More on synchronizations

There are two situations where you want to sort and compare ring width curves to make them match together in a family:
  • When you have lots of sample-files from one site and want to build a reference curve of them all. A single sample from your site does normally not match other reference curves. But a sum of samples - a reference curve for your site - can often be matched to a reference curve from a place quite far away.
  • When you want to create a serie of ring width mean values out of several radii measured from one and the same log. Using several radii makes your data more representative for the log as a whole, and so makes data easier to match against other ring width data series.

For these two cases, you can use the same procedure in CDendro to find out how curves fit together! (i.e. use the Add best member.. button)

Synchronizing radii from the same stem

In the case of synchronizing a group of radii from the same stem, there are some things to remember:

  • See that "No Detrending" is selected when creating the Mean value sample! Samples from the same stem usually have some characteristics in common that disappears when the ring width curves are detrended.
  • "Sum by stem" should be unchecked, otherwise the standard deviation curves will not be shown for the calculated mean value sample! (Provided that the sample name for each radius follow the CDendro naming standard)
With these settings the "Add best members to target collection" will probably be a good help to synchronize your collection!
How to continue an interrupted synchronization work?

If  you have to stop your work when you are halfway through the job, just save your target collection as e.g. a .fil or a .fh collection. You should also save your unsynchronized collection. You may very well save that unsynchronized collection with all the members that you have already added to your target collection!

When you later have time to continue, open that target collection and select it as your target. Then also open that unsynchronized collection and click the button "Add best members to target collection".

Within your unsynchronized collection, CDendro will then automatically uncheck all members  which have already been added to the target collection! This way you can resume your process of successively adding members to your target collection!

What to do if the adding process stops?
If samples are many and they do not match good enough, the process of successive additions will stop after a while telling that the next best candidate has not those good correlation values asked for.

How to find best matching samples?

After an "Add best members..." operation, there may still be some samples left which could be added after a visual inspection.
To find out how all samples left matches your new target collection:
1. For the target collection, click "Create mean value sample"
2. See that the resulting sample is selected as the Reference!
3. See that the Settings/Options for normalization and matching/Least overlap... value is set to 50
3. In your "many-samples" (unsynchronized) collection click "Test towards reference"

This will give you a list where all still not added samples (those still checked) are listed with their best and next best matching points towards the reference. If you find any sample which you really believe could be set at its right position, then double-click it and do a visual inspection. If it really fits, consider adding it to the target collection.

How to restart the addition process?

Sometimes it happens that you can build a second - in time parallel - reference curve (separate collection) from a site. This could be because some samples are taken from one place and others from another place. In such a case you may restart the addition process to create a second collection.

Building the first collection:

1. Start from an "unsorted many-samples" (unsynchronized) collection containing those samples you want to find the right offsets for.
Get a new empty target collection and start the addition process as described above.
When there are no more good looking samples to add, save your new collection, name it e.g. BEST1.fil. Close all windows.

Restarting - Building the second collection:
Now, we want to build another collection based on samples not in your BEST1.fil-collection.
2. See that BEST1.fil is your target collection.
3. Click the button Check all in that "unsorted many-samples" collection.
4. Select the command Collections/Uncheck members already in target collection.
5. Create a new target collection!
6. Click on that "unsorted many-samples" collection.
7. Click Show cross correlations.
8. Inspect the list at the end of the correlation table and select one sample which is quite long and has high correlation values in relation to other samples. You might also find out this from the "chocolate matrix".
9. Find this sample in the list with samples above the report frame and double-click on it to open a sample-window with a diagram for it.
10. Add this sample to your target collection at offset 0! Close the sample window!
11. In your "unsorted many-samples" collection, click on "Check all" to make all members of that collection available in the adding process towards the new target collection. That way the new collection may grow to cover parts of the BEST1 collection.
12. Click on "Add best members to target collection" and the addition process starts up.

How to select those members not within two already created collections, BEST1 and BEST2?
1. Click "Check all" in that "many-samples" collection.
2. Open the BEST1 collection and select it as the target collection.
3. Click in that "many-samples" collection to make it the top window.
4. Select the command "Collections/Uncheck members already in target collection"
5. Open the BEST2 collection and select it as the target collection.
6. Click in that "many-samples" collection to make it the top window.
7. Select the command "Collections/Uncheck members already in target collection"
Now the members checked in your "many-samples" collection are only those not within the BEST1 or BEST2 collections.
You may save this limited collection as e.g. NOTBEST1BEST2.fil - only checked members will be written!

How to add two collections?

When  you have two collections BEST1 and BEST2 and want to add them together:
1. Click on "Create mean value sample" in both collections.
2. Select the BEST1 mean value window as the Reference sample.
3. Do a whole sample correlation analysis in BEST2. If the collections do not fit together, stop here!
4. See that the BEST1 collection is your target collection.
5. From BEST2, select the command Collections/Copy collection members to target collection. Accept the proposed offset based on your previous correlation analysis.
6. The members of BEST2 will be copied into BEST1. Save BEST1 as BEST12.fil. Members of BEST2 which are already members of BEST1 will not be copied.


Quality control

There are four buttons in a collection window for quality control calculations:

1. "Show cross correlations" with various options and the chocolate graphics to show the quality of the matches.

2. "Test towards reference" which gives you a table showing how each member matches towards the current reference. If you see that the box "With block checking" is checked, you will get a list like that shown above. If a block of a sample has a high correlation while the other blocks have low correlations, this is an indication of a missed ring. This can be checked by double-clicking on the sample in the collection list and then inspecting the diagram as shown above. Here a year-ring is obviously missing between year 80 and 90 (when counted from left to right).
Note: By unchecking all and then checking certain members you can limit the length of the list when you want to inspect only those members.

3. "Test towards rest of collection". In this case each member is removed from the collection before a reference is created out of the rest of the collection and that member is compared to the reference. When there are few members in a collection this is an important tool for analysis.
Note: Also this command can be run in block checking mode. The block length and the block distance parameters for lists are set by the command Settings/Options for normalization of ring widths and for matching

4. "Show time lines" which gives you a plot showing how the samples overlap each other over time. Any zero values within a sample is marked with an "x" in the time line.

Collection reports can be saved in a file. You can copy (Ctrl-C) selected contents of a report and paste (Ctrl-V) it into a document.

Time line plots can be sent to the printer and then cut and taped together in a long strip.

Note: The "Create mean value sample" button operates only on checked members of a collection. This way it is easy to exclude a member from the reference curve of a collection to e.g. compare that member visually to the rest of the collection. To do this, uncheck that member. Click on the "Create mean value sample"-button. Double-click on the unchecked member to open it and click for a correlation analysis in its window.


Reviewing reference curves


With the automatic mechanism for adding best members to a collection in CDendro, you can quite soon rebuild an independent reference curve from its components.

An independent or freestanding curve is a curve
1. where the cutting year of the youngest sample is known and
2. where all samples date each other by matching each other well and by
3. overlapping each other in a reasonable way.
Though such a curve is freestanding, it should be crossdated towards curves from other places to check that there are no missing or extra years.

Other (published) curves which are not independent are those with e.g. a block of zero-values somewhere in the middle. Or with a younger part overlapping the older part with 30 years and a correlation coefficient of 0.14. An example of a zero block curve is the ITRDB file swed023w.rwl from Jämtland in Sweden. It covers the time 1827 - 1107 but has 25 years without values around 1327. The older part of this curve is no doubt dated towards a reference curve from a nearby place. Also the younger part of the curve must have been dated that way.

This creates a methodological problem: If many curves depend on the dating of other curves which depend on the dating of yet another curves we may run into trouble if an error is introduced somewhere on the way. This is especially true the longer back in time we are dating.

Therefore, it is essential to document how a curve is created and dated and on what material it is based and how samples are related to each other. This is especially important when some samples are short or has not so good correlation to other samples in the curve.


Building a reference curve by seeding with a known curve

If you already have a long reference curve from a nearby place, you can use that curve as a basis for a new curve.

Start with a cross correlation analysis of your sample data and start building a curve with the "Add best members to target collection" button. When you have a number of overlapping samples in your new collection, check its correlation towards that curve from the nearby place. If there is a strong correlation between the two curves, you can use that nearby curve as a "seed" or starting point when building on.

Save your nearby reference curve as a normalized data file (.d12) or a .wid file and add that file to your new collection. Then build on. When you are ready, you can easily remove the "nearby curve" either by unchecking that member or by deleting it. Using a seeded curve makes you work much faster than otherwise. Though if  you have any errors in that nearby curve, they will surely be transmitted to your new curve - so be careful!

One more word of caution: Climate may change over years and there may be cases where two places have the same climate during one period, but not during another. Especially when going far back in time, be very careful and validate (and document) what you are doing. Guessing is not a tool to use!

Note: When trying to rebuild an already existing curve, it may be suitable to have the "Tell bad dating" checkbox checked. On the other hand, when you are building from your new own samples which surely are not already dated (or if dated, you know their dating is wrong) , the box should be unchecked to avoid unnecessary warnings.



Copyright © 2011, Cybis Elektronik & Data AB, www.cybis.se