Kai Keller

Small islands -- especially developing states -- are among the regions most vulnerable to climate change impacts, yet their heterogeneous climates and often complex topographies are not captured by traditional global climate models. In this study, we assess the next-generation kilometer-scale (km-scale) global climate simulations from the Climate Change Adaptation Digital Twin (ClimateDT), implemented in the Destination Earth initiative of the European Commission, focusing on islands in the Canary, Mascarene, and eastern Caribbean archipelagos. We evaluate global km-scale climate historical simulations from IFS-NEMO, IFS-FESOM, and ICON coupled models, along with available CORDEX regional simulations, reanalysis products, and station and gridded observations over 1990–2014. The evaluation focuses on the variables most relevant for impact studies; 2 m temperature and precipitation, including mean climate, diurnal and seasonal variability, and extremes. We find that the km-scale models consistently represent island-scale spatial structure, particularly the sharp gradients associated with elevation and windward–leeward contrasts, evidenced most strongly over the mountainous islands of Tenerife, Gran Canaria and Réunion. The km-scale simulations also generally capture the phase and structure of diurnal cycles, despite systematic cold and dry biases in many regions. Over the eastern Caribbean, the ClimateDT simulations are among the first global models to explicitly represent small island land–sea contrasts, although precipitation remains challenging and several simulations retain strong dry biases. Overall, these results highlight that one of the main benefits of km-scale modeling is the improved representation of the physical processes that shape local climate, rather than a uniform reduction in error.

The Climate Change Adaptation Digital Twin (Climate DT), developed as part of the European Commission's Destination Earth (DestinE) initiative, sets up an operational system for producing multi-decadal, multi-model global climate projections and translating climate data into climate impact information to support adaptation efforts. This system delivers data with local granularity at spatial resolutions of 5–10 km and hourly outputs, leading to globally consistent information at scales that matter for decision-making. It also enables the testing of what-if scenarios such as high-resolution storylines, which are physically consistent global simulations of extreme events under different climate conditions and provide contextual insights to support concrete adaptation decisions. They support the generation of more equitable (understood as accessible and relevant across regions) climate information. The Climate DT is built on cutting-edge infrastructure, expert collaboration, and digital innovation. It is designed to support on-demand responses to policy questions, with quantified uncertainty. It will foster interactivity by allowing users to influence simulation design, model output portfolios, and application integration through co-design. AI-based tools, including emulators and chatbots, are being developed in parallel to enhance climate information access. Sector-specific applications are embedded in the system to synchronously translate climate data into tailored climate-impact indicators, with examples provided for energy, water, and forest management. The applications have been co-designed with informed users. A unified, cross-platform workflow defines the orchestration of all components, which is handled by a single workflow manager and relies on containerised components, facilitating automation, portability, maintainability, and traceability. Data management is unified using standard grids (HEALPix), ensuring consistency and easing data usability under a strict governance policy. Streaming enables real-time data use by the data consumers and unlocks access to the unprecedented data wealth produced by the high-resolution simulations. Monitoring tools provide real-time quality control of data and model outputs and enable continuous assessment of the realism of the climate simulations during Climate DT operation. The compute-intensive system is powered by world-class supercomputing capabilities through a strategic partnership with the European High Performance Computing Joint Undertaking (EuroHPC). Despite high computational demands, the Climate DT sets a new benchmark for delivering equitable, credible, and actionable climate information. It complements existing initiatives like CMIP, CORDEX, and national and European climate services, and aligns with global climate science goals to support climate adaptation.

A suite of high-resolution configurations of the coupled climate model IFS-NEMO to investigate recent and future climate variability and change has been recently developed within the project EERIE (European Eddy-Rich Earth System Models) and the Destination Earth initiative. This contribution emphasises the climate responses emerging from these simulations and their sensitivity to spatial resolution and the experimental protocol considered.The model hierarchy combines eddy-permitting (∼25 km) and eddy-rich (∼9 km) ocean components with convection-parameterised (∼25 km) and convection-permitting (∼4.5 km) atmospheric configurations, enabling a systematic assessment of resolution-dependent processes and feedbacks. Particular attention is given to how differences in model physics, resolution-aware tuning strategies, scenario forcing (i.e. SSP1-2.6 vs SSP3-7.0) and experimental design influence the simulated climate variability across configurations.Ongoing analyses of historical simulations show enhanced performance for the higher-resolution configurations in the representation of mean-state properties, with especially clear improvements in dynamical fields. We further assess the extent to which the shorter spinup approach employed in Destination Earth, compared to EERIE, can reliably capture internal variability and externally forced responses while substantially reducing computational cost.A systematically stronger future response of the Atlantic Meridional Overturning Circulation to external forcings is found in the eddy-resolving configurations compared to the eddy-permitting ones. Idealised control simulations with quadrupled CO2 forcing – inspired by the CMIP6 DECK experiments – also show a more pronounced temperature response at the highest resolution compared with the ∼25 km configuration, thus yielding stronger climate sensitivity.More generally, we also briefly outline emerging applications of the kilometre-scale IFS-NEMO model in other European research projects, including TerraDT, which focuses on land–atmosphere coupling, and PREDDYCT, which investigates the role of mesoscale ocean eddies in seasonal-to-decadal climate prediction. Together, these efforts highlight the added scientific value of high-resolution climate modelling for understanding forced responses and informing future climate projections.

As climate models advance toward higher resolutions, they become increasingly capable of resolving key Earth system processes, which in turn raises the need for robust and quantitative evaluation methods. In response to this challenge, we present pyhanami, an open-source Python package developed within the HANAMI project to assess the replicability and scientific skill of Earth System Models (ESMs) using statistical testing and objective, scalar-based metrics. Besides, to facilitate the practical application of these evaluations, pyhanami features a structured data interface that efficiently loads and inspects compatible model outputs.An ESM is considered replicable if the same experiment run on different computing environments or with different compilers produces identical results, i.e., representing the same climate. This ensures that differences between simulations reflect only the intended scientific changes in the model setup. Because bit-for-bit replicability is often unattainable across environments due to the chaotic nature of climate models, our practical goal is to achieve statistical indistinguishability. Building on existing methodologies, pyhanami provides an ensemble-based replicability test that combines multiple statistical tests and metrics to determine whether two simulated ensembles are statistically indistinguishable, as described in (K.Keller et al., 2025; doi.org/10.5194/gmd-18-10221-2025). To the best of our knowledge, automated and standardized replicability assessment is not currently supported in model evaluation tools, despite its importance for climate model development, validation, intercomparison, and porting. Complementing replicability, the scientific skill of an ESM describes its ability to accurately reproduce observed features of the climate system, from regional patterns to large-scale teleconnections. Many existing tools to evaluate this skill rely on visualization-based diagnostics, which often require expert knowledge and can be biased by subjective interpretation. In contrast, scalar metrics and scores provide quantitative and comparable measures of scientific skill, which are essential for interpreting climate projections, guiding model development, and model intercomparison. However, diagnostics for physical processes that require km-scale, high-resolution global climate models to be properly resolved remain underrepresented in state-of-the-art diagnostic suites. Although several metrics have been proposed for such small-scale processes, many lack standardized and widely available implementations. As high-resolution climate simulations become more common, the demand for objective diagnostics to support model tuning and improvement is increasing. pyhanami addresses this need by providing a growing set of scalar scientific skill metrics that enable quantitative and easily interpretable evaluation of phenomena such as Tropical Cyclones and the Tropical Intraseasonal Oscillation (ISO), including the Madden-Julian Oscillation and the Boreal Summer ISO modes. By integrating replicability testing, scientific skill metrics, and visualization tools into a single, self-contained package with a generic data interface, pyhanami streamlines evaluation workflows and supports the development of reliable climate projections, advancing the quality and reproducibility of geosciences research.

Climate models are an important tool to address challenges we face due to the changing climate. The global warming of the atmosphere and ocean leads to an increase in energy accessible to foster more frequent and intense tropical cyclones, extreme precipitation, and heatwaves, causing increasingly larger economic and non-economic damage. Many of those events are caused or influenced by small-scale convective processes that are not resolved in the typical CMIP-style models with resolutions of about 100 km.Models capable of resolving deep convection and large turbulent eddies in the atmosphere require horizontal resolutions between 1 km and 10 km. Turbulent processes in the atmosphere play a major role in distributing the energy within the atmosphere, and it has been shown that atmospheric models at resolutions of about 10 km or less significantly improve the resemblance to observations, for instance, regarding the magnitude of maximum wind gusts and the statistics and characteristics of tropical cyclones. Similarly, ocean models require resolutions in the order of 10 km or finer to explicitly resolve mesoscale ocean eddies and their contributions to the transport of salinity and heat and their effects upon the global system. Before we can make the transient historical simulations from which future projections are typically initialized with climate models, based on a certain emission scenario, the models need to achieve a climate state that is consistent with the boundary conditions at the initial time. For this, the model needs to be gradually spun up to reach a balanced state. Traditional approaches for model tuning and spinup used for coarse resolution models cannot be applied at very high resolutions. Typical spinup times to reach model equilibrium are around 1000 years, which remains unrealistic to achieve for km-scale models until today. This work presents the analysis of alternative cost-efficient spinup protocols and evaluates their associated initial shocks and drifts and how efficiently the coupled model approaches the equilibrium. We also contribute to answering the question of how reliable future projections are when initialized from a transient model state. The analysis is based on a series of ensemble simulations performed with the coupled IFS-NEMO climate model at about 25 km atmospheric and ocean resolution, i.e., Tco399/eORCA025 grids, and on different combinations of ocean-only spinup and coupled spinup lengths. Our analysis focuses on spinup designs that are optimized to initialize climate projections and historical simulations of 50 to 100 years with a minimal initial adjustment and “well-behaved” model trends, contrasting them to the existing multi-decadal km-scale simulations from initiatives like Destination Earth and EERIE.

Climate simulations with Earth System Models (ESMs) constitute the basis of our knowledge about the projected climate change for the coming decades. They represent the major source of knowledge for the Intergovernmental Panel on Climate Change (IPCC), and are an indispensable tool in confronting climate change. Yet, ESMs are imperfect and the climate system that they simulate is highly non-linear. Therefore, small errors in the numerical representation, initial states, and boundary conditions provided for future scenarios, quickly develop large uncertainties in the trajectories of their projected climates. To improve the confidence and minimize the uncertainty, future projections use large ensembles of simulations with the same model, and ensembles with multiple models. Using these two types of ensembles concurrently addresses two kinds of uncertainty in the simulations, (1) the limited spatiotemporal accuracy of the initial states, and (2) the uncertainty in the numerical representation of the climate system (model error). The uncertainty for the future development of the anthropogenic climate drivers is addressed by projecting different Shared Socioeconomic Pathway (SSP) scenarios. Organizing multimodel ensembles to make confident statements about the future climate addressing different SSP scenarios is a tremendous collaborative effort. The Coupled Model Intercomparison Project (CMIP) addresses this challenge, with the participation of 33 modeling groups in 16 countries. As one among numerous challenges that such undertaking poses, we are addressing model replicability in this article. The anticipated number of simulated years in the 6th CMIP phase (CMIP6) accumulated to about 40 000 years. With typical values for the computational throughput of about 1 to 15 simulated years per day (SYPD), it is clear that the simulations needed to be distributed among different clusters to be completed within a reasonable amount of time. Model replicability addresses the question, whether the climate signal from different scientific scenarios generated by the same model, performed on different clusters, can be attributed exclusively to the differences in the scientific drivers. It has been shown, that even changing specific compiler flags, leads to significant changes in certain climatological fields. Model replicability holds, when the model climatologies derived from the same model under a different computing environment, are statistically indistinguishable. If replicability does not hold, we cannot be certain that differences in the model climate are exclusively attributed to differences in the scientific setups. In this article, we present a novel methodology to test replicability. We further establish an objective measure of what constitutes a different climate based on Cohen's effect size. We provide a thorough analysis of the performance of our methodology and show that we can improve the performance of a recent state-of-the-art method by 60 %. We further provide an estimate of the ensemble size that is required to prove replicability with confidence. We find that an effect size of d=0.2 can be used as a threshold for statistical indistinguishability. Our analysis, based on the Community Earth System Model 2 (CESM2) Large Ensemble Community Project (LENS2) 100-member ensemble, shows that with 50 members, we can resolve effect sizes of ∼0.3, and with ensembles of 20 members, we can still resolve effect sizes of ∼0.35. We further provide a robust methodology to objectively determine the required ensemble size, depending on the purpose and requirement of the replicability test.

The Climate Change Adaptation Digital Twin (Climate DT), developed as part of the European Commission’s Destination Earth (DestinE) initiative, is a pioneering effort to build an operational climate information system in support of adaptation. This system produces global climate simulations with local granularity, providing information at scales that matter for decision-making. The Climate DT delivers multi-decadal climate simulations at spatial resolutions of 5–10 km, with hourly outputs, offering globally consistent, frequently updated data. The km-scale simulations address some limitations of current climate models, improving local granularity and reducing longstanding biases, supporting more equitable (understood as accessible and relevant across regions) and credible climate information. The Climate DT is built on cutting-edge infrastructure, expert collaboration, and digital innovation. It supports real-time, on-demand responses to policy questions, with quantified uncertainty. It fosters interactivity by allowing users to influence simulation design, model outputs, and applications through co-design. AI-based tools, including emulators and chatbots, are being developed to enhance flexible scenario exploration and ease climate information access. Sector-specific applications are embedded in the system to generate tailored climate-impact indicators, with examples for energy, water, and forest management. The applications have been co-designed with informed users. An important innovation is the use of high-resolution storylines. These are physically consistent simulations of extreme events under different climate conditions that provide contextual insights to support concrete adaptation decisions. A unified workflow across platforms orchestrates all components ensuring automation, containerisation for portability, and traceability. The unified data management ensures consistency and eases the usability of the data. Data is delivered using standard grids (HEALPix) at high-frequency (hourly) and follows a strict governance policy. Streaming enables real-time data use and unlocks access to the unprecedented data produced by the high- resolution simulations. Monitoring tools provide real-time quality control of both data and models and provide diagnostics during the Climate DT operation. The compute-intensive system is powered by world-class supercomputing capabilities through a strategic partnership with the European High Performance Computing Joint Undertaking (EuroHPC). Despite high computational demands, the Climate DT sets a new benchmark for delivering equitable, credible, and actionable climate information. In this way it complements existing initiatives like CMIP, CORDEX, and national and European climate services, and aligns with global climate science goals for climate adaptation.

The Climate Change Adaptation Digital Twin (Climate DT), developed as part of the European Commission’s Destination Earth (DestinE) initiative, is a pioneering effort to build an operational climate information system in support of adaptation. This system produces global climate simulations with local granularity, providing information at scales that matter for decision-making. The Climate DT delivers multi-decadal climate simulations at spatial resolutions of 5–10 km, with hourly outputs, offering globally consistent, frequently updated data. The km-scale simulations address some limitations of current climate models, improving local granularity and reducing longstanding biases, supporting more equitable (understood as accessible and relevant across regions) and credible climate information. The Climate DT is built on cutting-edge infrastructure, expert collaboration, and digital innovation. It supports real-time, on-demand responses to policy questions, with quantified uncertainty. It fosters interactivity by allowing users to influence simulation design, model outputs, and applications through co-design. AI-based tools, including emulators and chatbots, are being developed to enhance flexible scenario exploration and ease climate information access. Sector-specific applications are embedded in the system to generate tailored climate-impact indicators, with examples for energy, water, and forest management. The applications have been co-designed with informed users. An important innovation is the use of high-resolution storylines. These are physically consistent simulations of extreme events under different climate conditions that provide contextual insights to support concrete adaptation decisions. A unified workflow across platforms orchestrates all components ensuring automation, containerisation for portability, and traceability. The unified data management ensures consistency and eases the usability of the data. Data is delivered using standard grids (HEALPix) at high-frequency (hourly) and follows a strict governance policy. Streaming enables real-time data use and unlocks access to the unprecedented data produced by the high- resolution simulations. Monitoring tools provide real-time quality control of both data and models and provide diagnostics during the Climate DT operation. The compute-intensive system is powered by world-class supercomputing capabilities through a strategic partnership with the European High Performance Computing Joint Undertaking (EuroHPC). Despite high computational demands, the Climate DT sets a new benchmark for delivering equitable, credible, and actionable climate information. In this way it complements existing initiatives like CMIP, CORDEX, and national and European climate services, and aligns with global climate science goals for climate adaptation.

Climate simulations with Earth System Models (ESMs) constitute the basis of our knowledge about the projected climate change for the coming decades. They represent the major source of knowledge for the Intergovernmental Panel on Climate Change (IPCC), and are an indispensable tool in confronting climate change. Yet, ESMs are imperfect and the climate system that they simulate is highly non-linear. Therefore, small errors in the numerical representation, initial states, and boundary conditions provided for future scenarios, quickly develop large uncertainties in the trajectories of their projected climates. To improve the confidence and minimize the uncertainty, future projections use large ensembles of simulations with the same model, and ensembles with multiple models. Using these two types of ensembles concurrently addresses two kinds of uncertainty in the simulations, (1) the limited spatiotemporal accuracy of the initial states, and (2) the uncertainty in the numerical representation of the climate system (model error). The uncertainty for the future development of the anthropogenic climate drivers is addressed by projecting different Shared Socioeconomic Pathway (SSP) scenarios. Organizing multimodel ensembles to make confident statements about the future climate addressing different SSP scenarios is a tremendous collaborative effort. The Coupled Model Intercomparison Project (CMIP) addresses this challenge, with the participation of 33 modeling groups in 16 countries. As one among numerous challenges that such undertaking poses, we are addressing model replicability in this article. The anticipated number of simulated years in the 6th CMIP phase (CMIP6) accumulated to about 40,000 years. With typical values for the computational throughput of about 1 to 15 simulated years per day (SYPD), it is clear that the simulations needed to be distributed among different clusters to be completed within a reasonable amount of time. Model replicability addresses the question, whether the climate signal from different scientific scenarios generated by the same model, performed on different clusters, can be attributed exclusively to the differences in the scientific drivers. It has been shown, that even changing specific compiler flags, leads to significant changes in certain climatological fields. Model replicability holds, when the model climatologies derived from the same model under a different computing environment, are statistically indistinguishable. If replicability does not hold, we cannot be certain that differences in the model climate are exclusively attributed to differences in the scientific setups. In this article, we present a novel methodology to test replicability. We further establish an objective measure of what constitutes a different climate based on Cohen's effect size. We provide a thorough analysis of the performance of our methodology and show that we can improve the performance of a recent state-of-the-art method by 60 %. We further provide an estimate of the ensemble size that is required to prove replicability with confidence. We find that an effect size of d = 0.2 can be used as a threshold for statistical indistinguishability. Our analysis, based on the Community Earth System Model 2 (CESM2) Large Ensemble Community Project (LENS2) 100-member ensemble, shows that with 50 members, we can resolve effect sizes of about 0.3, and with ensembles of 20 members, we can still resolve effect sizes of ~ 0.35. We further provide a robust methodology to objectively determine the required ensemble size, depending on the purpose and requirement of the replicability test.

Climate simulations with Earth System Models (ESMs) constitute the basis of our knowledge about the projected climate change for the coming decades. They represent the major source of knowledge for the Intergovernmental Panel on Climate Change (IPCC), and are an indispensable tool in confronting climate change. Yet, ESMs are imperfect and the climate system that they simulate is highly non-linear. Therefore, small errors in the numerical representation, initial states, and boundary conditions provided for future scenarios, quickly develop large uncertainties in the trajectories of their projected climates. To improve the confidence and minimize the uncertainty, future projections use large ensembles of simulations with the same model, and ensembles with multiple models. Using these two types of ensembles concurrently addresses two kinds of uncertainty in the simulations, (1) the limited spatiotemporal accuracy of the initial states, and (2) the uncertainty in the numerical representation of the climate system (model error). The uncertainty for the future development of the anthropogenic climate drivers is addressed by projecting different Shared Socioeconomic Pathway (SSP) scenarios. Organizing multimodel ensembles to make confident statements about the future climate addressing different SSP scenarios is a tremendous collaborative effort. The Coupled Model Intercomparison Project (CMIP) addresses this challenge, with the participation of 33 modeling groups in 16 countries. As one among numerous challenges that such undertaking poses, we are addressing model replicability in this article. The anticipated number of simulated years in the 6th CMIP phase (CMIP6) accumulated to about 40,000 years. With typical values for the computational throughput of about 1 to 15 simulated years per day (SYPD), it is clear that the simulations needed to be distributed among different clusters to be completed within a reasonable amount of time. Model replicability addresses the question, whether the climate signal from different scientific scenarios generated by the same model, performed on different clusters, can be attributed exclusively to the differences in the scientific drivers. It has been shown, that even changing specific compiler flags, leads to significant changes in certain climatological fields. Model replicability holds, when the model climatologies derived from the same model under a different computing environment, are statistically indistinguishable. If replicability does not hold, we cannot be certain that differences in the model climate are exclusively attributed to differences in the scientific setups. In this article, we present a novel methodology to test replicability. We further establish an objective measure of what constitutes a different climate based on Cohen's effect size. We provide a thorough analysis of the performance of our methodology and show that we can improve the performance of a recent state-of-the-art method by 60 %. We further provide an estimate of the ensemble size that is required to prove replicability with confidence. We find that an effect size of d = 0.2 can be used as a threshold for statistical indistinguishability. Our analysis, based on the Community Earth System Model 2 (CESM2) Large Ensemble Community Project (LENS2) 100-member ensemble, shows that with 50 members, we can resolve effect sizes of about 0.3, and with ensembles of 20 members, we can still resolve effect sizes of ~ 0.35. We further provide a robust methodology to objectively determine the required ensemble size, depending on the purpose and requirement of the replicability test.