Gannon and Gannon: Passive acoustic assessment of soniferous fish density 
109 
normal distribution in a Q-Q plot than did the raw data. 
To investigate whether acoustic variables were related 
to local densities of Atlantic croaker or other environ- 
mental variables, we performed a MANCOVA with the 
two continuous acoustic parameters (received sound level 
and peak frequency) as dependent variables, with habitat 
and month as factors, and with CPUE, dissolved oxygen, 
and temperature as covariates. MANCOVA was run in 
SPSS vers. 16 (SPSS, Inc., Chicago, IL). 
We used Williamson’s index of spatial overlap (Wil- 
liamson, 1993) to characterize the extent to which At- 
lantic croaker calling was spatially correlated with At- 
lantic croaker density. Williamson’s index is customarily 
used to measure spatial overlap of the distribution of 
two species, such as that between a predator and its 
prey (Williamson, 1993; Garrison, 2000; Garrison et 
al., 2002; Link and Garrison, 2002). Here we used it 
to compare two independent measures of distribution 
for the same species. The index (0 ;/ ) is calculated as 
follows: 
m 
2=1 
m m 
2=1 2=1 
where m 
z 
N, 
the number of samples; 
a discrete sampling location; 
the relative abundance of Atlantic croaker 
as determined by trawl CPUE; and 
the relative abundance of Atlantic croaker 
as determined by passive acoustic methods 
(either calling index or relative received 
amplitude of calls). 
Index values >1 indicate spatial overlap greater than 
would be expected by chance, and values <1 reflect less 
spatial overlap than expected. 
We applied Williamson’s index to compare 1) CPUE 
to calling index and 2) CPUE to received sound level. 
We determined the significance of 0 ;/ using a random- 
ization procedure developed by Garrison (2000). The 
test statistic 0 ;/ was compared to a random distribu- 
tion of overlap values in which each value of N was 
randomly paired with a value for n t to calculate a 
randomized O tJ . This was done 4999 times, and the ob- 
served test statistic O (the 5000 th instance) was then 
compared to the generated distribution. Significance of 
the value was judged by the proportion of randomized 
O rj values that were greater than the observed 0 ;; (a 
value was judged to be significant if fewer than 5% of 
the randomized values were greater than 0 (/ ). 
Classification and regression tree (CART) analysis 
was used to explore relationships between the depen- 
dent variable CPUE and eight independent variables 
(day of year, dissolved oxygen concentration, tempera- 
ture, salinity, depth, calling index, received sound level, 
and distance to nearest creek). The CART method tests 
the global null hypothesis of independence between 
the dependent variable and each of the independent 
variables and identifies critical threshold values for 
the significant independent variables (Urban, 2002). 
Our study area was spatially structured with regard to 
habitat type (i.e., the mid-river habitat was a contiguous 
region at the center of the study area, the creeks were 
at the outer edges of the study area, and the mid-river 
habitat was a contiguous ring separating the mid-river 
from the creeks). Thus, the three habitat types were 
spatially correlated. To avoid multicollinearity in our 
analyses, we used a single variable (“distance to creek”) 
to represent the position of each sampling station in 
relation to each of the three habitats (i.e., any station 
close to the creeks was necessarily far from the mid- 
river and vice-versa). ArcGIS vers. 9.2 (Environmental 
Systems Research Institute, Redlands, CA) was used 
to calculate the distance from each sampling location 
to the creek habitat. We ran the CART analysis using 
the “Party” library in the R software environment 
(vers. 2.6.1, R Development Core Team; Hothorn et al., 
2006). P-values were calculated by using a quadratic 
test statistic. 
Results 
From June to October of 2000, we performed 14 simul- 
taneous trawl and passive acoustic surveys, for a total 
of 224 paired samples. Sixty four samples were excluded 
from analysis because of a high sea state (Beaufort 
scale), rain, boat noise, or the presence of dolphins. 
Noise from rough sea conditions, rain, and passing boats 
may not have affected the calling behavior of Atlantic 
croaker, but these noise sources masked the croakers’ 
calls, making it difficult to reliably measure their acous- 
tic characteristics. Recordings made in the presence of 
bottlenose dolphins appeared to have long periods of 
silence punctuated by short, sporadic outbursts of croaker 
calls. Therefore we excluded these samples because 
there appeared to be a change in Atlantic croaker call- 
ing behavior when dolphins were present. Exclusion of 
these 64 samples left a total sample size of 160; 70 in 
the tributary creeks, 53 along the river edge, and 37 in 
middle of the river (rough sea conditions and boat noise 
were most prevalent in the mid-river habitat). All Atlan- 
tic croaker caught in the study were young of the year. 
CPUEs were highest in all habitats during June 
(Fig. 2A) and the general trend was for declining densi- 
ties throughout the five-month study, likely the result 
of a large pulse of recruitment in spring, followed by 
mortality or emigration throughout summer and fall. 
Atlantic croaker densities were lowest in the mid- 
river habitat during July and August, coinciding with 
the occurrence of hypoxic conditions. Atlantic croaker 
recolonized the mid-river habitat when dissolved oxy- 
gen conditions improved in September and October. 
Standard lengths of croaker increased throughout the 
period from June to October in all habitats but were 
consistently lower in the mid-river habitat than in the 
other habitats (Fig 2B). Calling indices, representing 
