A disaster of a summer forecast...

[Weather-history]

I don't like criticising people but why do the media give credence to amateur weather forecasters? I find most of the time they are wrong and more often than not spectacularly

Here is one from my neck of the woods, Harry Kershaw, our local version of the late Bill Foggitt and he went for a very warm summer based on similarities between 1976 and this year at the start.

http://www.thisischeshire.co.uk/news/22931...ather_forecast/

A disaster overall his forecast and I don't rate him at all.

Here was his disaster forecast for summer 2004

http://www.bbc.co.uk/manchester/weather/harry_forecast.shtml

He was suggesting a cold winter and a wet September, last year. It wasn't

http://www.manchestereveningnews.co.uk/new...tal_winter.html

Edited August 20, 2008 by Mr_Data

[Nick H]

You could tell it was going to be a crap forecast just by looking at his shirt.

[johnholmes]

the media quietly forget when they are wrong and just trumpet the forecast on the fairly infrequent occasions they are right.

Always have done and I suspect they always will.

[multi cellular thunderstorm]

I always thought that could be a good technique in some circumstances, matching the current setup to one from recent years and seeing how things panned out then.

In theory it seems like it could work but I imagine just tiny differences would change the whole development even at close range.

If you had an exact match, which would most likely never happen, then it could probably work well I think.

I would like to see a computer programme where you feed in current charts and it matches them as close as possible to charts from the past. It would be interesting to see what level of accuracy you would get.

If only I was a software designer :huh:

[johnholmes]

been tried numerous times and it rarely and I mean rarely works.

If it did then the major centres would have been using it for years.

comparison is okay but its no use just using it and nothing else.

[The PIT]

Also makes the classic mistake of putting dates on. Guarantees you to be wrong as soon as you do that anyway.

Pattern matching has never been very successful. I think they used to do on the beeb for the month ahead forecast on TV before they scrapped the idea.

[Weather-history]

been tried numerous times and it rarely and I mean rarely works.

If it did then the major centres would have been using it for years.

comparison is okay but its no use just using it and nothing else.

It would help for a start if Harry Kershaw brushed up on his weather history! He cited the gales at the end of the Januaries of 1976 and 2008.

Well the last weeks of January 1976 and 2008 couldn't have been more different. One was cold to very cold and wintry (1976) and the other was very mild although it turned colder right at the end. (2008).

If you are basing your forecast on some form of pattern matching then he needs to get the very basics right!

[VillagePlank]

I would like to see a computer programme where you feed in current charts and it matches them as close as possible to charts from the past. It would be interesting to see what level of accuracy you would get.

If only I was a software designer

A very difficult problem with, probably, no findable solution. Occasionally you might see this sort of problem as referred to as NP complete What might surprise you is the sheer number of NP complete (problems which are known to have NO provable mathematical solution) there actually are: there are loads of the little buggers.

Essentially the problem is exactly what you intimated. Tiny differences. That little surreptitious kink in that low pressure, which turned out to be nothing in October 1974 turned out to be a lifethreatening storm for SE Britain in October 1987. So the question becomes which little (tiny) features are important and which ones are not, and how do I classify them enough so that they can be useful in prediction?

A neural network would be my approach, I guess, but even then I could see the weights be stuck on local maxima (certainly by using backpropogation) and almost certainly by evolving the weights using GA's.

There are other heuristic techniques that could be employed, and, indeed, I believe there is some merit in such techniques. As for reliability? At best .... uncertain.

Be wary of automated analogue approaches and associate them with a high degree of uncertainty.

comparison is okay but its no use just using it and nothing else.

Perhaps ... but I would only say in reducing the number of likely outcomes from the current pattern. Say reducing the likelyhood of an event occuring by about 10% in some cases, and increasing 10% in other (differential of 1/5th isn't actually that bad, but it's nowhere near MetO 5 day prediction rate of about 4/5)

Edited August 28, 2008 by VillagePlank

[J07]

Comparing MSL charts is obviously going to end in bad results.

If you're using pure model output, then that's wrong to begin with. How often does the model sfc pressure field agree with *every* available ob? Never (unless you're in the tropics). And if you're in the Southern Hemisphere you're faced with other problems. Namely, where are the obs? You'll be lucky to find a ship and a handful of buoys in the Tasman Sea and doubtless they will be not where you "want" them.

If you're using charts altered by man, then there's still issues. What spacing of the isobars? If you use a 4hPa spacing then you miss out on a lot of detail, especially localised orographic ridging and troughing. You also miss this on 2hPa spacing at times. So 1hPa spacing shows a heck of a lot....but are you missing more? How far do you go? Well no further than 0.1hPa because barometers are only accurate to this degree.

Then there's the upper fields. Forecasting surface features for the length of time he's attempting is never going work without knowing the 500hPa pattern, and you also would need to know the placement and strength of jets. And there would still be more information to gleen. Think of sparsely spaced are our atmospheric soundings. They give localised data and widely spaced intervals for a tiny spot. And if your balloon goes through a convective cloud it won't be representative anyway.

Going off on a tangent here anyway. I wonder about the senility of the gentleman in question. He might consider retirement from this.

[Weather-history]

Harry Kershaw has popped up again, this time in the Guardian

However, Harry Kershaw, 81, another amateur weather forecaster and former engineer from Sale, Manchester, is concerned. Using a technique known as "similarity forecasting" he scours the weather records, searching for years that had similar weather patterns to today. "From July 1 until now we seem to be following the weather of 1962," he says.

If the 1962 trend continues then we can expect an unusually cool and wet September. October will bring a brief respite, with temperatures reaching 21C, but after that it all goes downhill.

In 1963 we had one of the coldest winters of the century and Kershaw thinks we might be in for a repeat performance. Have your snow-shovels at the ready

Just a minute! This was the gentleman who said we are likely to have one of the top ten warmest summers on record and that this year was following similiarly to 1976?

Has it never occurred to him that his system is faulty?

[Thundery wintry showers]

It would help for a start if Harry Kershaw brushed up on his weather history! He cited the gales at the end of the Januaries of 1976 and 2008.

Well the last weeks of January 1976 and 2008 couldn't have been more different. One was cold to very cold and wintry (1976) and the other was very mild although it turned colder right at the end. (2008).

If you are basing your forecast on some form of pattern matching then he needs to get the very basics right!

I was also under the impression that the severe winds occurred on 3 January 1976, not towards the end of the month- another hole in the pattern matching?

Plus I don't think August 1962 was quite like this one- the stats suggest a much cooler month but with a considerably smaller shortage of sunshine.

I think pattern matching can work for short-term forecasts but is doomed for long-range forecasts as it only takes one weather variable to differ slightly and it throws the pattern match completely off course, the "snowball effect" etc.

[smich]

I was also under the impression that the severe winds occurred on 3 January 1976, not towards the end of the month- another hole in the pattern matching?

Plus I don't think August 1962 was quite like this one- the stats suggest a much cooler month but with a considerably smaller shortage of sunshine.

I think pattern matching can work for short-term forecasts but is doomed for long-range forecasts as it only takes one weather variable to differ slightly and it throws the pattern match completely off course, the "snowball effect" etc.

I agree, TWS re; short term forecasts but would add "even then on a very short timescale". It seems laughable that this "identikit" method of forecasting is udes on such a localised scale. If you were to look for a similar GLOBAL synoptic set up to the current one (at any given moment), I suspect you would search for a very long time! Hence any long range forecast would be doomed from the outset.

Steve M

[Terminal Moraine]

You would think the warning bells would ring for anyone thinking of long range forecasting using pattern matching, taking into consideration the Met' Office abandoned it in the 1970s as being little better than chance.

I still have many copies of the 'Monthly Weather Survey and Prospects' from the late 1960s and into the 1970s which contain a forecast for the month ahead and even at that relatively short range there was very little success from pattern matching.

Invariably when the forecast was reviewed and updated half way through the month there would be significant alterations because some parameter had altered early in the forecast period and thrown it into disarray.

I also tried pattern matching in the 1960s but soon abandoned it as useless.

Incidentally the gales in January 1976 occurred on the evening of the 2nd.

Edited September 3, 2008 by Terminal Moraine

[johnholmes]

I've said many times on here that pattern matching is of no sensible use be it short or long range. As TM comments above one of the main Met centres in the world abandoned it many years ago. It can be used carefully with other approaches but certainly not on its own.

[Weather-history]

He's called a meteorological marvel but I personally think he is hopeless! Harry Kershaw is going for a very mild January, one of the mildest on record followed by a much colder February ala Jan/Feb 1983

http://www.messengernewspapers.co.uk/news/...are_on_the_way/

January

1 Wet, gale force west winds 2 - 12 Wet and windy, gales 10 and 11 Gales

Way out!!

Edited January 5, 2009 by Mr_Data

[Methuselah]

If even 1% of Foggit's coldest-since-63, snowiest-since-47 predictions had come to pass, things would be better than they are!

Pattern-matched hedgehogs, that's never been done before...

[Weather-history]

It looks as though Harry has at least got the early part of February right!

[Thundersquall]

You could tell it was going to be a crap forecast just by looking at his shirt.

And the 'Craven-esque' red mac!

[Weather-history]

Bungler Harry Kershaw with yet another poor season forecast

http://www.messengernewspapers.co.uk/news/4325291.It_ll_be_warm_this_summer_predicts_Harry_/

Harrys summer predictions: May - An unusually cold month, that could be the fourth coldest since May 1902. Wintry showers and ground frost in the first half and wet and windy in the second.

1 - 3 Rain and cold wintry showers, sleet on hills.

4 - 8 Fine, cold, night frosts.

9 - 11 Wintry showers.

12 - 18 Fine, cool, night frosts.

19 - 20 Cool, heavy rain, gusty winds, sleet on hills.

21 - 27 More mild but heavy rain and strong winds.

28 - 31 Showery and windy but much warmer.

May was on the mild side!

June - Sunny and warm but chance of heavy thunder showers.

1 - 3 Cool and showery.

4 - 8 Much warmer and drier. Thunder possible.

6 - 7 Temperatures up to 26 degrees Celsius (80 Fahrenheit).

9 - 18 Warm and sunny up to 23 degrees Celsius (73 Fahrenheit).

19 - 22 Warm with some thunder and rain.

23 - 27 Dry, warm and sunny.

28 - 30 Cooler, wet and windy at 20 degrees Celsius (68 Fahrenheit).

Poor detail. The first part was pleasant, the middle section was cool, the last part warm.

July - After the first week dry, warm and sunny.

1 - 8 Cool, wet and windy.

9 - 22 Warm and sunny up tp 27 degrees Celsius (80 Fahrenheit).

23 - 24 Rain and thunder at 20 degrees Celsius (68 Fahrenheit).

25 - 27 Fine and warm.

28 - 31 Warm, rain and thunder.

Very poor. It was a washout, wetter even then July 1988 and 2007 for Manchester

---------------------------------------------------------------

August - First three weeks warm and sunny. The last week more cool and very wet.

1 - 5 Fine and very warm at 25 degrees Celsius (77 Fahrenheit).

6 - 12 Warm and wet, possible thunderstorms.

13 - 21 Fine, hot and sunny building to 27 degrees Celsius (81 Fahrenheit).

22 - 23 Heavy rain and thunder.

24 - 29 Cooler, wet and very windy with possible flooding in some areas.

30 - 31 Dry and cool, less windy, some fog.

September - Sunny, dry and warm for three weeks, last week wet and windy.

1 - 7 Warm and sunny, fodg during the early hours.

8 - 10 Fine, cooler, cloudy and some fog.

11 - 12 Showers.

13 -23 Warm and sunny, cool at night.

24 - 30 Wet and windy.

October - Warmer than average, rainfalll above average, strong winds at the end of the month.

Edited August 15, 2009 by Mr_Data

[Weather-history]

If even 1% of Foggit's coldest-since-63, snowiest-since-47 predictions had come to pass, things would be better than they are!

Its interesting point you make because is it just me or do these type of amateur weather forecasters tend to forecast extreme winters and summers?

Here's Harry Kershaw predicting the wettest April since 1818 for April 2008. The wettest April on record was actually 2000. April 2008 wasn't that wet by a stretch.

http://www.manchestereveningnews.co.uk/news/s/1042297_get_set_for_pour_april

Edited August 15, 2009 by Mr_Data

[Evening thunder]

the media quietly forget when they are wrong and just trumpet the forecast on the fairly infrequent occasions they are right.

Always have done and I suspect they always will.

I know this comment was almost a year ago, but indeed that seems to be the case.. and of course it has to be the other way round with the Met Office's forecasts! :lol:

[Solar Cycles]

I know this comment was almost a year ago, but indeed that seems to be the case.. and of course it has to be the other way round with the Met Office's forecasts!

So can any one tell me the difference between an amateur using seaweed and pine cones, and a multi million pound organastion with super computers!

[Methuselah]

Well, SC - if you cannot see the difference in accuracy between Foggit et al, and the Met - I cannot help you...

I'll ask you a simple question, though: how many 1947s have we had since 1947?????

[Solar Cycles]

Well, SC - if you cannot see the difference in accuracy between Foggit et al, and the Met - I cannot help you...

I'll ask you a simple question, though: how many 1947s have we had since 1947?????

Pete it's called irony! It was a tongue in cheek reply, to the MetOs p@@@ poor LRF abilties!

[Methuselah]

Pete it's called irony! It was a tongue in cheek reply, to the MetOs p@@@ poor LRF abilties!

:lol:

Too much vodka must have made me ironically challenged...