Summary
CausalEarth is an interdisciplinary project, aiming to improve our understanding of the interdependencies between major drivers (modes) of climate variability by developing novel statistical causal inference methods for both observations and model data.
Disentangling the interdependencies of the major modes, such as El Nino Southern Oscillation and the North Atlantic Oscillation, is key to understand regional climate, and essential for process-based climate model evaluation. The modes' interdependencies are characterized by common drivers, indirect effects, nonlinearities, regime-dependence, and heterogeneous spatio-temporal causal relations. Currently, observational analyses are mostly based on the correlation of scalar (one-dimensional) time series derived from regional averaging or principal component analysis, restricted to supposed causal regimes, e.g., the winter season or phases of multi-decadal climate indices, where dependencies are expected to be stationary. Such scalar correlation approaches fall short in capturing the modes' complex regime-dependent spatio-temporal causal interdependencies.
CausalEarth will develop innovative methods to move (1) from representing complex phenomena as scalar indices to learning spatio-temporal features, (2) from supposing causal regimes to learning them from data, and (3) from correlation to causal dependencies. To this end, CausalEarth will combine recent developments in machine learning with causal inference algorithms.
These methods will be used to infer the causal interdependencies and drivers of major climate modes from observations and to construct the next generation of causal metrics for climate model evaluation.
CausalEarth will push the limits of what can be learned from observational data about causal relations and drive model development towards breakthroughs in projecting our future climate.
Disentangling the interdependencies of the major modes, such as El Nino Southern Oscillation and the North Atlantic Oscillation, is key to understand regional climate, and essential for process-based climate model evaluation. The modes' interdependencies are characterized by common drivers, indirect effects, nonlinearities, regime-dependence, and heterogeneous spatio-temporal causal relations. Currently, observational analyses are mostly based on the correlation of scalar (one-dimensional) time series derived from regional averaging or principal component analysis, restricted to supposed causal regimes, e.g., the winter season or phases of multi-decadal climate indices, where dependencies are expected to be stationary. Such scalar correlation approaches fall short in capturing the modes' complex regime-dependent spatio-temporal causal interdependencies.
CausalEarth will develop innovative methods to move (1) from representing complex phenomena as scalar indices to learning spatio-temporal features, (2) from supposing causal regimes to learning them from data, and (3) from correlation to causal dependencies. To this end, CausalEarth will combine recent developments in machine learning with causal inference algorithms.
These methods will be used to infer the causal interdependencies and drivers of major climate modes from observations and to construct the next generation of causal metrics for climate model evaluation.
CausalEarth will push the limits of what can be learned from observational data about causal relations and drive model development towards breakthroughs in projecting our future climate.
Unfold all
/
Fold all
More information & hyperlinks
| Web resources: | https://cordis.europa.eu/project/id/948112 |
| Start date: | 01-02-2021 |
| End date: | 31-01-2026 |
| Total budget - Public funding: | 1 499 631,00 Euro - 1 499 631,00 Euro |
Cordis data
Original description
CausalEarth is an interdisciplinary project, aiming to improve our understanding of the interdependencies between major drivers (modes) of climate variability by developing novel statistical causal inference methods for both observations and model data.Disentangling the interdependencies of the major modes, such as El Nino Southern Oscillation and the North Atlantic Oscillation, is key to understand regional climate, and essential for process-based climate model evaluation. The modes' interdependencies are characterized by common drivers, indirect effects, nonlinearities, regime-dependence, and heterogeneous spatio-temporal causal relations. Currently, observational analyses are mostly based on the correlation of scalar (one-dimensional) time series derived from regional averaging or principal component analysis, restricted to supposed causal regimes, e.g., the winter season or phases of multi-decadal climate indices, where dependencies are expected to be stationary. Such scalar correlation approaches fall short in capturing the modes' complex regime-dependent spatio-temporal causal interdependencies.
CausalEarth will develop innovative methods to move (1) from representing complex phenomena as scalar indices to learning spatio-temporal features, (2) from supposing causal regimes to learning them from data, and (3) from correlation to causal dependencies. To this end, CausalEarth will combine recent developments in machine learning with causal inference algorithms.
These methods will be used to infer the causal interdependencies and drivers of major climate modes from observations and to construct the next generation of causal metrics for climate model evaluation.
CausalEarth will push the limits of what can be learned from observational data about causal relations and drive model development towards breakthroughs in projecting our future climate.
Status
SIGNEDCall topic
ERC-2020-STGUpdate Date
27-04-2024
Images
No images available.
Geographical location(s)
