Notice: Wiley Online Library will be unavailable on Saturday 27th February from 09:00-14:00 GMT / 04:00-09:00 EST / 17:00-22:00 SGT for essential maintenance. Apologies for the inconvenience.
1. Synchronous, episodic mast seeding is common in plant populations, and is thought to increase plant fitness through economies of scale, such as satiating seed predators, attracting seed dispersers and enhancing pollination success. Although mast seeding is easy to conceptualize, it has been quantified using a number of different metrics that reflect different features of pulsed reproduction.
2. We quantified spatio-temporal patterns of mast seeding across 36 populations of a high-elevation tree, Pinus albicaulis, for which perceived declines in cone production are a conservation concern. We tested for trends in mean cone production through space and time, and documented patterns of mast seeding using six different metrics: coefficient of variation, lag-1 autocorrelation, synchrony, average cone production by individual trees, and the frequency of high cone crops on absolute and relative scales.
3. Overall, we did not detect increasing or decreasing trends in cone production during our study period. Average cone production tended to increase from north-east to south-west. Population-level cone production tended to alternate between high and low years, but overall the coefficient of variation was low for a mast seeding species.
4. Metrics of mast seeding were not concordant across populations. The first principal component describing mast metrics separated populations with frequent high cone crops from those with high coefficients of variation. However, the second principle component was at least somewhat correlated with all metrics of masting, suggesting some ability to separate ‘masting’ from ‘non-masting’ populations.
5. In P. albicaulis, spatial variation in mast seeding could reflect differences in site productivity, differences in the importance of satiating generalist seed consumers versus attracting specialist seed dispersers, or recent invasion by an introduced pathogen.
6.Synthesis. Our research reinforces the conclusion that populations form a continuum of strategies between ‘masting’ versus ‘non-masting’ extremes. However, because different features of masting do not covary in space, understanding where populations fall along this continuum will depend on the features that are most important for mast seeding in a particular context.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Masting is fairly straightforward to conceptualize as a large, episodic pulse of reproduction (e.g. Taylor & Aarssen 1989). However, masting has proven surprisingly difficult to quantify. Across taxa, masting species have most often been defined as those where among-year variation in population-level seed production exceeds mean annual cone production (coefficient of variation > 1; Silvertown 1980; Kelly 1994; Herrera 1998). However, masting species have also been defined as those in which populations display negative autocorrelation in reproductive output (Koenig et al. 2003), or in which reproduction is synchronous among individual plants (Liebhold et al. 2004). In addition, ‘mast years’ have been variously defined as those where production exceeded the long-term mean by some pre-determined level (LaMontagne & Boutin 2009 and references therein) or where output surpassed an absolute threshold (e.g. McShea 2000).
In part, these different metrics of mast seeding arise because different features of masting affect different species interactions and may thus vary in importance across systems. For example, germination of dipterocarp seeds in mainland Malaysia and acorns in England may be limited by seed predation from rodents (Sun et al. 2007) and rabbits (Crawley & Long 1995) respectively. In these systems, absolute levels of seed production must be high enough to satiate local, generalist predators during mast years. But in Gunung Palung, Indonesian Borneo, absolute levels of dipterocarp seed production may be relatively unimportant since even large masting events that are discovered early by mobile seed predators suffer near complete predation (Curran & Leighton 2000). In this system, temporal variation in reproduction may be the most critical aspect of mast, so that masting events are rendered unpredictable in order to have a chance of escaping early detection. In boreal conifers, cycles in relative production may be more important than absolute production, to prevent the buildup of specialist seed predator populations in between masting episodes (Koenig & Knops 1998). If mast seeding increases pollination success, some absolute minimum density of flower production may be needed to attract animal pollinators (Crone & Lesica 2006), and some minimum density of pollen production may be needed to ensure wind pollination (Kelly, Hart & Allen 2001).
Any of these metrics are likely to capture differences among the most extreme cases of masting and non-masting species. However, there is no a priori reason to expect that these features change in concert across the continuum from masting to non-masting species and populations. Populations with the highest coefficients of variation are not necessarily those with the most frequent mast years on an absolute scale, nor are they necessarily those with most synchrony among individual trees (Buonaccorsi et al. 2003, Koenig et al. 2003). It is not clear which of these is the best metric of mast seeding in cases where ecologists want to pick up more subtle differences in masting among populations or species, such as differential responses to changing climate (McKone, Kelly & Lee 1998; Schauber et al. 2002) or species interactions (Schnurr, Ostfeld & Canham 2002). If the different features of mast covary strongly among species and populations, it matters little which measure is used in a given study; cross-study comparisons will be robust as the different measures are at least somewhat interchangeable. But if the features of mast do not covary consistently, comparative analyses will be more difficult, and it will be critical to identify the most appropriate measure of mast for each particular system.
Here, we assess patterns of cone production among populations of Pinus albicaulis Engelm. (whitebark pine), a widespread, but uncommon, conifer of western North America. Pinus albicaulis is widely referred to as a mast-seeding species (Tomback 1982; Mattson, Blanchard & Knight 1992, Kendall & Keane 2001; McKinney, Fiedler & Tomback 2009). However, many P. albicaulis populations are presently in decline, due to some combination of an introduced pathogen, white pine blister rust (Cronartium ribicola), increased frequency of bark beetle (Dendroctonus ponderosae) attack and climate change (Tomback & Resler 2007). Possible breakdown of mast seeding in this species has been a major concern for conservation policy; for example, a United States federal court ruled that grizzly bears in Yellowstone National Park were at risk of extinction due to effects of biotic and abiotic environmental change on P. albicaulis mast (United States Fish & Wildlife Service 2010). Therefore, it is particularly relevant to document spatio-temporal patterns of cone production in this species, in relation to commonly used metrics of mast seeding.
Specifically, we first tested whether average cone production differs through space and/or time, using a 20-year time series spanning 36 sites in western Montana, USA, and adjacent parts of Idaho, USA. We also compared spatial variation and covariation of six features associated with mast seeding: (i) variability in production at the population level, (ii) autocorrelation in population-level seed production, (iii) synchrony among individuals within populations, (iv) mean individual production, and the frequency of high-seed production years at the population level, measured both (v) relative to the time series of cone production for each site and (vi) in absolute terms. By comparing these metrics, we evaluated whether spatial variation in cone production leads to distinctly ‘masting’ versus ‘non-masting’ populations, or whether different features of mast seeding change independently across the landscape.
Materials and methods
Pinus albicaulis is a long-lived, stress-tolerant tree found in relatively cool climates of the Northwestern USA and Southwestern Canada, usually at high elevations with relatively low rainfall (Weaver 2001). Pinus albicaulis is an early successional species in much of its range and facilitates community development by mitigating an otherwise extreme environment. Trees reach reproductive age at about 20–30 years (McCaughey & Tomback 2001). Pinus albicaulis is monoecious; female cones take 2 years to develop, but plants are capable of initiating new cones every year. In Montana, male and female cones form as buds during spring (mid-June) of their first year, and wind pollination occurs in July, after which male cones senesce in early to mid-August (Weaver 2001). Female cones have visibly expanded by early July of their second year, and mature in late summer or early fall, i.e. c. 18 months after initiation. Cones produce large, non-winged seeds, rich in N, P and lipids (Lanner & Gilbert 1994). Like other stone pines, cones leave visible scars on tree branches; temporal patterns of P. albicaulis cone production can thus be observed from bud and cone scars (Forcella & Weaver 1986; Morgan & Bunting 1992; Nakashinden 1994). Pinus albicaulis cones contain large, nutritious seeds that are eaten by numerous animals such as grizzly bears (Ursus arctos), Clark’s nutcrackers (Nucifraga columbiana) and red squirrels (Tamiasciurus hudsonicus). The dominant, and possibly co-evolved, seed disperser of whitebark pine is the Clark’s nutcracker (Lanner 1982; Tomback 1982).
We sampled P. albicaulis at 36 sites across the northern US Rocky Mountains from 2004 to 2006. Sites were deliberately chosen to span a gradient of east and west aspects; prevailing winds from the west in our study region create maritime climates on west-facing slopes of ridges that run north-south and continental climates on east-facing slopes. Maritime climates have higher moisture and fewer extreme temperatures than continental climates. Therefore this sampling design partly decouples climate similarity from physical distance because sites on slopes with different aspects can be physically close but climatically different. In addition, we included sites on a rough gradient of time since blister rust invasion; in Glacier National Park and surrounding regions (the NW of our study area) mast seeding is believed to no longer occur due to blister rust infection (McKinney, Fiedler & Tomback 2009). In the Greater Yellowstone Ecosystem (the SE end of our study area) mast seeding was considered relatively intact throughout the period covered by our study (Kendall & Keane 2001), until the pine beetle outbreaks in 2007–2008. Along this gradient, sites were sampled opportunistically by identifying appropriate elevations that were accessible by 4WD vehicle and/or hiking and looking for P. albicaulis trees.
We sampled the largest climbable tree we encountered at each site and a subsequent sample of 1–12 of the largest climbable trees along a distance gradient from that first individual out to 5 km distant. Some sites (e.g. Glacier National Park) had low sample sizes (< 5 individuals) due to a dearth of live individuals. To verify that results were not due to sampling error from these small populations, we repeated analyses using only sites with ≥ 5 individuals, and obtained nearly identical results (E. Crone, unpubl. data). In each tree, we harvested 2–7 cone-bearing branches and estimated cone production over time by counting cone scars in relation to annual growth constrictions. We verified that two independent observers recorded perfect agreement of estimated years, and ∼5% error in the number of cones produced in a given year. We estimated total cone production in each year by counting the total number of cone-bearing branches on each tree, identified by their upright growth form. The estimated cone production per tree was then the product of the average cones per branch and the number of cone-bearing branches. As an index of population-level cone production, we also calculated the mean number of cones per tree in each year from our sample of trees in each population.
In 2007 and 2008, many stands in our study region were devastated by pine beetle (D. ponderosae) outbreaks. Our field surveys were completed in 2006, shortly before pine beetles became epidemic on whitebark pine in the northern US Rocky Mountains. The invasive pathogen, C. ribicola was present at low levels in all trees we sampled (E. McIntire, pers. observation).
We tested for increasing or decreasing trends of mean cone production through space and time using the estimated average number of cones produced per tree for each site. Analyses were conducted using linear mixed models (nlme package in R; Pinheiro et al. 2009) with year, latitude and longitude as fixed factors, site and Site × Year as random factors, and a lag-1 autoregressive (AR1) covariance matrix to account for the possibility that low years follow high years. Models were fit with a normal error distribution and identity link (verified by visual inspection of residuals; in contrast, use of a log link or removal of the AR1 covariance term both led to distinctly non-normal residuals). Independent variables were standardized to a mean of 0, so that the intercept term is an estimate of mean cone production. The initial model included all interactions of fixed factors; non-significant interactions were removed from the final model. Fixed factors were tested using likelihood ratio tests of models with each term removed versus the full model (i.e. marginal effects, analogous to Type II hypothesis tests in anova). We evaluated random effects and the autoregressive covariance term by comparing AICs of these models to models without these factors. See Bolker et al. (2008) for discussion of significance testing in mixed models.
We tested whether populations could be clearly described as ‘masting’ versus ‘non-masting’ by analysing the values of different metrics of masting, and quantifying their among-population covariance using Principal Component Analysis (PCA; princomp procedure in R, R Development Core Team 2009). This analysis included six metrics for each population: (i) annual variation in cumulative production by the population, measured as CVp = SD/, where is mean production at the population level and SD is the standard deviation of ; (ii) first-order autocorrelation, ACp, Pearson’s correlation coefficient of population-level cone production crop in consecutive years; (iii) synchrony, τ, Pearson’s correlation coefficient of cone production between individuals in the population (following Buonaccorsi et al. 2003); (iv) mean cone production of individual trees within each population (); (v) relative frequency of mast years, PSD, measured as the proportion of years where cone production exceeded the long-term mean by one standardized deviate (cf. LaMontagne & Boutin 2009); and (vi) absolute frequency of mast years, P36, where mast years are defined as those in which cone production exceeded 36 cones per tree (McKinney, Fiedler & Tomback 2009). We selected a threshold of 36 cones per tree based on work by McKinney, Fiedler & Tomback (2009); with data from three sites spanning our study area, they estimated that Clark’s nutcrackers were attracted to whitebark pine stands with ≥ 1000 cones ha−1. Based on the mean density of trees in our study area, this would be equivalent to a production rate of ≥ 36 cones per individual tree in order to constitute a mast year.
We reconstructed cone histories for 240 trees at 36 sites (Fig. 1). We climbed a median of 6.5 trees per site, with a range of 2–13 trees per site. We were able to reconstruct cone histories for a median of 20 years, with a range of 7–42 years.
Cone production by P. albicaulis was highly variable in space and time (Fig. 2). Trees produced an average of 23.4 cones year−1 (‘Intercept’ term in Table 1). Cone production tended to be higher in the south-east and lower in the north-west (latitude and longitude effects, Table 1), consistent with higher cone production in areas with drier climates and a shorter history of C. ribicola infection (Kendall & Keane 2001). Autoregressive mixed models indicated that cone production tended to be negatively autocorrelated over time within all populations (r = −0.15; Table 1); in other words, on average, high cone years tended to be followed by low cone years, as expected for a mast-seeding species. Overall, there were no trends toward increasing or decreasing cone production through time (main effect of year in Table 1). East–west differences in productivity declined through time (positive longitude by year interactions term in Table 1), although this interaction did not affect the direction of the response; the longitude effect evaluated for 1970 is −5.5, −3.6 for 1985 and −1.1 for 2005.
Table 1. Tests of spatio-temporal trends of cone production by whitebark pine (Coef – regression coefficient; SE – standard error)
Across sites, the average number of cones per tree ranged from 4.6 to 45.4 (mean = 23.4), and the coefficient of variation of mean cones per tree (CVp) at the site level ranged from 0.44 to 1.04 (mean = 0.76). Autocorrelation ranged from −0.59 to 0.54 (mean = −0.20) across sites; within-population synchrony ranged from 0.02 to 0.69 (mean = 0.35). Absolute production (P36) ranged from 0 to 0.57 (mean = 0.23); relative production (PSD) ranged from 0 to 0.25 (mean = 0.12).
The first principal component axis explained 34.6% of the variance in the data (Table 2b) and primarily separated populations with high variation in cone production (high CVp and PSD) from those with a high probability of having large cone crops (high P36 and high , Fig. 3). The second principal component explained an additional 31.9% of the variance and was strongly associated with high synchrony and negative autocorrelation versus low synchrony and zero or positive autocorrelation, and less strongly with high versus low CVp, PSD, P36, and . Because this second PC axis was negatively correlated with all of the features of masting, including being positively related to autocorrelation (masting populations exhibit negative autocorrelation), it could be seen as a single metric of relative tendency toward mast seeding based on all six metrics.
Table 2. (a) Pearson’s correlation coefficients between six features of masting and six principal component (PC) axes in Pinus albicaulis. (b) Importance of the six PC axes. CVP– coefficient of variation; ACP– lag-1 autocorrelation; τ– synchrony; = mean cone production; PSD– proportion of high-cone years (relative to site mean); P36 = proportion of years with > 36 cones per tree; PCx – principle component x
Proportion of variance
Cumulative proportion of variance
Cone production varies spatially among P. albicaulis populations. However, these differences do not correspond in a simple way to differences in mast seeding, as defined by all six metrics of masting. Instead, the first principal component in our analysis separated populations along a continuum with high CV of cone production and low absolute production ( and P36) at one end, and low CV with high absolute production at the other end. In other words, the populations with the most frequent large cone crops were not the populations with the most variable cone crops. The second principal component was associated with all six different measures, but it only explained one-third of the total among-population variance. This variation reinforces previous studies that have concluded that species cannot be grouped as ‘masting’ versus ‘non-masting’ (Herrera et al. 1998; Greene & Johnson 2004). Rather, the tendency toward masting varies continuously across species and populations (e.g. Koenig et al. 2003; Buonaccorsi et al. 2003). Similarly, in a study of seedfall by tropical trees in French Guyana, Norden et al. (2007) found that CVP varied somewhat between masting and non-masting trees (average CVs of 1.47 and 1.04 respectively), but that these ranges were broadly overlapping (range of 0.17–2.13 for masting species and 0.11–1.83 for non-masting species).
There are several possible, non-mutually exclusive, explanations for a lack of concordance between different aspects of mast seeding. First, in retrospect, a negative correlation of CVP with and P36 is intuitive algebraically. High CVp values occur in populations that have many years with zero reproduction and are much less influenced by the amount of reproduction in non-zero years. To illustrate this point, consider an heuristic example in which a population fluctuates between years with no cone production and years with XH cones that occur with some probability, p, leading to mean cone production of pXH. In this extreme example, CVP would depend only on p, not XH:
More generally, as CV is (by definition) normalized by mean cone production, it is relatively insensitive to reproductive output during years when reproduction occurs. Therefore, in systems where the benefits of mast seeding depend on some minimum absolute amount of flower or seed production, CVP would be a poor metric of mast seeding.
Second, researchers have often speculated that mast seeding is driven by the need for recovery of stored resources after depletion during masting events (Janzen 1974, formalized algebraically by Isagi et al. 1997). Intuitively, this hypothesis implies that more productive sites (e.g. those with higher ) should have less tendency toward masting than less productive sites. This expectation holds at global scales, in that tropical systems tend to have higher productivity and lower CVs of seed production through time (Kelly & Sork 2002; Wright et al. 2005). In the northern US Rocky Mountains, net primary productivity tends to increase from north-east to south-west, although these trends might not be evident at our specific sampled populations (e.g. P. albicaulis tends to occur near treeline, so climate may be relatively similar across populations). Mean cone production in our sites tended to increase from north-west to south-east (see Table 1), and the second principle component (our integrated measure of mast seeding) was correlated with latitude (r = −0.42, P = 0.019) but not longitude (r = 0.20, P = 0.293). Similarly, masting by rowan trees (Sorbus aucuparia) in Norway varies spatially, albeit not entirely in concordance with linear responses to differences in productivity (Satake & Bjornstad 2008). In that study, rowan trees had higher mean fruit production, lower CVs and longer inter-mast intervals in less productive, eastern sites than more productive, western sites (Satake & Bjornstad 2008; see their fig. 1c,d).
A third possibility is that decoupling of different features of masting in the P. albicaulis populations we studied could arise because different features affect fitness in different ways. For P. albicaulis, the key benefits of mast seeding seem to be attracting Clark’s nutcrackers for seed dispersal and satiating other consumers such as squirrels and grizzly bears (which consume seeds from squirrel middens) (Tomback 1982; Mattson, Blanchard & Knight 1992; McKinney, Fiedler & Tomback 2009). Relative production may be important for preventing numerical increase in generalist seed predators during non-mast years, while absolute production may satiate predators and/or affect the behavioural response of the nomadic seed dispersing nutcrackers during mast years (McKinney & Tomback 2007). Therefore, it may be that differences in the tendency toward high absolute cone production versus high variance in relative cone production reflect historical differences in abundance of nutcrackers, squirrels and grizzly bears. Such a contrast would not hold in other systems with less variable responses of consumers. For example, eastern oaks may be both dispersed and consumed by resident species (Ostfeld, Jones & Wolff 1996). Seeds of the highly masting dipterocarp trees of Southeast Asia are not dispersed by animals at all; masting here appears to be solely related to predation avoidance (Curran & Leighton 2000). Having to account for different spatial scales of attracting dispersers versus satiating predators could allow the decoupling of absolute and relative mast in our system but not in oaks or dipterocarps.
Finally, decoupling of features of P. albicaulis mast could also reflect responses to spatial variation in the introduced pathogen, blister rust. A few of our populations clearly did not mast, i.e. no (or positive) autocorrelation, low cone output and low CVp. These sites (small circles in Fig. 1) tended to be in the northern part of our study region, near Glacier National Park, where blister rust was first introduced and which now has the highest incidence of blister rust among our study sites (Kendall & Keane 2001). Therefore, our results support the well-established observation that blister rust tends to break down all aspects of mast seeding. However, if blister rust were the only factor affecting spatial variation in cone production, we would expect all the features of mast seeding to covary across sites. Blister rust is also thought to be intensifying in the southern part of our study region (Kendall & Keane 2001), which would presumably have led to significant declines in cone production through time, or a Time × Latitude interactions (both absent, although the non-significant interaction is in this direction, see Table 1). We lack detailed information on the dates of blister rust invasion in our study area, so it may also be that the interaction of longitude and time reflects differences in blister rust infection. It is tempting to speculate that this pathogen might differentially affect some aspects of mast seeding in the early stages of infection, in ways that could lead to discordance among measures of mast seeding, and maladaptation to current ecological conditions, rather than local differences in site productivity or local adaptation to different selective forces.
Like other mast-seeding species, P. albicaulis plays a key role in community dynamics by providing resource pulses for animal consumers (Ostfeld & Keesing 2000). Therefore, effects of environmental change on mast-seeding species may be particularly important for changes in community dynamics (McKone, Kelly & Lee 1998; Schnurr, Ostfeld & Canham 2002). From this perspective, our results provide the sobering message that different features of mast seeding may not change in concert. Simple measures of variation in seed output, such as the coefficient of variation, will surely distinguish clearly masting from non-masting populations, but they may not be a good measure of where populations sit along a continuum from historic (non-degraded) to degraded conditions. A more optimistic message would be that understanding of community interactions can guide selection of appropriate metrics for particular situations. CVp is one of the most commonly used metric of masting in the literature (Kelly & Sork 2002), but, in our system, this statistic is not a good predictor of ‘mast years’ as used in practice, i.e. large cone crops that might attract seed dispersers (McKinney, Fiedler & Tomback 2009) or prevent bears from looking for food in human-used areas (Mattson, Blanchard & Knight 1992). P36 and/or are probably the best metrics for these responses. In other systems, or for comparative analysis of mast seeding across populations or species, the most appropriate metric of mast seeding may not be any single measure, but will more likely depend on the functional relationships between masting plants and their interacting consumers, dispersers and pollinators.
We are grateful to J. Anderson, D. Thomson and especially J. Nowak for assistance with sampling. Our ideas benefited from discussion with A. Sala. This research was supported by an RJVA from the US Forest Service Rocky Mountain Research Station, a Natural Sciences and Engineering Research Council of Canada postdoctoral fellowship to E.M., and a National Science Foundation research grant (DEB 05-15756).