Research Focus ATMOSPHERE

The AI4PEX project aims to enhance the representation of atmospheric processes, namely, cloud feedbacks, in Earth System Models (ESMs) using machine learning (ML). AI4PEX will be leveraging new observation-based datasets and ML approaches to focus on the representation of processes occurring at scales smaller than ESMs' model grids. Overall, AI4PEX focuses on:

1. Enhancing knowledge: of the dynamics of diurnal evolution and variability of stratocumulus and cumulus clouds over the ocean using climate data records and AI.
2. Improving Cloud Feedback Representation: by learning ML-based parameterizations from global storm resolving simulations (GSRMs) and enhanced observations.
3. Model Evaluation: leveraging explainable AI (xAI) principles to deliver interpretable model evaluation and pave way to more effective bias reduction approaches.

Enhancing the representation of cloud feedbacks and model uncertainty will deliver significant improvements in projecting Equilibrium Climate Sensitivity (ECS) and future global warming. Through these efforts, AI4PEX seeks to transform the modelling of atmospheric processes and improve climate projections.

Here's a brief overview of where and how is planned:

Task T1.1: Cloud Regime and Process-dependent Exploitation of EO

[lead: SMHI]

• Objective: Disentangle the couplings between meteorology and low cloud regimes.
• Approach: Develop an AI/ML framework to analyse satellite-based cloud data and meteorological information from reanalysis.
• Focus Area:
  1. Diurnal evolution of stratocumulus and cumulus clouds over oceans.
  2. Cloud controlling factors such as sea surface temperatures (SSTs), inversions, humidity, and winds.
  3. Climatological insights from the satellite-based cloud climate data records.

Task T2.2: ML-based Parametrizations for the Atmosphere

[lead: UNIL; contributions: MF-CNRM, DLR, METO, UNIVLEEDS]

• Objective: Improve representations of uncertain atmospheric processes.
• Approach: Combine high-resolution simulations and observations to develop ML-based parametrizations for atmospheric models.
• Focus Area:
  1. Cloud feedbacks.
  2. Atmosphere-ocean heat exchanges.

Task T3.2: Constraining and Quantifying Model Uncertainty for the Atmosphere

[lead: DLR; contributions: METO, UNIL]

• Objective: Calibrate and quantify uncertainties in atmospheric models.
• Methods: Use of a wide range of Earth Observation, re-analyses datasets and emulation to develop automatic tuning methods.
• Focus Area:
  1. Atmospheric processes affecting cloud feedbacks.

Task T4.2: Evaluation of ML-enhanced Atmosphere Models

[lead: METO; contributions: DLR, UNIL, UVEG]

• Objective: Evaluate ML-based parametrizations in simulating subtropical low clouds and their radiative sensitivities.
• Tools: Use process-level diagnostics, explainable AI, and causality methods to assess cloud feedbacks and their climate invariance.
• Focus Area:
  1. Subtropical low clouds processes.
  2. Cloud feedbacks.

Task T6.3: Coupling ML-enhanced Land-Atmosphere Models

[lead: CNRS-IPSL; contributions: MPG, DLR]

• Objective: Couple the ML-enhanced atmosphere and land models to study changes in climate extremes.
• Approach: Conduct ensemble runs to study climate variability and evaluate the coupled model against observations by using prescribed SST and sea ice cover.
• Focus Area:
  1. Climate Extremes.
  2. Regional interactions and impacts on energy, water, and C cycles.

AI4PEX aims to drive a paradigm shift towards data-driven, physics-aware hybrid models, ultimately improving projections of Equilibrium Climate Sensitivity (ECS) and future global warming.