KEMMERER ET AL.: ERTS- 1 MENHADEN EXPERIMENT 



different for each model even though the distribu- 

 tions overlap without a wide margin of difference 

 between means (Table 5). 



A number of factors probably contributed to the 

 failure of the models to group fishing data closer to 

 1. It should be pointed out first, however, that no 

 seasonally caused variation in products was 

 noted, suggesting that the nonparametric group- 

 ing was caused by factors prevalent throughout 

 the June through September commercial fishing 

 sampling period. One of these factors may have 

 been the effect of commercial fishing operations on 

 the distribution offish as evidenced by visual ob- 

 servations made during the photographic surveys 

 of the study area. Menhaden schools frequently 

 were observed being chased by purse boats 

 through waters of varying visual qualities (i.e., 

 turbidity). In addition, oceanographic parameter 

 measurements generally were taken from the 

 mother vessel rather than the purse boats, which 

 often was several kilometers distant from the ac- 

 tual site offish capture. Another of these factors is 

 that there is no biological reason to suspect 

 menhaden distribution to be wholly a determinis- 

 tic function of environmental conditions; rather, 

 there most likely is a probability associated with 

 how and where fish are distributed in response to 

 these conditions. Also, there were errors as- 

 sociated with all of the parameter measurements 

 used to develop and test the models as well as a 

 distinct possibility that other parameters having 

 a direct influence on menhaden distribution might 

 not have been measured (e.g., zooplankton 

 biomass, presence or absence of predators, oxygen 

 tensions, etc.). And finally, there is the linear ad- 

 ditive nature of the models which at best probably 

 only approximates the real world situation. 



Selection of a best model was difficult in that 

 they all provide similar products. On the basis of 

 sample size, number of parameters (minimum), 

 and difference between means (Table 5), Model D8 

 would have to be given selection priority, how- 

 ever. 



A number of interpretations and presentation 

 methods can be applied to model products as long 

 as they recognize the imprecision of the models. 

 An example of one method applied to Model D8, for 

 7 August 1972 sea-truth data, is presented in 

 Figure 5. The categorization of model products 

 was done by dividing the values shown in Figure 4 

 for Model D8 into three ranges based upon a direct 

 comparison of fishing and nonfishing histograms: 



high potential = > 0.2 



moderate potential = -1.0 to 0.2 

 low potential = <-1.0 



The interpretation applied to high, moderate, and 

 low potential areas is related to relative probabil- 

 ity. In high potential areas, the probability offish 

 capture is higher than in moderate or low poten- 

 tial areas and higher in moderate than in low 

 potential areas. Incomplete commercial fishing 

 reports from 7 August 1972 indicate that most, if 

 not all, fishing occurred in the high potential 

 areas. 



An additional analysis was performed on the 

 commercial fishing data to determine if relation- 

 ships could be demonstrated between catch size 

 and the four oceanographic parameters which 

 made up the models. Catch size ranged from 5 to 

 225 and averaged about 38 thousand fish. Catch 



Table 5. — Tests of empirical models played with oceanographic data taken near sites of 

 commercial fish capture (with fish) and during main day events, the latter stratified to include 

 only those areas where fish were not detected photographically (without fish). 



'Coefficient of variation. 



'r-test for populations with unequal variances (Ostle, 1963). 



387 



