The danger of leaving weather forecasts to artificial intelligence


Humans have tried Predicting climate change over thousands of years, using early legends—”the red sky at night” is a sign of optimism for weather-weary sailors. This is actually related to the dry air and high pressure in an area—and from Roofs, hand-drawn maps and local rules of thumb. These future weather forecast guides are based on years of observation and experience.

Then, in the 1950s, the famous mathematician John von Neumann who assisted the Manhattan Project many years ago and the atmospheric physicist Jule Chaney, often regarded as the father of dynamic meteorology, Charney) led a group of mathematicians, meteorologists and computer scientists-tested the first computerized automatic prediction.

Charney and a team of five meteorologists divided the United States into (by today’s standards) fairly large regions, each of which is more than 700 kilometers in area. By running a basic algorithm that obtains the real-time pressure field in each discrete unit and predicts it within a day, the team created four 24-hour atmospheric forecasts covering the whole country. It took 33 days and nights to complete the forecast. Although far from perfect, the results are encouraging enough to revolutionize weather forecasting and push the field to computer-based modeling.

In the decades that followed, billions of dollars in investment and the development of faster and smaller computers led to a surge in predictive power. Models are now able to explain the dynamics of atmospheric masses as small as 3 kilometers, and since 1960, these models have been able to include increasingly accurate data sent from meteorological satellites.

In 2016 and 2018, GOES-16 and -17 satellites were launched into orbit, providing a series of improvements, including higher resolution images and accurate lightning detection. The most popular numerical models, the United States’ Global Forecast System (GFS) and the European Center for Medium-Range Weather Forecast (ECMWF) have released major upgrades this year, and the development of new products and models is faster than ever. With just one tap, we can access amazingly accurate weather forecasts about our exact location on the surface of the earth.

Today’s lightning forecasts are the product of advanced algorithms and global data collection, and seem to be only one step away from full automation. But they are not perfect yet. Despite expensive models, advanced satellite arrays, and large computers, human forecasters have their own unique set of tools. Experience—their ability to observe and map connections in the algorithm’s inability—provides these forecasters an advantage to continue to surpass the dazzling weather machines in the highest risk situations.

Although very useful According to Andrew Devanas, an operational forecaster at the Office of the National Weather Service in Key West, Florida, for macro forecasts, the model does not respond to small updrafts in the small land quadrants that indicate that waterspouts are forming. sensitive. Devanas lives near one of the most active waterspout areas in the world. Ocean tornadoes can damage ships that pass through the Florida Strait# and even land ashore.

The same restrictions hinder the prediction of thunderstorms, extreme precipitation, and land-based tornadoes, like those Tear It crossed the Midwest in early December, killing more than 60 people. However, when tornadoes occur on land, forecasters can usually find them by looking for the characteristics of the tornado on the radar; waterspouts are much smaller, and there is usually no such signal. In a tropical environment like the Florida Keys, the weather does not change much every day, so Devanas and his colleagues had to manually check for changes in the atmosphere, such as wind speed and available moisture-these changes are not always considered by the algorithm-look Are there any correlations between certain factors and the higher risk of water tornadoes.They compared these observations with a simulated probability index, which indicates whether the waterspout is likely to be Established Through the correct combination of atmospheric measurements, human forecasts It is “better” than the model in every index of predicting water sprouts.

Similarly, Research The report published by NOAA Weather Forecast Service Director David Novak and his colleagues showed that although human forecasters may not be able to “beat” models in typical sunny weather, they are still better than algorithmic calculators in severe weather. Forecast the weather more accurately. In the two decades of information researched by the Novak team, humans’ accuracy in predicting recent precipitation is 20% to 40% higher than the most commonly used national models, the Global Forecast System (GFS) and the North American Mesoscale Forecast System (NAM). Humans have also made statistically significant improvements in temperature prediction under the guidance of the two models. “Usually, we find that in larger events, predictors can make some value-added improvements to automated guidance,” Novak said.



Source link

Recommended For You

About the Author: News Center