INTRODUCTION

Historically, most high mountain lakes in western North America were devoid of fish (Bahls 1992; Dunham et al. 2004), but during the past century or more, many of these lakes have been stocked with salmonids—usually as fry—to expand and diversify angling opportunities. As a result, many high mountain lakes now provide self-sustaining salmonid fisheries, although stocking programs continue in many U.S. states and Canadian provinces. Such unique fisheries provide solitude, dramatic scenery, and a backcountry experience that are seldom found in other settings. Not surprisingly, anglers visiting high mountain lakes typically express high levels of satisfaction with their angling experience (Wyoming Game and Fish Department 2002; Koenig 2020).

Although angling effort at remote high mountain lakes is presumably diffuse (see McCormick 2015) compared to more accessible fisheries at lower elevations, fisheries managers must still make decisions regarding harvest regulations, which lakes and species to stock, and the rates and frequency of stocking. Numerous factors have contributed to the lack of detailed information regarding fisheries management of high mountain lakes. They include the aforementioned diffuse fishing effort in addition to difficulties in accessibility, a short angling season, more extreme seasonal weather variability, and a relatively low cost to stock lakes (US$250/lake; M. Koenig, Idaho Department of Fish and Game [IDFG], personal communication). Nevertheless, in any managed fishery, it is important to garner information on annual exploitation by anglers—that is, the proportion of fish that are removed from the population annually via fishing harvest. A common technique for estimating exploitation is to release a known number of tagged fish, which relies on anglers to report the tags as a means of generating exploitation estimates (Pollock et al. 2001). Although salmonid anglers often choose to release their catch (Policansky 2002), annual rates of catch and release (C–R) can also be estimated from reported tags by asking anglers whether they harvested or released the tagged fish. Combining estimates of exploitation and C–R provides fisheries managers with a measure of angler utilization of a fishery. To our knowledge, assessing the efficacy of a fish tagging program in high mountain lakes to estimate rates of exploitation and C–R has not been conducted, and our first objective was to fill this information gap.

Although basic information on rates of exploitation and C–R in high mountain lakes is long overdue, so too is information on factors affecting the catch of salmonids that are either stocked or naturally reproducing in such lakes. In Wyoming, angler accessibility was an important factor affecting salmonid size structure in high mountain lakes (Bailey and Hubert 2003); those authors concluded that fisheries located farther from roads were less likely to experience appreciable angler harvest. However, even for high mountain lakes in closer proximity to roads, an angler's interest in fishing a particular lake—and the angler's ability to land fish there—can be influenced by numerous factors besides fish abundance. Other potential factors include lake size (Ashe et al. 2014; Cassinelli and Meyer 2018) and morphology (Arlinghaus et al. 2017), the species (e.g., Brauhn and Kincaid 1982; Dwyer 1990) and size (Aas et al. 2000) of fish in the lake, and lake esthetics (Hampton and Lackey 1976). Our second objective was to evaluate factors associated with whether anglers caught tagged fish in high mountain lakes.

METHODS

There are about 3000 high mountain lakes scattered across Idaho, with over 1000 lakes now containing some species of salmonid and over 600 lakes on some sort of stocking rotation, the most common being a 3-year rotation (Meyer and Schill 2007). Fish are generally stocked from fixed-wing aircraft as 40–60-mm fry, with stocking size varying annually depending on flight scheduling as well as flight delays related to weather and fire conditions. Because fish are stocked as fry, they generally cannot be distinguished from wild fish. Consequently, whether tagged fish in the present study were of wild or hatchery origin was unknown. However, prior research indicates that self-sustaining wild salmonids comprise the vast majority (≥80%) of salmonids that are encountered in high mountain lakes in Idaho (Koenig et al. 2011; Cassinelli et al. 2019) and other locations, such as Wyoming (Wiley 2003). Catchable-sized fish were not stocked in any of the study lakes.

We tagged fish at 103 high mountain lakes (Figure 1). The lakes ranged from 1610 to 3158 m in elevation and from 0.4 to 21.9 ha in surface area (Table 1). Hiking distance to access these lakes ranged from 0 km (i.e., accessible by vehicle) to 28 km. Fishing regulations at all lakes included a six-fish daily harvest limit except at one lake that had a two-fish daily harvest limit. There were no size or gear restrictions.

Shore-based angling with artificial flies or lures was used to capture fish for tagging. Tagging was initiated in August 2017 and continued through August 2022, with tags being implanted in fish generally between late June and late September, depending on the year. Several of the 103 lakes were visited in more than 1 year over the entire study period, with a total of 114 tagging events. Tags were dispersed haphazardly across lakes, with an average of 10 tags released per tagging event. Species tagged in this study included Cutthroat Trout Oncorhynchus clarkii, Rainbow Trout O. mykiss, Rainbow Trout × Cutthroat Trout hybrids, Arctic Grayling Thymallus arcticus, and Brook Trout Salvelinus fontinalis (Table 2).

Landed fish were identified to species, measured to the nearest millimeter total length (TL), and implanted with a T-bar anchor tag at the base of the dorsal fin (Dell 1968). Tags were implanted in fish as small as 150 mm TL, but 98% of tagged fish were 175 mm TL or larger (mean = 272 mm; maximum = 510 mm). Anchor tags were fluorescent orange, 70 mm in length (51 mm of tubing), and treated with algicide to prolong readability. Tags were labeled on two sides, with one side stating the agency (IDFG) and phone number and the other side listing the agency website and a unique tag number.

Angler reporting of tagged fish was summarized from August 5, 2017 (the first day of fish tagging), to November 29, 2022. Anglers could report tags via the phone number or website, by traditional mail, or by visiting an IDFG office. Regardless of how the tags were reported, anglers were asked a series of questions, including the date on which the fish was captured and whether they harvested the fish or released it. Three tagged fish were reported twice by different anglers, and we included each reporting event as separate information for those fish. Among the 113 anglers who reported tags, eight anglers reported more than one tag from the same lake.

We measured several lake-specific characteristics that we felt might influence whether anglers caught a tagged fish in high mountain lakes (Table 1). Elevation (m) and lake surface area (ha) were estimated using Google Earth Pro version 7.4.6 (Google 2022). The shoreline development index, defined as the ratio of a lake's shore length to the circumference of a circle with that lake's area (Wetzel 2001), was also estimated from measurements in Google Earth Pro. Human population size living within 100 km of the lake was included based on the presumption that more people living in the vicinity of the lake might increase angler effort (Post et al. 2008); geospatial human population data were obtained from Environmental Systems Research Institute, Inc. based on the 2022 U.S. Census block-level population data.

Three measures of lake access difficulty were also included as predictor variables. First, hiking distance (km) from the nearest road to the lake was estimated by tracing the likely route using U.S. Geological Survey 1:24,000-scale digital topographical maps. The hiking route normally followed trails, but where trails were absent, we tracked what we assumed was the most likely route that hikers would take. We assumed that all angler access was by foot, though all trails were open to horseback riding and some trails were open to two-wheel or four-wheel all-terrain vehicles. Using the same digital maps, we also estimated the proportion of the hike that was on a trail and the cumulative gain in elevation (m) during the hike. Cumulative gain was included (rather than net gain) to account for instances in which the hike required passing over a series of one or more elevational gains during the hike. For the few instances in which the lake was lower in elevation than the starting point of the hike, we used the cumulative loss in elevation instead of the gain because the return hike from the lake to the vehicle would comprise the same amount of cumulative gain in elevation. Three final predictor variables were the species and length of fish being tagged and the Julian date of tagging. We included tagging date because fish that were tagged later in a field season would not have equal vulnerability to anglers in any given year relative to fish that were tagged earlier in the field season.

Statistical analyses were conducted using SAS (SAS Institute 2009). Multicollinearity among continuous predictor variables was assessed by calculating Pearson's correlation coefficients (r) between independent variables, which were less than $\left|0.40\right|$ for all but one comparison (total hiking distance versus cumulative elevation gain; r = 0.73), indicating a general lack of collinearity in the data. We used the GLIMMIX procedure in SAS to construct a generalized linear mixed model for binary data with random effects (essentially a logistic regression with random effects) to relate predictive variables to the catch of tagged fish by anglers. Each tagged fish was considered the unit of observation, and the dependent variable in the model was a dummy variable representing whether a tag was reported as caught by anglers (1 = reported; 0 = not reported). All predictor variables were included as fixed effects except for lake, which was included as a random effect. For the lone categorical predictor variable of species, Rainbow Trout was the reference species in the models.

All models were constructed with the inclusion of one predictor variable in addition to the random effect for lake; the null model included only the random lake effect. We also constructed all combinations of two-factor models; the random lake effect was included in all two-factor models as well. Models were ranked using Akaike's information criterion corrected for small sample size (AIC_c; Burnham and Anderson 2002), and we considered the most plausible models to be those with AIC_c scores that were within 2.00 units of the best model (Burnham and Anderson 2004). We used AIC_c weight (w_i) to assess the plausibility of a particular model relative to all other candidate models. Coefficient estimates were exponentiated for easier interpretation and are presented only for the most plausible models. Coefficients were only considered influential if their 95% CIs did not include unity. Interaction terms were originally included in all two-factor models; however, AIC_c scores were better in all instances for models without the interaction terms and the 95% CIs overlapped unity for all interaction terms, so the interactions were discarded. To assess model fit and classification accuracy, we calculated the area under the receiver operating characteristic curve (AUC) for the most plausible models.

RESULTS

In total, 1163 tags were implanted in various species of salmonids in 103 lakes scattered across Idaho (Figure 1); 125 of the tags from 52 different lakes were reported by anglers. Tagged fish were caught by anglers as early as June 9 and as late as October 17 in any given year, but the vast majority (>80%) were reported as having been caught between July 1 and September 10. Forty-nine tagged fish (39%) were reported as caught during the same year in which tagging occurred; the mean number of days at large for tagged fish was 318 days, with a range from 1 to 1465 days after tagging.

Tagged fish were more often released after capture (n = 79) than harvested (n = 46), with Cutthroat and Rainbow trout having the highest proportions of tagged fish being released by anglers (Table 2). Across all species combined, annual exploitation was 5.8 ± 7.1% (mean ± 90% CI), annual C–R was 8.5 ± 8.5%, and total annual catch (i.e., exploitation plus C–R) was 14.2 ± 11.0% (Table 3). The annual total catch percentage was highest for Rainbow Trout (21.7%) and lowest for Brook Trout (6.4%), but the 90% CIs overlapped for all species comparisons (Table 3).

The most plausible models explaining the variation in angler catch of tagged salmonids in high mountain lakes indicated that among the predictor variables included in our analyses (besides the random lake effect), the likelihood of a fish being caught by an angler was a function of hiking distance, shoreline uniformity, and fish size (Table 4). In the most plausible model, the odds of a fish being caught by an angler decreased by 11% for every additional kilometer of hiking distance and increased by 46% for every 0.1-unit increase in the shoreline development index (Table 5). A second plausible model indicated that the odds of a fish being caught by an angler decreased by 10% for every additional kilometer of hiking distance and increased by 0.4% for every 10-mm increase in fish TL. Model accuracy was 80% for both of these plausible models based on the AUC. There was little support for any other candidate models.

DISCUSSION

Due to the remote nature of many high mountain lakes in Idaho and throughout the Rocky Mountains and due to the rigorous hiking required to access many of these areas, it has been assumed that angling pressure is relatively low in such fisheries compared to more easily accessed lowland waters. However, empirical support for this assumption has been lacking. Our results indicate that annual rates of exploitation and C–R for salmonids occupying high mountain lakes in Idaho are indeed relatively low, with less than 15% of salmonids residing in these lakes estimated as being landed each year by anglers and less than 50% of landed fish estimated as being harvested. Although we cannot be certain that the low rates of annual exploitation and C–R were the result of low angling pressure and not low catch rates, we presume that the former explains our findings. Future studies that characterize catch rates and angling pressure at high mountain lakes would help to support or refute our presumption. The low exploitation rates that we observed would not be expected to markedly affect the abundance or size structure of salmonid populations in these types of fisheries, although error bounds were admittedly wide and we encourage continued tagging during routine lake surveys in these and adjacent lakes to improve the precision in our estimates of exploitation and C–R. Similar work in other regions of North America and beyond would allow for broader comparisons to our work.

We observed that tagged fish residing in high mountain lakes farther from access locations were less likely to be caught by anglers, presumably due to less angling effort. Although such a relationship is not surprising given that angling effort and accessibility are also inversely related in lowland lake fisheries (Post et al. 2008; Hunt et al. 2011), this effect was likely confounded in our study by two factors. First, fish residing in more distant waters with less fishing pressure likely had higher vulnerability to angler catch (van Poorten and Post 2005; Askey et al. 2006), and this would have diminished the accessibility–catch relationship that we observed. Second, the full impact of lake accessibility (as we characterized it) may also have been diminished because the remoteness of some lakes was likely circumvented by a segment of the angling population. For example, at the lake with the farthest hike (22 km) in our study, an angler reported capturing one of the 18 fish that we tagged in the lake. It is possible that the angler hiked to this lake, but it is more likely that access was achieved via a nearby wilderness airstrip. Similarly, some lakes that are far from a road could be accessed via trails open to the use of all-terrain vehicles, but only a portion of the angling public owns such vehicles; thus, access was not equivalent among all anglers. In the future, asking anglers (when they report tagged fish) how they accessed high mountain lakes would provide better information about the effect of accessibility on fish catch in such remote settings.

In addition to the importance of lake accessibility, our results suggest that a less uniform perimeter to the lake improved the likelihood that fish would be caught by anglers. A simple explanation is that a high mountain lake with a more sinuous shoreline inherently offers more projections of land out into the lake, making it easier for anglers to cast (and, thus, to catch fish) while avoiding shoreline trees and other casting obstructions. In addition, a more convoluted lake has more littoral habitat relative to pelagic habitat (Dolson et al. 2009), and littoral habitat has long been associated with increased lake productivity for fish (Moyle 1949; Northcote and Larkin 1956), including for salmonids in high-elevation waters (Chamberlain and Hubert 1996). Nonuniform shorelines may also contribute more terrestrial input into the lake (Wetzel 2001; Solomon et al. 2011), which can be important for salmonids in high-elevation lentic environments (Milardi et al. 2016). Taken collectively, these benefits may result in a higher carrying capacity in lakes with less uniform shorelines, increasing the number of fish that are available for anglers to target. Shoreline uniformity can be scouted by anglers using satellite images prior to choosing a hiking destination. If anglers are aware of the potential benefits of nonuniform shorelines, such an awareness may affect which lakes anglers choose to fish and what level of angling success they have upon arrival.

Our results suggest that larger fish were more likely to be captured by anglers, as has been repeatedly established in the literature for salmonids (e.g., van Poorten and Post 2005; Askey and Johnston 2013; Tsuboi et al. 2021; but see Meyer and Schill 2014). We speculate that larger trout in high mountain lakes may be inherently more likely to be caught by anglers due to behavior or habitat preferences. For example, larger fish are less vulnerable to predation (Penaluna et al. 2016; Sherker et al. 2021) and therefore may be more willing to occupy shallower, nearshore habitat, making them more exposed to shoreline anglers. Additionally, larger salmonids may also agonistically exclude smaller fish from more productive littoral habitat (Amundsen et al. 1993), which could diminish their vulnerability to angling.

The random lake effect was considered influential (in terms of error bounds around the effect size) in both of the most plausible models, which indicates that angler catch was affected by among-lake differences that were not captured in our study design. Obvious omissions are the esthetics and amenities of a particular lake, such as the grandeur of the setting or the ability to locate flat, soft spaces that are large enough for overnight camping. Myriad other catch-related and non-catch-related determinants likely influence whether an angler chooses to expend angling effort or succeeds in landing fish in remote alpine fisheries. Such determinants include angler catch rates, angling avidity, backcountry proficiency, shoreline vegetation, angling regulations, and the length of time for which a lake and the surrounding area are free of snow and ice conditions. What motivates anglers to fish certain waters is not fully understood, is difficult to quantify (Fedler and Ditton 1994; Hunt et al. 2019), and was outside the scope of our study.

Estimates of annual exploitation and C–R using reporting of tagged fish by anglers assume that all tagged fish are equally vulnerable to anglers. Our study design violated this assumption because fish were tagged throughout the angling season, and fish in high mountain lakes are not vulnerable to anglers for most of autumn, winter, and spring due to ice cover and harsh winter conditions. Energy deficits during winter for salmonids residing in such conditions at high elevation likely result in elevated mortality compared to other times of the year (Biro et al. 2021). Consequently, fish that were tagged at the end of a field season were not as vulnerable to angler catch as those tagged at the beginning of that field season. To assess how much this might have biased the estimates of annual exploitation, we divided the tags that were reported by anglers as harvested within 365 days of the tagging date into (1) tags reported in the same year and (2) tags reported in the following year. For the latter group, we assumed that total annual mortality (A) was 53.0% (Roth et al. 2022); u was 5.8% (this study); and, therefore, the expectation of natural death (v) was 47.2%. The sum for the latter group was divided by v and added to the number of tags reported as harvested during the same calendar year in which they were released, and this total was used to estimate an annual rate of exploitation that was now ostensibly adjusted for unequal tag vulnerability. The same adjustment was made for tags that were reported as caught and released. These adjustments changed our results very little: annual exploitation increased from 5.8% to 8.7%, and the annual C–R rate increased from 8.5% to 10.4%.

Besides fish being tagged throughout a prolonged angling season (rather than prior to the angling season) and the overarching issue of small sample size for exploitation and C–R estimates, a final limitation of our study was that at the time of tagging, fish origin (hatchery or wild) could not be distinguished. Indeed, fish were stocked as fry and lake surveys generally occurred 2–3 years after the last stocking event, thus allowing hatchery fish a very long period to become acclimated to their environment. Nevertheless, differences in vulnerability to angling between stocked and wild salmonids can persist for many years (Champigneulle and Cachera 2003). Because most (≥80%) salmonids residing in Idaho high mountain lakes are old, wild-origin individuals (Koenig et al. 2011; Cassinelli et al. 2019), we assume that our results primarily apply to wild fish. Since the hatchery Rainbow and Cutthroat trout stocked in Idaho are triploid fish, any continued tagging in this program could incorporate fin clips into the study design prior to releasing fish, and genetic ploidy analyses could be used to differentiate wild fish from hatchery fish.

At present, anglers do not appear to be overharvesting fish in most high mountain lake fisheries within Idaho, and considering the increasing trend in C–R angling behavior in salmonid fisheries (Policansky 2002), overharvest impacts appear to be unlikely for the foreseeable future. Nevertheless, we encourage continued tagging by biologists during routine high mountain lake fishery surveys as a cost-effective method of monitoring spatiotemporal changes in angler exploitation and C–R rates in such remote fisheries, where other methods of gathering this type of information (e.g., creel surveys) are less feasible. Such a tagging program could also be used as another means of contact with a small but important segment of the angling public.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflict of interest.

ETHICS STATEMENT

There were no ethical guidelines applicable to this study.