Climate model uncertainty has changed little over the past few decades despite advances in model complexity and resolution, extensive observational datasets, and considerable resources dedicated to model intercomparison and evaluation projects. In this presentation we review how perturbed parameter ensembles (PPEs) are helping to address this long-term uncertainty challenge. There have been around a hundred PPE studies across climate, weather, atmospheric chemistry, clouds and aerosols using process-based high-resolution models through to global-scale models. Over the last few years, the number of PPE studies has grown rapidly, as has the range of challenges that they are being applied to. Building on the successful applications of PPEs that have emerged over the last few years, we define several research priorities that would accelerate our understanding of model uncertainty and ultimately help to reduce it. Opportunities include:
- Providing robust uncertainty estimates and more fully characterizing the plausible spread in climate projections, which is vital to better communicate current knowledge to downstream science, impacts and decisions.
- Defining model development priorities by efficiently identifying model structural model deficiencies.
- Diagnosing the causes of inter-model spread within MIPs and to enable statistically rigorous multi-model constraint,
- Identifying new observations and new ways of using existing observations to provide tighter constraints on model uncertainty.
- Contributing to the development of parameterizations by disentangling complex processes and sensitivities across a hierarchy of models.
We also highlight how fully exploiting the potential of PPEs requires closer collaboration of the modelling and observational communities to address the particular challenges of using observations in model uncertainty quantification and constraint.
How to cite: Carslaw, K. and the co-authors: The importance and future development of perturbed parameter ensembles in climate science, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-6525, https://doi.org/10.5194/egusphere-egu26-6525, 2026.
Recently, satellite observations were successfully used to constrain and quantify uncertainty in the NASA GISS ModelE Earth system model; the development and successful application of this machine-learning accelerated Bayesian parameter estimation approach has been mirrored by similar developments at Earth system modeling centers worldwide. The result of this enterprise yields what we label a "CPE": a Calibrated Physics Ensemble. It is differentiated from Perturbed Parameter Ensembles in that it contains only observationally plausible parameter and model configurations. We comment on the previous CPE, as well as the next generation of ModelE CPEs, being prepared for CMIP7. We also present a critical application of the CPE methodology towards quantifying observational information content in a method analogous to an Observing System Simulation Experiment (OSSE). In contrast to traditional data-assimilation based OSSEs, our approach quantifies the uncertainties most relevant for climatic projections and impact assessments: model physics uncertainties. We demonstrate a proof of concept focusing on the value of reduced uncertainty in PBL water vapor retrievals, toward supporting design for a future NASA Planetary Boundary Layer satellite mission currently in incubation.
How to cite: van Lier-Walqui, M., Elsaesser, G., Loftus, K., Hu, A., Fridlind, A., Cesana, G., Tselioudis, G., and Schmidt, G.: Calibrated Physics Ensembles and Climate OSSEs: Tools for Constraining Uncertainty and Quantifying the Societal Value of New Observations, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-15585, https://doi.org/10.5194/egusphere-egu26-15585, 2026.
Aerosol–cloud and aerosol–radiation interactions remain among the dominant sources of uncertainty in estimates of effective radiative forcing (ERF). Perturbed parameter ensembles (PPEs) are now increasingly used to evaluate climate model forcings and to diagnose sources of uncertainty. PPEs systematically sample uncertainty by performing large sets of simulations in which key model parameters are perturbed, allowing the sensitivity of model outcomes to individual processes to be quantified. When combined with Gaussian process emulators, PPE outputs can be efficiently extended to millions of model surrogates, enabling robust statistical assessments of model uncertainty. Here, we focus on aerosol-related sources of uncertainty in ERF.
This work applies a PPE–emulator framework in a two-model, one-to-one configuration to study both parametric and structural uncertainties in two Earth system models: OpenIFS/AC cycle48r1 (EC-Earth4) and ECHAM6.3-HAM2.3. Parameters are selected based on aerosol ERF uncertainty analyses in ECHAM6-HAM (Bhatti et al., 2026), with corresponding perturbations applied in OpenIFS/AC using identical parameter ranges.
Both model ensembles are evaluated against satellite observations from MODIS/Terra and POLDER-3/PARASOL for the year 2010, focusing on annual mean aerosol optical depth, single-scattering albedo, and Ångström exponent as key observables linking aerosol microphysics to ERF.
In addition to the two-model comparison, we perform a detailed evaluation of the OpenIFS/AC PPE in its own right. This includes an assessment of regional patterns in aerosol properties and ERF, as well as a quantification of the relative contributions of individual parameters to model uncertainty. From the parametric uncertainty within OpenIFS/AC, we can identify model-specific sensitivities and regional responses for parameter constraining and model development.
Despite identical parameter perturbations, the two models exhibit systematic differences in their climate responses, associated with differences in aerosol life-cycle representation, cloud microphysics, and radiative coupling. Initial results indicate that sea-salt emissions contribute significantly to the largest global uncertainties in AOD at 550 nm in both models. The ERF uncertainties are driven by a more diverse set of parameters between the models, with fossil fuel, SO₂, dimethylsulfide (DMS), and biomass-burning emissions among the dominant contributors. The resulting inter-model spread can provide a quantitative measure of structural uncertainty that is not captured by single-model PPE studies. This two-model framework adds a structural dimension to previous PPE approaches by isolating structural effects under controlled parametric sampling.
Bhatti, Y. A., Watson-Parris, D., Regayre, L. A., Jia, H., Neubauer, D., Im, U., Svenhag, C., Schutgens, N., Tsikerdekis, A., Nenes, A., Irfan, M., van Diedenhoven, B., Arifi, A., Fu, G., and Hasekamp, O. P.: Uncertainty in aerosol effective radiative forcing from anthropogenic and natural aerosol parameters in ECHAM6.3-HAM2.3, Atmos. Chem. Phys., 26, 269–293, https://doi.org/10.5194/acp-26-269-2026, 2026.
How to cite: Svenhag, C., Bhatti, Y. A., Holopainen, E., Jia, H., Hasekamp, O. P., Nenes, A., and Im, U.: Structural and Parametric Contributions to Aerosol Effective Radiative Forcing Uncertainty in a Two-Model Perturbed Parameter Ensemble, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-9246, https://doi.org/10.5194/egusphere-egu26-9246, 2026.
Computer models of physical systems are often expensive to run and have large numbers of unknown parameters, with emulators trained on the model output for use as a cheap approximation of the true model. Using an emulator, we can efficiently predict the model output at unseen inputs, including a measure of uncertainty on this prediction, and search for not implausible matches to real-world observations via history matching. We usually have many high-dimensional spatial and/or temporal fields as outputs, and we consider how to efficiently emulate and calibrate such outputs.
There are many sources of uncertainty in this procedure, and in particular when calibrating we must address the critical issue of model discrepancy (the mismatch between the real world and the model that cannot be removed by better tuning the inputs). Using simulations of the land surface model JULES, and in particular considering the uncertainty in projections of the land carbon sink under climate change scenarios, we explore the impact that different assumptions regarding model discrepancy can have on inference we make about model parameters, and on the resulting uncertainty regarding future model behaviour.
We provide an efficient emulation and calibration framework that enables modellers to input their beliefs about various land surface model outputs, and thereafter explore calibrated-model world conditional on these judgements. In particular, considering the impact such choices have on the calibration of different input parameters, identifying trade-offs and potential structural errors, and how uncertainty on the calibrated inputs propagates through to uncertainty on projections of the carbon sink up to 2100.
How to cite: Salter, J., McNeall, D., Robertson, E., and Wiltshire, A.: Quantifying uncertainty in land surface model projections under varying calibration assumptions, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-14007, https://doi.org/10.5194/egusphere-egu26-14007, 2026.
A model is always a simplification of reality, in particular for systems as complex as the climate. Therefore, building models requires choices, including simplifications and assumptions. However, many such choices remain epistemically underdetermined, meaning that no scientific reason per se dictates a choice, be it between two model resolutions or a one- and two-moment cloud microphysics scheme. In other words, there is generally uncertainty around what option is best, but in setting up a model, one has to commit to one option. Moreover, there are consecutive model versions. Newer versions are generally supposed to lead to improvements in the representation of processes and/or the similarity of historical simulations and observed climate. For example, newly added modules or higher resolutions permitting convection may be considered a step change in model development. However, these changes do not always improve results. At least when combining metrics and considering match to observations, physical basis and usability, there is no clearly superior climate model. Thus, when choosing an already-built model or model version as a basis of a scientific study, this choice itself is also epistemically underdetermined.
Seeing that choices in setting up and choosing a model are epistemically underdetermined, yet need to be made, what is their effect on conclusions, and how are such choices made? To address these questions we conducted two separate analyses.
First, we address the effect of choices in model construction with a variance analysis of CMIP models, comparing variances due to model choice, model version choice, and forced climate response over a diverse range of output variables. We find that for many variables, much of the inter-generational ensemble variance origins from variance between different versions of one and the same model. Variance from historical climate change between 1979 and 2005, as represented in AMIP simulations of 29 models, is negligible in almost all 36 investigated global annual mean variables.
Second, we investigate drivers of model selection. With a bibliometric analysis of more than 7000 papers we show a strong correlation between the model used in a study and the study’s first author's institution. In other words, institutions display model attachment, with a median attachment of over 60 % to their favorite model. This shows that model selection is largely driven by context rather than by epistemic considerations.
That model choice is influenced by contextual factors but matters for study conclusions motivates further exploration of model variants, for example using perturbed parameter ensembles to explore the space of possible models. The institutional attachment shows how model space is currently sampled unequally: each institute largely samples only part of the model space. When single institutes concentrate research power, other parts of the model space remain undersampled. We give examples of ways to address these issues: the climate-scientific community could acknowledge contextual factors and study their effects, reconsider choices where pragmatically possible, and strengthen efforts to identify where results may generalise beyond the specific model or model version and the contextual factors in effect.
How to cite: Proske, U., Brunner, L., Fischer, S., Melsen, L. A., Undorf, S., and Bender, F.: Which model do you choose? Effects and drivers of climate model selection, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-3397, https://doi.org/10.5194/egusphere-egu26-3397, 2026.
Changes in aerosols since the preindustrial era have altered the top-of-the-atmosphere radiation balance by directly scattering solar radiation and indirectly interacting with clouds, known as aerosol effective radiative forcing (ERFaer). ERFaer persistently remains one of the most uncertain components in global climate model simulations, due to the imperfect representations of aerosol and cloud properties and processes. Perturbed parameter ensembles (PPEs) are increasingly used to quantify these sources of uncertainty and to constrain models with observations.
Here, we first present a single-model PPE using the ICON-A-HAM2.3 model, designed to identify key sources of ERFaer uncertainty. This PPE comprises 383 simulations for both preindustrial and present-day conditions, in which 42 parameters related to aerosol emissions, aerosol properties and processes, cloud microphysics, convection, and turbulence are perturbed simultaneously. Gaussian process emulators are trained on model outputs to enable efficient sampling of this high-dimensional parameter space. Our analysis focuses on uncertainty quantification and attribution for aerosol and cloud properties as well as ERFaer, along with comparisons against satellite observations from SPEXone/PACE and MODIS. Our results show a global mean ERFaer of −1.10 W m⁻² (5–95 percentile: −1.54 to −0.68 W m⁻²), with the overall uncertainty dominated by aerosol-related processes, particularly aerosol emissions.
Building on this single-model framework, we further propose a Multi-Model PPE (MMPPE) initiative within the AeroCom Phase IV experiments. This multi-model approach allows us to simultaneously address structural and parametric uncertainties across models, providing a coordinated pathway toward reducing ERFaer uncertainty in current climate models. An overview of the MMPPE design and objectives will be presented.
How to cite: Jia, H., Watson-Parris, D., Neubauer, D., Bhatti, Y., Schulz, M., Regayre, L., Stier, P., Quaas, J., Partridge, D., Arifi, A., Kubin, A., Nenes, A., Im, U., Schutgens, N., van Diedenhoven, B., Ferrachat, S., Lohmann, U., Tegen, I., Henkes, A., and Hasekamp, O. and the PPE Team: Quantifying and Constraining Aerosol Forcing Uncertainty: From Single-Model to Multi-Model Perturbed Parameter Ensembles, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-15980, https://doi.org/10.5194/egusphere-egu26-15980, 2026.
Clouds play a significant role in determining surface radiation budgets, with associated uncertainties largely stemming from aerosol-cloud interactions. Improving the representation of cloud-forming aerosols in Global Climate Models (GCMs) is essential for reducing these uncertainties, particularly in climatically significant regions such as the Arctic, where aerosol populations are expected to change in the future as Arctic Amplification processes unfold.
This study utilises observations of accumulation-mode (cloud-forming) aerosol size and number terms to evaluate and constrain a GCM two-dimensionally. Accumulation-mode mean diameter and number concentration were obtained from fitting log distribution functions to size-distribution observations of five ground-based Arctic sites. Size and number terms were used simultaneously to evaluate and constrain the UK Earth System Model (UKESM), coupled with UKCA (United Kingdom Chemistry and Aerosols) and the aerosol microphysical scheme GLOMAP (Global Model of Aerosol Processes), which demonstrated a consistent seasonal bias in simulated accumulation-mode aerosols across Arctic sites. To investigate the drivers of these biases, a Generalised Additive Model was applied to assess the relative importance of parameters within a Perturbed Parameter Ensemble. Here, we present the dominant parameters controlling simulated accumulation-mode size and number terms, and their spatial and temporal variation across the Arctic. Using a dual-constraint method, we identify optimal parameter ranges that yield observationally representative size distributions at each Arctic site.
Through inter-seasonal and -spatial application of constrained parameter ranges, we identify clear inconsistencies with little to no shared parameter space between winter-spring and summer-autumn months across all Arctic sites. This work identifies areas for future model development to further constrain physical processes and natural and anthropogenic emissions in the Arctic to remedy biases, and identifies a structural error in the representation of accumulation-mode aerosols in UKESM-UKCA-GLOMAP.
How to cite: Wadlow, I., Carslaw, K., and Neely III, R.: Dual observational constraint of Arctic cloud-forming aerosols reveals structural error in UKESM , EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-20996, https://doi.org/10.5194/egusphere-egu26-20996, 2026.
Uncertainty in aerosol radiative forcing remains high, which limits confidence in climate projections. Climate models depend on uncertain input parameters, and observations are used to constrain this uncertainty by ruling out parameter values that are unlikely. However, it is often not possible to tune a model to agree with multiple observations at the same time. Even when parameter uncertainty is sampled widely, some observational constraints are mutually inconsistent; i.e., they require opposing parameter values and therefore cannot be used together. These inconsistencies in observational constraints indicate structural issues in the way that aerosol processes are modelled, and limit how far parameter uncertainty ranges (and thus forcing estimates) can be reduced.
In this presentation, we describe how targeted model developments can help address inconsistencies in observational constraints and thereby help to reduce forcing uncertainty. In previous work, we introduced a workflow to detect potential structural inconsistencies by using a perturbed parameter ensemble (PPE) of the UK Earth System Model. The workflow revealed inter‑region and inter‑variable inconsistencies. For example, sulfate aerosol concentrations in different regions could not be consistently constrained, and constraint of aerosol optical depth degraded model performance for sulfate concentrations.
Here, we extend that approach using a new PPE that includes targeted structural changes designed to address the identified deficiencies. We reapply the same observational constraints and test whether the consistency of constraints improves, and whether there are any remaining structural deficiencies. We also assess the connection to aerosol radiative forcing: whether constraints that previously led to opposing aerosol radiative forcing values now align, and whether that alignment leads to more consistent forcing (and less uncertain) values across observational constraints.
This work demonstrates a practical path to directly target two interlinked causes of model uncertainty (parametric and structural): use model-observation inconsistencies to diagnose potential structural errors, implement targeted model developments, and iterate. The outcome is an evidence‑based development cycle that aims to make more observations usable simultaneously, reduce parametric and structural uncertainty, and ultimately contribute to reducing uncertainty in climate projections.
How to cite: Prévost, L., Regayre, L., Ghosh, K., Johnson, J., Grosvenor, D., Rostron, J., Turnock, S., Schulz, M., Haugvaldstad, O., Rumbold, S., Dalvi, M., Ge, Y., McNeall, D., Milton, S., and Carslaw, K.: Targeted model developments to improve consistency of observational constraints and reduce aerosol forcing uncertainty, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-12289, https://doi.org/10.5194/egusphere-egu26-12289, 2026.
Improving aerosol and cloud descriptions in ESMs can increase the confidence in estimates of climate impacts of changing anthropogenic aerosol emissions. In the FORCeS project funded by the Horizon 2020 framework programme, we combined experimental and theoretical approaches to bridge the current key gaps in the fundamental understanding of essential aerosol and cloud processes and their descriptions in selected European ESMs. Regarding aerosols, we focused on organic aerosol, particulate nitrate, absorbing aerosols, and ultrafine aerosol sources including new particle formation and growth. In terms of cloud microphysics, we targeted cloud droplet activation, hydrometeor growth and evaporation, ice formation and multiplication as well as aerosol processing and scavenging by clouds. The selection was made combining identified knowledge gaps in scientific understanding of these processes and/or their current representation in ESMs with a perturbed parameter ensemble approach to detecting potential structural deficiencies in an ESM. In this presentation, I will summarize a recently published overview article (Riipinen et al., Tellus B, 2026) where we provide recommendations applicable in models operating at the Earth system scale. Overall, the findings highlight the need for continuous efforts towards smart ways for representing the aerosol number size distribution as well as consistent representations of key parameters (e.g. liquid water content and cloud droplet number concentration). Furthermore, we provide guidance for future ESM evaluation emphasizing, in particular, the need for exploring the consistency of key parameters, process-based (as opposed to parameter-based), and the complementarity of in-situ and remote-sensed measurements for model evaluation.
How to cite: Riipinen, I. and the FORCeS project team: Recommendations for Treating Key Aerosol and Cloud Microphysics and Chemistry in Earth System Models , EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-20804, https://doi.org/10.5194/egusphere-egu26-20804, 2026.
Internal climate variability (ICV) is an important source of climate change projections uncertainty. The estimate of ICV can serve as a useful benchmark for assessing climate model performance and the emergence of anthropogenically forced climate change. Nevertheless, estimating the magnitude of ICV is challenging, especially on a regional scale. The ICV itself is state-dependent, which further complicates the assessment under transient climate change conditions. This study aims to quantify the magnitude of ICV using different types of data, i.e., earth system model simulations and observation-based datasets. We focus here on seasonal mean temperature over Europe and confront different methodological approaches: comparison of variability inferred from pre-industrial control simulations with the spread of single-model initial-condition large ensemble, separation of uncertainty sources in CMIP6 transient simulations, and forcing attribution in observed time series. In case of the large ensemble data, we also pay attention to the temporal development of ICV magnitude under changing external forcing.
How to cite: Randriatsara, H. H.-R. H., Holtanová, E., and Mikšovský, J.: Combining different views on internal climate variability of temperature over Europe, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-5489, https://doi.org/10.5194/egusphere-egu26-5489, 2026.
The effects of aerosols (small particles suspended in the air) on the Earth’s energy balance since pre-industrial times (aerosol radiative forcing) has significantly and repeatedly dominated the uncertainty in reported estimates of global temperature change from the Intergovernmental Panel on Climate Change (IPCC). Climate models are used to simulate the global distribution of aerosols and predict the aerosol radiative forcing. However, these models are extremely computationally expensive to run and such predictions are very uncertain since the values of the models' many inputs (parameters) are unknown. Expert elicited model parameter ranges form a multi-dimensional parameter uncertainty space of the climate model to explore. It is not feasible to densely sample this space directly, but by using Perturbed Parameter Ensembles (PPEs) and statistical methodologies (emulation, uncertainty quantification, and history matching) we can rigorously explore the effects of parametric uncertainty and then look to constrain it to a set of plausible models (parameter combinations) using real-world observations.
Constraining the uncertainty in aerosol forcing is a substantial challenge as the forcing itself cannot be observed directly. Hence, we must constrain model uncertainty using other observable quantities, feeding these constraints through to forcing predictions. Previous studies have shown limited success in this endeavour due to the differences in parameter sensitivity between observable variables and forcing, and the effects of ‘equifinality’ (compensating errors). Even if parameter sensitivities are shared, this does not automatically mean that constraint will feed through from the observable to the forcing, as the connections between the inputs and these outputs (observable / forcing) may align differently in the corresponding multi-dimensional response surfaces over the parameter space.
In this work, we propose an 'alignment measure' as an approach to determine the potential of a constraint on an observable quantity to provide constraint on the forcing. This measure involves analysing response surface alignment over parameter uncertainty space for a pair of variables through comparison of the surface partial derivatives. We will introduce the measure and show the application of it for the constraint of aerosol forcing in a large PPE from the UK Earth System Model. By understanding surface alignment in the model, this measure can lead to strategic observational constraint and improved uncertainty reduction of this complex climate response.
How to cite: Johnson, J., Webb, I., Owen, J., Oakley, J., Regayre, L., Ghosh, K., Prévost, L., and Carslaw, K.: Surface alignment to improve observational constraint and reduce predicted aerosol radiative forcing uncertainty, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-11904, https://doi.org/10.5194/egusphere-egu26-11904, 2026.
Earth System Models (ESMs), integrating atmosphere, ocean, land, ice, and biosphere, are vital in climate science to study drivers of climate change; quantify uncertainties in future climate projections; and to guide policy decisions. This often entails the analysis of Perturbed Parameter Ensembles (PPEs) formed by evaluating an ESM over a carefully constructed design of input parameter combinations. However, ESMs exhibit a complex structure, possess high-dimensional input and output, including spatial-temporal fields, and have long evaluation times. Further challenges arise due to model stochasticity and the numerous sources of uncertainty inherent within the modelling process. Combined, these severely inhibit the direct analysis of ESMs and the size of PPEs which may be constructed. Bayesian statistical and uncertainty analysis methodology are employed to overcome these limitations.
In this research a PPE for the UK Met Office UKESM1 model is used to investigate natural and anthropogenic aerosol emission interactions with clouds, which yields large Effective Aerosol Radiative Forcing (ERF; the temporal change in Earth’s energy balance due to aerosols) induced uncertainty in historical climate change. ERF is unobservable, thus model-observation comparison to calibrate ESMs is essential to robustly constrain ERF uncertainty. Moreover, ERF is key to accurately predicting future climate, yet research has resulted in little uncertainty reduction in over 30-years of IPCC reports.
Bayesian history matching, an efficient procedure for model-observation comparison, is performed to resolve parametric uncertainty and obtain all parameter combinations which produce ESM output consistent with observation data. This yields a greater constraint on ERF uncertainty. An efficient global parameter search is enabled by Bayesian emulators; fast statistical approximations for ESM outputs, providing both predictions at new parameter settings, along with a corresponding statement of the uncertainty, which are built for a carefully selected set of model outputs. These are embedded within an uncertainty quantification framework which includes structural model discrepancy linking ESM and the real-world, as well as representation and observation errors. In addition, the Bayesian paradigm enables the interrogation of how prior beliefs and uncertainties propagate through history matching, including discerning and evaluating implicit prior beliefs within studies.
How to cite: Owen, J., Johnson, J., Oakley, J., Webb, I., Regayre, L., Ghosh, K., Prévost, L., and Carslaw, K.: Bayesian History Matching and Uncertainty Analysis in Atmospheric Modelling, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-10365, https://doi.org/10.5194/egusphere-egu26-10365, 2026.
Quantifying parametric uncertainty is important for effective Earth system model development and calibration. Here, we present a 75-member perturbed parameter ensemble (PPE) with the coupled Earth system model NorESM3. As part of operational model calibration, we used the PPE to evaluate parameter sensitivities, identify bias‑reducing parameter choices and trade‑offs, and estimate parametric uncertainty within the model. The PPE was constructed by branching preindustrial (year 1850) simulations from a baseline run, perturbing parameters in the atmosphere, land, ocean, and sea ice modules within expert-informed ranges using Latin hypercube sampling. To study long-term behaviour, we continued a subset of members that met energy-balance criteria while preserving ensemble spread. We evaluated ensemble members against observational and reanalysis datasets to help distinguish biases responsive to parameter tuning from ones likely stemming from model structure.
How to cite: Fuglestvedt, H., Fjeldså, J., Sandstad, M., Gjermundsen, A., Sanderson, B., Debernard, J., Bentsen, M., Fisher, R., Seland, Ø., Olivié, D., and Schulz, M.: A perturbed parameter ensemble for the coupled Earth system model NorESM3, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-17495, https://doi.org/10.5194/egusphere-egu26-17495, 2026.
Earth system models are essential tools for projecting future climate change, yet their performance is limited by uncertainties in the parameterization. One of the most persistent biases is the double Intertropical Convergence Zone (ITCZ) problem. Here, we apply a machine-learning-based history matching approach to an atmosphere–ocean coupled model (GRIMs-NEMO) to reduce ITCZ-related biases while maintaining the global radiative balance. Radiative fluxes, precipitation, sea surface temperature (SST), and cloud fraction are selected as target variables, and a Gaussian Process emulator is used to efficiently explore the parameter space. The optimized parameter set reduces global-mean biases in outgoing shortwave and longwave radiation and alleviates the double ITCZ bias in the model. However, SST and cloud biases increase in parts of the tropical Pacific, which is interpreted as a consequence of enhanced cloud formation that reduces shortwave radiation and amplifies surface cooling. This limitation suggests that future tuning should include parameters related to ocean vertical mixing and cloud convection to better represent atmosphere-ocean interactions. This study demonstrates that ML-based history matching is an effective tool for reducing persistent biases in complex Earth system models and can contribute to improving the reliability of future climate projections.
※ This work was supported by the Korea Environment Industry & Technology Institute (KEITI) through the “Climate Change R&D Project for New Climate Regime” funded by the Korea Ministry of Environment (MOE) (2022003560001)
How to cite: Yun, S., Lee, S., and Moon, B.-K.: Machine-Learning-Based History Matching for Parameter Tuning of an Atmosphere-Ocean Coupled Model: Reducing the Double ITCZ Bias, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-17760, https://doi.org/10.5194/egusphere-egu26-17760, 2026.
Clouds, like other small-scale Earth-system processes, have to be approximated by simple functions in climate models. Such parameterizations often include uncertain constants. These parameters are estimated in a procedure called tuning where the model output is optimized with respect to observations [1]. Most models are tuned against present-day reanalyses [1]. However, recent studies [2,3] have demonstrated that certain parameter values which produce climate states in good agreement with present-day observations, are not well-suited for simulating climate states very different from present-day, such as the substantially colder Last Glacial Maximum (LGM, about 21.000 years ago). This implies equifinality, in which case better parameter values may be identifiable, or state-dependency, which should be taken into account when the model is used to extrapolate beyond the range of observations.
Here, we present a multi-state iterative Bayesian parameter estimation procedure. We use it to tune PaleoPlaSim [4], a coupled Atmosphere-Ocean General Circulation Model of intermediate complexity. It is a a paleoclimate-enhanced version of PlaSim-LSG [5]. We start by creating a Perturbed Parameter Ensemble (PPE). We vary 12 model parameters relating mainly to ocean mixing, cloud properties, and land surface properties. For each PPE member, we initialize a present-day (PD) and an LGM simulation. Across the initial PPE, we find that, globally, colder ensemble members exhibit a larger LGM-PD anomaly and higher temperature variability. This is consistent with palaeoclimate data and theoretical expectation. However, this relationship is weak and may be of opposing sign regionally, notably in the tropics. We hypothesize that this is due to different local climate feedback amplified or weakened by the perturbed parameters. This indicates that regional temperature variability is not necessarily fully coupled to global temperature, and climate sensitivity, indicated here by the LGM-PD anomaly.
To explore these degrees of freedom, we perform multiple tuning runs. We vary tuning targets, including weighted combinations of present-day observations, LGM climate reconstructions, and a temperature variability term. We test whether and how well this exploratory approach can identify state-dependent model parameters. Finally, we identify pathways to generalize our approach for complex climate model developments under computational constraints, for example by the use of machine-learning based emulators.
[1] Hourdin et al., “The Art and Science of Climate Model Tuning”, Bulletin of the American Meteorological Society, 98, 589–602, 2017.
[2] Sherriff-Tadano et al., “Southern Ocean Surface Temperatures and Cloud Biases in Climate Models Connected to the Representation of Glacial Deep Ocean Circulation”, Journal of Climate, 36, 3849–3866, 2023.
[3] Mikolajewicz et al., “Deglaciation and abrupt events in a coupled comprehensive atmosphere-ocean-ice sheet-solid earth model”, Climate of the Past Discussions, 1–46, 2024.
[4] Racky et al., “PaleoPlaSim 1.0: An Earth System Model of Intermediate Complexity for Paleoclimate Modeling and Large Ensemble Studies”, in prep., Proceedings of the 11th bwHPC Symposium 2025, 2025.
[5] Fraedrich et al., “The Planet Simulator: Towards a user friendly model”, Meteorologische Zeitschrift, 299–304, 2005.
How to cite: Racky, M. and Rehfeld, K.: Exploring Climate Feedbacks and Variability in an Objective Tuning of a Climate Model of Intermediate Complexity, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-19715, https://doi.org/10.5194/egusphere-egu26-19715, 2026.
The approximate representation of subgrid-scale processes in atmospheric General Circulation Models through parameterizations - such as convection and cloud microphysics - introduces significant parametric uncertainty. As a consequence, model tuning remains a crucial step in model development and in recent years the tuning procedure has evolved from an exclusively manual, expert-guided task, into a more rigorous scientific phase essential for reducing systematic biases and for constraining the global energy balance. We introduce ECtuner, a semi-automatic optimization software tool in python for the tuning of GCMs, developed for the tuning of the EC-Earth4 GCM. ECtuner uses global optimization algorithms to minimize a cost function based on the weighted distance between simulated fields and multiple observational targets. The tool identifies an optimal parameter set that best aligns the model with the present-day climate by computing the sensitivity of radiative fluxes to various atmospheric parameters from a set of perturbed simulations where one parameter is changed at a time. ECtuner offers flexibility, including a choice of global minimization algorithm, the introduction of a penalty for distance from default parameters and a choice of different tuning targets (such as TOA or surface fluxes), which can be weighted by season or latitudinal band. We present some results from the application of the tool to the tuning of the EC-Earth4 model, demonstrating how a significant reduction in the global TOA net imbalance can be achieved in AMIP simulations with a small change in essential tuning parameters.
How to cite: Albanese, M., Fabiano, F., and von Hardenberg, J.: ECtuner: a semi-automatic GCM tuning tool and its application to the EC-Earth4 model, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-12785, https://doi.org/10.5194/egusphere-egu26-12785, 2026.
Accurate predictions of future climate change are vital to limit the harmful impacts of global warming on society and ecosystems, and to inform effective policymaking. Nevertheless, the inherent limitations of Earth System Models (ESMs), even when employing multi-model ensembles, continue to engender considerable uncertainties in future climate projections. The emergent constraint (EC) approach has the potential to assist in reducing these uncertainties by establishing a linkage between models and observational data of the current climate. However, the EC method remains imperfect, and predicting temperature continues to present significant challenges.
Recent studies have demonstrated that transient Emergent Constraints (EC), particularly the Kriging for Climate Change (KCC), offer better predictive skill compared to traditional trend-based EC methods. However, existing KCC applications have largely been restricted to either Global Average Temperature (GAT) or simple joint GAT-local temperature predictions, often overlooking the complex spatial correlations inherent in climate data. The specific impact of spatial structure on future climate projections remains unexplored. To bridge this gap, this project introduces an innovative spatiotemporal EC approach.
Building on this, we introduced the localized KCC (EC) method to minimize prediction uncertainty by leveraging regional observational data. Specifically, to mitigate biases arising from non-warming factors, we implemented a joint framework that integrates GAT with region-scale adjustments for future temperature projections.
The validity of our approach was verified using an imperfect model test. We demonstrated that no matter which model is used as pseudo-observations, the bias in the posterior estimates is reduced in most regions compared to the prior. Overall, the global uncertainty is reduced by about 18% which is better than only using local temperature information. This enhanced robust method ultimately results in more reliable regional projections.
After validating the robustness of our method, we use HadCRUT5 observational data to primarily analyze and predict global temperature changes for a 20-year lead time (2040–2060) and a 50-year lead time (2070–2090).
How to cite: Li, Y., Okazaki, A., Nitta, T., Cauquoin, A., and Yoshimura, K.: Constraining Future Temperature Projections using a Localized Spatiotemporal Emergent Constraint Approach, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-1168, https://doi.org/10.5194/egusphere-egu26-1168, 2026.
Climate projections are widely used to inform about potential future climate. The current cutting edge in decadal-scale projections is at kilometre-scale global atmosphere-ocean models, which resolve finer parts of motion spectra as compared to the previous generation models. This imposes new challenges for evaluation of climate simulations. The question is which reference materials are adequate to evaluate the rich process-level variability present in the new-generation models. Common reanalyses, for instance, do not provide proper quidance in this respect. Here, we advocate the use of raw Earth observations for this purpose.
The architecture in the DestinE Climate DT (https://destination-earth.eu/) in highly synergetic with numerical weather prediction (NWP). Specifically, the so-called streaming approach enables run-time access for observation models to consume the near-native state vector and compute observation-space quantities, such as brightness temperature. Thereby, Climate DT opens the pathway for direct observation-space evaluation using, in principle, any set of Earth observations and potentially resolves the open question about adequante process-level reference materials. In Climate DT, these are used for online monitoring of ongoing simulations and their posterior evaluation. The synergy aspect here is the extensive sharing of observation modelling infrastructure with data assimilation in NWP, foremost with ECMWF.
We showcase early examples of statistical Earth observation-based evaluation of kilometre-scale climate simulations produced in Destination Earth Climate DT. All three kilometre-scale models contained in Climate DT simulate the mean climate well. At process-level, however, significant systematic errors appear. For instance, the diurnal range of 2-metre temperature and 10-metre wind speed variability are not well simulated. The examples demonstrate how direct use of Earth observations help to evaluate simulation results, especially at the process level. Such finding can accelerate model development, especially regarding the physical process parametrizations.
Finally, observation data base files, augmented with model counterparts, are generated online for all Climate DT simulations and they are accessible via the DestinE data lake. This presentation has a companion in ESSI1.8 about the technical implementation aspects.
How to cite: Tuppi, L., Bouvier, C., Räisänen, J., and Järvinen, H.: Observation-based evaluation of DestinE Climate DT simulations, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-4700, https://doi.org/10.5194/egusphere-egu26-4700, 2026.
Posters virtual: Fri, 8 May, 14:00–18:00 | vPoster spot 4
EGU26-19525 | ECS | Posters virtual | VPS7
Optimizing Aerosol Emissions over Europe using Surface Black Carbon MeasurementsFri, 08 May, 15:03–15:06 (CEST) vPoster spot 4
One approach to reduce the uncertaintites in the black carbon (BC) emissions estimated using the bottom-up inventories is by integrating the atmospheric models with observational data. In this study, we estimate aerosol emissions over Europe (15◦W–35◦E, 33–73 ◦N) by assimilating surface observations of BC from EBAS network using Local Ensemble Transform Kalman Filter (LETKF) in the LOTOS-EUROS chemical transport model. Sensitivity experiments indicate that an ensemble size of 24 and a localization distance of 300 km provide optimal performance. Furthermore, we assess the influence of CAMS BC boundary conditions on the emission estimates and find that these boundary conditions tend to overestimate BC concentrations near the domain boundaries.
Our results show that the bottom-up approach generally overestimates BC emissions across Europe. Quantitatively, the posterior emissions are found to be 21% and 30% lower than the prior emissions for the years 2011 and 2021, respectively. A reduction in both emissions and associated uncertainties is observed over central Europe, where the observations are dense. Seasonal analysis reveals that emission decreases are most pronounced over the central domain during autumn and winter. Finally, the validation of optimized BC concentrations with independent observations showed a decrease in bias and RMSE, however the correlation remains poor compared to the background concentrations.
How to cite: George, B., Thomasson, A., Roldin, P., Segers, A., and Schutgens, N.: Optimizing Aerosol Emissions over Europe using Surface Black Carbon Measurements, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-19525, https://doi.org/10.5194/egusphere-egu26-19525, 2026.
