On Tuesday, April 8, 2014 1:34:43 PM UTC+1, Dave Cornwell wrote:
Dawlish wrote:
On Monday, April 7, 2014 10:45:59 PM UTC+1, Dave Cornwell wrote:
Well I did, with the proviso that there was no consistency in the
charts. I should have specified the date as 10 days from 11.50pm on the
26th could be construed as the 5th, 6th or 7th April really depending.
Anyway I said "on the cool side with a northerly element, not very
windy, quite damp and not very sunny."
This would have been wrong for the 5th as it was still quite warm here
and under the influence of the fading HP. The 6th was indeed showery and
cloudy but not that cool. The 7th today has been damp with intermittent
rain, not sunny and much cooler, but not cool for the time of year. It
is stretching it to say there was a Northerly element but a cooler
westerly now. It's not been very windy for any of those days.
From what I remember of the predicted charts they are quite different
but not hopelessly so, but the pattern is brief and transient.
So all in all it may appear not that bad but as the evolution has been
quite different I would say any accuracies were down to luck, which is
what I believe anyway unless there is a good deal of consistency in the
output from the models. That is why I stick to a five day forecast on my
website.
Dave, S.Essex
www.laindonweather.co.uk
Yes, it would have been wrong for the 5th. Such forecasts have to be date specific, or we get into Piers territory, stretching the outcome date to days before and days after the originally stated date, to claim a correct forecast. My own are always for 10 days after the forecast, not 11, or 12, or even 9!.
You were right to say what you did about inconsistency in the charts making forecasting at 10 days pretty impossible when I set this challenge and it was the reason I also didn't think it was possible at the time to forecast with an 80% chance of being correct.
----------------------------------------------------------------------------------------------------
My recollection was the charts showed a more or less complete breakdown
of the HP leading to a slack LP set up, mainly over the North Sea,
giving a northerly. (Generally speaking). The truth was a more transient
and short lived westerly with HP returning.
Yes, that's right about the 5th.
I can't remember if you state the actual day the forecast is for but if
not do you then count the next day as day one and day 10 as the day of
the forecast? You can see the ambiguity risk.
Yes; accepted. I try to avoid that by sometimes forecasting in the morning and sometimes (though more often) in the evening. However, most times I judge my own forecasts at exactly 10 days, say, after the 06z, or the 18z output at T+240. There is bound to be a little overlap, to 10 days, at forecast, but I always state the day for which I'm forecasting in the **Forecast** title" and I do try to come back to every forecast 240 hours after I've made it.
Of course the other problem is that this is just really proving that
statistics work! The less the standard deviation between the models the
greater the confidence level in the forecast.
If only the MetO would use confidence levels, but I actually never see less than spaghetti at 10 days on the ensembles, so i don't think that using measures of dispersion around the mean would help, in this respect, would it? I do use ensembles alongside the operationals, but I don't think the ensembles show any narrowing of the plume on days that I forecast. All you could do (and I do) is to say that the most likely pattern at 10 days has consistency from the last few runs. I get my 80% accuracy from the percentage accuracy outcomes over 140+ forecasts.
I suspect that Student and his T test told us just that already!
Dave