Large tropical volcanic eruptions have been hypothesized to induce surface warming over Eurasia in winter. Recent studies, however, have questioned the robustness of this claim and highlighted the very large internal variability in that season, which can easily overshadow any forced signal caused by such eruptions. To address uncertainties arising from sparse winter proxy data in previous reconstructions, we here examine a recently developed “online” paleo-data assimilation dataset, LMR Seasonal, in which the analysis from one season is propagated forward as the prior for the next season, allowing information from the dense network of summer-sensitive proxies to better constrain winter climate. We examine Eurasian and Northern European temperatures first following the 20 largest tropical eruptions of the last millennium, each exceeding the magnitude of the 1991 Pinatubo event and then, separately, following the largest extra-tropical eruptions over the same period. Irrespective of latitude, we find no evidence of statistically significant wintertime post-eruption warming – or cooling – and correlation analyses show no significant relationship between volcanic stratospheric sulfur injection and subsequent winter temperature anomalies. These findings are corroborated with two additional datasets (WinTEDA and ModE-RA). In addition, analysis of individual eruptions reveals considerable spread across datasets for many events, leading us to question the validity of superposed epoch analysis, which can yield erroneous conclusions even on continental scales.
Vessel Speed Reduction (VSR) lowers ship CO2 emissions, but its aerosol consequences remain poorly constrained. Combining unmanned aerial vehicle (UAV) plume interception, single-particle mass spectrometry (SPMS), and population-balance inversion across 14 ship plumes, we reconstructed ultrafine and coarse modes and identified two marker gaps. Only 35.8% of the reconstructed particle-number population carried detectable V (29.5% at 4.9 versus 39.7% at 13.4 knots in a matched-vessel pair; 0–70.8% across a speed-blind manifold), showing operation-dependent marker coverage. Approximately 20% of analyzed particles met SPMS biomass-burning criteria, indicating potential false positives in ship-influenced samples. Plume-stratified analysis found 2.3–10.7-fold higher median sulfate-related ion signals in V/VO-bearing particles. Independent V–S co-location supports an association potentially arising from V-associated sulfur chemistry during exhaust cooling, pre-existing SO3 uptake, or terminal co-deposition; laser desorption/ionization matrix effects may contribute to the contrast. In the matched-vessel case, VSR’s CO2 benefit coincided with a carbon-rich particle shift, altered V/VO–sulfur signatures, and lower strict detectable-V coverage. This indicates a potential aerosol-composition and marker-reliability trade-off missed by bulk-mass metrics, with implications for hazard-relevant compositional indicators.
The Köppen–Geiger climate classification is widely used to represent global eco-climatic patterns, yet its application to high-altitude regions such as the Qinghai–Tibet Plateau (QTP) remains limited because tundra climates (ET) are defined primarily by temperature, overlooking moisture constraints on vegetation. Here we present a revised Köppen–Geiger framework that incorporates the aridity index (AI) as an intermediary to represent coupled hydrothermal conditions while preserving the original temperature–precipitation structure. This refinement subdivides the ET climate into three sub-classes: arid (ETw), semi-arid (ETm), and semi-humid (ETf) tundra. Validation against satellite-derived NDVI shows improved correspondence with vegetation patterns. The revised classification reveals systematic reorganization of climate zones across the QTP, highlighting the role of coupled thermal and moisture constraints in shaping eco-climatic gradients under recent climate change. More broadly, this approach provides a transferable pathway to refine climate classifications in high-altitude and cold–dry regions where temperature-based schemes alone are insufficient.
Understanding changes in global tropical cyclone (TC) intensity and their long-term trends is critical for enhancing societal resilience. Given their disproportionate impacts, shifts in strong TCs, which are defined here as Category 4 and above, are of particular concern. We examined long-term trends in the lifetime maximum intensity (LMI) of strong global TCs and found that their increasing trend is only 43% of the average across all TCs. This discrepancy points to a wind-dependent negative feedback mechanism that restricts the intensification of strong TCs. Further analysis identified enhanced interaction between strong TCs and the deep ocean subsurface as the key driver. Stronger winds induce more intensive mixing and a deeper mixed layer depth (MLD), entraining cold subsurface waters that strongly constrain TC intensification. Our findings highlight the need to improve understanding of intense TC feedback mechanisms to enhance future intensity projections, which is critical for mitigating societal and environmental impacts.
The Tibetan Plateau (TP) plays a crucial role in the Asian hydrological cycle, yet the dominant moisture sources and their governing physical processes remain debated. Here, we employ the Eulerian water tracer technique embedded in Community Earth System Model version 1 (CESM1), together with Lagrangian tracking and isotopic analysis, to quantify moisture sources of summer and winter precipitation over the TP during 2003–2021. We show that, within the Eulerian AWT source-attribution framework, terrestrial sources dominate both climatological mean and interannual variability of TP precipitation, with much stronger terrestrial control over the northern TP (NTP) than over the southern TP. Terrestrial sources account for 78.9 ± 12.1% of summer and 68.4 ± 10.0% of winter precipitation over the NTP, mainly from Western Europe (WEU), Central Asia (CA), Northern Africa (NAF), and TP evapotranspiration. This terrestrial dominance arises because most of the moisture originating from remote oceans is depleted before directly reaching the TP. Specifically, remote ocean-derived moisture (e.g., North Atlantic Ocean) is first deposited as precipitation over intermediate continental regions (WEU–CA and NAF), then re-enters the atmosphere through evapotranspiration, and is subsequently transported farther downstream to the TP. We identify this process as a terrestrial “moisture relay” mechanism. This relay mechanism also regulates interannual precipitation variability over the NTP, favoring enhanced convective precipitation during wet NTP summers and increased large-scale precipitation during wet NTP winters. Our results suggest that precipitation over the TP, especially over the NTP, is primarily controlled by the partitioning between precipitation loss and re-evaporation over upstream continental regions.
Ski resorts across the western U.S. are challenged by global warming, with fewer days of sufficient natural snow during the season and more frequent weather conditions unsuitable for artificial snowmaking early in the season. While this applies to all ski resorts, the magnitude of snow security reduction varies widely. Regional Earth system simulations of sufficient resolution to capture the weather and seasonal snowpack at ski areas in the Rocky Mountain region demonstrate that in the next few decades, high-elevation places, and those historically receiving more snowfall, will be impacted less, while those resorts that currently have relatively low snow security will suffer more. Even though resorts in the Colorado Rockies will experience more warming in the next few decades than ski areas closer to the Pacific coast, they generally are less at risk, including a smaller decrease in the ski season length and in snowmaking potential, and fewer rain-on-snow events.
Summer Arctic sea-ice melt exhibits strong interannual and regional variability, yet the cloud-mediated pathways linking atmospheric circulation to melt remain poorly quantified. Using satellite observations, reanalysis data, and sea-ice products for 2000–2024, we examine the Arctic Oscillation (AO) and two dipole-like circulation modes, DA1 and DA2. These modes organize distinct altitude- and phase-dependent cloud anomalies and coherent responses in net, shortwave, and longwave cloud radiative effects. The net response is governed primarily by shortwave cloud–surface-albedo interactions, while longwave effects provide a weaker but regionally relevant contribution. Snowfall and rainfall also vary systematically and can modify the snow–ice surface state and albedo. Across nine Arctic subregions, diagnosed net energy-equivalent responses from LWCRE, cloud-related SWCRE, and albedo-related SWCRE range from −1.73 to 7.95 cm for AO, −8.22 to 4.95 cm for DA1, and −4.69 to 8.25 cm for DA2 per standard deviation. Comparison with the corresponding regional summer sea-ice thickness responses indicates that these pathways provide a measurable but partial modulation of melt variability. These results highlight the importance of cloud-radiative, surface-albedo, and precipitation-related processes for understanding and predicting regional Arctic sea-ice change.
Tropical western Pacific upper ocean circulation is traditionally regarded as a passive ENSO response, yet its active role in ENSO preconditioning remains unclear. Using observations and model experiments during 1993–2025, we show that pre-winter intraseasonal zonal current variability (U ISV) systematically intensifies before El Niño and weakens before La Niña. Approximately 80% of ISV amplitude is explained by Indian Ocean Madden–Julian Oscillation (MJO) related wind forcing. Repeated MJO-driven U ISV generates a persistent low-frequency response through cumulative upper-ocean dynamical adjustment, enhancing Kelvin-wave-induced zonal heat advection and central-Pacific heat accumulation. U ISV intensifies approximately 8 months before ENSO onset, provides complementary long-lead upstream predictive information for ENSO preconditioning beyond monthly-mean circulation alone, and improves Niño-3.4 reconstruction (r = 0.73). Our findings identify a cross-timescale ocean dynamical bridge through which repeated MJO-driven intraseasonal ocean variability contribute to seasonal-to-interannual climate variability and marine heat waves.
Marginal warm permafrost is nearing 0 °C, but its transition across this threshold remains poorly constrained by limited long-term observations. We analyzed up to two decades of ground-temperature records from 59 boreholes on the northeastern Qinghai–Xizang Plateau, including four sites undergoing advanced degradation near or across 0 °C. Bias-corrected ERA5-Land data showed rapid regional warming (median 0.56 °C decade−1), consistent warming across the 4100–4600 m elevation range containing most boreholes, and an upward shift in mean annual air temperature beginning in 2021. Several boreholes remained near 0 °C for prolonged periods despite atmospheric warming, indicating latent-heat buffering and delayed warming at depth. As buffering weakened, ground temperatures shifted upward and warmed more rapidly. At the most rapidly degrading site, a late-season subzero layer approximately 13 m thick disappeared between 2021 and 2023 after several years of near-zero temperatures. Profile changes were consistent with talik development and limited thermal recovery, although annual late-season measurements cannot confirm year-round talik conditions. Contrasting vegetation greenness coincided with differing ground-temperature responses, suggesting that local surface conditions influenced degradation timing. These observations show that marginal permafrost can undergo rapid, nonlinear degradation near 0 °C and that continued regional warming may promote similar transitions.
Extreme windstorms cause substantial socioeconomic damage across many Northern Hemisphere regions, yet uncertainty remains whether their frequency will increase or decrease under global warming. Here, using a multi-member ensemble of HighResMIP global models (low- to high-resolution, coupled and atmosphere-only) together with regional dynamically downscaled simulations, we investigate future changes of extreme windstorms and their dynamical origins. Results suggest extreme windstorms over the northeastern USA and southeastern Canada will increase significantly, regardless of model resolution or coupling strategy, and consistently across global and regional simulations. This increase occurs despite a decline in overall mid-latitude cyclone numbers and is driven by more intense cyclones — including transitioning tropical cyclones — tracking into the region. Low-resolution simulations produce more frequent and larger windstorms than high-resolution ones, due to the lower 99th-percentile wind speeds and hence exceedance threshold and their larger eddy length scale which implies overestimation of the Rossby deformation radius. Low-resolution simulations also exhibit higher low-level baroclinicity and available potential energy (APE), yet weaker eddy kinetic energy (EKE) than simulations performed at high resolution. This indicates less efficient APE-to-EKE conversion. Our findings reveal a robust projected increase in extreme windstorm risk for northeastern North America and underscore that quantitative projections of synoptic-scale hazards require atmospheric grid spacing of ~25–50 km, finer than the ~100 km traditionally considered adequate for synoptic-scale modeling.
Eurasian autumn snow has been widely regarded as a source of East Asian winter predictability, yet its influence is strongly non-stationary. Here we show that this non-stationarity reflects a reorganization of the dynamical pathway rather than a simple loss of snow forcing. Using multiple snow products, ERA5 diagnostics, and ECMWF SEAS5 hindcasts, we identify a spatially organized October Eurasian snow mode that is more closely linked to East Asian winter temperature than basin-mean snow loading. Observations indicate that snow memory partly persists into early winter, whereas the intermediate atmospheric pathway is substantially less stable: tropospheric circulation adjustment weakens, and the snow-WAFz and WAFz-AO10 relationships are weak, seasonally and vertically dependent, and not consistently significant across periods. Despite this variability in the intermediate pathway, the October snow-East Asian winter temperature relationship remains comparatively robust across periods. The final East Asian temperature response is limited by a reorganized downward/regional projection. SEAS5 reproduces the upstream snow predictor reasonably well but fails most clearly at the WAFz-AO10 stage, where none of the ensemble members reproduces the observed sign of the relationship. This contrast identifies dynamical pathway fidelity, rather than snow representation alone, as a key constraint on seasonal predictability.
The 5th ECMWF-ESA Machine Learning Workshop (Bologna, Italy; 13-17 April 2026) convened the Earth System Observation and Prediction (ESOP) community to assess progress and new directions in the general field of the application of Machine Learning to ESOP and the specific thematic areas of Machine Learning (ML) for Digital Twins, hybrid ML-physics systems, ML for Earth observation (EO), end-to-end Data Assimilation and Prediction and high-performance computing. The 5-day workshop was held in a hybrid format at the Tecnopolo Data Manifattura (DAMA) in Bologna (IT) with an interactive online component, featuring 57 expert talks and 88 poster presentations. The workshop emphasised both the increasing scale and sophistication of ML for ESOP—including potential end-to-end workflows for future operations—and the parallel shift from CPU-centric to GPU/TPU-centric computing, with additional paradigms (quantum, edge, neuromorphic) emerging. Building on the reporting structure of the previous workshop report (ref. 1), this document summarises outcomes across the five thematic areas covering current topics and emerging trends. All the workshop presentations and recordings are available online at https://ecmwfevents.com/i/5th-ecmwf-esa-machine-learning-workshop/public/agenda .
El Niño events exert a profound influence on the global carbon cycle by imposing widespread heat and water stress on terrestrial ecosystems. Plant isoprene emissions respond rapidly to such stress, yet it remains unclear whether this response can track the spatiotemporal evolution of El Niño’s impacts over land. Here, we used satellite-derived global isoprene emissions for the first time to assess the west-east progression of the 2015–2016 El Niño. We observed that isoprene emissions increased by up to approximately 30% across tropical ecosystems relative to the climatological mean, with pronounced anomalies emerging during the event. The spatiotemporal evolution of these anomalies closely aligns with the El Niño progression inferred from sea surface temperature (SST) anomalies in the equatorial Pacific. In contrast, commonly used satellite vegetation products, including leaf area index (LAI) and solar-induced chlorophyll fluorescence (SIF), show weaker and spatially incoherent responses. These results demonstrate that satellite-derived isoprene provides a sensitive and mechanistically grounded tracer of ecosystem stress, offering a complementary perspective for monitoring the intensity and propagation of extreme climate events across terrestrial ecosystems.
The rime-splintering (RS) process, which is one of the most important secondary ice production mechanisms, plays a crucial role in regulating ice crystal number concentrations and precipitation formation in convective clouds. The initiation and efficiency of the RS process are modulated by the spatial distributions of liquid and ice particles, which are highly heterogeneous. However, numerical models are developed based on the idealized assumption of homogeneous liquid-ice mixing in a given volume, which causes large uncertainties in modeling ice production in convective clouds. Based on in-situ measurements from The Convective Precipitation Experiment (COPE), this study investigates how the heterogeneous particle distribution affects the RS process by defining an impact factor FHM. The liquid-ice mixing homogeneity (χ) in a given cloud is quantified using information-theoretic entropy, and the spread of riming rate (ΔRrim) is quantified using the maximum Rrim minus the minimum one. The results show that FHM increases with χ at a given temperature, indicating that a greater liquid-ice mixing homogeneity enhances the occurrence of RS process. Since the RS process is non-linear, in most of the observed clouds FHM exceeds 1, and increases with increasing ΔRrim, suggesting the heterogeneous particle distribution enhances the mean ice production rate compared to that assuming homogeneous particle distribution. Among the various meteorological factors that influence χ and ΔRrim, vertical wind shear (WS) and condensed water content (CWC) show the strongest yet opposing correlations with χ, while temperature (T) and WS contribute the most to ΔRrim. These insights provide a new perspective for advancing subgrid-scale parameterization of the RS process.
The Tibetan Plateau (TP) is characterized by frequent mixed-phase clouds, in which riming exerts a crucial control on precipitation formation. However, conventional bulk microphysics schemes generally ignore the regulatory impact of sub-grid liquid-ice mixing state on riming processes. This study improves the riming parameterization within the WRF Morrison scheme by adopting constant reduction coefficients and introducing a diagnostic mixing homogeneity parameter χ for snow and graupel, which varies with condensed water content and temperature. Reduced riming coefficients correct simulated precipitation distribution and spatial deviation, while χ-based dynamic regulation further optimizes temporal variation of regional precipitation, reducing RMSE by 23.8%. The modified scheme shows superior performance across all rainfall grades, particularly for heavy and torrential rain. Physically, χ inhibits excessive riming, maintains supercooled liquid droplets, and effectively alleviates the overestimation of surface precipitation. One-month simulations validate that considering liquid-ice mixing homogeneity markedly improves WRF precipitation simulation capability over the TP.
Greenland and Antarctic ice sheets are losing mass rapidly, contributing significantly to sea level rise and altering global climate patterns. Freshwater input from these ice sheets influences tropical climate through ocean-atmosphere interactions, but their distinct impacts remain unclear. Here, we use coupled climate model simulations with imposed meltwater fluxes from Greenland, Antarctica, and both combined under global warming scenarios to investigate tropical responses. We find that Greenland meltwater induces a pronounced southward shift of the Intertropical Convergence Zone (ITCZ) and strengthens the Walker circulation, while Antarctic meltwater causes weaker, opposite-signed effects. Combined meltwater forcing partially offsets these responses, with Greenland effects dominating. These asymmetric impacts arise from differential perturbations to ocean heat transport, atmospheric energy transport, and Atlantic–Pacific pressure-wind adjustments, with cloud-radiative changes acting as secondary modulators. Our results suggest that, under symmetric meltwater perturbations, Northern Hemisphere ice sheet melt may disproportionately influence tropical rainfall and circulation patterns, underscoring the need to represent freshwater forcing pathways in climate projections accurately.
Aerosol–cloud interactions (ACI) are a leading uncertainty in radiative forcing, yet aerosol particle size is poorly quantified over South Asia. Using 2.7 million satellite observations (2015–2021), we test whether the Ångström Exponent (AE) modulates cloud droplet effective radius (CER) responses to aerosol index (AI) and aerosol optical depth (AOD) over the subcontinent and adjacent ocean. Three methods target distinct estimands: AE-stratified binned regression, eXtreme Gradient Boosting (XGBoost) with SHapley Additive exPlanations (SHAP), and Double Machine Learning (DML) causal forests. Binned regressions reveal a robust land–ocean contrast. Oceanic coarse-mode regimes show negative sensitivities of − 2.6 to − 5.3 μm per AI unit, consistent with the Twomey effect. Continental fine-mode regimes show positive sensitivities of + 0.4 to + 1.3 μm per AI unit. This contrast persists across cloud water path (CWP) strata. XGBoost–SHAP corroborates it, ranking AE among the leading predictors. DML causal forests recover confounder-adjusted loading sensitivities ∂CER/∂AOD strengthening from polluted land to pristine ocean (+ 0.82 to + 3.99 μm per unit AOD); distinct from the negative CER–AI Twomey slope, these sign-uniform positive values reflect a cleaner marine CCN-loading response. Aerosol size thus shapes ACI regionally, with robust oceanic Twomey-like signals but specification-dependent continental sensitivities reversing under alternative AE proxies, precluding a continental anti-Twomey interpretation.
Reactive nitrogen (Nr) management links climate policy, air quality and food production, yet its combined impacts remain poorly quantified. Here, we conduct ensemble simulations with two global climate models, Goddard Institute for Space Studies (GISS) and Community Earth System Model (CESM), to evaluate the air quality and climate co-benefits of Nr mitigation. By mid-century, both models simulate positive aerosol radiative forcing (RF) and negative ozone RF, yielding a net positive RF of 150 mW m−2 (95% CI: 120–180 mW m−2) in GISS and 60 mW m−2 (10–120 mW m−2) in CESM. Global air quality also improves, with global area-weighted mean PM2.5 concentrations decreasing by 0.16 µg m−3 (0.07–0.25 µg m−3) and 0.13 µg m−3 (0.05 to 0.21 µg m−3) and global area-weighted mean maximum daily 8-h average ozone decreasing by 1.69 ppb (1.48–1.89 ppb) and 0.99 ppb (0.88–1.10 ppb) in GISS and CESM, respectively. Regional PM2.5 reductions exceed 5 µg m−3. Moreover, more than 80% of avoided premature deaths occur across 0–60°N. Although efficient nitrogen management induces a modest, transient near-term climate penalty, long-term climate benefits from lower N2O emissions eventually offset this effect and deliver major air quality, health, and ecosystem co-benefits.
Mitigation of anthropogenic emissions can influence mineral dust not only by limiting global warming but also by reshaping regional circulation and hydroclimate. Here, we use fully coupled CESM2.1.3 simulations to compare future African springtime dust under a China-aligned global emissions pathway consistent with the Paris Agreement’s 2 °C target (SSP2-com) and a baseline experiment in which anthropogenic emissions are held at their 2023 levels (Fix2023). By the late twenty-first century, dust aerosol optical depth (AOD) increases over much of Africa in both experiments, but the increase is substantially weaker over the southern Sahara–Sahel region under SSP2-com. Within the main 10°–20°N source belt, Fix2023 produces a stronger North Africa–North Atlantic thermal contrast, a larger circulation response, and greater strengthening of low-level easterly winds, favoring dust mobilization and transport. Farther south, where the correspondence between local wind and dust AOD is weaker, smaller increases in the frequencies of high-temperature and dry days under SSP2-com point to a complementary hydroclimatic influence through meteorological conditions conducive to surface drying. The regional contrast therefore reflects spatially heterogeneous dynamical and hydroclimatic processes rather than a uniform wind-speed control. These results demonstrate that a mitigation pathway aligned with the 2 °C target can substantially limit projected African springtime dust intensification.
Ultrafine particles can penetrate sensitive human organs and are strongly associated with adverse health effects. This study presents the first high-temporal-resolution chemical characterization of PM0.1 at a Mediterranean background site under conditions in which photochemistry and transport are dominant processes. PM0.1 mass averaged 0.5 μg m−3 and exhibited sharp midday increases from frequent SO2-driven new particle formation events occurring within 50 km of the site. PM0.1 exhibited weak correlation (R2 = 0.25) with PM1, though organic aerosol and sulfate dominated both. Elemental analysis identified elevated Ca and Fe in PM0.1, potentially linked to shipping, whereas PM2.5 Ca was mostly associated with dust/soil. Volcanic plumes from Mt. Etna (750 km away) were also captured. During these events, SO2 levels increased tenfold and were accompanied by elevated levels of Zn, Pb, and As in PM0.1, indicating persistence of heavy metals in PM0.1 even over long-range transport.
