Posted February 16, 2009
That is the typical time span covered by weather forecasts shown by broadcast meteorologists and the standard followed by the National Weather Service at this point as well. A viewer named Traci asked why meteorologists only forecast for seven days, and while the answer could get very long, the mercifully brief version is that seven days is very roughly a point at which the average accuracy of forecasts (at least those specific enough to include a low and high temperature and a statement as to expected cloud cover, potential for precipitation and precipitation type) drops off to about that of several alternatives to a forecast.
Those include 1) Persistence, or assuming that whatever the weather is now will remain the same day after day, 2) Climatology, in which you would assume that the weather on date XX will be the average of what has been observed in past years on that same date, or 3) Guessing, even if it is a non-random guess constrained by realism. I know some of you are already thinking to yourselves that sometimes it seems like option C sometimes applies anyway! Obviously, we (meteorologists as a profession) would like to put out forecasts that, on average, are better than any of those three alternatives.
In some circumstances, that can be difficult to accomplish even for a three or four day period, while under some other, more static conditions, reasonably good forecasts that achieve the goal of improving over 1/2/3 above out to nine or ten days are possible. Most often, though, this breakpoint of utility falls somewhere in the 5-7 day range, and seven days as a typical forecast period has emerged in the last year or two as a fairly standard compromise between the desire to have the forecast accurate as much as possible (which argues more toward a shorter period) and the desire for viewers/web visitors to always be able, for example, to see a forecast on Monday that extends into the following weekend.
This is a small slice of what goes into a forecast, but I thought one of today's forecast model outputs would provide a nice example of why we wouldn't routinely extend forecasts to 9 or 10 days. What I'm showing in the attached images is output from the Global Forecast System (GFS) model ensemble run. In an ensemble, the same model is run many different times, with many of the runs "perturbed" slightly, meaning that small differences are introduced in some of the variables the model is initialized with, differences that are consistent with the level of error that might occur in measuring atmospheric variables at weather stations, with radiosondes, aircraft and so on. It can then be instructive to see how much difference these small perturbations cause in the evolution of each "member" of the ensemble.
In the attached images, the first is the initial state of all the ensemble members, in this case showing two height contours of the 500 mb pressure surface, a representation of pressure at about 18,000 feet above the ground. The map shows two of the height contours, and their shape helps illustrate the location and intensity of upper air troughs and ridges. The yellow lines are the "control" model run, that is with no perturbations applied, and the gray lines are results from the same model run 12 hours earlier (this allows checking for consistency from one run to the next, or conversely for sudden changes that might lead to a loss of confidence in the results). The green lines are long term climate averages for the position of those contours, and can give a rough idea of whether pressures aloft are lower than average or vice versa. This can also give a sense of whether temperatures may be lower or higher than average in that vicinity. All of the blue and red lines are the results from the perturbed models runs.
The general idea here is that if many of the perturbed results all remain in good agreement, and the previous model run (gray lines) and current model run (yellow) continue to agree closely with most of those blue and red ones, you have some reason, though not a guarantee, that the results are probably pretty meaningful and characterize the evolution of the atmosphere rather well. On the other hand, if or when the results turn to a jumble (hence the name "spaghetti plot" for this display), whether in a particular region or across the whole model domain, that confidence is good deal lower. I'll leave it to you to page through the four attached images from last night's model run and see what happens to the pattern. The images show the initial state, a three-day forecast, a six-day forecast, and a nine-day forecast, respectively...