Abstract Machine learning (ML)‐based models have demonstrated high skill and computational efficiency, often outperforming conventional physics‐based models in weather and subseasonal predictions. While prior studies have assessed their fidelity in capturing synoptic‐scale atmospheric dynamics, their performance across timescales and under out‐of‐distribution forcing, such as +3K or +4K uniform‐warming forcings, and the sources of biases remain elusive, to establish the model's reliability for Earth science. Here, we design three sets of experiments targeting synoptic‐scale phenomena, interannual variability, and out‐of‐distribution uniform‐warming forcings. We evaluate the Neural General Circulation Model (NeuralGCM), a hybrid model integrating a dynamical core with ML‐based component, against observations and physics‐based Earth system models (ESMs). At the synoptic scale, NeuralGCM captures the evolution and propagation of extratropical cyclones with performance comparable to ESMs. At the interannual scale, when forced by El Niño‐Southern Oscillation sea surface temperature (SST) anomalies, NeuralGCM successfully reproduces associated teleconnection patterns but exhibits deficiencies in capturing nonlinear response. Under out‐of‐distribution uniform‐warming forcings, NeuralGCM simulates similar responses in global‐average temperature and precipitation and reproduces large‐scale tropospheric circulation features similar to those in ESMs. Notable weaknesses include overestimating the tracks and spatial extent of extratropical cyclones, biases in the teleconnected wave train triggered by tropical SST anomalies, and differences in upper‐level warming and stratospheric circulation responses to SST warming compared to physics‐based ESMs. The causes of these weaknesses were explored. Despite the noted weaknesses, NeuralGCM reproduces responses across experiments reasonably and performs comparably to ESMs. By integrating a dynamical core with ML, NeuralGCM shows potential for developing ML‐based ESMs.

Arctic moisture intrusions (MIs), narrow filaments of strong moisture transport, are key drivers of poleward moisture flux and Arctic weather extremes, yet their representation in climate models is poorly understood. Using a new Arctic MI detection algorithm, we document persistent biases across three CMIP generations (CMIP3–CMIP6): models overestimate MI occurrence over the Pacific sector and underestimate it over the Atlantic sector. These errors stem from misrepresented midlatitude westerly jets, with an equatorward North Atlantic jet associated with too few Atlantic MIs, and a poleward, weakened North Pacific jet linked to too many Pacific MIs. Experiments that correct sea surface temperature and sea ice concentration biases and increase atmospheric resolution improve jet structure and MI statistics, while a cloud-locking simulation indicates that better high-frequency cloud–radiation–circulation interactions can yield further gains. Our results clarify pathways to reducing long-standing MI and jet biases, providing guidance for improving simulations of Arctic and midlatitude climate.

Land surface relative humidity (RH) is a key variable in the coupled land-atmosphere system that profoundly influences terrestrial hydroclimate and ecosystems. Yet historical changes in land RH are not well understood due to limited observations, biased reanalyses, and the lack of a framework for interpreting RH changes under multiple influencing factors. Here, we show that the spatiotemporal variability of land RH and its distinct historical trends among observations, reanalyses, and Earth system models are captured by a simple index based on the ratio of precipitation (P) to a modified potential evapotranspiration formulated independently of RH ([Formula: see text]). The index provides a physical calibration of biased land RH in reanalyses and a quantitative framework for interpreting land RH changes. Over 1973-2024, land RH has decreased substantially, owing to the intrinsic rise in [Formula: see text] with temperature and little increase in land precipitation. Reanalyses overestimate the observed RH decrease, consistent with exaggerated surface warming and precipitation decline. The index captures this coherent bias and enables a calibration using observed precipitation and temperature. Models simulate a wide range of land RH trends, but nearly all runs underrepresent the historical drying. The index captures the model spread and discrepancy and attributes them to contributions of precipitation and [Formula: see text]. Weaker land RH decreases in models arise mainly from weaker subtropical precipitation declines, linked to muted intensification of subtropical highs and biased subtropical climatology. The model-observation discrepancy is unlikely explained by internal variability, implying model underestimation of forced RH decrease and a drier land future than current projections.

Abstract Simulations of the slab ocean configuration of the coupled Energy Exascale Earth System Model (E3SM) were used to isolate the role of poleward ocean heat transport (OHT) in shaping the climate and its response to CO2 forcing. Imposed changes to mean‐state OHT produce compensating changes in atmospheric heat transport (AHT) that are mediated by changes in surface evaporation. A reduction of maximum OHT by 0.56 PW (32%) reduces the global mean surface air temperature by 3.6°C. However, this cooler mean state exhibits 1.2°C more warming under CO2 quadrupling, with the largest differences occurring at high latitudes. The amplified warming arises from stronger surface albedo and lapse rate feedbacks in polar regions and a shortwave cloud feedback in the southern midlatitudes. These results highlight the critical role of mean‐state OHT in modulating mean‐state climate, the partitioning between the OHT and AHT, and climate sensitivity.

Coarse resolution, imperfect parameterizations, as well as uncertain initial states and forcings limit Earth System Model (ESM) predictions. Traditional bias correction via data assimilation improves constrained simulations but offers limited benefit once models run freely. We introduce an operator-learning framework that maps instantaneous model states to bias correction tendencies and applies them online during integration. The present study primarily targets developing architectures that can extract useful correction signals from limited training data while remaining stable under online integration. Building on a UNet backbone, we develop two operator architectures, Inception UNet (IUNet) and Multi-scale & Multi-branch (M&M) network, that combine diverse upsampling and receptive fields to capture multiscale nonlinear features under runtime constraints of the atmospheric component of the Energy Exascale Earth System Model version 2 (EAMv2). Trained on two years of EAMv2 simulations nudged toward ERA5 reanalysis, the operators generalize across height levels and seasons. Both architectures outperform standard UNet baselines in offline tests, indicating that functional richness rather than parameter count drives performance. In online hybrid EAMv2 runs, M&M delivers the most consistent bias reductions across variables and vertical levels. The ML-augmented configurations remain stable and computationally feasible in multi-year simulations, providing a practical pathway for scalable hybrid modeling. Our framework emphasizes long-term stability, portability, and cadence-limited updates, demonstrating the utility of expressive ML operators for learning structured, cross-scale relationships for online bias corrections in ESMs.

Nudging techniques are commonly employed to constrain atmospheric simulations toward observed states, facilitating model evaluation and sensitivity studies. However, if applied improperly - particularly to thermodynamic variables such as temperature and humidity - nudging can distort physical processes and introduce spurious biases, undermining the credibility of the simulations. This study presents an improved nudging implementation that applies vertically modulated tendencies to reduce adverse impacts on model physics. The framework is tested in version 2 of the Energy Exascale Earth System Model (EAMv2) using a suite of hindcast simulations nudged toward ERA5 reanalysis. We systematically evaluate the individual and combined effects of nudging wind, temperature, and humidity fields on the model's ability to represent large-scale atmospheric states and high-impact weather systems. Results show that the revised strategy - particularly when nudging temperature and humidity at selected levels - enhances hindcast skill by improving agreement with ERA5 without degrading the hydrological cycle or precipitation processes. Additional improvements in surface temperature, outgoing longwave radiation, and precipitation biases are achieved through targeted nudging of land-surface variables. The proposed approach strengthens the representation of large-scale conditions relevant to tropical cyclones, atmospheric rivers, and extratropical cyclones in the low-resolution EAMv2. These findings demonstrate that carefully designed thermodynamic nudging, especially of temperature and humidity, improves the realism of constrained simulations and broadens the utility of nudged EAMv2 for atmospheric modeling, machine learning, and high-impact weather research.

The U.S. Department of Energy's Energy Exascale Earth System Model (E3SM) version 2.1 builds on E3SMv2 with several changes, with the most notable being the addition of the Fox-Kemper et al. (2011) mixed-layer eddy parameterization. This parameterization captures the effect of finite-amplitude, mixed-layer eddies as an overturning streamfunction and has the primary function of restratification. Herein, we outline the changes to the mean climate state of E3SM that were introduced by the addition of this parameterization. Overall, the presence of the submesoscale parameterization improves the fidelity of the v2.1 simulation by reducing the ocean surface biases in the North Atlantic present in v2, as illustrated by changes in the climatological sea surface temperature and salinity and the Arctic sea-ice extent. Other impacts include a slight shoaling of the mixed-layer depths in the North Atlantic and a small improvement in the Atlantic Meridional Overturning Circulation (AMOC). We note that the expected shoaling due to the parameterization is regionally dependent in our coupled configuration. In addition, we investigate why the parameterization and its impacts on mixed-layer depth have little impact on the simulated AMOC: despite increased dense-water formation in the Norwegian Sea, only a small fraction of the water formed makes its way south into the North Atlantic basin. Version 2.1 also exhibits small improvements in the atmospheric climatology, with smaller biases in many notable quantities and modes of variability.

Modern climate projections lack adequate spatial and temporal resolution due to computational constraints, leading to inaccuracies in representing critical processes like thunderstorms that occur on the sub-resolution scale. Hybrid methods combining physics with machine learning (ML) offer faster, higher fidelity climate simulations by outsourcing compute-hungry, high-resolution simulations to ML emulators. However, these hybrid physics-ML simulations require domain-specific data and workflows that have been inaccessible to many ML experts. This paper is an extended version of our NeurIPS award-winning ClimSim dataset paper (Yu et al., 2024). The ClimSim dataset includes 5.7 billion pairs of multivariate input/output vectors spanning ten years at high temporal resolution, capturing the influence of high-resolution, high-fidelity physics on a host climate simulator's macro-scale state. In this extended version, we introduce a significant new contribution in Section 5, which provides a cross-platform, containerized pipeline to integrate ML models into operational climate simulators for hybrid testing. We also implement various baselines of ML models and hybrid simulators to highlight the ML challenges of building stable, skillful emulators.

Emissions‐driven (prognostic CO 2 ) simulations are essential for representing two‐way carbon‐climate feedback in Earth System Models. We present an emissions‐driven land–atmosphere coupled biogeochemistry (BGC) configuration (BGCLNDATM_progCO2) in version 2.1 of the Energy Exascale Earth System Model (E3SMv2.1). This is the first E3SM configuration that performs land‐atmosphere emission‐hindcasts. Here, we document its implementation, evaluate the model's performance against observations and other models, and propose a structured evaluation protocol for such emissions‐driven simulations. We conducted transient historical simulations (1850–2014) with BGCLNDATM_progCO2 and compare them to reference simulations—a land‐atmosphere coupled simulation without BGC and a standalone land simulation with BGC, both using prescribed CO 2 concentrations—and to observations. BGCLNDATM_progCO2 overestimates atmospheric CO 2 concentrations by 11–23 ppm yet stays within the 40‐ppm spread CMIP6 emission‐driven models and retains physical climate properties comparable to the reference runs. The CO 2 biases are partly attributed to underrepresented oceanic CO 2 uptake and inadequate representations of some terrestrial processes. In general, introducing prognostic CO 2 did not change physical climate metrics at the global scale but had larger regional effects, particularly over land where spatially heterogeneous CO 2 and prognostic leaf area index influenced surface energy balance. Finally, we propose a general evaluation protocol including spin‐up assessment, atmospheric CO 2 benchmarking, physical climate evaluation, and land biogeochemical analysis to support scientific rigor and facilitate inter‐model comparisons. The new configuration lays the groundwork for future enhancements, including improved terrestrial biogeochemical processes, integrated marine biogeochemistry, and additional human–Earth system interactions. These developments advance E3SM toward fully coupled emissions‐driven simulations, enabling more accurate carbon–climate feedback projections and informing mitigation policy by providing physically consistent carbon‐budget metrics for mitigation scenarios.

Boreal summer intraseasonal oscillation (BSISO) is a key component of tropical climate variability, characterized by eastward and northeastward propagation of organized convection across the Indo-Pacific region. BSISO influences weather and climate extremes through atmospheric teleconnections, but its future changes and related extremes remain unclear. Here, based on Coupled Model Intercomparison Project Phase 6, we find that under a high-emission scenario, BSISO convection will shift eastward by ~3° (~300 km) in 2065–2099, driven by moisture profile changes associated with sea surface temperature warming over central-to-eastern Pacific. This eastward-shifted BSISO convection, along with a northward expansion of the westerly jet, strengthens the BSISO teleconnections that extend into broader areas of North America. These changes will increase the BSISO-related heatwave risks by 23% and wildfire risks by 3.5 times over northern North America compared to present-day conditions. The findings underscore the amplification of BSISO’s influence on extreme risks of extreme weather, emphasizing the need for improved adaptation and mitigation strategies for future climates.

The role of cloud feedbacks in Arctic amplification (AA) of anthropogenic warming remains unclear. Traditional feedback analysis diagnoses the net cloud feedback as strongly positive in the tropics but either weak or negative in the Arctic, suggesting that AA would be amplified if cloud feedbacks were suppressed. However, in cloudlocking experiments using the slab ocean version of the Energy Exascale Earth System Model (E3SM), we find that suppressing cloud feedbacks results in a substantial decrease in AA under greenhouse gas forcing. We show that the increase in AA from cloud feedbacks arises from two main mechanisms: 1) the additional energy contributed by positive cloud feedbacks in the tropics leads to increased poleward moist atmospheric heat transport (AHT) which then amplifies Arctic warming; and 2) the additional Arctic warming is amplified by positive noncloud feedbacks in the region, together making extrapolar cloud feedbacks amplify AA. We also find that cloud changes can modify the strength of noncloud feedback, but that modification has a small effect on Arctic warming. We further examine the role of cloud feedbacks in AA using a moist energy balance model, which demonstrates that interactions of cloud feedbacks with moist AHT and other positive feedbacks dominate the influence of clouds on the pattern of surface warming. However, the contribution of cloud-induced changes in noncloud feedbacks on AA is relatively minor. These results demonstrate that traditional attributions of AA, that are based on local feedback analysis, overlook key interactions between extrapolar cloud changes, poleward AHT, and noncloud feedbacks in the Arctic.

Reliable simulation, prediction, and complete theoretical understanding of atmospheric blocking remain challenging despite its significant socio-economic impacts. Generations of climate models have notoriously underestimated blocking frequency, particularly over the Euro-Atlantic sector. Identifying factors controlling blocking frequency and dynamics is therefore essential for improving its simulation. Here, using a cloud-locking experiment, we show that cloud radiative effects (CREs) significantly increase the frequency of Euro-Atlantic blocking. CREs enhance upstream diabatic source of wave activity, both directly through longwave heating and indirectly through their feedback on latent heating, with the latter playing the dominant role. The resulting increase in the upstream diabatic source feeds into local wave activity downstream and promotes blocking formation. Qualitatively similar results are shown by multi-model experiments with radiatively inactive clouds to longwave radiation, albeit with a larger impact from mean-state changes. The results underscore the necessity of accurately representing cloud-radiation interactions in weather and climate models for improved prediction of blocking events.

The Southern Annular Mode (SAM) is the most dominant natural mode of variability in the mid-latitudes of the Southern Hemisphere (SH). However, both the sign and magnitude of the feedbacks from the diabatic processes, especially those associated with clouds, onto the SAM remain elusive. By applying the cloud locking technique to the Energy Exascale Earth System Model (E3SM) atmosphere model, this study isolates the positive feedback from the cloud radiative effect (CRE) to the SAM. Feedback analysis based on a wave activity-zonal momentum interaction framework corroborates this weak but positive feedback. While the magnitude of the CRE feedback appears to be secondary compared to the feedbacks from the dry and other diabatic processes, the indirect CRE effects through the interaction with other dynamical and thermodynamical processes appear to play as important a role as the direct CRE in the life cycle of the SAM. The cross-EOF analysis further reveals the obstructive effect of the interactive CRE on the propagation mode of the SH zonal wind directly through the CRE wave source and/or indirectly through modulating other diabatic processes. As a result, the propagation mode becomes more persistent and the SAM it represents becomes more predictable when the interactive CRE is disabled by cloud locking. Future efforts on inter-model comparisons of CRE-denial experiments are important to build consensus on the dynamical feedback of CRE. The annular mode is the most dominant mode of variability in the mid-latitude atmospheric circulation system. Its origin, maintenance, and feedback mechanisms have long been the focus of atmospheric dynamics research. While its dry mechanisms are well understood, the role of the diabatic processes in the midlatitude storm tracks, especially the radiative effects of the cloud fields that evolve together with the storms, have not been quantified. Even the sign of the diabatic feedback to the annular mode remains a topic of debate. By disabling the cloud radiative feedback through a cloud-locking technique that decorrelates the cloud fields with other dry and moist components in an atmospheric model, we are able to isolate the cloud radiative effect (CRE) throughout the life cycle of the Southern Annular Mode (SAM). Compared to the case with interactive CRE, the propagating SAM becomes more persistent and predictable in the cloud-locking run, although the overall diabatic feedback to the standing dipole representation of the SAM is somewhat attenuated in this run. The diabatic and CRE feedbacks to the SAM identified here epitomize the sensitivity of the leading mode of the atmospheric variability to the model representation of the diabatic processes. Cloud locking technique is used to isolate the interactive cloud radiative effect (CRE) on the Southern Annular Mode (SAM) Cloud radiative effect boosts the persistence of SAM as the leading EOF of the Southern Hemisphere zonal wind The interactive CRE feedback acts to weaken the cross-EOF interaction and counter the poleward propagation of the zonal wind

Due to the rapidly changing climate, the frequency and severity of extreme weather is expected to increase over the coming decades. As fully-resolved climate simulations remain computationally intractable, policy makers must rely on coarse-models to quantify risk for extremes. However, coarse models suffer from inherent bias due to the ignored "sub-grid" scales. We propose a framework to non-intrusively debias coarse-resolution climate predictions using neural-network (NN) correction operators. Previous efforts have attempted to train such operators using loss functions that match statistics. However, this approach falls short with events that have longer return period than that of the training data, since the reference statistics have not converged. Here, the scope is to formulate a learning method that allows for correction of dynamics and quantification of extreme events with longer return period than the training data. The key obstacle is the chaotic nature of the underlying dynamics. To overcome this challenge, we introduce a dynamical systems approach where the correction operator is trained using reference data and a coarse model simulation nudged towards that reference. The method is demonstrated on debiasing an under-resolved quasi-geostrophic model and the Energy Exascale Earth System Model (E3SM). For the former, our method enables the quantification of events that have return period two orders longer than the training data. For the latter, when trained on 8 years of ERA5 data, our approach is able to correct the coarse E3SM output to closely reflect the 36-year ERA5 statistics for all prognostic variables and significantly reduce their spatial biases.

Abstract While some previous studies examined the contribution of Eastern Pacific (EP) hurricanes toward precipitation in the arid Southwest US (SWUS), their potential to influence wildfires in that region has not been explored. Here we show, using observations and simulations from the Energy Exascale Earth System Model (E3SM), that recurving EP hurricanes modulate the wildfire environment in the SWUS by increasing precipitation and soil moisture, and reducing the vapor pressure deficit. This is especially the case during late season months of September–October when the likelihood of storms to recurve and make landfall increases. Further, analysis of burnt area observations reveals that for the months of September–October, recurving EP hurricanes may significantly reduce the prevalence of wildfires in the SWUS. Finally, E3SM simulations indicate that late season EP hurricanes have been on the decline, with important implications for wildfires in the SWUS.

The interaction between clouds and radiation is a key process within the climate system, and assessing the impacts of that interaction provides valuable insights into both the present-day climate and future projections. Many modeling experiments have been designed over the years to probe the impact of the cloud radiative effect (CRE) on the climate, including those that seek to disrupt the mean CRE effect and those that only disrupt the covariance of the CRE with the circulation. Seven such experimental designs have been added to the Energy Exascale Earth System Model version 1 (E3SMv1) of the US Department of Energy. These experiments include both the first and second iterations of the Clouds On/Off Klimate Intercomparison Experiment (COOKIE) experimental design, as well as the cloud-locking method. This paper documents the code changes necessary to implement such experiments and also provides detailed instructions for how to run them. Analyses across experiment types provide valuable insights and confirm the findings of prior studies, including the role of cloud radiative heating toward intensifying the monsoon, intensifying rain rates, and poleward expansion of the general circulation owing to cloud feedbacks.

AbstractLarge‐scale dynamical and thermodynamical processes are common environmental drivers of high‐impact weather systems causing extreme weather events. However, such large‐scale environmental conditions often display systematic biases in climate simulations, posing challenges to evaluating high‐impact weather systems and extreme weather events. In this paper, a machine learning (ML) approach was employed to bias correct the large‐scale wind, temperature, and humidity simulated by the atmospheric component of the Energy Exascale Earth System Model (E3SM) at ∼1° resolution. The usefulness of the ML approach for extreme weather analysis was demonstrated with a focus on three high‐impact weather systems, including tropical cyclones (TCs), extratropical cyclones (ETCs), and atmospheric rivers (ARs). We show that the ML model can effectively reduce climate bias in large‐scale wind, temperature, and humidity while preserving their responses to imposed climate change perturbations. The bias correction is found to directly improve water vapor transport associated with ARs, and representations of thermodynamical flows associated with ETCs. When the bias‐corrected large‐scale winds are used to drive a synthetic TC track forecast model over the Atlantic basin, the resulting TC track density agrees better with that of the TC track model driven by observed winds. In addition, the ML model insignificantly interferes with the mean climate change signals of large‐scale storm environments as well as the occurrence and intensity of three weather systems. This study suggests that the proposed ML approach can be used to improve the downscaling of extreme weather events by providing more realistic large‐scale storm environments simulated by low‐resolution climate models.