Predicting the past

[doodab]

The high-top model was devised in time for the winter of 2010-2011.

Using its data, the Met Office forecast in autumn 2010 that there was a 40% chance of a cold start to the winter, with a 30% chance of a mild start, and a 30% chance of an

That's an improvement?

[IR35FanClub]

I think the only thing that you be certain about in forecasting is that putting some real data into the model will expose the limitation of your model so you need more hardware and a revised model to run it.

The only way to get a truely accurate forecast is to build a life size earth and create the starting conditions as they are now, but you'll only be able to forecastabout 1 minute ahead as you'll need to update the data in real time with actual inputs, like sun and butterflys flapping their wings from the real earth.

Anything else is an approximation and liable to fail to predict correctly once in a while.

What we can say is that if a model is given a set of historical starting figures and can accurately predict past weather over longer periods, then it is likely to be fairly good at near to mid term forecasts, but not necessarily long term (thats where you get into climate modelling instead of weather forecasting).

Boinc have the climatepredicition screen saver - where volunteer computers have been running thousands of climate models with slightly different algorithms. They start them back in the 1800's and see if they end up with 2012 having the artic seas boil off and the uk at abosulte zero etc. They can then discard those models and refine others. The one that don't generate carp, not a fish, are left to see what happens in future.

I've got one but it's only up to 1880. there's a visualiser which show a resolution of 5km iirc, the weather looks farily normal so far. UK is cloudy and raining in summer, and theres still ice in the arctic.

that's the same for any kind of forecasting - economic - whatever. To try and forecast stock market predictions, you'll need to have several billion people on hand and the news and weather for the next 20 or so years.

[IR35FanClub]

I think the only thing that you be certain about in forecasting is that putting some real data into the model will expose the limitation of your model so you need more hardware and a revised model to run it.

The only way to get a truely accurate forecast is to build a life size earth and create the starting conditions as they are now, but you'll only be able to forecastabout 1 minute ahead as you'll need to update the data in real time with actual inputs, like sun and butterflys flapping their wings from the real earth.

Anything else is an approximation and liable to fail to predict correctly once in a while.

What we can say is that if a model is given a set of historical starting figures and can accurately predict past weather over longer periods, then it is likely to be fairly good at near to mid term forecasts, but not necessarily long term (thats where you get into climate modelling instead of weather forecasting).

Boinc have the climatepredicition screen saver - where volunteer computers have been running thousands of climate models with slightly different algorithms. They start them back in the 1800's and see if they end up with 2012 having the artic seas boil off and the uk at abosulte zero etc. They can then discard those models and refine others. The one that don't generate carp, not a fish, are left to see what happens in future.

I've got one but it's only up to 1880. there's a visualiser which show a resolution of 5km iirc, the weather looks farily normal so far. UK is cloudy and raining in summer, and theres still ice in the arctic.

[DimPrawn]

Met Office better placed to predict 'big freezes' - Telegraph

They can now predict the freezes in the past couple of years with some certainty. These guys are amazing.