web analytics

How can AI make weather forecasts cloudless

Not to mention that he personally enjoys the storms that constantly plague the state or the common legislature with hailstones the size of a golf ball. “I’m on the third roof in 15 years,” he laughed.

But that harsh weather is his reason for coming: A computer scientist who once worked in robots, was employed by the University of Oklahoma School of Meteorology. And last fall, at a cost of $ 20 million from the National Science Foundation, has launched one of the country’s leading artificial intelligence systems and climate change. As new methods of machine learning become ubiquitous and produce alarming programs, such as facial recognition or mimicking human writing, his or her center is part of a new initiative to see if they can learn clouds.

Dr. McGovern’s Institute, which includes six other universities and a variety of independent partners, is part of that effort. In addition to developing artificial intelligence systems to improve weather forecasts and coastal oceans, they are working to ensure that the tools they build are considered reliable for human forecasters who will eventually use them. “We’re testing the whole cycle,” he said. “We will save lives and save property.”

AI is already making existing predictive systems more efficient and contributing to the increase in speed and accuracy of forecasting, and demonstrates the promise of tracking bad weather patterns such as storms and hail with great accuracy. Technology will not replace traditional weather but will also add to and strengthen existing systems.

Promoting efficiency

There is great potential for more accurate forecasts and better prepared weather forecasts. According to the National Oceanic and Atmospheric Administration, by 2020 there would have been a record 22 climate and weather disasters that caused more than $ 1 billion in damage. The actors estimate that a recent ice rink in Texas destroyed infrastructure and disrupted supply chains worth $ 90 billion. And with all the improvements in forecasting over the years, there is still much we can know. According to Drs. McGovern, the government is extensive in providing advanced warnings about hurricanes, expecting 80% of them, but it tends to get false, and 80% warnings turn out to be wrong.

Since the beginning of modern weather forecasts in the 1950’s, meteorologists have relied heavily on “weather forecasts” – mathematical models that mimic the earth and the atmosphere in terms of physics of water, wind, earth, and sunlight, and how they interact in pursuit of detailed interpretation. about 100 million pieces of data each day, a complex process similar to the imitation of the human brain or the birth of the universe.

For decades, this has resulted in a steady improvement in the accuracy of the forecast. But in recent years, the proliferation of satellite observations as well as new sensors, such as air-pressure monitors on billions of cell phones, has overwhelmed scientists’ ability to adapt to the weather. And even scratching a fraction of that data has required a clear increase in computer power to make timely predictions.

New intelligence techniques work in a completely different way from the old techniques by training neural networks in this data flood and not in the laws of physics. Instead of using brute-force computation to predict the weather based on current conditions, these networks review data on past weather and improve their understanding of how conditions change. AI technological techniques have been used in climate and climate for decades – the National Oceanic and Atmospheric Administration sponsored its first conference on AI back in 1986 – but recent advances in in-depth learning and greater access to computing computers have provided faster research power.

AI techniques are not used to generate predictions themselves, at least for now. This is because the traditional methods are very good: Two weeks before the winter storm that hit Texas in mid-February, the Fort Worth office of the National Weather Service began advising that extreme cold weather was on its way, and on Sunday many models measured their intensity within a few degrees. Ted Ryan, a meteorologist there, says that they sometimes carry out predictions using a sophisticated machine learning algorithm to see if they can produce very different results, but they do not always incorporate them into their predictive and messaging activities. “It’s between a curiosity and something new.”

Another challenge for AI is that it is much better at predicting common patterns within the data it has been trained in – but the weather is even more important when it is out of the ordinary, like the Texas hurricane, which plagued the coldest temperatures since 1899.

Just because AI has not changed traditional predictors does not mean that it will not change immediately, however. The current weather forecast is already so powerful with the power of computing that it can only work on the fastest computers in the world, and scientists continue to try to force boundaries. Mechanical learning methods can reduce energy consumption by mimicking parts of global climate models but with simpler calculations and less power – and with minimal changes in accuracy.


Sid Boukabara, chief scientist at the National Oceanic and Atmospheric Administration’s Center for Satellite Applications and Research, thinks those benefits will be significant. “In some parts, it can work ten times more than a thousand times.” But it is too early to know how much this will improve the accuracy of the overall weather forecast.

For example, the European Center for Medium-Range Weather Forecast currently compares the earth to a 9-square-kilometer grid on the side (about 5 miles) and covered 137 layers high in space. Deputy Director Peter Bauer says each step up in detail requires an increase in power: The world’s largest computer computer in Bologna, Italy, will use an estimated 6,000 homes. He and his colleagues quickly approached the limits of what they could not afford or forgive, he said.

This year – relying on AI’s methods to improve efficiency – the center will begin developing a new global model with a 1-kilometer solution that will bring storms and marine systems to a better focus, said Dr Bauer. “Big and fast machines give us a lot of computing power, but we need to completely change the code we use so that we can use it properly.”

Near home

Researchers see the promise of AI in small applications, for example in using machine learning to create hyperlocal predictions that are more relevant to people who discover them. Even the best weather models in the world are compared to a local resolution of several miles, so they can accurately predict rain in your region but do not predict rainfall in your area. Researchers say that an in-depth study of the subject would allow us to delve into more complex geography in order to predict more details, such as how to use high-resolution images. In this regard, it does this by adding weather forecasts from conventional methods with details such as geography to explain how extensive weather patterns will be fulfilled in certain areas.

Machine learning can also be seen as important in accurate “current delivery” – short-term forecasts compiled at a fast pace that is not possible with traditional methods. Last spring, two scientists at Google Research showed that, in addition to the laws of nature that were explicitly imposed on them, deep neural networks could predict rain in the next eight hours better than other high-end models. Nal Kalchbrenner, one of the investigators, said they were working diligently to increase the rate and duration of the forecast. “This vision opens the way for AI to make broader and greater contributions to the field of astronomy and climate science.”

Even promoting an accurate weather for an hour or two can have a profound effect on businesses, which face a wide range of risks. “If I tell you it’s going to be a hailstorm, you can put your car in the garage,” said Dr. McGovern. “If you are a car manufacturer or a car dealer, and you have a thousand cars at the waist, you cannot move a thousand cars at that time. “Last year, he co-authored a study of mechanical studies that improved short-term hailstones, and overcame a limited number of weather data by analyzing and analyzing thousands of reports for each storm. He is now working with NOAA to develop strategies into an effective product.

Governments and big business have recognized the power of AI. Last October, Google for NOAA and Alphabet Inc. announced a partnership to explore how machine learning can help an agency to make better use of satellite and environmental data. In January, the state-run National Center for Atmospheric Research invested $ 35 million in new computers equipped with state-of-the-art AI technology. And both NOAA and its European Center partners have recently published strategies to incorporate artificial intelligence into their work.

However, beware of the hype that closes all AI objects, some leading researchers are wary of excessive prominence. Stephan Rasp is not a listener; co-authored one of the first studies showing that AI can mimic certain predictive features, and “artificial intelligence” is still in the first place where he works as a major data scientist, Climate.ai. you see, a lot of ongoing research is in the process of proving the concept, as it has successfully used AI in simplified data sets but there are few examples yet to improve climate and climate.

He thinks it could take ten years for the field to decide where AI can be useful and where it can’t. “When you have a hammer it all looks like a nail. And I feel like that’s what we’re doing right now. “

This article was published from a wire agency feed without text editing.


 

 

 

 

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.