Using numbers to project...following on from Snowmaiden's points

[Stratos Ferric]

Point taken, you are dismissing analogues altogether.

However in terms of odds, whilst the historic data supports them, probabilities are of no use in forecasting are they? They can only provide indicative figures based on the preceeding (fill in the blank) years.

If the warming trend were to reverse (I am not saying it will at all), then the probabilities are seriously flawed until corrected, the time it takes for them to be corrected by recent historical analysis makes them unwieldy at best.

Similarly an acceleration skews them.

Statistics are very useful at showing that the world is warming, at proving beyond doubt that we are in a warming trend. They are utterly useless at telling us what the winter will be like, or the summer, the spring or the autumn.

Statistics - is/was

Analogues, synoptic patterns, teleconnections - why

Forecasting has to be based on more than probabilities, the two are indeed incongruous, one is by definition forward looking, the other utterly reliant on the historic.

SM posted this in the CET thread, and raises interesting questions about using trends to project, some of which I agree with.

A reasonable analogy here is financial markets, which operate in ways that are broadly - though not specifically - predicatable in the short term but largely unknowable beyond 10-15 days out. One branch of market analysis is charting, which uses a number of rlling averages (and some other mor complex comparisons and "rules") to inform stock trading decisions based on recent performance. Not a perfect science at all, but sometimes a good indicator.

The problem with charting is that short term blips are hard to forecast. You can easily say that Chelsea should win the premiership again this year, but not know whether they will win on any given weekend, though on balance they are likely to, and if they lose they are very unlikely to get hammered. So it is at present with winters; they are likely to be warmer than the recent past, but individual ones may be cooler, though it's highly unlikely that they'll be as cold as they were occasionally in, say, the 60's or early 80's.

Interestingly, I recall some analysis I read a year or so ago that was a scientific assessment of forecast accuracy in the US, and the conclusion was that a rule of thumb something along the lines of "if the weather today is the same as yesterday then the odds are tomorrow will be the same, if not then expect it to change again" was as accurate, and more, than detailed forecasts. And if you think about it that kind of assessment will tend both to stack up, and extrapolated several orders of magnitude explains SP's argument regarding the problem of using trend to predict change. The fact is that a curve has to start somewhere, and looking back down the line behind you is not the best indicator of when it might happen.

Where I have to disagree with SP is the suggestion that simply because stats are historic they cannot project the future (I use "project" rather than "predict" quite specifically there), and I wonder how much of this is driven by understanding and how much by hope. Why do I say that? Well, SP I'm sure hopes for a cold winter, but retro fitting recent trends makes a cold (in absolute terms) winter improbable, if not unlikely, hence perhaps the tendency to debunk when numbers imply an outcome we don't personally favour. Returning to the start of the curve, there can sometimes be indications that a cycle has topped out, but this requires knoweldge of how the trend has behaved in the past. Hence, at present, there is some excitement about the potential for this year to be theofourth consecutive year of cooling. A chartist might look at this on the basis that such a trend indicates the start of a long drop. Interestingly, the CET has NEVER fallen for more than four consecutive years. However, magnitude of change is as important as direction, and on the basis the current reversal is really little more than a pause, the total drop across the past three years is less than around 90% of the SINGLE YEAR drops recorded.

The reason why I KNOW stats have some validity is because they do indicate a big picture direction. We know winters, in particular, are notably warmer than they used to be, and that this is no longer a statistical quirk; we can say this with some certainty because the trend has been maintained for a long time, and because it can also be explained by other driving factors (e.g. SSTs), and is corroborated (ruling out measurement error) by, for example, changes in flora, fauna, and growing cycles. We also kow that sudden and dramatic changes in either direction, without some forcing of the system (e.g. major volcanic eruption), are pretty much unknown in the measured record.

So, what do we have with trends? At present, hereabouts, we have a warming climate, particularly so overnight and in winter, but we know that month to month we can have fluctuations. Statistically, therefore, we can say with some certainty that whilst we don't know whether this winter will be warmer or cooler than the the past five or ten, say, we can say with a high degree of certainty that it will not be anything like as cold as the ones I remember from my youth in the late 70s and early 80s. As to SP's inference that forward forecasting is better, the one thing that can be said about the stats is that they are indisputable; given that our ability to forecast forward more than 5 days based purely on synoptics is at best tenuous, then extrapolation based on something that is known is at the very least a useful starting point. In practice, the sensible forecaster will use both. I want to know how my team is playing, but I also need to know who I'm drawn against. In this context hoping for a cold winter this year is something like being, say, a Cardiff City supporter ahead of the third round of the FA Cup; my team is playing as well as it has for years, but when I see I'm drawn against Chelsea it doesn't mean I'm about to advance.