Olson and Galvan Magaha: Food habits and consumption rates of Co/yphaena hippurus 



281 



levels of taxonnmic resolution. To facilitate analysis, 

 we also grouped the prey laxa by order (e.g. Tetra- 

 odontiformes), family (e.g. Carangidae), genus (e.g. 

 Auxis spp. ), or functional group (e.g. flyingfishes). 



Because the three diet indices provide different in- 

 sights into predation habits, we applied a graphical 

 representation of these measures, proposed by Cortes 

 (1997), to help interpret the data. We made three-di- 

 mensional scatter plots of '/i O, "^ W, and '7(N for all 

 samples and for the data pooled by sampling area to 

 help evaluate the degree of dominance of particular 

 prey and the feeding strategy (generalized vs. special- 

 ized) of the dolphinfish. Although we measured the 

 three components of the index of relative importance 

 (IRI) (Pinkas et al., 1971), we did not calculate IRI 

 values because the index is dependent upon the taxo- 

 nomic resolution of the prey (Hansson, 1998). Also, for 

 a predator that consumes a large size range of prey 

 (see heading "Prey size," below), the IRI is overly in- 

 fluenced by numerous small prey. 



We examined diel feeding characteristics by stratify- 

 ing the data according to stomach fullness of the pred- 

 ator and digestion state of the prey. The scheme for 

 grouping the data, patterned after Calliet ( 1976 ), is dia- 

 grammed in Figure 2. Prey in digestion states 1 and 2 were 

 categorized as from "recent" feeding events, whereas prey 

 in states 3 and 4 were categorized as from "previous" feed- 

 ings. These two strata were further subdivided according to 

 stomach fullness. Prey from stomachs <50'7f full were cate- 

 gorized as "low" fullness or empty, whereas those from stom- 

 achs >50'7f full were categorized as "high" fullness (Fig. 2). 

 We plotted the percent occurrence of the prey items in these 

 tour digestion and fullness strata by area and the time of 

 day the sets were made: "early morning" (05:12-09:00), "late 

 morning"(09:01-12:00), "early afternoon"( 12:01-15:00), and 

 "late afternoon" (15:01-18: 16 hours). 



We fitted regression trees (Breiman et al., 1984) to the 

 gravimetric data for each prey group to detect statisti- 

 cally important differences by area and dolphinfish size. 

 Regression trees are well suited for detecting and extract- 

 ing important relations and complex interactions in multi- 

 variate ecological (Death and Fabricius, 2000) and fisher- 

 ies data (Walters and Deriso, 20001. We used a two-step 

 process. For both steps, the %W of each prey group in 

 the stomach contents was the response variable. For the 

 first step, defining area strata (see next paragraph), we 

 used latitude and longitude as the predictor variables. For 

 the second step, modeling the importance of area and dol- 

 phinfish size in explaining variation in the %W for each 

 prey group, we used area designations (north, west, east, 

 southwest, and southeast) and fork length as the predictor 

 variables. We used the tree functions in S-Plus (MathSoft 

 Inc., 1999) and cross-validation to prune fully grown trees 

 so that only important splits remained. Prediction errors 

 were used as pruning criteria (Breiman et al., 1984; De'ath 

 and Fabricius, 2000). 



We stratified the data by area (Fig. 1) according to two 

 criteria. Latitude divisions at 15°N and 0° were based on 

 the spatial and seasonal heterogeneity of the purse-seine 

 sets that provided the samples. All the sets sampled from 



3&4 



1 &2 



0-50° 



51-100° 



Stomach (ullness 



Figure 2 



Schematic diagram showing prey digestion-state and predator 

 stomach-fullness criteria for four categories used to analyze diel 

 feeding characteristics of common dolphinfish in the eastern 

 Pacific Ocean. 



May through November each year were made north of the 

 equator, and most sets sampled during December through 

 April were made south of the equator. Also, the regression 

 trees indicated that latitude and longitude were important 

 in explaining the variability in the gravimetric data for 

 several prey taxa. Epipelagic-cephalopod taxa were most 

 important in the diet of the common dolphinfish caught 

 east of 82°41'W and south of 1° 46'S, and flyingfishes were 

 most important in the diet of those caught west of 81°W. 

 Therefore, we stratified the data from south of the equa- 

 tor into "southwest" and "southeast" areas separated at 

 82°41'W (Fig. 1). Similarly, we stratified the data from 

 samples collected between 0° and 15°N into "west" and 

 "east" areas divided at 111°W because a regression tree fit- 

 ted to the gravimetric data for the Tetraodontiformes indi- 

 cated that this meridian was important in explaining vari- 

 ation in '7(W for this taxon. Tetraodontiformes were more 

 important in the diet of the fish caught west of 111°W. 



Consumption rates 



We employed a method described by Olson and Mullen 

 (1986) to calculate preliminary estimates of daily rates of 

 food consumption by common dolphinfish. The model pre- 

 dicts feeding rate ( r, grams per hour) by dividing the mean 

 weight of the stomach contents per predator (W, grams) 

 by the integral (A. proportion x hours=hours) of the func- 

 tion that best fits experimental gastric evacuation data. 

 For a predator that consumes a variety of prey that are 

 evacuated at different rates. 



,=0 ^ 



where subscripts / refer to each of/ prey types. 



(1) 



