Dressel and Norcross: Using poststrafication to improve abundance estimates from multispecies surveys 



485 



design, developing an index of relative abundance from 

 all samples, or samples in the habitat or HFD areas, 

 is an easy and effective way to estimate statistically 

 significant changes in abundance among years. To de- 

 termine which tows should be included in an index to 

 effectively approximate the variations in the annual 

 total abundance estimates, it is helpful to compare the 

 size of the habitat area over years and to study the dis- 

 tribution of species density within the habitat area. The 

 goal of creating an index should be to include the most 

 information possible, while avoiding undue influence 

 from the haphazard distribution of sample sites. 



If the total study area is the same in each year, the 

 choice of whether to use the all-site index should depend 

 on whether the size of the habitat area is constant over 

 the compared years. In this study, the defined habitat 

 area for each species was the same over the six years 

 compared. Therefore, for an index of relative abundance, 

 the habitat index retained all necessary information 

 and reduced possible bias due to the disproportionate 

 distribution of haphazard samples between habitat and 

 nonhabitat areas. When a temporally dependent strati- 

 fication variable, such as temperature, is used to define 

 the placement of stratum boundaries, however, the size 

 of the habitat area may vary between years. If the an- 

 nual size of the habitat area varies, some common size 

 would need to be chosen for the relative index to ap- 

 proximate the annual changes in the total abundance 

 estimates. The all-site index could be used for this 

 purpose, but the index will be affected by any dispro- 

 portionate distribution of samples between habitat and 

 nonhabitat areas. Another possible way to do this would 

 be to include all tows from the habitat area each year, 

 plus as many zero catches from the nonhabitat area 

 necessary to be proportional to the annual size of the 

 nonhabitat area. Such an approach would not depend 

 on actual tows in nonhabitat area but would depend on 

 the estimated size of the habitat and nonhabitat areas 

 and the sample size in the habitat area. 



If the size of habitat area is the same in each year, 

 the choice of whether to use the habitat index should 

 depend on whether the distribution of species density 

 is constant throughout the habitat area. If a species' 

 density distribution is approximately constant across 

 the habitat area, a haphazard distribution of sample 

 sites should have little influence. Constructing an index 

 from all habitat tows may then be desired to retain the 

 largest sample size and the most information possible. 

 Alternatively, if a species has a strong density gradient 

 within its habitat area, a disproportionate distribution 

 of sites in relation to the size of high and low fish-den- 

 sity areas may provide an unrepresentative estimate of 

 abundance from the habitat index. In this case, if a suf- 

 ficient number of samples are taken in the HFD area, 

 constructing an index from samples within the species' 

 HFD area alone may provide an effective index while 

 minimizing the effect of a disproportional distribution 

 of haphazard samples within the habitat area. 



A comparison of the number of zero catches and the 

 mean nonzero catch between the high and low fish- 



density areas provides information about the density 

 distribution of species within a habitat area. The pro- 

 portion of zero catches of rock sole, yellowfin sole, and 

 flathead sole and the mean nonzero catch between high 

 and low fish-density areas indicated density gradients 

 within the habitat areas. Unlike these three species, 

 the proportion of Pacific halibut zero catches was ap- 

 proximately the same in the HFD area as across the 

 entire habitat area and the difference in mean nonzero 

 catch between low and high fish-density areas was only 

 approximately half that of the other species. Therefore, 

 it appears that the Pacific halibut density distribution 

 across the defined habitat area varied little compared 

 with the other three species. 



In this study, we suggest that the habitat index was 

 the most appropriate for all four species. For each spe- 

 cies in our study, the size of the habitat area remained 

 the same across all six years. Thus, the habitat index 

 eliminated the influence of disproportionately allocated 

 samples in habitat and nonhabitat areas. For Pacific 

 halibut, the relatively homogenous distribution of abun- 

 dance across the habitat area indicates that the effect 

 of disproportionate samples between high and low fish- 

 density areas is small and that samples across the 

 entire habitat area are helpful in describing annual 

 differences in abundance. For rock sole, yellowfin sole, 

 and flathead sole, the difference in the proportion of 

 zero catches and nonzero mean abundance between the 

 high and low fish-density areas was considerable. As a 

 result, differences in annual abundance suggested by 

 the habitat index may be affected by the inconsistent 

 disproportion of samples between high and low fish-den- 

 sity areas over years. Although it would be preferable to 

 use the HFD index in these cases, annual sample sizes 

 in the HFD area were so small that we recommend the 

 habitat index instead. Recognizing that the habitat 

 index will not account for the annual disproportion of 

 samples between the high and low fish-density areas, 

 we used the comparison of the size and the number of 

 samples taken in high and low fish-density areas to flag 

 differences in annual index abundance estimates that 

 might be over- or underestimates. If this method is ap- 

 plied in a management context, the levels of the factors 

 describing the density distribution of the species (i.e., 

 difference in the percent of zero catches and the percent 

 difference in mean nonzero catch between years) can be 

 set as criteria and kept constant over years to elimi- 

 nate subjectivity between years or between species. For 

 example, if the percent of zero catches in high and low 

 fish-density regions differ by 40% and the mean nonzero 

 catch in the HFD area is 30% greater than that in the 

 LFD area, the HFD index should be used. Otherwise, 

 the habitat index should be used. 



For many surveys, identifying habitat and fish-density 

 areas for poststratification and index construction is pos- 

 sible with currently available information. The estima- 

 tion methods used in the present study can be applied 

 to any survey for which abundance and environmental 

 measurements are available for each sampled site and 

 the environmental measurements are related to species 



