Combining Field Data With Computer Simulations to Determine a Representative Reach for Brook Trout Assessment

Fisheries biologists often infer the status of a stream fishery by sampling a representative stream reach or site with backpack electrofishing gear to conduct population estimates. Multiple pass depletion techniques and removal estimators (e.g., Zippin 1956, 1958; Carle and Strub 1978; White et al. 1982) are commonly used to estimate abundance, density, or biomass of targeted species at the sample site. This type of information at single or multiple sites can be used by itself or in conjunction with other information (i.e., physical stream characteristics) to classify streams or watersheds for management purposes such as assessing management strategies and to identify and afford greater protection of those streams showing exceptional quality (Modde et al. 1991; Brenden et al. 2008). Fish are generally not distributed uniformly within a stream because of heterogeneity in the distribution of preferred habitat, competitive interactions among individual fish occupying preferred habitat, and seasonal fish movements to feeding, spawning, and refuge habitat. Given that fish are not uniformly distributed within a stream, estimates of fish density obtained through dividing abundance estimates by the area sampled are highly dependent upon the amount of stream habitat sampled. If the length of stream sampled at a selected representative site is too short, density estimates could be either negatively or positively biased depending on whether the selected site encompassed marginal or optimal habitat for the species of interest.

The appropriate length of a representative reach is a pervasive question in stream fish management and depends on the component of the stream system examined. For stream habitat assessments, a representative reach length of 40 times the average stream width has been recommended (Simonson et al. 1994; Kaufmann and Robison 1998). Mason (2009) suggested a 250-m-long reach of stream be sampled for habitat assessment based upon variation in thalweg profiles for small Pennsylvania streams containing brook trout Salvelinus fontinalis. Others have provided recommended reach lengths for estimating fish species richness or an index of biotic integrity (e.g., Angermeier and Smogor 1995; Reynolds et al. 2003; Hughes and Herlihy 2007). However, when the sampling objective is to estimate fish abundance, density, or biomass of a targeted species, there is a paucity of information in the literature as to an appropriate representative reach length, and it is likely species and location specific.

The objective of this study was to determine how bias in estimates of brook trout density (no./100 m²) and biomass (kg/ha) change with increases in the length of stream sampled. Brook trout were chosen as our study species because of their socioeconomic importance and much effort is currently being put forth by federal, state, and nongovernmental organizations for brook trout restoration. The sampling effort represented the amount of effort normally used by a typical field crew of three to four persons in the course of a single day when assessing brook trout fisheries in small headwater streams (2- to 5-m wetted width) using backpack electrofishing gear during summer base flow conditions. When combined with a desired level of accuracy needed for management purposes, this information will yield recommendations for an appropriate representative reach length for assessing brook trout populations.

This study was conducted on seven streams in the Sinnemahoning Creek watershed in north-central Pennsylvania (Figure 1). These streams are representative of wild brook trout streams within the Appalachian Plateau physiographic province of Pennsylvania. Streams were first to third order at the sampling site. Cumulative drainage areas and slopes at the sampling site were obtained from NHDPlus data (http://www.horizon-systems.com/nhdplus/) and ranged from 3.8 to 18.5 km² and 1.9% to 5.9% slope. Mean wetted widths ranged from 2.5 to 4.6 m (Table 1). The streams can be characterized as step-pool type streams although long sections (∼50 m or more) of simple riffle habitat with a lack of large woody debris exist, which is typical of Appalachian streams that have undergone previous logging activities within the past century. Stream temperatures rarely exceed 20°C in these streams during summer, total alkalinity values ranged from 8 to 34 mg/L, and pH ranged from 6.5 to 7.7. All streams, with the exception of Driftwood Branch, are located on public land under the ownership of the Pennsylvania Department of Conservation and Natural Resources and are forested with mixed hardwoods. Three streams (Fourmile Run, Lick Island Run, and Rock Run) were sampled in 2010, and the remaining four streams (Bell Draft, Driftwood Branch, Montour Run, and Prouty Run) were sampled in 2011. Species richness was low in these streams with only two to three other species being present in each. Other species included slimy sculpin Cottus cognatus, blacknose dace Rhinichthys atratulus, and longnose dace R. cataractae. Low numbers of brown trout Salmo trutta were found in Driftwood Branch, Prouty Run, and Lick Island Run, but brook trout were the dominant trout species by numbers and biomass.

A 600-m-long study reach was established on each stream during summer baseflow conditions. We chose a 600-m reach length because this was the maximum length of stream that two electrofishing crews could sample using multiple pass depletion backpack electrofishing techniques during the course of a single day. The 600-m reach was then divided into twelve 50-m subreaches. One electrofishing crew started at meter 0 (the downstream end of the study reach) while the other electrofishing crew started at meter 300. Block nets were placed at the upstream and downstream end of each 50 m subreach to meet the assumption of a closed population. Crews then electrofished all available habitats within the 50-m subreach. A three-person crew conducted sampling with a DC battery powered Appalachian Aquatics backpack electrofisher outfitted with electrodes consisting of 30.45-cm–diameter and 0.95-cm–thick stainless steel rings. One person carried the electrofisher, a cathode, and a dipnet, a second person carried an anode and a dipnet, and a third person followed with a dipnet and a bucket to transport fish. Specific conductance in the sampled streams ranged from 34 to 64 µS/cm (18 to 49 µS/cm ambient conductivity) and voltages on the electrofisher ranged from 250 to 350 V and was adjusted as per manufacturer recommended amperage (8–9 A) across the electrofishing unit. Appropriate voltage settings were verified by electrotaxis and narcosis of brook trout. Multiple electrofishing passes were made in each subreach until either no brook trout were collected or the electrofishing crew conducted five passes (a minimum of three passes were conducted on each subreach). Electrofishing crews would sample two consecutive 50-m subreaches on a given pass so as to increase the time between electrofishing passes and thus allowing time for remaining fish to return to normal behavior. Total length (mm) and weight (g) were collected from each captured brook trout and the data were kept separate for each pass within each 50-m subreach. Once an electrofishing crew had completed sampling on two consecutive 50-m subreaches, they moved upstream to begin sampling the next two consecutive 50-m subreaches. A block net was kept in place at meter 300 the entire day to ensure that there was no movement of fish collected and released by the upstream electrofishing crew into areas of the stream the downstream electrofishing crew had not yet sampled. Wetted widths were measured to the nearest 0.1 m at three locations within each subreach.

We used computer simulations to explore how bias of brook trout density and biomass estimates changed with increasing reach length. The 600-m reach was considered the sampling universe from which samples could be drawn. Two types of simulations were conducted. The first type of simulation (hereafter termed contiguous subreaches) mimicked electrofishing procedures used to estimate brook trout density and biomass by a field crew during a typical visit to a single reach on a stream. A starting subreach was chosen and additional adjacent subreaches upstream of the starting subreach were included such that the total simulated reach equaled some desired total reach length. For example, if the simulated total reach length was to be 200 m and the starting subreach began at meter 100, then subreaches starting at meters 100, 150, 200, and 250 were included. Catches of brook trout in each subreach were combined according to electrofishing pass for the first three electrofishing passes. Brook trout collected during the fourth and fifth electrofishing passes were not included in the simulation so as to simulate typical three-pass depletion techniques. This type of simulation was conducted for all possible starting locations within the total 600-m reach that allowed a desired reach length (50–600 m total) to be completely contained within the total 600-m reach. The second type of simulation (hereafter referred to as random subreaches) was a Monte Carlo analysis in which total reach lengths of 50 to 600 m were simulated by randomly choosing (without replacement) 1 to 12 subreaches such that the combined length equaled a desired total reach length. This process was repeated 500 times for each desired total reach length. We simulated without replacement because this more accurately reflects how sampling would take place in the field. Furthermore, sampling with replacement could have inflated the bias of density and biomass estimates if a 50-m subreach that had unusually high or low numbers of brook trout was included multiple times in a single simulation. Again, catches of brook trout in each subreach were combined according to the first three electrofishing passes. Randomly choosing subreaches to combine decreased any effect of similarity in brook trout density or biomass among adjacent sub-reaches and allowed a more rigorous examination of the possible variation in estimates of density and biomass.

Population estimates of brook trout within each simulation were made using the Carle and Strub (1978) maximum likelihood removal estimator:

where Z  =  an intermediate statistic in the Carle and Strub (1978) estimator, Ŷ  =  estimated population size, K  =  number of capture events, T  =  total number of animals in event i, ⁠, C_i  =  the number of animals captured in event i, and α and β  =  parameters of the prior beta distribution, both set to 1.0 for a uniform prior. The population size was found by searching for the lowest Ŷ that produced a Z-value of 1.0 or less. The total number of animals captured, T, was used as a lower bound for Ŷ in the search. When estimating density, population estimates were made for age 1+ and young-of-the-year (YOY) brook trout separately and combined (combined ages), and when estimating biomass, population estimates were made separately for 25-mm–length classes. Age classes were determined by examining length frequency histograms for each stream.

Density (no./100 m²) was estimated as

where Ŷ_i  =  the population estimate of age 1+, YOY, or all brook trout, –—–ww  =  the mean wetted width of the simulated sampling reach (m), and l  =  the length of the sampling reach (m). Biomass (kg/ha) was estimated as

where Ŷ_j  =  the population estimate of length class j and w¯_j  =  the mean weight (kg) of fish in length class j.

Bias was the difference between estimated density and biomass and the “true” density and biomass. The true density was equal to the total number of fish (age 1+, YOY, and combined ages) captured during all electrofishing passes in all subreaches divided by the total wetted area of the 600 m reach. The true biomass was equal to the total weight of all fish captured during all electrofishing passes in all subreaches divided by the wetted area of the 600-m reach and converted to kilograms per hectare. To make comparisons across streams, absolute bias was converted to percent bias and mean percent bias and 5th and 95th percentiles of percent bias (creating a 90th percentile range) were calculated for each stream and total simulated reach length. Percent bias was summarized across streams by averaging mean percent bias and 5th and 95th percentiles of percent bias.

Brook trout were captured from all 50-m subreaches in all streams except for one subreach on Montour Run. The total number of brook trout captured within all other 50-m subreaches ranged from 1 to 66. Across the seven streams, a total of 84 subreaches were sampled. Of these 84 subreaches, no brook trout were captured during the third electrofishing pass on 32 subreaches and no additional passes were conducted. Of the 52 subreaches where at least four electrofishing passes were conducted, brook trout were found on the fourth pass in 18 subreaches. Of these 18 subreaches where brook trout were captured on the fourth electrofishing pass, nine had brook trout captured on the fifth pass (Data S1, Supplemental Material). Estimated capture efficiency from the Carle and Strub (1978) estimator over the first three electrofishing passes averaged 0.79 ± 0.17 (± standard deviation) for age 1+ brook trout and 0.77 ± 0.18 for YOY brook trout. (These estimates of capture efficiency are likely positively biased [Sweka et al. 2006]).

True densities and biomass varied considerably among streams (Table 1). Densities of YOY in streams sampled in 2010 were generally greater than YOY densities in 2011; whereas, age 1+ densities were generally greater in 2011 compared with 2010. Likewise, biomass of the streams sampled in 2011 was greater than those sampled in 2010. Age 1+ brook trout total length ranged from 79 to 250 mm and YOY brook trout total length ranged from 43 to 84 mm.

The contiguous subreach simulations showed that the potential range for bias decreased with increasing stream length electrofished. Mean percent bias in estimates of YOY, age 1+, and total brook trout density fluctuated with increasing reach length but remained near 0 in all cases (Figures 2–4) when examining individual streams. The 90th percentile range of percent bias could be quite large when small reach lengths were simulated, but it decreased rapidly as the simulated reach length increased. The 90th percentile range of bias was generally greater for YOY density (Figure 3) compared to age 1+ (Figure 2) or total brook trout density (Figure 4) and could range from −100% to greater than 300% when reach lengths were ≤100 m. Percent bias of biomass estimates showed similar patterns (Figure 5) and mean percent bias fluctuated near 0% over all simulated reach lengths and the 90th percentile range decreased with increasing stream length.

The random subreach simulations gave similar results when compared to the contiguous subreach simulations (Figures 6–9). Mean percent bias was again near 0% and the 90th percentile range decreased with increasing simulated reach length electrofished for density of all age classes as well as biomass. There were fewer fluctuations in the 90th percentile ranges (as represented by the dashed lines corresponding to the 5th and 95th percentiles in Figures 6–9) than in the contiguous subreach simulations due to the higher sample sizes used in the Monte Carlo simulations.

In general, 90th percentile ranges of percent bias reached −25% to +25% at around 400–450 m of simulated stream length when considering all streams as a group (Table 2). The exception was mean percent bias of biomass estimates in the contiguous subreach simulations which tended to show the potential for more negative bias and even at a reach length of 550 m, the 90th percentile range of percent bias ranged from −33.04% to 1.17%. Average values of percent bias from the case of a 600-m simulated reach reflect bias due to the removal estimator itself with only three electrofishing passes.

The appropriate reach length for estimating stream fish density or biomass ultimately depends upon the purpose for the assessment and the desired level of accuracy. Robson and Regier (1964) suggested ±50% may be adequate for preliminary studies or in fisheries management surveys where only a rough idea of population size and composition is needed; whereas, ±25% may be required for more accurate management work, and ±10% should be routine for research purposes. Under these general guidelines, our simulations suggest a representative reach length of 200–250 m for coarse management purposes and 400–450 m for more accurate management purposes when estimating brook trout density or biomass from a representative reach. We consider these reach length recommendations fairly robust for brook trout streams in Pennsylvania because our field sampling covered a range of stream sizes and brook trout abundances typically encountered in Pennsylvania under base flow conditions. Also, our sampling encompassed interannual variation in year class strength, with 2010 showing higher densities of YOY compared with 2011. Our results should not be interpreted as a universal sampling recommendation for brook trout assessment because sampling designs can, and should, vary depending on the research or management questions being asked. Our results are informative for situations in which time and personnel constraints limit sampling to a single site on a stream to infer population status with commonly used DC backpack electrofishing gear.

A large uncertainty in our study was the estimate of true values of density and biomass within the 600-m study reaches. Although the majority of subreaches (89%) were electrofished until no brook trout were found on the last electrofishing pass, 9 out of 84 subreaches had at least one remaining brook trout by the fifth electrofishing pass and it is likely that additional brook trout remained in the stream following the fifth electrofishing pass, especially in stream reaches with more complex habitat. This makes our estimates of true density and biomass themselves negatively biased. Nevertheless, we feel that we did remove the vast majority of brook trout from the streams, allowing valid assessment of how the potential for bias changes with electrofishing site length. An alternative study design could have used a mark–recapture approach whereby brook trout were electrofished, marked, and released back into the 600-m study reaches. Bias in density and biomass could have then been assessed based on the number of marked fish. However, such a study would also have associated error due to possible mortality from electrofishing, handling, and marking individuals, especially YOY brook trout.

Removal estimators are generally negatively biased and this bias is a function of population size (Riley and Fausch 1992; Sweka et al. 2006) and differences in capture probability due to fish size, stream size or depth, and habitat complexity (Peterson et al. 2004; Meyer and High 2011). In our simulations, we did not account for the effects of stream habitat on capture probability, but we did account for the effects of fish size to a degree by stratifying density and biomass estimates according to age and length classes, respectively. Some researchers have recommended developing and using capture probability models (Peterson et al. 2004) or using mark–recapture methods (Peterson and Cederholm 1984) in place of removal estimators to avoid the negative bias associated with them. Although efforts should be made to minimize or correct for the bias associated with removal estimators, Meyer and High (2011) suggested these biases will be relatively small and probably adequate for routine monitoring purposes. The additional effort needed to obtain unbiased estimates, such as mark–recapture estimates, may be outweighed by the increased cost associated with repeated trips to the same site (Meyer and High 2011), which may be large for remote brook trout streams throughout the species's range. If the purpose of sampling is to infer the status of a stream fish population, site selection likely has as much or more influence on conclusions about that population than bias in the removal estimator.

Representative reach lengths recommended for assessing stream habitat may be too small for adequate representation of stream fish population density or biomass. For example, if a reach length of 40 times the mean wetted width (Simonson et al. 1994; Kaufmann and Robison 1998) had been used to assess brook trout density and biomass in our study streams, electrofishing site lengths would have been 100–184 m based on the average wetted widths (Table 1). At these reach lengths percent bias could range from −62% to +99% for biomass estimates and −55% to +70% for total brook trout density estimates. Mason (2009) suggested a site length of 250 m for typical brook trout streams in Pennsylvania based upon variation in thalweg profiles. At 250 m, percent bias could range from −38% to +44% for biomass estimates and −30% to +37% for total brook trout density estimates.

It is not surprising that length of a representative reach to assess brook trout density or biomass would need to be greater than a length of a representative reach to assess stream habitat. Brook trout prefer pool habitat (Hartman and Logan 2010; Mollenhauer 2011) and also show greater preference for pools with abundant cover such as large woody debris (Flebbe and Dollof 1995; Hartman and Logan 2010). It is not uncommon for brook trout streams in Pennsylvania to have reaches of 50 m or more with little to no pool habitat, resulting in a high degree of variation in the number of brook trout per unit area along a stream gradient. This was observed in the wide range of total numbers of brook trout found in a 50-m subreach in our study (Table 1). In particular, Montour Run had one 50-m reach with no brook trout captured, yet its true biomass over the 600-m total reach was near the average biomass for all streams sampled. Preferred brook trout habitat is patchily distributed along a stream reach and brook trout themselves are even more patchily distributed among preferred habitats.

We encourage other researchers and managers to conduct pilot studies prior to embarking on larger monitoring efforts to determine if the proposed sampling effort will be adequate to meet the data needs for management decisions. The methodology we employed in determining how the accuracy of density and biomass estimates changes with reach length can be used as a model for determining appropriate reach lengths to sample in other systems and for other species.

Please note: The Journal of Fish and Wildlife Management is not responsible for the content or functionality of any supplemental material. Queries should be directed to the corresponding author for the article.

Data S1. Brook trout representative reach data for fish and wetted widths.

Found at DOI: http://dx.doi.org/10.3996/032012-JFWM-027.S1 (62 KB XLSX).

Reference S1. Kaufmann PR, Robison EG. 1998. Physical habitat characterization. Section 7 in Lazorchak JM, Klemm DJ, Peck DV, editors. Environmental monitoring and assessment program- Surface waters: field operations and methods for measuring the ecological condition of wadeable streams. U.S. Environmental Protection Agency Report 620/R-94/004F, Cincinnati, Ohio.

Found at DOI: http://dx.doi.org/10.3996/032012-JFWM-027.S2; also available at http://www.epa.gov/emap/html/pubs/docs/groupdocs/surfwatr/field/Sec07.PDF (403 KB PDF).

Reference S2. Mason DA. 2009. Influence of stream channel morphology, stream habitat, and landscape features on brook trout densities in central Pennsylvanian streams. M.S. Thesis, Pennsylvania State University, University Park, PA.

Found at DOI: http://dx.doi.org/10.3996/032012-JFWM-027.S3 (851 KB PDF).

Reference S3. Mollenhauer R. 2011. Seasonal movement and habitat use of wild brook trout (Salvelinus fontinalis). M.S. Thesis, Pennsylvania State University, University Park, PA.

Found at DOI: http://dx.doi.org/10.3996/032012-JFWM-027.S4 (874 KB PDF).

Reference S4. White GC, Anderson DR, Burnham KP, Otis DL. 1982. Capture-recapture and removal methods for sampling closed populations. Los Alamos National Laboratory, LA-8787-NERP, Los Alamos, New Mexico.

Found at DOI: http://dx.doi.org/10.3996/032012-JFWM-027.S5 (19 MB PDF).

We would like to thank Pennsylvania Fish and Boat Commission staff (Andy Leakey, Amidea Daniels, and Dave Nihart), Northeast Fishery Center staff (Rebecca Lynch, Steve Krum, Steve Davis, Lyndsie Wszola, Greg Jacobs, and Brian Layton), and research assistants from The Pennsylvania State University (Rebecca Wagner and David Stainbrook) for their help in electrofishing for brook trout. Funding for this study was provided by the Pennsylvania Fish and Boat Commission and the U.S. Fish and Wildlife Service, Northeast Fishery Center, Lamar, Pennsylvania. This manuscript was improved with the comments of Dr Mike Millard (U.S. Fish and Wildlife Service, Northeast Fishery Center), Julie Nieland (NOAA), Joan Trial (Maine Department of Marine Resources), Alan Temple (U.S. Fish and Wildlife Service, National Conservation Training Center) and an anonymous reviewer.

Any use of trade, product, or firm names is for descriptive purposes only and does not imply endorsement by the U.S. Government.