The Danger of Leaving AI Time Prediction

People tried to predict climate change for millennia, using early tradition— “red skies at night” is an optimistic sign for weather-weary sailors actually associated with dry air and high pressure over the area — as well as observations taken from roofs, hand-drawn maps and local rules. These guides for future weather forecasts are based on years of observation and experience.

Then, in the 1950s, a group of mathematicians, meteorologists, and computer scientists — led by John von Neumann, a renowned mathematician who had helped Project Manhattan years earlier, and Jule Charney, an atmospheric physicist often considered the father of dynamic meteorology — tested the first computerized automatic forecast.

Charney, with a team of five meteorologists, divided the United States into (by today’s standards) fairly large plots, each covering more than 700 kilometers. By running a basic algorithm that took a real-time pressure field in each discrete unit and predicted it in advance over one day, the team made four 24-hour atmospheric forecasts covering the entire country. It took 33 full days and nights to complete the forecasts. Although far from perfect, the results were encouraging enough to start a revolution in weather forecasting, shifting the field towards computer modeling.

Over the next decades, billions of dollars of investment and the evolution of faster, smaller computers led to an increase in predictability. The models are now able to interpret the dynamics of atmospheric particles only 3 kilometers in size, and since 1960, these models have been able to include increasingly accurate data sent from weather satellites.

In 2016 and 2018, the GOES-16 and -17 satellites were launched into orbit, providing a number of improvements, including higher resolution images and accurate lightning detection. The most popular numerical models, the US Global Forecast System (GFS) and the European Center for Medium-Range Weather Forecasts (ECMWF), have announced significant upgrades this year, and new products and models are evolving faster than ever. With the touch of a finger, we can access an astonishingly accurate weather forecast for our exact location on the Earth’s surface.

Today’s lightning speed predictions, the product of advanced algorithms and global data collection, seem one step away from full automation. But they are not perfect yet. Despite expensive models, a number of advanced satellites and mega-computers, human forecasters have a unique set of their own tools. Experience – their ability to observe and draw connections where algorithms cannot – gives these forecasters an advantage that continues to outperform brilliant weather machines in the most important situations.

Although extremely useful with big picture forecasting, the models are not sensitive to, say, small rising currents in one small land quadrant suggesting a water outflow is forming, says Andrew Devanas, an operational forecaster at the National Weather Service office in Key West, Florida. Devanas lives near one of the world’s most active regions for water spills, sea tornadoes that can damage ships passing through the Florida Sea # and even coming ashore.

The same restriction makes it impossible to predict thunderstorms, extreme rainfall and tornadoes on land, such as those tore through the Midwest in early December, killing more than 60 people. But when tornadoes appear on land, forecasters can often spot them looking for their signature on the radar; outbursts are much smaller and often lack this signal. In a tropical environment like the Florida Keys, the weather doesn’t change much from day to day, so Devanas and his colleagues had to manually look at variations in the atmosphere, such as wind speed and available humidity – variations not seen by algorithms. always consider – to see if there is a correlation between certain factors and a higher risk of water spills. They compared these observations with a modeled probability index that indicates whether outbursts are likely and found that with the right combination of atmospheric measurements, human prognosis “surpassed” the model in every metric of predicting shoots.

Similarly, research published by NOAA Director of Weather Forecasting David Novak and colleagues show that while human forecasters may not be able to “beat” models on your typical sunny day in fine weather, they still give more accurate predictions than algorithms in poor condition time. During the two decades of information that Novak’s team studied, people were 20 to 40 percent more accurate in predicting precipitation in the near future than the Global Forecast System (GFS) and the North American Mesoscale Forecast System (NAM), the most commonly used national models. Humans also made statistically significant improvements in temperature prediction compared to the guidelines of both models. “We often find that in larger events, forecasters can make some improvements with the added value of automated management,” says Novak.

Source link

Be the first to comment

Leave a Reply

Your email address will not be published.