Length of melting season

[SteveMDFP]

Ice Cool Kim, et al:

I'd guess you'd call this approach systems analysis.  Mostly over my head, but I think I can see the gist of where you're going.  A thought to add to your words:

"I'm thinking that there may be a resonant frequency in ice physics that is responding to AO patterns. (Maybe it's the other way around)."

My hunch would be the other way around.  In the arctic, open water tends to associate with low-pressure systems (at least in the colder months).  These low-pressure systems mix cold upper-atmosphere temps with surface air and waters, cooling them, tending to permit freezing.  Essentially, a short-term regional negative feedback on surface ice. 

Now, increased surface currents are also so generated in these areas, which will tend to promote mixing in the upper tens of meters of water, potentially enhancing some mixing of warmer, deeper waters and locally reducing the halocline gradient, which may promote more release of warmth to the surface, a slower positive feedback to ice melt in the same regions.
 
I imagine interplay between a faster negative feedback and a slower positive feedback might produce the oscillations observed.

I'm pretty fuzzy, however, on how you'd actually test this hypothesized mechanism against measured data.

[Ice Cool Kim]

Thanks Steve,

systems analysis is exactly what it's about.

It may make more sense ocean driving atmosphere but I suspect there must be a fair amount of interplay.

Your comment about low pressure feedback is useful. There is a strong short feedback that seems to be recognised as 1y anti-correlation in the literature.

The apparent full-wave form has me a bit puzzled, I could imagine a case for half wave (horizon) but can't think of anything likely to be sensitive to magnitude but not sign. Yet the form looks pretty clear. I doubt that it's coincidental.

Maybe it's a processing artefact, I'll have to look into that possibility , though it seems unlike.

It could be an aliasing effect due to the very low sampling rate ( 6 months) implicit in just looking twice a year. My longest filter was 30 days which is approx same cut-off as a 60 running mean so the data is not fully filtered for aliasing effects.

Of course if I do filter it sufficiently there'll be no melt season left to look at

[frankendoodle]

The one good thing about this melt season will be seeing what conditions are like when the sea ice is nearly all the first year thin stuff. Approximately 80% of the ice from March is less than a year old. We will get an idea of what a winter recovery looks like after an ice free summer and what a melt season looks like with ice that is only months old (not years). We will see how large a factor wind and sea currents play in transporting the ice, the major factor of Antarctic sea ice change.

[Ice Cool Kim]

In a month or so I'll have another point on my graph for the newly completed freeze period. That should be interesting to see if it bounces back up or continues down.

Previous form would indicate it goes back up to about 0.495, I've estimated.  But it will be interesting to see whether that holds up or the post-2007 shortening continues.

[Ice Cool Kim]

In view of the close match between melt and (1-freeze) since 2000, I'm doubtful that the interleaving is correct.

Here I have taken the average of each years melting and (1_freezing) periods centred on June and December.

There is still a short period from 1996-99 that looks out of phase but this even shows up as anomalous on Chris' simple one day spot min/max method. That feature is persistent and seems to be real.

As does the difference in 2012.

Other than that, there seems to be a clear correlation with AO across the whole record. Notably following quite closely the abrupt changes around 1987 and 2007. 

[ChrisReynolds]

If looking for periodicities I typically have thousands of cycles of data available, and new cycles adding to that number. If I want to find periodic features I use a spectrum analyser, if that's not available I can use a scope triggered from a signal generator and look for stationary features as I change the sig gen's frequency.

However here we have 35 cycles.

So any periodicities I've found that don't have a ready physical explanation remain loose ends looking for a physical mechanism to tie them to.

For example in "Large Decadal Decline of the Arctic Multiyear Ice Cover" (2011) Comiso notes a possible 8 to 9 year cycle in MYI extent/area. I couldn't see a mechanistic explanation for this, when Kevin O'Neill emailed Dr Comiso he couldn't, he suggested it could be due to a process analogous to the Antarctic Circumpolar Wave (period ~4yrs). In the end we couldn't get to the bottom of the matter. So it remains a loose end looking for a mechanism, and the most parsimonious explanation is that it isn't really a periodic signal but is an outcome of many processes in a short dataset acting to create the appearance of a periodic signal.

Another example would be in the 1990s when some researchers viewed the changes in the Arctic as being natural, with the 1930s and 1990s declines in ice and warming being two peaks of that cycle. 20 years later and nobody serious buys that interpretation.

[Ice Cool Kim]

Yes, it is obviously very difficult to detect any periodicity with the amount and crudity of the data available. But by the same reasoning, not being able to clearly detect them does not justify concluding the signal is "stochastic.".

The year to year alternation is clear. Also between 1990 and 2005 there is a two up - two down repetition (circa 4 years period). The way it breaks at each end could be indicative of the presence of another close periodicity, perhaps the 5.38 I found a suggestion of.  I will have to try fitting a model to see whether the idea actually fits in reality.

There are enough points in AO to do an FFT but since it is a derived function I'm not sure how meaningful the result would be. Similar processing to derive PDO really does not represent the 1974-75 flip in the SST data all that well.  I'm a little suspicious of these techniques, though seeing the degree of correlation here shows they do have some physical meaning.

I think the plot above shows that my processing is pulling something valid out of the ice data which is interesting.  This would seem to justify my approach of filtering out the short term 'noise' to get a better indication of melting / freezing periods and the value of using both periods. For some reason all the papers I've seen so far only examine melting period. I think this shows that neither F or M is more indicative on its own but the average is better than either one.

Effectively this is doubling the sampling frequency. With one point per year this has to be advantageous.

[ChrisReynolds]

Separating out melt and freeze as distinct but mutually related periods is a reasonable thing to do.

Sorry, but my position is that unless you have a) mechanism, or b) a cycle that's very unlikely to be stochastic, or c) cause to suspect mechanism even if you can't define it. The default position must be that it is interesting, may be worth more investigation, but should be presumed to be the outcome of random (stochastic) processes. If you had a longer series that default position could be dispensed with as it would be covered by option 'b'. If you are using an annual or biennial cycle I'd suggest 'b' is the best position for that. But for longer cycles, certainly around 9 years (e.g. Comiso) I'm not convinced it would be so unusual that 'b' would cover it. So it remains interesting etc etc

Regards 'c' - for example an 11 year cycle in phase with the solar cycle (or with a near constant phase offset) is very likely to be related to the solar cycle, despite the small number of cycles in a 35 year period).

[Ice Cool Kim]

 

The default position must be that it is interesting, may be worth more investigation, but should be presumed to be the outcome of random (stochastic) processes.

I generally agree with your above comments but this sentence I see as wrong.

There is no reason to assume any result.

This reminds me of exo-planets. At school we were told quite clearly that there were no planets outside the solar system. We were alone and unique.

The fact that we would not have been able to see if they did exist did not seem to matter. They were not there.

Now we have better resolution we have found hundreds already.

The fact the we do not have the length and resolution in this data to clearly detect any cyclic signals is in no way a justification for assuming there aren't any.

Assuming randomness is no more justified than assuming periodicity.

In a system full of interacting +ve and -ve feedbacks where the major fluctuation is hard period oscillation , I would have though assuming randomness was highly questionable.

[ChrisReynolds]

Assuming randomness (no structure) should be the default position, except in the three cases I outline.

To adopt Russell's teapot analogy:

If I were to suggest that between the Earth and Mars there is a china teapot revolving about the sun in an elliptical orbit, nobody would be able to disprove my assertion provided I were careful to add that the teapot is too small to be revealed even by our most powerful telescopes. But if I were to go on to say that, since my assertion cannot be disproved, it is intolerable presumption on the part of human reason to doubt it, I should rightly be thought to be talking nonsense

i.e. we cannot go around suspecting periodic behaviour unless we have evidence, especially in a dataset as short as 35 years.

If you can't show that the apparent periodicity fits the three criteria I outlined, the default position must be that it is an accidental outcome of stochastic process, if it is due to interaction or modulation of other periodic processes this must be shown, although this latter point is a possibility I unintentionally omitted from my previous comment.

But our default position cannot be that there are true periodic processes everywhere. Arguably there is but one true periodic process - the Earth's relationship with the sun. Aside from that many processes claimed to be periodic are in fact pseudo-periodic with a stochastic element, e.g. ENSO. If you want to show they're real (like the teapot), you must demonstrate that, the default position must be that they are not there (like the teapot).

And yes this applies to exo-planets, until there is evidence of their existence, or mechanistically based reason to suppose they exist, setting off on a one-way mission to colonise one is idiotic. Just because they are then found does not disprove the principle.

If fairies are found to exist that does not prove the existence of god, ghosts, or indeed teapots orbiting the sun.

[Ice Cool Kim]

teapot:

But if I were to go on to say that, since my assertion cannot be disproved, it is intolerable presumption on the part of human reason to doubt it, I should rightly be thought to be talking nonsense

It's a clever game of words to arrive a the desired conclusion.  Clearly it is not "intolerable presumption " to doubt, However, it does not follow that it is not "intolerable presumption " to assume there are no teapots or similar sized objects that have not been detected.

Stochastic is just a fancy word for a process that is too complex to be explained by the current level of observation or understanding.  Or at least is often misused in that sense.

The only conclusion that can be made from a lack of observations is that you do not have sufficient information to make a judgement about the system.

And yes this applies to exo-planets, until there is evidence of their existence, or mechanistically based reason to suppose they exist, setting off on a one-way mission to colonise one is idiotic. Just because they are then found does not disprove the principle.

A ridiculous misrepresentation of my point. I was not suggesting a one-way mission. I was saying making definitive statements about the existence of something based only on a total lack of knowledge on which to base a decision is illogical.

[Ice Cool Kim]

If you are using an annual or biennial cycle I'd suggest 'b' is the best position for that. But for longer cycles, certainly around 9 years (e.g. Comiso) I'm not convinced it would be so unusual that 'b' would cover it. So it remains interesting etc etc

Regards 'c' - for example an 11 year cycle in phase with the solar cycle (or with a near constant phase offset) is very likely to be related to the solar cycle, despite the small number of cycles in a 35 year period).

I would have expected to polar regions to be particularly susceptible to solar fluctuations but the sample is too short to be that much use. Solar is also a lot more complex than 11y so maybe we're not looking in the right place or not recognising a signal this is there.

If anything I would suggest a circa 15y repetition but two cycles is insufficient to have any confidence, it could indeed be coincidental or combinations of something else.

[Jim Williams]

The default position must be that it is interesting, may be worth more investigation, but should be presumed to be the outcome of random (stochastic) processes.

I generally agree with your above comments but this sentence I see as wrong.

There is no reason to assume any result.

Occam's Razor: "Entities should not be multiplied unnecessarily."

The description which fits the available data and proposes the fewest independent actors is to be preferred.  Until the data requires it there is no good reason to assume there is anything there.  So until the signal is demonstrated to be there you shouldn't attempt to explain the potential signal.

[ChrisReynolds]

Yes, but if you can't demonstrate the applicability of either a, b or c as I outlined then you can't claim periodicity.

[Ice Cool Kim]

Yes, but if you can't demonstrate the applicability of either a, b or c as I outlined then you can't claim periodicity.

agreed. Neither can you claim anything else.

In the face of a lack of information, you don't assume stochasicity , you conclude you do not have enough information.

enough philosophy for me for one day , back to the data.

[Ice Cool Kim]

I've fixed the problematic oddity in the Antarctic and corrected a scaling error in the filter that may have had some small effect on Arctic too so here's updated Arctic plot.

As before longer filters very similar but with a little less variation.

The correlation with AO is now quite striking across the whole period of the satellite record.

It is a fair bit higher in 1997-98 which may reflect the unusually strong El Nino, however the bumps are still in the right place.

This uses very latest data and seems to show that earlier reports of the correlation breaking down were premature.  (Most recent I saw on that used data ending in 2007).

I think I've got most of the bugs (never say all) out of my processing and the resulting correlation suggests my filtering approach is doing a good job of extracting longer term signal from the weather disturbance and the circa 14d pseudo periodicity.