The development of generative models is wide-ranging, and its benefits – along with risks – extend across many sectors. A project recently unveiled by Google is another example of such broad reach. Near SEEDS, a model using artificial intelligence in meteorology. to obtain more accurate forecasts. In addition, the method is more accessible compared to those currently used.
The name is clever. SEEDS are “scalable diffusion sampler with ensemble envelope“, which in Spanish can be translated as “scalable ensemble diffusion sampler”. For others, “seedsmeans “seed”, and this is logical: the model strives to become an embryo a new way to forecast the weather. “This is a new artificial intelligence technology designed to speed up and improve weather forecasts using diffusion models to better characterize rare or extreme weather events,” commented researchers from Google Research.
One of the most interesting aspects of this tool, which combines artificial intelligence and meteorology, is its design, similar to massive language models (LLM). In other words, it works in a similar way to how OpenAI’s ChatGPT and Google’s own Gemini do it. That is, they generate results from text prompts.
How SEEDS works and why artificial intelligence in meteorology promises a revolution
“AI is the key to better climate forecasting,” he says in a conversation with Hypertext Google Researcher and Meteorologist, Robert Carver. The expert notes that most forecasts are quite accurate for more common conditions, such as mild or warm summer days. But he notes that creating enough models to determine the likely outcome of an extreme event is beyond the reach of most services.
When did artificial intelligence begin to be used in meteorology? Does the visibility of systems like Gemini affect this?
Research groups have been using neural networks, one of the technologies behind artificial intelligence, to improve weather forecasting since the 1980s. As technology advanced, Google began working on artificial intelligence to improve hourly forecasting. It is called “nowcasting“, and indicates whether it will rain in the next hour or not.
This technology now allows us to predict Google search results and impacts people’s lives every day. Present tense We continue to advance the application of AI for meteorological research., using several different methods to further develop this technology. SEEDS is one such project.
What new variables does the use of artificial intelligence bring to meteorology? In this sense, what changes do these technologies offer in relation to the methods currently used to predict climate and possible severe events?
AI is key. This allows us to work faster and more efficiently, which is important for something as sensitive as the weather. In particular, the project SEEDS allows us to create many more possible forecasts for a given time at a lower cost.. This allows us to better assess the likelihood of weather events, including severe weather events such as heat waves, that can seriously affect people.
Imagine you interviewed 1,000 people instead of 10.. Surveying 1,000 people to study something like the prevalence of depression can take a long time. This may limit me to studying only 10. But I will gain greater accuracy by doing more surveys because I will be able to better estimate the prevalence of this disease in a larger group. So if I could invent a technology that would allow me to do this, I could do something pretty powerful. Thus, conduct research quickly and provide solutions, policies, and assistance faster. SEEDS is doing it for the climate.. This gives us a wider range of forecasts to analyze, allowing us to better determine the likelihood of different weather events.
Why do you emphasize that with SEEDS, the use of artificial intelligence in meteorology is more cost-effective compared to other methods used in the field?
Currently, the best practice for making probabilistic forecasts is to run a computer weather model on a supercomputer 30 to 50 times. It uses slightly different initial conditions. We call this “ensemble prediction.” If it only rains in 50% of the different forecasts, then the chance of rain is 50%. It takes a lot of computing power to create all of this. We show that SEEDS will allow us to use only 10% of the computing power of the traditional approach. a supercomputer to produce the same number of weather forecasts.
They noted that SEEDS is an artificial intelligence for meteorology included among the generative models. Based on this, how is it similar to ChatGPT or Gemini?
Broadly speaking, generative AI models predict the most likely response to a given input. They allow us to create one text message (for example, “image of a cat driving a car”) and generate many images corresponding to it. Likewise, SEEDS “Message” – Unique Climate Prediction future. You can then create many more possible forecasts for the same time period. This gives us greater statistical confidence about what weather we can expect in the real world.
In a sort of beginner’s section on artificial intelligence in meteorology, could you explain to us in simple terms how SEEDS works?
If only two possible weather forecasts are entered into SEEDS, it will return an even larger set of weather forecasts that could occur at that time. Previously, this took hours, maybe even days, but we have reduced this time by 90%. And since we base the likelihood of a weather event (such as rain or snow) on how many of our forecasts show it will occur, such a large number of plausible scenarios means that the probability of its occurrence is much more accurate.
When will the promise of SEEDS be realized?
The initiative they are working on, combining artificial intelligence with meteorology, is one of Google’s many research projects, Carver said. This means that this is research that is not yet being used at a general level. “Our mission as Google researchers is to develop cutting-edge technologies that can have a positive impact on people’s lives.. Climate is one such variable,” he concludes.
Carver’s team recently published their research in a journal. Science achievements. The effort may truly be worth it. According to Google, “As climate change worsens and extreme weather events become commonplace, Accurate weather prediction can save lives“
Source: Hiper Textual
