362 
Fishery Bulletin 107(3) 
Table 1 
Location-accuracy classes for marine mammal location estimates obtained with satellite telemetry and assigned by Service 
Argos, and reported in other studies. Also included are the proportions of the final, filtered harbor seal (Phoca vitulina richar- 
dii) location data set from the present study that fell into each location class. Vincent et al.(2002) calculated accuracy for tags 
deployed on captive grey seals (Halichoerus grypus) in France and calculated accuracy separately for the latitude and longitude of 
position estimates. Hays et al. (2001) calculated accuracy using tags located in fixed positions in Brazil and on Ascension Island, 
as mean deviation from true tag location. 
Location 
accuracy 
class 
Proportion of 
total locations 
used in this study 
Service Argos 
accuracy estimate (m) 
Vincent et al. 2002 
(unfiltered; lat./long.) (m) 
Hays et al. 2001 
(mean deviation from true) (m) 
3 
0.20 
<150 
157/295 
270 
2 
0.25 
<350 
259/485 
540 
1 
0.21 
<1000 
494/1021 
1330 
0 
0 
>1000 
2271/3308 
10,100 
A 
0.34 
None assigned 
762/1244 
990 
B 
0 
None assigned 
4596/7214 
7000 
where tow area = distance towed (in meters) x door 
spread of tow (3.42 m); and 
MWT CPUE = 
( number caught/tow volume) x 10,000, (2) 
where tow volume = number of flowmeter revolutions 
x 0.0269 m/rev x net mouth area 
(10.7 m 2 in this case). 
Crab / shrimp CPUE = 
number caught per 5 minute tow, (3) 
CPUE from the 39 sampling stations was used to create 
maps of the relative abundance of harbor seal prey spe- 
cies in SFB, by using the inverse distance weighting 
interpolation method (Geostatistical Analyst extension 
to ArcGIS 9.2, ESRI, Redlands, CA). Inverse distance 
weighting is a deterministic interpolation method and 
makes no assumptions about the input data; this was 
important given the patchy nature of fish distributions 
in SFB. Given seasonal differences in prey species’ 
abundance and distribution, and in harbor seal behav- 
ior related to breeding and molting, we created four 
maps for each prey species, one for each harbor seal 
“season” (spring: March-May; summer: June-August; 
fall: September-November; winter: December-Febru- 
ary). In SFB, harbor seals pup during the spring and 
molt during the summer. Only records for those months 
and years when we had tagged harbor seals active in 
SFB waters were included in the analyses. 
Using the Hawth’s analysis tools extension (avail- 
able online at http://www.spatialecology.com/htools) 
for ArcGIS 9.2 (ESRI, Redlands, CA), a 1-km grid was 
laid over a map of the entire study area, consisting of 
all waters from the mouth of SFB, to the eastern edge 
of Suisun Bay (Fig. 1). All harbor seal locations and 
environmental data sets in the GIS were reprojected 
to Universal Transverse Mercator (UTM) coordinates, 
using the North American Datum of 1927 (NAD 27), 
zone ION, and resampled to an initial grid resolution of 
1-km. For each season, an average CPUE of each prey 
species was assigned to each 1-km grid cell, by using 
the area-weighted mean of the values falling within 
that grid cell. In addition, we counted the number of 
harbor seal locations falling within each grid cell; be- 
cause the number of tagged animals was limited, data 
from individual harbor seals were pooled for this analy- 
sis (see Erickson et al., 2001). The minimum scale of 
analysis was 1 km 2 , well within the estimated average 
accuracy of the filtered harbor seal location data (Table 
1; see also Bekkby et al., 2002). 
To vary the scale of analysis, data from the 1-km grid 
cells were combined into progressively larger grid cell 
sizes, ranging from 2 to 10 km. Given the size of SFB 
(and the fact that the sample size decreased with each 
successively larger grouping), we did not consider scales 
larger than 10 km. Because of the irregular shoreline 
of SFB, some grid cells overlapped land; therefore, we 
removed grid cells that represented primarily land from 
the analyses. For all remaining grid cells, we calculated 
the number of harbor seal locations per km 2 . 
For each spatial scale (1 to 10 km) and each season, 
we calculated the Pearson’s correlation coefficient be- 
tween the number of harbor seal locations per grid cell 
and the CPUE for each potential prey species in that 
cell. We plotted correlation coefficients versus scale 
for each season to assess the effects of scale on the 
strength of the spatial relationships between harbor 
seals and potential prey. 
To estimate the availability of foraging habitat during 
each season, we used regression tree analyses (Breiman 
et al., 1984) to identify threshold values of prey CPUEs 
that would most strongly differentiate between grid cells 
with greater use by harbor seals and cells with lesser 
use by harbor seals, for each season. In other words, 
this threshold value indicated the minimum CPUE for 
prey in the grid cells representing areas that were fre- 
