Advanced spatio-temporal causal inference for climate research

Informations projet

CausalEarth

N° de convention de subvention: 948112

DOI

10.3030/948112

Date de signature de la CE 13 Janvier 2021

Date de début 1 Février 2021

Date de fin 31 Juillet 2026

Financé au titre de

• EXCELLENT SCIENCE - European Research Council (ERC)

Coût total

€ 1 499 631,00

Contribution de l’UE

€ 1 499 631,00

Coordonné par

UNIVERSITAET POTSDAM
Germany

Periodic Reporting for period 4 - CausalEarth (Advanced spatio-temporal causal inference for climate research)

Période du rapport: 2024-06-01 au 2025-02-28

Résumé du contexte et des objectifs généraux du projet

CausalEarth is an interdisciplinary project, aiming to improve our understanding of the causal interdependencies between major drivers of climate variability from observational and climate model data. To this end, CausalEarth develops novel machine learning-based causal inference methods that can account for common drivers, indirect effects, nonlinearities, nonstationarity, and the complex spatio-temporal nature of the underlying phenomena. The objectives are, firstly, to develop novel methods and algorithms, analyze their theoretical properties, and test these in numerical experiments. Secondly, these novel methods will be applied to climate model data in order to understand their limits and intercompare different climate models. Finally, applications to observational data will facilitate to learn about causal relations and evaluate how well climate models can reproduce them in order to contribute and improve our understanding of the climate system and climate change.

Travail effectué depuis le début du projet jusqu’à la fin de la période considérée dans le rapport et principaux résultats atteints jusqu’à présent

The focus of this first phase was on the development of causal inference and machine learning methods able to deal with the complexities of climate data, in particular their spatio-temporal nature. To test these methods, a toymodel for climate data alongside with benchmarks was developed. An important further step was a theoretical investigation of the spatio-temporal dependency analysis paradigm, that is, how well can one in principle learn causal relations from such complex data. Further works covered more technical aspects such as dealing with high-dimensional conditional independence tests and how to learn from multiple data sources. In the coming period, these methods will be further advanced and applied to climate model and observational data. To foster the application of causal methods in climate science and beyond, the project also published a review / guide article in Nature Reviews Earth and Environment. Additionally, a guide blog post series on causal inference and machine learning more targeted to the lay audience was started this year and continuous to be published at the Medium blog post channel.

Progrès au-delà de l’état des connaissances et impact potentiel prévu (y compris l’impact socio-économique et les conséquences sociétales plus larges du projet jusqu’à présent)

My foundational work on how to efficiently estimate causal effects from scarce data was a major breakthrough because it was quite unexpected that such a strong theoretical result was possible. Until the end of the project I expect further methodological advancements that enable to learn and understand causal relations in climate dynamics and how well these are represented in climate models.

CausalEarth schematic