KEMMERER ET AL.: ERTS-1 MENHADEN EXPERIMENT 



in attempts to capture schools because the 

 schools often inhabit waters too shallow for 

 efficient boat and net operations (negative 

 correlation associated with depth). Spotter pilots 

 tend to concentrate their fish-searching efforts 

 on turbid waters because of a relatively high 

 frequency of fish encounter in these waters 

 (negative correlation associated with secchi disc 

 transparency). The positive correlation associated 

 with chlorophyll a seems reasonable in that 

 menhaden are plankton feeders. Salinity is a 

 questionable concomitant factor although, 

 because these fish are euryhaline organisms and 

 inhabit estuarine waters throughout most of 

 their lives, a preferred association with waters of 

 low salinity seems plausible (negative correlation 

 associated with salinity). Christmas and Gunter 

 (1960) reported that 70% of the catch from 87 

 sets in the Mississippi Sound came from waters 

 ranging from 5 to 24 ppt salinity, suggesting also 

 a menhaden preference for low salinity waters. 

 No biological significance can be attached directly 

 to Forel-Ule color (negative correlation) yet, 

 although this color may manifest water trans- 

 parency and chlorophyll content. Correlation 

 coefficients between Forel-Ule color and secchi 

 disc transparency and chlorophyll a were -0.404 

 and 0.369, respectively, significant at the 99% 

 confidence level. 



The lack of statistical significance for several of 

 the parameters listed in Table 1 should not 

 necessarily be construed as meaning that no 

 such correlations exist. For example, surface 

 water temperature was relatively constant 

 spatially throughout the study period and there- 

 fore its effect, if any, on the distribution and abun- 

 dance of menhaden may not have been sufficient 

 to gain statistical significance. In the long run, 

 however, temperature may be a very important 

 parameter. One also should be reminded that the 

 correlations were developed from linear expres- 

 sions for the sake of statistical tractability. 

 The correlations, therefore, may not factually 

 represent real world situations where most 

 responses probably are nonlinear. 



The concern over a possible significant sensor 

 bias in the menhaden distribution estimates 

 prompted attempts to substantiate the results 

 through other approaches. The set of commercial 

 fishing data which included measurements of 

 selected oceanographic parameters provided the 

 only avenue through which substantiation could 

 be accomplished. However, these data were notice- 



ably biased in that environmental measurements 

 were obtained only from areas where catches were 

 made or attempted. In addition, the boats did not 

 fish randomly throughout the study area; rather, 

 they fished according to fish availability, distance 

 from home port (minimized to reduce operating 

 expense), day of the week (tendency to fish 

 farther from home port as the fishing week 

 progressed), and water depth (usually about 2 m 

 for efficient boat operation). Nevertheless, if 

 caution is used in the analysis, the data can be 

 used to substantiate some of the results gained 

 through photographic sensing of the menhaden 

 stocks. 



In the classical statistical situation, one gener- 

 ally attempts to differentiate between two pre- 

 sumably different populations, e.g., with and 

 without menhaden. As noted previously, the 

 principal problem with the commercial fishing 

 data is that data were not obtained from areas 

 without fish. However, if the assumption is 

 made that all other environmental measurements 

 collected throughout the study period (main and 

 secondary day events) were taken at random in 

 terms of temporal and spatial coverage, then it 

 is logical to assume that these latter measure- 

 ments included areas with and without menhaden. 

 The commercial fishing data can then be handled 

 as a "with fish" subset of the total data population, 

 i.e., with and without fish. 



The difficulty in this approach is that differences 

 are difficult to demonstrate with a high level of 

 statistical significance because the subset (with 

 fish) is not discrete from the total population 

 (with and without fish). The hypotheses which can 

 be tested are that the means (x) and standard 

 deviations (s) of the subset and total population 

 are different, resulting in the following four 

 general conditions and accompanying conclusions: 



1. Means and standard deviation are not 

 significantly different; conclusion: fish dis- 

 tribution is not related to the parameter 

 tested. 



2. Means are significantly different but stan- 

 dard deviations are not; conclusion: fish dis- 

 tribution is related to the parameter tested. 



3. Means are not significantly different but 

 standard deviations are; conclusion: fish dis- 

 tribution is related to the parameter tested. 



4. Means and standard deviations are both 

 significantly different; conclusion: fish dis- 

 tribution is related to the parameter tested. 



381 



