Join Climate Change AI for an open webinar on Machine Learning for Climate Science and Earth Observation - Tuesday October 19th, 17:00 CET/Zurich time; 11:00 AM US Eastern Time.
Register (free) here: Machine learning for Climate Science and Earth Observation
Speakers:
Gustau Camps-Valls (Professor, Universitat de València)
Maike Sonnewald (Associate Research Scholar, Princeton University)
Description:
Physics-aware ML for Earth sciences
Most problems in Earth sciences aim to do inferences about the system, where accurate predictions are just a tiny part of the whole problem. Inferences mean understanding variables relations, deriving models that are physically interpretable, that are simple, parsimonious, and mathematically tractable. Machine/deep learning models are excellent approximators, but very often do not respect the most elementary laws of physics, like mass or energy conservation, so consistency and trustworthiness are often compromised. I will review the main challenges ahead in the field, and introduce several ways to live in the Physics and machine learning interplay. This is a collective long-term AI agenda towards developing and applying algorithms capable of discovering knowledge in the Earth system.
A robust blueprint for trustworthy AI for climate analysis
Using AI for climate analysis means applications should ensure that their skill is not due to chance. To develop trustworthy and transparent AI, rooting work in physical understanding allows interpretability and enhanced explainability of predictive skill. Here, I present a blueprint for a transparent ML application, engineered to elucidate its source of predictive skill. The application reveals 3D ocean current structures from surface fields in climate models. I apply this to predict ocean current changes to understand the variability of global heat transport under climate change. A labeled data set is engineered using an explicitly interpretable equation transform and k-means to model data, allowing theoretical inference. A multilayer perceptron is then trained, explaining its skill using a combination of layerwise relevance propagation and oceanographic theory. Such interplay between interpretable and explainable AI can deliver actionable insight in support of climate decision making and beyond.
Register (free) here: Machine learning for Climate Science and Earth Observation
Speakers:
Gustau Camps-Valls (Professor, Universitat de València)
Maike Sonnewald (Associate Research Scholar, Princeton University)
Description:
Physics-aware ML for Earth sciences
Most problems in Earth sciences aim to do inferences about the system, where accurate predictions are just a tiny part of the whole problem. Inferences mean understanding variables relations, deriving models that are physically interpretable, that are simple, parsimonious, and mathematically tractable. Machine/deep learning models are excellent approximators, but very often do not respect the most elementary laws of physics, like mass or energy conservation, so consistency and trustworthiness are often compromised. I will review the main challenges ahead in the field, and introduce several ways to live in the Physics and machine learning interplay. This is a collective long-term AI agenda towards developing and applying algorithms capable of discovering knowledge in the Earth system.
A robust blueprint for trustworthy AI for climate analysis
Using AI for climate analysis means applications should ensure that their skill is not due to chance. To develop trustworthy and transparent AI, rooting work in physical understanding allows interpretability and enhanced explainability of predictive skill. Here, I present a blueprint for a transparent ML application, engineered to elucidate its source of predictive skill. The application reveals 3D ocean current structures from surface fields in climate models. I apply this to predict ocean current changes to understand the variability of global heat transport under climate change. A labeled data set is engineered using an explicitly interpretable equation transform and k-means to model data, allowing theoretical inference. A multilayer perceptron is then trained, explaining its skill using a combination of layerwise relevance propagation and oceanographic theory. Such interplay between interpretable and explainable AI can deliver actionable insight in support of climate decision making and beyond.