How to build a synchronized collection from unsynchronized members

In this lesson we will see how you can easily synchronize a number of collection members towards each other.
This collection contains 10 members which are not synchronized. Currently their offsets are all set at zero!
You can download this collection from unsync.rwl

We will now see how to easily synchronize these members using the "Add best members to target collection" button of CDendro. First we have to find a member to start from, i.e. one member that matches well towards several other members of the unsynchronized collection!

We start off by requesting a table showing how each member matches best to all other members. To make the matrix small for this demonstration, we only look at the best TTest values, but you may very well check several other items, e.g. correlation coefficient and overlap.
Note that we order the matrix sorted by the best matching members.
Here we can see that NMBS08A could be a good starting point when building a new synchronized collection. It has six matches against other members of the collection with TTest values at 6 or above.
So let us start with an empty target collection where we put NMBS08A as the only member. It will then work as a "seed" that will attract other matching members.

To copy the NMBS08A into the target collection:
1. Either double click the NMBS08A member so it opens in a separate window. Then use the "Add to target collection" button.
2. or see that only NMBS08A is checked and then use the menu command Collections/Copy collection members to target collection.

We then turn to our unsynchronized collection where we click the button "Add best members to target collection". We then set up the values as shown and click OK.
CDendro will quickly find that NMBS11 gives a good match. Click OK to continue!
NMBS11 will then be opened within a new "sample window" and compared towards the mean value of the members of the target collection, in this case just towards NMBS08A. If you accept this as a good match, you may very well click on the button "Add to target collection".
Though here we wanted to investigate that match carefully, so we clicked the "Workbench" tab (1).

From a comparison with several normalization methods, the table shows that this can be considered a very good match (2). So we order the sample to be added to the target collection at offset -3 as shown above (3 and 4). Then we close the NMBS11 sample window.

Note: With CDendro 7.4 there is a third button Add, close and show next candidate to the left of the Ok-button above to be used for quicker handling.

We now have two synchronized members in our target collection.
We then turn back to our unsync.rwl collection and click "Add best members to target collection". Please note that both NMBS08A and NMBS11 have been automatically unchecked, as they are already added to the target collection.
Still one more matching candidate, 6671, is found!
The new candidate for addition, 6671, is now compared to a mean value of NMBS08A and NMBS11, i.e. to a mean value of the target collection.
After we have added it to the target collection, we have three synchronized members, all at different offsets, within our target collection.
We will now see how this adding process may run automatically!
See that the required TTest-value is set high enough, here at 8.1. Also see that the lowest allowed correlation coefficient for a 30 years long segment/block is set high enough, here at 0.35
Uncheck the "Stop before adding a candidate..." box and click OK!
CDendro then found the next four best matching members of the unsynchronized collection and added them to the target collection at synchronizing offsets.
Note that only the remaining - non-added - members above are still checked!
In this simple case there are only three members left, though with a bigger unsynchronized collection it can be quite difficult to find the next best matching member.
We will now match each remaining checked member of the unsynchronized collection towards a mean value of the target collection. To make CDendro look for the best match (and not the current match with current (wrong) datings) we should use the "(Test towards)...undated ref." button as shown above.
Also see that the checkbox "With condensed report" is checked so you get a sorted list of matches from the remaining members. (With many remaining members, this list really matters!)
The tables show that NM049 is now the best matching member to add to the target collection.

(Note: The very bottom row above is not relevant for this type of matching. It is actually intended for the case when many properly dated reference curves are stored in the collection which is then compared to a reference of unknown origin, e.g. ships timber. Then the bottom row shows the best possible dating years, each with a count showing how many references that had its best match at that year together with the mean TTest value of these matches.)

After adding the remaining members at synchronized offsets we have all members synchronized.
Running "Test towards rest of collection" will show very good correlation values.

Note that by clicking on the headers of the CC or TT columns you can sort the collection members by increasing or decreasing Corr-coeff- or T-values!

It is always a good habit to check how the individual members crossdate towards each other.

With an error within the first member added to an empty collection (e.g. an erroneous reference!) you may very well end up with a "synchronized" collection giving you good values when running "Test towards rest of collection". Though that collection is anyhow wrong, which could be revealed by a matrix as above where some members would then give conspicuously low correlation values.

The original unsynchronized collection above was provisionally dated to 8000. You may now create a mean value sample from your new collection and compare that to a known reference so you can set a proper endyear to your synchronized collection.

In this case it was compared to a reference from the island of Nämdö and found to match at year 1895, see sync.rwl.

More options

Requirement for a Minimum Chi2 value Since the first versions of the "Add best member mechanisms", the new requirement "Minimum Chi2 value for selection" has been added. This is because - for some normalization methods - extremely narrow rings at different positions in a sample and in a reference might "attract each other" and thus fool the correlation calculus to find a false match (i.e. where the narrow rings overlap each other), see the section Methodology/Fooled by narrow rings and the normalization method. We have found that in such cases of a false match, we have always also found a low Chi2 value. With the requirement of a reasonably high Chi2 value for a match, we have thus reduced the risk of fooling this Add-best-member tool.

Blockwise checking When blockwise checking is turned on, CDendro will see that all blocks (segments) of the best found member do have at least a minimum correlation coefficient value against the current reference.
The With extended messaging option gives you a chance to see where these block-checkings fail (i.e. where and for which member they give a too low value).

Blockwise matching When there are some very long members with an error somewhere in the middle of the ring width sequence, there will be no good match to find towards a reference, though a block (segment) at one side of the error might give a high T-value when placed at the correct matching point. To overcome this and anyhow find such an erroneous member within a very big unsynchronized collection, you can turn on "Blockwise matching". Depending on the block length and overlap specified you might have to set the T-value somewhat lower than normally, e.g. at 6.0 to find a block of 100 years in length with a correlation coefficient above 0.52.

August 24 2013 (first version Sept. 17 2010), Lars-Åke Larsson, www.cybis.se



Copyright © 2017, Cybis Elektronik & Data AB, www.cybis.se