@prefix vivo: . @prefix edm: . @prefix ns0: . @prefix dcterms: . @prefix dc: . @prefix skos: . vivo:departmentOrSchool "Science, Faculty of"@en, "Zoology, Department of"@en ; edm:dataProvider "DSpace"@en ; ns0:degreeCampus "UBCV"@en ; dcterms:creator "Nicol, Sandra Diane"@en ; dcterms:issued "2008-01-03T19:21:40Z"@en, "2007"@en ; vivo:relatedDegree "Master of Science - MSc"@en ; ns0:degreeGrantor "University of British Columbia"@en ; dcterms:description "Determining the physical and biological habitat variables that influence the abundance of juvenile salmonids in British Columbia streams will improve management practices. Habitat models are tools that provide insight into organisms’ habitat needs and provide a more efficient mechanism for estimating population abundance than direct measurement. Models have been developed for salmonids in other jurisdictions, but very few have included invertebrate drift (a primary food source for juvenile salmonids) as a predictive variable. This is because temporal and spatial variation of drift abundance are widely assumed to be so high that drift cannot be reliably estimated without unreasonable effort. This thesis investigates the temporal and spatial variability of invertebrate drift and the impact of its inclusion in habitat models for juvenile salmonid abundance in two chapters. The first objective of the first chapter was to evaluate the temporal variability of invertebrate drift by comparing the seasonal and day-to-day variation in drift abundance to spatial variation within and between sites. The second objective was to develop predictive models for invertebrate drift abundance. Aquatic, terrestrial and total invertebrate drift abundances varied primarily between sites and very little between days or months at the same site, indicating that a single day of sampling is sufficient to assess drift abundance for comparison among sites. The abundance of invertebrate drift was related to productivity- and flow-related habitat variables. The objectives of the second chapter were to develop predictive models for juvenile salmonid abundance in southwestern BC using physical and biological habitat variables, to determine whether habitat variables differ between the Coast and Interior regions of BC, to determine the contribution of invertebrate drift to the relative predictive ability of the models, and to determine cost:benefit ratios for the predictive models and their component variables. The final models for predicting abundance of all young-of-year salmonids combined, and rainbow trout (Oncorhynchus mykiss) and coho salmon (O. kisutch) individually, included variables related to stream structure and productivity, and the models for rainbow and coho showed regional differences. Invertebrate drift did not improve model fit."@en ; edm:aggregatedCHO "https://circle.library.ubc.ca/rest/handle/2429/256?expand=metadata"@en ; dcterms:extent "1324210 bytes"@en ; dc:format "application/pdf"@en ; skos:note "INFLUENCE OF PHYSICAL AND BIOLOGICAL HABITAT VARIABLES ON JUVENILE SALMONID AND INVERTEBRATE DRIFT ABUNDANCE IN SOUTHWEST BRITISH COLUMBIA STREAMS by SANDRA DIANE NICOL B.Sc., The University of British Columbia, 2003 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE in THE FACULTY OF GRADUATE STUDIES (Zoology) THE UNIVERSITY OF BRITISH COLUMBIA December 2007  Sandra Diane Nicol, 2007 ii Abstract Determining the physical and biological habitat variables that influence the abundance of juvenile salmonids in British Columbia streams will improve management practices. Habitat models are tools that provide insight into organisms’ habitat needs and provide a more efficient mechanism for estimating population abundance than direct measurement. Models have been developed for salmonids in other jurisdictions, but very few have included invertebrate drift (a primary food source for juvenile salmonids) as a predictive variable. This is because temporal and spatial variation of drift abundance are widely assumed to be so high that drift cannot be reliably estimated without unreasonable effort. This thesis investigates the temporal and spatial variability of invertebrate drift and the impact of its inclusion in habitat models for juvenile salmonid abundance in two chapters. The first objective of the first chapter was to evaluate the temporal variability of invertebrate drift by comparing the seasonal and day-to-day variation in drift abundance to spatial variation within and between sites. The second objective was to develop predictive models for invertebrate drift abundance. Aquatic, terrestrial and total invertebrate drift abundances varied primarily between sites and very little between days or months at the same site, indicating that a single day of sampling is sufficient to assess drift abundance for comparison among sites. The abundance of invertebrate drift was related to productivity- and flow-related habitat variables. The objectives of the second chapter were to develop predictive models for juvenile salmonid abundance in southwestern BC using physical and biological habitat variables, to determine whether habitat variables differ between the Coast and Interior regions of BC, to determine the contribution of invertebrate drift to the relative predictive ability of the models, and to determine cost:benefit ratios for the predictive models and their component variables. The final models for predicting abundance of all young-of-year salmonids combined, and rainbow trout (Oncorhynchus mykiss) and coho salmon (O. kisutch) individually, included variables related to stream structure and productivity, and the models for rainbow and coho showed regional differences. Invertebrate drift did not improve model fit. iii Table of Contents Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .vii Co-Authorship Statement. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii Chapter 1: Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Chapter 2: Spatial and temporal variability of invertebrate drift in southwest British Columbia streams. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 Discussion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 Chapter 3: Influence of physical and biological habitat factors on juvenile salmonid abundance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 Discussion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 Appendix 1: Chapter 1 Habitat Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69 Appendix 2: Chapter 2 Habitat Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 Appendix 3: UBC Research Ethics Board Certificate of Approval . . . . . . . . . . . . . . . . . . 79 iv List of Tables Table 2.1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 Table 2.2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 Table 2.3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 Table 2.4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 Table 3.1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 Table 3.2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 Table 3.3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 Table 3.4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 Table 3.5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 vList of Figures Figure 2.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 Figure 2.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 Figure 2.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 Figure 2.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 Figure 2.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 Figure 2.6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 Figure 2.7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 Figure 2.8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 Figure 3.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 Figure 3.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 Figure 3.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 Figure 3.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 vi Acknowledgements Funding for this study was provided by the Forest Science Program of British Columbia. Thanks to Richard Bailey of Fisheries and Oceans Canada, Allen Hanson of the British Columbia Conservation Foundation, David Coutlee of the Nicola Watershed Stewardship and Fisheries Authority, Mike Wallis of the Salmon River Watershed Roundtable, Peter Tschaplinski of BC Ministry of Forests, Ron Ptolemly of BC Ministry of Environment, Scott Decker and Dennis Morgan for assistance selecting sites and/or for sharing their electrofishing data. Many thanks to the multitude of landowners who allowed us to work on their property. Field assistance was provided by Tessa Richardson, Tom Hicks and Brock Ramshaw. Laboratory assistance was provided by Geoff Cheung, Mikaela Davis and Patrick Carrier. Statistical advice was provided by Lorne Rothman. Thanks to the Shurin, Srivastava and Harley labs for support throughout the project. Many thanks to those who reviewed the manuscript: John Richardson, Jackie Ngai, Christian Veenstra, and the Shurin, Srivastava and Harley labs. vii For Scott viii Co-Authorship Statement Sandra Nicol was the primary researcher responsible for identifying and designing this research program, performing the research, analyzing the data and preparing the manuscript. However, assistance with research program identification and design, data analysis, and manuscript preparation was provided by Jordan Rosenfeld and Jonathan Shurin. 1Chapter 1: Introduction Management and conservation of species requires accurate and efficient estimation of populations sizes. Habitat models are one way of obtaining these estimates. Models that predict population size using habitat measurements have many applications for population management and conservation. These predictive models can improve understanding of species’ habitat requirements, including how these requirements interact, and they can require fewer resources to apply than direct field measurements of organism abundance. Habitat models are used to identify high quality habitat for protection, plan habitat restoration projects and monitor resource extraction activities (e.g. Binns and Eiserman 1979, Clark et al. 1993). Freshwater fish such a salmonids (Family Salmonidae) are well suited to habitat modeling due o their restricted habitat (lakes and streams, rather than patches of grassland or ocean floor) and due to the potential utility of predictive models based on simple habitat metrics relative to labour-intensive fish sampling. Salmonid habitat is often categorized into physical features – such as the area of pools and the number of pieces of large woody debris – and biological variables that influence food abundance such as the amount of detritus. Most studies of juvenile salmonid habitat have emphasized physical variables like channel width, depth, and structure, and reach characteristics like gradient and sediment size (see Coulombe-Pontbriand and LaPointe 2004, Davies 1989, Rosenfeld et al. 2000, Sharma and Hilborn 2001) which are all related to salmonid habitat quantity or quality. The most direct measurement of food availability is invertebrate drift. Salmonids primarily feed on drifting invertebrates (Wipfli 1997) that are delivered by two mechanisms: in situ invertebrate production based on algal- or detrital- based food chains (Vannote et al. 1980), or input of terrestrial invertebrates from the riparian zone (Cada et al. 1987). Canopy cover and composition, organic sediment, temperature, and water conductivity (see Binns and Eiserman 1979, Johansen et al. 2005, Scarnecchia and Bergersen 1987) are additional variables that may influence the abundance of invertebrate prey. Although invertebrate drift is the preferred food of many juvenile salmonids, it is 2conspicuously absent from most habitat models and is rarely used to assess habitat quality (Fausch et al. 1988). Drifting invertebrates, both aquatic and terrestrial, are a primary food source for juvenile salmonids (Wipfli 1997), and elevated drift contributes to increased juvenile salmonid growth (Imre et al. 2004, Nielsen 1992, Rosenfeld et al. 2005, Slaney 1972), density (Slaney 1972) and survival (Imre et al. 2004, Nislow et al. 1998). Food supply is often estimated by proxy through measurements of canopy cover or other variables (Johansen et al. 2005). However, including a direct measurement of invertebrate drift abundance as a candidate variable in habitat models may improve the predictive ability of the models compared to indirect estimates, and may help determine the degree to which food availability limits salmonid abundance. Invertebrate drift is perceived as difficult to quantify (Binns and Eiserman 1979, Brittain and Eikeland 1988) due to high temporal and spatial variability. This variability has led some authors to suggest that multiple samples (4-7) need to be collected on multiple days to accurately estimate invertebrate drift abundance (Allan and Russek 1985, Matthaei et al. 1998, Shearer et al. 2002). Unfortunately, such resource-intensive sampling protocols discourage investigators from using drift to assess habitat quality for juvenile salmonids. However, samples collected on a single day may be sufficient depending on the degree of temporal variation between days relative to spatial variation between sites. As one of the reasons to use habitat models is their improved efficiency, the sampling protocol for invertebrate drift will impact its inclusion in models; if drift sampling requires multiple visits to a site it will not be viewed by fisheries managers as a practical index of habitat quality. Habitat models are often a more time-efficient way to estimate salmonid abundance than direct measurements. The predictive variables that biologists choose to include in regression analysis are usually based on expert opinion as to which variables are most effective. However, variables are rarely selected in an explicit cost-benefit framework. Assessing the cost-effectiveness of habitat models and their terms (the habitat variables) should allow fisheries scientists to choose which habitat models and component variables maximize predictive power. A cost-benefit analysis should quantify the advantages and disadvantages of various habitat variables and whole models, allowing investigators to choose to use 3direct measurements of fish abundance, an existing model, or a new combination of habitat variables to estimate habitat quality. This thesis is organized into two related studies. Chapter 2 examined the temporal and spatial variability of invertebrate drift to determine the level of sampling intensity (single visit or multiple visit) necessary to compare drift abundance between sites. If multi-day sampling is required then it is unrealistic to expect fisheries managers to use invertebrate drift as an index of habitat quality. The objectives of this study were: 1) to determine whether accurately measuring drift requires sampling on multiple days by comparing the temporal variation in drift abundance on a seasonal and day-to-day basis to spatial variation between sites, 2) to determine the range of drift between sites in Coast and Interior regions of southwestern British Columbia, 3) to develop predictive models between invertebrate drift abundance and a suite of physical and biological habitat variables, and 4) to determine the variables associated with drift abundance at a landscape scale. Chapter 3 involved developing habitat models for young-of-the-year (YOY) rainbow trout (Oncorhynchus mykiss), coho salmon (O. kisutch), and all YOY combined using physical and biological habitat variables, including invertebrate drift. The objectives of this study were 1) to develop predictive models for juvenile salmonid abundance in southwestern BC using physical and biological habitat factors, 2) to determine whether habitat variables differ between the Coast and Interior regions, 3) to determine the contribution of invertebrate drift to the relative predictive ability of models, and 4) to determine benefit:cost ratios for predictive models and their component variables. 4References Binns, N.A., and Eiserman, F.M. 1979. Quantification of fluvial trout habitat in Wyoming. Transactions of the American Fisheries Society 108: 215-228. Cada, G.F., Loar, J.M., and Cox, D.K. 1987. Food and feeding preferences of rainbow and brown trout in southern Appalachian streams. The American Midland Naturalist 117: 374-385. Clark, J.D., Dunn, J.E., and Smith, K.G. 1993. A multivariate model of female black bear habitat use for a geographic information system. Journal of Wildlife Management 57(3): 519-526. Coulombe-Pontbriand, M., and LaPointe, M. 2004. Landscape controls on boulder-rich, winter habitat availability and their effects on Atlantic salmon (Salmo salar) parr abundance in two fifth-order mountain streams. Canadian Journal of Fisheries and Aquatic Sciences 61(4): 648-658. Davies, P.E. 1989. Relationships between habitat characteristics and population abundance for brown trout, Salmo trutta L., and blackfish, Gadopsis marmoratus Rich., in Tasmanian streams. Australian Journal of Marine and Freshwater Research 40: 341- 359. Denoël, M., and Lehmann, A. 2006. Multi-scale effect of landscape processes and habitat quality on newt abundance: Implications for conservation. Biological Conservation 130: 495-504. Dorgelöh, W.G. 2006. Habitat suitability for tsessebe Damaliscus lunatus lunatus. Journal of African Ecology 44: 329-336. Fausch, K.D., Hawkes, C.L., and Parsons, M.G. 1988. Models that predict standing crop of stream fish from habitat variables: 1950-1985. Gen Tech Rep, U.S. Department of Agriculture Forest Service Pacific Northwest Research Station, Portland, Oregon. Gavashelishvili, A., McGrady, M.J., and Javakhishvili, Z. 2006. Planning the conservation of the breeding population of cinereous vultures Aegypius monachus in the Republic of Georgia. Oryx 40(1): 1-8. Imre, I., Grant, J.W.A., and Keeley, E.R. 2004. The effect of food abundance on territory size and population density of juvenile steelhead trout (Oncorhynchus mykiss). Oecologia 138: 371-378. Johansen, M., Elliott, J.M., and Klemetsen, A. 2005. A comparative study of juvenile salmon density in 20 streams throughout a very large river system in northern Norway. Ecology of Freshwater Fish 14: 96-110. McAlpine, C.A., Rhodes, J.R., Callaghan, J.G., Bowen, M.E., Lunney, D., Mitchell, D.L., Pullar, D.V., and Possingham, H.P. 2006. The importance of forest area and configuration relative to local habitat factors for conserving forest mammals: A case study of koalas in Queensland Australia. Biological Conservation 132: 153-165. Nielsen, J.L. 1992. Microhabitat-specific foraging behavior, diet, and growth of juvenile coho salmon. Transactions of the American Fisheries Society 121: 617-634. 5Nislow, K.H., Folt, C., and Seandel, M. 1998. Food and foraging behavior in relation to microhabitat use and survival of age-0 Atlantic salmon. Canadian Journal of Fisheries and Aquatic Sciences 55: 116-127. Rosenfeld, J., Porter, M., and Parkinson, E. 2000. Habitat factors affecting the abundance and distribution of juvenile cuttrhoat trout (Oncorhynchus clarki) and coho salmon (Onchorhynchus kisutch). Canadian Journal of Fisheries and Aquatic Sciences 57: 766-774. Rosenfeld, J.S., Leiter, T., Lindner, G., and Rothman, L. 2005. Food abundance and fish density alters habitat selection, growth, and habitat suitability curves for juvenile coho salmon (Oncorhynchus kisutch). Canadian Journal of Fisheries and Aquatic Sciences 62: 1691-1701. Scarnecchia, D.L., and Bergersen, E.P. 1987. Trout production and standing crop in Colorado’s small streams, as related to environmental features. North American Journal of Fisheries Management 7: 315-330. Sharma, R., and Hilborn, R. 2001. Empirical relationships between watershed characteristics and coho salmon (Oncorhynchus kisutch) smolt abundance in 14 western Washington streams. Canadian Journal of Fisheries and Aquatic Sciences 58: 1453-1463. Slaney, P.A. 1972. Effects of prey abundance on distribution, density and territorial behaviour of young rainbow trout in streams, University of British Columbia. Vannote, R.L., Minshall, G.W., Cummins Kenneth, W., Sedell, J.R., and Cushing, C.E. 1980. The river continuum concept. Canadian Journal of Fisheries and Aquatic Sciences 37: 130-137. Wipfli, M. 1997. Terrestrial invertebrates as salmonid prey and nitrogen sources in streams: contrasting old-growth and young-growth riparian forests in southeastern Alaska USA. Canadian Journal of Fisheries and Aquatic Sciences 54: 1259-1269. Zharikov, Y., Lank, D.B., Huettmann, F., Bradley, R.W., Parker, N., Yen, P.P.W., McFarlane- Tranquilla, L.A., and Cooke, F. 2006. Habitat selection and breeding success in a forest-nesting Alcid, the marbled murrelet, in two landscapes with different degrees of forest fragmentation. Landscape Ecology 21: 107-120. 6Chapter 2: Spatial and temporal variability of invertebrate drift in southwest British Columbia streams1 Introduction Drifting invertebrates of aquatic and terrestrial origin are a primary food source for juvenile salmonids (Wipfli 1997), and elevated drift contributes to increased juvenile salmonid growth (Imre et al. 2004, Nielsen 1992, Rosenfeld et al. 2005, Slaney 1972) , density (Slaney 1972) and survival (Imre et al. 2004, Nislow et al. 1998). Invertebrates are a critical component of stream food webs, yet drift is seldom used as a measure of habitat quality in favour of aspects of habitat structure like woody debris (Fausch et al. 1988). This is largely because drift is perceived as difficult to quantify (Binns and Eiserman 1979, Brittain and Eikeland 1988), with high temporal and spatial variability. This variability has led some authors to suggest that multiple samples (4-7) need to be collected on different days to accurately estimate invertebrate drift abundance (Allan and Russek 1985, Matthaei et al. 1998, Shearer et al. 2002). Such resource-intensive sampling protocols (laboratory time required to sort and quantify a single drift sample can vary from 3 to 10 hours) have likely contributed to the limited use of drift to assess habitat quality for juvenile salmonids. However, fewer samples collected on a single day may be sufficient depending on the degree of temporal variation between days relative to spatial variation between sites. Understanding how invertebrate drift biomass varies in time and space will allow investigators to better evaluate the number of samples required to accurately characterize drift biomass, and to evaluate the potential of drift as a predictor of salmonid abundance. The scale and magnitude of temporal variation in drift abundance will affect the sampling effort required to accurately assess prey abundance, and therefore prey availability for drift- feeding fish. Temporal variation in drift occurs at a hierarchy of scales, ranging from well- documented diel variation to variation between days and longer seasonal trends (Resh and Rosenberg 1989). Aquatic invertebrates are vulnerable to visual predators and preferentially drift at night (Flecker 1992), so that drift has relatively uniform low daytime abundance and 1 A version of this chapter will be submitted for publication: Nicol, SD, JS Rosenfeld and JB Shurin. Spatial and temporal variability of inverte- brate drift in southwest British Columbia streams. 7peaks shortly after dark (Waters 1962), whereas terrestrial invertebrates fall into streams more frequently during the day (Rincón and Lobón-Cerviá 1997). Variation between days has also been studied but reports of the magnitude of day-to-day variation differ (Shearer et al. 2002, Williams 1980). Williams (1980) found that between-day variation in drift abundance varied up to 6 times, suggesting that investigators mush sample a site on more than one day in order to estimate drift abundance. Other studies suggest that day-to-day variation is no greater than variation between replicates (Shearer et al. 2002). Day-to-day variability in drifting aquatic invertebrates is affected by nocturnal light levels (aquatic drift is suppressed on bright nights that expose invertebrates to visual predators; Holt and Waters 1967), temporal lags that cause compensatory drifting (high night time aquatic drift after a night of suppressed drifting), life history events such as mass emergence of adults (Williams 1980) and variable discharge. Day-to-day variability of terrestrial invertebrates is generally higher than for aquatic invertebrates (Nakano et al. 1999), indicating higher variance in the factors (eg. wind) that deliver terrestrial prey from riparian zones. Total drift abundance also shows longer seasonal trends. Invertebrate drift biomass (per m3) in temperate regions tends to be higher in the summer than the winter (Brittain and Eikeland 1988, Kawaguchi and Nakano 2001, Shearer et al. 2002). Allan et al. (2003) found that drift abundance peaked in July, and others (eg. Bacon et al. 2005, Rincón and Lobón-Cerviá 1997, Slaney 1972) have observed spring peaks in drift abundance followed by summer declines. Understanding the drivers of temporal variation in drift allows improved sampling protocols for examining the mechanisms that influence prey abundance for drift-feeding fish. Differences in average drift abundance between streams leads to spatial variation in drift at regional scales, which could be a major driver of differences in habitat capacity for juvenile salmonids. Aquatic drift abundance should be influenced by variables from one or both of two categories: those related to biotic productivity factors and those related to invertebrate entrainment in the water column. In principle, aquatic invertebrate drift should be higher in streams with greater productivity, in which case aquatic drift biomass may be associated with habitat variables that affect benthic productivity, such as temperature, stream nutrients (Richardson 1993), algae (chlorophyll-a) concentrations (Schell 1999, Shearer et al. 2003), 8detritus density or quality (Chadwick and Huryn 2007), or factors such as riparian canopy cover that influence benthic primary production in light-limited streams (Wallace et al. 1997, Wipfli 1997, Wipfli and Musselwhite 2004). Flow-related physical factors that affect entrainment of benthic invertebrates into the drift, such as water velocity (Harvey et al. 2006), channel gradient, the availability of riffle habitat (Rader 1997), or flow regime (Lancaster 1999) may be associated with drift. Biomass of drifting terrestrial invertebrates is affected by the chance occurrence of an invertebrate falling into a stream; terrestrial invertebrates may be influenced by factors that affect their activity level such as temperature (Edwards and Huryn 1995), or their overall abundance in the riparian area such as canopy cover and composition. However, evidence for the importance of canopy cover composition on terrestrial invertebrate abundance is mixed (Johansen et al. 2005, Mason and MacDonald 1982, Wipfli 1997, Wipfli and Musselwhite 2004). Identifying the factors that influence drift abundance will improve understanding of how habitat influences salmonid food supply, and consequently could lead to improved management of fish habitat. The primary objectives of this study were: 1) to determine whether accurately measuring drift requires sampling on multiple days by comparing the temporal variation in drift abundance on a seasonal and day-to-day basis to spatial variation between sites at a subset of 4 intensively sampled streams; 2) to determine the range of drift between sites in Coast and Interior regions of southwestern British Columbia, 3) to develop predictive models between invertebrate drift abundance and a suite of physical and biological habitat variables, and 4) to determine the variables associated with drift abundance at a landscape scale. Methods Temporal variation: magnitude of day-to-day and seasonal variation relative to site effects Four sites were chosen for repeated sampling to determine the magnitude of day-to-day variation in drift biomass: one small and one large stream from the west coast of British Columbia (Coast region, Husdon and Chapman Creeks, respectively), and one small and 9one large stream from the southern interior of BC (Interior region, Senn and Yard Creeks, respectively). Sites were chosen that differed in size and geographic area so that any conclusions are not limited to a specific type of stream, such as small streams in the Coast region. Sites were chosen from the pool of sties recommended by local fisheries biologists because they were representative of streams in their size classes in their regions. Each site was sampled twice per month over four months (June to September 2005), with within- month samples 1-3 days apart to minimize effects of seasonal trends on daily variation and to allow us to examine variation at two temporal scales (days and months). No sampling was conducted during or soon after precipitation to standardize the effect of weather on drift abundance. Two replicate drift samples were collected on each sampling date, yielding 16 samples per stream except for Yard Creek, where a second visit in June was cancelled due to weather. Invertebrate Drift Sampling Invertebrate drift samples were collected at the upstream end of pools immediately below riffles to standardize habitat effects on drift and to measure the quantity of drift that a dominant fish would experience at the head of a pool. The nets were placed in separate pools, and as much as possible nets were positioned so that they did not filter the same water. At sites with no suitable pools nets were placed in riffles. Samples were collected with a 1 m long, 250 μm mesh net with an opening 0.2 m wide, for 180-300 minutes, depending on water velocity. The depth and velocity of the water in the net opening were measured in order to calculate the volume of water sampled, and thereby the biomass of drifting invertebrates per m3. We conducted daytime sampling rather than 24-hour sampling because daytime drift abundance is more uniform (Waters, 1962), and most juvenile salmonids are visual predators that benefit from day time drift abundance. Arguably, daytime drift samples more accurately represent prey abundance for juvenile salmonids than 24 hour drift samples where biomass is dominated by nocturnal peaks. Drift samples were collected at least two hours after sunrise and two hours before sunset to reduce daily variability in light intensity, and were not collected during or after rain events. Samples were preserved in 5% formalin and returned to the lab for sorting. 10 Invertebrates were sorted from detritus at 10X magnification in the laboratory and preserved in 70% ethanol for later identification and measurement using a digitizer (Roff and Hopcroft 1986). Invertebrates were identified to order with some exceptions; non-arthropods such as nematodes were identified to phylum, and common arthropods such as chironomids were identified to sub-family. Each taxon was categorized as aquatic or terrestrial in origin. Adult insects with aquatic larvae, such as Nematocera (suborder of Diptera), were counted as aquatic. The dry mass (mg) of invertebrates in each sample was calculated using published taxa-specific length-weight regressions (Benke et al. 1999, McCauley 1984, M. Wipfli unpubl., Meyer 1989, Sabo et al. 2002, Sample et al. 1993, Smock 1980). Temporal Variation Statistical Analysis The magnitude of day-to-day, monthly, and between- and within-site variance was compared with mixed model analysis of variance using the lme4 package (Bates 2007) in R 2.4.1 (www.r-project.org). The models were fit using restricted maximum likelihood (REML). Candidate models with the four combinations of fixed factors were compared to determine which best explained the variability in the drift density (mg dry mass per m3 water). The full model included two fixed factors (region (Coast vs. Interior) and month (June-September)) and two random grouping factors (Julian day and site). Month was analyzed as an ordered factor to allow polynomial analyses (by default R examines the polynomial terms of an ordered factor). Site was considered to be a random factor because the four sites were chosen to be representative of the population of available sites, not for specific interest in those particular sites. The region model included region, Julian day, and site. The month model included month, Julian day and site. The random factors model included only the random factors Julian day and site. The analysis was not performed as a nested ANOVA due to complications arising from different dates; using Julian day addresses daily variation. Drift densities were log transformed so that model residuals would meet assumptions of normality. The models were compared using Akaike weights based on the Second-Order Information Criterion (AIC c , for small samples). The weights were used to calculate evidence ratios to select the best fit model (Burnham and Anderson 2002). Variance components were 11 calculated for the random factors to assess their relative contribution to total variance in drift abundance, with the residual variance representing variance among replicate samples collected on a single day. This analysis was applied to total invertebrate biomass, aquatic invertebrates and terrestrial invertebrates separately. Spatial variation: factors influencing drift abundance across multiple streams at a landscape scale Thirty sites (including the four sites used for the temporal study) in the Coast and Interior regions of British Columbia were sampled for drift abundance and habitat features using the protocols described above in the summer of 2005 or 2006 (Figure 2.1). At the four temporal variation sites a third sample was collected on the first August sampling date, and the three samples from that date were used for the spatial analysis. Three drift samples were taken at all other sites, and average biomasses of replicate samples were used in analysis. The 30 Figure 2.1: Locations of study streams in south western British Columbia, Canada. Some sites are close enough together that they share a point. Number labels refer to row labels in Appendix 1. Universal Transverse Mercator coordinates are also available in Appendix 1. 12 sites were chosen to cover a wide range of productivity levels and stream sizes (1.7-33.1 m channel width; Appendix 1). Measurements of habitat features were made on the same day as samples were collected when possible. Habitat Measurements Spatial variation in drift, between and within streams, can also be considerable and is influenced by measurable habitat factors. However, the measurement of habitat effects is complicated by the impact of upstream invertebrate density, which can be detected long distances downstream (Waters 1965, Wipfli and Gregovich 2002). Habitat factors must therefore be consistent for a great enough distance that upstream invertebrate production levels do not mask their effects. That is, the impact of a short stretch of stream with no canopy cover probably cannot be detected downstream or even within that stretch due to mixing with invertebrate drift originating upstream. For this reason habitat surveys were conducted over 80-250 m and average values were calculated for the surveyed area. Invertebrate drift samples were collected within the survey area. Habitat surveys were conducted at all 30 sites (see Moore et al. 1997) to collect information on reach-level factors that may influence invertebrate drift abundance. Average summer temperature was estimated for each site using ClimateBC V.3.2 (Wang et al. 2006) and UTM coordinates and elevation measurements for each site. Conductivity was measured once at each site. The intermediate dimension (median length of the x, y, and z axes) of the five largest water-moved particles (Hogan 1996), and the percent canopy cover and the canopy composition (percent cover of conifer, alder, and other deciduous trees) were measured at 3-5 locations spaced throughout each site. Note that the sum of the conifer, alder and other deciduous canopy cover was often greater than the total canopy cover estimate when the different types of canopies overlapped. The length and width of each channel unit (riffle, pool, glide, run, and cascade) were measured, including channel units in secondary and backwater channels. These measurements were used to calculate mean bankfull width and the percent area of each site by channel unit type. Substrate composition (fines, gravel, cobbles, boulders, bedrock, percent particulate organic matter (POM)), gradient, and depth 13 of each channel unit were weighted by channel unit area to calculate average values for each site. Spatial Variation Statistical Analysis Regression models to predict total, aquatic, and terrestrial drift abundance were fit using linear models in R 2.4.1. Models were fit using ordinary least squares on transformed, centred data. In order to apply these models to new data sets the new data must be centred using the same constants as this study (see Appendix 1 to calculate variable means). Data were log transformed for normality (Appendix 1) and centred to reduce colinearity between main effects and interaction effects (Quinn and Keough 2002). After examining correlation between the measured habitat variables, explanatory variables with no correlation coefficients greater than 0.7 were selected as a starting point for model selection (if two variables were strongly correlated then the model selection process was started twice, once with each variable). These variables were average summer temperature (temp), conductivity (cond), largest particle (lp), percent alder cover, gradient, percent cover of gravel sediment (grav), and percent cover of organic sediment (org). Site elevation (elev) was related to region (region), and percent canopy cover (can) was correlated with bankful width (wb). One of elev and region (strongly related), one of can and wb, and the other seven target variables were included in four starting models for each invertebrate biomass value (total, aquatic, terrestrial). Interaction terms were added to the starting models based on improved AIC values. Terms were then removed from the starting models in a stepwise fashion until removing terms no longer improved AIC values. The models with the lowest AIC values were selected for comparison. The models were compared using Akaike weights based on the Second-Order Information Criterion (AICc for small samples). The weights were used to calculate evidence ratios to select the best fit model (Burnham and Anderson 2002). Tolerance, Cook’s Distance, and normality of the residuals were checked for all selected models. Interactions included in final models were investigated using simple slopes (Quinn and Keough 2002). The simple slope of a variable is its slope when the variable that it interacts with is held constant. Generally the simple slopes are calculated at the mean of the other variable, and one standard deviation above and below the mean (low, average and high 14 values of the interacting variable). Simple slope calculations allow greater understanding of trade-offs and other interactions. Habitat measurements from the two regions were compared using two-sample Student’s t-tests to examine differences between the regions. Results Temporal Variation Total invertebrate drift biomass in a sample varied from 1.59 mg/m3 in Husdon Creek (the small Coast stream) to 180.60 mg/m3 in Yard Creek (the large Interior stream; Figure 2.2). The four sites, Chapman, Husdon, Yard and Senn had averages (over all samples) of 26 ± 19 (standard deviation) mg/m3, 19 ± 24 mg/m3, 92 ± 44 mg/m3, and 44 ± 27 mg/m3, respectively. Average drift biomass in the Interior region was three times the biomass in the Coast region (66 ± 43 vs. 22 ± 21 mg/m3, respectively). Aquatic invertebrates made up 78% of the total invertebrate biomass averaged across all samples (Figure 2.3); this proportion was consistent between sites. Terrestrial invertebrates made up the remaining 22% (Figure 2.4). Terrestrial and aquatic invertebrates also showed the same relative drift densities between sites as total invertebrates; Yard Creek had the highest drift density followed by Senn, Chapman, and Husdon. The average terrestrial and aquatic invertebrate drift densities in the Interior region greater than in the Coast region (aquatic: 51 ± 41 mg/m3 and 17 ± 18 mg/m3, respectively; terrestrial: 15 ± 16 mg/m3 and 5 ± 10 mg/m3, respectively). The best fit model for total, aquatic, and terrestrial invertebrates based on AICc Akaike weights (ω i ) was the region model (including region fixed , site random , and Julian day random ; Table 2.1). The region model had an Akaike weight 2.9 times higher than the random factors model for total invertebrate density, 1.9 times higher than the random factors model for aquatic invertebrate density, and 2.2 times higher than the full model for terrestrial invertebrate density, indicating that the region model was approximately twice as likely as the next best model for all invertebrate categories. These results suggest that there was no consistent 15 Figure 2.2: Total invertebrate drift biomass of samples collected at four sites (A Senn Creek, B Yard Creek, C Husdon Creek, D Chapman Creek) over four months. Open and filled circles indicate the first and second visits to the site each month. No consistent between day variation is apparent; variation between within the sites appear to be the primary sources of variation. 0 50 10 0 15 0 20 0 A B 0 50 10 0 15 0 20 0 Jun Jul Aug Sep C Jun Jul Aug Sep D to ta l d rif t b iom as s ( m g m 3 ) 16 Figure 2.3: Aquatic invertebrate drift biomass of samples collected at four sites (A Senn Creek, B Yard Creek, C Husdon Creek, D Chapman Creek) over four months. Open and filled circles indicate the first and second visits to the site each month. No consistent between day variation is apparent; variation between within the sites appear to be the primary sources of variation. 0 50 10 0 15 0 20 0 A B 0 50 10 0 15 0 20 0 Jun Jul Aug Sep C Jun Jul Aug Sep D aq ua tic d rif t b iom as s ( m g m 3 ) 17 Figure 2.4: Terrestrial invertebrate drift biomass of samples collected at four sites (A Senn Creek, B Yard Creek, C Husdon Creek, D Chapman Creek) over four months. Open and filled circles indicate the first and second visits to the site each month. No consistent between day variation is apparent; variation between within the sites appear to be the primary sources of variation. 0 50 10 0 15 0 20 0 A B 0 50 10 0 15 0 20 0 Jun Jul Aug Sep C Jun Jul Aug Sep D te rre str ial d rif t b iom as s ( m g m 3 ) 18 Model AIC c Di lik ωi ωmax / ωi Total region + month + Julian day + site 170.77 6.57 0.04 0.03 26.7 region + Julian day + site 164.2 0 1 0.71 1 month + Julian day + site 172.63 8.43 0.01 0.01 67.5 Julian day + site 166.31 2.11 0.34 0.25 2.9 Aquatic region + month + Julian day + site 168.07 6.57 0.04 0.02 26.7 region + Julian day + site 161.5 0 1 0.63 1 month + Julian day + site 169.13 7.63 0.02 0.01 45.3 Julian day + site 162.81 1.31 0.51 0.32 1.9 Terrestrial region + month + Julian day + site 237.11 1.6 0.45 0.29 2.2 region + Julian day + site 235.51 0 1 0.65 1 month + Julian day + site 242.66 0.03 0.02 0.02 35.5 Julian day + site 241.12 0.06 0.06 0.04 16.5 Table 2.1: Likelihood ratio calculations for the temporal candidate models for total, aquatic and terrestrial drift. AIC c is the Second-Order Information Criterion (used for small sam- ples), Δ i is the difference between the AIC c and the minimum AIC c , lik is the likelihood of the model given the data (e^(- ½ * Δ i )), ω i is the likelihood of the model divided by the sum of the likelihoods for the competing models (see Burnham and Anderson 2002 ch 2). The ω max / ω i ratio indicates how much more likely the best model (ω max ) is compared to the candidate model (ω i ). seasonal effect on any of the invertebrate groups, but that drift biomass was higher in the Interior. There was strong support for the inclusion of region in the final model; the region- only model had a likelihood weight up to 16 times higher (was 16 times more likely; Table 2.1) than the random factors model that rejected region. The coefficients for region for total, aquatic, and terrestrial invertebrates confirm that invertebrate drift is higher in the Interior than the Coast region (Table 2.2). Variance components analysis indicated that the random variation contributed by Julian day was negligible for total and aquatic invertebrate drift and low (8%) for terrestrial invertebrates (Figure 2.5). Between-site variation is negligible for terrestrial invertebrates but contributes 28% of the random variation for total invertebrate drift and 47% for aquatic drift. Within- 19 Invertebrates Intercept (SE) Region (Interior) coefficient (SE) Total 2.68 (0.39) 1.29 (0.55) Aquatic 2.39 (0.54) 1.24 (0.76) Terrestrial 0.46 (0.31) 1.58 (0.44) Table 2.2: Model coefficients and standard errors for the intercept and fixed factor (region) for total, aquatic and terrestrial invertebrates temporal models. Figure 2.5: Cumulative variance components of site, day, and residual (within-site) variation. The variance components show the proportion of random variation that can be attributed to site, day, and residual variation for total, aquatic and terrestrial invertebrate drift. Total Aquatic Terrestrial Pr op or tio n of V ar ian ce 0. 0 0. 2 0. 4 0. 6 0. 8 1. 0 Julian Day Site Residual 20 site variation (residual variation due to spatial replication (n = 2 samples) within a site) contributed the majority of the random variation for total, aquatic and terrestrial invertebrates (72%, 51% and 92%, respectively). The high residual variation of terrestrial invertebrates indicates greater variation between replicate samples than between days, possibly due to the small number of replicates (2) collected on each day, the effects of single large terrestrial invertebrates that contribute disproportionately to biomass (e.g. a single large beetle can skew a dry weight measurement) or greater accumulations of leaves bearing terrestrial invertebrates in one drift net than another. Spatial Variation Invertebrate biomass at the 30 sites varied over an order of magnitude from 7.9-283 mg/ m3, except for one site with very high biomass (High Falls with 1062 mg/m3; Appendix 1). Average invertebrate drift biomass was 97 ± 191 mg/m3 for total invertebrates, 52 ± 46 mg/m3 for aquatic invertebrates, and 45 ± 169 mg/m3 for terrestrial invertebrates. Similar to seasonal samples, aquatic invertebrates made up 76% ± 20% of total invertebrate drift biomass; aquatic invertebrate mass was greater than terrestrial invertebrate mass at all sites except for one Coast stream (High Falls) and one in the Interior (Criss) due to the presence of several large terrestrial invertebrates (caterpillars). Total invertebrate biomass was higher in the Coast than the Interior region (100 mg/m3 and 92 mg/m3), but this is largely due to a single site (High Falls) with particularly high biomass due to very large terrestrial inputs (several caterpillars fell into the stream). When this site was identified as an outlier (using a box plot) and excluded the average total invertebrate drift biomass in the Coast region fell to 44 ± 34 mg/m3. Aquatic invertebrate drift biomass was 41 ± 38 mg/m3 in the Coast region and 70 ± 52 mg/m3 in the Interior, while terrestrial invertebrate drift densities in the Coast and Interior were 8 ± 9 mg/m3 and 23 ± 34 mg/m3, respectively, with the High Falls Coast site excluded as an outlier (60 ± 218 mg/m3 with High Falls site included). The High Falls site was excluded from multiple regression analysis if its Cook’s Distance exceeded 1 indicating high influence (Quinn and Keough, 2002), otherwise it was included in analysis. Total canopy cover was significantly higher in the Coast region (39% ± 29% versus 17% ± 21 20%, Welch two-sample t-test on untransformed data: t = 2.48, p = 0.02). The proportion of total canopy in each category (alder, other deciduous, and canopy) did not differ between regions. Alder constituted 52% ± 38% of total canopy cover at Coast sites, and 50% ± 34% at Interior sites. Other deciduous trees were 25% ± 24% and 35% ± 49%, and coniferous trees were 23% ± 27% and 28% ± 37% in the Coast and Interior, respectively. Elevation (199 m and 517 m, Welch two-sample t-test: t = -6.74, p = 1.13 x 10-6) and conductivity (65.7 μS ± 41.6 μS and 130.0 μS ± 91.2 μS, Welch two-sample t-test: t = -2.3, p = 0.04) were lower at Coast sites. The final models predicting total, aquatic, and terrestrial invertebrate densities using habitat measurements were selected using the same evidence ratio method as for temporal variation (Table 2.3). All final models had acceptable Cook’s Distance and tolerance values and normally distributed residuals. The best models for total, aquatic, and terrestrial invertebrate densities used different variables (Table 2.4). The best model for total invertebrates had an r2 of 0.28 (F 4,24 = 2.38, p = 0.08), and included variables related primarily to aquatic and riparian productivity: conductivity, alder cover, canopy cover, and the interaction between conductivity and alder cover (Figure 2.6). Simple slopes analysis (Quinn and Keough 2002) of the interaction between conductivity and percent alder shows that the slope of the alder coefficient was positive (2.64) when conductivity was low (one standard deviation below the centred average of 0, log(conductivity) = 0.898). The alder coefficient was positive but lower (0.79) at average conductivity, and negative at high conductivity (-1.06 at one standard deviation above average conductivity). The same effect occurred for the slope of conductivity when alder was held at low, average and high values; the slope for conductivity was 0.65 at low alder cover (SD asin√alder = 0.216), 0.20 at average alder cover and -0.24 at one standard deviation above average. The best model for aquatic invertebrate drift had greater predictive power (r2 = 0.44; F 2,25 = 4.94, p = 0.004), and included region and flow- related habitat variables: gradient, bankfull width, and the interaction between region and gradient (Figure 2.7). Simple slopes analysis showed a coefficient for gradient of 0.62 in the Coast region and -0.34 in the Interior, indicating a positive effect of gradient on drift in the Coast but not in the Interior. The High Falls site did not have high influence on these two models and was included throughout the analysis. However, it had high influence for 22 Model AICc Δ j lik ω i ω max / ω i Total grad + wb + cond + ald + cond*ald 9.08 4.91 0.09 0.04 11.63 wb + cond + ald + cond*ald 7.05 2.88 0.24 0.11 4.23 cond + ald + cond*ald 6.56 2.39 0.3 0.14 3.31 reg+ reg*can + can + cond + ald + cond*ald 7.75 3.58 0.17 0.08 6 reg+ can + cond + ald + cond*ald 7.06 2.89 0.24 0.11 4.24 can + cond + ald + cond*ald 4.17 0 1 0.46 1 org + can + cond + ald + cond*ald 7.79 3.62 0.16 0.07 6.11 Aquatic cond + wb + reg + grad + reg*grad -3.49 4.48 0.11 0.04 9.4 wb + reg + grad + reg*grad -7.97 0 1 0.34 1 reg + grad + reg*grad -7.7 0.27 0.87 0.3 1.15 elev + grad + elev*grad -6.86 1.11 0.57 0.2 1.74 wb + elev + grad + elev*grad -5.93 2.04 0.36 0.12 2.77 Terrestrial wb + grad + lp + org + ald + grav + cond + cond*ald + lp*wb 31 15.15 0 0 1949.16 org + grav + elev + cond + elev*cond 20.08 4.22 0.12 0.07 8.27 ald + grav + elev + cond + elev*cond 19.89 4.03 0.13 0.07 7.51 grav + elev + cond + elev*cond 17.03 1.18 0.56 0.31 1.8 elev + cond + elev*cond 15.85 0 1 0.55 1 elev + grad + elev*grad -6.86 1.11 0.57 0.2 1.74 wb + elev + grad + elev*grad -5.93 2.04 0.36 0.12 2.77 Table 2.3: Likelihood ratio calculations (see Table 2.1) for the spatial candidate models for total, aquatic and terrestrial drift. Abbreviations are: grad (gradient), wb (bankfull width), cond (conductivity), ald (alder cover), reg (region), can (canopy cover), org (organic sedi- ment), elev (elevation), lp (largest particles). 23 log(total drift) = 3.8 + 0.2*log(cond) – 1.6*asin(√can) + 0.8*asin(√ald) – 2.1*log(cond)* asin(√ald) r2 = 0.28 SS df F p Partial r2 conductivity 0.374 1 0.414 0.526 0.017 canopy 3.547 1 3.927 0.059 0.141 alder 1.705 1 1.888 0.182 0.073 conductivity*alder 3.891 1 4.307 0.049 0.152 residuals 21.678 24 log(aquatic drift) = 3.4 + 0.6*reg(Coast) + 0.3*grad + 0.3*wb – 1.0*reg(Coast)*grad r2 = 0.44 SS df F p Partial r2 region 1.947 1 3.365 0.08 0.118 gradient 0.102 1 0.175 0.68 0.069 bankfull width 1.765 1 3.026 0.094 0.108 region*gradient 6.660 1 11.412 0.002 0.313 residuals 14.591 25 log(terrestrial drift) = 1.9 + 0.0005*elevation + 0.2*log(cond) + 0.0022*elev*log(cond) r2 = 0.31 SS df F p Partial r2 elevation 1.699 1 1.762 0.197 0.068 conductivity 3.022 1 3.135 0.089 0.116 elevation*conductivity 4.837 1 5.017 0.035 0.173 residuals 23.139 24 Table 2.4: Model coefficients and partial r2 values for total, aquatic and terrestrial spatial models. Abbreviations are same as for Table 2.3. 24 Figure 2.6: Scatter plots of the habitat factors included in the total drift model – conductivity, alder canopy cover and total canopy cover – against total drift. The lines indicate the model slopes, not the simple regression slopes. The solid lines indicate the model slope without any interaction effect (i.e. when all other values are held at their means). The dashed and dotted lines on the alder and conductivity graphs indicate the slope when the interacting value is low and high, respectively (i.e. the dashed line on the alder graph represents the slope of alder when conductivity is low). Analyses were performed on transformed data. Open symbols indicate Interior region sites, filled symbols indicate Coast sites. Figure 2.7: Scatter plots of the habitat factors included in the aquatic drift model – bankfull width and gradient – versus total drift. The lines indicate the model slopes, not the simple regression slopes. The solid line on the bankfull width graph indicates the model slope. The dashed line on the gradient graph is the slope in the Coast region. The dotted line is the slope in the Interior. Analyses were performed on transformed data. Open symbols indicate Inte- rior region sites, filled symbols indicate Coast sites. 5 10 15 20 25 30 10 20 50 10 0 bankfull width (m) 0 1 2 3 4 5 10 20 50 10 0 gradient (%)a qu at ic dr ift bio m as s ( m g m 3 ) 50 100 150 200 250 10 50 20 0 50 0 conductivity(�S) 0 10 20 30 40 50 10 50 20 0 50 0 alder canopy cover (%) 0 20 40 60 80 10 50 20 0 50 0 total canopy cover (%) to ta l d rif t b iom as s ( m g m 3 ) 25 Figure 2.8: Scatter plots of the habitat factors included in the terrestrial drift model – con- ductivity and elevation – against total drift. The lines indicate the model slopes, not the simple regression slopes. The solid lines indicate the model slope without any interaction effect (i.e. when all other values are held at their means). The dashed and dotted lines on the conductivity graph indicate the slopes when elevation is low and high, respectively. Analy- ses were performed on transformed data. Open symbols indicate Interior region sites, filled symbols indicate Coast sites. Data from High Falls, the site with high drift biomass, was ex- cluded from the final model due to high influence and is therefore not shown on these graphs. 50 100 150 200 250 conductivity(�S) 1 5 20 10 0 200 400 600 800 elevation (m) 1 5 20 10 0 te rre str ial d rif t b iom as s ( m g m 3 ) 26 terrestrial invertebrates and was therefore excluded. The best model for terrestrial drift included elevation, conductivity, and the interaction between elevation and conductivity (Figure 2.8) (r2 of 0.31, F 3,24 = 3.62, p = 0.03). Simple slopes analysis of the interaction between conductivity and elevation shows that conductivity has a neutral or negative effect on terrestrial drift at low elevation and a positive effect at high elevation (coefficient of -0.39 at elevation = -249, 0.16 at average elevation and 0.71 at elevation = 249). These final models all had tolerance values above 0.1, Cook’s Distance values below 1, and normally distributed residuals Discussion Day-to-day variation of daytime samples is much less than spatial variation between and within sites (that is, the difference in the average drift biomass between days is less than the difference between samples taken on one day), suggesting that sampling effort should be focused on collecting more samples from each site, rather than more visits to each site. This result addresses our first objective: to determine whether drift must be measured on multiple days. Shearer et al. (2002) found that day-to-day and within site variation were similar for 24 hour samples, whereas this study found greater within site variation than day- to-day variation. However, Shearer et al. (2002) came to the same conclusion that sampling effort should be concentrated on one day when the objective is to compare between sites. The daytime-only sampling used in this study probably resulted in lower variability than in Shearer et al. (2002) as only the daytime drift was captured, rather than the more variable night peaks. These results contrast with Williams (1980) and Gibson and Galbraith (Gibson and Galbraith 1975), who found high day-to-day variability, although the relative magnitudes of day-to-day, between and within site variability were not measured. Williams (1980) used 24 h samples and Gibson and Galbraith (Gibson and Galbraith 1975) used 48 h samples, capturing two successive nights of peaks, potentially increasing between sample variation. None of these studies reported controlling for rain events as we did in this study, possibly contributing to high day-to-day variation. 27 The majority of invertebrate drift in this study was of aquatic origin, and aquatic invertebrate drift had lower day-to-day variability than terrestrial drift (1% and 8% respectively, Figure 2.5). Other studies have found that terrestrial invertebrates constitute the majority of invertebrate drift (eg. Nakano et al. 1999), but also found greater variability in terrestrial invertebrates (Kawaguchi and Nakano 2001, Nakano et al. 1999). The greater day-to-day variability of terrestrial invertebrates may account for the higher temporal variability found in other studies compared to this study. In streams where terrestrial invertebrates contribute more to drift samples, the relatively stable levels of aquatic invertebrates may be masked by variation in terrestrial inputs. The higher temporal variability of terrestrial invertebrates may be a sampling artefact due to the lower number and larger individual biomass of terrestrial prey – a single large beetle or caterpillar is much larger than the majority of aquatic invertebrates caught in drift nets – or temporal invertebrates may be influenced by random events such as high wind. Our samples collected on windy days captured many leaves, and these leaves could have been vectors for the large numbers of Hemipterans (aphids and leaf hoppers) found in these samples. The absence of a strong seasonal effect in this study may be caused by sampling protocol, but this result was well supported by the data. Other studies found greatest invertebrate density in midsummer (Elliott 2002, Shearer et al. 2002) and in late spring (Hansen and Closs 2007, Rincón and Lobón-Cerviá 1997). This study used small sample sizes and had large residual variation; these effects may have masked a seasonal effect. However, the Akaike weights for the model with month effect but without region effect indicate that it has low likelihood. Adding month to the random factors only model decreased the likelihood of the random factors model by 25, 32 and 2 times for the total, aquatic and terrestrial invertebrates respectively (Table 2.1), indicating strong support for dropping month from the model. The absence of a seasonal effect indicates that daytime samples collected throughout the summer months can be compared to each other, lengthening the sampling window. Total drift biomass was most influenced by habitat factors that control food availability. Lower invertebrate densities were found in areas with more canopy cover. It is also possible that the negative influence of canopy cover was related to bankfull width; wider streams 28 (with less canopy cover) tended to have greater densities of drifting invertebrates, probably due to higher relative discharge. However, the models that included bankfull width instead of canopy cover were less likely according to their AIC c values (Table 2.3), supporting the interpretation that direct shading effects are a more probable cause of the negative relationship with invertebrate densities. Alder cover has a positive influence on invertebrate drift density, indicating that detritus is a major source of food for aquatic invertebrates as found by Wipfli and Musselwhite (2004) and Rosenfeld (2000), or a direct source for terrestrial invertebrates (Mason and MacDonald 1982), or both. Conductivity also has a positive influence on total invertebrate density, suggesting that the periphyton is a major food source for aquatic invertebrates, or that conductivity and terrestrial invertebrate drift biomass are related. Shearer et al. (2003) found a density dependent relationship between chlorophyll a concentration and invertebrate drift, indicating that greater periphyton abundance can support more invertebrates and therefore higher baseline drift levels as found in this study. The significant negative interaction between conductivity and alder density is most likely a result of the accelerating impact of conductivity on decomposition of detrital food sources (Greenwood et al. 2007). Both alder and conductivity have positive coefficients when the other is low (Table 2.4); when stream nutrient levels are low there is food available from the detritus-based food chain and when canopy levels are low there is abundant light for the periphyton-based chain. However, when alder densities are high there is less light available for periphyton. Some periphyton species, such as diatoms, are able to compensate for low light levels and produce nearly the same biomass as in full light (Rier et al. 2006), while others, such as chlorophytes, require more light (Richardson et al. 1983). Therefore when light is restricted there can be a drop in primary productivity, depending on what species of algae dominate the periphyton. In these streams the detritus-based food chain becomes more important. A large proportion of the detritus in streams is deposited in the fall, and consumed throughout the year; when nutrient levels (conductivity) are high decomposition rates increase (Greenwood et al. 2007), due to the nutrient needs of decomposers such as bacteria, and the detritus may not last through the year. Invertebrates dependent on detritus will not be able to sustain high densities if the detritus supply is too low due to decomposition. For these reasons the negative interaction between conductivity and alder density is not surprising. 29 Aquatic invertebrate biomass was most correlated with physical habitat variables, not the productivity variables that impacted total invertebrates. It is surprising that productivity variables were not included in the aquatic model; the physical variables may simply overwhelm the productivity variables when only aquatic invertebrates are considered. However, conductivity was included in a candidate model (Table 2.3); its exclusion may be primarily related to low sample size. Region was included in the final model as a factor variable and explained 12% of the total variation of aquatic invertebrates (Table 2.4). Invertebrate density was consistently higher in the Interior than the Coast. The inclusion of region in the final model indicates that the habitat factor(s) that caused greater density in the Interior were not measured in this study. Bankfull width had a positive but non- significant influence on aquatic invertebrate drift density, as did gradient averaged over all of the sites. Water velocity rises with gradient and bankfull width; greater water velocity results in longer invertebrate suspension distances, and greater drift densities because the time that each invertebrate spends in the drift increases (Elliott 2002). The negative interaction between region and gradient shows that the slope for gradient is significantly higher in the Coast region than in the Interior, where it is actually slightly negative (Figure 2.7). These results contrast with other examinations of invertebrate biomass that have found that food availability has the greatest impact (Richardson 1993, Rosenfeld 2000, Wipfli and Musselwhite 2004). However, flow has been shown to influence invertebrate drift biomass as well (Elliott 2002, Harvey et al. 2006). Further research into the relationship between benthic invertebrate biomass, water velocity, and drift invertebrate biomass could clarify the sources of variation of aquatic invertebrate biomass. Surprisingly, terrestrial invertebrate drift biomass was not influenced by canopy cover or composition, but rather by elevation and conductivity. The relationship between canopy cover and terrestrial invertebrate input to streams is well studied but inconsistent (positive effect: Mason and MacDonald 1982, no effect: Wipfli , Wipfli and Musselwhite 2004). The relationship between conductivity and terrestrial invertebrate density is probably indirect. Riparian vegetation composition may influence conductivity as well as the input of terrestrial invertebrates to the stream. Alternatively, water conductivity and terrestrial 30 invertebrates may both be influenced by a factor that was not measured in this study such as underlying geology. Site elevation was probably included in the final model because of the significant interaction effect with conductivity. At low elevations conductivity has a negative coefficient, but at average and high elevations the coefficient is positive. The temporal variation component of this study showed that terrestrial invertebrates had more day to day variability than aquatic invertebrates; this variability makes it more difficult to determine what spatial factors influence terrestrial invertebrate density, as there is more unexplained variation than for total and aquatic invertebrates. Physical habitat features that are not usually measured for stream ecology studies such as average wind speed may influence terrestrial invertebrate drift density. The coefficients of determination for the three final models were relatively low, but within the typical range for habitat capacity models (r2 for the total, aquatic, and terrestrial drift were 0.28, 0.44, and 0.31, respectively). Multiple regression habitat capacity models often yield similar coefficients of determination for other species (e.g. Davies 1989, Scarnecchia and Bergersen 1987). The model for total drift includes taxonomically diverse invertebrates with different habitat preferences, so the low explanatory power is not surprising. The taxa were divided into aquatic and terrestrial types, each of which is less taxonomically diverse than all taxa combined. This decreased diversity may have increased the predictive power for aquatic invertebrates but not for terrestrial. Aquatic and terrestrial biomass were affected by different habitat variables as well, and showed different levels of day-to-day variability. The terrestrial model did not have increased predictive ability like that of the aquatic model, possibly because of the greater day-to-day variability found in the temporal study. Nevertheless, these models provide some insight into the processes that affect drift biomass and suggest specific habitat factors that merit further research. Measurable habitat factors that influence productivity and water velocity affect the abundance of juvenile salmonid prey. As prey abundance is positively correlated with growth and survival (Imre et al. 2004, Nielsen 1992, Nislow et al. 1998, Rosenfeld et al. 2005) it is an important consideration for juvenile salmonid habitat managers and fisheries scientists. Many of the habitat factors measured in this study are also used in salmonid habitat capacity 31 models (eg. Davies 1989, Scarnecchia and Bergersen 1987) and habitat assessment protocols (Moore et al. 1997). These habitat factors impact juvenile salmonids directly and indirectly their impact on prey abundance. As this study shows, the impact on prey abundance is complex, involving interactions between habitat factors. Because of the complexity of the relationship, the impact of habitat on invertebrate drift should be considered directly by managers and scientists. Prey habitat requirements could be incorporated into management decisions – for example, (Romero et al. 2005) discuss the practise of removing deciduous riparian vegetation in the Pacific Northwest to encourage conifers. The practise is intended to increase the supply of large woody debris, an important structural element of stream habitat, but may simultaneously decrease prey abundance. Direct measurement of invertebrate drift and a better understanding of habitat relationships could improve salmonid management practises in the future. 32 References Allan, J.D., and Russek, E. 1985. The quantification of stream drift. Canadian Journal of Fisheries and Aquatic Science 42: 210-215. Allan, J.D., Wipfli, M.S., Caouette, J.P., Prussian, A., and Rodgers, J. 2003. Influence of streamside vegetation on inputs of terrestrial invertebrates to salmonid food webs. Canadian Journal of Fisheries and Aquatic Sciences 60: 309-320. Bacon, P.J., Gurney, W.S.C., Jones, W., McLaren, I.S., and Youngson, A.F. 2005. Seasonal growth patterns of wild juvenile fish: partitioning variation among explanatory variables based on individual growth trajectories of Atlantic salmon (Salmo salar) parr. Journal of Animal Ecology 74: 1-11. Bates, D. 2007. The lme4 Package. Benke, A.C., Huryn, A.D., Smock, L.A., and Wallace, J.B. 1999. Length-mass relationships for freshwater macroinvertebrates in North America with particular reference to the southeastern United States. Journal of the North American Benthological Society 18(3): 308-343. Binns, N.A., and Eiserman, F.M. 1979. Quantification of fluvial trout habitat in Wyoming. Transactions of the American Fisheries Society 108: 215-228. Brittain, J.E., and Eikeland, T.J. 1988. Invertebrate drift -- a review. Hydrobiologia 166: 77- 93. Burnham, K.P., and Anderson, D.R. 2002. Model Selection and Multimodel Inference: A practical information-theoretic approach. Springer, New York. Chadwick, M.A., and Huryn, A.D. 2007. Role of habitat in determining macroinvertebrate production in an intermittent-stream system. Freshwater Biology 52: 240-251. Davies, P.E. 1989. Relationships between habitat characteristics and population abundance for brown trout, Salmo trutta L., and blackfish, Gadopsis marmoratus Rich., in Tasmanian streams. Australian Journal of Marine and Freshwater Research 40: 341- 359. Edwards, E.D., and Huryn, A.D. 1995. Annual contribution of terrestrial invertebrates to a New Zealand trout stream. New Zealand Journal of Marine and Freshwater Research 29: 467-477. Elliott, J.M. 2002. The drift distances and time spent in the drift by freshwater shrimps, Gammarus pulex, in a small stony stream, and their implications for the interpretation of downstream dispersal. Freshwater Biology 47: 1403-1417. Fausch, K.D., Hawkes, C.L., and Parsons, M.G. 1988. Models that predict standing crop of stream fish from habitat variables: 1950-1985. Gen Tech Rep, U.S. Department of Agriculture Forest Service Pacific Northwest Research Station, Portland, Oregon. Flecker, A.S. 1992. Fish predation and the evolution of invertebrate drift periodicity - evidence from neotropical streams. Ecology 73(2): 438-448. Gibson, R.J., and Galbraith, D. 1975. The relationships between invertebrate drift and 33 salmonid populations in the Matamak River Quebec, below a lake. Transactions of the American Fisheries Society 104: 529-535. Greenwood, J.L., Rosemond, A.D., Wallace, J.B., Cross, W.F., and Weyers, H.S. 2007. Nutrients stimulate leaf breakdown rates and detritivore biomass: bottom-up effects via heterotrophic pathways. Oecologia 151: 637-649. Hansen, E.A., and Closs, G.P. 2007. Temporal consistency in the long-term spatial distribution of macroinvertebrate drift along a stream reach. Hydrobiologia 575: 361- 371. Harvey, B.C., Nakamoto, R.J., and White, J.L. 2006. Reduced streamflow lowers dry-season growth of rainbow trout in a small stream. Transaction of the American Fisheries Society 135: 998-1005. Hogan, D. 1996. Channel Assessment Procedure Field GuideBook. Holt, C.S., and Waters, T.F. 1967. Effect of light intensity on the drift of stream invertebrates. Ecology 48: 225-234. Imre, I., Grant, J.W.A., and Keeley, E.R. 2004. The effect of food abundance on territory size and population density of juvenile steelhead trout (Oncorhynchus mykiss). Oecologia 138: 371-378. Johansen, M., Elliott, J.M., and Klemetsen, A. 2005. Relationships between juvenile salmon, Salmo salar L., and invertebrate densities in the River Tana Norway. Ecology of Freshwater Fish 14: 331-343. Kawaguchi, Y., and Nakano, S. 2001. Contribution of terrestrial invertebrates to the annual resource budget for salmonids in forest and grassland reaches of a headwater stream. Freshwater Biology 46: 303-316. Lancaster, J. 1999. Small-scale movements of lotic macroinvertebratees with variations in flow. Freshwater Biology 41: 605-619. Mason, C.F., and MacDonald, S.M. 1982. The input of terrestrial invertebrates from tree canopies to a stream. Freshwater Biology 12: 305-311. Matthaei, C.D., Werthmuller, D., and Frutiger, A. 1998. An update on the quantification of stream drift. Arch. Hydrobiol. 143(1): 1-19. McCauley, E. 1984. The estimation of the abundance and biomass of zooplankton in samples. In A manual on methods for the assessment of secondary productivity in freshwaters. Edited by J.A. Downing and F.H. Rigler. Blackwell Scientific Publications, Oxford. pp. 228-265. Meyer, E. 1989. The relationship between body length parameters and dry mass in running water invertebrates. Arch. Hydrobiol. 117(2): 191-203. Moore, K., Jones, K., and Dambacher, J. 1997. Methods for Stream Habitat Surveys: Aquatic Inventory Project. Oregon Department of Fish and Wildlife Natural Production Program, Corvallis, OR, 97330. Nakano, S., Kawaguchi, Y., Tanaguchi, Y., Miyasaka, H., Shibata, Y., Urabe, H., and Kuhara, 34 N. 1999. Selective foraging on terrestrial invertebrates by rainbow trout in a forested headwater stream in northern Japan. Ecological Research 14: 351-360. Nielsen, J.L. 1992. Microhabitat-specific foraging behavior, diet, and growth of juvenile coho salmon. Transactions of the American Fisheries Society 121: 617-634. Nislow, K.H., Folt, C., and Seandel, M. 1998. Food and foraging behavior in relation to microhabitat use and survival of age-0 Atlantic salmon. Canadian Journal of Fisheries and Aquatic Sciences 55: 116-127. Quinn, G.P., and Keough, M.J. 2002. Experimental design and data analysis for biologists. Cambridge University Press. Rader, R.B. 1997. A functional classification of the drift: traits that influence invertebrate availability to salmonids. Canadian Journal of Fisheries and Aquatic Sciences 54: 1211-1234. Resh, V.H., and Rosenberg, D.M. 1989. Spatial-temporal variability and the study of aquatic insects. The Canadian Entomologist 121: 941-963. Richardson, J.S. 1993. Limits to productivity in streams: Evidence from studies of macroinvertebrates. In Canadian Special Publication of Fisheries and Aquatic Sciences. Edited by R.J. Gibson and R.E. Cutting. pp. 9-15. Richardson, K., Beardall, J., and Raven, J.A. 1983. Adaptation of unicellular algae to irradiance: an analysis of strategies. New Phytologist 93(2): 157-191. Rier, S.T., Stevenson, R.J., and LaLiberte, G.D. 2006. Photo-acclimation response of benthic stream algae across experimentally manipulated light gradients: a comparison of growth rates and net primary productivity. Journal of Phycology 42: 560-567. Rincón, P.A., and Lobón-Cerviá, J. 1997. Temporal patterns in macroinvertebrate drift in a northern Spanish stream. Marine and Freshwater Research 48: 455-464. Roff, J.C., and Hopcroft, R.R. 1986. High precision microcomputer based measuring system for ecological research. Canadian Journal of Fisheries and Aquatic Sciences 43: 2044- 2048. Romero, N., Gresswell, R.E., and Li, J.L. 2005. Changing patterns in coastal cutthroat trout (Oncorhynchus clarki clarki) diet and prey in a gradient of deciduous canopies. Canadian Journal of Fisheries and Aquatic Sciences 62: 1797-1807. Rosenfeld, J. 2000. Effects of fish predation in erosional and depositional habitats in a temperate stream. Canadian Journal of Fisheries and Aquatic Sciences 57: 1369-1379. Rosenfeld, J.S., Leiter, T., Lindner, G., and Rothman, L. 2005. Food abundance and fish density alters habitat selection, growth, and habitat suitability curves for juvenile coho salmon (Oncorhynchus kisutch). Canadian Journal of Fisheries and Aquatic Sciences 62: 1691-1701. Sabo, J.L., Bastow, J.L., and Power Mary, E. 2002. Length-mass relationships for adult aquatic and terrestrial invertebrates in a California watershed. Journal of the North American Benthological Society 21(2): 336-343. 35 Sample, B.E., Cooper, R.J., Greer, R.D., and Whitmore, R.C. 1993. Estimation of insect biomass by length and width. American Midland Naturalist 129(2): 234-240. Scarnecchia, D.L., and Bergersen, E.P. 1987. Trout production and standing crop in Colorado's small streams, as related to environmental features. North American Journal of Fisheries Management 7: 315-330. Schell, C. 1999. The response of periphyton chlorophyll A and invertebrate drift to the end of nutrient enrichment, University of British Columbia. Shearer, K.A., Hayes, J.W., and Stark, J.D. 2002. Temporal and spatial quantification of aquatic invertebrate drift in the Maruia River, South Island, New Zealand. New Zealand Journal of Marine and Freshwater Research 36: 529-536. Shearer, K.A., Stark, J.D., Hayes, J.W., and Young, R.G. 2003. Relationships between drifting and benthic invertebrates in three New Zealand rivers: implications for drift- feeding fish. New Zealand Journal of Marine and Freshwater Research 37: 809-820. Slaney, P.A. 1972. Effects of prey abundance on distribution, density and territorial behaviour of young rainbow trout in streams, University of British Columbia. Smock, L.A. 1980. Relationships between body size and biomass of aquatic insects. Freshwater Biology 10: 375-383. Wallace, J.B., Eggert, S.L., Meyer, J.L., and Webster, J.R. 1997. Multiple trophic levels of a forest stream linked to terrestrial litter inputs. Science 277: 102-104. Wang, T., Hamann, A., Spittlehouse, D., and Aitken, S.N. 2006. Development of scale- free climate data for western Canada for use in resource management. International Journal of Climatology 26(3): 383-397. Waters, T.F. 1962. Diurnal periodicity in the drift of stream invertebrates. Ecology 43(2): 316-320. Waters, T.F. 1965. Interpretation of invertebrate drift in streams. Ecology 46: 327-334. Williams, D. 1980. Invertebrate drift lost to the sea during low flow conditions in a small coastal stream in western Canada. Hydrobiologia 75: 251-254. Wipfli, M. 1997. Terrestrial invertebrates as salmonid prey and nitrogen sources in streams: contrasting old-growth and young-growth riparian forests in southeastern Alaska USA. Canadian Journal of Fisheries and Aquatic Sciences 54: 1259-1269. Wipfli, M.S., and Gregovich, D.P. 2002. Export of invertebrates and detritus from fishless headwater streams in southeastern Alaska: implications for downstream salmonid production. Freshwater Biology 47: 957-969. Wipfli, M.S., and Musselwhite, J. 2004. Density of red alder (Alnus rubra) in headwaters influences invertebrate and detritus subsidies to downstream fish habitats in Alaska. Hydrobiologia 520: 153-163. 36 Chapter 3: Influence of physical and biological habitat factors on juvenile salmonid abundance1 Introduction Accurate population size estimation is fundamental to species management, from conservation of endangered species’ habitat to harvest of exploited species. Models that predict abundance based on habitat characteristics serve two primary purposes in population management: they improve understanding of species’ habitat requirements, and they can be a more cost-effective way to estimate abundance than direct field measurement. Habitat models are often applied to identify high quality habitat for protection, a primary objective of many conservation projects (e.g. Clark et al. 1993). They are also useful indicators of potential habitat capacity when fish abundance is depleted due to over-harvest. Designing habitat restoration projects and monitoring the effects of resource extraction on habitat quality can be addressed using predictive models as well (Binns and Eiserman 1979). Such models have been developed for a variety of taxa including mammals (Dorgelöh 2006, McAlpine et al. 2006), amphibians (Denoël and Lehmann 2006), birds (Gavashelishvili et al. 2006, Zharikov et al. 2006), and freshwater fish (see Fausch et al. 1988). Freshwater fish such as salmonids (Family Salmonidae) are well suited to habitat modeling due to their restricted habitat in lakes and streams and the difficulty of obtaining direct population estimates. However, they are often regionally specific (Bowlby and Roff 1986) – a model developed in one location is rarely a good predictor in another – although similar variables are included in most models. Variables used in regression analysis can be divided into those that are related to the physical structure of available habitat, and those that are correlated with the abundance or productivity of prey. For example, physical habitat variables like channel width, percent pool habitat, and LWD abundance are commonly used as correlates of habitat structure. Stream gradient and sediment size (see Coulombe- Pontbriand and LaPointe 2004, Davies 1989, Rosenfeld et al. 2000, Sharma and Hilborn 1 A version of this chapter will be submitted for publication: Nicol, SD, JS Rosenfeld and JB Shurin. Influence of physical and biological habitat factors on juvenile salmonid abundance. 37 2001) are additional variables that correlate with channel size and velocity, which can be expected to influence habitat availability for fish. Invertebrate drift is the most direct index of prey abundance for drift-feeding fishes. Invertebrate prey are delivered to fish by two primary mechanisms: in situ invertebrate production based on algal- or detrital-based food chains (Vannote et al. 1980), or input of terrestrial invertebrates from the riparian zone (Cada et al. 1987). Canopy cover and composition, organic sediment, temperature, and water conductivity (see Binns and Eiserman 1979, Johansen et al. 2005, Scarnecchia and Bergersen 1987) affect the abundance of invertebrate prey delivered from terrestrial and aquatic sources. Invertebrate drift is conspicuously absent from the majority of juvenile salmonid habitat models (Fausch et al. 1988) even though drifting aquatic and terrestrial invertebrates are a primary food source for juvenile salmonids (Wipfli 1997), and elevated drift contributes to increased juvenile salmonid growth (Imre et al. 2004, Nielsen 1992, Rosenfeld et al. 2005, Slaney 1972), density (Slaney 1972) and survival (Imre et al. 2004, Nislow et al. 1998). Benthic chlorophyll-a, riparian canopy cover, water column nutrients (or surrogates like conductivity) are often used as expected correlates of the abundance of prey for drift-feeding fish. Benthic invertebrate abundance is sometimes used in models as a more direct measure of food availability (eg. Bowlby and Roff 1986, Jowett 1992). Invertebrate drift should be a more direct measurement of food availability, as benthic density and drift biomass are not strongly correlated (Shearer et al. 2003). Including invertebrate drift biomass as a candidate variable in habitat models may therefore improve the predictive ability of habitat capacity models and help determine the degree to which food availability limits salmonid abundance. The predictive models or variables that biologists include in regression analysis are usually based on expert opinion and professional judgement as to which are most effective and practical to collect. However, models are rarely selected in an explicit cost-benefit framework, where the benefit is the proportion of variance in fish abundance explained by a model (or variable) and the cost is the time and effort required to collect the associated data in the field or laboratory Assessing the cost-effectiveness of habitat models and their terms (the habitat variables) should allow fisheries scientists to choose which habitat models and component variables maximize predictive power. This improved efficiency is important for 38 fisheries scientists with limited resources (i.e. personnel, funding), and should allow scientists to make more informed decisions about the optimal variables to include in habitat models. For example, a variable with high predictive ability that takes more time to measure than fish abundance is clearly less efficient than a direct measurement. The objectives of this study were 1) to develop reach-scale predictive models for juvenile salmonid abundance in southwestern BC using physical and biological habitat factors, 2) to determine whether habitat variables differ between the Coast and Interior regions, 3) to determine the contribution of invertebrate drift to the relative predictive ability of models, and 4) to do a cost-benefit analysis for predictive models and their component variables. Methods Study Sites Fifty stream sites in the Coast and Interior regions of British Columbia were sampled for drift abundance and habitat features in the summers of 2005 or 2006 (Figure 3.1). Sites were chosen to span a wide range of productivities and stream sizes (1.3 – 33.1 m channel width; Appendix 2). Every effort was made to find sites with reliable accounts of spawning adults to ensure that sites would be saturated with juveniles and rearing habitat would be near carrying capacity. Sample reach lengths were 20-30 bankfull widths (80-250 m) with no major tributaries or changes in valley form, vegetation, or land use (as in Moore et al. 1997). Fish density Data on salmonid density were either collected in the field (n = 19) or from existing multi- pass removal data sets (Hagen 2005, Hanson 2005, Morgan 2001, Rosenfeld et al. 2000, Tschaplinski 2006). Regression analysis showed that the year fish were collected had no significant impact on fish density (fish/m2, F 1,48 = 1.9, p = 0.17) or biomass (g wet mass/m2, F 1,45 = 0.9, p = 0.35). For streams without existing fish abundance data, a subset of channel units were triple- or double-pass depletion electrofished in August of 2006 or 2007. Removal 39 data were used to calculate an estimate of fish abundance as described by Schnute (1983). The wetted area electrofished was measured for each site to determine the density of fish (fish per m2 wetted area). Biomass (g wet mass per m2 wetted area) was calculated by multiplying the average wet mass of fish at a site by the estimated number of fish present. When only fish lengths were available biomass was estimated using the following power functions generated from the data set of fish lengths and masses collected at all sites: YOY: log(mass) = -11.73 + 3.07*log(length) . . . . . . . . . (F 1,3380 = 1.1 x 105, p = 2.2 x 10-16) rainbow: log(mass) = -11.54 + 3.04*log(length) . . . . . . . (F 1,1065 = 4.1 x 104, p = 2.2 x 10-16) coho: log(mass) = -12.05 + 3.14*log(length) . . . . . . . . . (F 1,1311 = 2.2 x 104, p = 2.2 x 10-16) Figure 3.1: Locations of study streams in south western British Columbia, Canada. Some sites are close enough together that they share a point. Number labels refer to row labels in Appendix 2. Universal Transverse Mercator coordinates are also available in Appendix 2. 40 To control for seasonal effects on fish size associated with different collection dates (July 12- September 22), we tested for a relationship between calendar day and average fish biomass. There was no relationship for YOY or coho, so biomass on the day of sampling was used in regression analysis. However, there was a significant day effect for rainbow trout (F 1,26 = 5.7, p = 0.02). Rainbow trout data were corrected to September 1st biomass by multiplying the number of days between the fishing date and September 1st by 0.013 (the slope of the mass- day relationship) and adding this correction to the measured mass of juvenile rainbow trout. This method maintained the existing variation in the data while correcting for the day effect. Invertebrate Drift Sampling Invertebrate drift samples were collected at the upstream end of pools immediately below riffles to standardize habitat effects and to measure the quantity of drift that a dominant fish would experience at the head of a pool. At sites with no suitable pools, nets were placed in riffles (approximately one third of the sites). Samples were collected with a 1 m long, 250 mm mesh net with an opening 0.2 m square for a duration of 180-300 minutes. The depth and velocity of the water in the mouth of the net were measured in order to calculate the volume of water sampled, and thereby the biomass of drifting invertebrates per m3. We sampled drift in the daytime rather than overnight because daytime drift abundance is more uniform (Waters 1962), and most juvenile salmonids are visual predators that feed during daylight. Arguably, daytime drift samples more accurately represent prey abundance for juvenile salmonids than 24 hour drift samples where biomass is dominated by nocturnal peaks (Waters 1962). Drift samples were collected at least two hours after sunrise and two hours before sunset to reduce daily variability in light intensity, and were not collected during or after rain events until flow had returned to pre-flood levels. Samples were preserved in 5% formalin and returned to the laboratory for sorting. Invertebrates were sorted from detritus at 10X magnification in the laboratory and preserved in 70% ethanol for later identification and measurement using a digitizer (Roff and Hopcroft 1986). Invertebrates were identified to Order with some exceptions: non-arthropods such as nematodes were identified to Phylum, and common arthropods such as chironomids were 41 identified to sub-Family. Each taxon was categorized as aquatic or terrestrial in origin. Adult insects with aquatic larvae, such as Nematocera (Diptera), were counted as aquatic in origin. The dry mass (mg) of invertebrates in each sample was calculated using published taxa- specific length-mass regressions (Benke et al. 1999, McCauley 1984 ; M. Wipfli lab pers. comm., Meyer 1989, Sabo et al. 2002, Sample et al. 1993, Smock 1980). Habitat Measurements Habitat surveys were conducted at all 50 sites (see Moore et al. 1997) to collect information on reach-level factors that may influence juvenile salmonid density. Average summer temperature was estimated using ClimateBC V.3.2 (Wang et al. 2006) and UTM coordinates and elevation measurements for each site. Conductivity was measured once at each site. The intermediate dimension (middle length of the x, y, and z axes) of the five largest particles likely to have been moved by the peak flow (Hogan 1996), percent canopy cover, and canopy composition (percent cover of conifer, alder [Alnus rubra], and other deciduous trees) were measured at 3 to 5 locations spaced throughout each site. The length and width of each channel unit (riffle, pool, glide, run, and cascade) were measured, including channel units in secondary and backwater channels. These measurements were used to calculate site mean bankfull width and the percent area of each site by channel unit type. The percent substrate composition – fines, gravel, cobbles, boulders, bedrock, and particulate organic matter (POM) – was estimated for each channel unit, as were gradient and the percent of the channel unit with protective cover. Measurements made in each channel unit were weighted by channel unit area to calculate average values for each site. All large woody debris (LWD; greater than 10cm diameter or 1m length) at each site were counted and identified as pool- forming or non-pool forming. Statistical Analysis: Habitat Capacity Modeling Habitat capacity models were developed using ordinary least squares multiple regression with R 2.6.0. Models were fit for young of the year salmonids (all taxa combined), and for 42 YOY rainbow trout (Oncorhynchus mykiss) and YOY coho salmon (O. kisutch) seperately. Models were fit using log-transformed density (number/m2 wetted area) and log-transformed biomass (grams wet mass/m2 wetted area) as response variables. All variables were centred to control for colinearity between model main effects and interaction effects (Quinn and Keough 2002). In order to apply these models to new data sets the new data must be centered using the same constants as this study (i.e. a constant must be added or subtracted to the data so that the mean equals zero; see Appendix 1 for variable means). Correlation between the habitat variables was examined to ensure that the subset selected for regression were not correlated (r < 0.7), as strongly correlated variables make it difficult or impossible to perform the matrix inversion required to calculate regression coefficients (Zar 1999). Fourteen non-correlated Category Variable Productivity organic substrate (proportion) org conductivity (mS) cond average summer temperature (°C) temp alder cover (proportion) alder canopy cover (proportion) canopy * invertebrate drift (dry mg m-3) drift ! Physical Habitat largest particle (cm) lp pool area (proportion) pool riffle (proportion) riffle gravel substrate (proportion) grav gradient (%) grad bankfull width (m) wb * percent cover for fish cover LWD pieces (m-1 of stream) lwd pool forming LWD pieces (m-1) lwd_pf Regional Effects region – Interior or Coast region elevation (m) elev ! drift was not included in the starting models * only one of can and wb was included in a starting model Table 3.1: Variables included in the starting models, with units, and abbreviations in bold. Variables marked with (*) are correlated with each other so only one at a time was used in any starting model. The variable marked with (!), drift, was not included in the starting mod- els for the full model selection, but was included in the starting models for the subset model selection. 43 variables were chosen to begin model selection (Table 2.1). Canopy cover and bankfull width were correlated with each other (r = 0.7), so two starting models were used: one with the 14 variables and canopy cover, and one with the 14 variables and bankfull width for a total of 15 variables. Two-way interaction terms were added to the starting models based on improved Akaike Information Criterion (AIC) values. Terms were then removed from the starting models in a stepwise fashion until removing terms no longer improved AIC values. The models with the lowest AIC values were selected for comparison. The models were compared using Akaike weights based on the Second-Order Information Criterion (AIC c for small samples). Weights were used to calculate evidence ratios to select the best fit model (Burnham and Anderson 2002). Tolerance, Cook’s Distance, and normality of the residuals were checked for all selected models. Interactions included in final models were investigated using simple slopes (Quinn and Keough 2002). The simple slope of a variable is its slope when the variable that it interacts with is held constant. Simple slopes were calculated at the mean of the other variable, and one standard deviation above and below the mean (low, average and high values of the interacting variable). Simple slope calculations allow greater understanding of trade-offs and other interactions by revealing changes in the sign or magnitude of the target variable coefficient, at different levels of the interacting variable. A best model was determined for each fish species – young-of-the-year of all species combined (YOY), of rainbow only (rainbow) and of coho only (coho) – for density and biomass separately all of the available sites (the full habitat model); not all sites had both coho and rainbow, so the sample size was lower for these models. Because invertebrate drift was collected at only 30 of 50 sites, best models were then run using the subset of sites that also had invertebrate drift measurements (the habitat subset model). To determine the benefit of including invertebrate drift in the best model, invertebrate drift was then added to the subset habitat model (the habitat + drift subset model) and the improvement in model fit assessed using evidence ratios. Finally, models were developed using only the subset of drift sites and including drift in the starting models (the reduced habitat + drift model). The main function of this last analysis was to determine if the drift term was dropped during 44 model selection or included in the final model. If drift was dropped then the resulting habitat model was considered to be inferior to the full habitat model (based on all available data), because the reduced + drift model was based on a subset of the data. In this case only the results of the full habitat models were presented. Region (Coast or Interior), was included in the starting models to determine if there was a region effect that was not captured by the habitat variables. Habitat measurements from the two regions were compared using two-sample Student’s t-tests to assess regional differences in average habitat characteristics. Statistical Analysis: Cost-Benefit Assessment To determine which variables explained the most variance in fish abundance for the least effort, a benefit:cost ratio was calculated for each variable included in the full habitat model, for drift from the habitat + drift subset model, and for the full habitat models themselves. Model benefit was the full or partial r2 of the model or variable, and cost was the sum of the time (in minutes) required to assess all of the variables in the model. The partial r2, a measure of a variable’s individual contribution to the predictive ability of the model, was calculated for each variable included in a final model. Field effort measurements were made at 18 of the 50 sites. We used time (effort) as the metric of cost since labour is the primary cost for all of the variables measured. The cost for each variable was the total time at the site required to measure a variable (including any subsequent laboratory processing if needed). Field activities were timed each summer after a training period, so that all estimates of field time are based on experienced field staff. All timed field work was done with teams of two people; if the two field staff worked together to make a measurement (e.g. channel unit length) then the number of minutes for that activity were doubled. We did not include travel time to the site because it made an equal contribution to all variables, and because it would be necessary to travel to the site in order to estimate fish abundance directly. However, at more remote sites with greater travel time, the relative cost of time spent on site is lower, and therefore the benefit of reduced field time 45 is lower than at sites with low travel time; this trade-off was not addressed in this cost-benefit analysis. Results Study Sites Invertebrate drift and conductivity (a correlate of nutrient levels) were higher in the Interior than the Coast whereas total canopy cover and alder cover were higher in the Coast (Table 3.2), suggesting higher prey availability at Interior sites compared to the more heavily shaded, nutrient-poor Coast. Habitat in the Coast was generally more structured with greater pool area, gravel substrate, and large woody debris. Despite the greater number of LWD pieces in the Coast region, there was no difference in the amount of pool-forming LWD between regions (0.007 pieces/m in the Coast, 0.004 pieces/m in the Interior; t = 1.0, p = 0.3), indicating that a higher proportion of Interior LWD pieces function in pool-formation. Interior sites also had significantly higher elevation (616 m compared to 102 m in the Coast; t = -10.2, p = 5.2 x 10-12). Density of young-of-the-year (YOY) salmonids averaged 1.0 ± 0.93 (standard deviation) fish per m2 of wetted habitat. Six salmonid species were collected: coho salmon, rainbow trout, cutthroat trout (O. clarki), chinook salmon (O. tshawytscha), bull trout (Salvelinus confluentus) and mountain whitefish (Prosopium williamsoni). Rainbow trout were present at 31 sites (Appendix 1) at an average density of 0.47 ± 0.39 fish/m2. Coho salmon were present at 39 sites at an average density of 0.59 ± 0.73 fish/m2. Fish biomass (g/m2) showed the same pattern; YOY had the greatest biomass (1.88 ± 2.01 g/m2), followed by coho (1.57 ± 2.05 g/m2), and rainbow (0.56 ± 0.49 g/m2). Mean fish densities and biomasses did not differ between regions (Table 3.2). Habitat Capacity Models All models included terms relating to both stream productivity and physical structure, and 46 Coast mean Interior mean Welch’s t p Fish Abundance YOY (fish/m3) 1.10 (0.90) 0.84 (0.97) 0.92 0.36 YOY (g/m3) 1.95 (1.38) 1.76 (2.95) 0.24 0.81 rainbow (fish/m3) 0.47 (0.42) 0.47 (0.39) -0.05 0.96 rainbow (g/m3) 0.51 (0.48) 0.60 (0.51) -0.5 0.62 coho (fish/m3) 0.60 (0.54) 0.57 (1.04) 0.1 0.92 coho (g/m3) 1.44 (1.01) 1.87 (3.53) -0.4 0.7 Productivity org 0.12 (0.13) 0.07 (0.14) 1.11 0.27 cond 66.2 (37.8) 121.1 (83.7) -2.69 0.01 temp 15.5 (0.93) 15.9 (0.95) -1.44 0.16 alder 0.23 (0.24) 0.09 (0.12) 2.64 0.01 canopy 0.44 (0.26) 0.19 (0.20) 3.78 <0.01 drift 43.7 (34.2) 92.4 (74.5) -2.11 0.05 Physical Structure lp 23.9 (26.9) 37.4 (27.9) -1.63 0.11 riffle 0.32 (0.23) 0.29 (0.26) 0.41 0.69 grav 0.34 (0.23) 0.23 (0.15) 2.02 0.05 grad 1.70 (1.60) 2.09 (1.04) -1.03 0.31 wb 7.68 (5.69) 10.5 (7.54) -1.38 0.18 pool 0.20 (0.18) 0.07 (0.11) 3.1 <0.01 cover 0.11 (0.06) 0.09 (0.05) 1.47 0.15 lwd 0.08 (0.08) 0.02 (0.03) 3.76 < 0.01 lwd_pf 0.01 (0.01) 0.004 (0.01) 1.04 0.3 Regional Effects elev 102 606 -9.85 <0.01 Table 3.2: Salmonid abundance and habitat variable means (SD) in the Coast and Interior regions and the results of Welch’s t-test comparing the means. Significant p-values are in bold, and show several significant differences among the habitat variables, but none for the salmonid abundances. Abbreviations are listed in Table 3.1. 47 log(YOY density) = 0.3 + 0.01(lp) + 2.6(pool) + 0.4(temp) – 0.003(cond) + 0.0003(elev) – 3.1(cover) + 1.6(grav) + 2.9(org) + 0.03(grad) + 36.6(lwd_pf) + 0.01(pool)(elev) – 0.0007(elev)(grad) + 54.0(cover)(org) + 1.52(grav)(grad) r2 = 0.75 Coefficients SS df F P partial r2 lp 0.01 2.02 1 6.65 0.01 0.17 pool 2.56 2.55 1 8.42 0.01 0.21 temp 0.44 3.61 1 11.9 < 0.01 0.27 cond -0.003 0.88 1 2.90 0.10 0.08 elev <0.01 0.00 1 0.01 0.94 0 cover -3.13 0.49 1 1.60 0.21 0.05 grav 1.55 0.2 1 0.67 0.42 0.02 org 2.86 3.46 1 11.42 < 0.01 0.26 grad 0.03 1.00 1 3.30 0.08 0.09 lwd_pf 36.56 4.35 1 14.34 < 0.01 0.31 pool:elev 0.01 2.6 1 8.59 0.01 0.21 elev:grad -0.001 1.83 1 6.03 0.02 0.16 cover:org 53.97 4.24 1 13.97 < 0.01 0.3 grav:grad 1.52 2.26 1 7.44 0.01 0.19 residuals 9.71 32 log(YOY biomass) = 0.2 – 0.003(lp) + 0.3(temp) – 0.005(cond) – 0.001(elev) – 0.1(grad) – 0.02(wb) – 0.01(lp)(grad) + 0.001(cond)(wb) r2 = 0.65 coefficients SS df F p partial r2 lp -0.003 0.34 1 0.76 0.39 0.02 temp 0.27 1.95 1 4.34 0.04 0.11 cond -0.005 5.04 1 11.22 < 0.01 0.24 elev -0.001 3.28 1 7.31 0.01 0.17 grad -0.13 0.82 1 1.82 0.19 0.05 wb -0.02 6.31 1 14.04 < 0.01 0.29 lp:grad -0.01 4.00 1 8.89 0.01 0.2 cond:wb 0.001 2.68 1 5.96 0.02 0.15 residuals 15.73 35 Table 3.3: Model coefficients, sums of squares (SS), degrees of freedom (df), F-, p-, and partial r2 values for YOY, rainbow and coho full models. Significant p-values are in bold. Model r2 values are listed following the equations. Abbreviations are listed in Table 1. Type II ANOVA results are presented. 48 log(rainbow density) = 0.7 + 0.03(lp) + 6.3(alder) + 0.01(cond) – 17.1(cover) + 41.2(lwd) + 0.2(region) – 5.2(canopy) – 0.04(lp)(region) r2 = 0.83 Coefficients SS df F P partial r2 lp 0.03 1.73 1 5.7 0.03 0.23 alder 6.3 8.11 1 26.71 < 0.01 0.58 cond 0.01 7.16 1 23.58 < 0.01 0.55 cover -17.09 10.3 1 33.92 < 0.01 0.64 lwd 41.23 13.0 1 42.93 < 0.01 0.69 region 0.16 0.26 1 0.85 0.37 0.04 canopy -5.22 9.17 1 30.19 < 0.01 0.61 lp:region -0.04 6.42 1 21.13 < 0.01 0.53 residuals 5.77 19 log(rainbow biomass) = 0.3 + 0.03(lp) + 6.2(alder) + 0.01(cond) – 15(cover) + 33(lwd) + 0.7(region I) – 4(can) – 0.03(lp)(region I) r2 = 0.76 coefficients SS df F p partial r2 lp 0.03 2.92 1 7.41 0.01 0.3 alder 6.17 8.03 1 20.39 < 0.01 0.55 cond 0.01 1.8 1 4.56 0.05 0.21 cover -15.2 7.47 1 18.95 < 0.01 0.53 lwd 32.65 7.18 1 18.23 < 0.01 0.52 region 0.73 0.37 1 0.93 0.35 0.05 can -3.99 5.14 1 13.04 < 0.01 0.43 lp:region -0.03 4.19 1 10.62 < 0.01 0.38 residuals 6.7 17 Table 3.3: continued 49 log(coho density) = 0.3 – 0.8(alder) – 1.7(pool) – 0.9(riffle) + 0.07(temp) – 0.01(cond) + 0.001(elev) + 2.3(grav) + 1.0(org) + 0.04(grad) + 2.0(lwd) + 0.4(region) – 0.002(elev)(grad) + 23.2(org)(region) – 6.3(tavsm)(lwd) r2 = 0.92 Coefficients SS df F P partial r2 alder -0.83 0.51 1 1.68 0.21 0.07 pool -1.68 1.25 1 4.13 0.05 0.15 riffle -0.92 1.18 1 3.89 0.06 0.14 temp 0.07 0.67 1 2.22 0.15 0.09 cond -0.01 2.19 1 7.23 0.01 0.24 elev 0.001 0.21 1 0.69 0.41 0.03 grav 2.31 5.57 1 18.42 < 0.01 0.44 org 1.04 1.52 1 5.03 0.03 0.18 grad 0.04 1.56 1 5.15 0.03 0.18 lwd 2.0 0.53 1 1.76 0.2 0.07 region 0.39 0.12 1 0.4 0.54 0.02 elev:grad -0.002 4.33 1 14.33 < 0.01 0.38 org:region 23.24 11.23 1 37.14 < 0.01 0.62 tavsm:lwd -6.33 4.57 1 15.11 < 0.01 0.4 residuals 6.96 23 log(coho biomass) = – 0.1 – 1.4(riffle) – 0.003(elev) + 2.2(grav) – 0.7(org) – 0.1(grad) + 1.4(region I) – 0.002(elev)(grad) + 26(org)(region I) r2 = 0.81 coefficients SS df F p partial r2 riffle -1.39 2.64 1 5.25 0.03 0.16 elev -0.003 1.67 1 3.33 0.08 0.11 grav 2.23 5.9 1 11.74 < 0.01 0.3 org -0.66 0.17 1 0.33 0.57 0.01 grad -0.1 2.02 1 4.02 0.06 0.13 region 1.38 2.15 1 4.28 0.05 0.14 elev:grad -0.002 2.92 1 5.82 0.02 0.18 org:region 26.2 17.22 1 34.3 < 0.01 0.56 residuals 13.56 27 Table 3.3: continued. 50 passed Cook’s Distance, tolerance and normality tests. The full habitat model for YOY density explained a large proportion of the variance in fish abundance (Table 3.3; r2 = 0.75, F 8,39 = 7.1, p < 0.0001). The biomass full habitat model had a poorer fit (r2 = 0.57) and is not considered in detail since it shared most terms with the density model; it is not discussed further here. The interaction terms in the density model suggested that intermediate gradients are associated with the greatest salmonid abundances (gradient and gravel for the density model). Simple slopes analysis shows that the slope of gradient was positive (0.6) when there was a lot of gravel sediment (one standard deviation above average, gravel SD = 0.2) and negative when there was very little gravel (-0.004). The model also suggested that more complex habitats are associated with greater salmonid density (coefficients for pool-forming LWD, largest particles and pool area were all positive [Figure 3.2]). Two interactions indicated regional effects: the pool-elevation interaction suggested that impact of pools was greater in the Interior (coefficient = 4.3) than the Coast (0.8), possibly because pool area was lower in the Interior (Table 3.2); the gravel-elevation interaction showed that the higher gradients (within the range of the data) were more favourable at low elevations in the Coast (elevation SD = 297) and negative at high elevations in the Interior (0.2 and -0.2, respectively). The density full habitat model for rainbow trout included productivity and physical habitat related variables and a regional effect (Table 3.3, Figure 3.3; r2 = 0.83, F 8,19 = 11.5, p < 0.01). The biomass model had the same terms as the density model, with the same signs and similar magnitudes, so only the density model is discussed here. The rainbow trout density model included both alder and total canopy cover; while the alder coefficient was positive, the total canopy cover coefficient was negative (6.3 and -5.2). The coefficient of conductivity was also positive (0.01), in contrast with the YOY model. The model showed that rainbow abundance was greater in streams with more woody debris, but lower in streams with more cover from predators, indicating a mixed response to habitat complexity. The impact of largest particle size was slightly negative in the Interior (-0.01), but positive in the Coast (0.03). The density full habitat model for coho salmon included more physical structure than 51 Figure 3.2: Scatter plots of the key habitat factors included in the YOY density full model (average summer temperature, organic sediment, pool-forming LWD and pool area) against YOY density. YOY density was log transformed for the analyses and is displayed on a log scale. The solid lines indicate the full model regression best fit lines, not the simple regres- sion best fit lines. Dashed and dotted lines indicate interaction effects: in the organic sedi- ment graph the dashed line is the slope of organic sediment when cover is high and the dotted line is the slope when cover is low. The dashed and dotted lines in the pool area graph repre- sent the slopes at high and low elevations, respectively. Filled symbols indicate Coast region sites, the open symbols indicate Interior region sites. 14.0 15.0 16.0 17.0 0. 1 0. 2 0. 5 1. 0 2. 0 summer temperature (ϒC) 0.0 0.1 0.2 0.3 0.4 0.5 0. 1 0. 2 0. 5 1. 0 2. 0 organic sediment 0.00 0.01 0.02 0.03 0.04 0. 1 0. 2 0. 5 1. 0 2. 0 pool−forming LWD m2 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0. 1 0. 2 0. 5 1. 0 2. 0 stream pool area yo un g− of −y ea r s al m on id d en si ty (f ish m 2 ) productivity variables, and both variable types showed regional differences (Table 3.3, Figure 3.4; r2 = 0.91, F 14,23 = 18.4, p < 0.01). The biomass model only included a subset of the terms that were present in the density model so it is not discussed here. The coho models shared many similarities with the YOY models, such as the negative gradient-elevation interaction and conductivity coefficients, and the positive average summer temperature coefficient. The impact of stream structure (pool and riffle area, sediment size) on the density of coho salmon was complex and somewhat contradictory (riffle and pool both had negative coefficients but gravel had a positive coefficient). There was a regional difference in the impact of organic sediment (Figure 3.5): the coefficient of organic sediment was much larger in the Interior (24.3) than the Coast (1.0), and there was also an interaction between average summer temperature and large woody debris. The impact of an increase in temperature in areas with 52 Figure 3.3: Scatter plots of the key habitat factors included in the rainbow density full model (alder cover, canopy cover, conductivity, largest particle size, LWD, and cover) against rain- bow density. Rainbow density was log transformed for the analyses and is displayed on a log scale below. The solid lines indicate the full model slopes, not the simple regression slopes. The dashed and dotted lines in the largest particles graph indicate the slopes in the Coast and Interior regions, respectively. The biomass model distributions are very similar to these and are therefore not presented. Filled symbols indicate Coast region sites, the open symbols indicate Interior region sites. 0.0 0.2 0.4 0.6 0.8 0. 02 0. 05 0. 20 0. 50 canopy cover 0.0 0.2 0.4 0.6 0.8 0. 02 0. 05 0. 20 0. 50 alder canopy cover 50 100 150 200 250 0. 02 0. 05 0. 20 0. 50 conductivity(�S) 20 40 60 80 0. 02 0. 05 0. 20 0. 50 largest particle size (cm) 0.00 0.02 0.04 0.06 0.08 0.10 0.12 0. 02 0. 05 0. 20 0. 50 LWD m2 0.00 0.05 0.10 0.15 0.20 0. 02 0. 05 0. 20 0. 50 proportion of stream with cover yo un g− of −y ea r r ai nb ow d en si ty (f ish m 2 ) abundant woody debris (LWD SD = 0.07) was negative, but was positive in areas with low LWD (-0.38 and 0.53). Comparisons of the habitat subset and habitat + drift subset models for the three species showed that adding drift to the full models did not significantly improve their predictive ability (Table 3.4). Evidence ratios ranged from 3.8 for the YOY biomass model to 3.9 x 1014 for the coho density model. That is, the full habitat model run on the subset of sites with drift data ranged from 3.8 to 3.9 x 1014 times more likely than the habitat + drift subset model with drift added. The r2 values of the habitat + drift subset models were not greatly improved compared to the full-subset models, despite the addition of an extra term (Table 3.4). None of the models developed using only the subset of sites that included drift data (the reduced 53 Figure 3.4: Scatter plots of the key habitat factors included in the coho density full model (organic sediment, gravel sediment, and riffle area) against coho density. On the organic sedi- ment graph the filled symbols indicate Coast region sites, the open symbols indicate Interior region sites. Coho density was log transformed for the analyses and is displayed on a log scale below. The solid lines indicate the full model slopes, not the simple regression slopes. The dashed and dotted lines in the organic sediment graph indicate the slopes in the Coast and Interior regions, respectively. The dashed and dotted lines in the gradient graph indicate the slopes of gradient at high and low elevations. The biomass model distributions are very similar to these and are therefore not presented. Filled symbols indicate Coast region sites, the open symbols indicate Interior region sites. habitat + drift models) included drift as a term in the final model. Thererfore, drift was never retained when it was present as a starting variable. The subset models are not presented here. The cost-benefit analysis revealed that the full habitat models are more efficient (have a higher benefit:cost ratio [r2/minute]) than a typical triple-pass depletion estimate of salmonid density. Typical triple pass depletion electrofishing takes approximately four person-hours (240 minutes) depending on stream size. The benefit:cost ratio for a direct measurement of fish abundance is therefore 1/240 minutes, or 0.004 r2/min, which is lower than the full models and many of the individual variables in this study (Table 3.5). The full habitat model ratios were all near 0.007 (Table 3.5). The total cost in person-minutes was approximately 110 minutes. Conductivity had the greatest individual term ratio at 0.64 for the rainbow density model, and drift the lowest (2.7 x 10-5 for YOY density). The ratio for drift was based on partial r2 values from the habitat + drift subset models, and was low (Table 3.5) because the cost was high for both field and laboratory time. Setting and collecting three drift nets took approximately 45 person-minutes (this time does not include the time that the nets were filtering the water), and laboratory processing of a single sample could take more than 0.0 0.1 0.2 0.3 0.4 0.5 0. 01 0. 05 0. 20 1. 00 organic sediment 0.2 0.4 0.6 0.8 0. 01 0. 05 0. 20 1. 00 gravel sediment 0 1 2 3 4 0. 01 0. 05 0. 20 1. 00 gradient (%)yo un g− of −y ea r c oh o de ns ity (f ish m 2 ) 54 10 hours. The minimum time for laboratory processing was approximately 1 hour for one sample. The minimum cost for one site was therefore 225 minutes (3 samples), much higher than the other individual variables (1-73 minutes). Discussion One of the more robust inferences that could be made from the model results is that juvenile salmonids are generally positively associated with habitat that is more structurally complex; LWD or pool-forming LWD had significantly positive relationships with all species. More complex habitat may support more fish because increased visual isolation reduces predation and aggression (Dolinsek et al. 2007). As the visual isolation of the fish increases, their territory size decreases (less competition for foraging space) and more fish can be supported in a habitat. Several other inferences and conclusions can be drawn from these habitat models. However, although our habitat models explained a high proportion of the variance in both density and biomass of juvenile salmonids, some of the regression coefficients and interactions were difficult to interpret in terms of underlying mechanisms. The negative influence of conductivity in the YOY and coho models was difficult to interpret. Other studies have found positive relationships between conductivity and salmonid abundance (e.g. Coughlan et al. 2007, Scarnecchia and Bergersen 1987). Conductivity was expected to be positively related to salmonid abundance due to the positive impact of conductivity on stream productivity (e.g. Chételat et al. 1999). However, conductivity may also influence the timing of food availability by affecting decomposition rate. Greenwood et al. (2007) found that decomposition rate was positively associated with conductivity, and suggested that in streams with high conductivity detritus decompose earlier in the year, reducing the food supply for detritivorous invertebrates, resulting in lower food availability by the end of the summer growing season. The negative impact of conductivity may also have been due to unseen correlations with other variables. Habitat variables related to stream discharge suggest that the relationships between flow and density differ between combined YOY salmonids, rainbow and coho. The YOY model 55 Model AIC c Δ j lik ω i ω max / ω i r2 YOY fish/m2 full-subset 40.1 0 1 1 1 0.82 full-subset-drift 58 17.9 1.30 x 10-4 1.30 x 10-4 7692 0.82 YOY g/m2 full-subset 6.6 0 1 0.79 1 0.69 full-subset-drift 9.29 2.69 0.26 0.21 3.76 0.74 rainbow fish/m2 full-subset 25.4 0 1 1 1 0.85 full-subset-drift 49.3 23.9 6.49 x 10-6 6.49 x 10-6 1.54 x 105 0.85 rainbow g/m2 full-subset 46.15 0 1 1 1 0.77 full-subset-drift 80.2 34.04 4.05 x 10-8 4.05 x 10-8 2.47 x 107 0.79 coho fish/m2 full-subset 116 0 1 1 1 0.96 full-subset-drift 183 67.21 2.54 x 10-15 2.54 x 10-15 3.94 x 1014 0.97 coho g/m2 full-subset 29.92 0 1 0.81 1 0.68 full-subset-drift 32.86 2.94 0.23 0.19 4.26 0.76 Table 3.4: Likelihood ratio calculations for the full-subset (best model constructed using the full data set, run on the subset of sites with invertebrate drift data) and full-subset-drift (same model with drift added) models for YOY, rainbow and coho. AIC c is the Second-Order Infor- mation Criterion (used for small samples), Δ i is the difference between the AIC c and the mini- mum AIC c , lik is the likelihood of the model given the data (e^(- ½ * Δ i )), ωi is the likelihood of the model divided by the sum of the likelihoods for the competing models (see Burnham and Anderson 2002 ch 2). The ω max / ω i ratio indicates how much more likely the best model (ω max ) is compared to the candidate model (ω i ). 56 Y O Y Y O Y ra in bo w ra in bo w co ho co ho de ns ity bi om as s de ns ity bi om as de ns ity bi om as s C os t (m in ) be ne fi t (p ar tia l or f ul l r 2 ) be ne fi t / c os t (r 2 / m in ) be ne fi t (p ar tia l o r fu ll r 2 ) be ne fi t / c os t (r 2 / m in ) be ne fi t (p ar tia l or f ul l r 2 ) be ne fi t / c os t (r 2 / m in ) be ne fi t (p ar tia l or f ul l r 2 ) be ne fi t / co st ( r2 / m in ) be ne fi t (p ar tia l or f ul l r 2 ) be ne fi t / c os t (r 2 / m in ) be ne fi t (p ar tia l or f ul l r 2 ) be ne fi t / c os t (r 2 / m in ) lp 5 0. 17 0. 03 0. 02 0. 00 4 0. 23 0. 05 0. 3 ca no py 5 0. 61 0. 12 0. 43 0. 09 al de r 5 0. 58 0. 12 0. 55 0. 11 0. 07 0. 01 co nd 1 0. 08 0. 08 0. 24 0. 24 0. 64 0. 64 0. 21 0. 21 0. 24 0. 24 te m p 3 0. 27 0. 09 0. 11 0. 09 0. 03 el ev 3 0 0 0. 17 0. 06 0. 03 0. 01 0. 11 0. 04 gr ad 9 + 64 0. 09 0. 00 1 0. 05 0. 00 1 0. 18 0. 00 2 0. 13 0. 00 2 co ve r 9 + 64 0. 05 0. 00 1 0. 64 0. 01 0. 53 0. 01 or g 9 + 64 0. 26 0. 00 4 0. 18 0. 00 2 0. 01 0 gr av 9 + 64 0. 02 0. 00 0 0. 44 0. 00 6 0. 3 0. 00 4 w b 64 0. 29 0. 00 5 ri ffl e 64 0. 14 0. 00 2 0. 16 0. 00 3 po ol 64 0. 21 0. 00 3 0. 15 0. 00 2 lw d 29 0. 69 0. 23 0. 52 0. 02 0. 07 0. 00 2 lw d_ pf 29 0. 31 0. 01 fu ll m od el 11 1 0. 75 0. 00 7 0. 65 0. 00 6 0. 83 0. 00 8 0. 76 0. 00 7 0. 92 0. 00 8 0. 81 0. 00 8 10 8 10 6 dr if t 22 5 0. 00 6 2. 7x 1 0- 5 0. 14 6. 2x 10 -4 0. 00 7 3. 1x 10 -5 0. 06 2. 7x 10 -4 0. 12 4 5. 5x 10 -4 0. 26 1 .2 x 10 -3 T ab le 3 .5 : C os ts ( pe rs on -m in ut es p er s it e) a nd b en efi ts ( pa rt ia l r 2 ) v al ue s fo r th e te rm s in cl ud ed in th e fu ll d en si ty a nd b io m as s m od el s fo r Y O Y , r ai nb ow a nd c oh o. A bb re vi at io ns a re li st ed in T ab le 3 .1 . T he p ar ti al r 2 v al ue s fo r al l o f th e va ri ab le s ex ce pt f or d ri ft c om e fr om th e fu ll m od el s, w he re as th e pa rt ia l r 2 v al ue s fo r dr if t c om e fr om th e fu ll -s ub se t- dr if t m od el r un o n th e su bs et o f si te s w it h in ve r- te br at e dr if t d at a. 57 includes an interaction effect (gravel-gradient, Table 3.3) which suggests that intermediate water velocities are most beneficial for YOY abundance. The rainbow trout models include only one variable related to flow: largest particle size. The impact of largest particle size was positive in the Coast region but negative in the Interior region, probably due to differences in the habitat preferences of Oncorhynchus mykiss in the Coast and Interior. Steelhead trout, the anadromous variety of rainbow trout, are abundant in the Coast region, where the positive largest particle size slope indicated that rainbow (mostly juvenile steelhead) were associated with larger, more powerful streams with larger particle sizes like cobbles (as in Hill et al. 2006).The effect of discharge on coho density was unclear. Pool area had a significant negative coefficient, contrasting with other studies (e.g. Nickelson et al. 1992, Roni and Quinn 2001). The negative pool coefficient may be an anomaly specific to this data set. Most research has examined coastal coho populations; coho may respond differently to discharge in the Interior than the Coast for reasons unexamined in this study. Further research on Interior coho populations may help determine whether this result was unique to this data set, or if there are regional differences in coho abundance patterns. The models for rainbow and coho showed regional interactions, indicating that habitat effects observed in one region cannot be generalized to others. Sites sampled in the Interior and Coast regions differed in several key attributes – notably, conductivity and drift were higher in the Interior, but alder cover, canopy cover, percent gravel substrate, percent pool, and LWD were higher in the Coast. Salmonid abundance did not differ between regions, but region was involved in significant interactions with largest particle size in the rainbow model, and organic sediment in the coho model. As discussed above, the regional interaction in the rainbow model may be related to a difference in habitat use between steelhead and rainbow juveniles. The regional interaction in the coho model was unexpected and difficult to explain. The effect may be a Type I Error, or may be related to unexplored differences in the food webs between regions (organic sediment is related to invertebrate production). The presence of regional interactions in the models confirms observations by earlier authors (e.g. Fausch et al. 1988) that regional differences affect the predictive ability of habitat models. Invertebrate drift abundance did not significantly improve the fit of the habitat + drift 58 subset models using the Second-Order Information Criterion (AIC c ) for small sample sizes (Burnham and Anderson 2002). However, assessing model improvement following drift inclusion using the standard Akaike Information Criterion indicated that drift significantly improved the YOY and coho biomass models. This suggests that drift may have a positive effect on salmonid biomass, but the effect was neither large nor consistent enough to be significant in our study. Invertebrate drift positively impacts juvenile salmonid growth (Imre et al. 2004, Nielsen 1992, Rosenfeld et al. 2005, Slaney 1972), survival (Imre et al. 2004, Nislow et al. 1998) and density (Imre et al. 2004, Slaney 1972), so it seems unlikely that prey abundance has no real effect on fish abundance. The absence of invertebrate drift from the models could be an artifact of imprecise characterization of prey abundance (drift) at the different sites. We characterized prey abundance using only three replicate invertebrate drift samples, collected on a single day from each site between June and September. Other studies suggest a May peak in drift abundance (Slaney 1972), and that salmonid growth peaks in early spring (Bacon et al. 2005), before our drift sampling started, so we may not have accurately characterized differences in prey availability between sites. Within-site variability of invertebrate drift is also high; other studies suggest collecting four or more samples (e.g. Allan and Russek 1985, Shearer et al. 2002), although drift sampling (Chapter 2) suggests that two replicate samples are sufficient to detect major differences between sites. The final potential sampling issue that could have masked a relationship between drift and fish abundance was that salmonid populations may have been below carrying capacity at some sites due to low spawner abundance or high predator density. However, the models showed strong relationships with between juvenile abundance and habitat variables other than drift, suggesting that there was generally sufficient recruitment for salmonid populations to be habitat limited. Finally, drift may not be sufficiently variable among our sites to emerge as a significant factor constraining salmonid populations. Our initial expectation of a positive relationship between drift and juvenile salmon implicitly assumes bottom-up control of fish abundance, and minimal top-down effects of fish predation. However, it is possible that fish deplete drift to varying degrees depending on site 59 characteristics, thereby obscuring any relationship between drift biomass and fish abundance across sites. A more realistic expectation would be a positive relationship between fish abundance and invertebrate production, rather than standing crop. Depletion of drift biomass (standing crop) through predation may cause drift concentration to poorly represent drift production. Salmonid populations have been shown to deplete stream invertebrate populations (Huryn 1998), so it is likely that some degree of top-down control occurred at some of our sites. Dependence on prey sources other than drift would also weaken the relationship with fish abundance. Although drift is the primary food source of stream salmonids, they can switch to benthic invertebrates when drift is unavailable (Nislow et al. 1998) or if forced through competition (Nielsen 1992). Finally, the range of variation in prey abundance may be small relative to other drivers of habitat quality in our data set, so that it correlates weakly with fish abundance. The final objective of this study was to examine the cost effectiveness of different models and variables for assessing or monitoring habitat quality for juvenile salmonids. Predictive models of habitat capacity in this sense are most useful when stream habitats are not saturated (e.g. because of habitat-independent factors like over-harvesting of adults), and fish density cannot be used as a direct index of habitat quality. For streams that are heavily under-recruited, habitat-based estimates of capacity are more reliable indices of capacity than measured fish abundance (as well as potentially more cost-effective). Deviation of observed fish abundance from that predicted by habitat capacity models is also a useful way of identifying site-specific management issues (e.g. low spawner abundance from over harvest, or poor spawner access). The cost-benefit analysis in this study showed that habitat models with strong predictive ability (r2 = 0.65 to 0.92) can be developed and the associated field data collected after less than two hours of field work, less time than typical triple-pass electrofishing. Other fish collection methods, such as single-pass electrofishing (Jones and Stockwell 1995), require less field time than triple-pass electrofishing and may therefore have higher benefit:cost ratios than the full habitat models presented here. Most of the habitat variables used in this study took small enough amounts of time to measure that they had high benefit:cost (r2/min) ratios. 60 The exception was invertebrate drift, which took more field time and much more laboratory time than any of the other variables (Table 3.5). While the benefit:cost ratios for individual variables calculated in this study are good indicators of their general predictive ability, their exact efficiency (r2/min) could be different when combined with other variables in regression. That is, the partial r2 values of model terms depend in part on what other terms are included in the model. Therefore, while they provide a useful starting point for evaluating the cost- effectiveness of different predictive variables, they should be considered context dependent. This study suggests that physical habitat variables appear to be more robust predictors of fish abundance than direct measures of invertebrate drift. Future research on stream food webs, particularly on the impact of salmonid density on benthic and drifting invertebrate abundance, and on the degree to which drift accurately represents prey availability for drift feeding fish, should improve understanding of the factors that limit the abundance of stream salmonids. Further investigation of juvenile salmonid foraging behaviour (drift versus benthic feeding), especially in relation to the abundance of drift and benthic prey, could lead to more precise capacity models and improve the ability of habitat managers to consider the role of food availability during planning, monitoring, and restoration activities. 61 References Allan, J.D., and Russek, E. 1985. The quantification of stream drift. Canadian Journal of Fisheries and Aquatic Science 42: 210-215. Bacon, P.J., Gurney, W.S.C., Jones, W., McLaren, I.S., and Youngson, A.F. 2005. Seasonal growth patterns of wild juvenile fish: partitioning variation among explanatory variables based on individual growth trajectories of Atlantic salmon (Salmo salar) parr. Journal of Animal Ecology 74: 1-11. Benke, A.C., Huryn, A.D., Smock, L.A., and Wallace, J.B. 1999. Length-mass relationships for freshwater macroinvertebrates in North America with particular reference to the southeastern United States. Journal of the North American Benthological Society 18(3): 308-343. Binns, N.A., and Eiserman, F.M. 1979. Quantification of fluvial trout habitat in Wyoming. Transactions of the American Fisheries Society 108: 215-228. Bowlby, J.N., and Roff, J.C. 1986. Trout biomass and habitat relationships in southern Ontario streams. Transactions of the American Fisheries Society 115: 503-514. Burnham, K.P., and Anderson, D.R. 2002. Model Selection and Multimodel Inference: A practical information-theoretic approach. Springer, New York. Cada, G.F., Loar, J.M., and Cox, D.K. 1987. Food and feeding preferences of rainbow and brown trout in southern Appalachian streams. The American Midland Naturalist 117: 374-385. Chételat, J., Pick, F.R., Morin, A., and Hamilton, P.B. 1999. Periphyton biomass and community composition in rivers of different nutrient status. Canadian Journal of Fisheries and Aquatic Science 56: 560-569. Clark, J.D., Dunn, J.E., and Smith, K.G. 1993. A multivariate model of female black bear habitat use for a geographic information system. Journal of Wildlife Management 57(3): 519-526. Coughlan, S.M.C.J., Connerton, M.J., Ringler, N.H., Stewart, D.J., and Mead, J.V. 2007. Survival and growth responses of juvenile salmonines stocked in eastern Lake Ontario tributaries. Transaction of the American Fisheries Society 136: 56-71. Coulombe-Pontbriand, M., and LaPointe, M. 2004. Landscape controls on boulder-rich, winter habitat availability and their effects on Atlantic salmon (Salmo salar) parr abundance in two fifth-order mountain streams. Canadian Journal of Fisheries and Aquatic Sciences 61(4): 648-658. Davies, P.E. 1989. Relationships between habitat characteristics and population abundance for brown trout, Salmo trutta L., and blackfish, Gadopsis marmoratus Rich., in Tasmanian streams. Australian Journal of Marine and Freshwater Research 40: 341- 359. Denoël, M., and Lehmann, A. 2006. Multi-scale effect of landscape processes and habitat quality on newt abundance: Implications for conservation. Biological Conservation 130: 495-504. 62 Dolinsek, I.J., Biron, P.M., and Grant, J.W.A. 2007. Assessing the effect of visual isolation on the population density of Atlantic salmon (Salmo salar) using GIS. River Research and Applications 23: 763-774. Dorgelöh, W.G. 2006. Habitat suitability for tsessebe Damaliscus lunatus lunatus. Journal of African Ecology 44: 329-336. Fausch, K.D., Hawkes, C.L., and Parsons, M.G. 1988. Models that predict standing crop of stream fish from habitat variables: 1950-1985. Gen Tech Rep, U.S. Department of Agriculture Forest Service Pacific Northwest Research Station, Portland, Oregon. Gavashelishvili, A., McGrady, M.J., and Javakhishvili, Z. 2006. Planning the conservation of the breeding population of cinereous vultures Aegypius monachus in the Republic of Georgia. Oryx 40(1): 1-8. Greenwood, J.L., Rosemond, A.D., Wallace, J.B., Cross, W.F., and Weyers, H.S. 2007. Nutrients stimulate leaf breakdown rates and detritivore biomass: bottom-up effects via heterotrophic pathways. Oecologia 151: 637-649. Hagen, J., Decker, A.S., and Bison, R.G. 2005. Calibrated rapid assessment methods for estimating total steelhead parr abundance in the Thompson River system, British Columbia - 2004 interim report. Report Prepared for BC Ministry of Water, Land, and Air Protection, Kamloops. Hanson, A.E. 2005. Juvenile density survey summaries. British Columbia Conservation Foundation, Surrey, BC. Unpublished Report. Hill, M.S., Zydlewski, G.B., and Gale, W.L. 2006. Comparisons between hatchery and wild steelhead trout (Oncorhynchus mykiss) smolts: physiology and habitat use. Canadian Journal of Fisheries and Aquatic Science 63: 1627-1638. Hogan, D. 1996. Channel Assessment Procedure Field Guide Book. Huryn, A.D. 1998. Ecosystem-level evidence for top-down and bottom-up control of production in a grassland stream system. Oecologia 115: 173-183. Imre, I., Grant, J.W.A., and Keeley, E.R. 2004. The effect of food abundance on territory size and population density of juvenile steelhead trout (Oncorhynchus mykiss). Oecologia 138: 371-378. Johansen, M., Elliott, J.M., and Klemetsen, A. 2005. A comparative study of juvenile salmon density in 20 streams throughout a very large river system in northern Norway. Ecology of Freshwater Fish 14: 96-110. Jowett, I.G. 1992. Models of the abundance of large brown trout in New Zealand rivers. North American Journal of Fisheries Management 12: 417-432. McAlpine, C.A., Rhodes, J.R., Callaghan, J.G., Bowen, M.E., Lunney, D., Mitchell, D.L., Pullar, D.V., and Possingham, H.P. 2006. The importance of forest area and configuration relative to local habitat factors for conserving forest mammals: A case study of koalas in Queensland Australia. Biological Conservation 132: 153-165. McCauley, E. 1984. The estimation of the abundance and biomass of zooplankton in samples. In A manual on methods for the assessment of secondary productivity in freshwaters. 63 Edited by J.A. Downing and F.H. Rigler. Blackwell Scientific Publications, Oxford. pp. 228-265. Meyer, E. 1989. The relationship between body length parameters and dry mass in running water invertebrates. Arch. Hydrobiol. 117(2): 191-203. Moore, K., Jones, K., and Dambacher, J. 1997. Methods for Stream Habitat Surveys: Aquatic Inventory Project. Oregon Department of Fish and Wildlife Natural Production Program, Corvallis, OR, 97330. Morgan, D. 2001. Coho surveys. Unpublished. Nickelson, T.E., Rodgers, J.D., Johnson, S.L., and Solazzi, M.F. 1992. Seasonal changes in habitat use by juvenile coho salmon (Oncorhynchus kisutch) in Oregon coastal streams. Canadian Journal of Fisheries and Aquatic Sciences 49: 783-789. Nielsen, J.L. 1992. Microhabitat-specific foraging behavior, diet, and growth of juvenile coho salmon. Transactions of the American Fisheries Society 121: 617-634. Nislow, K.H., Folt, C., and Seandel, M. 1998. Food and foraging behavior in relation to microhabitat use and survival of age-0 Atlantic salmon. Canadian Journal of Fisheries and Aquatic Sciences 55: 116-127. Quinn, G.P., and Keough, M.J. 2002. Experimental design and data analysis for biologists. Cambridge University Press. Roff, J.C., and Hopcroft, R.R. 1986. High precision microcomputer based measuring system for ecological research. Canadian Journal of Fisheries and Aquatic Sciences 43: 2044- 2048. Roni, P., and Quinn, T.P. 2001. Density and size of juvenile salmonids in response to placement of large woody debris in western Oregon and Washington streams. Canadian Journal of Fisheries and Aquatic Sciences 58: 282-292. Rosenfeld, J., Porter, M., and Parkinson, E. 2000. Habitat factors affecting the abundance and distribution of juvenile cutthroat trout (Oncorhynchus clarki) and coho salmon (Onchorhynchus kisutch). Canadian Journal of Fisheries and Aquatic Sciences 57: 766-774. Rosenfeld, J.S., Leiter, T., Lindner, G., and Rothman, L. 2005. Food abundance and fish density alters habitat selection, growth, and habitat suitability curves for juvenile coho salmon (Oncorhynchus kisutch). Canadian Journal of Fisheries and Aquatic Sciences 62: 1691-1701. Sabo, J.L., Bastow, J.L., and Power Mary, E. 2002. Length-mass relationships for adult aquatic and terrestrial invertebrates in a California watershed. Journal of the North American Benthological Society 21(2): 336-343. Sample, B.E., Cooper, R.J., Greer, R.D., and Whitmore, R.C. 1993. Estimation of insect biomass by length and width. American Midland Naturalist 129(2): 234-240. Scarnecchia, D.L., and Bergersen, E.P. 1987. Trout production and standing crop in Colorado's small streams, as related to environmental features. North American Journal of Fisheries Management 7: 315-330. 64 Schnute, J. 1983. A new approach to estimating populations by the removal method. Canadian Journal of Fisheries and Aquatic Sciences 40: 2153-2169. Sharma, R., and Hilborn, R. 2001. Empirical relationships between watershed characteristics and coho salmon (Oncorhynchus kisutch) smolt abundance in 14 western Washington streams. Canadian Journal of Fisheries and Aquatic Sciences 58: 1453-1463. Shearer, K.A., Hayes, J.W., and Stark, J.D. 2002. Temporal and spatial quantification of aquatic invertebrate drift in the Maruia River, South Island, New Zealand. New Zealand Journal of Marine and Freshwater Research 36: 529-536. Shearer, K.A., Stark, J.D., Hayes, J.W., and Young, R.G. 2003. Relationships between drifting and benthic invertebrates in three New Zealand rivers: implications for drift- feeding fish. New Zealand Journal of Marine and Freshwater Research 37: 809-820. Slaney, P.A. 1972. Effects of prey abundance on distribution, density and territorial behaviour of young rainbow trout in streams, University of British Columbia. Smock, L.A. 1980. Relationships between body size and biomass of aquatic insects. Freshwater Biology 10: 375-383. Tschaplinski, P. 2006. Carnation Creek electrofishing August 2006. Unpublished. Vannote, R.L., Minshall, G.W., Cummins Kenneth, W., Sedell, J.R., and Cushing, C.E. 1980. The river continuum concept. Canadian Journal of Fisheries and Aquatic Sciences 37: 130-137. Wang, T., Hamann, A., Spittlehouse, D., and Aitken, S.N. 2006. Development of scale- free climate data for western Canada for use in resource management. International Journal of Climatology 26(3): 383-397. Waters, T.F. 1962. Diurnal periodicity in the drift of stream invertebrates. Ecology 43(2): 316-320. Wipfli, M. 1997. Terrestrial invertebrates as salmonid prey and nitrogen sources in streams: contrasting old-growth and young-growth riparian forests in southeastern Alaska USA. Canadian Journal of Fisheries and Aquatic Sciences 54: 1259-1269. Zar, J.H. 1999. Biostatistical Analysis. Simon & Schuster, New Jersey. Zharikov, Y., Lank, D.B., Huettmann, F., Bradley, R.W., Parker, N., Yen, P.P.W., McFarlane- Tranquilla, L.A., and Cooke, F. 2006. Habitat selection and breeding success in a forest-nesting Alcid, the marbled murrelet, in two landscapes with different degrees of forest fragmentation. Landscape Ecology 21: 107-120. 65 Conclusion The underlying rationale for this study was to determine 1) the degree to which prey (invertebrate drift) for drift-feeding salmonids varies across time and space, and 2) whether drift abundance can be used as a significant predictor of fish abundance. Temporal variation of invertebrate drift at the day-to-day scale is lower than spatial variation within or between sites (Figure 2.5). Therefore it is unnecessary to resample a site on successive days in order to calculate a precise estimate of invertebrate drift abundance. However, variation within sites (residual variation) was high: up to 90% of the random variation. High within site variation made estimates of drift abundance less accurate, and probably contributed to the low predictive ability of the habitat models for drift in Chapter 2. The reduced precision of these estimates may also have contributed to the rejection of drift from the salmonid habitat models in the salmonid chapter. Therefore multiple samples are necessary to estimate invertebrate drift abundance, but there appears to be no need to repeat that sampling effort on additional visits to the site. Invertebrate drift measurements could be used in habitat models for juvenile salmonids in areas with lower within site variation, or by collecting more than three samples. Future research to characterize the variation of invertebrate drift would allow further refinement of sampling protocols, and a greater understanding of the timing of food availability for juvenile salmonids. It would be useful to determine the relationship between daytime drift samples, samples collected at dusk, and samples collected in the first hours after dark (the first night time peak, Waters 1962), in order to determine how representative daytime samples are of food availability for juvenile salmonids throughout the day and night. Examination of the environmental conditions that influence terrestrial invertebrate input (e.g. wind, rain) would also improve sampling protocols, and could explain some of the day-to- day variation of terrestrial drift (Figure 2.5). Physical and biological habitat variables were both included in the final models predicting the abundance of young-of-the-year coho salmon, rainbow trout, and all salmonids combined, 66 but invertebrate drift was not included. In general, habitats with more structure were better habitat for salmonids, but there were mixed responses to all of the candidate variables. Models for individual species had higher coefficients of variation than the YOY-combined model, suggesting real differences in habitat requirements or preferences between the species. The absence of invertebrate drift from the final models may have been related to sampling problems (see Chapter 3 Discussion), but other considerations, such as complex trophic interactions, are also possible explanations. Further research on complex trophic interactions such as the effect of habitat on foraging behaviour, prey-switching, and thresholds of food abundance that prompt changes in feeding behaviour would help explain how food availability affects salmonid abundance in a natural setting. Much of the previous research on the impacts of invertebrate drift abundance on juvenile salmonids has been conducted in artificial channels (e.g. Imre et al. 2004, Rosenfeld et al. 2005, Slaney 1972). In real stream channels, where benthic food is more abundant and other behaviour-influencing factors such as competition are present, the positive impact of invertebrate drift on salmonid growth, abundance and survival may be overwhelmed by other factors and by behaviour. The predictive models for juvenile salmonids developed in this study are directly applicable to fisheries management projects such as habitat assessment, monitoring, and restoration. The models could be used as presented, or selected variables could be applied based on available resources; the cost:benefit analysis in Chapter 3 can provide fisheries scientists with a good starting point for management decisions related to the power of potential inventory variables. The regional differences in the rainbow and coho models indicate that there is a need for increased regional specificity in habitat assessment protocols – paradigms of fisheries management developed in the Coast region are less applicable in other regions of British Columbia than has previously been assumed. Further research on regional habitat differences and their impacts on salmonid abundance would improve the management of these species. As do all Master of Science theses, this one had weaknesses and strengths. The scale of the 67 project caused weaknesses regarding sample size and data collection: it was not feasible to sort drift samples for all of the sites, or all three of the samples collected on each visit for the temporal study. Had all of the samples been sorted, invertebrate drift may have had a greater impact in the habitat models. Due to the scale of the project, data collected by other people were used extensively; the quality of the data is not being questioned, but there may have been slight differences in electrofishing method that resulted in differences between those data and my own. However, the scale of the project was also its greatest strength, and allowed this project to stand apart from many habitat modeling exercises. In an extensive review, Fausch (1988) found that the majority of habitat models were based on less than 20 sites, as were the majority of models that had coefficients of determination greater than 0.75. In contrast, the salmonid models in this study included up to 50 sites and five of the six coefficients of determination were 0.75 or more. The model with the greatest r2, for coho density, had an r2 = 0.92 and was based on 38 sites. Fausch (1988) also found that many of the reviewed studies did not place enough emphasis on the statistical procedures used to develop models, in particular regarding poor methods for choosing the best models. In this study a great amount of emphasis was placed on researching and applying the most objective and statistically sound model selection method available, resulting in greater confidence that the most appropriate models were chosen. 68 References Fausch, K.D., Hawkes, C.L., and Parsons, M.G. 1988. Models that predict standing crop of stream fish from habitat variables: 1950-1985. Gen Tech Rep, U.S. Department of Agriculture Forest Service Pacific Northwest Research Station, Portland, Oregon. Imre, I., Grant, J.W.A., and Keeley, E.R. 2004. The effect of food abundance on territory size and population density of juvenile steelhead trout (Oncorhynchus mykiss). Oecologia 138: 371-378. Rosenfeld, J.S., Leiter, T., Lindner, G., and Rothman, L. 2005. Food abundance and fish density alters habitat selection, growth, and habitat suitability curves for juvenile coho salmon (Oncorhynchus kisutch). Canadian Journal of Fisheries and Aquatic Sciences 62: 1691-1701. Slaney, P.A. 1972. Effects of prey abundance on distribution, density and territorial behaviour of young rainbow trout in streams, University of British Columbia. Waters, T.F. 1962. Diurnal periodicity in the drift of stream invertebrates. Ecology 43(2): 316-320. 69 Drift Invertebrate Density (mg/m3) site code region total aquatic terrestrial 1 Brohm brom C 71.2 63.9 7.2 2 Carnation carn C 15.8 12.0 3.8 3 Chapman chap C 14.7 8.7 6.0 4 Coho South cohs C 36.1 34.5 1.7 5 Colvin colv C 16.8 16.0 0.9 6 Cook cook C 38.8 32.6 6.2 7 Elk elkk C 91.5 54.6 37.0 8 Finish Early fini C 9.3 9.2 0.1 9 High Falls higa C 1062.0 130.8 931.2 10 Husdon husd C 10.8 6.6 4.2 11 Lost Shoe lost C 51.7 41.5 10.2 12 Rousseau rous C 51.1 33.1 18.1 13 Sandhill sand C 22.8 16.3 6.5 14 Shovelnose shov C 57.2 38.9 18.3 15 Silver Hope silh C 57.3 47.9 9.4 16 Slesse sles C 137.0 134.2 2.9 17 Wakefield wake C 52.2 45.2 7.0 18 Yola yola C 7.9 6.7 1.2 19 Crazy craz I 46.9 39.6 7.4 20 Criss cris I 84.2 34.6 49.7 21 Lemieux Main lemm I 61.2 48.0 13.2 22 Lemieux Side lems I 161.7 147.4 14.3 23 Lindquist lind I 77.9 61.1 16.8 24 Moore moor I 135.2 120.9 14.3 25 Senn senn I 34.8 15.5 19.3 26 Silver B silb I 12.6 9.9 2.6 27 Six Mile sixm I 57.9 52.3 5.6 28 Spius spiu I 283.6 161.9 121.7 29 Vermelin verm I 33.6 31.6 1.9 30 Yard yard I 119.2 114.4 4.9 average 97.1 52.3 44.8 Appendix 1: Habitat measurements from Chapter 2 sites. 70 Appendix 1: continued. UTM coordinates site easting northing datum zone elev (m) 1 Brohm 491286 5517692 WGS84 10 260 2 Carnation 353871 5420162 WGS84 10 20 3 Chapman 447858 5478537 WGS84 10 60 4 Coho South 430330 5498771 WGS84 10 80 5 Colvin 437426 5481030 WGS84 10 20 6 Cook 444195 5481677 WGS84 10 20 7 Elk 585091 5444106 WGS84 10 40 8 Finish Early 307631 5448910 NAD27 10 40 9 High Falls 478552 5531752 WGS84 10 60 10 Husdon 448842 5478484 WGS84 10 80 11 Lost Shoe 311047 5431297 NAD27 10 40 12 Rousseau 349310 5409882 WGS84 10 40 13 Sandhill 311047 5431297 WGS84 10 40 14 Shovelnose 476144 5545906 WGS84 10 160 15 Silver Hope 617821 5454219 WGS84 10 440 16 Slesse 594561 5436690 WGS84 10 240 17 Wakefield 441671 5479843 WGS84 10 20 18 Yola 617879 5453010 WGS84 10 480 19 Crazy 384614 5650320 WGS84 11 400 20 Criss 643436 5639327 WGS84 10 540 21 Lemieux Main693929 5701713 WGS84 10 400 22 Lemieux Side 693929 5701713 WGS84 10 400 23 Lindquist 696491 5688907 WGS84 10 400 24 Moore 681607 5569893 WGS84 10 640 25 Senn 374055 5645513 WGS84 11 380 26 Silver B 332562 5607099 WGS84 11 400 27 Six Mile 314538 5603249 WGS84 11 740 28 Spius 636528 5534836 WGS84 10 860 29 Vermelin 302950 5691650 WGS84 11 660 30 Yard 372188 5640897 WGS84 11 380 average 278 71 Appendix 1: continued. site average summer temperature (ºC) conductivity (mS) largest particle (cm) canopy cover (proportion) alder cover (proportion) 1 Brohm 16.0 17.2 60.8 0.00 0.59 2 Carnation 15.1 40.5 5.3 0.10 0.09 3 Chapman 16.3 20.5 84.0 0.14 0.10 4 Coho South 16.1 63.4 7.4 0.71 0.17 5 Colvin 17.0 116.5 2.6 0.59 0.09 6 Cook 16.6 129.2 4.2 0.81 0.11 7 Elk 17.1 119.7 24.4 0.58 0.08 8 Finish Early 15.3 NA NA 0.88 0.00 9 High Falls 15.9 10.0 45.0 0.19 0.24 10 Husdon 16.2 46.0 2.8 0.57 0.11 11 Lost Shoe 14.1 76.1 10.4 0.30 0.29 12 Rousseau 15.2 42.5 13.0 0.08 0.05 13 Sandhill 14.1 144.7 2.8 0.56 0.38 14 Shovelnose 15.2 28.9 43.1 0.24 0.24 15 Silver Hope 15.2 72.1 37.8 0.03 0.00 16 Slesse 15.2 53.9 78.7 0.04 0.02 17 Wakefield 16.9 92.9 36.8 0.62 0.21 18 Yola 14.7 42.8 85.4 0.03 0.01 19 Crazy 16.0 24.4 21.5 0.02 0.00 20 Criss 16.4 274.0 66.4 0.15 0.01 21 Lemieux Main17.3 201.0 NA 0.10 0.08 22 Lemieux Side 17.3 201.0 NA 0.10 0.08 23 Lindquist 17.1 66.0 24.6 0.13 0.08 24 Moore 15.9 220.0 2.1 0.14 0.12 25 Senn 16.7 54.7 5.0 0.31 0.31 26 Silver B 17.0 210.0 18.0 0.73 0.13 27 Six Mile 15.2 100.5 2.0 0.00 0.00 28 Spius 14.6 163.3 82.3 0.04 0.01 29 Vermelin 15.0 13.0 12.9 0.34 0.15 30 Yard 16.3 31.8 20.3 0.00 0.00 average 15.9 92.3 29.6 0.28 0.12 72 Appendix 1: continued. site bankfull width (m) pool area (proportion) riffle area (proportion) gravel sediment (proportion) organic sediment (proportion) gradient (%) 1 Brohm 5.0 0.1 0.6 0.84 0.00 5.1 2 Carnation 9.1 0.3 0.2 0.10 0.05 1.0 3 Chapman 20.1 0.0 0.3 0.68 0.00 1.3 4 Coho South 2.7 0.2 0.3 0.53 0.12 1.8 5 Colvin 2.8 0.1 0.2 0.54 0.30 0.5 6 Cook 2.4 0.2 0.5 0.14 0.14 0.7 7 Elk 4.2 0.0 0.2 0.00 0.00 3.4 8 Finish Early 1.7 0.7 0.2 0.19 0.00 0.0 9 High Falls 10.5 0.0 0.8 0.40 0.00 1.6 10 Husdon 2.6 0.2 0.1 0.26 0.14 0.4 11 Lost Shoe 7.8 0.3 0.0 0.60 0.37 1.3 12 Rousseau 10.1 0.3 0.5 0.21 0.10 0.8 13 Sandhill 4.5 0.2 0.0 0.30 0.13 0.0 14 Shovelnose 13.6 0.0 0.5 0.09 0.00 2.6 15 Silver Hope 17.4 0.0 0.4 0.20 0.06 1.6 16 Slesse 20.8 0.0 0.6 0.20 0.01 1.8 17 Wakefield 6.3 0.1 0.0 0.08 0.00 1.9 18 Yola 15.9 0.1 0.4 0.45 0.01 2.0 19 Crazy 17.5 0.0 0.3 0.08 0.01 2.3 20 Criss 8.0 0.1 0.6 0.27 0.01 3.3 21 Lemieux Main21.1 0.0 0.5 0.30 0.00 1.5 22 Lemieux Side 6.7 0.1 0.3 0.34 0.36 1.0 23 Lindquist 6.9 0.0 0.4 0.57 0.13 0.9 24 Moore 4.2 0.2 0.0 0.08 0.01 0.5 25 Senn 3.3 0.1 0.2 0.11 0.17 1.7 26 Silver B 5.7 0.0 0.3 0.10 0.04 3.6 27 Six Mile 3.1 0.0 0.0 0.03 0.50 1.0 28 Spius 16.7 0.0 0.0 0.46 0.00 1.6 29 Vermelin 6.5 0.2 0.2 0.09 0.04 2.5 30 Yard 33.1 0.0 0.6 9.03 0.01 2.2 average 9.7 0.1 0.3 0.58 0.1 1.7 73 Appendix 2: habitat measurements from Chapter 3 sites. Abbreviations follow Table 3.1. site region lp (cm) canopy (proportion) alder (proportion) 1 angu C 30.5 0.60 0.07 2 bole I 35.3 0.17 0.15 3 brom C 60.8 0.63 0.59 4 carn C 5.3 0.10 0.09 5 chap C 84.0 0.14 0.10 6 cohn C 2.5 0.83 0.05 7 cohs C 7.4 0.71 0.17 8 cold I 64.0 0.02 0.01 9 colv C 2.6 0.59 0.09 10 cook C 4.2 0.81 0.11 11 craz I 21.5 0.02 0.00 12 cris I 66.4 0.15 0.01 13 dead I 53.2 0.11 0.00 14 depo C 17.6 0.28 0.28 15 elkk C 24.4 0.58 0.08 16 fini C 29.0 0.88 0.88 17 fred C 4.6 0.48 0.43 18 guic I 23.0 0.28 0.00 19 hifa C 25.8 0.45 0.21 20 hunt C 27.4 0.40 0.32 21 husd C 2.8 0.57 0.11 22 kooa C 5.7 0.38 0.15 23 koob C 0.6 0.49 0.00 24 kooc C 1.0 0.38 0.00 25 lemm I NA 0.10 0.08 26 lems I NA NA NA 27 lind I 24.6 0.13 0.08 28 lost C 10.4 0.30 0.29 29 maka I 73.2 0.03 0.00 30 mash C 76.5 0.54 0.54 31 moor I 2.1 0.14 0.12 32 pach C 5.6 0.55 0.37 33 rous C 13.0 0.08 0.05 34 sand C 2.8 0.56 0.38 35 sari C 10.4 0.13 0.02 36 senn I 5.0 0.31 0.31 37 shov C 43.1 0.24 0.24 38 sila I 76.9 0.44 0.21 39 silb I 18.0 0.73 0.13 40 silh C 37.8 0.03 0.00 41 sixm I 2.0 0.00 0.00 42 sles C 78.7 0.04 0.02 43 spiu I 82.3 0.04 0.01 44 stag C 1.2 0.52 0.25 45 verm I 12.9 0.34 0.15 46 wake C 36.8 0.62 0.21 47 weee C 1.8 0.88 0.88 48 weym I 55.0 0.50 0.40 49 yard I 20.3 0.00 0.00 50 yola C 85.4 0.03 0.01 average 28.7 0.35 0.18 74 Appendix 2: continued site pool (proportion) riffle (proportion) tavsm (oC) cond (uS) 1 angu 0.35 0.65 16.2 55.4 2 bole 0.04 0.24 15.4 127.8 3 brom 0.08 0.63 16.0 17.2 4 carn 0.26 0.20 15.1 40.5 5 chap 0.04 0.26 16.3 20.5 6 cohn 0.17 0.52 16.1 53.6 7 cohs 0.23 0.28 16.1 63.4 8 cold 0.04 0.79 14.5 98.5 9 colv 0.05 0.22 17.0 116.5 10 cook 0.15 0.46 16.6 129.2 11 craz 0.00 0.30 16.0 24.4 12 cris 0.06 0.59 16.4 274.0 13 dead 0.14 0.16 15.6 234.0 14 depo 0.13 0.58 15.2 29.9 15 elkk 0.00 0.25 17.1 119.7 16 fini 0.70 0.19 15.3 NA 17 fred 0.19 0.56 15.4 50.9 18 guic 0.01 0.00 16.2 55.2 19 hifa 0.01 0.75 15.9 10.0 20 hunt 0.20 0.55 15.4 90.3 21 husd 0.23 0.13 16.2 46.0 22 kooa 0.36 0.03 14.0 141.0 23 koob 0.24 0.01 14.0 96.3 24 kooc 0.00 0.00 14.0 34.9 25 lemm 0.00 0.54 17.3 201.0 26 lems 0.10 0.26 17.3 201.0 27 lind 0.02 0.37 17.1 66.0 28 lost 0.25 0.03 14.1 76.1 29 maka 0.00 0.70 14.9 138.6 30 mash 0.05 0.00 15.9 47.7 31 moor 0.18 0.00 15.9 220.0 32 pach 0.20 0.49 14.9 34.8 33 rous 0.32 0.46 15.2 42.5 34 sand 0.21 0.00 14.1 144.7 35 sari 0.46 0.22 15.6 47.6 36 senn 0.12 0.23 16.7 54.7 37 shov 0.01 0.53 15.2 28.9 38 sila 0.00 0.00 14.6 1.5 39 silb 0.01 0.34 17.0 210.0 40 silh 0.02 0.41 15.2 72.1 41 sixm 0.00 0.00 15.2 100.5 42 sles 0.01 0.60 15.2 53.9 43 spiu 0.00 0.00 14.6 163.3 44 stag 0.70 0.17 14.0 111.9 45 verm 0.17 0.24 15.0 13.0 46 wake 0.13 0.00 16.9 92.9 47 weee 0.25 0.20 16.3 73.6 48 weym 0.45 0.06 15.2 85.7 49 yard 0.01 0.64 16.3 31.8 50 yola 0.09 0.43 14.7 42.8 average 0.15 0.31 15.6 87.5 75 Appendix 2: continued site elev (m) wb (m) cover (proportion) grav (proportion) 1 angu 20 8.59 0.16 0.37 2 bole 720 9.79 0.07 0.25 3 brom 260 4.99 0.12 0.06 4 carn 20 9.06 0.06 0.84 5 chap 60 20.07 0.06 0.10 6 cohn 40 1.48 0.21 0.32 7 cohs 80 2.73 0.14 0.68 8 cold 880 16.06 0.07 0.28 9 colv 20 2.89 0.15 0.52 10 cook 20 2.41 0.10 0.54 11 craz 400 17.55 0.02 0.45 12 cris 540 8.01 0.15 0.08 13 dead 660 11.42 0.08 0.25 14 depo 640 11.85 0.08 0.28 15 elkk 40 4.18 0.03 0.14 16 fini 40 1.74 0.08 0.31 17 fred 40 9.75 0.07 0.61 18 guic 600 8.20 0.11 0.12 19 hifa 60 10.55 0.04 0.19 20 hunt 20 8.11 0.11 0.24 21 husd 80 2.57 0.21 0.40 22 kooa 20 3.52 0.20 0.59 23 koob 20 3.20 0.12 0.07 24 kooc 20 2.37 0.23 0.17 25 lemm 400 21.10 0.02 0.27 26 lems 400 6.70 0.03 0.30 27 lind 400 6.87 0.06 0.34 28 lost 40 7.79 0.09 0.26 29 maka 820 9.39 0.04 0.34 30 mash 80 8.15 0.14 0.19 31 moor 640 4.21 0.18 0.57 32 pach 40 4.68 0.08 0.81 33 rous 40 10.14 0.07 0.60 34 sand 40 4.48 0.20 0.21 35 sari 20 15.37 0.08 0.71 36 senn 380 3.27 0.12 0.08 37 shov 160 13.56 0.05 0.30 38 sila 860 6.85 0.09 0.13 39 silb 400 5.66 0.08 0.11 40 silh 440 17.37 0.03 0.09 41 sixm 740 3.10 0.08 0.10 42 sles 240 20.77 0.05 0.20 43 spiu 860 16.66 0.05 0.03 44 stag 20 2.22 0.10 0.09 45 verm 660 6.52 0.14 0.46 46 wake 20 6.34 0.10 0.20 47 weee 40 1.30 0.10 0.31 48 weym 780 4.18 0.10 0.09 49 yard 380 33.11 0.18 0.09 50 yola 480 15.93 0.15 0.08 average 294 8.74 0.10 0.30 76 Appendix 2: continued site org (proportion) grad (%) lwd (pieces/m) lwd_pf(pieces/m) 1 angu 0.06 1.6 0.03 0.00 2 bole 0.04 2.1 0.01 0.00 3 brom 0.00 5.1 0.00 0.00 4 carn 0.05 1.0 0.04 0.01 5 chap 0.00 1.3 0.01 0.00 6 cohn 0.18 1.5 0.23 0.03 7 cohs 0.12 1.8 0.14 0.02 8 cold 0.00 1.4 0.01 0.00 9 colv 0.31 0.5 0.11 0.00 10 cook 0.14 0.7 0.14 0.02 11 craz 0.01 2.3 0.00 0.00 12 cris 0.01 3.3 0.04 0.04 13 dead 0.02 1.6 0.01 0.00 14 depo 0.05 0.9 0.08 0.00 15 elkk 0.00 3.4 0.02 0.00 16 fini 0.05 8.0 0.26 0.00 17 fred 0.23 0.7 0.02 0.00 18 guic 0.01 2.4 0.00 0.00 19 hifa 0.00 1.6 0.02 0.00 20 hunt 0.00 2.7 0.02 0.00 21 husd 0.14 0.4 0.24 0.03 22 kooa 0.20 0.8 0.20 0.04 23 koob 0.28 0.1 0.09 0.00 24 kooc 0.40 0.7 0.18 0.00 25 lemm 0.00 1.5 0.00 0.00 26 lems 0.36 1.0 0.01 0.00 27 lind 0.13 0.9 0.03 0.00 28 lost 0.37 1.3 0.03 0.00 29 maka 0.00 1.9 0.00 0.00 30 mash 0.00 1.3 0.02 0.00 31 moor 0.01 0.5 0.00 0.00 32 pach 0.04 1.2 0.03 0.01 33 rous 0.10 0.8 0.01 0.00 34 sand 0.13 0.0 0.13 0.01 35 sari 0.01 0.5 0.01 0.00 36 senn 0.17 1.7 0.05 0.02 37 shov 0.00 2.7 0.02 0.00 38 sila 0.01 4.0 0.02 0.00 39 silb 0.04 3.6 0.04 0.00 40 silh 0.06 1.6 0.01 0.00 41 sixm 0.50 1.0 0.00 0.00 42 sles 0.01 1.8 0.00 0.00 43 spiu 0.00 1.6 0.00 0.00 44 stag 0.49 1.2 0.11 0.00 45 verm 0.06 2.5 0.10 0.01 46 wake 0.00 1.9 0.03 0.00 47 weee 0.16 3.6 0.19 0.04 48 weym 0.00 4.1 0.00 0.00 49 yard 0.01 2.2 0.02 0.00 50 yola 0.01 2.0 0.01 0.00 average 0.10 1.8 0.06 0.01 77 Appendix 2: continued site shock_area (m2) YOY (#) rainbow (#) coho (#) 1 angu 185.4 178 NA 109 2 bole 96.1 117 74 5 3 brom 123.5 44 22 NA 4 carn 322.0 418 NA 414 5 chap 85.1 89 46 42 6 cohn 31.1 93 NA 29 7 cohs 28.8 108 NA 19 8 cold 320.3 139 87 11 9 colv 81.2 142 NA 28 10 cook 29.8 61 NA 3 11 craz 126.5 87 57 24 12 cris 118.6 90 80 1 13 dead 179.0 80 79 NA 14 depo 77.2 54 NA 35 15 elkk 74.8 123 NA 69 16 fini 35.6 20 NA NA 17 fred 207.3 130 29 95 18 guic 129.2 50 2 8 19 hifa 110.9 45 44 NA 20 hunt 188.1 86 32 54 21 husd 84.4 71 NA 23 22 kooa 78.7 194 NA 156 23 koob 90.5 71 NA 10 24 kooc 59.9 132 NA 125 25 lemm 48.3 22 20 2 26 lems 109.2 211 72 187 27 lind 88.9 298 NA 295 28 lost 371.3 208 NA 157 29 maka 99.1 18 16 1 30 mash 78.4 93 60 24 31 moor 108.0 23 23 NA 32 pach 135.2 90 17 67 33 rous 213.3 178 74 102 34 sand 192.1 49 NA 35 35 sari 304.5 91 9 78 36 senn 37.6 142 62 69 37 shov 150.4 290 227 48 38 sila 71.0 21 5 NA 39 silb 89.2 41 24 NA 40 silh 79.4 49 37 NA 41 sixm 60.0 28 22 NA 42 sles 167.0 232 169 48 43 spiu 135.0 94 73 2 44 stag 41.8 61 NA 4 45 verm 26.5 36 24 4 46 wake 110.2 105 NA 88 47 weee 29.5 180 NA 43 48 weym 70.7 55 44 NA 49 yard 168.6 28 9 4 50 yola 188.5 123 76 NA average 120.8 108 52 65 78 Appendix 2: continued. site YOY/m2 rainbow/m2 coho/m2 drift (dry mg/m3) 1 angu 0.96 NA 0.59 NA 2 bole 1.22 0.77 0.05 NA 3 brom 0.36 0.18 NA 71.16 4 carn 1.3 NA 1.29 15.80 5 chap 1.05 0.54 0.49 14.72 6 cohn 2.99 NA 0.93 NA 7 cohs 3.75 NA 0.66 36.14 8 cold 0.43 0.27 0.03 NA 9 colv 1.75 NA 0.34 16.83 10 cook 2.05 NA 0.1 38.80 11 craz 0.69 0.45 0.19 46.94 12 cris 0.76 0.67 0.01 84.21 13 dead 0.45 0.44 NA NA 14 depo 0.7 NA 0.45 NA 15 elkk 1.65 NA 0.92 91.55 16 fini 0.56 NA NA 9.26 17 fred 0.63 0.14 0.46 NA 18 guic 0.39 0.02 0.06 NA 19 hifa 0.41 0.4 NA 1062.02 20 hunt 0.46 0.17 0.29 NA 21 husd 0.84 NA 0.27 10.79 22 kooa 2.47 NA 1.98 NA 23 koob 0.78 NA 0.11 NA 24 kooc 2.2 NA 2.09 NA 25 lemm 0.46 0.41 0.04 61.24 26 lems 1.93 0.66 1.71 161.69 27 lind 3.35 NA 3.32 77.95 28 lost 0.56 NA 0.42 51.70 29 maka 0.18 0.16 0.01 NA 30 mash 1.19 0.77 0.31 NA 31 moor 0.21 0.21 NA 135.20 32 pach 0.67 0.13 0.5 NA 33 rous 0.83 0.35 0.48 51.14 34 sand 0.26 NA 0.18 22.79 35 sari 0.3 0.03 0.26 NA 36 senn 3.77 1.65 1.83 34.82 37 shov 1.93 1.51 0.32 57.16 38 sila 0.3 0.07 NA NA 39 silb 0.46 0.27 NA 12.55 40 silh 0.62 0.47 NA 57.29 41 sixm 0.47 0.37 NA 57.88 42 sles 1.39 1.01 0.29 137.04 43 spiu 0.7 0.54 0.01 283.62 44 stag 1.46 NA 0.1 NA 45 verm 1.36 0.9 0.15 33.59 46 wake 0.95 NA 0.8 52.20 47 weee 6.1 NA 1.46 NA 48 weym 0.78 0.62 NA NA 49 yard 0.17 0.05 0.02 119.20 50 yola 0.65 0.4 NA 7.88 average 1.20 0.47 0.60 97.11 79 Appendix 3: UBC Research Ethics Board Certificate of Approval                                          "@en ; edm:hasType "Thesis/Dissertation"@en ; vivo:dateIssued "2008-05"@en ; edm:isShownAt "10.14288/1.0066197"@en ; dcterms:language "eng"@en ; ns0:degreeDiscipline "Zoology"@en ; edm:provider "Vancouver : University of British Columbia Library"@en ; dcterms:publisher "University of British Columbia"@en ; dcterms:rights "Attribution-NonCommercial-NoDerivatives 4.0 International"@en ; ns0:rightsURI "http://creativecommons.org/licenses/by-nc-nd/4.0/"@en ; ns0:scholarLevel "Graduate"@en ; dcterms:title "Influence of physical and biological habitat variables on juvenile salmonid and invertebrate drift abundance in southwest British Columbia streams"@en ; dcterms:type "Text"@en ; ns0:identifierURI "http://hdl.handle.net/2429/256"@en .