The more stems the merrier: advantages of multi-stemmed architecture for the demography of understorey trees in a temperate broadleaf woodland


Correspondence author. E-mail:


1. Multi-stemmed trees are an understudied but common component of temperate broadleaf forests that can provide insight into how plants persist and regenerate within communities. In particular, multi-stemmed architecture may be an important trait for the growth and survival of trees in forest understoreys that have to cope with low light levels, but this idea has been rarely tested using long-term individual-level data.

2. We use measurements of 8527 individual woody stems from 1985, 1996 and 2008 to model the growth, survival and recruitment of hazel (Corylus avellana) and hawthorn (Crataegus laevigata and C. monogyna) trees as a function of neighbourhood competition in the understorey of a minimum-intervention, mixed-ash woodland in England. We test the effects of browsing by deer (Muntiacus reevesi) on woodland dynamics by comparing demographic rates during a period of high deer densities (1985–1996) with rates recorded from 1996 to 2008, when sustained culling substantially reduced deer densities.

3. Growth and survival of hazel and hawthorn trees increased with the number of stems they possessed, demonstrating clear benefits to multi-stemmed architecture. Surviving trees continued to accumulate basal area and stems without any clear upper limit after 23 years. However, increasing numbers of stems generally reduced the growth and recruitment of stems within multi-stemmed trees.

4. Temporal differences in deer browsing appeared to influence the strength of intra-specific (i.e. inter-stem) competition and, consequently, the growth, survival and recruitment of multi-stemmed trees. Most notably, stem survival declined with the number of stems in a tree only after deer culling, but not during a period of high deer densities, suggesting that intense deer browsing reduced resource competition among stems. Inter-specific neighbourhood competition had no detectable effect on hazel or hawthorn demography.

5.Synthesis. Multi-stemmed architecture is an advantageous trait for understorey trees in temperate woodlands relative to the allocation of resources towards the growth of a single stem. We suggest that the low light levels of forest understoreys favour ‘persistence’, through multi-stemmed growth, rather than ‘regeneration’ niches (i.e. periodic recruitment through seed), with the advantage of this life-history strategy influenced by herbivory and intra-specific competition.


Multi-stemmed trees are a common feature of many temperate and tropical forests (del Tredici 2001; Rackham 2003; Dietze & Clark 2008; Bellingham & Sparrow 2009; Poorter et al. 2010). Resprouting multiple stems from a common origin at or near the soil surface is thought to increase the persistence of trees, maximizing their long-term site occupancy and posing a contrasting life-history strategy to regeneration from seed (Bond & Midgley 2001). Other benefits may also be conferred by resprouting, e.g. minimizing respiring tissue per unit leaf area (Kobe 1997). However, competition for resources arising from variation in disturbance intensity and frequency, and/or site productivity, will influence the degree of resprouting (Iwasa & Kubo 1997; Bellingham & Sparrow 2000, 2009; Vesk & Westoby 2004). For example, where severe disturbances are relatively infrequent, such as within late-successional forests (Frelich & Reich 1999), above-ground competition for light will select for height growth by favouring single-stemmed trees (Midgley 1996; Bond & Midgley 2003). Specifically, tree height is maximized on productive sites by single-stemmed individuals because multi-stemmed trees have a larger crown area to support and thus require resources to be allocated to diameter expansion rather than stem elongation to prevent buckling (McMahon 1973; King 1981; Givnish 1984). Understorey shade-tolerant species are therefore more commonly observed to be multi-stemmed than trees that have to quickly reach the forest canopy to acquire light (del Tredici 2001; Bellingham & Sparrow 2009; Poorter et al. 2010).

Resprouting can provide an important mechanism to increase the accumulation of photosynthates and reproduction of tree species that remain in the understorey for the duration of their life (Aikawa & Hori 2006; Fujiki & Kikuzawa 2006). Light is often patchily distributed within the forest understorey owing to random small-scale gap creation, the transient nature of sunflecks and inter-specific variation in light transmission (Canham et al. 1990, 1994). Therefore, to have an increased probability of finding high levels of an unevenly distributed resource, plants can produce multiple stems growing in various directions, particularly if physical space in the understorey is not limited (de Kroon & Hutchings 1995). Clonal plants trade off growth within a single ramet for the production of additional ramets to forage for resource-rich patches, and multi-stemmed trees may function similarly (Bellingham & Sparrow 2000). The growth of stems along horizontal axes maximizes light interception, in part by forming a long shallow crown with little self-shading, and is favoured under the low light levels of forest understoreys (Aiba & Kohyama 1997; Poorter, Bongers & Bongers 2006). For clonal woodland plants, this ‘guerrilla’ strategy of widely spaced ramets that rapidly spread through the local environment increases the probability of growth into a sunfleck and hence survival, as opposed to the ‘phalanx’ growth strategy, where ramets form a tightly aggregated and slowly expanding patch that excludes other plants from competing for physical space and resources (Lovett-Doust 1981). However, these two strategies are extremes of a continuum, which has yet to be explicitly related to the persistence of multi-stemmed trees (but see Franco 1986; Coomes & Grubb 1998; Vaughan et al. 2007). Whether trees adopt a ‘guerrilla’ or ‘phalanx’ strategy may depend on the degree of competition in the local environment. Tree saplings have been found to adopt a guerrilla-like strategy under high resource levels, such that the spacing of plagiotropic branches along vertical stems increases with light availability (King et al. 1997; Coomes & Grubb 1998). In contrast, trees may be less likely to manifest a sprawling growth form, i.e. less resprouting, when inter-specific competition is high (Vilà, Weiner & Terradas 1994; Kabeya et al. 2003; Vesk & Westoby 2004).

Multi-stemmed trees may be as effective at competing and foraging for resources in the understorey as trees that invest in the growth of a single stem (Bellingham & Sparrow 2000; Bond & Midgley 2003), but top-down controls may be equally important in influencing tree persistence (e.g. herbivory, Kuijper et al. 2010). Under frequent disturbance that does not remove all available above-ground biomass (e.g. herbivory), resprouting is predicted to recover above-ground biomass rapidly (Bellingham & Sparrow 2000). Empirical data have supported this prediction by demonstrating that resprouting frequently occurs in small trees following simulated herbivory (Vesk 2006). Resprouting is maximized as a tolerance strategy in the presence of herbivores (Koop 1987; Herms & Mattson 1992; Rackham 2003) and would presumably allow trees to persist in the understorey without much evidence in the pollen record. Persistence in the presence of herbivores may be enhanced if (i) high densities of large stems protect younger and smaller stems from herbivores (i.e. ‘safety in numbers’, Fig. 1a; Ripple & Beschta 2005) and/or (ii) multi-stemmed trees function as distinct ‘stands’, whereby disturbance in the form of herbivory prevents competitive thinning from occurring among stems as trees increase in size (i.e. ‘competitive release’Fig. 1b; sensuCoomes & Allen 2007b).

Figure 1.

 Conceptual model of the growth, survival and recruitment of multi-stemmed trees in the presence of low (dashed line) and high (solid line) herbivore pressures. (a) Strictly a ‘safety in numbers’ effect, where high herbivore levels reduce demographic rates (i.e. lower intercept), but multi-stemmed trees are at a greater relative advantage than trees with few stems compared to when herbivore pressures are reduced (i.e. greater slope during high herbivore pressures). (b) Mutually inclusive ‘competitive release’ hypothesis, where herbivory prevents stems within multi-stemmed trees from reaching states of resource competition, which slow demographic rates in the absence of disturbance. Dotted grey line denotes number of stems in a tree at which competitive thinning commences.

Our objective was to model the influence of the number of stems in a tree on the growth, survival and recruitment of multi-stemmed trees in abandoned coppice woodland over a 23-year period. We expected that the multi-stemmed growth form was abundant for the dominant understorey trees at our site, Corylus avellana, Crataegus laevigata and C. monogyna, because it was associated with higher growth and/or stem recruitment rates, and/or lower mortality, compared with single-stemmed trees of the same species (e.g. Dietze & Clark 2008). The multi-stemmed growth form almost certainly originates primarily from coppicing up until 1920, but trees would have not persisted in that form for another 98 years without clear benefits to growth, survival and recruitment. Covariates of the number of stems in a tree, such as the basal area of trees and stems, and potentially plant age, make inferring causality difficult, but correlations between demographic rates and the number of stems in a tree have a strong theoretical basis. Specifically, if the mean annual change in basal area of a stem is g, the annual basal area growth of a tree with N stems would be equal to Ng. Tree growth will thus increase with N, if gains in basal area from additional stems are greater than reductions in stem growth arising from inter-stem competition (e.g. Fajardo & McIntire 2010). Similarly, if the mean probability of a stem surviving t years is st, where s is the annual probability of survival, the survival of a tree with N stems is 1 − (1 − st)N, increasing with N unless st strongly declines with N owing to inter-stem competition.

We also asked whether multi-stemmed trees are more tolerant of inter-tree (i.e. inter-specific) competition and herbivory than trees with few stems. The present number of stems in a tree at our site is likely influenced by competition for light and herbivory (Bond & Midgley 2003), as well as coppicing history prior to abandonment of traditional forest management (Peterken 1993; Rackham 2003). Our measurements span a period of woodland invasion by muntjac deer (Muntiacus reevesi), followed by a period of sustained deer population control. While other factors might have changed between our two measurements, e.g. climate or nitrogen deposition, deer are widely acknowledged as among important drivers of change in British woodlands over the 23-year time period of our study (Hopkins & Kirby 2007; Amar et al. 2010), including at our site (Cooke 2006), so we interpret differences between our two measurement periods as an effect of deer browsing.

We tested the following predictions:

  • P1. Growth of trees increases with the number of stems in a tree because of the additive growth of individual stems, which collectively exceed reductions in growth owing to inter-stem competition.
  • P2. Survival of trees increases with the number of stems in a tree owing to the multiplicative probability of survival of individual stems.
  • P3. Intra-specific (i.e. inter-stem) competition is low within multi-stemmed trees, as measured by the lack of a negative relationship between the number of stems in a tree and the growth, survival and recruitment of stems. This hypothesis is consistent with the ‘guerrilla’ strategy of clonal plants (Lovett-Doust 1981).
  • P4. Competition from neighbouring trees negatively affects growth, survival and recruitment of entire multi-stemmed trees and stems within multi-stemmed trees.
  • P5. Multi-stemmed trees better tolerate herbivore damage than single-stem trees of the same species, consistent with the ‘safety in numbers’ and ‘competitive release’ hypotheses (Fig. 1). We expected a steeper slope for the relationship between number of stems in a tree, and each of growth, survival and recruitment when deer densities were high, but a lower intercept, as deer browsing negatively affects demography.

Materials and methods

Study site

Our study site was Monks Wood National Nature Reserve, Cambridgeshire, England (157 ha, 52°24′N, 0°14′W; see Steele & Welch 1973; Mountford 2004 for details). The site is an ancient semi-natural mixed ashwood and mostly conforms to the W8 Fraxinus excelsiorAcer campestreMercurialis perennis community type in the British National Vegetation Classification (Rodwell 1991). Ash (F. excelsior) is the main overstorey tree along with scattered field maple (A. campestre) and oak (Quercus robur). Hazel (Corylus avellana) and hawthorn (Crataegus laevigata and Cmonogyna) dominate the understorey. Other minor woody species include blackthorn (Prunus spinosa), wild service (Sorbus torminalis) and common dogwood (Cornus sanguinea). The ground flora has changed over time from being characterized by dog’s mercury (M. perennis) to becoming dominated by graminoids, e.g. Brachypodium sylvaticum and Carex pendula (Crampton et al. 1998). Calcareous clay, loamy or gley soils (palaeosols) underlay the site and are mostly poorly drained except for the slightly elevated southern part of the site.

Monks Wood was traditionally managed as coppice with ash and oak standards until the wood was felled between 1914 and 1920, after which it was left to naturally regenerate (Steele & Welch 1973). The site attained protected status in 1953, and most of the site (91%, 143 ha) has been allowed to develop into closed-canopy woodland without intervention since 1920. Large oak and ash standards that were never felled remain in some areas.

Muntjac deer were first reported as present in about 1970 and established a resident population of c. 200 individuals by the mid-1980s (Steele & Welch 1973; Cooke & Farrell 2001; Cooke 2006). Concerns associated with the impacts of muntjac deer on woodland vegetation led to the introduction of stalking inside the wood in 1998. In the ten winters from 1998/99 to 2007/08, 48–106 deer were shot per winter, with a total of 821 being culled. Roe deer (Capreolus capreolus) and Chinese water deer (Hydropotes inermis) also occur within the region but at considerably lower densities (Cooke & Farrell 2001). Browsing by rabbits (Oryctolagus cuniculus) and brown hares (Lepus europaeus) may cause localized damage to low-lying ground flora (Cooke 1997, 2006).

Surveillance of deer population densities

We counted the number of deer observed at dusk (i.e. from 1 h before and after sunset) along transects totalling 3–8 km between January and May each year from 1986 to 2008. Transects varied in location to cover the whole wood reasonably uniformly each spring. Additionally, the mark–resighting method of Mayle, Peace & Gill (1999) was used to estimate deer density during 1998 and 1999 in a study area of 61 ha that included two of the four permanent forest transects described later. Briefly, eight deer were fitted with radio-collars, and the total deer population (N) was estimated as = 8(M + U + 1)/(+ 1), where M and U were the number of marked and unmarked deer observed during deer counts, respectively.

Permanent forest plots

Four permanently marked 180- to 270-m-long, 20-m-wide belt transects were established in 1985 (total area = 1.7 ha). The diameter at breast height (1.3 m; d.b.h.) of all live woody stems ≥1.6 cm d.b.h. along each transect was measured between June and August 1985, October and November 1996 and July and August 2008, and the spatial positions of stems were mapped. No stems that had a d.b.h. ≥1.6 cm were <1.3 m long, except for privet (Ligustrum vulgare), which we measured at ground level (0 cm height). Stems of multi-stemmed trees were denoted as originating from the same spatial coordinates. In 1996 and 2008, we also recorded the crown position of each stem as either canopy (emergent in the overstorey), subcanopy (mainly in the overstorey and partially overtopped by neighbouring trees) and understorey (entirely overtopped by overstorey trees). We also mapped the positions and sizes of canopy gaps, i.e. breaks in the overstorey, from ground-level observations in 1996 and 2008.

Demographic models of multi-stemmed trees

We fit models to predict the growth, survival and recruitment of individual multi-stemmed trees within a Bayesian framework. All models included two covariates: the number of stems in a tree and a traditional distance-dependent measure of neighbourhood competition, estimated as a function of the size and proximity of neighbouring trees (Bella 1971; Canham, LePage & Coates 2004; Canham & Uriarte 2006). By including the number of stems in a tree as a covariate, we tested whether the growth and survival of trees increased with the number of stems they possessed (P1 and P2). The inclusion of a measure of competition also allowed us to test whether competition from neighbouring trees negatively affected growth, survival and recruitment (P4). We tested whether the effects of model covariates differed between our two measurement periods, which corresponded with high and low deer densities, to determine whether the tolerance of multi-stemmed trees to herbivore damage differed from that of single-stemmed trees of the same species (P5). For all analyses, we treated the two separate hawthorn species (C. laevigata and Cmonogyna) as one, hereafter ‘hawthorn’, as hybrids occur frequently. The concept of two distinct species has been proposed to be irrelevant in south-east England as almost all populations lie along a continuum of introgression (Byatt 1975).

Our models do not consider tree age and size as factors influencing relationships between the number of stems in a tree and demographic rates. Traditional forest management would have coppiced all trees irrespective of their age (Peterken 1993; Rackham 2003) and so would have not created differences in the number of stems between young and old trees. Although tree age cannot be measured for our multi-stemmed trees (see Appendix S1 in Supporting Information for details), it unlikely to increase linearly with tree growth and survival (Ryan & Yoder 1997; Coomes & Allen 2007a) and thus will not provide an alternate hypothesis to explain any positive effects of the number of stems in a tree on growth and survival (as predicted by P1 and P2). A potential covariate of age, total tree basal area, is also not included in our models because it is closely related to the number of stems in a tree, which was our primary covariate of interest (Spearman’s rank test, ρ = 0.90 and 0.78, for hazel and hawthorn respectively, < 0.001). Inclusion of both of these variables in our models consequently introduced a high degree of covariation between parameter estimates. The largest stem in each tree was also positively correlated with the number of stems in a tree (Spearman’s rank test, ρ = 0.58 and 0.40, for hazel and hawthorn respectively, < 0.001). Irrespective of these correlations, exploratory analyses revealed that neither tree basal area nor the basal area of the largest stem in a tree strongly influenced variation in the number of stems in a tree (see Figs S1 and S2).


We tested whether the annual change in basal area of trees increased with the number of stems they possessed, by summing the basal area of all stems within each tree. We assumed that the net annual change in basal area (dB/dt) followed a power function that varied with the number of stems (N) in an individual and neighbourhood competition (C):

image(eqn 1)

where λ, θ and ξ are estimated parameters. Our neighbourhood competition index (C) for a focal tree considers the relative crowding effects of = 1…n neighbours within a 6-m radius (Canham, LePage & Coates 2004):

image(eqn 2)

where A is the basal area of neighbour i at the first measurement period and d is the distance of neighbour i to the focal tree. We also tested competition indices that were distance-independent, considered the size of neighbours relative to a focal tree, and only included trees larger than the focal tree (Tables S1–S3).

We fit the integrated (‘closed’) form of the growth function to our data because growth varies continuously and nonlinearly with size (Coomes & Allen 2007a):

image(eqn 3)

where B and B′ are the initial and final basal areas for individual i of species j over time period k, t is the time elapsed between measurement periods, v is the random effect associated with each individual i, such that viN(0, inline image), and inline image is the variability associated with individual trees. We modelled the coefficients λjk, γjk, θjk and ξjk with a different estimate for each species j randomly drawn from a normal distribution with a different mean for each time period k, e.g. λjkN(inline image, inline image). As net annual change integrates multiple stems, patterns of net basal area change can be obscured by stem mortality and growth of surviving stems. We thus used eqn 1 to separately model changes in basal area because of the growth of stems that survived between measurement periods and losses in basal area within each tree owing to stem mortality. We modelled increases in variance associated with larger basal area by allowing inline image to increase as a power function of B, where inline imageN(0, inline image + inline image) and δ is an estimated parameter.


We tested whether multi-stemmed trees had greater survival in the forest understorey. We observed the status (Sijk) of each tree i at the end of each measurement period (Sijk = 1 for alive or 0 for dead) and assumed SijkBernoulli (pijk), where pijk is the probability of individual i of species j being observed as alive over time period k. We assumed that p was constant among years as a tree with an initial size of B grew and could be related to the annual probability of survival, yijk, as (Sheil & May 1996):

image(eqn 4)

We modelled yijk using the logistic function:

image(eqn 5)

where α is the intercept, β and ϕ are estimated coefficients randomly drawn for each species from a normal distribution with a different mean for each time period (as for growth models), v is the random effect associated with each individual i (estimated as for growth), and ε is the normally distributed residual error. We standardized N and C to a mean of zero and standard deviation of one prior to model estimation.


There was recruitment of only seven and two trees between 1985 and 1996 and 1996 and 2008, respectively, and so, we did not model this process because there were insufficient data to conduct statistical analyses.

Stem-level demographic models

We modelled dynamics within each hazel and hawthorn individual by repeating tree-level analyses for each stem. Inclusion of the number of stems in a tree as a covariate allowed us to test whether there was competition within multi-stemmed trees, as measured by the direction of the relationship between the number of stems in a tree and stem growth, survival and recruitment (P3). Similar to tree-level models, models fit with a measure of neighbourhood competition as a covariate allowed us to test the effects of neighbouring trees on stem growth, survival and recruitment (P4).


We fit the integrated power function (eqn 3) to predict the growth of individual stems within each tree.


We modified analyses of survival (eqn 5) by treating stem survival as a binomial rather than Bernoulli process, where the number of trials was the total number of stems in a tree at the first measurement period and the number of surviving stems represented successes.


We treated stem recruitment as a random binomial process, where we observed T new recruits during each census interval. We assumed Tijk∼ B (Ωijk, Nijk), where Ωijk is the realized recruitment rate of individual i of species j over time period k, and Nijk is the number of stems at the end of a measurement period. Our observed data correspond to Ω as we only observed stems that recruited and survived the entirety of our measurement periods and not those that recruited and died before plots were remeasured in either 1996 or 2008. Ω can be related to the actual rate of recruitment (r) as Ωijk = 1 – (1 –rijk)t (Sheil & May 1996). We modelled r as:

image(eqn 6)

where a is the intercept, b and c are estimated coefficients, with parameters for each species j drawn from a normal distribution with a different mean for each time period k. As for survival, we standardized N and C to a mean of zero and standard deviation of one prior to model estimation. An empirical estimate of recruitment was calculated as 1 − [1 − (r/s)]1/t, where r is the number of recruits, s is the total number of stems at the end of the measurement period, and t is the length of the measurement period in years (Sheil & May 1996).

Estimation of parameters in demographic models

We fit all models within a hierarchical Bayesian framework using Markov chain Monte Carlo (MCMC) sampling by calling WinBUGS version 1.4.3 (Lunn et al. 2000) from R ver. 2.9 (R Development Core Team 2009) with the R2WinBUGS package (Sturtz, Ligges & Gelman 2005). Five MCMC chains of 200 000 iterations were simulated for each model, with a burn-in period of 200 000 runs. We modelled all regression parameters with relatively uninformative priors. We assumed the parameters in growth models, inline image and inline image, were ∼N(0, 10), and inline image, was ∼ ln N(0, 1) because it represented the intercept and could therefore only take on non-negative values. For all survival and recruitment models, priors for regression parameters were ∼N(0, 100). Variance parameters in all models were assigned priors that were ∼U(0,100), but preliminary analyses suggested we allow a narrower range for the exponent of the power variance function in growth models, where inline imageU(0, 10), to prevent the MCMC chain being stuck at unrealistically large values (Gelman & Hill 2007).

We estimated missing values of the competition index C using Bayesian imputation (Lunn et al. 2009), as competition effects were estimated only for a subset of our measurements, i.e. trees ≥6 m from plot boundaries (equal to 488 of 1368 and 1809 of 5173 observations for trees and stems, respectively). This approach allowed us to utilize our entire data set in analyses, rather than omitting trees for which C could not be calculated. We assumed that C values were missing at random, i.e. simply because a tree was near the plot edge and that there was no relation between whether C was estimated and any of our demographic measures. For growth models, missing values were drawn from a lognormal distribution with mean and standard deviation of a lognormal distribution fit to observed C values using the fitdistr function in the MASS package in R. We assumed CijkN(0, 1) for survival and recruitment models as observed values were standardized to this distribution prior to model estimation, and missing values were randomly drawn from this distribution. We used the cut function in WinBUGS to prevent imputed values of C from influencing model estimation (Lunn et al. 2009).

Model evaluation

For each parameter, we calculated posterior means and 95% credible intervals (CIs) by drawing a subset of 1000 simulations. We used these to test our predictions, such that the number of stems in a tree increased tree growth and survival if 95% CIs were greater than, and non-overlapping, zero (P1 and P2). We expected 95% CIs to overlap zero for the effect of stem number on stem growth, survival and recruitment, as predicted by P3, and 95% CIs to be less than, and not overlapping zero, for the effect of inter-specific competition (P4). We compared 95% CIs between the two time periods of our study to test for the effects of deer. The two time periods correspond with high and low deer browsing pressure, and we interpreted parameters with non-overlapping 95% CIs as significantly different across these two periods owing to deer (P5).

We used four approaches to evaluate the robustness of estimated models. Convergence was assessed visually by chain traces and by calculating the potential scale reduction factor, inline image, for each parameter from the 1000 simulation subsets. inline image predicts the extent to which a parameter’s confidence intervals will be reduced if models are run forever; all our inline image values were less than a threshold of 1.1, which is considered acceptable (Gelman & Hill 2007). We also ensured that the effective number of simulation draws, neff, a measure of the independence amongst the subset of 1000 simulations, always exceeded 100 (Gelman & Hill 2007). Lastly, model fit was summarized by calculating a Bayesian R2 at the level of our observed data, which is synonymous with the proportion of variance in classical linear regression (Gelman & Pardoe 2006):

image(eqn 7)

where E is the posterior mean, V is the variance, εk are the residual errors of the K observations, and yk are the predicted values for the response variable.

We also predicted mean tree growth and survival, and stem growth, survival and recruitment with increasing numbers of stems in a tree, and plotted these response curves against observed values. For all curves, the neighbourhood competition index was fixed at mean observed values for the respective species and time period. Observed values were plotted as empirical means averaged within size-class bins across both measurement periods, with each bin containing tree-level observations: c. 20 for tree growth, 60 for tree survival, 200 for stem growth and 50 for both stem survival and recruitment. Distributions of observed growth data had a long right-tail, and thus, we plot medians rather than means.


Overall patterns of woodland dynamics

Changes in the overall structure and composition of the wood from 1985 to 2008 were consistent with a maturing stand. The basal area of all woody species increased from 30.8 m2 ha−1 in 1985 to 33.6 m2 ha−1 by 2008, with the mean basal area per tree ± SE showing similar increases from 303 ± 12 to 483 ± 21 cm2. Ash was the dominant tree, accounting for 53% of total basal area in 1985 and increasing to 60% by 2008, whilst field maple and oak respectively comprised 16–19% and 10% of basal area over this time period. Hazel and hawthorn were always the dominant understorey species, accounting for 64% of basal area in the understorey layer in 2008. The overstorey remained largely closed with the coverage of canopy gaps changing little, from 9.5% in 1996 to 8.1% of transect area in 2008, and few gaps forming independently after 1996 (12%).

Multi-stemmed trees frequently occurred at our site, dominating the understorey in terms of both frequency of occurrence and basal area. We recorded 1077 multi-stemmed trees of a total 2181 trees. There was little change in their relative abundance over time (percentage of multi-stemmed trees in 1985, 1996 and 2008: 52%, 60% and 52%, respectively), likely due to the negligible levels of tree recruitment. Similarly, the mean number of stems in a tree was relatively constant (mean ± SE in 1985, 1996 and 2008, respectively: 4.5 ± 0.1 stems, 4.5 ± 0.1 stems and 4.2 ± 0.1 stems), suggesting that there may be little benefit to growth and/or survival in gaining more than five stems. Most multi-stemmed trees occurred within the understorey (739 of the 1077 trees with at least two live stems) and were primarily either hazel or hawthorn (36% and 62% of multi-stemmed trees, respectively). On average, multi-stemmed trees were larger than single-stemmed trees (mean basal area across all species ± SE: 423 ± 50 vs. 371 ± 39 cm2, respectively; Mann–Whitney test, U =3.2 × 105, < 0.001).

Changes in deer populations

The mean number of deer seen per hour in Monks Wood did not differ from a maximum of 22.5 in 1986 to 17.0 in 1998 (Mann–Kendall trend test for change in sighting frequency: τ = −0.03, = 0.903; Fig. 2). However, following the introduction of stalking in 1998, the mean number of deer seen per hour was significantly reduced to 2.1 by 2008, despite sampling intensity increasing (change in sightings between 1998 and 2008: τ = −0.63 = 0.006; Fig. 2). The mark–resighting technique similarly indicated that deer density declined from 1.1 deer per ha in 1998 to 0.4–0.7 deer per ha in 1999 following the initiation of stalking. Reductions in deer densities were associated with dramatic increases in the establishment of woody seedlings and suckers (height >30 cm but <130 cm) in our plots from 4 stems per ha between 1985 and 1996 to 657 stems per ha between 1996 and 2008, despite the canopy remaining largely closed.

Figure 2.

 Surveillance of muntjac populations at Monks Wood, England (1986–2008). Points denote mean number of deer seen per hour along surveillance transects ± SE, with number of surveillance visits increasing over time (grey bars, plotted on second y-axis). Broken vertical line indicates commencement of stalking in 1998.

Advantages of multi-stemmed growth (predictions 1 and 2)

There were clear benefits to being multi-stemmed in the woodland understorey, as predicted by P1 and P2. Gross annual change in tree basal area, arising from the growth of stems that survived between measurement periods, always increased with the number of stems in a tree, consistent with P1 (Fig. 3a,d; Table 1). Although the amount of basal area lost because of stem death similarly increased with stem number across measurement periods, net basal area change was positive during one measurement period for each species: 1985–1996 for hawthorn and 1996–2008 for hazel (Fig. 3; Table 1). For hawthorn, slower basal area growth in the absence of deer may have been related to stand development, i.e. greater basal area per unit woodland area between 1996 and 2008 slowed growth of hawthorn trees, but had no effect on hazel (Fig. 3f). Survival was also strongly linked to being multi-stemmed, such that increasing numbers of stems increased tree survival, supporting P2 (Table 1; Fig. 4). The mean number of stems of trees that survived was approximately twice that of trees that died (Fig. 5), and survival increased most over the range of 1–5 stems (Fig. 4).

Figure 3.

 Annual changes in basal area of multi-stemmed Corylus avellana and Crataegus trees (= 239 and 489, respectively) in Monks Wood as a function of the total number of stems in a tree (N). (a, d) Gain in basal area of trees owing to growth of stems that survived measurement periods; (b, e) loss of basal area within trees arising from stem death; and (c, f) net changes in tree basal area arising from growth, mortality and recruitment (which contributed minimally to annual changes). Curves denote mean model fit for periods of high (solid line, 1985–1996) and low deer densities (dashed line, 1996–2008). Points denote median change in basal area within size-class bins with bars representing inter-quartile range across both measurement periods (see text for details). *Effect of N significantly different between time periods (i.e. non-overlapping 95% credible intervals).

Table 1.   Credible intervals (95% CIs) for model intercepts and the effects of the number of stems in a tree (N) and competition (C) on tree and stem growth and survival, and stem recruitment, of Corylus and Crataegus at Monks Wood. Estimates for growth models represent exponents of respective effects in power function, with the exception of the model intercept. For survival and recruitment models, parameter values were standardized to a mean of zero and standard deviation of one prior to model estimation. Two sets of estimates are reported for each parameter, with values corresponding to high (1985–1996) and low (1996–2008) deer density periods, respectively. Bolded CIs are significantly different from zero, and italicized values are significantly different between time periods. Bayesian R2 values were calculated for each model at the observation level and thus summarize the proportion of variance captured by model predictions. Estimates of variance parameters are reported in Table S4
 R2Time periodCorylusCrataegus
Gross tree growth0.961985 to 19960.180.620.911.30−0.09 – 0.050.320.940.680.98−0.08 – 0.07
1996 to 20080.651.540.660.90−0.08 – 0.040.341.370.140.65−0.18 – 0.03
Tree basal area loss0.881985 to 19960.040.181.722.08−0.16 –−0.20 – 0.05
1996 to 20080.−0.16 –−0.19 – 0.03
Net tree growth0.811985 to 19960.181.28−0.69 – 0.33−0.11 –−0.12 – 0.12
1996 to 20080.341.500.290.74−0.09 –−0.22 – 0.12
Tree survival0.891985 to 19964.115.031.502.69−0.12 –−0.11 – 0.09
1996 to 20085.457.141.033.06−0.15 –−0.10 – 0.10
Stem growth0.961985 to 19960.441.11−0.17 – 0.04−0.07 – 0.050.340.940.24–−0.02−0.09 – 0.04
1996 to 20080.721.400.28–−0.13−0.06 – 0.030.591.340.45–−0.120.20–−0.09
Stem survival0.881985 to 19963.263.50−0.21 – 0.02−0.11 – 0.104.414.78−0.18 – 0.06−0.09 – 0.07
1996 to 20083.894.290.35–−0.11−0.09 ––−0.06−0.07 – 0.09
Stem recruitment0.501985 to 19964.33–−3.89−0.18 – 0.16−0.11 – 0.134.84–−4.480.34–−0.06−0.09 – 0.08
1996 to 20084.92–−4.420.49–−0.12−0.10 – 0.145.59–−5.130.47–−0.10−0.08 – 0.09
Figure 4.

 Mean estimated survival rate for trees of (a) Corylus avellana and (b) Crataegus in Monks Wood. Curve denotes mean model fit for periods of high (solid line, 1985–1996) and low deer densities (dashed line, 1996–2008). Points denote empirical survival averaged within size-class bins across both measurement periods (see text for details). Bars are 95% credible intervals, calculated for proportions using the approximation to the central limit theorem, where CI = inline image. = 321 and 543 for hazel and hawthorn trees, respectively.

Figure 5.

 Mean number of stems on (a) Corylus avellana and (b) Crataegus trees that died or survived in Monks Wood from 1985 to 2008 (= 321 and 543, respectively). Data from 1985 to 1996 and 1996–2008 were pooled together, and we plotted the number of stems per tree at the start of each measurement period.

Inter-stem competition reduces growth, survival and recruitment (prediction 3)

Stem growth and recruitment generally declined with increasing numbers of stems in a tree, suggesting that competition among stems was relatively high and providing no support for P3. Growth and recruitment of hawthorn stems always declined with the number of stems in a tree, likely because resources available for growth did not increase linearly with stem number (Fig. 6d,f; Table 1). For hazel, these rates only declined with increasing numbers of stems after deer densities were reduced, suggesting that greater stem survival after deer culling may lead to resource competition among stems (Fig. 6a,c; Table 1). In general, there was no indication of self-thinning or an upper limit to stem recruitment within trees as the number of stems in a tree did not consistently decline with increasing mean basal area of stems (see Appendix S1 in Supporting information and Fig. S3).

Figure 6.

 Mean estimated (a, d) stem growth, (b, e) stem survival, and (c, f) stem recruitment of Corylus avellana and Crataegus in Monks Wood. Curves denote model fit for periods of high (solid line, 1985–1996) and low deer densities (dashed line, 1996–2008). *Effect of N significantly different between time periods, i.e. non-overlapping 95% credible intervals (CIs). For (a, d), points denote median change in basal area within size-class bins, with bars representing inter-quartile range. For all other panels, points denote empirical means within size-class bins (see text for details), and bars are 95% CIs as calculated for proportions. All points and bars were calculated across both measurement periods. For hazel and hawthorn, respectively, stem growth: n = 1188 and 2323; both survival and recruitment: n = 239 and 489 tree-level observations.

Deer browsing likely controlled the strength of inter-stem competition (prediction P3) and its effect on stem survival. Stem survival did not decline with the number of stems in a tree during the period of high deer densities, suggesting an absence of density-dependent mortality in a period when foliage and stems were being removed and providing indirect support for P3 (Table 1; Fig. 6b,e). However, survival of stems of both species declined with increasing numbers of stems in a tree once deer densities were reduced (Table 1). This finding again suggested that deer browsing may prevent inter-stem competition. Estimates of tree survival predicted from stem survival were closely correlated with predicted values from tree-level data (Spearman’s rank correlation test: ρ = 0.90, < 0.001; see Appendix S1 in Supporting information for details; Fig. S4).

Absence of neighbourhood competition (prediction 4)

Neighbourhood competition did not affect any demographic rate, contrary to P4, with the exception of the growth of hawthorn stems (Table 1). When deer densities were reduced, the growth of hawthorn stems declined with increasing basal area of neighbouring trees, suggesting that deer browsing may have limited resource competition among trees (Fig. 6; Table 1). Models derived from other competition indices produced similar estimates of model parameters (Tables S1–S3; see Figs S5–S10).

Changes in demographic processes after deer culling (prediction 5)

Multi-stemmed trees showed some evidence of being able to better tolerate herbivore damage and responding negatively when deer densities were reduced. Slopes for the effects of the number of stems in a tree on tree growth and stem recruitment from four models significantly differed between measured periods, supporting prediction P5 (i.e. higher slope during high deer densities, Table 1). For example, hawthorn trees with many stems had larger net increases in basal area than trees with few stems when deer densities were high, but once deer densities were reduced, trees with few stems accumulated basal area as quickly as trees with many stems (Fig. 3f; Table 1). Hawthorn stem growth also declined less with the number of stems in a tree when deer densities were high, suggesting that deer may have reduced resource competition within multi-stemmed trees (‘competitive release’ hypothesis, P5). Additionally, the growth of surviving hazel stems and hazel tree and stem survival responded positively (i.e. higher intercept) to reductions in deer densities, consistent with prediction P5 (Table 1). For stem recruitment, however, model intercepts for both species were significantly lower following deer culling, suggesting that herbivores may interact with other processes that influence demography, e.g. intra-specific competition. Specifically, greater survival after deer population reductions potentially contributed to lower recruitment (Table 1). Mean predicted recruitment rates (±SE) were lower for 1996–2008 compared with those of the period between 1985 and 1996 for hazel and hawthorn, respectively: 2.0 ± 0.1% vs. 1.3 ± <0.1%; 1.3 ± <0.1% vs. 0.64 ± <0.1%. Recruitment was always exceeded by annual mortality for hazel and hawthorn stems between 1985 and 1996 and 1996 and 2008, respectively: 3.7 ± 0.1% and 2.0 ± 0.1% per year; 1.3 ± <0.1% and 1.7 ± 0.1% per year. However, despite the low and declining levels of recruitment over time, mortality did not consistently increase.


The big get bigger and the smaller die quicker

Our study demonstrates that multi-stemmed architecture is an advantageous life-history trait for growth and survival in the understorey of temperate woodlands, relative to trees of the same species with few stems. Consistent with prediction P1, gross tree growth increased with the number of stems in a tree because of the additive growth of individual stems, which exceeded density-dependent reductions in stem growth. Although positive responses of net tree growth depended on the amount of basal area lost owing to stem mortality, the survival of trees clearly increased with the number of stems they possessed, as predicted by P2.

The niche provided within temperate forest understoreys by patchily distributed light levels appears to favour resprouting, which is the primary mechanism for generating multi-stemmed trees (Koop 1987; del Tredici 2001). Feild et al. (2004) hypothesized that the earliest woody angiosperms grew in shaded, disturbed understoreys, and consequently, resprouting served as an important trait permitting recruitment in these habitats. Phylogenetic analyses have suggested that resprouting is an ancestral condition in angiosperms, which would have presumably been selected against in favour of more rapid elongation of a single stem as competition for light increased with the development of broadleaf forest canopies (Wells 1969; Bond & Midgley 2003; Feild et al. 2004). We therefore suggest that the regeneration–persistence niche continuum might actually be closely related to light availability and the canopy position of trees. For example, trees within forest understoreys might favour persistence strategies that allow regeneration once canopy gaps become available, whilst trees that reach the canopy should have relatively ample resources available to reproduce. These ideas also extend those of Kohyama (1987), which differentiated the regeneration niches of ‘pessimistic’ species that persist in the forest understorey awaiting gap formation from ‘optimistic’ species that invest in rapid upward growth to the canopy.

When you are your own worst enemy: inter-stem vs. neighbourhood competition

Declines in both stem growth and recruitment with increasing numbers of stems in a tree emphasize the influence of intra-specific competition on the demography of multi-stemmed trees, contrary to prediction P3. Competition within trees likely occurred for mineral nutrients as there were no concomitant declines in survival during high densities, which would be expected if carbohydrate reserves were limiting (Canham et al. 1999). Carbohydrate reserves are shared among stems of multi-stemmed trees (del Tredici 2001) and may be used to buffer against disturbances (e.g. herbivore defoliation) to increase survival (Kobe 1997). High levels of carbohydrate allocation to storage relative to photosynthetic tissue, a characteristic of multi-stemmed trees (Bond & Midgley 2001), will also consequently occur at the expense of growth (Kobe 1997). An additional explanation for the lack of a negative relationship between the number of stems in a tree and survival is that deer browsing prevents mortality from arising due to resource competition (sensuCoomes & Allen 2007b), and this is supported by declines in survival with the number of stems in a tree following deer culling. Experimental manipulations that compare the effects of intra-specific density dependence and carbohydrate remobilization from older to younger stems (e.g. Sakai & Sakai 1998) on the demography of multi-stemmed trees are clearly needed.

The lack of evidence for crowding from neighbouring trees (prediction P4) may be unsurprising given that physical space is abundant in the understorey relative to the canopy. Subcanopy and understorey gaps are common features of many forests (Connell, Lowman & Noble 1997), but their transient nature implies that as the resources they provide fluctuate, so too will the intensity of inter-specific competition. Although inter-specific competition was primarily weak, neighbourhood competition did negatively influence the growth of hawthorn stems, and we directly quantified the thresholds at which these responses occur – most notably where the summed ratio of basal area to distance from a focal hawthorn stem exceeds the observed mean of 751 cm2 m−1. This finding supports empirical observations that suggest that the management of large standard trees negatively influences the regrowth of coppice stools (Joys, Fuller & Dolman 2004).

Persistence of multi-stemmed trees despite herbivores

Multi-stemmed trees had higher demographic rates than single-stemmed individuals of the same species during periods of high herbivore pressures (e.g. hazel tree growth and survival), as predicted by P5. Once deer densities were reduced, we also found evidence that these rates declined with the number of stems in a tree. One potential explanation is that multi-stemmed trees can in fact function as distinct ‘stands’, whereby disturbances in the form of herbivory modulate the strength of intra-specific competition (i.e. generating ‘competitive release’, sensuCoomes & Allen 2007b). In contrast, we found little support for the idea that multi-stemmed trees protect young, small stems from herbivory and other disturbances (i.e. a ‘safety in numbers’ strategy against browsers, Ripple & Beschta 2005). Our results therefore suggest that interactions between herbivores and inter-stem competition can influence the persistence of multi-stemmed trees in woodland understoreys.

Our finding that hazel and hawthorn persist for >23 years beneath relatively closed-canopy woodland (≥90% cover in the canopy tier), and for decades longer after being overtopped by canopy trees, contradicts a popularized view of temperate forest succession. Specifically, the ‘Vera hypothesis’, which predicts that large mammalian herbivores control cyclic transitions from grassland to forest, depends on understorey trees being removed from closed-canopy woodland by competition for light and herbivory in order for woodlands to revert to grasslands (Vera 2002; Kirby 2004). Vera (2002) heavily relies on palynology to infer past vegetation history, arguing that the loss of understorey trees, such as hazel, from closed-canopy forest is supported by the fact that hazel seldom flowers and produces little pollen in shaded conditions. A major oversight of Vera’s (2002) model is that it focuses solely on regeneration through seed (i.e. the ‘regeneration’ niche) and neglects understorey tree persistence. While the shaded conditions of the forest understorey may limit growth and reproductive effort, our results demonstrate that trees can simply accumulate stems and grow larger within some types of forests (i.e. ash-maple woodland), leading to high numbers of multi-stemmed trees that persist rather than regenerate through seed. Persistence enables trees to exploit canopy gaps for reproduction, but this strategy, measured through growth and stem recruitment, is undetectable in palynological records. Hence, our findings do not suggest that closed-canopy forests lack a diverse and persistent community of understorey trees, but browsing by large herbivores can reduce densities of understorey trees within other closed-canopy forests (e.g. Mountford et al. 1999; Tanentzap et al. 2009, 2011; Kuijper et al. 2010).

Culling is frequently used as a management tool to reduce the impacts of deer on vegetation (Cooke & Farrell 2001; Tanentzap et al. 2009, 2011; Royo et al. 2010), but we know of no long-term data sets testing the effects of deer population reductions on the degree of competition within and among trees. Most studies of tree demography have focused on herbivore exclusion experiments, and these may have limited value for informing management actions (Hester et al. 2000). We found neither an increase in stem recruitment nor a consistent increase in stem survival across both study species following deer population reductions. Recent studies in other temperate forests have reported slow responses of forest size structures to much more intensive deer culling than reported in this study (Tanentzap et al. 2009, 2011). However, our results do provide a potential mechanism to help explain this general pattern, in addition to the slow and variable growth of trees, by suggesting that the re-initiation of density-dependent processes within stands and individual trees may impede expected rates of forest ‘recovery’ following deer culling. As such, we suggest that continued sustained culling can prove effective at reducing browse impacts and restoring ‘natural’ processes to forest stands for trees in the smallest size classes.

Disturbances aside from deer browsing are unlikely to influence demographic rates

We cannot discount the possibility that other disturbances might have changed between our two measurement periods in addition to deer, but this seems unlikely. Although changes in tree demography might have been influenced by severe weather events, wind damage was infrequent and canopy gaps remained scarce throughout our study (Mountford 2004). Drought appears to have had a similarly small effect on demographic processes during our study period, with only two minor events recorded in the summers of 1989 and 1990, and few noticeable effects on trees (Mountford 2004). However, contrasting patterns of growth between hazel and hawthorn trees, driven by differences in the amount of basal area lost due to mortality and gained by surviving stems, suggest that other factors might influence the demography of our two study species. Mountford (2004) remarked that several large hazel stems that had died or were noticeably decayed by 1996 may have suffered from earlier droughts (e.g. severe event in 1976) and/or infestation by honey fungus (Armillaria spp. as described by Rackham 2003).


Our study builds an understanding of why multi-stemmed trees persist within forest understories, by demonstrating clear benefits to survival, and to some degree, growth. These findings are consistent with theory that has predicted that multiple stems act to increase the persistence of individual trees (Bond & Midgley 2001, 2003) and may consequently provide an explanation for the prevalence of this life-history strategy in the shaded conditions of forest understoreys. While we did not find evidence that multi-stemmed trees strictly adopt a ‘guerrilla’ growth strategy through the forest understorey, competition may play an important role in influencing the extent to which this strategy is manifested (Lovett-Doust 1981; Vesk & Westoby 2004). Competition among stems within the same tree (i.e. intra-specific density dependence) negatively influenced demographic processes of the dominant understorey trees at our site, with stronger effects apparent after deer culling. As such, our results also point to inter-stem competition as a new mechanism to explain why deer-disturbed forest stands may be slow to respond once deer populations are reduced.


We thank George Peterken and Christa Backmeroff for having the foresight to establish the vegetation monitoring transects. We also thank Edmund Tanner, Richard Kobe, Peter Bellingham and an anonymous reviewer for generously providing helpful comments on an earlier version of the manuscript, and Elizabeth Cooke, Xuefei Yao, Lynne Farrell and Anne Mountford for fieldwork. Chris Gardiner and Natural England kindly provided logistic support.