Why is AI unable to foretell floods like these attributable to the Dana in Valencia prematurely? | Technology | EUROtoday

Artificial intelligence (AI) is able to figuring out tumors from 1000’s of photographs, intervening exactly in a surgical treatment, discovering antibiotics and proteins, evaluating market conduct or optimizing an industrial course of, amongst dozens of different makes use of. However, this technological capability may be very restricted within the face of antagonistic phenomena akin to floods, probably the most damaging catastrophes after earthquakes (together with tsunamis). Floods have affected 2.5 billion individuals in 20 years, killing 1 / 4 of 1,000,000 of them and inflicting harm estimated at 936 billion, in line with the worldwide database on pure disasters (EM-DAT) of the Catholic University of Louvain ( Brussels). However, developments to foretell and keep away from its results will not be but efficient, regardless of being one of many aims of technological giants and establishments around the globe. Because?

“Existing climate models are not very good for certain extreme weather events, which are increasing much faster in the real world compared to what the models tell us should happen. It is important to forecast extremes so that we can have early warnings,” explains Dim Coumou, a climatology skilled on the University of Amsterdam (Netherlands).

One of the obstacles to correct forecasts is the knowledge accessible to coach synthetic intelligence. Although decades-old info exists, it is probably not related to understanding probably the most antagonistic phenomena. “Extreme events are, by definition, rare. So you don’t always have many observations. That is a big obstacle if you want to use artificial intelligence methods,” Coumou particulars in Horizon.

AI supplies outcomes based mostly on patterns noticed by analyzing an enormous quantity of information. You can establish a tumor from a picture when you have 1000’s extra to match and these are high quality and already supported by sure diagnoses. “AI needs large amounts of high-quality data to be useful for science and databases that contain this type of data are scarce,” warns David Baker, latest Nobel Prize in Chemistry for creating open synthetic intelligence methods that create new proteins.

Added to the shortage of information on uncommon phenomena is the problem of precisely predicting a meteorological occasion as dynamic and sophisticated as a dana (remoted melancholy at excessive ranges), which causes heavy rains and storms that may be very localized and variable in depth. Its conduct is erratic relying on a number of elements, akin to surrounding temperatures, humidity, strain, winds, orography and interactions with different geographical and atmospheric components.

Even very early knowledge doesn’t assure a exact sample to know precisely when and the place it would hit. And even when these knowledge present a dependable precipitation mannequin, they’re inadequate to take away the uncertainties. “Very intense storms do not automatically cause damage. There are many other factors at play,” warns Kevin Collins, professor of Environment and Systems on the Open University of the United Kingdom, to the Science Media Center.

The professors of the University of Lleida Víctor Resco de Dios and Domingo Molina agree. “It is not enough to know how much and where it will rain, but we also need to establish how that rain will be transformed into floods and which areas will be potentially affected,” they write in The Conversation

The fleeting variability of all of the elements concerned, the dispersion and shortage of particular knowledge on extraordinary phenomena and the heterogeneous high quality of those restrict synthetic intelligence to establish a sample and generate a prevention and warning mechanism. Developing a system that had a warning capability of simply 24 hours, in line with the Global Commission on Adaptation, “would reduce damage by 30%.” These, in line with a examine in Nature communicationsrise to 143,000 million {dollars} yearly (about 133,000 million euros), greater than the typical recorded by EM-DAT when together with an analysis of human losses.

Developing a system that had a warning capability of simply 24 hours, in line with the Global Commission on Adaptation, would scale back harm by 30%

In this fashion, the hot button is not solely the reliability of the prediction but additionally the lead time. “A forecast, even if accurate, has no information value if it does not arrive early enough to significantly reduce casualties and property damage caused by flash floods,” explains Geon-Wook Hwang, a researcher on the Korea Institute of Engineering. Civil and Construction Technology (KICT) engaged on a system to forecast floods.

Despite the difficulties, technological giants akin to Google or IBM, in collaboration with NASA, and different European establishments and different continents, throughout the framework of the United Nations five-year plan, are working to develop synthetic intelligence instruments that facilitate dependable prediction. within the medium and long run and enhance early warning methods.

Google DeepMind, the factitious intelligence firm of the North American know-how big, confirmed in Science a machine learning-based climate forecasting mannequin to offer 10-day predictions “better, faster and more accessible than existing approaches,” in line with the examine. The mannequin, referred to as GraphCast, outperformed conventional methods in 90% of the instances examined.

The system that served as a reference for Google was the European Center for Medium-Range Weather Forecasts (ECMWF), which has a supercomputer in Bologna (Italy) with round a million processors and an influence of 30 petaflops. (30,000 trillion calculations per second). This heart, which makes use of synthetic intelligence in its Integrated Forecast System (AIFS) and gives long-term forecasts of local weather occasions, anticipated the torrential rains of September in central Europe.

GraphCast doesn’t require these capabilities and makes use of machine studying educated on historic knowledge to ship an correct 10-day forecast in lower than a minute. “We believe this marks a turning point in weather prediction,” say the authors, led by DeepMind scientist Remi Lam.

IBM can also be on this race, in collaboration with NASA, with a proposal, additionally for machine studying. “Foundational artificial intelligence models that use geospatial data [meteorológicos, de sensores y de satélite] “They can be game-changers because they allow us to better understand, prepare for and address the many climate-related phenomena that affect the health of our planet in a way and at a speed never seen before,” explains Alessandro Curioni, vp of Accelerated Discovery at IBM. .

For Kate Royse, director of the British Hartree supercomputing heart, these fashions “would allow smarter decisions to be made based on the accurate prediction and management of flood risk, which is essential for the future planning of cities.”

“A good use of AI-based weather forecasts would be to complement and enhance our prediction toolbox, perhaps allowing us to produce models for accurate assessment and interpretation of the probability of extreme events,” Andrew Charlton-Pérez, professor, advised Reuters. of meteorology on the University of Reading within the United Kingdom.

A proposal for a complementary system is being investigated by two facilities of the German scientific consortium Helmholtz and revealed in Nature Communications. This mannequin combines the meteorological service’s precipitation forecasts with knowledge on soil moisture, one of many crucial elements for the event of floods, and flows to foretell flood zones and depths, in addition to the affect on buildings, streets, railway sections, hospitals or different crucial components of infrastructure. “The responsible authorities and the population not only have information about a possible water level 30 kilometers upstream, but also a high-resolution flood map that shows the impacts of the flood. For example, they could know where people might be in danger or who needs to be evacuated,” says hydrologist Sergiy Vorogushyn, from the German Research Center for Geosciences and co-author of the examine.

https://elpais.com/tecnologia/2024-11-07/por-que-la-ia-es-incapaz-de-predecir-inundaciones-como-las-causadas-por-la-dana-en-valencia-con-antelacion-suficiente.html