The Edge of Predictability: Long-range Weather Forecasting

Some factors and complexities in predicting drought. The influence of ENSO (El Niño-Southern Oscillation) was discussed at a July 2007 NCAR colloquium. Useful weather forecasts were once limited to a period of two or three days. Now, thanks to improved computer models, there is measurable skill more than a week ahead. However, if a forecast period is extended much further, the outlook becomes no better than chance, because of the chaos effect—small errors that grow over time. Even when seasonal forecasts are skillful (see seasons to years), it's impossible to say what weather will materialize on a given day several months out.

Scientists at NCAR are working to pin down the edge of predictability. It once appeared the limit was 10 days. Computer models can now project large-scale weather features up to 15 days out, although the skill dwindles to negligible levels at the far end of that range.

The Science of Predictability



Click to enlarge illustration. This graph illustrates the concept of predictability. Weather events come in a great variety of sizes (spatial scales). Typically smaller-scale features evolve more rapidly and so are said to have smaller time scales. Conversely, larger-scale systems such as hurricanes, or the weather patterns that produce droughts, have much longer time scales. The theory of predictability says that there is an inherent finite time limit for a skillful weather forecast and that this time limit depends on the scale of the weather system of interest.

A fundamental objective of predictability research is to discover the predictability limits for important weather systems. The blue line in the graph illustrates a guess of where the predictability limit for the depicted weather systems, while the red line indicates our current skill level. The more accurately we can determine the position of the blue line relative to the red line, the better we will know how much room for improvement there is, or in other words, when a forecast system is performing as accurately as possible. Scientists at NCAR are striving to both discover the theoretical predictability limits and to push our current forecast skill level up towards those limits.


Ensemble techniques are a valuable way to extend forecast range and quality, especially for longer-range periods. At NCAR and elsewhere, ensembles are created using several simultaneous runs of the same model. Researchers randomly tweak the initial conditions for each run, spanning the range of error known to be present at the starting line. There is no way to tell in advance which of the 10 forecasts in an ensemble will wind up being closest to correct. Still, the actual weather usually ends up within the ensemble range. Forecast centers in several countries now produce ensembles as part of their standard lineup of products. Through retrospective studies of ensembles, scientists can gain insight on how model errors grow.

Longer-term forecasts of one to three weeks may gain skill as they focus on regime shifts, the transitions into and out of weather patterns that persist for a week or more. NCAR scientists are examining whether there might be preferred modes of the atmosphere, such as configurations of the jet stream that tend to be locked into place—a long-sought guide for extended forecasting. However, early results hint that such modes may not serve as a useful forecast tool. Scientists are also exploring how nonlinear events (those that grow more quickly than the features usually tracked by large-scale models) affect regime shifts and dictate the limits of predictability.