Artificial intelligence is looking for tipping points in the climate system – Market Research Telecast

as Tipping points in the climate system are variables that cause a drastic change in the climate above a certain threshold value. Because beyond these tipping points, so the idea, a self-reinforcing mechanism is set in motion that accelerates climate change ever more. Well-known tipping points are, for example, the thawing of the Arctic permafrost, the collapse of the oceanic current systems or the thawing of the ice sheets at the poles whether and how many more such points there are is the subject of current research.

This article is from issue 7/2021 of the Technology Review. The magazine will be available from September 30th, 2021 in stores and directly in the heise shop. Highlights from the climate booklet:

Chris Bauch from the University of Waterloo and his colleagues have now trained a deep, neural network to Identify tipping points in climate systems and give warnings when the system approaches a dangerous tipping point. The approach is based on an abstract description of complex, dynamic systems: The system analyzes the auto-correlation of time series values and learns to recognize specific patterns that herald a bifurcation, a qualitative change in state.

However, Bauchs team is by no means the only one trying to get better predictions about climate change with the help of machine learning and artificial intelligence, reports MIT Technology Review in its current issue 07/2021. For example, a team led by Tapio Schneider from the California Institute of Technology is working on eliminating a central weakness of current climate models with the help of machine learning: the extremely simplified modeling of clouds.

Because the global models that were used for the current IPCC report, for example, model the climate system in a grid with an edge length of 100 kilometers. Clouds are much smaller they will therefore be parameterized, that is, one cell of the model is calculated as 20 percent cloudy, for example. Schneider and his colleagues therefore take the basic physical equations of physical climate models and coarsen them by using, for example, averaged values on a large energy grid. In order to still be able to model the small-scale, dynamic processes of the clouds, they add additional functions to the equations that cover these processes. These functions, which are essential for dynamics, are learned by neural networks from high-resolution, local cloud simulations and weather data.

Others like Jakob Runge from the TU Berlin use the methods of causal inference to identify cause-effect relationships in climate data with the help of AI. We defined variables such as temperature, pressure and so on in certain regions. Then, when we apply that to the observation data, we see what the causal network looks like, says Runge. Some processes are interconnected, others are not. You get a network of dependencies a kind of fingerprint. Then we take the same variables in models, learn the causal network in the model data and compare. Are they the same? Where do the models not form the reality so well? And not necessarily in the absolute values, but in the causal relationships. The method can also be used to calculate the reliability of a model, not only on the current, but also on future data.

(wst)

Disclaimer: This article is generated from the feed and not edited by our team.

Visit link:
Artificial intelligence is looking for tipping points in the climate system - Market Research Telecast

Related Posts

Comments are closed.