We reach a new year and so time for a brand new annual CET competition in addition to the normal monthly competition. I will start with a reminder of how the competition works and also set out some new rules to accommodate those who miss a month or two during the year. This should enable as many people as possible to remain in the competition throughout the whole year.
How do the competitions work?
Each month a new thread is opened a few days in advance of the start of a new month. Your challenge is to predict the mean Central England Temperature (CET) for the following month. The CET is the longest running temperature series and goes back to the mid 1600's. Each month I post some statistics about the CET for that month in recent years as well as a summary of some of the latest model output and long range forecasts to provide some context to assist you in making your prediction.
Predictions can be made either to one or two decimals (the actual data is published to two decimals). The closest person or persons to the final figure at the end of the month are declared the winner. There are no prizes. This competition is just a bit of fun and the chance to amaze your fellow posters with your skill and foresight in predicting the weather (or you can just make a random guess!).
We use the Hadley data for the purposes of this competition which is maintained by the Met Office. Provisional figures are published daily with a final adjusted number usually issued on the first day of the following month. Other CET series are available (such as that maintained by Philip Eden) but are not always updated daily which is the main reason why we use Hadley.
In addition to the monthly competition there is an annual competition which runs throughout the year (this is different to the one off up-front annual CET prediction competition which is in a separate thread). The annual competition keeps a running total of the cumulative prediction errors from each month for each person. Absolute differences are used so it makes no difference if you overpredict or underpredict the CET in any given month. At the end of the year the person with the lowest cumulative prediction error after totalling each of the 12 months is declared the winner for the year. This is always a hard fought contest and requires a consistency of good predictions in order to win.
Summary of important rules
- Predictions must be made by 23:59 of the final day of the month in order to avoid penalties
- All predictions should be made in the CET thread for the relevant month where possible (except for the final two months of the year - see below). As the thread is usually only opened 4 days before the end of the month, if you expect to be away or unable to post for any reason you can send me your prediction earlier by way of private message and I will post it in the thread when it is opened.
- Predictions for November and December are to be made by private message to me. This is to avoid tactical predicting at the end of year rather than predictions based on expectations of the weather. The purpose of the competition is to predict what you think the temperature will be not to protect your position in the league table.
- Once you have posted your prediction you are only permitted to change it once. Any subsequent amendments will be ignored. No amendments are allowed after 23:59 on the last day of the month
- You may make a prediction to either one or two decimals. You can predict the same figure as someone else.
- Late entries will be accepted up to 23:59 on the 2nd of the month. Late entries are subject to a penalty for the purposes of the annual competition. For 2013 the penalties will be 0.2C for each day the entry is late (i.e. an entry made on the 2nd of the month will incur a penalty of 0.4C).
- Where two members have the same cumulative prediction error for the year at the end of any month, the order of ranking in the competition table will be determined by which person has had the closest prediction to the actual CET in any month. If there is a still a tie I then look to the second closest prediction for each person and so on until the tie is broken. If the tie cannot be broken (very unlikely by December but can happen in the first couple of months of the year) then each person impacted is awarded the same position in the table.
New rule for 2013
In previous years anyone who did not make a prediction in every month of the year became ineligible for the annual competition. There have been a number of requests for this to be changed as there are sometimes unforeseeable reasons that prevent people making a prediction in time. I have devised a new rule to accommodate this request which hopefully does not give those missing predictions any sort of unfair advantage. The new rule will work as set out below.
If a person in the annual CET competition does not make a prediction in any one month (including January) their prediction error for that month will be calculated as follows:
- the upper quartile (75th percentile) of the prediction errors of all the people in the annual competition who did make a prediction for that month will be calculated;
- the normal penalty for a late prediction on the 2nd of the month (i.e. 0.4C) will be added to the figure above
The result of the above calculation will be used as the CET prediction error for any person who misses a prediction in a particular month subject to one further point:
- If, for any person, their average monthly prediction error for all the previous months in the year to date is higher than the 75th percentile of errors (for those that did make a prediction) in the month the person in question made no prediction, then the average prediction error will be substitued instead of the 75th percentile figure. The 0.4C penalty will still be added on top.
This override is only likely to apply to those people towards the foot of the table. Let me give you an illustration from the 2012 numbers. Lets say vince had failed to make a CET prediction in November. The 75th percentile of errors for those that did make a prediction was 1.30C. The average prediction error for vince in the months January to October was 1.86C. So in this situation vince would be given a prediction error of 1.86C+0.4C=2.26C for November if he had failed to make a prediction. In fact he did make a prediction and had an error of 1.45C.
The above may sound a little complicated but it is actually quite simple to apply as I can easily calculate the upper quartile figure in my spreadsheet. I didn't think it would be fair to just give anyone who missed a prediction the same error as the worst prediction that month. Equally going with the median error is not really fair to everyone else in my view. Hence why I have gone for the upper quartile figure.
One final important point. No one person may miss more than two predictions during a calendar year. If a third prediction is missed that person will be eliminated from the competition (otherwise it would mean 25% or more of the annual error being calculated on non-existent predictions which would reduce the credibility of the competition too much).
Lets see how this works out in 2013. Hopefully it strikes a reasonable balance but it can be changed again next year if necessary. I will not make any changes to the rules during the year.
January historic data summary
Here is the usual rundown of historic data for previous years.
12 of the last 20 January's have seen a CET more than 0.5C above the 1971-2000 mean. Only 3 were more than 0.5C below average including 2009 and 2010.
Long run averages:
1971-2000: 4.2C
1981-2010: 4.4C
1988-2012: 4.8C
In 2012 the January CET was 5.45C. in 2010 it was just 1.39C. In 2007 we saw a CET of 7.0C which was the warmest since 1921.
Here is a chart of the January CET since 1961 with a 10 year moving average
Here are some details of what the models and forecasters generally are saying at the moment
GEFS
http://www.wetterzentrale.de/pics/MT8_London_ens.png
http://www.wetterzentrale.de/pics/MT8_Manchester_ens.png
The first 10 days of the month look to be well above average but this is not necessarily reflective of what will happen in the rest of the month. This time last month the GEFS showed below average figures for the whole of the first 12 days of December and this did indeed turn out to be correct although since then it has been very mild.
ECM ensembles (De Bilt)
http://www.knmi.nl/exp/pluim/vijftiendaagse/index.html
Very mild throughout.
Weatheronline
http://www.weatheronline.co.uk/cgi-app/reports?MENU=Month-ahead
No recent updates. But their forecast looks quite accurate for the start of January - i.e. mild. Possibly turning cooler mid month.
Netweather
http://www.netweather.tv/index.cgi?action=winter-forecast-2012;sess=
Prolonged cold spells
Met office
http://www.metoffice.gov.uk/media/pdf/c/2/A3-plots-temp-JFM.pdf
Slightly cooler than average most likely scenario but high degree of uncertainty.
http://www.metoffice.gov.uk/public/weather/forecast/
Average or above average but possibly turning cooler late in the month
Pattern matching (just for fun)
I looked briefly at years since 1900 with mean Oct / Nov temperatures of 8C to 8.6C (2012 8.26C) and a December CET between 4.3C and 5.3C.
There are only 8 matches so this year's pattern is quite unusual.
3 years had a January CET within +-0.5C of the 1971-2000 mean. One year was very mild (1930 at 6.3C). 3 years had a CET between 2.5C and 3.4C. Finally there was one very cold year (1958 at 1.6C).
So not much to go on there but overall a slight bias towards cooler than average conditions (very similar to the Met Office contingency planners summary).
Low confidence though and given the very mild first 10 days expected from the latest model guidance I would be surprised if the CET finished more than 0.5C below average. More likely to be above average I would think.
Edited by moderator
01 January 2013 15:44:06
|
Reason: Not specified