The importance of ice cover for energy turnover in juvenile Atlantic salmon


Anders G. Finstad, Norwegian Institute for Nature Research, Tungasletta 2, NO-7485 Trondheim, Norway. E-mail:


  • 1Under benign laboratory tank conditions we compared food consumption and metabolism of Atlantic salmon (Salmo salar) juveniles exposed to simulated ice cover (darkness) with fish in natural short, 6 h light, day length (without ice). Three different populations along an ice-cover gradient were tested (59°N−70°N).
  • 2Resting metabolism was on average 30% lower under simulated ice cover (6·6 J g−1 day−1) than under natural day length (9·4 J g−1 day−1), and the response was similar for all populations. Northern salmon grew equally well in dark and light conditions, whereas the southern grew significantly poorer in the dark. Fish from all populations fed more under natural day length than in the dark and the northern population had higher consumption than the southern. The relative high growth of fish from the northern population in the dark compared to the southern populations was due partly to higher consumption and partly to higher growth efficiency. Fish from the southern populations had negative growth efficiency in the dark.
  • 3We also studied the importance of ice cover under more hostile conditions in stream channels using the northern population only. Juveniles held in channels with simulated ice cover lost less energy (20 J g−1 day−1) than those held in channels with transparent cover (26 J g−1 day−1). This difference in energy loss was due partly (50%) to higher food consumption under simulated ice (4·5 and 1·6 J g−1, respectively) and partly (30%) to light-induced differences in resting metabolic rate.
  • 4In conclusion, both experiments showed lower metabolic costs in darkness under simulated ice cover than without ice. Under benign laboratory conditions the response to light (ice cover) varied among populations and only the northern population were able to attain positive growth in the dark. Under semi-natural conditions the lack of ice cover induced strong negative effects on the energy budget. Because energetic deficiencies are assumed to be an important cause of winter mortality, our study indicates that ice break-ups or removal following climatic change may affect winter survival significantly, particularly in northern populations.


Ecological processes are influenced by prevailing climatic conditions and there is now ample evidence that recent climatic changes influence ecological systems (Stenseth et al. 2002; Walther et al. 2002). Models of global warming predict further increase in winter temperatures on the northern hemisphere. Changes in temperature may affect physiological (metabolic and reproductive) processes directly, particularly among ectothermic animals. However, changes in environmental conditions are not predictable simply in terms of temperature responses. Other indirect effects of temperature increases may also affect biological processes severely. Ice conditions are particularly sensitive to small thermal changes, as minor changes in water temperature may induce freezing or thawing. In terrestrial ecosystems, mild winters leading to ice crust formation are known to restrict access to food resources and led to increased mortality due to starvation in Arctic ungulates (Forchhammer et al. 2002 and Solberg et al. 2001) and to decreased winter survival in tundra voles (Aars & Ims 2002). In temperate aquatic ecosystems variation in ice-cover conditions may have large impacts on both individual organisms and ecosystems (e.g. Adrian et al. 1999; Quayle et al. 2002). A large body of evidence derived both from historical trends and model predictions now points towards a future decline in ice cover of aquatic habitats in the northern hemisphere (e.g. Magnuson et al. 2000; Blenckner, Omstedt & Rummukainen 2002; Yoo & D’Odorico 2002; Assel, Cronk & Norton 2003).

The winter period is regarded commonly as a survival bottleneck for temperate freshwater organisms (Cunjak & Therrien 1998; Cunjak, Prowse & Parrish 1998), and in temperate fish, energy deficiency is one of the suggested main sources of winter mortality (e.g. Johnson & Evans 1991; Miranda & Hubbard 1994; Cunjak et al. 1998). Removal of ice cover may greatly increase fish vulnerability towards predators, particularly endothermic animals. This may shift the trade-off between feeding and risk of predation (Metcalfe & Thorpe 1992; Bull, Metcalfe & Mangel 1996; Metcalfe, Fraser & Burns 1999), causing poorer energy status (due to reduced food consumption), increased predation rates, or both. Further, the potential negative effects on the energy budget may be escalated due to increased metabolism caused by the effect of light on melatonin secretion mediated by the pineal gland (Wendelaar Bonga 1993). Thus, small changes in winter temperature may also have large impact on aquatic animals through their influence on ice conditions.

Atlantic salmon (Salmo salar L.) commonly spends from 2 to 5 years in freshwater prior to migrating to sea. The species is distributed over a wide geographical area along the east and west coast of the North Atlantic Ocean, occupying a diverse array of freshwater habitats (Elliott et al. 1998). In the northern areas of its distribution, winter constitutes large parts of the year and rivers are commonly ice-covered. In combination with snow cover, this creates a major contrast in habitat characteristics to summer conditions, particularly with regard to light. In contrast, southern populations are found in rivers which only rarely, if ever, are ice-covered during the winter.

Here, we test for consequences of altered winter ice conditions on the energy budget of juvenile Atlantic salmon. Metabolism, food consumption and energy balance were studied for fish held in laboratory conditions and fed ad libitum while manipulating light conditions simulating habitats with or without ice cover. The juvenile salmon were first-generation hatchery-reared fish from three populations originating from rivers differing in winter ice-cover conditions (59°N−70°N). Thus, we also explored the potential for adaptive variation in responses to altered ice-cover conditions. Finally, for fish from the northernmost population, we also tested for energetic consequences of ice-cover removal in semi-natural stream channels.

Materials and methods

experimental site and fish

All experiments were conducted at the NINA Research Station, Ims, in south-western Norway (59°N, 6°E) from January to March 2003. The fish used in the experiments were one summer old (0 +) first-generation hatchery-reared Atlantic salmon with parents originating from the Rivers Alta (70°N, 23°E), Suldal (59°N, 5°E) and Imsa (59°N, 6°E). The populations were selected to include fish from one cold and extensively ice-covered river, from one cold but usually not ice-covered river and from one warmer river without ice cover. The northern River Alta is ice-covered for 6–7 months (mean annual temperature 3·9 °C). In contrast, the River Imsa (mean annual temperature 8·6 °C) and the glacier-fed River Suldal (mean annual temperature 6·5 °C) are never (River Imsa) or only sporadically (River Suldal) ice-covered during the winter period. For each laboratory cohort, eggs were collected from a minimum of five females, and fertilized with one male per female. Prior to winter acclimatization the fish were maintained at the same ambient temperatures and light regimen. Two months before experimental start-up, the fish were held at gradually declining temperatures and day lengths (≈1 °C and 1 h per week) to simulate the natural autumn conditions, until the final temperature (0·9 °C) and photoperiod (6 h light and 18 h dark) were reached.

Temperature was maintained at ≈ 1 °C (mean 1·3 ± 0·5 SD) during the metabolic measurements and the common environment experiments. In natural habitats, ice cover in combination with snow considerably reduces perceived light levels during wintertime. In order to simulate light conditions experienced by fish under ice cover and without ice during winter, two similar tank compartments were created and light isolated as in a photo laboratory darkroom. The no-ice treatment received 6 h light (≈ 70 lux) and 18 h dark, while the ice-cover condition was simulated by rearing the fish in darkness. Light levels were measured both in the ice treatment (laboratory) and under the ice cover of the River Alta during wintertime (LI-COR Inc., LI-250 Light Meter with LI-190SA Quantum Sensor or LI-192SA Underwater Quantum Sensor). In springtime (22 April 2001), light levels under ice cover (at 50 cm water depth) in the River Alta varied from 6·3 µmol s−1 m−2 (day) to < 0·01 µmol s−1 m−2 (night). Daily averages are expected to resemble night-time values for most of the winter due to the polar night. Light intensity in the laboratory dark treatment was < 0·01 µmol s−1 m−2. Husbandry as well as operation of the respirometer in the ice treatment was conducted using red photo-laboratory darkroom light which is expected to be physiologically undetectable for salmonids (Ali 1961; Dodt 1963). Both light regimens received the same inflow water and the temperature was therefore identical in the two treatments.


Resting metabolic rates (Jobling 1994) were measured as oxygen consumption rates of unfed, inactive fish following largely the experimental setups of, for example, McCarthy (2000), O’Connor, Taylor & Metcalfe (2000) and Cutts, Metcalfe & Taylor (2002). However, a closed system was chosen because of the low temperature maintained in the present experiment and corresponding small difference between inflow and outflow water. The respirometry system consisted of a rack of 16 parallel acrylic chambers (180 mm length, 200 mm diameter), each connected with a small pump allowing recycling of the water within each chamber. The volume of the system (pump and tubes included) was approximately 360 mL. This system allowed the measurement of five individuals from each population as well as one blind chamber (control) simultaneously.

Prior to metabolic measurements, the fish were maintained in the experimental tanks for 6 weeks and deprived of food for the last ≈ 48 h. The fish were introduced into the chambers the day before the start of the experiment. Measurements of oxygen consumption were started the following morning and three recordings were conducted on each fish with at least 1 h break between. The oxygen concentrations in the chambers was recorded (OxyGuard Handy Mk III) at the beginning and the end of each trial (approximately 45 min) and it was not allowed to drop more than c. 25%. Volume and wet mass of the fish were recorded. Oxygen consumption was calculated as the decrease in oxygen concentration with time (mg l−1 h−1). The oxycalorific coefficient was set to 13·59 kJ g−1 (Jobling 1994). For each experimental treatment, at least 15 fish from each population were measured. To ensure that only inactive fish were included, the 10 fish with the lowest oxygen consumption were used for further analyses. Mean wet mass (± SD) of the fish included in the analyses was 16·10 (5·99), 18·58 (2·89) and 17·07 (2·78) g for the Rivers Alta, Suldal and Imsa, respectively.

growth and food consumption in tank environment

Using 12 tanks, two replicates in each of the two light regimens were run for each of the three populations. The tanks were 45 × 45 cm and 60 cm deep, had a water flow of 2 L per min and a water level of 30 cm. Experimental units were distributed randomly within each section to avoid systematic tank effects. Oxygen saturation was always close to 100% during the experiment. Mean size (± SD) at the start of the experiment was 17·24 (4·86), 19·02 (3·37), 17·44 (3·37) g wet mass for the Rivers Alta, Suldal and Imsa populations, respectively.

Ten individually marked (Alcian blue in fins and adipose fin clipping) salmon were used in each tank. Each fish was weighed after approximately 48 h of starvation at the beginning and the end of the experiment. Fish that died during the experiment were replaced by similarly sized fish to maintain densities, but the replacement fish were not included in the results. The fish were fed to satiation with CsCl enriched granulated fish food (Felleskjøpet, Sandnes, Norway, Cs concentration: 14·1 p.p.m. fresh mass) administered from automatic feeders. The experiments lasted for 48 days.

Cs enriched food was used to estimate food consumption in accordance with Forseth et al. (2001). The estimate of food consumption is based on estimating the intake of caesium from an observed change in caesium body burden with time. Based on known rates of assimilation and elimination, the food consumption is obtained by dividing the caesium intake by the concentration in food (Forseth et al. 1992; Forseth et al. 2001). Food consumption was estimated for a random subsample of five individuals from each experimental unit (in total 10 fish per population and treatment). These data were used when testing for population and light interaction effects on food consumption and growth efficiency. When comparing light and dark treatments within population, the Cs body burdens in fish not analysed were estimated using the relationship between Cs content and growth rate in order to increase the sample size and raise the test power. Further description of methods and calculations as well as evaluation of the method can be found in Forseth et al. (2001) and Jonsson et al. (2001).

Specific energy was determined from proximate composition of lipid and proteins determined by extraction and controlled combustion (e.g. Berg & Bremset 1998) with the exception of the individuals measured for Cs content, where energy was calculated from dry mass percentage (Hartman & Brandt 1995) obtained from the proximate analyses. Carbohydrates are a minor body component of teleost fishes (0·1–0·5% on wet mass-basis) (Craig 1977), and were ignored in the present analyses. Proteins were assigned a caloric value of 24·0 kJ g−1 and lipids a value of 39·0 kJ g−1 (Jobling 1994).

Growth was measured as the standardized mass-specific growth rate (Ω%) (Ostrovsky 1995)

image(eqn 1)

where M0 and Mt are the respective body mass (g) at the beginning and end of each experiment, t is the experimental period (days) and b is the allometric mass exponent for the relation between specific growth rate and body mass fixed at 0·31 (Elliott & Hurley 1997). Growth efficiency was calculated in energy terms as the change in energy content of the fish during 1 day divided by the daily food consumption. The energy value of the food pellets was 21·6 kJ−1 g.

The experimental setup was planned as a nested anova with tank (replicates) nested within population and light treatment. In accordance with Underwood (1997), post-hoc pooling of replicates (removal of the tank factor from the model) were conducted for α larger than 0·25.

ice cover in semi-natural habitat

To investigate the effect of ice cover on the energy use and feeding under conditions more similar to the natural environment experienced by wild juvenile salmon during winter, we used outdoor semi-natural stream channels. The experimental set-up consisted of 10 channels (485 × 50 cm), with water depth of approximately 30 cm and gravel substrate. Every second channel was covered with light impermeable plywood and the remaining with clear plastic. Clear plastic was used to prevent drift of exogenous material into the system while allowing natural daylight. Daylengths increased from ≈ 6 h at the start of the experiment to ≈ 9 h at the end. The outlets of the channels were blocked by a screen to prevent escapes and were cleaned daily. Mean temperature was 2·2 °C (± 0·75 SD) during the experimental period. Water was run in the channels for 10 weeks prior to the experiment to permit colonization of an invertebrate fauna. In addition, drifting invertebrates were introduced naturally with the supply water from a nearby lake throughout the experiment.

At the beginning of the experiment, 10 individually marked (Alcian blue in fins and adipose fin clipping) and weighed (precision: ±0·01 g) salmon from the River Alta population were released into each channel. Mean wet mass (± SD) at the start of the experiment was 15·57 (± 5·43) g. After 72 days, the channels were drained and the fish collected, killed, weighed and stomach content was removed. The fish were stored at −24 °C before specific energy was determined as described above. Mass specific growth rate was determined according to eqn 1.

Stomach contents were identified to various taxa, dried at 55 °C for 1 week and weighed (precision 0·0001 g). Daily food consumption (D) was estimated by the method of Eggers (1979):

D = 24MR(eqn 2)

where M is the mean mass of the stomach content and R is the gastric evacuation rate, estimated for brown trout (S. trutta L.) (Elliott 1972). The stomach content was given a caloric value of 18 kJ g−1 dry mass (grand mean of aquatic invertebrates; Cummings & Wuycheck 1971). Due to low gastric evacuation rates, and hence less relative daily fluctuations in stomach content, the number of daily samples needed to give reliable estimates of daily food consumption will decrease considerably with decreasing temperatures. At the low ambient temperatures observed in the present experiment (≈ 2 °C), one daily sample is sufficient to give a reliable estimate (A. G. Finstad, unpublished material).

The experimental setup was planned as a nested anova with individual channels (replicate) nested within ice cover. This yielded five replicates (channels) within each treatment. Post-hoc pooling of replicates was conducted in accordance with Underwood (1997). Some fish died during the study (16 of 100 fish) and to obtain a balanced design missing observations were replaced with a dummy variable set to the mean of observations within the given replication unit (Underwood 1997).



Overall, the resting metabolic rate of fish reared in darkness (simulated ice cover) was 30% lower than fish reared in 6 h daylight (Fig. 1). An initial ancova model, which in addition to light treatment and population as factors also included ln transformed body mass as covariate, did not reveal any significant mass effect on the resting metabolism (F1,25 = 0·041, P = 0·839). In lieu of the homogeneous size group selected for the present experiment, the covariate body mass was therefore excluded from the final analyses. Although the resting metabolic rate differed significantly between the two light regimens, no significant population or population by light interaction was apparent (anova: light, F1,24 = 5·46, P = 0·028; population, F2,24 = 1·15, P = 0·212; population × light, F2,24 = 0·56, P = 0·575). However, the corresponding test power (assuming observed variance and effect size and α = 0·05) was low (population, 1 − β = 0·116; population × light, 1 −β = 0·116). It is therefore not possible to interpret these results as evidence for the lack of population differences in resting metabolism or as lack of population differences in response to the experimental treatment.

Figure 1.

Mean (± SE) resting metabolism (J g−1 d−1) for the River Alta, Suldal and Imsa populations reared in darkness (black bars) or 6 h light (open bars) at 1·3 °C.

growth and food consumption in tank experiments

Mean mass-specific growth rates differed both between populations and light treatments (anova: light, F1,114= 28·17, P < 0·001; population, F2,114 = 26·11, P < 0·001; population × light, F2,114 = 4·65, P = 0·011) (Fig. 2a). Initially, replicates (tanks) were entered as a nested factor in the anova model, but was removed when not found significant at α = 0·25 (F1,114 = 0·44, P = 0·506). As indicated by the significant treatment by population interaction, the populations differed in their response to light regimen. Mean growth rates were higher in the 6-h light treatment than in the dark for fish from the two southern Rivers Imsa and Suldal populations (Fig. 2a). However, for the northern River Alta salmon there were no significant differences in growth between the two treatments (Fig. 2a). Further, the River Alta fish were the only ones that obtained significant positive mean growth rates under the ice treatment (Fig. 2a). In contrast, all populations maintained positive growth rates in the no-ice treatment.

Figure 2.

Mean (± SE) mass-specific growth (a), food consumption (b) and growth efficiency (c) for the River Alta, Suldal and Imsa populations reared in darkness (black bars) or in 6 h daylight (open bars). P-values (t-test, two sided) are given for comparison between light regimens within populations (N = 40).

Food consumption varied significantly both between populations and light treatments (light, F1,56 = 96·47, P < 0·001; population, F2,56 = 16·21, P < 0·001; population × light, F2,56 = 3·55, P = 0·035) and food consumption was significantly lower (≈ 10–40%) in the dark than in the light treatment for all populations (Fig. 2b). As indicated by the significant treatment by population interaction, the populations differed in their response to light regimen. Growth efficiency varied accordingly (Fig. 2c) and differed between the dark and light treatments for all populations, but in different directions (light, F1,56 = 20·57, P < 0·001; population, F2,56 = 10·16, P < 0·001; population × light, F2,56 = 9·27, P < 0·001). Whereas the northern River Alta salmon maintained higher growth efficiency in the dark than in the light treatment, there was a clear negative effect of the dark treatment on growth efficiency in the southern River Suldal and River Imsa populations (Fig. 2c).

energy loss in semi-natural environment

All fish reared in the semi-natural stream channels lost mass during the experimental period resulting in negative mass-specific growth rates. However, there was a clear effect of the presence or absence of ice cover on the mass loss as mass-specific growth rate in the dark channels (−0·80 ± 0·04 SE) was significantly higher (t81 = 2·10, P = 0·038), on average 20%, than in the channels with clear plastic cover (−0·99 ± 0·05 SE). The effect of replications (channels) did not prove significant at α = 0·25 (F4,93 = 0·610, P = 0·656) and was removed from the anova model before testing for ice treatment effect on mass-specific growth.

The differences in the loss of mass-specific energy were similar to the observed differences in mass loss and was on average 23% higher in the no-ice treatment than in the ice treatment (mean ± SE: 26·5 ± 1·3 and 20·2 ± 1·0 J g−1 day−1, respectively). The differences in energy loss between the two treatments were highly significant (t81 = 3·892, P < 0·001). As for mass-specific growth rates, the effect of replication (channel) was removed from further analyses when found not significant at α = 0·25 (F4,93 = 0·562, P = 0·691). No significant correlation between wet mass and specific energy loss was detected (r = 0·096, P = 0·386).

The stomach content consisted mainly of Trichoptera sp. as well as various unidentifiable insect parts. Both the high frequency of empty stomachs (50–75%), and an overall high degree of digestion of the stomach content which made accurate identification of taxa difficult, indicated a low frequency of feeding. Estimated average food consumption from stomach content was higher for juveniles reared under simulated ice cover than for fish reared under clear cover (mean ± SE: 4·5 ± 1·2 and 1·6 ± 0·7 J g−1 day−1, respectively). A significantly higher (χ = 4·30, P = 0·038) proportion of fish was found with stomach content in the ice-cover treatment (50%) compared to the clear cover treatment (25%) but mass-specific dry weight of stomach content (excluding empty stomachs) did not differ significantly between the two groups (t25 = 0·851, P = 0·403).


This study provides evidence for significant effects of ice cover on the energy budget of juvenile Atlantic salmon. The energy balance of the fish was affected both directly through light-induced physiological changes in metabolic rate and indirectly through behavioural differences influencing food consumption and activity metabolism. However, ice-cover treatment responses differed between populations. The direction of these responses concurred with predictions of adaptations to local variations in ice-cover conditions. The River Alta population experience up to 6 months of ice covered by snow, which results in very low light levels. In contrast the Rivers Suldal and Imsa have very few or no periods of ice-cover formation during winter. Only fish from the River Alta were able to maintain a significant mean positive growth rate under the ice treatment in the tank experiments. However, the adaptive significance of these findings is not conclusive. Low population replication makes the population effect potentially confounded by variation in other local environmental factors.

Temperate zone fishes such as Atlantic salmon normally deplete their energy resources during winter (e.g. Cunjak 1988; Berg & Bremset 1998; Hutchings et al. 1999; Finstad, Berg & Lohrmann 2003). In contrast to observations from the benign laboratory conditions, a net depletion of storage energy was observed for fish reared in the semi-natural stream channels. This discrepancy in the energy balance between the two rearing environments probably reflect differences in food type and feeding conditions as the fish in the tank environment where fed ad libitum, and had a higher energy intake than fish in the stream channels which relied on natural food items. Fish reared in channels with simulated ice cover lost on average 6 J g−1 day−1 less than fish reared in stream channels without ice cover. As revealed by the laboratory experiments, 30% (2 J g−1 day−1) of this difference in energy loss could be accounted for by differences in resting metabolism. Further, the estimated energy intake for the two treatments differed and another 50% (3 J g−1 day−1) of the difference in energy loss could be accounted for by differences in energy intake. The last 20% variation in energy loss between ice treatments are due accordingly to differences in other metabolic processes such as specific dynamic action (feeding metabolism) and activity metabolism. This latter figure is necessarily uncertain, as errors from both estimates of food consumption and resting metabolism accumulates in the residual part of the energy budget. However, the basic findings of a reduction in energy turnover costs with simulated ice cover for the River Alta salmon correspond with findings from the tank experiments.

The observed higher food consumption for fish reared in stream channels with simulated ice cover relative to fish reared in natural day length appear to contrast with the results from the tank experiments. In the tank environment, the fish decreased its energy intake in the ice treatment irrespective of population origin. This contrast is probably caused by a change in the fish's interpretation of the foraging to predation risk trade-off in the different rearing habitats combined with difference in feeding conditions. Fraser & Metcalfe (1997) demonstrated that night time foraging efficiency in juvenile Atlantic salmon may be less than 35% of their daytime efficiency. The lower energy intake in the ice treatment of the tank environment is therefore likely to be a direct result of reduced foraging efficiency in the dark. It is to be expected that the indoor tank environment, offering protection from all disturbance except from husbandry, is perceived as less risky than the semi-natural stream channels placed outdoors. Although the stream channels in the no-ice treatment were covered with clear plastic, this did not shield the fish from observing potential predators visually. Metcalfe et al. (1999) suggested that activity patterns in juvenile salmon is a result of complex trade-offs between food availability, food capture efficiency and predation risk. The fact that the reduced energy intake in the no-ice treatment of the stream channels resulted mainly from a lower frequency of feeding (higher occurrence of empty stomachs) support further that behavioural constraints were involved in creating the observed food consumption differences. It has been shown previously that juvenile salmon make fine-tuned adjustments to predation risk, responding to an increased risk by reducing their feeding activity (Dill & Fraser 1984; Metcalfe, Huntingford & Thorpe 1987; Angradi 1992). A functional explanation for a large part of the higher energy loss in stream channels without ice cover is therefore an increase in apparent predation risk followed by reduced time spent foraging.

Climate–ecology correlatives are now increasingly common and provide ample evidence that recent climatic changes influence ecological processes (e.g. Ottersen et al. 2001; Stenseth et al. 2002; Root et al. 2003). However, while these correlations are important first steps, mechanistically based explanations have to be provided in order to make inferences regarding underlying functional relationships and useful predictions. Ice covering of aquatic habitats is sensitive to changes in climatic variables (e.g. Magnuson et al. 2000) and variation in ice cover may influence strongly ambient light conditions as well as the vulnerability of the fish to endothermic predators. Both these factors will affect the feeding to predation risk trade-offs (sensuMetcalfe et al. 1999). In addition, the negative effect of removed ice cover on the energy balance of the fish is reinforced by light-induced increases in metabolism. It is evident that indirect effects of temperature changes through removed ice cover and altered habitat light regimens affect the energy budget significantly and accordingly may reduce survival of northern salmonids during a bottleneck period. Similar effects may also be evident in other freshwater ectothermic animals.


Sigurd Einum is acknowledged for comments on the manuscript. We also thank the staff at NINA Research Station for valuable assistance during the experiments and Anne Lohrmann for assistance with the proximate analyses. The study was supported financially by the Norwegian Research Council (NFR grant no. 145208/210), Statkraft SF, Norwegian Electricity Industry Association, the Norwegian Directorate for Nature Management and the Norwegian Institute for Nature Research.