FISHERY BULLETIN: VOL 76. NO. 3 



Most of the chlorophyll work in the central 

 Pacific has been based upon 12 or more sampled 

 depths. Thus, it was encouraging to find that as 

 few as 10 depths, regardless of the sampling 

 strategy, gave a generally satisfactory picture of 

 the amount of chlorophyll in the water column. Of 

 nearly 400 estimates from individual casts, IGOi 

 fell within ±109? of the true value. This percent 

 increased to 859? for stratified designs SR-1 and 

 SR-3 and to 949r for the 36 systematic casts. How- 

 ever, to the extent that these fluorescence profiles 

 underestimate the structural complexity of the 

 true chlorophyll distribution, these results proba- 

 bly overestimate the accuracies of the designs. 



Study B 



The results of the field study were remarkably 

 similar to those of the computer study (Table 4). 

 Bias and accuracy were investigated by assuming 

 that the entire population was exactly represented 

 by the 20 samples in one cast (systematic samples 

 plus stratified random samples plus "free" sam- 

 ples). The results of study A indicate that the dis- 

 crepancy is not likely to be severe. 



Table 4. — Results of study B, a field sampling study. The esti- 

 mated parameter is total chlorophyll above 180 m and the true 

 value ( 6) is estimated from the 20 combined samples of the two 

 designs; units are milligrams per cubic meter. 



indicates pair of replicate casts. 



Acciiracx 



Bias 



When the 18 casts are considered in chronologi- 

 cal sequence, it is evident that RSS tended to de- 

 viate from the true value in the same direction on 

 adjacent casts. The direction of bias was the same 

 within five of the six pairs of replicate casts 

 (0.05<P<0.10) and a run test over the entire 

 sequence was significant (P = 0.05). The absolute 

 magnitude of the bias which would result were the 

 five replicate pairs considered estimates of five 

 population totals ranged from 0.39r to3.7'~'f with a 

 mean of 2.09^ . Within the restricted spatial area of 

 this survey, between 2 and 8 days appear neces- 

 sary for the natural fluctuations of the population 

 to be sufficient to average out bias inherent in 

 RSS. 



Precision 



Precision was investigated by means of the six 

 pairs of replicate casts. Stratified random design 

 SR-3 with one sample per strata was more precise 

 than the design with two samples per strata 

 */3.3 - 6.5, P<0.10). The precision of the system- 

 atic design was intermediate and was not sig- 

 nificantly different from either. 



The accuracies of the two stratified random de- 

 signs, as measured against the total chlorophyll 

 estimated from all 20 samples, were similar, but 

 the systematic design RSS-1 was significantly 

 more accurate than either (signed rank test, 

 P<0.05). Possibly the greater accuracy of the sys- 

 tematic designs, seen in both studies, might be 

 partially attributable to the different arithmetic 

 formulae with which total chlorophyll was calcu- 

 lated since these give somewhat different weights 

 to the individual samples. To test this, the esti- 

 mates from both the restricted systematic and the 

 stratified random designs were calculated by 

 linear integration. This did not alter the relative 

 accuracies of RSS-1 and SR-3 in study B or of 

 RSS-1 and SR-1 in study A, nor did it eliminate the 

 bias apparent in RSS-1. Thus, it appears to be the 

 sample location rather than the formula by which 

 total chlorophyll is calculated that is responsible 

 for the greater accuracies of the systematic de- 

 signs. 



DISCUSSION 



These studies indicate that there is a potential 

 for biased estimates to be derived from systematic 

 samples collected from a planktonic ecosystem. 



624 



