Effect of wood density and water permeability on wood decomposition rates of 32 Bornean rainforest trees

Abstract

Aims

A better understanding of wood litter decomposition is essential for predicting responses of forest ecosystems to global climate change. Recent studies suggest that chemical properties of wood litters, rather than physical ones such as wood density, are more important for interspecific differences in wood decomposition rates. However, empirical data are still limited, especially for tropical trees. In addition, decomposition rate of wood litter often varies with time, which makes interspecific comparison difficult. We studied the wood decomposition of 32 rainforest trees to elucidate (i) the degree of interspecific variation in wood decomposition rate of a given size and configuration and (ii) if initial wood density and water permeability are consistent predictors of the overall decomposition rate and its pattern over time.

Methods

A common garden decomposition experiment was conducted in a tropical rainforest in Malaysian Borneo for 32 native tree species. Small wood sticks were set on the forest floor and the weight loss was monitored monthly for 2.7 years.

Important Findings

We found large variation in the wood decomposition rate (a 49-fold range), suggesting that we need to consider this variation when calculating community-level carbon dynamics of tropical rain forests. The physical traits of wood, i.e. wood density and water permeability, were related to wood decomposition rate and its pattern over time. Decomposition half-time related positively and negatively to initial wood density and water permeability, respectively. The time-dependent-rate model fitted better for 18 species (56% of the study species) that had higher water permeabilities than the others, suggesting that micelle porosity in wood relates to temporal changes in decomposition rate.

Sarawak, carbon dynamics, fine wood litter, wood density, tropical rain forest

INTRODUCTION

Decomposition of woody litter is an important component of the forest carbon cycle and a better understanding of its controlling factors is essential for predicting responses of forest ecosystems to global climate change (Cornwell et al. 2009; Weedon et al. 2009). Recent studies suggest that interspecific variation in decomposition rates is large for woody debris, implying a need to integrate the variation in models to estimate the forest carbon cycle (Cornwell et al.2008, 2009; Weedon et al. 2009). However, determinant factors of interspecific variation in wood decomposition rates are not well understood. Some recent studies (e.g., Cornwell et al. 2009; van Geffen et al. 2010; Weedon et al. 2009) suggest that wood chemical properties (i.e. N, P, and lignin concentrations, and C/N ratio) are more important than physical ones (e.g. wood density, conduit area, etc.). On the other hand, other studies reported that high-density wood is decomposed more slowly than low-density wood in tropical trees (Chambers et al. 2000; Chave et al. 2009; Takahashi and Kishima 1973).

There are many studies on species traits and decomposition rates of leaf litters, which provide a general agreement that chemical properties of leaf litters relate to variation in decomposition rate (e.g. Cornelissen 1996; Cornwell et al. 2008; Hobbie 2000; Zhang et al. 2008). In contrast, empirical data are still limited for wood litter, particularly in tropical forests (Weedon et al. 2009). As far as we know, there are no such studies in Asian rain forests. Thus, we need to accumulate more data to understand the general relation between wood traits and decomposition rates (Weedon et al. 2009). Studies in tropical forests are especially important because within-site variation in wood traits is larger in tropical than in temperate forests (Chave et al. 2006).

A technical constraint in interspecific comparison of wood decomposition rate is the fact that decomposition rates of woody debris are not always constant but rather they often vary with time (Brischke and Rapp 2008; Freschet et al. 2012; Harmon et al.1986, 2000; Laiho and Prescott 2004; Yoneda 1975a, 1975b; Yoneda et al.1977, 1990). If decomposition rate varies with time, we must take care during the interspecific and inter-site comparison of decomposition rates because species’ ranks in decomposition rates may vary with the length of study period. It is, therefore, important to understand the factors relating to the pattern of decomposition over time.

In this study, we measured wood decomposition rates of 32 tree species with a common garden experiment in a Bornean rain forest. We addressed the following questions in this study: (i) What is the extent of interspecific variation in wood decomposition rates of a given size and configuration within a tropical rain forest? (ii) Are initial wood density and water permeability consistent predictors of the overall decomposition rate and its pattern over time?

MATERIALS AND METHODS

Study site

This study was conducted in a Shorea splendida (Dipterocarpaceae) plantation at Semenggoh Nature Reserve (latitude ca. 1°23′N, longitude ca. 110°19′E), which is located ~19 km south of Kuching city, Sarawak, East Malaysia. The plantation was established by Sarawak Forest Department during the period 1927–40 in a riverine pan of the Semengoh river, which has alluvial soil (Tan et al. 1987). The average diameter and basal area in 1974 were 34.9cm and 26.0 m² ha⁻¹, respectively, and the average diameter growth rate was 0.72cm per year (Tan et al. 1987). The soil of the least disturbed natural forest in Semenggoh Nature Reserve was Ultisols, the acidic forest soils [pH(H₂O): 4.2–4.6], derived from argillaceous rocks (Hirobe et al. 2004). The mean annual temperature and rainfall during the experimental period were 26.4°C and 4581.0mm, respectively, at the Kuching airport, which was located ~10 km north of the nature reserve (Malaysian Meteorological Service, unpublished results).

Wood stick decomposition experiment

Sample wood sticks (1×1 × 15cm) were made from 32 species of air-dried timber wood (Table 1) purchased commercially from local timber companies. All the species are common in primary forests in the study region. Most of them are large trees of lowland rain forest, but some are restricted to peat swamp forest (i.e. Hopea pentanervia, Sandoricum beccarianum, Parastemon urphyllum, Gonystylus bancanus). As we used only commercially available species, which are often selected for their wood durability, the species set we used could be over-represented by slowly decomposing species comparing to a natural forest tree community. Therefore, interspecific variation in decomposition rate could be underestimated in this study. However, results on the relation between wood traits and decomposition rate may be less affected by species selection if the selected species had a considerable range in wood traits. Our species set included species with a wide range of wood density, i.e. 0.294–0.945g cm⁻³ (Table 1).

Table 1:

parameters of wood decomposition of 32 Bornean rainforest tree species

Family	Species	n	ρ0 (g cm−3)	U (g cm−3 h−1)	Logistic model				Exponential model
					β (year−1)	ID	T0.5 (year)	AIC	K (year−1)	T0.5 (year)	AIC
Anacardiaceae	Campnosperma auriculatum	36	0.517	0.028	1.30	0.87	1.7	−57.8	0.41	1.7	−42.6
Anacardiaceae	Gluta macrocarpa	35	0.702	0.025	0.15	0.00	4.7	−85.0	0.15	4.7	−87.0
Apocynaceae	Dyera sp.	36	0.394	0.050	1.86	0.85	1.1	−71.5	0.74	0.9	−43.7
Burseraceae	Dacryodes sp.	36	0.437	0.024	0.22	0.28	3.9	−108.2	0.17	4.1	−110.0
Dilleniaceae	Dillenia reticulata	36	0.956	0.020	0.71	0.89	3.2	−96.4	0.15	4.6	−89.0
Dipterocarpaceae	Dryobalanops aromatica	36	0.576	0.022	0.46	0.68	3.0	−95.2	0.20	3.5	−94.2
Dipterocarpaceae	Hopea pentanervia	36	0.915	0.014	0.10	0.00	7.1	−138.2	0.10	7.1	−140.2
Dipterocarpaceae	Hopea vesquei	36	0.556	0.016	0.26	0.19	3.1	−72.7	0.22	3.2	−74.7
Dipterocarpaceae	Shorea falciferoides	36	0.945	0.013	0.20	0.37	4.9	−166.1	0.13	5.3	−167.2
Dipterocarpaceae	Shorea superba	36	0.747	0.012	0.08	0.00	8.8	−174.1	0.08	8.8	−176.1
Dipterocarpaceae	Shorea sp. 1	36	0.450	0.030	1.22	0.83	1.6	−63.2	0.45	1.5	−50.1
Dipterocarpaceae	Shorea sp. 2	36	0.412	0.026	1.71	0.95	1.7	−73.7	0.38	1.8	−33.9
Fagaceae	Castanopsis foxworthyi	36	0.920	0.018	0.82	0.90	3.0	−118.7	0.17	4.2	−99.4
Hypericaceae	Cratoxylum arborescens	36	0.454	0.021	0.83	0.78	2.1	−102.9	0.32	2.2	−88.2
Lauraciae	Eusideroxylon zwageri	36	0.836	0.007	0.03	0.00	27.7	−145.5	0.03	27.7	−147.5
Leguminosae	Copaifera palustris	36	0.582	0.023	0.66	0.74	2.4	−36.7	0.27	2.6	−37.1
Leguminosae	Dialium indum	36	0.816	0.015	0.50	0.90	4.7	−139.4	0.08	8.6	−130.4
Leguminosae	Dialium kunstleri	32	0.855	0.023	0.17	0.00	4.0	−100.8	0.17	4.0	−102.8
Leguminosae	Dialium palatysepalum	36	0.806	0.013	0.46	0.84	4.3	−112.1	0.11	6.4	−110.3
Leguminosae	Koompassia excelsa	36	0.817	0.018	0.54	0.90	4.4	−127.9	0.09	7.5	−121.0
Leguminosae	Parkia speciosa	36	0.843	0.026	0.20	0.00	3.6	−135.6	0.20	3.6	−137.6
Leguminosae	Sindora leiocarpa	33	0.512	0.032	2.94	0.84	0.7	−68.0	1.23	0.6	−49.3
Meliaceae	Sandoricum beccarianum	36	0.294	0.026	1.17	0.69	1.2	−84.2	0.61	1.1	−70.8
Moraceae	Artocarpus nitidus	36	0.674	0.029	3.98	0.95	0.8	−40.3	1.07	0.6	−26.8
Myristicaceae	Myristica lowiana	36	0.488	0.027	0.31	0.09	2.4	−59.9	0.29	2.4	−61.9
Myrtaceae	Syzygium sp.	36	0.655	0.025	0.15	0.00	4.5	−142.3	0.15	4.5	−144.3
Rosaceae	Parastemon urophyllum	36	0.943	0.018	0.07	0.00	10.2	−116.6	0.07	10.2	−118.6
Rosaceae	Parinari oblongifolia	36	0.627	0.021	0.16	0.00	4.4	−92.1	0.16	4.4	−94.1
Stercuriaceae	Sterculia bicolor	36	0.328	0.100	1.31	0.58	0.9	−39.6	0.83	0.8	−38.3
Thymelaeaceae	Gonystylus bancanus	36	0.569	0.094	2.38	0.92	1.1	−72.6	0.73	1.0	−32.0
Thymelaeaceae	Gonystylus forbesii	35	0.600	0.060	1.99	0.88	1.1	−88.4	0.73	0.9	−48.1
Thymelaeaceae	Gonystylus sp.	36	0.611	0.058	2.16	0.95	1.4	−96.8	0.54	1.3	−30.2

Figures in bold italics are those used in the best-fit model.

n = number of wood sticks recollected in the common garden experiment; ρ₀ = initial wood density; U = index of water permeability; β, ID, k = coefficients of decomposition models; T_0.5 = decomposition half-life; AIC = Akaike Information Criterion.

We used a fine-blade table saw for cutting the timber wood, but the blade was not extremely fine and this might have caused some damage on the surface of the wood samples to promote microbial access. This may have increased the decomposition rates of each species comparing to naturally fallen wood debris of the same species.

The wood sticks of each species were taken from several (ca. 3–5) pieces of wood. As we did not check whether all the wood pieces of the same species were from one or more individual trees, we could not exclude the possibility that we used only one individual tree for some species and that the traits measured in this study were not within-species average but those of the individual tree. However, the overall pattern of this study is likely to be robust for this problem because traits relating to wood decomposition are generally smaller within species than between species (Chave et al. 2006; Freschet et al. 2012; van Geffen et al. 2010).

Each sample stick was weighed and labeled with an aluminum numbering tag for identification. To convert the initial air-dried weight to an oven-dried weight, 10 sample sticks of the same species were dried in a ventilated air oven at 70°C for a week. The remaining 36 sample sticks of a given species were laid on the forest floor of the study site. In order to minimize environmental heterogeneity, all the samples were set within a 2×3 m² flat area. Wood sticks of the same species were set along a line in an arbitrary direction under the litter layer. Distance between the nearest wood sticks of a same species was ≈1cm and the lines of different species were 10–50cm apart. As the distance between wood sticks of a same species was only 1cm, it was possible that the neighboring samples were connected by fungal networks via the soils. It should also be noted that environmental heterogeneity at very small spatial scale, if present, could have affected the results as we placed wood sticks of a same species together.

One stick per species was collected monthly from May 2001 to February 2004 (32 times in total). On the last collection date, we collected all the remaining sticks: five for most species except four species, for which 1–3 sticks were lost during the experiment. All collected sample sticks were washed, dried in a ventilated air oven at 70°C for a week and weighed.

Physical trait measurement

Initial wood density (ρ₀) was measured for the wood sticks that were used for air-to-oven-dry weight ratio. We calculated ρ₀ of each wood stick as the oven-dried weight divided by the volume (15cm³). The average of ρ₀ of the 10 samples was used as the initial wood density of each species.

Water permeability was measured for the initial wood samples of each species, according to the method of Hayashi and Nishimoto (1965). Four wood sticks of each species were kept in distilled water in a thermostatic water bath at 30°C for 4h to absorb the water. The permeability (U) [g cm⁻³ h⁻¹] of each wood sample was calculated as

(1)

where w₀, w_4h, v and t are the initial weight of a wood sample, the weight of the wood sample after the water absorption, the volume of a wood stick (15cm³) and the duration of experiment (4h), respectively. The average of U of the four samples was used as the permeability index of each species. Wood sticks used for the physical trait measurement were not used for the decomposition experiment.

Statistical analysis

We fitted two decomposition models, exponential and logistic models, to the results of the decomposition experiment. The exponential model assumes a constant decomposition rate and is formulated by the following equation:

(2)

where t is time [year] after laying sample sticks on the forest floor, w is the oven-dry weight of a wood stick [g] at time t, w₀ is the initial oven-dry weight of the wood stick [g] at t = 0 and k [per year] is the coefficient of the decomposition rate. This model was developed by Jenny et al. (1949) and discussed in detail by Olson (1963).

The logistic model (Yoneda 1975a) assumes that decomposition rate increases with time and is written in the form

(3)

where ID is an index of durability and β [per year] is the coefficient of decomposition rate when t is very large (Yoneda 1975a, 1975b). If ID = 0, Equation (3) is identical to Equation (2) (k = β). If ID > 0, the decomposition rate increases with t up to β. The larger the value of ID the slower the increase in decomposition rate.

The coefficients of the models (k, β and ID) were estimated for each species using the non-linear least squares method with the function ‘nls’ on R version 2.10.1 (R Development Core Team 2009). Initial values of k, β and ID were 0.004, 0.1, and 0.8, respectively. We used ‘port’ algorithm with the lower bounds of zero for k, β and ID. We confirmed that all the estimations had converged. The data of the last collection date (n = 2–5) were averaged for each species. Data were deleted from the analysis if w > 1.1w₀, which may be a result of our label handling errors (n = 6). We additionally deleted 61 data that were identified as outliers (25 upper and 36 lower ends of the data) by the function ‘lowess’ (Cleveland 1979, 1981) (online Supplementary Figure S1.). We used 0.5 for the span parameter and excluded the data from the analyses as outliers if the values deviated more than 20% from smoothing values. In order to select the best-fit model, we compared the Akaike Information Criterion of the two models. For comparison of species for which different models provided the best fit, we calculated the decomposition half-life (T_0.5; years needed to reach 50% mass loss) based on the estimated model parameters. The removal of the outliers produced small changes in the overall results and conclusion except that removing outliers changed the best-fit model for seven species. The fitting of the model was apparently worse in four of the seven species without the removal of the outliers (online Supplementary Figure S2).

An important difference in sampling design between the current and many previous decomposition experiments was the balance between frequency and number of replications of sample collection. In this study, we collected samples frequently, i.e. monthly (32 times during the 2.7-year experiment) but only one sample was collected for each species at each time. In many previous studies, samples were collected less frequently but multiple samples were collected at each collection time. Our sampling design (high frequency but no replication at each sampling time) may have had low statistical accuracy. Therefore, we conducted computer simulations to analyze the effect of sampling frequency and replications on estimation of decomposition rates, and confirmed that our sampling design had enough statistic power to estimate decomposition rate and its pattern over time (online Supplementary Figures S3 and S4).

Relationships between the physical wood traits (ρ₀ and U) and the decomposition half-life (T_0.5) were analyzed by the Pearson’s correlation coefficient and the multiple regression analysis. All variables were log-transformed in the multiple regression analysis due to their right-skewed distributions.

RESULTS

Interspecific variation in decomposition

Decomposition processes were highly variable among different species (Fig. 1, Table 1). After 2.7 years, Sindora leiocarpa had lost almost all of its initial dry weight (T_0.5 = 0.7 years, Table 1), whereas Eusideroxylon zwageri had lost only 5.7% (T_0.5 = 27.7 years). For 18 out of 32 study species, the best-fit model was the logistic model, indicating that the decomposition rates increased with time (Fig. 2, Table 1). For the other 14 species, the exponential model was selected, indicating a constant decomposition rate. Species whose best-fit model was the logistic model has a significantly shorter decomposition half-life (T_0.5) than those whose best-fit model was the exponential model, with means of 2.1 (standard abbreviation [SD] = 1.9) and 6.6 (SD = 6.5) years, respectively (Wilcoxon test, P < 0.01).

Figure 1:

changes in relative weights of wood sticks during a 2.7-year common garden experiment. Solid and dotted lines are values predicted by the logistic and exponential decay models, respectively. Species are arranged in order of increasing half-life (T_0.5) from left to right and from top to bottom.

Figure 2:

water permeability (A) and initial wood density (B) for woods of two species groups having different decomposition characters. EXP spp are species (n = 14) for which decomposition followed the exponential decay model, which has a constant decomposition rate. LOG spp are those (n = 18) that follow the logistic decay model, for which mortality rate increases with time. See text for details of the models.

Wood traits and decomposition

The mean water permeability of woods were significantly larger for species whose best-fit model was the logistic model than those that fitted well to the exponential model (Wilcoxon test, P = 0.019, Fig. 2A). In contrast, wood density was lower for the logistic-model species although the difference was statistically marginal (Wilcoxon test, P = 0.059, Fig. 2B).

T_0.5 was significantly correlated positively to initial wood density (Spearman’s rank correlation, r_s = 0.674, P < 0.001) and negatively to water permeability (r_s = −0.843, P < 0.001) (Fig. 3). Although initial wood density and water permeability were significantly negatively correlated (r_s = −0.630, P < 0.001), the multiple regression analysis showed significant effects in both traits (P = 0.014 and P < 0.001 for wood density and water permeability; R² = 0.729), indicating that both wood density and water permeability were significant predictors of the overall decomposition rate.

Figure 3:

relationship between decomposition half-life (T_0.5), initial wood density (A) and water permeability (B). Open and closed circles are species whose decomposition rates fitted to the logistic and exponential decay models, respectively. Spearman’s rank correlation coefficients (r_s) are also shown. Note that both x and y axes are in log scale.

DISCUSSION

Limitation of interpretation

An important limitation of interpretation of the present study is that it is difficult to estimate actual decay rates of woody debris in natural forests from decay rates of the small-sized wood samples we used. It is known that wood decomposition rates vary greatly with difference in size of the wood sample (van Geffen et al. 2010). Because our wood samples were small (1×1 × 15cm), the surface/volume ratio was relatively large and thus the decay rates and half-life times may be over- and underestimations, respectively, for decay of larger wood debris, such as tree stems.

It is also noted that presence of bark and its properties have strong effect on decomposition rate. As we used artificial wood sticks without bark, variation in bark properties of natural woody debris may well interfere with the species ranking in decomposition rate and pattern in natural woody debris with bark.

The effect of termites is another important factor to be considered because termites are one of the major factors in wood decomposition in many tropical forests (e.g. Abe 1980; Cornwell et al. 2009; Wood and Sands 1978). We observed visible evidence of termite attack only in a small proportion of the wood samples (3.0% of 1151) in this study. We should, thus, consider the decomposition rates of this study as those mostly without termite effects. The low termite attack could be by chance because we recorded more frequent termite attack in a larger field experiment using the same wood sticks in a tropical forest in Sarawak (Mori et al. unpublished results). We set 5200 wood sticks of Dryobalanops aromatica at 1300 points in a 52-ha area and we found evidence of termite attack in 49.5% of the samples. Termites could influence the species ranking in decomposition rate if the between-species variation in termite resistance is large. Further experiments controlling the termite attack (e.g. Arango et al. 2006; Takamura 2001) are needed to evaluate species’ specific termite resistance.

Finally, the wood traits we measured were not necessarily the (only) causal factors determining wood decomposition rates. Because we did not analyze chemical wood properties, it is possible that some chemical wood traits that relate to the physical properties could be a reason for the correlation observed in this study.

Despite these limitations, we think this study still allows for between-species comparisons and is helpful for studies of wood decomposition. As we used wood samples of same size, which have been decomposed under similar environmental conditions with little termite attack, the variation observed must be due to differences in the structural and/or chemical traits of the wood inside the bark.

Large interspecific variation in wood decomposition rates

We found a very large within-site interspecific variation in wood decomposition rates (Fig. 1, Table 1). Based on k values of the exponential model, our result showed a 49.1-fold range for all species and a 35.5-fold range for the middle 90% of species. These values are much larger than the variations in leaf decomposition rates. Cornwell et al. (2008) reported that the average range of leaf decomposition rates within a site was 18.4-fold for all species and 10.5-fold for the middle 90% of species for 14 sites over the world. Hirobe et al. (2004) found a 6.2-fold range in leaf decomposition rates (k = 0.38–2.36) of 15 tree species in the same forest that we studied. Our result strongly suggests a need to consider species composition of wood debris in estimating wood decomposition rates in tropical forests, as has been done for leaf decomposition (Austin and Vitousek, 2000; Cornwell et al. 2008; Swift et al. 1979; Takeda et al. 1984; Vivanco and Austin 2006).

Time dependency of wood decomposition rates

The majority of species (18 of 32) had a time-dependent decomposition rate. For these species, decomposition rates were initially small and then increased as decomposition proceeded. Similar patterns of wood decomposition have been reported for many temperate trees (e.g. Brischke and Rapp, 2008; Freschet et al. 2012; Harmon et al.1986, 2000; Laiho and Prescott 2004; Yoneda 1975a). Although the mechanisms of the increase in decomposition rate in woody litter are not yet clear, this study suggests that water permeability may relate to the difference in decay models because the logistic model fits better than the exponential model for species with higher water permeability.

The water permeability may relate to the physical structure of wood material. Yamamoto and Hong (1994) reported that wood of Gonystylus species had one of the largest water permeabilities among 24 tropical species. They also found that Gonystylus had a high proportion of micelle porosity, or fiber gap, in the wood, suggesting that water permeability could be used as an index of the proportion of micelle porosity. Micelle porosity is a collection of very tiny gaps that are difficult for decomposers to access from the outer surface of the wood. As decomposition proceeds, the micelle pores may connect to each other, resulting in a gradual increase in the decomposition surface inside the wood, which would enhance the surface-to-volume ratio and hence also the decomposition rate (Harmon et al. 1986). In contrast, for the wood with only larger pores, such as vessels, decomposers may enter the inside of the wood relatively easily from the early phase of decomposition. This may result in a relatively constant decomposition rate, expressed by the exponential model.

However, the above-mentioned mechanism determining the temporal pattern is still very speculative. There may be many other possible (non-exclusive) factors relating to the temporal pattern, such as loss of initial heartwood defenses over time (Harmon et al. 1986), gradual leaching of constitutive compounds (Spears and Lajtha 2004) and enhanced activity of microbial decomposers as decomposition progresses (Cornwell et al. 2009). We need more detailed studies on the effects of chemical and physical wood traits, as well as the activity of decomposers to elucidate the temporal pattern of wood decomposition and its mechanisms.

Wood density and decomposition rate

We observed a significant positive correlation between initial wood density and decomposition half-life, indicating that higher density woods were decomposed more slowly. Our results support the previous view that higher density woods decompose more slowly (e.g. Chave et al. 2009) but are in contrast with the recent studies showing no correlation between wood density and decomposition rate (van Geffen et al. 2010; Weedon et al. 2009). There may be several possible explanations for this discrepancy.

First, as briefly mentioned above, decomposition of woods with and without bark may differ considerably. We used artificial wood materials without bark, whereas Weedon et al. (2009) and van Geffen et al. (2010) used natural woody debris with bark. If variation in species’ bark properties affects decomposition considerably, the effect of wood density may be masked and the species rank in decomposition rate may differ in wood materials with and without bark.

Second, the wider range in wood density used in our study made the correlation statistically significant. Weedon et al. (2009) is based mainly on species from temperate forests with a limited number of tropical tree species, very few of which had high wood density (only two species had wood densities >0.8g cm⁻³). The range of wood density was 0.22–0.69g cm⁻³ in van Geffen et al. (2010), which lacked heavy wood species in comparison to our study species (0.29–0.98g cm⁻³). According to Chave et al. (2009), the range of density of tropical trees was as wide as 0.08–1.39g cm⁻³. Many tropical forests may include species with heavier wood than those studied by van Geffen et al. (2010). For example, we could not find significant correlations between wood density and decomposition half-life when we classified the species into two groups with smaller and larger wood density than 0.69g cm⁻³ (n = 19 and 13, r_s = 0.239 and −0.187, and P = 0.32 and 0.54 for the smaller and larger groups, respectively). The relationship between wood density and decomposition rate in tropical forest is likely to be driven by the presence of dense wood species.

CONCLUSIONS

This study showed a very large interspecific variation in wood decomposition rates of Bornean rainforest tree species. This strongly suggests a need to incorporate species-specific decomposition rates when calculating ecosystem-level carbon dynamics for tropical forests. Our results suggest that a significant relation between wood density and decomposition rate among Bornean trees is likely to be driven by dense wood species. This needs to be tested in other tropical forests with different species sets. The novel finding of this study is that the water permeability of wood consistently explains interspecific variation not only in the overall rate of wood decomposition but also in its temporal pattern. Further experimental studies are required to elucidate the mechanisms behind this relationship.

SUPPLEMENTARY MATERIAL

Supplementary Figures S1–S4 are available at Journal of Plant Ecology online.

FUNDING

Grants-in-Aid for Scientific Research from Japan Society for the Promotion of Science (20405011).

ACKNOWLEDGEMENTS

We are grateful to the Forest Department of Sarawak, the Sarawak Forestry Corporation, and the Management Office of the Semenggoh Nature Reserve for permitting and helping us to survey in Sarawak, Malaysia. We wish to thank Dr T. Yoneda for his advice in the decomposition experiments. We thank to anonymous reviewers for their comments that improved the paper.

Conflict of interest statement. None declared.