Slater's thread

[Freegrass]

I've added my own projection. What do you think? All will depend on the weather of course, but I can see us approximating the average trend, while ending up a little lower than projected by Slater.

Lets see how far off I'll be in the end... 

Edited my drawing a little. I think this is more probable. But it's always a gamble of course...
I think we'll end up somewhere around 4.75. Where would that put us in the list?

I wasn't too far off... I just missed the rapid slowdown earlier on.

[El Cid]

Slater's model was pretty right once again yet many still look down on it

[Tor Bejnar]

Of course, Andrew Slater is looking down on his model from heaven. 
 
May his spirit continue to RIP.

[Renerpho]

I've been archiving the output of Slater's model for the best part of 2021 (you can find some of their daily updates on archive.is). Here's my own diagram of the data, including its predictions up to the end of the year.

On average, Slater's prediction was 0.21 million km² too high this year, but never deviating more than 0.77 million km² from the observed extent. That is considerably better than any other model I am aware of (including the predictions made in the freezing season thread on this forum ).

Since late October, their prediction closely follows the 2011-2020 average. We'll see if that comes true.

[Freegrass]

Did anyone else notice that the Slater model stopped working this year? So sad...

https://cires1.colorado.edu/~aslater/SEAICE/

[Renerpho]

Did anyone else notice that the Slater model stopped working this year? So sad...

https://cires1.colorado.edu/~aslater/SEAICE/

Yeah... Still checking regularly, just in case something happens.

Since regular updates stopped on 1 December 2021, the page has been updated exactly once (on 6 October 2022). I will continue to save regular (about weekly) snapshots at https://archive.is/cires1.colorado.edu. However, if the website hasn't been updated by May, I will abandon that.

[Steven]

I was thinking about the temperature graphs that used to be available on Andrew Slater's website:

https://web.archive.org/web/20171206065355/http://cires1.colorado.edu/~aslater/ARCTIC_TAIR/

His graphs were very useful, in my opinion much more so than the DMI 80N graph.

From the description on his website, I'm not sure which numeric data Slater was using. 

Anyway here is my first attempt to create something similar.  I used NCEP reanalysis data for surface temperature, which I'm averaging over a region in the Arctic Ocean very similar to the region Slater used.  The red parts in the image below show Slater's region:



Here is a quick summary of the results I get (see attached csv-file for details):



But there are several caveats for now:

1) I'm using an approximation of Slater's region in the Arctic Ocean.  But the NCEP data resolution is only 2.5 degrees for both latitude and longitude, which is too crude to include the channels in the Canadian Archipelago or exclude islands etc.  Moreover I'm using a crude weighting: NCEP data points are either included in the region or not.  This could be improved by using a smooth continuous weighting rather than just a binary yes/no choice.

2) I'm using surface temperature data for now.  I'd like to switch it to 925mb temperature, but there's a practical problem: the NCEP website puts all 17 layers (1000mb, 925mb, 850mb, etc) in a single huge file, which takes forever to download on this old laptop.  Will have to find a workaround for this.

3) NCEP reanalysis is an old product, there may be better alternatives available.  But I'm not sure where to find the numeric data for those.

4) Due to the above remarks, I haven't automated anything yet in the calculations (and maybe never will).  Might be continued...

[Steven]

2) I'm using surface temperature data for now.  I'd like to switch it to 925mb temperature, but there's a practical problem: the NCEP website puts all 17 layers (1000mb, 925mb, 850mb, etc) in a single huge file, which takes forever to download on this old laptop.  Will have to find a workaround for this.

Update: I was able to download the files anyway, after rebooting the laptop, and reprocessed all calculations to use the temperature at 925 millibars instead of surface temperature.  Here are the updated results:



Moreover the results for the year 2016 match pretty well with the old graphs on Slater's site, despite him using a different dataset.

[gerontocrat]

So much better than the DMI 80 measurements.

[Steven]

Thanks, gero.

Meanwhile I automated the graphs and made them available here:

https://sites.google.com/view/arctic-sea-ice/home/temperature

It will be updated frequently (probably daily, at least if the automatic updates run smoothly given that they include a huge file download).  Maybe I'll make some further refinements to the region over which the temperature is averaged.  But that would probably only affect the results by at most a few percent, so it's low priority.

[John_the_Younger]

Yeah!  A-Team couldn't stand DMI-80N but suggested no practical replacement (that I know of).  You've done it, Stephen: congratulations!

I find it interesting that the after-freeze-starts temperature spike shown by DMI is not apparent in the 925mb temp data.

[uniquorn]

Thanks Steven, here's a gif of the individual years and a montage.

[Steven]

Yeah!  A-Team couldn't stand DMI-80N but suggested no practical replacement (that I know of).  You've done it, Stephen: congratulations!

I find it interesting that the after-freeze-starts temperature spike shown by DMI is not apparent in the 925mb temp data.

The main credit goes to Slater, I just tried to reproduce what he did.  Zack Labe also has some Arctic temperature graphs, I think those are updated monthly and he averages over the regions >80°N or >70°N.

If the after-freezing plateau is caused by the ocean venting heat in the late melt season and early freezing season, I guess this venting is mainly affecting the surface and the impact will be less pronounced at higher levels in the atmosphere.

Thanks Steven, here's a gif of the individual years and a montage.

Thanks. Just a small suggestion (feel free to ignore): if you add a graph for the 1980-2010 average (or some other fixed period) to each frame in the collage, it would give a reference frame and would help to identify colder or warmer-than-normal periods for each year.

[uniquorn]

Thanks. Just a small suggestion (feel free to ignore): if you add a graph for the 1980-2010 average (or some other fixed period) to each frame in the collage, it would give a reference frame and would help to identify colder or warmer-than-normal periods for each year.

good idea

[be cause]

I shall slip off Neven's gag long enough to say well done Steven !

[oren]

Indeed, very well done. An important addition to our arsenal of charts.

[Steven]

Thanks all.  A few more thoughts:

1) I'm thinking to add regional temperature data, by dividing the big region into smaller subregions.  Given the coarse resolution, it may be hard to get accurate data for the 13 traditional regions of the Arctic Ocean.  But it should be feasible to do it for a few larger regions, for example something like the 4 quadrants of the Arctic Ocean that Neven used in the past (i.e., the Siberian, Pacific, Canadian and Atlantic sectors).  I haven't made up my mind yet about the choice of regions...

2) Maybe it could be interesting to try to reproduce Slater's sea ice prediction model too?  It seems doable, although I'm not sure if there is enough information available to figure out all the details of his algorithm.

[oren]

I doubt reproducing Slater's model would be all that useful, but then I never thought it had such a good predictive value.
OTOH I for one would appreciate some regional temperature values. Regions of interest could be:
Siberian: ESS+Laptev+Kara.
Pacific/American: Chukchi+Beaufort.
CAA - an interesting region by itself with highly variable weather.
Central - just the CAB.

Of course this depends on the amount of effort you want to invest, and on the feasibility of getting this granularity of data.

[Steven]

I doubt reproducing Slater's model would be all that useful, but then I never thought it had such a good predictive value.

The model seems to have a decent track record for predictions.  Keep in mind that the main goal is to predict the September monthly average extent, not the daily fluctuations.  The 30-day average gets rid of much of the fluctuations. 

But I'm not sure if the extent predictions from Slater's model would be much different from using a simple linear regression with NSIDC sea ice area as predictor variable.  The main difference is that Slater's model does the predictions pixel-wise, using gridded NSIDC concentration data, and then it aggregates those pixel-wise results to get an overall prediction for extent.  So the model basically just reverses the normal order of the operations: it does the regressions first and the aggregating last, instead of the other way around. 
   Anyway, I'm not going to spend time on this for now.

Regarding the regional data: If I have some time the next days, I'll have a closer look at the coastline granularity and resolution issues and test a few ideas.

[uniquorn]

Thanks all.  A few more thoughts:

1) I'm thinking to add regional temperature data, by dividing the big region into smaller subregions.  Given the coarse resolution, it may be hard to get accurate data for the 13 traditional regions of the Arctic Ocean.  But it should be feasible to do it for a few larger regions, for example something like the 4 quadrants of the Arctic Ocean that Neven used in the past (i.e., the Siberian, Pacific, Canadian and Atlantic sectors).  I haven't made up my mind yet about the choice of regions...

2) Maybe it could be interesting to try to reproduce Slater's sea ice prediction model too?  It seems doable, although I'm not sure if there is enough information available to figure out all the details of his algorithm.

When thinking about regional temperature data it may be worth including the median of recent minimums which might be of interest for the 'When will the Arctic Go Ice Free?' thread. Maybe it's too close to the CAB area to be worth it though.

sic-lead v110 minimums 2012-2023
avg and median of minimums 2012-2023
std and min of minimums 2012-2023
avg freezing season temperature 79/80-22/23

[kassy]

Interesting graphics, thanks.

[Steven]

I made a few changes to deal with the coarse resolution of the NCEP data.  In the old version, each NCEP grid cell was either fully included (100%) or fully excluded (0%) in the calculation.  In the new calculation, each NCEP grid cell is weighted by the percentage of ocean pixels (vs land pixels) it contains.  So the data points near coastlines may be weighted by 40% or 60% etc depending on how many ocean pixels (vs land pixels) they have nearby.

This allows to include the CAA, which was excluded in the old version.

It turns out the old and new algorithm give very similar results: compare the blue and red lines in the graph below for the year 2016.  And both of them agree pretty well with the old Slater graphs, see the green line in the graph.  There's an offset of about 1.5°C since the NCEP reanalysis data seem to be a bit warmer than the data Slater used.  But the patterns of peaks and throughs are very similar, which is what matters.

[Steven]

For the regional data, I decided to use the 13 traditional regions of the Arctic Ocean after all.  The left image below shows the template regional definition that I used, and the right image is my implementation of it using the NCEP grid cells.  The right image was generated automatically by a piece of code I wrote.



Here is a more detailed view, showing the individual NCEP grid cells.  As mentioned, each grid cell is weighted in the calculations by the percentage of ocean pixels it contains, hence cells that cover mostly land have little impact in the calculations.



I ran the numbers and put the regional graphs and daily/monthly data here: 

https://sites.google.com/view/arctic-sea-ice/home/temperature

The graphs for the smaller regions look a bit noisy and hard to read (example for Beaufort Sea attached below).  Maybe I should try to smooth the data somewhat, or change the colors, or merge several regions together as original planned.

[uniquorn]

<>
The graphs for the smaller regions look a bit noisy and hard to read (example for Beaufort Sea attached below).  Maybe I should try to smooth the data somewhat, or change the colors, or merge several regions together as original planned.

Thanks again Steven
They look fine showing one year at a time, maybe show the latest 3 years and the decadal averages, or make the charts much bigger.

[kassy]

The first options cleans it up nicely.

[binntho]

Steven, thanks for your excellent work. But I have a question - some of the regions are far outside the Arctic circle or the Arctic proper. Are they all included in this graph? If so, the graph is misnamed, it is not in fact the "Arctic Ocean" but rather "oceanic waters in the Northern Hemisphere that can potentially contain sea ice during winter months".



Also I never have understood why people would want to use 925mb temperatures rather than surface temperatures. For me as an amateur, I have no idea what the temps at 925mb mean for the actual temperatures affecting the ice.

[oren]

Thanks a lot Steven. When I have time I will try and play with the data.
I think presenting a year and the average (or decadal averages) can be clearer, because of the inherent volatility of the numbers.

[Steven]

<>
The graphs for the smaller regions look a bit noisy and hard to read (example for Beaufort Sea attached below).  Maybe I should try to smooth the data somewhat, or change the colors, or merge several regions together as original planned.

Thanks again Steven
They look fine showing one year at a time, maybe show the latest 3 years and the decadal averages, or make the charts much bigger.

Indeed, 14 years was probably way too much for such noisy data.  4 lines seems to be the limit of what is readable without looking too overloaded.  I changed it like this for the regional graphs:



Will try to add regional temperature graphs for previous years somewhere on the website, but Google Sites requires uploading all images manually which would be impractical for 13 regions and 43 years = 559 graphs.  Even for only the last 10 years it would be too much, unless the graphs for previous years are combined into a single collage image like you did before.

But I have a question - some of the regions are far outside the Arctic circle or the Arctic proper. Are they all included in this graph? If so, the graph is misnamed, it is not in fact the "Arctic Ocean" but rather "oceanic waters in the Northern Hemisphere that can potentially contain sea ice during winter months".

Also I never have understood why people would want to use 925mb temperatures rather than surface temperatures. For me as an amateur, I have no idea what the temps at 925mb mean for the actual temperatures affecting the ice.

The main graph uses Slater's region, which is the red area shown here.  It seems reasonable to call this "Arctic Ocean".  The regional graphs cover a much larger area, even up to 50°N for Okhotsk and Labrador Sea, but those are not included in Slater's region.

925mb is used by NSIDC and other experts.  I am not an expert on any of this, but my impression is that 925mb temperature is strongly correlated with high pressure.  High-pressure-dominated summers like 2019 and 2020 have higher temperature than low pressure summers like 2012 and 2016.  Surface air temperature is useful too, but less so in autumn when the Arctic ocean is venting heat.  Anyway I'll add regional data for surface air temperature too, but maybe only in csv-format to avoid duplicating the regional graphs.

[Phil.]

"Also I never have understood why people would want to use 925mb temperatures rather than surface temperatures. For me as an amateur, I have no idea what the temps at 925mb mean for the actual temperatures affecting the ice."

One reason would be that as long as there is ice the surface temperature won't exceed 0ºC, the 925mb temperature will be more representative of the atmosphere.  The DMI temperature shows a summer max just over 0ºC every year.

[uniquorn]

"Also I never have understood why people would want to use 925mb temperatures rather than surface temperatures. For me as an amateur, I have no idea what the temps at 925mb mean for the actual temperatures affecting the ice."

One reason would be that as long as there is ice the surface temperature won't exceed 0ºC, the 925mb temperature will be more representative of the atmosphere.  The DMI temperature shows a summer max just over 0ºC every year.

Agreed, but I ran the oceanwide data anyway to see how much difference the open water area makes. Shows a similar trend at a higher temperature. 2019/20 really stand out.

[P-maker]

Recalling several discussions with former DMI colleagues, there are a number of pros and cons regarding the choice of surface versus 925 hPa temperatures.

First, the benefit of using ECMWF-generated surface (2m) temperatures is that there are at least a few relevant anchor points based on real observations from ships of oppportunity, bouys, ice island observatories, land stations etc.

The cons would be - as described - that the summer temperatures close to melting ice would always stay close to 0C. However, during winter - and particularly during inversions - the surface temperature over the ice surface tends to be extremely low - much lower than just a few hundred meters above the ice surface.  This will give an artificial low variability during summer, when most ASIF members follow the melting season. On the other hand, very few observers seem to care about the exaggerated temperature variability during winter.

In contrast to the traditional DMI 80N temperature indicator, the 925 hPa temperature index is an almost purely model-based number, since very few radiosonde measurements are taken over - or close to - the Arctic Ocean. This will most likely lead to a higher temperature variability during summer and comparatively lower variability during summer. Thus, some kind of filtrering is needed in order to retrieve a signal amongst all the noise.

Adding to this seems to be choice of geographical areas ( selecting separate basins), averaging techniques etc. Considering the high spatial variability of Arctic sea ice during the calendar year ( and thus the extent of open ocean with very few actual observations), it will most likely always be the personal taste of the programmer, whether such an 925 hPa index will suit all requirements. It will be extremely difficult to find objective criteria to come up with an authoritative indicator, which will eventually replace the DMI 80N one.

[binntho]

Thanks for the clarification - the map being used is this one here, which is eminently sensible:



But the use of 925mb rather than surface temperatures makes this a non-starter when we are looking at what the ice is doing. The DMI 80N graph is pegged to the ice during summer, yes we all know that. But is pegged to the ice and it is the ice that we are interested in primarily.

So does 925 temps of -15°C or -20°C mean that the ocean surface is freezing? I have no idea. I can't even begin to think of how to interpret those temperatures in any meaningful way as relates to what the ice or sea surface is doing.

[oren]

To each their own. Some swear by 925mb temps, some by surface temps.
In my layman's understanding, the 925mb temps above ice denote the energy going into or out of the ice, especially in summer, while the surface temps are pegged so do not give much indication of the situation. So high 925mb temps in summer might indicate a rapid loss of thickness (top and bottom melt), and low 925mb temps indicate a low loss or no loss, while in both cases surface temps remain at the ice/snow melting point and indicate nothing much.
In any case, Steven's initiative of providing the data is mighty welcome.

[gerontocrat]

In any case, Steven's initiative of providing the data is mighty welcome.

It certainly is. He is a magician. We are damn lucky to have him in the ASIF.
I am playing fast & loose with the data on the "when will the Arctic go free" thread.
I am using surface temp data. While I understand the argument for using 925mb data for the summer, I am looking for the upward spikes when the ice has gone, and also at the yearly trend of air temperatures at the surface.

my latest offering is at https://forum.arctic-sea-ice.net/index.php?topic=2348.msg387826#msg387826

[gerontocrat]

I am looking for the upward spikes when the ice has gone, and also at the yearly trend of air temperatures at the surface.

& here is a graph from the Chukchi data of daily surface air temperatures that shows the spike in temperatures in July.

It also shows the September to November period of highest temperature increases.

[Steven]

Meanwhile I added the regional graphs for 925mb to the automated daily pipeline, so they will be updated daily on the website.  Also added regional graphs for surface air temperature on a separate subpage.

There's still this issue:

3) NCEP reanalysis is an old product, there may be better alternatives available.  But I'm not sure where to find the numeric data for those.

I looked around for ECMWF ERA5 reanalysis data, which is a much more modern reanalysis product with a 10 times higher resolution than NCEP.  But I can only find monthly and hourly data to download.
If anyone knows an URL to download daily average gridded temperature data for ERA5 or for another modern reanalysis product, you're definitely welcome to post it here.  Otherwise I'll stick to the NCEP data.

[FishOutofWater]

Excellent work Steven. Thanks for carrying Slater's work forward.

As for why use 925mb temperatures, somewhere in the Arctic Report Card circa 2016, according to Google these words explained why:

Dec 12, 2016 — Temperatures are from slightly above the surface layer (at 925 mb level) to emphasize large spatial patterns rather than local features.

[binntho]

Excellent work Steven. Thanks for carrying Slater's work forward.

As for why use 925mb temperatures, somewhere in the Arctic Report Card circa 2016, according to Google these words explained why:

Dec 12, 2016 — Temperatures are from slightly above the surface layer (at 925 mb level) to emphasize large spatial patterns rather than local features.

Indeed, Steven is one of those members of this forum who is making a real contribution, to the benefit of the rest of us.

As for the 925mb discombobulation, I have no doubts that real scientists know what to do with it - but my smurflike brain gets very confused and has problems coping with anything more complicated than the most basic information.

But a great thanks to everybody who has helped clear up my confusion from last week!

[Steven]

I added a tool to generate regional temperature graphs for previous years.  It has drop-down lists to select a year (from 1979 to 2023) and to select a region:

https://sites.google.com/view/arctic-sea-ice/home/temperature

[Glen Koehler]

     Nice work Steven!

[gerontocrat]

Steven continues to throw out showers of solid gold data.

All I need is 30 hours a day to look at it, then see it, understand it and link it to other data.

[Freegrass]

Steven continues to throw out showers of solid gold data.

All I need is 30 hours a day to look at it, then see it, understand it and link it to other data.

Steven should change his name into data wizard.
Excellent job indeed!
I love the volume charts...
For me, they're the most important. How big is the ice cube that needs to be melted before the summer sun can start heating up the water?