Introduction

Freshwater fishes such as salmonids require a diversity of habitat throughout their life history. Such habitat is referred to as complementary habitat (Schlosser 1995), where loss of a habitat type (i.e., foraging, spawning, refugia) would result in the extirpation of a local fish population (Schlosser 1995). A variety of factors influence habitat selection for fishes, where heterogeneity in resources exist both spatially and temporally. For example, brook trout (Salvelinus fontinalis) require small headwater streams for reproduction in the fall, but can experience substantial gains in somatic growth if foraging in a larger main stem in the spring (Petty et al. 2005; Petty et al. 2014; Huntsman et al. 2016). In addition to foraging and spawning habitat, refugia (e.g., coldwater or warmwater, fish cover) are an equally important habitat type that must be accounted for when characterizing habitat needs (Gilliam and Fraser 1987). Although the importance of complementary habitat to fishes is well accepted, how fishes interpret habitat needs is species and location specific (Rosenfeld et al. 2014).

Ontogenetic shifts in habitat requirements and the dynamic nature of lotic systems make identifying ideal habitat for fishes difficult. Ontogenetic shifts in habitat and resource use (e.g., prey selection) is well documented in many fish species (Werner and Hall 1988; Persson and Hansson 1999; Anglin and Grossman 2013). For example, shifts in resource use may be the result of competition, where subordinate fish (often smaller fish) are forced into lower quality habitats (Hughes 1998; Petty and Grossman 2010; Ayllón et al. 2013). Issues associated with characterizing suitable habitat for fishes are exacerbated by changes in habitat through anthropogenic alteration to the landscape (e.g., altered flow regimes by dams) and a changing climate (García et al. 2011; Wenger et al. 2011).

Freshwater resources are especially susceptible to climate change in the Southwest United States (Seager et al. 2007) and access to suitable habitat for many fishes in this region has become a significant limiting factor. Limited precipitation, high temperatures, and land management practices (e.g., irrigation) specific to the Southwest already threaten persistence of many aquatic communities and these stresses will likely be exacerbated with a changing climate (MacDonald 2010). Indeed, Wenger et al. (2011) demonstrated that increasing temperatures and altered flow regimes anticipated by warming may drastically reduce suitable habitat for trout assemblages throughout much of the Southwest. Thus, it is imperative that aquatic ecologists and fisheries managers describe what constitutes suitable habitat for freshwater communities in this region.

Correlative studies often identify strong relationships between physical microhabitat and fish presence, however, these studies commonly exclude foraging habitat as an explanatory variable (Rosenfeld et al. 2014). Reliance solely on microhabitat characteristics for management may be insufficient or misleading. The consequence of losing a positively selected habitat is usually unclear because fish can often maintain positive growth in a habitat that is selected less than other habitats. Eventually, the less preferred habitat may act as a “metabolic sink” where fish lose weight and die (Sogard 1994; Rosenfeld 2003). Therefore, energetic consequences of selected microhabitats must be considered to understand how microhabitat affects fish metabolic needs.

For drift-foraging fishes such as trout, drift-foraging models provide a unique opportunity to assign profitability criteria to foraging habitat patches (Rosenfeld 2003). In addition, bioenergetic models (i.e., drift-foraging models) provide a mechanistic approach to complement microhabitat selection studies by offering insight into the causation underpinning microhabitat selection. Bioenergetic models have been informative in stream drift-foraging where the mechanisms underlying foraging are relatively easily modeled to predict microhabitat choice of stream salmonids based on energetic costs and benefits (e.g., Fausch 1984; Hughes and Dill 1990; Railsback and Harvey 2002; Jenkins and Keeley 2010). The limited application of bioenergetic models in explaining fish distributions is likely due to the effort required to collect the data necessary for model construction. However, application of bioenergetic models in stream ecology is increasing in popularity, as evident by the drift-foraging compendium published in Environmental Biology of Fishes (see Piccolo et al. 2014).

This study was motivated as part of a larger assessment of the Rio Ruidoso headwaters in southcentral New Mexico to determine the suitability of Rio Grande cutthroat trout (Oncorhynchus clarkii virginalis) repatriation. The headwaters reside on the Mescalero Apache Reservation where the Mescalero Tribe would like to restore this native trout to their lands. The goal was to determine how rainbow trout (O. mykiss) utilized habitat year round in an isolated area of the stream system where migration barriers limit fish movement during periods of presumed thermal stress (e.g., super-cooled water containing anchor and frazil ice) and low discharge. This study utilized microhabitat characteristics (i.e., water depth, water velocity, cover distance, and water temperature) and a mechanistic drift-foraging model to explain fish habitat selection and energetic profitability of selected habitat. The objectives were to 1) identify differences in stream habitat and habitat selection among seasons and size-classes of rainbow trout and 2) assess the energetic profitability of used and available habitats for each month and size-class.

Methods

Study area

The Middle Fork Rio Ruidoso is a small, second order headwater stream originating in the Sacramento Mountains in southcentral New Mexico (Fig. 1). The drainage area of the river is approximately 7.2 km2 and precipitation occurs largely as snow. Peak runoff occurs in late March–April with significant supplemental precipitation occurring in July and August in the form of large rain events during summer monsoons. The study area is characterized as high gradient (>5%) with a series of step-pool mesohabitats composed of both short riffles and runs between pools.

Fig. 1
figure 1

Map of the study area within the Middle Fork Rio Ruidoso, Mescalero Apache Reservation, New Mexico. The upper and lower bounds of the study reach were barriers and are represented by dark black lines. Temperature logger locations are depicted by open stars and the upper and lower bounds of each study reach are identified as closed triangles

Full size image

The study area encompassed 1.5 km between two natural fish barriers (i.e., waterfalls) on the Mescalero Apache Reservation. Elevation within the study area varies from 2450 m at the lower barrier to 2627 m at the upper barrier. There are no perennial tributaries within the study area. Rainbow trout is the dominant fish species in the study area (100–200 fish/100 m) with few brook trout present (<1 fish/100 m). Fishing pressure is negligible because public access is restricted to only Tribal members.

A temperature logger (ProV2; Onset, Inc.) was randomly positioned between 0 m and 240 m upstream of the lower barrier and five subsequent loggers were systematically placed every 240 m upstream of the previous logger (Fig. 1). Three 50 m long study reaches were selected downstream of the first, middle, and last temperature loggers to assist with the monthly assessment of habitat availability and use by rainbow trout (Fig.1).

Habitat availability

Available habitat was measured monthly at the three study reaches between December 2011 and November 2012. Cross-sections were systematically placed along the thalweg of each 50-m reach to measure available fish habitat. The first cross-section was randomly placed between 0.01 and 1.5 m upstream of the lower end of the reach and subsequent cross-sections were placed every 1.5 m upstream of the previous. Distance between cross-section intervals was increased to 3 m after March 2012 (no significant difference in the distribution of data) to reduce sampling effort. Stream depth (cm), focal water velocity (cm/s), mean water velocity (cm/s), and distance to cover (cm) were measured at 0.25 m intervals along each cross-section and values were then averaged for each cross-section. Stream depth (± 1 cm) and water velocity (± 0.1 cm/s) were measured with a portable velocity meter (Hach model FH950, Loveland, Colorado) using a wading rod along cross-sections. Focal water velocity was taken at the average depth of the snout of fish observed within the study area (2.5 cm from bottom). If stream depth was <0.5 m, mean water velocity was taken at 60% of the stream depth. If stream depth was ≥0.5 m, mean water velocity was calculated by averaging water velocities taken at 20% and 80% of the stream depth (Gordon et al. 2004). Cover for trout was defined as anything that can provide concealment (e.g., woody debris, boulders, undercut banks, overhanging vegetation, ice).

Habitat use

Foraging locations were similarly evaluated every month at all three study reaches between December 2011 and November 2012. Shallow water precluded underwater observation techniques (Heggenes et al. 1990); therefore, two observers performed the characterization of fish locations from positions along each stream bank before measuring use or available habitat variables. Prior to the monthly characterizations, the observers calibrated visual estimates of fish lengths by placing pieces of PVC cut to a range of fish lengths on the stream benthos (Dolloff et al. 1996); after which, observers cautiously moved upstream to minimize disturbance to fishes. Each fish was allowed a minimum of 10 min to re-adjust their positions as a precaution. Observations commenced when all fish selected a position and began foraging. The observers sketched fish locations relative to stream features and total length (±1 cm) and focal height of each fish (±1 cm; taken at snout of fish) were estimated in proximity to reference objects. Once each habitat unit was surveyed, an observer entered the stream to place colored markers under the position of each fish with the guidance of the second observer from the stream bank. After the reach was surveyed for fish, stream depth, focal water velocity (velocity at snout of fish), mean water velocity, and distance to cover were measured at every marked fish location as previously described for available habitat.

Bioenergetic model development

Drift-foraging model: Foraging profitability can be evaluated in the context of bioenergetic principles. For stream dwelling salmonids, gross energy intake (GEI) is the total energy ingested by fish from drifting prey and is calculated by estimating the amount of drifting prey passing through a fish’s effective foraging window, also known as the maximum capture area (MCA). The MCA can be modeled as a half circle area that is perpendicular to streamflow, within which a fish can capture prey (Fig. 2). The radius of the MCA is defined by the maximum capture distance (MCD) from which a fish can detect a prey item and react to the item (reaction distance, RD) before it drifts past the fish.

Fig. 2
figure 2

Diagram depicting the maximum capture area (MCA; half circle) with a radius equal to the maximum capture distance (MCD) of a foraging salmonid. RD = reaction distance, Vave = mean velocity of the MCA, and V = focal point velocity. Modified from Guensch et al. (2001)

Full size image

To assess profitability of both used and available fish positions, energetic costs were modeled using the concept of net energy intake (NEI). The bioenergetics foraging model of NEI developed by Hughes and Dill (1990) and modified by Addley (1993) and Jenkins and Keeley (2010) includes estimates of potential fish energy intake, physiological energy needs and losses, and estimates of effective foraging distance (i.e., the distance fish can and will detect and pursue prey) to assess the energetic value of fish habitats. Potential energy intake was calculated for each fish position noted by observers and each available habitat location averaged for each transect. Using depth, focal and mean velocities, and prey drift densities, a gross energy intake (GEI) value was calculated for each focal position and available location. Losses from swimming, cost of intercepting a prey item, cost of digesting a prey item, and losses of energy through feces were then subtracted from GEI to calculate NEI. The model of Jenkins and Keeley (2010) was further refined to include maximum sustained swimming velocity obtained from Harvey and Railsback (2009) that used maximum sustained swimming velocity data from rainbow trout (Table 1). Minimum prey size was also modified in this study and defined as the smallest prey item in which the benefit of capturing the prey item (E i ) outweighed the cost of capture (CC i ; Table 1). Due to gape limitations of age 0 fish, an equation from Keeley and Grant (1997) was used to calculate the maximum prey size that an age 0 fish could ingest (PL max ; Table 1). Estimates of fish weight (as a function of length), necessary for calculating the swimming cost (SC) equation, were estimated from a length-weight relationship developed from more than 500 rainbow trout collected throughout the study area (Table 1).

Table 1 Equations used to calculate net energy intake (NEI) of fish positions
Full size table

Net energy intake was estimated based on whether or not the estimated GEI of a foraging position exceeded the maximum consumption of the fish (C max ). Fish cannot eat an infinite amount of food; thus, a maximum food ration that a fish can ingest in a day was calculated (Elliott 1976; Table A1). The resulting maximum energy that a fish could consume in a day was used as the upper limit for GEI. For this study, estimates of C max , modified by foraging time, were based on the amount of daylight hours available for fish to forage at the time of sampling. While fish can forage continually, nighttime foraging assessment of drift was hazardous and thus not within the scope of this study. The following equation was used to estimate C max :

$$ {C}_{max}=\frac{1}{feeding\frac{hours}{day}}\bullet D\bullet 4.438\frac{ calorie s}{mg}\bullet 4.1868\frac{joules}{calorie}\bullet 0.58\kern0.24em $$
(1)

where the maximum food ration (D, mg/day) was calculated from Elliott (1976; Table A1), conversion to calories (4.438 cal/mg) was from Elliott (1976), and the energy assimilation fraction (0.58) was from Elliott (1976) and Brett and Groves (1979). The amount of feeding hours/day was assigned as the number of daylight hours from 0.5 h before sunrise to 0.5 h after sunset to the nearest 0.5 h. Net energy intake was estimated using the following equations:

$$ {\displaystyle \begin{array}{l} If\ GEI\ge {C}_{\mathrm{max}}, then:\\ {} NEI={C}_{\mathrm{max}}\times {E}_i-\left( SC\times 3600\right)\\ {} If\ GEI<{C}_{\mathrm{max}}, then:\end{array}} $$
(2)
$$ NEI=\frac{\sum \limits_{i=1}^n{DD}_i\bullet {MCA}_i\bullet {V}_{ave}\bullet \left({E}_i-{CC}_i\right)- SC}{1+\sum \limits_{i=1}^n{t}_i\bullet {DD}_i\bullet {MCA}_i\bullet {V}_{ave}}\bullet 3600 $$
(3)

where: E i was the energy acquired from a food item of size-class i, and SC was the swimming costs of maintaining a foraging position, DD i was the drift density of prey size-class i that enters the fish’s maximum capture area (MCA, see Table 1), V ave was the mean water velocity within the MCA, CC i was the cost associated with capturing a prey item of size-class i, and t i was the time spent handling a prey item of size-class i.

Macroinvertebrate collections: A preliminary assessment of macroinvertebrate drift within the Middle Fork Rio Ruidoso over a 24 h period demonstrated that drift had a diel pattern with peak drift rates at night and minor variation from 1 h before sunrise to 1 h after sunset. Therefore, drift nets (45 cm width × 27 cm height; 500 μm mesh) were set on every sample date for one hour within each reach, three times during the day beginning at 09:00, 12:00, and 15:00 to obtain drift density estimates for the drift-foraging model. Drift nets were secured into the substrate using metal stakes with the top of the net protruding from the water to collect macroinvertebrates drifting throughout the entire water column. Mean water velocity and depth were measured at eight evenly spaced points across the mouth of the drift net to calculate the volume of water passing through the net over time. A single drift net at each site was sufficient at sampling greater than 75% of the stream discharge within each sample reach. At the end of each hour, specimens were transferred from each net to a Nasco whirl-pak™ storage bag and preserved in 95% ethanol for identification and enumeration. Macroinvertebrates were classified by taxa (order or family), measured for length (± 0.25 mm), and assigned to 0.5 mm size-classes. Drift density for each size-class and net were calculated using the formula by Allan and Russek (1985):

$$ DD=\frac{prey\;}{Q\;}. $$
(4)

where DD is drift density (total number of prey/cm3), Q = volume of flow filtered by the drift net (cm3/h), and prey is the number of prey captured in one hour. A daily mean drift density was computed for each reach using the three daily drift samples and used in the bioenergetics model for each sampling period.

Model application: Net energy intake was estimated for individual fish observed at each foraging location. The bioenergetics model used individual fish length, focal water velocity, mean water velocity, depth, reach-specific macroinvertebrate drift density, and mean water temperature for that day. Net energy intake was similarly calculated at each site and each sample time to assign energetic values to available stream sites throughout each reach by using the available habitat data collected along cross-sections at each site and the respective drift densities for those stream reaches. The same calculations were made for estimates of available NEI, but the total length midpoints for five size-classes of fish representing mean lengths-at-age: 57 mm (age 0), 99 mm (age 1), 143 mm (age 2), 177 mm (age 3), and 210 mm (age 4+) were used. Size-classes were based on back-calculated lengths-at-age from fish scales collected throughout the study using the Fraser-Lee method as described by DeVries and Frie (1996). Average fish lengths observed for the age 0 and age 4+ size-classes were used for available NEI estimates instead of midpoints because there is no midpoint for these two size-classes (being the largest and smallest size-classes). Fish size has a large influence on NEI estimates and therefore it was necessary to use multiple size-classes. Net energy intake values were calculated for each size-class at each point along the cross-section and averaged over each cross-section. Estimates of NEI were used to evaluate the following: (1) proportion of fish meeting maintenance ration (NEI ≥ 0); (2) proportion of sites with available habitat meeting maintenance ration (NEI ≥ 0); (3) proportion of fish meeting maximum growth ration (C max ); and, (4) proportion of sites with available habitat meeting maximum growth ration (C max ) for each month.

Statistical analysis

The three study reaches in the Middle Fork Rio Ruidoso exhibited similar ranges of available microhabitat within each month, between months, and within a season (Kalb 2013). Therefore, all data were combined for each season for analyses. Seasons (i.e., winter, spring, summer, and fall) were delineated based on mean monthly temperatures and assigned to months. The coolest three months were classified as winter (December–February), and the warmest three months were classified as summer (June–August). Spring (March–May) and fall (September–November) were classified as the transitional periods between summer and winter.

Habitat use and habitat availability data were compiled in a presence/availability framework for logistic regression analysis. Only five variables were included in the regression analysis because of high correlation among predictor variables (Pearson’s correlation >0.5). Stream temperature, depth (square root transformed), focal velocity (log +1 transformed), distance to fish cover, and NEI were used for model development. A conventional presence/absence study design could not be used for resource selection probability function development because all fish in each study reach could not be located with certainty (Fielding and Bell 1997). However, resource selection probability functions from presence/absence data is proportional to resource selection functions (RSF) from presence/available data (Johnson et al. 2006), and likelihood methods should not be affected by using presence/available data (Boyce et al. 2002). A candidate set of models were constructed within a resource selection function (RSF) framework (Boyce et al. 2002) for each season and fish size-class. Twelve a priori hypotheses were constructed, where habitat use was modeled as a function of the five covariates (Table A2).

Logistic regression models were constructed with the “glmer” function in the “lme4” package (Bates et al. 2015) in program R. Each regression model included a similar random month effect to account for repeated sampling in each location every month. The variance-covariance matrices for the random effects were estimated using Powell’s BOBYQA algorithm option as a non-linear optimizer in the lme4 package. To help with model convergence, NEI and distance to fish cover were rescaled by subtracting the mean from each observation and then dividing by the standard deviation. Model selection was carried out by Akaike’s information criterion corrected for small sample size (AICc, Burnham and Anderson 2002) using the MuMIn package in program R (Bartoń and Bartoń 2015).

Model validation was assessed using the cross validation approach described by Boyce et al. (2002) and Johnson et al. (2006). For model validation, a training and testing data set of 67 and 33%, respectively, was used based on the following equation:

$$ Testing\kern0.17em data\; set={\left(1+{\left(p-1\right)}^{0.5}\right)}^{-1} $$
(5)

where p is the number of predictors in the model (Huberty 1994; Fielding and Bell 1997). The training data set was used to construct and rank candidate models during model selection, and the testing data set was used for model validation. The range of predicted RSF values were partitioned into 10 equally spaced bins and the frequency of RSF scores for each used location from the testing data set were placed into the bins to examine model performance. A Spearman-rank correlation between the RSF bin ranks and bin use frequencies was calculated for the best model of each candidate set (Boyce et al. 2002; Johnson et al. 2006). A model with good predictive performance would have a significant positive correlation, which would indicate fish were more frequently observed in habitats with high RSF scores.

Results

Habitat availability

Available habitat reflected seasonal thermal and hydrologic conditions typical of southcentral New Mexico headwater streams. Water temperature was warmest in the summer (15.0 °C) and coldest in winter (0.8 °C), with similar mean temperatures between spring and fall (8.1 °C and 8.2 °C, respectively; Table 2). Discharge was also lowest in summer (0.002 m3/s) and highest in late winter/early spring (0.006–0.007 m3/s) reflecting precipitation patterns (Table 2). Overall, higher discharge in winter and spring led to the highest focal velocities (7.1 cm/s and 6.8 cm/s, respectively) and greatest mean depths (14.0 cm and 13.7 cm, respectively; Table 2). Available distance to fish cover was lowest in the winter and spring when ice cover was commonly encountered (Table 2).

Table 2 Mean habitat measurements by season for available habitat and at locations used by fish of ach size-class within the Middle Fork Rio Ruidoso, Mescalero Apache Reservation, New Mexico. Standard errors are reported in parentheses
Full size table

Macroinvertebrate drift

Macroinvertebrate drift was strongly season- and temperature- dependent, affecting the amount of potential energy available at stream locations. Mean macroinvertebrate drift density and energy density was relatively low throughout the winter and spring months (<6 macroinvertebrates/m3; ≤46 joules/m3) before peaking in June (25 macroinvertebrates/m3; 219 joules/m3; Fig. 3). Drift rates declined substantially in July (five macroinvertebrate/m3; 15 joules/m3) but rebounded and remained relatively high throughout the fall months (5–18 macroinvertebrates/m3; 29–64 joules/m3) despite declining temperatures (Fig. 3).

Fig. 3
figure 3

Mean monthly macroinvertebrate drift density (macroinvertebrates/m3) and drift energy density (joules/m3) within the Middle Fork Rio Ruidoso, Mescalero Apache Reservation, New Mexico. Error bars represent standard error. Note: no data was collected in August due to high turbidity

Full size image

Habitat selection

Model selection explained resource selection of rainbow trout from the Middle Fork Rio Ruidoso, which varied substantially by size-class and by season (Table A3). Some candidate models for winter age 3 and summer age 4+ size-classes failed to converge and were subsequently not interpreted (Table A3). Depth appeared in the most parsimonious models and after further inspection revealed fish of all size-classes frequently selected greater depths (Table 3). Other variables revealed important secondary trends dependent upon season and size-class. During the winter, selection of positions close to cover was a significant factor for age 0, 1, and 2 fish and ages 1 and 2 fish were more frequently found in reaches with warmer temperatures (Table 3). Selection of positions that maximized NEI was an important factor for nearly all size-classes during the spring (Table 3). Positions within slower velocities were a significant factor for smaller size-classes (ages 0, 1, and 2) during the winter and spring, likely due to the higher discharges during these seasons (Table 3). Cross-validation for the best model in each candidate set revealed adequate fit for 12 of 20 best-fit models (Table 4).

Table 3 Effect size of regression coefficients for each resource selection function
Full size table
Table 4 Model validation for the best model identified by model selection for each season by size-class analysis, using the cross-validation approach (10 bins) of Boyce et al. (2002)
Full size table

Energetic use vs availability

Differences in temperature and macroinvertebrate drift among seasons directly affected the estimated metabolic demands of fish in the Middle Fork Rio Ruidoso. Low estimated metabolic rates during the winter months (December–February) led to relatively low energetic needs (Fig. 4). Mirroring stream temperatures and macroinvertebrate drift, NEI rates increased for most size-classes from winter through June (Fig. 4). Warm July water temperatures (15.1 °C) and low drift rates, generated low energetic potential for all size-classes throughout the study area (Fig. 4). Net energy intake rates for all size-classes declined throughout the fall (September–November) in accordance with declining water temperatures (Fig. 4). Greater than 62% of all energy estimated to be acquired by larger size-classes (ages 1+) throughout the year occurred during May and June (Fig. 4). Conversely, age 0 fish acquired much of their potential energy throughout the spring and fall months and just 26% of their yearly energy potential in May and June (Fig. 4a).

Fig. 4
figure 4

Potential net energy intake at locations where fish were observed (“Fish”) and at available habitat (“Available”). Error bars represent standard error for a) age 0, b) age 1, c) age 2, d) age 3, and e) age 4+ fish within the Middle Fork Rio Ruidoso, Mescalero Apache Reservation, New Mexico. Note: no data was collected in August due to high turbidity

Full size image

Throughout the year, fish in the Middle Fork Rio Ruidoso selected stream positions associated with higher energy potential than random stream habitats (Fig. 4). Of those stream positions, 81% of all fish selected locations predicted to meet maintenance rations, while 40% of selected positions were predicted to maximize growth. Winter and early spring were especially important for growth potential of smaller size-classes, where a high proportion of used habitat (>50%) could provide smaller fish (ages 0 and 1) with maximum rations (Fig. 5). In contrast, June was the only month that greater than 50% of larger size-classes (ages 3 and 4+) were predicted to meet maximum rations (Fig. 5). Overall, larger size-classes were energetically limited for much of the year with fewer fish predicted to meet maintenance rations than smaller size-classes. The highest energetic stress occurred in mid-summer (July) for all size-classes, where less than 50% of the population found sites that could meet maintenance rations (Fig. 5). Moreover, no used or available stream locations in July were capable of providing rations that could maximize growth (Fig. 5).

Fig. 5
figure 5

Monthly proportion of fish that met maintenance ration (NEI ≥ 0; black bars) and maximum ration (Cmax; white bars) along with the proportion of available habitat that met maintenance ration (black circles) and maximum growth ration (white circles) for a) age 0, b) age 1, c) age 2, d) age 3, and e) age 4+ fish observed within the Middle Fork Rio Ruidoso, Mescalero Apache Reservation, New Mexico. Note: no data was collected in August due to high turbidity

Full size image

Discussion

This study was designed to provide insight into factors that limit rainbow trout productivity and thus potential productivity of a native cutthroat trout if repatriated within the Middle Fork Rio Ruidoso. Resource selection functions and drift-foraging models were employed to demonstrate the importance of complementary habitat for a range of size-classes and seasons over one year. Results revealed a strong selection for habitat that met energetic and refugia criteria by fish of all size-classes throughout the year. Strong selection of stream depth was especially important for all size-classes, where foraging habitat and other forms of refugia (e.g., thermal) varied by size-class and by season. No doubt, access to suitable foraging habitat was important to rainbow trout making habitat selection decisions as evidenced by fish presence in positions of higher energetic potential than available sites. However, stream depth was the most limiting factor for rainbow trout in the Middle Fork Rio Ruidoso. This was likely due to the extremely low discharge (<0.007 m3/s) within the Middle Fork Rio Ruidoso. Stream depth affords refugia for fish from predation and thermal extremes. In the context of risk versus reward, absence of refugia would result in faster rates of mortality than would sub-optimal growth conditions. If Rio Grande cutthroat trout were repatriated into the Middle Fork Rio Ruidoso, limited refugia in the form of deep pools would result in sub-optimal protection from predation and thermal extremes.

Nearly all the rainbow trout in the Middle Fork Rio Ruidoso selected for deeper habitat than was available throughout the study. This was expected for larger size-classes, where larger salmonids in small streams are commonly found in deeper habitat (Maki-Petäys et al. 1997; Mitchell et al. 1998; Armstrong et al. 2003). However, it was surprising to find smaller size-classes selected deeper habitats in higher proportion than available, because young-of-year salmonids often select shallower habitats in pool margins and riffle/run habitats (Baltz et al. 1991; Bozek and Rahel 1991; Maki-Petäys et al. 1997; Anglin and Grossman 2013). This has been attributed to smaller fishes as inferior competitors, which are often forced into less ideal habitat (Schlosser 1982). In this study, however, size-related habitat partitioning based on stream depth was not observed.

Depth plays an important role in providing thermal refugia for fish during both summer and winter months. Selecting positions in deep pools may shelter fish from super-cooled waters as deeper water may be slightly warmer, thereby avoiding frazil and anchor ice (Huusko et al. 2007). This was especially true within the Middle Fork Rio Ruidoso, where frazil and anchor ice were commonly encountered throughout the winter. Indeed, warmer temperatures and deeper habitats were important habitat characteristics used in higher proportion than available for age 1 and 2 fish in winter. Conversely, coldwater refugia has also been found to be important to trout species during warmer months, where stratified pools provide thermal profiles below upper lethal temperatures (Elliott 2000; Tate et al. 2007). Many studies describe salmonids actively pursuing coldwater refugia during stressful summer temperatures (Keefer et al. 2009; Young et al. 2010; Petty et al. 2012). Interestingly, summer temperatures within the Middle Fork Rio Ruidoso did not approach the thermal limits of rainbow trout; although, age 1 fish used colder temperatures in higher proportion than available during the summer.

Fishes select habitat by balancing the risk of mortality versus the energetic profitability in that habitat (Werner et al. 1983; Gilliam and Fraser 1987). In this study, selection of deeper pools for all combinations of fish sizes and seasons suggests predation risks may be the most influential factor affecting habitat selection by rainbow trout in the Middle Fork Rio Ruidoso. Additionally, smaller size-classes (ages 0, 1, and 2) displayed strong selection for cover during the winter. Similar observations of stream salmonids using greater depths and positioning closer to cover during winter has been documented (Huusko et al. 2007) and may be in response to predator avoidance as swimming performance of salmonids is reduced at low temperatures causing slow acceleration to escape predators (Webb 1978; Johnson et al. 1996).

Stressful thermal conditions in the winter make deeper habitats ideal, but low discharge year-round within the Middle Fork Rio Ruidoso creates shallow riffle/run habitats (<15 cm) where risk of predation can be much higher than in pools (Gregory 1993). In addition to larger fish occupying deeper habitats, small fish similarly concentrated in deeper areas. With the exception of rainbow trout cannibalism, there were no other piscivorous fishes in deeper pools of the Middle Fork Rio Ruidoso (brook trout were incidental). Therefore, the strongest risk of piscivory was from terrestrial predators [specifically osprey (Pandion haliaetus) and American dippers (Cinclus mexicanus) were commonly observed during this study], which has been hypothesized as an influential mechanism forcing fish and other aquatic organisms (e.g., crayfish) into deeper habitats (Power 1987; Englund and Krupa 2000; Conallin et al. 2014; Kemp et al. 2017; Wolff et al. 2016). This likely explains the concentration of all fish size-classes in deeper habitats, where smaller fish would normally be forced into margins.

In addition to serving as important refugia habitat throughout the year, deeper habitats also provide potentially energetically favorable conditions for salmonid growth. Greater depths allow fish to maximize their capture area, resulting in higher rates of prey encounter (Piccolo et al. 2007). Fish selecting greater depths would also have access to slower focal velocities. While slower velocities decrease the prey encounter rate, the benefit of faster velocity is negated as swimming costs become greater and capture distances shrink (Hill and Grossman 1993). Habitat transition zones (e.g., interface between pools and riffles) provide an opportunity to maximize growth potential where fish occupying the head of pools reduce active metabolic costs via swimming while experiencing relatively high rates of prey delivery (Fausch 1984; Nakano 1995). In addition to selecting deeper habitats in higher proportions than available, these habitats had the potential to meet maintenance rations throughout the year. With the exception of July, these habitats could commonly meet maximum rations.

Energetic profitability was greatest from late spring to early summer for age 1+ fish. For rainbow trout, this timing directly follows the spawning season. Post-spawning mortality rates of trout are directly dependent on lipid reserves (Hutchings et al. 1999), suggesting replenishing lipid reserves post-spawning is an important component of rainbow trout life-history in the Middle Fork Rio Ruidoso. Energetic profitability during this period was not as substantial for age 0 fish. This could be due to intercohort competition for foraging, with larger spawning fish switching from spawning habitat to foraging habitat. However, it is more likely that macroinvertebrate cohorts are reaching larger sizes, escaping gape limitations of the smaller fish. Further support for this gape limitation constraint was seen by few available habitats meeting maintenance rations in age 0 fish during this period; in contrast, available habitat for larger fish at this time of the year was most profitable.

Although basic metabolic demand could be met for 81% of the fish throughout the year, the number of available sites that could meet maintenance rations were relatively limited for the largest size-classes (age 3 and 4+). In particular, less than half of available habitat for age 4+ fish had the potential to meet maintenance rations for seven months of the year. This suggests that there are few microhabitats able to support large rainbow trout (>190 mm) in the Middle Fork Rio Ruidoso. This may be due to isolation in cold headwater sites, where higher quality prey is needed to support larger fish. Small, isolated headwater streams often support smaller fish, where access to larger water bodies with a higher diversity in potential prey items (e.g., prey fish) can result in substantial benefits in somatic growth (Leeseberg and Keeley 2014; Petty et al. 2014; Huntsman et al. 2016). The Middle Fork Rio Ruidoso study reach is bounded by barriers at the upper and lower ends, so dispersal to and from habitats outside of the bounded reach was not possible. This not only removes the potential to exploit outside resources, but also subjects fish within the boundaries to intense intraspecific competition. For example, Petty et al. (2014) found brook trout confined to coldwater refugia in headwater tributaries during warm years experienced reduced instantaneous growth rates due to elevated intraspecifc competition. In cool years, however, strong density-dependent competition was alleviated by fish expanding distributions downstream into habitat once inhospitable due to temperature, a concept Petty et al. (2014) referred to as the temperature-productivity squeeze. Similar to observations by Petty et al. (2014) in the warmer years, the Middle Fork Rio Ruidoso also prevented fish from expanding distributions outside of the bounded stream reach, likely increasing the stress of density-dependent competition for foraging resources.

Although the development of drift-foraging models has been underway for more than 30 years (Piccolo et al. 2014), the complexity of mechanisms underpinning the model are difficult to assess. In this study, low macroinvertebrate drift in July led to relatively low estimates of NEI for most fish. Fausch et al. (1997) found that dolly varden (S. malma) switched from drift-feeding to active searching for benthic macroinvertebrates when prey drift rates were artificially reduced. Active search behavior was never observed in rainbow trout within the Middle Fork Rio Ruidoso. A diet analysis, however, may have provided evidence as to whether fish actively foraged on benthic macroinvertebrates. It is likely that if fish were consuming prey from the benthos, the inclusion of a benthic-foraging model in addition to the drift-foraging model may have aided in identifying when fish were more drift- or benthic-foragers (e.g., during low flows) and would have improved NEI estimates.

Fishes require foraging, refugia, and spawning habitat to persist in a dynamic riverscape (Schlosser 1995; Fausch et al. 2002). However, the concentration of such habitats within a riverscape is not always equal, where a limitation of one habitat establishes the population’s carrying capacity. In this study, refugia in the form of depth limited rainbow trout productivity in the Middle Fork Rio Ruidoso. During most seasons and for all but the largest size-classes (ages 3 and 4+), the majority of available foraging positions could meet rainbow trout maintenance rations. However, strong selection for deep pools, limited size-class partitioning of habitat based on depth, and observations of high densities of piscivorous terrestrial predators near the stream (e.g., osprey) indicate rainbow trout persistence was strongly dependent on acquiring refugia. Given that rainbow trout and cutthroat trout are phylogenetically and ecologically similar (Seiler and Keeley 2009), these results provided important insight on the successful re-introduction of Rio Grande cutthroat trout in the Rio Ruidoso. Microhabitat studies of various lotic cutthroat trout populations suggest a similar preference for deep pool habitats with slower moving water (Bozek and Rahel 1991; Heggenes et al. 1991; Spangler and Scarnecchia 2001). Although rainbow trout can tolerate slightly higher temperatures than cutthroat trout, temperature specific growth curves of both species are comparable and reflect similar energetic requirements (Bear et al. 2007; Zeigler et al. 2013). Given the similarities between the species, it is likely that repatriation of native cutthroat trout would be successful. However, due to the risk of hybridization and interspecific competition, repatriation of Rio Grande cutthroat trout should only occur after extirpation of the non-native fish.