Multi-stemmed trees in montane rain forests: their frequency and demography in relation to elevation, soil nutrients and disturbance


  • Peter J. Bellingham,

    Corresponding author
    1. Landcare Research, PO Box 40, Lincoln 7640, New Zealand
      *Correspondence author. E-mail:
    Search for more papers by this author
  • Ashley D. Sparrow

    1. Department of Natural Resources and Environmental Science, University of Nevada, Mail Stop 186, 1000 Valley Road, Reno, Nevada, 89152, USA
    Search for more papers by this author

*Correspondence author. E-mail:


  • 1Multi-stemmed trees are a common component of tropical and temperate montane rain forests, where they co-occur with single-stemmed trees. We hypothesized that multi-stemmed architecture should predominate in less productive sites (such as at high elevation or with low soil nutrient availability) and that, because it frequently results from sprouting, it should prevail in sites that are frequently disturbed. We also hypothesized that where multi-stemmed architecture predominates, there should be lower rates of mortality and recruitment of individuals.
  • 2We tested these hypotheses using data from permanent plots in tropical montane rain forests in Jamaica (14 years) and in temperate montane rain forests in New Zealand (19 years). The frequency of multi-stemmed trees varied across plots in both forests (4–34% in Jamaica; 0–21% in New Zealand) along gradients of elevation and soil nutrients.
  • 3Analyses examined the direct and indirect effects of the various environmental variables that are likely to drive site productivity and disturbance. We showed contrasting predictors of the frequency of multi-stemmed trees in the tropical and temperate forests. Multi-stemmed trees in Jamaican forests were most frequent on sites of low soil nutrient status (especially available soil phosphorus, P) whereas in New Zealand they were most frequent at high elevation sites, especially towards tree line. In both forests there was no relationship between multi-stemmed tree frequency and slope.
  • 4Turnover (the mean of mortality and recruitment rates) of multi-stemmed trees in Jamaican forests was lowest on sites of low soil nutrient status (especially available P) but was unrelated to environmental predictors in New Zealand. In both forests, turnover rates of multi-stemmed trees overall were 60% lower than those for co-occurring single-stemmed trees, offering support for the hypothesis that multi-stemmed architecture favours persistence.
  • 5Synthesis. Our study demonstrates that multi-stemmed trees can predominate in low productivity sites in montane rain forests. Their low turnover compared with co-occurring single-stemmed trees confirms the importance of evaluating the persistence niche as a mechanism promoting woody species coexistence in forests.


Multi-stemmed trees are common in many forests (Johnston & Lacey 1983; Del Tredici 2001). Multiple stems originate at or near the ground (Givnish 1984; Bellingham & Sparrow 2000) and often arise by sprouting (Vesk & Westoby 2004; Fujiki & Kikuzawa 2006). Disturbance can promote multi-stemmed architecture by stimulating growth of existing or new sprouts after damage to existing stems (Peterson & Jones 1997); for example, after wind storms (Bellingham et al. 1994; Miura & Yamamoto 2003) and after fire (Bond & van Wilgen 1996). Disturbance is not a requirement for multi-stemmed architecture to develop (Johnston & Lacey 1983; Woolley et al. 2008) and, in the absence of disturbance, replacement of stems by new sprouts can benefit tree fitness through enhanced carbon gain or sexual reproductive output (Chamberlin & Aarsen 1996; Fujiki & Kikuzawa 2006). Whether promoted by disturbance or not, multi-stemmed architecture often enables persistence at a site (Pigott 1989; Bond & Midgley 2001; Del Tredici et al. 1992).

Multi-stemmed trees are particularly prevalent in tropical and temperate montane rain forests (van Steenis 1935; Midgley & Cowling 1993). Low productivity constrains maximum canopy height in many montane rain forests below the 12-m threshold above which multi-stemmed architecture is less structurally stable (McMahon 1973; Givnish 1984, 1995). Because height growth is relatively unconstrained in productive lowland forests, multi-stemmed architecture becomes less competitive because multi-stemmed trees are shaded (Midgley 1996). Multi-stemmed architecture in lowland forests is usually competitive in early succession where canopy heights are low (Lutz & Halpern 2006), but can be maintained in taller forests that are frequently disturbed such as floodplain forests (e.g. Cercidiphyllum japonicum maintains a sprout bank around a tall emergent stem in Japanese floodplain forests; Kubo et al. 2005). Multi-stemmed architecture can persist in lowland forest subcanopies when the canopies above are thin (Grier & Logan 1977) and also when they are intact, in which case multi-stemmed trees can trade off the smaller maximum height they attain against more rapid growth rates and earlier maturation (Kohyama 1992).

Dominance of multi-stemmed trees in low-productivity sites was predicted in a theoretical model (Bellingham & Sparrow 2000) and is consistent with empirical evidence for forests (e.g. Kruger et al. 1997). However, the model requires further testing because, in some moisture- or nutrient-limited shrublands and grasslands, the capacity to sprout (often a prerequisite for multi-stemmed architecture) predominates in relatively resource-rich sites (Clarke et al. 2005; Dalgleish & Hartnett 2006; but see Pausas & Bradstock 2007). There is little evidence to evaluate demographic patterns along environmental gradients in forests where multi-stemmed architecture is prevalent. Most demographic studies in these forests report the dynamics of stems only, even though many arise from the same root stock and are not independent (e.g. Tanner & Bellingham 2006). There are few demographic evaluations of multi-stemmed woody plants (e.g. Tanner 2001; Fujiki & Kikuzawa 2006; Alfonso-Corrado et al. 2007) and the significance of stems of sprout origin to plant demography is often speculative (Harper 1977). A recent review suggested that assessments of the demography of woody plant stems of sprout origin require evaluation over at least 5-year intervals (Cornelissen et al. 2003).

In this study, we used long-term data from tropical (Jamaica, 14 years) and temperate (New Zealand, 19 years) montane rain forests to examine the environmental drivers of dominance by multi-stemmed trees. Tropical montane rain forests and cool-temperate Southern Hemisphere rain forests are physiognomically very similar (Richards 1952; Dawson & Sneddon 1969) but our sites have contrasting disturbance regimes (see Methods). We present the first evaluation of mortality and recruitment of multi-stemmed and single-stemmed individuals in rain forests. Our aim is to determine whether there are consistent relationships between the dominance and dynamics of multi-stemmed trees, and site productivity and disturbance.

A conceptual model for multi-stemmed trees in montane rain forests

Bellingham & Sparrow (2000) predicted greater frequency of sprouting, and of multi-stemmed trees, in forests as productivity declines and frequency of disturbance increases. In a montane rain forest, productivity is linked to gradients of elevation (Kitayama & Aiba 2002; Moser et al. 2007) and frequency of disturbance to topography, especially slope (Nagamatsu & Miura 1997). A conceptual model (Fig. 1) outlines expected influences on the frequency of multi-stemmed trees in rain forests.

Figure 1.

Complete conceptual model to test hypotheses of drivers of dominance and dynamics of multi-stemmed trees in rain forests.

Elevation exerts a direct influence on forest productivity because temperature decreases with increasing elevation. On temperate mountains, declining temperature limits tree growth (Körner 1999; Körner & Paulsen 2004) such that at tree line forest stature is often dominated by stunted multi-stemmed architecture (krummholz). New Zealand tree lines are very low (c. 1200–1300 m) compared with continental sites at the same latitudes (c. 1800–3300 m; Körner & Paulsen 2004; Wardle 2008), thus forest stature reduces over a narrow elevational range. Declining forest stature with increasing elevation on wet tropical mountains, as in Jamaica, is driven by the combination of declining temperature and light, the latter linked to increasing cloudiness with elevation (Bruijnzeel & Veneklaas 1998). Increasing cloudiness (and decreasing light) is further likely to limit forest stature on some wet temperate mountains (Wardle 1986). Elevation also exerts indirect influences on forest stature. In particular, nutrient availability, especially N and P, declines with increasing elevation on wet tropical mountains (Grubb 1977) mediated by increasing precipitation. Slower rates of decomposition with increasing altitude also lead to increased soil C : N and declining pH (Grubb 1977). These indirect effects of elevation are likely to be most pronounced on tropical mountains. On temperate mountains nutrients are not necessarily limiting at tree line; disturbance history (notably past glaciation) is a major determinant of nutrient supply rates (Prescott et al. 1992; Tonkin & Basher 2001; Bowman et al. 2003).

As slope increases there is a higher incidence of ground–surface disturbances. Landslides occur on steep slopes in both the Jamaican and New Zealand sites (Dalling 1994; Reif & Allen 1988), but are more frequent in the New Zealand sites, driven by high rainfall events (Tonkin & Basher 2001) and periodic movements of the Alpine fault (Wells et al. 1999). Multi-stemmed architecture can be more prevalent as slope instability increases (Tang & Ohsawa 2002). Damage to trees on steep slopes caused by erosion of bedrock can promote sprouting; resources are remobilized from damaged above-ground parts to form new stems (Del Tredici et al. 1992; Sakai et al. 1997). Slope is also likely to exert indirect effects on the proportion of multi-stemmed trees by influencing nutrient availability. On ridge crests with little slope, soil organic matter accumulates and C : N increases; P supply can be particularly deficient (Grubb 1977; Richardson et al. 2008). On steep slopes, where primary successions occur, as on landslides, P availability is relatively high and N availability very low (Walker & del Moral 2003). Differential erosion of bedrock under forest canopies according to steepness of slopes and local deposition also results in different levels of soil N and P availability (Vitousek et al. 2003).

Other disturbances promote multi-stemmed architecture in montane forests, in particular sprouting after wind disturbances. Hurricanes affect Jamaican montane forests on average every 25 years and promote growth of sprouts to form new stems (Bellingham et al. 1994). In Jamaica, hurricane winds can affect forests on ridge crests more than those on slopes (Bellingham 1991) and this could result in a greater frequency of multi-stemmed trees on ridge crests. Less intense wind disturbances, including cyclones, affect New Zealand montane forests and many dominant trees sprout after wind damage (Martin & Ogden 2006), but unlike Jamaica, the return interval to most sites is > 100 years. Fire is part of the disturbance regime in some temperate montane forests; some species sprout in response and form multi-stemmed trees (e.g. riparian angiosperms in largely coniferous North American forests; Pettit & Naiman 2007). However, globally very few montane rain forests are affected by fire. Fire disturbance regimes in those forests are predominantly anthropogenic and have altered the forests’ composition (e.g. Corlett 1984). Fire is not part of the disturbance regime in New Zealand temperate montane rain forests (Ogden et al. 1998). Jamaican montane rain forests also lack a history of fire disturbance, and anthropogenic fires at their margins result in their destruction (McDonald et al. 2002).

As a step towards a comprehensive test of the model (Fig. 1), we tested some of the key components, that is the effects of elevation, soil nutrients, and slope (the latter a proxy for one component of the disturbance regime) on multi-stemmed architecture and the dynamics of multi-stemmed trees (Fig. 2).

Figure 2.

Models to test hypotheses for relationships between quantified environmental variables and (a) frequency of multi-stemmed individuals and (b) turnover (average of mortality and recruitment) of multi- and single-stemmed individuals.

  • 1Multi-stemmed architecture in rain forests will become more prevalent as productivity declines (Bellingham & Sparrow 2000). Multi-stemmed architecture will also be prevalent in sites subject to frequent disturbance (Nzunda et al. 2007a), such as steep, unstable slopes where multiple stems are likely to be a response to uprooting and physical damage (Sakai et al. 1997; Nzunda et al. 2007b).
  • 2Population turnover (the mean of mortality and recruitment rates) of both single- and multi-stemmed individuals will be lowest on the lowest productivity sites (Russo et al. 2005; Stephenson & van Mantgem 2005). In low-productivity sites, where we expect multi-stemmed architecture to predominate, there will be lower population turnover of multi-stemmed individuals than of single-stemmed individuals. If the turnover of multi-stemmed trees is lower than that of single-stemmed individuals in the same site (Bond & Midgley 2003; Zobel 2008), then this would give further support to multi-stemmed architecture being an expression of a ‘persistence niche’ (Bond & Midgley 2001).

Our approach to examining these hypotheses was to fit structural equation models (SEMs; Grace 2006) because this analysis is appropriate for examining the direct and indirect effects of the various environmental variables that are likely to drive site productivity and disturbance. In species-rich rain forests, a possible confounding influence is that the propensity of species to sprout or form multi-stemmed individuals is not uniform across species (Bellingham et al. 1994) and species are not uniformly distributed along environmental and disturbance gradients (Tanner 1977; Reif & Allen 1988). Therefore we also tested whether there are phylogenetic gradients that confounded our ability to test our hypotheses.


study sites

Tropical montane forests in this study were in the western Blue Mountains, Jamaica (18°05′ N, 76°38–40′ W). The geology of the mountains is complex, with at least eight rock types in highly faulted strata including schists, shales, amphibolites and, locally, limestone. Rainfall is c. 2800 mm year−1, with a relatively dry period during July–August (Shreve 1914). Composition of the lower and upper montane rain forests reflects elevation, variation in soils, and moisture gradients according to the prevailing trade winds, (Shreve 1914; Grubb & Tanner 1976). Primary successions occur on landslides (Dalling 1994) and secondary successions occur after hurricanes (Tanner & Bellingham 2006).

Temperate montane forests in this study were in the Whitcombe River valley on the western side of the Southern Alps, South Island, New Zealand (43°3–7′ S, 170°58′–171°3′ E). Geology in the site is schist and rainfall is c. 7000–8000 mm year−1, distributed evenly throughout the year with many high-intensity events. Forest composition changes with elevation, variation in soils with topography, and with disturbance. Primary successions on landslides and floodplains are commonplace (Reif & Allen 1988) and secondary successions occur after death of old-growth canopies (Bellingham & Lee 2006).

permanent plot measurements

We used data from permanently-tagged stems, all identified to species, ≥ 3 cm d.b.h. (diameter at 1.3 m) in permanent plots in both sites. In each plot, measurements were recorded for all tagged stems, including single-stemmed individuals and multi-stemmed individuals, that is, those that arose from a common rootstock and which were separate below the point of measurement (1.3 m). Thus we recorded diameter data, mortality and recruitment for stems and for individuals (both single- and multi-stemmed).

Jamaican tropical montane forests were sampled using 16 permanent 200-m2 plots measured in 1990 and 2004 (mean census interval 14.0 ± 0.02 year). The plots were placed at regular intervals, sampling the northern windward slopes, the crest, and the southern leeward slopes of the mountains within c. 2.5 km2 (1300–1900 m a.s.l.; locations mostly shown in Bellingham 1991). There were no major disturbances to the forests during the census period, but Hurricane Gilbert in 1988 had disturbed forests in the plots (Bellingham 1991) and this is likely to have affected demographic processes during the current study (Tanner & Bellingham 2006). The New Zealand temperate montane forests were sampled using 400-m2 plots established along random compass lines from valley floor to tree line: 23 plots along 8 lines (250–850 m), measured in 1980 and 1999 (mean census interval 18.9 ± 0.02 year). Tropical montane forests in Jamaica were located 1100 m higher in elevation than the New Zealand forests (Table 1) but the spanned elevation range was similar. The Jamaican tropical forests had higher tree diversity than the New Zealand forests (higher Fisher's α, Shannon H′ and higher β diversity; Table 1 and Table S1 in Supporting Information).

Table 1.  Soil and site characteristics, tree diversity, characteristics of multi-stemmed individuals and demography of individuals in tropical (Jamaica) and temperate (New Zealand) montane rain forests: mean ± SE (range)
 JamaicaNew Zealand
  • *

    Soil P = Bray-extractable P for Jamaica and Olsen-extractable P for New Zealand.

  • Calculated excluding three plots in Jamaica and nine plots in New Zealand where multi-stemmed individuals n ≤ 10 at t0.

Soil and site characteristics
 Elevation1680 ± 44 (1300–1920)560 ± 40 (250–850)
 Slope29 ± 3.5 (6–45)26 ± 3.0 (2–44)
 Soil pH4.2 ± 0.21 (3.1–5.7)4.1 ± 0.09 (3.3–5.2)
 Soil C : N13.8 ± 0.51 (10.6–17.4)20.3 ± 0.92 (13.8–31.6)
 Soil N (%)1.26 ± 0.09 (0.54–1.86)0.71 ± 0.11 (0.17–1.79)
 Soil P (mg kg−1)*24.4 ± 3.7 (10.0–54.0)18.4 ± 2.3 (8.8–50.6)
Tree diversity
 Alpha10.95 ± 1.23 (4.91–20.09)3.81 ± 0.22 (1.78–6.37)
 Shannon H′2.65 ± 0.09 (1.89–3.13)1.92 ± 0.22 (1.48–2.35)
 Beta (DCA length; half-change units)3.71533.3496
Multi-stemmed individuals
 Percentage of multi-stemmed individuals15.7 ± 1.9 (3.8–34.0)8.6 ± 1.1 (0–20.6)
 Stems/multi-stemmed individuals2.6 ± 0.07 (2.3–3.5)2.3 ± 0.07 (2.0–3.2)
Demography of individuals
 Mortality (% per year)1.90 ± 0.21 (0.92–3.20)1.56 ± 0.21 (0.77–3.82)
 Recruitment (% per year)2.09 ± 0.26 (0.69–3.59)1.95 ± 0.42 (0.16–5.69)

In all plots, we recorded elevation and slope, and sampled soils. Soils, including organic horizons, were sampled to 10 cm depth at a single random point in each Jamaican plot, and to 15 cm depth in each New Zealand plot (mean values from four random samples per plot). Soils were air-dried, sieved then analysed for pH, N, C and available P (Bray 1-extractable P from Jamaica, Olsen P from New Zealand).

data analysis

We computed the percentage of individuals that were multi-stemmed in each plot in both forests at the earliest census, and computed the mean number of stems per multi-stemmed individual in each plot. We carried out detrended correspondence analysis (DCA) of the basal area contributed by plant orders in each plot at the earliest census (using CANOCO for Windows version 4.54, ter Braak & Šmilauer 1997) to determine whether there were phylogenetic gradients along major environmental gradients (orders follow Heywood et al. 2007).

Mortality and recruitment rates of stems and individuals between censuses were determined for each plot in each forest. We also determined these rates for both multi-stemmed and single-stemmed individuals in each plot; 3 Jamaican plots and 9 New Zealand plots, where multi-stemmed individuals n ≤ 10 at the earlier census, were excluded from these analyses. Mortality as percentage is 100 (m = 1 − [1 − (N0 − N1)/N0]1/t), where N0 = the number at the earliest census, N1 = the number of survivors at the most recent census, and t = time in years. Recruitment as percentage is 100 (r = 1 − (1 − Nr/Nt)1/t), where Nr = the number recruited between censuses, Nt = the number at the most recent census. We also report annual turnover rates as the average of annual mortality and recruitment rates in each plot (e.g. Phillips & Gentry 1994; Stephenson & van Mantgem 2005).

Hypotheses were tested by fitting data to the conceptual model as SEMs (Grace 2006) using AMOS student version 5.0.1 software (AMOS Development Corp., 2003). Before analysis, variables were log-transformed as required for meeting assumptions of normality. Before fitting the SEMs, environmental data were tested for correlation (Table S2) and analysed by principal components analysis, separately for both forests, using S-plus version 8.0.4 software (Insightful Corp. 2007). Separate model predictions were made for abundance of multi-stemmed individuals (hypothesis 1; Fig. 2a) and turnover of multi-stemmed and single-stemmed individuals and the ratio of these turnovers (hypothesis 2; Fig. 2b). Analyses were conducted for all species combined. We also tested within-species patterns in the frequency of multi-stemmed individuals. Only two species, both in New Zealand, were sufficiently widespread and abundant (with n ≥ 10 individuals in n ≥ 10 plots at the earliest census) to conduct these tests. These were the canopy tree Weinmannia racemosa (Cunoniaceae) and the understorey tree Pseudowintera colorata (Winteraceae). No species was sufficiently frequent to examine within-species relationships in Jamaica. SEM pathways were forced simplifications of the conceptual model that excluded all latent (i.e. unmeasured, but inferred) variables because the relatively small number of plots in each forest constrained the total degrees of freedom available for model fitting. Models were fitted using maximum likelihood estimation and additional linkages were added to models where overall model diagnostics indicated significant covariance not accounted for by the preliminary model (Grace 2006; Table S2). For ease of comparison between forests, fitted SEMs were not further simplified by removal of non-significant relationships. SEMs are presented as diagrams in the same manner as in Vile et al. (2006), with the unstandardized regression coefficient and its significance level shown for each relationship, and the proportion of total variance explained (R2) given for the final response variable (i.e. frequency of multi-stemmed trees or turnover of trees).


multi-stemmed trees and relationships with elevation, soil nutrients and disturbance

Multi-stemmed trees were nearly twice as frequent in tropical montane forests in Jamaica as they were in temperate montane forests in New Zealand (Table 1). The frequency of multi-stemmed trees (%) was strongly positively related to elevation in temperate montane forests in New Zealand (all species combined, Fig. 3b) but was unrelated to elevation in Jamaican tropical montane rain forests (Fig. 3a). In temperate montane rain forests in New Zealand, multi-stemmed P. colorata trees increased in frequency with elevation (as predicted), but the opposite was the case for W. racemosa (Fig. S1).

Figure 3.

Structural equation models testing hypothesized relationships from the conceptual model for the proportion of multi-stemmed individuals in rain forests in (a) Jamaica; (b) New Zealand. Path coefficients between variables are un-standardized partial regression coefficients. Thick arrows indicate significant paths (P < 0.05); solid arrows are significant paths consistent with the model (Fig. 2a) and dashed arrows are significant paths opposite to prediction. Arrows not originating from a variable represent residual error variances and are the effects of unexplained causes. Intercept values are given in italics next to variable name boxes. The final response boxes contain the intercept and R2 value in italics. Variables marked with asterisks were log-transformed before model fitting. Overall goodness-of-fit for Jamaica: inline imageP = 0.345; for New Zealand inline imageP = 0.011.

The frequency of multi-stemmed trees was negatively related to Bray-extractable soil P in tropical montane forests in Jamaica, as predicted by the conceptual model (Fig. 3a), but in temperate montane forests in New Zealand the frequency of multi-stemmed trees (across all species) was positively related to soil Olsen P (Fig. 3b), contrary to prediction. Soil N and other soil variables were unrelated to the frequency of multi-stemmed trees in both forests. In New Zealand, relationships between the frequency of multi-stemmed W. racemosa trees and soil pH, C : N and Olsen P were consistent with our model predictions (although against prediction in relationship to soil N; Fig. S1). In contrast, multi-stemmed P. colorata tree frequency was positively related to soil pH and negatively to soil C : N ratio (both contrary to prediction). Soil C : N was lower in Jamaican tropical montane forests than in New Zealand temperate montane forests, and soil N% was greater in Jamaica (Table 1). Soil pH was similar in both forests. Available soil P was not comparable between the Jamaican and New Zealand sites because different measures of extractable P were used (Bray-extractable in Jamaica, Olsen in New Zealand).

Slope, a proxy for frequency of ground–surface disturbance, was similar in both forests and there was no relationship between multi-stemmed tree frequencies across all species combined and slope in either forest (Fig. 3). Of the two common New Zealand trees, the frequency of multi-stemmed W. racemosa trees increased with slope (consistent with the model) but there was no relationship in the case of P. colorata (Fig. S1).

phylogenies of multi-stemmed trees and relationships with environment

Plant orders differed in their percentages of multi-stemmed individuals in both forests (range in Jamaica 0.2–23.8%; New Zealand 0–22.2%; Table S1, Fig. S2). However, the percentage of multi-stemmed individuals in plant orders was not correlated with the DCA Axis 1 score of the basal area of orders in any of the forests (Jamaica r = 0.36, P = 0.22; New Zealand r = 0.32, P = 0.44; Fig. S2). We interpret this to mean that although different plant orders dominated along major environmental gradients in each of the forests, there was no phylogenetic bias in the frequency with which multi-stemmed trees occurred along the same gradient.

effects of elevation, soil nutrients, and disturbance on tree turnover

The turnover of both multi- and single-stemmed trees in tropical montane forests in Jamaica was positively related to elevation (contrary to prediction; Fig. 4), but as elevation increased, the turnover of multi-stemmed trees relative to single-stemmed trees in the same site declined (Fig. 4), consistent with our predictions. In temperate montane rain forests in New Zealand there were no relationships between tree turnover and elevation (single- or multi-stemmed trees).

Figure 4.

Structural equation model testing hypothesized relationships from the conceptual model for the turnover of multi-stemmed individuals in Jamaican montane rain forests (no significant relationships were found in New Zealand rain forests). Path coefficients between variables are un-standardized partial regression coefficients. Thick arrows indicate significant paths (P < 0.05); solid arrows are significant paths consistent with the model (Fig. 2b) and dashed arrows are significant paths opposite to prediction. Other SEM diagram annotations as in Fig. 3. Overall goodness-of-fit inline imageP = 0.159.

Turnover of multi- and single-stemmed trees was positively related to soil Bray-extractable P (as predicted) in tropical montane forests in Jamaica (Fig. 4). Turnover of multi- and single-stemmed trees in Jamaica was negatively related to soil C : N (as predicted), although the relationship for multi-stemmed tree turnover relative to that of single-stemmed trees increased, contrary to prediction (Fig. 4). In New Zealand temperate montane forests, the SEMs fitted poorly to tree turnover data: soil Olsen P and C : N had only weak positive effects on turnover of single-stemmed trees, both contrary to model predictions. Multi- and single-stemmed tree turnover was unrelated to slope in tropical and temperate montane forests.

Mortality and recruitment rates of individuals (combining single- and multi-stemmed individuals) were similar in the tropical and temperate rain forests (Fig. S3). Multi-stemmed individuals had much lower turnover (average of mortality and recruitment) rates than single-stemmed individuals in both forests (Fig. 5). Typically, turnover rates are reported for stems rather than individuals. In both forests, turnover rates were similar for stems and for all individuals (combining single- and multi-stemmed trees) because more abundant single-stemmed individuals diluted the contribution of slow turnover of rarer multi-stemmed individuals, although individual turnover rates were lower than stem turnover rates in Jamaica (Fig. S3).

Figure 5.

Annual mortality and recruitment rates of single-stemmed individuals (closed symbols) and multi-stemmed individuals (open symbols) in Jamaican (circle; n = 13 plots) and New Zealand (triangle; n = 14) montane rain forests. Values are means across plots ± SEM. In both forests, turnover rates (average of mortality and recruitment rates) of multi-stemmed individuals were less than those of single-stemmed individuals (paired t-tests; P ≤ 0.005).


multi-stemmed trees and relationships with elevation, soil nutrients and disturbance

The frequency of multi-stemmed trees in both the tropical and the temperate montane rain forests was driven by just one key environmental variable and one SEM pathway. However, the single environmental determinant differed between the tropical and the temperate forest. As elevation increased in temperate montane forests in New Zealand, multi-stemmed individuals were more frequent. This is consistent with our expectation that temperature (declining with elevation) rather than soil nutrients would be the principal determinant in temperate forests. It is also consistent with expectations of greater frequency of krummholz architecture as temperature limits forest growth near tree line (Körner 1999). This pattern was apparent for one of two common trees in New Zealand (P. colorata) and, similarly, multi-stemmed Nothofagus cunninghamii trees increased in frequency with elevation in cool temperate Australian rain forests (Johnston & Lacey 1983).

In contrast, elevation had no effect on the frequency of multi-stemmed trees in tropical montane forests in Jamaica. Instead, multi-stemmed architecture was more frequent as available P (mediated by pH and slope) declined. This is consistent with the theory that P, as well as N, limits productivity and forest stature on tropical mountains (Grubb 1977; Tanner et al. 1992, 1998; Moser et al. 2007). That the pathway is mediated by slope (Fig. 3a) is again consistent with general models of P availability in montane forests. On flat or gently sloping terrain, high precipitation and poor drainage lead to soil podsolisation and low P availability (Grubb 1977), whereas on steeper terrain, erosion exposes bedrock and thereby brings new supplies of P to the surface (Vitousek et al. 2003; Porder et al. 2007). Our results also suggest that in some temperate forests P supply can dictate the frequency of multi-stemmed trees at lower elevations. Multi-stemmed W. racemosa trees in New Zealand were most frequent on low P soils and at low elevation, the latter in contrast to the trend across all species combined. More investigations are needed in low-stature but less temperature-limited temperate forests to determine whether soil nutrient limitation, especially of P, determines architecture, for example across nutrient gradients or soil chronosequences (Wardle et al. 2004).

Multi-stemmed architecture can be a response to chronic disturbance regimes (Smale 1994). For example, on forests on unstable dunes in subtropical South Africa, 39% of individuals were multi-stemmed (Nzunda et al. 2007a), a much greater frequency than in either of the forests in our study. Multi-stemmed trees are often common on steep slopes, where many trees have a high sprouting capacity to counteract ground–surface disturbances and can obtain resources for sprouting from their above-ground parts that often survive disturbances (Del Tredici et al. 1992; Sakai et al. 1995, 1997; Nzunda et al. 2007b). However, slope was not a direct predictor of abundance of multi-stemmed trees in either the tropical or the temperate montane forest in our study; instead its effects were indirect (by mediating nutrient availability, see above). Our interpretation of slope as a driver of disturbance is hampered by our inability to discriminate between long-term frequency of disturbance and time since last disturbance, two variables that are not orthogonal and that are often confounded in non-experimental studies of disturbance ecology (e.g. Connell 1978). More detailed investigations of soil profiles or analysis of isotopes in our study sites could reveal time since large-scale primary disturbance in plots (e.g. Hewitt 1996), a superior measure of disturbance frequency and, in turn, perhaps a predictor of multi-stemmed architecture.

effects of elevation, soil nutrients, and disturbance on tree turnover

Our data from forests, where up to 34% of individuals are multi-stemmed, demonstrate that there is a small difference in reporting stand dynamics at an individual level vs. at a stem level (Fig. S3). In contrast to overall stand dynamics patterns, it is clear that turnover of multi-stemmed individuals was substantially less than that of single-stemmed individuals (Fig. 5). This result offers strong support for the hypothesis that multi-stemmed architecture in trees, especially when maintained by sprouting, enables long-term persistence of individuals (Bond & Midgley 2001; Vesk & Westoby 2004). Persistence after major disturbance has been shown in montane forests: a hurricane caused low turnover of individuals in the very nutrient-limited Jamaican Mor Ridge forest because most trees were multi-stemmed and death of these stems was offset by recruitment of newly sprouted stems (Tanner & Bellingham 2006). Multi-stemmed architecture results in lower population dynamism and greater persistence, but an implication of our study is that population dynamics are likely to be variable within species. Multi-stemmed architecture in two widespread trees in New Zealand montane forests ranged in frequency among plots (0–23% for P. colorata and 0–35% for W. racemosa), and this is also the case for other trees (Cao & Peters 1998; Nanami et al. 2004). Moreover, within species there can be genetic differences between populations that are predominantly multi-stemmed compared with those that are single-stemmed, thus multi-stemmed architecture in local populations may be reinforced by limited gene flow (Steinke et al. 2008). Because of variability within species, assignment of species traits with respect to persistence may not be as readily achieved in montane rain forests as in other systems where sprouting is a common response after disturbance, especially fire (e.g. Cornelissen et al. 2003; Pausas & Lavorel 2003; Klimešova & Klimeš 2007).

In generally P-limited tropical montane forests (see above), sites with low available soil P are likely to be among the least productive sites. Thus our finding that the lowest turnover of trees, both single- and multi-stemmed, in Jamaican montane forests is in sites with the lowest soil Bray-extractable P, is consistent with other studies which have shown lowest turnover rates in forests on sites of low productivity (Russo et al. 2005; Stephenson & van Mantgem 2005). Higher mortality of multi-stemmed trees in more productive sites, as in Jamaica (i.e. with high soil Bray-extractable P), is also consistent with studies which have shown that there can be high mortality, even of tree species that sprout, in high-productivity sites (Lutz & Halpern 2006). Overall, however, there was little support for our conceptual model linking multi-stemmed tree turnover to environmental variables and in the New Zealand temperate forests the variance explained was low. More data on turnover of multi-stemmed trees from other forests will be needed to determine links between persistence (Bond & Midgley 2001) and site productivity.

light and infrequent disturbances as influences on multi-stemmed trees

Our original model emphasized that site productivity is a key determinant of sprouting in woody plants (Bellingham & Sparrow 2000). In the test of the model for multi-stemmed trees in rain forests, we have only examined one component of site productivity – soil nutrients – as a driver of multi-stemmed architecture and turnover (Fig. 1). Light availability is another key component of productivity which probably explains some of the variance not accounted for in our tests of the model (Figs 3 and 4). Elevation can provide a coarse surrogate for light availability in temperate and tropical mountains because as elevation increases there is increasing cloudiness that results in lower light levels and dominance of diffuse radiation (Wardle 1986; Bruijnzeel & Veneklaas 1998). This oversimplifies matters because light competition is mediated by soil resource availability (sensu Grime 2002). Forest stature, leaf size and leaf area indices in tropical montane rain forests all decline with increasing elevation as a result of nutritional constraints (Grubb & Tanner 1976; Tanner et al. 1998; Kitayama & Aiba 2002; Moser et al. 2007), thus light levels can be higher below canopies on mountaintops with lowest soil nutrient availability, as is the case in Jamaica (Bellingham et al. 1996). Similarly, leaf area and stature decline with elevation and decreasing temperature towards tree lines on temperate mountains. While our model predicts that low light levels and reduced productivity should favour multi-stemmed architecture, Dunphy et al. (2000) proposed that high light levels promote multi-stemmed architecture in tropical dry forests. Therefore further tests of our model will require direct measurements of light availability in plots.

Disturbance is also a central driver of sprouting and multi-stemmed architecture in our original model (Bellingham & Sparrow 2000; Fig. 1). Chronic, low-level disturbance as a result of soil movement on steep slopes did not predict the incidence of multi-stemmed architecture in our two sites, but large periodic disturbances are probably influential. Hurricanes promote sprouting and favour persistence of multi-stemmed trees in some Jamaican sites (Tanner & Bellingham 2006). They also change resource availability, for example increased light levels for up to 2 years afterwards (Bellingham et al. 1996), which may alter competitive balances between growth of single-stemmed vs. multi-stemmed trees. Less frequent but larger-magnitude tectonic disturbances in both sites are likely to favour multi-stemmed architecture during early succession (Reif & Allen 1988) and also alter site productivity substantially (Walker & del Moral 2003). Periodic fire, too, is likely to drive multi-stemmed architecture in many other forests. Further tests of our model would benefit through incorporation of documented or modelled frequency and intensity of periodic disturbances at each plot.


We have shown that multi-stemmed architecture is driven principally by soil nutrient availability on tropical mountains and by elevation as a likely index of temperature on temperate mountains. It may be that lower light availability, because of higher elevation, has a stronger effect on multi-stemmed architecture than we have realised. Future tests of our model (Fig. 1) will require determination of the circumstances in which the frequency of multi-stemmed trees is related to light competition and/or low-temperature stress. This could be determined along latitudinal gradients and should consider forests where cold-temperature extremes are greater than those encountered in New Zealand forests (e.g. rain forests at high latitudes in South America, Barrera et al. 2000).

Multi-stemmed trees have lower turnover than single-stemmed trees and hence longer generation times. When longer generation times result in long-term persistence of species with poor litter quality and slow litter decomposition rates, then this is likely to affect the rate of carbon and nutrient cycling. For example, the montane Mor forests in Jamaica are dominated by multi-stemmed trees of species that produce litter that decomposes more slowly than that of nearby forests with fewer multi-stemmed trees (Tanner 1981; Tanner & Bellingham 2006) and in the Mor forests c. 50-cm deep, acidic humus accumulates with a soil C : N twice as that of soils in nearby forests (Tanner 1977; Tanner & Bellingham 2006). Longer generation times could also result in lower speciation rates in genera or families with high frequency of multi-stemmed architecture compared with shorter generation times for co-occurring genera with high frequency of single-stemmed architecture (Rohde 1992; see also Wells 1969). There is no evidence that this is the case in fire-prone Mediterranean shrublands worldwide; speciation rates of persistent sprouters, with long-generation times, are not different from those of seeders (Bond & Midgley 2003; Lamont & Wiens 2003). However, persistence through multi-stemmed architecture may be a reason for low levels of speciation in some genera in montane forests where the disturbance regimes are less severe and frequent than in Mediterranean shrublands. Monotypic genera in montane forests with high frequencies of multi-stemmed architecture include Cyrilla (Jamaica), and Davidia, Cercidiphyllum and Tetracentron (China; Tang & Ohsawa 2002). This is unlikely to be a general explanation because they co-occur with other species-rich genera that also have high frequencies of multi-stemmed architecture (e.g. Vaccinium in Jamaica and Prunus in China; Tang & Ohsawa 2002). Future studies could examine the genetic differences among co-occurring multi-stemmed and single-stemmed species in forests to ascertain the evolutionary consequences of these different life-history strategies (Bond & Midgley 2003).


New Zealand rain forest data derive from the National Vegetation Survey databank ( The Jamaican plots were re-measured in 2004 with funds from a Manaaki Whenua fellowship and from Gonville and Caius College, Cambridge (to Edmund Tanner). This study received financial support from the Marsden Fund, Royal Society of New Zealand and the Foundation for Research, Science and Technology (Ecosystem resilience OBI). We thank Dave Kelly for his support of the study. Sarah Richardson, Matt McGlone and Edmund Tanner made valuable criticisms of the manuscript.