Arctic ice data conflict, example of software tool usage

Moderator Tim Channon writes: This article is unusual for the Talkshop. I may for personal reasons be leaving the Talkshop, a consequence of recent serious troubles, why I have been quiet. I hope this proves untrue.
I want to get some items out in the world for others. From my point of view leave a legacy. I am releasing the Synth software, done rapidly so none of this is neat or complete.

This is a simple example of usage with perhaps some current interest. No answers are intended for more the whole idea is about gaining insight by you playing around with data and ideas.

Figure 1

According to published datasets the characteristics of sea ice changed around 2006/7. Dataset used here is NSIDC/NOAA D02135 monthly. ( if you try and decode the published data, read the notes, has offsets which need compensation, do not use as is)

Based on experience I think this change is in the data gathering, perhaps geometry, not actual ice change. In the past in various articles I’ve mentioned some of the dubious practices.

Only tools used here are text editor, Openoffice Calc for post processing and the Synth software. Done under W8.1 but could be just as well under Linux.You will be able to clone this, doesn’t take rocket science skills with a PC, hopefully discover more.

Figure 2

This is what happens if the tool is told to remove annual.
fs=12 annual=1 interp=nk origin=1978.875 temp.txt

temp is the casual filename I’ve used for the data, a single column of ascii in a text file. A couple of months of data are missing, marked as nk (software ignores these). There was satellite failure yet some published dataset have miraculously filled the gap, there is no data. (older post on my blog mentions the details)

Output files tem.met and temp.rem are plotted.

Notice the obvious problem 2007 onwards, why?

As a simple ploy I fed in subsets of the data, 1978.875 .. 2006.875 and 2006.875 to latest. Using the exact same start date and given the accuracy of the software computing the annual change is trivial. For annual the periods are literally locked.
fs=12 annual=1 interp=nk origin=1978.875 temp.txt (data excludes newest)
fs=12 annual=1 interp=nk origin=2006.875 temp1.txt

Putting this together, relatively complicated to do, easy once you know how.

Figure 3, this may or may not be accurate, look for yourself

Figure 1 shows annual for early and late, using both produces figure 3. The software give the amplitudes and phases.

How? The output .asc files where imported, time series created, result combined and subtracted from original data, producing the remainder. One twist. the offset is different so the late simulation uses the offset from the early data, otherwise there is a step.

Valid? Who knows. Why the data change? Needs finding out. I’ve posted I give up on sea ice after very detail analysis of daily data, been working on this stuff for years. Concluded the data is very dubious. Linear trend is probably human error.
Now a lot of data is being hidden from public. Why?

Can more be done with the above?

Of course. Lets take the created data used for figure 3 as the input for more processing, making sure the invalid data is marked appropriately.(the use of an irregular timebase, extra data column could be used instead but isn’t)

fs=12 gens=3 interp=nk origin=1978.875 temp3.txt

This time in normal mode, not annual. Use 3 terms (generators).

Figure 4, matched, output allowed to extend to 2020.

There is no party trick here, you will be able to clone these results, if you think there is cheating all the C sources are provided. Here the software is stepping outside of the capability of DFT, data is too short in time. This software will match shape, even a tiny part of a longer curve. This result is ambiguous, not to be trusted, why you have a brain but if this leads to insight, good.

Is there more periodic kind of detail in the data? Not in my opinion of any use, you could look. A rough rule is KISS, keep it simple stupid. (there is no definite limit on the term count, speed will limit things first, I’ve tried 100+, program uses little memory, mostly a few megabytes)

Now lets do something which might open some eyes.

Figure 5. Output timebase has been altered. Very easy to do this.

Changed cell C17 to 1950 and C16 to 4, output start date and 4 samples a year, aliasing is not a problem in this case. Data from figure 3 is plotted to its own separate timescale.

Here I am illustrating the effect of inferring waves. The waves periods are irrationally related so the phase varies, hence might add or subtract, leading to the greater shape complexity.

The waves here are approximations.

There is a long period wave and a common one in climatic data, 21 years or so. In my opinion probably of solar magnetic origin.

Following is a guess.

Figure 6, plot from a different simple example, 3 term again.

Solar heat enters mostly in the middle latitudes, flows north and south to polar regions.

If we flip this or the ice data upside down there is roughly speaking a fit after a delay of roughly a decade.

I expect a refined work could be done, this example is supposed to be leading to ideas, insight, not formal results.

Please ask questions whilst you can.

Putting this in the published examples.

Software: Talkshop top menu, Portal, Tim’s software

The texts, documents are poorly written, I wish this was not so. Contribute and help if you can.

Stupid questions are fine, in private if you need to. No question is stupid, nor the person asking. Maybe I am in yoda mode

Post by Tim