As a legacy of WMO's HIWeather programme, we also invite discussion of the interdisciplinary challenges, gaps, and opportunities in evaluating the warning value chain from observing, nowcasting and forecasting to warning and response. Understanding the true added value that each contribution brings to decision-making and community outcomes is critical.
Meanwhile, ongoing rapid developments in machine learning bring both opportunities and challenges for the warning process, and with the conference theme in mind contributions at this intersection point are also particularly welcome this year.
Topics may include:
• Nowcasting systems
• Links to severe weather and severe weather impacts
• Automated first guess warning systems
• Post-processing techniques
• Seamless deterministic and probabilistic forecast prediction
• Integrating systems and information within a forecast and warning value chain
• Use of machine learning and other advanced analytic techniques
• Can output of data-driven (AI) models contribute to warning systems?
In response to the critical global challenge of bridging the early warning capability gap—particularly for developing and least developed countries—this paper presents the Cloud-based Global Meteorological Early Warning Support Platform (C-EWS) developed by China Meteorological Administration (CMA). C-EWS embodies a novel operational paradigm designed to overcome the persistent scientific and logistical barriers that hinder effective early warning operations in resource-constrained settings.
C-EWS integrates a lightweight, cloud-based architecture to ensure universal accessibility. Through server-side rendering, asynchronous data streaming, and a unified API gateway, the platform shifts computational burden to the cloud, enabling smooth interactive exploration of multi-gigabyte datasets even over low-bandwidth connections. The platform consolidates multi-source observational data (including Fengyun satellite and WMO GTS observations), global NWP models (CMA-GFS, ECMWF-IFS, NCEP-GFS, ICON, JMA-GSM), and AI forecasts ('Fengqing', ECMWF-AIFS) into a unified, interactive web environment.
Unlike existing international platforms that offer either single-model expert products or basic multi-model visualization, C-EWS uniquely enables comprehensive multi-model diagnostic analysis, integrating both physics-based and AI-driven forecasts for enhanced uncertainty assessment. The platform embeds over 20 interactive analytical tools for multi-model comparison, forecast stability evaluation, and vertical profiling, enabling forecasters to rapidly analyze three-dimensional circulation patterns and assess model biases.
C-EWS delivers a suite of multi-model-based objective early warning products covering multiple hazard types—including tropical cyclones, sand and dust storms, extreme heat, heavy precipitation, gale winds, and floods. Verification against 2025 forecast products demonstrates substantial skill improvements: for heavy rainfall warnings, TS reached 0.226–0.158 for short-range forecasts—improvements of over 26% compared to NWP model forecasts. Operational efficiency has also been significantly enhanced: the time required to produce comprehensive Global Hazardous Weather Bulletins has been reduced from approximately 3.5 hours to 1 hour (a 71% improvement), while the volume of early warning products issued has increased by 136%. The platform also employ a co-development framework, enabling partner countries to co-create tailored solutions.
Since its operational deployment, C-EWS has supported forecasting for over 50 severe weather events across more than 30 countries and regions. Feedback from stakeholders confirms its practical value. The platform also supports multilateral mechanisms including WMO's Multi-Model Integrated Forecasting and Application (MMIFA) pilot project, the WMO Coordination Mechanism's (WCM) Hydrometeorological Weekly Scan, and the South Asia Hydromet Forum (SAHF) forecast consultations.
This work establishes a scalable and scientifically robust pathway to broaden access to advanced forecasting capabilities for developing nations, directly contributing to the United Nations' "Early Warnings for All" (EW4ALL) initiative and strengthening global climate resilience. Future developments will focus on expanding data source integration, operationalizing hybrid NWP+AI approaches for sub-seasonal to seasonal forecasting, and developing self-service configuration tools for more scalable customization.
How to cite: Yang, S.: China’s Global Meteorological Early Warning Support Platform: Development and Applications, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-43, https://doi.org/10.5194/ems2026-43, 2026.
Providing early information about severe weather events is crucial for authorities and the public to take adequate measures. These measures will depend on the type and intensity of predicted severe weather and require different lead times to be put into action. Since current numerical weather prediction at the German Meteorological Service (Deutscher Wetterdienst, DWD) reaches forecast horizons of up to seven days, warning information can theoretically be provided at equally long lead times. This possibility is leveraged by a prototype for automated warnings that is being developed within the DWD RainBoW program ("Risk-based, Application-oriented and INdividualizaBle Provision of Optimized Warning Information"). In addition to the longer leadtimes, it has the bonus of retaining a high and consistent update frequency.
Prototypical warnings are produced in parallel processing chains for each warning element. At present the prototype can generate warnings for wind gust, frost and rain based on meteorological data from in-house numerical weather prediction and nowcasting systems. Each processing chain is triggered as soon as new input is available and transforms said input into probabilistic warnings plus additional meteorological information for context. This includes wind direction and possible worst case wind speeds for wind gust warnings; possible worst case temperatures for frost warnings; and expected accumulated precipitation and length of a rain event for rain warnings.
A processing chain consists of a series of modules for the individual production steps. Steps that are necessary for all meteorological warning elements have been generalized, so that the modules can be reused across all processing chains. If additional production steps are required, further modules can easily be added. The modular approach proved very valuable because it simplified building tailored processing chains, as well as extending them to new functionalities. Moreover, it allows for central controlling and monitoring of processes.
The research prototype is currently running continuously for the warning elements wind gust, frost and rain, for which it proved to be stable and robust. In this contribution, the prototype's setup and results for the three warning elements will be presented.
How to cite: Felsberg, A., Königer, L., Hammelmann, J., Sauter, C., Baumgartner, M., and Klink, M.: Automated Warnings: Ongoing developments of a prototype for wind gust, frost and rain, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-224, https://doi.org/10.5194/ems2026-224, 2026.
At the German Meteorological Service (Deutscher Wetterdienst, DWD), thunderstorm warnings are currently issued on-detection. In a manual workflow, detected individual thunderstorms are used as a base to delineate warning regions. Depending on the synoptic situation, this can result in a significant workload that may lead to inconsistencies in the spatio-temporal granularity of delineated thunderstorm events. Automating the production of thunderstorm warnings can ameliorate this issue. Therefore, within the RainBoW program ("Risk-based, Application-oriented and INdividualizaBle Provision of Optimized Warning Information"), the automation of thunderstorm warnings is currently implemented using the in-house nowcasting model NowCastMIX1. It provides thunderstorm warning regions with lead times up to one hour, including information about the associated hazards (such as gusts and precipitation) and their derived warning level. Given the model’s update rate of five minutes, the system must process warning information at a high frequency. This high update rate can lead to undesirable fluctuations in the warning levels prior to an event, potentially confusing end users.
One approach to stabilize the warning level over multiple model updates is temporal smoothing. In this work, we present a prototype for temporal smoothing of automated thunderstorm warnings based on NowCastMix. The smoothing is applied individually to the warning level of each associated hazard on an event basis. The overall thunderstorm warning level is then derived as the maximum warning level among all associated hazards. To further enhance stability, warning durations are binned to prevent rapid changes in the start and end times. The smoothing is applied to gridded data which is subsequently converted to polygons to facilitate mapping onto administrative regions.
To evaluate the impact of the smoothing procedure, we employ an object-based verification approach. Various performance scores are used to determine optimal smoothing parameters for our purposes. These results provide a starting point for further evaluations using administrative regions as a base. Preliminary results suggest that this methodology reduces warning fluctuations while maintaining the spatial and temporal accuracy for effective warnings from a nowcasting system.
1Paul James et al. DOI:10.1175/WAF-D-18-0038.1
How to cite: Hammelmann, J., Brune, S., Baumgartner, M., Klink, M., and Feige, K.: Closing the gap between nowcasting and warning: A prototype showcase, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-327, https://doi.org/10.5194/ems2026-327, 2026.
Issuing warnings for severe weather events is a core duty of national weather services to help protect society and critical infrastructure. At MeteoSwiss, forecasters carry the responsibility to issue such warnings for large-area rain events. To support them in this task, they are provided with a range of targeted automatized severe weather products.
In this contribution, we focus on a newly available rain warning proposal product that we operationalized at the end of 2025. Our algorithm summarizes noisy severe weather information from ensemble forecasts into smooth large-scale rain warning proposals and delivers them to our forecasters once a day in the morning, as a basis for their assessment of the warning situation. The algorithm was developed in close collaboration with forecasters to ensure that the resulting warning proposals align with their operational needs. Each warning proposal consists of an estimate of the warning level, the affected regions, the affected time span, and the expected precipitation amount. The forecasters can choose to accept, modify, or reject the proposals. If they accept or modify a proposal, they need to add some additional text information, regarding e.g., potential impacts and behavior recommendations, to the proposal before issuing it as an official warning.
Here, we will share our lessons learned from the operationalization process and provide an overview of the forecasters’ feedback we gathered and its consequences for our continued algorithm developments. Additionally, we will highlight emerging potential application areas of our warning proposal algorithm beyond its primary use, ranging from warning verification to numerical weather prediction (NWP) model development and climate change assessments.
How to cite: Beusch, L., Knirsch, L., Mühlhofer, E., and Attinger, R.: Insights from operationalizing a rain warning proposal algorithm for forecasters at MeteoSwiss, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-415, https://doi.org/10.5194/ems2026-415, 2026.
The growth and advances in Numerical Weather Prediction (NWP), post-processed output and new forecasting tools over the last 20 or 30 years has led to a greater need for dialogue between operations and research. This has led to the birth and subsequent development of more formal Operations-to-Research/Research-to-Operations collaboration within the broader meteorological community.
The Met Office is no exception and since the late 2010s/early 2020s a formal role has existed whereby a Principal Operational Meteorologist leads O2R on behalf of the operational community and, in more recent years, also leading a small, highly-motivated team of OpMets to support this work.
In this presentation, processes and methodologies (co-)led by O2R and by which the operational community at the Met Office feeds back to research areas are outlined. These processes are intended to promote two-way flow of dialogue and provide operational perspectives, use cases and customer benefits of NWP and post-processed output, in particular, and help inform the future direction and development of these outputs.
Among the processes highlighted are the value of a log – the Daily Forecast Assessment (DFA) – completed by operational meteorologists when on-duty to capture their perceptions of model performance. Feedback submitted to the DFA informs known model characteristics and helps prioritise work to improve Met Office NWP.
Meanwhile, another particular focus of this presentation concerns the role of testbeds and intensive evaluations, where scientists, developers, operational meteorologists and other users work together collaboratively over a period of few weeks on a series of forecasting and subjective verification exercises. This is a proven concept for accelerating the development of new tools and forecasting techniques. Here, an intensive evaluation run during autumn 2025 investigating aspects of the Met Office’s latest model upgrade, which went live in January 2026, is drawn upon to illustrate how operational meteorologists assessed new science to help improve the forecasting of extreme and impactful weather and, by extension, consider the impact of this new science on the Met Office’s public weather warning service, the National Severe Weather Warnings Service (NSWWS).
How to cite: Suri, D. and Nation, R.: Operations-to-Research (O2R) at the Met Office – A Gateway for Operational Meteorologists to Inform Met Office Science and Help Improve Forecasting of Extreme and Impactful Weather, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-423, https://doi.org/10.5194/ems2026-423, 2026.
Impact-based warning systems increasingly aim to translate meteorological forecasts into sector-specific risk information. However, a critical gap persists for the energy sector: the weather situations that cause the most severe grid stress — compound renewable generation failures, supply-demand mismatches, and cascading events — often do not exceed conventional meteorological warning thresholds. A Dunkelflaute or a storm-to-icing cascade may be meteorologically unremarkable yet operationally devastating. This contribution presents an automated, end-to-end framework that extends the forecast and warning value chain to energy system impacts, demonstrating how multi-hazard detection, ensemble-based probabilistic forecasting, and causal modeling can generate first-guess impact warnings for a sector not yet systematically served by warning services.
The framework implements automated hazard detection for (currently) five energy-relevant weather hazards — wind speed ramping, storm gusts, precipitation and icing, heatwaves, and cold spells — using configurable threshold-based methods with multi-level severity classification. For the critical forecast-to-warning step, an ensemble-based wind power ramping detection system processes ECMWF IFS 51-member ensemble forecasts through timing-based clustering that preserves extreme events typically destroyed by spatial averaging. This yields five (adjustable to more) probabilistic scenarios with severity classifications from −4 to +4, explicitly accounting for modern turbine storm control regimes (25–35 m/s). A complementary solar ramping system uses a three-component approach decoupling atmospheric convective threat from photovoltaic vulnerability, eliminating nighttime false alarms. A compound event module identifies co-occurring and cascading hazard sequences that amplify grid stress beyond what any single-hazard warning would indicate.
Weather pattern classification provides the seamless bridge between synoptic-scale NWP guidance and local impact probability, enabling regime-conditioned automated first-guess warnings. We explore the application of the Cause-Trigger framework (Hlaváčková-Schindler et al., 2025) to distinguish atmospheric processes that cause energy system hazards from those that merely trigger them — with direct implications for achievable warning lead times. The framework combines physics-informed machine learning with transparent threshold-based detection, addressing the session's question of whether data-driven model output can meaningfully contribute to warning systems.
Input data span from ERA5 and ARA high-resolution reanalysis through operational IFS and AIFS forecasts to Destination Earth Digital Twins, enabling seamless analysis across climate, seasonal, and operational timescales. We present results for the Austrian Alpine domain, where complex terrain amplifies forecast uncertainty, and show how reanalysis-derived hazard climatologies serve as the reference baseline for evaluating forecast-based impact warnings — closing the loop from observing through forecasting to warning and response within the energy sector.
How to cite: Schicker, I., Lexer, A., and Lehner, S.: Automated impact-based warnings for energy system stress: Integrating multi-hazard detection, ensemble probabilistics, and causal analysis into the forecast and warning value chain, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-616, https://doi.org/10.5194/ems2026-616, 2026.
With the program RainBoW (Risk-based, Application-oriented and INdividualizaBle Provision of Optimized Warning Information), the German Meteorological Service (Deutscher Wetterdienst, DWD) is currently renewing the national weather warning system. RainBoW aims to increase the practical value of weather warnings for a wide range of users by improving their comprehensibility, extending their forecast horizon, and enabling individualization for specialized users with needs beyond those of the general public.
The new warning system will be deployed in stages, with the first release targeting warnings for the general public. Key changes include a standardization to three consistent warning levels across all weather warning elements, and a more temporally differentiated approach to thunderstorm warnings. The latter introduces a thunderstorm potential warning that bridges the gap between an uncertain pre-information and an acute warning triggered by already-observed convective activity. Finally, warning texts are restructured to increase their communicative effectiveness.
Translating the initial theoretical concepts into practice presents several challenges, as many design decisions do not have obvious solutions and require trade-offs. For example, it is necessary to balance precision and the usability of warning information by recipients – particularly when formulating the contents of warning texts, but also in specifying rules for proactively sending warning notifications in a way that avoids alert fatigue. Further challenges lie in the definition of rules for handling concurrently active weather warnings and keeping consistency with other (non-meteorological) national warning systems, which may use other warning level definitions.
This talk provides an overview of the ongoing developments within RainBoW and discusses selected implementation dilemmas encountered while moving from conceptual ideas into reality.
How to cite: Feige, K. and the RainBoW Team: From concept to reality: Implementing a new weather warning system at the German Meteorological Service, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-672, https://doi.org/10.5194/ems2026-672, 2026.
The capacity of overhead transmission lines is fundamentally governed by atmospheric conditions, primarily wind speed, ambient temperature, and solar radiation, that determine the rate of conductor cooling and heating. Dynamic Line Rating (DLR) exploits this weather dependency to safely increase or decrease the permissible current beyond conservative static ratings. However, translating raw numerical weather prediction output into actionable operational warnings for transmission system operators (TSOs) requires a carefully designed warning value chain that bridges the gap between meteorological forecasts and grid management decisions.
We present a multi-scale automated detection and warning system developed within the Destination Earth (DestinE) Extremes Digital Twin initiative, co-designed with three European TSOs/DSOs. The system implements a cascading forecast-to-warning chain across three scales:
(1) Continental-scale screening using the DestinE Global Digital Twin (ECMWF IFS ensemble forecasts at ~9 km) identifies regions where compound meteorological events, specifically the co-occurrence of high temperature (>30°C), low wind speed (<2 m/s), and high solar radiation (>600 W/m²), pose a risk to transmission capacity. A five-level severity classification provides an automated first-guess warning product across the European domain.
(2) Regional refinement using AROME/ALARO forecasts (~2.5 km) extracts detailed atmospheric profiles along specific transmission corridors at multiple heights (10 to 100 m above ground) and lateral grid points, capturing the spatial heterogeneity critical in Alpine terrain where sheltered valley sections may experience unfavorable conditions while exposed ridge crossings remain favorable.
(3) On-demand hectometric detail (100 to 500 m) from the DestinE Extremes Digital Twin is triggered only when the continental screening identifies compound events, providing sub-kilometer resolution for bottleneck identification along critical line segments.
At each scale, the meteorological detection is coupled with CIGRE and IEEE 738 thermal models to translate weather conditions into quantitative ampacity forecasts (maximum permissible current in Amperes) and identify the specific line segment that limits the entire corridor's capacity, i.e. the operational bottleneck.
Ensemble-based probability statements are derived through k-means clustering of IFS ensemble members, producing operationally meaningful scenarios (e.g., "64% probability of mixed DLR conditions with a 200 A capacity reduction at the bottleneck segment") rather than simple mean/spread summaries. The compound event detection specifically addresses the interdisciplinary gap between meteorological severity and grid impact: high temperatures combined with calm winds can be more operationally critical than either extreme in isolation.
The system has been validated against archived AROME forecasts for Austrian Alpine transmission lines and verified against TSO operational data. We discuss the challenges of evaluating the true added value of each element in this warning chain, from the continental early warning that triggers expensive high-resolution runs to the segment-level ampacity forecast that informs real-time dispatch decisions, and the role of user co-design in ensuring that meteorological expertise translates into improved grid operation outcomes.
How to cite: Schicker, I., Deckmyn, A., Termonia, P., van den Bergh, J., van Oyen, T., and Le Maire, V.: From Weather Forecast to Grid Decision: A Multi-Scale Compound Event Detection and Warning System for Dynamic Line Rating of Overhead Transmission Lines, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-764, https://doi.org/10.5194/ems2026-764, 2026.
Reliable feature identification, long time series of identified features, and tools to explore them provide substantial benefits for weather nowcasting—including warning generation—medium-range forecasting, process understanding, climate information provision, and the evaluation of climate model outputs. Moreover, expert use of these features within an established feedback loop enables the creation of high-quality training datasets for further application development and machine learning (ML) model training.
With the advent of new methods enabled by cloud services and machine learning, the hydro-meteorological community has launched numerous projects to identify meteorological features from remote-sensing data, including satellite imagery. EUMETSAT and its Member States are building a collaborative environment for joint manual annotation, model development, and the Earth System Feature Database within the European Weather Cloud (EWC). The EWC is a cloud-based collaboration platform for meteorological application development and operations in Europe and to enable the digital transformation of the European Meteorological Infrastructure. It consists of data-proximate cloud infrastructure, alongside with the EWC Community Hub which enables collaborative development, sharing of code and ML models and the exploitation of meteorological applications.
EUMETSAT also plans to compile a database of long time series of meteorological features identified from various satellite datasets. This database will support the analysis of the development and interrelationships of these features, enabling new insights for all timescales from nowcasting to climate and downstream models such as impact predictions. Initial work has begun with a feasibility study for additional feature types.
This presentation will introduce the collaborative working environment for feature identification, how to take part in the collaboration, and provide example use cases. It will also present early results from the feasibility study, demonstrating the potential for performing feature identification on long time series of Earth-observation data.
How to cite: Tervo, R., Biermann, L., Karatosun, A., Huckle, R., and Hogervorst, F.: Identifying Earth System Features from Satellite Data for Nowcasting, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-647, https://doi.org/10.5194/ems2026-647, 2026.
The ThundeR rawinsonde processing package is a free, open-source R package designed for the visualization of atmospheric soundings and hodographs, as well as for the rapid calculation of convective parameters which are widely used in both research and operational forecasting of severe convective storms.
The package can calculate over 300 parameters in realtime (~1/100 s), enabling efficient processing of large numerical datasets. In recent years, ThundeR has been applied to global reanalysis datasets, operational numerical weather prediction models, and studies of environments associated with lightning observations and severe weather reports across Europe, North America, South America, and Australia.
ThundeR has also contributed to the development of ESSL’s AR-CHaMo models and has been utilized by severe storm researchers in multiple countries, including national hydrometeorological services. The construction of environmental datasets collocated with severe storm observations from various regions has provided a robust platform for evaluating the predictive skill of hundreds of convective parameters. This framework has also supported the iterative development and testing of new parameter concepts.
In this work, we present modifications to the calculation procedures of several established parameters, leading to improved identification of environments conducive to lightning, large hail, tornadoes, and severe winds, including significant severe events. These updates include refinements to the computation of convective inhibition, lifted index, and storm-relative helicity, as well as the introduction of a new ventilation parameter and two novel parcel-lifting methods: most-unstable mean-layer (MUML) and most-unstable above 500 m (MU5).
Finally, we demonstrate how these advancements can be applied in operational forecasting of severe convective storms and in modeling their climatology on a global scale.
How to cite: Szuster, P., Taszarek, M., Nixon, C., Pucik, T., Groenmeijer, P., Battaglioli, F., and Czernecki, B.: Improvements of convective parameters calculation algorithms in thundeR, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-295, https://doi.org/10.5194/ems2026-295, 2026.
Thunderstorms pose serious risks through lightning, heavy rainfall, hail, and strong winds. These events develop rapidly over highly localized areas, making timely short-term forecasts essential. Deep learning has proven particularly effective for thunderstorm nowcasting by rapidly learning spatiotemporal patterns from diverse observational datasets, enabling precise multi-hazard predictions within seconds - well-suited to operational early warning systems.
COALITION-4 is a deep learning nowcasting algorithm based on an encoder-forecaster architecture with recurrent convolutional layers. It nowcasts thunderstorm-related hazards - accumulated precipitation, lightning occurrence, and hail probability - up to 60 minutes ahead at 1 km and 5 minute resolution over the full domain of Switzerland. The operational model ingests data from the Swiss dual-polarization radar network, Météorage lightning observations, and a digital elevation model. The integration of NWP forecasts and satellite imagery has also been tested at the development stage, but is not used in the first version of operational implementation. Extensive preprocessing, data augmentation, and GPU-accelerated training with an adaptive learning rate ensure robust generalization across diverse convective conditions.
We present a comprehensive validation covering six convective seasons, enabling a statistically robust assessment of model skill across a wide range of convective regimes. The evaluation framework has been extended beyond standard grid-based metrics to explicitly quantify the effective lead time of warnings prior to thunderstorm onset - a metric of direct relevance to user preparedness and protective action. Results are compared both quantitatively and qualitatively against the previous operational nowcasting system at MeteoSwiss.
Since its operational deployment, robustness and reliability have been substantially improved through continuous quality monitoring, automated fallback mechanisms for degraded or missing input data, and iterative refinements informed by forecaster feedback. Building on this foundation, an automated warning pipeline is being developed that will deliver push notifications directly to the general public via the MeteoSwiss mobile app, alongside existing support for civil protection agencies, fire brigades, and aviation ground operations. This service is planned to become operational in summer 2026, representing a key step in closing the warning value chain from nowcast to societal response.
Looking ahead, we describe a significant advancement in the ML training procedure: a substantially expanded training dataset combined with a revised sample selection strategy that better represents high-impact convective events. This update yields a marked and consistent improvement in nowcasting skill across all hazard types and lead times, and points toward further enhancements in the accuracy and reliability of automated thunderstorm warnings.
How to cite: Hamann, U., Buzzi, M., Pacey, G., Miralles, O., Rombeek, N., Tarin Burriel, N., van Thor, J., Grochal, P., Kavachevich, D., Nasirifard, P., Juda, P., Germann, U., and Leinonen, J.: COALITION-4: Probabilistic Multi-Hazard Thunderstorm Nowcasting with Deep Learning for Robust Automated Early Warnings, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-302, https://doi.org/10.5194/ems2026-302, 2026.
The Meteorological Service of Canada (MSC) is actively modernizing how it produces weather forecasts and warnings to better serve Canadians. This transformation is driven by the necessity to adapt to evolving client needs, integrate the rapid pace of scientific and technological innovation, including advances in AI and machine learning, and harness the phenomenal growth in the volume of environmental information now available. Output from high-resolution, ensemble, and environmental prediction systems is increasingly accessible, and significant work is underway to bring far more of this information into MSC's routine forecast production process.
A key pillar of this modernization is the Weather Elements on Grid (WEonG) system — a new approach to producing the weather and environmental elements that underpin MSC's forecast programs, including Public, Marine, and Air Quality forecasting. WEonG move away from the legacy point-based approach, making forecast elements available across a full grid and enabling forecasts to be generated at any location in Canada. This improved offering and packaging of post-processed model data is central to MSC's broader service transformation, including efforts to increase automation where appropriate in the forecast production process.
This presentation provides an overview of WEonG and the methods used to produce it, including how the system leverages the full suite of Canadian numerical weather and environmental prediction (NWEP) models and advanced post-processing techniques. By enabling the flexible integration of detailed information, including uncertainty, WEonG makes it easier to tailor products to a wide range of user needs, ultimately delivering better, more timely information to Canadians.
How to cite: Gagnon, S., Gheti, R., and Barszcz, A.: Transforming Forecast Production at MSC: The Weather Elements on Grid System, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-827, https://doi.org/10.5194/ems2026-827, 2026.
MOSMIX is a point-based, global forecasting system providing predictions up to 14 days ahead for a wide range of meteorological variables. It features a seamless transition from observations to forecasts by downscaling and combining output from the global NWP models IFS and ICON over the previous three days. Among forecasters at Deutscher Wetterdienst (DWD), MOSMIX has earned a strong reputation for its high quality, particularly for variables such as wind and temperature, and is therefore widely used in operational forecasting and warning processes.
Despite being operational for decades and having reached a mature state, MOSMIX is based on increasingly outdated technology. The system relies primarily on multiple linear regression, with nonlinearity represented through a large number of situation-dependent equations. Specifically, separate equations are defined for each model run, location, variable, lead time, and season. While these millions of equations ensure high forecast quality, they also necessitate additional measures to maintain consistency across variables and lead times. As a result, the overall complexity of MOSMIX has become a bottleneck for further development.
Moreover, MOSMIX produces high-quality forecasts only at locations where observations are available. Although interpolation is possible for sites near observation stations, the system cannot generate fully gridded forecast fields.
To address these limitations, DWD is currently developing a new system intended to replace MOSMIX. This presents a significant challenge, as the new system is expected to match or exceed the performance of the current operational system while retaining its key features. At the same time, it should provide gridded forecasts, at least over the European domain.
In this presentation, we introduce a prototype of a hybrid grid- and point-based system that aims to achieve MOSMIX-level performance at station locations while simultaneously producing gridded forecasts. The approach extends the station embedding methodology of Rasp and Lerch (2018) and Schulz and Lerch (2022), enabling the neural network to be applied consistently on a spatial grid. Model performance is evaluated using out-of-sample verification, in which the stations used for validation are excluded from the training process. We also address the issue of consistency across variables.
Preliminary results for wind gusts indicate that, while it remains challenging to outperform MOSMIX in terms of standard metrics such as RMSE, the new approach can achieve comparable performance.
Literature:
Rasp, S., and S. Lerch, 2018: Neural Networks for Postprocessing Ensemble Weather Forecasts. Mon. Wea. Rev., 146, 3885–3900, https://doi.org/10.1175/MWR-D-18-0187.1.
Schulz, B., and S. Lerch, 2022: Machine Learning Methods for Postprocessing Ensemble Forecasts of Wind Gusts: A Systematic Comparison. Mon. Wea. Rev., 150, 235–257, https://doi.org/10.1175/MWR-D-21-0150.1.
How to cite: Schröder, G., Brune, S., and Trepte, S.: Modernization of the seamless point-based forecasting system MOSMIX using AI, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-326, https://doi.org/10.5194/ems2026-326, 2026.
The Korea Meteorological Administration (KMA) has been operating GraphCast, Pangu-Weather, and FourCastNet based on initial conditions from the Korean Integrated Model (KIM), the Unified Model (UM), and the ECMWF model since 2024.
A comparative evaluation between AI-based and NWP-based forecasts indicates that, when the same initial conditions are used, GraphCast and Pangu-Weather generally outperform traditional NWP models. While AI forecasts initialized with ECMWF analyses outperformed the ECMWF NWP model, AI forecasts initialized with KIM analyses do not surpass the performance of the ECMWF NWP system, highlighting the critical role of initial condition quality. These results demonstrate that, despite rapid advancements in AI models, forecast skill remains strongly dependent on the accuracy of the driving initial fields. Therefore, improving the quality of initial conditions through continued advancement of NWP systems is essential for maximizing the performance of AI-based forecasts.
In addition, the superior performance of AI forecasts initialized with ECMWF analyses may not be solely attributable to the higher quality of the initial conditions. It is also likely influenced by the fact that AI models have been trained on datasets generated by the ECMWF model, implying that consistency between the training data and the initial conditions plays a significant role in enhancing forecast skill.
Several limitations of AI models have become evident through years of operational use. In particular, during summer heavy rainfall events, AI models tend to underestimate precipitation intensity, as has been widely demonstrated in numerous previous studies. To enhance the operational applicability of AI-based forecasts, it is necessary to develop post-processing and bias-correction techniques that mitigate such underestimation. A synergistic approach combining AI model development, NWP improvement, and targeted bias correction is expected to further advance forecast accuracy and reliability in operational settings.
Keywords: AI Weather Forecasting, GraphCast, Pangu-Weather, Initial Conditions, KIM, Precipitation Underestimation, bias correction
How to cite: Shin, H., Choi, W.-J., and Park, B.-K.: Operational Application and Performance Evaluation of AI-based Weather Prediction Models: Impacts of Initial Conditions and Heavy Rainfall Biases, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-346, https://doi.org/10.5194/ems2026-346, 2026.
IMPROVER (Integrated Model Post-Processing and Verification) has been developed by the Met Office as an open-source probability-based post-processing system to fully exploit our convection permitting, hourly cycling ensemble forecasts. This toolbox of post-processing steps is used to produce the Met Office’s operational forecasts for the public.
Screen temperature is a key diagnostic for public weather and one that high resolution numerical weather prediction models are good at forecasting. However, forecasts on a km-scale grid are still unable to represent the variation of temperature across specific sites, for example those in valleys unresolved by the model orography. The production of site-specific forecasts therefore requires the addition of sub-grid detail which may be added through physical corrections and / or using statistical methods.
In operational Met Office site-specific temperature forecasts IMPROVER is used to apply a temperature lapse rate adjustment to account for site displacement from the grid cell average altitude. In addition, the forecasts are calibrated using Ensemble Model Output Statistics (EMOS) to remove model biases and adjust the spread of our probabilistic forecasts. EMOS coefficients are calculated by pooling all sites together, allowing the application of calibration to both observed and unobserved sites. The resulting forecasts are more accurate than the unadjusted gridded forecasts at these locations, but there remains scope for improvement, particularly at more challenging sites.
In this talk I will present the results of applying a whole host of calibration techniques available within IMPROVER to screen temperatures across the UK area. These include EMOS with differing configurations, Standardised Anomaly Model Output Statistics (SAMOS), Quantile regression Random Forests (QRF), and Reliability Calibration. Further machine learning approaches will also be discussed. A constraint on all techniques is that they must be applicable to all forecast sites, not just those returning observations. The relative performance, as well as the pros and cons of the different methods will be discussed, as will the future of our approach to site-specific screen temperature forecasts.
How to cite: Ayliffe, B., Evans, G., Hooper, B., and Grant, K.: Finding more skill in UK temperature forecasts, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-466, https://doi.org/10.5194/ems2026-466, 2026.
Users of gridded weather forecasts, whether derived from ensemble or deterministic sources, often require realistic weather scenarios or “realizations” that represent the range of possible forecast outcomes. This need is especially acute for hydrological applications, where precipitation realizations are essential inputs to hydrological models that subsequently generate deterministic or ensemble river flow forecasts. However, weather forecasts are available from many sources, spanning a range of spatial scales, resolutions, and lead times—from high‑resolution, short‑range forecasts covering small domains to coarse, long‑range global forecasts. Combining forecasts across timescales whilst minimising the appearance of undesirable artefacts is a continued challenge.
The IMPROVER project addresses this by converting each forecast source into exceedance probabilities and blending them into a seamless multi‑model probabilistic forecast. This blending step produces smooth, coherent probability forecasts across the full range of lead times and serves as the foundation for generating multi‑model blended realization forecasts.
The generation of these realizations follows a multi‑stage Ensemble Copula Coupling (ECC) procedure designed to preserve spatial and temporal structure, where possible. A subset of raw ensemble members is first selected using clustering to act as templates. These templates are then enhanced through temporal interpolation and the injection of stochastic noise to make them suitable for generating precipitation fields. Probability forecasts are sampled to reflect the spatial characteristics of the raw ensemble members, and an additional step ensures that intensity peaks in the generated realizations remain consistent with those present in the underlying members.
This study presents the implementation of this blended‑realization approach and its application to driving a national‑scale hydrological model covering England and Wales. We discuss the performance of the generated realizations and highlight the benefits of this probabilistic, multi‑model strategy for operational hydrological forecasting.
How to cite: Evans, G., White, M., Wright, B., and Beaver, J.: Implementation of realization generation from a multi-model probabilistic blended forecast, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-482, https://doi.org/10.5194/ems2026-482, 2026.
IMPROVER (Integrated Model Post-Processing and Verification) has been developed by the Met Office as an open-source probability-based post-processing system to fully exploit our convection permitting, hourly cycling ensemble forecasts. Post-processed MOGREPS-UK model forecasts are blended with deterministic UKV model forecasts and data from the coarser resolution global ensemble, MOGREPS-G as well as ECMWF, to produce seamless probabilistic forecasts from now out to 14 days. For precipitation, an extrapolation nowcast is also blended in at the start. Forecasts are converted to probabilities at the start, and all initial stages of post-processing are performed on gridded data, with site-specific forecasts extracted as a final step, helping to ensure consistency. Data are processed on a 10km global grid and on a 2km UK-centred grid. Physical and statistical corrections are applied to the data to ensure the probability distribution functions for each source model are sufficiently similar for blending into a seamless probabilistic forecast.
In order to blend data from different model sources, it is necessary that the data represent the same quantity. The most common inconsistency is altitude - different model configurations have different representations of the orography of mountains and valleys which can lead to very large differences in altitude-dependent diagnostics such as temperature, wind and lying snow, which is particularly relevant for driving hydrological models.
In this talk we present a novel method for adjusting lying snow data from one orography representation to another and demonstrate the effectiveness of this in creating data with similar characteristics from ensemble numerical models with different horizontal resolution so that they can be blended together into a single probabilistic data set.
How to cite: Moseley, S.: A method for regridding lying snow for blending of multiple forecast sources in IMPROVER , EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-541, https://doi.org/10.5194/ems2026-541, 2026.
Delivering seamless weather forecasts has long been a central objective, reflecting both user expectations and sustained development efforts. At present, multiple specialized forecasting systems are operated across different lead times and subsequently combined to produce integrated products and services. Because this integration is often tailored to specific applications, inconsistencies may arise between forecast products, complicating their interpretation and potentially reducing their overall utility. In response, MeteoSwiss has initiated a project to develop a seamless forecasting system designed to support a wide range of applications, including public weather forecasts, impact-based warnings, and climate services. Part of the project explores the possibility to use data-driven forecasts as recent advances in Machine Learning-based forecasting methods offer new opportunities to achieve genuine seamlessness directly within a unified forecasting framework. The envisioned system is designed to deliver ensemble forecasts with lead times of up to ten days while enabling high-frequency updates on the order of ten minutes to address nowcasting requirements. Beyond system development, the project also aims to transition this new approach into operations and progressively replace components of the current baseline systems.
This contribution summarizes progress achieved during the first two years of the project. Development of the ML-based forecast system has taken place within the Anemoi framework, in close collaboration with broader European initiatives in data-driven prediction. As an initial demonstrator, a deterministic forecast with a five-day horizon, 1 km spatial resolution, and hourly temporal resolution has been implemented, running every six hours. The system employs a stretched-grid configuration and has been trained using ERA5 data, a 20-year regional ICON reanalysis dataset, and several months of operational ICON analyses. Evaluation shows that, in terms of deterministic skill scores, the demonstrator equals or outperforms the operational ICON limited-area model. Current developments focus on enabling more frequent forecast updates and incorporating observational data during inference to support nowcasting, representing the next key milestone toward operational deployment.
We conclude by summarizing the main insights gained to date—spanning scientific and methodological advances as well as technical, operational, and organizational experience—and by outlining the principal challenges that remain on the path toward full operational implementation.
How to cite: Spirig, C. and the SeamlessWeather Team: Seamless Weather Forecasting through Machine Learning, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-587, https://doi.org/10.5194/ems2026-587, 2026.
Met.3D (documentation including installation instructions available at https://met3d.readthedocs.org) is an open-source research software with the goal to make interactive, 3-D, feature-based, and ensemble visualization techniques accessible to the meteorological community. At the EMS 2025, we introduced Met.3D version 2.0, featuring, for example, a new user interface and a batch mode. This year, we report on the use of Met.3D during the international North Atlantic Waveguide and Downstream Impact Campaign (NAWDIC, https://www.nawdic.kit.edu/). NAWDIC took place in January and February 2026 with the aim of providing detailed observations of mid-latitude atmospheric dynamics and involved, amongst other aircraft and ground stations, the German HALO research aircraft.
As part of the campaign’s forecasting team, we deployed Met.3D in batch mode (for pre-defined 3-D forecast products) and in an interactive remote visualization setup to support weather forecasting and flight planning for HALO. Visualization servers by the German Climate Computing Centre (DKRZ) were remotely accessible from the campaign site in Shannon, Ireland, and used to analyze open data ICON forecasts from the German Weather Service (DWD). Interactive 3-D visualizations, including vertical cross-sections, Lagrangian particle trajectories, and feature-based representations of atmospheric structures, enabled the detailed analysis of air mass origins and transport pathways throughout the troposphere. These capabilities proved valuable for identifying scientifically relevant flight paths under various constraints.
We present selected case studies from NAWDIC to illustrate how interactive 3-D visualization complements traditional 2-D analysis methods and enhances situational awareness in complex meteorological scenarios. Furthermore, we highlight how the Met.3D batch and interactive setups are further developed for use, e.g., in future campaigns.
How to cite: Rautenhaus, M., Fischer, C., Vogt, T., and Dorff, H.: Use of Met.3D for interactive 3-D visual analysis for forecasting during the NAWDIC field campaign, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-694, https://doi.org/10.5194/ems2026-694, 2026.
The presentation will provide an overview on the status and future plans for DWD's new Seamless Integrated Forecasting system (SINFONY) for very-short-range high resolution forecasts, which consists of several components:
1) The radar Nowcasting ensembles STEPS-DWD and convective cell objects ensemble KONRAD3D-EPS including hail and life cycle information.
2) The ICON-RUC-EPS with hourly LETKF assimilation of 3D radar volumes, Meteosat VIS and IR channels, hourly new forecasts using the Seifert-Beheng 2-moment microphysics scheme with prognostic hail and advanced forward operators for radar (EMVORADO) and VIS/IR satellites RTTOV-MFASIS).
3) Optimal combinations ("blending") of the Nowcasting and NWP ensemble forecasts in observation space, which constitute the seamless forecasts. The areal INTENSE precipitation- and reflectivity ensemble is targeted towards hydrologic warnings. It also serves as input to quickly estimate a pluvial flash-flood potential for very small river catchments on short leadtimes (AREA). The combined Nowcasting- and NWP cell object ensemble KONRAD3D-SINFONY helps evolve DWD’s warning process for convective hazards.
4) Common Nowcasting and NWP verification systems for precipitation, reflectivity and cell objects help to continuously improve the SINFONY components.
After several years of intensive development in a project framework, "fitness for operations" of the components ICON-RUC, STEPS-DWD and INTENSE has been achieved, especially for small-scale strong convective events from observation time up to 12h. Weather radar is at the heart of it, having DWD's warning process (forecasters, automated systems) and the hydrological authorities in mind. The ICON-RUC-EPS already served 1 1/2 years in operations and received very positive feedback from it's users. INTENSE is close to operationalisation and KONRAD3D-SINFONY is strongly evolving. Their long-term maintenance and further development is a cross-cutting activity of several disciplines and sections within DWD.
In parallel, the systems will be further improved in a new project phase SINFONY-3.0 (2025-2029) towards seamless forecasts for other parameters/phenomena and other user groups:
- Temperature, wind, cloudiness, fog, solar radiation, visibility, ceiling for
- aeronautical forecasts, renewables, customer data portals,
- with good year-round performance,
- and seamless forecasts beyond 12h.
For this, we try to improve the ICON-RUC forecasts for these parameters, and, by using AI-methods, we blend ICON-RUC-EPS into ICON-D2-EPS and we integrate MTG satellite data into our Nowcasting and combined products.
How to cite: Blahak, U. and the Team SINFONY & Friends: Quo vadis SINFONY - the seamless combination of Nowcasting and ICON Rapid Update Cylce NWP at DWD, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-759, https://doi.org/10.5194/ems2026-759, 2026.
Severe weather events can arise from fundamentally different atmospheric processes, posing distinct challenges for forecasting and warning systems. This study compares two high-impact events in Zagreb: a large-scale windstorm in March 2026 and a convective storm in July 2023, both of which caused significant damage in an urban area despite their different origins. In both cases, the damage in Zagreb was primarily caused by wind. In March 2026, this occurred in combination with heavy rainfall, although wind speeds did not reach record values. In contrast, the July 2023 event featured a prolonged episode of extreme wind gusts, reaching record values for the Zagreb area, where such intensities are uncommon.
The March 2026 event was associated with a well-developed synoptic-scale system, characterized by strong pressure gradients and widespread wind fields. Such conditions were captured by global and regional numerical weather prediction (NWP) models, with ALADIN-HR providing a reliable baseline for forecasting precipitation and wind intensity as well as spatial extent several days in advance. In contrast, the July 2023 event was driven by deep convection, featuring localized but extremely intense wind gusts and rapid storm evolution, which proved significantly more difficult to predict using standard NWP approaches. For the convective case, short-term forecasting tools and observationally driven systems, such as INCA, can play a crucial role in capturing storm development and evolution at relevant spatial and temporal scales. Additional value for both can be provided by post-processing approaches, including machine-learning-based methods, that help refine local-scale impact estimates.
By analysing these two contrasting cases, the study highlights how the usefulness of forecasting tools depends strongly on the dominant atmospheric processes. The results demonstrate the importance of a multi-scale forecasting framework that combines information from global and regional models, nowcasting systems, and post-processing methods, while acknowledging the inherent limitations of each approach.
How to cite: Odak, I., Kuzmic, J., Stanesic, A., Vujec, I., and Lozuk, J.: Multi-Scale Forecasting of Severe Wind Events: Insights from Two Contrasting Cases in Zagreb, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-787, https://doi.org/10.5194/ems2026-787, 2026.
Weather-related disruptions in aviation are inevitable, but their operational impact often depends on how effectively forecast information is translated into decisions. In January 2026, MeteoSwiss launched operationally adMET (Aerodrome Meteorological Forecast Information Service), an ML-enhanced aviation forecasting system designed to strengthen the entire forecast-to-alerting-to-decision chain within airport and air traffic control operations.
The system integrates data from high-resolution ensemble Numerical Weather Prediction (NWP) and real-time observations with machine-learning-based predictions to generate calibrated probabilistic guidance from the immediate nowcasting range (0-2 h) up to 30 hours ahead. Increased temporal resolution in the short range and statistical refinement of ensemble output allow the system to provide rapidly updated impact indicators for low visibility, ceiling constraints, wind limitations, and convective activity. Rather than focusing on raw meteorological variables, adMET delivers threshold-based, operationally interpretable information that directly links forecast uncertainty to capacity and safety implications.
The transition from development to daily operations revealed that the critical challenge lies not only in forecast skill, but in embedding probabilistic information within established workflows. Air traffic controllers, apron coordinators, and dispatchers operate within tightly constrained decision timelines. For probabilistic forecasts to add value, they must be trusted, intuitively visualized, and clearly connected to operational consequences. Extensive user engagement, iterative interface adjustments, and targeted training were therefore essential components of the implementation process.
Early operational feedback indicates tangible benefits across the aviation decision chain. Enhanced short-term calibration supports earlier recognition of potential bottlenecks, smoother coordination among stakeholders, and more consistent management of weather-induced capacity reductions. At the same time, the system maintains human oversight, positioning ML-driven post-processing as an automated, yet transparent, first-guess layer within a broader decision-support framework.
This contribution discusses the interdisciplinary lessons learned from embedding machine-learning-based probabilistic forecasting into an operational decision environment. It highlights how the seamless integration of deterministic and probabilistic prediction, combined with user-centered communication strategies, can strengthen the link between advanced forecasting systems and real-world response. This demonstrates how data-driven models can meaningfully contribute to impact-oriented warning processes in high-stakes aviation contexts.
How to cite: Landmann, J. M., Attinger, R., Aznar Siguan, G., Barras, H., Hamann, U., Reiniger, T., Wehrli, K., Exterde, S., Jordi, T., and Stocker, C.: adMET: Integrating Probabilistic Weather Information into Aviation Decision-Making , EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-108, https://doi.org/10.5194/ems2026-108, 2026.
In the context of aviation planning over Europe, a two-day forecast of the convective activity is required. Such a product, named the Cross-Border Convection Forecast (CBCF), is regularly produced during the convective season by several meteorological services across Europe and coordinated by EUMETNET. Using a collaborative editing platform, human forecasters generate forecast polygons to quantify the expected convective situation. In particular, the forecast includes the expected degree of organization of the convective cells and the expected likelihood of occurrence.
The goal of this work is to provide an automatic guidance product based on AI methods to assist the forecasters. Moreover, the guidance product should directly provide the final CBCF-forecast category, i.e. degree of organization and its likelihood of occurrence. We report on the development of such a guidance product, where the basic forecast data are taken from the ICON-EU ensemble forecasts that are operationally produced at the German Meteorological Service and cover the whole desired European forecast domain. Based on this ensemble data, several derived statistical quantities are computed and serve as input to the AI method.
One significant challenge in the development of the guidance product is the definition of the target for training of the AI model since direct observations are not available for the final forecast categories. As a result, another quantity needs to be used as target for the training. In our approach we employ the area fraction of lightning, being a derived product of lightning observations and defined as the area covered by lightning within a radius of 50km. We show preliminary results for the AI methods that use the area fraction of lightning as target. Moreover, we discuss possible modifications of this quantity to adapt it better to the desired final product.
How to cite: Baumgartner, M., Schröder, G., and Primo, C.: AI-aided Convection Forecast for Aviation Planning, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-442, https://doi.org/10.5194/ems2026-442, 2026.
Safe airport operations depend on accurate wind measurements to assess the wind conditions above the runway and along the flight paths of aircraft. At Amsterdam Schiphol Airport, wind speed and direction are measured using cup anemometers and wind vanes installed at multiple locations for each runway. The placement of these instruments involves a trade-off between proximity to the runway, distance from nearby obstructions, site availability, and the need to share measurements across runways, all while complying with ICAO regulations. Previous computational fluid dynamics (CFD) simulations have highlighted potential disturbances affecting these measurements.
Wind conditions can vary spatially due to terrain and nearby structures, making the precise location of cup anemometers and wind vanes crucial for accurately representing runway conditions. Examining the spatial distribution of wind can validate instrument placement and reveal any localized deviations from expected conditions.
To support this, the Royal Netherlands Meteorological Institute deployed a Leonardo Skiron3D Doppler wind lidar at Schiphol Airport. The lidar will be positioned at four locations across the airport to cover all eight in-situ wind observation sites. This study introduces a method for validating the representativeness of in-situ wind measurements specifically for the Buitenveldertbaan (09/27) runway.
A very low elevation angle of 0.5° is selected to capture near-surface wind observations across the entire runway. Since the Doppler wind lidar measures only the radial wind component along its line of sight, direct comparison with in-situ measurements is not feasible. To address this, the in-situ data are decomposed and spatially mapped to estimate the radial wind component assuming a uniform wind field. To account for variations in the altitude of the lidar beam, the decomposed wind field is adjusted using a standard power-law wind gradient with a surface roughness exponent of 1/7.
Results indicate a very small difference between the decomposed and height-adjusted in-situ measurements and the lidar observations. Additionally, lidar data reveal that the terminal building influences wind conditions above the runway during southerly winds, although these effects do not impact the in-situ measurements.
How to cite: Mathijssen, T.: Assessment of runway wind conditions at Amsterdam Airport Schiphol using a Doppler wind lidar, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-611, https://doi.org/10.5194/ems2026-611, 2026.
Modern numerical weather prediction systems produce unprecedented volumes of data up to hundreds of terabytes every day. While advances in computing capabilities have improved overall forecast skill, they also introduce technical challenges for forecasters. Transforming large amounts of complex ensemble data into useful information for rapidly evolving weather situations is difficult because forecasters have limited time to analyse all available data. Furthermore, raw model output is also difficult to interpret for the public, who often rely on a single forecast provided by their local weather provider. The European Center for Medium-Range Weather Forecasts (ECMWF) produces data from the physics-based Integrated Forecasting System (IFS), but recently they have also started to offer data-driven products from the Artificial Intelligence/Integrated Forecasting System (AIFS). ECMWF already offers products that track and quantify extratropical cyclones (ETC) through an ETC database (CDB) on their website, but for now this does not yet include AIFS data and mainly focuses on greater Europe and the North Atlantic region.
This project presents real-time ensemble forecast products designed to identify, track and quantify approaching ETCs affecting northern Europe using both IFS and AIFS data. The products are co-designed with operational forecasters at the Finnish Meteorological Institute (FMI) to ensure they meet the criteria for integration into operational use. Automated cyclone tracking software TRACK is applied to each forecast member to identify and quantify the impact-relevant metrics of approaching systems. Storms are assessed according to metrics ranging from traditional dynamical measures like maximum vorticity and minimum mean sea level pressure to impact-relevant metrics such as wind footprint and storm severity index, which offer more value for the user.
Ensemble data are condensed and presented as forecast products delivered on a publicly accessible website which updates in real-time. The forecasts are compared against climatology and previous high-impact storms, which places approaching systems into a historical context and helps inform how unusual and potentially impactful they are. The provided ensemble-based forecast products have not previously been implemented for ETCs in northern Europe. The project offers a novel method for monitoring and communicating the likelihood and severity of ETCs.
How to cite: Strömberg, J., Bouvier, C., Cornér, J., Sinclair, V., and Solin, K.: R2O STORMS: Real-Time Ensemble Forecast Products For Extratropical Cyclones In Northern Europe, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-152, https://doi.org/10.5194/ems2026-152, 2026.
The Korea Meteorological Administration (KMA) has been operating the Korean Integrated Model (KIM), an indigenous global numerical weather prediction system, since April 2020. In May 2025, KMA began the operational service of a high-resolution KIM, which features improved horizontal resolution to enhance overall forecast performance and prediction accuracy. Despite these technical advancements, KIM revealed continuous systematic temperature biases in the high latitudes of the Northern Hemisphere (≥70°N), particularly over sea-ice areas.
In this study, we evaluated the systematic errors of KIM by comparing its forecasted fields against ERA5 reanalysis data. We focused on the diagnostic analysis of surface skin temperature (Tsfc), 2m temperature (T2m) and lower-tropospheric air temperature to identify the specific characteristics and seasonal trends of these biases. Preliminary results showed that KIM underestimated surface temperatures over sea-ice areas throughout most of the year, although a transition to overestimation was observed during the spring period from mid-April to May. These biases suggest that the initialization and physical processes for surface conditions in the polar areas within KIM require further research and development.
To diagnose the sources of these errors, we analyzed the relationships between temperature biases and radiation/cloud-related variables using time-series analyses. The results indicated that KIM generally underestimated surface downward and net longwave radiation, while overestimating surface net shortwave radiation. In addition, cloud fraction was overestimated, whereas total cloud condensate was generally underestimated and the fraction of cloud ice was relatively large. This study aims to examine how these discrepancies in hydrometeors and radiation variables contribute to the observed temperature biases, especially over sea-ice areas. The findings from this evaluation will provide a basis for improving physical parameterizations and surface-related processes in KIM to enhance the predictive skill of the high-resolution KIM in high-latitude areas.
How to cite: Park, K.-H., Lim, J.-O., Lee, S., and Lee, Y.-H.: Evaluation of high-latitude temperature biases over sea-ice areas in the Korean Integrated Model (KIM), EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-425, https://doi.org/10.5194/ems2026-425, 2026.
Typhoon-induced storm surges pose the most severe marine disaster threat to the densely populated and economically vital Guangdong-Hong Kong-Macao Greater Bay Area (GBA). Accurate and timely forecasting of these events remains a significant operational challenge. To address this, we developed and validated an advanced, real-time storm surge prediction system for the GBA—the Greater Bay Area Storm Surge Prediction System (GBASSP). This operational system features a tightly coupled framework, integrating the Global/Regional Assimilation and Prediction System (GRAPES) atmospheric model with the Finite-Volume Coastal Ocean Model (FVCOM). The GBASSP achieves an exceptionally high horizontal resolution of up to 80 meters in critical coastal zones, enabling detailed simulation of complex coastlines and estuary dynamics. The performance of GBASSP was rigorously verified against observational data. Key findings demonstrate its robust forecasting capability: (i) The system provides reliable early warning and surge forecasts with a lead time of at least two days prior to typhoon landfall. (ii) For 24-hour forecasts during typhoon events, the model exhibits high accuracy, with a maximum storm surge error as low as 5 cm for specific cases and a mean absolute error for maximum surge heights of 19.7 cm across evaluated events. Furthermore, the timing error for the predicted peak surge is consistently within one hour of observations. (iii) In a comprehensive comparative analysis with other established storm surge prediction models, GBASSP shows superior skill. It achieves the smallest relative error (5.9%) and root mean square error (21 cm) among all models compared. Its average absolute error also falls within the range of the best-performing benchmarks. In conclusion, the GBASSP establishes itself as a high-precision, operational tool for real-time storm surge forecasting in the GBA. Its coupled atmosphere-ocean design, very high resolution, and demonstrated accuracy in predicting both surge magnitude and timing make it a valuable asset for disaster prevention, mitigation, and emergency response, ultimately contributing to enhanced coastal resilience against typhoon threats.
How to cite: Zhou, M., Liu, C., Dai, G., Huang, H., Song, Q., and Li, M.: Real-Time Prediction of Typhoon-Induced Storm Surge in the Greater Bay Area: A Model Setup and Validation Study, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-28, https://doi.org/10.5194/ems2026-28, 2026.
The Basque Country currently operates a regional coastal flooding Early Warning System (EWS) designed for exposed open-coast environments, with a primary focus on urban coastal areas backed by beaches and coastal protection structures. This system serves as a short-term coastal management tool to forecast potential storm hazards and assist coastal authorities in deploying appropriate mitigation solutions. However, the Basque Coast also contains several confined coastal environments, such as enclosed bays, where hydrodynamic behaviour differs markedly from that of the open coast, requiring tailored approaches to forecast accurately the level of hazard.
Pasaia Bay, hosts several urban centres (Pasai San Juan, Pasai San Pedro and Pasai Antxo) located within a confined estuarine port environment. Pasai San Juan, in particular, experiences recurrent passive flooding events, which are expected to increase due to sea level rise, requiring the development and implementation of a specific Early Warning System to anticipate the hazard.
The first step of this work was to review the existing historical data and relate it to the prevailing hydrodynamic conditions. Next, the work proceeded with the design, calibration, and validation of a coastal flooding Early Warning System specifically developed for the Pasaia Bay. The system is based on the Total Water Level indicator that combines: (i) sea level (astronomical tide and meteorological contribution), and (ii) port agitation or longwave contribution. Using water level time series collected during local measurement campaigns within the bay, functional relationships are established between offshore wave forcing and the bay’s internal hydrodynamic response. This enables the anticipation of the Total Water Level indicator in the critical zone (Pasai San Juan) and the definition of operational activation thresholds.
The system is designed to provide alerts to the Pasai San Juan City Council more than 72 hours in advance, supporting preventive management, the protection of urban areas, and informed decision-making during flooding events. This development represents an initial step towards extending early warning capabilities tailored to confined coastal environments to other localities in the Basque Country.
How to cite: Abalia, A., Gaztelumendi, S., Egaña, J., Epelde, I., Liria, P., Del Campo, A., Gonzalez, M., de Santiago, I., Nieto, A., and Garnier, R.: Towards an Operational coastal flooding Early Warning System in a confined bay: The case of Pasaia Bay , EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-220, https://doi.org/10.5194/ems2026-220, 2026.
As of 1 January 2025, the Czech Republic has enacted a new Act on the Public Hydrometeorological Service, which, among other provisions, formally defines the Integrated Warning Service System (IWSS) and introduces the single‑voice principle for all warnings issued within this framework. The detailed structure of the warning system is specified in the implementing decree issued by the Ministry of the Environment of the Czech Republic. The new IWSS became fully operational on 1 July 2026.
The redesigned warning system, jointly operated by the Czech Hydrometeorological Institute (CHMI) and the Hydrometeorological Service of the Armed Forces of the Czech Republic, incorporates key elements of impact‑based warning approaches. Until now, the issuance of warnings relied on fixed mandatory criteria expressed as specific thresholds—such as precipitation totals over defined intervals, air‑temperature limits, probabilities of severe thunderstorms with a given hail diameter, or wind‑gust values. The decision‑making process was purely hazard‑oriented and based on a warning matrix evaluating the probability of occurrence and the forecast intensity of the hazardous phenomenon.
In the new warning system, threshold criteria serve only as indicative guidance, while the emphasis shifts from hazard to risk. The previous warning matrix has been replaced with a new one that jointly assesses the probability of occurrence (hazard level) and the expected impacts of the phenomenon (risk level).
Another major innovation in the CHMI warning system is the introduction of human thermal discomfort, represented by the newly established heat stress and cold stress warning groups. Further details on this embedded Heat Health Warning System are presented by the authors in a poster contribution in session OSA 2.4.
Acknowledgement: This work was supported by the Institutional Support for the Long-Term Conceptual Development of Research Organisation CHMI (DKRVO) 2023-2027 by the Ministry of the Environment of the Czech Republic.
How to cite: Novák, M., Došková, L., Hrubý, J., and Hynčica, M.: A new version of the Integrated Warning Service System in the Czech Republic, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-529, https://doi.org/10.5194/ems2026-529, 2026.
Title: IMPROVER tools for physically consistent realizations
Authors: Max White, Jasmine Beaver, Gavin Evans
IMPROVER (Integrated Model Post-Processing and Verification) is an open-source, Python-based toolbox developed by the UK Met Office for post-processing ensemble weather forecasts. It provides a wide range of capabilities, including physical and statistical corrections, model blending, regridding, thresholding, and ancillary data generation. To date, much of the development has focused on producing skilful probabilistic forecast products. However, many applications – such as driving hydrological and impact models – also require spatially and temporally coherent forecast scenarios, often referred to as physically consistent realisations.
To address this need, we have implemented a suite of tools within IMPROVER for generating physically realistic realisations from post-processed, multi-model ensemble forecasts across different spatial domains, resolutions, and lead times. These methods aim to retain the calibrated statistical properties of probabilistic forecasts while producing coherent forecast evolutions suitable for downstream applications.
We will present the open-source, reusable Python functionality added to the IMPROVER toolbox to support this capability, including:
- Generation of additional realisations: expansion of a dataset’s realisation dimension to increase ensemble size while maintaining physical consistency.
- Stochastic noise generation: the addition of spatially correlated noise using a short-space Fourier transform to break grid point ties (for example, zero precipitation occurring in multiple ensemble members). This avoids physically unrealistic, spatially uncorrelated artifacts.
- Realisation clustering and matching: k-medoids clustering of realisations (ensemble members) from a primary forecast source (for example, a lower-resolution global model used at longer lead times), followed by matching to realisations from secondary forecast sources such as nowcasts or higher resolution ensembles. This defines a trajectory for each realization across the lead time range. Jumps between different forecast sources can be smoothed using temporal interpolation.
- ECC-Q and ECC-T mapping: calibration using Ensemble Copula Coupling, either by sampling at evenly spaced quantiles (ECC-Q) or using a parametric gamma distribution (ECC-T), allowing the spatial characteristics of the raw ensemble members to be better represented after sampling.
- Period disaggregation and temporal interpolation utilities: tools for disaggregating accumulated diagnostics (for example, from 3-hourly to 1-hourly) and temporally interpolating forecast fields using linear methods or machine-learning-based approaches such as Google FILM.
- Deterministic realisation selection: extraction of a deterministic realisation from a clustered ensemble, providing a physically plausible single forecast trajectory.
How to cite: White, M., Beaver, J., and Evans, G.: IMPROVER tools for physically consistent realizations, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-396, https://doi.org/10.5194/ems2026-396, 2026.
Weather warnings issued by national meteorological services are commonly derived through manual interpretation of numerical weather prediction output. While this approach allows for expert judgement, it limits update frequency and lead time and introduces variability between forecasters. Within the RainBoW program ("Risk-based, Application-oriented and INdividualizaBle Provision of Optimized Warning Information"), the German Meteorological Service (Deutscher Wetterdienst, DWD) is developing a prototype system that derives warning-relevant weather events from ensemble forecast data. This contribution presents the implementation and refinement of this prototype, with focus on frost and rain. It demonstrates how automated event detection can be used to supplement analysis that have traditionally been performed manually.
The prototype system integrates ensemble data of multiple configurations of the ICON numerical weather prediction model as input, to maximise the provided forecast lead time and accuracy. The different setups are ICON-D2 Rapid Update Cycle (RUC), ICON-D2, ICON-EU, and ICON with forecast lead times ranging from 14 hours up to seven days. These forecast datasets are supplemented by additional information, such as radar-based precipitation data. Event detection is performed per warning element, but similar across various input data sources. A rule-based approach is used to detect events in each ensemble member individually. Detected events are subsequently aggregated across ensemble members and model configurations in order to derive consistent event signals that are suitable for warning generation. This approach enables the combination of information from multiple models while also utilizing the ensemble character of the input data.
A key achievement of this work is the development of a processing chain that automatically generates updated event information whenever new model data become available. This enables a substantially higher update frequency than workflows based on human forecasters, while maintaining longer lead times. Currently, it operates as a research prototype and is not yet part of the operational warning workflow at DWD. Selected case studies demonstrate the detection of frost and rain events from ensemble forecasts. The results are compared with warning information derived from observational datasets to investigate the behaviour and interpretability of the prototype.
Future work will focus on comprehensive statistical verification, further refinement of the methodology, and the extension of the system to additional warning elements.
How to cite: Königer, L., Felsberg, A., Baumgartner, M., and Klink, M.: From Ensembles to Alerts: Deriving Event-Based Information from Forecasts, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-438, https://doi.org/10.5194/ems2026-438, 2026.
Nowadays weather forecasts are available from a wide range of sources. Combining these weather forecasts into a form that is digestible by operational meteorologists and other users is key for decision-making. IMPROVER takes the approach of converting the individual forecast sources into exceedance probabilities and combining these exceedance probabilities to create a multi-model blended probabilistic forecast.
At the Met Office, the operational IMPROVER implementation blends a deterministic nowcast, a deterministic UK domain model (UKV), an ensemble UK domain model (MOGREPS-UK), an ensemble global domain model (MOGREPS-G) with ensemble forecasts from ECMWF’s IFS to create a seamless probabilistic forecast out to 14 days. However, recent advances in AI weather models present new opportunities to further enhance performance. This study evaluates the added value of incorporating ECMWF’s AIFS‑CRPS ensemble AI model into the IMPROVER blending framework. We assess the impact on forecast skill across spatial scales, lead times, and exceedance thresholds, and analyse patterns of improvement to understand the mechanisms underpinning any added benefit from AI weather forecasts.
We also investigate how multi-model blending interacts with forecast calibration. Calibration methods can improve both mean bias and ensemble spread, however the net effect of these calibration approaches when combined with multi-model blending needs assessment. By comparing calibration applied before and after blending, we explore how best to maximise future forecast quality.
This work provides an early assessment of how AI-based ensemble models can be integrated into an operational probabilistic forecasting system, informing the Met Office’s strategy for augmenting traditional NWP with emerging AI weather prediction capabilities.
How to cite: Sop, N. H., Evans, G., Siegert, S., Ferro, C., and Kwasniok, F.: Value of including AI weather models within a multi-model probabilistic blend, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-671, https://doi.org/10.5194/ems2026-671, 2026.
The Korean Peninsula is characterized by a diverse range of precipitation types, including hail, heavy rainfall, sleet, snow, and freezing rain. These conditions necessitate accurate Hydrometeor Classification (HC) to ensure public safety and effective infrastructure management. The Weather Radar Center (WRC) of the Korea Meteorological Administration (KMA) operates a dual-polarization radar network and provides real-time HC products. In this study we introduce an advanced HC techniques optimized for the meteorological characteristics of South Korea based on dual-polarization radar.
The HC algorithm is based on fuzzy logic, integrating dual-polarization variables with numerical model temperature data. To ensure the reliability of the dual-polarization variables, we corrected the attenuation in reflectivity (ZH) and differential reflectivity (ZDR) caused by beam blockage and radome attenuation. Furthermore, as the correlation coefficient (ρhv) tends to decrease with radar range, it was adjusted in low Signal-to-Noise Ratio (SNR) regions. Bilateral filtering was also employed to suppress observational noise. Based on these quality-controlled inputs, the fuzzy logic algorithm utilizes membership functions (MBFs) and weights for each radar variable and temperature to determine the most probable hydrometeor type for each radar bin. The MBFs were localized and optimized using drop size distribution data from a 2-Dimensional Video Disdrometer (2DVD). To mitigate discontinuities in classified hydrometeors caused by model-derived temperature, the melting layer height derived from dual-polarization variables was incorporated, enabling more precise discrimination between liquid and solid phases. Additionally, spatial continuity was enhanced through a mode filter technique.
The improved HC algorithm was evaluated using ground-based hail observations from weather stations and more accurately distinguished hail from heavy rain. Furthermore, in temperature advection cases, the incorporation of the melting layer height allowed for more consistent classification of both liquid and solid phases. These results suggest that the proposed method improves the relibality of HC and has the potential to enhance real-time weather monitoring and hazard mitigation.
How to cite: Choi, H.-J., Kwon, S., and Suk, M.-K.: Optimization of Fuzzy Logic-based Hydrometeor Classification using 2DVD Observations over the Korean Peninsula, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-131, https://doi.org/10.5194/ems2026-131, 2026.
Increasing localized heavy rainfall due to climate change necessitates proactive severe weather alerts. Since June 2022, the Korea Meteorological Administration (KMA) has provided a weather radar-based real-time severe weather alert service via a mobile application for public safety. Targeting 3,510 administrative districts nationwide, the service operates based on user location and consists of four alert types: Rainfall Start, Heavy Rain, Lightning, and Hail.
Issuance conditions are subdivided according to radar data characteristics. Rainfall Start alerts use the MOTION (Motion vector estimation and extrapolation of radar echo for Integrated Operation Nowcasting) model, triggered when predicted 1-hour intensity is ≥0.5 mm/h. Lightning alerts are provided for predictions ≥0.01 kA within one hour based on MAPLE motion vectors. Heavy Rain alerts utilize HSR (Hybrid Surface Rainfall) data, issued when 15-minute accumulation exceeds 15 mm and instantaneous intensity exceeds 50 mm/h. Hail alerts integrate storm cell thickness, vertically integrated liquid (VIL), and temperature.
To reduce user fatigue and enhance effectiveness, operational standards have been improved. Winter alerts are suspended to control notifications from snow flurries, and the Rainfall Start threshold was raised from 0.1 to 0.5 mm/h to prevent false alarms from mid-level precipitation. Despite these efforts, issues remained with missed alerts due to long re-sending restriction intervals or redundant notifications during continuous precipitation.
To resolve these problems, radar precipitation data (Sep 2022 – Aug 2025) were analyzed. Results identified a median system duration of 4 hours and confirmed that a 1-hour non-precipitation duration is an appropriate factor for distinguishing independent systems. Accordingly, the re-sending restriction interval was optimized from 6 to 4 hours, and a condition to prevent duplicate alerts within one hour was added. Re-analysis confirmed a significant reduction in unnecessary alerts. This optimization enhances service reliability and contributes to mitigating meteorological disasters.
Keywords: Weather Radar, Severe Weather Alert Service, Public Safety
Acknowledgements: This research was supported by the "Development of convergence and application technology for ground-based remote sensing (KMA2026-00222)" of "Development of Severe Weather Nowcasting and Convergence Technology for National Radar" project funded by the Weather Radar Center, Korea Meteorological Administration.
How to cite: Kim, M.-G., Kim, K.-H., Kim, Y.-H., and Nam, K.-Y.: Development for a Radar-Based Severe Weather Alert Service for Public Safety Using the KMA Mobile Application, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-422, https://doi.org/10.5194/ems2026-422, 2026.
The increasing frequency of localized convective precipitation events demands higher precision in short-range forecasting to support early warning and public safety. The Korea Meteorological Administration (KMA) utilizes the MOtion vector estimation and extrapolation of radar echo for Integrated Operational Nowcasting (MOTION) system as its primary tool. While MOTION provides a stable operational baseline through kinematic extrapolation, its performance is fundamentally limited by a linear formulation that cannot fully represent the rapid, nonlinear intensity changes characteristic of convective systems.
This study enhances the MOTION framework—which conventionally integrates multi-resolution motion vectors, variational optimization, and Semi-Lagrangian extrapolation—by replacing numerical Growth and Decay Rate (GDR) estimation with a data-driven AI approach. We rigorously compared three architectures: Multi-Layer Perceptron (MLP), Deep Generative Model of Rain (DGMR), and RainNet. Among these candidates, the RainNet model consistently achieved the highest accuracy across all validation metrics. To ensure computational efficiency, SHapley Additive exPlanations (SHAP) analysis was applied to evaluate feature importance, revealing that the GDR from the previous time step was the dominant input variable. This result enabled a significant reduction in computational load while maintaining high forecast performance.
The proposed framework was validated over consecutive convective seasons. Quantitative results from June to September 2024 showed an 11% reduction in Mean Absolute Percentage Error (MAPE) compared to the original numerical method for lead times up to 120 minutes. Subsequent updates in 2025, featuring higher-resolution data and model optimization, led to an additional 10% improvement in MAPE over the initial AI version. These cumulative improvements demonstrate the effectiveness of the AI-integrated approach. Following a successful trial, the enhanced system has been formally integrated into the KMA’s operational environment, providing a more reliable foundation for high-resolution precipitation nowcasting.
This research was supported by the "Development of radar-based technology for nowcasting information computation (KMA2026-00221)" of "Development of severe weather response technology based on National Radar Convergence" project funded by the Weather Radar Center, Korea Meteorological Administration.
How to cite: Ko, K., Kim, K.-H., and Nam, K.-Y.: Enhanced AI-Based Growth and Decay Rate Estimation for Radar Precipitation Nowcasting in Korea, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-435, https://doi.org/10.5194/ems2026-435, 2026.
In order to strengthen the application of deep learning in precipitation nowcasting over North China, a three-year dataset of 10-minute quantitative precipitation estimation (QPE) observations was employed to develop a minute-level nowcasting model based on the U-Net architecture. This model enables rolling precipitation forecasts with a 10-minute update interval for the next 0 to 2 hours. To evaluate its performance, we conducted verification using long-term series from June to September in 2020 and 2025, and further analyzed four heavy precipitation events that occurred on August 12, 2020, July 1, 2021, July 30, 2024, and August 27, 2025. A set of evaluation metrics—including threat score (TS), bias score (BIAS), probability of detection (POD), success ratio (SR), and false alarm rate (FAR)—was adopted for comprehensive assessment.
The results demonstrate that the U-Net model produces predictions close to observations, albeit with some degree of false alarms. Its overall forecasting performance is markedly superior to that of the optical flow method, persistent forecast, and the CMA-MESO numerical model. Specifically, when the minute-level precipitation intensity does not exceed 10 mm per 10 minutes, the U-Net model outperforms both the optical flow method and the persistent forecast. Similarly, for hourly precipitation not exceeding 25 mm·h⁻¹, the U-Net model shows better performance than the CMA-MESO model and the optical flow method. It should be noted, however, that heavy precipitation events with intensities exceeding these thresholds are relatively scarce in the training dataset, resulting in insufficient samples for the model to adequately capture such extreme patterns; thus, special attention is required when applying the model to heavy rainfall scenarios.
Despite these advantages, the U-Net model exhibits certain limitations, including forecast blurring and an excessively broad precipitation area. To mitigate these issues, we further introduced multimodal meteorological variables by incorporating 10-minute 850 hPa temperature fields and u/v wind components into the model input. Experimental results indicate that this multimodal approach effectively reduces the extent of false alarms, enhances positional accuracy, and yields more realistic precipitation intensity, ultimately leading to a further improvement in the TS score.
How to cite: xu, C.: A U-Net-Based Minute-Level Precipitation Nowcasting Model with Multimodal Meteorological Variables over North China, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-180, https://doi.org/10.5194/ems2026-180, 2026.
Due to the impact of climate change, the likelihood of short-duration intense rainfall events is increasing. In particular, urban areas are particularly vulnerable to cascading and compounding flood impacts when torrential rains occur. Therefore, to effectively respond to urban flooding, it is crucial to produce rapid inundation information alongside real-time rainfall forecasting. However, existing physics-based flood models have limitations in real-time prediction due to long computation times and high dependence on detailed topographic and drainage network data. To address these limitations and enhance flood prediction accuracy, this study utilized a deep learning-based flood prediction model (CRU-Net) and radar-based rainfall nowcasting.
CRU-Net is a U-Net-based deep learning model designed to combine Residual Blocks and Convolutional Block Attention Module (CBAM) to simultaneously capture the spatiotemporal variability of rainfall and the complex characteristics of urban topography. Furthermore, by utilizing input parameters of physics-based flood models such as topography, land cover, and drainage networks, it serves as a surrogate model for flood prediction that reflects the physical structure of urban watersheds. In this study, CRU-Net was trained using flood scenarios generated with SWMM and a 2D flood analysis model. Additionally, to secure a lead time for flood prediction, the KICT-RAIN-AI model was used to generate predicted rainfall with a lead time of 10 to 180 minutes.
The study area was Gwanak-gu, Seoul, South Korea, which suffered flood damage in August 2022. Rainfall occurring between August 8 and 9, 2022, was predicted, and flood depths were predicted at 10-minute intervals. Based on the time-series prediction results, flood inundation extent maps were created, and flood reproducibility was examined by comparing them with the actual flood trace maps from 2022. This study demonstrates the potential to overcome the limitations of conventional physics-based approaches and to secure the lead time necessary for evacuation during urban flood events
Acknowledgements:
This work is financially supported by Korea Ministry of Climate, Energy, Environment(MCEE) as Climate Resilient R&D Project for Water-Related Disaster Management (RS-2026-25502323).
How to cite: Doh, E., Yoon, S.-S., and Kim, H.-J.: Improving Urban Flood Predictability Using Deep Learning-Based Radar Rainfall Nowcasting and an Inundation Model, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-786, https://doi.org/10.5194/ems2026-786, 2026.
MET Nordic is an hourly gridded dataset developed by the Norwegian Meteorological Institute (MET Norway), providing near-surface meteorological variables at 1 km resolution for Scandinavia, Finland, and the Baltic countries. Variables include temperature at two metres, precipitation, sea-level pressure, relative humidity, wind speed and direction, global radiation, long-wave downwelling radiation, and cloud area fraction.
The dataset integrates forecasts from the MetCoOp Ensemble Prediction System (MEPS) and various observational sources, including crowdsourced temperature and precipitation data from citizen-managed weather stations. These additional data sources improve the analysis and short-term forecasts.
The MET Nordic dataset is produced in real time (MET Nordic RT), serving civil protection and public weather services (e.g. Yr.no). Additionally, when new methods are introduced, we rerun the dataset back to 2012 in order to create an updated archive of historical analyses and forecasts. MET Nordic rerun can be used in hydrological models, case studies, and also for training machine learning models. In January 2026 we officially released rerun version 4 (MET Nordic rerun v4). This poster describes the input data, methods, and results for version 4 of the MET Nordic analysis, with a focus on hourly precipitation.
In MET Nordic v4, observational data from multiple rain gauge types are adjusted for wind undercatch as well as for systematic differences between crowdsourced and conventional observations. These adjustments aim to reduce systematic errors in hourly precipitation analysis – however they may also increase uncertainty in individual cases. The corrected observations are then quality-controlled using our in-house library, Titanlib, available at https://github.com/metno/titanlib.
The spatial analysis method has also been updated in MET Nordic v4. The new method, Ensemble-based Statistical Interpolation (EnSI), combines model output and observations in a multi-scale framework. A “started Box-Cox transformation” is applied when analyzing variables that deviate from Gaussian distributions. EnSI was evaluated using 231 heavy precipitation events, including a reconstruction of hourly precipitation and temperature during the 2023 “Hans” extreme weather event in Scandinavia. Results show that the multi-scale approach improves both accuracy and precision compared to a single-scale scheme.
MET Nordic is publicly available and documented on https://github.com/metno/NWPdocs/wiki/MET-Nordic-dataset. The EnSI spatial analysis method is implemented in the GridPP post-processing tool, available at https://github.com/metno/gridpp.
How to cite: Neuville, A., Båserud, L., Nipen, T. N., Seierstad, I. A., and Lussana, C.: MET Nordic analysis of hourly precipitation over Scandinavia, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-696, https://doi.org/10.5194/ems2026-696, 2026.
Accurate forecasting of cold season precipitation in the Eastern Mediterranean, particularly over the complex terrain of Cyprus Island, remains a persistent operational challenge. In a novel study for this region, we evaluate the impact of assimilating high-resolution, state-of-the-art remote sensing observations from the CYGMEN project (Cyprus GNSS Meteorology Enhancement) infrastructure (CyMETEO) on short-range numerical weather prediction.
Using the Weather Research and Forecasting (WRF) model configured at a convection-permitting resolution of 2 km, we employ the WRFDA 3D-Var system to assimilate dense, localized observation networks. The assimilated datasets comprise Zenith Total Delay (ZTD) from a regional GNSS network, kinematic profiles from a Wind Profiler Radar, and vertical temperature and relative humidity profiles from a Microwave Radiometer (MWR), alongside conventional meteorological data from the Global Telecommunication System (GTS). Prior to assimilation, all non-conventional observations underwent a rigorous Quality Assurance and Quality Control (QA/QC) protocol via intercomparison with reference radiosonde profiles. Crucially, the standard deviations derived from this validation were explicitly utilized to define the observational error covariances within the DA framework, ensuring an optimal weighting of the ingested data. To quantify the added value of these advanced observations, a full cyclic assimilation suite was contrasted against an open-loop control run across four distinct heavy precipitation events during the 2025–2026 cold season. Model performance was systematically evaluated against an independent network of Automatic Weather Stations (AWS). Objective verification demonstrates a substantial improvement in Quantitative Precipitation Forecasts (QPF), notably reflected in higher Fractional Skill Scores (FSS) and a reduced false alarm ratio at higher rainfall thresholds. The data assimilation cycling successfully mitigated the severe overestimation of precipitation prevalent in the control experiments by correcting a pervasive low-level moist bias. Specifically, the assimilation of GNSS-ZTD effectively constrained the Integrated Water Vapor (IWV) field, leading to a much sharper spatial localization of orographic rain bands over the terrain. Furthermore, the combined ingestion of MWR thermodynamic profiles and wind profiler kinematics optimized the Convective Available Potential Energy (CAPE) and improved the representation of low-level wind shear. This resulted in a far more accurate depiction of boundary layer moisture convergence and the precise temporal initiation of convection. These findings highlight the critical value of the CYGMEN network in improving regional numerical weather prediction over the Cyprus Island.
How to cite: Parde, A. N., Oikonomou, C., and Haralambous, H.: Impact of CyMETEO Data Assimilation on Cold-Season Precipitation Forecasts over Cyprus, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-782, https://doi.org/10.5194/ems2026-782, 2026.
Supporters & sponsors
