![[logo] US EPA](http://www.epa.gov/epafiles/images/logo_epaseal.gif){kind=logo}
Premise 19 - Sampling protocols
Click below to view some of the premises from Karr and Chu (1999).
Sampling protocols are well defined for fishes and invertebrates
FROM "Restoring Life in Running Waters" by James R. Karr and Ellen W. Chu
(Reprinted with permission from Island Press)
The utility of any measure of biological condition in a stream depends on how accurately the original sample represents the fauna present in that stream--that is, how successful it is in avoiding statistical "bias." Indeed, a fundamental assumption of the fish IBI is that the sample on which it is based reflects the taxa richness and relative abundances of the stream's fauna, without bias toward taxa or size (Karr et al. 1986). Implicit in this assumption is that sampling effort is standardized. Any fish-sampling protocol must therefore be consistent, comprehensive, and representative of the stream's microhabitats, including pools, riffles, margins, and side channels. A study of fish of the Muese River in Belgium reinforced this point, revealing that sampling multiple mesohabitats (rifles, pools, and runs) was needed to accurately evaluate river quality (Didier and Kestemont 1996).
Many researchers have helped to refine the protocols for sampling fish to evaluate or implement an IBI (Ohio EPA 1988; Lyons 1992a,b; Lyons et al. 1995; Lyons et al. 1996). Other protocols for sampling fish and invertebrates have also been described, although their goals and applications vary somewhat from development of an IBI (Klemm et al. 1990, 1993, for USEPA's Environmental Monitoring and Assessment Program [EMAP]; Cuffney et al. 1993 and Meador et al. 1993 for US Geological Survey's National Water Quality Assessment [NAWQA]).
Early work on the fish IBI identified sampling gear, the range of microhabitats in a stream, and stream size as important factors affecting sampling accuracy (Karr et al. 1986; Ohio EPA 1988). These researchers showed that, with standard procedures, it is feasible to sample virtually all fish from all microhabitats in small- to medium-size streams. Boat-mounted electrofishing gear is the most effective and most efficient in the widest variety of stream types. Early work by Angermeier and Karr (1986) suggested that fully sampling from two entire meanders typically captures the variety of stream microhabitats, yielding enough individual fish to calculate taxa richness and relative abundances for IBI metrics. Later work in several geographic areas suggests about 40 channel widths as the appropriate length of sampling efforts (Lyons 1992b; Paller 1995a,b; Angermeier and Smoger 1995). In relatively homogeneous systems (e.g., low-gradient streams), longer distances may be needed (Angermeier and Smogor 1995).
Large rivers, lakes, reservoirs, and coastal and estuarine environments contain a diversity of habitats. No single sampling method is appropriate to every one of those habitats, yet using multiple sampling methods is difficult, expensive, and thus impractical. As a result, selective sampling protocols, which measure biological condition on the basis of one or a few local microhabitats, have been developed for these systems (Thoma 1990; Weaver et al. 1993; Jennings et al. 1995; Deegan et al. 1997; Whittier et al. 1997b; Whittier 1998).
Benthic invertebrates, such as insects, crayfish, and worms, pose different sampling challenges: more species to deal with than among fishes, more microhabitats, more sampling techniques and protocols appropriate for the variety of microhabitats. Therefore, you must either use many different protocols to get a representative invertebrate sample or first test whether sampling from a single microhabitat accurately represents stream condition.
In their study of streams in the Tennessee Valley, Kerans et al. (1992) sampled invertebrates from pools (Hess sampler) and riffles (Surber sampler) and evaluated 18 invertebrate attributes as indicators of human influence. They concluded that monitoring designs "that quantitatively sample multiple habitats, are spatially replicated, and use many different attributes for assessment provide a good method for determining biological condition" (Kerans et al. 1992: 388). Although a number of invertebrate attributes behaved similarly for pools and riffles, others (e.g., mayfly taxa richness, caddisfly taxa richness) matched expected stream health rankings better for pools than for riffles. When the researchers combined metrics to create a B-IBI, patterns were stronger for pools than for riffles. Rankings were not always consistent for pool and riffle data (Kerans and Karr 1994), perhaps because these studies were done in relatively large rivers with substantial sedimentation, which might be detected more readily in pool environments (B. L. Kerans, pers. commun.).
Debate still rages over whether single- or multiple-habitat sampling is best with invertebrates. Some contend that a single habitat is adequate; others insist that sampling multiple habitats is essential. Our experience suggests that sampling a single habitat is appropriate and adequate, although our reasons for this conclusion do not always agree with others'. Sampling riffles, for example, is often justified on the grounds that riffles are the most diverse, the most productive, or the dominant habitat (Plafkin et al. 1989; Barbour et al. 1996b; Barbour et al., in press). We are not convinced that these claims are true or even at issue. Still, because we have successfully and cost-effectively used single-habitat samples to discern human effects on small streams (Kerans et al. 1992; Kerans and Karr 1994; Kleindl 1995; Rossano 1995, 1996; Patterson 1996), we recommend a single-habitat sampling protocol that concentrates on riffles.
Because a Surber sampler samples only part of a riffle, a single sample may not be precise enough to judge stream condition. We therefore tested the effects of replicate sampling of invertebrates, using data from the John Day River basin of north-central Oregon (Fore and Karr, unpubl. manuscript). Five replicates were collected, and their contents were identified for each of seven sites (Tait et al. 1994). Using a bootstrap resampling algorithm, Fore and Karr simulated the effects of taking one, three, or five replicates at a site. Fore and Karr changed the number of replicates for each site to test whether metric precision varied as a function of the number of replicates (Figure 32). With only one replicate, a metric could either increase or decrease depending on which of the five replicates was chosen by the bootstrap algorithm. In practice, therefore, the numerical value of a metric calculated using a single Surber sample at a site would depend on where in the riffle that sample had been taken. When the mean of three replicates is plotted, however, the relationship between metric scores and human influence is more consistent (see Figure 32). Metric precision increases little if five replicates are collected instead of three. Thus we conclude that the increased costs of sample collection and analysis for three replicates over one are justified, but not those for five replicates.
For invertebrates, therefore, we recommend a standard sampling area of approximately 0.1 square meter (a Surber sampler frame of 0.3 meters by 0.3 meters) and three replicate samples for each site. We also recommend collecting from riffles for three reasons: (1) riffles are easier to define and identify by field crews than are pools or margins; (2) riffles are more uniform than other stream microenvironments and thus easier to compare across watersheds; and (3) riffles are shallow, and the current through them is fast, making sampling with kicknets or Surber samplers easier. We also take all replicates in a single riffle; this strategy characterizes one site more fully than does the alternative of sampling once in each of several riffles, as some protocols propose (e.g., EMAP; R. M. Hughes, pers. commun.).
Figure 32
Figure 32: Results of bootstrapping analysis (random sampling with replacement) of the relative abundance (percentage) of predators for seven stream sites along a gradient of grazing intensity in the John Day Basin, Oregon. For each site, one, three, or five replicates were randomly selected, and least-fit regression lines (100 in each graph) were plotted. The lines in the upper graph are based on means for one replicate (our of five possible) per site; in the middle, for three replicates per site; in the bottom graph, for five replicates per site. Precision increases with number of replicates, especially between one and three replicates; in fact, the relationship between site condition and proportion of predators may appear either negative or positive with only one replicate. Note, however, that precision increases relatively little from three to five replicates. The lower two graphs clearly show that the relative abundance of predators increases as resource condition improves.
These methods work well in small to midsized streams. For large rivers--where Surber and similar methods are difficult to use, or it is necessary to minimize the effect of habitat differences among sampling sites--sampling using artificial substrates, such as Dendy samplers as used by Ohio EPA, may be preferable (Cao et al. 1996).
It is especially important to collect and count a sufficient number of insects to characterize the biota in multiple dimensions. If sampling fails to yield a total of 500 or more organisms (for example, in regions where natural invertebrate densities are low), the number of replicates or the sampled area may need to be increased. We believe that sampling enough organisms is far more important than how sampling is organized (e.g., single or multiple riffles, composite samples, or no composite samples). Subsampling that counts only 100, 200, or even 300 organisms, as recommended by RBP and some other protocols, tends to reduce the utility of many metrics that have become standard in multimetric assessments (Doberstein 1998; see Premise 29).
It is probably not always necessary to identify insects to species; strong patterns emerge from samples where most insects are identified only to genus (except for chironomids). Identification to genus provides distinct advantages over identification only to family, however--in particular, by strengthening the ability to discriminate among sites of intermediate quality (Figure 33). In California coastal marine environments, for example, family identifications were optimal for detecting the effects of organic pollution on benthic taxa richness, dominance, and diversity (Ferraro and Cole 1990, 1995).
Figure 33.
Figure 33:. Number of clinger taxa present in samples of benthic invertebrates from 65 Japanese streas ranked in order of intensity of human influence (see Figures 4 and 5). The pattern is consistent across the influence gradient. regardless of the level of taxonomic identification, but the slope becomes smaller from species to genus to family, reducing the metric's usefulness for discriminating among sites at higher taxonomic levles (data provided by E.M. Roassno).
Using standard methods for sampling invertebrates (2), we have been able to detect changes in biological condition caused by a whole range of human influences from the Grand Tetons (Patterson 1996) to streams in several areas of Oregon and Washington (Kleindl 1995; Karr, Morley, and Adams, unpublished data).
Finally, for both fishes and invertebrates, timing of sampling is important. Karr et al. (1986) recommended periods of low to moderate stream flow for sampling fishes. For benthic invertebrates, recent experience leads us to recommend late summer, before autumn rains begin. We sample stream insects in the Pacific Northwest in September. Water flows are generally stable and safe for field work at that time of year, and invertebrates are abundant. Sampling at this time also minimizes disturbance to the redds, or nests, of anadromous fish. Optimal sampling period will, of course, vary regionally and should be set based on knowledge of the regional biota, precipitation patterns, and other relevant factors.