English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Climate models are essential to understand and project climate change, yetlong-standing biases and uncertainties in their projections remain. This islargely associated with the representation of subgrid-scale processes,particularly clouds and convection. Deep learning can learn these subgrid-scaleprocesses from computationally expensive storm-resolving models while retainingmany features at a fraction of computational cost. Yet, climate simulationswith embedded neural network parameterizations are still challenging and highlydepend on the deep learning solution. This is likely associated with spuriousnon-physical correlations learned by the neural networks due to the complexityof the physical dynamical system. Here, we show that the combination ofcausality with deep learning helps removing spurious correlations andoptimizing the neural network algorithm. To resolve this, we apply a causaldiscovery method to unveil key physical drivers in the set of input predictorsof atmospheric subgrid-scale processes of a superparameterized climate model inwhich deep convection is explicitly resolved. The resulting causally-informedneural networks are coupled to the climate model, hence, replacing thesuperparameterization and radiation scheme. We show that the climatesimulations with causally-informed neural network parameterizations retain manyconvection-related properties of the original high-resolution climate model andclearly outperform the non-causal approach, while retaining similargeneralization capabilities to unseen climates. The combination of causaldiscovery and deep learning is a new and promising approach that leads tostable and more accurate climate simulations and paves the way towards morephysically-based causal deep learning approaches also in other scientificdisciplines.

Causally-informed deep learning to improve climate models and projections