OCTOBER PATTERN INDEX (OPI) MONITORING WINTER SEASON 2014-2015

[VillagePlank]

I am not data analyst but I would of thought filtering out a possible neutral value OPI would be valid method. After all if there is anything to the OPI it stands to reason that stronger positive or negative results would give stronger correlation and therefore we are simply filtering out possible noise. 

 

What evidence is there for filtering (or banding) the OPI? It is not a discrete index, and therefore the technique you suggest has no merits apart from curve fitting, which is a real bad no-no. It is possible that other indices provide evidence for step-change behaviour, but I can't find any.

[VillagePlank]

Errata: The Nov/CET graph shows Jan on y-axis on my last post. Sorry!!

[snowking]

I am not data analyst but I would of thought filtering out a possible neutral value OPI would be valid method. After all if there is anything to the OPI it stands to reason that stronger positive or negative results would give stronger correlation and therefore we are simply filtering out possible noise. 

 

Absolutely - I think people are perhaps (understandably so) getting confused about how we should look at using the OPI from a raw data standpoint.

 

From the analysis Steve carried out it becomes fairly obvious that in years where the OPI is either greater than 1 or less than -1, there is a strong correlation to milder/colder winters, particularly it would seem for our part of the world.

 

As proof of this, here is a table showing all the years that BFTV very kindly extracted OPI values for, listed from most negative to most positive, along with CET figures, colour coded for above and below the 71-00 average:

 

 

What we can see from this is that, approximately, this >1, <-1 stance gives us consistently the above hypothesis.

 

What nobody is saying here is that we, with just the raw OPI figure, are able to calculate the likely winter CET (in terms of above or below average). With just the raw OPI data we would be unable to determine any sort of likely figure for this if we finished between 1 and -1 (though seemingly for +OPI years this number comes down a little further to 0.7, though bare in mind of course the small sample size).

 

Clearly the OPI team themselves have some sort of methodology for taking this data and producing some composites, such as the hauntingly accurate forecast last year for N. America, and once their research is released it will be fascinating to see how these are produced.

 

But what nobody can deny here is that, at face value, and with an admittedly small sample size, we can make an assumption that with an OPI figure above or below 1 or -1 we can project a milder or colder than average winter - I will also qualify this by saying that making assumptions in any branch of science is fraught with danger, but certainly at face value this is the case.

 

So, nobody (hopefully at least) is suggesting that we will be able to use the OPI as a new miracle tool every single winter, but this year we seem to be in a position where by we can make a relatively informed hypothesis about a colder than average winter.

 

A hypothesis is there to be proved or disproved, and the conclusion will not be available until March 1st 2015, but my expectations based upon this and many other factors are fairly high for this year.

 

SK

[VillagePlank]

Isn't there the danger of further back engineering of a product that is already based on hindcasting. We could probably filter to give something that looked plausible but would it be of value as a forecast tool?

 

Unfortunately, we have very limited dataset, so any sort of analysis, including mine, comes with a huge pinch of salt. Saying that, though, my analysis is good insofar as the quantity of data is available.

[snowking]

What evidence is there for filtering (or banding) the OPI? It is not a discrete index, and therefore the technique you suggest has no merits apart from curve fitting, which is a real bad no-no. It is possible that other indices provide evidence for step-change behaviour, but I can't find any.

 

I don't think Mucka necessarily is looking at curve fitting, I rather more think he is referring more to what I just posted about above - the fact that the OPI looks like a more useful tool (in terms of raw data output) in years with a figure above or below 1 or -1. I think many others have accepted that with values in between this range it is not as useful.

 

But on another note, some really wonderful research you have presented there and the stronger links to January, whilst not unexpected from the research I've been carrying out in recent weeks, is great to see

 

SK

[VillagePlank]

 

A hypothesis is there to be proved or disproved, and the conclusion will not be available until March 1st 2015, but my expectations based upon this and many other factors are fairly high for this year.

 

SK

 

Absolutely. Bog standard statistics vs the experts. I have my prediction in for January (between 0degC and 4degC, with 95% certainty assuming the validity of the OPI) what's yours?

I don't think Mucka necessarily is looking at curve fitting, I rather more think he is referring more to what I just posted about above - the fact that the OPI looks like a more useful tool (in terms of raw data output) in years with a figure above or below 1 or -1. I think many others have accepted that with values in between this range it is not as useful.

 

But on another note, some really wonderful research you have presented there and the stronger links to January, whilst not unexpected from the research I've been carrying out in recent weeks, is great to see

 

SK

 

What evidence is there for banding apart from 'it fits' ?

Edited November 2, 2014 by Sparkicle

[snowking]

Absolutely. Bog standard statistics vs the experts. I have my prediction in for January (between 0degC and 4degC, with 95% certainty) what's yours?

 

My expectation for January at present is somewhere around the 3-3.5c mark as an upper bound. There is the potential for this to slip a little lower, but as a nice round figure I think 3c would be where my thoughts are right now, given data from previous eQBO, El Nino years.

 

SK

[VillagePlank]

My expectation for January at present is somewhere around the 3-3.5c mark as an upper bound. There is the potential for this to slip a little lower, but as a nice round figure I think 3c would be where my thoughts are right now, given data from previous eQBO, El Nino years.

 

SK

 

The stats support your prediction

[snowking]

 

What evidence is there for banding apart from 'it fits' ?

 

You're right, there is none other than the raw statistics. Unfortunately with such a small sample size the evidence is not absolutely compelling, but I am talking purely from a face value point of view, and within this the suggestion of <-1, >1 (though this could be taken down to 0.7 strictly speaking) clearly works.

 

Had the OPI this year finished at say -0.7, I don't think that anybody would be expecting a cold winter outright. But we can only go for now with the limited data we have, and that certainly suggests cause for optimism.

 

SK

[VillagePlank]

 and that certainly suggests cause for optimism.

 

SK

 

The evidence (and only just) suggests that only January would be restricted in CET value. The rest is pure chance re:OPI vs CET.

Edited November 2, 2014 by Sparkicle

[snowking]

The evidence (and only just) suggests that only January would be restricted in CET value. The rest is pure chance re:OPI vs CET.

If you take the dataset as a whole, certainly yes.

 

However once again if you take the <-1, >1 stance, all winter CET's finished below/above the 71-00 average as a whole, and that is what I was trying to get at - which has been the suggested use of the OPI ever since Steve Murr's initial research earlier last month.

 

SK

[VillagePlank]

If you take the dataset as a whole, certainly yes.

 

However once again if you take the <-1, >1 stance, all winter CET's finished below/above the 71-00 average as a whole, and that is what I was trying to get at - which has been the suggested use of the OPI ever since Steve Murr's initial research earlier last month.

 

SK

 

Which suggests a step-change, of which there is no evidence for ie curve fitting. I'll think I'll bow out now, unless someone has something to say about my analysis.

[snowking]

Which suggests a step-change, of which there is no evidence for ie curve fitting. I'll think I'll bow out now, unless someone has something to say about my analysis.

I would have thought the raw data presented in the table is the evidence - as stated before, that is making assumptions with a small sample size, which is never overly-scientific, but the evidence presented is that so far (sorry for sounding like a stuck record) every time we have seen an OPI above to below the 1/-1 range, it has resulted in an above/below average CET.

 

You are correct IF you include all the years in between that range there is then no correlation evident, but the point I was making (and I think we are both making the same point but with different mindsets) is that up until now the 'extreme' ends of the OPI scale result in predictable results...allegedly at least.

 

SK

[SomeLikeItHot]

If you got the data, I can do the analysis.

 

It's unlikely we will see step change behaviour, although it is possible.

 

NDJ. February isn't winter. It could well be that the restrictions of the UK definition of winter holds up analysis here.

 

Since when did Febuary stop being Winter and November start being winter?

[johnholmes]

Since when did Febuary stop being Winter and November start being winter?

 

the WMO and Met O definitions of northern hemisphere winter are quite clear, 1 December to end of February.

[Dancerwithwings]

Since when did Febuary stop being Winter and November start being winter?

 

Absolutely... Don't mind him adding a month but he definitely ain't taking a month

[starstream]

Absolutely... Don't mind him adding a month but he definitely ain't taking a month

And as it can be the coldest one....definitely not!!

[Summer Sun]

Forecast should be available within the coming week according to Gavin Partridge: http://www.theweatheroutlook.com/twocommunity/default.aspx?g=posts&m=641026#post641026

[Barry12]

Since when did Febuary stop being Winter and November start being winter?

 

Next he'll be telling us May is Summer and August is Autumn.

[SomeLikeItHot]

To complete this for the actual winter months.

 

Running correlation and regression on Feb gives a correlation of 0.48. The regression line puts a -2.2 OPI (if that the final OPI figure) giving around a 2.8 +/- 1.6 (1 StDev) for a Feb CET (vs 4.3 for the 1976-2013 avg) so pretty wide range but on the cold side. 

 

The winter (DJF) CET Mean correlation is 0.68 actually higher than for any individual winter month, regressing forecasts that a -2.2 OPI gives  a CET 3.0 +/- 0.9.

 

So based on the small data set we have it would appear to be forecasting a below average winter. Past performance is no guarantee of future returns. 

[Weather-history]

I take the OPI's negative value is a plus then,

[VillagePlank]

Next he'll be telling us May is Summer and August is Autumn.

 

My mistake: I was taking the astronomical winter (a by product of other research I am doing) insofar as NDJ being the months with the shortest days and least solar insolation. Whoops. Still, at least we now know that the November CET definitively is not reckoned upon by the OPI!

 

I'll redo the analysis, tomorrow, for the peculiar (UK) meteorological definition. Hopefully, it'll tally up much more, but we'll see.

[johnholmes]

 

I'll redo the analysis, tomorrow, for the peculiar (UK) meteorological definition.

not peculiar to UK at all, see my earlier post re definition of meteorological  winter in the northern hemisphere

[VillagePlank]

Here's the charts for Feb, and the updated Mean and Median,

 

 

Better correlations for the mean and median and February comes in quite modestly.

[Mucka]

Since I have insomnia I thought I would write a pointless and unscientific post about the OPI statistics provided in the excellent chart in snowking's post.

Now all of this is somewhat tongue in cheek, especially since the sample size is far too small to be statistically significant but it is kind of fun playing with data so bare with me.

By the way when I set out on this little task I had absolutely no idea what the results might be.

 

There are 38 years OPI values listed but of those

63% (24) returned a negative value

and only 37% (14) a positive value.

This may suggest the OPI formula is more likely to produce a negative value than a positive one. (Or it may not)

 

Also the average negative value is somewhat higher than the average positive value;  -0.97 against 0.76

If the neutral value were considered to be 0 the mean average OPI bias per year would be -0.33

 

If we took this absurd premise as reality and add 0.33 to all the OPI values we end up with 18 positive values and 20 negative values.

 

If we then use <-0.6 negative and >0.6 positive as definitive values we get 9 negative OPI years and 11 positive and 18 neutral.

 

Of the 9 negative years all 9 had below average CET (100%)

Of the 11 positive years 10 of a possible 11 had an above average CET. (91%)

Of the 18 neutral years there were 9 above average CET (50%) and 9 below average CET (50%)

 

Now that's correlation baby!

 

Lies, damn lies and statistics. 

Edited November 3, 2014 by Mucka