A study by German scientists in Hamburg and Jena, published today in the journal Nature, shows that artificial intelligence may improve our understanding of the climate and the Earth system. Especially the possibility of profound learning has only partly been exhausted thus far. Particularly, complex processes like hurricanes, fire propagation, and vegetation dynamics may be better explained with the support of AI. Consequently, climate and Earth system models will be improved, with new models combining intelligence and physical modeling. In the past decades mostly characteristics have been researched using machine learning strategies, like the distribution of land properties from the neighborhood to the global scale.
For a while now, it’s been possible to handle more energetic processes by using sophisticated profound learning techniques. This enables for instance to quantify the global photosynthesis on property with simultaneous consideration of seasonal and short term variations.
Deducing inherent laws from observation data
From plenty of detectors, a flood of earth system information became available, but so far we have been lagging behind in diagnosis and interpretation, explains Markus Reichstein, controlling your stresses director of the Max Planck Institute for Biogeochemistry in Jena, directory board member of the Michael Stifel Center Jena and original author of the publication.
This is where profound learning techniques become promising instrument, beyond the classical machine learning applications like image recognition, natural language processing or AlphaGo, adds co writer Joachim Denzler from the Computer Vision Group of the Friedrich Schiller University Jena and member of MSCJ. Examples for application are intense events like fire spreads or hurricanes, that are very complicated processes influenced by local conditions, but additionally by their temporal and spatial context. This also applies to atmospheric and sea transport, soil movement, and vegetation dynamics, and a number of the classic subjects of Earth system science.
Intellect improve the climate and the Earth system models
Deep learning strategies are hard. All information driven and statistical procedures don’t ensure physical consistency in themselves, are dependent on information quality, and might experience difficulties with extrapolations. In addition to, the requirement of information processing and storage capability is quite significant. The publication discusses all of these requirements and barriers and develops a strategy to effectively combines machine learning with physical modeling. If both techniques are brought together, so called models are made. They can for instance be used for modelling the movement of sea water to predict the sea surface temperature. Whilst the temperatures are modelled physically, then the sea water motion is represented by a machine learning approach. The idea is to combine the best of two worlds, the consistency of real models with the flexibility of machine learning, to get significantly improved models, Markus Reichstein further explains. The scientists contend detection, early warning of extreme events in addition to seasonal and long term prediction and projection of weather and climate will greatly benefit from the talked profound learning and hybrid modeling approaches.