If you’re not convinced by current weather forecasting systems, you might want to consider whether you’d like AI to do the job instead of traditional options. You’d better start thinking about it quickly, as Google has already created the model to do the job.
Google’s project is called NeuralGCM and is a combination of the latest machine learning technology combined with more traditional weather prediction processes. The company’s goal is twofold. On the one hand, they want to make the processes used to obtain weather forecasts more efficient. On the other hand, they also aim to reduce costs so that less investment is required to obtain accurate information.
Solving a problem
Machine learning techniques have been used in weather forecasting, but experts have encountered some problems. The main one is in long-term forecasts. Although these systems work well in short periods of time, when they are used over long periods of time, the situation changes. And, of course, one of the main objectives of weather forecasting is to be able to do it in advance. What they do excel at, however, is their speed and short-term effectiveness.
Because of this, Google wanted to combine the best of the most current models with traditional systems in order to break down previous limitations. From Google Research, Stephan Hoyer comments that they have truly achieved a combination of physics and artificial intelligence. As he says, it is no longer a confrontation between both technologies.
How does NeuralGCM work?
First, a traditional system is used to analyse climate data, especially atmospheric changes, with the aim of making a forecast. Once this part of the process is complete, AI is used. It is used in those areas where classic models are weaker or fail more. For example, small-scale forecasts, forecasts in microclimates such as the fog that occurs in certain parts of the world and forecasts that must take cloud formation into account. The expert says that what they do is introduce AI in a very selective way and that, in this way, they can correct common errors.
But does it really work? So far, NeuralGCM is giving promising results. According to its developers, it is capable of making forecasts as effective as those of the ECMWF (European Centre for Medium-Range Weather Forecasts) from 1 to 15 days, which is an important step forward compared to other AI models. It is particularly noted that an AI model like this does not overturn all previous systems or the knowledge that has been used in weather forecasting for years. What it does is complement them to achieve more precise results and a faster and more efficient way of obtaining some data that would otherwise be difficult to obtain. For now, they are satisfied to see, with the tests carried out, that the AI ​​works and makes good forecasts.
Now they want to look at the long-term weather forecasting environment. Because that is where they really believe the big payoff lies, from which all of humanity could benefit. If we had accurate long-term weather forecasts, there would be many ways to prepare ourselves to avoid unfortunate incidents or situations.
Not only that, but they want to use this weather forecasting AI to be able to anticipate larger events, such as tropical cyclones. It is good to know that NeuralGCM is also going to be an open source project that will allow all interested scientists to work with the model and take advantage of it in different ways. This means that, very possibly, we will start to see many AI and systems that will start to offer us weather forecasts. Now, it will be up to each one to decide whether they are trustworthy or not. But, for the moment, they are on the right track.
