Determinants of the pathways of litter chemical decomposition in a tropical region



  • Litter decomposition is a key ecosystem process, yet our understanding of the drivers in chemical changes in leaves during decay is limited. Our aim was to determine the comparative differences (chemical divergence or convergence) between sites and the drivers of decay pathways.
  • We used the litterbag method (‘in situ’ litterfall and standardized ‘control’ leaves) in Australian tropical rainforests and near-infrared spectrometry to show the chemical pathways during decomposition (c. 360 d; 12 control sites; 17 in situ sites). Chemical convergence/divergence was determined from spectral dissimilarity and quantile regression along a mass loss moving average. The influence of environment (climate and soil) and litter quality on decay pathways was determined between sites using correlation analysis.
  • Throughout the region, litter composition in both treatments converged chemically during decay. However, divergent chemical pathways were shown for some samples/sites (especially with high initial lignin, phenolics and carbon (C), poor soil phosphorus (P), sodium (Na) and more seasonal moisture), and the diversity of decay residues increased with mass loss despite overall chemical convergence.
  • Our study suggests that there is general chemical convergence of leaf litter during early decay, but also that divergent chemical pathways occur in locations that experience more intense seasonal drying, and contain species or conditions that promote poor-quality litter.


The drivers of the direction of chemical changes in leaf litter during decomposition on forest soils are relatively poorly understood when compared with rates of decay (Wickings et al., 2012), yet their comprehension is essential in understanding forested ecosystem processes and biogeochemical responses to climate change. The general consensus is that leaf litter decomposes and chemically transforms more rapidly in its home environment (‘home field advantage’) as a result of specialization of decomposers (Ayres et al., 2009; Wallenstein et al., 2013), and that, overall, the chemistry of diverse litters converges during decay (Fierer et al., 2009; Homann, 2012; Wallenstein et al., 2013). However, recent work using modern laboratory techniques has highlighted the fact that divergent chemical properties may also emerge during decay, particularly as a result of differences in soil biological activity (Wickings et al., 2012) and different microbial residues, which make up the bulk of soil organic matter (SOM) (Miltner et al., 2011). While both divergent and convergent chemical decomposition pathways exist on different sites (Wickings et al., 2012), in order to fully understand decomposition processes there is a need to explore the factors that affect the trajectory of litter chemistry during decay (Wallenstein et al., 2013).

It is important to fully comprehend the influence of environment (e.g. climate, soil) and biology (e.g. plant species composition/litter chemical quality, soil activity) on the pathways of decay in tropical forests, because of the large impact these environments have on global biogeochemical cycles (Malhi & Wright, 2004) and the potential for climatic change (i.e. warmer temperatures with more seasonal moisture) in coming years (Suppiah et al., 2007). Decay rates may increase in these forests with warmer temperatures, but increased rainfall seasonality would generally have the opposite effect (Parsons et al., 2012). However, the dynamics of this are not straightforward, because a higher concentration of recalcitrant components in leaves (e.g. tannins/polyphenolics) makes decay more sensitive to changes in both moisture and temperature (Fierer et al., 2005; Suseela et al., 2013). Climatic influences on plant traits may also affect litter chemical recalcitrance (Bais et al., 2011). This may be coupled with a succession of species with increased degrees of leaf toughness and sclerophylly (Read et al., 2009). These changes will have an impact on litter decay and nutrient and carbon (C) cycles (Hättenschwiler & Vitousek, 2000; Caldwell et al., 2003; Fierer et al., 2005).

It is suggested that climate change will result in varying shifts in litter quality and decay dynamics, owing to the uneven climatic effects across sites with varying litter quality. However, the importance of different drivers in determining variability in chemical properties and pathways during decay remains largely unresolved (Wallenstein et al., 2013). Modern chemical techniques can improve insights into ecological processes that were once too complex to define (Foley et al., 1998). However, to date, most work following broad chemical changes in leaf litter during decay across multiple sites has been restricted to temperate and managed habitats (Gillon & David, 2001; Preston et al., 2009a,b; Homann, 2012; Wickings et al., 2012; Wallenstein et al., 2013). Considering that the current body of literature on the pathways of decay is limited for the tropics, the goal here was to compare chemical information on leaf litter during decomposition throughout a biodiverse and seasonally wet, tropical rainforest region. The aim was to determine the patterns of chemical changes in the region and the relative pattern of convergence or divergence in litter chemical properties.

In this study, we used near-infrared spectra (NIRS) as indices of chemical composition to understand the drivers of decay pathways of leaf litterbag samples decomposed at sites throughout the Australian wet tropics. We then compared the chemical differences and influences of litter quality and environment (climate and soil) on the patterns seen. We did so because climate and litter quality broadly determine the effects of soil fauna (Yang & Chen, 2009; García-Palacios et al., 2013) and microbial activity (Schimel et al., 2007; Bray et al., 2012; Wallenstein et al., 2013), so we expected these elements to control the divergence or convergence of litter properties in determinable ways. Untangling these complex interactions is important, as the impact of future climates on decomposition processes will depend greatly on the interaction of environmental variability and litter quality (Suseela et al., 2013).

Materials and Methods

Our leaf material came from the regional litterbag study of Parsons et al. (2012), using locations throughout wet, tropical north Queensland, Australia. This work has already highlighted litter chemical quality (especially phosphorus (P), total C and phenolics), temperature and moisture seasonality as primary drivers of leaf decomposition rates in the region (Parsons et al., 2012), and determined methods to show chemical compositional changes and chemical decomposability in the material with NIRS (Parsons et al., 2011). In the present work, we extended the NIR study of Parsons et al. (2011), reanalysing the spectra to explore how the leaves change relative to each other during decay, in order to show how the chemical pathways of decay either diverge or converge.

Field study of leaf litter decay and NIR spectral analysis

Near-infrared spectra contain comprehensive information on the number and type of C–H, N–H and O–H bonds in the material, and provide accurate estimates of organic chemical differences in plants and soils, particularly of biological origins (Foley et al., 1998; Soriano-Disla et al., 2014), including decay residues (Gillon et al., 1999; Gillon & David, 2001; Ono et al., 2003; Terhoeven-Urselmans et al., 2006; Parsons et al., 2011; Soriano-Disla et al., 2014). Other techniques may provide somewhat higher chemical resolution than the NIR spectra, but NIR spectral analysis is generally of lower cost and provides relatively faster throughput. This allows larger sample sizes while still accurately estimating a wide variety of chemical attributes. NIR spectral analysis is also particularly useful for showing relative chemical differences (Foley et al., 1998).

We used the NIR spectra of the litterbag treatments in Parsons et al. (2011, 2012): ‘in situ’ and ‘control’ litters. For the in situ and control treatments, 17 and 12 sites were used, respectively. Fewer sites were used for the control treatments because of limitations in material; however, sites used for both treatments covered the same range in climate and soil fertility (Parsons et al., 2012). The Supporting Information (Table S1) shows the characteristics of the sites used and the locations of the treatments. This was a large portion of the variability in the region for temperatures (23.5–17.7°C mean annual temperature), mean annual rainfall (3419–1594 mm yr−1), precipitation seasonality and soil types (e.g. old oligotrophic granite to oligotrophic-mesotrophic basaltic formations). In this region, annual rainfall is generally high across sites, with regional variance in rainfall patterns resulting largely from varying totals in the winter months (dry season; Parsons, 2010).

The in situ samples comprised leaf litterfall collected at the site and decomposed in litterbags on the soil near the location where they were collected. These samples were from the peak litterfall time for the region (beginning of the wet season) (Parsons et al., 2014) and consisted of mixed, naturally senesced leaves characterizing the litterfall of the site (Parsons et al., 2012). The aim was to follow the decomposition of this peak litterfall event to understand decomposition relative to the site. Litterfall samples were pooled from 10 litterfall traps and then dried and the leaf component isolated. Only fresh, naturally senesced material was placed in the bags; that is, green leaves and partially decomposed material were excluded (Parsons et al., 2012). Plant species richness varied between plots/in situ litterbag placements, but was generally high and no single species dominated the study material for any of the locations (Parsons et al., 2014).

Our control litter consisted of the naturally senesced leaves of the semideciduous Australian rainforest species, Archidendron vaillantii (F. Muell.) F. Muell. The control treatment allowed for site differences to be exposed without the influence of variability in litter quality. The in situ treatment allowed for ‘home field advantage effects’ (e.g. see Ayres et al., 2009) to be reduced and decomposition to be studied relative to the site and the composition of litter produced therein (Parsons et al., 2012). Despite the control litter being a legume species with relatively high N, it also had, on average, higher total phenolics and lignocellulose contents than most of the in situ samples (Parsons et al., 2012), potentially limiting decay rates.

To lower the chances of variability in litter mixes in the bags, we mixed/homogenized leaves as best as possible before placing them in the litterbags, for each in situ site and the control sample. We placed c. 5 g of dried leaves inside litterbags for decomposition (2 mm mesh size with c. 3 cm gaps on the sides to allow faunal movement through the bags) (Parsons et al., 2012). Samples were placed on the soil in the middle of the wet season and removed on five occasions for chemical analysis until around midway through the second wet season (Parsons et al., 2012). Here, to determine chemical composition with NIRS and to show the relative chemical dynamics, we used the NIR spectra from five litterbags per site per treatment from time intervals 0, c. 20, c. 50, c. 110, c. 230 and c. 360 d (Parsons et al., 2011). The time zero/initial compositions were determined from sets of c. 5 g of leaf litter kept aside from exposure in litterbags. For the control, this consisted of 10 samples per site, and for the in situ treatment it consisted of five samples per site.

Standard methods for NIRS are well documented and were adhered to in this study (Shenk & Westerhaus, 1991). Each litterbag and initial composition sample was dried, ground and scanned with NIRS in replicate. Details of the approach for NIR spectral data preparation and determination of chemical compositions can be found in Parsons et al. (2011). This included spectral pretreatments and smoothing, wet chemistry on a representative calibration set (as described earlier), regressions/calibrations for each chemical component, and then predictions of the full sample set (i.e. before and after decomposition) using these calibrations (Shenk & Westerhaus, 1991; Parsons et al., 2011). The mean of the replicate spectra (as first derivative spectra of 1100–2500 nm) for each bag/sample was used in analyses (Parsons et al., 2011). All initial chemical compositions for both treatments are shown in Table S2 and are discussed in Parsons et al. (2012). Mean initial nitrogen (N), P, C, calcium (Ca), lignin, α-cellulose and phenolics were 1.89, 0.02, 49.8, 0.25, 41.8, 29.0 and 0.81%, respectively, in the control litter, and were in the ranges 0.85–1.7, 0.02–0.08, 47.0–50.8, 0.25–2.12, 30.5–44.0, 17.8–23.0 and 0.33–1.0%, respectively, in the in situ litter (Table S2).

Patterns of chemical changes during leaf decay

To improve on the principal component analyses (PCAs) in Parsons et al. (2011) and better show chemical patterns during breakdown, we reanalysed the spectra using kernel PCA (kPCA) (Mika et al., 1999). kPCA is a nonlinear ordination method and is better at maximizing the identification of chemical differences in the spectra than conventional PCA (Wu et al., 1997; Langeron et al., 2007; Labbé et al., 2008; Williams et al., 2009). Langeron et al. (2007) and Labbé et al. (2008) provide extensive background on the mathematics and intricacies of kPCA application to NIR data. Here we followed this approach to produce a kPCA map showing broad chemical differences in the leaf litter during decomposition. The method included standard approaches to NIR spectra as mentioned earlier and in Parsons et al. (2011), and for kPCA application, it included the selection of the optimal kernel shape parameter (radial basis Gaussian projection) (Langeron et al., 2007; Labbé et al., 2008). To maximize model accuracy of the kPCA, randomized k-fold ‘leave-10-out’ cross-validation was used in model building (Baumann, 2003). All outlier detection and PCAs were achieved using the ‘chemometrics’ (Filzmoser & Varmuza, 2011) and ‘kernlab’ (Karatzoglou et al., 2004) R packages (R Development Core Team, 2013). The optimal kernel parameter selection was undertaken using grid search in the Unscrambler X software (CAMO, Norway 2012) in classification mode.

To show the patterns with regard to known variables during decay, the sample scores for the kPCA axes were correlated (Spearman rank) with chemical composition: total N, P, C, Ca, acid-unhydrolysable residue (AUR) and acid-hydrolysable residue (AHR) as determined with the NIRS calibration approach. AUR and AHR portions are defined, respectively, as the lignin and α-cellulose components of leaves after exposure on the soil from the acid detergent fibre method (Rowland & Roberts, 1994).

Patterns of convergence or divergence in litter chemistry

To show comparative chemical changes in litter during mass loss (i.e. changes relative to mass loss) and further explore the drivers of any deviations seen, we synthesized the multivariate data of the kPCA by determining the dissimilarity (Bray–Curtis) between points (i.e. litterbag samples) in the full score space (Strukelj et al., 2013). The Bray–Curtis values denoted the relative chemical differences and were scaled to between 0 and 1 before analysis (i.e. most similar to most dissimilar). Following this, a moving average approach similar to time-lag analysis (Collins et al., 2000) was used to enable the most robust use of the available data to view comparative trends in decay pathways between sites – this is termed ‘masslag’. Analysing dissimilarity values with this approach is a powerful way of viewing patterns and variability in ordination data (Collins et al., 2000). In this manner, smoothing was achieved by rounding mass loss values for each litterbag sample (as % dry mass remaining) to 10%. Masslag grouping was then created as, for example, 30% masslag smoothing groups: 100–70, 90–60, 80–50, … 40–10%; 40% smoothing: 100–60, 90–50, 80–40% etc. Patterns were initially tested with 20, 30, 40 and 50% masslag groupings to find the smoothing level to best show patterns, while including sufficient samples to show trends (i.e. if the smoothing level was too low, relatively few samples would potentially be included in some of the groups, especially mass ranges towards the lower end of mass loss) (Collins et al., 2000). After an appropriate smoothing level was chosen for each treatment, we analysed the relationship between masslag and chemical dissimilarity to show the relative changes in leaf litter during decay.

Considering the complex nature of the data, it was likely that more than one (e.g. mean) linear relationship was present between masslag and chemical dissimilarity. In particular, if chemical divergence had occurred, we would expect some samples to show deviated or anomalous changes relative to other samples. To gain a more comprehensive picture of the relationship, we used quantile regressions. This works by splitting the data into percentiles (tau, τ); here, the data were split into Bray–Curtis chemical dissimilarity percentiles. We set τ to 0.05, 0.10, 0.3, 0.5, 0.7, 0.9 and 0.95; that is, from most similar to most dissimilar (Koenker, 2005). Regressions were then run for each quantile. A positive regression slope suggested chemical divergence, and a negative relationship chemical convergence, during decay. Normality of the dissimilarity data was improved with square root transformations of the distance measures before analysis.

Litter quality and environmental influences on chemical patterns

To determine what litter quality (in situ treatment) and environmental factors (control and in situ treatments) drove any patterns present (e.g. divergence, uniform trends/no relationship or convergence), we used correlation analyses on the masslag/dissimilarity relationships, making use of the quantile groupings. The raw environmental and initial litter quality data used for this can be found in Tables S1 and S2. For the correlation analysis, a matrix of dissimilarity for each environmental/litter quality variable for comparisons between litterbag samples was calculated. For example, for initial N dissimilarity: sample 1 compared with sample 2 (Ndiss(1–2)) = Nsample1 – Nsample2, where Ndiss(1–2) is the dissimilarity measure and Nsample values are the initial N contents of the respective samples. Doing this for all combinations, we produced an n × n dissimilarity matrix for each variable. A correlation value for each variable with masslag was then obtained, doing so for τ groups: upper 95%, mean and lower 5%. A negative correlation for a variable within a quantile group suggested influence on convergent chemical patterns between sites (i.e. decreasing chemical differences between samples during decay). Conversely, a positive correlation denoted influences on divergent patterns (i.e. increasing chemical differences during decay). For the in situ litter, this was done for all initial litter chemical quality variables, and for in situ and control litter samples, it was done for climate (total rainfall over the experiment, mean annual temperature and dry season moisture) and soil (total N, total P, organic C and total Na) (Table S1). Soil data came from three randomly chosen samples around the litterbag placements. Average values of 0–10 and 20–30 cm depths were used (Parsons, 2010). The moisture seasonality variable is from leaf wetness sensors deployed at the sites (Parsons et al., 2012).

Finally, to further explain the masslag chemical dissimilarity patterns using the control litter, we created another dissimilarity matrix using the actual chemical compositions during decay (N, P, Ca, C, AUR and AHR) with the same technique as before (i.e. for each dissimilarity measure, calculating the relative difference in concentrations between the two samples). Absolute difference values and square root transformations were used. This matrix was then used to show the pattern of convergence or divergence in the chemical variables.


A total of 467 in situ and 294 control leaf litter spectral samples remained for comparisons between the NIR spectra after spectral outlier removal. The range of data remaining after this is shown in Table S3. For the in situ litter, this represented 100 (i.e. initial contents) to 9.6% original dry mass remaining. For the control treatment, slower decay was seen with a range of 100–33.7%. For both treatments, changes in N, P and Ca over decay can be summarized as immobilization (i.e. increases in comparison to original contents) with some mineralization of N in later samples; AUR increased then decreased, while AHR and total C decreased as decay progressed (axis correlations in Fig. 1a and Table S3). For both the control (Fig. 1a) and in situ litter (Fig. 1b), patterns of chemical change were represented by eight kPCA components, together explaining 98% of total spectral variance, for each treatment.

Figure 1.

Leaf litter chemical changes in control (a) and in situ (b) treatments, from 12 and 17 sites, respectively, during 1 yr decomposition in a tropical rainforest region. The first three principal component axes (from kernel principal component analysis (kPCA) of a total of eight axes) are shown, representing chemical differences derived from near-infrared spectrometry over decay. Points show individual litterbag samples (these are graduated in colour according to % mass remaining, from blue to red). Correlations with the kPCA axes are shown for contents of carbon (C), dry mass (mass), phosphorus (P), nitrogen (N), calcium (Ca), acid-unhydrolysable residue (AUR) and acid-hydrolysable residue (AHR).

Comparative changes in chemical components in the during decay (control litter)

The relative patterns of chemical indices during decay yielded different results, denoting uniform, increasing or decreasing similarity in variables over mass loss (Fig. 2). For N, we noted relatively uniform changes throughout the region over mass loss (Fig. 2a and Table 1; no linear relationship between relative N differences and masslag, = 0.52). For P, AUR and AHR, concentrations converged (Table 1, Fig. 2b,e,f; negative linear relations, P < 0.001). Conversely, divergence occurred in the proportions of Ca and C during decay (positive linear relationships, < 0.001). For C and Ca, dissimilarity in compositions decreased until it was between the 80–40% and 70–30% mass loss smoothing groups, after which it increased, largely driving the divergence seen (Fig. 2c,d).

Table 1. Linear regression statistics for mass loss vs the absolute relative differences in chemical concentrations in decomposed leaf litters in Australian tropical rainforest (standard control litter; see Fig. 2 for plots of these trends)
 SlopeInterceptP (slope)
  1. Significant negative or positive slopes relate to convergence or divergence, respectively, in chemical composition during mass loss between samples. Mass loss refers to masslag, which is a smoothed/moving average, as defined in the text.

  2. AUR, acid unhydrolysable residue; AHR, acid unhydrolysable residue.

Phosphorus−0.0030.109< 0.0001
Carbon0.0891.02< 0.0001
Calcium0.0480.53< 0.0001
AHR−0.182.18< 0.0001
Figure 2.

Absolute relative differences (%) in litter chemical compositions during leaf decay in mass loss smoothing groups (masslag) for nitrogen (a), phosphorus (b), calcium (c), total carbon (d), acid unhydrolysable residue (AUR, e) and acid hydrolysable residue (AHR, f), from the control leaf litter decomposed at sites in Australian tropical rainforest. The slopes of linear regressions around the mean are shown (ns, no relationship; +ve, positive regression slope/divergence in concentration; −ve, negative slope/convergence; ***, < 0.001). See Table 1 for regression statistics.

Comparative chemical dynamics during decay of the control litter (convergence or divergence)

In the control litter, a 40% smoothing provided sufficient samples in masslag groups to show trends. A mean chemical convergence trend was seen in the control litter, with the 70–30% and the 80–40% groups showing lower median dissimilarity than the 100–60% group (5.0 and 5.9% decreases in dissimilarity from 100 to 60%, respectively; Fig. 3(a) and Table 2; τ = 0.50: slope = −0.009, = 0.0005). However, divergent patterns over mass loss occurred for the most spectrally/chemically dissimilar samples (Table 2, τ = 0.95: slope = 0.006, < 0.001). The most similar samples diverged slightly over mass loss (Table 2, Fig. 3a, τ = 0.05, slope = 0.004); however, this relationship was not significant (= 0.31) and was probably driven by only a few samples, as the 10th and 30th percentiles of dissimilarity corresponded with more uniform patterns of change (Table 2, Fig. 3a, τ = 0.1, = 0.44) and chemical convergence (τ = 0.3, slope = −0.007, = 0.002). The range of chemical compositions (diversity of chemical residues) in masslag groupings steadily increased over decomposition (error bars in Fig. 3a; e.g. SD of dissimilarity in the 100–60% group = 0.10, compared with 0.13 in the 70–30% group). This suggests, in the time/range of mass loss seen in this study, that decayed leaves became more chemically diverse than fresh litter, while still, on average, converging chemically. Correlations explaining these patterns in terms of environment did not show significant relationships for divergence in the 95th percentile (Table 3; no strong positive relationships). Dry season moisture, soil Na and soil P correlated significantly with the mean dissimilarity pattern (Table 3; Spearman rank correlation: dry season moisture, 0.112; soil P, 0.127; soil Na, 0.085; < 0.001 for all three), with positive correlations suggesting these factors contributed most to chemical variance.

Table 2. Linear regression statistics for mass loss (as masslag smoothing) vs chemical dissimilarity (between samples within masslag groups) for the control and in situ leaf litters decomposed in litterbags at different sites in Australian tropical rainforest (see Fig. 3 for plots of these relationships)
  1. Results shown are for fifth to the 95th quantiles (τ), corresponding to regressions using the most similar to the most dissimilar samples. Significant positive slopes relate to chemical divergence and negative slopes to chemical convergence during decay/mass loss.

Control litter
P 0.310.440.0020.00050.110.049< 0.001
In situ litter
P < 0.0001< 0.0001< 0.0001< 0.0001< 0.0001< 0.0001< 0.0001
Table 3. Correlations (Spearman ranks, control litter) of environment (as environmental dissimilarity; see text for definition) with chemical dissimilarity (Bray–Curtis between near-infrared spectra) between litterbag samples in mass loss (i.e. masslag) smoothing groups
  1. Data are split into chemical dissimilarity quantiles, τ: upper 95% (most dissimilar), 50% (mean), and lower 5% (most similar) (i.e. in Table 2 and Fig. 3a). Significant positive correlations suggest influence on divergent chemical properties during mass loss for the relevant quantile group.

  2. ***, < 0.001; **, < 0.01; *, P < 0.05.

Dry season moisture−0.156*0.112***0.003
Soil nitrogen0.0210.0070.035
Soil phosphorus−0.202**0.127***0.099
Soil carbon−0.0470.0140.035
Soil sodium0.0050.085***−0.050
Figure 3.

Relative changes in the chemical dissimilarity of leaf litter grouped by mass loss (i.e. along a smoothing series, termed ‘masslag’) for control litter (a) and in situ leaf litter (b) decomposed in litterbags from multiple sites in Australian tropical rainforest. Points/boxplots show Bray–Curtis dissimilarities of near-infrared spectra comparing litterbag samples in masslag groupings. Quantile regression lines are shown for τ = 0.05, 0.1, 0.3, 0.5, 0.7, 0.9 and 0.95 (bottom to top along y-axis), corresponding to regressions using the most similar to the most dissimilar samples. Negative slopes suggest chemical convergence and positive slopes suggest chemical divergence during mass loss for the respective τ. See Table 2 for the regression statistics relating to each τ grouping. The length of the error bars denotes the diversity of chemical residues in groupings (see text for relevant errors). For the control, 40% smoothing groups were used, while for the in situ litter 30% groups were used (see the Materials and Methods section for a detailed definition of masslag).

Comparative chemical dynamics during decay of the in situ litter (convergence or divergence)

For the in situ litter, a 30% smoothing provided a sufficient spread of samples in groups, while still showing trends. For this treatment, mean Bray–Curtis dissimilarity decreased during decay (Fig. 3b; 7.7% decrease in the median dissimilarity when comparing 100–70% with 50–20% masslag). Similarly, the mean linear model showed overall convergence (Fig. 3b, slope = −0.008, < 0.0001, Table 2). The convergence trend (negative linear regressions) was most prevalent in the middle to lower quantiles of chemical dissimilarity. This suggests that the most similar samples at the start of the study became even more similar chemically during mass loss (Fig. 3b). Despite this, as for the control, the upper 90% of dissimilarity measures in the in situ litter saw positive linear trends of dissimilarity and masslag (slope = 0.004, < 0.0001 for τ = 0.95), or chemical divergence (Table 2, Fig. 3b). The range of chemical compositions in masslag groupings (error bars in Fig. 3b) increased over decay (e.g. SD of dissimilarity in the 100–70% group = 0.10; that in the 50–20% group = 0.13), while similarity increased, similar to the control.

Poor litter quality, especially high initial lignin, C and phenolics, aligned with the samples that showed chemical divergence (i.e. positive correlations) in the upper 95% of the data (Table 4): initial total phenolics (Spearman rank correlation = 0.299, = 0.0001), initial C (0.229, = 0.0001), initial lignin (0.120, = 0.0002, Table 4). Environmental variability (soil and climate) did not explain chemical divergence in the in situ litter with any significance (Table 4); however, all environmental variables were important in driving chemical changes, notably towards convergence in chemical properties (Table 4; τ = 0.5, all < 0.001 correlations and negative slopes with masslag), with soil P and dry season moisture being the strongest drivers (correlation −0.182 and −0.165, respectively; < 0.001).

Table 4. Correlations (Spearman ranks, in situ litter) of litter quality and environment (as dissimilarity; see text for definition of how dissimilarity was calculated) with chemical dissimilarity (Bray–Curtis between near-infrared spectra) between litterbag samples in mass loss (i.e. masslag, see text) smoothing groups
  1. Quantiles are defined in the same way as in Table 3. Significant positive correlations suggest influence on divergent chemical properties during mass loss for the relevant quantile group.

  2. ***, < 0.001; *, < 0.05.

Initial litter quality
Nitrogen (N)−0.212***−0.146***−0.010
Calcium (Ca)−0.230***−0.158***−0.019
α- cellulose−0.308***−0.234***−0.069***
Phosphorus (P)−0.217***−0.174***−0.048*
Carbon (C)0.229***0.180***0.047*
Dry season moisture−0.139***−0.165***−0.046*
Soil N−0.180***−0.072***−0.005
Soil P−0.276***−0.182***−0.050*
Soil C−0.122***−0.088***0.004
Soil sodium (Na)0.055−0.077***−0.085***


Chemical changes in the Australian rainforest region followed common patterns for litter on poor soils in the tropics (Tian et al., 1992; Parsons & Congdon, 2008), with general increases (immobilization) in nutrient contents (N, P and Ca) in the early to mid-stages of the study and decreases (mineralization) in C and cellulose. Generally, litter chemistry converged, but not uniformly, which is in line with other studies in different biomes (Wickings et al., 2012; Wallenstein et al., 2013). This convergence was still only slight throughout the region in the time period of this work, suggesting that chemical differences were maintained over decay. Some chemical components (C and Ca) diverged in concentrations with mass loss. This was probably a result of different mineralization rates (irrespective of amount of mass lost) of these components at different sites. We would expect that, as more mass loss took place (beyond the temporal extent of our study), chemical differences in litter in the region would have converged further (i.e. divergent samples would converge with longer time on the soil). However, the increasing diversity of chemical residues seen here (i.e. error bars in Fig. 3), compared with fresh litter, suggests that fundamentally different chemical compositions are produced in the formation of SOM from different plant communities/soil types/climates regardless of commonality in broad chemical composition (Wickings et al., 2012), seen here in litter also having very different turnover times (Parsons et al., 2012).

In general, deviations in the chemical pathways of litter decay are driven by differences in soil biological activity, with litter quality (as seen in the in situ study for AUR and phenolics) performing important regulatory effects (Wickings et al., 2012). Our study supports the proposition that concentrations of C compounds, such as condensed tannins and lignin, are primary determinants of litter decay pathways in tropical rainforests (Coq et al., 2010; Hättenschwiler et al., 2011); that is, these components explained the divergence pattern seen for the most dissimilar samples of our in situ litter. Our study also showed the abiotic factors controlling deviations in decay pathways in a relatively variable (environmentally) group of tropical rainforest sites. This was particularly related to variance in dry season moisture and soil fertility (especially P and Na); that is, it most strongly defined variability in the control litter dynamics. These determinants of variability in litter chemical changes during decay also control decomposition rates and C cycling in tropical forests (Vitousek, 1984; Parsons & Congdon, 2008; Kaspari et al., 2009; Parsons et al., 2012). In turn, these factors lead to alterations in soil community composition and activity, greater specialization of soil biota (Wallenstein et al., 2013) and SOM formation and residues, with important broad-scale effects on C and nutrient cycles that impact on the entire biosphere (Wickings et al., 2012).

Although chemical convergence seems to be the general trend during decay, our results suggest that this may be an oversimplification of what occurs when comparing the early stages of decay for litter on varying soils and climates, even within a single region. This is probably a result of the imprints of initial litter chemistry and of different decomposer communities remaining in the material as microbial residues build up (Wickings et al., 2012; Wallenstein et al., 2013). Certainly, in the long term, as humic fractions take over dominance in residues, convergence occurs; however, in the early decay stages (as seen in the present study), leaves at drier sites and in poor-quality litters may diverge chemically, suggesting more diverse residues from these conditions. Here, by showing the divergent decay pathways relative to initial litter chemistry, we reiterate the remarkable importance of variability in litter chemical quality in determining decomposition and soil biotic patterns in natural systems (Bray et al., 2012).

Although we did not directly quantify soil biotic activity in this study, the importance of variability in soil communities cannot be overlooked. Importantly, it has been shown in the north Queensland region that strong seasonal drying in the winter months decreases CO2 and N2O outputs from soils, corresponding to falls in decomposer activity (Kiese & Butterbach-Bahl, 2002). Similarly, for more seasonally dry sites, mass loss rates from leaf litter slow substantially during the dry season, only to increase again during the wet season (Parsons & Congdon, 2008; Parsons et al., 2012). Here, in seasonally dry tropical forests, relative divergence of litter decay chemical pathways occurs in locations that experience more intense seasonal drying and contain species or conditions that promote poor quality litter. This promotion of litter recalcitrance also occurs as a result of low soil fertility, which alone reduces microbial growth rates with direct effects on decay and decomposer community specialization (Torsvik & Øvreas, 2002). Litter chemical quality is spatially and temporally varied in the Australian wet tropical region (Parsons, 2010; Parsons et al., 2014). However, litter produced from our sites was of a recalcitrant nature, including the control litter (e.g. compared with Wickings et al., 2012, who used grass and corn leaves, which are generally of higher quality than rainforest litter), even for tropical forests (Vitousek, 1984; Hättenschwiler et al., 2011), as a result of poor to extremely poor soils (Parsons, 2010; Parsons et al., 2014).

Interestingly, poor-quality litter may be, at least partially, climate-driven in seasonal rainforests, especially for some inhibitory phenolic compounds, which increase in concentrations with photodamage/higher solar radiation (e.g. under seasonal drought) and nutrient stress (Hättenschwiler & Vitousek, 2000; Close & McArthur, 2002). This climate–soil litter quality link produces a negative feedback on decay, and not only is it an important determinant of litter quality, but it also limits decomposition rates, increases litter standing crops (Parsons, 2010; Parsons et al., 2012, 2014) and determines deviations in decay pathways and litter residue formation (this study).

For the seasonally wet tropics, the potential for future increases in radiation and decreased rainfall in the transition from the dry to the wet season, and increased temperatures (Suppiah et al., 2007), suggests alterations in decay pathways that may be partly understood through viewing present-day responses to variability. This is likely to be led by a succession of species more adapted to increased temperatures and rainfall seasonality, with concurrent changes in leaf decomposability (Read et al., 2009; Parsons et al., 2014). The increasing effects of temperature and precipitation on poor-quality litter (Suseela et al., 2013) also point to greater degrees of change on poorer soils and in more stressed (e.g. moisture, solar radiation) and relatively cooler environments (e.g. uplands in the tropics) (Fierer et al., 2005). Our trend of divergent chemical pathways of poor-quality litter and in drier rainforests suggests that in soils we may see altered chemical residues from litter decay under future climate conditions. However, any alterations will be tightly coupled with shifts in microbial functionality and composition (Allison et al., 2013), and the complex effects of atmospheric pollutants (e.g. CO2, N deposition) (Lindroth, 2010) and disturbance on C and N cycles (Bernal et al., 2012; Parsons et al., 2014). How this balances out is a complex problem for modellers of C cycles and soil processes (Zhou et al., 2009).


The diversity of decomposition residues is not fully understood in complex natural systems. In tropical forests, the chemical pathways during decay are strongly correlated with initial chemistry (especially lignin, P and phenolics), climate (especially moisture seasonality in the seasonal tropics) and soil fertility (especially P and Na). Our study supports the idea that convergence of chemistry is a general trend (Wallenstein et al., 2013). Deviations from these pathways occur as a result of variability in both the decomposer community and the effects of poor litter quality (Wickings et al., 2012), especially C quality and annual moisture availability. Further, more diverse chemistries develop during decay than are present at litterfall (Wickings et al., 2012), but this is in line with overall convergence patterns. Extensions of this study could look at the compositions of unique residues quantified with the NIR spectra, or use other higher-resolution chemical techniques to add to our knowledge of the fate of leaf litter in SOM formation.


Stephen Williams is acknowledged for important contributions to the broader project on decomposition, nutrient cycling and climate change in Australian tropical rainforests. Luke Shoo, Collin Storlie, Vanessa Valdez-Ramirez, Yvette Williams, Jessica Cheok, David Coates, Joseph Holtum, the JCU-DPI Rapid Assessment Unit, the JCU Advanced Analytical Unit and the Queensland National Parks and Wildlife Service offered invaluable support at different stages of this work. This project was funded by the James Cook University School of Marine and Tropical Biology and the JCU Research Advancement Program, the Marine and Tropical Science Research Facility, the Skyrail Rainforest Institute and the Earthwatch Institute. Three anonymous reviewers improved earlier drafts of this manuscript.