By that I mean: how far into the future can we forecast specific weather features--perhaps the intensity and position of a low pressure center or the position of a front or whether they will be rain during the early morning hours. Such forecasts contrast with climate prediction, where we attempt to predict the general nature of the weather over a period...say the average temperatures for the spring.
Lately we have increasingly seen forecast skill extend well past a week...and in some cases, extending to nearly two weeks....and in this blog I will show you a few examples.
A lot of research was done on this topic--determining the limits of predictability-- both from theoretical and modeling perspectives. For example, meteorologists such as Edward Lorenz and Joseph Smagorinsky would run weather forecast models many times, each starting from a SLIGHTLY different beginning--well within observational error. They would see how long it would take before the forecasts would become very different. Generally, it took about two weeks, but there is nothing unique about that period. Sometimes forecasts would diverge after a few days, while in other situations the forecasts held together much, much longer. The implication is that during some periods we find that forecast skill, even with a perfect forecast model, is lost quickly, while in other periods we might skillfully predict the weather for weeks.
 |
| The ability of a small change in the initial forecast state to eventually cause a big change in a forecast is known as the butterfly effect, since the atmospheric disturbance of a butterfly flapping its wings would eventually alter a forecast a long time into the future. |
In the early days of forecasting, when our forecast models were crude and our observational resources were few, weather prediction could not get close to the theoretical limits. But during the last decade or so, with the availability of satellite data, fast computers, and far better computer models, prediction skill has extended further and further out in time. In the 60s-80s we were lucky to skillfully forecast out 2-3 days. In the 90s 3-4 days. And in the 2000s, 4-5 day forecasts were often quite skillful. But recently, we have seen highly skillful forecasts consistently at 5-7 days, and occasionally approaching two weeks.
Want to see some examples? Here is a 360 hr forecast (15 DAYS) from a National Weather Service ensemble weather prediction system for the height of the 500 mb pressure surface. Specifically, it shows the predicted anomaly (difference) from normal of the mean of a large collection of forecasts (the ensemble). Note the HUGE (orange color) high height anomaly in the north Pacific.
Next, here is a 192 hour (8 day!) forecast valid the for same time. The feature is still there.
got a number of other features correct...or nearly correct...as well.
Or examine the 240 h (10 day) forecast by the European Center for Medium Range Forecasting and compare it to the 24 hour forecast for the same time. The images below show 500 hPa heights and 850 hPa (around 1500 meters) temperatures. Not perfect, but a number of features are very similar...useful forecast skill there!
Or a recent example--a forecast at 500 hPa made 7.5 days ago for today at 5 PM....and a 12-h prediction for the same time. A lot of the same features, although there are some minor differences in phasing and amplitude.
The general consensus is that we will see such skill extend out further in time, by at least one day per decade. There is a limit to this extension of forecast skill...but there is much we can do to improve our predictions more--so improvements should continue for a while.
So when people ask you how far into the future meteorologists can predict the weather, a good answer is:
2-3 days with excellent skill
3-4 days with moderate, but useful skill
5-6 days with marginal skill
..and occasionally skill extending out 7-10 days.
Yes...sometimes forecasts go wrong in less than a day...but is considerably less frequent than a decade ago.
