Background & Summary

Wood density has long been recognised as an essential and variable property of wood1, resulting from wood tissue complexity and environmental variations2. Since wood density is an accurate predictor of the mechanical properties of wood material, its variability has received particular attention for industrial purposes1. Wood density, which measures the amount of dry matter per unit volume, is also one of the two key factors, along with wood volume, governing the amount of biomass contained in a tree and is thus a crucial parameter in forest ecosystem biomass and carbon assessment3,4. Wood density variation has hence been shown to be a major driver of the spatial distribution of forest biomass5,6 and must be taken into account when assessing it7.

Finally, as wood density integrates the proportion, dimension and variation of diverse specialised cells devoted to different functions, it has emerged as a major functional trait in ecophysiological research8. Wood density has been interpreted as an integrative trait reflecting the trade-offs between the different biological functions of wood tissue, including water transport efficiency and safety9,10,11,12,13, mechanical strength14 and storage of water and nutrients15,16. Being proportional to wood construction costs and mechanical strength, wood density has also been interpreted as reflecting the broad trade-off between growth and durability, with higher wood density providing greater mechanical stability on the one hand, but incurring greater construction costs and slower growth on the other17,18 (but see critical view of this theory19). Wood density is therefore a trait associated with life-history strategies encountered across tree species20.

Despite all these crucial implications and uses, the extent and drivers of wood density variations remain partially understood. One of the reasons for this lack of knowledge is that wood density is an inner characteristic of trees, which cannot be derived from external measurements. Wood density measurements are therefore costly and time-consuming, and their availability remains limited. This is why the main existing datasets were obtained by progressive data accumulation over time.

The Global Wood Density (GWD)8,21 and TRY22 databases provide wood density records for many species (about 8,000 for GWD and 11,000 for TYR, which integrates GWD data) distributed throughout the world. These records have however been obtained using heterogeneous and sometimes poorly described methodologies. Of equal concerns, they are based on small samples (from a single tree to several dozen) that cannot take into account the various sources of wood density variation (tree ontogeny, age, silviculture, climate, etc.) and therefore poorly represent forest tree populations. Moreover, such databases are not immune to missing values in a given study area, notably for rare species23,24. Because of this, such wood density datasets are invaluable for analysing between-species, or regional to global patterns of wood density variation, but do not allow for fine analysis of interspecific and intraspecific variations in a given area.

More recently, the European GenTree dataset has been constituted on several individual tree characteristics25. GenTree data include wood density measurements acquired using a homogenous methodology on wood cores from 3,098 trees from seven main European temperate tree species. Wood cores were collected across most of the geographical range of species distribution and from dominant and co-dominant trees only. As a result, GenTree data cover a wide range of ecological conditions, but only a small fraction of the diversity of European tree species and populations.

The XyloDensMap project26 aimed to establish a comprehensive set of wood density records within and among multiple European tree species with a consistent methodology. To achieve this goal, a collaboration assembling skills in forest inventory and wood metrology was elaborated that combined field sampling of wood cores by the French national forest inventory (NFI), a statistical survey based on a probabilistic sampling design, and a high-flow method of wood density measurement by X-ray Computed Tomography (CT) (Fig. 1).

Fig. 1
Fig. 1
Full size image

Overview of the generation of wood density data along the XyloDensMap project. The process starts with the collection of wood cores in the forest plots of the French National Forest Inventory, through the measurement of wood density along samples by CT scanning. (a) field forest plot, (b) manual sampling of a tree at breast height with a 4.3 mm diameter increment borer, and reading of the recent radial increment, (c) storage of a field plot’s cores in a polycarbonate honeycomb box, (d) mailing of core boxes to INRAE Grand-Est research centre (Champenoux site) equipped with CT measurement facilities, (e) core drying and assemblage of boxes by blocks and longitudinal superblocks for massive scanner CT measurement, (f) measurement, (g) processing and segmentation with the Carden software, (h) calculation of the mean wood density for each wood increment core.

The XyloDensMap collaboration has assembled a dataset of 110,763 individual tree wood density records from 156 species, across a broad range of ecological conditions in France, reflecting a large set of European climates. Data include the measurement of the mean anhydrous wood density (mass of dry wood divided by the dry volume) and an estimate of the mean basic wood density (mass of dry wood divided by the green volume) of radial wood cores extracted from tree stem at the conventional breast height. Additional information on stand’s geographical position, climatic conditions, tree age, height and diameter at breast height are provided. Further, wood shrinkage has been measured around one spatially systematic fraction (1/4th) of the collected cores, providing a dataset for further refined modelling of this shrinkage, in addition to reference values available in the literature. Here we present the XyloDensMap dataset and the method used to obtain it.

Methods

The French territory to sample a wide diversity of European tree species

The French territory encompasses a large set of climate types (oceanic, continental, mountain and Mediterranean climates), elevation gradients (Pyrénées and Alps) and soil conditions27, favouring a diversity of native tree species. After centuries of forest cover loss, the territory has been continuously afforested since the mid-19th century, with a 60% increase in forest area28, contributing to a net enrichment in exotic tree species over one century29. Accordingly, French forests show the greatest diversity of tree species in Europe: (i) 13 out of the 14 forest types of the European classification are encountered across the territory30, with exception of the boreal forest type, (ii) 12 tree species occupy > 500,000 hectares each as dominant species of the forest canopies31, (iii) half the area is occupied by tree species mixtures32, favouring the capture of several tree species at a single sampling location. The French territory therefore offered a unique opportunity for encompassing a large number of European tree species with a unified sampling and data collection process.

Sampling design and wood core collection

Wood cores were collected on field plots of the French national forest inventory (NFI), operated on the metropolitan territory. Since 2005, the French NFI has adopted the standard of a grid-based systematic annual sampling design, such that annual samples homogenously cover the whole territory and form a regular systematic spatial cover after five years. In forest inventory, this spatio-temporal design is known as the “interpenetrating panel design”33,34,35,36. This strategy has two immediate advantages for the purpose of covering the maximum number of tree species: (i) the systematic nature of each annual sample allows to cover the full diversity of forests across the territory in a single year, (ii) the spatial complementarity of successive annual samples further favours the capture of rare tree species.

On each field forest plot, collected wood cores were those routinely taken by the French NFI for both age and increment measurements. Cores were extracted at a conventional height of 1.30 m with increment borers (core diameter = 4.3 mm) in the radial direction. The associated protocols comprise the coring of two dominant trees per plot, plus one tree per diameter size class per species (five diameter classes starting from the census diameter of 7.5 cm: [7.5, 12.5], [12.5, 22.5], [22.5, 37.5], [37.5, 52.5] and [52.5, +∞]). Therefore, sampling of the maximum tree size gradient and tree species diversity is favoured in each plot. In total, 110,763 wood cores were collected on 20,697 NFI forest field plots and from 156 tree species (which account for 99.99% of the growing stock in mainland France, as estimated with the French NFI data from the 2016–2019 collection years and the NFI estimation process, which can be run online at https://ocre-gp.ign.fr/) during the 2016–2019 period. The 110,763 wood cores comprised 36,249 “pith cores” and 74,514 “peripheral cores”. The “pith cores” are taken from the bark to as close as possible to the pith in order to estimate tree age (by counting all tree rings from the pith) and the average annual growth of the last five tree rings. The “peripheral cores” are shorter radial cores (representing on average 63% of tree radius, against 100% for pith cores) routinely taken without concern for being as close as possible to the pith, as they are used to measure annual growth of the last five rings in the French NFI program. Nevertheless, in the context of this project, field crews were given the formal instruction to drill trees deeper than usually consented for these cores, hence their coverage of most of the tree radius.

As a major consequence of the design, tree species were sampled proportionally to the extent of the spatial distribution and prevalence of their tree populations. Since the abundance of tree species is J-distributed over the forest territory27, so is the number of wood cores collected by tree species, found in remarkable proportion to the growing stock of tree species (Fig. 2). The number of wood cores per species hence varied across four orders of magnitude, ranging from one for several tree species (e. g., Cedrus libani) to 12,778 for Quercus robur. It exceeded 1,000 cores for 23 tree species. This has implications for the capture of wood density distribution and population parameter estimations (see Usage notes below).

Fig. 2
Fig. 2
Full size image

Relationship between the number of wood cores collected by tree species and the importance of these tree species in the French forests as measured by their growing stock. Growing stock (stem volume of trees >7.5 cm census diameter from the basis of the stem up to 7 cm) was estimated with the national sampling and estimation process of the NFI, for the data collection years 2016–2019. The black line represents the OLS regression line of the logarithm of the number of collected increment cores against the logarithm of the growing stock.

Cores were taken over the entire area covered by the forest in metropolitan France (Fig. 3): from plains to the altitudinal limit of the forest (i.e., from 0 to 2,300 m above sea level), and from the Mediterranean zone in the South to the oceanic regions of the West and the North and the semi-continental regions of the East.

Fig. 3
Fig. 3
Full size image

Maps of the French NFI plots in which the wood cores of the XyloDensMap dataset were collected. Points are coloured according to the species (broadleaf species on top map, conifer species on bottom map) demonstrating the largest percent canopy cover in the corresponding NFI field plot.

As a consequence, data encompass a broad range of soil and climatic conditions. Regarding the climatic normals, annual mean temperature ranges between 3 °C and 17 °C across all plots, while annual precipitation ranges between 500 and 2,200 mm (Fig. 4). The range of covered climatic conditions logically strongly depends on species distribution. For widely distributed species such as Quercus petraea, Fagus sylvatica and Abies alba, wood cores were taken from forest plots covering most of the full range of climatic conditions observed in the XyloDensMap dataset. For species with a much more restricted distribution, such as Pinus Halepensis, found mainly in the Mediterranean region, samples covered narrower climatic conditions.

Fig. 4
Fig. 4
Full size image

Climatic conditions covered by the XyloDensMap dataset. Each point represents a NFI plot in which wood cores were taken for wood density measurement. Points are coloured according to the species with the highest relative canopy cover rate in the corresponding forest NFI plot. For each plot, mean annual temperature and annual precipitation were obtained from the French meteorological institute (Météo France) and correspond to 1991–2020 climate normals.

However, it is important to consider the spatial distribution of species in the XyloDensMap dataset in relation to their more global distribution. Indeed, the NFI-based sampling of tree species found across the French territory may reflect contrasted positions in the broader distributional range of these species (for instance, while Quercus ilex and Pinus pinaster are found at their northern margin in the Mediterranean area, continental and boreal tree species such as Pinus sylvestris or Picea abies are encountered at their southwestern limits). Although our dataset provides very detailed information on the wood density of the main European tree species, its limited spatial coverage within a specific region of Western Europe might not fully represent the species’ broader range. As a result, conclusions drawn from the dataset may not be applicable or accurate for the entire geographic range of these species. The Euforgen initiative (European Forest Genetic Resource Program) portal can be used to get a rapid and comprehensive overview of the continental distribution of the sampled tree species at https://www.euforgen.org/species.

Technical operations

As of field data collection, wood cores were placed in polycarbonate honeycomb boxes (12 increment cores per box, on average) and referenced by a unique identifier (box number and row position of the wood core in the box), used to link the core wood density measurements with the NFI field plots, and associated measurements on the corresponding tree and stand. Boxes were weekly sent by post to the wood density measurement location, and stored in cold conditions (between 4 and 6 °C). Cores were dried at 103 °C by placing the boxes in an oven during 48 hours. After drying, the boxes were stored at room temperature and sealed with an adhesive tape to prevent wood moisture increase until X-ray scanning.

Measurement of longitudinal profiles of core anhydrous density by X-ray Computed tomography

Boxes containing the cores at anhydrous state were assembled into packages (17 boxes per package) and passed through X-ray Computed Tomography (CT) using a medical X-ray scanner (GE BrightSpeed Excel) located at INRAE Champenoux, France (Fig. 1). Due to a breakdown during the project, another medical scanner (GE Revolution EVO) located at the hospital in Essey-lès-Nancy, France, was also used for part of the samples (13% of increment cores). For each core, one CT-scan image was produced every 0.625 mm in the longitudinal direction. Each image is composed of 512 × 512 pixels (pixel size of 0.32 mm) characterised by a grey level (Hounsfield unit, measuring radiodensity), converted into a wood density value using a published procedure37.

While high-speed CT scanning of large wood samples has gained substantial recognition38, the technology for rapid acquisition of wood density from large sets of unprepared, packaged tree cores is an entirely new approach, which required the original development of the CARDEN software as part of the XyloDensMap project39,40. For each scanned package, the software first isolates the different increment cores, and then calculates for each an average of the pixels for images taken every 0.625 mm along each core, resulting in a wood density profile. Full details on the methodological process and associated errors in wood density estimates are provided in the CARDEN-specific article39.

It should be noted that the longitudinal resolution of the wood density profiles (one measurement every 0.625 mm) is well suited for estimating the mean wood density of each increment core, but is too weak to identify ring limits, and properly assess wood density variations at ring or intra-ring resolutions41. Wood density profiles could be used to study radial and temporal variations in wood density (but not annual variations), but this would require in-depth data processing (including cross-dating of samples in the case of time series), which has not been done to date.

Estimate of longitudinal profiles of core basic density by accounting for volumetric shrinkage

Because the drying led to a “volumetric shrinkage” (i.e. a loss of volume) of wood cores, wood density as measured by X-ray CT is an anhydrous density (dry mass of wood divided by the dry volume). The total volumetric shrinkage of wood from green (or saturated) state to anhydrous state is usually calculated as follows:

$${\rm{WS}}=100\times \frac{{V}_{S}-{V}_{0}}{{V}_{S}}$$
(1)

with WS being the total volumetric shrinkage of wood (in %) from the saturated state to the anhydrous state, VS the volume of wood in the saturated state and V0 the volume of wood in the anhydrous state.

From the CT measurements, the profiles of basic wood density could finally be obtained using the following equation:

$${{WD}}_{b}={{WD}}_{0}\times \left(1-\frac{{WS}}{100}\right)$$
(2)

with WDb (kg m−3) the basic density of wood, WD0 (kg m−3) the density of dry wood as measured by X-ray CT, and WS the wood volumetric shrinkage (%).

Two approaches were considered for estimating volumetric shrinkage. First, we assembled a comprehensive set of shrinkage values from the available literature (Table S1). Second, we implemented an experimental approach where volumetric shrinkage was directly estimated on XyloDensMap wood cores from CT-scan images. For that, a subset of 23,828 wood cores from 120 species were scanned twice, in the initially saturated state and in the anhydrous state. The volume of wood cores in both states was then calculated from the total number of pixels for the whole core, the fixed size of each pixel (0.32 mm) and the longitudinal scanner pitch (0.625 mm).

In view of the experimental character of the approach, the default basic wood density released in the present dataset was calculated from the shrinkage values of the literature. The shrinkage values estimated from the CT-scan images are also provided in a separated table (see Data records). Of note, the volumetric shrinkage values gathered from the literature tend to be smaller than those estimated from the CT-scan images, resulting in larger basic wood density values than if these CT-scan estimates had been used, especially for broadleaf species (see Fig. S1 for a visualisation of the impact of the choice of the volumetric shrinkage reference used, and Table S2 for a comparison of shrinkage values by species). This companion dataset obtained at an individual tree level offers the potential of modelling their dependence from several core properties (tree species, anhydrous density, core type, …), as reported in the literature42.

Estimate of mean anhydrous/basic density of increment cores

While the arithmetic mean of the successive 0.625 mm measurements may be first considered as an obvious representative estimate of the mean stem wood density, it does not take into account the relative importance, or weight, of each segment of the core in the tree section. Owing to the radial size expansion of trees along their lifespan, parts located on the core periphery (near the bark) represent a greater surface area and volume in the stem than the internal parts (towards the pith).

Were wood density remained constant throughout the life of the trees, this would not pose any particular problem. However, wood density is known to vary along the radial direction of a tree stem as resulting from physiological processes along tree life-history43. For conifers, wood density tends to increase from the pith to the periphery, while for deciduous trees the variations are more complex: wood density can decrease, remain constant or increase depending on the species44. This radial effect has been repeatedly demonstrated for several tree species of the forest territory under scrutiny, for example on Norway spruce45, on sessile oak46, or on common beech47.

In order to account for this size effect, a weighted mean wood density (for both anhydrous and basic wood densities) of tree core was therefore computed by weighting each CT density measurement along the longitudinal core profile by the assumed circular area associated to the radial position in the stem, as expressed from the final tree diameter measured by the field inventory48. The weighted mean anhydrous and basic wood densities were computed as follows:

$${WD}_{w}=\frac{{\sum }_{i=1}^{n}(((\Pi \times {(r-0.625\times (i-1))}^{2})-(\pi \times (r-0.625\times i{)}^{2}))\times {WD}_{i})}{{\sum }_{i=1}^{n}(\Pi \times {(r-0.625\times (i-1))}^{2})-(\pi \times {(r-0.625\times i)}^{2})}$$
(3)

with WDw (kg m−3) the weighted mean anhydrous or basic density of wood, n the total number of CT-scan images along the increment core from the bark side, i the ith CT-scan image along the increment core, r the stem radius, WDi (kg m−3) the mean anhydrous or basic density of wood for the ith CT-scan image.

As increment core length is rarely equal to the tree radius, the last wood density measurement (on the bark side) was set at the radial position corresponding to the tree radius. If the core length exceeded the tree radius, measurements beyond the pith were not considered for the calculation of the weighted mean. For peripheral cores, the innermost part of the tree outside the range defined by the core length was not considered, knowing that this internal part represents a small proportion of the whole tree volume.

In any of these configurations, the weighted mean wood density is therefore computed according to the fraction of the tree radius it represents. To allow users a possible stratification of data according to the relative length of wood cores, core length is also provided in the dataset (see Data records). The core length is calculated as follow:

$${Core}{\rm{\_}}{Length}=\frac{{n}_{{CT}{\rm{\_}}{images}}\times 0.625}{1-\frac{{LS}}{100}}$$
(4)

with Core_Length (mm) the estimated length of wood core in the green state, nCT_images the number of CT-scan images along the core, 0.625 (mm) the scanner pitch, and LS the mean wood core length shrinkage (%) for the corresponding botanical class (broadleaf or conifer), as estimated from the measurements of core length in both states (green and dry) on a subset of increment cores (same subset as for volumetric shrinkage measurements).

Overall, the weighting had little influence on the estimated wood density values, as shown in Fig. S2, with for example a mean absolute difference between the mean basic wood density and the weighted mean wood density of 9.3 kg m−3 (1.6%).

As increment cores can contain residual bark tissue, we checked that this did not introduce any bias into the estimation of the mean wood density. First, the presence of bark was assessed by visual observation on one third of the samples (~38,000 increment cores). This evaluation showed that around 45% of the increment cores had bark. Second, for the 38 species with more than 100 observations on the presence/absence of bark, we compared the mean wood density according to the presence or absence of bark in the wood core. This comparison revealed that the presence of bark had very little influence on the estimation of the mean wood density of wood core (Fig. S3). On average, at species level, we observed an absolute difference of 12 kg m−3 (2%) in the mean anhydrous wood density between the two samples (with or without bark). In fact, even if bark could be included in the measurements, this had virtually no effect on the calculated mean wood density, due to the high number of measurements carried out on each increment core (143 measurements per core on average: 94 for peripheral cores and 244 for pith cores).

Data Records

The XyloDensMap dataset is openly available at Figshare repository49. It is composed of three comma-separated files, one R code file and one metadata file. A mirror version of the dataset is also available on French Recherche Data Gouv repository50.

The first file (xdm_individual_density_data.csv) provides tree- or stand-level information for each of the 110,763 sampled trees, with the identifier of the NFI sampling plot, the identifier of the tree on the NFI plot, the geographical coordinates of the centre of the 1 km × 1 km square cell in which the NFI plot was inventoried, the botanical class and the tree species, the type and estimated length of wood core, the estimated tree age (only available for “pith cores”), the stem diameter at breast height, the tree height, the mean annual temperature (normal of the 1991–2020 period), the total annual precipitation (normal of the 1991–2020 period), the mean anhydrous wood density (WD0), the weighted mean anhydrous wood density (WD0w), the mean basic wood density (WDb), the weighted mean basic wood density (WDbw), the wood volumetric shrinkage (WS) used to convert anhydrous density into basic wood density, and the name of the scanner model used. Missing values are denoted by NA. These data can be supplemented by additional tree or stand characteristics, available at https://inventaire-forestier.ign.fr/dataIFN/.

The second file (xdm_species_density_data.csv) provides statistics at species level for the weighted mean basic wood density (WDbw). The mean is provided for the 156 species; the variance, the standard deviation, the coefficient of variation, the 95% bilateral confidence interval for the species mean and standard deviation, the precision around mean and standard deviation, along with the p-values of the Shapiro-Wilk and Jarque-Bera normality tests, are provided for the 105 species with at least 10 wood density measurements (see Usage notes).

The third file (xdm_individual_shrinkage_data.csv) provides the wood volumetric shrinkage values measured on a subset of 23,828 wood cores. These values can be used as an alternative to the default literature shrinkage values used to estimate the basic wood density from anhydrous density, either by estimating volumetric shrinkage means by tree species (see example in the fourth file below), or by a more sophisticated modelling approach taking into account individual tree or core attributes.

The fourth file (xdm_R_import_and_manipulate_data.R) is an R code designed to guide the user through data import and manipulation, with examples of data operations and visualizations, a theoretical example of biomass calculation using the species wood density means provided in the second file, and an example of recalculation of the basic wood density using the shrinkage values contained in the third file instead of the shrinkage values of the literature.

The fifth file (xdm_metadata.txt) is a metadata file containing all definitions and units for each table and each variable.

Technical Validation

Comparison of CT measurements with gravimetric estimates

During the project, X-ray scanning of increment cores was carried out in batches, with around 1,500 cores being scanned within one or two hours. To ensure the accuracy of the measurements, the medical scanner was calibrated before each tomography session using the same calibration set consisting of 120 dry wood samples machined to be perfectly cylindrical. The volume of these dry wood cylinders was measured with a caliper (accuracy of 10−2 mm), and their weight was measured on an electronic balance (accuracy of 10−3 g), thus providing a reference wood density for calibrating the wood density values estimated from CT.

To cover a wide range of wood types and densities, the dry wood cylinders came from 20 species (Acer pseudoplatanus, Diospyros ebenum, Fagus sylvatica, Ginkgo biloba, Heritiera utilis, Juglans regia, Juniperus communis, Liriodendron tulipifera, Olea europaea, Populus tremula, Picea abies, Pinus sylvestris, Pinus taeda, Pseudotsuga menziesii, Quercus ilex, Quercus petraea, Sequoia sempervirens, Sorbus torminalis, Taxus baccata, Ulmus minor). The wood density range covered by these calibration wood cylinders extends from 200 up to 1100 kg m−3.

We fitted a calibration equation between the values estimated from CT and the reference densities established from weight and volume measurements. Figure 5a shows the result of a calibration carried out between the two anhydrous densities. In general, good agreement was observed between the CT measurements and the values estimated from weight and volume measurements. For calibration shown in Fig. 5a, the mean absolute difference was 36.6 kg m−3 (6%), with an average negative bias of −29.6 kg m−3 of CT measurement as compared to the gravimetric one.

Fig. 5
Fig. 5
Full size image

Cross-technique and cross-scanner validation of anhydrous wood density measurements. (a) relationship between CT and gravimetric density values for a subset of 120 perfectly cylindrical wood cores from 17 tree species, (b) relationship between CT density values measured on the same cores using two different scanners (same sample). Full black line: OLS regression line between two measurement sources, dashed purple line: reference Y = X.

Comparison of CT measurements using two different scanners

The same subsample of 120 reference dry wood cylinders was also used to compare the two scanners used for XyloDensMap measurements, namely the INRAE scanner (GE BrightSpeed Excel) and the hospital scanner (GE Revolution EVO). To do this, the 120 wood cylinders were scanned by each scanner model and the wood density values thus obtained were compared (Fig. 5b). This comparison showed that the wood density values obtained using the two scanners were highly coherent, with a mean absolute difference of 15.7 kg m−3 (3%) between the two measurements, and a residual standard error of 17.3 kg m−3.

Usage Notes

The XyloDensMap dataset contains four different types of wood density data, which can be used for a variety of purposes. The four wood density variables, together with their potential uses and calculation methods, are shown in Fig. 6.

Fig. 6
Fig. 6
Full size image

Presentation of the various data provided in the XyloDensMap dataset, with their potential uses. Based on CT-scan measurements of a wood density value every 0.625 mm along the increment cores, different calculation steps were used to calculate the four wood density data provided in the XyloDensMap dataset. In addition, a subset of increment cores was used to estimate volumetric wood shrinkage values, also provided in the XyloDensMap dataset. The different data can be used for a variety of purposes, including comparison with other datasets or biomass estimation.

In what follows, we develop an example of comparison with another dataset and an evaluation of the statistical moments of wood density, bearing in mind that we also provide with the dataset an R code to assist users, including an example of calculating biomass from volume using the mean wood density by species, among other examples of data manipulation and visualization (see Data records).

Comparison with the European GenTree dataset

A dataset (the GenTree dataset) containing wood density measurements for 3,098 trees of seven species across Europe has recently been published25. The GenTree dataset is based on fewer trees and species than the XyloDensMap dataset, but assesses wood density on a broader spatial scale. In GenTree, basic wood density was measured according to gravimetric methodology, i.e. through wood core volume and weight measurements.

A comparison of the XyloDensMap and GenTree mean basic wood densities (WDb variable in the XyloDensMap dataset) revealed consistent patterns at interspecific and intraspecific levels for the seven species common to the two datasets (Fig. 7). However, differences could be observed, the direction and extent of which depend on the species. Difference in species means between the two datasets was only 1% for Fagus sylvatica, 2% for Quercus petraea, 4% for Betula pendula and Pinus sylvestris, 5% for Populus nigra, 11% for Picea abies and 12% for Pinus pinaster. These differences could be explained by numerous factors, including geographical location of the sampling sites, characteristics of the sampled trees, sample size, volumetric shrinkage hypotheses, etc.

Fig. 7
Fig. 7
Full size image

Comparison of basic wood density distributions between the European GenTree dataset25 and the French XyloDensMap dataset (this paper) for the seven species common to the two datasets. For XyloDensMap, the arithmetic mean basic wood density of the cores (WDb variable in the dataset) is used.

For example, for Pinus pinaster, basic wood density values tend to be larger in GenTree than in XyloDensMap (species mean of 503 kg m−3 and 444 kg m−3, respectively). The GenTree dataset contains data for older (55 years in GenTree vs 32 years for trees with age measurement in XyloDensMap) yet only slightly bigger trees (diameter at breast height of 40 cm in GenTree vs 35 cm in XyloDensMap), implying narrower tree rings, and a larger wood density in view of the strong positive correlation between ring age and wood density as well as the strong negative relationship between tree ring width and wood density for this species (Fig. S4); this is consistent with the greater wood density observed for Pinus pinaster in GenTree than in XyloDensMap.

By contrast, for Picea abies, values tend to be lower in GenTree than in XyloDensMap (species mean of 344 kg m−3 and 388 kg m−3, respectively) and here the explanation cannot be found in the characteristics of the sampled trees, as GenTree data come from trees with narrower tree rings, which should lead to larger wood densities in view of the negative relationship between tree ring width and wood density also observed for this species (Fig. S4). However, GenTree’s sampling sites are at much higher latitudes or altitudes, and wood density is known to correlate positively with summer temperatures51,52, which supports a smaller wood density for Picea abies in GenTree.

An in-depth statistical analysis of the differences between the two datasets, taking into account tree sample attributes (size, dimensions and age, as well as genetic information if available) and methodological differences, has not been carried out as part of this data paper, but remains available to interested users given the open status of both datasets.

Assessing the mean and variability of wood density by species, along with their precision

Estimating forest biomass has become a prominent challenge in the context of climate change and the monitoring of policies aimed at mitigating the greenhouse effect or developing the bioeconomy. In most cases, forest biomass is calculated by multiplying the green volume of trees by the mean wood density at species level. Knowing the precision of the mean values by species used is therefore an important factor to consider when assessing the uncertainties associated with biomass and carbon estimates. Here, we provide information on the precision (using the 95% bilateral confidence intervals) around the mean wood density calculated at species level from XyloDensMap data (mean of the weighted mean basic wood density of increment cores, WDbw variable in the dataset).

Due to the uneven distribution of XyloDensMap density records across tree species that extend over four orders of magnitude, the precision of the mean varies largely. For this reason, means of basic wood density (WDbw) and their 95% bilateral confidence intervals were computed for the 105 tree species with N ≥ 10 (negative association between the interval range and log N cores, R2 = 0.63, p < 10−10; see the second provided file).

Second, this dataset gives an unprecedented view on the variability of wood density within tree species (intraspecific variation accounting for 27.3% of total variation as measured by an analysis of variance), that may also be of major relevance for uncertainty analyses, e.g. in the field of carbon accounting. Standard deviations of basic wood density and their 95% bilateral confidence intervals were also computed (same negative association between the interval range and log N cores, R2 = 0.61, p < 10−10; see the second provided file). In addition, the relationship between the number of tree cores and the standard deviation of wood density at species level showed a strong cross-species increase in this variability below ca. 100 cores, suggesting a lack of convergence for this parameter below this sampling depth (Fig. S5).

A breakdown of tree species with respect to these precisions (i.e., the 95% interval ranges) are provided in Table 1. It shows that roughly one third of tree species are secured with a remarkable precision of 20 kg m−3 for each of these two moments (mean and standard deviation). Further, more than half the tree species are also covered with a precision of 50 kg m−3.

Table 1 Number of species in each precision class for mean and standard deviation of weighted mean basic wood density across tree species of the XyloDensMap record.
Full size table

Assessing the distribution by species

Along with the mean and variability by species, the distribution of wood density within tree species populations is also an important characteristic to consider, for example for simulation or ecological purposes53, or simply when applying parametrical tests to compare species. Here, we investigated whether wood density followed a Gaussian distribution. In view of the strongly asymmetric structure of the XyloDensMap wood density sample, results must be considered with caution, as small samples cause tests unable to reject the null hypothesis, and very large samples are at the opposite risk of rejecting it, even for small deviations from the assumption. The Shapiro-Wilk test, which targets the distribution variance, was implemented due to its statistical power54,55 using the shapiro.test function of the stats package in R software56. It was complemented by the Jarque-Bera test, whose statistics targets the distribution skewness and kurtosis, and has high-power for long-tailed distributions55, and which was performed using the jarque.test function of the moments package in R software57. We discourage the use of the Kolmogorov-Smirnov test in the present context, as theoretical values for the distribution moments are not known, and in view of the frequent presence of tie values in the dataset. Tests were not implemented for tree species with N < 10 (51 species out of 156) in view of their restricted ability to reject the null-hypothesis. Also, the Shapiro-Wilk test could not be computed for N > 5000 in R (the eight most abundant tree species), leading to compute an average p-value over 100 random samples without replacement of size N = 5000.

Outcomes of these tests were strongly comparable. At a level of p = 0.001 (to avoid over rejection with large samples), the normality assumption hypothesis was rejected by one of the two tests for 48 species out of the 105 tested species (N ≥ 10), with concordance of the two tests for 44 species; for the 4 remaining species, rejection was based on the Jarque-Bera test alone for three species and by the Shapiro-Wilk test alone for one species. In more detail, the normality assumption was rejected (i) for seven tree species out of 48 showing 10 ≤ N < 100 (rejection in Prunus domestica only from Jarque-Bera test), (ii) for 19 tree species out of 34 showing 100 ≤ N < 1000 (Pinus nigra only from Shapiro-Wilk test and Arbutus unedo only from Jarque-Bera test), (iii) for 22 out of the 23 tree species showing 1000 ≤ N (Quercus petraea only from Jarque-Bera test), illustrating the size effect. In total, the normality assumption could therefore be accepted for 57 out of 105 tree species tested. These preliminary results open the door to further analyses to better understand the distribution of wood density within tree species populations and its ecological significance.