WinDENDRO, an image analysis system for annual tree-rings analysis
Image Analysis for Plant Science
img0
Regent Instruments Inc. since 1991

 

 

WinDENDRO

 


For sales informations,
contact us by email

sales@regentinstruments.com

 
Analysis Process
Image Acquisition
Ring Detection
Rings Validation
The Ring-Widths Graphic & Cross-Dating
Density Analysis
Data
Stem Analysis
Miscellaneous
Image Acquisition

The first step of tree-ring analysis is image acquisition. WinDENDRO has been designed and is optimised for optical scanners but can also analyse images from cameras and digital filmless x-ray systems. It can open image files produced by these hardware manufacturers programs (provided they are saved in a standard format such as tiff, jpeg, jpeg2000, bmp or png), but most of the time WinDENDRO is used to acquire images directly from such devices (when they are TWAIN compatible*). Optical scanners are particularly well adapted for tree-rings analysis. They are very easy to use and produce high quality images over large areas, free of illumination problems (such as non-uniformity). They have a built-in permanent calibration (unlike cameras), so the operator does not have to calibrate their images, neither have to adjust their lighting system or lens aperture and focus.

WinDENDRO deals with scanners in a very efficient way. It has two methods of image acquisition for them, one is optimised for ease of use and requires just a mouse click to get an image. Time is saved in bypassing the standard "Preview" step (which can take tens of seconds per sample) and by using positioning accessories which allows to repetitively place the samples at the right place on the scanner. The other scanning method is more complex and powerful, it uses the Preview step to optimise the scanning parameters.

   

Regent’s optional core holder and positioning system eases and accelerates cores scanning by allowing the operator to rapidly position the samples at the same place on the scanner glass (thus eliminating the need to preview before a scan) and allows manipulation of cores during their preparation. This system can be rapidly added or removed at any time.

Note: Scanner illustrated is not a model we sell.

 

Disks are placed face down on the scanner for image acquisition. You can scan the whole disk (although this takes lots of memory) or you can scan narrow paths, a few millimeters or centimeters in width, from pith to bark. A scan typically takes between 10 and 60 seconds to complete depending on the image size (scan area and resolution). Note that scanning whole disks or using very high dpi (such as 2400 or more) will take more time (and a fast computer with plenty of memory). Right after scanning, the image is displayed on screen.

 

* WinDENDRO can work with almost any TWAIN (32 & 64 bit) scanners that you can buy locally. Our scanners come with a calibration for higher precision and optional accessories such as core holder and positioner. Note that we do not suggest or recommend scanners.

** To learn more about sample preparation required for scanning and WinDENDRO see the "Regarding wood species, narrow rings and ring contrasts" note next.

Ring Detection

After image acquisition, you indicate WinDENDRO where to measure rings in the image. This is done by tracing paths interactively. Straight line paths (such as those extending from a disk pith to bark) can be created automatically with a single mouse click (up to a few hundreds in a single click). More complex shape paths (see below) are created manually by clicking at different places.

  In its simplest form a path can run across a sample (core or disk radius) on a straight line. Hundreds of paths can be created with a single mouse click in pre-defined directions around the clicked position. Paths can also be created by clicking at their beginning and ending points.  
       
 
You can trace paths with a perpendicular trajectory to ring boundaries (as in manual dendrochronological measurement methods).
 
Paths may contain discontinuities to avoid damaged areas in order to move perpendicular to ring boundaries.
 
  Ring-widths can be measured taking into account their boundary orientation relative to the path. This allows to increase the precision when using straight paths (to achieve similar results as to path made perpendicular to ring widths).  
   
  One or more cores can be analysed per image. In this example multiple cores have been scanned side by side to save scanning time. They are then analysed in the same image by clicking their respective beginning and ending (for curved cores intermediary clicks are required).  
   

After paths have been created, rings are automatically detected by WinDENDRO and their presence is indicated over the image with lines and text. Lines indicate the rings boundary position and orientation and the text indicates the year and ring number. Earlywood width can also be displayed along with other ring features (explained next). Close to the image, a profile of the light variations inside the path is also displayed along with ring and earlywood widths. This region is also used to adjust the sensitivity of the automatic ring detection. The sensitivity of this initial detection can be adjusted in function of the rings appearance. Narrow and low contrasts rings require more attention.

One or four methods of ring detection are provided in function of the WinDENDRO software model. The first method is based on light intensity differences. It is simple, efficient and works fine for contrasted rings like those of conifers. Another ring detection method is called Teach & Show and is a method where you show to WinDENDRO what a tree ring is in the image so that it can automatically detect them after. This method plus two others are well adapted for low contrasted rings like those of deciduous trees. There are limits however and minimal contrast is required in order to get a good level of automatic detection. When contrast is too low, you can work in manual mode where you indicate the position of rings in the image by clicking them.

 

The image content inside the path (not only the central line) is used to calculate the intensity profile (and for the density version, the Density or Blue Reflectance profile) displayed parallel to the path. The path width is adjustable so that you can choose what the profile is made of (paths should only contain valid tree-rings information, they should not encompass the core holder for example).

The WinDENDRO Density version also uses ring orientation to produce more precise profiles. Each time a ring is moved or reoriented, the intensity profile is recalculated using a virtual slit that matches the rings boundary.

   
 

Regarding Wood Species, Narrow Rings & Ring Contrasts

When rings are well contrasted, such as with medium to large rings (0.5 mm and over) of coniferous species, the automatic detection rate is in the range of 85 to 100% requiring little modifications from the operator so the productivity gain over manual methods is very high. Rings with lower contrasts such as those from deciduous (hardwood) species or narrow rings, require more attention to preparation, scanning and analysis settings. The more time is spent on obtaining a good image, the less time is spent on their analysis (this is true for manual methods also). Low contrast samples can be analysed with WinDENDRO but the productivity gain over manual methods is lower than with conifers because more operator corrections are needed. Low contrasted rings requires paying attention to:

1) Sample preparation. It has a great influence on the automatic detection rate. As rings get narrower, the finer the preparation has to be (0.01 mm rings require a finer method than 1.0 cm rings). Visible mechanical marks (like scratches) should be avoided as they tend to be more visible in digitized images than to the naked eye. They can trigger false ring detections or wrong orientations. There are no universal method accepted for preparation but sanding is very popular for dry samples.

2) Scanning. Contrasts enhancements done during scanning or after in WinDENDRO allows to see the rings more easily. Narrow rings require higher resolution (DPI) than large rings. As a rule the practical minimum number of pixels per ring is four and this number increases (up to ten typically) as ring contrasts lower. A good quality scanner is also mandatory. Good quality is not only related to the theoretical dpi claimed by its manufacturer, it is also a matter of good optics and electronics. Regent Instruments tests and compares all scanner models it sells. It also ensures they are compatible with WinDENDRO and its accessories. Regent Instruments can take the time to look at your samples before recommending a system, so do not hesitate to contact us.

3) Analysis settings. They must also be fine tuned for low contrasts rings. You can experiment with the two methods of ring detection provided and adjust their parameters to optimise the automatic detection. Some samples are better done in the complete manual mode.

 

 

 

Rings can be tagged with observations that you define (you choose their name, meaning and the symbol used to indicate their presence close to a ring). For example you could define two features called "narrow" and "frost" and select which rings have these characteristics. These ring features are then displayed on the graphic and in the image close to rings which have these features.

No modifications are made to the original image which is always available for future reference or analyses. The images can also be exported to other software programs (to create a report for example). They can also be saved with their analyses and later be retrieved and edited or validated by WinDENDRO.

Rings Validation

After the initial automatic rings detection, a validation must be done to consider the possibility of the presence of false, frost, locally absent rings or simply misclassification done by WinDENDRO. In this regard we say WinDENDRO is a semi-automatic rather than an automatic system. This is done by browsing the image and looking for missing or false rings. Contrary to systems based on positioning tables, you can switch back and forth along the ring paths without precision loss (due to gears backlash). Previously identified rings can be reviewed at any moment even years later. Adding or deleting rings is easily done interactively. When the mouse is held over a ring, the latter is highlighted (in yellow as illustrated). Clicking it deletes the ring, clicking at a place there is no ring adds one. Rings can be deleted in groups, moved or reoriented. Earlywood-latewood boundaries position can also be overriden. As modifications are done, the rings number and year are automatically updated in the image and the ring-widths graphic.

 
Detected rings can be reoriented to match the ring boundary for more precise measurement.
Rings can be moved.
 
Rings can be deleted (or added) by clicking them one by one or by using an image selection.
 
 
The Ring-Widths Graphic & Cross-Dating

 

A graphic of ring-widths in function of the year is displayed during the analysis and is automatically updated as rings are edited during the validation phase (Reg and Density versions). This graphic is also used for visual and numerical cross-dating. It can display simultaneously master chronologies and the ring widths of the sample under analysis and correlate some of them to help find mistakes in the analysis. Ring-width series can be detrended (converted to indices) using the smoothing spline method.

The smoothing spline can be displayed to help determine its filtering strength.

Skeleton plots can be displayed during the measurements of one or more series simultaneously with master chronologies. These are used by dendrochronologists to identify rings that are smaller or larger than their neighbors for visual cross-dating.

Ring widths can be displayed unmodified (as measured in millimeters) or as index values (ring widths for which long term non-climatic variations such as those due to tree aging have been removed) and can also be converted to a logarithm scale to increase the effects of narrow rings variations.

There are many interactive commands associated with the graphic. For example, when you click a year on it, the image is scrolled to display the part of the image where that ring is. The ring-width series can be splitted and shifted at different places (and the correlation updated) to help find missing or false rings. Modifications done by adding rings at splits points can be ported to the analysed sample by activating a command. Three split points where the data series can be independently shifted. More points can be added or removed.

Density Analysis

WinDENDRO is available with or without density analysis capability. The principles behind WinDENDRO’s density analyses are well known and have been applied for years in tree ring analysis. Different methods can be applied to perform density analysis in WinDENDRO:

  • The conventional film-based method
    This method is the oldest and most widely used for tree ring density analysis by dendrochronologists. It is well documented in Dr. Schweingruber book Basic and Applications of Dendrochronology. Wood samples are cut into thin slices (1 to 2 mm thick), brought to a predetermined humidity level and exposed to x-rays over a film. The film is then developed. In WinDENDRO, the conventional analog densitometer is replaced by a digital method that consists of scanning the x-ray film and then measuring light that passes through a virtual slit scanned over the rings. Unlike the analog densitometer, parameters such as slit size can easily be changed. When computing density, WinDENDRO automatically sets the slit angle tangent to the ring boundary in order to produce accurate measurements. The slit angle linearly changes its orientation gradually between rings so that it is tangent to all rings boundaries. The operator can interactively override the slit angle estimated by WinDENDRO.

  • The Blue Reflectance method
    This method does not produce density measurements but a surrogate of it. It estimates density-like data based on the assumption that wood color (or reflectance in the blue band) is related to density. In recent years some researchers obtained interesting results using the blue channel of color images. One major advantage of the reflected-light method is that it does not require costly radiation equipment. It requires a good understanding of its limitations however.

  • The filmless digital x-ray imaging method
    It is possible to use the newest filmless digital x-ray cameras or scanners (not sold by Regent) to acquire x-ray images of pieces of wood and then measure their density. Some systems require that the pieces of wood still be prepared as described above. Density is measured in WinDENDRO from these images exactly as it is from digitized films (except for calibration).

Multiple density measurements can be saved on a ring or pixel basis. Available per ring: ring width, earlywood and latewood width (in mm or percentage of ring width), ring maximum density, ring minimum density, ring mean density, earlywood mean density, latewood mean density and ring boundary orientation. Available per pixel along a ring path: pixel density or light intensity (calibrated or not) and slit orientation.

Data

When an analysed image is saved to a file, the analysis is automatically saved with it. This analysis can later be retrieved,validated or modified simply by loading the image in WinDENDRO. The analysis data such as ring width, minimum density etc, are also saved to standard text files that can be read by many programs including spreadsheet style software like Microsoft’s Excel. WinDENDRO has its own format (documented in its user guide) to store these data but can also convert files to the decadal (Tucson) format used by dendrochronologists (at 1/100 or 1/1000 of a mm precision). Unlike the decadal format, the WinDENDRO format allows to store the analysis settings, date and time, image information along with the rings measurements.

Stem Analysis

XLSTEM is a stem analysis program (set of Macros) that runs within Microsoft Excel. It allows to visualize data produced by WinDENDRO and to do standard stem analysis (like reconstituting tree growth as a function of age, measuring tree volume...). Calculations can be done interactively on selected trees or in batch.

XLSTEM calculates the following information from ring width data produced by WinDENDRO:

  • Mean radius (quadratic method), diameter and area per disk (cumulative or incremental)
  • Tree height and volume as a function of age or year (cumulative or incremental)
  • Basal and summary information about the tree

It has three height interpolation methods: Linear, Carmean and Newberry.

Miscellaneous

WinDENDRO comes with a printed (and digital pdf file) color illustrated manual and prompt and competent technical support (via e-mail). The typical answering time is within one hour. It is also done by competent persons, people close to and which can rely on WinDENDRO programmers for technical advices.

WinDENDRO is a member of a family of related products for plant science research and production. Among them you will find: WinCELL for wood anatomical cell analysis, WinRHIZO & WinRHIZO Tron for root analysis (extracted and in-soil) WinSCANOPY for canopy and radiation analyses from fisheye hemispherical images WinCAM for color analysis WinFOLIA for broad leaf analysis WinSEEDLE for seed and needle analysis

The steep increase in scanner and computer performances and prices decrease over the years have made WinDENDRO systems more than an affordable solution to set up a tree-ring facility. Its cost compares favorably with binocular and positioning table based manual systems.