenever summary 

 statistics are presented, results are given for a 

 range of values for the clustering parameter. 



Table 2. — Examples of the effect of changes in the value of the clustering parameter for 5 arbitrarily selected 



cruises. 



359 



