The radiation parameterization is one of the computationally most expensive components of Earth system models (ESMs). To reduce computational cost, radiation is often calculated on coarser spatial or temporal scales, or both, than other physical processes in ESMs, leading to uncertainties in cloud-radiation interactions and thereby in radiative temperature tendencies. One way around this issue is the emulation of the radiation parameterization using machine learning which is usually faster and has good accuracy in a high dimensional parameter space. This study investigates the development and interpretation of a machine learning based radiation emulator using the ICOsahedral Non-hydrostatic (ICON) model with the RTE-RRTMGP radiation code which calculates radiative fluxes based on the atmospheric state and its optical properties. With a Bidirectional Long Short-Term Memory (Bi-LSTM) architecture, which can account for vertical bidirectional auto-correlation, we can accurately emulate shortwave and longwave heating rates with a mean absolute error of $0.049~K/d\,(2.50\%)$ and $0.069~K/d\,(5.14\%)$ respectively. Further, we analyse the trained neural networks using Shapley Additive exPlanations (SHAP) and confirm that the networks have learned physical meaningful relationships among the inputs and outputs. Notably, we observe that the local temperature is used as a predictive source for the longwave heating, consistent with physical models of radiation. For shortwave heating, we find that clouds reflect radiation, leading to reduced heating below the cloud.

This study presents the first attempt to directly simulate a full cycle of the quasi-biennial oscillation (QBO) in a global storm-resolving model (GSRM) that explicitly simulates deep convection and gravity waves instead of parameterizing them. Using the Icosahedral Nonhydrostatic (ICON) model with horizontal and vertical resolutions of about 5 km and 400 m, respectively, we show that an untuned state-of-the-art global storm-resolving model is already on the verge of simulating a QBO-like oscillation of the zonal wind in the tropical stratosphere for the right reasons. ICON shows overall good fidelity in simulating the QBO momentum budget and the downward propagation of the QBO jets in the upper QBO domain (25 km–35 km). In the lowermost stratosphere, however, ICON does not simulate the downward propagation of the QBO jets to the tropopause. This is the result of a pronounced lack of QBO wave forcing, mainly on planetary scales. We show that the lack of planetary-scale wave forcing in the lowermost stratosphere is caused by an underestimation of the planetary-scale wave momentum flux entering the stratosphere, which is too weak by 20%–40%. We attribute this lack of planetary-scale wave momentum flux to a substantial lack of convectively coupled equatorial waves in the tropical troposphere. Therefore, we conclude that in the present global storm-resolving model, simulating a realistic spatio-temporal variability of tropical deep convection, in particular convectively coupled equatorial waves, is currently the main roadblock towards simulating a reasonable QBO.

The domain nesting of the icosahedral non-hydrostatic (ICON) model has been used operationally at Deutscher Wetterdienst for several years. Now it was also made available for the atmospheric part of the ICON Earth system model. With this new climate configuration, regionally higher resolved simulations without the additional use of a separate regional climate model (RCM) are possible. Simulations were performed for the years 1979-2010 at a global resolution of about 80 km and a subdomain over Europe at 40 km resolution. Two simulations with this setup were evaluated and compared: one with a feedback from the regional subdomain to the global domain (two-way nesting) and one without feedback (one-way nesting). The mean atmospheric state of both simulations on the global scale is only slightly different compared to a reference experiment. However, comparisons to reanalyses show regionally distinct biases. The feedback from the subdomain to the global domain has a similar impact over Europe as a globally higher resolution, indicating a stronger North-Atlantic Oscillation at higher horizontal resolution. Over Europe, the skill is higher in the subdomain than in the global domain, but no systematic advantages can be attributed to the feedback. Artifacts at the lateral boundaries of the regional subdomain, as they are known from RCM simulations, also occur strongly in the simulation without feedback and are eliminated by allowing the feedback. A further reduction of resolution dependency of model physics is supposed to improve particularly the simulation with feedback.

• This work documents ICON-ESM 1.0, the first version of a coupled model based 19 on the ICON framework 20 • Performance of ICON-ESM is assessed by means of CMIP6 DECK experiments 21 at standard CMIP-type resolution 22 • ICON-ESM reproduces the observed temperature evolution. Biases in clouds, winds, 23 sea-ice, and ocean properties are larger than in MPI-ESM. Abstract 25 This work documents the ICON-Earth System Model (ICON-ESM V1.0), the first cou-26 pled model based on the ICON (ICOsahedral Non-hydrostatic) framework with its un-27 structured, icosahedral grid concept. The ICON-A atmosphere uses a nonhydrostatic dy-28 namical core and the ocean model ICON-O builds on the same ICON infrastructure, but 29 applies the Boussinesq and hydrostatic approximation and includes a sea-ice model. The 30 ICON-Land module provides a new framework for the modelling of land processes and 31 the terrestrial carbon cycle. The oceanic carbon cycle and biogeochemistry are repre-32 sented by the Hamburg Ocean Carbon Cycle module. We describe the tuning and spin-33 up of a base-line version at a resolution typical for models participating in the Coupled 34 Model Intercomparison Project (CMIP). The performance of ICON-ESM is assessed by 35 means of a set of standard CMIP6 simulations. Achievements are well-balanced top-of-36 atmosphere radiation, stable key climate quantities in the control simulation, and a good 37 representation of the historical surface temperature evolution. The model has overall bi-38 ases, which are comparable to those of other CMIP models, but ICON-ESM performs 39 less well than its predecessor, the Max Planck Institute Earth System Model. Problem-40 atic biases are diagnosed in ICON-ESM in the vertical cloud distribution and the mean 41 zonal wind field. In the ocean, sub-surface temperature and salinity biases are of con-42 cern as is a too strong seasonal cycle of the sea-ice cover in both hemispheres. ICON-43 ESM V1.0 serves as a basis for further developments that will take advantage of ICON-44 specific properties such as spatially varying resolution, and configurations at very high 45 resolution. 46 Plain Language Summary 47 ICON-ESM is a completely new coupled climate and earth system model that ap-48 plies novel design principles and numerical techniques. The atmosphere model applies 49 a non-hydrostatic dynamical core, both atmosphere and ocean models apply unstruc-50 tured meshes, and the model is adapted for high-performance computing systems. This 51 article describes how the component models for atmosphere, land, and ocean are cou-52 pled together and how we achieve a stable climate by setting certain tuning parameters 53 and performing sensitivity experiments. We evaluate the performance of our new model 54 by running a set of experiments under pre-industrial and historical climate conditions 55 as well as a set of idealized greenhouse-gas-increase experiments. These experiments were 56 designed by the Coupled Model Intercomparison Project (CMIP) and allow us to com-57 pare the results to those from other CMIP models and the predecessor of our model, the 58 Max Planck Institute for Meteorology Earth System Model. While we diagnose overall 59 satisfactory performance, we find that ICON-ESM features somewhat larger biases in 60 several quantities compared to its predecessor at comparable grid resolution. We empha-61 size that the present configuration serves as a basis from where future development steps 62 will open up new perspectives in earth system modelling. 63

A promising approach to improve cloud parameterizations within climate models and thus climate projections is to use deep learning in combination with training data from storm-resolving model (SRM) simulations. The Icosahedral Non-Hydrostatic (ICON) modeling framework permits simulations ranging from numerical weather prediction to climate projections, making it an ideal target to develop neural network (NN) based parameterizations for sub-grid scale processes. Within the ICON framework, we train NN based cloud cover parameterizations with coarse-grained data based on realistic regional and global ICON SRM simulations. We set up three different types of NNs that differ in the degree of vertical locality they assume for diagnosing cloud cover from coarse-grained atmospheric state variables. The NNs accurately estimate sub-grid scale cloud cover from coarse-grained data that has similar geographical characteristics as their training data. Additionally, globally trained NNs can reproduce sub-grid scale cloud cover of the regional SRM simulation. Using the game-theory based interpretability library SHapley Additive exPlanations, we identify an overemphasis on specific humidity and cloud ice as the reason why our column-based NN cannot perfectly generalize from the global to the regional coarse-grained SRM data. The interpretability tool also helps visualize similarities and differences in feature importance between regionally and globally trained column-based NNs, and reveals a local relationship between their cloud cover predictions and the thermodynamic environment. Our results show the potential of deep learning to derive accurate yet interpretable cloud cover parameterizations from global SRMs, and suggest that neighborhood-based models may be a good compromise between accuracy and generalizability.

This work documents the ICON-Earth System Model (ICON-ESM V1.0), the first coupled model based on the ICON (ICOsahedral Non-hydrostatic) framework with its unstructured, isosahedral grid concept. The ICON-A atmosphere uses a nonhydrostatic dynamical core and the ocean model ICON-O builds on the same ICON infrastructure, but applies the Boussinesq and hydrostatic approximation. The oceanic carbon cycle and biogeochemistry is represented by the HAMOCC6 module and the terrestrial biogeophysical and biogeochemical process are integrated in the new JSBACH4 module. We describe the tuning and spin-up of a base-line version at a resolution typical for models participating in the Coupled Model Intercomparison Project (CMIP). The performance of ICON-ESM is assessed by means of a set of standard CMIP6 simulations. Achievements are well-balanced top-of-atmosphere radiation, stable key climate quantities in the control simulation, and a good representation of the historical surface temperature evolution. The model has overall biases, which are comparable to those of other CMIP models, but ICON-ESM performs less well than its predecessor, the MPI-ESM. Problematic biases are diagnosed in ICON-ESM in the vertical cloud distribution and the mean zonal wind field. In the ocean, sub-surface temperature and salinity biases are of concern as is a too strong seasonal cycle of the sea-ice cover in both hemispheres. ICON-ESM V1.0 serves as a basis for further developments that will take advantage of ICON-specific properties such as spatially varying resolution, and coupled configurations at very high resolution.