'Sample towards sample matching STANDARDDEF std.def 'should set up colors BASEPATH cdendro1/ RTHUMBPATH pic/ 'Set naviginfo to empty to avoid "if you click on a picture ...." NAVIGINFO NOSLIDES NOCENTER LIBPATH ' set as empty, then requires slides.htm, picnavig.htm and some .gif-files on this directory NOCENTER TITLE 1. Matching two samples in CDendro NEWLINE TEXT Last update: Nov 26 2023 (a tip at the end) NEWLINE PIC openpos.gif NEWLINE TEXT

Use the CooRecorder program to measure the two samples NMBS01.jpg and NMBS05.jpg That's the two look alike logs from the Bergvik house at Nämdö. (To find these files, see the note below!)

Start measuring from the outermost ring and measure towards the inner part of the tree.

When you have the coordinate files as your versions of NMBS01.pos and NMBS05.pos stored somewhere, then start CDendro.

1. Select one of the open commands shown above. Navigate to the proper directory and click on Nmbs01.pos.

2. Then hold down the control key and click also on Nmbs05.pos. Then click the Open button.

Note: You will find the files mentioned here in the directory "demo" in that directory where you have installed CDendro, normally "C:\Program Files (x86)\Cybis\CDendro92\demo".

NEWLINE FILL 1 PIC dofirst.gif NEWLINE TEXT
The uppermost window (the one that is not hidden) is now Nmbs05.pos
1. Click on the button "Select as reference"
2. Click on the top bar of Nmbs01.pos to select that window.

NEWLINE PIC saymakecorr.gif NEWLINE TEXT
Now we are at the Nmbs01 window.
Click on the button "Make corr. analysis" to do correlation calculations towards the reference Nmbs05
NEWLINE FILL 0 PIC matching.gif NEWLINE TEXT
The sample Nmbs01 has now been compared to the reference Nmbs05. The program automatically places the curves over each other at the best matching point, which is here at offset 0, indicating that both trees were cut the same year.

The upper two curves show normalized ring width values, the lower two curves show actual ring width values. The red and green curves belong to the sample Nmbs01, the black and blue curves to the reference Nmbs05.

You can shift the curves relative to each other by clicking on the up/down arrows to the right of the offset field... (Or use the right/left arrow keys, but then you first have to click within the curves window to make it "focused".)
You can also change the offset value directly to e.g. 35.

To see a text report on the matching done, click on the Workbench tab!

NEWLINE TEXT

The report window on the Workbench tab shows:

The two asterix (**) over the rightmost column indicate that the TTest values of the P2YrsL method are used for sorting the best matches. The table shows that these two samples match when their first year (0) overlap each other. The correlation coefficient between the curves is then 0.68 with the P2YrsL method and 0.61 with Baillie/Pilcher's method. The overlapping is then 84 years long. (The Nmbs01 sample is not more than that long.)

The next best match has a coefficient of not more than 0.27 when the Nmbs01 sample is positioned as if it was 35 years older than the reference Nmbs05 and with (only) 67 years overlapping.

The TTest value is based on the correlation value but it also takes into account that a match with a short overlap is less worth than a match with a longer overlap when the correlation coefficient values are the same.
The TTest value is calculated as where c is the correlation coefficient value and n is the number of overlapping years.

Note: The contents of the table above may differ on your machine depending on your setting of least overlap in "Options for normalization of ring widths and for matching". Here we had a 50 year least overlap. With a setting of e.g. 70 years least overlap for comparisons, the 35 year alternative above will not be found.

NEWLINE TEXT
The process of dating is based on:
  • Finding two samples which match properly together.
  • Adding these two together and then finding more samples which match the first two.
There are tools in CDendro to make this process automatic or semi-automatic!

Note: If you are comparing a group of curves where some are quite short say 50 years, then see that Settings/Options for normalization.../Least overlap in years between samples when correlating is set at a value which is considerably lower than those 50 years! (30 is fine!) Otherwise the program will not find the match where a 50 year long sample is 10 years younger than the reference.

NEWLINE TEXT
Tip:When comparing correlation values from different normalization methods like P2Yrs and BaPi:
Check that overlaps are the same!

Example: Two curves compared. The reference is longer than the sample tested which match at offset=0.

Table sorted by P2Yrs: Proportion of last two years growth

--Rel Over P2Yrs ----** BaPi ------ (year)
-year lap CorrC TTest CorrC TTest
0 130 0.45 5.7 0.44 5.5 (1870)

Overlap for P2Yrs** is 130 with a TTest value of 5.7
For BaPi the series was not long enough to calculate it for overlap=130, so 129 was selected (not shown).

Table sorted by Baillie Pilcher Corr coeff:...

--Rel Over P2Yrs ----- BaPi -----** (year)
-year lap CorrC TTest CorrC TTest
0 129 0.48 6.2 0.44 5.5 (1870)

The longest overlap for BaPi** is 129. When that shorter overlap was used also for P2Yrs, its TTest value increased to 6.2.

When P2Yrs is the master method (**) the maximum overlap is 130 giving a TTest value of 5.7 Though when BaPi is the master method (**) the maximum overlap is only 129 giving a TTest value of 6.2

Beware of different overlaps when comparing CC or TTest values of different normalization methods!