Markus G. Donat’s research while affiliated with Barcelona Supercomputing Center and other places

Initialized climate predictions have shown success in predicting interannual to decadal climate variations in some regions. However, the initialized predictions also suffer from different issues arising from imperfect initializations and inconsistencies between the model and the real world climate and processes. In particular, a so-called signal-to-noise paradox has been identified in recent years. The paradox implies that models can predict observations better than they predict themselves despite some physical inconsistencies between modeled and real world climate. This is often interpreted as an indicator of model deficiencies. Here we present results of a perfect-model decadal prediction experiment, where the predictions have been initialized using climate states from the model's own transient simulation. This experiment avoids issues related to model inconsistencies, initialization shock and the climate drift that affect real-world initialized climate predictions. We find that the perfect-model decadal predictions are highly skillful in predicting the near-surface air temperature and sea level pressure of the reference run on decadal timescales. Interestingly, we also find signal-to-noise issues– meaning that the perfect-model reference run is predicted with higher skill than any of the initialized prediction members. This counterintuitive result suggests that the signal-to-noise paradox may not be due just to model deficiencies in representing the observed climate in initialized predictions. We illustrate that this signal-to-noise problem, on multi-annual to decadal timescales, is related to analysis practices that concatenate time series from different discontinuous initialized simulations, which introduces inconsistencies compared to the continuous transient climate realizations and the observations. In particular, the concatenation of predictions initialized independently into a single time series breaks its auto-correlation.

Analysis of overshoot scenarios, characterized by a peak in radiative forcing levels followed by a decline, show that changes during the CO2 increasing phase are not necessarily compensated during the CO2 decreasing phase, particularly at regional level. Even if the global mean temperature may recover after the overshoot, at the regional level the situation post-overshoot may differ from the situation pre-overshoot, with spatial patterns characterized by large-scale temperature asymmetries. These asymmetries, found between Northern (NH) and Southern Hemisphere (SH), between high and mid-latitudes of the NH, and between western and eastern areas of the Southern Ocean, alter atmospheric dynamics and through it the hydroclimate. Changes in the sea ice, changes in the ocean circulation and heat transport, and thermal inertia of the ocean have been identified as potential sources of hysteresis, highlighting the impact of oceanic changes in the behavior of atmospheric variables in case of overshoot. This work analyzes SSP5-3.4OS and SSP1-1.9 overshoot experiments from the Coupled Model Intercomparison Project Phase 6 (CMIP6) to assess how well these mechanisms can explain the large-scale temperature asymmetries that characterize the difference between pre-overshoot and post-overshoot states. These analyses show that the relative contribution of each mechanism strongly depends on the model. Certain models like MRI-ESM2-0 are mainly impacted by changes in the Atlantic Meridional Overturning Circulation (AMOC), others like CNRM-ESM2-1 show a relevant impact of sea ice changes in high-latitude areas, and others like IPSL-CM6A-LR show also a relevant impact of changes in the Southern Meridional Overturning Circulation (SMOC). Inter-model differences in the contributions of the meridional overturning can be associated with different climatologies of Mixed Layer Depth (MLD) in the northern North Atlantic (NNA) and in certain areas of the Southern Ocean. Despite these differences across models, all the mechanisms contribute to shape the regional temperatures after overshoot, with the temperature asymmetries between NH and SH mainly explained by changes in the AMOC, those between high and mid-latitudes of the NH by sea ice changes, and those between western and eastern areas of the Southern Ocean by the SMOC. These results highlight the importance of model intercomparison and analysis of ocean dynamics to understand the regional impacts of an overshoot, and more generally the responses to forcing changes.

Near-Term Climate Forcers (NTCFs) play a crucial role in shaping Earth's climate, yet their effects are often overshadowed by long-lived greenhouse gases (GHGs) when addressing climate variability. This study explores the climatic impact of elevated non-methane NTCF concentrations from 1950 to 2014 using CMIP6-AerChemMIP simulations. We analyse data from four Earth System Models with interactive tropospheric chemistry and aerosol schemes, leveraging a twelve-member ensemble to ensure statistical robustness. Unlike single-species or idealised radiative forcing studies, our approach captures the combined effects of co-emitted NTCF species. Our results show that the negative radiative forcing of aerosols dominates the overall NTCF impact, offsetting the warming effects of absorbing aerosols and tropospheric ozone. Multi-model mean analyses reveal three key climate responses: (1) a global cooling, amplified in the Arctic, where autumn temperatures decrease by up to 5 °C, (2) a 38 % increase in Labrador Sea ocean convection, and (3) changes in tropical precipitation, including a 0.6° southward displacement of the Intertropical Convergence Zone (ITCZ). This research addresses the mechanisms driving these climatic changes and underscores the importance of incorporating interactive NTCFs in climate projections. As inferred from their historical impact, future NTCF reductions could amplify regional responses to increasing GHG concentrations, thus requiring more ambitious mitigation strategies.

Initialized climate predictions are routinely carried out at many global institutions that predict the climate up to next ten years. In this study we present 30 year long initialized climate predictions and hindcasts consisting of 10 ensemble members. We assess the skill of the predictions of surface air temperature on decadal and multidecadal timescales. For the 10 year average hindcasts, we find that there is limited added value from initialization beyond the first decade over a few regions. However, no added value from initialization was found for the third decade (i.e. forecast years 21–29). The ensemble spread in the initialized predictions grows larger with the forecast time, however, the initialized predictions do not necessarily converge towards the uninitialized climate projections within a few years and even decades after initialization. There is in particular a long-term weakening of the Atlantic Meridional Overturning Circulation (AMOC) after initialization that does not recover within the 30 years of the simulations, remaining substantially lower compared to the AMOC in the uninitialized historical simulations. The lower AMOC mean conditions also result in different surface temperature anomalies over northern and southern high latitude regions with cooler temperature in the northern hemisphere and warmer in the southern hemisphere in the later forecast years as compared to the first forecast year. The temperature differences are due to less transport of heat to the northern hemisphere in the later forecast years. These multi-decadal predictions therefore highlight important issues with current prediction systems, resulting in long-term drift into climate states inconsistent with the climate simulated by the historical simulations.

Hot-dry compound extremes have recently gained increasing attention due to their potential impacts on environments and societies. For these reasons, assessing climate predictions is essential to providing reliable information on such extremes. However, despite several studies focusing on compound extremes in the past and climate projections, little is known on a multi-annual timescale. At this regard, decadal climate predictions have been produced to provide useful information for this specific timescale. Thus, we evaluate the ability of the CMIP6 multi-model decadal climate hindcast in predicting hot-dry climate extremes, as well as their hot and dry univariate counterparts for the forecast years 2–5. The multi-model skillfully predicts hot-dry compound extremes and hot extremes over most land regions, while the skill is more limited for dry extremes. However, we find only small and spatially limited improvements from the initialisation of the hindcasts, especially for the hot-dry compound extremes, with most of the skill coming from external forcings, especially long-term trends. Finally, we find the decadal hindcast to be able to reproduce the connections between the compound extremes and their hot and dry univariate components. Evaluations such as this of decadal hindcast are an essential tool to establish the potential and the limits of these products, a necessary step to provide reliable and useful information regarding such impactful extremes.

Most operational climate services providers base their seasonal predictions on initialised general circulation models (GCMs) or statistical techniques that fit past observations. GCMs require substantial computational resources, which limits their capacity. In contrast, statistical methods often lack robustness due to short historical records. Recent works propose machine learning methods trained on climate model output, leveraging larger sample sizes and simulated scenarios. Yet, many of these studies focus on prediction tasks that might be restricted in spatial extent or temporal coverage, opening a gap with existing operational predictions. Thus, the present study evaluates the effectiveness of a methodology that combines variational inference with transformer models to predict fields of seasonal anomalies. The predictions cover all four seasons and are initialised one month before the start of each season. The model was trained on climate model output from CMIP6 and tested using ERA5 reanalysis data. We analyse the method's performance in predicting interannual anomalies beyond the climate change-induced trend. We also test the proposed methodology in a regional context with a use case focused on Europe. While climate change trends dominate the skill of temperature predictions, the method presents additional skill over the climatological forecast in regions influenced by known teleconnections. We reach similar conclusions based on the validation of precipitation predictions. Despite underperforming SEAS5 in most tropics, our model offers added value in numerous extratropical inland regions. This work demonstrates the effectiveness of training generative models on climate model output for seasonal predictions, providing skilful predictions beyond the induced climate change trend at time scales and lead times relevant for user applications.

Previous studies have suggested that Arctic sea-ice loss can have a profound influence on atmospheric circulation far away from the Arctic. However, there is little scientific consensus on the features of these remote responses, with the opposite impacts reported. Here we present a multi-model analysis of the decadal climate response to Arctic sea-ice loss using state-of-the-art energy conserving methodologies to isolate the impacts of sea-ice decline. We observe weakening of the Aleutian Low and development of a geopotential ridge in the North Pacific, associated with drier winter conditions over the southwest United States. Over the Atlantic, a negative NAO-like response drives wetter winter conditions across the western Mediterranean. These decadal-scale impacts substantially differ from reported centennial-scale responses to Arctic sea-ice loss simulated using non-energy conserving methodologies. Factors such as the timescale of the response and methodologies used to isolate the impacts of disappearing sea-ice cover should be carefully considered when consolidating scientific understanding on the future impacts of changing Arctic.

Seamless climate predictions integrate forecasts across various timescales to provide actionable information in sectors such as agriculture, energy, and public health. While significant progress has been made, there is still a gap in the continuous provision of operational forecasts, particularly from seasonal to multi-annual time scales. We demonstrate that filling this gap is possible using an established climate model analog method to constrain variability in CMIP6 climate simulations. The analog method yields predictive skill for surface air temperature forecasts across timescales, ranging from seasons to several years, consistently outperforming the unconstrained CMIP6 ensemble. Similar to operational climate prediction systems, standardized precipitation index forecasts are less skillful than surface air temperature forecasts, but still systematically better than the CMIP6 unconstrained simulations. The analog-based seamless prediction system is competitive compared to state-of-the art initialised climate prediction systems that currently provide forecasts for specific time scales, such as seasonal and multi-annual. While the current prediction systems provide only 1–2 initialisations per year, the analog-based system can easily provide seamless predictions with monthly initialisations, delivering seamless climate information throughout the year currently not available from traditional seasonal or decadal prediction systems. Furthermore, due to analog-based predictions being computationally inexpensive, we argue that these methods are a valuable and viable complement to existing operational prediction systems.

Overshoot scenarios, in which the forcing reaches a peak before starting to decline, show non-symmetric changes during CO2-increasing and CO2-decreasing phases, producing persistent changes in climate. Irreversibility mechanisms, associated with (among other factors) lagged responses of climate components, changes in ocean circulation and heat transport, and changes in the ice cover, bring hysteresis to the climate system. This work analyzes simulations from the Coupled Model Intercomparison Project Phase 6 (CMIP6) to explore the relevance of these mechanisms in overshoot scenarios with different forcing conditions (SSP5-3.4OS and SSP1-1.9) and their impact on regional climates, with a particular focus on the degree to which changes in regional extremes are reversible. These analyses show that in scenarios with strong forcing changes like SSP5-3.4OS, the post-overshoot state is characterized by a temperature asymmetry between the Northern Hemisphere and Southern Hemisphere associated with shifts in the Intertropical Convergence Zone (ITCZ). In scenarios with lower forcing changes like SSP1-1.9, this hemispheric asymmetry is more limited, while temperature changes in polar areas are more prominent. These large-scale changes have an impact on regional climates, e.g., temperature extremes in extratropical regions and precipitation extremes in tropical regions around the ITCZ. Differences between pre- and post-overshoot states may be associated with persistent changes in the heat transport and a different thermal inertia depending on the region, leading to a different timing of the temperature maximum in different regions. Other factors like changes in aerosol emissions and ice melting may be also important, particularly for polar areas. Results show that irreversibility of temperature and precipitation extremes is mainly caused by the transitions around the global temperature maximum, when a decoupling between regional extremes and global temperature generates persistent changes at regional level.