Abstract Uncertainty in land model projections remains high and the roles of parametric and structural uncertainty are difficult to disentangle. To compare parametric sensitivity across model structures we present two parameter perturbation ensembles using the Community Land Model (CLM) operating in satellite phenology mode. The ensembles contrast two vegetation modules: (a) the default CLM vegetation module and (b) the Functionally Assembled Terrestrial Ecosystem Simulator (CLM‐FATES). We perturbed over 300 parameters and quantified their effects on biophysical fluxes globally and across biomes. Most parameters have minimal impact on biophysical fluxes, with only a few substantially influencing results. While both models exhibit similar parameter sensitivity for some fluxes, CLM‐FATES shows larger spread in gross primary productivity (GPP), driven by strong sensitivity to carboxylation rate. CLM‐FATES also shows a weaker GPP response to soil hydrology parameters and exhibits higher water use efficiency (WUE). Cross‐model comparisons reveal similar sensitivities for some parameters (e.g., leaf dimension) but divergent responses to others (e.g., stomatal intercept), highlighting underlying structural differences. Differences in WUE and sensitivity to hydrology and stomatal conductance parameters underscore how model structure fundamentally alters parametric sensitivity. The data sets generated from these ensembles can be used to identify influential parameters and guide future calibration efforts.

Atmospheric rivers (ARs) are long, narrow conduits transporting water vapor poleward. ARs are often associated with extreme precipitation and can have significant impacts on humans and the environment when they make landfall. Here we use a previously developed deep learning model, CGNet, to detect ARs in high resolution Community Earth System Model (CESM) historical and future simulations and evaluate the results using multiple reanalysis products. CGNet is a convolutional neural network trained on expert-labeled AR and tropical cyclone masks from the ClimateNet project. We probe the sensitivities of CGNet by testing how different characteristics of the input (variables, normalization values, spatial and temporal resolution) impact the resulting AR detection. Global and regional AR frequency magnitudes are strongly dependent on source data. To understand future changes, we compare features of the detected ARs across CESM simulations using historical and Representative Concentration Pathway (RCP) 8.5 forcing. Focusing on western North America, we find that ARs largely increase over California and decrease over the Pacific Northwest and western Canada in response to projected warming. We leverage built-in spatiotemporal tracking with CGNet to connect regional changes in ARs with shifts in lifecycle characteristics. We further investigate precipitation associated with ARs and find that AR precipitation increases in California are largely driven by changes in AR precipitation intensity, whereas AR frequency provides an important contribution to AR precipitation changes at more northerly latitudes. AR-associated extreme precipitation is projected to increase over much of the region, with important implications for impacts.

Identifying the causes of Earth's extremes is challenging because counterfactual experiments are not possible in the observed world, while numerical experiments are computationally expensive and subject to biases. Data-driven causal discovery offers a complementary path, but existing approaches can fail in undersampled, high-dimensional regimes, and may not recover multi-timestep, multivariate pathways leading to particular events. We introduce Tracer of Causal Evolutions in Space and Time (TraCE-ST), a probabilistic Lagrangian approach that produces event-conditioned causal trajectories in multivariate gridded data. In synthetic experiments and real-world extreme events, TraCE-ST recovers known causal drivers and estimates their relative contributions, while also highlighting less-studied drivers, including orography-driven vorticity for Tropical Storm Debby (2006) and anomalous ocean-surface fluxes for the 2021 Pacific Northwest heatwave. Here, we propose causal tracking as an efficient data-driven framework for synthesizing causal evidence and generating testable hypotheses, complementing association analyses and numerical modeling while accelerating the study of high-impact events.

Climate models are increasingly used to inform water availability projections at regional scales. However, the models’ own runoff sensitivities—the change in runoff per unit change of precipitation or temperature—are often biased, which can degrade their projections of runoff change. Specifically, models tend to underestimate the runoff decline in response to a temperature increase. Here, we conduct a comprehensive analysis with multiple observational datasets, two climate model generations, and large ensemble sampling of internal climate variability to assess these biases and to constrain future runoff projections across major river basins globally. For basins that can be robustly constrained by available observations, the constraint indicates stronger runoff declines than raw model projections. The constrained projections thus indicate more severe impacts of climate change on water resources than indicated by direct climate model output. Global river runoff projections constrained by observations indicate stronger declines in runoff under climate change than raw model outputs, according to a study combining multiple observational datasets, two generations of climate models, and large ensemble sampling of internal variability.

Abstract Land surface models (LSMs), such as the Community Land Model version 5 (CLM5), represent complex vegetation processes; however, systematic biases persist between modeled and observed leaf area index (LAI) because of parameter uncertainties and knowledge gaps. The high computational cost of CLM5 is a barrier to extensive sensitivity analyses and ensemble simulations at the global scale. This study addresses these limitations by training an evidential deep neural network (EDNN) emulator on a 500‐member CLM5 perturbed parameter ensemble generated via Latin hypercube sampling of 32 key plant physiological parameters. The EDNN employs cyclic temporal encoding to preserve seasonal periodicity and to predict LAI anomalies, thereby emphasizing variability. It also quantifies predictive uncertainty by jointly learning aleatoric and epistemic components in a single forward pass, yielding well‐calibrated probabilistic outputs without requiring computationally intensive ensembles. Across the contiguous United States, the EDNN reproduces CLM5‐simulated LAI with a median R2≈0.8 on held‐out members and years while requiring substantially less computation. The EDNN emulator captures seasonal cycles, interannual variability in LAI, and uncertainties (aleatoric and epistemic) in a single pass. The sensitivity analysis highlights photosynthetic capacity and the leaf carbon‐to‐nitrogen ratio as dominant controls on LAI variability, with seasonal shifts in their influence reflecting phenological dynamics. These capabilities enable comprehensive parameter‐sensitivity studies and more efficient calibration and tuning of land surface models by supporting rapid probabilistic forecasting and adaptive model refinement, thereby paving the way for scalable Earth‐system modeling, robust parameter exploration, and uncertainty‐informed projections of land‐surface processes.

Accurately predicting terrestrial ecosystem responses to climate change over long-timescales is crucial for addressing global challenges. This relies on mechanistic modeling of ecosystem processes through land surface models (LSMs). Despite their importance, LSMs face significant uncertainties due to poorly constrained parameters, especially in carbon cycle predictions. This paper reviews the progress made in using data assimilation (DA) for LSM parameter optimization, focusing on carbon-water-vegetation interactions, as well as discussing the technical challenges faced by the community. These challenges include identifying sensitive model parameters and their prior distributions, characterizing errors due to observation biases and model-data inconsistencies, developing observation operators to interface between the model and the observations, tackling spatial and temporal heterogeneity as well as dealing with large and multiple data sets, and including the spin-up and historical period in the assimilation window. We outline how machine learning (ML) can help address these issues, proposing different avenues for future work that integrate ML and DA to reduce uncertainties in LSMs. We conclude by highlighting future priorities, including the need for international collaborations, to fully leverage the wealth of available Earth observation data sets, harness ML advances, and enhance the predictive capabilities of LSMs.

The large‐scale atmospheric circulation is a key driver for regional climate extremes, yet its response to anthropogenic forcing remains uncertain. The Pacific trough (PT) regime is a persistent circulation pattern modulating temperature, precipitation, and fires over North America. We show that the observed boreal winter‐spring (December to May) PT frequency and duration have increased significantly over the past 76 years, contributing to amplified extreme anomalous heat over western and central Canada. These observed changes are not well represented in the climate simulations analyzed herein. However, our results indicate that rising greenhouse gas concentrations likely contribute to increased winter‐spring PT frequency, which is further modulated by sea surface temperatures (SSTs). While the recent La Niña‐like and negative Pacific Decadal Oscillation‐like SST trends have dampened this increase, our results suggest that if an eventual emergence of the modeled El Niño‐like response to elevated were to occur in reality, it would reinstate the increase in PT frequency, duration, and downstream amplification of regional extreme heat. However, the occurrence, timing, and magnitude of this shift remain uncertain, given the complex, interlaced role of external forcings and internal variability in modulating historical trends, as well as models' inability to reproduce them. Additionally, modeling decisions regarding future trajectories for anthropogenic emissions, including aerosols and greenhouse gases, play a critical role in projecting future changes in PT frequency. Our findings underscore the need for a better understanding and modeled representation of long‐term changes in the atmospheric circulation to inform climate adaptation and risk assessment.

Previous research has demonstrated that specific states of the climate system can lead to enhanced subseasonal predictability (i.e., state-dependent predictability). However, biases in Earth system models can affect the representation of these states and their subsequent evolution. Here, we present a machine learning framework to identify state-dependent biases in Earth system models. In particular, we investigate the utility of transfer learning with explainable neural networks to identify tropical state-dependent biases in historical simulations of the Energy Exascale Earth System Model version 2 (E3SMv2) relevant for midlatitude subseasonal predictability. Using a perfect model framework, we find transfer learning may require substantially more data than provided by present-day reanalysis datasets to update neural network weights, imparting a cautionary tale for future transfer learning approaches focused on subseasonal modes of variability.

Comprehensive land models are subject to significant parametric uncertainty, which can be hard to quantify due to the large number of parameters and high model computational costs. We constructed a large parameter perturbation ensemble (PPE) for the Community Land Model version 5.1 with biogeochemistry configuration (CLM5.1-BGC). We performed more than 2,000 simulations perturbing 211 parameters across six forcing scenarios. This provides an expansive data set, which can be used to identify the most influential parameters on a wide range of output variables globally, by biome, or by plant functional type. We found that parameter effects can exceed scenario effects and that a small number of parameters explains a large fraction of variance across our ensemble. The most important parameters can differ regionally and also based on the forcing scenario. The software infrastructure developed for this experiment has greatly reduced the human and computer time needed for CLM PPEs, which can facilitate routine investigation of parameter sensitivity and uncertainty, as well as automated calibration.

Climate modelling and analysis are facing new demands to enhance projections and climate information. Here we argue that now is the time to push the frontiers of machine learning beyond state-of-the-art approaches, not only by developing machine-learning-based Earth system models with greater fidelity, but also by providing new capabilities through emulators for extreme event projections with large ensembles, enhanced detection and attribution methods for extreme events, and advanced climate model analysis and benchmarking. Utilizing this potential requires key machine learning challenges to be addressed, in particular generalization, uncertainty quantification, explainable artificial intelligence and causality. This interdisciplinary effort requires bringing together machine learning and climate scientists, while also leveraging the private sector, to accelerate progress towards actionable climate science. Machine learning methods allow for advances in many aspects of climate research. In this Perspective, the authors give an overview of recent progress and remaining challenges to harvest the full potential of machine learning methods.

Crucial to the assessment of future water security is how the land model component of Earth System Models partition precipitation into evapotranspiration and runoff, and the sensitivity of this partitioning to climate. This sensitivity is not explicitly constrained in land models nor the model parameters important for this sensitivity identified. Here, we seek to understand parametric controls on runoff sensitivity to precipitation and temperature in a state-of-the-science land model, the Community Land Model version 5 (CLM5). Process-parameter interactions underlying these two climate sensitivities are investigated using the sophisticated variance-based sensitivity analysis. This analysis focuses on three snow-dominated basins in the Colorado River headwaters region, a prominent exemplar where land models display a wide disparity in runoff sensitivities. Runoff sensitivities are dominated by indirect or interaction effects between a few parameters of subsurface, snow, and plant processes. A focus on only one kind of parameters would therefore limit the ability to constrain the others. Surface runoff exhibits strong sensitivity to parameters of snow and subsurface processes. Constraining snow simulations would require explicit representation of the spatial variability across large elevation gradients. Subsurface runoff and soil evaporation exhibit very similar sensitivities. Model calibration against the subsurface runoff flux would therefore constrain soil evaporation. The push toward a mechanistic treatment of processes in CLM5 have dampened the sensitivity of parameters compared to earlier model versions. A focus on the sensitive parameters and processes identified here can help characterize and reduce uncertainty in water resource sensitivity to climate change.

Terrestrial processes influence the atmosphere by controlling land-to-atmosphere fluxes of energy, water, and carbon. Prior research has demonstrated that parameter uncertainty drives uncertainty in land surface fluxes. However, the influence of land process uncertainty on the climate system remains underexplored. Here, we quantify how assumptions about land processes impact climate using a perturbed parameter ensemble for 18 land parameters in the Community Earth System Model (CESM2) under preindustrial conditions. We find that an observationally-informed range of land parameters generate biogeophysical feedbacks that significantly influence the mean climate state, largely by modifying evapotranspiration. Global mean land surface temperature ranges by 2.2°C across our ensemble (σ = 0.5°C) and precipitation changes were significant and spatially variable. Our analysis demonstrates that the impacts of land parameter uncertainty on surface fluxes propagates to the entire Earth system, and provides insights into where and how land process uncertainty influences climate.

Machine learning (ML) is a revolutionary technology with demonstrable applications across multiple disciplines. Within the Earth science community, ML has been most visible for weather forecasting, producing forecasts that rival modern physics-based models. Given the importance of deepening our understanding and improving predictions of the Earth system on all time scales, efforts are now underway to develop forecasting models into Earth-system models (ESMs), capable of representing all components of the coupled Earth system (or their aggregated behavior) and their response to external changes. Modeling the Earth system is a much more difficult problem than weather forecasting, not least because the model must represent the alternate (e.g., future) coupled states of the system for which there are no historical observations. Given that the physical principles that enable predictions about the response of the Earth system are often not explicitly coded in these ML-based models, demonstrating the credibility of ML-based ESMs thus requires us to build evidence of their consistency with the physical system. To this end, this paper puts forward five recommendations to enhance comprehensive, standardized, and independent evaluation of ML-based ESMs to strengthen their credibility and promote their wider use.

Abstract Stratospheric aerosol injection (SAI) would potentially be effective in limiting global warming and preserving large‐scale temperature patterns; however, there are still gaps in understanding the impact of SAI on wildfire risk. In this study, extreme fire weather is assessed in an Earth system model experiment that deploys SAI beginning in 2035, targeting a global temperature increase of 1.5°C above pre‐industrial levels under a moderate warming scenario. After SAI deployment, increases in extreme fire weather event frequency from climate change are dampened over much of the globe, including the Mediterranean, northeast Brazil, and eastern Europe. However, SAI has little impact over the western Amazon and northern Australia and causes larger increases in extreme fire weather frequency in west central Africa relative to the moderate emissions scenario. Variations in the impacts of warming and SAI on moisture conditions on different time scales determine the spatiotemporal differences in extreme fire weather frequency changes, and are plausibly linked to changes in synoptic‐scale circulation. This study highlights that regional and spatial heterogeneities of SAI climate effects simulated in a model are amplified when assessing wildfire risk, and that these differences must be accounted for when quantifying the possible benefit of SAI.

The Arctic hydrological system is an interconnected system that is experiencing rapid change. It is comprised of permafrost, snow, glacier, frozen soils, and inland river systems. In this study, we aim to lower the barrier of using complex land models in regional applications by developing a generalizable optimization methodology and workflow for the Community Terrestrial Systems Model (CTSM), to move them toward a more Actionable Science paradigm. Further end-user engagement is required to make science such as this "fully actionable. " We applied CTSM across Alaska and the Yukon River Basin at 4-km spatial resolution. We highlighted several potentially useful high-resolution CTSM configuration changes. Additionally, we performed a multi-objective optimization using snow and river flow metrics within an adaptive surrogate-based model optimization scheme. Four representative river basins across our study domain were selected for optimization based on observed streamflow and snow water equivalent observations at 10 SNOTEL sites. Fourteen sensitive parameters were identified for optimization with half of them not directly related to hydrology or snow processes. Across fifteen out-of-sample river basins, 13 had improved flow simulations after optimization and the mean Kling-Gupta Efficiency of daily flow increased from 0.43 to 0.63 in a 30-year evaluation. In addition, we adapted the Shapley Decomposition to disentangle each parameter's contribution to streamflow performance changes, with the seven non-hydrological parameters providing a non-negligible contribution to performance gains. The snow simulation had limited improvement, likely because snow simulation is influenced more by meteorological forcing than model parameter choices.

This study focuses on assessing the representation and predictability of North American weather regimes, which are persistent large-scale atmospheric patterns, in a set of initialized subseasonal reforecasts created using the Community Earth System Model, version 2 (CESM2). The k-means clustering was used to extract four key North American (10 degrees-70 degrees N, 150 degrees-40 degrees W) weather regimes within ERA5 reanalysis, which were used to interpret CESM2 subseasonal forecast performance. Results show that CESM2 can recreate the climatology of the four main North American weather regimes with skill but exhibits biases during later lead times with overoccurrence of the West Coast high regime and underoccurrence of the Greenland high and Alaskan ridge regimes. Overall, the West Coast high and Pacific trough regimes exhibited higher predictability within CESM2, partly related to El Ni & ntilde;o. Despite biases, several reforecasts were skillful and exhibited high predictability during later lead times, which could be partly attributed to skillful representation of the atmosphere from the tropics to extratropics upstream of North America. The high predictability at the subseasonal time scale of these case-study examples was manifested as an "ensemble realignment," in which most ensemble members agreed on a prediction despite ensemble trajectory dispersion during earlier lead times. Weather regimes were also shown to project distinct temperature and precipitation anomalies across North America that largely agree with observational products. This study further demonstrates that unsupervised learning methods can be used to uncover sources and limits of subseasonal predictability, along with systematic biases present in numerical prediction systems.

Climate variability and weather phenomena can cause extremes and pose significant risk to society and ecosystems, making continued advances in our physical understanding of such events of utmost importance for regional and global security. Advances in machine learning (ML) have been leveraged for applications in climate variability and weather, empowering scientists to approach questions using big data in new ways. Growing interest across the scientific community in these areas has motivated coordination between the physical and computer science disciplines to further advance the state of the science and tackle pressing challenges. During a recently held workshop that had participants across academia, private industry, and research laboratories, it became clear that a comprehensive review of recent and emerging ML applications for climate variability and weather phenomena that can cause extremes was needed. This article aims to fulfill this need by discussing recent advances, challenges, and research priorities in the following topics: sources of predictability for modes of climate variability, feature detection, extreme weather and climate prediction and precursors, observation-model integration, downscaling and bias correction. This article provides a review for domain scientists seeking to incorporate ML into their research. It also provides a review for those with some ML experience seeking to broaden their knowledge of ML applications for climate variability and weather.

Abstract. Extreme weather events have been demonstrated to be increasing in frequency and intensity across the globe and are anticipated to increase further with projected changes in climate. Solar climate intervention strategies, specifically stratospheric aerosol injection (SAI), have the potential to minimize some of the impacts of a changing climate while more robust reductions in greenhouse gas emissions take effect. However, to date little attention has been paid to the possible responses of extreme weather and climate events under climate intervention scenarios. We present an analysis of 16 extreme surface temperature and precipitation indices, as well as associated vegetation responses, applied to the Geoengineering Large Ensemble (GLENS). GLENS is an ensemble of simulations performed with the Community Earth System Model (CESM1) wherein SAI is simulated to offset the warming produced by a high-emission scenario throughout the 21st century, maintaining surface temperatures at 2020 levels. GLENS is generally successful at maintaining global mean temperature near 2020 levels; however, it does not completely offset some of the projected warming in northern latitudes. Some regions are also projected to cool substantially in comparison to the present day, with the greatest decreases in daytime temperatures. The differential warming–cooling also translates to fewer very hot days but more very hot nights during the summer and fewer very cold days or nights compared to the current day. Extreme precipitation patterns, for the most part, are projected to reduce in intensity in areas that are wet in the current climate and increase in intensity in dry areas. We also find that the distribution of daily precipitation becomes more consistent with more days with light rain and fewer very intense events than currently occur. In many regions there is a reduction in the persistence of long dry and wet spells compared to present day. However, asymmetry in the night and day temperatures, together with changes in cloud cover and vegetative responses, could exacerbate drying in regions that are already sensitive to drought. Overall, our results suggest that while SAI may ameliorate some of the extreme weather hazards produced by global warming, it would also present some significant differences in the distribution of climate extremes compared to the present day.