Call For Forecasts -- Predict 20 March Before 4 March

[BLAST FROM THE PAST]

Deep LP 975-985 on SW/NE track across Ireland and the NW UK/Scotland, mainly NW parts getting gale force winds and Scotland up to severe gale force winds with general SW low. HP over Iberia/Azores building N over UK behind the LP which heads into Norway so a switch to warm, carm conditions.

BFTP

[The PIT]

An improving day as Low pressure to the north east moves slowly away into Europe allowing a ridge of high pressure to build slowly from the south west.

Western areas will start with strong NNW winds and light showers which die away as the day goes leading pleasant sunny spells as the winds moderate.

Eastern England and Scotland. Heavy showers becoming more confined to the east coast as the day progress. The mountains may also see some snow. Winds close to gale force on the east and Scottish coasts easing as day goes on.

Southern England mainly dry with a brisk breeze early on but coming pleasant and sunny in the afternoon.

Temps below average Scotland, Northern England and the east coast.

Average to above in Western and Southern areas.

[Roger J Smith]

Just a note to those who have submitted verbal forecasts without a map, the only way to be included in the scoring (correlation) would be to provide a map-substitute grid of pressure readings for 12z on 20 March, for each 5 deg interval in the grid from 65 N 20 W east to 65 N 20 E, and from that boundary south to 45 N 20 W across to 45 N 20 E.

Your grid would therefore have 5 rows of 9 pressures.

If you want to save yourself some typing, you can go like this:

20.18.16.13.10.06.03.97.94

etc

(make sure there are four more similar rows)

and rely on the fact that I am not totally daft to work out the full readings (no maps are likely to stray outside 950 to 1050 now that MB has entered, so all such values are defined).

Please note, these 5 deg intervals are the lat-long lines on the base map, so it's not that hard to do if you can just print out the base map. I don't need to tell you this, surely, but 55N 0W for your orientation would be the lat-long intersection point in the North Sea east of Newcastle. I am dropping 70N from my validation grid partly because it's somewhat off the base map in the west, but also to avoid some needless work (the correlations are going to be significant enough in the grid from 65 south to 45).

I may attempt to estimate such a grid for some of you without a map, but no promises plus you're at the mercy of my mind-reading abilities which are only moderately good.

One can note that with a verbal forecast, while it may verify well within the time frame, you have a bit more flexibility than a pressure forecast at specific hour, as your pattern could be 12 hours "out" without affecting very much the accuracy of your verbal details.

By the way, I enter the values like this (from the example above) 20.18.16.13.10.6.3.-3.-6 so that format is good too if you wish.

We are rapidly approaching the dawn of "fantasy island" GFS model comparison, so be forewarned, the entry deadline is 4 March at 1500 GMT ... anything after that will not count as the opportunity to consult the 16d GFS will be available.

If you have a map entered but now have a different idea, there's no rule against entering an updated map, we will score both of them, but limit of one per person.

This may sound like a ton of work but once I have the map grids in a computer file, the correlations take very little time.

Edited March 2, 2011 by Roger J Smith

[Glacier Point]

Final call..

Latest MJO forecast tools suggest only a weak amplitude event so angular momentum not so positve and La Nina persistence in the atmosphere. Assume Global Wind Oscillation in phase 3 around the 20th.

Forescast MSLP>20/04/11

Edited March 3, 2011 by Glacier Point

[bluearmy]

spookily, there are others not too far away from this. will be interesting to see if the ones drawn weeks ago are less accurate than those done in the final days which, no doubt, use some nwp input, albeit, a long way away.

[Roger J Smith]

I have input the data for 13 forecast maps and will add any last minute entries submitted today. Anything after 1500h today will not be included in the validation.

I was able to run an inter-forecast correlation to see what clusters of forecasts there might be among these thirteen entries. In general terms, this showed a cluster of about half the forecasts including two from GP. I should mention here that I have included the first and the third of GP's three maps and left out the middle (time range) submission mainly because I see a gradual progression of a zonal WSW pattern in all three and we can assume that correlations for the missing middle time frame will lie between maps one and three. The two GP maps had a correlation of .89 which means they have only changed slightly. It would be interesting to get more details about the model from which these maps were submitted, as I am guessing some of us reading this thread may be more familiar with the system being used than others are.

My own question about the model, GP, would be whether or not it ever shows organized weather systems with a tighter pressure gradient, as these maps look more like ensemble maps that deal with probable large-scale patterns rather than attempting to show transient features as most of our contest entries have done.

An ensemble type forecast will normally verify over a narrower range of correlations, say -0.4 to +0.6, than detailed maps with sharper features which could verify in a range like -0.9 to +0.9 at random.

I've created my own ensemble forecast from the 13 entries, based on pressure averages at the 45 defined grid points. This is the result (it is not much different from the two GP entries, even when compared to the average of the other eleven which yields almost the same result):

Later today or tonight, I will have an analysis of correlations of all entries with the 16-day GFS for "map time" on 20 March.

[Glacier Point]

Thanks for the work Roger.

The forecast maps are derived from the NCEP/NCAR reanalysis tool using a custom data set uploaded to the NOAA server. The data sets I've been using are for respective Global Wind Oscillation phases (from the spider graph plot) and I've refined these in the last week or so for a sub-set where the Arcitc Oscillation is positive as I think this will be a signficant variable, particularly mean pressure around Greenland and Iceland at that time given the lagged state of the polar stratosphere.

Readers might get an inclining into the 'spread' of possible solutions presented by the GWO phases 'on the table'. Phases 0 and 4 which represent angular momentum in a neutral or weakley negative state are much more supportive of the mean ridge being centred over the mid latitudes. Phases 1-2-3 are driven by angular momentum operating a much lower and negative base state and these are associated with a flatter pattern with the mean ridge displaced southward and an upper flow generally west and west-south-west. Interestingly all three low angular momentum phases depict low pressure around the central Mediterranean and low pressure over Iceland.

The forecast plot is analogous to an ensemble mean and highlights the weakness in attempting to predict a single day at long range. Features such as fronts will not be picked out using this methodology. I'm operating off the balance of probablility here as the GWO forecast plot usually captures a mean across 3-5 days, so if the single day is representative of the mean, I'll be OK. It might for research purposes be worth producing a second evaluation for a 5 day mean pressure centred on the 20th March to see how we got on with our 'super ensemble mean' for entries.

Just as a footnote to this, should angular mometum increase sufficiently upwards, the evolution towards phase 4 would be on the cards around or just after the 20th so we should see a rise in pressure from the south in line with the phase composite:

Edited March 4, 2011 by Glacier Point

[bluearmy]

looking through the gfs 12z suite op and ens mean, i think we declare this over and award the prize to me and iceberg

can only go downhill from here onwards ....................

[Roger J Smith]

Thanks for the above GP, very informative ... I have input the 16d GFS and as suspected by Bluearmy, several of the experimental forecasts score quite well against it. Here's an overview of how I plan to report on the predictions as we move towards validation time.

As mentioned, there are 13 experimental forecasts in play. These are listed below in the order they were presented and the numbers represent days before validation date. Each forecaster is reduced to a two or three letter code as follows:

COA 32 (Coast)

RIC 32 (Richie V)

ICE 32 (Iceberg)

TWS 32 (Thundery Winter Showers)

GP1 32 (Glacier Point, first entry)

LOM 32 (LomondS)

PB. 32 (PaulB)

STM 31 (Stormmad26)

RJS 27 (Roger J Smith)

SFT 26 (Stationary Front)

MB. 20 (Murcie Boy)

GP2 17 (Glacier Point, third entry)

BLU 17 (Bluearmy)

and also

NW. 16 (derived Netweather ensemble from the above)

Now, a correlation analysis was performed against the GFS 16d output at 12z. Note that correlation can be very high when pressure patterns are similar even though there could be a range of extreme values. Anyone who had the highs and lows in the right places for today's map would score high on correlation, but there could be a substantial difference in the intensity function on the maps. I won't overburden the reader or the thread with intensity analysis in these early stages, just for now it will suffice to mention this point, as it may come into play looking at overall forecast similarity as we move closer to map time.

The correlations with the GFS-16d are listed below in order (consult the list above if you can't pick out your identifier) ...

ICE .76

BLU .73

MB. .70

SFT .61

GP1 .61

GP2 .61

NW. .54 (this is our group ensemble)

RIC .42

PB. .31

COA .22

RJS .13

TWS .03

STM .02

LOM -.40

At the present time, with the GFS-16d showing basically an Icelandic low moving east towards strong blocking high over the Baltic region that ridges to strong high pressure west of France, entries with these basic features with some variation have scored higher than what I consider to be the "random" thresh-hold of .40 and forecasts that show different circulations such as northerlies or easterlies have scored either random or as in one case negative correlation. Our experience with the GFS-16d is such that nobody should feel too badly about being down on this list, and so now the interesting part begins, watching the GFS evolve and then picking up the ECM at 10d. Looking at the ECM at days 9 and 10 today, it would seem to be evolving towards a somewhat more southerly storm track than the GFS but that's a very subjective opinion.

I will confine my validations to the 12z model runs each day, but perhaps other readers will post intermediate output for comparison.

[Roger J Smith]

The 15d GFS has shifted considerably and correlates at -0.12 with its previous run (24h). Now it shows low pressure generally over Scandinavia and the Icelandic low further southeast than before. The strong Biscay high is further north closer to Ireland.

As a result, our pack of forecasts have shifted around in their correlations to day 15. Compare with day 16 below:

ICE .76 now .11

BLU .73 now -.29

MB. .70 now .35

SFT .61 now .58

GP1 .61 now .21

GP2 .61 now .51

NW. .54 now .43

RIC .42 now .10

PB. .31 now .51

COA .22 now -.32

RJS .13 now .12

TWS .03 now -.36

STM .02 now .65

LOM -.40 now .51

Forecasts currently making gains are those showing lower pressure over the Baltic region.

This scrambling around of correlations is typical of early "FI" during which the GFS is only picking up faint low-correlation signals (based on average performance, we never know in specific cases until later). There is really no particular trend emerging from the GFS considering all four runs. The ECM is trending towards a solution more similar to the 16d GFS. Back again on Sunday with the 14d results.

[Roger J Smith]

The 14d GFS has changed considerably from the previous day (corr -.75) to more of a blocked pattern with high pressure retreating through the Baltic and a weak SSE flow over the western part of our grid. The correlation with day 16 GFS was 0.32, so a flip-flop pattern somewhat in evidence at this point.

This has changed all the correlations considerably with our experimental forecasts, most of which now show a negative correlation with the 14d GFS. The three-day evolution of these correlations now looks like this:

ICE .76 .11 .12

BLU .73 -.29 .55

MB. .70 .35 -.11

SFT .61 .58 -.36

GP1 .61 .21 -.09

GP2 .61 .51 -.30

NW. .54 .43 -.36

RIC .42 .10 -.08

PB. .31 .51 -0.20

COA .22 -.32 .07

RJS .13 .12 -.32

TWS .03 -.36 .69

STM .02 .65 -.83

LOM -.40 .51 -.78

We'll see where the long and winding road through Fantasy Island leads next.

[Roger J Smith]

The 13d GFS has now returned to a more zonal appearance with parent low pressure east of Iceland linked to a moderately developed low over Ireland. This begins to look a lot more like some of our zonal type forecast maps and has changed the correlations back to similar values to day 16 (where we started this exercise) as can be seen in the list that follows. Meanwhile, the 13d GFS correlates at .82 with the 16d, and at only .09 with the previous day, .05 with two days previous.

Here's the updated forecast map correlation study showing the four day evolution. Remember, these started on day 16 in the order of their correlation, check out the similarity today.

ICE .76 .11 .12 .61

BLU .73 -.29 .55 .35

MB. .70 .35 -.11 .60

SFT .61 .58 -.36 .78

GP1 .61 .21 -.09 .82

GP2 .61 .51 -.30 .80

NW. .54 .43 -.36 .70

RIC .42 .10 -.08 .61

PB. .31 .51 -0.20 .57

COA .22 -.32 .07 .37

RJS .13 .12 -.32 .50

TWS .03 -.36 .69 -.15

STM .02 .65 -.83 .21

LOM -.40 .51 -.78 -.30

Several of our experimental forecasts are doing quite well at this point, with the two offered by Glacier Point holding a slight lead over Stationary Front and RichieV, Iceberg, Murcie Boy, PaulB and RJS all in a fairly narrow range behind this leader pack. The idea of a second low near Ireland is the reason why my forecast has shown a somewhat larger increase in correlation than other zonal type forecasts did (from day 16 to day 13). The Netweather composite or ensemble is doing very well really at 0.70 ... we'll see which one is closer at time zero.

Meanwhile, I've played around with some averages of groups of forecasts, and one that's quite interesting at this point is the average of the three most "developed" zonal maps, from Ric, RJS and SFT. These all have sub-970 mb lows between Scotland and Iceland, and strong westerly flow over parts of the British Isles. The composite of these three maps, which correlates above the NW mean, looks like this:

[Roger J Smith]

The 12d GFS continues the theme of strong low pressure near Iceland but is a few hours slower than the previous run. Still, it correlates at .84 with the previous day. Going back to the first GFS map four days earlier, the correlation is also high, .81 ... the maps for days 15 and 14 are now out of phase entirely with low correlations with today's model run (.35, -.02).

The experimental forecasts continue to do quite well as a group with the NW composite map at .83 correlation with the 12d GFS. In fact the composite map is almost the best of the 14 maps we are comparing, but SFT leads the pack now with an almost identical map showing similar pressure extremes and shaping, and a correlation of .90.

The scores for today are as follows, and note I have scrambled the pack to reflect today's order of merit.

SFT .61 .58 -.36 .78 .90

NW. .54 .43 -.36 .70 .83

ICE .76 .11 .12 .61 .81

GP1 .61 .21 -.09 .82 .81

GP2 .61 .51 -.30 .80 .80

MB. .70 .35 -.11 .60 .80

RIC .42 .10 -.08 .61 .72

PB. .31 .51 -0.20 .57 .72

RJS .13 .12 -.32 .50 .37

BLU .73 -.29 .55 .35 .37

STM .02 .65 -.83 .21 .32

COA .22 -.32 .07 .37 .10

TWS .03 -.36 .69 -.15 -.09

LOM -.40 .51 -.78 -.30 -.25

We will soon have the ECM on board for extra comparisons. Back again tomorrow with day 11 GFS.

[MurcieBoy]

Hi RJS, its fascinating again seeing the flip flop of the GFS and the individual movements of the correlations. Re my forecast, not that it may make much of a difference, but I would just like to say it was based on 00:00hrs charts (rather than 12:00hrs).

Re suggestion for another challenge: each year I host a BBQ and big Fun Day for the kids (and adults) in the summer in my business and invariably it rains (sometimes with a lot of wind too!) or it is too cold for a "summer day out" in a T shirt and shorts. This year, I have decided to set the date now (in March) and undertook my research into which week will be the hottest for this summer with the likelihood of sunshine (and no rain!). I have found the week and have set the date for my guests! It will definitely (edit: or should I say, according to my calculations) be excellent BBQ weather - perhaps the hottest in recent years!

If you fancy a challenge (open to all, as per 20 March challenge) for guessing the hottest day/week in London (so as we can be quite specific) this summer and what the likely temperature will be, I would be up for that. Perhaps folk could submit predictions by 31 March 2011?

Edited March 9, 2011 by MurcieBoy

[Roger J Smith]

Probably that 12 hour difference would not affect correlation very much, especially looking at your map which has no fast-moving features on it. However when the time rolls around I will compare the 00z map to the 12z map and let you know if it makes any significant difference.

Meanwhile, the 11-day GFS has once again changed considerably, and seems to be fairly similar to what the 11-day ECM might be if we had that extra day (based on linear extrapolation from day-9 and day-10 ECM). There has basically been a slowing down of cyclogenesis near Iceland and the high centered over Russia (a common feature of all long-range maps we've been seeing this week) now ridges more strongly back through the U.K. into the Atlantic. Because of that change, most of the correlations have been altered considerably. Once again, I have scrambled the pack to show this. But before getting to the pack, let's start a new line in this report, the ongoing GFS correlation with its own output. For day 11, this goes back five days and starting from day 16, we have:

.57 -.03 .48 .21 .37

What this shows is that the map grid has shifted to preserve some correlation but today's map more closely resembles that from five days ago (day 16).

Now in terms of the workshop entries, we have this new set of results:

BLU .73 -.29 .55 .35 .37 .63

ICE .76 .11 .12 .61 .81 .48

TWS .03 -.36 .69 -.15 -.09 .48

MB. .70 .35 -.11 .60 .80 .41

GP1 .61 .21 -.09 .82 .81 .20

PB. .31 .51 -.20 .57 .72 .18

GP2 .61 .51 -.30 .80 .80 .17

SFT .61 .58 -.36 .78 .90 .15

NW. .54 .43 -.36 .70 .83 .00

RIC .42 .10 -.08 .61 .72 -.11

COA .22 -.32 .07 .37 .10 -.12

STM .02 .65 -.83 .21 .32 -.34

LOM -.40 .51 -.78 -.30 -.25 -.48

RJS .13 .12 -.32 .50 .37 -.53

The NW ensemble has now dropped back to a totally random "look" while the pack looks more similar to where we started five days ago.

We're a little far from the "tipping point" yet so I don't feel that this current situation will necessarily begin to lock in. Hold on for updates. Tomorrow we pick up the ECM for comparison.

[Coast]

Roger, how does the 'Supermoon' of the 19th influence your forecast - if at all?

[Roger J Smith]

Supermoon -- the full moon has been given that title because it is perigeean this month. That fact has only a slight influence on my model index values, the subset of data for perigeean full moons is not very different from the index values in general for full moons. So while the timing of lunar events has a bearing on my forecast, the fact that this is a "supermoon" is not a big factor in and of itself.

Meanwhile, I have some computer problems today, my home internet computer won't boot up so I am off-site to access the internet, but my data for this and many other projects are stored in an older (still working) computer at home, so I have the 10-d GFS and ECM data written down and will do the correlations later. The report won't show up here unless I fix the problems today, until my next visit to this (library) terminal at 0400 GMT. So the day-10 reports will be posted then. Just looking at the maps though, things won't change much from day-11 and the ECM looks to be in the 0.7 correlation range with the GFS, just as an estimate. Both are showing the gradual breakdown of strong high pressure blocking over the Baltic regions. The strong low pressure and tight gradients of days 12-13 GFS are no longer present. Still, ten days out can provide significant changes as we saw in the earlier study.

[Roger J Smith]

This is the belated update for day 10 (yesterday). Will update today (day 9) later -- looks a bit less blocked than yesterday on GFS.

The GFS 10d correlated with the ECM at .39, visually it looked higher.

The updated correlations with day 16 to 10 GFS (d10 ECM in brackets) are:

fcst .. Days 16-11 GFS .......... Day 10

_______________________________________________

BLU .73 -.29 .55 .35 .37 .63 ..... .80 (.52)

ICE .76 .11 .12 .61 .81 .48 ........ .29 (.44)

TWS .03 -.36 .69 -.15 -.09 .48 ..... .51 (.36)

MB. .70 .35 -.11 .60 .80 .41 .... -.13 (.45)

GP1 .61 .21 -.09 .82 .81 .20 ..... .05 (.21)

PB. .31 .51 -.20 .57 .72 .18 .... -.10 (.39)

GP2 .61 .51 -.30 .80 .80 .17 .... -.06 (.29)

SFT .61 .58 -.36 .78 .90 .15 .... -.07 .36)

NW. .54 .43 -.36 .70 .83 .00 .... -.18 (.21)

RIC .42 .10 -.08 .61 .72 -.11 .... -.16 (-.07)

COA .22 -.32 .07 .37 .10 -.12 .... -.02 (-.22)

STM .02 .65 -.83 .21 .32 -.34 .... -.66 (-.09)

LOM -.40 .51 -.78 -.30 -.25 -.48 .. -.68 (-.29)

RJS .13 .12 -.32 .50 .37 -.53 .... -.26 (-.11)

A report on day 9 will follow later, probably with the order re-scrambled.

Edited March 11, 2011 by Roger J Smith

[Thundery wintry showers]

It's interesting that some of the model runs are increasingly modelling the key features that I had gone for (Scandinavian High, relatively high pressure over Greenland, low pressure mid-Atlantic, low pressure in the Mediterranean) when it was honestly just a wild "stab in the dark". Early in the analysis it looked like I was out on a limb with very low correlation scores. The models are still changing quite a lot so we could still end up with a completely different scenario.

[Roger J Smith]

My computer troubles have been overcome for the time being, so I am back on schedule here with the day 9 update.

And the ECM has decided to wander back to the more developed solution with low pressure near the Shetlands, although not as strong as the "NW3" composite that I showed earlier, actually quite similar to the NW 13-forecast average at about 1000 mbs.

This means that the ECM has correlated at -0.21 with its day 10 output, quite a change over the 24 hours. Meanwhile the GFS has only changed somewhat to a less blocked appearance, and correlates at .78 with its previous offering. the inter-model correlation is negative (-0.12) meaning that the two main models have less similarity than even random output might show. They are not totally opposite, however.

These developments have changed the correlations of our 13 forecast maps considerably. For today, I will continue on with the grid in place yesterday, planning to construct a new chart altogether tomorrow that will summarize some of this data and get us into the transition from Fantasy Island to more reliable model output.

the updates ... day 9 correlations show GFS then ECM in brackets

fcst .. Days 16-11 GFS .......... Day 10 ................ Day 9

__________________________________________________________

BLU .73 -.29 .55 .35 .37 .63 ..... .80 (.52) ...... .65 (.18)

ICE .76 .11 .12 .61 .81 .48 ........ .29 (.44) ...... .81 (.33)

TWS .03 -.36 .69 -.15 -.09 .48 ..... .51 (.36) ...... .38 (-.54)

MB. .70 .35 -.11 .60 .80 .41 .... -.13 (.45) ...... .64 (.43)

GP1 .61 .21 -.09 .82 .81 .20 ..... .05 (.21) ...... .59 (.47)

PB. .31 .51 -.20 .57 .72 .18 .... -.10 (.39) ...... .47 (.07)

GP2 .61 .51 -.30 .80 .80 .17 .... -.06 (.29) ...... .53 (.46)

SFT .61 .58 -.36 .78 .90 .15 .... -.07 (.36) ...... .51 (.45)

NW. .54 .43 -.36 .70 .83 .00 .... -.18 (.21) ...... .44 (.61)

RIC .42 .10 -.08 .61 .72 -.11 .... -.16 (-.07) ...... .44 (.39)

COA .22 -.32 .07 .37 .10 -.12 .... -.02 (-.22) ...... -.04 (.66)

STM .02 .65 -.83 .21 .32 -.34 .... -.66 (-.09) ...... -.15 (.52)

LOM -.40 .51 -.78 -.30 -.25 -.48 .. -.68 (-.29) ...... -.69 (.50)

RJS .13 .12 -.32 .50 .37 -.53 .... -.26 (-.11) ...... .03 (.43)

As one can see, many of the forecasts are reasonably correlated with the ECM. But with the difference between the two models, these results remain very tentative. Now we get into the interesting part of the validation as the solution begins to emerge from the "clouds" of obscurity. Let me know if there are any other types of correlation study that you would like to see in this discussion, as I find them quite easy to calculate once the model data are input.

[stewfox]

My computer troubles have been overcome for the time being, so I am back on schedule here with the day 9 update.

the updates ... day 9 correlations show GFS then ECM in brackets

fcst .. Days 16-11 GFS .......... Day 10 ................ Day 9

__________________________________________________________

BLU .73 -.29 .55 .35 .37 .63 ..... .80 (.52) ...... .65 (.18)

Great thread.

Could you just clarify for me and maybe many others what the figures mean.

In terms of correlation in simple terms ie 0.8 or -0.3etc

So for example to move from .73 to -0.29 in one day means a big divergence in the pressure maps for that day ?

[Roger J Smith]

The correlation co-efficient is a measure of the similarity of two sets of data, so when applied to a forecast against another forecast, a measure of how similar they are. A value of 1.0 would indicate an identical pattern (the scales could be different). A value of about 0.4 is the thresh-hold for significant similarity, it means there is at least some broad similarity between two maps. One would like to see 0.6 or 0.7 for a working model at five days, and quite often the major models achieve this. Negative values can be interpreted the same way for differences, a value of -0.4 is starting to get broadly "different" and -0.7 would indicate almost the reverse pattern.

The one weakness in this approach is that a high correlation can be achieved by a much less intense pattern than reality, or vice versa, with features in the right places on the grid. To give an example, you would likely get a high correlation if reality showed a 980 mb low near Shetlands and a 1030 mb high over southern France, but a model showed 1000 mb to 1016 mb as the pressure gradient, but the shape and wind direction were essentially the same. That suggests a second index, pressure difference at grid points, would need to be studied for high correlation cases (there is not much point in comparing them for low correlations because they would be quite high by definition).

I noticed the 18z GFS was trending more like the 12z ECM today. But the study will only deal with 12z model runs, we have enough data already without adding three more layers of complexity.

[Roger J Smith]

Day 8 report

______________________

There is somewhat more consensus today with the GFS and ECM correlating at .58, and generally trending towards a high over the Biscay region with low pressure either west of (GFS) or near (ECM) Iceland. What seems to be vanishing from the model runs in recent days is the former tendency to strong blocking high pressure in Scandinavia and Russia. These features have been vastly deflated to more of a seasonal zonal flow pattern.

In general, many of the forecasts correlate very well today with the ECM and not so well with the GFS which has higher pressures in the west-central and northwest portions of the validation grid. In fact the NW composite correlates at .90 with the ECM and differs from it only in showing a deeper low north of the UK.

Starting today, I am going to simplify the correlation table by removing the days 16 to 11 GFS data which can be found in any of the previous days. These now become somewhat irrelevant to our study as the models begin to converge on an actual solution. The UKMO and GEM which go out to 6d are showing broadly similar outcomes.

Before discussing the various forecasts, I should note that the ECM correlates at 0.5 with its previous day's run, and the GFS at 0.73.

These are the correlation stats for each of our forecasts from day 10 to day 8 (today), and I have rescrambled the order again using today's ECM as the guide.

Fcst ... day 10 ....... day 9 ....... day 8 ...... all GFS(ECM)

SFT .... -.07 (.36) ... .51 (.45) ... .61 (.96)

GP2 .... -.06 (.29) ... .53 (.46) ... .61 (.94)

NW. .... -.18 (.21) ... .44 (.61) ... .38 (.90)

GP1 .... .05 (.21) ... .59 (.47) .... .54 (.82)

MB. .... -.13 (.45) ... .64 (.43) ... .47 (.80)

PB. .... -.10 (.39) ... .47 (.07) ... .78 (.73)

ICE ..... .29 (.44) ... .81 (.33) .... .53 (.70)

RIC .... -.16 (-.07) ... .44 (.39) ... .30 (.65)

STM .... -.66 (-.09) .. -.15 (.52) .. -.03 (.64)

RJS .... -.26 (-.11) ... .03 (.43) ... -.03 (.39)

TWS .... .51 (.36) ... .38 (-.54) .. .34 (.32)

BLU ..... .80 (.52) ... .65 (.18) ... .19 (.14)

COA ... -.02 (-.22) ... -.04 (.66) .. -.40 (.12)

LOM ... -.68 (-.29) ... -.69 (.50) .. -.49 (.12)

The order changes somewhat for the GFS correlation, with PB leading at .78.

At eight days, there is still room for a major change in model evolution -- in some past cases, these strong Biscay highs turn out to be overdone and this could allow the storm track to shift south. Another update to follow this time Sunday.

[Roger J Smith]

The day 7 maps have been analyzed, and the situation remains about the same today as yesterday. Both ECM and GFS show strong high pressure over the southern part of the UK into the Baltic region, with the Icelandic low fairly suppressed and a little to the west of Iceland.

The inter-model correlation today is 0.64, not as high as this will become in the next few days, but quite typical of this time range. The ECM correlates at .88 with its previous day's 12z run, and the GFS has shifted more with only 0.5 correlation to day 8.

The Net-weather forecasts correlate today as follows (order taken from last value, ECM correlation):

Fcst ... day 10 ....... day 9 ....... day 8 ...... day 7 (all days GFS (ECM))

SFT .... -.07 (.36) ... .51 (.45) ... .61 (.96) ... .39 (.91)

GP2 .... -.06 (.29) ... .53 (.46) ... .61 (.94) ... .42 (.87)

MB. .... -.13 (.45) ... .64 (.43) ... .47 (.80) ... .39 (.82)

ICE ..... .29 (.44) ... .81 (.33) .... .53 (.70) ... .57 (.82)

NW. .... -.18 (.21) ... .44 (.61) ... .38 (.90) ... .24 (.80)

GP1 .... .05 (.21) ... .59 (.47) .... .54 (.82) ... .50 (.77)

PB. .... -.10 (.39) ... .47 (.07) ... .78 (.73) ... .51 (.75)

RIC .... -.16 (-.07) ... .44 (.39) ... .30 (.65) ... .31 (.59)

BLU ..... .80 (.52) ... .65 (.18) ... .19 (.14) ... .56 (.39)

STM .... -.66 (-.09) .. -.15 (.52) .. -.03 (.64) ... -.36 (.37)

RJS .... -.26 (-.11) ... .03 (.43) ... -.03 (.39) ... .05 (.32)

TWS .... .51 (.36) ... .38 (-.54) .. .34 (-.32) ... .48 (-.05)

COA ... -.02 (-.22) ... -.04 (.66) .. -.40 (.12) ... -.23 (-.06)

LOM ... -.68 (-.29) ... -.69 (.50) .. -.49 (.12) ... -.86 (-.24)

(note, typo on day 8 TWS corr with ECM changed to -.32)

At this stage, it looks as though the main question to be answered is the actual strength of the European high and whether any of the more developed forecast maps will improve in their correlation due to weakening of that high and subsequent development of frontal waves closer to the UK than the current model consensus. The JMA also goes to 7d and looks broadly similar today, and the GEM output looks slightly more "developed" but this model has the characteristic of being more zonal. The "NW" correlation of 0.80 with the ECM but from three weeks (on average) earlier is very interesting, no?

[Richie V]

My guestimate was based around the massive peak energy signal around the end of the month along with the steepening temperature gradient in the NA at this time of year. Does that make sense