Photo courtesy of NBC4 Washington
Yes and no. It depends on what type of prediction you're looking for. If you're willing to settle for predictions of the near future and can tolerate predictions in terms of probabilities, the goal is already at hand. But if you want definitive, long-range predictions of the precise weather, the answer is no—we'll never be able to do it perfectly. Weather prediction struggles with two enormous problems. First, any prediction of the future is no better than our understanding of the present. Just knowing that it's sunny or raining outside doesn't count for much; to make a reasonable prediction of the future, we need lots of data about the present. We need detailed measurements of air pressure, temperature, wind speed, humidity, and so on at an enormous number of locations and altitudes. We even need to know what people are doing because people influence weather. But the biggest problem in weather prediction is that the atmosphere is essentially chaotic.
In this context, chaos means that a very slight alteration in the atmosphere's present arrangement will lead to a radically different arrangement just days or weeks down the road. Chaotic systems are exquisitely sensitive to initial conditions and the futures of two nearly identical arrangements of a chaotic system will diverge exponentially with the passage of time.
In the case of the atmosphere, it is said that just one flap of a butterfly's wings will eventually alter the weather of the entire Earth. The time-scale over which these chaotic effects cripple weather prediction are days and weeks. The better you understand the present, the longer your prediction can hold out against chaos. However, exponential growth eventually beats down even the most ambitious efforts—each additional day of prediction accuracy requires that you measure the present many times more accurately than before. Beyond about one week, the measurement requirements become overwhelming. At that point, the best anyone can do is to predict weather probabilities. Such statistical predictions are all that's possible based on imperfect understandings of the present.
Answered by Lou A. Bloomfield of the University of Virginia.