optimizing scanning techniques for biomass estimation, tree growth, and anatomical analysis at different scales

High-Throughput Tree Ring Analysis and Density Estimation

Wood density is an easy-to-measure variable1 that reflects both the anatomical and chemical properties of the wood2. In biomass estimations of aboveground biomass, wood density is an important weighing variable 3,4,5, that is multiplied with the dimensions of the tree and a factor representing the carbon content of the wood. Wood density is tightly linked to the mechanical properties of timber6 and reflects the life history of a tree7.
Cell wall density is measured as being approximately 1500 kg/m&#179; and is considered fairly constant8, however intra-ring cell wall density variations should be considered as well8,9.Woody cells (in general tracheids in conifers, vessels, parenchyma and fibers in hardwoods) are oriented/shaped in different ways and cell wall thickness and lumen size of these cells varies10. Therefore, wood density varies between trees, within a tree (axial and transversal) and within short intervals within a tree ring11,12. In many cases the wood density variation at the ring scale also delimits the tree ring boundary13. Wood density and ultimately tissue fractions are generated and in this paper are broadly put into three categories (i.e., three different resolution scales), depending on the study goal (Figure 1) as described below.
Inter-ring scale: By measuring pieces of wood, a single value is obtained for that sample. This can be done through water immersion or geometrically14. This way, general biomass or wood technological variables can be obtained. To include pith-to-bark variation, these pieces of wood can be further divided into blocks that are measured manually to obtain information on the life history strategy15. When switching to low-resolution X-ray CT such as in medical scanners17,18, TRW data on medium-to-wide rings can be made in an efficient way on many samples18,19,20. This is also the scale that can be used to assess biomass from pith-to-bark from both temperate and tropical trees4,22, typically ranging in resolutions from 50 &#181;m to 200 &#181;m.
Ring scale: Wood is a recorder of past environmental conditions. The best known parameter is tree-ring width (TRW), but for global temperature reconstructions, maximum latewood density (MXD) records are proven to be a better proxy for temperature22. MXD is an easy-to-measure variable23, and a proxy for cell wall thickness and cell size on the last cells of a tree ring, and are at tree line and boreal sites positively linked to seasonal air temperature24: the warmer and longer the summers, more cell wall lignification occurs which thus increases the density of these last cells. Traditional measurements such as immersion and geometry are less accurate to determine this ring-level density. A previous work developed a toolchain for using X-ray film on thin-cut samples25. This sparked a revolution in both forestry and later paleoclimatology15,18, defining maximum latewood density (MXD), i.e., the peak density value often at the end of a ring, as a proxy for summer temperature. The basic principle is that the samples are sawn (approximately 1.2 mm to 7 mm13) to be perfectly parallel to the axial direction, and the sample is put on a sensitive film exposed to an X-ray source. Then these radiography films are read out through a light source that detects the intensity and saves the profiles and the annual tree ring parameters. These tools, however, require a significant amount of sample preparation and manual work. Recently this has been developed for X-ray CT in a more standardized way or based on mounted cores26. Resolution here ranges between 10 &#181;m and 20 &#181;m. TRW is measured on this scale as well, especially when dealing with smaller rings.
Anatomical scale: At this scale (resolution &#60; 4 &#181;m), the average density levels become less relevant as the main anatomical features are visualized and their width and proportions can be measured. Typically, this is done through making microsections or high-resolution optical scans or &#181;-CT scans. When the ultrastructure of the cell walls needs to be visualized, scanning electron microscopy is the most commonly used method27. At the anatomical scale, the individual tissue fractions become visible so that physiological parameters can be derived from the images. Based on the individual anatomical parameters and the cell wall density of wood, anatomical density can be derived for comparison with conventional estimators of wood density24.
Due to improved sectioning techniques and image software29,30, dendro-anatomy30 has been developed to have a more accurate record of the wood, both to have a closer estimate of the MXD in conifers and to measure several anatomical variables from broadleaf trees. On this scale, actual anatomical parameters are measured and related to environmental parameters31 . With &#181;CT this level can be obtained as well32,33.
As wood is inherently hygroscopic and anisotropic, wood density needs to be carefully defined and the measurement conditions need to be specified, either as oven-dry, conditioned (typically at 12% moisture content) or green (as felled in the forest)34. For large samples and technical purposes, wood density is defined as the weight divided by its volume at given conditions. However, the value of wood density is strongly dependent on the scale at which it is measured, for instance from pith-to-bark wood density can double, and on a ring scale (in conifers) the transition of earlywood to latewood results in a significant rise in wood density as well, with a peak at the ring boundary.
Here, an X-ray CT scanning protocol of increment cores is presented in order to measure features at the aforenoted 3 scales (Figure 1). Recent developments in X-ray CT can cover most of these scales, due to a flexible set-up. The research goals will determine the eventual protocol for scanning.
A crucial limiting factor (which is inherently connected to the scaled nature of wood density and wood in general) is the resolution and time necessary for scanning. Examples demonstrate how to: (i) obtain inter-ring tree scale wood density profiles for biomass estimations in Terminalia superba from the Congo Basin, (ii) obtain density records from Clanwilliam cedar (Widdringtonia cedarbergensis) based on helical scanning on a HECTOR system35, and (iii) measure vessel parameters on sessile oak, on the Nanowood system.&#160;Both scanners are part of the suite of scanners at the UGent Center for X-ray Tomography (UGCT,
<img alt="Figure 1" class="xfigimg" src="/files/ftp_upload/65208/65208fig01.jpg" /> 
Figure 1: General methodological decision tree for X-ray CT scanning. The rows indicate the steps to take, starting from the research goal all the way to the final data format. White boxes are the steps that are relevant for this toolchain. Greyed-out boxes are steps that can be performed with other software or R packages, such as dplr47 and Treeclim48 for tree-ring analysis, and ROXAS44 as well as ImageJ42 or other (commercial) applications for deriving wood anatomical parameters based on the CT images. <a href="https://www.jove.com/files/ftp_upload/65208/65208fig01large.jpg" target="_blank">Please click here to view a larger version of this figure.</a>
X-CT research on wood 
Set-up of a scanner: A standard X-ray CT scanner consists of an X-ray tube, an X-ray detector, a rotation stage, and a set of motors to move the rotation stage, and in most cases also the detector, back and forth (Figure 2).
<img alt="Figure 2" class="xfigimg" src="/files/ftp_upload/65208/65208fig02.jpg" /> 
Figure 2. The HECTOR scanning system. The system35, showing the source detector distance (SDD) and the source object distance (SOD). <a href="https://www.jove.com/files/ftp_upload/65208/65208fig02large.jpg" target="_blank">Please click here to view a larger version of this figure.</a>
Most lab-based systems have a cone-beam geometry, which means that the produced X-rays are distributed from the tube&#39;s exit window in a cone-beam shape, implying that by changing the distance between the object and the tube (SOD = Source-Object-Distance) and the detector and the tube (SDD = Source-Detector-Distance), the magnification is controlled (see the discussion on resolution). Due to the penetrating power of X-rays, they pass through the object, and the intensity of the attenuation beam is a function of the energy of the X-ray beam, the chemical composition of the object (the atomic number of the elements present) and the density of the material. Given a constant energy spectrum&#160;and a constant material composition of wood, the attenuation of the X-ray beam is highly dependent on the density of the material, which explains its use for densitometry. The attenuation (or transmission) can be expressed by the Beer-Lambert law:
<img alt="Equation 1" src="/files/ftp_upload/65208/65208eq01.jpg" style="margin: auto;" />
with I0 the incoming X-ray beam exponentially which decays to a transmitted X-ray beam Id when propagating through the material over a distance d. The linear attenuation coefficient &#181; depends on a series of interactions with the material of the object. The projections are thus recordings of the transmitted beam.
Practically, the object is mounted on the rotation stage, a proper SOD and SDD are selected, a certain power is selected as well (related to object size, density, and composition) and the object is rotated 360&#176; and during that rotation multiple projections are taken. These projections are then used to reconstruct the interior structure of the object. There are several reconstruction algorithms available, of which the most used ones are still based on the analytical framework developed decades ago, relying on the Radon transformation and the Fourier slice theorem. For more details, the reader is referred to specialized literature36.
Conundrum of resolution, data volume and sample size: Resolution is key in X-ray CT scanning. In systems with inverse geometry, or parallel beam geometry such as synchrotron beamlines, other considerations play a role. This protocol only discusses standard lab-based X-ray CT scanning with cone-beam geometry. Here, the concept of magnification, detector pixel size and spot size are essential. Magnification is defined as the ratio of SDD/SOD. Next, the pixel size of the detector obviously also impacts the resolution: the smaller the pixel size, the higher the resolution, but in most cases the field of view (FoV) is also directly related to pixel size and size of the detector (smaller pixel size, smaller FoV for the same number of pixels). Further, the spot size of the X-ray beam is also important: the larger the spot size, the lower the resolution, which means that less details can be seen.
It is important to address that one could obtain a resolution that is higher than what is possible according to the abovementioned limits, therefore it is better to use the term voxel size (a voxel is a volume pixel) instead of resolution. Moreover, there are other factors at play, such as detector sharpness, which further limit the true resolution at which an object is scanned. Only a true calibration of the system, using established targets, supplies the truthful answer.
In most cases, however, the voxel size at which an object can be scanned is mostly limited by the size of the object. This means that the larger the object, the lower the voxel size will be. If the object does not fit within the FoV of the detector for a certain voxel size, then the voxel size can be reduced, e.g., by limiting the magnification.
Scan time and data volume are important to consider when deciding the desired voxel size. In general, the smaller the voxel size means the higher the detail one wants to see, the smaller the sample or the fewer samples that can be scanned at once, more time is required and larger the data volumes will be collected. Imagine the following theoretical example: one can scan a sample measuring 10 cm x 10 cm x 10 cm at 50 &#181;m at once with a certain X-ray CT system and would like to scan that same volume at 10 &#181;m, the volume that fits within the FoV would only be 2 cm x 2 cm x 2 cm, assuming that this is physically possible. This means that 125 scans are needed (5&#179; = 5 times higher resolution, scales to the power of 3 due to the volumetric nature of the imaging technique) to cover the entire volume, and that the data volume would increase likewise. Of course, this is only a thought experiment, and one needs to consider much more than only resolution. For more information, the reader is referred to an overview of the scanning possibilities37.
Flexibility of instruments for scanning of wooden objects: In the last decade, many companies have delivered X-ray CT systems with a similar assembly as HECTOR35. An overview of several CT systems, especially assessed for their temporal resolution, is given in38.
In all, the flexibility and user-friendliness of X-ray CT systems has improved considerably. Many systems allow scanning of a diverse range of objects, which is also the case with the systems at UGCT. The protocol below is demonstrated for the HECTOR system, which is suitable for tree-ring analysis. The protocol however is valid for any other available system if resolution and data format allows for it.
These systems allow the scanning of a variety of objects. A few pictures of different wooden objects scanned with the HECTOR system are given in Figure 3. It is this flexibility that comprises the three scales we present in Figure 1, ranging from a coarse resolution to a very fine resolution.
<img alt="Figure 3" class="xfigimg" src="/files/ftp_upload/65208/65208fig03.jpg" /> 
Figure 3.&#160;Scanning set-up examples. (A) A log, (B) a cello49, (C) sample holders (type 1) with tree cores for batch scanning and (D) sample holder type 2 with increment cores for helical scanning mounted on the rotation stage of HECTOR. <a href="https://www.jove.com/files/ftp_upload/65208/65208fig03large.jpg" target="_blank">Please click here to view a larger version of this figure.</a>

Author Spotlight: Advancements in X-ray CT Tool Chain for Tree Core Analysis

Tree Core Analysis with X-ray Computed Tomography

The scope of my research is to study wood growth and its response to climate. The questions I'm trying to answer is, how can we reconstruct the climate of the past and what is the tree response to current climate change? So the current experimental challenges that we face are among others, the large data volumes that we obtain with x-ray CT scanning.So we need to find a proper balance between the number of samples that we want to scan and the required resolution. The X-ray CT techniques has allowed us to scan three core at an unprecedented rate. We have density profiles both from tropical and temperate regions.So the main advantages that we have with our technique is that we can do a lot of high throughput scanning and analysis, but also that all the indications and manipulations that we do are traceable along the entire tool chain, including the tree ring indications. This tool will allow us to update current to maximum later density chronologies, and also measure wood properties on tree course from all over the world, especially in regions where there's little data available.

Optimizing Scanning Techniques for Biomass Estimation, Tree Growth, and Anatomical Analysis at Different Scales

To begin core scanning, depending on the research purpose, select the proper sample holder type, load the tree cores in paper straws into the sample holder, and record the position of each core in the downloadable spreadsheet template. Perform X-ray micro CT scanning following proper settings and scanning protocol. To obtain density values, install CoreProcessor, RingIndicator, and CoreComparison toolboxes using the given link.Next for pre-processing the core volumes with the CoreProcessor, select the folder with the reconstructed 16-bit cross-sectional TIFF slices in the spreadsheet file. Select a couple of slices for inspection. Then select dark and white references as indicated in the spreadsheet file for wood density calculations.In a new popup window, following the information from the spreadsheet file, select every core separately by drawing a circle or ellipse around it. Then click on Mass core extraction to extract all the cores in a given cylinder. In the CoreProcessor toolbox, click on Manual TG correction then select the extracted folder and ensure proper orientation of the transversal and radial plane of every core volume.To visualize grain direction from the bottom right plot, observe the presented slice and draw a line indicating grain direction. Double click to automatically rotate the core. Then crop the core volume to get only wood and no other material.Open the RingIndicator toolbox and select a multi-page TIFF file. Then under a Volume tab, inspect the graphical user interface in the transverse and radial plane. For structural correction on the transverse and radial plane, click on the image to insert green bars starting with the pith and ending with the most recent ring.Once a green bar is placed on one plane, it automatically generates on other planes. Click Densitometry, then Densitometry plot to calculate the density profile. Create and plot a density profile by selecting Overlay plotting, and then Plot density profile.Once the density profile is visualized, indicate all tree rings. Move the nodes at the end of the bar to change the angle. Then change the bar position using the middle node and adjust the size of the nodes.Press Data, then Export. Press Rings followed by Export rings to ensure ring and fiber indications are written in the proper txt files. Change the felling date to the year when the cores were taken from the living trees, or choose an appropriate date.In Overlay plotting, select Plot rings to display the years. Also choose the plane to plot rings or the density profile. After indicating two cores, open the CoreComparison toolbox for comparing the ring width series.Select the txt files to compare. Then a screen displaying ring widths and cross dating and statistical parameters, such as glysophikite and spearman correlation between individual series will open. Open a ring indicator instance to adjust the green bars and export them again.Then maximize the core comparison screen and press the Refresh button to reevaluate the metrics. Next, select Plotting and exporting, then Ring width to visualize the tree ring width or TRW data. Click Clotting and exporting, then export RW data to export TRW data in a spreadsheet or Tucson format.Finally, select the Other plotting, then Export cluster data for obtaining mean density, MXD, MND, and quartile data per tree ring. This study demonstrates three levels of biomass estimation or tree growth increment at the inter ring, ring, and anatomical scale. Radial and axial wood density trends in Terminalia superba demonstrated an increasing density trend from pith to bark and a higher wood density in the upper stem.An example of chronology development with minimum density and maximum latewood density from Widdringtonia cedarbergensis is shown. A high resolution oak core scan demonstrated that earlywood and latewood vessels are segmented.

Research

JoVE Journal

Environment

An X-ray computed tomography (CT) toolchain is presented to obtain tree-ring width (TRW), maximum latewood density (MXD), other density parameters, and ...