And we begin... Part Deux

[usedtobe]

I've been under the impression that the PDO lags and is influenced by ENSO in the means? ENSO itself doesn't correlate with the NAO state but it does correlate to the AP index very well, and the PDO anomaly also correlates well with the AP index value, which makes me wonder if the PDO anomaly is forced by same driver modulating the NAO? Is the consensus here that the strong correlation between the PDO, AP index, and NAO is just a freak Wtf coincidence? I find that hard to believe, the statistical odds of that being a coincidence seem quite low.

The PDO does seem to lag the AP index but I find thateven more interesting. The AP index versus the NAO would make sense if indeed it is stratospheric connection in the PV structure, but how and why that signal is showing up in the PDO anomaly seems like anyones guess since ENSO and PDO are connected, so the Geomag forcing on ENSO to get than PDO anomaly doesn't seem farfetched.

This is something I'm very interested in, my older brother who got me into the wx tells me it's "clearly" the geomag sun, so I'm biased, but it seems like there really is something there.

How many times have I told you that I have read the papers.....? I mean wow really. Point me to where it is proven that snowcover directly forces blocking in the AO/NAO area. Not that it correlates, not that there is a plausible mechanism...but that the mechanism that is proposed directly forced the AO/NAO blocking.

You can use the same argument about the geomag sun and ENSO or the NAO. Low values of the AP index do correlate pretty well with a negative NAO but the mechanism is just as loose or looser that snow cover.

[BethesdaWX]

You can use the same argument about the geomag sun and ENSO or the NAO. Low values of the AP index do correlate pretty well with a negative NAO but the mechanism is just as loose or looser that snow cover.

Oh yes I know the geomag idea is theory as well with a lesser known mechanism too. I think of it as if the PDO, NAO, Snowcover, and AP index all correlate fairly well, couldn't it inferred that they all have a common driver?

[SVT450R]

Looking good latest year to year comparison.

[skierinvermont]

Since the entire mechanism for snowcover is warming/disruption of the stratosphere I think it's worthwhile to continue to track the stratosphere. Thus polar stratosphere continues to plummet remaining near record lows:

[skierinvermont]

I did some further research on the connection between stratospheric temperatures and blocking which I thought I would share although it's probably not news to most peope:

It appears that a "cold" November stratosphere has an AO which is .75 higher in Dec and 1.5 higher in January than a "warm" November stratosphere. Given this November's stratosphere is near record cold, that is good reason to expect below average blocking this December and January.

When the stratosphere remains cold in December, the correlation strengthens slightly.

I used 50mb stratospheric temperatures 1964-1997 (what was available from KNMI climate explorer). Below are the AO values associated with the indicated months.

Keep in mind all the values are skewed negative, especially in January, because for some reason the AO averages negative in the long-run. The important thing is the difference between warm and cold stratospheres.

Nov stratosphere

10 warmest: Dec AO -.5, Jan AO -1.26

aboveaverage: Dec AO -.19, Jan AO -.57

below average: Dec AO +.03, Jan AO -.26

10 coldest: Dec AO +.13, Jan AO -.13

Nov+Dec stratosphere

10 warmest: Dec AO -.74, Jan AO -1.14

above average: Dec AO -.38, Jan AO -1.05

below average: Dec AO +.22, Jan AO +.22

10 coldest: Dec AO +.36, Jan AO +.26

[skierinvermont]

I found another stratospheric temperature data set that goes back to 1950. This time I focused on the area 50-90N 90E-150W (basically Russia and surrounding areas) where the correlation is strongest.

I used 70mb stratospheric temps from the NCEP reanalysis: http://www.esrl.noaa.../timeseries1.pl

Based on this I would give us at least a 75% chance of having a +AO in December+January. If the stratosphere remains near record cold for the next 2-3 weeks, our chances diminish even further. There has never been a significantly -AO January following a top 10 cold stratosphere.

November stratosphere

-The December AO was 1.09 higher following the 10 coldest than the 10 warmest.

-The January AO was 1.75 higher following the 10 coldest than the 10 warmest.

-The December AO was .63 higher following cooler than average than warmer than average

-The January AO was 1.71 higher following cooler than average than warmer than average

November + December stratosphere

-The December AO was 1.82 higher following the 10 coldest than the 10 warmest

-The January AO was 2.81 higher following the 10 coldest than the 10 warmest

-The December AO was .72 higher following cooler than average than warmer than average

-The January AO was 1.41 higher following cooler than average than warmer than average

- 8/10 coldest had +AO Dec, only 1 below -.1

- 8/10 coldest had +AO Jan, none below -.2

-9/10 warmest had -AO Dec, none above +.2

-9/10 warmest had -AO Jan, none above +.3

For reference this is what this Oct 15 - Nov 15 looked like this year:

Which matches very well to the November stratosphere preceding +AOs. The following is a correlation between the November stratosphere temperature and the subsequent January AO.

In addition, if we look higher in the stratosphere at 20mb the spatial pattern of anomalies 10/15-11/15 has matched that of other Novembers preceding +AO December and Januaries. The similarity is astonishing. The current stratosphere is in a classic "pre +AO pattern."

10/15-11/15 20mb temps this year:

November 20mb temp correlation to subsequent AO:

[ORH_wxman]

The stratosphere temps thus far in November resemble that of 1975 and 1956 the most looking back through each year Nov 1-15. 1999 isn't a terrible match either. ..and surprisingly 2010 wasn't a terrible match either. But the first two seemed like the closest just eyeballing the N hemisphere charts.

[skierinvermont]

The stratosphere temps thus far in November resemble that of 1975 and 1956 the most looking back through each year Nov 1-15. 1999 isn't a terrible match either. ..and surprisingly 2010 wasn't a terrible match either. But the first two seemed like the closest just eyeballing the N hemisphere charts.

2010 has a pretty big switch around mid month November that doesn't show up in the full month means where the stratospheric pattern basically did a complete 180 and Russia ends up warmer than average (which matches the -AO correlation you would expect). And then the stratosphere remained pretty favorable through December. I guess that's what we would have to hope for this month. Agree this year pretty good match to 75, 56 and 99.. and I looked it up they all had +AO Dec-Jans (I have trouble remembering all these years).

first half of Nov 2010:

second half:

[WxUSAF]

Week 45 snowcover is now second highest ever observed. 36.63M km^2. Record snowcover for Eurasia. North America is above average, but not near record levels.

http://climate.rutgers.edu/snowcover/table_area.php?ui_set=0&ui_sort=1

[bluewave]

This new snow advance index looks interesting.

http://web.mit.edu/j...Jones_GRL11.pdf

[MN Transplant]

This new snow advance index looks interesting.

http://web.mit.edu/j...Jones_GRL11.pdf

Damn, hell of a statement.

The fact that we discovered a single predictive

index that explains close to 75% of the variance of

the winter AO (though the period is short and the degraded

SAI over a longer time period explains less of the AO

variance) is inconsistent with this thinking and demonstrates

that the AO, while thought to be unpredictable, may in fact

be one of the most easily predicted phenomenon known in

the climate system.

[skierinvermont]

Damn, hell of a statement.

The stratospheric temperature data I presented a few posts back can probably explain just as much if not more of the variance if I had to guess. The correlation was very strong. I don't see why so many papers focus on snow cover when you can cut straight to the proposed mechanism (stratospheric temperatures and pattern). I guess the lead time is a little longer with snow cover since it is farther up the causal chain.

Also notice that when they use the full longer data set the correlation degrades.

[QVectorman]

This new snow advance index looks interesting.

http://web.mit.edu/j...Jones_GRL11.pdf

Great stuff...so I'm guessing that this year has a negative SAI forecasting a -AO since the snow cover south of 60N advanced rapidly near the end of the month?

[okie333]

Great stuff...so I'm guessing that this year has a negative SAI forecasting a -AO since the snow cover south of 60N advanced rapidly near the end of the month?

Actually positive SAI... they inverted the SAI in the graph.

[skierinvermont]

Great stuff...so I'm guessing that this year has a negative SAI forecasting a -AO since the snow cover south of 60N advanced rapidly near the end of the month?

The index is the slope of the line of best fit for Eurasian snowcover during the month of October. I would say the slope this October was near average, perhaps slightly above average. The end of the month did finish above average, but the 20-28th was pretty far below average which would lower the slope. So the AO signal is probably near neutral perhaps slightly negative.

But I think the period they study is too short and that the mechanism (the stratosphere) is more important than the snow itself. For this reason I would forecast at least a 75% chance of a +AO in the December January period, with a good chance of a very +AO.

[skierinvermont]

Notice how the % of variance explained falls from 72% to 40% when they use the longer data set. They seem to blame this on the fact that the longer data set uses weekly instead of daily data, but this doesn't seem like a very satisfactory answer to me. More likely, IMO, is that some of the correlation over the short 13 year period was due to chance. As they show, removing just 1 year, 2009, reduces the R-squared value from 72% to 56%.

[QVectorman]

Actually positive SAI... they inverted the SAI in the graph.

Ya I accounted for that, I factored a +SAI and multiplied by -1 resulting in a -SAI

[BethesdaWX]

I like this paper because it does not assert that the snowcover is the causative mechanism for changes in the AO and NAO, but rather use it as a correlating device. October snowfall, and snowfall in general, also correlate well to you know what. I'm out, but statistical correlation may lead you an accurate anomaly, but not necessarily to the forcing that created that anomaly, but rather a reflection of the forcing that was the culprit for that anomaly.

The fact that we discovered a single predictive

index that explains close to 75% of the variance of

the winter AO (though the period is short and the degraded

SAI over a longer time period explains less of the AO

variance) is inconsistent with this thinking and demonstrates

that the AO, while thought to be unpredictable, may in fact

be one of the most easily predicted phenomenon known in

the climate system.

But I think the period they study is too short and that the mechanism (the stratosphere) is more important than the snow itself. For this reason I would forecast at least a 75% chance of a +AO in the December January period, with a good chance of a very +AO.

This should lead to some answers to our disagreement. Lets see if the NAO goes negative in late December as I think it will. We also have 8 months left before we come to the end of the solar lag, get ready.

[QVectorman]

Notice how the % of variance explained falls from 72% to 40% when they use the longer data set. They seem to blame this on the fact that the longer data set uses weekly instead of daily data, but this doesn't seem like a very satisfactory answer to me. More likely, IMO, is that some of the correlation over the short 13 year period was due to chance.

Where are you getting those percentages? I searched the whole paper and neither 72 or 40 come up. Plus, all of their data produces a p<.01 which means that there is only a 1% chance that the results are due to random chance. Secondly, the SAI actually has a higher r value for the longer data set.

[CAD_Wedge_NC]

Where are you getting those percentages? I searched the whole paper and neither 72 or 40 come up. Plus, all of their data produces a p<.01 which means that there is only a 1% chance that the results are due to random chance. Secondly, the SAI actually has a higher r value for the longer data set.

Yes. That is what I gathered from the paper as well. Also, it suggests that we should see a -AO winter due to the pronounced advance in snow cover late in Oct. I would not be so quick to discount this article as rubbish. We all acknowlege that there is a connection between Siberian snow cover and a -AO winter. We just assumed that it was the extent of the snow cover and not the rate of advancement.

Interesting paper......... It will be a major indicator for our winter weather forecasting if it is proven to be a valid tool.

[skierinvermont]

Where are you getting those percentages? I searched the whole paper and neither 72 or 40 come up. Plus, all of their data produces a p<.01 which means that there is only a 1% chance that the results are due to random chance. Secondly, the SAI actually has a higher r value for the longer data set.

The 72 number actually is referenced indirectly when they say snow explains near 75% of the variance. They never mention where they get that number, but it is derived from squaring the r correlation coefficient of .844 seen on the graph Fig 2c.

The 40% number comes from squaring the r coefficient from Fig 2a (.626) which is the full 1975-present dataset.

Given this, I remain highly skeptical that the snowcover index they present actually will explain near 75% of the variance long-term. They seem to ignore the 40% explanatory power of the 1975-present data because it used weekly instead of daily data. But I highly doubt that using daily instead of weekly data would have that much more explanatory power. It's also worth noting that removing 2009 from the 1997-present daily data reduces the % of variance explained from 72% to 60%. 13 years is just too short.

I also don't see any plausible theoretical mechanism whereby Oct snow explains 75% of the variance in the winter AO. That would say the AO is by and large just a reflection and mostly predictable by the Oct snow index. Correlations that strong are rare in meteorology as far as I can tell. Too many variables for one single variable to be so powerful.

[usedtobe]

Great stuff...so I'm guessing that this year has a negative SAI forecasting a -AO since the snow cover south of 60N advanced rapidly near the end of the month?

I'm not sure, I think the snowcover increase needs to be more uniform through the month for it to mean we get a negative ao but don't know for sure. Cohen is or has issued a forecast recently based on the snow cover.

[QVectorman]

I'm not sure, I think the snowcover increase needs to be more uniform through the month for it to mean we get a negative ao but don't know for sure. Cohen is or has issued a forecast recently based on the snow cover.

Does anyone have a link to his forecast yet?

[usedtobe]

The 72 number actually is referenced indirectly when they say snow explains near 75% of the variance. They never mention where they get that number, but it is derived from squaring the r correlation coefficient of .844 seen on the graph Fig 2c.

The 40% number comes from squaring the r coefficient from Fig 2a (.626) which is the full 1975-present dataset.

Given this, I remain highly skeptical that the snowcover index they present actually will explain near 75% of the variance long-term. They seem to ignore the 40% explanatory power of the 1975-present data because it used weekly instead of daily data. But I highly doubt that using daily instead of weekly data would have that much more explanatory power. It's also worth noting that removing 2009 from the 1997-present daily data reduces the % of variance explained from 72% to 60%. 13 years is just too short.

I also don't see any plausible theoretical mechanism whereby Oct snow explains 75% of the variance in the winter AO. That would say the AO is by and large just a reflection and mostly predictable by the Oct snow index. Correlations that strong are rare in meteorology as far as I can tell. Too many variables for one single variable to be so powerful.

I'm also skeptical the higher correlations are with a very small data set. Using the full data set actually has an r squared of .39. That means there is a whole lot of variance not explained by the index.

[bluewave]

I'm also skeptical the higher correlations are with a very small data set. Using the full data set actually has an r squared of .39. That means there is a whole lot of variance not explained by the index.

They may be overdoing the actual correlation, but the idea using the rate of increase is an interesting.

[skierinvermont]

I'm also skeptical the higher correlations are with a very small data set. Using the full data set actually has an r squared of .39. That means there is a whole lot of variance not explained by the index.

yep (i rounded the .39 to .4). I very much doubt there is a silver bullet.. the .39 makes more sense than .72. Still not bad though and the rate does seem to correlate better than absolute extent.

[ayuud]

Does anyone have a link to his forecast yet?

“So what’s Cohen’s forecast for this winter?

Snow cover last month in Siberia was average. New Englanders should expect precipitation near average and temperatures slightly above average.

In other words, this winter won’t be as severe as last winter.

http://www.thebostonchannel.com/weather/29803952/detail.html#ixzz1e7BPXbt4

[okie333]

35-55N snowcover anomaly is off the charts

[HM]

35-55N snowcover anomaly is off the charts

Anyone jumping off the cold forecasts for mid-winter will regret it. Okay okay...I'm being a dbag. But the build up of snow cover and ozone is enough evidence to me, along with the active MJO, to suggest the AO is heading into the tank at some point--solar notwithstanding.

[skierinvermont]

At first I thought that chart was global and that the world was ending.

Sounds like a good argument HM, I hope you are right.