Linking seed dispersal to cache protection strategies


  • Alberto Muñoz,

    Corresponding author
    1. CREAF, Universitat Autònoma de Barcelona, Edificio C, 08193 Bellaterra, Spain
    2. Instituto de Ciencias Ambientales (ICAM)-Área Zoología, Universidad de Castilla-La Mancha, E-45071 Toledo, Spain
      Correspondence author. E-mail:
    Search for more papers by this author
  • Raúl Bonal

    1. Grupo de la Biodiversidad Genética y Cultural, Departamento de Ecología, Instituto de Recursos Cinegéticos (CSIC-UCLM-JCCM), E-13005 Ciudad Real, Spain
    2. Instituto de Ciencias Ambientales (ICAM)-Área Zoología, Universidad de Castilla-La Mancha, E-45071 Toledo, Spain
    Search for more papers by this author

Correspondence author. E-mail:


1. The spatial distribution of dispersed seeds results from the combined action of the caching strategies followed by different granivores. Hence, it is essential to study the factors that influence seed predation and caching decisions to achieve a better understanding of the dispersal process.

2. In this study, we document how seed dispersal and the spatial patterns of natural recruitment are linked to the strategies used by granivores to protect their cached seeds from pilferage. We present a theoretical model showing that those strategies may convey benefits for both seed cachers and plants.

3. We studied the relationships among seed production, seed predation/caching, cache pilferage and plant recruitment in a savanna-like landscape of oaks dispersed by scatter-hoarding rodents.

4. Our results show that acorn-dispersing rodents were concentrated under the canopies of scattered oaks, where the theft of cached acorns increased by 77% as compared to that of the surrounding open landscape. Acorns were thus cached selectively in the open areas to reduce pilferage; in fact, none of the few seeds cached beneath tree canopies survived predation by granivores (pilferage + recovery). Meanwhile, some acorns cached in the surrounding open areas were neither pilfered nor recovered and then recruited successfully. Accordingly, natural recruitment of newly emerged seedlings was higher outside than under canopies, suggesting that rodent caching strategies have direct implications for the directed dispersal of oaks.

5.Synthesis. The spatial patterns of seed dispersal shape the fitness of both the plant because they influence dispersal and recruitment efficiency, and the granivores that cache and predate its seeds because they influence their foraging efficiency. Cache protection strategies reduce pilferage significantly and enhance seed recovery rates by the cache owner. At the same time, more seeds remain dispersed and unrecovered. Thus, cache protection strategies can provide net benefits to the plant in terms of effective directed dispersal.


Many studies have analysed the effects of seed-caching granivores on seed dispersal patterns and plant recruitment (e.g. Vander Wall 1993, 2008; Hollander & Vander Wall 2004; Muñoz & Bonal 2007). In this context, it has been suggested that seeds are usually not moved randomly; rather, they are frequently dispersed to certain sites that increase their recruitment chances, what is known as ‘directed dispersal’ (Wenny & Levey 1998; Wenny 2001; Pearson & Theimer 2004; Briggs, Vander Wall & Jenkins 2009). In order to assess the role of granivores in the directed dispersal of seeds, it is necessary to identify the causes ruling the spatial distribution of caches. In this sense, some studies have highlighted the importance of behavioural research for understanding the patterns of seed dispersal and plant recruitment (see classic studies by Stapanian & Smith 1978, 1984, 1986). For example, Scofield, Sork & Smouse (2010) have recently shown that the social behaviour and territoriality of acorn woodpeckers (Melanerpes formicivorous) influence the dispersal distances and directions of Quercus agrifolia acorns in their territories.

The causes of food-caching behaviour have received a great deal of attention from the perspective of animal cognition (e.g. Heinrich & Pepper 1998; Brotons 2000; Bugnyar & Kotrschal 2002; Emery, Dally & Clayton 2004; Dally, Clayton & Emery 2006; Hopewell & Leaver 2008; Pfuhl et al. 2009). We know that food-caching animals have developed abilities to select food items (Muñoz & Bonal 2008a) and to cache and recover them (Clayton & Dickinson 1998; Emery & Clayton 2001; Roberts et al. 2008; Zinkivskay, Nazir & Smulders 2009). These capabilities can also lead them to pilfer caches made by other conspecific individuals, and have led to the evolution of an array of cache protection strategies against pilferage (reviewed in Dally, Clayton & Emery 2006). In general, caching animals avoid the sites and situations where pilferage risk is high, such as the black-capped chickadees (Poecile atricapillus), that prefer to cache food items in sites with low pilferage risk (Hampton & Sherry 1994). When Merriam’s kangaroo rats (Dipodomys merriami) observe conspecifics pilfering their caches, they switch from scatter-hoarding to larder-hoarding (Preston & Jacobs 2001), and cache in sites less susceptible to pilferage (Preston & Jacobs 2005). In the presence of conspecifics, grey squirrels (Sciurus carolinensis) make caches farther apart (Leaver et al. 2007). When conspecific thieves are present, corvids cache behind large obstacles (Bugnyar & Kotrschal 2002), in shady or distant sites (Dally, Emery & Clayton 2004, 2005). They are also known to re-cache the food in new sites unavailable to the pilferer (Emery, Dally & Clayton 2004), or delay caching (Heinrich & Pepper 1998) to reduce pilferage.

All of these studies highlight the importance of knowing the motivations of seed-caching behaviour, with special emphasis on cache protection strategies, to understand the spatial patterns of seed dispersal and seedling recruitment. However, cache protection strategies have not been linked explicitly to their potential consequences on seed dispersal, partly because ‘cognitive approaches’ and ‘ecological approaches’ are still two often-separated areas of research (see Dukas 2004). In the present study, we explicitly analyse how the spatial patterns of seed dispersal and recruitment are linked to the cache protection strategies used by seed-dispersing granivores. We firstly present a simple model which proposes that cache protection strategies may have benefits for both partners: the plant and the seed disperser. At the same time that they reduce pilferage allowing the cache owner to recover its stored food, more seeds remain dispersed and unrecovered (i.e. uneaten) and may finally germinate and become established. Under this hypothetical scenario, we have studied the seed caching and pilfering patterns by small rodents, and the recruitment patterns in a savanna-like landscape with scattered oaks within an extensive grassland matrix. In a 2-year study, we first tested whether abundance and activity of rodents are influenced by tree canopies. We expected rodents to be concentrated beneath canopies and trunk surrounding resprouts, the only protective cover in the study area (Muñoz, Bonal & Díaz 2009). Then, we measured whether pilferage of cached seeds was higher beneath canopies than in the surrounding open areas due to the high rodent concentration beneath tree canopies. We monitored seed production, seed predation, seed dispersal and the patterns of natural recruitment in relation to canopy cover. We expected more seeds to be cached outside canopies to reduce pilferage and more seeds to remain unrecovered there, thus increasing the probabilities of seedling establishment in the open areas surrounding the oaks. Consequently, the strategy of dispersing and caching seeds selectively outside the mother tree canopy to reduce pilferage may have direct implications for oaks’ directed dispersal, since it could enhance recruitment by decreasing predation risk and other density-dependent negative consequences.

Materials and methods

The theoretical approach

To analyse how seed dispersal and recruitment are linked to the cache protection strategies by seed-dispersing granivores we used an approach (see Fig. 1), which takes into account both the foraging efficiency of cachers (C) and plant recruitment efficiency (P). When a seed is available, predation is positive for the cacher (C+, i.e. food intake) but not for the plant (P−, i.e. seed killed). Caching the seed can be positive for the cacher (C+, seed is hidden to avoid pilferage) and the plant (P+, the seed passes the bottleneck of being killed on the ground or by pilferers). If the caching strategy is efficient against pilferers, the survival likelihood of the seed cached will increase, assuming that pilfered caches are predated and not re-cached. From a cacher perspective, pilferage avoidance increases food availability. From a plant perspective, in addition to predation avoidance, it is also important whether the cache is located at a site where the environmental conditions (e.g. soils, nutrients, light) are suitable for germination and establishment (Wenny & Levey 1998; Wenny 2001). At this point, granivore and plant pathways diverge, as those seeds surviving pilferage can be either recovered by the cache maker (C+, P−) or not recovered (C−, P+).

Figure 1.

 Different steps of a cacher’s decisions once a seed is available and the consequences for its own foraging efficiency (C) and the recruitment efficiency of the plant (P) (see materials and methods: ‘the theoretical approach’). The figure shows how caching strategies can shape both cache pilferage and cache recovery, and also determine plant recruitment efficiency. The upper graph compares the number of seeds pilfered (dark grey), recovered (grey) and not recovered (black) of A seeds cached in sites with low (Sl) and high (Sh) conspecific activity. For each site the number of seeds in grey represents the fitness benefit for the cacher, and the number of seeds in black the benefit for the plant. This hypothetical situation shows that following a caching strategy Sl can provide benefits for the cacher and the plant: caching in Sl reduces significantly pilferage, and then this strategy can prevail over Sh even when more seeds remain unrecovered, thus also benefiting the plant.

In a hypothetical context in which a granivore can cache A seeds either in a site in which conspecific activity is high (Sh) with a caching probability p, or alternatively in a site with low conspecific activity (Sl) with a probability (1-p), the benefits for the cacher and the plant would be:

Cacherinline imageinline image
Plantinline imageinline image

Where Ph and Pl are the number of seeds pilfered in Sh and Sl, respectively. In the same way, Nh and Nl represents the number of seeds not recovered by the cacher in Sh and Sl. The benefit for a cacher would be p(1−Ph/A−Nh/A) + (1−p)(1−Pl/A−Nl/A), which is equal to 1p(Ph/A + Nh/A)−(1−p)(Pl/A + Nl/A). The maximum cacher benefit is 1, which occurs when pilferage rate is 0 in either site (Ph/A Pl/A = 0) and all seeds are recovered by the cacher (Nh/A Nl/A = 0). In that case, the benefit for the plant [pNh/A + (1−p)Nl/A] would be zero, as all seeds would be finally recovered and predated by the cacher. If the benefits cachers received by dispersing seeds in Sl are higher than in Sh, then [1−Ph/A−Nh/A] < [1−Pl/A−Nl/A], which can be simplified as [Pl/A + Nl/A[Ph/A + Nh/A]. Given that pilfering rate is likely to be higher where conspecific activity is high, then caching seeds in Sl can significantly reduce pilferage (Ph/A >> Pl/A) and thus the strategy of caching in Sl can prevail over the strategy Sh even when more seeds remain not recovered (i.e. Nh/A Nl/A). Consequently, when a caching strategy Sl reduces pilferage rate in a higher extent than reductions in recovery rate (i.e. [Ph/APl/A] > [Nl/ANh/A]) the strategy could spread in the cacher population with benefits for the plant too, as their dispersed seeds would be protected from pilferers and some of them would be unrecovered, being able to germinate and establish. The upper graph of Fig. 1 shows an example of this situation, in which caching seeds in Sl increases both the number of seeds recovered by the cacher and the number of seeds left for the plant (N/A) thanks to reductions in pilferage rates compared to Sh (i.e. Pl/A << Ph/A and Nh/A < Nl/A). Using this approach we designed a field study to document and quantify the link between cache protection strategies, seed dispersal and recruitment.

Study area and species

The field work was carried out at Cabañeros National Park (Ciudad Real province, central Spain; 39°24′ N, 3°35′ W) during the acorn-fall periods (from September to February) of 2003–2004 and 2004–2005. The study area is a grassland (c. 5000 ha) with scattered and quite isolated (14 trees ha−1) holm oaks, Quercus ilex, L. (see Muñoz, Bonal & Díaz 2009 for details on vegetation structure). The holm oak is the most widespread oak of the western Mediterranean basin, extending over more than 5 million ha in the Iberian Peninsula (Blanco et al. 1997). Holm oak is a vigorous resprouter (Ortego, Bonal & Muñoz 2010), and the trunks of each of our study trees were surrounded by a surface of resprouting stems no taller than one metre. It is an evergreen oak that can produce very large crops (up to 38 000 acorns per tree), but with a strong inter-annual variability (Herrera et al. 1998; Bonal, Muñoz & Díaz 2007; Bonal & Muñoz 2007; coefficient of variaton across years = 2.22 Espelta et al. 2008). Its acorns are an important food source for small rodents, constituting the majority of their diet in autumn and winter (den Ouden, Jansen & Smit 2005; Muñoz & Bonal 2007, 2008a,b).

In the study area, previous live-trappings revealed that the Algerian mouse Mus spretus is the most abundant small rodent (> 95% of captures), followed by the wood mouse Apodemus sylvaticus (Muñoz, Bonal & Díaz 2009). Both species are nocturnal and are prominent acorn predators, but they disperse and cache them as well (Muñoz & Bonal 2007, 2008b). Other acorn consumers include the red deer Cervus elaphus and wild boar Sus scrofa (Bonal & Muñoz 2007; Muñoz, Bonal & Díaz 2009). In turn, rodents are preyed upon by an array of small carnivores and birds of prey such as red foxes (Vulpes vulpes), genets (Genetta genetta), stone martens (Martes foina), wild cats (Felis sylvestris), barn owls (Tyto alba), little owls (Athene noctua) and tawny owls (Strix aluco) (Muñoz & Bonal 2007).

Experimental design

Monitoring acorn production, rodent abundance and rodent activity

We recorded both acorn production and rodent abundance and activity during the acorn-fall seasons of 2003–2004 and 2004–2005, to analyse whether cache protection strategies and pilferage were influenced by the level of competition for seeds (i.e. the ratio number of acorns : number of rodents). Acorn production was recorded in a sample of 24 focal trees using seed traps (plastic buckets with an opening of 0.12 m2 and 0.5 m in depth) randomly hung from the lower branches. The number of traps placed at each tree ranged from 3 to 11 and was proportional to the surface of the canopy, covering between 1.5% and 2% of canopy area in all trees (Bonal, Muñoz & Díaz 2007). Canopy areas were calculated from the average of three random measures of the canopy diameter, considering canopies as roughly circular (Bonal, Muñoz & Díaz 2007). We assessed whether acorns were removed from traps by placing 100 pairs of acorns marked with a tiny incision in the proximal pole inside 100 traps (for a similar procedure see Bonal, Muñoz & Díaz 2007); none of them disappeared during the course of the study. Once acorn rain ended, the traps were emptied and their content taken to the laboratory, where the acorns were counted and classified. Acorn production was estimated as the number of mature acorns in each tree (Bonal & Muñoz 2007).

Rodent abundance was estimated by live-trappings. In each study period, we performed a rodent census in November placing 160 traps distributed in groups of 10 traps in different sections of the study area. Each trapping station consisted of 10 Sherman live-traps laid out in two parallel lines of five traps (the distance between lines and between traps within lines was 10 m). Each trapping session extended over three nights during the new moon in order to avoid any confounding effect of moonlight on rodent activity and trappability (Muñoz, Bonal & Díaz 2009). Traps were set between 12:00 pm and 4:00 pm GMT of the first day and checked between 7:00 am and 9:00 am GMT on the following 3 days. Each trap was baited with a piece of apple and a paste made of tuna in oil and flour. A piece of waterproof cotton was also added to protect captured rodents from cold and rain (Muñoz, Bonal & Díaz 2009). Rodents were identified to species, sexed, weighed using a 100-g spring balance, marked by cutting a small patch of hair from their head, and released. Data from live-trapping provide information on the identity and abundance of rodents, but also on their spatial distribution regarding canopy cover. Our aim was to have two distinct areas, one of high rodent activity (Sh) and one of low activity (Sl), so as to fit the structure of our model (see above). To do so, we placed some traps under the canopies of the isolated oaks while others were located within the surrounding grassland matrix around each of these trees, as we expected to find differences between both microhabitats (see Muñoz, Bonal & Díaz 2009). Moreover, in November 2003 we captured five M. spretus (three males and two females) that were marked with radiotransmitters (Mammal Collar Cable-tie Ag337 Pip; 0.90 g, Biotrack Ltd, Wareham, UK) in an adjacent area with a similar savanna-like landscape. The aim of tracking the activity of these rodents was to estimate the proportion of time that they spent under and outside the canopy of scattered trees. This species is nocturnal; hence radiotracking sessions were performed during 15 nights at 2-h intervals from 8:00 pm GMT to 8:00 am GMT. In each location, we noted if the collared rodent was located outside or beneath tree canopies.

Monitoring acorn predation and caching

Acorn predation and caching by small rodents were monitored in the 24 focal holm oaks. Distance between any two focal study trees was at least 35 m. This design ensures independence between trees in terms of rodent foraging patterns, as the distance between focal trees was much larger than the home ranges recorded for rodents in the same area (for similar procedure see Muñoz & Bonal 2007, 2008a). At each study tree, we monitored acorn predation and caching by placing feeding plots (50 × 50 cm) on the ground below the canopy. The number of feeding plots per tree varied between two and four and was proportional to the surface of its canopy. Feeding plots were protected from other acorn consumers (mainly ungulates) using 50 × 50 × 20 cm wire cages with a mesh size of 5 × 5 cm, which only allowed the entrance of rodents (Muñoz & Bonal 2007, 2008a). At each feeding plot, we placed three sound acorns collected from the oak beneath which the cages were located. We discarded those acorns infested by the chestnut weevil Curculio elephas, the most prominent pre-dispersal predator of holm oak acorns (Bonal, Muñoz & Díaz 2007), as weevil infestation is known to influence rodent foraging decisions (Sork & Boucher 1977; Steele, Hadj-Chikh & Hezeltine 1996, Muñoz & Bonal 2008a). Infested acorns were easily identifiable due to the small oviposition scar on the seed coat (Bonal & Muñoz 2007, 2009). Acorns placed in the feeding plots were marked with 50 cm of fluorescent fishing line with 5 cm of fluorescent pink flagging tape at one end to facilitate acorn location after removal. This methodology is not considered to influence rodent foraging behaviour (Forget & Wenny 2005), and by using this procedure we were able to locate more than 95% of the removed acorns (for similar procedure see Forget 1990; Jansen et al. 2002; Muñoz & Bonal 2007, 2008a). Marked acorns were handled using fresh gloves to avoid the effects of human contact on rodent foraging decisions (Wenny 2002). During the acorn drop seasons (September–February) of 2003–2004 and 2004–2005 we checked all marked acorns every 10 days. At each checking, all marked acorns were classified as exploited or unexploited by rodents. Those acorns exploited were in turn classified as (i) eaten at the feeding plots, (ii) moved and eaten or (iii) moved and cached. At each observation we searched intensively for every removed acorn, recording the dispersal distance and whether they were moved beneath the tree canopy or not. Eaten acorns were replaced with freshly marked acorns at the feeding plots at each revision. We compared the number of natural caches that were not recovered or pilfered under (Nh) and outside (Nl) canopies.

Monitoring cache pilferage and the patterns of natural recruitment

Artificial caches were used to estimate cache pilferage rates (Vander Wall et al. 2006). Cache pilferage rates were measured by burying 126 simulated caches (containing one acorn each) by the end of the seeding season (i.e. January 2004 and 2005). Simulated caches were distributed among 21 of the 24 focal trees where rodent foraging patterns were monitored. There were six caches per focal tree: three beneath the tree canopy (within a radius of approximately 1 m around the trunk), and the other three buried 7 m away from the outermost perimeter of the canopy, in a straight line with those placed under the oak cover. This distance is approximately the average distance recorded for the natural caches made by rodents outside tree canopies (see results). Simulated caches were observed every 25 days and we noted whether the acorns had been stolen by rodents or not to compare whether the pilferage rates differed under (Ph) and outside (Pl) canopies. Pilfered acorns were easily identifiable as rodents leave a characteristic hole and shard acorn coat after predation. We also recorded the number of seedlings that recruited from the artificial caches in order to compare the establishment success between the two microhabitats (beneath canopy vs. open areas). In addition, at each focal tree, we counted the number of natural current-year oak seedlings emerged under oak cover and within a radius of 7 m from the canopy border to test whether trees have more surviving seedlings under or outside their canopies. Counts were made in autumn 2004 because in Mediterranean areas this is the most reliable period to measure recruitment as germination takes place in early spring (Muñoz, Bonal & Díaz 2009). In autumn, seedlings have already gone through the major bottleneck, which is the mortality due to desiccation during the critical first summer drought (Pulido & Díaz 2005).

Data analysis

Due to the spatial isolation of the focal trees, each one was considered to be an independent replicate of rodent foraging behaviour. In fact, the minimum distance between focal trees was much larger than the home ranges recorded for rodents at our study site, so that we could ensure that the same rodent did not forage in different focal trees (Muñoz & Bonal 2007, 2008a). We used loglinear models to test whether the canopy cover of scattered trees influenced the rodent decisions regarding seed predation and caching. These analyses are particularly useful when analysing selection experiments (Heisey 1985; Muñoz & Bonal 2008a). In the loglinear models, the response variable was the ‘frequency of acorns’, and the factors were the ‘rodent decision’ (consumption vs. caching) and the ‘canopy cover’ (under vs. outside). A significant interaction between the factors ‘rodent decision’ and the ‘canopy cover’ would indicate that the decision of rodents on whether to cache or predate a seed is influenced by canopy cover. Once an acorn disappeared (i.e. it was cached or consumed), we simulated a new acorn being dropped from the tree by replacing the original one in the feeding plot. Since our objective was to assess the final outcome for the plant and the cacher at the end of the seeding season, we pooled the originally placed and the replaced acorns of each tree in the variable ‘frequency of acorns’, given that the replacement rate was similar in all the focal trees. We performed the loglinear analyses also considering the focal tree as factor to avoid that the patterns documented could be only attributed to a few trees. Between-year comparisons of the production of acorns by focal trees, emergence of seedlings, rodent abundance and pilferage rates were performed with multi-factorial and repeated-measures analyses of variance.


Acorn production, rodent abundance and rodent activity

Acorn production was much higher in 2004–2005 (mean ± SE = 6797.1 ± 1026.4 per tree; range: 800–17 867; = 24) than in 2003–2004 (mean ± SE = 533.3 ± 156.1; range = 0–2267; = 24; repeated-measures anova, F1,22 = 37.92, < 0.0001). The variance in acorn production among the focal trees was higher in 2004–2005 (Levene’s test, F1,30 = 21.15, < 0.0001). The Algerian mouse (M. spretus) was the only rodent species captured in the two study periods. The number of rodents captured per trapping station (consisting of 10 Sherman traps) was significantly higher in 2003–2004 (mean ± SE = 1.31 ± 0.27; = 21) than in 2004–2005 (mean ± SE = 0.25 ± 0.27; = 4; F1,30 = 7.87, < 0.01). Thus, during the 2003–2004 season there was a higher abundance of rodents that foraged on a lower abundance of acorns compared to the 2004–2005 season. Rodents were more abundant beneath the canopy cover, as the number of captures in traps placed under canopies was higher than that expected by chance (χ21 = 4.49, = 0.03). We also obtained 33 locations of the five M. spretus that were radiotracked. In 30 of 33 independent contacts, rodents were located under the tree canopies (90.9% of time), although the canopy covers represented only 3% of the total surface of the study area. These results clearly differentiated the two areas regarding rodent activity illustrated in our model: one with high rodent activity (under canopies, Sh) and the other with low rodent activity (outside canopies, Sl).

Caching decisions, pilferage and seed dispersal by rodents

In both study periods most marked acorns were exploited by rodents. However, the percentage of acorns exploited per tree was higher in 2003–2004 (95.83 ± 5.39), the low production year, than in 2004–2005 (62.21 ± 5.39; F1,44 = 19.40, < 0.0001). Most of these acorns were eaten, either at the feeding plots or shortly after being removed. However, a proportion of them was removed and cached (Table 1), always singly, at a depth of 1.5–2 cm underground. Acorns were moved longer distances in the acorn-fall season of 2003–2004 (3.8 ± 0.4 m; range: 0.04–40 m, = 275 acorns) than in 2004–2005 (1.8 ± 0.4 m; range: 0.05–30.4 m, = 104 acorns; F1,375 = 10.84, = 0.001). Dispersal distances were longer for acorns carried to open areas (8.7 ± 0.9 m) than for acorns dispersed within the oak cover (1.5 ± 0.2 m; F1,375 = 87.05, < 0.0001).

Table 1.   Proportions of acorns exploited by rodents per focal tree that were eaten at the feeding plots, moved and eaten, and cached
 Percentage eaten at the feeding plotPercentage moved and eatenPercentage cached
  1. Data are presented as mean ± SE.

2003–2004 (= 23 trees, 578 acorns)49.48 ± 3.3348.27 ± 3.352.24 ± 0.75
2004–2005 (= 21 trees, 248 acorns)57.51 ± 5.5734.97 ± 4.637.52 ± 4.81

The decision whether to cache or consume the acorns was sharply influenced by canopy cover in both study periods (Table 2, interaction ‘canopy cover’ × ‘seed eaten vs. seed cached’), with no differences between focal trees (Table 2, interaction ‘canopy cover’ × ‘seed eaten vs. seed cached’ × ‘tree’). When rodents removed the acorns for consumption, they did so more commonly beneath the canopy than outside it (2003–2004: F1,44 = 98.63, < 0.0001; 2004–2005: F1,34 = 19.25, < 0.001; Fig. 2). By contrast, acorns were cached much more frequently outside the tree cover (F1,80 = 19.05, < 0.0001; Fig. 3) in both study years (interaction year × canopy F1,80 = 1.15, = 0.29; Fig. 2). Thus, rodents followed a strategy of caching seeds selectively in the areas of low conspecific activity (Sl) where, in addition, the pilferage rates (Pl) were lower than in the high activity areas (Ph). Specifically, the number of seeds pilfered was on average 77% higher in high activity sites than in low activity ones. Consequently, caching preferentially in sites with a low presence of conspecifics Sl may notably increase the number of seeds saved from pilferage because Ph >>> Pl.

Table 2.   Results of the fitting of loglinear models to the three-way contingency tables generated by the factors ‘tree’, ‘canopy cover’ and ‘eaten vs. cached’
Eaten vs. cached1235.06<0.001151.98<0.001
Tree × Canopy2256.35<0.0011017.260.068
Tree × Eaten vs. cached2216.590.78106.100.81
Canopy × Eaten vs. cached117.78<0.000117.820.005
Canopy × Tree × Eaten vs. cached228.420.99107.080.72
Figure 2.

 Proportion of acorns per tree moved and eaten, and moved and cached under and outside tree canopy in the acorn-fall seasons of 2003–2004 and 2004–2005. Whiskers show SE.

Figure 3.

 Number of caches stolen by rodents per focal tree outside and under tree canopy in the acorn-fall seasons of 2003–2004 and 2004–2005. Whiskers show SE.

As expected, pilferage rates were higher in the acorn-fall season of 2003–2004 (when rodents were more abundant and seeds more scarce) than in 2004–2005 (F1,80 = 4.38, = 0.039; Fig. 3). Changes between years in pilferage risk affected the spatial location of caches, and rodents followed a more strict Sl strategy (i.e. caching outside canopies) in the year when the probability of pilferage under canopies was higher. In 2003–2004 acorn availability was low and rodent abundance high, and thus the pilferage rate was also high under the tree cover (Fig. 3). In that period, a higher proportion of acorns was cached outside than under the canopy cover (F1,16 = 4.57, = 0.04; Fig. 3). In 2004–2005, by contrast, pilferage under the canopies decreased and differences between microhabitats were not significant, although more caches were still made outside the canopy (F1,8 = 1.28, = 0.29; Fig. 3).

Compared to artificial caches, acorns in natural caches (those made by rodents) had a higher likelihood of being consumed. That is because in the case of natural caches not only pilferage, but also recovery by cache makers, takes place. Nonetheless, cache location also had an effect on the fate of acorns. In both study periods, 100% of natural caches made by rodents under the canopy cover were recovered or stolen by conspecifics. That is, in the high activity area the number of seeds that finally remained not consumed after dispersal (Nh) was zero. In contrast, there were natural caches that remained non-pilfered or recovered in the low-activity area (Nl > 0). In fact, in the period 2003–2004 three natural caches made outside the tree canopy remained intact and the acorns germinated: one produced a viable seedling. In 2004–2005, two natural caches made outside the tree canopy also remained intact and produced viable seedlings. As our model illustrates, no seed was found to be re-cached after being recovered or pilfered from the first cache. In the case of the simulated caches buried by us, seed survival likelihood also differed depending on oak cover. The number of seedlings recruiting in open areas was significantly higher than beneath the canopy (repeated-measures anova, F1,28 = 6.17, = 0.019; Interaction with the study period was not significant F1,28 = 0.68, = 0.41; Fig. 4). Finally, the patterns of natural regeneration arising from the natural seedling census showed the same result. The number of 1-year seedlings emerging outside the canopy cover was higher than beneath the oaks (repeated-measures anova, F1,15 = 12.65, = 0.002; Fig. 4).

Figure 4.

 Comparisons of the number of seedlings per tree (mean ± SE) emerged from acorns planted under and outside oak canopies (left part), and those recruited naturally (right part). Grey points = under canopy cover; Black points = outside canopy cover.


We have shown, using a theoretical approach and with the support of field data that cache protection strategies can benefit both seed-caching granivores and plants. Our model considers a series of variables when a granivore make a number of caches (A): the number of cached seeds that are pilfered by other conspecifics (P) and the number of cached seeds not pilfered or recovered by the cacher (N). These latter are those seeds that remain dispersed and uneaten and that can finally recruit. The values of P and N can vary depending on whether the caches are made in sites with high activity of granivores (Sh) or low activity (Sl). Pilferage rate is expected to be higher in the areas of high activity than in the low-activity areas (Ph/A >>> Pl/A). Consequently, caching in the low-activity areas may save a considerable number of seeds from pilferage providing a net benefit for the rodents higher than caching in high activity areas even when their recovery efficiency in the low-activity areas is lower (i.e.Nl/A Nh/A; see Fig. 1 upper graph). Accordingly, a cache protection strategy can benefit both the granivore and the plant when the reductions in pilferage are higher than the reductions in recovery rate, that is, when (Ph/APl/A) > (Nl/ANh/A).

We approached this issue with field data in a savanna-like system, where we have analysed the interplay among cache protection strategies by small rodents, cache pilferage, acorn dispersal and oak recruitment. First, we found that seed-dispersing rodents showed a patchy distribution, and were concentrated under the canopies and resprouts of scattered oaks, what differentiated clearly an area of high rodent activity (under canopies, Sh) from other with a low rodent activity (outside canopies, Sl). Second, and as predicted theoretically, rodents consumed most acorns (> 50%) and pilfered more caches under the canopies (66.7%) than outside the canopies (22.2%). Consequently, changing from caching under canopies (Sh) to caching outside canopies (Sl) implies reducing cache pilferage by 77% (Ph/A >>> Pl/A). Third, we found that rodents cached acorns preferentially in the open areas surrounding the trees, where pilferage rates were lower. Also, all caches made by rodents under canopies were pilfered or recovered, so that the number of caches that remained uneaten under canopies (Nh) was zero. However, in the open sites there were several caches which were neither pilfered nor recovered (i.e. Nl > 0). This difference suggests that selection of the best sites to avoid pilferage can reduce recovery efficiency (i.e. there is a trade-off between P and N; see Fig. 1). Our results show that P>> Pl and that Nh Nl. We know that Nh < Nl because Nh = 0 and Nl > 0, but explicitly quantifying Nh and Nl involves differentiating in the field whether natural caches are recovered by the cacher or pilfered by other conspecific individual. Thus, caching in sites less frequented by conspecifics (i.e. Sl) can be a quite successful strategy to avoid cache pilferage, more efficient than caching where rodents are abundant (Sh) since seed losses for the cache owner (i.e. seeds pilfered + seeds not recovered) in the high activity areas are higher than in the low-activity areas (i.e. Ph NPl + Nl). In addition, it benefits the oak as well, since Nl Nh, which means that at least some acorns will not be recovered and will thus have the chance of establishment.

It is important to note that the seed dispersal behaviour in one acorn-fall season has important consequences for short-lived rodents and their lifetime fitness (Muñoz & Bonal 2008a). However, for long-lived oaks, even if only a few acorns are not pilfered or recovered each season (low Nl) and manage to establish, this number should be multiplied by numerous seeding seasons, which could imply a high overall fitness gain for the plant from those few ‘forgotten’ seeds. Moreover, this phenomenon could lead to the evolution of masting for some species if in mast years cached seeds are less likely to be retrieved (Kelly & Sork 2002). From the plant perspective it is also necessary to consider whether cache location fulfils other requirements for seed establishment (Wenny & Levey 1998; Wenny 2001; see Fig. 1). In our system, seeds cached in open sites increase the survival to pilferage by 77%, but also avoid other negative density-dependent factors under the mother trees (as suggested by Janzen 1970; Connell 1971). For example, ungulates in the study area can eat 100% of available seeds under the oak canopies (Bonal & Muñoz 2007). Accordingly, we found a higher abundance of natural seedlings and those that emerged from the experimental caches in open areas compared to the ones under canopies.

Our results also suggest that the model parameters are sensitive to inter-annual variability in acorn production and granivore population sizes, as the values of variables changed between seeding seasons. Pilferage rates were higher in the fall season of 2003–2004 (when rodents were more abundant and seeds more scarce) than in 2004–2005. This can explain why rodents followed a stricter strategy of caching outside canopies in the year when the probability of pilferage under canopies was higher. These inter-annual variations suggest that pilferage avoidance strategies are flexible behavioural patterns influenced by temporal variability of cache pilferage risk, which in our system seems to depend on the density of conspecifics and acorn production (see Kotler, Brown & Hickey 1999; Van der Merwe, Burke & Brown 2007; Schmidt & Ostfeld 2008; Hopewell, Leaver & Lea 2008). When competition is lower, seed cachers do not need to avoid the sites with high risk of pilferage, and this has subsequent consequences for seed dispersal and recruitment.

The caching decisions of seed-dispersing rodents are likely to be determined by rodent abundance and activity, which is higher under oak cover and generates a higher risk of pilferage. The mechanisms underlying these contrasting activity areas could differ between systems, and spending time in the low-activity areas may be costly for granivores due to a variety of factors such as poorer microclimatic conditions, higher risk of predation and lower food abundance. The higher abundance and activity (including seed predation) of small rodents under vegetation cover is often explained by the increased predation risk in open sites (Manson & Stiles 1998; Russell & Schupp 1998; Mohr et al. 2003; Matías, Mendoza & Zamora 2009). This tendency could explain the aggregation of rodents under trees and resprouts in our savanna-like system, which, in fact, are strongly aggregated under the cover of oaks all year-round, and not only in autumn when acorns are produced (Muñoz, Bonal & Díaz 2009). The costs of spending time in low-activity areas may be different for cachers and pilferers because, due to their spatial memory, cachers can adjust their behaviour to reduce the time spent in the open areas when recovering their own caches as compared to pilferers (Jacobs 1992; Vander Wall & Jenkins 2003; Vander Wall et al. 2006). This integrated information on ‘where’, ‘when’ and ‘how’ managed by cache makers is not available for potential pilferers, which would incur higher costs while searching for the caches of others. Potential pilferers may quickly abandon the search (see Pfuhl et al. 2009), thus explaining the lower pilferage rates in the open sites far from the protective cover of the tree canopy and resprouts. Pilferers, in turn, could use olfaction to detect the caches, thus reducing the costs of searching in open sites. However, most rodents are unable to detect the weak olfactory signals released by seeds when buried and the soils are very dry (Vander Wall 1998, 2000), as occurs in our arid Mediterranean savanna-like landscape. In addition, humidity is even lower in open sites than in the shady microhabitat under the canopies (Vetaas 1992; Manning, Fischer & Lindenmayer 2006), which could enhance pilferage efficiency under canopies.

Our results suggest that by caching the seeds preferentially outside the canopies of scattered trees seed-dispersing rodents reduce pilferage rates and also increase the survival of cached seeds. Significant reductions in the pilferage rates of cached seeds can allow the cacher to recover more seeds, but can also allow more seeds to remain dispersed and unrecovered, as was predicted by our model. Under such circumstances, granivores can provide a net benefit to the plant in terms of directed dispersal and can thus enhance recruitment. Moreover, in our system, rodents not only reduced pilferage of dispersed seeds by conspecifics, but also promoted the escape of seeds from many other negative density-dependent factors under the mother trees. It should be possible to apply the model presented here to other systems in which seed-dispersing granivores show aggregated distribution. This type of distributions can readily generate differences in the spatial distribution of pilferage risk and may potentially affect seed-caching patterns as well. It is quite plausible that reduced pilferage rates due to cache protection strategies lead to increases in survival of dispersed seeds and enhance recruitment, although the effectiveness of the strategy in reducing pilferage and the effectiveness of cachers in recovering their own caches need to be considered in different plant–granivore assemblages.


B. Nicolau and L. Arroyo helped during field work and the staff of the Cabañeros National Park provided all the facilities to carry out this study. M. T. Monaghan, J. M. Aparicio, J. M. Espelta, L. Muñoz, S. Rutherford and two anonymous referees provided helpful comments on the manuscript. This study received financial support from the projects REN2003-07048/GLO (MCYT), PAC-02-008 (JCCM), 096/2002 (MMA), Consolider – Ingenio MONTES (CSD 2008-00040) and PII1C09-0256-9052 (JCCM). A.M. and R.B. were supported by JCCM and ‘Juan de la Cierva’ contracts.