Assemblage Stability in Stream Fishes
Transcription
Assemblage Stability in Stream Fishes
Assemblage Stability in Stream Fishes: A Review GARY D. GROSSMAN JOHN F. DOWD MAURICE CRAWFORD School of Forest Resources University of Georgia Athens, Georgia 30602, USA ABSTRACT/We quantified the stability of nine stream fish assemblages by calculating coefficients of variation of population size for assemblage members. Coefficients of variation were high and averaged over 96%; indicating that most assemblages were quite variable. Coefficient of variation (CV) estimates were not significantly affected by: (1) years of study, (2) mean abundance, (3) familial classification, or (4) mean interval between collections. We also detected minor regional differences in CVs. The high variability exhibited by many stream fish assemblages suggests that it may be diffi- Annual and season variations in flow regimes (i.e., droughts and floods) can produce substantial fluctuations in the physical environment of many lotic ecosystems. Because droughts and floods occur with a relatively high frequency, especially when compared to many other natural disturbances (e.g., El Nino, hurricanes), lotic environments are excellent systems for tests of equilibrium and nonequilibrium ecological theories. Implicit tests of such theories occurred as early as 1951, when William Starrett (1951) observed large variations in species abundances in an Iowa riverine fish assemblage and attributed these variations to unpredictable hydrologic events. Similar results were obtained by later researchers (Larimore 1954, Metcalf 1959, Paloumupis 1958, Larimore and others 1959, John 1964, Lowe and others 1967, Rinne 1975, Harrell and others 1967, Harrell 1978, Mills and Mann 1985, Moyle and Li 1979). Prompted by the general ecological debate regarding the importance of equilibrium and nonequilibrium processes to assemblage dynamics, Grossman and others (1982) reviewed the literature on stream systems. This review, coupled with a reanalysis of assemblage structure data from an Indiana stream, led them to reiterate Starrett's hypothesis and suggest that hydrologic variability may facilitate coexistence in assemblages of many stream organisms. The proposed KEY WORDS: Community; Structure; Assemblage structure; Assemblage stability; Community stability; Population variability; Stream fishes Environmental Management Vol. 14, No. 5, pp. 661-671 cult to detect the effects of anthropogenic disturbances using population data alone. Consequently, we urge managers to exercise caution in the evaluation of the effects of these disturbances. More long-term studies of the ecological characteristics of undisturbed stream fish assemblages are needed to provide a benchmark against which disturbed systems can be compared. We suggest that CVs are a better estimator of population/assemblage stability, than either Kendall's W or the standard deviation of the logarithms of numerical censuses. This conclusion is based on the following reasons. First, CVs scale population variation by the mean and, hence, more accurately measure population variability. Second, this scaling permits the comparison of populations with different mean abundances. Finally, the interpretation of CV values is less ambiguous than either of the aforementioned metrics. mechanism for coexistence is that mortality associated with the occurrence of floods and droughts acts to prevent resource limitation or competitive exclusion within lotic systems. Species residing in these environments can utilize similar resources and coexist: a contradiction of several theoretical predictions (MacArthur 1972). If this mechanism is a general one for lotic assemblages, ecological theory, especially equilibriumbased ecosystem stability and recovery models, may have little relevance to streams and rivers (but see DeAngelis and Waterhouse 1987). Grossman and other's conclusions did not go unchallenged (Herbold 1984, Rahel and others 1984, Yant and others 1984). Nonetheless, many investigators now agree that floods and droughts can have a pervasive influence on both structural and functional characteristics of lotic ecosystems (Resh and others 1988). The variability of lotic assemblages, coupled with the potentially restricted applicability of many theoretical models, poses a special problem for agencies charged with the detection and mitigation of anthropogenic disturbances in streams and rivers (e.g., toxicant spills, dams, channelization, etc.). Although some stream taxa apparently recover quickly from disturbance (Yount and Niemi 1990), recovery rates are strongly affected by factors such as: (1) persistence of the effects of disturbance, (2) species' differential abilities to survive disturbance and recovery (Kelly and Harwell 1990, Yount and Niemi 1990), (3) presence of refugia (Sedell and others 1990), and (4) hydrologic conditions (Cairns 1990, Yount and Niemi 1990). In 11990 Springer-Verlag New York Inc. 662 G. D. Grossman and others addition, considerable disagreement exists over the definitions of disturbance and recovery (see Resh and others 1988). Is recovery merely the reappearance of species comprising the original assemblage, or the reestablishment of these species in their prior relative abundances (i.e., previous assemblage structure)? Regardless of these problems, assessment of both disturbance and recovery in lotic ecosystems is dependent upon characterization of the variability present in undisturbed streams and rivers. With this goal in mind, we have quantified the variability of North American stream fish assemblages through an analysis of data from papers published since Grossman and others (1982). Our purpose here is threefold: first, to ascertain the progress made since 1982 with respect to the debate over assemblage organization in lotic fishes, second, to suggest improvements in the methodologies currently used to assess assemblage stability in stream organisms, and third, to relate this general topic to the detection of disturbance and facilitation of recovery in lotic ecosystems. Design of Stream Fish Assemblage Organization Studies At present, there are 10 published studies (Table 1) that test for mechanisms determining the organization of lotic fish assemblages. The basic design of these studies is to delineate permanent station(s) along a stream and then repeatedly sample the stations over many years. Specific spatial and temporal requirements are necessary to assure the validity of these results (Grossman 1982, Grossman and others 1982, Connell and Sousa 1983). First, sampling should include the minimum home-range sizes of the dominant species. This increases the probability that population variability is dominated by mortality and recruitment, rather than by movement in and out of the station. Secondly, sampling should comprise at least one mean generation time of assemblage dominants to ensure that stability is not an artifactual consequence of low adult mortality coupled with great longevity and low recruitment (Frank 1968, Davis and van Blaricom 1978). Although Connell and Sousa (1983) state that the minimum temporal requirement for an assemblage stability study is one complete turnover of the assemblage, this is unnecessary for species with a quantifiable age structure (e.g., many fishes, trees, etc.). Because an investigator can age individuals of such species, population stability caused by relatively equal levels of recruitment and mortality (i.e., true stability), can be differentiated from that due to great longevity coupled with low adult mortality and recruitment (Grossman 1982, Warner and Chesson 1985). After satisfying the spatial and temporal requirements for assemblage organization studies, most investigators tested for concordance of ranked abundances of assemblage members (i.e., assemblage stability) using Kendall's W. Unfortunately, this test possesses methodological limitations that affect its interpretation (Grossman and others 1985, Rahel and others 1984). Because W presently is being used for analyses of assemblage stability, it seems worthwhile to describe the consequences of these problems. First, confusion exists over W's null and alternative hypotheses. The null hypothesis for W is that concordance of ranks is not significantly different from 0 (i.e., random). Conversely, the alternative hypothesis merely states that concordance of ranks is significantly different from 0 (i.e., nonrandom). Hence, when one rejects the null hypothesis, it does not necessarily mean that ranked abundances are stable, but that the relationship is significantly different from 0 or random. To determine the level of stability present in an assemblage, it is necessary to examine the magnitude of W, which ranges from 0.0 to 1.0. If W is high (e.g., >0.75) and significant, one could conclude that ranked abundances are stable because the relationship is strong. Depending on the number of rows and columns in the calculation, however, there are cases where the null hypothesis for W is rejected, yet the value of W is low (e.g., <0.50, see Ross and others 1985, Matthews and others 1988). In these cases, investigators have concluded that ranked abundances were highly stable, a conclusion that may not be warranted given W's low value. Kendall's W possesses an additional problem; it is strongly affected by the continued low abundances of rare species (Grossman and others 1982, Rahel and others 1984). Consequently, a statistically significant result can be obtained if rare species remain rare while common species fluctuate substantially. To control for this possibility, it is necessary to sequentially delete the rarest species from the analysis and recalculate W. Table 2 presents an example where an artifactual conclusion of stability was reached as a result of this problem. Given the interpretational difficulties of W, perhaps a different analytical tool is needed for the quantification of assemblage stability. We are currently using the coefficient of variation (CV) of population size to assess assemblage stability patterns in Coweeta Creek, North Carolina (Freeman and others 1988). This metric has a variety of advantages over W: (1) it directly measures the parameter of interest (i.e., population variability); (2) it standardizes variability (i.e., vari- Stream Fish Assemblage Stability Table 1. 663 Review of stream fish assemblage organization studies published since 1982a Study Site Station Seasons (N) Species (N) Study time span (yr) Annual samples (N) 1 2 5 9 3 4 6 1 5 7 1 1 1 1 1 1 1 1 1 1 1 3 15 15 15 15 9 13 14 13 13 12 12 15 15 15 15 18 18 18 6 6 6 10 6 6 6 6 5 5 5 3 3 3 5-7b 1 1 1 1 1 1 1 3 3 3 3 3 4 4 4 4 3 4 6 5 3 9-22b 10 11 4 6 5 5 5 5 5 4 9-13b 10 11 4 6 5 5 5 5 5 4 4-12b 3 ll-19 b 5-9b 3-5b Matthews and others (1988) Piney Creek Matthews and others (1988) Brier Creek Matthews and Others (1988) Kiamichi River Meffe and Berra (1988) Cedar Fork Card and Flittner (1974) Sagehen Creekc Moyle and Vondracek (1985) Martis Creek0 Freeman and others (1988) Coweeta Creek Grossman and others (1982) Ross and others (1987) Erman (1986) Otter Creek 8 9 10 lOa 2 3 4 1 2 3 1 Black Creek 1 Meffee and Minckley (1987) Sagehen Creek d Aravaipa Creek e Authors' conclusions Ranks concordant, deterministic organization Ranks concordant, deterministic organization Ranks concordant, deterministic organization Ranks concordant, some equilibrium characteristics present Not relevant Ranks concordant, deterministic organization Ranks concordant, populations moderately fluctuating Ranks not concordant, stochastic Ranks concordant, populations stable Ranks concordant, some population fluctuation Ranks concordant, population data not given a We omitted some studies (e.g., Ross and others 1985, Matthews 1986) because later papers (e.g., Matthews and others 1988) included these data. Varied by season. 'Species subjected to substantial angling mortality (e.g., Salmo gairdneri) were excluded from calculations. d Due to changes in the stream fauna produced by an impoundment (Erman 1973), the data used in our reanalysis are the preimpoundment data of Card and Flittner (1974). "Because population data were not presented we could not include data from this study in our analysis. b ance is expressed as a percentage of the mean), which facilitates comparison of species with different mean abundances; and (3) because population variation is expressed as a percentage of the mean, its interpretation is relatively unambiguous. Of course, we are left with the problem of classifying CV values, and Freeman and others (1988) have proposed the following classification scheme: (1) CV =£ 25% = highly stable, (2) 25% < CV =£ 50% = moderately stable, (3) 50% <CV ^ 75% = moderately fluctuating, (4) CV ^ 76% = highly fluctuating. A determination of assemblage stability then is made by examining CV values for all assemblage members. We use this classification in our paper, but recognize its arbitrary nature. It is our hope that the classification system will be refined through dialectical exchanges and further comparative study. Finally, for illustrative purposes, we present assemblage stability analyses based on W and CV values in Table 3. Methods To determine the variability present in stream fish assemblages, we calculated the CV of population size 664 G. D. Grossman and others Table 2. Example of effects of rare species on calculation of Wa Table 3. Comparison of Kendall's W and CV values for three midwestern lotic fish assemblages3 Year Species A B C D E F 1 2 3 4 1604 481 1263 240 1 1 240 882 80 161 1 1 2791 6459 3987 1771 557 239 702 531 3126 439 1 1 0.83 Sig <0.01 Total abundance (%) 100 0.71 <0.01 99 0.43 >0.20 97 W All species Delete rarest species (F) Delete two rarest species (E and F) a Data are from a North American stream and have been multiplied by a constant. for assemblage members from nine drainages (Table 1). These data sets comprise the assemblage organization studies previously discussed and should be consulted for habitat descriptions and collecting methodologies. (For purposes herein, a site represents a stream or river, whereas a station is the subsection of a site within which collections were made.) Rather than reanalyze data from all stations in all sites, we selected data representing a longitudinal gradient along a stream (e.g., Piney Creek stations 1, 2, 5, and 9), which also generally contained a majority of species present. An assemblage member was included in analyses if it occurred in at least 50% of the annual censuses in a site and was not potentially subjected to substantial angling mortality (i.e., salmonids and Micropterus spp.). Although five sites were sampled only during summer months, four additional sites were sampled in at least two seasons (Table 1). Because of possible seasonal differences in population variability, it was necessary to use two seasonal data sets, one based on summer data only and a second including data from all seasons. To calculate population CVs, we selected one census from each set of collections made within a given year. Because some data sets consisted of multiple collections from multiple stations within a given year, it was necessary to further partition the data set. Consequently, we calculated CVs based on censuses from the station with the highest mean abundance for a species, as well as from the station with the lowest mean abundance. We used this partitioning to test for the effects of habitat suitability (i.e., we assumed that Site Brier Creek Station 3 Station 4 Station 6 Piney Creek Station 1 Station 2 Station 5 Station 9 Kiamichi River Station 1 Station 5 Station 7 X abundance (SD) XCV (SD) W 14(17) 29 (50) 30 (23) 135 (38) 112(33) 133 (55) 0.347b NS 0.478C 0.434C 58 (112) 59 (70) 42 (40) 28 (17) 91 (31) 89 (36) 99 (33) 86 (57) 10(7) 16(17) 60 (105) 103 (44) 70 (33) 99 (47) 0.591d 0.502e 0.439= 0.463e 0.317 NS 0.542 NS 0.671' Presented are mean abundance values ± 1 SD, mean values for the coefficient of variaion of population size (± 1 SD) for assemblage members, and W for each assemblage. We used the data and criteria of Matthews and others (1988) for calculations of W. Conclusions based on CV values indicate that these assemblages are composed of highly fluctuating species whereas conclusions based on W values yield a different result (Matthews and others 1988). b Some of our values for W differ slightly from those of Matthews and others (1988) due to minor computational errors in the aforementioned paper. These small errors do not affect the conclusions of Matthews and others (1988), however. C P < 0.05. d P < 0.001. °P < 0.01. estimates from the station with the species' highest mean abundance represent its most favorable habitat, whereas the station with the lowest mean abundance represented its least favorable habitat). Thus, we conducted our analyses on four data sets: (1) CV estimates based on censuses from the station with the highest mean abundance during all seasons (set All-High), (2) CV estimates based on censuses from the station with the lowest mean abundance during all seasons (set AllLow), (3) CV estimates based on censuses from the station with the highest mean abundance during summer (set Summer-High), and (4) CV estimates based on censuses from the station with the lowest mean abundance during summer (set Summer-Low). These data sets are not independent, and as a consequence, we used the 0.01 level of significance to control for experiment-wise error. Because all four data sets approximated a normal distribution (Figure 1), parametric statistics were used in most analyses. We employed distribution-free statistics, however, when deviations from normality were large because of small sample sizes. Besides our interest in quantifying population vari- Stream Fish Assemblage Stability o m O UJ EC 20 191817161514131211 • 109- . l l I... COEFFICIENT OF VARIATION Figure 1. The distribution of coefficient of variation of population size estimates for the All-High data set. Distributions for the three remaining data sets were very similar. ation within stream fish assemblages, we also examined the effects of several methodological techniques on CV estimates. First, it has been suggested that stream fish populations fluctuate greatly because the dynamics of abundant, but variable, young-of-the-year (yoy) overwhelm stable adult populations (Yant and others 1984). We tested this hypothesis in two ways. First, we calculated seasonal CVs for adults only and compared these estimates to those based on all individuals in the population, at three stations in Coweeta Creek. A t test for paired samples was used for hypothesis testing. Second, we compared CV estimates based on spring and autumn samples for four sites. If inclusion of yoy does significandy affect CV values, then there should be a significant difference between CV estimates made in seasons when yoy have the smallest (i.e., spring) and the greatest (i.e., autumn) influence on population size. We tested this hypothesis using the signed ranks test. We also tested for the overall effect of mean abundance on CV estimates by dividing species into 11 abundance classes (mean abundance classes: (1) 1-2 individuals, (2) 3-4, (3) 5_6, (4) 7-8, (5) 9-10,..., (11) >20 individuals) and performing an ANOVA on this data set. Abundance classes containing fewer than five estimates were deleted from analyses. To assess the effects of sampling regime on CVs, we classified studies according to the number of annual censuses in an estimate (classes = 3—5 years, 6—8 years, and >8 years). In addition, we examined whether the mean number of years between censuses 665 significantly affected CV values (not all stations were sampled annually). For this analysis, treatment classes were: 0-1 years, 1+-2 years, 2 + — 3 years, and >3 years. We evaluated both effects statistically using ANOVA and Tukey-Kramer a posteriori tests. We also determined whether within-site, regional, and taxonomic factors affected CV values. We tested for seasonal differences in CV values within a station using either Friedman's test (more than two seasons) or the signed ranks test (two seasons), with species as blocks and season as the treatment (Sokal and Rohlf 1981). A similar analysis was conducted to examine among-station differences in CVs within a site. We examined regional effects on CV estimates by assigning sites to geographical regions and using ANOVA and Tukey-Kramer tests for significance testing. Regional classifications were as follows: (1) midwestern—Brier Creek, Kiamichi River, and Piney Creek; (2) northern —Cedar Fork Creek and Otter Creek; (3) southeastern—Coweeta Drainage; (4) southern—Black Creek; and (5) western—Martis Creek and Sagehen Creek. Finally, we investigated the effect of taxonomic classification on CVs by testing for significant differences among six families (Cyprinidae, Cottidae, Percidae, Catostomidae, Centrarchidae, and Cyprinodontidae). In a secondary analysis, we tested for differences between Notropis and non-Notropis cyprinids using a Mann-Whitney U test. Both regional and taxonomic hypotheses were evaluated using ANOVA and Tukey-Kramer tests. Hydrologic variation is a major cause of variation in stream fish assemblages. To illustrate how hydrologic variation may be compared within and among drainages, we constructed flow duration curves for several rivers in Georgia (Inman 1971). The total period of record for US Geological Survey gauging stations was used to determine the cumulative frequency distribution of the daily mean discharge (Searcy 1959). We scaled the discharge associated with each frequency class by the mean discharge for the sampling period to remove the effects of basin size and climatic variation. Discharge then was plotted on a four-cycle log scale (y axis); the frequency was plotted as the percent time the discharge was equaled or exceeded on a normal probability scale (x axis). Results Mean CV values were high for all data sets (AllHigh X ± SD = 99 ± 42, All-Low = 105 ± 40, Summer-High = 97 ± 41, Summer-Low = 102 ± 42). In addition, neither mean CV values nor variances differed significantly among data sets (X F = 666 G. D. Grossman and others 1.04, P > 0.05, Var F = 1.12, P > 0.05). Not surprisingly, mean abundances were significantly higher in All-High and Summer-High data sets than in the remaining two data sets (All-High, X ± SD = 38 ± 68, All-Low = 10 ± 19, Summer-High = 35 ± 67, Summer-Low = 16 ± 29, X F = 10.92, P < 0.001). Variances in mean abundance also differed significantly among data sets (Fmax = 12.29, P < 0.001). Redundancy among data sets ranged from a low of 21 % (All-High - All-Low) to a high of 69% (All-Low Summer-Low) and averaged 54%. Redundancy occurred when an estimate fell into more than one classification (e.g., when there was only one CV estimate for a species in a site, it occurred in both All-High and All-Low data sets). Inclusion of yoy did not significantly affect CVs at station 1 in Coweeta Creek (Table 4). Because of insufficient sample sizes, tests were not possible for Coweeta stations 2 and 3, but we observed an identical pattern for these stations. Spring—autumn comparisons for five stations at four sites also indicated a lack of significant differences between spring and autumn CV estimates (Table 5). These results suggest that inclusion of yoy in CV estimates does not significantly affect CV values, at least in the sites examined. Neither the number of samples in an estimate nor the mean interval between samples significantly affected CV Values for any data set (Table 6). Coefficient of variation estimates also did not differ significantly among mean abundance classes (Table 6). Within a station, season only had a significant effect on CV values in one of six stations from four sites (Table 7). Within a site, among-station comparisons of CVs did not yield significant results, regardless of the site (Table 8). Regional analyses indicated that southern and southeastern CV estimates were significantly lower then northern CVs in the All-High data set (Table 6). Taxonomic effects were not apparent for any data set (Table 6). This was true even when the Cyprinidae were separated into Notropis and non-TVotropis species: All-Low Summer-High Summer-Low All-High T T T T = -1.26 = 0.35 = -0.37 = 0.19 P P P P = = = = 0.21 0.72 0.70 0.85 Flow duration curves represent the flow characteristics of a stream throughout the entire range of discharge (Searcy 1959). If coexistence in stream fish assemblages is attributable to hydrologic variation (Grossman and others 1982), then flow duration curves may enable us to identify the systems in which this mechanism is important. Flow duration curves are illustrated for several Georgia streams in Figures 2—4. Plotting the flow duration curve on a probability axis expands both ends of the curve. This enhances the differences between sites for both high flows and low flows. Curves with steep slopes have flows that are more highly variable, whereas a flat slope represents a stream with regulated flows. Systems exhibiting steep flow duration curves may be more likely to contain assemblages that are affected by disturbance. Stabilization of flows in such streams may have a greater effect on coexistence than it would have in less variable systems. Flow duration curves varied both within and among drainage systems. The curves for three locations on the Oconee River have similar slopes for the range of flows between 50% and 90% exceedence. However, extreme flows for the basins are different. Allen Creek, the smallest basin with a drainage area of 44.8 km2, was most variable for high flows but less variable for low flows. The middle-sized basin, the Oconee River near Athens, Georgia, had a drainage area of 1030.8 km2. Its extreme flows fell between those of Allen Creek and the 2823.1-km2 drainage area of the Oconee River near Greensboro, Georgia. These flow differences are caused by differences in storage capacity of the systems. Headwater streams tend to be more variable, especially for high flow events, because they possess less channel storage. Downstream gauges also usually display less variability because they integrate different headwater subwatersheds. Curves for four different rivers in Georgia illustrate the differences one can expect among systems. All four streams have drainage areas around 1036 km2. The Etowah River is in the Mountain province, the Yellow River and Oconee River are in the Piedmont province, and the Ogeechee River is substantially in the Coastal Plain province. Flow duration curves for these streams do not have the same slope between the 50% and 90% exceedence values. The Etowah River and Oconee River are flatter than the other two. Overall, there are more differences among rivers in low flows than in high flows. Discussion Our findings suggest that many stream fish assemblages are composed of species that vary substantially in population size (i.e., CV values for all four data sets averaged over 96%). In addition, rare species did not fluctuate more than abundant species because: (1) mean CV values were virtually identical for low and high data sets, and (2) mean abundance did not significandy affect CV values. Finally, sampling error probably did not produce the high CV values observed, be- Stream Fish Assemblage Stability 667 Table 4. Coefficient of variation of population size estimates for all population segments and for adults only, for station 1, Coweeta Creek, North Carolina3 Spring Summer Autumn Species All Adults only All Adults only All Adults only Co. bairdi Rh. cataractae Cl. funduloides On. mykiss Ca. oligolepis 23 62 47 100 117 16 67 47 148 117 34 81 60 65 b 33 84 60 105 b 17 37 125 77 68 17 49 125 b 68 "There were no significant differences in CV estimates for all versus adults only comparisons ((test for paired samples, on each set of seasonal data). b Absem. Table 5. Estimates of coefficient of variation of population size for species present in a site during multiple seasons3 Seasons Species (N) T Spring vs autumn Spring vs autumn Spring vs autumn 6 15 13 8.0 65.0 44.0 Spring vs autumn Spring vs autumn 13 10 38.0 21.5 Site Cowetea Creek Station 1 Cedar Fork Creek Otter Creek Black Creek Station 1 Station 2 a H0: CV estimates do not differ by season. None of the test values are significant at the 0.05 level (signed ranks test). Table 6. Significance tests for the effects of numbers of samples, mean interval (years) between samples, mean abundance, region, and family, on CV estimates3 Data set All-High All-Low Summer-High Summer-Low Number of samples (F) Region Mean interval (F) Mean abundance (F) F 3.15 3.05 1.21 4.24C -0.16 -0.25 -0.83 2.77 2.94 2.39 1.66 2.40b 1.27 1.93 1.22 3.04d Pairwise differences south, southeast, < north north, southwest, west > southeast Family (F) 1.19 1.36 1.15 0.91 "Results are for F tests and Tukey-Kramer a posteriori tests. None of the hypothesis tests for differences in mean CV values between Notropis and Non-Notropis cyprinids were significant. b /> = 0.0244. Abundance classes: 11 < 7. C P < 0.005. d P = 0.0194. cause studies in which different collecting methods were used (i.e., seining, draining, and electrofishing) generally did not exhibit significant differences in CV values (i.e., midwestern, northern, and southern versus western and southwestern). These results support the findings of Grossman and others (1982). Although we would not postulate that these systems are organized solely by stochastic mechanisms (Grossman and others 1982), the application of equilibrium-based models to these assemblages may be inappropriate. The high variability present in stream fish assemblages poses a significant problem for resource managers. First, this variability may make the detection of many anthropogenic disturbance difficult. We are not suggesting that managers take a passive role with respect to this problem; quite the contrary. What we do 668 G. D. Grossman and others Table 7. Friedman's test results for seasonal differences in CV values within a station Site Coweeta Creek Station 1 Station 2 Station 3 Cedar Fork Creek Otter Creek Black Creek Station 1 — Allen Creek - - near Athens Season Species (N) Test value All All All All All 5 4 3 15 13 0.40 7.60a All 13 .154 .133 3.57 "P < 0.05. .01 .1 1 10 50 90 99 99.999.99 Percent Time Discharge Equaled or Exceeded Figure 3. Flow duration curves for three portions of the Oconee River (Georgia) drainage. Table 8. Significance tests for among-station differences in CV values within a site3 Site Black Creek Autumn Spring Kiamichi River Brier Creek Piney Creek Martis Creek Sagehen Creek Coweeta Drainage Spring Summer Stations (N) Species (N) Test value 2 9 3 3 4 3 4 9 13 5 7 10 2 3 19.00 45.00 0.40 2.00 3.00 3.00 1.80 3 3 3 3 0.67 4.67 >0.05). .01 .1 1 10 50 90 99 99.999.99 Percent Time Discharge Equaled or Exceeded Figure 2. A flow duration curve for the Oconee River near Greensboro, Georgia. recommend is that managers recognize that the effects of even moderate levels of anthropogenic disturbance may be masked by the natural variability of lotic fish assemblages (for a similar conclusion, see Vaughan and Van Winkle 1982). This necessitates very cautious judgement regarding whether or not an impact has Oconee R. — Yellow R. -- Oqeechee R. --- Etowoh R. 1.00-; .01 .1 1 10 5O 90 99 99.9 99.99 Percent Time Discharge Equaled or Exceeded Figure 4. Flow duration curves for four Georgia rivers. occurred! It also calls for the development of more sophisticated numerical techniques to enable researchers to better identify disturbance-induced population trends. The use of indices of environmental condition such as the index of biotic integrity also show promise for the identification of disturbance in stream fish assemblages (Fausch and others 1988, Hughes 1990), although recognition of the variability present in these assemblages should be incorporated in such indices. One prediction regarding disturbance can be made from our results. If variability in stream fish populations is produced by hydrologic variation, and this variation prevents competitive exclusion through its negative effect on population size, then any action that reduces hydrologic variability (e.g., hydroelectric and flow-control structures) may cause a loss of species. Unfortunately, because of the substantial habitat modifications caused by these structures, it will be difficult to know whether a loss of species is due to: (1) flow stabilization, (2) habitat modification, or (3) a combination of both factors. We suggest, however, that structures that reduce hydrologic variability may cause a loss of species independent of habitat modifications. Stream Fish Assemblage Stability Although flow duration curves may aid us in identifying systems in which flow stabilization potentially may have strong physical or biological effects, the precise relationship between these curves and population variability or species richness of fish assemblages is not well known. Connell and Sousa (1983) proposed that the standard deviation of the logarithms (base 10) of sequential censuses be used as an estimator of population stability. They recognized that the CV of population size could be used for this purpose, but stated that it was sensitive to high values (Connell and Sousa 1983, p. 800). Connell and Sousa (1983) noted that their index was sensitive to low values, but because they "were more interested in population variation at low numbers," they did not use the CV. Nevertheless, we believe that CV values are a more relevant estimator of population stability for the following reasons. First, CVs express the standard deviation as a percentage of the mean: the type of variability most relative to a population stability study. Connell and Sousa (1983) attempt to remove the effects of differential mean abundances by using logarithms as a scaling factor. Although this technique collapses the variability observed, it does not scale variability by the original mean abundance. This has important consequences for the measurement of population variability (Table 9). Species of low abundance that also have standard deviations greater than the mean may obtain low values for Connell and Sousa's index (e.g., Table 9, Oncorhynchus mykiss). This cannot happen with CV estimates. Although a more complete analysis of the behavior of these two estimators is in progress (Grossman, unpublished data), we suggest that examination of assemblage members' CV values better represent the parameter of interest in a population and/or assemblage stability study. Second, because CV estimates are calculated by dividing the standard deviation of population estimates by mean abundance, their interpretation is simple and unambiguous. A CV value of 50% means that the standard deviation is one-half the mean abundance. In contrast, the interpretation of the standard deviation of the logarithms of sequential censuses is less clear. In fact, because of the use of logarithms, population variability estimates are compressed, even though they are the parameter of interest. Although we prefer CV values over W for tests of population and/or assemblage stability, this technique is not without error. First, because CV values are ratios, they may possess unusual distributional properties, especially if the variance is correlated with the mean. Neither of these problems influenced our data set (e.g., Figure 1, Table 8), but they could affect 669 others. Second, by decomposing an assemblage into its component populations, we are no longer examining "assemblage level" behavior. Third, the classification system proposed for CV values does not have a strong a priori foundation. Fourth, CV values cannot distinguish between sampling variability and actual population variability. Fifth, CV values cannot detect time-dependent trends in population variation (i.e., long-term increases or decreases or cyclical fluctuations) or correlated trends among assemblage members. Despite these problems, most of which are shared by W (i.e., 3, 4, and 5), we still believe that CV values are a valuable tool for quantifying assemblage stability. In addition, some of these limitations (i.e., 1 and 5) can be addressed by examination of the abundance data upon which CV estimates are based. Finally, the use of similarity indices also shows promise for tests of assemblage stability (see Matthews and others 1988); however, their behavior must be investigated better before we can evaluate their efficacy. We did not restrict our analyses to censuses taken at frequencies representing at least one turnover of the assemblage, as suggested by Connell and Sousa (1983). We also did not heed our own criteria and directly examine population age structures, because, with one exception (Coweeta drainage), these data were not available. We did find, however, that increasing the time interval between censuses did not significantly affect CV estimates. Our analyses indicate that familial classification did not have a strong effect on CVs. Centrarchid species were just as variable as cyprinids and percids. Although more data are needed to examine the generality of this finding, our data base does include estimates from a variety of regions. Geographical analyses showed that CV estimates varied significantly among regions for All-High and Summer-Low data sets. These findings must be viewed as tentative, because some regions were represented by only one (i.e., south and southeast) or two (western and northern) streams. In fact, Coweeta Creek, the sole southeastern stream, is a southern Appalachian trout stream. It is certainly not representative of the majority of lotic systems in the region (i.e., Piedmont and Coastal Plain systems). Although our data indicate that stream fish populations are quite variable, we will not propose an organizational mechanism for these systems. Like Meffe and Berra (1988), we believe that more detailed environmental data are necessary for causal inferences regarding the factors determining population levels and variability. It is clear, however, that the assemblages examined here are probably not in equilibrium. In addition, species within a given system may respond to biological and physicochemical variation in a species- 670 G. D. Grossman and others Table 9. Comparison of population variability estimtes made using CV of population size and standard deviation of the logarithms (base 10) of sequential censuses (Connell and Sousa 1983)a Species Abundance (X ± 1 SD) CV Standard deviation of log (n + 1) Calostomus tahoensis Coitus bairdi Salmo trutta Rhinichthys osculus Rhinichthys cataractae Oncorhynchus rnykiss Clinostomus funduloides 245a 54 24 22 17 8 7 160 20 123 191 24 111 49 0.66 0.08 0.76 0.75 0.10 0.43 0.17 ± ± ± ± ± ± ± 393 11 29 42 4 9 4 "Data rounded to the nearest individual, fractional values to ±0.1 were included in calculations. specific manner (Mills and Mann 1985, Schlosser 1985, Freeman and others 1988). The elucidation of these mechanistic responses should yield insights into the maintenance of assemblage structure in stream fishes. In conclusion, in contrast to other authors (Herbold 1984, Yant and others 1984, Matthews 1986, Ross and others 1987, Matthew and others 1988), we suggest that populations comprising stream fish assemblages vary substantially. Our purpose here is not to criticize the conclusions of earlier investigators but to identify how different analytical techniques may yield differing conclusions. These conclusions also may strongly affect how resource professionals manage lotic systems. For example, we would urge resource managers to be cautious with respect to the evaluation of anthropogenic impacts on stream systems, because even substantial impacts may be difficult to detect. A similar caveat applies to the detection of recovery in damaged streams. Mitigation effects should not be halted until mean abundances and variability approximate predisturbance levels. Of course this requires data on predisturbance population dynamics, data which are lacking not only for individual streams (but see Erman 1973, 1986) but also for entire geographical regions. Consequendy, we would urge management agencies to undertake more long-term studies of stream fish assemblages in undisturbed watersheds to provide a benchmark against which disturbed systems can be compared. Acknowledgments We gratefully acknowledge T. Berra, D. Erman, W. Matthews, G. Meffe, P. Moyle, S. Ross, and J. Whitaker, Jr., for providing access to their data sets on stream fishes, and R. Ratajczak for data analysis. This manuscript benefited from the comments of several of the aforementioned investigators as well as those of L. Barnthouse, J. Barrett, V. Boule, M. Freeman, P. Harper, J. Hill, H. Li, G. Niemi, D. Stouder, and D. Yount. We also appreciate the patience and support of B. Dowd and B. Mullen. Finally, we thank the Environmental Protection Agency and the Natural Resources Research Institute, University of Minnesota, for inviting us to participate in the Lode Ecosystems Recovery Workshop. Portions of this work were supported by Mclntire-Stennis grant GEO-0035-MS to the senior author. Literature Cited Cairns, J., Jr. 1990. Theoretical basis for predicting rate and pathways of recovery. Environmental Management 14:517526. Connell, J. H., and W. P. Sousa. 1983. On the evidence needed to judge ecological stability or persistence. American Naturalist 121:789-824. Davis, N., and G. R. van Blaricom. 1978. Spatial and temporal heterogeneity in a sand bottom epifaunal community of invertebrates in shallow water. Limnology and Oceanography 23:417-427. DeAngelis, D. L., and J. C. Waterhouse. 1987. Equilibrium and nonequilibrium concepts in ecological models. Ecological Monographs 57:1-21. Erman, D. C. 1973. Upstream changes in fish populations following impoundment of Sagehen Creek, California. Transactions of the American Fisheries Society 102:626—629. Erman, D. C. 1986. Long-term structure of fish populations in Sagehen Creek, California. Transactions of the American Fisheries Society 115:682-692. Fausch, K. D., J. R. Karr, and P. R. Yant. 1984. Regional application of an index of biotic integrity based on stream fish communities. Transactions of the American Fisheries Society 113:39-55. Frank, P. 1968. Life histories and community stability. Ecology 49:355-357. Freeman, M. C., M. K. Crawford, J. C. Barrett, D. E. Facey, M. G. Flood,]. Hill, D.J. Stouder, and G. D. Grossman. 1988. Fish assemblage stability in a Southern Appalachian stream. Canadian Journal of Fisheries Aquatic Sciences 45:1949-1958. Card, R., and G. A. Flittner. 1974. Distribution and abundance of fishes in Sagehen Creek, California. Journal of Wildlife Management 38:347-358. Stream Fish Assemblage Stability Grossman, G. D. 1982. Dynamics and organization of a rocky intertidal fish assemblage: the persistence and resilience of taxocene structure. American Naturalist 119:611-637. Grossman, G. D., P. B. Moyle, and J. O. Whitaker, Jr. 1982. Stochasticity in structural and functional characteristics of an Indiana stream fish assemblage: a test of community theory. American Naturalist 120:423-453. Grossman, G. D., P. B. Moyle, and J. O. Whitaker, Jr. 1985. Stochasticity in structural and assemblage organization in an Indiana stream fish assemblage. American Naturalist 126:275-285. Harrell, H. L. 1978. Responses of the Devil's River (Texas) fish community to flooding. Copeia 1978:60—68. Harrell, R. C., B.J. Davis, and T. C. Dorris. 1967. Stream order and species diversity of fishes in an intermittent stream. American Midland Naturalist 78:428—436. Herbold, B. 1984. Structure of an Indiana stream fish association: choosing an appropriate model. American Naturalist 124:561-572. Hughes, R. M. 1990. Use of ecoregions to develop biological criteria for assessing recovery of aquatic ecosystems. Environmental Management. Inman, E. 1971. Flow characteristic of Georgia streams. United States Geologic Survey Open File Report, Atlanta, Georgia. 262 pp. John, K. R. 1964. Survival of fish in intermittent streams of the Chiricahua Mountains, Arizona. Ecology 45:112-119. Kelly, J. R., and M. A. Harwell. 1990. Indicators of ecosystem recovery. Environmenal Management 14:527—546. Larimore, R. W. 1954. Minnow productivity in a small Illinois stream. Transactions of the American Fisheries Society 84:110— 116. Larimore, R. W., W. F. Childers, and C. Heckrotte. 1959. Destruction and reestablishment of stream fish and invertebrates affected by drought. Transactions of tlie American Fisheries Society 88:261-285. Lowe, C. H., D. S. Hinds, and E. A. Halpern. 1967. Experimental catastrophic selection and tolerances to low oxygen concentration in native Arizona freshwater fishes. Ecology 48:1013-1017. MacArthur, R. 1972. Geographical ecology. Harper and Row, New York. 269 pp. Matthews, W.J. 1986. Fish faunal structure in an Ozark stream: stability persistence and a catastrophic flood. Copeia 1986:388-397. Matthews, W. J., R. C. Cashner, and F. P. Gelwick. 1988. Stability and persistence of fish faunas and assemblages in three mid-western streams. Copeia 1988:947-957. Meffe, G. K., and W. L. Minckley. 1987. Persistence and stability of fish and invertebrate assemblages in a repeatedly disturbed Sonoran desert stream. American Midland Naturalist 117:177-191. Meefe, G. K., and T. M. Berra. 1988. Temporal characteristics of fish assemblage structure in an Ohio stream. Copeia 1988:684-690. Metcalf, A. L. 1959. Fishes of Chautauqua, Cowley and Elk counties, Kansas. University of Kansas Publications, Museum of Natural History 11:345-400. Mills, C. A., and R. H. Mann. 1985. Environmentally-in- 671 duced fluctuations in year-class strength and their implications for management. Journal of Fish Biology 27(suppl A):209-226. Moyle, P. B., and H. W. Li. 1979. Community ecology and predator-prey relations in warm water streams. Pages 171-180 in H. Clepper; (ed.), Predator-prey systems in fisheries management. Sport Fishing Institute, Washington, DC. Moyle, P. B., and B. Vondracek. 1985. Persistence and structure of the fish assemblage in a small California stream. Ecology 66:1-13. Paloumupis, A. A. 1958. Responses of some minnows to flood and drought conditions in an intermittent stream. Iowa State Journal of Science 32:547-561. Rahel, F.J., J. D. Lyons, and P. A. Cochran. 1984. Stochastic or deterministic regulation of assemblage structure? It may depend on how the assemblage is defined. American Naturalist 124:583-589. Resh, V. H., A. V. Brown, A. P. Covich, M. E. Gurtz, H. W. Li, G. W. Minshall, S. R. Reice, A. L. Sheldon, J. B. Wallace, and R. Wissmar. 1988. The role of disturbance in stream ecology. Journal of ttie North American Benthological Society 7:433-455. Rinne, J. N. 1975. Changes in minnow populations in a small desert stream resulting from naturally and artificially induced factors. Southwestern Naturalist 20:185—198. Ross, S. T., W.J. Matthews, and A. A. Echelle. 1985. Persistence of stream fish assemblages: effects of environmental change. American Naturalist 126:24-40. Ross, S. T., J. A. Baker, and K. E. Clark. 1987. Microhabitat partitioning of Southeastern stream fishes: temporal and spatial predictability. Pages 42-51 in W.J. Matthews and D. C. Heins (eds.), Community and evolutionary ecology of North American Stream fishes. University of Oklahoma, Norman, Oklahoma. 310 pp. Schlosser, I.J. 1985. Flow regime, juvenile abundance, and the assemblage structure of stream fishes. Ecology 66:1484-1490. Searcy, J. K. 1959. Flow duration curves. United States Geologic Survey Water Supply Paper 1542-A. Sedell, J., R. Hauer, C. P. Hawkins, and J. Stamford. 1990. The role of refugia in recovery from disturbance: modern fragmented and disconnected river systems. Environmental Management 14:711-724. Starrett, W. C. 1951. Some factors affecting the abundance of minnows in the Des Moines River, Iowa. Ecology 32:13-27. Vaughan, D. S., and W. Van Winkle. 1982. Corrected analysis of the ability to detect reductions in year-class strength of the Hudson River white perch (Morone americana) population. Canadian Journal of Fisheries and Aquatic Sciences 39:782-785. Warner, R. R., and P. L. Chesson. 1985. Coexistence mediated by recruitment fluctuation a field guide to the storage effect. American Naturalist 125:769-787. Yant, P. R., J. A. Karr, and P. L. Angermeier. 1984. Stochasticity in stream fish communities: an alternative explanation. American Naturalist 124:573-582. Yount, J. D., and G. J. Niemi. 1990. Recovery of lotic communities and ecosystems from disturbance—a narrative review of case studies. Environmental Management 14:547— 570.