EL PAÍS offers the América Futura section openly for its daily and global information contribution on sustainable development.
If you want to support our journalism, subscribe
here
.
The American Nobel Prize in Physics Richard Phillips Feynman announced in 1985, in a conference at Harvard, that one day there would be machines that would be better than humans at predicting the weather and that would do so faster.
What seemed like science fiction almost 40 years ago is today a reality.
Weather predictions are made with complex models that combine data from the atmosphere and oceans stored in supercomputers.
Google, for example, uses artificial intelligence (AI) through GraphCast to offer global weather diagnosis in just one minute and much more accurately than was done before.
Technology allows us to work with large volumes of information and bring it closer to users: it is common for us to carry a cell phone in our pocket with the temperature on the home interface, and we increasingly have more tools at our disposal with geolocated data to make decisions when faced with a situation. changing climate;
In addition, specialized journalists are trying to make news more attractive with augmented reality and there are even televisions that have left their weather segments to artificial intelligence.
In Argentina, the Prevenir (Forecasting and Warning of Flash Flood Events) project, a joint initiative between Argentine and Japanese entities that uses different methodologies, including AI, seeks to develop an early warning system for unforeseen urban flooding.
In an initial stage, Prevent has been designed for two of the most populated cities in the country that have highly vulnerable basins: that of the Sarandí and Santo Domingo streams, in the city of Buenos Aires, and that of the Suquía River, which runs along mountain slopes. of the province and city of Córdoba.
Prevent is a pioneering project in the region.
Although in Brazil some researchers are working on the use of AI to improve forecasts through learning from systematic errors, the Argentine project will provide useful tools and recommendations for similar systems in other parts of the world.
Additionally, in Argentina, the National Meteorological Service is also working with AI techniques to estimate precipitation using satellite data and to forecast phenomena such as fog or electrical activity.
The researcher at the National Scientific and Technical Research Council (Conicet) and professor at the Department of Atmospheric and Ocean Sciences at the University of Buenos Aires (UBA), Juan Jose Ruiz, explains that Prevent works like Google's tool, but on a smaller scale.
“The idea is similar to GraphCast, but the model is simpler because it focuses on precipitation and is specialized to reproduce the dynamics of the atmosphere in two particular regions,” explains Ruiz.
The GraphCast platform was trained with weather data that the European Center for Medium-Range Weather Forecasts (Ecmwf) has collected since 1979. It analyzed more than 40 years of weather history to make weather predictions.
In the Argentine case, although the learning processes are more limited, according to Ruiz, there are several years of information provided by weather radars: “From this set of data, AI-based models can learn how it will evolve in the future. future a certain precipitating system,” he points out.
In addition to improving flash flood forecasts and warnings to be more accurate and communicated more efficiently, Prevent seeks to expand and integrate observation networks.
This will also serve to raise awareness among the population about risk prevention and train researchers.
According to Ruiz, the operations carried out by computers to simulate the transformations of the atmosphere entail a very large computational cost.
They must do a lot of math to predict the weather.
There are artificial intelligence algorithms, for example, that can be between 1,000 and 10,000 times more effective than current ones.
Prevent's artificial brain is supplied with information from different supercomputers that have better performance when solving calculations and tasks such as training AI models.
The Clementina XXI supercomputer is one of them.
Hosted at the National Weather Service headquarters, it can develop and train models.
In addition, the project also has access to the Japanese supercomputer Fugaku, the fourth most powerful in the world.
“Extreme phenomena continue to be a challenge”
According to the Conicet researcher, algorithms can also learn how errors that affect simulations and predictions behave, which helps to better detect and correct them.
But there is still much to investigate to more accurately predict some meteorological phenomena that cause great damage.
“Extreme events continue to be a challenge,” acknowledges Ruiz.
Depending on how AI-based systems are trained, according to the professor, they can underestimate the frequency of extreme events.
However, in his opinion, they can be very useful, especially when they receive specific training, such as models that “provide the probability of occurrence of a given event,” he says.
Although Prevenir is only two and a half years old, the researcher highlights the benefit of collecting information on a large number of events with different tools such as radar, satellite and operational numerical simulations, which grows the database that feeds the algorithms. the AI.
However, he explains that these conclusions were obtained from synthetic data sets that are used for research as they allow for testing.
“The next step in this research is to make the leap towards using real data to train AI models to generate prototypes that can be applied in real life.”
It is estimated that the project will end in 2027, the year in which they expect the prototype to be working.
At the moment, AI applied to meteorology is in the research stage and far from being a tool open to the public, such as Chat GPT.
Therefore, it will not be available for users who want to know what will happen each day with the weather.
The final objective of the project and of others that seek to integrate artificial intelligence into weather forecasts, explains Ruiz, "is to make more precise tools available to forecasters that allow them to improve the quality of the service provided to society."