- Challenges in developing high-resolution mesoscale models with a focus on end-users and the EUMETNET forecasting programme. This involves cooperative operational systems as well as developments of different parts of the operational modeling chain, from nowcasting to forecasting, deterministic and probabilistic, and post-processing. Observation impact studies to assess the importance of different parts of the observing system for global and limited area NWP models are also welcome.
- Numerics and physics-dynamics coupling in weather and climate models. This involves the development, testing and application of novel numerical discretization techniques and sub-grid models, variable-resolution modelling, as well as performance aspects on current and emerging computing architectures.
Additionally, we welcome contributions to hybrid approaches that extend physical models with data-driven machine learning (ML) techniques. Specifically (but not exclusively), we are interested in how ML techniques can be used to identify and address deficiencies in physical models.
Operational forecasting systems routinely blend outputs from multiple numerical weather prediction (NWP) models, yet these blends often rely on static weights that do not reflect how model skill fluctuates in time and space. This study explores a simple dynamic weighting approach that updates the contribution of two global NWP systems - ECMWF and ACCESS, using both their deterministic and ensemble variants - based on their recent performance. By assigning weights proportional to inverse RMSE over a rolling 30-day window, the method adapts automatically to evolving model behaviour across regions, lead times, and variables.
The experiment focuses on 2-m temperature, 2-m dewpoint, and 10-m wind speed during two contrasting seasonal periods (December-February and June-August). Model fields are regridded to a common domain, and no bias correction or calibration is applied. This design isolates the effect of the weighting strategy itself, providing a clean assessment of how dynamic weighting compares to static, fixed-weight blends. Verification is performed against the Bureau of Meteorology's gridded Mesoscale Surface Analysis System (MSAS) and an extensive network of Automatic Weather Stations (AWS).
The largest gains occur for temperature and dewpoint at longer lead times, where the dynamic blend outperforms both individual models and the static blends. A distinctive result from the MSAS-based evaluation is the reduced amplitude of the diurnal RMSE cycle for temperature and dewpoint. This behaviour reflects the method's ability to exploit complementary biases - such as opposing warm and cold tendencies across models at different valid times - resulting in partial cancellation of systematic errors without explicit bias correction.
Wind speed forecasts also benefit, though improvements are more modest, reflecting the inherently higher variability and sensitivity of wind fields. Small regions of degradation emerge in areas with sparse observations or complex terrain, highlighting the limitations of retrospective RMSE when reference data are uncertain or unrepresentative.
Overall, this work demonstrates that a transparent, performance-based dynamic weighting strategy can deliver meaningful forecast improvements using only raw model output. Its low operational cost and ability to adapt to evolving model skill make it a promising candidate for next-generation blending approaches and a foundation for future regime-dependent or probabilistic adaptive systems.
How to cite: Abellan, E. and Johnson, R.: When the Models Compete, the Forecast Wins: Dynamic Weighting for NWP, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-258, https://doi.org/10.5194/ems2026-258, 2026.
In mid-2026 ECMWF introduced into operations a new cycle (50r1) of its physics-based IFS (Integrated Forecast System). One key feature of 50r1 was an upgrade to the convection scheme. This is particularly beneficial for bringing SST-triggered convection inland across coastal regions, but also has positive displacement impacts over and downstream of other convectively active areas. These improvements were achieved, in model code, primarily by commuting a proportion (typically 40%) of precipitation particles from the convection scheme into the large scale scheme. That 40% can then advect with the wind; in the previous cycle it fell out instantaneously. This change has had a big impact, worldwide, on the ratio of convective to total precipitation (cpr). Now small values of cpr are much more common, and vice versa.
With all this and other model changes in mind a comprehensive, supervised ML-style re-assessment has been made of relationships between short range forecast rainfall (F) and observed (gauge-based) rainfall (O), for cycle 50r1, using ranges of cpr and other variables to create ‘weather type’ classes. This is revealing, showing a large range of noteworthy situation-dependant biases, which are useful for both forecasters and model developers. For example, there is a diurnal cycle in bias, with afternoon rainfall systematically over-predicted by the IFS, whilst night-time rainfall is less erroneous. This bias difference is amplified in CAPE-rich environments. Meanwhile, in some different scenarios, the model consistently overpredicts rainfall by a large margin (O/F ~ 0.25), whilst in others, it markedly underpredicts (O/F ~ 3). We have found cpr, CAPE and low level dewpoint depression to be key predictors. Numerous results of this assessment will be presented in this talk. The results are based on over 10 million O:F pairs. Moreover with the new convection scheme advecting precipitation into regions where previous cycles would have failed, we now have the capacity to improve those forecasts even further, to account for inadequacies in the 40% transfer factor incorporated.
For users, ECMWF can now provide related situation-dependant bias-corrected forecast products, via the "ecPoint" post-processing approach (along with estimates of sub-grid variability). This post-processing and the related conditional verification calibration activity discussed above were all streamlined in short time in 2026 by using agentic AI to (i) convert a pre-existing containerised calibration tool into an open-source (localhost) web application, (ii) markedly accelerate processing speed therein and (iii) add numerous other user-oriented functions to it. With this tool others can now do similar assessments with their own models. A brief overview of this development will be given.
How to cite: Hewson, T. and Pillosu, F.: Evolving Systematic Errors in Precipitation Forecasts from ECMWF, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-740, https://doi.org/10.5194/ems2026-740, 2026.
In-situ observations constitute an essential component of the atmospheric observing system and play a key role in improving Numerical Weather Prediction (NWP), both through their direct assimilation and through their use in diagnostics. By assimilating these observations, the representation of the atmospheric state in the forecast model leads to more accurate Numerical Weather Predictions. However, the availability of in-situ observations is characterized by significant temporal and spatial gaps, particularly in regions of the Global South.
In this study, we investigate the potential of a novel type of observation for the global model of the German Weather Service (DWD): bio-logging data collected from small tags attached to white storks (Ciconia Ciconia), originally deployed to study migratory movements and other behavioural aspects of these birds. These tags provide high-frequency measurements of both horizontal and vertical position, but could also provide meteorological information. Here, we assimilate estimated wind speed and direction from the thermalling flight behaviour of white storks using high-frequency GPS recordings from the bio-logging dataset. The data covers a region extending from Germany over France and Spain to West Africa, mainly in a range up to 700 hPa.
Preliminary results from a 10-day experiment indicate that the observation-minus-first-guess and observation-minus-analysis statistics fall within physically reasonable ranges. Analysis of vertical profiles and comparison with neighbouring aircraft observations show a general consistency between the assimilated stork and aircraft statistics. In this small sample, the stork observations even show improved agreement with the model at certain flight levels, in terms of both bias and standard deviation.
How to cite: Anesiadou, A., Cress, A., Schomburg, A., Flack, A., and Kirchner, T. M.: Use of bio-logging data from thermalling white storks as potential atmospheric observations in a global data assimilation system, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-156, https://doi.org/10.5194/ems2026-156, 2026.
Accurate representation of ionospheric electrodynamic forcing is essential for modeling the coupled thermosphere–ionosphere system, particularly at high latitudes where Joule heating and ion drag dominate the energy and momentum budget. These processes play a central role in controlling thermospheric temperature, neutral winds, and density variability. In the ICOsahedral Nonhydrostatic (ICON) model, ionospheric forcing is typically parameterized using empirical inputs or simplified formulations, which do not fully capture the complex spatial and temporal variability associated with magnetosphere–ionosphere coupling.
In this work, we develop a machine learning–based parameterization of Joule heating and ion drag using a convolutional neural network architecture. The model is trained on outputs from the Whole Atmosphere Community Climate Model with thermosphere–ionosphere extension (WACCM-X), enabling it to learn nonlinear relationships between electrodynamic drivers and the resulting thermospheric response. A U-Net architecture is employed to capture both local structures and large-scale spatial patterns, which are essential for representing high-latitude electrodynamic variability.
To enable deployment within a different dynamical core, a grid conversion framework is developed to map data between the structured latitude–longitude grid of WACCM-X and the unstructured triangular grid used in ICON. This mapping ensures physical consistency of the input and output fields while preserving spatial coherence across resolutions. The trained model is evaluated on independent time periods outside the training dataset and demonstrates strong skill in reproducing both the spatial distribution and magnitude of Joule heating and ion drag across multiple altitude levels.
The proposed approach provides a computationally efficient alternative to traditional parameterizations, reducing reliance on empirical inputs while retaining high accuracy. By capturing complex electrodynamic variability, this method offers improved representation of high-latitude forcing in global models. This work represents a step toward hybrid modeling frameworks in which machine learning augments first-principles approaches to enhance the fidelity of upper-atmosphere simulations.
How to cite: El Zaatari, L., Schwabe, M., and Stephan, C.: AI-driven approach to ionospheric modelling, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-233, https://doi.org/10.5194/ems2026-233, 2026.
The domain of weather forecasting is currently undergoing a significant transformation driven by advances in machine learning, where Data-Driven Models (DDMs) have demonstrated equal or superior performance compared to traditional Numerical Weather Prediction models in predicting various variables, while operating at a fraction of the computational cost (Bouallegue et al., 2024). Probablistic DDMs have the potential to provide a computationally cheap solution for ensemble modelling at hectometric scale (with a spatial resolution of 500 to 750 m), motivating Task 330141 of the Destination Earth Weather-Induced Extremes Digital Twin (Extremes DT, DE330) project (ECWMF, 2024). An important step towards hectometric ensemble modelling is high-resolution (km-scale) modelling over a regional domain, for which Nordhagen et al. (2025) have already shown promising results for the Nordic area with their Bris CRPS-FFT model, a model using a stretched-grid approach, transfer learning and a Continuous Ranked Probability Score (CRPS) loss function with Fast Fourier Transform (FFT) to combine good model skill with spatial coherence, also at the smaller scales. On-demand extreme forecasting involves potentially high-impact events, which requires the flexibility to respond fast to triggered events and provide a tailored forecast on the domain of interest, including crucial uncertainty information. In this work, we combine the kilometer-scale multi-domain training approach -based on dynamical graph training- with the Bris CRPS-FFT approach, with the aim of providing on-demand ensemble forecasts of extremes at a hectometric scale. Probablistic multi-domain DDMs have the potential to provide the forecasting speed, the domain flexibility and the uncertainty quantification necessary to handle this complex task. Following Nipen et al. (2024), a global model at 0.25-degree resolution is pre-trained (stage A and B), after which transfer learning is applied to integrate the higher-resolution data in stage C. In this stage, dynamical graph training is employed, where multiple regional datasets are alternated as batch input to increase the generalizability of the model. In the final stage, we finetune the model on 200+ hectometric datasets of the Extremes DT triggered by forecast extreme events, resulting in the first on-demand autoregressive ensemble model of its type at hectometric scale. The results reflect the potential of multi-domain training combined with transfer learning for on-demand highly flexible domains with sparse data availability.
How to cite: Buurman, S., Salihi, A. F. S., Nordhagen, E. M., Santa Cruz, M., van Ginderachter, M., Schönach, D., and Nils Nipen, T.: Towards a data-driven weather model for forecasting on-demand extremes at hectometric scale , EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-581, https://doi.org/10.5194/ems2026-581, 2026.
UTCIcast is an operational framework for forecasting urban thermal stress that combines numerical weather prediction with high-resolution urban climate modeling. At its core lies GLIDE-SOL, a fully scripted and globally deployable Python workflow that enables rapid, consistent, and repeatable simulations of urban thermal conditions based on the SOLWEIG model.
GLIDE-SOL builds upon the SOLWEIG radiative balance formulation but redesigns the full modeling pipeline—including automated input generation, execution, and post-processing—to operate exclusively with globally available datasets. All required inputs, such as terrain, building morphology, vegetation structure, land cover, and meteorological forcing, are automatically derived from harmonized global products. This removes the need for local preprocessing and allows consistent applications across cities worldwide, from neighborhood to metropolitan scales.
Within UTCIcast, GLIDE-SOL is driven by short-range numerical weather prediction (e.g., ICON-EU forecasts up to 72 hours), enabling near-real-time simulations of the urban radiative environment. The model is implemented on GPUs, allowing meter-scale resolution simulations with reduced computational time. It produces key fields such as mean radiant temperature (Tmrt), shadow patterns, and near-surface meteorological variables, which are combined to compute the Universal Thermal Climate Index (UTCI), a physically consistent indicator of outdoor thermal stress.
To improve performance under coarse meteorological forcing, GLIDE-SOL incorporates lightweight physical diagnostics that capture key urban processes. These include a directional wind attenuation scheme based on urban roughness and obstacles, and diagnostic air temperature corrections that combine a simplified urban heat island (UHI) cycle with elevation-based adjustments derived from high-resolution digital elevation models. These additions enhance the representation of ventilation, nocturnal warming, and local temperature gradients.
Scalability is achieved through domain tiling with cross-tile synchronization, preserving radiative consistency while enabling simulations over large urban areas at fine spatial resolution. Outputs are generated as compressed georeferenced rasters and integrated into interactive web maps with hourly time navigation and pixel-level inspection.
The workflow is structured into three reproducible components: automated global input generation, a GPU-accelerated SOLWEIG execution engine, and a post-processing module for systematic analysis and visualization. An operational application in Dortmund, based on hourly observations from 25 stations and simulations at 2 m resolution over more than one year, demonstrates substantial improvements in UTCI accuracy, with RMSE reduced from 9.9°C to 2.7°C when including wind and temperature diagnostics.
By integrating global data, GPU-accelerated urban physics, and scalable processing, UTCIcast—powered by GLIDE-SOL—provides a flexible platform for near-real-time heat monitoring, forecast-based risk assessment, and consistent multi-city urban climate analyses.
How to cite: Zonato, A., Milelli, M., and Monaco, L.: UTCIcast: A Scalable Urban Thermal Stress Forecasting System Powered by GLIDE-SOL, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-112, https://doi.org/10.5194/ems2026-112, 2026.
Precipitation is an important variable of concern in weather forecast and climate prediction, its simulation and prediction performance have always been one of the indicators of the numerical models. Precipitation prediction is influenced by many factors, including not only the dynamic framework and various process parameterization schemes of the model, but also the initial conditions (i.e., initial values) and boundary conditions. Through case studies with the climate model of CMA Climate Prediction System, this study explores the impacts of initial and boundary conditions in land model on the precipitation prediction, as well as the possible land-atmosphere interactions involved.
The AMIP-kind experiments have been conducted with the CMA climate model for participating the LS4P (Impact of initialized Land Surface temperature and Snowpack on S2S Prediction) project. By evaluating the deviation between climate model simulation results and observations, a temperature mask was created and introduced in the model to ‘correct’ the initial conditions of spring soil temperature (ST), and its impact on the temperature in May and precipitation in downstream areas in June was evaluated. The primary results indicate that adjusting the initial soil temperature of spring in key regions (such as the Tibet Plateau) can effectively improve the model predictive accuracy for summer precipitation in downstream areas. And the initial ST correction of the different large-scale terrains (Tibet Plateau and Rocky Mountains) has different impacts on the prediction of precipitation in the middle-lower reaches of the Yangtze River Valley of China. We will also introduce the impacts of boundary conditions, mainly including sea surface temperature and land cover datasets. Overall, the contributions on precipitation prediction can reach as much as 1/3. Therefore by appropriate configuring and updating initial and boundary conditions in the model, precipitation prediction can be effectively improved.
How to cite: Shi, X. and Zhang, Y.: Contributions of land model initial and boundary conditions to improving the precipitation prediction of CMA climate model, EMS Annual Meeting 2026, Utrecht, Netherlands, 6–11 Sep 2026, EMS2026-299, https://doi.org/10.5194/ems2026-299, 2026.
Supporters & sponsors
