2. A missed or extra ring...

Last update: 16 January 2014. See red marked section below.
We start with inserting an extra incorrect ring in the Nmbs01.pos file.

Open CooRecorder and insert an extra ring below coordinate number 38 as shown in the picture above.
Save the coordinate file as "falsering.pos". (There is already such a file in the demo directory!)

We will now see how this incorrect ring can be detected.

Note: You may as well use CDendro to insert a zero ring into the Nmbs01.pos file! To do this, right-click on the curve diagram where you want your extra ring (shift-right-click to remove it). Then save the sample with the zero ring as a .wid-file!

1. Use the menu command Collections/Create reference curve from big Decadal file" and select the file swed302.rwl mentioned in the section "Match sample to reference curve"

2. In the new window, click the button "Select as reference" (if it is not already selected as the reference)!
3. Open the falsering.pos file: (Click Samples/Open ... single sample files..., then select "falsering.pos" which we created above).
4. Click on the Workbench tab
5. Click on the button "Make a whole sample correlation analysis"!

The following table is shown - There is no sign of a match in it!

C:\Program Files\Cybis\CDendro74\demo\falsering.pos using No detrend 
compared to the reference 
C:\Documents and Settings\Ake\My Documents\swed302 of length 414 using No detrend  Dated to 1995 
Minimum overlap used when finding best match: 50 
Table sorted by Proportion of last two years growth (2,0,T (P2Yrs)/TTest 
--Rel Over  *P2Yrs------  BaPi-------  C84F-------  BesIE------  Skel-  GLK--  (year) 
-year  lap   CorrC TTest  CorrC TTest  CorrC TTest  CorrC TTest   Chi2    GLK 
  342   71    0.39   3.5   0.31   2.7   0.29   2.5   0.34   2.9    1.1   0.58  (1653) 
  314   85    0.35   3.4   0.37   3.6   0.32   3.1   0.27   2.5    1.6   0.62  (1681) 
  311   85    0.31   3.0   0.32   3.0   0.24   2.2   0.23   2.1    8.1   0.59  (1684) 
  119   85    0.30   2.9   0.35   3.4   0.40   3.9   0.45   4.5   17.8   0.63  (1876) 
  291   85    0.29   2.7   0.27   2.5   0.27   2.5   0.22   2.1    1.1   0.55  (1704) 
   82   85    0.28   2.7   0.18   1.6   0.14   1.3   0.09   0.8    0.5   0.68  (1913) 
  190   85    0.28   2.7   0.30   2.8   0.28   2.7   0.18   1.6    5.3   0.46  (1805) 
  336   77    0.29   2.6   0.28   2.5   0.19   1.6   0.13   1.1    5.3   0.59  (1659) 
  218   85    0.27   2.5   0.23   2.1   0.19   1.7   0.17   1.5    0.8   0.61  (1777)
We have two buttons to handle this situation:
1. The "Find match despite a missing ring!" button which sometimes works well when there is only one or possibly two incorrect rings and when the reference curve is not too long.

2. The "Make block correlation analysis" button which gives a list of possible matching points for separate parts of the curve. With also the "Extended block length test" checked it might sometimes indicate a matching point.

Click the "Find match despite a missing ring!" button.

We get the following table - which tells there is an incorrect extra ring somewhere around ring number 38:

Table sorted by Proportion of last two years growth (2,0,T (P2Yrs)/TTest 
--Rel Over  *P2Yrs------  BaPi-------  C84F-------  BesIE------  Skel-  GLK--  (year) 
-year  lap   CorrC TTest  CorrC TTest  CorrC TTest  CorrC TTest   Chi2    GLK           Approx position of inserted/removed extra rings 
  120   84    0.81  12.4   0.81  12.3   0.81  12.4   0.76  10.4   42.4   0.84  (1875)   -38 
  118   83    0.70   8.7   0.65   7.8   0.67   8.1   0.62   7.0   32.0   0.73  (1877)   +18  +19  -38 
  122   82    0.64   7.5   0.56   6.0   0.61   6.8   0.60   6.6   22.4   0.76  (1873)   -18  -19  -38 
  121   83    0.62   7.1   0.63   7.4   0.65   7.8   0.66   7.8   28.4   0.73  (1874)   -18  -38 
  119   84    0.59   6.7   0.57   6.3   0.59   6.7   0.58   6.3   26.2   0.73  (1876)   +18  -38 
  117   83    0.48   4.9   0.45   4.5   0.38   3.7   0.37   3.6   20.3   0.69  (1878)   +38  +39 
  198   84    0.43   4.3   0.41   4.0   0.37   3.6   0.35   3.4    9.9   0.61  (1797)   -60 
  115   81    0.44   4.3   0.42   4.1   0.38   3.6   0.39   3.8   11.8   0.63  (1880)   +28  +29  +38  +39 
   20   82    0.41   4.0   0.36   3.4   0.33   3.1   0.17   1.5    6.7   0.56  (1975)   -18  -19  -38  +60 
There is a match at relative year 120  i.e. at  1875

In this case the "Find match despite a missing ring" worked very well - though it is not always so!

Next, we will use the Make block correlation analysis button to look for possible matches.


Click on the button: "Make block correlation analysis" We get the table shown above!

The "falsering.pos" sample has been divided into (overlapping) chunks of a lenght of 30 years. These six blocks have all been separately correlated towards the reference.

Five of these blocks have a best match which would set the whole sample to relative year 120 or 119, i.e. 1875/76. The block, covering rings between ring 20 and 50, contains (as we know) an incorrect extra ring so this block does not match any good at its right place. Though there is another match for this block at relative year 1 meaning that if it was a correct match, then the whole sample should date to year 2014. When this was originally written that was in the future...

Update Jan 16 2014:
Another way to find the 1875/76 match is to look at a line below a "SetsSampleTo alternatives..." line (see above) and look for two adjacent years which will give you 1875 and 1876. At the last yellow marked line above this gives 2 blocks matching the reference so that the whole sample should match to 1875 and 3 blocks matching the reference so that the whole sample should match to 1876. I.e. a good chance that there is a false ring somewhere in the middle of these 5 blocks. - You might also experiment with changing the blocklength from 30 to e.g. 40 or 50.

Still another way is to check "Extended block length test" before running the "Make block correlation analysis". This test is run exactly as the test described above, but now when a block has been tested for a match it is then extended by 10 more years (=the length of the blockdistance) and then again tested for a match, then again extended with 10 years etc. as long as the block can be extended within the sample. I.e. we have successively growing blocks being tested for their best and next best matches.

In cases when you know that the sample is not younger than the reference, you can avoid too young nonsense datings, by unchecking the box "Sample may be younger than the reference". This holds for both whole sample correlation analysis and (from version 7.8) also for block correlation analysis!

Note: Another way of finding errors in your data is to look at the "Curve display" window where you can shift your sample over the reference curve to find where you have missed or added an extra ring. This is especially useful when you have measured several radii from the same log and want to match them together and make a mean value curve of ring widths. In that case look carefully at both the lower ring width curves and the upper normalized curves when searching for an error.

Note: Also consider that when your sample is long enough you can easily cut it in two halves and test them separately for correlation. To cut them, use the settings and "Create sample" button within the "Look at blocks..." frame in the lower left corner below the curve diagrams.

Note: Yet another way of finding errors in your data is to have several samples in a "collection". Then check the box "With block checking" and click the button "Test towards reference". This will give you a table like that shown above for each sample in the collection. In that case, please note that the settings for blocklenght and blockdistance is done in "Settings/Options for normalization and matching"



Copyright 2014, Cybis Elektronik & Data AB, www.cybis.se