- Email: [email protected]
Designing long-term fish community assessments in connecting channels: Lessons from the Saint Marys River
JGLR-00704; No. of pages: 8; 4C: Journal of Great Lakes Research xxx (2014) xxx–xxx
Contents lists available at ScienceDirect
Journal of Great Lakes Research journal homepage: www.elsevier.com/locate/jglr
Designing long-term fish community assessments in connecting channels: Lessons from the Saint Marys River Jeffrey S. Schaeffer a,⁎, Mark W. Rogers a, David G. Fielder b, Neal Godby c, Anjanette Bowen d, Lisa O'Connor e, Josh Parrish f, Susan Greenwood g, Stephen Chong g, Greg Wright g a
U. S. Geological Survey Great Lakes Science Center, 1451 Green Road, Ann Arbor, MI 48105, USA Michigan Department of Natural Resources and Environment, Alpena Fisheries Research Station, 160 E. Fletcher St., Alpena, MI 49707, USA Michigan Department of Natural Resources and Environment, Northern Lake Huron Management Unit, 1732 W. M-32, Gaylord, MI 49735, USA d U. S. Fish and Wildlife Service, Alpena Fish and Wildlife Conservation Office, 24 W. Fletcher St., Alpena, MI 49707, USA e Fisheries and Oceans Canada, Great Lakes Laboratory for Fisheries and Aquatic Sciences, 1 Canal Drive, Sault Ste. Marie, ON P6A 6W4, Canada f Ontario Ministry of Natural Resources, 1235 Queen St., Sault Ste. Marie, ON P6A 2E5, Canada g Chippewa/Ottawa Resource Authority, 179 W. Three Mile Rd., Sault Ste. Marie, MI 49783, Canada b c
a r t i c l e
i n f o
Article history: Received 1 March 2013 Accepted 13 February 2014 Available online xxxx Communicated by Edward Roseman Index words: Saint Marys River Long-term data Cisco Northern pike Walleye Yellow perch
a b s t r a c t Long-term surveys are useful in understanding trends in connecting channel fish communities; a gill net assessment in the Saint Marys River performed periodically since 1975 is the most comprehensive connecting channels sampling program within the Laurentian Great Lakes. We assessed efficiency of that survey, with intent to inform development of assessments at other connecting channels. We evaluated trends in community composition, effort versus estimates of species richness, ability to detect abundance changes for four species, and effects of subsampling yellow perch catches on size and age-structure metrics. Efficiency analysis revealed low power to detect changes in species abundance, whereas reduced effort could be considered to index species richness. Subsampling simulations indicated that subsampling would have allowed reliable estimates of yellow perch (Perca flavescens) population structure, while greatly reducing the number of fish that were assigned ages. Analyses of statistical power and efficiency of current sampling protocols are useful for managers collecting and using these types of data as well as for the development of new monitoring programs. Our approach provides insight into whether survey goals and objectives were being attained and can help evaluate ability of surveys to answer novel questions that arise as management strategies are refined. Published by Elsevier B.V. on behalf of International Association for Great Lakes Research.
Introduction Long-term surveys are an essential component of fisheries management in the Laurentian Great Lakes that allow examination of temporal changes in fish communities and exploration of mechanisms underlying observed changes. The availability of multi-decade annual surveys has been an asset for documenting a variety of responses ranging from community change via biological invasion (e.g., Bunnell et al., 2006; Riley et al., 2008) to trends in physiological condition (e.g., Madenjian et al., 2000). Long-term data have also been used to develop predictive models of year class strength (e.g., Kocovsky et al., 2010), determine prey availability for stocked piscivores (e.g., Rand et al., 1993), and as inputs for setting harvest quotas (e.g., WTG, 2012). While long-term data have proven useful throughout the Great Lakes, nearly all standardized
⁎ Corresponding author. E-mail address: [email protected] (J.S. Schaeffer).
monitoring programs have focused on lacustrine components of the system with data from connecting channels being comparatively scarce. The connecting channels of the Laurentian Great Lakes include the Saint Marys River (SMR), Saint Clair and Detroit River System (SCDRS), the Niagara River (NIA), and the Saint Lawrence River (SLR). These connecting channels are among the world's largest rivers, with all annual mean discharges exceeding 2100 m3/s and represent either between-lake connections (SMR, SCDRS, NIA) or outflow to the Atlantic Ocean (SLR). Connecting channels are recruitment sources for multiple lake-dwelling species (e.g., Roseman et al., 2007) and are centers of biological diversity (Edwards et al., 1989; Pratt and O'Connor, 2011). Surprisingly, little long-term monitoring has occurred within connecting channels in comparison to the lakes-proper, despite their perceived importance to ecosystem function. To date, only the SMR has received spatially broad (river-wide) monitoring of adult fish populations with periodic gill net surveys that occurred about once every six years during 1975–2009 or periodic creel surveys (1999–2001, 2005–2009) (Greenwood et al., 2011; Schaeffer et al., 2011). The Great Lakes ecosystem depends partially on the health of connecting channels (Great
http://dx.doi.org/10.1016/j.jglr.2014.03.002 0380-1330/Published by Elsevier B.V. on behalf of International Association for Great Lakes Research.
Please cite this article as: Schaeffer, J.S., et al., Designing long-term fish community assessments in connecting channels: Lessons from the Saint Marys River, J Great Lakes Res (2014), http://dx.doi.org/10.1016/j.jglr.2014.03.002
2
J.S. Schaeffer et al. / Journal of Great Lakes Research xxx (2014) xxx–xxx
Lakes Restoration Initiative, 2010), and thus, increased monitoring in connecting channels is desirable. The SMR time series provides an opportunity to assess the efficiency of the most comprehensive connecting channels sampling program within the basin for informing the design of long-term monitoring programs at other connecting channels. The SMR typifies the challenges of conducting long-term assessments in connecting channels. The SMR flows for 112 km between Lakes Superior and Huron and is comprised of lotic, lentic, and wetland habitat complexes throughout the system (Ripley et al., 2011). Along the river's continuum, there are a series of natural lakes and embayments that provide distinct and unique habitats. Due to river length and habitat heterogeneity, the initial sampling during 1975 resulted in the establishment of 32 fixed-site gill net stations to provide broad spatial coverage and encompass available habitats (Gebhardt et al., 2002). Subsequent surveys followed the original design, but the survey was broadened to 45 sites in 1995 using a multi-agency approach (Fielder, 2002, Fig. 1). All surveys provided species composition, size structure, and relative abundance data. From 1995 forward, age structures were collected from several target species (i.e., angler-preferred), most notably walleye and yellow perch. In addition, angler creel surveys were conducted during most years after 1999 (Greenwood et al., 2011). The SMR gill net survey was initiated to provide a broad community assessment for a region about which little was known (Schorfhaar, 1975). As the survey was repeated, it became to be considered a core component of a larger fish community assessment and a desirable source for stock assessment data (Gebhardt et al., 2002). The evolution and expansion of SMR monitoring ultimately resulted in insufficient
resources (i.e., time and funding) to successfully complete both gill net and creel surveys within the same year. During 2006, the only year when both surveys were performed, Schaeffer et al. (2011) reported that there were substantial inconsistencies between the two surveys for northern pike (Esox lucius) and cisco (Coregonus artedii). Furthermore, gill net survey frequency gaps were large enough (i.e., up to 8 years) to allow individual year classes for some species to pass through the population without detection. While the SMR gill net survey has followed standard protocols and not experienced serious changes (e.g., shifts in gill net materials), it has been confronted with a new dilemma: can the historical survey effort meet current resource management's data needs, and could the design be modified to meet those needs with less effort? We examined survey data collected between 1975 and 2009 to answer these questions. Methods Gill net survey We analyzed data from seven gill net surveys performed during August, 1975–2009. Surveys occurred about every six years and occurred in 1975 (32 net sets), 1979 (32 net sets), 1987 (27 net sets), 1995 (53 net sets), 2002 (44 net sets), 2006 (34 net sets), and 2009 (43 net sets). Gill net sites spanned the system from above Sault Ste. Marie to the head of the outflow into Lake Huron near Drummond Island (Fig. 1). The most recent collection (2009) followed methods established in earlier surveys to allow complete temporal comparability
Fig. 1. Saint Marys River and location of gillnet sets made in August 2009. Surveys from previous years spanned either the same sites or sites within these locations.
Please cite this article as: Schaeffer, J.S., et al., Designing long-term fish community assessments in connecting channels: Lessons from the Saint Marys River, J Great Lakes Res (2014), http://dx.doi.org/10.1016/j.jglr.2014.03.002
J.S. Schaeffer et al. / Journal of Great Lakes Research xxx (2014) xxx–xxx
3
among years from 27 to 53, with a mean of 37 sets annually. Effort varied because local conditions sometimes prevented sampling, and data from eight net sets in 2006 were excluded because catch processing deviated from assessment protocols. Nets were fished overnight on the bottom for all surveys for approximately 24 h. Fish from each mesh size within each net lift were identified, measured, and counted. Length data from surveys conducted between 1975 and 1987 no longer exist, but for surveys conducted between 1995 and 2009 all fish were measured (mm) and weighed (g). Structures (i.e., cleithra, scales, spines) were collected for age estimation from target species that supported the majority of harvest, which included: cisco, northern pike, salmonines (primarily Pacific salmon Oncorhynchus spp.), smallmouth bass (Micropterus dolomieu), walleye (Sander vitreus), and yellow perch. In the case of yellow perch, either spines or scales were used for age estimation depending on agency protocol. Analysis Fig. 2. Total annual mean catch-per-effort (CPE, number per gillnet set) and species composition in gill net surveys, Saint Marys River, 1975–2009. Other species category includes mean pooled CPE of up to 25 species.
(Fielder and Waybrant, 1998; Fielder et al., 2003; Gehbardt et al., 2002; Grimm, 1989; Miller, 1981; Schorfhaar, 1975). All surveys used multifilament nylon gill nets and from 1995 onward were performed by a multi-agency consortium (Fielder, 2002). Survey nets in 1975, 1979, 1987, and 1995 were 122 m in length and 1.8 m high, with four mesh sizes in 30.5 m panels: 50.8 mm, 63.5 mm, 76.2 mm and 114.3 mm stretched mesh (measured). During 2002, 2006, and 2009, nets had the same depth, but were lengthened to 304.8 m by adding additional 30.5 m panels of 38.1 mm, 88.9 mm, 101.6 mm, 127.0 mm, 139.7 mm, and 152.4 mm stretch measure mesh. Catch-per-effort (CPE) was expressed as total number of each species per net set across only those mesh sizes consistently fished each year to ensure comparability among years. Arithmetic mean CPE's for each species were calculated from all individual sets fished within a year. Number of sets varied
We assessed differences in species-specific CPE among years. Analysis of variance (ANOVA) was performed on relative abundance data at the significance level of alpha = 0.05 and followed the methods of Sokal and Rohlf (1981). CPE data were log transformed to meet ANOVA assumptions of normality variance homogeneity, and Tukey's test was used to examine post-hoc CPE differences among years for individual target species. Analysis was performed with SAS 9.2 computer software (SAS, 2007). We used 2009 gill net data to evaluate the power to detect differences in between two estimates of CPE for target species (i.e., cisco, northern pike, walleye, and yellow perch). We estimated the statistical power (ability to detect a change when one has occurred) to detect small, medium, and large changes (± 20%, ± 50%, and ± 80%, respectively; Cohen, 1988) in relative abundance across a range of gill net effort. Power was estimated using a z-approximation to the noncentral-t for a two-tailed test of differences between two means (Gerow, 2007). We also estimated the increase in number of species captured with
Table 1 Mean CPE (catch per 305 m overnight set), capture frequency, mean length, and length sample size (Nl), for fish species caught in 45 305-m gill net sets deployed during 2009, St. Marys River, Michigan-Ontario. Fish species ranked in order of descending CPE. Common name
Scientific name
Mean CPE
Capture frequency (N)
Mean length (mm)
N
Yellow perch White sucker Cisco Walleye Rock bass Brown bullhead Northern pike Round whitefish Smallmouth bass Longnose sucker Lake whitefish Rainbow smelt Redhorse spp. Freshwater drum Pumpkinseed White bass Silver redhorse Alewife Lake trout Chinook salmon Burbot Channel catfish Lake sturgeon Shorthead redhorse Black bullhead White perch Rainbow trout Sea lamprey Lake chub
Perca flavescens Catostomus commersoni Coregonus artedi Sander vitreus Ambloplites rupestris Ictalurus nebulosis Esox lucius Prosopium cylindricum Micropterus dolomieu Catostomus catostomus Coregonus clupeaformis Osmerus mordax Moxostoma spp. Aplodinotus grunniens Lepomis gibbosus Morone chrysops Moxostoma anisurum Alosa pseudoharengus Salvelinius namaycush Oncorhynchus tschawytscha Lota lota Ictalurus punctatus Acipenser fulvescens Moxostoma macrolepidotum Ictaluris melas Morone americana Oncorhynchus mykiss Petromyzon marinus Couesius plumbeus
38.06 18.48 6.79 5.09 4.23 1.93 1.83 1.81 1.76 1.65 1.51 0.86 0.60 0.41 0.39 0.30 0.30 0.23 0.16 0.15 0.13 0.11 0.11 0.07 0.05 0.05 0.02 0.02 0.02
36 34 16 31 27 7 21 7 25 11 12 9 7 10 5 5 5 4 2 2 5 5 2 3 2 2 1 1 1
179.6 388.9 315.9 440.2 167.6 272.1 539.9 385.4 313.1 366.1 411.3 123.7 394.1 453.6 145.9 335.4 517.5 152.4 633.7 141.0 534.0 538.0 734.4 485.0 258.0 186.6 – 345.0 155.0
1637 795 292 219 182 83 79 78 76 71 65 37 26 18 17 13 13 10 7 7 6 5 5 3 2 2 1 1 1
Please cite this article as: Schaeffer, J.S., et al., Designing long-term fish community assessments in connecting channels: Lessons from the Saint Marys River, J Great Lakes Res (2014), http://dx.doi.org/10.1016/j.jglr.2014.03.002
4
J.S. Schaeffer et al. / Journal of Great Lakes Research xxx (2014) xxx–xxx
Fig. 3. Relationship between number of species collected and number of gill net sets, Saint Marys River, 1975–2009. Dashed and dotted lines are 80% and 90% of total species collected, respectively. Error bars are empirical 95% confidence intervals from 1000 bootstrap resamples of data.
increased effort (i.e., species accumulation curve) using a jackknife procedure with random data entry using the Vegan Community Ecology Package (Oksanen et al., 2013) in R (R Development Core Team, 2013). Lastly, we used 2009 yellow perch data to evaluate the effects of subsampling on estimates of the age and size structure of the population as an alternative to the current practice of aging all fish collected. We simulated subsampling with replacement (i.e., bootstrap; Dixon, 1993) that ranged from retaining 5 yellow perch per cm group to 25 yellow perch per cm group for subsequent age estimation. We repeated the
simulation 1000 times for each subsampling regime and assumed that all fish from length bins where catches were less than the subsample number would be assigned ages. For each subsample replicate, we used size-at-age information to create an age–length key and assigned ages to fish not selected in the subsample following methods described by Isermann and Knight (2005). Subsampling simulations were constructed in R and used the FSA package (Ogle, 2012) for constructing age-keys. Our goal was to determine if survey efficiency could be increased via subsampling relative to the total number that was assigned
Please cite this article as: Schaeffer, J.S., et al., Designing long-term fish community assessments in connecting channels: Lessons from the Saint Marys River, J Great Lakes Res (2014), http://dx.doi.org/10.1016/j.jglr.2014.03.002
J.S. Schaeffer et al. / Journal of Great Lakes Research xxx (2014) xxx–xxx
5
Fig. 4. Relationship between statistical power to detect differences using a t-test and number of gill net sets for four target species and three levels of abundance increase, Saint Marys River, 2009.
ages in 2012 (i.e., 1233 individual fish). We calculated the coefficient of variation in estimated numbers at age and estimated mean length at age from the 1000 simulations for each subsample sample size. We also calculated the coefficient of variation in the slope of catch curves (see Ricker, 1975) from the 1000 simulations for each subsample sample size. Catch curve slopes resulted from linear regressions of logtransformed numbers at age from expanded age-key estimates (y dependent variable) at ages 4–8 (x independent variable). Results The SMR gill net survey suggests that community composition varied little among years. Yellow perch, white sucker, cisco, northern pike, rock bass (Ambloplites rupestris), and walleye were always among the 10 most common species captured and always comprised a large proportion of all individuals captured (72–91%) (Fig. 2). The only major temporal difference we detected was that northern pike were less abundant during 2002, 2006, and 2009 compared with other years (Tukey's multiple comparison, P b 0.05). The 2009 gill net survey captured 29 species across 43 sites; yellow perch and white sucker (Catostomus commersoni) were the most common species and comprised about 64% of all individuals captured (Table 1). Community composition during 2009 was similar to other surveys in terms of species captured and relative abundance among species. Reduced gill net effort had little effect on numbers of species caught. Effort required to capture 80% of all species observed during surveys ranged from 11 to 21 net sets among years (Fig. 3). Our results showed that setting 25 gill nets per survey year would have captured 90% of the
species caught in full surveys except in 2002 when 32 net sets would have been required (Fig. 3). Reduced effort relative to previous survey years would have little impact on analyses of community composition or species richness. Using 2009 data, the SMR survey had low power to detect a 50% increase in relative abundance (0.2 for walleye and cisco, 0.3 for yellow perch and northern pike), and power to detect an 80% increase in relative abundance was higher only for yellow perch and northern pike (0.6 and 0.5, respectively; Fig. 4). Detecting a 50% increase in relative abundance with a power equal to 0.5 would have required 100 net nights for yellow perch and northern pike, and more than 200 net nights for walleye and cisco. Sample sizes have resulted in good power (i.e., approximately 0.80) to detect 80% reductions in relative abundance between two samples for walleye, northern pike, and yellow perch, but low power for cisco (approximately 0.50) (Fig. 5). Smaller changes in relative abundance among species would not likely be detectable without more gill netting effort. Thus, our power analysis indicated that the present gill net survey does not provide a strong ability to detect significant relative abundance changes for target species. During 2009, 1233 yellow perch were assigned ages for obtaining estimates of mean length-at-age. Simulations indicated that subsampling of these fish would have little influence on estimates. At 5 fish per cm group, CVs in estimated number at age varied from 50 to 100% excluding age 8 which had a CV of zero (no age-8 fish were captured). As numbers of fish increased, the range of CVs in estimated number at age decreased and approached 20% across all ages. Coefficients of variation in estimated mean length at age were close to 10% when 10 fish per cm group were assigned ages and less than 10% at larger subsamples
Please cite this article as: Schaeffer, J.S., et al., Designing long-term fish community assessments in connecting channels: Lessons from the Saint Marys River, J Great Lakes Res (2014), http://dx.doi.org/10.1016/j.jglr.2014.03.002
6
J.S. Schaeffer et al. / Journal of Great Lakes Research xxx (2014) xxx–xxx
Fig. 5. Relationship between statistical power to detect differences using a t-test and number of gill net sets for four target species and three levels of abundance decrease, Saint Marys River, 2009.
per cm group. Coefficients of variation in the slope estimates from simulation catch curves were smaller than 10% across the subsampling scenarios we evaluated. Thus, subsampling would have had little effect on mortality estimates. Reducing the number of fish retained to assign ages had the potential to drastically decrease age estimation effort relative to the realized effort in 2009 (Table 2) and using age-key extrapolations would allow subsample data to reflect assessment metrics from larger samples. Thus, assigning ages to a subsample of yellow perch in 2009 would have reduced laboratory effort with little cost to stock assessment metrics. Discussion Great Lakes connecting channels are important ecologically because despite over two centuries of invasion, harvest, habitat loss, and
changes in water quality they retain a semblance of a native fish community. Monitoring them may be vital for both stock preservation and to help maintain a source of recruits for lake populations (Edwards et al., 1989; Schaeffer et al., 2011). To date, the SMR is the only connecting channel with long-term fish community data, but it illustrates that managers are faced with difficult choices when designing and implementing these types of surveys. The ability to detect changes in diversity and relative abundance is often the primary goal of monitoring programs (Yoccoz et al., 2001) and programs strive to maximize efficiency in reaching that goal. Our analysis demonstrated options for improving SMR survey efficiency and highlighted trade-offs associated with varying survey effort. The current SMR assessment program produces multiple valuable metrics for management (Schaeffer et al., 2011), but our results suggest that more intensive effort would be required to detect 50% abundance
Table 2 Variability in estimated numbers at age resulting from simulated subsampling of a fixed number of yellow perch per length bin (i.e., 1 cm) for age estimation and subsequent age–length key expansion. Data from 2009 survey only, parentheses indicate 95% confidence intervals from 1000 nonparametric bootstrap resamples without replacement. Number subsampled
5/cm 10/cm 15/cm 20/cm 25/cm
Coefficient of variation (*100) in number at age Age 1
2
3
4
5
6
7
8
78.9 44.5 41.2 32.6 18.6
23.3 16.4 13.1 10.9 9.8
16.0 11.4 9.1 7.6 6.7
17.5 11.5 9.0 7.8 7.0
54.0 34.3 28.2 23.7 19.8
64.0 41.8 33.2 27.8 24.0
89.2 54.1 45.7 33.3 19.4
0.0 0.0 0.0 0.0 0.0
Median number aged
CV of slope estimate
101 177 243 306 306
6.59 4.69 3.77 3.11 2.14
Please cite this article as: Schaeffer, J.S., et al., Designing long-term fish community assessments in connecting channels: Lessons from the Saint Marys River, J Great Lakes Res (2014), http://dx.doi.org/10.1016/j.jglr.2014.03.002
J.S. Schaeffer et al. / Journal of Great Lakes Research xxx (2014) xxx–xxx
changes between 2 samples in our four target species. Power to detect relative abundance changes was generally low and varied among target species. With historical gill net effort, power to detect large changes in relative abundance (i.e., 80%) never exceeded 0.5 and effort required to improve statistical power is not likely feasible due to time and worker limitations. Our power analysis for 2009 data indicated that despite having low power, the sampling program was comparatively more efficient for monitoring northern pike, walleye and yellow perch relative abundances than for cisco. Analyses evaluating sample size requirements for detecting relative abundance changes commonly result in exorbitantly high estimates (Quist et al., 2006). For example, Paukert (2004) estimated it would often require over 1000 electrofishing and 500 trammel net samples to detect 25% relative abundance changes in two Colorado River sucker species with power ≥ 0.8. However, power can also be increased by reducing variability and biases. Biased sampling can be useful for identifying population trends if biases are consistent among sampling periods (Reynolds, 1996), for example using consistent gill net mesh sizes in the SMR survey. However, in the SMR survey there may be other factors such as sites annually selected for sampling (with their potential among-site differences in CPE) that could lower power to detect among-year differences. Removing variability is commonly accomplished via a stratified design, but partitioning variance within statistical analysis can prove beneficial for understanding influences on power and survey design efficiency (e.g., Wagner et al., 2009). Future analyses of the SMR monitoring program could use a multi-level analysis to consider site-dependent and time-dependent effects on population and community indices. Species accumulation curves indicated that SMR gill net effort should almost always be adequate for indexing species richness. It is important to recognize that species richness estimates from gillnet sampling are biased relative to the whole community's true composition (Hamley, 1975), and this is true for the SMR gill net survey as well (Schaeffer et al., 2011). However, the SMR survey is the most spatially comprehensive and continuous data source for the system and does allow for determination of community composition in dominant native species (Schaeffer et al., 2011). Thus, the current sampling protocol is useful for evaluating community stability, and we showed that the survey would continue to be useful with reduced effort. Random subsampling of total catch is commonly used to make inferences about the total population age composition and reduces processing times or number of fish collected (Bettoli and Miranda, 2001). Our results revealed opportunities to increase efficiency via subsampling the yellow perch catch for estimating population size structure and age composition. Similarly, Coggins et al. (2013) concluded that subsampling with 10 fish per length bin would achieve accuracy and precision of multiple population assessment metrics, including growth and mortality, based on simulations across multiple life histories. Coggins et al. (2013) suggested subsampling 10 fish per length bin as a guideline for most studies. An additional advantage to decreasing sample size for the SMR assessment is that it could allow for continuity in aging structures, whereas currently ages are estimated using both spines and scales. We restricted our analyses of reduced age and growth sampling to yellow perch during 2009. We chose to use 2009 data because yellow perch sample sizes were large and age estimation effort was extensive. Further simulations would be required to determine if subsampling is feasible for other species of interest that are less frequently captured. One intriguing consideration resulting from our analysis is that although reduced effort would exacerbate current power limitations for comparing catch rates, lower effort could still provide an index of species richness and samples for target species population metrics (i.e., age and size structure). Thus, reducing field and laboratory effort might allow the gill net survey to be annualized. Annual surveys could allow managers to have a better understanding of population characteristics such as cohort-specific analyses that would obviate assumptions of equal recruitment. Tetzlaff et al. (2011) reported that following cohorts through time provided a more reliable index of year-class strength
7
compared to using catch curve residuals. Furthermore, in some cases surveys have been separated temporally such that percid year classes could pass through the population entirely without detection. Annual surveys in connecting channels would also be consistent with most other Great Lakes monitoring programs. The SMR data provide a tool for informing future long-term survey designs. Agencies considering establishing long-term surveys on other connecting channels might consider using a similar approach, starting with identical gill nets fished at multiple sites to obtain broad spatial coverage. While using gill nets for monitoring community composition and relative abundance has inherent biases, employing similar methods across connecting channels would facilitate cross-system comparisons. Field experiments such as localized mark-recapture studies could estimate species-specific detection probabilities among habitat types to further improve value of gill net survey data. The present gill net survey is likely an insufficient stock assessment tool, but adequate as a fish community assessment tool. If a stock assessment is desired, then gill net effort needs to be increased 2.5 to 5 times. If it is the latter, then reduced effort could be considered for providing community composition data and population metrics for specific species. Managers could avoid that choice by reducing survey effort, but performing directed and well-designed stock assessments that might more efficiently collect unbiased data from target species, using gears that might increase sample size and reduce by-catch. The SMR assessment was designed with broad objectives to gain community-wide information on a largely unknown system; as the survey was repeated there became a tension between a need for data continuity versus a need for species-specific information as new management strategies evolved. This situation is common in long-term monitoring programs, many of which began in an exploratory fashion and sometimes without specific objectives. We suggest that analyses of statistical power and efficiency of current sampling protocols are a useful step that benefits managers collecting and using these types of data for decision making as well as for the development of new monitoring programs. Evaluating monitoring program data robustness provides insight into whether survey goals and objectives are being attained, and they can help evaluate the ability of surveys to answer novel questions that arise as management strategies are refined.
Acknowledgments We thank the many dedicated individuals who participated in field sampling. The Ontario Ministry of Natural Resources upper Great Lakes management unit and U.S. Fish and Wildlife Service personnel processed structures and assigned ages. This is contribution 1831 of the Great Lakes Science Center.
References Bettoli, P.W., Miranda, L.E., 2001. A cautionary note about estimating mean length at age with subsampled data. N. Am. J. Fish. Manag. 21, 425–428. Bunnell, D.B., Madenjian, C.P., Claramunt, R.M., 2006. Long-term changes of the. Lake Michigan fish community following the reduction of exotic alewife (Alosa pseudoharengus). Can. J. Fish. Aquat. Sci. 63, 2434–2446. Coggins, L.G., Gwinn, D.C., Allen, M.S., 2013. Evaluation of age–length key sample sizes required to estimate fish total mortality and growth. Trans. Am. Fish. Soc. 142, 832–840. Cohen, J., 1988. Statistical Power Analysis for the Behavioral Sciences, 2nd ed. Laurence Earlbaum Associates, Hillsdale, NJ. Development Core Team, R., 2013. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. Dixon, P.M., 1993. The bootstrap and the jackknife: describing the precision of ecological indices. In: Scheiner, S.M., Gurevitch, J. (Eds.), Design and Analysis of Ecological Experiments. Chapman and Hall, New York, pp. 290–318. Edwards, C.J., Hudson, P.L., Duffy, W.G., Nepszy, S.J., McNabb, C.D., et al., 1989. Hydrological, morphometrical, and biological characteristics of the connecting rivers of International Great Lakes: a review. Proc. Int. Large River Symp. Can. Spec. Publ. Fish. Aquat. Sci. 106, 240–264. Fielder, D., 2002. St. Marys River Task Group: the St. Marys River and challenges of multijurisdictional fisheries. Fisheries 27, 32–34.
Please cite this article as: Schaeffer, J.S., et al., Designing long-term fish community assessments in connecting channels: Lessons from the Saint Marys River, J Great Lakes Res (2014), http://dx.doi.org/10.1016/j.jglr.2014.03.002
8
J.S. Schaeffer et al. / Journal of Great Lakes Research xxx (2014) xxx–xxx
Fielder, D.G., Waybrant, J.R., 1998. Fish population surveys of St. Marys River, 1975–95, and recommendations for management. Fisheries Research Report 2048Michigan Department of Natural Resources, Ann Arbor. Fielder, D.G., Borgeson, D.J., Bowen, A.K., Koproski, S.R., Greenwood, S.J., Wright, G.M., 2003. Population Dynamics of the St. Marys River Fish Community 1975–2002. Great Lakes Fishery Commission, Miscellaneous publication, Ann Arbor http:// www.glfc.org/lakecom/lhc/SMR2002rpt.pdf. Accessed July, 2012. Gebhardt, K., Fielder, D., Greenwood, S., Robbins, H., Sutton, T., 2002. St. Marys River Fisheries Assessment Plan. Great Lakes Fisheries Commission (http://www.glfc.org/ lakecom/lhc/smrfinal.pdf. Ann Arbor. Accessed July, 2012). Gerow, K.G., 2007. Power and sample size estimation techniques for fisheries management: assessment and a new computation tool. N. Am. J. Fish. Manag. 27, 397–404. Greenwood, S.J., Fielder, D.G., Godby, N., Kolb, T., 2011. A Synthesis of Sport Fishing Activity in the St. Marys River, 1999–2001 and 2005–2009. Great Lakes Fishery. Commission. Miscellaneous Publication (http://www.glfc.org/lakecom/lhc/SMRHarvestRpt19992009FINAL.pdf. Accessed July 2012.). Grimm, K.S., 1989. A fisheries survey of the St. Marys River, Chippewa County, August– October, 1987. Mich. Dept. Nat. Res. Tech. Rep. No. Hamley, J.M., 1975. Review of gillnet selectivity. J. Fish. Res. Board Can. 32 (11), 1943–1969. Isermann, D.A., Knight, C.T., 2005. A computer program for age–length keys incorporating age assignment to individual fish. N. Am. J. Fish. Manag. 25, 1153–1160. Kocovsky, P.M., Stapanian, M.A., Knight, C.T., 2010. Night sampling improves indices used for management of yellow perch in Lake Erie. Fish. Manag. Ecol. 17 (1), 10–18. Lakes Restoration Initiative, Great, 2010. Great Lakes Restoration Initiative Action Plan FY 2010–2014. (http://glri.us/pdfs/glri_actionplan.pdf. Accessed June, 2012). Madenjian, C.P., Elliot, R.F., DeSorcie, T.J., Stedman, R.M., O'Connor, D.V., Rottiers, D.V., 2000. Lipid concentrations in Lake Michigan fishes: seasonal, spatial, ontogenetic, and long-term trends. J. Great Lakes Res. 26, 427–444. Miller, B.R., 1981. A 1979 fisheries survey of the St. Marys River system in Chippewa County, and comparisons with 1975 results. Mich. Dept. Nat. Res. Tech. Rep. No. Ogle, D., 2012. Functions to Support Simple Fisheries Stock Assessment Methods. (http:// www.rforge.net/src/contrib/Documentation/FSA.pdf. Accessed June, 2012.). Oksanen, J., Guillaume Blanchet, F., Kindt, R., Legendre, P., Minchin, P.R., O'Hara, R.B., Simpson, G.L., Solymos, P., Stevens, M.H.H., Wagner, H., 2013. Vegan: Community Ecology Package. (http://vegan.r-forge.r-project.org. Accessed March, 2012.). Paukert, C.P., 2004. Comparison of electrofishing and trammel netting variability for sampling native fishes. J. Fish Biol. 65, 1643–1652.
Pratt, T.C., O'Connor, L.M., 2011. An assessment of the health and historical changes of the nearshore fish community of the St. Marys River. J. Great Lakes Res. 37 (Suppl. 2), 61–69. Quist, M.C., Gerow, K.G., Bower, M.R., Hubert, W.A., 2006. Random versus fixed-site sampling when monitoring relative abundance of fishes in headwater streams of the Upper Colorado River basin. N. Am. J. Fish. Manag. 26, 1011–1019. Rand, P.S., Stewart, D.J., Seelbach, P.W., Jones, M.L., Wedge, L.R., 1993. Modeling steelhead population energetic in Lakes Michigan and Ontario. Trans. Am. Fish. Soc. 122, 977–1001. Reynolds, J.B., 1996. Electrofishing, In: Murphy, B.R., Willis, D.W. (Eds.), Fisheries Techniques, 2nd edition. American Fisheries Society, Bethesda, Maryland, pp. 221–253. Ricker, W.E., 1975. Computation and interpretation of biological statistics of fish populations. Fish. Res. Board Can. Bull. 191. Riley, S.C., Roseman, E.F., Nichols, S.J., O'Brien, T.P., Kiley, C.S., Schaeffer, J.S., 2008. Deepwater demersal fish community collapse in Lake Huron. Trans. Am. Fish. Soc. 137, 1879–1890. Ripley, M.P., Arbic, B., Zimmerman, G., 2011. Environmental history of the St. Marys River. J. Great Lakes Res. 37 (Suppl. 2), 5–11. Roseman, E.F., Kennedy, G.W., Boase, J., Manny, B.A., Todd, T.N., Stott, W., 2007. Evidence of lake whitefish spawning in the Detroit River: implications for habitat and population recovery. J. Great Lakes Res. 33, 397–406. SAS Institute Inc., 2007. OnlineDoc 9.1.2. SAS Institute, Inc., Cary, NC. Schaeffer, J.S., et al., 2011. Long-term trends in the St. Marys River open water fish community. J. Great Lakes Res. 37 (Suppl. 2), 70–79. Schorfhaar, R., 1975. A Fisheries Survey of the St. Marys River, 1975. Michigan Department of Natural Resources, Fisheries Division, Marquette. (Unpublished). Sokal, R.R., Rohlf, F.J., 1981. Biometry, Second Edition. W. H Freeman and Company, San Francisco, California. Tetzlaff, J.C., Catalano, M.J., Allen, M.S., Pine III, W.E., 2011. Evaluation of two methods for indexing fish year-class strength: catch-curve residuals and cohort method. Fish. Res. 109, 303–310. Wagner, T., Vandergoot, C.S., Tyson, J., 2009. Evaluating the power to detect temporal trends in fishery-independent surveys: a case study based on gill nets set in the Ohio waters of Lake Erie for walleyes. N. Am. J. Fish. Manag. 29, 805–816. WTG (Walleye Task Group), 2012. Report for 2011 by the Lake Erie Walleye Task Group. Annual Report to the Lake Erie Committee. Great Lakes Fishery Commission, Windsor, Ontario. Yoccoz, N.G., Nichols, J.D., Boulinier, T., 2001. Monitoring of biological diversity in space and time. Trends Ecol. Evol. 16, 446–453.
Please cite this article as: Schaeffer, J.S., et al., Designing long-term fish community assessments in connecting channels: Lessons from the Saint Marys River, J Great Lakes Res (2014), http://dx.doi.org/10.1016/j.jglr.2014.03.002
Contents lists available at ScienceDirect
Journal of Great Lakes Research journal homepage: www.elsevier.com/locate/jglr
Designing long-term fish community assessments in connecting channels: Lessons from the Saint Marys River Jeffrey S. Schaeffer a,⁎, Mark W. Rogers a, David G. Fielder b, Neal Godby c, Anjanette Bowen d, Lisa O'Connor e, Josh Parrish f, Susan Greenwood g, Stephen Chong g, Greg Wright g a
U. S. Geological Survey Great Lakes Science Center, 1451 Green Road, Ann Arbor, MI 48105, USA Michigan Department of Natural Resources and Environment, Alpena Fisheries Research Station, 160 E. Fletcher St., Alpena, MI 49707, USA Michigan Department of Natural Resources and Environment, Northern Lake Huron Management Unit, 1732 W. M-32, Gaylord, MI 49735, USA d U. S. Fish and Wildlife Service, Alpena Fish and Wildlife Conservation Office, 24 W. Fletcher St., Alpena, MI 49707, USA e Fisheries and Oceans Canada, Great Lakes Laboratory for Fisheries and Aquatic Sciences, 1 Canal Drive, Sault Ste. Marie, ON P6A 6W4, Canada f Ontario Ministry of Natural Resources, 1235 Queen St., Sault Ste. Marie, ON P6A 2E5, Canada g Chippewa/Ottawa Resource Authority, 179 W. Three Mile Rd., Sault Ste. Marie, MI 49783, Canada b c
a r t i c l e
i n f o
Article history: Received 1 March 2013 Accepted 13 February 2014 Available online xxxx Communicated by Edward Roseman Index words: Saint Marys River Long-term data Cisco Northern pike Walleye Yellow perch
a b s t r a c t Long-term surveys are useful in understanding trends in connecting channel fish communities; a gill net assessment in the Saint Marys River performed periodically since 1975 is the most comprehensive connecting channels sampling program within the Laurentian Great Lakes. We assessed efficiency of that survey, with intent to inform development of assessments at other connecting channels. We evaluated trends in community composition, effort versus estimates of species richness, ability to detect abundance changes for four species, and effects of subsampling yellow perch catches on size and age-structure metrics. Efficiency analysis revealed low power to detect changes in species abundance, whereas reduced effort could be considered to index species richness. Subsampling simulations indicated that subsampling would have allowed reliable estimates of yellow perch (Perca flavescens) population structure, while greatly reducing the number of fish that were assigned ages. Analyses of statistical power and efficiency of current sampling protocols are useful for managers collecting and using these types of data as well as for the development of new monitoring programs. Our approach provides insight into whether survey goals and objectives were being attained and can help evaluate ability of surveys to answer novel questions that arise as management strategies are refined. Published by Elsevier B.V. on behalf of International Association for Great Lakes Research.
Introduction Long-term surveys are an essential component of fisheries management in the Laurentian Great Lakes that allow examination of temporal changes in fish communities and exploration of mechanisms underlying observed changes. The availability of multi-decade annual surveys has been an asset for documenting a variety of responses ranging from community change via biological invasion (e.g., Bunnell et al., 2006; Riley et al., 2008) to trends in physiological condition (e.g., Madenjian et al., 2000). Long-term data have also been used to develop predictive models of year class strength (e.g., Kocovsky et al., 2010), determine prey availability for stocked piscivores (e.g., Rand et al., 1993), and as inputs for setting harvest quotas (e.g., WTG, 2012). While long-term data have proven useful throughout the Great Lakes, nearly all standardized
⁎ Corresponding author. E-mail address: [email protected] (J.S. Schaeffer).
monitoring programs have focused on lacustrine components of the system with data from connecting channels being comparatively scarce. The connecting channels of the Laurentian Great Lakes include the Saint Marys River (SMR), Saint Clair and Detroit River System (SCDRS), the Niagara River (NIA), and the Saint Lawrence River (SLR). These connecting channels are among the world's largest rivers, with all annual mean discharges exceeding 2100 m3/s and represent either between-lake connections (SMR, SCDRS, NIA) or outflow to the Atlantic Ocean (SLR). Connecting channels are recruitment sources for multiple lake-dwelling species (e.g., Roseman et al., 2007) and are centers of biological diversity (Edwards et al., 1989; Pratt and O'Connor, 2011). Surprisingly, little long-term monitoring has occurred within connecting channels in comparison to the lakes-proper, despite their perceived importance to ecosystem function. To date, only the SMR has received spatially broad (river-wide) monitoring of adult fish populations with periodic gill net surveys that occurred about once every six years during 1975–2009 or periodic creel surveys (1999–2001, 2005–2009) (Greenwood et al., 2011; Schaeffer et al., 2011). The Great Lakes ecosystem depends partially on the health of connecting channels (Great
http://dx.doi.org/10.1016/j.jglr.2014.03.002 0380-1330/Published by Elsevier B.V. on behalf of International Association for Great Lakes Research.
Please cite this article as: Schaeffer, J.S., et al., Designing long-term fish community assessments in connecting channels: Lessons from the Saint Marys River, J Great Lakes Res (2014), http://dx.doi.org/10.1016/j.jglr.2014.03.002
2
J.S. Schaeffer et al. / Journal of Great Lakes Research xxx (2014) xxx–xxx
Lakes Restoration Initiative, 2010), and thus, increased monitoring in connecting channels is desirable. The SMR time series provides an opportunity to assess the efficiency of the most comprehensive connecting channels sampling program within the basin for informing the design of long-term monitoring programs at other connecting channels. The SMR typifies the challenges of conducting long-term assessments in connecting channels. The SMR flows for 112 km between Lakes Superior and Huron and is comprised of lotic, lentic, and wetland habitat complexes throughout the system (Ripley et al., 2011). Along the river's continuum, there are a series of natural lakes and embayments that provide distinct and unique habitats. Due to river length and habitat heterogeneity, the initial sampling during 1975 resulted in the establishment of 32 fixed-site gill net stations to provide broad spatial coverage and encompass available habitats (Gebhardt et al., 2002). Subsequent surveys followed the original design, but the survey was broadened to 45 sites in 1995 using a multi-agency approach (Fielder, 2002, Fig. 1). All surveys provided species composition, size structure, and relative abundance data. From 1995 forward, age structures were collected from several target species (i.e., angler-preferred), most notably walleye and yellow perch. In addition, angler creel surveys were conducted during most years after 1999 (Greenwood et al., 2011). The SMR gill net survey was initiated to provide a broad community assessment for a region about which little was known (Schorfhaar, 1975). As the survey was repeated, it became to be considered a core component of a larger fish community assessment and a desirable source for stock assessment data (Gebhardt et al., 2002). The evolution and expansion of SMR monitoring ultimately resulted in insufficient
resources (i.e., time and funding) to successfully complete both gill net and creel surveys within the same year. During 2006, the only year when both surveys were performed, Schaeffer et al. (2011) reported that there were substantial inconsistencies between the two surveys for northern pike (Esox lucius) and cisco (Coregonus artedii). Furthermore, gill net survey frequency gaps were large enough (i.e., up to 8 years) to allow individual year classes for some species to pass through the population without detection. While the SMR gill net survey has followed standard protocols and not experienced serious changes (e.g., shifts in gill net materials), it has been confronted with a new dilemma: can the historical survey effort meet current resource management's data needs, and could the design be modified to meet those needs with less effort? We examined survey data collected between 1975 and 2009 to answer these questions. Methods Gill net survey We analyzed data from seven gill net surveys performed during August, 1975–2009. Surveys occurred about every six years and occurred in 1975 (32 net sets), 1979 (32 net sets), 1987 (27 net sets), 1995 (53 net sets), 2002 (44 net sets), 2006 (34 net sets), and 2009 (43 net sets). Gill net sites spanned the system from above Sault Ste. Marie to the head of the outflow into Lake Huron near Drummond Island (Fig. 1). The most recent collection (2009) followed methods established in earlier surveys to allow complete temporal comparability
Fig. 1. Saint Marys River and location of gillnet sets made in August 2009. Surveys from previous years spanned either the same sites or sites within these locations.
Please cite this article as: Schaeffer, J.S., et al., Designing long-term fish community assessments in connecting channels: Lessons from the Saint Marys River, J Great Lakes Res (2014), http://dx.doi.org/10.1016/j.jglr.2014.03.002
J.S. Schaeffer et al. / Journal of Great Lakes Research xxx (2014) xxx–xxx
3
among years from 27 to 53, with a mean of 37 sets annually. Effort varied because local conditions sometimes prevented sampling, and data from eight net sets in 2006 were excluded because catch processing deviated from assessment protocols. Nets were fished overnight on the bottom for all surveys for approximately 24 h. Fish from each mesh size within each net lift were identified, measured, and counted. Length data from surveys conducted between 1975 and 1987 no longer exist, but for surveys conducted between 1995 and 2009 all fish were measured (mm) and weighed (g). Structures (i.e., cleithra, scales, spines) were collected for age estimation from target species that supported the majority of harvest, which included: cisco, northern pike, salmonines (primarily Pacific salmon Oncorhynchus spp.), smallmouth bass (Micropterus dolomieu), walleye (Sander vitreus), and yellow perch. In the case of yellow perch, either spines or scales were used for age estimation depending on agency protocol. Analysis Fig. 2. Total annual mean catch-per-effort (CPE, number per gillnet set) and species composition in gill net surveys, Saint Marys River, 1975–2009. Other species category includes mean pooled CPE of up to 25 species.
(Fielder and Waybrant, 1998; Fielder et al., 2003; Gehbardt et al., 2002; Grimm, 1989; Miller, 1981; Schorfhaar, 1975). All surveys used multifilament nylon gill nets and from 1995 onward were performed by a multi-agency consortium (Fielder, 2002). Survey nets in 1975, 1979, 1987, and 1995 were 122 m in length and 1.8 m high, with four mesh sizes in 30.5 m panels: 50.8 mm, 63.5 mm, 76.2 mm and 114.3 mm stretched mesh (measured). During 2002, 2006, and 2009, nets had the same depth, but were lengthened to 304.8 m by adding additional 30.5 m panels of 38.1 mm, 88.9 mm, 101.6 mm, 127.0 mm, 139.7 mm, and 152.4 mm stretch measure mesh. Catch-per-effort (CPE) was expressed as total number of each species per net set across only those mesh sizes consistently fished each year to ensure comparability among years. Arithmetic mean CPE's for each species were calculated from all individual sets fished within a year. Number of sets varied
We assessed differences in species-specific CPE among years. Analysis of variance (ANOVA) was performed on relative abundance data at the significance level of alpha = 0.05 and followed the methods of Sokal and Rohlf (1981). CPE data were log transformed to meet ANOVA assumptions of normality variance homogeneity, and Tukey's test was used to examine post-hoc CPE differences among years for individual target species. Analysis was performed with SAS 9.2 computer software (SAS, 2007). We used 2009 gill net data to evaluate the power to detect differences in between two estimates of CPE for target species (i.e., cisco, northern pike, walleye, and yellow perch). We estimated the statistical power (ability to detect a change when one has occurred) to detect small, medium, and large changes (± 20%, ± 50%, and ± 80%, respectively; Cohen, 1988) in relative abundance across a range of gill net effort. Power was estimated using a z-approximation to the noncentral-t for a two-tailed test of differences between two means (Gerow, 2007). We also estimated the increase in number of species captured with
Table 1 Mean CPE (catch per 305 m overnight set), capture frequency, mean length, and length sample size (Nl), for fish species caught in 45 305-m gill net sets deployed during 2009, St. Marys River, Michigan-Ontario. Fish species ranked in order of descending CPE. Common name
Scientific name
Mean CPE
Capture frequency (N)
Mean length (mm)
N
Yellow perch White sucker Cisco Walleye Rock bass Brown bullhead Northern pike Round whitefish Smallmouth bass Longnose sucker Lake whitefish Rainbow smelt Redhorse spp. Freshwater drum Pumpkinseed White bass Silver redhorse Alewife Lake trout Chinook salmon Burbot Channel catfish Lake sturgeon Shorthead redhorse Black bullhead White perch Rainbow trout Sea lamprey Lake chub
Perca flavescens Catostomus commersoni Coregonus artedi Sander vitreus Ambloplites rupestris Ictalurus nebulosis Esox lucius Prosopium cylindricum Micropterus dolomieu Catostomus catostomus Coregonus clupeaformis Osmerus mordax Moxostoma spp. Aplodinotus grunniens Lepomis gibbosus Morone chrysops Moxostoma anisurum Alosa pseudoharengus Salvelinius namaycush Oncorhynchus tschawytscha Lota lota Ictalurus punctatus Acipenser fulvescens Moxostoma macrolepidotum Ictaluris melas Morone americana Oncorhynchus mykiss Petromyzon marinus Couesius plumbeus
38.06 18.48 6.79 5.09 4.23 1.93 1.83 1.81 1.76 1.65 1.51 0.86 0.60 0.41 0.39 0.30 0.30 0.23 0.16 0.15 0.13 0.11 0.11 0.07 0.05 0.05 0.02 0.02 0.02
36 34 16 31 27 7 21 7 25 11 12 9 7 10 5 5 5 4 2 2 5 5 2 3 2 2 1 1 1
179.6 388.9 315.9 440.2 167.6 272.1 539.9 385.4 313.1 366.1 411.3 123.7 394.1 453.6 145.9 335.4 517.5 152.4 633.7 141.0 534.0 538.0 734.4 485.0 258.0 186.6 – 345.0 155.0
1637 795 292 219 182 83 79 78 76 71 65 37 26 18 17 13 13 10 7 7 6 5 5 3 2 2 1 1 1
Please cite this article as: Schaeffer, J.S., et al., Designing long-term fish community assessments in connecting channels: Lessons from the Saint Marys River, J Great Lakes Res (2014), http://dx.doi.org/10.1016/j.jglr.2014.03.002
4
J.S. Schaeffer et al. / Journal of Great Lakes Research xxx (2014) xxx–xxx
Fig. 3. Relationship between number of species collected and number of gill net sets, Saint Marys River, 1975–2009. Dashed and dotted lines are 80% and 90% of total species collected, respectively. Error bars are empirical 95% confidence intervals from 1000 bootstrap resamples of data.
increased effort (i.e., species accumulation curve) using a jackknife procedure with random data entry using the Vegan Community Ecology Package (Oksanen et al., 2013) in R (R Development Core Team, 2013). Lastly, we used 2009 yellow perch data to evaluate the effects of subsampling on estimates of the age and size structure of the population as an alternative to the current practice of aging all fish collected. We simulated subsampling with replacement (i.e., bootstrap; Dixon, 1993) that ranged from retaining 5 yellow perch per cm group to 25 yellow perch per cm group for subsequent age estimation. We repeated the
simulation 1000 times for each subsampling regime and assumed that all fish from length bins where catches were less than the subsample number would be assigned ages. For each subsample replicate, we used size-at-age information to create an age–length key and assigned ages to fish not selected in the subsample following methods described by Isermann and Knight (2005). Subsampling simulations were constructed in R and used the FSA package (Ogle, 2012) for constructing age-keys. Our goal was to determine if survey efficiency could be increased via subsampling relative to the total number that was assigned
Please cite this article as: Schaeffer, J.S., et al., Designing long-term fish community assessments in connecting channels: Lessons from the Saint Marys River, J Great Lakes Res (2014), http://dx.doi.org/10.1016/j.jglr.2014.03.002
J.S. Schaeffer et al. / Journal of Great Lakes Research xxx (2014) xxx–xxx
5
Fig. 4. Relationship between statistical power to detect differences using a t-test and number of gill net sets for four target species and three levels of abundance increase, Saint Marys River, 2009.
ages in 2012 (i.e., 1233 individual fish). We calculated the coefficient of variation in estimated numbers at age and estimated mean length at age from the 1000 simulations for each subsample sample size. We also calculated the coefficient of variation in the slope of catch curves (see Ricker, 1975) from the 1000 simulations for each subsample sample size. Catch curve slopes resulted from linear regressions of logtransformed numbers at age from expanded age-key estimates (y dependent variable) at ages 4–8 (x independent variable). Results The SMR gill net survey suggests that community composition varied little among years. Yellow perch, white sucker, cisco, northern pike, rock bass (Ambloplites rupestris), and walleye were always among the 10 most common species captured and always comprised a large proportion of all individuals captured (72–91%) (Fig. 2). The only major temporal difference we detected was that northern pike were less abundant during 2002, 2006, and 2009 compared with other years (Tukey's multiple comparison, P b 0.05). The 2009 gill net survey captured 29 species across 43 sites; yellow perch and white sucker (Catostomus commersoni) were the most common species and comprised about 64% of all individuals captured (Table 1). Community composition during 2009 was similar to other surveys in terms of species captured and relative abundance among species. Reduced gill net effort had little effect on numbers of species caught. Effort required to capture 80% of all species observed during surveys ranged from 11 to 21 net sets among years (Fig. 3). Our results showed that setting 25 gill nets per survey year would have captured 90% of the
species caught in full surveys except in 2002 when 32 net sets would have been required (Fig. 3). Reduced effort relative to previous survey years would have little impact on analyses of community composition or species richness. Using 2009 data, the SMR survey had low power to detect a 50% increase in relative abundance (0.2 for walleye and cisco, 0.3 for yellow perch and northern pike), and power to detect an 80% increase in relative abundance was higher only for yellow perch and northern pike (0.6 and 0.5, respectively; Fig. 4). Detecting a 50% increase in relative abundance with a power equal to 0.5 would have required 100 net nights for yellow perch and northern pike, and more than 200 net nights for walleye and cisco. Sample sizes have resulted in good power (i.e., approximately 0.80) to detect 80% reductions in relative abundance between two samples for walleye, northern pike, and yellow perch, but low power for cisco (approximately 0.50) (Fig. 5). Smaller changes in relative abundance among species would not likely be detectable without more gill netting effort. Thus, our power analysis indicated that the present gill net survey does not provide a strong ability to detect significant relative abundance changes for target species. During 2009, 1233 yellow perch were assigned ages for obtaining estimates of mean length-at-age. Simulations indicated that subsampling of these fish would have little influence on estimates. At 5 fish per cm group, CVs in estimated number at age varied from 50 to 100% excluding age 8 which had a CV of zero (no age-8 fish were captured). As numbers of fish increased, the range of CVs in estimated number at age decreased and approached 20% across all ages. Coefficients of variation in estimated mean length at age were close to 10% when 10 fish per cm group were assigned ages and less than 10% at larger subsamples
Please cite this article as: Schaeffer, J.S., et al., Designing long-term fish community assessments in connecting channels: Lessons from the Saint Marys River, J Great Lakes Res (2014), http://dx.doi.org/10.1016/j.jglr.2014.03.002
6
J.S. Schaeffer et al. / Journal of Great Lakes Research xxx (2014) xxx–xxx
Fig. 5. Relationship between statistical power to detect differences using a t-test and number of gill net sets for four target species and three levels of abundance decrease, Saint Marys River, 2009.
per cm group. Coefficients of variation in the slope estimates from simulation catch curves were smaller than 10% across the subsampling scenarios we evaluated. Thus, subsampling would have had little effect on mortality estimates. Reducing the number of fish retained to assign ages had the potential to drastically decrease age estimation effort relative to the realized effort in 2009 (Table 2) and using age-key extrapolations would allow subsample data to reflect assessment metrics from larger samples. Thus, assigning ages to a subsample of yellow perch in 2009 would have reduced laboratory effort with little cost to stock assessment metrics. Discussion Great Lakes connecting channels are important ecologically because despite over two centuries of invasion, harvest, habitat loss, and
changes in water quality they retain a semblance of a native fish community. Monitoring them may be vital for both stock preservation and to help maintain a source of recruits for lake populations (Edwards et al., 1989; Schaeffer et al., 2011). To date, the SMR is the only connecting channel with long-term fish community data, but it illustrates that managers are faced with difficult choices when designing and implementing these types of surveys. The ability to detect changes in diversity and relative abundance is often the primary goal of monitoring programs (Yoccoz et al., 2001) and programs strive to maximize efficiency in reaching that goal. Our analysis demonstrated options for improving SMR survey efficiency and highlighted trade-offs associated with varying survey effort. The current SMR assessment program produces multiple valuable metrics for management (Schaeffer et al., 2011), but our results suggest that more intensive effort would be required to detect 50% abundance
Table 2 Variability in estimated numbers at age resulting from simulated subsampling of a fixed number of yellow perch per length bin (i.e., 1 cm) for age estimation and subsequent age–length key expansion. Data from 2009 survey only, parentheses indicate 95% confidence intervals from 1000 nonparametric bootstrap resamples without replacement. Number subsampled
5/cm 10/cm 15/cm 20/cm 25/cm
Coefficient of variation (*100) in number at age Age 1
2
3
4
5
6
7
8
78.9 44.5 41.2 32.6 18.6
23.3 16.4 13.1 10.9 9.8
16.0 11.4 9.1 7.6 6.7
17.5 11.5 9.0 7.8 7.0
54.0 34.3 28.2 23.7 19.8
64.0 41.8 33.2 27.8 24.0
89.2 54.1 45.7 33.3 19.4
0.0 0.0 0.0 0.0 0.0
Median number aged
CV of slope estimate
101 177 243 306 306
6.59 4.69 3.77 3.11 2.14
Please cite this article as: Schaeffer, J.S., et al., Designing long-term fish community assessments in connecting channels: Lessons from the Saint Marys River, J Great Lakes Res (2014), http://dx.doi.org/10.1016/j.jglr.2014.03.002
J.S. Schaeffer et al. / Journal of Great Lakes Research xxx (2014) xxx–xxx
changes between 2 samples in our four target species. Power to detect relative abundance changes was generally low and varied among target species. With historical gill net effort, power to detect large changes in relative abundance (i.e., 80%) never exceeded 0.5 and effort required to improve statistical power is not likely feasible due to time and worker limitations. Our power analysis for 2009 data indicated that despite having low power, the sampling program was comparatively more efficient for monitoring northern pike, walleye and yellow perch relative abundances than for cisco. Analyses evaluating sample size requirements for detecting relative abundance changes commonly result in exorbitantly high estimates (Quist et al., 2006). For example, Paukert (2004) estimated it would often require over 1000 electrofishing and 500 trammel net samples to detect 25% relative abundance changes in two Colorado River sucker species with power ≥ 0.8. However, power can also be increased by reducing variability and biases. Biased sampling can be useful for identifying population trends if biases are consistent among sampling periods (Reynolds, 1996), for example using consistent gill net mesh sizes in the SMR survey. However, in the SMR survey there may be other factors such as sites annually selected for sampling (with their potential among-site differences in CPE) that could lower power to detect among-year differences. Removing variability is commonly accomplished via a stratified design, but partitioning variance within statistical analysis can prove beneficial for understanding influences on power and survey design efficiency (e.g., Wagner et al., 2009). Future analyses of the SMR monitoring program could use a multi-level analysis to consider site-dependent and time-dependent effects on population and community indices. Species accumulation curves indicated that SMR gill net effort should almost always be adequate for indexing species richness. It is important to recognize that species richness estimates from gillnet sampling are biased relative to the whole community's true composition (Hamley, 1975), and this is true for the SMR gill net survey as well (Schaeffer et al., 2011). However, the SMR survey is the most spatially comprehensive and continuous data source for the system and does allow for determination of community composition in dominant native species (Schaeffer et al., 2011). Thus, the current sampling protocol is useful for evaluating community stability, and we showed that the survey would continue to be useful with reduced effort. Random subsampling of total catch is commonly used to make inferences about the total population age composition and reduces processing times or number of fish collected (Bettoli and Miranda, 2001). Our results revealed opportunities to increase efficiency via subsampling the yellow perch catch for estimating population size structure and age composition. Similarly, Coggins et al. (2013) concluded that subsampling with 10 fish per length bin would achieve accuracy and precision of multiple population assessment metrics, including growth and mortality, based on simulations across multiple life histories. Coggins et al. (2013) suggested subsampling 10 fish per length bin as a guideline for most studies. An additional advantage to decreasing sample size for the SMR assessment is that it could allow for continuity in aging structures, whereas currently ages are estimated using both spines and scales. We restricted our analyses of reduced age and growth sampling to yellow perch during 2009. We chose to use 2009 data because yellow perch sample sizes were large and age estimation effort was extensive. Further simulations would be required to determine if subsampling is feasible for other species of interest that are less frequently captured. One intriguing consideration resulting from our analysis is that although reduced effort would exacerbate current power limitations for comparing catch rates, lower effort could still provide an index of species richness and samples for target species population metrics (i.e., age and size structure). Thus, reducing field and laboratory effort might allow the gill net survey to be annualized. Annual surveys could allow managers to have a better understanding of population characteristics such as cohort-specific analyses that would obviate assumptions of equal recruitment. Tetzlaff et al. (2011) reported that following cohorts through time provided a more reliable index of year-class strength
7
compared to using catch curve residuals. Furthermore, in some cases surveys have been separated temporally such that percid year classes could pass through the population entirely without detection. Annual surveys in connecting channels would also be consistent with most other Great Lakes monitoring programs. The SMR data provide a tool for informing future long-term survey designs. Agencies considering establishing long-term surveys on other connecting channels might consider using a similar approach, starting with identical gill nets fished at multiple sites to obtain broad spatial coverage. While using gill nets for monitoring community composition and relative abundance has inherent biases, employing similar methods across connecting channels would facilitate cross-system comparisons. Field experiments such as localized mark-recapture studies could estimate species-specific detection probabilities among habitat types to further improve value of gill net survey data. The present gill net survey is likely an insufficient stock assessment tool, but adequate as a fish community assessment tool. If a stock assessment is desired, then gill net effort needs to be increased 2.5 to 5 times. If it is the latter, then reduced effort could be considered for providing community composition data and population metrics for specific species. Managers could avoid that choice by reducing survey effort, but performing directed and well-designed stock assessments that might more efficiently collect unbiased data from target species, using gears that might increase sample size and reduce by-catch. The SMR assessment was designed with broad objectives to gain community-wide information on a largely unknown system; as the survey was repeated there became a tension between a need for data continuity versus a need for species-specific information as new management strategies evolved. This situation is common in long-term monitoring programs, many of which began in an exploratory fashion and sometimes without specific objectives. We suggest that analyses of statistical power and efficiency of current sampling protocols are a useful step that benefits managers collecting and using these types of data for decision making as well as for the development of new monitoring programs. Evaluating monitoring program data robustness provides insight into whether survey goals and objectives are being attained, and they can help evaluate the ability of surveys to answer novel questions that arise as management strategies are refined.
Acknowledgments We thank the many dedicated individuals who participated in field sampling. The Ontario Ministry of Natural Resources upper Great Lakes management unit and U.S. Fish and Wildlife Service personnel processed structures and assigned ages. This is contribution 1831 of the Great Lakes Science Center.
References Bettoli, P.W., Miranda, L.E., 2001. A cautionary note about estimating mean length at age with subsampled data. N. Am. J. Fish. Manag. 21, 425–428. Bunnell, D.B., Madenjian, C.P., Claramunt, R.M., 2006. Long-term changes of the. Lake Michigan fish community following the reduction of exotic alewife (Alosa pseudoharengus). Can. J. Fish. Aquat. Sci. 63, 2434–2446. Coggins, L.G., Gwinn, D.C., Allen, M.S., 2013. Evaluation of age–length key sample sizes required to estimate fish total mortality and growth. Trans. Am. Fish. Soc. 142, 832–840. Cohen, J., 1988. Statistical Power Analysis for the Behavioral Sciences, 2nd ed. Laurence Earlbaum Associates, Hillsdale, NJ. Development Core Team, R., 2013. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. Dixon, P.M., 1993. The bootstrap and the jackknife: describing the precision of ecological indices. In: Scheiner, S.M., Gurevitch, J. (Eds.), Design and Analysis of Ecological Experiments. Chapman and Hall, New York, pp. 290–318. Edwards, C.J., Hudson, P.L., Duffy, W.G., Nepszy, S.J., McNabb, C.D., et al., 1989. Hydrological, morphometrical, and biological characteristics of the connecting rivers of International Great Lakes: a review. Proc. Int. Large River Symp. Can. Spec. Publ. Fish. Aquat. Sci. 106, 240–264. Fielder, D., 2002. St. Marys River Task Group: the St. Marys River and challenges of multijurisdictional fisheries. Fisheries 27, 32–34.
Please cite this article as: Schaeffer, J.S., et al., Designing long-term fish community assessments in connecting channels: Lessons from the Saint Marys River, J Great Lakes Res (2014), http://dx.doi.org/10.1016/j.jglr.2014.03.002
8
J.S. Schaeffer et al. / Journal of Great Lakes Research xxx (2014) xxx–xxx
Fielder, D.G., Waybrant, J.R., 1998. Fish population surveys of St. Marys River, 1975–95, and recommendations for management. Fisheries Research Report 2048Michigan Department of Natural Resources, Ann Arbor. Fielder, D.G., Borgeson, D.J., Bowen, A.K., Koproski, S.R., Greenwood, S.J., Wright, G.M., 2003. Population Dynamics of the St. Marys River Fish Community 1975–2002. Great Lakes Fishery Commission, Miscellaneous publication, Ann Arbor http:// www.glfc.org/lakecom/lhc/SMR2002rpt.pdf. Accessed July, 2012. Gebhardt, K., Fielder, D., Greenwood, S., Robbins, H., Sutton, T., 2002. St. Marys River Fisheries Assessment Plan. Great Lakes Fisheries Commission (http://www.glfc.org/ lakecom/lhc/smrfinal.pdf. Ann Arbor. Accessed July, 2012). Gerow, K.G., 2007. Power and sample size estimation techniques for fisheries management: assessment and a new computation tool. N. Am. J. Fish. Manag. 27, 397–404. Greenwood, S.J., Fielder, D.G., Godby, N., Kolb, T., 2011. A Synthesis of Sport Fishing Activity in the St. Marys River, 1999–2001 and 2005–2009. Great Lakes Fishery. Commission. Miscellaneous Publication (http://www.glfc.org/lakecom/lhc/SMRHarvestRpt19992009FINAL.pdf. Accessed July 2012.). Grimm, K.S., 1989. A fisheries survey of the St. Marys River, Chippewa County, August– October, 1987. Mich. Dept. Nat. Res. Tech. Rep. No. Hamley, J.M., 1975. Review of gillnet selectivity. J. Fish. Res. Board Can. 32 (11), 1943–1969. Isermann, D.A., Knight, C.T., 2005. A computer program for age–length keys incorporating age assignment to individual fish. N. Am. J. Fish. Manag. 25, 1153–1160. Kocovsky, P.M., Stapanian, M.A., Knight, C.T., 2010. Night sampling improves indices used for management of yellow perch in Lake Erie. Fish. Manag. Ecol. 17 (1), 10–18. Lakes Restoration Initiative, Great, 2010. Great Lakes Restoration Initiative Action Plan FY 2010–2014. (http://glri.us/pdfs/glri_actionplan.pdf. Accessed June, 2012). Madenjian, C.P., Elliot, R.F., DeSorcie, T.J., Stedman, R.M., O'Connor, D.V., Rottiers, D.V., 2000. Lipid concentrations in Lake Michigan fishes: seasonal, spatial, ontogenetic, and long-term trends. J. Great Lakes Res. 26, 427–444. Miller, B.R., 1981. A 1979 fisheries survey of the St. Marys River system in Chippewa County, and comparisons with 1975 results. Mich. Dept. Nat. Res. Tech. Rep. No. Ogle, D., 2012. Functions to Support Simple Fisheries Stock Assessment Methods. (http:// www.rforge.net/src/contrib/Documentation/FSA.pdf. Accessed June, 2012.). Oksanen, J., Guillaume Blanchet, F., Kindt, R., Legendre, P., Minchin, P.R., O'Hara, R.B., Simpson, G.L., Solymos, P., Stevens, M.H.H., Wagner, H., 2013. Vegan: Community Ecology Package. (http://vegan.r-forge.r-project.org. Accessed March, 2012.). Paukert, C.P., 2004. Comparison of electrofishing and trammel netting variability for sampling native fishes. J. Fish Biol. 65, 1643–1652.
Pratt, T.C., O'Connor, L.M., 2011. An assessment of the health and historical changes of the nearshore fish community of the St. Marys River. J. Great Lakes Res. 37 (Suppl. 2), 61–69. Quist, M.C., Gerow, K.G., Bower, M.R., Hubert, W.A., 2006. Random versus fixed-site sampling when monitoring relative abundance of fishes in headwater streams of the Upper Colorado River basin. N. Am. J. Fish. Manag. 26, 1011–1019. Rand, P.S., Stewart, D.J., Seelbach, P.W., Jones, M.L., Wedge, L.R., 1993. Modeling steelhead population energetic in Lakes Michigan and Ontario. Trans. Am. Fish. Soc. 122, 977–1001. Reynolds, J.B., 1996. Electrofishing, In: Murphy, B.R., Willis, D.W. (Eds.), Fisheries Techniques, 2nd edition. American Fisheries Society, Bethesda, Maryland, pp. 221–253. Ricker, W.E., 1975. Computation and interpretation of biological statistics of fish populations. Fish. Res. Board Can. Bull. 191. Riley, S.C., Roseman, E.F., Nichols, S.J., O'Brien, T.P., Kiley, C.S., Schaeffer, J.S., 2008. Deepwater demersal fish community collapse in Lake Huron. Trans. Am. Fish. Soc. 137, 1879–1890. Ripley, M.P., Arbic, B., Zimmerman, G., 2011. Environmental history of the St. Marys River. J. Great Lakes Res. 37 (Suppl. 2), 5–11. Roseman, E.F., Kennedy, G.W., Boase, J., Manny, B.A., Todd, T.N., Stott, W., 2007. Evidence of lake whitefish spawning in the Detroit River: implications for habitat and population recovery. J. Great Lakes Res. 33, 397–406. SAS Institute Inc., 2007. OnlineDoc 9.1.2. SAS Institute, Inc., Cary, NC. Schaeffer, J.S., et al., 2011. Long-term trends in the St. Marys River open water fish community. J. Great Lakes Res. 37 (Suppl. 2), 70–79. Schorfhaar, R., 1975. A Fisheries Survey of the St. Marys River, 1975. Michigan Department of Natural Resources, Fisheries Division, Marquette. (Unpublished). Sokal, R.R., Rohlf, F.J., 1981. Biometry, Second Edition. W. H Freeman and Company, San Francisco, California. Tetzlaff, J.C., Catalano, M.J., Allen, M.S., Pine III, W.E., 2011. Evaluation of two methods for indexing fish year-class strength: catch-curve residuals and cohort method. Fish. Res. 109, 303–310. Wagner, T., Vandergoot, C.S., Tyson, J., 2009. Evaluating the power to detect temporal trends in fishery-independent surveys: a case study based on gill nets set in the Ohio waters of Lake Erie for walleyes. N. Am. J. Fish. Manag. 29, 805–816. WTG (Walleye Task Group), 2012. Report for 2011 by the Lake Erie Walleye Task Group. Annual Report to the Lake Erie Committee. Great Lakes Fishery Commission, Windsor, Ontario. Yoccoz, N.G., Nichols, J.D., Boulinier, T., 2001. Monitoring of biological diversity in space and time. Trends Ecol. Evol. 16, 446–453.
Please cite this article as: Schaeffer, J.S., et al., Designing long-term fish community assessments in connecting channels: Lessons from the Saint Marys River, J Great Lakes Res (2014), http://dx.doi.org/10.1016/j.jglr.2014.03.002