A computational model that combines conventional weather forecasting technology with machine learning developed by Google outperformed other AI-based tools for forecasting of the climate climate conditions and long-term climate trends.
Machine learning was the cutting edge technology in its field artificial intelligence Before the emergence of genetic intelligence (programs like ChatGPT) that revolutionized the technology industry, but also in recent times machine learning has once again started to be a tool for research and development of AI technologies. In fact, a new and even more sophisticated version of machine learning called “deep learning” has emerged, which is starting to gain more and more space in the technology industry.
The model that It was presented in the journal Nature is the first to generate accurate global weather forecasts. Its development opens the door to forecasts that are faster and less energy-intensive than existing tools, and more detailed than purely AI-based approaches.
“Traditional climate models have to run on supercomputers. This is a model you can run in minutes,” says Steven Hoyer, a deep learning researcher at Google who led the team that developed the new model.
How Artificial Intelligence Improves Climate Forecasting
Current forecasting systems are typically based on general analysis climate process models (GCMs), programs that rely on the laws of physics to simulate processes in Earth’s oceans and atmosphere and predict how they might affect weather and climate. However, GCM models require a lot of computing power, and advances in machine learning have begun to provide a more efficient alternative. “We have terabytes or petabytes (a million times larger than a gigabyte) of historical weather data. By learning from these patterns, we can build better models,” Hoyer says.
There are already some machine-learning predictive models available, such as Pangu-Weather, made by Chinese tech giant Huawei, and GraphCast from DeepMind, a London-based AI technology developer that Google acquired in 2014. These models have similar levels of accuracy to standard GCMs for deterministic forecasts that produce a single weather prediction. But GCMs are not as reliable for complex or long-term climate predictions.
“The problem with pure machine learning approaches is that you only train a system on data from events that have already happened. The climate is constantly changing, we are heading into the unknown, so machine learning models need to be able to extend into this unknown future. By introducing physics into the model, we can ensure that our models are physically constrained and won’t be able to do anything unrealistic,” says Scott Hoskigg, who researches artificial intelligence and environmental data at the Alan Turing Institute in London.
Hybrid
Hoyer and his team developed and trained NeuralGCM, a model that combines “the capabilities of a traditional physics-based atmospheric calculation system with some elements of artificial intelligence,” as the researcher explains. The model developed by the researchers has been used to produce short- and long-term weather forecasts, as well as climate projections. To assess the accuracy of NeuralGCM, the researchers compared its predictions with real-world data, as well as results from other models, including GCMs and those based solely on machine learning.
Like current machine learning models, NeuralGCM can produce accurate, deterministic short-term weather forecasts—one to three days in advance—while consuming a fraction of the power required by GCMs. It also made far fewer errors than other machine learning models when producing long-term forecasts beyond seven days. In fact, NeuralGCM’s long-term forecasts were similar to forecasts made by the European Centre for Medium-Range Weather Forecasts (ECMWF-ENS) Ensemble Model, a GCM widely regarded as the best model for weather forecasting.
The research team also looked at how well the model could predict different weather events, such as tropical cyclones. They found that many of the pure machine learning models produced inconsistent and inaccurate predictions compared to NeuralGCM and ECMWF-ENS. The researchers even compared NeuralGCM to very high-resolution climate models, known as global storm-resolving models. NeuralGCM was able to produce more realistic tropical cyclone numbers and tracks in a shorter time frame.
Being able to predict such events is “very important for improving decision-making capabilities and preparedness strategies,” says Hosking.
Naftemporiki.gr