Earth system models are the most important tools for describing the physical state of the Earth including climate models which can predict how it might change in the future under the influence of human activities. AI is used increasingly to help improve these forecasts.
The development of Earth is a complex exchange of many factors, including the land surface with flora and fauna, the oceans with their ecosystem, the polar regions, the atmosphere, the carbon cycle and other biogeochemical cycles and radiation processes. Researchers refer to this as the Earth system.
The use of AI in the Earth system is now being investigated by a team led by Christopher Irrgang from the German Research Centre for Geosciences Potsdam. Their proposal is to develop a self-learning ‘neural Earth system modeling.’
Classical Earth system models are based on both well-known and lesser-known physical laws. With the help of mathematical and numerical models, the state of a system at a future time is calculated from what is known about the state of the system at a present or past time.
However, the price is that the increasingly complex ESMs require immense computational resources. Despite recent developments, even the predictions of the latest models contain uncertainties. For example, they tend to underestimate the frequency and strength of extreme events. Researchers are worried that abrupt changes could occur in certain subsystems of Earth, so-called tipping elements in the climate system, which classical modeling approaches cannot predict accurately. Many key processes, such as type of land use or the availability of water and nutrients, cannot be represented well.
The challenges of classical ESM methods, but also the ever-increasing amounts of available Earth observations, open up the field for the use of artificial intelligence. The advantage is that these self-learning systems do not require knowledge of the possibly very complex and not even fully known physical laws and relationships. Instead, they are trained on large datasets and learn the underlying systematics themselves. This powerful and flexible concept can be extended to almost any desired complexity. For example, a neural network can be trained to recognize and classify patterns in satellite images, such as ocean eddies, cloud structures or crop quality.
https://phys.org/news/2021-09-opportunities-limits-ai-climate.html