Interspecific differences in wood decay rates: insights from a new short-term method to study long-term wood decomposition


  • Grégoire T. Freschet,

    Corresponding author
      Correspondence author. E-mail:
    Search for more papers by this author
    • Present address: Department of Forest Ecology and Management, Swedish University of Agricultural Science, 901 83 Umeå, Sweden.

  • James T. Weedon,

    1. Department of Systems Ecology, Faculty of Earth and Life Sciences, Institute of Ecological Science, VU University, de Boelelaan 1085, 1081 HV Amsterdam, The Netherlands
    Search for more papers by this author
  • Rien Aerts,

    1. Department of Systems Ecology, Faculty of Earth and Life Sciences, Institute of Ecological Science, VU University, de Boelelaan 1085, 1081 HV Amsterdam, The Netherlands
    Search for more papers by this author
  • Jurgen R. van Hal,

    1. Department of Systems Ecology, Faculty of Earth and Life Sciences, Institute of Ecological Science, VU University, de Boelelaan 1085, 1081 HV Amsterdam, The Netherlands
    Search for more papers by this author
  • Johannes H. C. Cornelissen

    1. Department of Systems Ecology, Faculty of Earth and Life Sciences, Institute of Ecological Science, VU University, de Boelelaan 1085, 1081 HV Amsterdam, The Netherlands
    Search for more papers by this author

Correspondence author. E-mail:


1. While the importance of wood decay for the global carbon balance is widely recognized, surprisingly little is known about its long-term dynamics and its abiotic and biotic drivers. Progress in this field is hindered by the long time-scales inherent to the low decay rates of wood and the lack of short-term methods to assess long-term decomposition dynamics in standardized field conditions.

2. Here, we present such a method, which relies on the sampling and short-term incubation of wood from several decay stages covering the entire decay process. Together these short-term decay steps are used to model and discriminate between three potential decay dynamics (linear, exponential and sigmoid) using an iterative optimization procedure. We applied this method to analyse long-term wood decay of six subarctic tree species (six stems and two roots) and test the hypotheses that (i) different wood species follow distinct decay dynamics and (ii) interspecific variation in wood traits controls variation in wood decay rates in a standardized environment.

3. We found interspecific variation in long-term wood decay dynamics: decay of Alnus and Salix stems was best described by exponential models, whereas decay of Sorbus stems and Betula and Pinus roots was best fitted by linear models and Betula, Pinus and Populus stems each displayed a sigmoid decay dynamics (up to 5-year initial lag phase). A six-fold variation was observed between the decomposition half-lives of all eight wood types, from 6.8 years (6.1–7.5, 95% C.I.) for Alnus stems to 41.3 years (34.5–51.8) for Pinus roots. Initial wood traits such as pH (R2 = 0.92), dry matter content (R2 = 0.79) and lignin (R2 = 0.73) were good predictors of long-term wood decay rates.

4.Synthesis. Our findings suggest changing decay dynamics across wood species and types that are likely to arise from changing underlying wood decay processes (i.e. varying wood functional traits/decomposer community interactions). Our new method, which combines advantages of direct observations and the chronosequence approach, allows reliable comparisons of species contributions to long-term wood decay rates and provides future opportunities to experimentally disentangle intrinsic and external abiotic and biotic drivers of long-term wood decay processes.


Dead wood is a key driver of forest ecosystem functioning through its impact on carbon (C) and nutrient cycling (Harmon et al. 1986) and its beneficial role to microbial, animal and plant diversity (Freedman et al. 1996). The substantial stocks of woody debris (WD) – forested ecosystems constitute as much as 50% of total C in the terrestrial biosphere (Malhi 2002) and WD represents up to 20% of the total in old-growth forests (Delaney et al. 1998) – and its ecological importance have stimulated studies on various aspects of their role in ecosystem processes (see reviews by Harmon & Sexton 1996; Wirth, Gleixner & Heimann 2009). Moreover, the need for understanding and predicting the consequences of current global climate and land-use changes has put increasing pressure on ecologists to quantify ecosystem C stocks and fluxes, with a special focus on forests and peatlands (IPCC 2000). In this context, accurately measuring WD turnover and its sensitivity to (biotic and abiotic) environmental factors would be highly valuable with respect to the improvement of C cycling models and consequently global climate predictions (Sitch et al. 2003; Cornwell et al. 2009).

Wood decomposition is a complex process which, like litter decomposition in general (Swift, Heal & Anderson 1979), involves biotic and abiotic influences (Harmon et al. 1986; Boddy & Watkinson 1995; Chen et al. 2000; Progar et al. 2000), as well as the mechanical and chemical properties of the wood itself (Harmon et al. 1986; Taylor et al. 1991; Chambers et al. 2000). Micro-organisms, particularly fungi, and invertebrates are the main agents of wood decay (Käärik 1974). However, results from field studies examining the relative importance of these factors on decomposer activity and wood decay show contrasting results for both environmental (Harmon, Krankina & Sexton 2000; Yatskov, Harmon & Krankina 2003; Kueppers et al. 2004) and wood quality parameters (Chave et al. 2009; Cornwell et al. 2009; Weedon et al. 2009; van Geffen et al. 2010). These context-dependent results illustrate the difficulties of singling out species trait effects, environmental effects and their interactions in heterogeneous environments with multiple wood types and multiple interactions among decomposition drivers. The common extrapolation of long-term wood decay dynamics from short-term results may also be partly responsible for these discrepancies. Indeed, while most studies use negative exponential functions to describe WD decay, a few studies have suggested the inadequacy of assuming similar processes across tree species (Harmon et al. 1995; Hale & Pastor 1998; Harmon, Krankina & Sexton 2000). Whether such variability in decomposition processes across species is marginal or the rule still needs to be resolved. In this respect, long-term experimental, standardized manipulations of WD decomposition are much needed.

Studying long-term wood decomposition has remained a great challenge to date because of the extended timescale involved, up to centuries for complete wood decay under certain conditions (Kueppers et al. 2004). To describe long-term WD decay dynamics and obtain estimates of decay rates, researchers have used direct and indirect methods. Direct approaches, such as those generally used in natural wood durability studies (see Scheffer & Morrell 1998), involve the monitoring of newly fallen wood pieces for varying periods (e.g. Romero, Smith & Fourqurean 2005; Eaton & Lawrence 2006). However, long-term decomposition studies are often incompatible with the time frame of individual research projects and the current need for extensive comparative measurements of WD decay rates but, when implemented over short time periods, they usually capture the first stages of wood decay only and therefore lack relevance for the entire course of decomposition. Direct approaches are nevertheless particularly helpful for comparing different treatments and disentangling multiple influences. Indirect approaches, using chronosequences relying on historic records on extreme events (e.g. logging and massive windfall) or on other evidence indicating duration of WD decay (e.g. age of regenerating stands, age of trees growing on the log, age of fire scars and radio carbon dating), allow the estimation of long-term decomposition dynamics within a much shorter time frame (e.g. Krankina & Harmon 1995; Chen, Harmon & Griffiths 2001; Yatskov, Harmon & Krankina 2003; Kueppers et al. 2004; Hérault et al. 2010). Recently, promising progress has been made in comparing species contributions to wood decomposition rates using the chronosequence approach (van Geffen et al. 2010; Hérault et al. 2010), but the resolution is coarse because of the great uncertainties related to historic reconstruction and spatial and temporal heterogeneity (Harmon, Nadelhoffer & Blair 1999), as compared with standardized experiments.

Here, we present a new method that combines the advantages from the direct observation and chronosequence approaches. This method uses wood relative density (RD; WD density divided by initial WD density) as a proxy for decay stage (Christensen 1984) and, by analogy to the decomposition vectors’ method (Harmon, Krankina & Sexton 2000), relies on the sampling of several decay stages to cover the entire decay process. Our key innovation is to incubate multiple WD samples, covering a large range in decay stages (characterized by their RD) for each of a range of species, simultaneously for a set duration in a ‘common-garden’ environment (cf. Cornelissen 1996). Then, for each species, we use a modelling procedure to combine all component short-term decay curves from all these decay stages into a long-term mass loss curve against time and test which decay model (linear, exponential and sigmoid) provides the best fit (Fig. 1). Thus, by relating the RD decrease of several decay stages of wood, this method enables direct and standardized comparison of long-term mass loss dynamics of different species using a short-term experiment. We demonstrate the applicability of this technique in a case study of six dominant tree species of Swedish subarctic forests, Betula pubescens (Ehrh.) ssp. czerepanóvii, Pinus sylvestris (L.), Alnus incana (L.), Populus tremula (L.), Sorbus aucuparia (L.) and Salix caprea (L.) and test the hypotheses that (i) different wood species follow distinct decay dynamics in a similar environment and that (ii) interspecific variation in wood traits controls variation in WD decay rates.

Figure 1.

 Outline of the method used to model long-term wood decay dynamics based on short-term incubations of wood in various decay stages. Only the four main steps are shown (a–d). Further details are provided in the text. RD stands for woody debris’ relative density and RD′ for RD after incubation. AIC is Akaike Information Criterion.

Materials and methods

Study Area

Both plant material collection and experimental setup took place around the Abisko Research Station, North Sweden (68°21′N, 18°49′E) within the low altitude (350–400 m a.s.l.) forested area. Climatic data from the recent decade (1999–2008) showed a mean annual rainfall of 352 mm and mean January and July temperatures of −9.7 and 12.3 °C, respectively, with average daily temperatures ranging from −39.0 to 21.3 °C (meteorological data, Abisko Research Station). Throughout the Abisko valley, forests are dominated by polycormic mountain birch trees (Betula pubescens ssp. czerepanóvii), both in dry and wet areas. Four other deciduous tree species and one evergreen are commonly found within the area. Pinus sylvestris and Populus tremula are characteristic of the dry forest, while Salix caprea, Sorbus aucuparia and Alnus incana are mainly found in riparian areas. Following two successive episodes of severe tree defoliation by caterpillars of the moth Epirrita autumnata in 1954 and 1955, which killed a large number of birch stems, the forested area of Abisko valley consists mainly of rejuvenated birch stands of low stature (Tenow et al. 2004).


Our new method of estimating long-term wood decay requires the sampling of dead wood at various stages of decay. Thereto, all six tree species were sampled for dead woody stems, and two of them, Betula pubescens and Pinus sylvestris, were also sampled for dead woody roots. The sampling was performed from mid-August to mid-October 2007 and complemented in April 2008. Because of the young age and low stature of the forest, the average diameter of WD was relatively low. To ensure balanced species comparisons, only the common WD diameter class of 5 ± 1.2 cm was sampled, while the minimum length was 50 cm. From each log sample, we obtained four subsamples of 10 cm for the decomposition experiment and a central cylinder of 10 cm for wood density, wood traits and residual humidity measurements. To ensure the representativeness of the WD central cylinder for the neighbouring subsamples used for the decomposition experiment, WD with heterogeneous external and internal aspect were discarded. In particular, WD with large branching knots, irregular shapes or uneven bark cover were avoided.

For each tree species and material (stem and root), between two and three samples of newly dead material were identified and sampled in the field, on standing trees, according to both internal and external aspects of the wood. After measurements of wood density on subsamples of each material, only the denser wood sample of each species was kept as representative of newly dead wood and other samples were discarded. For each species and material, WD pieces were also sampled from various stages of decay. In this way, we aimed to cover the widest range of wood decay stages. WD samples that were in an advanced state of fragmentation, preventing the direct estimation of volume for initial wood density calculation, were avoided in this study. Nevertheless, to cover the latest stages of WD decay and therefore improve the range of decay models, such fragmented samples can still be used by estimating the initial stem diameter from less degraded parts of the stem, as described by Harmon & Sexton (1996). After sawing WD samples into five 10-cm-long cylindrical subsamples, these were cleaned of alien material, then air-dried and stored in paper bags pending further treatment and analyses.

Decomposition Study

The central cylinder of each WD sample was used to estimate density, residual humidity and, on the newly dead WD sample only, WD chemical and structural traits of the sample. We estimated WD density (mg cm−3) as the ratio of wood oven-dry mass (mg; 96 h at 60 °C) to volume (cm3). This measure of wood oven-dry mass could, however, lead to slight biases as, in contrast to vapour phase technique, it does not have the potential to extract all bound water from wood tissues (Williamson & Wiemann 2010). The volume was estimated both by volume displacement in a graduated glass column (note that better accuracy could be achieved with the water displacement method; Williamson & Wiemann 2010) and by measurement of length and diameter (taken as the average of three cross-section measures, i.e. six measurements) and taken as the mean of both estimates. Residual humidity of each sample was estimated on the central cylinder as the ratio of the difference between air-dry and oven-dry mass to air-dry mass (the latter measured before volume estimation).

For each WD sample, the four remaining air-dry subsamples were weighed separately and sealed into nylon litter bags of 1-mm mesh size, which allowed exchange of micro-organisms and small soil invertebrates. Measured air-dry masses were corrected for residual humidity to obtain true dry mass. The litter-bag samples were incubated in outdoor litter beds in an experimental garden surrounded by birch forest at Abisko Research Station, on 23 April 2008, following Cornelissen et al. (2004). The litter beds consisted of rectangular wooden frames sunk into the ground, including a layer of grit stones (particle sizes 10–20 mm) as a free-draining foundation on top of the original soil profile. They were covered by a 20-mm layer of mixed fresh and old litter collected in September 2007 from the surrounding dry and wet forests and ponds. The litterbags were laid out flat, without overlap, and covered by a 10-mm layer of the same mixed litter. Four separate blocks (c. 2 × 1 m) were used to host the four distinct groups of subsamples per sample. The litter beds were covered with a 10-mm metal grid mesh to protect them from birds and rodents. The litter bags were subject to the local climatic influences and did not receive any additional treatment.

The litter bags were harvested after two full years of incubation on 23 April 2010, while still frozen and stored at −16 °C pending further processing. After defrosting, adhering soil, soil fauna and other alien material were removed from the decomposed litter by gentle brushing and rinsing with tap water. Litter subsamples were then dried (60 °C, 96 h) and weighed. The percentage mass loss of each subsample was calculated as the ratio of its dry mass after decomposition to its original dry mass. For each replicate, WD density after decomposition was estimated by recalculating the pre-incubation WD density of the related sample central cylinders after accounting for the replicate percentage mass loss. These calculations assumed a constant volume before and after incubation to avoid biases related to material fragmentation leading to post-incubation density underestimation (Christensen 1984).

Wood Litter Trait Measurements

All newly dead WD types were measured for C, N, P and lignin content, as well as pH. For these analyses, air-dried subsamples were ground and subsequently oven-dried for 24 h at 60 °C. Carbon and nitrogen concentrations were measured by dry combustion on a NA 1500 elemental analyser (Carlo Erba, Rodana, Italy). Phosphorus was measured by acid digestion as referred to in Freschet et al. (2010). Lignin concentration was determined by the extraction of non-ligneous compounds as described in Freschet et al. (2010). For pH measurements, 0.15 mL of each ground sample was shaken with 1.2 mL demineralized water in an Eppendorf tube for 1 h at 250 rpm. After centrifugation at 9000 g for 5 min, pH of the supernatant solution was measured (Cornelissen et al. 2006). Additionally, an extra subsample of each newly dead WD type was measured for dry matter content (DMC). Subsamples were immersed in tap water for 9 days to ensure homogeneous filling of air spaces then wiped gently and measured for their water-saturated weight. Subsequently, dry weight was measured after drying for 96 h at 60 °C. DMC was expressed as the ratio between dry weight (mg) and water-saturated weight (g).

Modelling Wood Decomposition

An overview of the method is presented in Fig. 1. For each WD type, all density measurements (n‘WD decay stage’ × pre- and post-incubation × four replicate subsamples) were standardized, i.e. divided by the density of the densest newly dead WD, to give relative wood density (RD) values, which ranged from 1 to 0 (100% to 0%). For each pair of pre- and post-incubation RD, the set of four replicate subsamples was averaged to produce one single average ‘2-year-decay-vector’ (sensuHarmon, Krankina & Sexton 2000). We therefore have, for each WD type, a vector Y of n observed ‘decay vectors’ (Fig. 1a) which we assume can be modelled by a statistical decomposition model that relates Y to a vector of latent (unobserved) t values X via model-specific parameters θ. For each model and WD type, we are specifically interested in a) the estimates of θ and b) the likelihood of the fitted model so that different decomposition models can be compared for each WD type. We generate maximum likelihood estimates for θ using an algorithm conceptually similar to the expectation maximization (EM) algorithm (Dempster, Laird & Rubin 1977; Gupta & Chen 2011), that is, an iterative procedure (Fig. 1c) is used to alternate between estimates of X (t values in our case) and θ, until a stable maximum likelihood is obtained. Our approach differs slightly from classic EM in that we use a quasi-Newton optimization algorithm, ‘L-BFGS-D’ in the optim() function of R (Byrd, Lu & Nocedal 1995; R Development Core Team 2009), to find the vector of t values X that maximizes the likelihood of a (non) linear regression of Y on X. We found this method to be more effective at avoiding local optima than the classic EM implementation.

This method was used to estimate the parameters of the following three models for each WD type:

1. Linearinline image
2. Negative exponentialinline image
3. Negative sigmoidinline image

where t is time in years and m, k, a and b are parameters to be estimated. All models assume RD = 1 at t = 0 (i.e. the wood fragment with highest initial density represents ‘fresh’ litter at the start of the incubation), and further that the parameters are subject to the constraints m, k and > 0, b ≥ 1 and t  0. Files containing example scripts for performing the optimization are provided as Supporting Information (Appendix S1). These can easily be modified to allow the fitting of other, arbitrary decomposition models.

Initial tests on simulated data showed that the success or failure of the nonlinear model fitting for model 3 was partly dependent on the choice of the initial set of t values, and the starting parameter estimates for the nonlinear regression. Best performance was achieved with a self-starter function (Ritz & Streibig 2009) that generated initial t values by arranging the vectors along a straight line determined by the mean slope of all n decomposition vectors (Fig. 1b). Similarly, model failure was avoided when the mean slope of the decomposition vectors was used as the starting value for a, and the starting value of b was set to 1 (i.e. when using realistic starting values of model parameters in the iteration procedure). For consistency, the same initial values of t were used for all three models, and the mean slope was also used as a starter value for k in model 2.

Empirical time series of wood mass loss with sufficient temporal resolution are not available to independently test our estimation method, so we used Monte Carlo simulations (Hilborn & Mangel 1997) to determine the range of decomposition dynamics over which our method can reliably estimate decay parameters. Simulated ‘WD decay curves’ were generated using a large range of parameter values. These decay curves were then ‘sampled’, that is, a number of ‘decay vectors’ were generated from each decay curve to which we added a random component with variance based on the observed within-log variance seen in our dataset. These simulated data were then submitted to the same iterative optimization procedure as each of our real ‘WD type’ dataset. The estimated parameters were compared with those used to generate the simulated ‘WD decay curve’ to determine the accuracy and any biases in our estimation method. Briefly, we found that our procedure produced reliable estimates for linear and exponential models except for very low m and k parameter values and started to produce unreliable estimates in the sigmoid model only for very slowly decomposing WD with long lag phases (i.e. generated with a combination of low a and/or high b with the sigmoid model) that result in initial 2 year RD change of <2% (further details are given in Appendix S2). Based on the final selection of best-fit model for each WD type and the observed decay rate of newly dead wood litter in this experiment, all our WD types show only small potential relative error except Pinus stems (sigmoid model: 2.2% mass loss after 2 years) and Pinus roots (linear model: = 0.012). This suggests that, while the modelled output is reliable for most WD types, the duration of the lag phase of Pinus stem sigmoid model and the overall decay rate of Pinus roots might be slightly under- or over-estimated. Since the ‘newly dead’ WD sample strongly influences the model for which the best fit is obtained, as well as the eventual estimate of lag phase duration in the case of sigmoid decay dynamics, it is advisable to increase replication of this initial decay class, and perhaps sample at more regular intervals, so as to obtain well-constrained estimates of the decay rate in the initial period.

The output for each model fit is a set of modelled t values representing estimates of the age of the different decay stages and model parameters (Fig. 1d). For each WD type (stem or root species), once optimization procedures had been performed for each distinct model (linear, exponential and sigmoid), the model fits were compared using the value of Akaike Information Criterion (AIC) calculated by the following formula:


where L is the maximized value of the likelihood function for each model, and k is the number of parameters fitted by the model (thus allowing for comparison of models with different numbers of parameters). The decay model that provided the lowest AIC was considered the best description of the underlying decay dynamics and used to estimate the WD type decomposition half-life (T1/2; time (years) needed to reach 50% mass loss) for subsequent analyses. Using T1/2 allows comparison of species for which different decay curves provided the best fit. Ninety-five per cent bootstrap confidence intervals (C.I.) for T1/2 for each WD type and decay model combination were calculated from the 2.5 and 97.5 percentiles of the distributions of T1/2 estimated from 1000 bootstrap samples of the wood decay stages within each WD type (Efron & Tibshirani 1993). This measure includes uncertainty in our fitted values of t for each combination of WD type and model and avoids the unrealistically narrow C.I. that would result from calculations using the best fit only. The relatively large bootstrap C.I. found here stress the need to ensure an even distribution of wood decay stage samples across the whole decay process. Ordinary least square regressions were used to test predictions of WD T1/2 from initial wood litter traits (to comply with normality assumptions, T1/2, P, C/N and lignin/N were log10-transformed before analysis). To assess the degree of correlation between explanatory variables, Pearson’s correlation coefficients were calculated for the relationships between lignin, DMC and pH.


Our model results showed that the wood decomposition dynamics differed across species as none of the three models tested could consistently provide the best fit for the decomposition process of all WD types (species-organ combinations; Table 1; Fig. 2). Long-term decay of three WD substrates was best fitted by a negative sigmoid function, three by a negative linear function and two by an exponential decrease function. Thus, only Alnus and Salix coarse stems displayed a constant relative mass loss over time (exponential model), while the relative mass loss of Sorbus coarse stems and Betula and Pinus coarse roots increased instead regularly as decomposition advanced (linear model) and that of Betula, Pinus and Populus coarse stems was initially low, then increased and finally stabilized (sigmoid model) (Fig. 3).

Table 1.   Model fits and predictions for each wood substrate
SpeciesnLower limit of model prediction (% RD)LinearSingle exponentialSigmoid
AICmT1/295% C.I.AICkT1/295% C.I.AICbaT1/295% C.I.
  1. n is the number of distinct decay samples used in models. RD is relative wood density. AIC is Akaike Information Criterion for a given decay model. m, k, b and a are predicted model parameters. T1/2 is decomposition half-life. C.I. is bootstrap confidence interval. For each substrate, best-fit models are indicated in bold.

 Alnus incana1137−104.70.0657.76.5–8.6121.00.1026.86.17.5−–7.5
 Betula pubescens1330−128.60.03913.011.1–18.4−119.50.05412.710.9–20.5134.51.790.09312.210.515.3
 Pinus sylvestris1136−122.50.03116.114.5–24.3−105.50.03917.715.0–32.0128.12.530.08217.514.320.1
 Populus tremula1223−94.50.0598.57.0–12.4−75.30.0729.66.8–16.8106.22.790.1997.66.59.1
 Salix caprea1434−111.70.0717.15.7–7.9114.60.0987.15.7–8.4−–8.3
 Sorbus aucuparia1642183.60.0529.68.510.4−165.80.06910.18.7–11.5−–11.2
 Betula pubescens1144128.10.03713.611.716.2−120.80.05013.912.2–17.6−126.71.310.06713.211.1–15.9
 Pinus sylvestris53977.90.01241.334.551.8−67.00.01448.540.3–66.5−70.41.300.02437.320.8–52.5
Figure 2.

 Comparison of linear, exponential and sigmoid best-fit decay model estimates for each substrate (a–h). Model parameters’ and residual variance estimates are provided in Table 1. Two-year decomposition vectors are displayed for each model type. Initial decomposition vectors are displayed in black because they are set at t = 0 for all three model types. Estimated models do not extend beyond data range.

Figure 3.

 Comparison of each substrate’s best-fit decay model on a common time axis.

A 2.5-fold variation in T1/2 was observed across species for coarse stems, ranging from 6.8 years (6.1–7.5, 95% C.I.) for Alnus incana to 17.5 years (14.3–20.1, 95% C.I.) for Pinus sylvestris, and a 3-fold difference for coarse roots, from 13.6 years (11.7–16.2, 95% C.I.) for Betula pubescens to 41.3 years (34.5–51.8, 95% C.I.) for Pinus sylvestris (Table 1; Fig. 3). Across all species and organs, a 6-fold variation in T1/2 was observed. A major difference in shape between root versus stem decay curves arose from the lack of initial lag phase of both Betula and Pinus roots as compared with their respective stems (Fig. 3). However, while the T1/2 of Pinus roots was double that of Pinus stems (19.0 vs. 39.9 years), Betula stems and roots had rather similar T1/2 (14.5 and 13.6 years, respectively; Table 1).

Wood debris T1/2 was strongly positively related to initial WD lignin content (R2 = 0.73; < 0.01) and DMC (R2 = 0.79; < 0.01) and negatively related to WD initial pH (R2 = 0.92; < 0.001) but not significantly related to initial N, P, C/N, lignin/N and wood density (Fig. 4). Lignin content, DMC and pH were strongly correlated (r ≥ 0.84 and < 0.01 for each pairwise correlation).

Figure 4.

 Predictions of woody debris decay rates by initial wood lignin, dry matter content (DMC) and pH.


Our new short-term method provides reliable long-term estimates of wood decay dynamics (following different mass loss models) and comparative decay rates across species and substrate types. Common litter bed conditions included a broad pool of fungi (WD gathered from a number of locations within the studied area with their respective biota), even contact of WD with the soil and consistent log size, together potentially minimizing the interactive effects of environmental covariables on tree species differences. Litter beds allow several extensions testing for the effects of such covariables (e.g. log diameter, surrounding substrate type, fauna exclusion, humidity, light or temperature conditions). Though, our approach may also be implemented in the field (e.g. for between-site comparison) to directly represent in-situ conditions. Therefore, our relatively easy and rapid method may be applied in any biome with slow-turnover litter. In environments where wood borers are predominant drivers of wood decay, particular attention will, however, need to be paid to the representativeness of the WD sampling. Here, we used this method to single out species contributions to long-term variation in decomposition processes and rates of coarse WD. Although further empirical evidence for more WD types is needed to back up our initial conclusions, our results do support our hypotheses indicating (i) substantial interspecific and inter-organ differences in the mass loss dynamics of coarse WD in a subarctic flora and (ii) good predictive powers of several wood functional traits on long-term wood decay rates.

Interspecific Variation in Long-Term Wood Decay Dynamics

The substantial variation in wood decay dynamics between species (e.g. Populus vs. Sorbus stems) and organs (e.g. Pinus roots vs. stems) contradicts the widespread assumption that WD decay typically follows single (Olson 1963) or multiple (Minderman 1968) negative exponential functions (Table 1, Fig. 2). Our standardized incubation environment revealed species-specific and organ-specific differences in wood decay dynamics that should theoretically arise from differences in wood traits and/or their interactions with this environment. These differences are most pronounced at the early stages of WD decay. Thus, our model estimations indicated a 5-year initial lag time for Pinus stems and 2-year lag times for Populus and Betula stems while this lag was absent in other WD types. Previous studies have also observed such initial lag phase in both tropical and boreal ecosystems on Coccothrinax readii (Harmon et al. 1995), Picea abies (Naesset 1999), Pinus contorta (Laiho & Prescott 1999) and Pinus sylvestris (Harmon, Krankina & Sexton 2000), up to 5 years, before decay rate increased rapidly. These observations need to be differentiated from observed lag arising from prolonged periods where stems remained standing before downing (e.g. Yatskov, Harmon & Krankina 2003; Mäkinen et al. 2006; Tuomi et al. 2011). Nevertheless, the slow initial decomposition observed here may also relate to the preliminary phase of microbial and/or invertebrate colonization necessary to initiate decomposition (Swift, Heal & Anderson 1979; Harmon et al. 1986; Laiho & Prescott 1999; Yatskov, Harmon & Krankina 2003). Indeed, the complete spread of decomposers throughout large WD can take several years (Harmon 2009). In the absence of mechanical injury, colonization of sapwood may only begin once bark has dried and cracked (Käärik 1974), galleries have been dug by invertebrates (Ausmus 1977; Schowalter et al. 1992) or bark defences have been overcome by microbes (Käärik 1974). Differences in bark structure and composition between WD species and organs might therefore explain differences in initial decay rates and thus in the decomposition model that describes these processes. Besides, while the resistance of sapwood to decay is low for most tree species, heartwood resistance is highly variable across species (Käärik 1974) owing to large variations in their non-structural secondary metabolite contents (50-fold in angiosperms, 20-fold in gymnosperms; Cornwell et al. 2009). Therefore, once decomposing organisms have at least partly overcome bark defences and further colonized the relatively nutrient-rich and less chemically defended sapwood, their inward progression rate will depend on the content of those highly species-specific heartwood secondary metabolites (Käärik 1974). Faster decomposer colonization rates of angiosperm versus gymnosperm logs have also been tentatively attributed to the morphology of sap conduits (Carlquist 2001; Cornwell et al. 2009) and hydrophobic properties of gymnosperm resins (Pearce 1996).

In contrast to the common assumption of constant relative (i.e. instantaneous) mass loss rate throughout, dynamics of intermediate wood decay stages also appeared variable across WD types. Most WD progressively showed increasing relative mass loss rates, at least down to 50–60% mass loss (Fig. 3), possibly because of the progressive loss of initial heartwood defences over time (Harmon et al. 1986). In contrast, fast decomposing WD such as Alnus and Salix stems, with typically little heartwood defence, did not show such acceleration in relative mass loss rate (exponential model). Several other factors could simultaneously account for an increase in relative decomposition rate from early to intermediate stages of decomposition. First, leaching of constitutive compounds of wood (e.g. dissolved organic carbon, polyphenols; Spears & Lajtha 2004) should progressively increase with decay stage as permeability to water and microbial colonization rises, and the highly polymeric wood compounds are degraded into soluble fractions (Harmon et al. 1986). This should be more pronounced in WD of high wood density or in species producing hydrophobic resins (e.g. Pinus WD). Second, as wood decay advances, microbial decomposers gather and accumulate nutrients limiting microbial growth and activity (such as N) through a variety of mechanisms (Cornwell et al. 2009). This strong initial nutrient acquisition by microbial decomposers may be progressively reinvested towards lignocellulolytic enzyme production as wood decay advances (Sinsabaugh et al. 1993; Weedon et al. 2009). Finally, wood fragmentation by invertebrate decomposers and microbial action reduces wood particle size, and the consequently higher surface-to-volume ratio should support faster decomposition (Harmon et al. 1986).

After 60–75% mass loss, the state of WD fragmentation makes it difficult to identify species and measure WD density accurately (Christensen 1984). Nevertheless, methods exist to overcome this constraint (see Methods). Despite lacking observations from the final phase of decay, our results suggest that most WD types may already show a stabilization in their relative mass loss rate within our data range. Besides, the three substrates that did not show any inflection (linear model) also displayed the highest lower limit of model prediction (around 60%), which does not preclude a later stabilization in their relative mass loss rate. Here, absence of data beyond 60–75% mass loss limits our interpretation of the dynamics of final WD decay. The very few studies on the latter stage of WD decay documented results varying from strongly decreasing decay rates beyond 50% mass loss (e.g. Harmon et al. 1995; Krankina & Harmon 1995) to no clear decrease in the latter phases of decay (Harmon, Krankina & Sexton 2000; Mäkinen et al. 2006). These observations suggest that late-phase decomposition rates may also be species-dependent (see also Harmon, Krankina & Sexton 2000).

Interspecific Variation in Wood Mass Loss Rates

Modelling, as applied here, can only provide estimations of mass loss curves and loses accuracy in the final phase of WD decay. To minimize the impact of these factors on estimates of WD decomposition rates, we used decomposition half-life (T1/2) as a measure of mass loss rate that provides comparable values across models (Table 1) without extrapolating beyond our data range. The large differences in T1/2 across species and organs of similar WD shape and diameter, when incubated in a common environment, support our first hypothesis of great interspecific differences in the initial and intermediate phases of WD decomposition. Although more WD types need to be compared, also in other regions, to make our conclusion more robust, the importance of wood traits in driving wood decay processes (second hypothesis) was confirmed by the good relationships found between decomposition rates and several wood functional traits such as lignin content, DMC and pH. Wood pH, which is generally negatively correlated with the nutritional status of decomposing material and positively with the initial amount of antimicrobial organic acids (Cornelissen et al. 2006), was a strong predictor (R2 = 0.92) of WD decay rates. Indeed, non-structural secondary metabolites such as organic acids are known to provide WD heartwood with high decay resistance (Harmon et al. 1986). As one of the most recalcitrant compounds of wood (Käärik 1974), lignin content represents the ratio of recalcitrant to more degradable tissues and this is reflected in its strong correlation with DMC (Garnier & Laurent 1994). Besides, the distribution of lignified structures throughout woody tissues likely constrains access of non-lignin-degrading microbes to more degradable materials such as cellulose (Scheffer & Cowling 1966). Nevertheless, these results contrast with recent studies (e.g. van Geffen et al. 2010) or literature surveys (Cornwell et al. 2009; Weedon et al. 2009), which did not find very strong predictive power of lignin with respect to WD decomposition rates. This discrepancy could stem from the relatively low secondary metabolite contents of our WD types compared with the strong defensive effect of secondary metabolites that may have masked the lignocellulose effect in other studies. Besides, the WD types represented in our study featured a wide range of lignin contents and did not show the potential threshold in wood decay rate observed by Taylor et al. (1991) at high levels of lignin content, which can potentially hide the overwhelming impact of lignin in regression analyses (Prescott 2010).

Initial wood nutrient content (N, P and C/N) was non-significantly related to WD T1/2, confirming that, while initial wood nutrient content generally matters for decomposition (Weedon et al. 2009), its influence on wood decomposition is rather complex, like in leaf litter (e.g. Hobbie 2005). For instance, van Geffen et al. (2010) proposed that wood nutrient content might not be a strong control on decomposition rates when wood C/nutrient ratios are within the range of those from wood decaying fungi – e.g. C/N of 40–400 (Dix & Webster 1995) as compared with C/N of 70-395 in our study. Besides, microbial decomposers may gather nutrients from external sources, thereby alleviating their potential nutrient limitation (Cornwell et al. 2009).


This study has demonstrated the potential for a novel short-term common-garden wood decomposition experiment to provide estimations of long-term wood decay dynamics and rates at the tree species level. As such, it also opens up promising perspective for comparing woody versus non-woody materials of highly contrasting decomposition rates and distinguishing between environmental (moisture or temperature regime, soil macro-fauna presence/absence, etc.) and wood functional trait effects, while reducing environmental noise (Freschet, Aerts & Cornelissen 2011).

Our model estimates have also provided support for contrasting WD mass loss dynamics, displaying between one and three distinct phases depending on wood species and organ (stem versus root). As such, they suggest that the widespread use of exponential functions to model WD decay should be regarded more critically as it likely overlooks the complexity and diversity of wood decay processes. The 6-fold difference in decomposition rates across WD of similar shape and diameter decomposing in the same environment suggests, as shown for other litter types (Silver & Miya 2001; Cornwell et al. 2008), an important role for wood functional traits as drivers of wood decomposition. Here, initial wood pH, lignin and DMCs were good predictors of wood decay rates. These results suggest that the wood decomposition processes underlying changes in wood decay dynamics may vary across wood species and types, for instance under influence of wood functional traits or decomposer community interactions.


Our acknowledgments go to the Abisko Scientific Research Station (ANS) for hosting this experiment. We are grateful to Bob Douma for helpful discussion on decomposition modelling and to Bruno Hérault and Jacques Beauchêne for their outstanding review of this manuscript. We also would like to thank Amy Austin and Cindy Prescott for providing a number of constructive comments on this study. G.T.F. was supported by EU Marie Curie training network MULTIARC contract MEST-CT-2005-021143; R.A. by EU ATANS grant Fp6 506004; and J.H.C.C. by grants 047.017.010 and 047.018.003 of the Netherlands Organisation for Scientific Research (NWO).