September Arctic sea ice extent: 1935-2014

A couple of days ago a paper was published in the journal Revista de Climatología by Diablobanquisa, a Spanish blogger who regularly writes about Arctic sea ice on his blog. In this paper Diablobanquisa, together with his co-author, presents a new time series of September Arctic sea ice extent, including data for the Siberian sector. I think the work looks very solid and gives us an improved view of pre-satellite Arctic sea ice.

Below is the English abstract of the paper, preceded by the graph showing the results of the new time series (updated up to 2015):

A new time series of September
Arctic sea ice extent: 1935-2014

Abstract

Since 1979 satellite-borne passive microwave sensors have provided a continuous and consistent record of Arctic sea ice extent. This record shows a significant downward trend, particularly at September, when Arctic sea ice reaches its annual minimum. Records before 1979 exist, but are not consistent with the satellite record and have limited reliability, specially before 1953. We present a new time series of September Arctic sea ice extent from 1935 to 2014 that includes data for the Siberian sector (AARI operational charts) not used previously in the Arctic wide existing time series (Walsh, HadISST). The new record has been adjusted to be consistent with satellite data. The trend for 1935-2014 is -3.5% decade, while the trend for the satellite era is -13.3% decade. However, the trends since 1935 until early 1980s are positive and statistically significant. The trends turn negative in the 1990s, reaching statistical significance from 2006 onwards. The lowest annual minimum in the pre-satellite era is higher than any annual minimum after 2001.

Median September Arctic sea ice extent for the lowest decade of the pre-satellite era (1935-1944, left) and for the lowest decade of the satellite era (2005-2014, right):

– Download full text (pdf, in Spanish)

– Download extent data values (csv)

– Download gridded data (netcdf)

Methodology and several graphs comparing the results to previous work, can be found over on Diablobanquisa (Spanish version including conclusion here). There's also a separate thread over on the Arctic Sea Ice Forum.

Congratulations and thank you, Diablo + co-author!

Comments

[Rob Dekker]

September historic reconstructions are more difficult than August, because of the sparse observations available.

But since this thread started with Diablo's reconstruction for September 1935 - present, I owed a September reconstruction using my Match-and-Merge spatial/temporal fill-in algorithm over that period.

Note that I started with the Walsh reconstruction, then removed Walsh's spatial/temporal fill-in source, and I removed the Walsh&Johnson source, but let the other observations (such as the Navo yearbooks, the Dehn collection, AARI, DMI etc) stand. Then I let my match-and-merge algorithm fill it the rest. Here is the result :

Click for a larger image.

First thing to note is that the September reconstruction more closely follows Walsh reconstruction than for August. That's probably since September does not use Kelly fields, and this suggests that the August Kelly fields in Walsh reconstruction are really no good.

Second, I wanted to get some sort of "uncertainty" metric for these reconstructions, and the best I could come up with is to plot the difference between August and September. Remember that both months are based on independent observations, so if there is a wild difference between the months, then there is probably a large uncertainty in the reconstruction.

But as you can see, the difference between August and September is rather stable (hovers around 900 k km^2) and this gives confidence that the uncertainty in the August and September reconstruction is rather small (maybe in the order of a few hundred k km^2 SD).

Finally, let's compare the reconstructions.
Diablo, I do not have the numbers for Meier et al and your reconstruction, so could you please provide a 'final' overview plot of how our reconstructions of Arctic sea ice in September from 1935 to present day differ ? If you need my exact numbers for 1935-1978, please let me know and I can send them to you (if I know your email) or post them here.

[Rob Dekker]

Second plot here is for the entire 1850-present period, for the month of September :

Notice that with my fill-in algorithm, there appears a small uptrend from 1850 to about 1900. I'm not sure if that is real, but I suspect it is not. After all, with very sparse observations my algorithm tends to amplify the 'open water' area ('open-water bias' as explained earlier) and with the very sparse observations of the pre-1900 era, that effect may amplify.

Another indication that that 'up-trend' in the 19th century is not real is the August-minus-September graph plotted in blue : before 1900 the difference between August and September appears to go up, which is very unlikely to be realistic.

In all, for the full 1850-present period, I prefer my August reconstruction (presented earlier). There are more observations in August, and it results is a robust and realistic result, especially when you analyze individual years as I did in this thread.

[Rob Dekker]

Final note : Because of the "open water" bias of my fill-in algorithm, I consider my September reconstruction a lower bound for ice extent since 1850. Maybe the difference with Walsh' original which is (for pre-1900) more based on a fixed 'climatology', the difference between the graphs is an indication of the uncertainty in observation.

[Diablobanquisa]

Thank you for your work, Rob.

Diablo, I do not have the numbers for Meier et al and your reconstruction, so could you please provide a 'final' overview plot of how our reconstructions of Arctic sea ice in September from 1935 to present day differ ? If you need my exact numbers for 1935-1978, please let me know and I can send them to you (if I know your email) or post them here.

I have extracted your extent numbers from the graph you posted above. It's not exact, but I think it should be accurate enough for these comparison purposes. Just to check I did it right: your lowest pre-satellite value is 1941 around 6.42 M km^2? And the highest is 1963 with 8.78 M km^2?
(Alternatively, you can download our numbers from: https://doi.org/10.5281/zenodo.44758 and plot them yourself).

The graph below compares Walsh et al. 2016 (blue), my results (red) and yours (black), for September 1935-2013 (click for a larger version):

I think that the three time series look pretty consistent.

Regarding some of your comments:

First thing to note is that the September reconstruction more closely follows Walsh reconstruction than for August. That's probably since September does not use Kelly fields, and this suggests that the August Kelly fields in Walsh reconstruction are really no good.

I agree.

Another indication that that 'up-trend' in the 19th century is not real is the August-minus-September graph plotted in blue : before 1900 the difference between August and September appears to go up, which is very unlikely to be realistic.

I consider my September reconstruction a lower bound for ice extent since 1850. Maybe the difference with Walsh' original which is (for pre-1900) more based on a fixed 'climatology', the difference between the graphs is an indication of the uncertainty in observation.

I think that Walsh et al. and you have done a great work reconstructing the 1850-1934 period. However, I think that even August data before 1935 must be taken with a grain of salt. And I wouldn't put too much faith in any results before 1900: there are simply no enough direct observations available. I think the uncertainties and error margins before 1900 are huge.

[Diablobanquisa]

And this is the graph comparing with Meier et al. (click for a larger version):

[Diablobanquisa]

I've also extracted your August numbers and plotted them against your and our results for September (click for a larger version):

[Rob Dekker]

Thanks Diablo,
It is remarkable how closely the post-1935 September reconstructions match. Especially the difference between your and my reconstruction is surprising, since we used different methods, and had only AARI observations in common. Apologies for being so skeptical at the start of this thread.

Regarding the full 1850-present reconstruction, Walsh and his team deserve all the credit. All I did was take out the suspect Kelly fields and filled them in with a regional match-and-merge spatial/temporal fill-in algorithm. In that regard, I feel that my best contribution to this work is my August reconstruction :

My August reconstruction simply matches much better with the September numbers from Meiers, you and Walsh own numbers, so
if anything came out of this analysis, it is the hope that Walsh et al takes the Kelly fields source out of the next version of their historical Arctic sea ice extent reconstruction and replace it by their spatial/temporal fill-in algorithm.

I'm still disappointed that I have not been able to come up with a statistically sound method to determine "uncertainty" in the reconstruction, but if I find a method, I'll surely post it here.

All in all, it has been marvelous thread, and I appreciate your clarity of insight and attention to detail from you, Diablo. If you ever think of publishing another paper, please let me know and I'd be happy to contribute. Neven has my contact info.

[Diablobanquisa]

Thanks for everything, Rob, I have really enjoyed this thread too.

[Rob Dekker]

When you look at the images I posted, it will tell to upgrade the photobucket account to one that supports 3rd party hosting. Photobucket charges $400/year for that and I'm not going to pay that amount of money.

So, to look at the images I posted, right-click the image and open in a new window. Thanks !

[Jim Hunt]

I tried that Rob, but as a 3rd party I still only saw the "Please update your account" image.