Storing food is an evolutionary strategy that allows many species of animals to overcome times of food scarcity (Andersson and Krebs 1978, Smith and Reichman 1984). Food is stored either short-term (days/weeks) or long-term (weeks/ months) and species may store small amounts in many locations (scatter hoarding) or larger amounts in one or two locations (larder hoarding) (Vander Wall 1990). While food stores can allow individuals to minimize predation risks while foraging or invest in other behaviors, food storage is often related to variable environmental conditions that either reduce the amount of available food, limit access to food or both (Clarkson et al. 1986, Vander Wall 1990, Steele et al. 2006, Delgado et al. 2014, Jakopak et al. 2017, Pitera et al. 2018).

The ability to collect and store food at one time and use it at another time is an important evolutionary adaptation and allows species to successfully cope with changing food supply and energy demands. Animals living in temperate climates experience periods of low plant productivity and freezing temperatures during winter and may store food to help meet nutritional demands. For example, the honeybee Apis mellifera in the northern part of its range stores up to 40% of the hive honey production for use in winter (Seeley 1985), and red squirrels Tamiasciurus hudsonicus make caches of conifer cones that are necessary for successful winter survival (Rusch and Reeder 1978).

The two extant species of beavers, the Eurasian beaver, Castor fiber, and the North American beaver, C. canadensis, are found across a large latitudinal range in the Holarctic biogeographical region occurring in most types of wetland habitats including large and small lakes, ponds, large and small rivers, streams, and in human modified systems such as agricultural drainage and irrigation ditches and fish ponds (Busher and Hartman 2001, Rosell et al. 2005, Sjöberg and Ball 2011, Busher 2016). Beavers in alpine habitats and at higher latitudes face periods of food scarcity during winter when plant productivity is low and mobility can be restricted by snow and ice. Under these conditions beavers store branches of woody species in a food cache usually placed in water outside the lodge or burrow where the beaver family will overwinter (Jenkins and Busher 1979, Busher 1991, 1996, 2003, Hartman and Axelsson 2004). The food cache along with fat storage (Aleksiuk 1970, Smith and Jenkins 1997) reduced activity (Aleksiuk and Cowan 1969, Potvin and Bovet 1975, Lancia et al. 1982), construction behaviors and social behaviors (Semyonoff 1953, Stephenson 1969, Buech et al. 1989) allows the beaver family to meet metabolic demands and overcome reduced food availability during the winter.

Beaver food caching is facultative (Wilsson 1971, Lavsund 1989, Osmundson and Buskirk 1993, Hartman and Axelsson 2004) depending on altitude, latitude and climatic regime. For example, in North America, beavers in the southeastern United States (below approximately 35 degrees N latitude) do not generally build food caches (Swafford 2002, Baker and Hill 2003), while farther north (above approximately 38 degrees N latitude) and at higher elevations they usually build caches (Hodgdon 1978, Busher 1980, 2016). The pattern in Europe, including European Russia, may be less consistent with beavers reported to show more variation in cache construction behavior (Wilsson 1971, Djoshkin and Safonov 1972, Hartman and Axelsson 2004). As an illustration of the facultative nature of cache construction beavers on the Rhone river typically do not cache food, but when a pair from the Rhone was translocated to Switzerland, they did build a food cache (Blanchet 1959). The construction of the winter food cache is related to other construction behaviors such as dam and lodge building and has an innate component (Richard 1967, 1983).

In North America aspects of woody species selection (Busher 1991, 1996), general caching behavior (Hodgdon 1978, Slough 1978, Busher 2003) and the relationship between family group size and food cache size (Kafcus 1987, Easter-Pilcher 1990, Osmundson and Buskirk 1993) have been investigated. Additionally, there is evidence that the North American species increases caching behavior as environmental temperature declines during autumn and become less selective in species stored in the cache as autumn progresses (Hodgdon 1978, Busher 1996, 2003).

In Eurasia, Dzięciołowski and Misiukiewicz (2002) documented the woody species found in food caches in Poland, Hartman and Axelsson (2004) examined the prevalence of caches and physical parameters of the location of the caches in Sweden, and Vasin (2001) in Siberia, documented the volume stored per family member. Wilsson (1971) studied food caching by captive beavers in enclosures and beavers living free along a river in northern Sweden. He considered food storage to be associated with increased tree felling and suggested that both are potentially stimulated by a decline in temperature.

However, for the Eurasian species, less is known about the temporal and spatial development of the food cache, response to potential environmental stimuli, and any relationship between food cache size and family size.

We investigated the physical characteristics, spatial and temporal development of food caches at sites in northern Europe to better understand this evolutionarily important behavior. Our primary goal was to compare food caching behavior between populations in three European countries living in a variety of wetland habitats, something that had not been done before. We examined water depth at food caches, initial cache size and date, temporal growth of caches and the potential relationship between cache initiation and cache size and ambient temperature.

Moreover, knowledge of group size, age of the adult pair, duration of territory occupancy and relative water flow along three rivers in Norway allowed us to examine the relationship between food cache size and these variables. At the Norway sites, because accurate family sizes were known, we were able to test the hypothesis that families with more individuals build larger food caches since a larger food cache should be able to support more individuals during the energetically stressful winter period of confinement. Additionally, if a positive relationship between food cache size and family size exists then food cache size could be used to estimate population size. Tests of this hypothesis have been equivocal in North America with Kafcus (1987) and Easter-Pilcher (1990) reporting significant relationships between food cache size and family size while Osmundson and Buskirk (1993) reported no significant relationship. Our study is the first to test this hypothesis in the Eurasian beaver.

Methods

Data were collected at study areas in Sweden, Norway and Lithuania during autumn 2015 (Fig. 1) and again in Norway during autumn 2016. The Swedish sites were located near Uppsala and at the Grimsö Wildlife Area (latitude of both = 59 degrees) along small streams, a mid-sized river and a small lake. The Swedish sites were first visited 17–23 October (late week 42/week 43) and weekly visits were made through 17 November (week 47).

In Norway (latitude = 59 degrees) active sites were along three midsized river systems; Lunde on the Straumen river, Gvarv on the Gvarv River and Patmos on the Sauar river. This beaver population is part of a multi-dimensional, longitudinal research program (Rosell et al. 1998, Rosell and Hovde 2001, Mayer et. al. 2017a). At this study area family group size, individual age and length of territory occupancy were known due to annual intensive live capture and marking activities (Mayer et al. 2017b). In 2015, initial observations were made between 31 October and 3 November (week 44/45). These sites were revisited 11–18 December (week 50/51). In 2016 data were collected on 22 October (week 42; initial visit) and 13–25 November (week 46/47; revisit).

In Lithuania (latitude = 54 degrees) the active sites were usually associated with soviet era drainage ditches abutting agricultural land, or natural depressions in the morainic landscape. Many sites had well developed ponds, although others were essentially small water impoundments in marshy areas. These sites have been included in a number of long-term studies on beaver ecology (Ulevičius 1999, Ulevičius et al. 2009). Data were collected 9–13 November (week 46).

Figure 1.

Beaver lodge with a large food cache from Bø, Norway (a), and a map illustrating the locations of the research sites (b).

Vegetation at all areas was mixed, with regions of Salix spp., Alnus spp. and Populus spp. Other species such as bird cherry Prunus padus, common ash Fraxinus excelsior, rowan Sorbus aucuparia, birch Betula spp., Norway spruce Picea abies and Scots pine Pinus sylvestris were present in smaller amounts. Woody species composition in caches was determined by visual examination of branches. Dominant species in the cache was estimated by either visual means or taking subsamples of branches. All branches removed for sampling were returned to the cache.

Active sites were identified by the presence of fresh mud on the lodge, fresh cutting in the area or the presence of a food cache. In 2015, cache volume (m³) was calculated by measuring the length, width and depth of each cache using a calibrated avalanche probe. Depth was measured in multiple areas of the cache and averaged. In Norway, only cache area (length × width) was measured in 2016 so any comparisons between 2015 and 2016 uses area not volume for cache size.

Temperature data were obtained from the Norwegian Meteorological Institute (< www.eklima.met.no>), the Swedish Meteorological Hydrological Institute (< www.smhi.se>) and other international weather databases.

In addition to descriptive statistics we used ANOVA, Tukey's HSD test, Pearson's correlation coefficient and the Mann–Whitney U test to examine relationships between variables.

In Norway, to estimate variation in final food cache area (length × width at last measurement) within and between active sites, we calculated the standard deviation (SD) of food cache area between 2015 and 2016 separately for each individual active site (same family group) and combined for all active sites. Finally, we investigated the effect of water flow, family group size and duration of territory occupancy on final food cache area (dependent variable; log-transformed to meet the assumption of residual normality) using a linear mixed-effects models of the R package ‘lme4' (Bates et al. 2015). We ran three separate analyses to avoid overfitting models. The first model included the water flow (categorized as ‘none', ‘slow’, ‘medium’ and ‘fast') and year (2015 versus 2016), because we measured food cache area at slightly different times between years). We did not include water depth in this model because it was only measured in autumn 2015. The second model included family group size and the third model included the duration of territory occupancy (in years; averaged for the male and female pair member). Duration of territory occupancy and the average age of the beaver pair were highly correlated (r = 0.91, p < 0.001), so we did not include pair age in the analyses. Site ID was included as random intercept to account for multiple observations. An alpha of 0.05 was used as the cutoff for significance, and 0.1 as cutoff for a trend.

Results

Number of caches and mean volumes

In Sweden, eight active sites were investigated beginning 17–20 October. Beavers at four sites (50%) had caches at this time (two near Uppsala and two at Grimsö), while two sites had caches by 2–8 November and a final site had a cache by 17 November (week 47). Mean initial cache volume in Sweden (n = 7) was 10.75 m³ (range 0.5–23.6 m³, SD = 7.71). By week 47 seven of eight active sites (87.5%) had food caches and the mean volume was 41.47 m³ (SD = 20.13). Sample sizes are small (3 at Uppsala and 4 at Grimsö) making statistical comparison difficult. However, both initial (Uppsala = 12.93 m³, Grimsö = 9.11 m³) and final (Uppsala = 46.03 m³, Grimsö = 38.05 m³⁾) mean cache volumes were larger at Uppsala sites than at Grimsö sites. The number of woody species in the caches ranged from one to three (Table 1).

Table 1.

Descriptive information on active sites including volume at time of initial visit, number of and dominant woody species in the food cache and mean water depth of the food cache. nc indicates no cache was present on the date of first visit.

In Norway 19 sites were visited between 31 October and 3 November (week 44) and 13 (68%) had food caches. When sampled again in December, 17 of 19 sites (89%) had caches. However, two sites could not be visited due to ice so this is a minimum estimate. Initial cache sizes ranged from 1.35 m³ to 49.45 m³ with a mean of 17.06 m³ (SD = 15.58) (Table 1). Food caches at Lunde (n = 5) were the smallest (mean = 8.48 m³, SD = 5.26), while caches at Patmos (n = 4, mean = 19.95 m³, SD = 20.01) and Gvarv (n = 4, mean = 24.90 m³, SD = 17.82) were larger. Twelve of the sites with caches in November were sampled again in December and the pattern was similar with Lunde the smallest (mean = 23.8 m³ SD = 17.8) and Patmos (mean = 42.2 m³, SD = 47.6) and Gvarv (mean = 47.6 m³, SD = 30.7) larger. The number of woody species in caches ranged from 1 to 4 (Table 1). When comparing food cache area between 2015 and 2016, the variation within the same individual active sites was 3.7-fold lower (SD = 7.7) compared to the variation across all active sites (SD = 28.2).

Seventeen of twenty (85%) active sites in Lithuania had food caches by week 46. The mean cache volume was 52.11 m³ (range = 1.62–149.00 m³, SD = 39.57). Five caches were less than 10 m³ suggesting an initial stage of development, while 10 sites had caches larger than 50 m³ indicating an earlier initiation date. The number of woody species in the cache ranged from 1 to 7 (Table 1).

By week 47 (mid-November) 37 of 47 (79%) active beaver sites in Sweden, Norway and Lithuania had food caches and by December, 41 of 47 (87%) active sites had food caches. This is a minimum estimate since sites in Sweden and Lithuania were not sampled in December and additional food caches may have been initiated by then.

Water flow, demographic factors and cache size in Norway

In Norway food cache area tended to be larger at sites with no water flow compared to sites with slow and medium water flow (water flow (none): estimate ± SE: 0.68 ± 0.38, p = 0.087; Fig. 3a). No food caches were constructed at sites with fast water flow. No correlation was found between water depth and water flow (r = 0.13, p = 0.67) and means were similar (none = 1.38 m, slow = 1.53 m, medium = 1.49 m).

The final food cache area at Norwegian sites (n = 37 in 2015 and 2016 combined) ranged from 3 to 158 m² (mean = 22.8 m², SD = 28.2) and the final food cache area was larger in 2015 compared to 2016 (estimate ± SE: –0.54 ± 0.11, p < 0.001). The mean family group size at sites with a food cache (n = 19 in 2015 and 18 in 2016) was 4.1 (range 2–9, SD = 2), with no difference between years (Mann–Whitney U test: U = 115, p = 0.221). The duration of territory occupancy ranged between 1 and 13 years (mean = 4.7, SD = 3.6).

Family group size and duration of territory occupancy had no significant effect on the food cache area (p > 0.2; Fig. 3b–c).

Figure 2.

(a–c) Estimated and observed initiation dates of new food caches and mean weekly minimum temperatures at research sites. Dotted line is mean weekly temperature (degrees C) and solid bars are number of new food caches. The Norway data includes two caches sampled in December but not in November.

Discussion

Our data are unique in that we compared food caching behavior in three populations in three countries in northern Europe. We observed both similarities as well as plasticity within and between these Eurasian beaver populations in food caching behavior. We found general similarities in water depths at caches, percentage of active sites with caches, and the pattern of cache initiation associated with declining autumn temperature between the research areas. Variability existed in observed cache sizes both within and between the populations. We also found no relationship between cache size and group size or duration of territory occupancy in the Norwegian population.

Figure 3.

The effect of water flow (a) on the final food cache area (m²) at Norwegian sites. The 95% confidence intervals are given as bars. (b) shows the relationship between food cache area and family group size and (c) between food cache area and duration of territory occupancy (both not significant).

The comparison of water depth of the food caches suggests reasonable similarity in choice of location for the winter lodge with the water depth near the entrance being critical to avoid freezing, allowing access to the food cache and allowing winter mobility. The results of our depth analysis found that no cache was located in water shallower than 1 m or in water deeper than 2.3 m. The only comparable data to ours is from Hartman and Axelsson (2004), who studied beaver food caching in Värmland, Sweden. They measured the water depth outside the lodge (not directly in the cache) and reported the mean depth at sites with a cache to be 1.1 m. This mean depth is less than either the total mean depth from all our study sites (1.6 m) or the mean depth for our Swedish sites (1.9 m). Hartman and Axelsson (2004) also reported finding caches in water less than 1.0 m, which differs from our observations in Sweden, Norway and Lithuania.

Hartman and Axelsson (2004) also found that caches in shallower water were significantly smaller than in deeper water and their overall mean cache volume was approximately half the mean volume we observed in our Swedish sites. However, since many of their caches were in shallow water this may partially account for volume differences between our studies. Wilsson (1971) also reported that caches on smaller watercourses (with an assumed shallower depth) were smaller than caches on larger rivers. Differences between Hartman and Axelsson (2004) and our study may be due to the specific habitat types they studied in Värmland, could indicate plasticity in selection of sites for the winter cache, or reflect differences in data collection.

Given the specific local habitat differences between our three major sites the similarity in choice of cache location is compelling. Water depths were significantly greater at the Swedish sites, but very similar between the sites in Norway and Lithuania, which are very different habitats (river versus small streams and ponds). However, the ranges were similar between all three areas. While certainly habitat dependent, it seems reasonable to assume that most beaver food caches will occur in water at least 1 m deep. Additional data to document this across the Eurasian range could provide an additional management tool to access beaver habitat quality.

Our three study areas were generally similar in percentage of active sites with food caches. Only one site in Sweden, two sites in Norway and three sites in Lithuania did not have food caches at the time of the last observation in 2015. Since we did not sample in Sweden and Lithuania in December and some sites in Norway were not visited due to ice cover, it is possible that our final percentage (approximately 90%) is low. In Sweden and Norway approximately 70% of sites had caches by the end of October (week 44/45). In 2016, in Norway, we also observed a lower percentage of sites with food caches in October, but by November (week 46/47) caches were observed at all sites.

Hartman and Axelsson (2004) reported a lower percentage of sites with food caches (67%) than we observed, but 50% of their lodges without caches had water depths of 0.5 m or less suggesting water depth as a limiting factor for cache construction. Wilsson (1971), studying farther north in Sweden (approximately 5° north of our sites) suggested that most, if not all, beavers constructed caches, even animals he kept in enclosures. Based on our results it appears that a high percentage of, if not all, beaver families living in Northern Europe should construct food caches. Specific site characteristics (type of wetland, water depth, water flow rate) may also influence this behavior.

We observed differences in final cache volume among the areas, with caches in Lithuania the largest and caches in Norway the smallest. The sites in Norway were on mid-sized rivers while the sites in Lithuania were on smaller streams, drainage ditches or ponds, which may have influenced cache size. However, caches in Sweden were also larger than those in Norway and some were on mid-sized rivers. The monthly autumn–winter temperature pattern for 2015–2016 shows Lithuania recording consistently colder temperatures than Sweden and Norway, especially during the winter months. Lithuania is influenced by cold, dry continental air masses during winter while Norway is influenced more by warmer maritime air masses associated with the Norwegian current. Our Swedish sites, with caches smaller than Lithuania but larger than Norway, were between 50 and 150 km from the Baltic coast and winter temperatures were generally colder than Norway but warmer than Lithuania. Interestingly, both Lithuania and Sweden recorded mean March temperatures below freezing while Norway had a mean March temperature above freezing. Our data, which appear to be unique in the literature, suggest that ambient air temperature is associated with larger food cache volume.

Timing of food cache construction depends on specific environmental conditions at the research location. For example, for the North American beaver, Aleksiuk (1970), in northern Canada, reported caching to begin as early as late August, Osmundson and Buskirk (1993) in Wyoming (at elevations between 2300 and 3500 m) reported caches were begun as early as mid-September, while in Massachusetts, in the northeastern United States, cache construction may not occur until early October (Hodgdon 1978). In Varmland, Sweden, Hartman and Axelsson (2004) reported cache construction beginning as early as mid-September (week 38) and continuing through late November, while Vasin (2001), in northwestern Siberia, reported that beavers begin caching in mid-August with freeze up in mid-October.

Initiation of cache construction at our research locations was fairly consistent. The earliest estimates suggest cache construction began as early as late September/early October (week 39/40) in Sweden and Lithuania, but not until mid-October (week 42) in Norway. There is extensive variability in timing of cache initiation at individual sites and it may reflect behavioral plasticity in individual animals. Moreover, in Norway, food cache size within active sites varied much less between years compared to the variation between active sites. This suggests that individual family groups may show consistent food caching behavior. More research is required to determine why some families begin caching early while others wait until later and if there is a consistent annual pattern.

While many variables may influence cache construction (photoperiod, deciduous leaf fall, changing woody species nutritional value) the relationship to declining minimum temperatures and the first period of below freezing temperatures during the autumn is compelling. This relationship has been suggested by Wilsson (1971) in Sweden and Busher (2012) in Massachusetts, USA. Wilsson (1971) noted that beavers in outdoor enclosures began cache construction at similar times to beavers living free along the river, but that beavers held in indoor terraria started caching at different times. Wilsson (1971) goes on to suggest that temperature may be the environmental stimulus since beavers in terraria, under more controlled (warmer) temperature regimes, showed more variability in caching behavior. Our data connect decreasing minimum temperatures and the time of first period of below freezing temperatures with the onset of caching behavior and imply that declining temperature during the autumn may act as an environmental stimulus for initiation of food caching behavior in beavers.

No other study of the Eurasian beaver has documented the temporal development of beaver food caches and our results from Sweden over a five-week period provide insight into this aspect of cache development. Cache development was relatively slow early in the autumn but increased as the season progressed. Mean cache volume in mid-October was significantly smaller than in mid-November. These data are relatively consistent with data from the North American beaver where caches have an initial slow rate of development, followed by a more rapid phase and then a slower growth phase near the end of the caching period (Busher unpubl.).

Our Norway data allowed us to compare the relationship between water flow speed, family size, duration of territory occupancy and cache size. Although not significant, we found that beavers at sites with no water flow (lakes, oxbow lakes) tended to build larger caches compared to sites with moving water (main river). This seems reasonable, since these sites froze over first limiting access to food resources other than from the cache. Conversely, the one active site with fast flowing water did not have a food cache in either year of the study. This could indicate that beavers in calmer water bodies (lakes, ponds) require larger caches to sustain them throughout winter due to prolonged periods of ice cover. Alternatively, the slow water flow allows for the construction of larger caches compared to more fast flowing water bodies (streams and rivers), whereas the two hypotheses are not mutually exclusive.

Water flow and water depth at active cache sites may be related and together influence caching behavior. In Norway, the precise location of the lodge and cache were variable but tended to be on areas of the river systems with slower moving water. We found no significant correlation between water depth and water flow and cache depths were similar regardless of flow. However, additional data are required to better understand any relationship between water flow rate and water depth and how these variables influence food caching behavior.

We found no significant effect of either family group size or length of territory occupancy and food cache size, which was surprising since we assumed that a larger family group and a more experienced adult pair would construct larger caches. Our results are similar to what Osmundson and Buskirk (1993) reported for the North American beaver, but different from the significant relationship between food cache size and family group size reported by both Kafcus (1987) and Easter-Pilcher (1990) also for the North American beaver. These different findings on the relationship between group size and cache size may be due to methodological differences (cache area versus cache volume, population sampling techniques), specific habitat differences or behavioral differences between populations. However, the lack of agreement on the relationship between family group size and food cache size suggests that caution should be used when relating cache size to number of beavers present in a family.