Can we forecast the weather more accurately?

I don't know if you have noticed, but the weather forecast has made great progress recently. In the past, some big storms were unpredictable, and they ravaged several counties as soon as they came; Today's forecasts are often released a few days in advance, and even the daily weather forecast is getting more and more accurate. In the past few decades, rain has not been predicted. How many times have you met?

Historically, weather forecasting has always been regarded as an inaccurate art. Before19th century, it was impossible to collect data in a large area by telegraph, and weather forecasting did not exist at all. Until 1950s, weather forecast was based entirely on experience and historical data. Forecasters compare what they see on the weather map with the image data in a series of meteorological records, and then find out the overlapping places, and then make predictions.

Many folklore about climate is based on the above experience. When I used to keep bees in central Virginia, my own method was to record the date when bees drove drones out of the hive every autumn. The sooner the drone is expelled, the colder this winter will be! How ridiculous! It seems that any stupid method can be used as a weather forecast!

In the 1950s, the complex equations governing various movements of the atmosphere could be solved by computers for the first time. Scientists don't expect the weather to be the same forever, so they only pay attention to history and experience, but use various physical and chemical laws to predict what will happen in the weather. In the second half of the 20th century, the scale and accuracy of computer prediction have been continuously improved.

However, two basic problems of computer prediction have not been completely solved: first, until recently, there were not enough data input programs. Temperature and speed are only recorded at a few very distant points (such as airports), so it is not enough for computers to make accurate predictions based on these data. Today, at least in industrialized countries, we can get more and better data because of widely distributed ground observation networks and satellite observation stations.

The second problem is more important and seems to be related to the system established in the atmosphere. I mean, all kinds of movements in the atmosphere can be chaotic. One of the best metaphors is fame? Butterfly effect? (Butterfly effect), that is to say (at least in principle), the atmosphere is highly sensitive. For example, a butterfly flapping its wings in Kolkata can produce a series of chain events, which may eventually lead to a storm in Rio de Janeiro.

From a practical point of view, the chaotic behavior of the atmosphere means that it is difficult to make long-term weather forecasts. For example, if we calculate the seven-day weather forecast based on the model at six o'clock in the afternoon and do it again at nine o'clock in the afternoon, the two predicted climate models may be different because the starting point of calculation is different. This is not a problem that can be solved by a better computer, it is a problem built in the earth's atmosphere. Because of this, it was not until 1995 that the National Weather Service was willing to make a detailed forecast for the next three to five days.

But by 1996, the meteorological bureau began to increase these forecasts to seven days, and may even exceed this limit. Obviously, these new predictions are based on sufficient new technologies. Scientists no longer let an ultra-accurate computer run with the best data, but execute the same program many times, changing the startup time or making minor changes to the startup conditions every time.

For example, instead of choosing from 6 pm or 9 pm-based forecasts, they execute these two forecasts. All in all, the program of the National Weather Service needs 46 different computers to perform the operation. If all the results predict rain (for example), it is assumed that it will rain. The idea is that most or all execution times show better performance than the results of stand-alone operation. This technique is called "ensemble averaging", and it puts forward a very reasonable method to extract useful information from chaotic systems.

Similar efforts have been made in long-term climate prediction. At present, the longest forecast made by the National Weather Service is 15 months. Here, on average, the same program is no longer executed many times from different starting points, but goes through three very different programs, each of which needs to describe some aspect of long-term weather. Next time, if all three methods have the same prediction, then this prediction is considered to come from the basic dynamics of the atmosphere, rather than obsessing over the details of the analysis method. So, in the next few years, we should be able to see whether the weather forecast of the National Weather Service is better than my bee.