1. Herbivores make decisions at different spatial levels in response to food plant quality and quantity. Although experiments have examined the responses of herbivores at lower levels, few have examined how herbivores respond to such variation at the stand level. We assessed the response of a herbivore community in a boreal forest to manipulations of food resources at the stand level by performing a large replicated experiment where we fertilized young forest stands and then followed their use by herbivores during the subsequent year.
2. We used stands at 25 sites in an area 35 × 45 km. At each site we had one fertilized plot, a close control (100 m away) and a distant control (300 m away), all 50 × 50 m.
3. Before the growing season we fertilized treatment plots with calcium–ammonium–nitrate at 600 kg ha−1 (200 kg N ha−1). Fertilization significantly improved browse quality (nitrogen concentration) in both downy birch Betula pubescens and Scots pine Pinus sylvestris. Furthermore, the amount of browse significantly increased for birch, and Scots pine showed a similar trend.
4. Considering the animal community as a whole, we found that 10 of 13 animal species/groups made more tracks during winter in fertilized than in control plots. The total number of tracks was greater in fertilized plots, followed by close and then distant controls.
5. In the summer following fertilization, moose Alces alces strongly selected fertilized plots over controls. Furthermore, during the following winter moose again selected fertilized plots over controls. Hares Lepus timidus similarly left more pellets in the fertilized plots. Other mammals used the fertilized and the close controls similarly, and both were used more than distant controls. The number of grouse pellets did not differ among the treatments, although they followed a similar trend.
6. Several lines of evidence suggest that moose browsed fertilized plots more. Although not statistically significant, both heavy and light browsing were more common in fertilized than in control plots. Trees with the top shoot removed, or with bark stripped from the stem, did not differ significantly, but once again fertilized plots were browsed more than controls.
7. Our results are discussed in light of understanding both how herbivores in general respond to changes in food quality and quantity, and if fertilization may be a useful tool in modern forestry to manage herbivores.
8. Fertilization might be useful as a tool to alter the location of herbivore feeding, but buffer strips around any fertilized areas appear necessary, and other potential environmental effects should be evaluated.
Herbivore behaviour is generally closely linked to the food plants. In many systems, food resources change predictably in quantity and quality with the seasons, and concomitant with this the behaviour of the herbivores changes. At the extreme, herbivores migrate long distances among feeding areas: wildebeest and caribou are among the best-known seasonal migrations (Fryxell 1991; Belovsky 1991).
Given the obvious logistical difficulties, it is perhaps not surprising that there are very few large-scale field experiments on the impact of food resources on the behaviour of large- and medium-sized herbivores. However, herbivores in general do respond positively to nitrogen content (White 1993), and manipulation of this characteristic has been used in some elegant large-scale experiments with snowshoe hares Lepus americanus Erxleben. (Krebs et al. 1995).
Boreal forests are generally nitrogen limited (Tamm 1991). Consistent with this, the application of nitrogen is a common forestry management practice in Fennoscandia, and is employed to increase economic yields. This practice led us to consider that it might be possible to use the application of nitrogen fertilizer to create certain ‘sacrificial’ stands as a practical way to manipulate herbivore distribution, by attracting herbivores away from more valuable forest stands. In many areas in Fennoscandia, mammalian herbivores (moose Alces alces L. in the north; moose and roe deer Capreolus capreolus L. in the south) cause significant economic damage to the forest industry by browsing heavily in newly regenerating forest stands (Lavsund 1987).
Here, our objective was to test if the herbivore community of the boreal forest responds to manipulations of its food resources at the stand level. To do this, we performed a large replicated experiment in which we fertilized young forest stands and then followed the subsequent use by herbivores during winter and summer. As one of few large-scale experimental field studies, our results are discussed in light of understanding both how herbivores in general may respond to changes in food quality and quantity, and, more specifically, if fertilization may prove to be useful in modern forestry as a tool to manage the damage caused by herbivores.
The study was performed in coastal northern Sweden (64°N, 21°E; Fig. 1). The area belongs to the middle boreal forest vegetation zone (Ahti, Hämet-Ahti & Salas 1968). The coniferous forests are generally dominated by Norway spruce Picea abies L. Karst., but on drier sites Scots pine Pinus sylvestris L. is more frequent. Common deciduous species are silver birch Betula verrucosa Ehrh. and downy birch B. pubescens Ehrh., and less frequently aspen Populus tremula L. rowan Sorbus aucuparia L. and willows Salix spp.
Moose are the dominant large herbivore, with a winter density of about 0·8 moose km−2 (data from helicopter surveys conducted during the 1997/98 winter by J.P. Ball). Reindeer Rangifer tarandus L., which are semi-domesticated in Sweden, are present in the study area only in winter and are not common. Mountain hare Lepus timidus L. is fairly common in the study area, but information on their density is lacking.
In early 1996 we located stands of young Scots pine (with some deciduous trees) scattered over the study area in northern Sweden (Fig. 1). Habitat selection by animals for feeding may vary with many factors (for example their sex, Miquelle, Peek & Van Ballenberghe 1982; competitor density, Fretwell 1972; Houston & McNamara 1988); and the knowledge an animal has Gotceitas & Colgan 1991). Therefore, we wished to evaluate the responses of as many individuals of the animal community as possible. Additionally, to avoid psuedoreplication in study sites (cf. Hurlbert 1984) and to ensure that our results would have greater generality, the use of a large geographical area was desirable. We therefore used 25 sites spread over an area of 35 × 45 km. All stands were regenerating from clear-cut logging 7–15 years ago, and were thus of an age when they are heavily utilized by moose and other herbivores for feeding (Lavsund 1987).
At each site we selected two homogenous plots, each 50 × 50 m and separated by 100 m from each other. One of these two plots was randomly assigned to a fertilization treatment (‘fertilized’) and the other left as a ‘close control’. An additional control plot with similar vegetation about 300 m away from the other two plots was selected at random (‘distant control’). The reason for this second, more distant, control plot was to test the hypothesis that the fertilized plot would attract herbivores and would also increase herbivore utilization in the area immediately around the fertilized plot (i.e. to test for a potential ‘spill-over’ effect). Thus, we had one fertilized plot, a close control and a distant control plot at each of 25 sites.
Although we assigned treatments at random, any experiment can be confounded by pre-existing differences (Green 1979; Hurlbert 1984). Therefore, to test for any pre-existing differences in food resources, we counted the number of trees and shrubs by species that were over 1·2 m in height at 10 circular subplots (100 m2) at the fertilized, close control and distant control plots at each of the 25 sites. The height of 1·2 m was chosen because this area commonly receives about 1 m of snow, which covers shorter browse; counts of animal tracks were begun when snow depths were approaching 70 cm.
In April 1996 (before the growing season), we fertilized the treatment plots by hand by dosing with calcium–ammonium–nitrate at 600 kg ha−1, which provides 200 kg N ha−1. This treatment approximates the application rate commonly used today in Swedish forestry in coniferous stands (Melin & Nömmik 1988).
To test if the fertilization treatment was of sufficient magnitude to really alter the quantity and/or quality of available browse (and thus to have achieved an experimental perturbation of sufficient magnitude to consider this a ‘real’ experiment), we conducted two measurements. First, after the growing season we measured the length of the top shoots (current annual growth only) of both birch and Scots pine in the fertilized, close and distant control plots, to assess if fertilization increased the amount of browse available. Secondly, to test if fertilization altered the quality of the winter food, we collected twigs of these same species in April 1997 for analysis of the nitrogen concentration. However, to ensure that these collected twigs were representative of those actually eaten by herbivores, we first conducted preliminary sampling to establish the diameter at point of browsing (‘bite diameter’). We measured the terminal diameter of approximately 500 browsed twigs of both birch and Scots pine (systematically collected from all plots). We then clipped twigs at this diameter for the subsequent nitrogen analysis. For the nitrogen analysis per se, we collected shoots from 30 randomly selected trees of each species in the fertilized and control plots at each of the 25 sites in April 1997 (i.e. 1 year after fertilization). These shoots were dried to constant mass at 40 °C, milled and analysed for total nitrogen in a CHN-analyser (Elemental analyser CHN 2400; Perkin-Elmer, Norwalk, CT).
Estimates of herbivore use
The use of the fertilized, close control and distant control plots by vertebrate herbivores was estimated during three periods: (i) at the end of the first growing season, i.e. 4 months (September 1996) after the fertilization, by counting faecal pellets of moose; (ii) during the winter 1996/97 (December 1996–April 1997), by counting tracks of the entire animal community in the snow; and (iii) in spring (May–June 1997), by counting pellets of moose, hare, roe deer and grouse (hazelhen Bonasia bonasia L. willow grouse Lagopus lagopus L., black grouse Tetrao tetrix L. and capercaillie Tetrao urogallus L.). In spring (May–June 1997) we also collected data on winter browsing by moose (the most important herbivore) on Scots pine (an important moose food during winter).
Because moose are the dominant herbivore in our study area, we collected data on moose pellets at three critical times during the study: pretreatment, at the end of the summer following fertilization, and at the end of the winter following fertilization. To prepare for the pellet counts, 16 (in a 4 × 4 grid) circular subplots were laid out systematically in each 50 × 50 m plot. Ten of these 16 subplots were randomly selected, and prepared by removing all old faecal pellets in the spring of 1996; this not only improved the precision of our subsequent pellet counts, but also provided a pretreatment estimate of the use of the experimental plots by moose. These circular subplots were 100 m2 for moose and roe deer, and 2 m2 for hare. As pellets were found they were crushed in order to prevent accidental re-counting.
Animal tracks were counted during the winter of 1996/97 (December–April). Initially, we started to count tracks 5 days after the latest snowfall (to allow reasonable numbers of tracks to accumulate), but frequent snowfalls posed difficulties in finding snow-free periods of this length. Therefore, after January we started to count tracks 2 days after a snowfall. The weather remained unco-operative, but at each of the 25 locations we always counted the fertilized, close and distant control on the same day, so the differences in scheduling do not confound our evaluation of the treatments. For each species we counted all tracks passing the four sides of the square plot (50-m sides) and three 50-m transects laid out at equal distance inside the quadrat. Thus, for each plot we counted tracks along a total transect length of 350 m.
Moose browsing during winter on Scots pine (neither reindeer nor roe deer browse Scots pines of this size during winter to any appreciable extent, and hare browsing could be differentiated by the cleanly cut shoots) was evaluated in the spring of 1997. We recorded browsing on Scots pine in the following classes: heavy, i.e. more than 10 bites per tree; light, 1–10 bites per tree; the top shoot browsed; and the bark stripped from the stem.
Analysis of residuals (Tabachnick & Fidell 1983) revealed that the assumptions of parametric tests were not met for many variables, and data transformation did not fully rectify this. Therefore, for statistical analysis we used non-parametric tests: the Kruskal–Wallis test (KW) (Siegel & Castellan 1988) for analyses considering three groups (e.g. the number of moose tracks in the fertilized plots and close and distant control plots). For the statistical comparison of two groups (e.g. nitrogen content in fertilized vs. close control plots), we used the Wilcoxon rank-sum test (W) (Siegel & Castellan 1988). To evaluate the tracks left by the entire animal community in the fertilized, close and distant control plots (and to give each species or group of species equal weight), we used the Sign test (Conover 1980).
There were no significant differences (KW < 3·2, P > 0·20, n = 75 for all) in the density of any browse species (Pinus sylvestris, Betula pubescens, B. verrucosa, Salix spp. Populus tremula, Alnus incana L. Moench., Picea abies and Juniperus communis L.) prior to the fertilization treatment among the various plots. So randomization was effective and our results were not confounded by pre-existing differences among treatment plots.
Fertilization effects on food quantity and quality
The mean diameter at browsing of downy birch and Scots pine taken by moose was 2·4 and 3·9 mm, respectively (95% confidence limits were 2·4–2·5 mm for downy birch, n = 550, and 3·7–4·0 mm for Scots pine, n = 511). We used twigs of these sizes for nutrient analysis to ensure that our conclusions regarding nutrient quality applied to the twigs actually consumed by moose.
Browse quality (as indexed by nitrogen concentration; Danell, Edenius & Lundberg 1991) was significantly increased by the fertilization treatments in both downy birch and Scots pine (W = 5·1, P < 0·05, n = 50, and W = 32·3, P < 0·0001, n = 50, respectively; Fig. 2a). Furthermore, the amount of total browse available (as indexed by the length of the current top shoot) was significantly greater in the fertilized plots for downy birch (KW = 20·2, P < 0·0001, n = 75; Fig. 2b), and, although not significant, Scots pine showed a similar trend (KW = 4·8, P = 0·09, n = 75). Thus, downy birch in fertilized plots responded to fertilization primarily by increasing the amount of browse available, whereas Scots pine responded mainly by increasing the quality of the browse available (Fig. 2a,b).
Over the entire winter we counted 1839 tracks of herbivores or their predators (we hoped the latter group might provide additional insight into the distribution of their herbivorous prey). This total included moose (488 tracks), reindeer (379), hare (434), microtine rodents (100), fox Vulpes vulpes L. (241), weasel Mustela nivalis L. (127) and a few other species. Note, however, that for the unit of statistical independence we used the experimental plot (75 d.f. = 25 sites × 3 treatments). For example, even though we counted 379 reindeer tracks at the 75 plots, 69 of these plots had zero tracks.
As assessed by counts of tracks in the snow, willow grouse was the only animal species that differed at P < 0·05 in its use of the treatment plots (fertilized plots selected, KW = 6·1, P < 0·05, n = 75; Fig. 3). All other animals for which we observed tracks (moose, reindeer, roe deer, hare, squirrel Sciurus vulgaris L., microtine rodents, capercaillie, black grouse, hazelhen, fox, weasel and ermine Mustela erminea L.) did not differ significantly among the treatments (KW < 3·1, P > 0·20, n = 75 for all; Fig. 3). Variation was high, however: the average coefficient of variation per species was 886%.
Most importantly, if we considered the animal community as a whole (i.e. testing how many species used the fertilized plots more than the control plots), we found that the community made significantly more tracks in the fertilized areas than in the control areas (10/13 species/groups of species, P < 0·05, Sign test). The three exceptions to this pattern were microtine rodents, squirrels and ermine, which utilized control areas more than fertilized plots, but none did so significantly (KW < 3·1, P > 0·20, n = 75 for all; Fig. 3). Overall, however, the tracks clearly indicated that the animal community used the fertilized plots most, followed by the close controls and then the distant controls (Fig. 3).
In contrast to the extremely variable track data, pellet group counts were not influenced by weather: pellets accumulated over the summer after fertilization and were counted in late autumn for moose, and accumulated over the winter and were counted in early spring after snow melt for all species. Thus, for the dominant herbivore (moose) we had pellet group data for both summer and winter, but for other animals we had data for pellets integrated over the entire year following fertilization. We counted a grand total of 2490 pellet groups (e.g. moose) or pellets (e.g. hares). Hares exhibited a significant difference in the amount of pellets left in the three treatment plots (KW = 22·9, P < 0·001, n = 75; Fig. 4), with more pellets in the fertilized plots. Mammals other than hare and moose used the fertilized and the close control plots similarly, but both were used more than the distant control (Fig. 4). Pellets left by grouse (capercaillie, black grouse and hazelhen) did not differ among the treatments (KW < 2·0, P > 0·36, n = 75; Fig. 4).
There were no significant differences in the density of moose pellet groups prior to our experimental manipulations (KW = 0·23, P = 0·89, n = 75). Note, however, that in the summer following fertilization moose strongly selected fertilized plots over both control plots (KW = 12·3, P < 0·005, n = 75; Fig. 5). Furthermore, during the winter following fertilization, moose again selected the fertilized plots in preference to control plots (KW = 6·5, P < 0·05, n = 75; Fig. 5).
Consideration of the direct evidence of herbivory revealed several lines of evidence that moose browsed more in the fertilized plots (Fig. 6). Light browsing (1–10 shoots browsed per tree) was more common in the fertilized plots (KW = 5·6, P = 0·06, n = 75), and heavy browsing (greater than 10 shoots browsed per tree) was also more common in fertilized plots than in control plots (KW = 5·2, P = 0·07, n = 75). Trees with the top shoot removed or where moose stripped the bark from the stem did not differ among treatments (KW < 2·2, P > 0·20, n = 75 for both), but even here the same trend, for the fertilized plots to be browsed more than controls, was seen (Fig. 6).
There were no significant (P > 0·20) pre-existing differences among our treatments for the density of any browse species or for the use of these plots by moose prior to fertilization. Furthermore, our fertilizations altered browse quality and quantity of the two most important browse species (Pinus sylvestris and Betula pubescens;Fig. 2a,b). Although we restricted our nitrogen analysis to the two species most browsed in winter, we suspect that many other plant species showed similar responses. We conclude that there were no differences among the plots before fertilization, and that fertilization provided an increase of sufficient magnitude in browse quality and quantity for us to proceed.
Although the number of tracks recorded was large, on a species-by-species basis the track data revealed only one species (willow grouse) that exhibited a difference between treatments that was significant at the P < 0·05 level. It seems possible that the lack of statistical differences for individual species might be attributable to the extremely high variation in the track data (partly due to weather). Nevertheless, most species (or species groups) showed exactly the same trend for tracks, i.e. highest in fertilized plots, close control plots were intermediate, and the fewest tracks were found in distant control plots. In fact, if we consider the animal community as a whole, we do in fact find a significant difference between treatments, with fertilized plots being used more than control plots.
In contrast to the snow track data, pellets accumulated regularly and were thus not subject to the vagaries of weather. Moose pellets provided rather clear evidence that there were no pre-existing differences in moose use of our plots to confound our experiment (Fig. 5). After fertilization, moose clearly did select habitats at the stand level based on browse quality and/or quantity. Hare (the second most important herbivore in our area) showed a similar response to the fertilization, as indexed by pellet counts (Fig. 4).
Whereas tracks and pellets suggest that moose and hares (and indeed the entire animal community) spent more time in fertilized plots, less in close control plots, and even less in distant control plots, it must be noted that this does not prove that they were actually eating there. However, the direct assessment of moose herbivory on Scots pine confirms this ranking of use of the three treatments (Fig. 6). Although these differences were close to significant only for the browse estimate classes ‘heavy shoot herbivory’ and ‘light shoot herbivory’, we again see the general pattern for all four direct measures of herbivory: the fertilized plots were used most, then the close controls, and the distant controls were used least.
Some individual tests were significant at the 0·05 probability level, and many others were close to significant (e.g. P = 0·06). Overall, the consistency of all these criteria (i.e. the ranking of the use of the plots was almost always fertilized > close control > distant control) leads us to conclude that the animal community in general (and moose in particular) did in fact select habitats at the stand level by the criteria of browse quality and/or quantity.
This general response of the herbivores to increased nitrogen concentration in their food plants is in accordance with earlier reports in the wildlife literature (Miller 1968; Behrend 1973; but see Grenier, Bernier & Bedard 1977). The increase in herbivory that we observed in response to fertilization is also in accordance with the carbon/nutrient balance hypothesis (Bryant, Chapin & Klein 1983). Specifically, nitrogen fertilization is predicted to decrease the resistance to herbivory of slowly growing plants equipped with carbon-based defences by decreasing the amount of plant secondary compounds, which, in turn, deters feeding (Bryant, Chapin & Klein 1983).
Counts of grouse pellets and tracks were low and highly variable, but the trend we observed was similar for both and mirrored the pattern exhibited by most other species. The positive response by grouse to nitrogen fertilization was probably because fertilization improved the palatability of downy birch shoots (which are eaten by hazelhen, and black and willow grouse) and Scots pine shoots (eaten by capercaillie).
It is perhaps surprising that the microtine rodents (with the bank vole Clethrionomys glareolus L. as the dominant species) showed a negative correlation with the addition of nitrogen (Fig. 3). This may be attributable to the weakness of track counts for these small inconspicuous rodents. In addition, note that microtine rodents inhabit the subnivean space and only occasionally appear on the surface of the snow (Ylönen & Viitala 1985) where, of course, we counted tracks. If fertilization improved the quality and/or quantity of microtine food plants (including the bark of Scots pine and birch) buried under the snow (as it did for the birch and Scots pine shoots that we collected above the snow), these rodents might have to make fewer foraging trips on the snow surface where they are susceptible to predators. This might be the reason for the divergent response of microtine rodents as indexed by our track count data.
It is interesting to note that the predatory fox and weasel also showed a positive response to the added nitrogen. We suggest that this is a response to an increase in abundance of their herbivorous prey, and it supports our data showing increased densities of herbivores in the fertilized plots. Overall, even though many differences were not statistically different at P < 0·05, 10 of 13 species (or groups of species, such as microtines) were positively associated with fertilization. Moose browse on birch and Scots pine during winter, and strip birch leaves during summer (Bergström & Hjeljord 1987). Because birch and Scots pine increased in both quality and quantity, our experiment does not allow us to separate the effects of browse quality from quantity. Logically, herbivores might respond to both criteria, and our conclusions do not rest on identifying which is more important.
There is much recent interest among ecologists in applying spatially explicit models to understanding how landscape configuration and herbivore populations interact to influence a variety of important ecological mechanisms (Kareiva & Wennergren 1995; Bascompte 1998; Pastor et al. 1998 and references therein). Such models implicitly or explicitly assume that large mobile herbivores make certain decisions based on food quality/quantity on scales ranging from bites to the landscape level (Senft et al. 1987). Relevant observational studies are not rare, but experimental tests are much less common, and those that do exist are understandably at lower spatial scales. For example, some experimental studies of captive animals have found that browsers do select individual bites in relation to browse quality (Gross et al. 1993). At a slightly larger scale, moose altered their bite size and browsing intensity in an experiment with free-ranging moose (Edenius 1993). At a larger scale again, Danell, Edenius & Lundberg (1991) exposed artificial tree stands to free-ranging moose and found that food selection appeared to occur primarily at the tree level within stands, with moose being able to identify single trees with a slightly higher nitrogen concentration than average trees. At the largest scale, in the forestry literature there are reports of the effect of fertilization of forest stands with respect to damage by browsers (Löyttyniemi 1981). Extending these previous experimental studies to an even larger scale, our experiment suggests that moose (and indeed the entire herbivore community during winter) are similarly selective at the stand level.
We conclude that fertilization with nitrogen has the potential to be a useful tool to ameliorate forest damage by herbivores in nitrogen-limited ecosystems, or at least to manipulate where this damage occurs in winter. Because herbivores often cause severe damage to young forest stands, and because fertilization is already used in forestry, the distribution of these winter feeding sites in the landscape may be altered by this practice. Note, however, that our analysis suggests that one must also be cognizant of the fact that moose (and other herbivores) did not exhibit perfect discrimination of the borders of the fertilized plots, but invariably exhibited some ‘spill over’ into the close control a full 100 m away from the fertilized plot. If fertilization is employed as a technique to alter the location of herbivore feeding sites, a buffer strip (wider than 100 m) must be incorporated around the fertilized plot. This should avoid conflicts with neighbouring landowners whose adjoining forest might be adversely impacted by this ‘spill over’ herbivory. Finally, other potential environmental effects might result from adding nitrogen to the boreal forest ecosystem, e.g. microtine rodents may increase in number following improvements in their food supply, water quality may decline, or the abundance of various invertebrates may change. These should be considered prior to implementing any management through nitrogen fertilization.
We thank Assi Domän (Lycksele), MoDo (Robertsfors), Georg Adamsson, Kjell, Sune, and Tage Forsmark, Börje Johansson, and Karin Linné for allowing us to conduct this experiment on their land. Ola Gärdemalm provided valuable technical assistance in the field and laboratory. Careful reviews by Göran Ericsson, Margareta Bergman and two anonymous referees improved the paper. Financial support was provided by the Swedish Environmental Protection Agency and the Kempe Foundation.
Received 24 August 1998; revision received 15 November 1999