Other indices were down; NCEP/NCAR by 0.056°C. As often recently, polar variations played a big part; both poles were quite warm, and GISS and TempLS mesh respond to this. I expect NOAA and HADCRUT will show larger reductions.
I'll show the map comparisons below the fold. The updated comparison plots with 1998 are here
Here is the GISS map for the month
The Moyhu spherical harmonics map is here:
Basically the same features, though GISS shows more emphatic warming in the Arctic.
It's important to note these two comments
ReplyDeletePeter Green November 4, 2016 at 11:51 AM
"Couch it in such as way as to make it very clear that it enacting the will of the people..."
and
Albatross November 11, 2016 at 2:05 PM
"Nick,
Great tool! Is there any way to save a file of the animation? Your tool nicely shows the loss of stations reporting in Canada, and it would be great to have a movie showing that to share. Tks."
Rubber duckie to anyone that gets the significance of this
It is interesting to compare these two charts with the data obtained by processing UAH's 2.5° gridded TLT data, by using Nick's globe viewer:
ReplyDeletehttps://s3-us-west-1.amazonaws.com/www.moyhu.org/maps/webgl/grid.html
and enntering as data the grid values for october 2016:
http://m.uploadedit.com/ba3s/1479333780466.txt
To compare this in turn with UAH's own representation
https://i2.wp.com/www.nsstc.uah.edu/climate/2016/October/OCTOBER%202016.png
is a bit difficult as Nick uses a different coloring mechanism.
I have the impression of an increasing similarity between TempLS and UAH's gridded data in the arctic regions.
GISS arctic is a lot warmer then the TempLS arctic. Maybe we can look at the NCEP Global Forecast System and Reanalysis data from here
ReplyDeletehttp://www.karstenhaustein.com/climate.php
For October 2016 (down on this site) the Arctic is even about 1°C warmer then the GISS. So I conclude the TEMPLS is too cold there, and even GISS may underestimate the temperature over the arctic ocean.
TempLS too cold? It may well be. The more detailed map is here. A large part of the Arctic ocean is covered by just two abutting triangles. Three of the four corners are relatively cool. But there really isn't better data - GISS would have done similar. Reanalysis may be the best here.
DeleteI had looked on the GISS data with 250km smoothing and there are many warm station near the arctic ocean. But your TempLS grid goes far more in the north using many data points in the ocean.
DeleteMaybe that's the reason. Your are using ocean temperature, where the ocean is not covered by sea ice. The ocean temperature anomaly is close to zero there, but in the reference period 1951-1980 a large part of this was covered by sea ice in October. So the air temperature was far below 0°C there, and the air temperature anomaly should be large there. May be it would be a good idea not to use the ocean anomaly at points that are free of sea ice now but was at least partial covered by sea ice in the reference period 1951-1980.
Yes, there is a risk of having too many sea points. That is a cool anomaly here, since water can't vary much, while the air is anomalously warm. It's a problem at this time of year, when there is still a fringe of water above the North coasts. Once it is clear that that has frozen, I exclude the data, though the SST sources still have it as -1.8°C.
DeleteDoes than mean you exclude the SST data only, if it is frozen in 2016?
DeleteOr do you exclude it too if it is not frozen in 2016 but frozen in a year of the reference period 1951-1980?
Uli,
DeleteThese are tricky issues. SST datasets gives a temperature of -1.8°C (freezing point seawater) for iced areas. I view that as an irrelevant temperature for climate, and generally enter it as NA. So if that is the current reading, then yes, it is eliminated. I don't myself use an anomaly base period, since I use least squares fitting. But there is a minimum record requirement (for the month). This creates a small bias, because fringe areas that are rapidly warming are not used because of lack of data in colder days. But that is usually just for a few fringe months. Also messy is where just part of the month is frozen, so the temp showing isn't -1.8.
It occurred to me while writing this that it is really a censored data problem (skeptics sometimes get excited by that phrase). I'll look at that.
Nick,
Delete"I don't myself use an anomaly base period, since I use least squares fitting."
what does then the anomaly mean? I'm confused. Absolute temperature could it not be ...
"This creates a small bias, because fringe areas that are rapidly warming are not used because of lack of data in colder days."
I think the problem is opposite. your are using too much SST data in areas that was ice covered in previous years. If I look at your data on the globe with show stations and mesh here:
https://moyhu.blogspot.de/p/blog-page_24.html
I see that the Arctic has got a very low anomaly despite of very large SAT anomaly in the NCEP data here
http://www.karstenhaustein.com/climate.php
If I look at the mesh, the regular ocean mesh on the circles around the North Pole, the points there have a very low anomaly compared the very warm land stations. But If I compare it to an October in 1976 f.e. there are far less ocean data used, almost no one in the arctic circle. I suppose it is due the presence of sea ice there.
But that mean to me that the TempLS algorithm does evaluate the SST on now sea ice free but ones ice covered ocean areas as near zero and interpolates this even over the yet ice covered North Pole area. Would this not introduce a really big bias?
Uli,
Delete"what does then the anomaly mean?"
I've written quite a lot about anomaly over the years. Basically it is the difference (for a given measurement at a given time) between what is observed and what might have been expected a priori. Statistically, you want anomalies to have similar distributions, especially mean, for averaging. So you try to make the mean zero. That is very different from absolute temperature.
I use a statistical model fitted by least squares - my most recent description is here.
"Would this not introduce a really big bias?"
As I said in my previous response, dealing with the sea ice/open sea transition is difficult. Yes, I think mine may be biased there to SST. The effect of that is to damp movements, since SST can't change (for warm or cold) by nearly as much as polar air. As with most indices, a big reason for this bias is lack of air stations. TempLS mesh upweights those few land stations by letting them be the vertices of large triangles, whose area decides their weight. That is much better than leaving them unweighted; in effect the air stations are used to approximate large areas of air over sea ice, which is not ideal, but there is nothing better, and much that is worse.
And so it usually does work out that much but not all of the sea ice is interpolated by land. But at certain times like now, it doesn't work out so well. The sea is still quite far north, and the mesh is likely to make more use of SST locations than we might wish. Last month, this combined with the fact that some of the air stations that were upweighted by the mesh were in fact among the cooler ones, led to what is I think probably a too cool outcome.
TempLS is a method, and I don't want to intervene to impose ad hoc judgments. You'll notice that NOAA is out now, and is down by 0.12°C. They do much less Arctic reweighting than TempLS.
Nick, I'm not sure, if I explained good. (NOAA has no data in the arctic ocean. And miss this region too. I'm not sure if it's not better that get a too cold anomaly there.)
DeleteI think there may be a big problem, not only in TempLS but in all other temperature indices too, that using SST as proxy for SAT over ocean. I think they miss and will miss a large part of arctic warming.
Your TempLS method and visualization has helped me a lot to recognize this problem.
Let's me try to explain again.
There is a large difference in the temperature over the arctic ocean between GISS and TempLS as seen in the figures above. TempLS is cold blue, but GISS is hot red. According to the NCEP data even GISS is biased low, but not so much as TempLS. TempLS shoes a blue cold anomaly over the arctic ocean despite a record low sea ice cover.
I asked me how this could be and looked at you triangular grid for interpolation. This grid is good because it allows to find the source of the cold bias.
I don't think that the interpolation over the current sea ice is the main problem. (But it is a problem too).
I don't think TempLS is especial bad. NOAA, HadCruT, all have this problem. I don't think the reason the lack of air stations, but the use of too much (or wrongly use of) SST data in wrong places.
I now think almost all estimates for global SAT temperature that use SST are suffer a cold bias over regions, that are now ice free (or have reduced sea ice cover) but was covered by sea ice in the past. GISS is less effected (up to now, but see below), because it uses no SST near land stations even over ocean and it interpolates over sea ice by land stations only, not using SST for interpolation there. Leaving out this regions as unknown is no option, because this too would cause in a cold bias for global temperatures because this regions warm much faster than the global average.
End first part, because longer than 4096 Characters.
Let's consider a hypothetical example:
DeleteOn the start of October 2016 all sea ice in the northern hemisphere vanishes completely and the SST there rose slightly above the threshold for open water, slightly above -1.8°C, so that all that count as ice free and all the SST are used in the arctic ocean. The SAT over the open arctic ocean is assumed to have now the same temperature as the sea.
The land stations are almost not affected, because they lie all close to open water, as the sea ice reached not the coast of the arctic ocean yet.
This may not so hypothetical in the future, if the arctic sea ice completely vanishes.
What would be the SAT above the arctic ocean? The absolute SAT is about -1.8°C or so. According to the Reanalysis data the temperature in the 20th century with lot of sea ice there was about -20°C.
So the SAT anomalie, that a good method should give in this case over the arctic ocean, is about +18°C or so.
But I suppose that almost all temperature indices will give either no data there, because there is now no sea ice there but it was ice covered in the reference period, so they don't use SST (which would only available now). Or they use SST because the algorithm say that there is no sea ice now. Then the result may be an anomaly close to +0°C over the arctic ocean. Both may result in a global anomaly which may be far too cold. Also the GISS (Ts+SST) algorithm does not help here, because there is no sea ice, where land stations SAT anomaly can be interpolated. GISS Ts may be better but if the SAT chance for the land stations near the Arctic coast is much less then the SAT chance form ice covered to open ocean then it too may underestimate the warming there.
What would the TempLS result in this hypothetical case? You can check. I guess close to +0°C.
If the algorithms switching (maybe unintentional) from SAT to SST anomaly in areas ones covered by sea ice, the may miss the large temperature chance there and may miss a warming of the global SAT +0.538°C if ones all sea ice is gone. The difference is the difference in the absolute temperature difference using air and water temperatures from BEST.
I think that's a problem. May be it may account for a large chunk of the difference between "measured" and projected SAT temperature anomaly.
Your TempLS method and visualization has helped me a lot to recognize that the problem lies in the areas that currently ice free but was covered with sea ice in the past.
What do you think?
Nick,
ReplyDeleteAny comments on this thread regarding where 2016 will end up finishing in comparison with 2015?
https://twitter.com/andyskuce/status/799067508154114048
RB,
DeleteI've updated the plots on my records post here. They mostly, for surface, indicate 2016 exceeding 2015 by about the same amount as 2015 exceeded 2014 - ie about 0.08°C. Some a little less, and the TLT measures quite close, relying on 2016 not emulating the late 1998 dip. I'll update that post when HADCRUT Oct is in.
Everyone is following the bouncing ENSO ball.
ReplyDeleteI posted this to the Azimuth Project forum today:
https://forum.azimuthproject.org/discussion/comment/15582/#Comment_15582
http://imageshack.com/a/img922/7710/G5NCF7.jpg
That bouncing ball is highly deterministic and likely has been following a strict schedule for centuries.
BEST l/o just dropped in. October is up 0.12 from Sept!! 0.913 vs 0.79
ReplyDeleteHas it more correctly picked up the polar warmth?
Olof,
ReplyDeleteY, GISS always a little bit too cool/warm when it comes to extreme anonmalies, because (dont know the source anymore) there algorithm kicked out some data, when its seems unrealistic to the data from before. It such extreme month like this, this will cause a systematical cool bias.
Look out Karstens Webpage for Anomalies, in times where GFS or Reanalysis show a extraordinary warm artic, GISS comes 1-2K cooler out then GFS or other would suggest
Nick - unrelated, the arctic NCEP temp lat-lon applet has an X-axis alignment problem. Tried a couple different browsers, same result in each (I normally use Chrome). Result can be viewed here ASIB Forum comment
ReplyDeleteKevin,
DeleteI can't see it. Do you mean the x-axis marked in days? It has data to day 328 (23 Nov) and that looks about right to me. There might be a one day leap year issue. I doubt if it is browser-dependent.
Nick. the data is starting to plot to the left of where 0 would be on the X-axis at what would amount to the first data point being day -25
Deletehttps://3.bp.blogspot.com/-V_BCRJhoA5w/WDh0JTf3FKI/AAAAAAAAAM0/P7earNWnZrwaDO3369iJF013c0oRG2NEACLcB/s1600/N70_80_pacific.JPG
Is a screen capture.
Another commenter on the forum used the applet and his screen capture shows the same X-axis misalignment.
A pixel count of the png from the applet shows data starting at column 58, but the X-axis doesn't even start until pixel 70.
Perhaps it's just because I expect the Y-axis to intersect at day 0 and the X-axis to extend all the way to day 0.
Kevin,
DeleteThanks for explaining. I don't think it's a misalignment - it just does show data before day 0, which is of course from last year. I think I originally allowed that because it gives some context in the first few days of the year. I could take it out, or maybe color it differently, but I actually think it does sometimes help.