A comparison of 12 tree species of Amazonian caatinga using growth rates in gaps and understorey, and allometric relationships



1. The relative growth rate of saplings of 12 species from an oligotrophic lowland rain forest were measured in treefall gaps and understorey. Mean relative height growth (RH) within treefall gaps was found to be slowest for tall-tree species with branched saplings, intermediate for subcanopy trees and fastest for tall-tree species with unbranched saplings. Most species had similar RH within the understorey. RH values were not related to leaf mass per unit area (LMA) or foliar N concentrations.

2. Allometric relationships between the total leaf area (TLA) and height were dependent upon light conditions; in general saplings of a given height had a greater TLA in treefall gaps than in understorey. The species with the largest estimated TLA values in gaps tended to have the greatest RH values in gaps; no such trend emerged in the understorey. The values of the allometric coefficients were not related to foliar properties.

3. The relationship between stem diameter and height was only weakly dependent on light conditions and the relationship between the growth rates in these dimensions was also weak. The lack of plasticity may reflect the fact that the height–diameter relationship has little bearing on a sapling’s tolerance of shade.

4. One way of accommodating the dependence of allometry upon irradiance is to add RH as a covariate. We derive a relationship between growth rates from this resource-dependent allometric equation and show that it reasonably describes measurements taken in the caatinga forest.


There have been many attempts to link various types of tree architectures with other aspects of their ecology, such as shade tolerance and potential height growth (Horn 1971; Givnish 1978; 1984; Kohyama 1987; King 1990), but a simple paradigm remains elusive (Kohyama & Hotta 1990). Givnish (1984) argued that the unbranched habit is characteristic in the juvenile stage of more light-demanding, faster-growing species and Kohyama (1987) added the idea that shade-tolerant juveniles of this type are ‘optimists’, favoured by early formation of gaps, while highly branched juveniles are ‘pessimists’, favoured where gap formation is delayed. Givnish (1978) argued that compound leaves, especially pinnate leaves, are typical of light-demanding, fast-growing species and provided the rationale that they benefit from throwing away the axes giving immediate support to the leaflets at an early stage, so minimizing respiratory load. Popma, Bongers & Werger (1992) and Reich, Waters & Ellsworth (1992) have suggested that fast-growing, light-demanding species in tropical rain forest tend to have leaves with relatively low dry mass per unit area and high N concentration.

There have been very few critical tests of these ideas, at the level of seeking correlations between measured growth rates under various conditions and sapling form (O’Brien et al. 1995; Aiba & Kohyama 1996). Our first objective was to seek correlations between sapling growth rates (measured in gaps and understorey) and the branch form and leaf properties of 12 species from an Amazonian caatinga forest in southern Venezuela. Caatinga trees are short statured with narrow crowns and the forest type is known to be limited by a meagre nitrogen supply (Medina & Cuevas 1989; Coomes & Grubb 1996; Coomes 1997). The tree species of caatinga vary considerably in the measured growth rates of their saplings in natural gaps (Coomes & Grubb 1998) and in their branch and leaf form (Coomes & Grubb 1996), despite the fact that forest lacks completely the most light-demanding type of tree, characterized as ‘pioneer’ by Swaine & Whitmore (1988).

The architecture of saplings changes with ontogeny and the best way of following these changes is by fitting allometric relationships; these allow morphology to be examined from a dynamic perspective (King 1990). Species are compared by estimating the parameters bo and b1 in allometric equations of the form:

inline image

where x and y are dimensions of a plant (e.g. crown area and height). The common use of allometric relationships is grounded on the historical observation that biological relationships between x and y are independent of the supply of resources, often conforming to Huxley’s ‘allometric law’y = exp (bo) xb1 (see Huxley 1932). However, a cursory examination of trees growing in closed woodland and open fields reveals the marked differences in the shapes of their canopies, strongly suggesting that allometric parameters are not immutable descriptors of species but affected by the supply of resources (Coleman, McConnaughay & Ackerley 1994). Of course, we would expect the dependence of allometry upon the supply of resources to be more pronounced among plants than animals, because plants are known to be far more plastic in allocation patterns. As a second objective, we sought to contrast the allometric relationships of saplings growing in gaps and understorey. We hypothesized that species responsive to light availability would also be plastic in morphology.

Some of the 431 saplings described in this paper were used in a trenching experiment to test the effect of above- and below-ground competition on growth in height, branch length, stem thickness and leaf number (Coomes & Grubb 1998). These measurements allowed us to regress relative growth rates in height against relative growth rates in other dimensions. Theoretically, allometric relationships and regressions of relative growth rates should lead to similar conclusions about ontogenetic changes in sapling form, because differentiation of the allometric equation gives dy/ydt = b1 dx/xdt; we sought evidence for such a link. We consider how including a resource-dependence term into allometric relationships affects the growth rate equations.

Materials and methods


Coomes & Grubb (1996) have given a full account, from which the following summary is taken. La Esmeralda (3° 10’N, 65° 33’W, 99 m a.s.l.), on the Río Orinoco in southern Venezuela, had a mean annual rainfall of 2633 ± 86 mm in 1970–1986. Rainfall exceeds pan evaporation from April to November but not from December to March. The soils under the caatingas at La Esmeralda are albic arenosols, characterized by a mor humus layer overlying a bleached white-sand E horizon in which there are few fine roots. The soil is strongly acidic (pH 4·0 for 0–10 cm soil) and has a highly porous humus horizon (≈ 53 ± 9% air by volume). The soil surface is composed of hollows and hummocks around 1 m apart and 30 cm tall, and heavy rain brings the water table within 20 cm of the tops of the hummocks during the wet season. There is a paucity of species in all life-forms. We characterized 46 species of tree > 5 m tall, 14 shrubs and treelets < 5 m tall, nine herbs, nine climbers and eight epiphytes and one hemi-epiphyte. The total shrub density was moderate (2000 ha–1) but herbs were scarce (700 ha–1) and very small.


The species given in Table 1 may be categorized as tall trees (15–25m) with (1) branched saplings, (2) unbranched saplings or as (3) subcanopy species (5–15m). We use generic names to refer to all species except the two species each of Caraipa and Protium. Among the species in category (1) Caraipa longipedicellata and Iryanthera have many plagiotropic branches, while Eperua has few stubby branches that are non-deterministic in position. The subcanopy trees include saplings (0·5–2m) that are unbranched, orthotropically branched and plagiotropically branched, but all develop into mature trees with spreading canopies supported by long branches. In contrast, all of the tall trees listed, except Parkia, have characteristically narrow crowns when mature. All species in category (2) have compound leaves, while those in categories (1) and (3) are mostly simple leafed.

Table 1.  . Details of 12 Amazonian caatinga species, including height when mature, leaf form (S, simple; C, compound; P, pinnate), branch form (P, plagiotropic; O, orthotropic; U, unbranched) and the replication, leaf mass per unit area (LMA) in gaps (G) and understorey (U), foliar N concentrations (N), basal area of all trees >5 cm d.b.h. (BA) and sapling density (D) Thumbnail image of

We do not have sufficient data on persistence in deep shade to classify the species by this criterion but suggest that Eperua, C. longipedicellata and Iryanthera are very shade-tolerant, being common as seedlings, saplings and trees, and exhibiting reverse-J frequency distributions for height (Coomes & Grubb 1996).


Saplings were selected, within the understorey and in 12 treefall gaps in 3 ha of Amazonian caatinga, only if they were 0·5–2·0m height, >0·5 m from a mature tree, >2 m from another labelled sapling and near free of herbivore damage. Each sapling was marked with a ring at 50cm height on the main stem and in April 1993 the following measurements were recorded for all saplings: height from the ring to the leading apex, the length of each branch, the mean of two orthogonal measurements of trunk diameter immediately above the ring and the total number of leaves. Some of the saplings were trenched as a part of a study to determine the relative importance of above- and below-ground competition (Coomes & Grubb 1998). Square trenches (0·75 m × 0·75 m × 0·40 m deep) were cut around randomly assigned saplings and trenches were re-cut in December 1993. Saplings were re-measured in April 1994 to give estimates of relative growth rates in height, total length of all branches, net number of leaves and trunk thickness (RH, RB, RL and RT, respectively) using Rx = (ln x2– ln x1)/t, where x1 and x2 were the initial and final value of the measurements and the time, t, was 1 year. The increases in RH after trenching were small and not significantly different among species (analysis of variance in Coomes & Grubb 1998) and for these reasons we pool the RH values of trenched and untrenched saplings to give mean RH values in gaps and understorey.

Three fully expanded leaves (or leaflets of compound-leafed species) were sampled from each sapling in April 1994. The leaves were placed between sheets of newspaper, oven dried in a press at 80°C for 2 days and later measured with a digitizing area metre (Delta T Devices, Burwell, Cambridge, UK). Leaves were further dried after leaf-area determination and weighed to 0·01 g. The drying process caused some shrinkage of leaves (9% for 12 Eperua leaflets whose margins were traced onto paper prior to drying) but there was a close 1:1 relationship (r2=0·98) between values of leaf mass per unit area (LMA) recorded in this study and values recorded in a separate study using areas determined from fresh leaves (P. J. Grubb & D. A. Coomes, unpublished data). The mean area per leaf (MLA) was calculated in gaps and understorey for each species, and multiplied by the number of leaves on each sapling to give an estimate of its total leaf area (TLA); strictly total lamina area as we did not include the area of petioles or rhachides. Further details are given in Coomes & Grubb (1998).


A generalized linear modelling package (Glim 4·0, Royal Statistical Society, London, UK) was used to test the significance of terms of the allometric and growth-rate models outlined in the text. For a more conservative comparison, allometric parameters were determined by least squares linear regressions for saplings of each species in gaps and understorey. Standard errors of the slopes and intercepts were estimated and compared using standard methods given in Zar (1984: pp. 268–275 & 292–299). Harvey & Pagel (1995) suggest methods of regression that are appropriate when the ‘explanatory variable’ is subject to error but inaccuracy should be relatively small when ln H0 is used, because height measurement error (c. 2 mm) is small compared with the range of heights sampled (600–2000 mm). The error may be larger for relative growth-rate models, but alternative techniques could not be adopted because multiple regression methods were being used.



In gaps, three of the four species with the slowest height growth rates (RH) were tall trees with branched saplings, three of the four species with the fastest height growth rates were tall trees with unbranched saplings, while four of the five subcanopy trees had intermediate values (Fig. 1). The mean penetration of daylight into gaps was 5·5 ± 2·1%, estimated as a mean diffuse site factor (sensuAnderson 1964) on overcast days using 50 measurements from paired quantum sensors (Skye Instruments Ltd, Llandrindod Wells, Powys, UK). In the understorey, where only 1·7 ± 0·3% daylight penetrated, Eperua and C. longipedicellata had significantly slower RH than the other 10 species but otherwise there were no significant differences. The mean RH values of species in gaps were not rank correlated with leaf mass per unit area (Spearman’s R = 0·31, P = 0·32), nor with mean area per leaf (R = 0·17, P = 0·58), nor (for nine species) with the foliar N concentration (R = – 0·23, P = 0·56): the mean values are provided in Table 1.

Figure 1.

. Relative growth rates in height of saplings growing in understorey (unshaded bars) and gaps (dark bars), pooling data from trenched and untrenched plots (Table 6, Coomes & Grubb 1998): B, tall trees with branched saplings; U, tall trees with unbranched saplings; S, subcanopy trees.


Allometric relationships were affected by light conditions and adjustments were made by incorporating (1) height growth and (2) whether the sapling was found in gaps or understorey (i.e. a ‘light’ factor) into generalized linear models. Because growth rate and ‘light’ are highly correlated (mean RH in understorey and gaps 0·05±0·0034 and 0·15±0·0076, respectively), the significance of ‘light’ and RH depend greatly on the order in which they are entered into an analysis of variance. When terms are entered in the order given in Table 3, it can be seen that the effect of light is non-significant over and above that of RH. On the other hand, when ‘light’ is entered before RH it is highly significant, explaining nearly as much of the variance as the model given in Table 3. RH is marginally significant, even when added after ‘light’, but accounts for only an additional 1% of the variance. Clearly, it makes little difference whether we define the effect of light on allometry by simply stating whether a sapling is growing in a gap or in the understorey, or by including RH as a covariate. For the purposes of highlighting differences between caatinga species it is more convenient to give allometric relationships in gaps and understorey but in the section discussing the theoretical implications of resource-dependent allometric relationship, we use RH.

Table 3.  . Analysis of variance of the allometric relationships between stem thickness (To), number of leaves (Lo) and height (Ho). Interaction terms with P > 0·10 are not shown, nor are P-values > 0·05 Thumbnail image of

Total leaf area with height

Regression lines had common slopes in gaps and understorey but significantly different intercepts for 11 of the 12 species (Table 2). We back-transformed the allometric relationships to predict the total leaf area (TLA) of saplings of 200 cm height. Among the tall-tree species, those with unbranched saplings tended to have greater TLA values than branched saplings by 200 cm height in gaps (Fig. 2) but no pattern emerged among subcanopy trees, with the highly branched Caraipa punctulata having the largest TLA of any species and the sparsely branched Cybianthus having the smallest (Fig. 2). All five unbranched species (subcanopy and tall) had significantly greater TLA in gaps than in understorey but this was true of only the two Caraipa species among the seven branched species. The Caraipa species had a greater TLA principally because they had considerably larger MLA values in gaps than in understorey (Table 1), while the greater TLA of the unbranched species was caused by greater numbers of leaves in gaps. We also back-transformed the allometric relationships to calculate TLA values of saplings of 60 cm height but the values in gaps and understorey were not different and the rank order of species by TLA was not related to our classification; differences in TLA emerge only later in the ontogenetic process. The allometric intercept values were not correlated with the LMA values or foliar N-values given in Table 1 (P > 0·25 for all comparisons).

Table 2.  . Intercept values (b0) of allometric relationships between total leaf area and height (mean r2 0·45), and stem thickness and height (mean r2 0·86), in gaps (G) and understorey (U) for 12 species in an Amazonian forest Thumbnail image of
Figure 2.

. The canopy area in understorey (unshaded bars) and gaps (shaded bars) of saplings of 2·0 m height, estimated from allometric relationships. B, tall trees with branched saplings; U, tall trees with unbranched saplings; S, subcanopy trees.

Stem thickness with height

For all species, lines generated for saplings in gaps and understorey had significantly different intercepts but no significant differences in slope were detected (Table 2). However, back-transformation of the data suggested that allocation to height and diameter growth were not strongly dependent on light availability: the estimates of stem diameters were not significantly different in gaps and understorey, except for those of Dendropanax and Eperua at 60 cm, and Bombacopsis and Cybianthus at 200 cm, and even these species were not consistently thicker- or thinner-stemmed in gaps. The allometric intercept values were not correlated with the LMA values or foliar N-values given in Table 1 (P > 0·25 for all comparisons).

Branch length with height

There were no significant differences among highly branched species in allometric relationships between branch length and height, and the relationships were not different in gaps and understorey. The overall relationship was: ln B = 1·24 ln H– 0·97, with standard errors 0·27 and 1·3 for b1 and bo (r2 = 0·51).


Differentiation of the standard allometric equation leads to a directly proportional relationship between relative growth rates but we have shown that the allometric relationship depends on how fast saplings are growing, i.e.

inline image

and differentiation of this equation leads to:

inline image

In other words, the relative growth rate of y (RY) is expected to be a linear function of Rx, Rx2 and 1/x (d2x/dt2), the ‘relative’ rate of acceleration. Unfortunately, acceleration of sapling height is very difficult to determine, requiring the measurement of height on at least three occasions, with sufficiently long time intervals between samplings to ensure that inaccuracies of measurement account for little of the variance. It does, however, seem reasonable to assume that the principal factor affecting the acceleration is resource availability, in which case eqn 2 states that Ry is expected to depend on Rx, Rx2 and the availability of light and nutrients. In the following section we test whether these additional terms improve the fit of the relationships between growth rates.

When relationships between RT, RL and RH were investigated by generalized linear modelling, it was found that regressions containing Rx2 explained a significantly larger amount of variance than those containing Rx alone, exactly as predicted from eqn 3 (Table 4). Species identity and the availability of above- and below-ground resources (‘light’ and ‘nutrients’) affected intercept values but had no effect on slope, again as predicted by eqn 2. These improvements of models relating RH to RL somewhat justify the use of RH as a surrogate for irradiance in the allometric equations. However, saplings recorded in gaps may well have grown for many years in the understorey, and since we have based our relationships on the sizes of many individuals, rather than on the allometric growth of a single tree, we acknowledge that the result may not be general (Weiner & Thomas 1992).

Table 4.  . Analysis of variance of the relationships between relative growth rates, with terms added to the model in the order given. Interaction terms with P > 0·10 are not shown, nor are P-values > 0·05 Thumbnail image of

Ignoring species effects for the moment, the overall relationships for untrenched saplings in the understorey are:

inline image

inline image

The effect of growth in gaps is to increase the intercept value by 0·137 and 0·056 in eqns 4 and 5, respectively, while the effect of increased nutrient availability is to increase the intercept of eqn 4 by 0·292. These relationships, plotted in Fig. 3 (RH range includes 90% of values recorded in field), demonstrate that untrenched saplings in the understorey are expected to be far less leafy than those growing in trenched plots or gaps. It can be seen that untrenched saplings growing in the understorey have RH > RL, while saplings that are trenched or growing in gaps have RH < RL. This effect is not a case of etiolation, the short-term escape mechanism exhibited by many shade-intolerant species, because (1) all species persist in shade for long periods, (2) all have greater growth rates in gaps than understorey and (3) saplings are affected by nutrient supply as well as light. There were too few data to compare the effects of light and trenching for individual species but significant relationships between RH and RL were found for seven species when all treatments were considered together (Fig. 4).

Figure 3.

. Predicted relationships between relative growth rate in stem thickness (RT), number of leaves (RL) and height (RH), for untrenched saplings in understorey (no symbols), untrenched saplings in gaps (diamonds) and trenched saplings in understorey (squares).

Figure 4.

. Predicted relationship between relative growth rate in the number of leaves (RL) and height (RH) of saplings of seven species (acronyms given in Table 1).



Some time ago it was suggested that species producing apical leaves on long, steeply inclined petioles are most able to maximize upward growth in canopy gaps by (1) overtopping other saplings that are competing for light and (2) allocating assimilate to height growth instead of branch growth (Givnish 1978). The mean height growth rates of caatinga species in gaps support this paradigm, as the tall-tree species with unbranched saplings exhibited the fastest growth, tall trees with branched saplings exhibited the slowest growth, and subcanopy trees occupied the middle ground (Fig. 1).

In contrast, most species did not differ significantly in height growth in the understorey. Inevitably the field measurement of very slow growth has a large degree of error associated with it, but we believe that the lack of difference has an ecological basis. It may be that height growth in the understorey is less important than survival (Canham 1989). Persistence in deep shade has been shown to trade-off against height growth in gaps in two studies of tree species in Panamanian lowland rain forest (Hubbell & Foster 1992; King 1994) and in a study of tree species in North American temperate deciduous forest (Pacala et al. 1996); similarly there was a trade-off between persistence in deep shade and responsiveness of dry mass increment among European tall shrubs in a garden experiment (Grubb et al. 1996). The reason for the poor survival of the potentially fast-growing Panamanian species studied by King (1994) was that leaf life spans were too short to maintain a positive whole-plant carbon budget in the understorey. While we do not have sufficient data on persistence to pursue this point convincingly, it is notable that Eperua and C. longipedicellata and Iryanthera, reckoned on the basis of population structure to be strongly shade tolerant, had the slowest growth rates in shade and were among the least responsive to light (Fig. 1). Another reason for the lack of difference in deep shade may be that height growth depends upon the degree of branching; while branching initially leads to some reduction in allocation to height growth (as found for Acer saccharum by Bonser & Aarssen 1994), the lateral exploration of the understorey may later lead to optimal height growth in deep shade (as suggested from simulations of West Sumatran species by Kohyama 1991).


Caatinga tree species do not differ significantly in the slopes of their allometric relationships, the same result as that reported for saplings of comparable height ranges in a Japanese warm–temperate rain forest (Kohyama 1987) and in an Indonesian tropical lowland rain forest (Kohyama & Hotta 1990). We believe it is misleading to attach evolutionary significance to the apparent convergence of slopes among species, because our study was not able to accept the null hypothesis with P = 0·05. In fact, the observed differences in the slope of individual species give rise to an estimated fivefold range of stem thickness by 10 m height! We strongly suspect that the height range of 0·6–2·0 m used in these studies is insufficient to test for significance differences among slope parameters; indeed, significant differences were found among eight Panamanian tree species when a height range of 1–50 m was used (O’Brien et al. 1995).

The slope of the allometric relationship between total leaf area and height is lower for the caatinga species studied than for those of Japanese warm–temperate rain forest (Kohyama 1987) or Indonesian tropical lowland rain forest (Kohyama & Hotta 1990); the mean slopes are 0·66, 1·35 and 1·72, respectively. This result shows that the accumulation of canopy area with height is relatively slow in caatinga, agreeing with the observation that caatinga trees are generally small crowned when mature (Coomes & Grubb 1996). The back-transformed values of total leaf areas of 2·0 m saplings give values of 990–3300 cm2 for caatinga species, compared with values of 2400–3500 cm2 for warm–temperate rain forest and 2500–7200 cm2 for tropical lowland rain forest.


Oligotrophic forests are often dominated by thin-boled trees (Richards 1952; Tracey 1982; Coomes & Grubb 1996) and this is reflected in low slopes of the allometric relationship between height and stem thickness; the mean slope from RMA regressions was 0·70, compared with 1·45 for 375 species of dicotyledonous tree in Niklas (1994). Unlike the relationships between height and TLA, those between height and stem thickness were virtually unaffected by light. Furthermore, the relationships between growth rates were unaffected by trenching and only marginally affected by light (Fig. 3). The lack of plasticity may reflect the fact that the height–diameter relationship has little bearing on a sapling’s tolerance of shade or nutrient shortage but is very closely related to the tensile strength of the wood and resistance to wind damage (reviewed by Niklas 1994).


In general, species with potentially high rates of height growth (i.e. RH in gaps) were most able to develop large canopies in gaps. Evidence for this comes from (1) the correlation between total leaf area (TLA) and RH in gaps (Figs 1 and 2, Spearman’s R = 0·53, P = 0·075) and (2) the fact that all species have a common RH parameter in the allometric relationships (the S×RH term is not significant in Table 3), implying that the species with high mean RH will have large b3×RH terms in eqn 2. The inflexibility of slow-growing species to changes in light conditions has no apparent advantage but may be the inevitable consequence of adaptations to growth in extreme environments, as suggested for nutrient shortage by Chapin (1980).

Growth-rate relationships and allometric relationships should lead to similar conclusions about ontogenetic changes in sapling form but there are, in fact, many inconsistencies between the results of the two approaches. The first problem is that the RHRL relationships were non-significant for five species, presumably because the saplings increased in height only sporadically. For example, the leading meristems of Caraipa spp. and Eperua were observed to die back and then recommence height growth from subapical buds (cf. Hallé, Oldeman & Tomlinson 1978). Second, we would expect species with potentially high RL (i.e. species with high values of RL when RH is high) to have large TLAs in gaps, but no such correlation between these attributes was found among the seven species given in Fig. 4 (Spearman’s R = 0·25, P = 0·58). There may be several reasons why allometric and growth rate predictions do not agree at the species level, including (1) the indubitably large errors associated with the measurement of slow growth rates and (2) the fact that plants recorded in canopy gaps may have developed in the understorey for many years. More encouragingly, the convergence of RL values at low RH does agree with the fact that TLA values are more similar in understorey than in gaps and the unimportance of nutrients or light on RHRT relationships does concur with the similarity of allometric equations in gap and understorey.


All tall-tree species with unbranched saplings had compound leaves, in support of Givnish (1978) who argued that compound leaves are typical of faster-growing species. The fact that Eperua, one of the slowest-growing species, also has pinnate leaves indicates that the rule is by no means ubiquitous. We found no correlation between LMA or foliar N concentration and the growth rates, or the allometric coefficients, of caatinga saplings. For a wide range of vegetation types, light-demanding, faster-growing species tend to have lower LMA and higher foliar nitrogen concentration (Reich et al. 1992), reflecting a low investment in generalized protection against physical and biological hazards (Grubb 1984; Turner 1994). These traits are often associated with shorter leaf life and higher maximum rates of net assimilation (Reich et al. 1992). The fact that we found no relationship between growth rates and leaf properties, despite finding that LMA is correlated with N (r = – 0·72, t = – 2·8, P = 0·03), reflects the importance of allocation and plant architecture in determining growth rates. It may be for this reason that other studies in single communities have not revealed the trends found in large-scale comparisons (cf. Reich 1993).


Farnsworth & Niklas (1995) used a simulation to show that a variety of branching arrangements with similar fitness may evolve when natural selection is driven by more than one factor. If growth and survival of saplings have a strong effect on the overall fitness of species (as argued by Grubb 1977) then multifactorial selection for the following criteria may be important in explaining the diversity of architecture among shade-tolerant species: (1) ability to compete with neighbouring saplings, (2) ability to respond to lateral as well as vertical irradiance and (3) plasticity in the initiation of branching.

1. Competition among saplings

Competition for light is strongly assymetric (Weiner 1990) and species that produce apical leaves on long, steeply inclined petioles are likely to have a competitive advantage. For example, in Japanese subboreal forest, widely spaced saplings of the conically-crowned Picea jezoensis are faster growing than flat-crowned Abies sachalinensis, but Picea is overtopped and outcompeted by Abies in dense stands (Kubota & Hara 1996). We have suggested elsewhere that juveniles are weakly affected by competition with neighbouring juveniles in caatinga but are strongly affected by the impact of mature trees on the availability of nitrogen and light (Coomes & Grubb 1998). Another indicator of weak above-ground competition among saplings in the understorey is that their total leaf area per unit area of ground is estimated to be only 0·49–1·1. This estimate was calculated by multiplying a mean sapling density of 6 m–2 (Coomes & Grubb 1996) with the mean TLA values at 60 and 200 cm height for C. longipedicellata and Eperua, which account for 75% of all saplings.

It might be thought that competition for light by unbranched saplings would inhibit the canopy expansion of plagiotropically branched species. The fact that both Caraipa species had substantially larger TLAs in gaps than in the understorey provides further evidence that competition with surrounding saplings is weak (Fig. 2). We would expect competition between saplings to become important in the centres of large gaps, where dense stands of saplings are commonly found, but such gaps are rare in caatinga forests (Coomes & Grubb 1996). It is the four tall-tree species with unbranched saplings that are most likely to benefit from large gaps and it is notable that these species had significantly larger TLAs in gaps than in the understorey.

2. Plasticity in the initiation of branching

The effect of light on the height at which saplings initiate branching may depend strongly on whether a species develops into a canopy or subcanopy tree. Saplings of subcanopy tree species were mostly unbranched in the understorey (37 of 52) but mostly branched in gaps (31 of 51). This contrasts with the fixity of unbranched saplings of taller-tree species, none of which initiated branching until a height of 4–6 m. Similarly, King (1996) reports that subcanopy trees of 6–15 m height are wider-crowned than tall-tree species of comparable height in a Costa Rican lowland tropical rain forest.

3. Responsiveness to lateral light

The influence of a gap may extend far beyond its physical boundary, partly because saplings can respond to the lateral fluxes of light that permeate the understorey (Popma et al. 1988). Species with deep canopies (e.g. conical forms) are able to intercept a large proportion of lateral radiant flux and we speculate that C. punctulata is responsive to gaps because its hollow cone arrangement of leaves allows it to capitalize on lateral light at gap edges.


Of the tall-tree species studied, those with unbranched saplings have faster height growth in gaps than those with branched saplings. Fast growth in gaps did not trade-off against fast growth in the understorey but we were unable to test whether it traded-off against persistence. Generally, species with fast height growth in gaps had the largest total leaf areas at 2·0 m in gaps but we have little idea about the causes of differences in flexibility. Because plants are able to change allocation patterns in response to the availability of resources, allometric parameters are not immutable descriptors of species. Future studies need to document the flexibility of the allometric parameters of different species more closely, and interpret the finding in terms of growth vs storage (e.g. Pacala et al. 1996), and in terms of flexibility in the life spans of organs, rates of gas exchange and allocation of resources.


We thank the staff of Servicio Autonomo para el Desarrolo Ambientale del Amazonas (SADA-Amazonas) for logistic support, the Yek’wana villagers of La Esmeralda for assistance throughout the project, and Professor E. Medina at Instituto Venezolano de Investigacíones Científicas for advice and facilities. The work was funded by a research studentship from the Natural Environment Research Council with additional support from Gonville and Caius College. Dr O. Huber provided helpful suggestions during the field work, and Dr T. Kohyama provided useful comments on the manuscript.


  1. Present address: Landcare Crown Research Institute, PO Box 69, Lincoln, New Zealand.