Forecasting is still all about chance
Study thoroughly and then play the probability game.
By Karsten Shein
Comm-Inst, Climate Scientist
Boeing 737 on final approach in the shadow of developing storms. Forecasters can usually give pilots a pretty good idea that storms are likely, but pinpointing where they will pop up is not as easy.
We've all heard the TV weather forecaster call for, say, a 70% chance of rain. Then either it rains on you or it doesn't. If it doesn't, we joke that meteorology is the only job you can have where you can be wrong half the time and still get paid.
But just because it didn't rain on you, did the forecaster get it wrong?
How about if a briefer told you there was a 70% chance of encountering thunderstorms along your route? You fly the route and, although you don't have to dodge any cells, you see them popping up in the distance.
Would you say the briefer was right or wrong? What if the briefer gave you instead a 30% chance, but you had to deviate around an anvil-top monster? Did they get it wrong then?
In both cases, the briefer was right on. There was a chance of storms, and storms did in fact materialize. Unfortunately, like just about every forecast—weather, stocks, sports—we are not at the point where we know everything we need to consider.
Perhaps in the next 100 years we will know exactly how a molecule behaves, and have computers that can track and predict the motion of every molecule in the atmosphere—but even then we're still faced with the probability of some random input that will mess things up.
For example, all indications might point to the best team in the league beating the worst. If all inputs remain known, that will happen. But what if the best team is the best largely because of one star player? What happens to the odds of winning if that player is injured a few minutes into the game?
Likewise, weather forecasting is probabilistic rather than absolute because there are many atmospheric controls that are not completely known, are too complex to model with present computing capabilities, or are simply random.
We have a pretty good handle on the conditions necessary to create a thunderstorm, but what we don't have are enough observations of the atmosphere to map every square kilometer of it. And if we did have those observations, we don't yet have the computing power to run models that would be able to take all those observations and tell us quickly where those storms might pop in the next hour.
Even then, small errors in observations or calculations would toss a tiny amount of error into the model, which at the scale of a square kilometer could easily throw the forecast off—with a storm either not developing, or developing a few kilometers away from where it was forecast.
These are the basic reasons why all forecasts are probabilities instead of certainties. We have good quality high-frequency (sub-daily) surface observations at perhaps 10,000 weather stations spread across the roughly 501 million square kilometers of Earth's surface, and many of those stations are hundreds of kilometers from their nearest neighbor.
Upper air observations are even sparser, with the several billion cubic kilometers of the troposphere and stratosphere served by fewer than 1000 weather balloon stations worldwide.
Graphical Airmet display from ADDS. Airmets and Sigmets present important information about weather that may affect flight safety during the valid times, but they don't contain information regarding the probability of occurrence.
Between these surface and upper air stations, the computer models or weather forecasters must guess at what the conditions are.
These guesses are in most cases going to be reasonably close, but they will never be exactly accurate. Similarly, we often do not fully know the complete behavior of some things in the atmosphere, so we come up with approximations that account for much, but not all, of that behavior.
In other cases, we know a lot about how something behaves—like clouds—but the physics are so complex that, in the interest of providing a timely forecast, the weather models use a scaled-down version of the physics, which produces reasonable, but not exact, simulations.
As computing power and scientific understanding increase, the models' ability to account for the atmosphere's behavior accurately will improve. But, unless we are able to know the present conditions between weather stations more accurately, there will remain a source of uncertainty in the forecast.
Unfortunately, in addition to there being uncertainty in the forecasts, there is a lot of uncertainty among the public about what those probabilities mean. What does a 40% chance of rain mean? In a point forecast (ie, one for a specific location), it means exactly what you might think—that it is slightly more likely not to rain than to rain.
But the forecaster can't say for certain that it won't rain, perhaps because the atmosphere is exhibiting a clash between conditions that would suppress rain and conditions that would favor it. Or perhaps they've looked at similar days in the past, and about 40% of the time those similar days have produced rain.
In an area forecast (ie, one that covers a larger area of Earth's surface, such as a county or a river basin), the probability means something slightly different. It means a 40% chance of rain falling somewhere within the area. The probability of it falling at a specific location within the area may be far smaller. This is why, even if the area forecast called for a 70% chance of rain, you still might not see any where you are.