As part of a broader effort to develop next-generation models for numerical weather prediction and climate applications, a hydrostatic atmospheric dynamical core is developed as an intermediate step to evaluate a finite-difference discretization of the primitive equations on spherical icosahedral grids. Based on the need for mass-conserving discretizations for multi-resolution modelling as well as scalability and efficiency on massively parallel computing architectures, the dynamical core is built on triangular C-grids using relatively small discretization stencils. This paper presents the formulation and performance of the baseline version of the new dynamical core, focusing on properties of the numerical solutions in the setting of globally uniform resolution. Theoretical analysis reveals that the discrete divergence operator defined on a single triangular cell using the Gauss theorem is only first-order accurate, and introduces grid-scale noise to the discrete model. The noise can be suppressed by fourth-order hyper-diffusion of the horizontal wind field using a time-step and grid-size-dependent diffusion coefficient, at the expense of stronger damping than in the reference spectral model. A series of idealized tests of different complexity are performed. In the deterministic baroclinic wave test, solutions from the new dynamical core show the expected sensitivity to horizontal resolution, and converge to the reference solution at R2B6 (35 km grid spacing). In a dry climate test, the dynamical core correctly reproduces key features of the meridional heat and momentum transport by baroclinic eddies. In the aqua-planet simulations at 140 km resolution, the new model is able to reproduce the same equatorial wave propagation characteristics as in the reference spectral model, including the sensitivity of such characteristics to the meridional sea surface temperature profile. These results suggest that the triangular-C discretization provides a reasonable basis for further development. The main issues that need to be addressed are the grid-scale noise from the divergence operator which requires strong damping, and a phase error of the baroclinic wave at medium and low resolutions.

Climate and weather variability in the North Atlantic region is determined largely by the North Atlantic Oscillation (NAO). The potential for skillful seasonal forecasts of the winter NAO using an ensemble‐based dynamical prediction system has only recently been demonstrated. Here we show that the winter predictability can be significantly improved by refining a dynamical ensemble through subsampling. We enhance prediction skill of surface temperature, precipitation, and sea level pressure over essential parts of the Northern Hemisphere by retaining only the ensemble members whose NAO state is close to a “first guess” NAO prediction based on a statistical analysis of the initial autumn state of the ocean, sea ice, land, and stratosphere. The correlation coefficient between the reforecasted and observation‐based winter NAO is significantly increased from 0.49 to 0.83 over a reforecast period from 1982 to 2016, and from 0.42 to 0.86 for a forecast period from 2001 to 2017. Our novel approach represents a successful and robust alternative to further increasing the ensemble size, and potentially can be used in operational seasonal prediction systems.

Seasonal prediction is one important element in a seamless prediction chain between weather forecasts and climate projections. After several years of development in collaboration with Universität Hamburg and Max Planck Institute for Meteorology, the Deutscher Wetterdienst performs operational seasonal forecasts since 2016 with the German Climate Forecast System, now in Version 2 (GCFS2.0). Here, the configuration of the previous system GCFS1.0 and the current GCFS2.0 are described and the performance of the two systems is compared over the common hindcast period of 1990–2014. In GCFS2.0, the forecast skill is improved compared to GCFS1.0 during boreal winter, especially for the Northern Hemisphere where the Pearson correlation has increased for the North Atlantic Oscillation index. Overall, a similar performance of GCFS2.0 in comparison to GCFS1.0 is assessed during the boreal summer. Future developments for climate forecasts need a stronger focus on the performance of interannual variability in a model system.

We develop a new Earth System model configuration framed into the ICON architecture, which provides the baseline for the next generation of climate predictions and projections (hereafter ICON XPP – where XPP stands for eXtended Predictions and Projections). ICON XPP comprises the atmospheric component of the numerical weather prediction (ICON NWP), the ICON ocean and land surface components, and an ensemble-variational data assimilation system, all adjusted to an Earth System model for pursuing climate research and operational climate forecasting. Two baseline configurations are presented: (1) a 160 km atmosphere and a 40 km ocean resolution, and (2) 80 km atmosphere and 20 km ocean resolution. A CMIP DECK (Diagnostic, Evaluation and Characterization of Klima) experimentation framework is used for a first evaluation. ICON XPP depicts the basic properties of the coupled climate. The pre-industrial climate shows a top-of-atmosphere balanced radiation budget and a mean global near-surface temperature of 13.8–14.0 °C. The ocean shows circulation strengths in the range of the observed values, such as the AMOC at 16–18 Sv and the flows through the common passages. The current climate is characterized by a trend in the global mean temperature of ∼ 1.2 °C since the 1850s, similar to reference datasets. Regionally, the hydroclimate differs greatly from observed conditions. For example, the inter-tropical convergence zone (ITCZ) has a double peak and a wet southern subtropical branch across the oceans. Further, the Southern Ocean sea surface temperature has a strong positive mean bias with temperatures up to 5 °C higher than observations. Dynamical processes, such as El Niño/Southern Oscillation (ENSO) performs similarly to CMIP6-like coupled models. Tropical waves and the Madden-Julian Oscillation are well captured, and the 40 km atmospheric configuration has a spontaneous weak quasi-biennial oscillation. The atmospheric dynamics in the northern extra-tropics of both configurations represent well the position of the jet stream as well as the influences of the transient momentum transports and their feedbacks on the jet stream. Overall, ICON XPP performs similarly to climate models performed in CMIP6 making it a good basis for climate forecasts and projections, and climate research.

The variability of the sea level pressure in the North Atlantic sector is the most important driver of weather and climate in Europe. The main mode of this variability, the North Atlantic Oscillation (NAO), explains up to 50% of the total variance. Other modes, known as the Scandinavian index, East Atlantic, and East Atlantic/West Russian pattern, complement the variability of the sea level pressure, thereby influencing the European climate. It has been shown previously that a seasonal prediction system with enhanced winter NAO skill due to ensemble subsampling entails an improved prediction of the surface climate variables as well. Here, we show that a refined subselection procedure that accounts both for the NAO index and for the three additional modes of sea level pressure variability is able to further increase the prediction skill of wintertime mean sea level pressure, near‐surface temperature, and precipitation across Europe.

Variability of the North Atlantic Oscillation (NAO) drives wintertime temperature anomalies in the Northern Hemisphere. Dynamical seasonal prediction systems can skilfully predict the winter NAO. However, prediction of the NAO‐dependent air temperature anomalies remains elusive, partially due to the low variability of predicted NAO. Here, we demonstrate a hidden potential of a multi‐model ensemble of operational seasonal prediction systems for predicting wintertime temperature by increasing the variability of predicted NAO. We identify and subsample those ensemble members which are close to NAO index statistically estimated from initial autumn conditions. In our novel multi‐model approach, the correlation prediction skill for wintertime Central Europe temperature is improved from 0.25 to 0.66, accompanied by an increased winter NAO prediction skill of 0.9. Thereby, temperature anomalies can be skilfully predicted for the upcoming winter over a large part of the Northern Hemisphere through increased variability and skill of predicted NAO.