To identify studies in peer-reviewed journals that provide species-specific stem wood C content data, we searched three databases (Web of Science, Web of Knowledge, Google Scholar) using the search terms “carbon” and “tree”, and for “carbon” and major tree tissue types (including “wood”, “bark”, “root” and “stem”). Additional studies were then identified in the reference sections of papers found in the databases. We included papers that were published or in press as of April 1, 2012. To ensure that only reliable and comparable data was used, studies incorporated into our analyses: (1) provided Latin botanical species names in order to avoid uncertain species assignments; (2) provided data from trees that were ≥10 cm diameter at breast height (dbh) to avoid confounding interspecific variation with size-dependent variation (e.g., differences in wood C between saplings and large conspecific trees [34]. See also discussion on within-stem variation below); and (3) explicitly described field- and lab methodologies. In cases where a single publication had two or more wood C records for the same species [35] we calculated the average value for use in our analysis. When conducting our search, we also found a small number of studies provided wood C data taken from different trunk heights. Here we chose to report data taken from 1.3 m dbh because this was the most commonly replicated sample methodology, and because the effect of sample height location on C content tended to be small (<1%) and non-significant [36,37,38]. In nearly all cases data was taken directly from tables, with two exceptions: figures were used to estimate wood C content for 32 tropical species (see Figure 1 in [4]), and to estimate C_vol for 7 temperate species (see Figure 4 in [1]).

For each published record we classified each species “type” as (1) angiosperm, or (2) conifer, and recorded species “biome” (or provenance) as (1) tropical, (2) subtropical/Mediterranean or (3) temperate/boreal. Provenance was determined based on the location where the trees were sampled, and not the species’ biogeographical origin (e.g., tropical angiosperm species sampled by Telmo et al. [39] in subtropical/Mediterranean Portugal, were classified as subtropical/Mediterranean). Temperate and boreal species were combined under one classification due to boreal records being found strictly in general analyses of North American trees [1] or Scandinavian studies that sampled south of the boreal zone [40].

In addition to reviewing published estimates of stem wood C content, we compiled data from studies presenting tissue-specific C content. In addition to the criteria noted above, we considered only studies that presented data on two or more of the following tissue types: (1) stem wood; (2) branch wood (as described by authors); (3) twigs (as described by authors); (4) bark; (5) coarse roots (>5 mm diam.); (6) fine roots (<5 mm diam.); (7) leaves. Some studies provide analyses of additional tree tissues and wood anatomical features, but insufficient data were reported to support a cross-species analysis.

All statistical analyses were conducted using R v. 2.10.1 (R Foundation for Statistical Computing, Vienna, Austria). Our first analysis step was to evaluate variation in species wood C content as a function of biome and species type. Our literature review found uneven sample sizes (in terms of species) and sampling effort (in terms of number of references) across different biome/type groupings (Table 1). Therefore we tested the effect of biome and type on wood C content using analysis of variance with F-tests based on Type III sum of squares, in the “car” R package [41]. Our next analysis step was to directly compare average wood C values among the biome/type groupings. Again, because of unbalanced sample sizes these comparisons were done by calculating a least squares mean (or the “population marginal mean” sensu Searle et al. [42]) for total wood C for each of the six biome/type groups. Least squares means were calculated by performing a generalized least squares regression in the “nlme” R package [43], and using the regression parameters to calculate predicted values for each biome/type groupings.

We then compared our estimated wood C mean values for each biome/type species groups to corresponding values suggested by IPCC [30] forest C accounting protocols. Agreement between wood C values derived from the literature and those from the IPCC [30] were statistically evaluated by testing the linear hypothesis that IPCC values perfectly predict mean wood C values from the literature (where the y intercept = 0 and the slope (or regression coefficient for IPCC values) = 1). This test was performed in the “car” R package, and was performed on a total of seven data points: six biome/type group values, and overall mean wood C value (corresponding to the “default” value of the IPCC).

We also found uneven sample sizes for records of C_vol, therefore we conducted the same least squares means calculations and analysis of variance tests on C_vol data. In a final analysis step, we also tested if C_vol (when grouped by biome/type) made a significant contribution to total wood C content. This was done using a one-sided t test that compared mean C_vol values against a population with a mean of 0.

As in the case of stem wood, we compiled data from studies presenting values for C content of other plant tissues, in all cases using data presented in tables or text. Where multiple tissue-specific values were presented within a given study, we computed an average across all values presented in table form. Where multiple studies presented values for the same species (and tissue type) we computed a simple unweighted mean across studies (using the subset of studies presenting C content values across multiple tissue types). For each tissue type with sufficient sample size (N > 9) we tested significance of the relationship between tissue-specific C content and the corresponding species mean stem wood C content, using standard major axis regression. We also tested for non-linearity of these relationships on the basis of a second-order polynomial term in a least squares regression in which stem wood was considered the independent variable. To test for deviations of the relationship from a linear 1:1 pattern, we computed the standardized major axis regression to test for deviations of the intercept from zero and slope from 1. Differences in tissue-stem C relationships between conifers and angiosperms were also evaluated using analysis of covariance in which C content of a given tissue type was the dependent variable, and stem wood C content and tree type were independent variables; an interaction term was also included in the model to test for heterogeneity of slopes.

We found a total of 31 published studies reporting species-specific C content data for stem wood, representing 253 species-specific C records (Table 1, see also [44]). Of these, only seven studies provided data for ≥10 species, with the largest dataset including 59 species-specific records (Table 1). Overall subtropical/Mediterranean regions are the most poorly represented, with only 28 total species-specific records available, owing to five peer-reviewed studies (Table 1 and Table 2, Figure 1). Tropical angiosperm species were the most-well represented group with 134 species-specific records found from seven published studies (Table 2), while tropical conifers had the poorest representation with only one species-specific record for Pinus caribaea available (Table 2). In total, angiosperms were better represented compared to conifer species, with 206 vs. 47 species-specific records available, respectively (Table 2).

In all biomes wood C content varied across species by ~10–15%, ranging from 41.9–51.6% in tropical species, 45.7–60.7% in subtropical/Mediterranean species, and 43.4–55.6% in temperate/boreal species (data not shown, but see [44]). Mean wood C content varied significantly as a function of both biome (F_{2, 249} = 12.02, p < 0.0001) and species type (F_{1, 249} = 29.21, p ≤ 0.0001; Table 3). Tropical species showed significantly lower wood C content than subtropical/Mediterranean and temperate/boreal species, while subtropical/Mediterranean and temperate/boreal species tested were not statistically different in terms of mean wood C content (Table 4). Our analysis showed that across the whole dataset, conifer species exhibited mean wood C content (50.8 ± 0.7%, 95% C.I.) that was ~3% greater on average than angiosperm species (47.7 ± 0.3%; Table 4), a trend consistent in all biomes when considered independently (Figure 1). Subtropical/Mediterranean and temperate/boreal conifers showed the highest mean wood C content at 50.54 ± 2.8% and 50.8 ± 0.6% respectively (Figure 1). Tropical angiosperms maintained the lowest wood C fractions with an average of 47.1 ± 0.4% (Table 2).

Taken together, the IPCC suggested values [30] did not significantly match wood C values found in the literature (linear hypothesis test F_0,7 = 0.499, p = 0.634; Figure 2); the degree of discrepancy between suggested values differed substantially among biome/type classifications (Figure 2). Specifically, IPCC [30] C fractions were within 1% of observed wood C values, only slightly underestimating wood C content in tropical conifers (0.3% underestimate), subtropical/Mediterranean angiosperms (0.9% underestimate), and temperate/boreal angiosperms (0.8% underestimate). IPCC values only provided a slight overestimate (0.2%) for wood C for temperate/boreal conifers (Table 2, Figure 2). Differences between IPCC [30] wood C fractions and observed C fractions were much larger for tropical angiosperms where IPCC values represent a 1.86% overestimate, and in subtropical/Mediterranean conifers where IPCC values represent a 1.54% underestimate (Table 2, Figure 2). The global wood C value of 47% suggested by the IPCC [30] represents a 1.2% underestimate of the global average wood C value (48.2 ± 0.3%) calculated from available literature (Table 2, Figure 2).

We found only 70 species-specific records for C_vol, owing to three studies [1,2,3]. Of these, our previous assessment of C_vol in 59 tropical trees [2] is the only to provide data for more than seven species (Figure 3); Chinese and North American temperate/boreal species comprised the remaining 11 species-specific records for C_vol [1,3]. We found C_vol varied significantly as a function of biome (F_1,67 = 6.74, p = 0.012; Table 3), but did not differ significantly between angiosperm and conifer species (F_1,67 = 0.97, p = 0.328; Table 3). Results from one-sided t tests showed contributions of C_vol were significantly greater than 0 in all groups tested (tropical angiosperms, t₅₈ = 14.85, p < 0.0001; temperate/boreal angiosperms, t₇ = 4.45, p = 0.002; temperate/boreal conifers, t₂ = 2.94, p = 0.05). When grouped by biome and type, tropical species exhibited the highest mean C_vol (2.5 ± 0.3%), followed by temperate/boreal conifers (2.1 ± 1.4%), and temperate/boreal angiosperms (1.3 ± 0.6%; Table 2, Figure 3).

In analyses of tissue-specific C content values, all tissues with the exception of leaves showed statistically significant linear relationships with stem wood C content (Figure 4); in no case was a significant non-linear trend detected. Considering pooled data sets (angiosperms and conifers), all of the standardized major regressions did not differ significantly from a 1:1 relationship (showing a y intercept not significantly different from 0 and a slope not significantly different from 1). A number of relationships are remarkably tight: 80% of the variation in coarse root C content, and 75% of the variation in branch C content is explained by stem C content. In spite of this, there is some evidence for systematic differences in these relationships between angiosperms and conifers. There was significant heterogeneity of slopes detected for branch and leaf C (p = 0.05 and p = 0.03, respectively); and a difference in intercept for twig C (p = 0.028), coarse root C (p = 0.03), and fine root C (p = 0.003). The cases of heterogeneity of slopes are driven entirely by a single outlying species (Pinus massoniana), for which very high stem C is reported [63]. In all cases showing differences in intercept value, the pattern is consistently of higher C content values for a given tissue in conifers than in angiosperms, given the observed stem C value.

Analyses of tissue-specific wood C values indicate that stem wood C provides a surprisingly good direct approximation for C content in other tissues (Figure 4), particularly those tissues that represent the most important biomass fractions in large trees after stem wood (namely branches and coarse roots). This result suggests that in spite of tissue-specific functional demands, there are important constraints on genetic determination of the key chemical traits of woody tissues (such as lignin-cellulose ratio and non-structural carbohydrate content) that determine C content. Although the pooled relationships between stem wood C and other tissues did not differ from a 1:1 linear function, we did find significant differences between conifers and angiosperms in several cases. In each instance, conifers show higher than expected tissue-specific C content than do angiosperms, at a given value of stem wood C content (Figure 4). Likely explanations for this pattern are a higher relative degree of lignification of tissues in conifers [64], differences in conifer vs. angiosperm lignin chemistry [65], and/or higher relative concentrations of non-structural carbohydrates in stem wood of angiosperms [66].

One important consideration not addressed quantitatively in this review is differences in wood C content within stems, among different wood types: in particular differences between sapwood and heartwood, between juvenile wood and mature wood, and changes due to reaction wood formation. Heartwood is expected to show high C content due to chemical transformations occurring during formation, including deposition of phenolic compounds and reductions in starch and other non-structural carbon compounds [38]. Observations that heartwood can have an appreciably higher C content than sapwood date back to the late 19th century [67]; however, we were able to locate only six peer-reviewed studies on seven tree species that present relevant data, and differences/trends in wood C between the tissue types are not consistent. For example, in Pinus pinaster, Bert and Danjon [38] found heartwood to contain ~1–4% higher C content than sapwood; a trend that depended strongly on height of sampling. Similar differences in the 1–7% range have found in both conifers and angiosperms, including Pinus sylvestris (52.3% vs. 45.3% C for heartwood and sapwood, respectively [50]), Pinus pinaster (49.5% vs. 45.8% [50]), and Nothofagus antarctica (53.0% vs. 51.5% [58]). However, other studies have reported only small differences (<1%): Sequoiadendron giganteum (55.2% vs. 54.7% [1]), Quercus petraea (46.0 vs. 45.5% [46]), and Quercus pyrenaica (45.8 vs. 45.6% [46]). Thus, while heartwood may have appreciably higher C content then sapwood, this appears to be strongly taxon-specific, with by far the largest differences reported for pines, and the lowest differences generally reported for angiosperms [see also 37]. Moreover, data from Lamlom and Savidge [1] suggests the trend of increased wood C in heartwood vs. sapwood is a generalization of a much more complex pattern: in Sequoiadendron giganteum, the zone of transition wood (mean wood C 52.5%) between heartwood (mean wood C 55.2%) and sapwood (mean wood C 54.7%) was not intermediate in terms of wood C content. Instead, in Sequoiadendron giganteum changes in wood C from pith to bark were non-monotonic [1]. Since studies quantifying chemical changes from pith to bark tend to overlook this transition zone [68], we suggest a careful evaluation of the expected pattern that wood C decreases linearly from pith (i.e., heartwood) to bark (i.e., sapwood) is needed. Extractives deposited in heartwood may commonly include volatile C constituents, and thus analyses that include the volatile C fraction are particularly important for this question. Along these lines, Lamlom and Savidge [1] report sapwood C values slightly higher than heartwood values for conventionally dried samples of Thuja occidentalis, but the reverse pattern for desiccated wood samples prepared to retain volatile C.

Another important confounding factor we do not address is differences between juvenile wood and mature wood. Predictable changes in wood properties are often assumed to occur between juvenile and mature wood, with much of the evidence for these patterns coming from commercially viable North American conifer species [reviewed by 68]. With respect to wood C, juvenile wood generally has a higher lignin: cellulose ratio, and is thus expected to have a higher C content [34,38,68]. However, here we also caution against generalizations, as expected chemical differences between juvenile and mature wood have tended to be determined based on categorical comparisons that overlook the relatively large transition zone between pith and bark [68]. Lastly, an additional source of within-stem variation in wood chemistry that has received little attention is the effect of reaction wood (compression wood in conifers and tension wood in angiosperms). Development of reaction wood is expected to have large effects on wood chemistry [69], but we are unaware of data on C content in relation to reaction wood formation.

Our review has some important implications to high-level forest C accounting protocols as supported by the IPCC (“Tier 2” or “Tier 3” carbon accounting [30]). In documenting that wood C ranges by ≥10% across species of the same provenance, and finding that wood C varies across biomes and species types, our data support IPCC protocols suggesting that region- or species-specific wood C fractions should be used when converting AGB to C (Figure 1; [44]). Recent studies have shown it is technically tractable to incorporate species-specific C fractions into forest C accounting when detailed inventories are available [2,70]. However, it would clearly be challenging to incorporate species-specific C fractions into large-scale forest C accounting, especially in diverse tropical forests (such as western Amazonian or Bornean rainforests) that may contain ~250 tree species ha⁻¹ in the ≥10 cm dbh size class alone [71,72]. Incorporation of species-specific wood C data may be also particularly challenging, given that recent progress in estimating forest AGB relies heavily on remotely sensed data [73,74]. Such technologies may be able to discriminate species in temperate forests [75], but are likely to perform only limited species- or tree functional type discrimination in diverse tropical forests [76].

Despite these limitations, biome/type-specific wood C fractions found in this review represent the most current data that can be incorporated into forest C accounting. Currently, the largest discrepancies between IPCC values and existing data are primarily attributable to the IPCC conflating tropical and subtropical/Mediterranean species (Figure 2). In doing so, IPCC protocols may result in an overestimate of ~3.7% in wood C of tropical forests that have little or no conifer component. This finding is consistent with a previous study where we found IPCC wood C fractions overestimated Panamanian forest C stocks by 3.3% [2]. Based on our review here, we also suggest the IPCC wood C fractions may currently underestimate C stocks in subtropical/Mediterranean forests by 3.7% (Figure 2). However, examining the magnitude of this error with forest inventory data (similar to [2]) would be needed to confirm this. Our review indicates that the general “default value” of 47% C suggested by the IPCC is low, since existing data suggest a value of 48.2 ± 0.3% (95% C.I.). However, given the large consistent differences between conifers and angiosperms, and availability of wood C data for both groups in all three provenances defined here (Table 2), there appears no reason to continue use of generalized global wood C values.

Consistent with previous studies [1,2,3], our analysis indicates that C_vol is a non-negligible component of total wood C content. Overlooking C_vol in analysis of woody tissues underestimates total wood C content by 1.3–2.5% (Figure 3): in tropical forests, this error leads to underestimates of forest C on the order of 6.6 Mg C ha⁻¹ [2]. By convention, tree biomass is expressed in terms of dry mass, the standard protocol involving drying tissues to constant mass at 65 °C [77,78,79,80]. Biomass stocks are then converted to C stocks by multiplying the oven-dried biomass (i.e., biomass without the volatile constituents) by species- or forest-type-specific carbon fractions. Accurate C fractions determined from wood samples should represent both the volatile- and non-volatile C mass as a percentage of oven-dried biomass. Therefore, to correct for the effects of differences in drying treatments (and to account for the contribution of C_vol) it is important to account for the mass loss associated with loss of volatile constituents during the drying process.

C determinations based on elemental analysis are calculated as:

    (1)

where C_tot is the percentage mass of C in a wood sample (mass/mass basis); M_C is the mass of C in the sample (dried at a given temperature; most commonly 60–70 °C but see [44]), and M_S is the total sample mass (dried at the same temperature as M_C).

Therefore to make these values applicable to AGB estimates based oven-dried samples, we have proposed a “carbon conversion factor” (C_conv) that corrects for mass and C loss during sample drying [2]. Specifically, we suggest for the purposes of estimating forest C stocks and fluxes, wood C content should be calculated as:

    (2)

where M_C is the mass of C in a sample following freeze-drying; M_F represents the mass of the freeze-dried sample, and VMF represents the volatile mass fraction, calculated as:

    (3)

where M_F is as in Equation (2); and M_H represents mass of the same sample used for M_F determination, after oven-drying to constant mass (at 65 °C).

This proposed C_conv calculation has been used in only two studies to date [2,34], both of which only analyze stem wood. We reiterate our suggestion that if volatile C is to be included in forest carbon assessments, this should be done on the basis of the C_conv parameter described above to correctly express total live C in tree tissues as a proportion of conventionally determined measures of biomass. In the absence of additional comparative studies, data from our tropical comparative study [2] may be used to derive a function to approximate C_conv on the basis of C content determined from conventionally dried samples (C_heat):

    (4)

(Least squares regression constrained to have y intercept = 0; N = 59; adjusted r² = 0.74; p < 0.001).

Although our review found more than 200 species-specific wood C records (Table 1, Figure 1), there is still clearly a need to accumulate more species-specific wood chemical information. There currently exists 8, 412 species-specific records of wood density, which are readily (and freely) available in large databases [81]. This database resource has proven critical for refining estimates of forest- and tree AGB, by providing wood density data that is readily incorporated into tree allometric models [78,82]. While we provide a starting point for creating a similar database that contains species-specific wood C data (see [44]), clearly more sampling effort is needed. For example, in the wood density database [81], the number of Mexican species alone (N = 221) closely rivals the entire wood C dataset here (N = 253; Table 2).

In addition, our biome classifications may mask important variation among forest regions and taxonomic groups that are especially poorly represented (if not completely unrepresented) in terms of species-specific wood C data. Common rainforest trees of southeast Asia, such as those in canopy-dominant Dipterocarpaceae family (that constitute ~40% of total forest basal area [72]), are represented by only five species in our dataset (Table 1). Southeast Asian tropical forests are likely the most carbon-dense pantropically [7,15], and most dynamic in terms of forest C stocks and fluxes [32]. Similarly, although Becker et al. [35] provide data for Kenyan and Ugandan trees (Table 1), our review found no species-specific wood C data for trees from the Congo basin, a forested region that is likely the most C-rich in continental Africa [7,15]. Geographical discrepancies in wood C data availability are even more pronounced for the subtropical/Mediterranean region: of the five references presenting data for 28 subtropical/Mediterranean species, only one study representing five species was outside of Mediterranean Europe [63]. In addition, our review did not find published species-specific C content values for many important taxonomic and functional groups of woody plants including lianas, hemi-epiphytes, palms, tree ferns, and non-conifer gymnosperms (such as cycads and Gnetum), some of which may be increasing with respect to their contribution to total forest biomass [83].

The overarching goal in this review was to evaluate existing literature in order to amass and analyze wood C content data across a global dataset. We also identify methodological considerations that should be taken into account when measuring wood C content. We suggest that information on wood C content from a wider range of species is needed to inform forest C accounting in a number of forest types. Our synthesis and data compilation provides a starting point for a database on tree C content. However, there is a pressing need for additional species-specific wood C data. Amassing such data represents an opportunity to refine our understanding of the role forests may play in C capture and storage at spatial scales ranging from individual trees to the biosphere.