9 Implications of temporal and spatial scale Atlantic salmon (Sarmo sa/ar) research Carol L. Folt, Keith H. Nislow, and Mary E. Power for Abstract: The Atlantic salmon (Sulmo sulur) is a model species for studying scale issues (i.e., the extent, duration, and resolution of a study or natural process) in ecology. Major shifts in behavior and habitat use over ontogeny, along with a relatively long life span and large dispersal and migration distances, make scale issues critical for effective conservation, management, and restoration of this species. The scale over which a process occurs must be linked to the research design and we illustrate this with a discussion of resource tracking by Atlantic salmon. Identifying scale inconsistencies (e.g.. when a process is evident at one scale but not another) is shown to be an effective means by which some scale-dependent processes are understood. We review the literature to assess the temporal and spatial scales used in Atlantic salmon research and find most current studies appear to sacrifice spatial and temporal extent for increased resolution. Finally, we' discuss research strategies for expanding the temporal and spatial scales in salmon research, such as conducting multiple scales studies to elucidate scale inconsistencies, identifying mechanisms, and using techniques and approaches to generalize across studies and over time and space. RCsumC : Le saumon de I'atlantique (Sulmo sulur) est une esptce modtle pour l'ttude des probltmes d'kchelle (c'est- A-dire ]'&endue, la dur6e et la rksolution d'une Ctude ou d'un processus naturel) en Ccologie. Les changements majeurs dans le comportement et I'utilisation de I'habitat durant la vie du saumon atlantique, ainsi que la IongtvitC relativement grande et l'importance de la dispersion et de la migration de ce poisson, rendent critiques les probltmes d'6chelle pour la conservation, la gestion et le rttablissement de cette esp&ce. L`tchelle sur laquelle se dCroule un processus doit like au protocole de recherche, et nous examinons cette question en rapport avec la recherche de ressources par le saumon atlantique. On montre que le rep6rage du manque de cohtrence entre les tchelles (p.ex., quand un processus est tvident A une tchelle donnte mais pas A une autre) est un moyen pour comprendre les processus dCpendants de I'tchelle. Nous avons examint la documentation scientifique pour tvaluer les Ccheiles temporelles et spatiales utilisks dans la recherche sur le saumon atlantique et avons trouvt que la plupart des ttudes actuelles semblent Ctendues spatiales et temporelles pour obtenir une plus grande rksolution. Enfin, nous traitons des strategies de recherche permettant d'ttendre les tchelles temporelles et spatiales dans la recherche sur le saumon; on propose notamment d'effectuer des ttudes A Cchelles multiples pour tlucider les incohtrences d'tchelle, de reptrer les mkcanismes, et d'utiliser des techniques et des permettant de faire des gtntralisations A partir des diverses ttudes et dans le temps et l'espace. [Traduit par la Rtdaction] Introduction Ecologists increasingly recognize the importance of scale (Le., the extent, duration, and resolution of studies and natu- ral processes) to many issues, including salmonid ecology, conservation, and restoration. Because different biotic and abiotic processes act at different spatial and temporal scales, the conclusions of research studies often are influenced by the scale of investigation (Allen and Hoekstra 1992; Levin 1992; Fahrig 1992; Ray and Hastings 1996). Interest in the effect of scale on diverse topics can be seen from the num- ber of recent reviews addressing scale in aquatic habitats. These reviews have considered the importance of scale to questions such as the design and interpretation of research (Frost et al. 1988), fish habitat conservation and restoration I Received March 11, 1998. Accepted September 20, 1998. 1 514485 C.L. Folt and K.H. Nislow. Department of Biological Sciences, Dartmouth College, Hanover, NH 03755, U.S.A. M.E. Power. Department of Integrative Biology, University I of California at Berkeley, Berkeley. CA 94720. U.S.A. Can. J. Fish. Aquat. Sci. 55(Suppl. 1): 9-21 (1998) programs (Lewis et al. 1996), and the effect of environmen- tal constraints on species distributions and abundances (Poff 1997) The Atlantic salmon (Sulmo sulur) is a good model spe- cies for evaluating the importance of scale because pro- cesses affecting its growth, reproduction, survival, and evolution span a wide range of temporal and spatial scales. Atlantic salmon undergo major ontogenetic shifts in behav- ior and habitat use (Mills 1991). are relatively long-lived (2- 10 years), occupy a broad geographical range (circum North-Atlantic from southern New England to the Iberian peninsula) and disperse and migrate over extremely long distances (up to lOOOs of kilometres). As a result of these processes, a number of important questions about Atlantic salmon cannot be adequately ad- dressed without taking temporal or spatial scale into consid- eration. For example, effects of logging on salmonid habitat clearly vary over different spatial and temporal scales (e.g., are scale dependent). Consider the temporal and spatial scales of impact resulting from clear-cutting the forest along a 500 m stream reach. Effects of increased light penetration on instream temperature and primary productivity due to 0 1998 NRC Canada 10 Can. J. Fish. Aquat. Sci. Vol. 55(Suppl. l), 1998 canopy opening are likely to be evident at relatively limited temporal (in the year following logging) and spatial (-500 m) scales (Hicks et al. 1991). In contrast, effects of the clear-cut on stream sediment load, channel morphology, and large woody debris loading are likely to extend beyond the stream reach where logging has occurred (larger spatial scale) and not be fully manifest until years following the logging event (larger temporal scale). Moreover, effects at the large scale can also feed back over longer time periods to influence small-scale food web processes through their in- fluences on life cycles and interactions among species (Power et al. 1996; Wootton et al. 1996). Despite its recog- nized importance, few studies of Atlantic salmon have spe- cifically included more than one scale in their design. To understand fundamental processes underlying Atlantic salmon ecology requires linking physical and biological pro- cesses at appropriate temporal and spatial scales. In this re- view, we address scale-dependent processes that can affect the dynamics of salmon and other fish species. This paper is an overview, using a few key examples to illustrate how scale can affect the interpretation of pattern and process. We review the literature to assess the scales of investigation most frequently used in Atlantic salmon research and dis- cuss how scale can affect research conclusions. Finally, we emphasize the benefits of combining observations with ex- perimental and manipulative approaches for elucidating mechanisms and identifying appropriate scales in the study of Atlantic salmon. What is scale? "Scale" is the spatial and temporal dimension of a process or an entity (Lewis et al. 1996). For example, habitat selec- tion by salmon is a process whereby individuals make cer- tain "decisions" based on information they have integrated over particular distances and times. The size of a study area and the duration of a study are attributes of an etitify (the re- search study) whose characteristics are in part defined by these dimensions (e.g., diversity may increase with plot size up to a certain point; more behaviors are observed as obser- vation period increases). Allen and Hoekstra (1992) define three dimensions of scale: (1) spatial extent (the size of a process or entity), (2) temporal extent (the duration of a process or entity), and (3) grain (the finest level of spatial and temporal resolution, often determined by the frequency and density of samples or observations, or by levels of data aggregation). For example, Elliott's (1994) long-term study of brown trout population dynamics at Black Brow's Beck is of small spatial extent (conducted in a single stream reach of approximately 60 m2), large temporal extent (stretching more than 20 years, encompassing multiple generations), and fine-grained (the population was sampled multiple times each year). Extent and grain tend to be related in study design. Fine- grained (high-resolution) studies generally are performed over limited spatial and temporal scales. The terms "small- scale" tend to refer to processes or entities that are small in extent and (or) relatively fine-grained, and "large-scale" for processes or entities that are large in extent and (or) rela- tively coarse-grained. Given finite research resources, in- creased resolution (fine grain) is generally achieved by reducing the spatial or temporal extent of the study, or vice versa. We illustrate these aspects of scale with three hypothetical studies of salmon summer growth rates and show that the questions that can be adequately addressed differ among the designs due to differences in extent and grain. First, in a sin- gle stream, growth is measured only once at the end of the season, for 10 consecutive years (small spatial extent, large temporal extent, coarse-grained). This study could be used to examine the influence of between-year variation in cli- mate and discharge on growth of a single population. Sec- ond, also in a single stream, growth is measured at 10 intervals over a single growing season (small spatial extent, small temporal extent, fine-grained). This study could be used to examine the importance of short-term critical growth periods, but only for one population (among basin compari- sons could not be made). Third, in 10 different streams, growth is measured once at the end of a single growing sea- son (large spatial extent, small temporal extent, coarse- grained). This study could be used to address effects of stream-scale differences in geomorphology and water chem- istry on annual first-year growth, but not within season pat- terns of growth within a site or the importance of yearly variation. Extending the extent or grain can reveal poten- tially important processes that operate at different scales. Review of the Atlantic salmon literature reveals cases where information gathered at a particular scale is broadly applied without considering implications of the scale of the original sampling design (i.e., when inferences drawn from a single site are assumed to apply over broader spatial or tem- poral extent). To avoid this problem, careful attention must be given to the scale over which a process occurs and is measured and over which it influences a specific aspect of salmon biology. Fahrig (1992) states this succinctly. "To un- derstand the relationship of an organism to its environment, one must understand the interactions between the intrinsic scales of heterogeneity within the environment and the scales at which the organism can respond to this heterogene- ity." In Table 1, we list three environmental factors (current speed, temperature, and prey abundance) that are known to have a strong influence on springkummer juvenile salmonid growth and survival and are highly variable in streams (Bjornn and Reiser 1991). These processes are driven by dif- ferent factors at different spatial scales (Table 1). For exam- ple, stream flow at small spatial scales (within in pool or riffle) is strongly affected by local substratum characteris- tics. Over the entire basin, stream flow (i.e., discharge) is in- fluenced by the geology, slope, regional land use, and precipitation. Similarly, the effects of temperature and prey abundance on salmon growth patterns can vary at different scales. Tem- perature, for example, may vary only slightly at a small spa- tial scale and hence produce no variation in growth rate among fish within a pool, riffle, or stream reach. Neverthe- less, temperature variation across basins may be great and drive large among-basin variation in salmon growth, physi- ology, and production (Elliott 1985; Holtby 1988; Nicieza et al. 1994; Filbert and Hawkins 1995). In seeking to under- stand growth patterns small-scale studies might obscure the overriding influence of temperature, even if temperature across a region was the single most important explanatory 9 0 1998 NRC Canada 11 Folt et at. -..- and small-scale channel geometry (widthldepth) Channel slope, sinuosity, Ground w~ltcr sources Larger sc;ile hahit;it unit\ (i-il`llc> Shaiiing cl`l'ects on priii1;u.y - 100 in macrohabitat transitions (riflle runs) to pool) production I 00 m - km Land use, impoundment Overhead cover. shading, Downstrc;ini dispersal imliotti~dincnt Temperature el`l`ccts on invcrtclir;itc production Allochtonous inputs km - 100 km Basin size, parent geology Basin-wide inlluences on Basin chemistry and geology runoll and temperature >IO0 kni Regional precipitation and Reg ional c I i mite Regional species pool runoff variable. Prey abundance also may vary less over small spa- tial scales than large scales. As for temperature, it could be difficult to isolate the influence of food abundance on dnion production even from numerous small-scale or high- resolution studies. Important processes also have influence at different tcm- poral scales (Table 2). For example, size related shifts in habitat and diet of Atlantic salmon may be explained in part by seasonal shifts in prey phenology and stream discharge that occur over periods of weeks to months (Cada et nl. 1987; Keeley and Grant 1997). Over the short term, we may expect the most productive reaches to be those with the most food and the best discharge conditions. Yet, over the long term, overall patterns of fish or prey abundance may be driven by infrequent extreme events (e.g., niassive dis- charges or dry periods) that are regulated by climate trends operating over years to decades and beyond. These long- term processes may result in low population numbers per- sisting in reaches that in any given year (or in a short-term study) appear to provide highly suitable habitat. In the next section, we illustrate the importance of scale by examining the relationship between stream dwelling salmon and their food resources. We then discuss inferences that can be made by looking at this question over different spatial and temporal scales and with different resolution. Resource tracking and the importance of scale It is common in fisheries science to predict fish produc- tion from the quantity of resources (e.g., food, physical habi- tat type) in an area. The term "resource tracking" has been used to describe situations where there is a significant posi- tive correlation between the abundance or performance of consumers and the abundance or productivity of their resources (Hart and Fonseca 1995). If resource tracking oc- curs, the resource levels can be used for goals such ;IS identi- lying habitats likely to yield high fish production. evaluating the efficacy of habitat remediation, and understanding how fish production will respond to variation in resource abun- dance. Much effort has been directed towards testing whether salnion abundance (density, standing stock biomass. produc- tion) and performance (growth, survival, reproduction) track the abundance of resources such as prey (Mnson 1976; Cada et al. 1987; Hinch 1991; Richardson 1993; Filbert and Hawkins 1995) or habitat (Fausch et al. 1988; Hicks et 91. 1991: Fausch et al. 1994). These studies have been con- ducted over a range of spatial and temporal scales and illus- trate how scale affects the examination of an important ecological process. We use studies from various systems rel- evant to and including Atlantic salmon to show that (I) dif- ferent types of resource tracking require different measurements, (2) resource tracking may be measurable over one spatial scale but not another (we term this "scale inconsistency"), and (3) understanding the mechanism un- derlying the resource tracking relationship or scale inconsis- tency helps identify the appropriate measurement scale. Types of resource tracking Fish production tracks resources in two, non-mutually ex- clusive ways. Understanding their differences is necessary to interpret patterns and to predict the extent and resolution necessary to detect resource tracking in nature. We term the first type, "numerical resource tracking" (NRT), which oc- curs when more individuals are found at the highest resource productivity. NRT arises from an increase in recruitment or by immigration into high resource areas. We term the second 0 1998 NRC Canada 12 Can. J. Fish. Aquat. Sci. Vol. 55(Suppl. l), 1998 Table 2. Temporal scale of important environmental processes and iliiplic;itions fijr Atl;inlic s;iliiion (lioiii review by I!iornn ;ind Reiscr 1991; Hicks et at. 1991). Scale Environiiientul patterns and processes Elfcclh on sillnlon Hour-weeks Diel cycles in oxygcii tempernture Diel cycles in prey avnilahilily (i.e., drift Ilypoxii;i for alevins ;ind incuh;ltilig eggs 1)ic.l 1i.ediiig ;111d iictiviiy cycles of fry :IMI prr periodicity temperature changes Small-scale (aseasonal) strean1 discharge and Weeks-year Prey phenology Size-rel;ited hahitat and diet shifts affects feeding and growth Senson;il 1iabit:it and hehnvior:il shifts and sltiiitiisrlwiiitsr Miglalion cucslpliysiological ch;inges for p;wr-smolts Redd de-watering ;ind scouring (eggshlcvins) Oceanic prey :ivail:ihility ;itid tenipcr;iturcs ;il`fccting growth Variation in precipitation and streanillow associ;ited with for fry and parr Seasonality in temperature and discharge at high latitudes hnhitat shifts for parr Years-decades Between-year climatic variability (ENSO, oceanic processes) ;itid surviv;il of ;idlilts ENSO. ocean circulation events (it11 freshw;ttcr life st;ip) Changes in thc timing. ni;ignitudc ;ind varihility of cxtreiiic Changes in stream habitat niorphology (LWD ;tv;iil;thility and >Decades Climate change. adaptation and changes in migra- tion patterns, fishing pressure. land use change events dyn;miics) scditiicnt/substrarum chiiractcristics. cliannel characteristics type, "performance resource tracking" (PRT), which occurs when individual performance (but not necessarily density) increases with resource productivity. When both NRT and PRT occur more individuals with greater individual perfor- mance are observed under high resources. If populations are dense, so that resources have become limiting, numerical resource tracking can preclude perfor- mance tracking (PRT) and vice versa. For example, under the "ideal-free distribution" niodel (IFD, sensu Fretwell and Lucas 1970), individuals fill habitats sequentially, filling the best habitats first, and occupying the second best habitat when crowding lowers the quality of the first habitat to that of the second. The outcome of this ideal free distribution is that abundance tracks habitat quality, but individual perfor- mance is similar across habitats. Algivorous catfishes in tropical streams provide an example of the IFD model in situ (Oksanen et al. 1995). Non-territorial catfish in a Pana- manian river showed evidence of NRT because there were more catfish in stream pools with higher algal productivity. However, there was no evidence for PRT, in that fish from high and low productivity pools exhibited similar individual growth and survival rates (Power 1984). Distinguishing between NRT and PRT has management implications. First, different measurements over different spatial and temporal scales are required to detect each pro- cess. For example, testing for a correlation between abun- dance and habitat characteristics, probably the most basic and often used metric in salmon habitat studies (Bovee 1986; Fausch et al. 1988), will not yield evidence of re- source tracking if only PRT occurs. Similarly, measures of individual performance do not correlate with resource abun- dance under an IFD/NRT-only situation. In either case, it could be erroneously assumed that a particular resource was not limiting. Second, the management of a particulnr system may depend on whether the performance (bigger or more fe- cund fish) or the abundance (more individuals) is the out- come to be maximized. Understanding the mechanistic link between resources and their specific effects on growth or abundance could be effective for improving assessment and management. Surprisingly, there is not a clear answer in the literature as to what happens to salmon production when food resources are increased. Do you get more ~1ni11l fish (NRT), few big fish (PRT), more big fish (NRT and PRT), or no change at all? There are plausible models to explain any of these out- comes, but few direct tests. The relationship between food resources and NRT or PRT for Atlantic salmon can be diffi- cult to detect and therefore easy to ignore for several rea- sons. First, food resource abundance and productivity can be extremely hard to quantify (Resh and Rosenberg 1979). Sec- ond, salmon may respond to an increase in resources in ways that result in either NRT or PRT. For example, territory size of individual salmon may decrease with an increase in prey flux. NRT could then arise by packing in more territo- ries at high resources. If, as in the IFD model, territory size declines in direct proportion to an increase in food density, consumption rates remain equal across territories, and there is no PRT. In contrast, salmon territory size may not de- crease with food resource density, if, for example, it has a fixed relationship with body size. Then more territories will not "pack in" at high resource levels, and NRT should not be observed. If consumption rates still increase with increasing food resources, there will be PRT without NRT. The literature includes studies that partially suppo,-t and reject NRT and PRT for salmonid systems. Elliott (1994) measured an increase in individual growth rates with an 0 1998 NRC Canada Folt et al. 13 increase in food resources, providing strong evidence for PRT by brown trout in Black Brow's Beck. His study also provided evidence against ill1 IFD/NRT situation, because territory size did not decline with ;iii increase in resources. 111 contrast, both Dill et al. Cjuvenile coho, 198 I ) and Keeley ;ind Gr;int (Atlantic salmon, 1995) found mild support for territory size compression i1t higher food, and thus indirect support for NRT and aspects of the IF11 niodel. However, the effect of food on territory size was very small, explain- ing -2% of the variance in territory size for Atlantic salmon parr (2.90-14.50 cm fork length; Keeley and Grant 1995). Instead. body size was a much stronger determinant of terri- tory size. Direct tests of both NRT and PRT are needed to assess their relative importance as mechanisms linking food and production. Increasing consistency in the methods for assessing food and territory size will expand our general un- derstanding of this issue. Scale inconsistencies Consideration of the extent and resolution of each study is critically important for interpreting the relationship between salmonids and their resources. If either salmon abundance or performance tracks resources at one scale but not another. studies at different scales will arrive at different answers (i,e., there will be scale inconsistencies). Scale inconsistcn- cies present another layer of complexity in testing for re- source tracking. Mason ( 1976) provides a good example where identifying scale inconsistencies can reveal important processes. In a stream fertilization experiment run over ;I pe- riod of several months (small temporal extent). juvenile coho salmon survival and growth strongly increased with in- creased prey productivity (NRT and PRT). Yet, when 101- lowed for several years (greater teniporal extent), all evidence of NRT or PRT was absent, because they were ap- parently offset by heavy winter niortality. The cvidence hr NRT and PRT depended on the temporal scitle or extent of the study, and examination of both temporal scnles was ncc- essary to fully assess the relative importance of food, perfor- mance, and a third key variable, winter survival. Additional examples from both riverine and marine envi- ronments also show that fish track resources at some scales, but not others. Rose and Leggett (I 990) found that the corre- lation between the abundance of Atlantic cod and its major prey, capelin, in the North Atlantic (NRT) differed strongly over different spatial scales. At the largest spatial scale niea- sured (4-10 km), predator abundance was strongly posi- tively correlated with prey abundance, and both species were concentrated in large-scale oceanic fronts. However, at smaller spatial scales (2-3.5 kni) prey and predator abun- dance were either negatively correlated or uncorrelated. They interpreted this scale inconsistency as evidence for ei- ther small-scale predator avoidance by capelin or small-scale prey depletion by cod. The importance of scale inconsisten- cies, in this and other examples, is often that they provide the strongest tool for assessing the scale over which differ- ent processes operate. In riverine habitats, Fausch et al. (1994) also found scale inconsistencies in the relationship between abundance of two species of char (genus Safvcfirius) with physical and bi- otic factors on Hokkaido Island, Japan. Char abundance cor- related positively with certain physical characteristics at large and intermedi;ite spatial scales (e.g.. w;ttcrsheds and trihutaries within watersheds), hut was uncorrel;ttctl with these features at smaller spatial scales (e.6.. within tributar- ics or stre;im reaches). In contrast, the presence of the con- generic species did not inllucnce species' presence or ahundance at larger spatial scales, but correlated with ;ititin- dance at small spalid scnles. By coinparing processes at scv- era1 spatial scales. the scale inconsistencies could bc identified. and the niechanisins underlying tlieni could be elucidated. Elucidating mechanisms to resolve scale inconsistencies We and others (Hart and Fonseca 1995; Lewis et at. 1096; Power et al. 1998) recommend ;I two-part approach to iden- tify ;tiid then resolve scale inconsistencies. This approach combines taking explicit nieasureiiieiits or observations at several scales itnd using experiments or models to test nlcch- ;inisins underlying the patterns and to probe inconsistencies that ;ire revealed. Moreover, ;IS Power et al. (1998) argue. if the spatial scale addressed increases (e.g., from micro- habitats to watersheds), carefully designed field nwsure- nients and direct observations beconic even more iniportant because large-scale manipulations are difficult to carry out and virtunlly impossible to replicate. They suggest that nest- ing experiments or models within observational field studies is most effective for clarilying mechanisnis and extrapolat- ing :icross systems or scales. We illustrate the strength ol' this coinbined approach for studying resource tracking with Power's study of algivorous catfish in large tropical rivers (Power 1084; Oksanen et al. 1995; see also Power et al. 1998 for additional examples). Resource tracking was explicitly compared at two spatial scnles. Power found NRT at the relatively large spatial scale (between river pools), but no evidence for NRT at the smaller scale (within pools). Catfish were more abundant in deep microhabitats Farther I'rom the riverbank, while algal standing crops were higher in shallow, nearshore sites. The mechanism underlying the inconsistency was elucidated in predator exclusion experiments. Catfish were found to avoid nearshore habitats to reduce predation risk from wading birds, who could not forage effectively in the deeper, off- shore water (Power et al. 1989). In contrast, catfish resource tracking was not constrained by predation at the between- pool scale because they spent relatively little time in shallow intervening riffles when moving between pools. Our second example involves resource tracking by age-0 Atlantic salmon. In a study of prey availability and foraging by Atlantic salmon parr in several New England streams, Nislow et al. (1998) assessed PRT at two spatial scales. They compared salmon foraging rates (by watching fish via snorkeling) as a function of potential food availability (I) among salmon occupying different feeding locations within a reach (small scale) and (2) among salmon living in different streams. Within each stream, territory-specific drift rates were predicted as the mean free stream velocity in each territory times the mean drift concentration within the reach. Between streams, mean foraging rates tracked food re- sources (PRT). Foraging rates were significantly reduced by experimental reduction of the mean density of drifting prey 0 1998 NRC Canada 14 Can. J. Fish. Aqiiat. Sci. Vol. 55(Suppl. 1). 1998 to tlie reach and were higher in the high-food stream. In con- trast, at the sniall scale (comparing aniong iish within ;I reach). there was no PRT (i.e., there weru not higher forag- ing rates in high-current-speed. high-food locations (territo- ries) within each stream). These results indicate a scale inconsistency in resource tracking by age-0 salmon. One explanation for this scale inconsistency derives from understanding the niechanisnis relating current speed, food ilux, capture success, and microhabitat selection by stream salmonids. Among microhabitats within a stream reach, in- vertebrate drift rates tend to be strongly positively correlated with microhabitat current speed (Waters 1972; Bjornn and Chapman 1968; Hill and Grossman 1993; Nislow et nl. 1998). However, high current speeds also dccrcase tlie abil- ity of fish to detect and capture dr ng prey. Because of this tradeoff, consumption rates itre predicted to be niaximized ;it intermediate current speeds (and thus at interniediate prey flux rates), and juvenile salmonids arc predicted to preferen- tially occupy those niicrohabitats where consumption is pre- dicted to be maximized (Hughes and Dill 1990; Hill and Grossman 1993). Some stream salmonids, including Atlantic salmon, appear to select habitats in accordance with tradeotfs pre- dicted by this model (Hughes and Dill 1990; Hill mid Grossman 1993; Nislow et al. 1999). Under these model conditions, it is unlikely that performance (as for;iging rate) tracks food resource abundance at the within-reach scale (Fig. I). for two reasons. First, optimal current velocities (i.e., the velocities predicted to produce within +10%1 of the highest feeding rates based on tradeol'ls between capture success and prey flux with current speed) can span ;I fairly wide range (i.e.. roughly the same feeding rates are predicted to result over a range of current velocities). Sec- ond, because capture success declines with current speed and prey drift rate increases with current speed, high prey drift rate is not the same as high resource availability. There- fore, food abundance alone is not a proper currency for its- sessing food availability, and feeding will not necessarily track food abundance. Like the catfishes studied by Power (1984). salmon do not track their food on some scales, due to the positive correlation between resource abundance and factors that impede resource use (predation risk in the case of the catfishes; reduced capture success at high current speed in the case of salmonids). In contrast, differences in resource abundance between salmon rearing streams are correlated with differences in performance (Nislow et al. 1998). In fact. any constraint at the within-reach scale that reduces within-reach variation in foraging may make it easier to detect differences between high and low resource reaches, and reinforce PRT at the larger spatial scale. This example therefore illustrates both how understanding mechanisms can help resolve "scale in- consistencies," and how processes that occur on small scales can influence processes at larger scales of observation. Lewis et al. (1996) provide other examples where multi- scale investigations were effective for determining the ap- propriate scale for restoration and management practices such as liming to increase pH (Lacroix 1996; Keller and Gunn 1995 in Lewis et al. 1996) or habitat management (Crowder et al. 1996). energy intake (--) current speed current speed capture success (- . 7) energy intake J prey flux (-) current speed Over which scales are most studies conducted? The need to consider temporal and spatial scale to address important issues for Atlantic salmon ecology is becoming widely appreciated. This volume includes a number of pa- pers that deal explicitly with scale issues at the physiologi- cal, population and community level, and in freshwater and oceanic phases. Major shifts in behavior and habitat use Over ontogeny, along with a relatively long life span and large dispersal and migration distances, make scale issues critical for effective conservation, management, and restoration of 0 1998 NRC Canada 1 I 3 Folt et al. Atlantic salmon. Nevertheless, relatively few studies have been conducted at more than one spatial or temporal scale, resolution varies across studies, and experimental field stud- ies to tease apart n~echanisms are still uncomnlon. TO determine the spatial and temporal scales over which most Atlantic salmon studies have been conducted, we ran an on-line computer search. We considered two components of scale: (I) extent - the number of years or streams in a study and (2) grain - the number of samples/year or sites/stream in a study. We used the on-line search engine (Ovid: Gateway) to search the major life sciences database (BIOSIS for the Life Sciences) for all citation abstracts con- taining the key words "Atlantic salmon" and "ecology" from 1990 through February 1997. Of these citations, we included only papers which assessed, using field manipulation, corre- lations or comparisons, factors affecting habitat selection, life-history strategies, growth, and survival of freshwater- phase anadromous Atlantic salmon. Laboratory and aqua- culture studies were excluded. Studies were classified by both spatial and temporal ex- tent and resolution. We distinguished four levels of temporal extent, equivalent to study duration: SI year, 2-5 years. 6- IO years, and >IO years. We distinguished two levels of tem- poral grain or resolution, equivalent to the within-year sam- pling frequency: one sampling period a year (coarse grain) or multiple sampling periods a year (fine grain). For spatial extent, we distinguished four categories: studies which took place within 1 stream or river, 2-5 rivers, 6-10 rivers, and greater than 10 rivers. For spatial grain, we classified studies on the basis of whether they sampled one site/river (coarse grain) or multiple siteshiver (fine grain). It was more difficult to classify the spatial scale of each study than the temporal scale. Depending on the design and objective of the study, studies conducted in multiple tributar- ies of the same river system can be considered as multiple rivers in our classification system or as multiple sites within a single drainage system. For example, if multiple tributaries of a single river were sampled, we classified this study as being conducted in more than one tributary or stream (i.e., increased spatial extent) if the aim of the study was to com- pare some factor among streams (grain depends on whether there were replicate sites sampled). However, if the study was conducted in multiple tributaries of a single river, and the object was to use the average of these sites to obtain a average for that single drainage system, then we classified the study as characterizing a single river system, but with finer grain (e.g., more replicates). TO illustrate our classification scheme, consider McMenemy's (1995) study of the effect of stocking density on juvenile Atlantic salmon survival in the West River, Ver- mont, U.S.A., over 7 years. Sampling sites were distributed among different tributaries of the West River, and were cen- sused once each year. The study tested the effects of varia- tion in stocking density across years averaged across tributaries and did not address between-tributary differences. Therefore, we considered the. spatial extent of this study to be a one river system (the West River), and the spatial grain to be multiple sitedriver (fine grain). The temporal extent is 6-10 years, and the temporal grain is one sample/year (coarse grain). 15 Fig. 2. Distrihution of Atlantic salinon studies ( 1990-1997) uncovered by literature search over geogr;iphic locations (Fig. 20) and general resenrcli topics (Fig. 21)). Additional derails 011 literature search given in text. N = 5 I studies. stme studies were pliiccd in more tlian one resciircli topic c;itrgory. O FREQUENCY a) 40 30 20 10 0 i4 5 O PERCENT 401 30 20 IO 0 We identified 51 field studies that fulfilled our selection criteria. Of these we were able to determine spatial extent of 50 studies, temporal extent of 41 studies, temporal grain of 32 studies, and spatial grain of 33 studies. These studies 0 1998 NRC Canada 16 60 - 50 - 40 - 30 - 20 - 10 - 0- Fig. 3. Frequency and percentage of Atlantic salmon studies conducted at 5 categories of temporal (Fig. 3n) and spatial (Fig. 30) extent. Details of the literature search and classification criteria given in text. FREQUENCY liQ PERCENT 40 30 20 10 0 0 0 I 8 r. v4 In A r. In 0 Ti h h YEARS 0 I r. 0 r. r. In rl A a A x Can. J. Fish. Aquat. Sci. Vol. 55(Suppl. Fig. 4. Frequency and percentage of Atlantic saln~on studies conducted at 5 categories of temporal (Fig. 4tr) and spatial (Fig. 40) resolution. Details of the literature seiirch and classification criteria given in text. a) FREQUENCY PERCENT single multiple SAMPLES/YEAR b) 100 75 50 25 0 single multiple SITES / STREAM STREAMS Implications of scale bias to Atlantic salmon research encompassed most of the geographic range of the Atlantic salmon in both North America and Europe (with the excep- tion of the Baltic region) and a wide range of ecological top- ics (Fig. 2). The majority of these studies of Atlantic salmon appear to trade increased resolution for reduced spatial and temporal extent. More than 75% of studies were conducted within a single river system, with fewer than 20% lasting for more than 5 years (Fig. 3). However, most studies were relatively fine grained. Greater than 85% of studies included samples taken from multiple locations within a river or single drainage system, and 44% of the studies included samples taken multiple times within the course of a single year (Fig. 4). The results of our search indicate that scale-bias is com- mon in Atlantic salmon research. Most studies are limited in their temporal extent and rarely span the life cycle of an in- dividual, or even the change of seasons and its associated change in physical and biological constraints. Choosing the appropriate time scale for each study requires understanding the nature of both the physical and biological features of the system. As pointed out by Frost et al. (1988) the danger lies in selecting a time frame that is either too short or too long. Inferences must be carefully evaluated \vithin a specific [em- poral framework, which makes comparisons among studies even more challenging. Limitations on temporal extent (study duration) are probably linksd in pan to limited resources and human-imposed temporal limitations (the D 1998 NRC Canada Folt et al 17 average duration of a graduate thesis or a funding period). However, the paucity of long-term investigation weakens in- ferences we can make about processes spanning longer peri- ods and thus our ability to predict long-term trends under changing environmental and human pressures. There is also a bias in the spatial scale of most Atlantic salmon field investigations. Most studies have concentrated on measuring particular systems (e.g., single river systems, or even single tributaries or reaches within a single tribu- tary) using a multiple locations per river (high-resolution) approach. This emphasis probably reflects the management orientation of a great deal of Atlantic salmon research, which is often directed towards river-specific, applied prob- lems (i.e., is a particular river good or bad for salmon, and why?), and the idiosyncratic nature of research (salmon biol- ogists are placed somewhat haphazardly across the globe, studying the systems nearest to their home base). These in- tensive studies have been extremely effective at elucidating individual behaviors, genetics, physiology, and many small- scale ecological processes (Schlosser 1991). Yet they are less effective for comparing across systems where different biotic and abiotic factors prevail. There are great differences in resolution or grain across studies. Methods for censusing populations differ greatly, and whether individuals, reaches, or entire streams are being characterized influences the appropriate spatial scale of the study. As pointed out by Grant et al. (1998), Folt and Schulze ( 1993), and others, small-scale patchiness, which is generally not characterized in sampling programs, may be fundamental to understanding outcome of important density- dependent interactions. This could be particularly important for understanding interactions among space holding territo- rial species like Atlantic salmon juveniles, where the precise numbers and sizes and nature of neighbors or food can have a strong impact on individual foraging and overall perfor- mance. Sampling programs that miss small-scale patterns may overlook the fundamental importance of density-dependent processes in salmon population regulation. Ray and Hastings (1996) demonstrate that identifying density-dependent rela- tionships may be especially biased by the spatial sampling design. They argue that the lack of density-dependence ob- served in many current studies may be an artifact of sarn- pling programs that average over regions of widely varying population density and obscure key relationships. Further, with a study of scale-related dynamics in 79 insect popula- tions, they found that short-time series (low temporal extent) or low test power (fewer samples) were far less important for detecting density-dependence than identifying the appro- priate spatial scale. The effect of differing spatial resolution can be illustrated by comparing predictions from two types of bioenergetic models. These models are used to examine the relationship between resources and performance. They predict fish growth rate potential as a function of both biotic and abiotic aspects of the environment, (e.g., prey density, physical hab- itat structure) and intrinsic characteristics of individual fish (Hewett and Johnson 1992; Ney 1993). Although they use the same input parameters and the same equations relating environment to growth, the two types of models operate at dit`ferent spatial scales. "Simple" bioenergetic models are based on average conditions across a like. river, or (at slightly higher resolution) a reach, while "spatially explicit" bioenergetic models (Brandt et al. 1992) require higher reso- lution information because they incorporate the actual spa- tial distribution of fish, their prey, and physical conditions that affect foraging and growth. Spatially explicit and site- averaged models can produce very different predictions. In a revealing example, Brandt and Kirsch (1993) found that for striped bass (Morone saratilis) in Chesapeake Bay, site- averaged growth rate potential was consistently greater than spatially explicit predictions. Their explanation was that prey spatial distribution was highly patchy, and that prey densities in high density patches often exceeded densities that produce the maximum consumption rates of predators. In the spatially explicit version, this important behavioral constraint is modeled. However in the site-averaged model, behavioral restrictions on consumption are not included. In this example, a behavioral mechanism (density dependence in foraging rate) sets a ceiling on consumption, so that reso- lution of small-scale patchiness becomes necessary to under- stand the predator-prey dynamics. Because of limited support for research, increasing sam- pling grain to the resolution necessary to follow individuals and their neighbors will restrict either the number of sites in- vestigated or the number of times the samples are taken. When populations are especially patchy in space or time, it may be necessary to increase resolution no matter what the "cost." Yet, increased resolution is certainly not always re- quired and in some cases may actually hide other large- scale processes (Frost et al. 1988). Obviously, future studies should be designed with the understanding that the resolu- tion has a strong influence on the inferences that are drawn. A review of new methods for identifying the appropriate scale for specific systems and questions is well beyond the scope of this paper, but methods for matching variance in bi- ological and physical processes are of growing importance to the design of research programs (Horne and Schneider 1994). Strategies for expanding scale As discussed above, a powerful technique for identifying the importance of scale is to conduct multiple scale studies, identify scale inconsistencies, and to use a combination of experiments or models to test the underlying mechanisms in order to extrapolate across systems. However, there are also other ways to confront scale issues that are being applied in a number of different research areas. We now address three possible methods to expand scale and test for generality across scales. Scaling up High-resolution, low-extent studies are valuable for cer- tain questions. But site and time specificity comes with some costs as it limits the detection of larger scale determi- nants of potentially important processes. There have been a number of calls recently to "scale up," emphasizing the im- portance of comparing ecological processes for stream fish across larger spatial or temporal scales (Lewis et aI. 1996; 0 1998 NRC Canada 18 Cunjak 1996). Concordance in the change in fish abundance among streams in Colorado led Gowan and Fausch (I 996) to emphasize the importance of regional (large-scale) processes to fish dynamics. However, comparisons across systems or regions are problematic for a number of reasons. nuking it clear that scaling up is not simple. Difficulties of scaling up are particularly apparent when increasing extent is associated with decreasing resolution (coarser grain). For example. salmonid researchers have been debating whether population density in a stream reach is determined by seasonal "self-thinning" processes in rela- tion to some maximum carrying capacity of the environment (Grant and Kramer 1990; Grant 1993; Grant et al. 1998) or, instead, is determined by mortality during a critical period for survival early in ontogeny (Elliott 1989. 1994). Arm- strong (1997) used statistical techniques to look at a single dataset (Elliott 1994) with two degrees of temporal resolu- tion - multiple measurements over a year (fine resolution) and single estimates for each year in the dataset (coarse res- olution). Armstrong found that the coarse resolution analy- ses could not isolate an effect of a critical period and suggested the importance of self-thinning processes to den- sity. In contrast, the finer-resolution approach indicated that self-thinning was largely a by-product of early season mor- tality, which supported the critical period theory for explain- ing density. Large-scale studies (e.g., watersheds) are particularly vul- nerable to problems in addition to loss of fine scale resolu- tion. Because of logistical considerations, comparable replicates are limited; they may be non-existent in the case of paired watershed studies (Power et al. 1998). At such large scales, it is not at all uncommon for unpredictable events (fires, floods) with high-magnitude impacts to affect either the "control" or the comparison watershed only, largely invalidating the study. These studies also tend to measure highly integrated "output" variables such as stream flow or total production, which are affected by a host of fac- tors, making it difficult to understand the mechanisms by which changes occur (Lewis et al. 1996). Perhaps as a con- sequence of these limitations, Hicks et al. (1991) found that while watershed scale variables have been associated with salmonid abundance, no two studies have identified the same set of variables as being important, severely limiting the generality of research findings. Despite these problems, large-scale comparisons are es- sential for certain questions. For example, recent large-scale studies have revealed important relationships between land use and ecological variables (see references in Schlosser 1991). An approach to conservation of Pacific salmon spe- cies has been strongly advocated that requires management of salmon at least at the basin-wide level (Allendorf et al. 1997). For basin-wide management to be effective requires an expansion from single or few sites to sites across the range of habitats and range of conditions within the basin (see Kocik and Ferreri 1998). Another powerful tool for resolving scale issues is to "scale down" using experiments and models to explore the mechanisms underlying scale inconsistencies and patterns. A number of recent papers have dealt with the appropriate use of experiments and modeling in ecology (Frost et al. 1988; Can. J. Fish. Aquat. Sci. Vol. 55(Suppl. 1). 1998 Schlosser 1991; Power et ill. 1995, 1098; Osenberg et al. 1998: and others (this issue)) and emphasized their strengths and weaknesses (especially with respect to scale). Lahora- tory experiments have been widely used to identify ;I nuni- ber of behavioral and physiological processes for Atlantic salmon, but field experinients are still relatively uncoiiiiiion (but see Kennedy and Strange 1086; Whalen and LeBar 1994; Nislow et al. 1998). To effectively test for mecha- nisms driving inconsistencies or similarities at different scales we should sample at several scales. scaling up where possible and necessary, and conibine these nie;isurenients with carefully designed experiments or models nested within the sampling program (see reviews by Mather 1998; Fausch 1998; Grant et al. 1998; and others (this issue) for more de- tails on experiments with salmonids). Synthesis across studies One of the primary goals of all papers in this issue was to include a review of the literature in order to synthesize among studies and thereby to expand the scale of inference. Most reviews seek to expand the extent and resolution that can be obtained by comparing numerous intensive, and usu- ally small-scale individual studies. There is a growing move- ment within the field of ecology to do this in a quantitative fashion (Gurevitch et al. 1992; Gurevitch and Hedges 1993; Fernandez-Duque and Valeggia 1993; Arnqvist and Wooster 1995). Different techniques are being proposed, debated. and refined in the literature, including nieta-analysis and applica- tion of Bayesian statistics (Ellison 1996; Dennis 1996; Osenberg et al. 1998). These techniques appear promising for synthesizing among disparate studies of Atlantic salmon and are likely to result in some fundamental advances in the field when appropriately applied to topics currently under debate. Meta-analysis has been used by investigators in many dif- ferent systems to analyze and synthesize a number of inde- pendent studies. A key goal of meta-analysis is to extract a quantitative metric (e.g., effect size) from each study and then to apply statistical techniques to the comparison of this metric across studies (Osenberg et al. 1998). Recent meta- analyses include comparisons of the effects of selective log- ging on density of birds (Fernandez-Duque and Valeggia 1993), of effects of benthic versus drift feeding predators on the density of stream invertebrates (Dah1 and Greenberg 1996), and strength of fish-snail interactions across lakes (Osenberg et al. 1998). A number of issues identified in this volume would be suitable for meta-analysis. One difficulty in conducting a meta-analysis arises when the investigations are run on different time or spatial scales, as shown in Osenberg et al. (1998). However, meta-analysis also can be used to test for scale inconsistencies among studies. This in- formation can then be coupled with other approaches (field sampling, experimental manipulation, or modeling). Following individuals over space and time Atlantic salmon travel vast distances over time and space, which makes it logistically difficult to follow any individual over its lifetime. Okubo (1980) distinguishes two viewpoints or approaches for investigating moving populations. The first is the LaGrangian approach, which follows a population 0 1998 NRC Canada Folt et al. 19 or an individual through space. The second is the Eulerian approach, which is to watch ilidividuids or populations flow past Movements by Atlantic salmon impose logistic con- straints on following individuals, even within a single habi- tat during a single life history stage (Arnistrong et al. 1993). These constraints severely restrict the use of the LaGrangian technique. In studies of Atlantic salmon, it has been more common to take a Eulerian approach and to observe a num- her of individuals in an "instant" in time (e.g., by censusing all individuals in a reach or counting smolt moving past a counting station, etc.). Even when reaches are sampled re- peatedly, individuals are rarely distinguished. The Eulerian approach is appropriate for many questions, and carries with it the need to niake many strategic decisions about extent and resolution in the sampling design. because the temporal and spatial scale of the study is essentially im- posed by the researcher (as discussed previously). Adopting a LaGrangian perspective would enhance our ability to approach a number of questions in population and commu- nity ecology that cannot be addressed by Eulerian methods, in large part because LaGrangian approaches allow the organism to define the relevant scale. For example, long- term studies by Connell et al. (1997) on coral reef coni- munities employ an essentially LaGrangian approach. He has followed individuals for more than 25 years, making it possible to examine the cumulative effects of species inter- actions and physical processes on growth and survival over the lifetime of an individual. This approach has been very successful in understanding the ecology of these sessile communities. In situations where it is difficult or impossible to mark and follow individuals, modeling dispersal processes can al- low researchers to simulate a LaGrangian approach for in- vestigating habitat and population dynamics. For example, Bozek and Rahel (1991) found a scale inconsistency in habi- tat resource tracking in juvenile cutthroat trout. Within stream reaches, individuals preferred slow, deep micro- habitats, but shallow reaches with abundant spawning gravel had the highest trout abundances. The low abundance or to- tal absence of juvenile trout in stream reaches that seemed to contain suitable microhabitats appeared to result from spawning gravel limitation, explaining the scale inconsis- tency. Reaction-diffusion, or diffusion/dispersal models (Turchin and Thoeny 1993), which model distance- dependent dispersal success from a point source (such as a salmonid redd or a rearing tributary mouth) may therefore allow integration across spatial scales and improve our abil- ity to predict salmonid abundance and survival in the con- text of these critical demographic events. Even when mobile individuals can be marked and fol- lowed, the intensity of effort per individual restricts sample sizes (limits resolution) and generally forces inferences to be drawn from a few individuals in a few circumstances. Yet new techniques are being applied to the study of Atlantic salmon that will allow a more LaGrangian approach to be applied to larger samples of fish. For example, microsatellite genetic markers have been already developed to mark large numbers of individuals (Letcher and King 1997). Identity (from parentage) of an individual can be determined from a fixed point or through a fixed area. single fin clip, so individuals Cilti be sampled repeatedly dur- ing their lives. This technique will eventually be applied to muny questions (e.g.. How far do individuals move during the stream phase'? Which individuals (based on parentage or stocking location) are the successfully returning adults`?) Fingerprinting salmon via stable isotopes of strontium and nitrogen has also been shown to be ef-fective for following movenients of groups of individuals over time (Kennedy et al. 1997; Harrington et al. 1998). Using organism structures that record the history of individuals, for example, growth rings on fish otoliths (Wright et al. 1990) is a third technique that can be used to follow an individual over time. Perhaps the most powerful developments will arise from combina- tions of these new techniques, which iiiay allow large nuni- bers of salmon to be followed from birth to death and to address questions not possible with other methods. Summary and conclusions The "great scale problem," as stated by Fahrig (1992) is, "what is the appropriate scale for study of a particular prob- lem?' This is not a simple question for Atlantic salmon, be- cause as we have seen, the mechanisms linking salmon and their environment vary over long and short tinies and over large and small spatial scales. We suggest that identifying scale inconsistencies via multiple scale observations and measurements is the first step for identifying the scales over which processes affecting distribution and abundance oper- ate. Strategic use of experiments and models and expanded quantitative analysis of different studies using meta-analysis are likely to greatly enhance our understanding. Finally, ap- plication and development of new technologies that allow new approaches to old questions are certain to reveal novel insights that will contribute to the effective conservation and management of the Atlantic salmon worldwide. Acknowledgments We thank the participants in the workshop held in Braemar, Scotland, in March 1997, entitled, "Integrating across scales: predicting patterns of change in Atlantic salmon" for interactions and discussion of these issues. M. Mather and two anonymous reviewers also provided helpful comments on the manuscript. We are very grateful for the support and vision of the Northeastern Forest Experi- ment Station of the USDA Forest Service, especially Rich- ard DeGraaf, Robert Lewis, and Keith Jensen. This work was supported by grants from the US Fish & Wildlife Ser- vice National Biological Service, the U.S. Forest Service Northeast Regional Experimental Station, and a U.S. Forest- Service-Green Mountain National Forest/Dartmouth College cost-share grant to C. Folt. References Allen, T.F.H., and Hoekstra, T.W. 1992. Toward ;I unified ecology. Columbia University Prcss, New York, N.Y. Allendorf, F.E., Bayles, D., Bottom, D.L.. Currens, K.P., Frissell. C.A., Hankin. D., Lichatowich, J.A.. Nehlsen, W., Trotter, P.C.. and Williams. T.H. 1997. Prioritizing Pacific salmon stocks for conservation. Conserv. Riol. 11: 140-152. 0 1998 NRC Canada 20 Armstrong J. D., Shackley, P.E.. and Gardiner, R. 1994. Redistribu- tion of juvenile salmonid fishes after localized catastrophic de- pletion. J. Fish. Biol. 45: 1027-1039. Armstrong. J.D. 1997. Self-thinning in juvenile sea trout and other salmonid fishes revisited. J. Anim. Ecol. 66: 5 19-96, Arnqvist. G., and Wooster, D. 1995. Meta-analysis: synthesizing research findings in ecology and evolution. Tr. Evol. Ecol. 10: 236-240. Bjornn, T.C., and Reiser. D.W. 1991. Habitat requirements of salmonids in streams. Am. Fish. SOC. Spec. Publ. 19: 83-138. Bjornn, T.C., and Chapman, D.W. 1968. Distribution of salmonids in streams, with special reference to food and feeding. In Sym- posium on salmon and trout in streams. Edifed by T.G. North- cote. University of British Columbia Press, Vancouver, B.C. Bovee, K.D. 1986. Development and evaluation of habitat suitabil- ity criteria for use in the instream flow incremental methodol- ogy. US Fish Wildl. Sew. Biol. Rep. 86(7). Bozek, M.A., and Rahel, F.J. 1991. Assessing habitat requirements of young Colorado River cutthroat trout by use of macrohabitat and microhabitat analyses. Trans Am. Fish. SOC. 120: 571-581. Brandt. S.B., and Kirsch, J. 1993. Spatially-explicit models of striped-bass growth in Chesapeake Bay. Trans Am. Fish. SOC. Brandt, S.B.. Mason, D.M., and Patrick, E.V. 1992. Spatially- explicit models of fish growth rate. Fisheries, 17: 23-35. Cada, G.F., Loar, J.M.. and Sale, M.J. 1987. Evidence of food limi- tation of rainbow and brown trout in southern Appalachian soft- water streams. Trans Am. Fish. SOC. 116: 692-702. Connell, J.H.. Hughes, T.P., and Wallace, C.C. 1997. A 30-year study of coral abundance, recruitment, and disturbance at sev- eral scales in space and time. Ecol. Monogr. 67: 461488. Crowder A.A., Smol, J.P., Dalrymple, R., Gilbert. A., Mathers, A., and Price, J. 1996. Rates of natural and anthropogenic change in shoreline habitats in the Kingston Basin, Lake Ontario. Can. J. Fish. Aquat. Sci. 53(Suppl. 1): 121-135. Cunjak, R.J. 1996. Winter habitat of selected stream fishes and po- tential impacts from land-use activities. Can. J. Fish. Aquat. Sci. Dahl, J., and Greenberg. L. 1996. Impact on stream benthic prey by benthic vs. drift feeding predators: a meta-analysis. Oikos, Dennis, B. 1996. Discussion: should ecologists become Bayesians? Ecol. Appl. 6: 1095-1 103. Dill, L.M., Ydenbexg, R.C., and Fraser, A.H.G. 1981. Food abun- dance and territory size in juvenile coho salmon. Can. J. Zool. Elliott, J.M. 1985. Population dynamics of migratory trout in a Lake District stream 1966-83, and their implications for fisher- ies management. J. Fish. Biol. 27: 3543. Elliott, J.M. 1989. The critical period concept and its relevance for population regulation in young sea trout. J. Fish. Biol. 35: 91- 98. Elliott, J.M. 1994. Quantitative ecology and the brown trout. Ox- ford University Press, Oxford. Ellison, A.M. 1996. An introduction to Bayesian inference for eco- logical research and environmental decision making. Ecol. Appl. 6: 1036-1046. Fahrig, L. 1992. Relative importance of spatial and temporal scales in a patchy environment. Theor. Popul. Biol. 41: 300-314. Fausch, K.D. 1998. Interspecific competition and juvenile Atlantic salmon (Salmo salar): on testing effects and evaluating the evi- dence across scales. Can. J. Fish. Aquat. Sci. 55(Suppl. l): 218- 231. 122: 845-869. 53(S~ppl. 1): 267-282. 77: 177-181. 59: 1801-1809. Can. J. Fish. Aquat. Sci. Vol. 55(Suppl. l), 199 Fausch, K.D., Hawks. C.L., and Parsons. M.G. 1988. Models tha predict standing crop of streaiii fish from habitat variables. Gen Tech. Rep. PNW-GTR-213. Portland, OR. U.S. Dep. Agric. Fo Serv. Fausch, K.D., Nakano. S., and Ishigaki, K. 1994. Distrihuiion o two congeneric charrs in streams of Ilokkaido Island. Japan I considering multiple factors across scnles. Oecologia, 100: I- 12. Fernandez-Duque, E., and Valeggia, C. 1903. Meta-analysis: a valuable tool in conservation research. Conserv. Biol. 8: 555- 561. Filbert, R.B.. and Hawkins. C.P. 1995. Variation in condition of rainbow trout in relation to food, temperature, and individual length in the Green River, Utah. Trans Am. Fish. SOC. 124: 824- 835. Folt, C.L., and Schulze, P.C. 1993. Spatial patchiness, individual performance, and predator impacts. Oikos, 68: 560-566. Fretwell, S.D., and Lucas, H.L. 1970. On territorial behavior and other factors influencing habitat distribution in birds. Acta Biotheor. 19: 16-36. Frost, T.M., DeAngelis, D.L., Bartell, S.M., Hall. D.J., and Hurlbert, S.H. 1988. Scale in the design and interpretation of aquatic community research. In Complex interactions in lake communities. Edited by S.R. Carpenter. Springer-Verlag. Berlin. Gowan, C., and Fausch, K.D. 1996. Mobile brook trout in two high-elevation Colorado streams: Re-evaluating the concept of restricted movement. Can. J. Fish. Aquat. Sci. 53: 1370-1381. Grant, J.W.A. 1993. Self-thinning in stream salmonids. Can. Spec. Publ. Fish. Aquat. Sci. 118. pp. 99-102. Grant, J.W.A., and Kramer, D.L. 1990. Territory size as a predictor to the upper limit to population density of juvenile salmonids in streams. Can. J. Fish. Aquat. Sci. 52: 186-196. Grant, J.W.A.. Steingrimsson, ~.6., Keeley, E.R.. and Cunjak. R.A. 1998. Implications of territory size for the measurement and prediction of salmonid abundance in streams. Can. J. Fish. Aquat. Sci. 55(Suppl. I): 181-190. the results of independent experiments. In Design and analysis of ecological experiments. Edited by S.M. Scheiner and J. Gurevitch. Chapman and Hall, New York, New York. N.Y. pp. 376-398 Gurevitch, J.. Morrow, L.L., Wallace, A., and Walsh, J.S. 1992. A meta-analysis of competition in field experiments. Am. Nat. Harrington, R., Kennedy, B.P., Chamberlain, C.P.. Blum, J.D., and Folt, C.L. 1998. "N enrichment in agricultural catchments: field patterns and applications to tracking Atlantic salmon (Snlmo safar). Chem. Geol. (In press.) Hart, D.D., and Fonseca, D.M. 1995. Relationships between ben- thic distributions and heterogeneous flow environments: pro- cesses and patterns at three spatial scales. J. N. Am. Benthol. SOC. 12: 155. Hewett, S.B., and Johnson, B.L. 1992. Fish bioenergetics model 2 WIS-SG-91-250. Hicks, B.J.. Hall, J.D., Bisson, P.A., and Sedell, J.R. 1991. Re- sponses of salmonids to habitat change. Am. Fish. SOC. Spec. Hill, J., and Grossman, G.D. 1993. An energetic model of micro- habitat use for rainbow trout and rosyside dace. Ecology, 74: 685-698. Hinch, S.G. 1991. Small- and large-scale studies in fisheries ecol- ogy: the need for cooperation among researchers. Fisheries. pp. 229-260. Gurevitch, J.. and Hedges, L.V. 1993. Meta-analy ' 140: 539-572. Publ. 19: 483-518. 16(3): 22-27. 0 1998 NRC Canada J Folt et al. 21 Holtby, L.B. 1988. Effects of logging on stream temperatures in Carnation Creek, British Columbia, and associated impacts on coho salmon. Can. J. Fish. Aquat. Sci. 45: 502-515. H~~~, J K., and Schneider, D.C. 1994. Analysis of scale- dependent processes with dimensionless ratios. Oikos 70: 20 I- 211. Hughes, N.F., and Dill, L.M. 1990. Position choice by drift-feeding salmonids: model and test for Arctic grayling (Thyrtrallrts nrcticus) in subarctic mountain streams, interior Alaska. Can. J. Fish. Aquat. SCi. 47: 2039-2048. Keeley, E.R., and Grant, J.W.A. 1995. Allometric and environmen- tal correlates of territory size in juvenile Atlantic salmon (Strlrrio salar). Can. J. Fish. Aquat. Sci. 52: 186-196. Keeley, E.R., and Grant, J.W.A. 1997. Allometry of diet selectivity in juvenile Atlantic salmon (Salmo salar). Can. J. Fish. Aquat. sci. 54: 1894-1902. Keller, W., and Gunn, J.M. 1995. Lake water quality improvements and recovering aquatic communities. In Restoration and recov- ery of an industrial region: progress in restoring the smelter- damaged landscape near Sudbury, Canada. Edited bv J.M. Gunn. Springer, New York, N.Y. pp. 67-80. Kennedy, G.J.A., and Strange, C.D. 1986. The effects of intra- and inter-specific competition on the distribution of stocked juvenile Atlantic salmon in relation to depth and gradient in an upland trout stream. J. Fish Biol. 29: 199-214. Kennedy, B.P., Folt, C.L., Blum, J.D.. and Chamberlain, C.P. 1997. Natural isotope markers in salmon. Nature, 387: 766-767. Kocik, J.F., and Ferreri, C.P. 1998. Juvenile production variation in salmonids: population dynamics, habitat, and the role of spatial relationships. Can. J. Fish. Aquat. Sci. 55(Suppl. I): 191-200. Lacroix, G. 1996. Long-term enhancement of habitat for salmonids in acidified running waters. Can. J. Fish. Aquat. Sci. Letcher, B.H., and King, T.L. 1997. Tools for evaluation of anadro- mous fish restoration: potential for large-scale genetic marking of Connecticut River Atlantic Salmon. Abstract. ICES Science Meeting, 1997. Levin, S.A. 1992. The problem of pattern and scale in ecology. Lewis, C.A. Lester, N.P. Bradshaw, A.D. Fitzgibbon, J.E. Fueller, K. Hakanson, L., and Richards, C. 1996. Considerations of scale in habitat conservation and restoration. Can. J. Fish. Aquat. Sci. 53(Suppl I): 440445. Mason. J.C. 1976. Response of underyearling coho salmon to sup- plemental feeding in a natural stream. J. Wildl. Manage. 40: Mather. M.E. 1998. The role of context-specific predation in un- derstanding patterns exhibited by anadromous salmon. Can. J. Fish. Aquat. Sci. 55(Suppl. I): 232-246. McMenemy, J.R. 1995. Survival of Atlantic salmon fry stocked at low density in the West River, Vermont. N. Am. J. Fish. Man- age. 15: 366-374. Mills, D.H. 1991. Ecology and management of Atlantic salmon. Chapman & Hall, London. 351 p. Ney, J.J. 1993. Bioenergetics modeling today: growing pains on the cutting edge. Trans Am. Fish. SOC. 122: 736-748. Nicieza A.G., Reyes-Gavilan, F.G.;and Brana, F. 1994. Differenti- ation in juvenile growth and bimodality patterns between north- ern and southern populations of Atlantic salmon Can. J. Zool. Nislow, K.H., Folt, C.L., and Parrish, D.L. 1999. Favorable forag- ing locations for age-0 Atlantic salmon: application to the resto- ration of populations and habitats. Ecol. Appl. (In press.) 53(S~ppl. I): 283-294. Ecology, 73: 1943-1967. 775-788. 72: 1603-1610. Nislow, K.H. Folt. C.L., and Seandel, M. 1998. Food and foraging behavior in relation to microhabitat use and survival of age-0 Atlantic salmon. Can. J. Fish. Aquat. Sci. 55: 116-127. Oksanen. T. Power, M.E., and Oksanen. L. 1995. Ideal-free habitat selection and consumer-resource dynamics. Am. Nnt. 146: 565- 585. Okubo. A. 1980. Diffusion and ec?logical problems: mathematical models. Springer-Verlag. Berlin. Osenberg, C.W., Sarnelle, 0.. and Cooper, S.D. 1998. Effect size in ecological experiments: The application of biological models in meta-analyis. Am. Nat. 150: 798-812. Poff. N.L. 1997. Landscape filters and species traits: Towards mechanistic understanding and prediction in stream ecology. J. N. Am. Benthol. SOC. 16: 391-409. Power, M.E. 1984. Habitat quality and the distribution of algae- grazing catfish in a Panamanian stream. J. Anini. Ecol. 53: 357- 374. Power, M.E., Dudley, T.L., and Cooper, S.D. 1989. Grazing cat- fish, fishing birds, and attached algae in a Panamanian stream. Environ. Biol. Fishes, 26: 285-294. Power, M.E.. Tilman. D.. Carpenter, S.R., Huntly. N., Leibold, M.. Morin, P., Menge, B.A.. Estes, J.A.. Ehrlich, P.R.. Hixon. M., Lodge, D.M.. McPeek, M.A., Fauth, J.E., Reznick, D., Crowder, L.B., Holbrook, S.J., Peckarsky, B.L., Gill, D.E.. Antonovics, J.. Polis, G.A., Wake. D.B., Orians, G., Ketterson, E.D., Marschall, E., and Lawler, S.P. 1995. The role of experiments in ecology. Science, 270: 561-561. Power, M.E., Dietrich. W.E., and Finlay. J.C. 1996. Dams and downstream aquatic biodiversity: potential food web conse- quences of hydrologic and geomorphic change. Environ. Man- age. 20: 887-895. Power, M.E., Dietrich, W.E., and Sullivan, K.O. 1998. Experimen- tation, observation and inference in river and watershed investi- gations. 111 Issues and perspectives in experimental ecology. Edited by W.J. Resetarits and J. Bernardo. Oxford University Press, Oxford. Ray, C., and Hastings, A. 1996. Density dependence: are we searching at the wrong spatial scale? J. Anim. Ecol. 65: 556- 566. Resh, V.H., and Rosenberg, D.H. 1979. Sampling variability and life-history features: basic considerations in the design of aquatic insect studies. J. Fish. Res. Board Can. 36: 289-321. Richardson, J.S. 1993. Limits to productivity in streams: evidence from studies of macroinvertebrates. Can. Spec. Publ. Fish. Aquat. Sci. 118. pp. 9-15. Rose, G.A., and Leggett, W.C. 1990. The importance of scale to predator-prey interactions. Ecology, 71: 3343. Schlosser, I.J. 1991. Stream fish ecology: a landscape perspective. Bioscience, 41: 704-712. Turchin, P., and Thoeny, W.T. 1993. Quantifying dispersal of southern pine beetles with mark-recapture experiments and a diffusion-model. Ecol. Appl. 3: 187-198. Waters, T.F. 1972. The drift of stream insects. Ann. Rev. Entomol. Whalen, K.G., and Labar, G.W. 1994. Survival and growth of At- lantic salmon (Sulmo salur) fry stocked at varying densities in the White River, Vermont. Can. J. Fish. Aquat. Sci. 51: 2164- 2169. Wootton, J.T., Parker, M.S., and Power, M.E. 1996. Effects of dis- turbance on river food webs. Science, 273: 1558-1561. Wright, P.J., Metcalfe, N.B., and Thorpe. J.E. 1990. Otolith and so- matic growth rates in Atlantic salmon: evidence against cou- pling. J. Fish Biol. 36: 241-249. 17: 253-272. 0 1998 NRC Canada