Fact-checked by Grok 2 weeks ago

European Centre for Medium-Range Weather Forecasts

The Centre for Medium-Range Weather Forecasts (ECMWF) is an independent intergovernmental organisation supported by 23 Member States and 12 Co-operating States, functioning as both a and a 24/7 operational service dedicated to producing and disseminating global numerical weather predictions. Established in 1975 through a ratified by initial nations to pool meteorological resources for advanced forecasting, ECMWF pioneered operational medium-range weather forecasts starting in 1979, initially hosted in Reading, , with additional sites now in , , for Copernicus services and , , for data dissemination. ECMWF generates deterministic and ensemble forecasts four times daily, extending from medium-range (up to 10 days) to seasonal outlooks up to a year ahead, leveraging sophisticated techniques, Earth system modeling, and one of Europe's most powerful supercomputing facilities to achieve high forecast skill. It maintains the world's largest meteorological data archive and operates key components of the European Union's , including the Atmosphere Monitoring Service for air quality and trace gases, the Service for reanalysis and projections, and contributions to . These efforts support national meteorological services, , energy sectors, and disaster preparedness across its supported states. Among its defining achievements, ECMWF introduced the first operational ensemble prediction system in 1992 to quantify forecast uncertainty, significantly advancing probabilistic weather prediction globally, and continues to lead in accuracy through ongoing research in predictability, integration, and high-resolution modeling. The organisation's models are benchmarked as outperforming many national counterparts, underpinning improvements in early warning for events and adaptation strategies. While technical challenges like model biases in near-surface parameters persist and are iteratively addressed through verification studies, ECMWF's emphasis on empirical validation and international collaboration has solidified its role as a cornerstone of European and global .

History

Founding and Early Development

The European Centre for Medium-Range Weather Forecasts (ECMWF) originated from a 1967 proposal by the Council of the European Communities for a shared "European Meteorological Computing Centre," driven primarily by political aims to foster in rather than purely scientific imperatives. This initiative culminated in a project study completed on 5 August 1971, followed by a decision from the in November 1971 to establish the Centre. The first draft of the ECMWF Convention was discussed on 9 and 10 December 1971 by representatives from 14 of the eventual 18 founding Member States. The Convention was signed in 1973 and entered into force on 1 November 1975, formally creating ECMWF as an independent intergovernmental organization headquartered in Reading, , selected for its proximity to the UK Met Office and the . The 18 founding Member States ratified the original Convention, pooling resources to advance medium-range weather forecasting (initially defined as 4-10 days ahead) through collaborative research and operations. Early efforts emphasized cost-benefit justification, with analyses estimating annual economic gains of 200 million units of account from improved forecasts against a 20 million unit setup cost, based on interviews across 15 countries. ECMWF began as an extension of European Cooperation in Science and Technology (COST) projects, focusing on numerical weather prediction models and data assimilation. Initial operations relied on high-speed data links to national meteorological services, with computing infrastructure evolving into one of Europe's largest dedicated systems. The Centre produced its first real-time medium-range forecast on 15 June 1979, coinciding with the official opening of its Shinfield Park headquarters. Operational production for Member States commenced on 1 August 1979 with 10-day forecasts issued five days per week, expanding to daily seven-day coverage by 1 August 1980; these early outputs demonstrated rapid skill improvements, with 1980s seven-day forecasts outperforming prior five-day accuracy benchmarks.

Key Operational Milestones

The European Centre for Medium-Range Weather Forecasts (ECMWF) initiated operational medium-range forecasting on 1 August 1979, delivering ten-day predictions five days per week via an N48 spectral grid model with approximately 200 km horizontal resolution. This marked the transition from experimental real-time forecasts conducted in June 1979 to routine production, enabling Member States to access global outputs for the first time. A pivotal advancement occurred on 24 November 1992, when ECMWF integrated the first ensemble predictions into its operational system, initially running the three days per week to quantify forecast uncertainty through perturbations in initial conditions and model physics. This probabilistic approach, expanded to daily operations by the mid-1990s, significantly enhanced medium-range reliability by representing the range of possible outcomes rather than deterministic single forecasts. In 1997, ECMWF pioneered the operational implementation of four-dimensional variational (4D-Var) , shifting from optimal interpolation to a method that optimizes the model state over a time window by minimizing discrepancies between observations and short-range forecasts. This upgrade, building on experimental trials from January 1996, improved accuracy and , particularly for atmospheric dynamics, establishing a for global centers. Subsequent upgrades to the Integrated Forecasting System (IFS) included resolution enhancements and coupled modeling; for instance, Cycle 48r1 in June 2023 increased medium-range horizontal to 9 km, unifying configurations across forecast ranges while incorporating advances in physics and methods. More recently, on 25 February 2025, ECMWF operationalized its Forecasting System (AIFS), running parallel to the deterministic IFS to leverage for faster, data-driven predictions without sacrificing skill in core variables. The variant of AIFS followed on 1 July 2025, extending probabilistic capabilities through AI-generated perturbations.

Institutional Expansion and Modern Era

In response to growing computational demands and uncertainties following the United Kingdom's , ECMWF's member states approved amendments to its in 2005 and 2010, facilitating the addition of new members and expanding the organization's reach to 23 member states and 12 cooperating states by the . This growth supported enhanced international collaboration in medium-range and , with contributions funding an annual budget primarily derived from state assessments. To address escalating requirements for and , ECMWF pursued infrastructural expansion, culminating in the establishment of a multi-site . In December 2020, the ECMWF Council selected , , for new premises dedicated to Copernicus Atmosphere Monitoring Service operations, with activities commencing in late 2021. Concurrently, in 2017, Italy's proposal to host a new data centre in was accepted, leading to the facility's formal opening on 14 September 2021; this site, refurbished from a former tobacco factory at the Tecnopolo complex, now accommodates advanced supercomputing systems, including installations operational from 2022. The transition to marked a pivotal modern development, with the migration of over 450 petabytes of meteorological archive data completed in 2022, followed by the full shift of operational forecast services from Reading, , on 18 October 2022. This relocation ensured continuity and scalability for ECMWF's activities, integrating cloud infrastructure for services like the and Atmosphere Data Stores. These changes positioned ECMWF as a distributed entity—maintaining headquarters in Reading while leveraging for computing and for specialized monitoring—enhancing resilience and alignment with European environmental programmes such as Copernicus.

Governance and Organization

Member and Co-operating States

The ECMWF is governed and primarily funded by its 23 Member States, which collectively provide the bulk of its financial resources through contributions scaled to their ; these states hold representation in the ECMWF Council, the Centre's principal decision-making body. In 2023, contributions from the 35 total Member and Co-operating States amounted to £62.5 million, underscoring their role as the Centre's principal financiers. Member States also receive enhanced access to ECMWF's computing resources, including supercomputers and permanent data storage in the meteorological archive.
Member States
Austria, Belgium, Croatia, Denmark, Estonia, Finland, France, Germany, Greece, Iceland, Ireland, Italy, Luxembourg, Netherlands, Norway, Portugal, Serbia, Slovenia, Spain, Sweden, Switzerland, Türkiye, United Kingdom
Co-operating States, numbering 12, participate without formal governance roles or scaled financial obligations equivalent to Members; they benefit from access to basic computing facilities, the meteorological archive, and temporary data storage, facilitating collaboration on weather prediction while relying on Member-funded infrastructure.
Co-operating States
Bulgaria, Czech Republic, Georgia, Hungary, Israel, Latvia, Lithuania, Montenegro, Morocco, North Macedonia, Romania, Slovakia
This tiered structure enables broader international engagement in medium-range forecasting while ensuring operational stability through committed Member State support.

Leadership and Decision-Making Processes

The constitutes the principal governing and decision-making authority of the European Centre for Medium-Range Weather Forecasts (ECMWF), composed of two representatives from each , with at least one designated from the respective national meteorological service. It convenes biannually—typically in and December—to deliberate and approve critical elements including the Centre's long-term strategy, annual programme of work, budget, financial regulations, and staff regulations, as stipulated in the ECMWF established on 1 November 1975 and amended on 6 2010. The also holds authority over admissions of new s, ratification of cooperation agreements, appointments of the Director-General and deputy, and major property acquisitions, ensuring alignment with the intergovernmental framework while incorporating input from observers such as the , , and EUMETNET. The Director-General serves as the chief executive officer, appointed by the Council for a renewable term, and is responsible for the operational management, preparation of draft programmes and budgets for Council approval, and reporting on Centre activities. Dr. Florence Rabier has held this position since January 2016, overseeing strategic implementation amid expansions in capabilities and services. Florian Pappenberger, currently Deputy Director-General and Director of Forecasts, has been elected as the next Director-General, effective 1 January 2026. The Council elects its own and Vice-President annually from among its members, with terms limited to one year and no more than two consecutive re-elections; as of 2024, Penny Endersby of the serves as , with Dr. Roar Skålin of as Vice-President. Decision-making is supported by a network of advisory committees that furnish specialized recommendations to the Council, including the Scientific Advisory Committee on research and programme drafts, the Finance Committee on budgetary and financial oversight with delegated powers, the Policy Advisory Committee on overarching policies, the Technical Advisory Committee on operational matters, the Advisory Committee of Co-operating States on programme and budget implications for non-members, the Advisory Committee for Data Policy on data dissemination, and the Joint Advisory Group on collaborative initiatives like the European Weather Cloud. These bodies ensure scientifically robust and fiscally prudent resolutions, with the Council retaining final authority on all substantive approvals to maintain consensus among Member States.

Mission and Strategic Objectives

Core Research and Forecasting Mandate

The European Centre for Medium-Range Weather Forecasts (ECMWF) was established with the primary purpose of developing a capability for medium-range and providing such forecasts to its Member States. This mandate, enshrined in Article 2 of its founding , emphasizes the operation of global numerical models and data-assimilation systems to generate predictions of the Earth's fluid envelope, including the atmosphere, , and land surface components. Forecasts are produced using numerical methods, with initial conditions derived from assimilated observational data, extending typically from a few days to two weeks ahead, though the Centre also supports extended-range predictions up to seasonal timescales. The Centre delivers these outputs four times daily, prioritizing high-resolution ensemble-based predictions to quantify and enhance reliability for national meteorological services. Complementing forecasting, ECMWF's mandate includes continuous scientific and technical research aimed at improving prediction accuracy and extending forecast lead times. This involves advancing (NWP) techniques, such as model physics refinements, ensemble methods, and integration of emerging technologies like , to push the limits of medium-range skill amid challenges like climate variability. efforts focus on monitoring the broader system, including atmospheric composition and dynamics, to inform not only weather but also environmental applications, while ensuring outputs remain complementary to Member States' short-range capabilities. The Centre collects and archives meteorological data to support these activities, making processed results accessible to Members under Council-defined conditions. ECMWF's forecasting mandate extends to collaborative functions, such as providing resources prioritized for NWP-related by Member States and contributing to (WMO) programs. Advanced training for scientific personnel in numerical forecasting techniques is also integral, fostering expertise across . Under its 2025–2034 strategy, the Centre reaffirms this core focus on delivering world-leading medium-range predictions through physical and data-driven innovations, while collaborating within the European Meteorological Infrastructure to address evolving demands like preparedness. These elements collectively position ECMWF as a hub for global-scale, grounded in rigorous model development and validation against observations.

Integration with Broader Environmental Programmes

The European Centre for Medium-Range Weather Forecasts (ECMWF) implements the Copernicus Atmosphere Monitoring Service (CAMS) and the Copernicus Climate Change Service (C3S), which form core components of the European Union's flagship programme, Copernicus, aimed at monitoring environmental changes using satellite, in-situ, and model-based data. CAMS, operational since July 2015, delivers global and regional forecasts and analyses of atmospheric composition, including air quality, greenhouse gases, and aerosol levels, leveraging ECMWF's infrastructure to support policy decisions on and forcing. C3S, also managed by ECMWF, provides authoritative climate data sets, including historical reanalyses from 1950 onward, sector-specific indicators, and seasonal-to-decadal predictions, enabling users to assess climate variability and extremes for adaptation strategies. This integration extends ECMWF's medium-range forecasting mandate into long-term environmental monitoring by assimilating diverse observational inputs—such as satellite measurements from missions—into coupled models that simulate system interactions, including atmosphere-ocean feedbacks and fluxes. ECMWF contributes to the Copernicus Service (CEMS) through rapid mapping and early warning products, such as the European Awareness (EFAS), which uses predictions to forecast river discharges up to 10 days ahead, aiding across . These services produce open-access data disseminated via standardized interfaces, fostering applications in , , and sectors, with over 10 years of operational delivery by 2025 demonstrating improved accuracy in air quality forecasts and climate projections through iterative model upgrades. Beyond Copernicus, ECMWF supports national collaboration programmes in EU member states to enhance the uptake of environmental data, including tailored training and toolkits for integrating CAMS and C3S outputs into local decision-making processes, such as urban air quality management and assessments. This broader alignment with environmental objectives, funded through delegated implementation agreements since , positions ECMWF as a bridge between operational and policy-relevant , while maintaining its independence as an intergovernmental organization.

Core Operational Activities

Medium-Range Numerical Weather Prediction

The European Centre for Medium-Range Weather Forecasts (ECMWF) specializes in (NWP) for the medium-range timeframe, defined as 3 to 15 days ahead, leveraging its Integrated Forecasting System (IFS) to generate global forecasts. The IFS, developed in collaboration with , integrates atmospheric, oceanic, land surface, and sea-ice components into a unified framework for and prediction, solving the governing equations of atmospheric dynamics and physics on a grid. Operational medium-range products include the high-resolution deterministic forecast (HRES), which provides detailed single-member predictions, and the forecast (ENS), comprising 51 perturbed members to quantify uncertainty in weather evolution. These forecasts are produced twice daily, assimilating observational data via a variational to initialize model states. The IFS employs triangular truncation for , achieving operational s of approximately 9 for both HRES and ENS in recent cycles, enabling simulation of mesoscale phenomena while maintaining computational efficiency for global domains. Vertical consists of 137 hybrid-sigma-pressure levels extending to 0.01 , capturing tropospheric and stratospheric processes critical for medium-range . Key physical parameterizations include schemes, microphysics, and models refined through ongoing to address biases in and temperature forecasts. The ensemble generation incorporates perturbations in initial conditions and model physics via schemes, providing probabilistic outputs such as probability maps for extreme events. ECMWF's medium-range NWP demonstrates leading global accuracy, with verification statistics showing anomaly correlations for 500 hPa exceeding 60% at 10-day lead times in the , outperforming many national centers due to advanced and model resolution. Continuous evaluation against observations informs upgrades, such as enhancements in cycle 48 (implemented 2023), which improved track forecasts and upper-air predictability. Despite challenges from atmospheric chaos limiting deterministic skill beyond 10 days, ensemble methods extend reliable probabilistic guidance, supporting applications in , energy, and disaster preparedness.

Ensemble and Probabilistic Forecasting

The European Centre for Medium-Range Weather Forecasts (ECMWF) pioneered operational (EPS) to address inherent uncertainties in , initiating them on 24 November 1992 with low-resolution integrations at T63 spectral truncation and 19 vertical levels, comprising one unperturbed control forecast and 32 perturbed members run to 10 days ahead. This approach sampled forecast probability distributions by introducing perturbations to initial conditions via singular vectors—optimized structures capturing rapidly growing instabilities—and bred-vector methods, alongside stochastic physics perturbations to represent model errors. By quantifying spread around the deterministic forecast, the EPS enabled outputs, such as the likelihood of exceeding thresholds or anomalies, improving decision-making for events like severe storms where single forecasts often underperformed. Evolution of the EPS has focused on enhancing , size, and realism to better resolve uncertainty. The current medium-range EPS employs 50 perturbed members plus a , at approximately 9 horizontal (TCo1279) and 137 vertical levels, extending to 15 days with twice-daily runs; perturbations now incorporate an of data assimilations (EDA) since , generating variability from 25 lower- analyses to propagate observational uncertainties. Stochastic parametrizations, refined over decades, perturb subgrid-scale processes like and clouds to mimic structural model deficiencies, yielding more reliable spread-skill relationships where variance correlates with . Probabilistic products derive from statistics, including tercile probabilities for variables like 2 m and mean sea-level , calibrated via techniques such as the for verification against observations, ensuring outcomes align with predicted frequencies. Recent innovations integrate to augment traditional physics-based ensembles, with the Forecasting System (AIFS) ensemble becoming operational on 1 July 2025, offering comparable skill to the physics-driven at reduced computational cost while maintaining probabilistic integrity through diffusion models that generate diverse trajectories. This hybrid approach has demonstrated superior performance in ensemble spread for medium-range variables, reducing biases in and forecasts, as validated against reforecast datasets. The 's impact extends to downstream applications, including extreme event warnings, where probabilistic guidance has measurably improved economic value via cost-loss analyses, outperforming deterministic methods by capturing low-probability high-impact scenarios.

Extended-Range and Seasonal Predictions

ECMWF's sub-seasonal-range forecasts, extending up to 46 days ahead, deliver probabilistic assessments of weekly mean anomalies relative to a model , emphasizing potential deviations such as warmer or colder than average conditions. Updated daily at approximately 36 km horizontal resolution, these predictions exploit sources of predictability including slowly evolving sea surface temperatures, , stratospheric circulation, and residual atmospheric memory from initial conditions. The forecasts employ the Integrated Forecasting System (IFS) in configuration to quantify , facilitating applications in sectors like and planning where medium-range determinism fades. Complementing these, ECMWF's seasonal forecasts address timescales from one to seven months using the SEAS5 system, operational since 5 November 2017, which generates monthly-updated 51-member ensembles for monthly means and quarterly updates for annual forecasts up to 13 months. SEAS5 couples the IFS atmospheric component (Cycle 43r1 at TCo319 resolution, ~36 km) with the NEMO ocean model (ORCA025 configuration), LIM2 sea-ice prognostic scheme, and land surface interactions, initialized via the ORAS5 ocean and sea-ice reanalysis ensemble. This setup enhances representation of air-sea coupling critical for phenomena like El Niño-Southern Oscillation, with hindcasts spanning 1981–2016 for calibration. SEAS5 demonstrates tangible advances over its predecessor System 4, including a larger ensemble (51 versus 15 members), refined tropical biases (e.g., ~2°C reduction in Niño3.4 RMSE beyond two months), and improved sea-ice edge predictions through interactive modeling. scores, such as correlations for 2 m , show gains in tropical and select extratropical regions like the North Pacific, though variability persists in areas like the North-West Atlantic due to initialization shifts; overall reliability supports probabilistic outputs for climate-sensitive decisions in water resource management and disaster preparedness. These systems contribute to ECMWF's seamless prediction framework, bridging medium-range with probabilistic long-range guidance.

Data Handling and Analysis

Observational Data Assimilation

The European Centre for Medium-Range Weather Forecasts (ECMWF) employs observational data assimilation to integrate diverse measurements into its numerical weather prediction models, producing an optimal estimate of the atmospheric state that minimizes discrepancies between forecasts and observations. This process combines short-range model forecasts—serving as a background—with incoming observations via statistical methods, correcting systematic errors and initializing subsequent predictions. The assimilation cycle operates sequentially every 12 hours, updating the model state to reflect the latest data while accounting for observational errors and model uncertainties. ECMWF's primary technique is four-dimensional variational (4D-Var), which iteratively adjusts initial conditions over a 12- or 24-hour window to fit both observations and model dynamics, incorporating in four dimensions (three spatial plus time). This method outperforms simpler three-dimensional variants by enforcing physical consistency across the assimilation period, enabling better handling of sparse or delayed observations. Complementing 4D-Var, an Ensemble of Data Assimilations (EDA) runs 51 parallel lower-resolution cycles to sample analysis uncertainties, which directly initialize probabilistic ensemble forecasts and quantify error covariances more realistically than static approximations. Observations assimilated span multiple platforms, including approximately 90 satellite instruments processed daily for radiance data, along with in-situ measurements from weather stations, ships, buoys, radiosondes, aircraft reports, and oceanographic profiles such as floats. Satellite data dominate due to global coverage, but rejects outliers based on model consistency and instrument metadata; for instance, hyperspectral sounders provide vertical profiles of and , while imagers contribute and surface parameters. Land surface assimilation separately incorporates from satellites like SMOS and ASCAT, using simplified variants tailored to non-Gaussian error distributions in variables like depth. Ocean components draw from altimetry, , and salinity profiles, though atmospheric remains the core driver for medium-range forecasting. Recent enhancements include coupled atmosphere-ocean-ice 4D-Var implementations since 2016, reducing initialization shocks, and ongoing experiments with for bias correction in radiances to address instrument degradation. These updates have incrementally improved , with diagnostics revealing that assimilated observations account for about 18% of global error reduction, the remainder stemming from forecast constraints. Despite advances, challenges persist in assimilating convective-scale and resolving mesoscale features, prompting into ensemble-variational approaches for future operational upgrades.

Reanalysis and Historical Datasets

The European Centre for Medium-Range Weather Forecasts (ECMWF) generates reanalysis datasets by integrating historical observations with advanced models through four-dimensional variational (4D-Var) , yielding consistent estimates of atmospheric, oceanic, and land surface states over decades. These products reconstruct past weather and climate conditions, enabling applications in climate research, model validation, and long-term , while addressing observational gaps via model physics. ERA5, the fifth-generation ECMWF reanalysis produced under the Copernicus Service, spans hourly data from 1940 to the present, offering global coverage at approximately 31 km horizontal resolution and 137 vertical levels extending to 80 km altitude. It incorporates diverse observations, including satellite radiances and in-situ measurements, using the Integrated Forecasting System Cycle 41r2 for 1950–present and earlier cycles for pre-1950 periods to handle sparse data. Compared to its predecessor ERA-Interim (1979–August 2019, at 79 km resolution), ERA5 demonstrates reduced biases in variables like and due to enhanced model resolution, improved humidity control, and inclusion of remotely sensed . Complementing ERA5, the ERA5-Land dataset provides a 9 replay of land-surface components from 1950 onward, driven by ERA5 atmospheric forcing but with uncoupled land modeling to enhance detail in , , and energy fluxes over land. It maintains hourly temporal and includes uncertainty estimates at coarser scales, supporting applications in monitoring and where fine-scale effects matter. Earlier reanalyses, such as ERA-40 (covering 1957–2002), laid foundational methodologies but suffered from coarser resolution (about 125 km) and fewer assimilated observations, limiting their utility for regional studies compared to modern products. ECMWF's datasets are accessible via the Copernicus Climate Data Store, with ongoing updates ensuring near-real-time extensions and quality improvements through bias corrections and expanded observation types. These resources underpin global climate assessments, though users must account for model-dependent uncertainties in underrepresented phenomena like aerosols or sub-grid .

Technological Foundations

Integrated Forecasting System and Model Evolution

The Integrated Forecasting System (IFS) constitutes ECMWF's primary operational framework for , unifying , dynamical modeling, and physical parameterizations within a coupled system representation. At its core, the IFS employs a four-dimensional variational (4D-Var) scheme to estimate the initial atmospheric, oceanic, sea-ice, and land-surface states by minimizing discrepancies between model trajectories and diverse observations, including radiances, radiosondes, and surface measurements. This analysis initializes subsequent forecast integrations using hydrostatic for atmospheric dynamics, solved through transforms and semi-Lagrangian schemes, coupled to , wave, and biogeochemical modules for comprehensive system simulations. The system supports deterministic and ensemble predictions across medium-range (up to 15 days), sub-seasonal (up to 46 days), and seasonal (up to 7 months) horizons, with resolutions varying by application—such as TL1279 (approximately 9 km grid spacing) for high-resolution atmospheric forecasts. Development of the IFS traces back to the , evolving from earlier models into a modular that integrated advanced and coupled components. A pivotal milestone occurred with IFS cycle 11r7, implemented on 2 March 1994, which represented a comprehensive rewrite of the forecast model, enabling semi-Lagrangian dynamics and paving the way for operational 4D-Var by the late . Subsequent cycles incrementally refined , physics (e.g., cloud microphysics, schemes, and boundary-layer ), and observation handling, driven by empirical against reanalyses and targeted to mitigate biases in phenomena like tropical and stratospheric variability. By the early 2000s, the system incorporated prediction systems () with perturbations for probabilistic outputs, enhancing reliability in medium-range guidance. Recent upgrades have focused on unifying configurations, elevating computational efficiency, and integrating elements for parameter . Cycle 47r3, deployed on 12 October 2021, enhanced moist physics representations and expanded of hyperspectral data, yielding measurable gains in and upper-air . Cycle 48r1, introduced on 27 June 2023, standardized medium-range resolutions across deterministic and streams while improving overall model and tropical cyclone track predictions through refined ocean-atmosphere coupling. The latest operational iteration, cycle 49r1 on 12 November 2024, prioritized refinements in boundary-layer processes and modeling, resulting in superior and near-surface forecasts, as validated against verification datasets. These evolutions reflect iterative empirical , with each cycle tested via parallel runs against predecessors to ensure and improvements exceeding 1-2% in scores for key variables like 500 .
IFS CycleImplementation DateKey Improvements
47r312 October 2021Enhanced moist physics schemes; improved of and observations for better and cloud analysis.
48r127 June 2023Unified resolutions in medium-range forecasts; advancements in for aerosols and land-surface states; reduced biases in extended-range predictions.
49r112 November 2024Optimized and temperature profiles; enhanced variational bias correction for data; improved ensemble spread calibration.
Ongoing modernization efforts emphasize a shift toward modular, open-source codebases to facilitate on platforms and integration of hybrid physics-machine learning approaches, addressing limitations in traditional parameterizations for sub-grid processes. Complementing this, ECMWF has operationalized the Forecasting System (AIFS) on 25 February 2025 as a data-driven counterpart, trained on decades of IFS outputs to emulate core dynamics and physics, achieving comparable skill in upper-air and surface variables while reducing computational demands—marking a hybrid evolution beyond purely physics-based modeling.

Computing Infrastructure and Emerging AI Applications

ECMWF's high-performance computing facility (HPCF), located in , , forms the backbone of its operations. The current system, an BullSequana XH2000 operational since 18 October 2022, comprises four self-sufficient clusters with 7,680 compute nodes and 448 general-purpose I/O (GPIL) nodes, powered by processors offering 1,040,384 total cores and 2.1 PiB of memory. This setup delivers approximately 30 petaflops of sustained performance, supporting daily global forecasts through intensive simulations of atmospheric dynamics. The facility employs direct liquid cooling for efficiency, hierarchical Lustre-based storage via DDN EXAScaler, and a high-bandwidth HDR network exceeding 300 Tbps , with upgrades typically occurring every four to five years under an €80 million service contract with . Recent enhancements to the HPCF have enabled significant advances in forecast resolution and ensemble size, such as increasing medium-range ensemble forecasts from 18 km to 9 km grid spacing in Integrated Forecasting System (IFS) 48r1, thereby improving predictive detail for phenomena like tropical cyclones and atmospheric rivers. ECMWF collaborates on broader European initiatives, including access to the exascale supercomputer at , which entered operation on 5 September 2025 and targets at least one exaflop for enhanced and modeling. These resources underpin , ensemble generation, and reanalysis tasks, though they face ongoing challenges in and amid rising computational demands from higher-resolution models. In parallel, ECMWF has pioneered (AI) applications to augment traditional physics-based modeling, culminating in the Artificial Intelligence Forecasting System (AIFS), a data-driven model based on graph neural networks and inspired by advancements like DeepMind's GraphCast. Launched operationally on 25 February 2025, AIFS runs four times daily alongside the IFS, generating global forecasts at approximately 1-degree resolution using pressure-level and surface fields trained on ERA5 reanalysis data from 1979–2018 and fine-tuned with IFS outputs from 2019–2020. An ensemble variant followed on 1 July 2025, enabling probabilistic predictions with reduced computational overhead compared to deterministic physics simulations. AIFS demonstrates superior skill in key metrics, such as anomaly correlation coefficient () for 500 hPa geopotential height, outperforming the IFS in medium-range forecasts for verification periods, though it trails in surface variables like 2 m at coarse resolutions. Built using the open-source framework with and PyTorch Geometric, it leverages ECMWF's HPCF GPU capabilities for training and inference, offering faster run times—potentially tens of times quicker than equivalent numerical integrations—while requiring less energy. Forecasts are publicly accessible via OpenCharts and ECMWF's policy, fostering into hybrid AI-physics systems and sub-seasonal predictions through initiatives like the AI Weather Quest . Future developments include higher-resolution AIFS variants and integration with exascale resources to address limitations in extreme event predictability and long-range forecasting.

Outputs, Services, and Dissemination

Forecast Products and Accessibility

The European Centre for Medium-Range Weather Forecasts (ECMWF) generates a range of forecast products through its Integrated Forecasting System (IFS), including high-resolution deterministic forecasts (HRES) extending up to 15 days ahead at approximately 9 km , ensemble forecasts (ENS) comprising 51 members for probabilistic predictions, and extended-range forecasts up to 46 days. These products cover atmospheric variables such as , , , and , alongside specialized outputs like wave forecasts and tracks. Additionally, ECMWF produces seasonal forecasts using coupled ocean-atmosphere models, providing multi-month outlooks for variables including anomalies and totals. Forecast outputs are disseminated in multiple formats to support diverse applications, including graphical charts (e.g., meteograms, pattern maps, and tracks), gridded datasets in and , and BUFR-encoded observations-integrated products. Real-time data from both the IFS and the newer Forecasting System (AIFS) are included, with AIFS offering experimental high-resolution predictions leveraging for efficiency. Verification statistics and historical comparisons accompany these products to aid user interpretation. Accessibility to ECMWF products is structured by user type and agreement level, with member states and licensed partners receiving full, low- access to high-resolution data via dedicated systems like the ECMWF Web Charting (EWC) and . Public access has expanded significantly; a subset of real-time forecast data has been freely available since prior years through platforms like AWS and the World Meteorological Organization's systems (WIS, GTS), in formats including and graphical products downloadable via FTP. As of 1 October 2025, ECMWF opened its entire real-time product catalogue to the public at no cost, initially at 25 km resolution with zero , distributed via public clouds and direct servers to enhance global usability while maintaining sustainability. Higher-resolution access is planned for free public release in 2026, alongside continued dissemination through EUMETCast for broadcast reception. This policy shift aligns ECMWF with principles, broadening applications in research, disaster preparedness, and commercial weather services without compromising core operational priorities.

Applications in Severe Weather Monitoring

The European Centre for Medium-Range Weather Forecasts (ECMWF) applies its medium-range prediction system to monitor and provide early warnings for events, including wind storms, heatwaves, floods, droughts, and tropical cyclones, typically several days in advance to support contingency planning. These forecasts leverage probabilistic outputs from the 51-member to quantify risks of extremes, outperforming deterministic models in capturing and low-probability high-impact scenarios. Central to these applications are the Extreme Forecast Index (EFI) and Shift of Tails (SOT), which compare ensemble forecast distributions against long-term climatologies to flag potential anomalies. The EFI, ranging from 0 to 1, indicates the proportion of the forecast differing from climate norms, with values above 0.6 signaling elevated risk, while SOT measures distributional shifts toward extremes. Verification shows these tools exhibit skill (relative operating characteristic area >0.5) for medium-range lead times up to 7 days or more, particularly when combining parameters like convective available potential energy (CAPE) with precipitation probabilities. For tropical cyclones, ECMWF ensembles generate probabilistic track and intensity forecasts, with verified improvements in accuracy; for instance, they contributed to predictions of Cyclone Dianne's path in early 2024. In severe , EFI based on CAPE-shear parameters provided signals 6 days ahead for events like the June 18, 2016, thunderstorms in and , and April 26, 2016, outbreaks across the U.S. from to . Extreme wind monitoring uses EFI to detect gust anomalies, aiding warnings for storms. For extremes linked to floods, EFI and integrated water vapor transport EFI prototypes identify heavy rainfall risks, as in a forecasting system prototype and assessments showing utility for week-2 lead times during events. Heatwave monitoring, such as the early summer 2025 event, employs EFI to highlight unusual temperature deviations. These products are disseminated via charts and datasets, enabling national services to tailor local alerts.

Impact, Verification, and Challenges

Forecast Accuracy and Global Influence

The European Centre for Medium-Range Weather Forecasts (ECMWF) operational forecasts demonstrate leading accuracy in medium-range predictions, as quantified by standardized verification metrics such as the anomaly correlation coefficient () for 500 and continuous ranked probability skill score (CRPSS) for and . For the deterministic high-resolution forecast (HRES), the ACC for 500 hPa height in the extratropical maintains high skill, with thresholds like 80% ACC achieved at lead times superior to those of peer models, based on ongoing evaluations against operational analyses. predictions (ENS) further enhance reliability through probabilistic outputs, with 2024 verifications showing a roughly one-quarter day extension in skillful for 24-hour forecasts in the extratropics, verified against station observations via metrics like CRPSS exceeding 10%. Comparisons with other global systems, such as the U.S. National Centers for Environmental Prediction's (GFS), consistently position ECMWF ahead in medium-range skill for variables like mean sea level pressure and upper-air patterns, as tracked through (WMO) standardized assessments. These accuracy advantages stem from advanced , high-resolution modeling (9 km grid spacing in HRES), and ensemble perturbations capturing , enabling reliable predictions up to 10-15 days for synoptic-scale features. Annual evaluations, including those for 2020-2021 and beyond, confirm steady improvements, with ECMWF maintaining the longest lead times for useful skill in extratropical regions despite challenges like tropical predictability limits. Independent benchmarks, such as those comparing against NCEP ensembles, highlight ECMWF's edge, where GFS achieves a 500 hPa ACC of 0.8 at about 8.7 days, while ECMWF extends comparable thresholds further through refined physics and initialization. ECMWF exerts substantial global influence, with its forecasts serving as a for national meteorological and hydrological services (NMHS) in 34 Member States (primarily ) and cooperating states, plus adoption by additional WMO members and non-European entities for operational guidance. Daily products, including deterministic and ensemble outputs, inform decision-making in weather-sensitive sectors worldwide, such as aviation routing, optimization, and agricultural planning, facilitated by free access to charts, datasets, and APIs. Collaborative initiatives, including impact experiments simulating enhanced observations, demonstrate ECMWF's role in bolstering global forecast quality, particularly in data-sparse regions, through shared reanalyses like ERA5 and partnerships that amplify skill across interconnected prediction systems. This influence extends to research and AI model training, where ECMWF data underpins advancements in adopted internationally.

Technical Limitations, Biases, and Improvement Efforts

The ECMWF's Integrated Forecasting System (IFS) faces inherent limitations stemming from the chaotic nature of atmospheric dynamics, where small errors amplify over time, leading to predictability horizons typically extending to about 10-15 days for medium-range forecasts. Computational constraints further restrict horizontal resolutions, with operational ensemble forecasts currently at approximately 9 km despite ambitions for kilometre-scale grids, as higher resolutions demand exponentially more resources without proportional gains in skill beyond certain thresholds. Model representations of sub-grid processes, such as and cloud microphysics, introduce approximations that degrade accuracy for mesoscale phenomena like localized thunderstorms or tropical cyclones. Systematic biases persist in IFS outputs, including cold sea-surface biases in seasonal forecasts over regions like the Equatorial Eastern and negative biases in high-latitude runoff simulations, partly attributable to land shortcomings. Near-surface forecasts exhibit regional cold biases in winter, while elevation-dependent errors distort altitudinal gradients in mountainous areas, affecting variables like and winds. These biases arise from mismatches between model physics and observations, as well as assumptions of bias-free models in , which overlook systematic errors in observation operators or forcing terms. In predictions, underestimation of can mask true , blending bias with probabilistic shortcomings. ECMWF addresses these through iterative IFS cycle upgrades, such as the June 2023 implementation that unified medium-range resolutions at 9 km for both deterministic and ensemble runs, yielding 2-6% skill improvements in surface variables. The 2024 upgrade enhanced two-metre temperature forecasts via refined 4D-Var assimilation of surface observations and updated vegetation/snow schemes, reducing stratospheric biases by up to 50% with weak-constraint 4D-Var configurations. Ongoing efforts incorporate machine learning for bias correction, including post-processing heavy precipitation and adaptive schemes for numerical weather prediction outputs, alongside ocean-wave model refinements for better air-sea interactions under strong winds. The 2021-2030 strategy emphasizes scalability for 5 km ensembles, advanced data assimilation to handle model errors, and hybrid AI-physics approaches to mitigate extrapolation limits in unseen regimes.

Geopolitical and Funding Dependencies

The European Centre for Medium-Range Weather Forecasts (ECMWF) derives the majority of its operational funding from annual contributions by its 35 Member and Co-operating States, which totaled £62.5 million in 2023 and are apportioned according to a scale reflecting each state's gross national income. These contributions support core activities such as numerical weather prediction research and infrastructure maintenance, creating a direct dependency on the fiscal priorities and economic stability of participating governments, including 23 Member States (e.g., Austria, Belgium, France, Germany, Italy, the Netherlands, Norway, Sweden, Switzerland, Turkey, and the United Kingdom) and 12 Co-operating States. Additional revenue streams include sales of forecast data and externally funded projects, but national contributions remain the foundational element, exposing ECMWF to risks from budgetary constraints or policy shifts in key contributors like Germany and France, which provide the largest shares due to their economic scale. A significant supplementary dependency arises from contracts for implementing the Copernicus programme's Atmosphere Monitoring Service (CAMS) and Service (C3S), renewed periodically with the and providing dedicated funding for climate and atmospheric data services since 2014. This funding, which supports specialized infrastructure like the Copernicus hub in , , ties ECMWF's expansion in environmental monitoring to the bloc's budgetary cycles and priorities, potentially vulnerable to shifts in EU cohesion policy or post-2027 programme reforms. While diversifying revenue reduces sole reliance on state dues, the Copernicus linkage amplifies exposure to intra-European political dynamics, as evidenced by the partial relocation of operations from the (Reading) to Italy () following to safeguard continuity amid uncertainties in UK-EU relations. Geopolitically, ECMWF's consensus-based governance—requiring agreement among diverse member states for strategic decisions—introduces dependencies on diplomatic alignment, with non-EU members like the , , , and providing balance against predominant influence but complicating unified responses to external pressures. Its forecasting models assimilate global observational data via (WMO) conventions, rendering accuracy contingent on uninterrupted international exchange; disruptions, such as those during the 2022 , where Western providers including EUMETNET curtailed meteorological data access to to prevent military misuse, have highlighted vulnerabilities to conflict-induced gaps in upstream observations from affected regions. ECMWF mitigated such risks by prioritizing humanitarian forecasting support for and emphasizing data gap closure experiments, which demonstrated forecast degradation without comprehensive inputs, underscoring the causal link between geopolitical tensions and operational efficacy. 's non-membership limits direct leverage but amplifies broader risks from sanctions or reciprocal withholdings in WMO data flows.

References

  1. [1]
    Who we are - ECMWF
    ECMWF is both a research institute and a 24/7 operational service, producing and disseminating numerical weather predictions to its Member States. This data is ...History · Governance · Staff profiles · Member StatesMissing: founding | Show results with:founding
  2. [2]
    History - ECMWF
    The first real-time medium-range forecasts were made in June 1979, and ECMWF has been producing operational medium-range weather forecasts since 1 August 1979.
  3. [3]
    Key facts and figures | ECMWF
    We use the scientific advances made in areas such as data assimilation, Earth system modelling, predictability and reanalysis to improve our forecasts. · We have ...
  4. [4]
    What we do
    ### Main Services
  5. [5]
  6. [6]
    Addressing biases in near-surface forecasts - ECMWF
    Turning next to daytime biases in summer, daytime T2m and humidity forecasts for Europe have an overall cold and dry bias. This could be an issue just in the ...Missing: controversies criticisms
  7. [7]
    [PDF] Background and history | ECMWF
    Nov 1, 1975 · Over the years, ECMWF had developed a capability for operational ocean wave prediction, seasonal forecast- ing and ensemble prediction. The ...Missing: founding | Show results with:founding
  8. [8]
    ECMWF celebrates 40 years of operational medium-range forecasting
    Jul 31, 2019 · On 1 August 1979 ECMWF produced its first operational medium-range forecast for Member States. This major achievement marked the first ...Missing: establishment | Show results with:establishment
  9. [9]
    30 years of ensemble forecasting at ECMWF
    Nov 24, 2022 · In December 1992 operational ensemble forecasts began at ECMWF. They began at the US National Centers for Environmental Prediction at around the ...
  10. [10]
    [PDF] The history of ECMWF forecasts and UEFs
    Operational forecasting on an experimental basis began on 1 August. 1979, with forecasts ten days ahead, five days per week, using an N48 grid model.
  11. [11]
    [PDF] Fifty years of data assimilation at ECMWF
    ECMWF was founded in 1975 with the primary mission to produce ten-day weather forecasts using state-of-the-art NWP systems. From the outset, the Centre ...
  12. [12]
    Integrated Forecasting System | ECMWF
    IFS cycle. Implementation date. News item ; Cycle 49r1, 12 November 2024, Forecast upgrade improves wind and temperature predictions ; Cycle 48r1, 27 Jun 2023.
  13. [13]
    ECMWF's AI forecasts become operational
    Feb 25, 2025 · ECMWF has taken the Artificial Intelligence Forecasting System (AIFS) into operations today, 25 February 2025, to run side by side with its ...
  14. [14]
    ECMWF's ensemble AI forecasts become operational
    Jul 1, 2025 · ECMWF has taken the ensemble version of the Artificial Intelligence Forecasting System (AIFS) into operations today, 1 July 2025.
  15. [15]
    Member States - ECMWF
    ECMWF has 23 Member States and 12 Co-operating States. ECMWF's annual budget is funded primarily by annual contributions from the Member and Co-operating ...
  16. [16]
    ECMWF becomes a multi-site organisation | Copernicus
    Sep 14, 2021 · ECMWF has officially become a multi-site organisation, holding opening ceremonies for offices in Bonn, Germany, where its Copernicus Services will be primarily ...Missing: institutional developments
  17. [17]
    [PDF] 14 September 2021: new ECMWF data centre opens in Bologna, Italy
    Sep 14, 2021 · The new ECMWF data center in Bologna, Italy, opened on September 14, 2021, housing a new Atos supercomputer, planned to be operational in 2022, ...
  18. [18]
    One year of operational service in Bologna | ECMWF
    On 18 October 2022, ECMWF successfully migrated all operational forecast services from its data centre in Reading (UK) to its new data centre in Bologna (Italy)
  19. [19]
    Automated weather data services deployed in Bologna | ECMWF
    In 2022, ECMWF migrated its 450+ PiB data archive (700 PiB when counting backup copies) from Reading, UK, to its new data centre in Bologna, Italy.Missing: history institutional
  20. [20]
    CloudFerro successfully relocates ECMWF cloud to Bologna
    Nov 22, 2022 · Cloud infrastructure for ECMWF's Climate and Atmosphere Data Store services have been successfully migrated from Reading, UK, to Bologna, Italy.Missing: history institutional<|separator|>
  21. [21]
    Council
    ### ECMWF Council Summary
  22. [22]
    Governance - ECMWF
    The governing bodies are the Council, the Director-General, and the Council's advisory committees, whose functions are defined in the Convention.Missing: processes | Show results with:processes
  23. [23]
    Florian Pappenberger - Director General (elect) & Deputy ... - LinkedIn
    Florian Pappenberger is the Director General (elect, starting 01.01.26) and the Deputy Director-General and Director of Forecasts and Services at ECMWF.
  24. [24]
    [PDF] CONVENTION ESTABLISHING THE EUROPEAN CENTRE FOR ...
    The ECMWF's primary purpose is to develop and provide medium-range weather forecasts, and its objectives include developing global models and data systems.Missing: making | Show results with:making
  25. [25]
    [PDF] ECMWF Strategy 2025-2034
    Feb 7, 2025 · I am grateful for all the support over almost a year of discussions, which have included all committees as well as the 35 Member and. Co- ...<|separator|>
  26. [26]
    Fact sheet: ECMWF and the EU's Copernicus programme
    Sep 11, 2023 · ECMWF runs two services that are part of the EU's Copernicus Earth observation programme. They are the Copernicus Atmosphere Monitoring Service (CAMS) and the ...
  27. [27]
    Environmental services and future vision | ECMWF
    We are implementing the Copernicus Atmosphere Monitoring Service (CAMS) and the Copernicus Climate Change Service (C3S). We also contribute to the Copernicus ...
  28. [28]
    The Copernicus Atmosphere Monitoring Service: From Research to ...
    Union's Earth observation program Copernicus, entered operations in July 2015. Implemented by the European Centre for Medium-Range Weather Forecasts (ECMWF) ...Cams Implementation · Global Products · Regional Products<|separator|>
  29. [29]
    New paper reviews the first ten years of Copernicus at ECMWF
    May 27, 2025 · ... numerical weather prediction. Systems utilising ERA5 demonstrate a 10-20% performance enhancement, potentially yielding annual benefits ...
  30. [30]
    Copernicus Climate Change Service - ECMWF
    The Copernicus Climate Change Service (C3S), implemented by ECMWF, provides climate information to inform policy and protect citizens from climate hazards. It ...
  31. [31]
    [PDF] Ten years of Copernicus at ECMWF
    Apr 1, 2025 · ECMWF implements Copernicus' CAMS and C3S, providing data for climate, air quality, and more. ECMWF's expertise and collaboration ensure high- ...
  32. [32]
    Copernicus Atmosphere Monitoring Service - ECMWF
    ECMWF implements the Copernicus Atmosphere Monitoring Service (CAMS) on behalf of the European Union. Why do we need to monitor the atmosphere?
  33. [33]
    European Centre for Medium-Range Weather Forecasts - GEO
    Furthermore, CAMS focuses on climate forcing and greenhouse gas surface flux inversions and is developing a European monitoring capacity for anthropogenic CO2 ...Missing: broader | Show results with:broader
  34. [34]
    ECMWF's societal impact through service provision, partnerships ...
    The European Centre for Medium-Range Weather Forecasts (ECMWF) mission is to deliver high-quality global medium-range numerical weather predictions and ...
  35. [35]
    National Collaboration Programmes strengthen engagement ...
    NCPs aim to enhance the operational use and integration of Copernicus data and services across EU Member States and other participating countries, including ...
  36. [36]
    Forecasts - ECMWF
    ECMWF is the European Centre for Medium-Range Weather Forecasts. We are both a research institute and a 24/7 operational service, producing global numerical ...Medium-range forecasts · Datasets · Charts · Seasonal forecasts
  37. [37]
    Modelling and Prediction | ECMWF
    All our forecasts and reanalyses use a numerical model to make a prediction. We have developed our own atmospheric model and data assimilation system.Forecast evaluation · Atmospheric physics · Marine · Atmospheric dynamics
  38. [38]
    Medium-range forecasts | ECMWF
    Our medium-range forecasts consist of a single forecast (HRES) and our ensemble (ENS) which together give detailed information about the evolution of weather ...
  39. [39]
    Quality of our forecasts | ECMWF
    We continually monitor the accuracy of our operational forecasts. This evaluation provides essential feedback to both users and model developers.
  40. [40]
    Forecast evaluation | ECMWF
    ECMWF maintains a comprehensive range of verification statistics to evaluate the accuracy of its forecasts.
  41. [41]
    Forecast performance 2023 - ECMWF
    ECMWF maintains a comprehensive range of verification statistics to evaluate the accuracy of its forecasts. Each year, a summary of verification results is ...
  42. [42]
    The ECMWF ensemble prediction system: Looking back (more than ...
    Aug 26, 2018 · And so, starting in November 1985, the Met Office quasi-operational probabilistic ensemble forecast system was born (Murphy and Palmer, 1986) ...<|control11|><|separator|>
  43. [43]
    [1803.06940] The ECMWF Ensemble Prediction System - arXiv
    Mar 16, 2018 · This paper has been written to mark 25 years of operational medium-range ensemble forecasting. The origins of the ECMWF Ensemble Prediction System are outlined.
  44. [44]
    Potential Forecast Skill of Ensemble Prediction and Spread and Skill ...
    After a month of trial, on 19 December 1992 ECMWF started running the EPS every Saturday, Sunday, and Monday. The singular vectors used to generate the ...
  45. [45]
    Section 5 Forecast Ensemble (ENS) - Rationale and Construction
    Nov 12, 2024 · Processing the ensemble of forecasts is computationally expensive. The medium range ensemble currently has 50 members and 9km resolution. The ...
  46. [46]
    A 50-member Ensemble of Data Assimilations | ECMWF
    Since 2010, ECMWF has run an Ensemble of Data Assimilations (EDA) to help determine the initial conditions for its ensemble forecasts and its higher-resolution ...Missing: innovations | Show results with:innovations
  47. [47]
    Section 12.B Statistical Concepts - Probabilistic Data
    Nov 12, 2024 · The most common verification method for probabilistic forecasts is the Brier score (BS) or the Mean Square Error of Probability Forecasts.Measurement Of Model... · Cost/loss Diagrams And... · Using Deterministic...
  48. [48]
    Probabilistic weather forecasting with machine learning - Nature
    Dec 4, 2024 · Here we introduce GenCast, a probabilistic weather model with greater skill and speed than the top operational medium-range weather forecast in the world, ENS.
  49. [49]
    Probabilistic Predictions of Precipitation Using the ECMWF ...
    Four seasons are analyzed in detail using signal detection theory and reliability diagrams to define objective measure of predictive skill.
  50. [50]
    Sub-seasonal-range forecasts - ECMWF
    Our sub-seasonal-range forecasts (previously known as extended-range and monthly forecasts) provide an overview of potential weather conditions up to 46 days ...
  51. [51]
    Seasonal forecasts - ECMWF
    They are produced every month with a 51-member ensemble at a horizontal resolution of around 36 km. ... Information for forecast users (forecast issues, severe ...
  52. [52]
    ECMWF's new long-range forecasting system SEAS5
    SEAS5 forecasts show substantial improvements in the tropics, in particular for sea-surface temperature in the equatorial Pacific.
  53. [53]
    Data assimilation | ECMWF
    Data assimilation is typically a sequential time-stepping procedure, in which a previous model forecast is compared with newly received observations, the model ...
  54. [54]
    [PDF] Overview of Data Assimilation Methods - ECMWF Events (Indico)
    The observations are used to correct errors in the short forecast from the previous analysis time (every 12 hours at ECMWF, 6-3 hourly in other global. NWP ...
  55. [55]
    Fact sheet: Earth system data assimilation | ECMWF
    Jul 15, 2025 · The method used at ECMWF for the atmosphere is called four-dimensional variational data assimilation (4D-Var). 4D-Var iteratively adjusts the ...
  56. [56]
    Observations | ECMWF
    ECMWF routinely processes data from around 90 satellite instruments as part of its operational daily data assimilation and monitoring activities.
  57. [57]
    Towards unified land data assimilation at ECMWF: Soil and snow ...
    Jul 15, 2024 · Weather centres use a variety of data assimilation schemes to analyze different land variables in their operational forecast systems.
  58. [58]
    Observation influence diagnostic of a data assimilation system
    Results show that, in the ECMWF operational system, 18% of the global influence is due to the assimilated observations, and the complementary 82% is the ...
  59. [59]
    The ERA5 global reanalysis - Royal Meteorological Society (RMetS)
    May 17, 2020 · This paper presents a comprehensive overview of the ERA5 global reanalysis from 1979 onwards, which is produced at ECMWF as part of the ...
  60. [60]
    Climate reanalysis | ECMWF
    The latest ECMWF reanalysis is ERA5, which is being produced by C3S. ERA5 provides a snapshot of the atmosphere, land surface and ocean waves for each hour from ...Missing: details | Show results with:details
  61. [61]
    ECMWF Reanalysis v5 (ERA5)
    ERA5 is the fifth generation ECMWF atmospheric reanalysis of the global climate covering the period from January 1940 to present.Temperature · Wind · Precipitation · Atmosphere
  62. [62]
    ERA5 atmospheric reanalysis - Climate Data Guide
    The ERA5 dataset provides hourly data on surface and upper-air parameters at roughly 31 km global resolution on 137 levels reaching up to 1 Pa.<|separator|>
  63. [63]
    Browse reanalysis datasets - ECMWF
    Browse reanalysis datasets ; ERA5-Land, 1950-present ; ERA-Interim, 1979-August 2019.
  64. [64]
    ERA5-Land | ECMWF
    ERA5-Land includes information about uncertainties for all variables at reduced spatial and temporal resolutions. The model used in the production of ERA5-Land ...
  65. [65]
    ERA5-Land: a state-of-the-art global reanalysis dataset for ... - ESSD
    Sep 7, 2021 · ERA5-Land describes the evolution of the water and energy cycles over land in a consistent manner over the production period.Assets · Article · Metrics · Related articles
  66. [66]
    ERA5 hourly data on single levels from 1940 to present
    Jun 14, 2018 · ERA5 is the fifth generation ECMWF reanalysis for the global climate and weather for the past 8 decades. Data is available from 1940 onwards.
  67. [67]
    About our forecasts | ECMWF
    Our Integrated Forecasting System (IFS). All our forecasts are produced with the ECMWF Integrated Forecasting System (IFS), which has a global numerical ...Integrated Forecasting System · Medium-range forecasts · Seasonal forecasts
  68. [68]
    Modernisation of the Integrated Forecasting System - ECMWF
    The Integrated Forecasting System (IFS) is mission-critical software for ECMWF. It fulfils our primary purposes of (a) developing a capability for medium-range ...Missing: mandate | Show results with:mandate
  69. [69]
    An update to ECMWF's machine-learned weather forecast model AIFS
    Sep 23, 2025 · The revised AIFS version delivers highly skilled forecasts, as shown by anomaly correlation scores for 2023 in the Northern Hemisphere (Figure 6) ...<|separator|>
  70. [70]
    ECMWF – Europe's leading centre for weather prediction makes ...
    Feb 24, 2025 · A newly operational model, known as the Artificial Intelligence Forecasting System (AIFS), has been launched by the European Centre for Medium-Range Weather ...
  71. [71]
    Supercomputer facility | ECMWF
    The main system from Atos is made up of four self-sufficient clusters, also called 'complexes'. Each cluster is connected to all the high-performance storage.Missing: details 2025
  72. [72]
    ECMWF high-performance computing facility boosts forecasts and ...
    Jul 24, 2023 · The new HPCF has enabled ECMWF to increase the resolution of its medium-range ensemble forecasts (ENS) from 18 km to 9 km in Cycle 48r1.
  73. [73]
    Reaching JUPITER: ECMWF celebrates the first European exascale ...
    Sep 9, 2025 · Europe's first exascale supercomputer, JUPITER, at Forschungszentrum Jülich, Germany, went into operation on 5 September 2025, ...Missing: infrastructure | Show results with:infrastructure
  74. [74]
    AIFS: a new ECMWF forecasting system
    ECMWF decided to implement a data-driven forecast model, the Artificial Intelligence/Integrated Forecasting System (AIFS), a homage to ECMWF's Integrated ...
  75. [75]
    AI Weather Quest • Overview - ECMWF
    The AI Weather Quest serves as a unique platform to benchmark state-of-the-art AI/ML-based sub-seasonal predictions. It focuses on operational forecasting using ...Evaluation System · About · Resources · LoginMissing: emerging | Show results with:emerging
  76. [76]
    Catalogue of ECMWF real-time products
    The Catalogue outlines the real-time meteorological and oceanographic products available from the ECMWF forecasting system. These products are published ...Atmospheric Model high... · Atmospheric model Ensemble... · ConfigureMissing: methods | Show results with:methods
  77. [77]
    Charts - ECMWF
    Our Integrated Forecasting System (IFS) provides forecasts and associated verification at different resolutions and for multiple time ranges.ENS Meteograms · Charts catalogue · Tropical Cyclones · Extra-tropical cyclonesMissing: numerical | Show results with:numerical
  78. [78]
    [PDF] User Guide to ECMWF forecast products
    Due to the limited number of EPS members, the verifying analysis will be outside the ensemble range. For a system with 50 members this will happen 2/. 51 (~4 ...
  79. [79]
    Datasets - ECMWF
    Archived data. These are the historical meteorological data available from the ECMWF archive.Catalogue of ECMWF Real · Browse reanalysis datasets · Search ECMWF datasets
  80. [80]
    Access to forecasts - ECMWF
    National meteorological services NMHS of ECMWF Member and Co-operating States and their authorised users are granted full access to ECMWF products and ...
  81. [81]
    Open data - ECMWF
    A subset of ECMWF real-time forecast data from the IFS and AIFS models are made available to the public free of charge.Data · AIFS forecast - AIFS Single · ECPDS and PREd
  82. [82]
    ECMWF real-time forecasts - Registry of Open Data on AWS
    Description. These products are a subset of the ECMWF real-time forecast data and are made available to the public free of charge.
  83. [83]
    ECMWF makes its entire Real-time Catalogue open to all
    Oct 1, 2025 · “Since 2019, together with the support of our 23 Member States and 12 Co-operating States, the team has worked very hard to transition our ...
  84. [84]
    In a landmark move, ECMWF is poised to announce a new phase in ...
    Sep 30, 2025 · Florence Rabier, ECMWF Director-General, comments: “Since 2019, together with the support of our 23 Member States and 12 Co-operating States, ...
  85. [85]
    Access to real-time open data | ECMWF
    ECMWF data is available via WIS, DCPC, EUMETCast, and GTS. Data formats include GRIB, BUFR, and graphical products, downloadable from ECMWF's DCPC FTP server.
  86. [86]
    Severe weather prediction | ECMWF
    We work to provide early warnings of weather events such as these – made a few days ahead of the event and based on medium-range forecasts.Missing: applications monitoring
  87. [87]
  88. [88]
    Early Warnings of Severe Convection Using the ECMWF Extreme ...
    ECMWF provides the ensemble-based extreme forecast index (EFI) and shift of tails (SOT) products to facilitate forecasting severe weather in the medium range.
  89. [89]
    Early warnings of extreme winds using the ECMWF Extreme ...
    Jul 10, 2012 · The Extreme Forecast Index (EFI) was developed at ECMWF as a tool to provide forecasters with general guidance on potential extreme events.
  90. [90]
    ECMWF develops flash flood forecast system
    ECMWF has developed a prototype flash flood forecast system based on the total precipitation Extreme Forecast Index (EFI) product.
  91. [91]
    An Assessment of the ECMWF Extreme Forecast Index for Water ...
    Early awareness of extreme precipitation can provide the time necessary to make adequate event prepa- rations. At the European Centre for Medium-Range ...
  92. [92]
    ECMWF weather and climate experts monitor early summer ...
    Jul 1, 2025 · Most of Europe has been forecasted to see temperatures well above average in the coming days according to ECMWF's medium range weather forecasts.Missing: broader | Show results with:broader<|separator|>
  93. [93]
    Forecast performance 2024 - ECMWF
    ECMWF maintains a comprehensive range of verification statistics to evaluate the accuracy of its forecasts. Each year, a summary of verification results is ...
  94. [94]
    Verification - ECMWF | Charts
    Verification of other centres (mean sea level pressure)​​ These diagrams compare recent ECMWF verification scores for Mean Sea Level Pressure with ...
  95. [95]
    ECMWF vs NOAA forecast data accuracy
    Dec 18, 2019 · You can compare verification scores using this application from ECMWF made from guidelines by the WMO. The example below fits nicely with ...
  96. [96]
  97. [97]
    [PDF] NCEP NOTES Performance of the New NCEP Global Ensemble ...
    is defined as a 500-hPa geopotential anomaly correlation of. 0.8. The forecast lead time for this threshold increases from. 8.4 to 8.7 days in the new GEFS.
  98. [98]
  99. [99]
    Scalability - ECMWF
    Efficiency gains in all parts of the forecasting system are required in order to make a goal such as a 5 km horizontal resolution for ECMWF's ensemble forecasts ...
  100. [100]
    On Some Limitations of Current Machine Learning Weather ...
    Jun 14, 2024 · The main conclusion is that these ML models are not able to properly reproduce sub-synoptic and mesoscale weather phenomena and lack the ...
  101. [101]
    New Strategy pushes limits of weather science | ECMWF
    The new ECMWF Strategy to 2030 foresees a pushing of the boundaries of weather science to achieve global ensemble forecasts at kilometre-scale resolution.
  102. [102]
    Understanding systematic error growth across timescales | ECMWF
    Operational seasonal forecasts (SEAS5) exhibit a cold sea-surface temperature (SST) bias and an easterly wind bias over the Equatorial Eastern Indian Ocean, ...
  103. [103]
    Precipitation Biases in the ECMWF Integrated Forecasting System
    Research by Zsótér et al. (2019b) has shown that the ECMWF land data assimilation system contributes to negative ERA5 runoff biases in the high northern ...
  104. [104]
    Elevation-driven biases in seasonal weather forecasts: Insights from ...
    Furthermore, the ECMWF-SEAS5 forecast model is characterized by significant elevation biases that can directly affect the altitudinal gradient of ...<|separator|>
  105. [105]
    [PDF] Biases in the ECMWF model
    In this paper, systematic errors in the ECMWF model are discussed for the short range. Temperature and moisture are discussed by comparing with the analysis, ...Missing: controversies | Show results with:controversies
  106. [106]
    IFS upgrade brings many improvements and unifies medium-range ...
    An upgrade of ECMWF's Integrated Forecasting System (IFS) was implemented on 27 June 2023. One highlight is a horizontal resolution increase of the medium- ...
  107. [107]
    Improved two-metre temperature forecasts in the 2024 upgrade
    At ECMWF, tests assimilating T2m in 4D‑Var gave large improvements in short-range T2m forecasts, but in some winter regions 850 hPa temperatures were degraded, ...New Vegetation Maps And Soil... · Snow Data Assimilation And... · Combined Impact
  108. [108]
    Improving the handling of model bias in data assimilation | ECMWF
    Apr 20, 2020 · Tests of the new WC-4DVar configuration have shown that analysis and first-guess temperature biases in the stratosphere are reduced by up to 50%.
  109. [109]
    Ocean wind wave model upgrade improves the forecasting system
    Jun 24, 2024 · The improvements concern the representation of air–sea momentum exchange under strong wind conditions, and the role of waves in heat and moisture exchange.
  110. [110]
    How to systematically develop an effective AI-based bias correction ...
    Apr 21, 2025 · This study introduces ReSA-ConvLSTM, an artificial intelligence (AI) framework for systematic bias correction in numerical weather prediction (NWP).
  111. [111]
    ESA-ECMWF Report on recent progress and research directions in ...
    Jun 21, 2022 · Challenges are the lack of trust in ML tools due to the inability to extrapolate into unseen weather regimes and limited generalisation. It ...<|separator|>
  112. [112]
    [PDF] ECMWF Financial Statements of Account 2024
    The main sources of revenue include contributions from member & co-operating states, externally funded revenue, and sales of forecasts and data. Total operating ...
  113. [113]
    ECMWF and European Commission renew contract for Copernicus ...
    ECMWF has signed an agreement with the European Commission to continue implementing the EU-funded Copernicus Climate Change Service (C3S) and Copernicus ...
  114. [114]
    West curbs Russia's access to weather data that some fear could aid ...
    Mar 22, 2022 · A major supplier of meteorological data from Western and other governments suspended cooperation with Russia on Tuesday, becoming the latest ...Missing: response | Show results with:response
  115. [115]
    Supporting the humanitarian effort in Ukraine - ECMWF
    Supporting the humanitarian effort in Ukraine · Main components of the Impact-Based Forecast bulletin · Weather forecasts to better plan humanitarian operations.
  116. [116]
    Closing data gaps improves global forecasts
    Jun 16, 2025 · Using data assimilation techniques, these simulated observations were fed into ECMWF's operational model. The results showed that the more ...