Session 9 | Numerical modelling, convection-allowing models, data assimilation, and machine learning

Session 9

Numerical modelling, convection-allowing models, data assimilation, and machine learning

Orals MO2

| Mon, 17 Nov, 09:30–10:45 (CET)|Room Hertz Zaal

Posters TU4

| Attendance Tue, 18 Nov, 14:30–16:00 (CET) | Display Mon, 17 Nov, 09:00–Tue, 18 Nov, 18:30|Poster area, P53–72

Orals: Mon, 17 Nov, 09:30–10:45 | Room Hertz Zaal

09:30–10:00

ECSS2025-90

keynote presentation

Improving Our Understanding of Global Severe Convective Storm Hazards

Andreas F. Prein, Praveen Pothapakula, Christian Zeman Zeman, and Boris Blanc

Severe convective storms (SCSs) are responsible for substantial societal and economic losses around the world. These impacts are rising—primarily due to increasing exposure—yet the underlying hazard remains deeply uncertain, even in regions with dense observational networks. This uncertainty complicates efforts to anticipate and manage risk. In this keynote, we will present recent advances aimed at enhancing our understanding of global SCS hazards through the use of both statistical and dynamical modeling approaches. Statistical models, including machine learning techniques, are increasingly used to estimate hail hazard at continental to global scales, providing valuable insights into large-scale patterns and regional variability. Complementing these efforts, convection-permitting simulations—run at kilometer-scale resolution on regional and global scales—allow for the explicit representation of severe convective processes. These include hail, straight-line winds, and supercell thunderstorms, often accompanied by diagnostic tools like HAILCAST, which helps better assess event intensity and frequency. We will discuss the strengths and weaknesses of these approaches and highlight how combining them provides a more comprehensive understanding of SCS behavior and its sensitivity to environmental drivers. We will conclude by highlighting open scientific questions and future research directions, particularly in consideration of ongoing and projected climate changes, to support improved hazard assessment and risk reduction strategies globally.

How to cite: Prein, A. F., Pothapakula, P., Zeman, C. Z., and Blanc, B.: Improving Our Understanding of Global Severe Convective Storm Hazards, 12th European Conference on Severe Storms, Utrecht, The Netherlands, 17–21 Nov 2025, ECSS2025-90, https://doi.org/10.5194/ecss2025-90, 2025.

10:00–10:15

ECSS2025-140

Postprocessing Global AI Weather Prediction Models for Severe Weather Forecasting

Aaron Hill and Evan White

Data driven global AI models have garnered significant attention of late owing to their competitiveness with state-of-the-art dynamics-based numerical weather prediction (NWP) systems. After training, the AI models can simulate synoptic-scale patterns with relative ease and speed, and recent quantitative evaluations have suggested they can simulate large-scale temperature, pressure, and winds with the same or better accuracy than NWP models. However, these AI models suffer from some of the same deficiencies as NWP models for applications in high-impact weather domains. Namely, their rather coarse resolution doesn’t lend itself well to explicit predictions of convection hazards. To get around this issue with traditional NWP models, postprocessing methods have been used to generate explicit forecasts of hazards. As an example, the Global Ensemble Forecast System Machine Learning Probabilities (GEFS-MLP) forecast system leverages random forests (RFs) and inputs from a global NWP ensemble to generate daily probabilistic forecasts at lead times of 1-8 days for excessive rainfall, tornado, severe hail, and severe wind hazards. Output forecasts mimic operational outlooks and are now operational in national forecast centers.

In a similar way, output fields from the data-driven AI models can be used to generate hazardous weather outlooks. In this work, we apply the previously detailed GEFS-MLP framework to global AI model inputs, to generate daily probabilities of tornadoes, hail, and wind out to 8 days. We explore inputs from PanguWeather, GraphCast, and FourCastNet to drive separate severe weather predictions, and consider combining AI outputs as a 3-member ensemble to drive RF training and forecasts. The operations-like outlooks are compared to similar operational GEFS-MLP products to examine how data driven AI models can be used for small-scale hazardous weather prediction. Additionally, we explore generating ensembles of MLP forecasts by applying trained RFs to individual GEFS ensemble members to understand the value added by generating probabilistic forecasts. Further, explainable AI techniques are leveraged to decipher how MLP systems respond to inputs from different forecast models and whether global AI systems could be used in an ingredients-based forecasting paradigm.

How to cite: Hill, A. and White, E.: Postprocessing Global AI Weather Prediction Models for Severe Weather Forecasting, 12th European Conference on Severe Storms, Utrecht, The Netherlands, 17–21 Nov 2025, ECSS2025-140, https://doi.org/10.5194/ecss2025-140, 2025.

10:15–10:30

ECSS2025-261

Improving Polarimetric Radar Simulations in AROME: Evaluation of Microphysics and Forward Operator Assumptions

Clotilde Augros, Cloé David, Benoit Vié, and François Bouttier

Accurate simulation of dual-polarization radar variables is essential for their assimilation into convective-scale NWP models. This requires both realistic microphysical parameterization and a reliable representation of hydrometeor scattering characteristics through a polarimetric radar forward operator.

This study evaluates the performance of two microphysics schemes in the AROME model: ICE3 (1-moment) and LIMA (2-moment), using the polarimetric radar forward operator developed by Augros et al. (2016). Simulations were compared with C-band dual-polarization radar observations over 34 convective days in 2022. An object-based analysis of convective cells focused on hydrometeor structure, vertical profiles, and polarimetric signatures such as differential reflectivity (ZDR) columns. LIMA more accurately reproduced low-level polarimetric variables, notably by simulating larger raindrops and stronger ZDR and specific differential phase (KDP) values within convective cell cores. It also generated a realistic number of ZDR columns, along with accurate lifetimes and areal extents. However, both microphysics schemes showed deficiencies within and above the melting layer, highlighting limitations in the radar forward simulations in these regions.

To address these limitations, ongoing work focuses on refining the assumptions used in the radar forward operator. A two-step methodology is applied. First, a theoretical sensitivity study explores how variations in hydrometeor mass–size relations, shape, particle size distributions, and melting state affect simulated radar variables. Initial results, in particular, show the strong influence of density–size assumptions for dry snow. Second, the refined assumptions are evaluated in AROME simulations of convective and stratiform events to quantify their impact on radar outputs. This work aims to reduce systematic biases and improve consistency with observed polarimetric radar fields.

References:
Augros, C. et al. (2016), QJRMS, 142(S1), 347–362. https://doi.org/10.1002/qj.2572
David, C. et al. (2025), EGUsphere. https://doi.org/10.5194/egusphere-2025-685

How to cite: Augros, C., David, C., Vié, B., and Bouttier, F.: Improving Polarimetric Radar Simulations in AROME: Evaluation of Microphysics and Forward Operator Assumptions, 12th European Conference on Severe Storms, Utrecht, The Netherlands, 17–21 Nov 2025, ECSS2025-261, https://doi.org/10.5194/ecss2025-261, 2025.

10:30–10:45

ECSS2025-176

Performance assessment of drone-based photogrammetry coupled with machine-learning for the estimation of hail size distributions on the ground

Jannis Portmann, Martin Lainer, Killian P. Brennan, Marilou Jourdain de Thieulloy, Matteo Guidicelli, and Samuel Monhart

Hail-producing thunderstorm are extreme weather phenomena that can have devastating impacts on the environment, agriculture and infrastructure, leading to large financial losses. Remote sensing techniques such as dual-polarimetric weather radar are the state-of-the-art for observing hail over large areas, but don’t necessarily represent conditions on the ground.

To verify weather radar-based products, current ground-based observations provide valuable information but exhibit various limitations: Crowd-sourced reports provide information over widespread areas, but only indicate the size of the largest hail stones and are biased towards populated areas, while automatic hail sensors and hailpads provide hail size distribution (HSD), based only on small observational areas of <1m², implying that only a partial HSD is retrieved. Drone-based hail photogrammetry can help to close this observational gap by sampling thousands of hailstones within a hail core across large areas of hundreds of square-meters at high resolution. Combining the approaches of previous studies, images captured during a drone flight are combined into a rectified image (orthophoto), from which hailstones are detected and their sizes are estimated using machine-learning models for object detection and image segmentation.

To assess the uncertainty of the hail size distribution retrievals from machine-learning models, we set up experiments on different grass surfaces using synthetic hail objects with a well-defined ground truth in terms of size and number. The results of the experiments are compared to drone-based HSD retrievals of a real hail event, which occurred in 2022 in Locarno, Switzerland. In the experimental setup and the real event, 98% of the synthetic hail objects and 81% of hailstones were correctly detected. Overall, the estimated size is in good agreement with the real size of the synthetic hail objects and only slight underestimations of the detected objects could be found. Across all different size classes, the underestimation is around -0.75mm for both synthetic hail and hailstones. The high performance in detection and size estimation, coupled with large sampling areas allow us to estimate representative HSDs on the ground under favorable conditions, but due to its time-intensive and challenging data collection process, it is best coupled with other measurement methods. In combination, a reliable ground dataset can be created to validate radar estimates and potentially improve forecasting of hail events. Additionally, we present past experiences of using drone-based hail photogrammetry and ongoing improvements of the method to extend its application in the field under challenging conditions.

How to cite: Portmann, J., Lainer, M., Brennan, K. P., Jourdain de Thieulloy, M., Guidicelli, M., and Monhart, S.: Performance assessment of drone-based photogrammetry coupled with machine-learning for the estimation of hail size distributions on the ground, 12th European Conference on Severe Storms, Utrecht, The Netherlands, 17–21 Nov 2025, ECSS2025-176, https://doi.org/10.5194/ecss2025-176, 2025.

Posters: Tue, 18 Nov, 14:30–16:00 | Poster area

Display time: Mon, 17 Nov, 09:00–Tue, 18 Nov, 18:30

P53

ECSS2025-38

Quality Control and Observation Thinning for 2-metre Temperature Assimilation in an Operational Convective Nowcasting Framework

Francesco Uboldi, Elena Oberto, Massimo Milelli, Andrea Zonato, Martina Lagasio, and Antonio Parodi

This study investigates the impact of assimilating 2-metre temperature (T2M) observations into an operational convective-scale nowcasting system. The system is based on the WRF model at 2.5 km resolution and already includes the assimilation of radar reflectivity via 3DVAR and lightning observations through nudging techniques, delivering high-frequency forecasts to support real-time decision-making in civil protection.

The analysis focuses on August 2024 and uses T2M observations from both the Italian Civil Protection Department and the MeteoNetwork association. A unified and fully automatic quality control (QC) procedure is applied jointly to both datasets before assimilation. This procedure includes metadata validation (META), elevation consistency verification using a high-resolution digital elevation model, a background field check (BF) based on Optimal Interpolation, a spatial consistency test (SCT) using leave-one-out cross-validation, and a DZMIN check, which rejects stations whose elevation differs significantly from the model orography. Additionally, the standard WRFDA internal consistency check is performed to filter out observations too divergent from the short-range forecast background.

To align the observational dataset with the spatial resolution of the model and avoid redundancy, a thinning strategy is applied after QC to retain only a subset of spatially representative observations.

Two assimilation configurations are compared: one using all raw T2M observations without QC, and another using only QC-passed and thinned observations. Both are evaluated against the current operational baseline, which includes only radar and lightning assimilation. The comparison assesses differences in assimilation stability, consistency, and potential benefits derived from the integration of surface temperature data.

This work highlights the importance of structured quality control pipelines and data representativeness in high-resolution convective-scale data assimilation. It also confirms the potential of integrating dense observational networks—including non-institutional sources—into early warning frameworks. The results contribute to the objective of enhancing forecasting methodologies for extreme hydrometeorological events in support of environmental protection strategies across complex and vulnerable territories.

This research is conducted within the PNRR RAISE initiative - Spoke 3, which develops innovative technologies for environmental safeguard in water, air, and soil domains over the Ligurian region.

How to cite: Uboldi, F., Oberto, E., Milelli, M., Zonato, A., Lagasio, M., and Parodi, A.: Quality Control and Observation Thinning for 2-metre Temperature Assimilation in an Operational Convective Nowcasting Framework , 12th European Conference on Severe Storms, Utrecht, The Netherlands, 17–21 Nov 2025, ECSS2025-38, https://doi.org/10.5194/ecss2025-38, 2025.

Discussion Post a comment

P54

ECSS2025-64

Deep Learning-Based Prediction of Severe Convective Storm Perils

Davide Panosetti and Leandro Masello

As a leading commercial insurer specializing in property and infrastructure - including the rapidly expanding renewables sector - FM faces growing exposure to the impacts of severe convective storms (SCS). These high-impact weather events, which encompass hail, tornadoes, straight-line winds, lightning, and heavy precipitation, pose significant risks to insured assets. In this presentation, we introduce a suite of in-house predictive models developed to estimate both the frequency and intensity of these SCS perils. Our approach integrates observational data with state-of-the-art deep learning techniques to process and interpret high-dimensional meteorological inputs.

The models are trained on a comprehensive set of predictors derived internally from ERA5 reanalysis data, ensuring consistency and scalability across spatial and temporal domains. Deep learning models are particularly well-suited for this task, as they efficiently capture spatial dependencies and patterns within gridded atmospheric fields, enabling robust identification of conditions conducive to severe weather events.

These predictive tools are used internally to support underwriting activities, risk pricing, and accumulation management. In addition to peril-specific risk metrics, we introduce cross-peril correlation maps—particularly between hail and straight-line wind—that are instrumental in supporting FM’s growing Renewables business.

By combining physical understanding with data-driven modeling, our framework offers a scalable and interpretable solution for SCS risk assessment. The presentation highlights model architecture, training methodology, and selected case studies demonstrating operational relevance and performance.

How to cite: Panosetti, D. and Masello, L.: Deep Learning-Based Prediction of Severe Convective Storm Perils , 12th European Conference on Severe Storms, Utrecht, The Netherlands, 17–21 Nov 2025, ECSS2025-64, https://doi.org/10.5194/ecss2025-64, 2025.

P55

ECSS2025-70

Sensitivity of Supercell Environments to Orographic Modifications

Robert Kvak, Petr Zacharov, Martin Vokoun, and Marek Kašpar

It is well established that more pronounced terrain asymmetries are associated with a greater potential for terrain-induced modifications of the adjacent atmosphere, influencing both thermodynamic and kinematic conditions. Such modifications, particularly when locally enhanced, may contribute to an increased likelihood of supercell storm development. Although terrain effects on convective environments have received growing attention over the past decade, the extent to which orographic features influence mesoscale environments conducive to supercell formation remains an open question. Building upon recent research focused on supercell environments over mountainous regions in Europe, which demonstrated that high-resolution numerical weather prediction models can capture localized terrain-induced enhancements of supercell-favorable conditions, this study aims to further investigate the role of topography using a targeted modeling approach. We employ the COSMO model to simulate a set of supercell events that occurred over Central Europe during the 2015–2019 period, selecting days with observed supercell activity. In each simulation, the orography of the Carpathian Mountains is systematically modified to isolate the effects of changing the height and volume of mountain ranges, intermontane basins, and surrounding lowlands. For each orographic scenario, we evaluate key diagnostic parameters known to characterize supercell environments, such as thermodynamic instability and vertical wind shear. The resulting changes in these parameters are analyzed to quantify the influence of terrain on the mesoscale environment. Additionally, we assess how these terrain alterations affect the modeled predictability of supercell occurrence, with implications for both operational forecasting and the understanding of terrain–storm interactions.

How to cite: Kvak, R., Zacharov, P., Vokoun, M., and Kašpar, M.: Sensitivity of Supercell Environments to Orographic Modifications , 12th European Conference on Severe Storms, Utrecht, The Netherlands, 17–21 Nov 2025, ECSS2025-70, https://doi.org/10.5194/ecss2025-70, 2025.

Discussion Post a comment

P56

ECSS2025-123

Synoptic Analysis and Simulation of the High-Shear, Low-CAPE (HSLC) F4-Tornado in Hautmont, France from August 03, 2008 using ERA5 data and Cloud Model 1

Oliver Heuser, Lisa Schielicke, and Petra Friederichs

On August 03, 2008, an F4 tornado struck the city of Hautmont, France, causing extensive damage across seven municipalities and injuring numerous individuals during its 14-minute lifespan. The tornado developed from a pre-frontal convective system within a high-shear, low-CAPE (HSLC) environment - a setting in which the occurrence of strong tornadoes is considered atypical. This study aims to analyze the synoptic-scale situation with a particular focus on tornado-favorable and convectively relevant parameters, utilizing ERA5 reanalysis data. A reconstructed atmospheric sounding and hodograph were used as initial conditions for idealized simulations with the Cloud Model 1 (CM1), incorporating various convective initiation triggers. The objective was to explore potential polarimetric signatures indicative of supercellular structures or bow echoes.

The analysis indicates that a jet coupling event triggered quasi-geostrophic cyclogenesis, which led to the development of a surface low and the formation of a low-level jet. Together with high moisture content near the surface and within the atmospheric boundary layer, these factors were identified as key contributors to tornadogenesis in this case.

Simulation results highlight limitations in representing soundings with high boundary layer moisture within CM1. Among the tested initiation triggers, updraft nudging proved to be the most effective and, in fact, the only one capable of producing a tornado-like vortex. This suggests that updraft nudging can enhance or even partially compensate for the inherently weak convective updrafts in HSLC environments, potentially enabling the formation of short-lived supercells following convection initiation. Based on these findings, it is proposed to conceptually differentiate between two categories of convective triggers: those that initiate convection and those that support and maintain it under marginal conditions.

As an outlook first attempts of creating a series of synthetic idealizied soundings for the use of numerical simulations were attemptend with the goal of testing the limits and boundaries of various numerical convective simulation models.

How to cite: Heuser, O., Schielicke, L., and Friederichs, P.: Synoptic Analysis and Simulation of the High-Shear, Low-CAPE (HSLC) F4-Tornado in Hautmont, France from August 03, 2008 using ERA5 data and Cloud Model 1, 12th European Conference on Severe Storms, Utrecht, The Netherlands, 17–21 Nov 2025, ECSS2025-123, https://doi.org/10.5194/ecss2025-123, 2025.

Discussion Post a comment

P57

ECSS2025-125

Benefit of hectometric simulations of the 18 August 2022 derecho-producing mesoscale convective system over Corsica

Marc Mandement, Didier Ricard, Clément Strauss, Christine Lac, Benoît Vié, and Clotilde Augros

On the night of 17 to 18 August 2022, disorganized thunderstorms located between Catalonia and the Balearic Islands (Spain) gradually strengthened, reaching the size of a mesoscale convective system in the middle of the night. Moving very rapidly over the Mediterranean Sea, this system organized into a squall line which suddenly bowed near Corsica. The system hit the island shortly after 06:00 UTC, accompanied by exceptionally violent winds. An anemometer in Marignana (Corsica) measured an instantaneous wind speed (over 0.5 s) of 225 km h^-1 (62.4 m s^-1), setting a new record under a thunderstorm in mainland France. This system, which lasted around 18 h, produced violent winds on a trajectory crossing 5 European countries from Spain to the Czech Republic, qualifying it as a derecho. Simulations exploring the contribution of increased resolution and testing the latest developments in physical parametrizations were carried out with the Meso-NH numerical research model, using initial and boundary conditions provided by the best AROME-France forecast (18 August 2022 00:00 UTC run). Among the simulations carried out, the one with 1 km horizontal grid spacing and 90 vertical levels, which has an effective resolution 2 to 3 times finer than AROME-France, simulates severe wind corridors more finely than the operational forecast, making it possible to get closer to the extreme wind observations. It also appears that the location of these corridors correlates with areas within the convective system where the vertical component of relative vorticity is very large. Simulations at 250 m horizontal grid spacing show even finer wind corridors and vortices. These results show that very fine scale phenomena occur within a derecho, which can only be reproduced by simulations at hectometric resolution.

How to cite: Mandement, M., Ricard, D., Strauss, C., Lac, C., Vié, B., and Augros, C.: Benefit of hectometric simulations of the 18 August 2022 derecho-producing mesoscale convective system over Corsica, 12th European Conference on Severe Storms, Utrecht, The Netherlands, 17–21 Nov 2025, ECSS2025-125, https://doi.org/10.5194/ecss2025-125, 2025.

Discussion Post a comment

P58

ECSS2025-145

Improved Understanding of Drivers of Upper-Level Updrafts and their Role in Producing Thunderstorm Overshooting Tops

Melinda Berman, Robert Trapp, and Stephen Nesbitt

Overshooting tops (OTs) are manifestations of deep convective updrafts that extend past the tropopause into the stratosphere. OTs can transport moisture and aerosols into the stratosphere and have been connected to the occurrence and intensity of severe weather hazards, such as tornadoes and hail. Recent work has shown a connection between OT characteristics, such as area (OTA) and depth (OTD), and midtropospheric updraft structure, but OTs apparently can also be modified by static stability in the lower stratosphere (LS). Here we use numerical simulations of OTs and their associated convective storms in idealized and real-case inspired environments to further our understanding of the impact of LS static stability on OTA and OTD. Consistent with our observational work, these simulations show that OTD depends significantly on LS static stability. In contrast, OTA tends to be relatively insensitive to LS static stability and is more closely related to midtropospheric updraft-core area, which in turn depends on the tropospheric vertical wind shear.

Further questions remain, however, on what drives updraft intensity aloft in ongoing OTs. Numerical simulations are being leveraged to understand the dynamical drivers of updrafts above the mid-troposphere. This will allow us to further develop the three-dimensional time-dependent model of a thunderstorm, including its extension into the stratosphere. Preliminary results indicate that parcels that spend longer amounts of time in the OT experience stronger dynamical pressure gradient forcings in the mid-to-upper troposphere (8-12 km) compared to parcels that spend short amounts of time or do not enter the OT.

How to cite: Berman, M., Trapp, R., and Nesbitt, S.: Improved Understanding of Drivers of Upper-Level Updrafts and their Role in Producing Thunderstorm Overshooting Tops, 12th European Conference on Severe Storms, Utrecht, The Netherlands, 17–21 Nov 2025, ECSS2025-145, https://doi.org/10.5194/ecss2025-145, 2025.

P59

ECSS2025-155

Lightning and precipitation forecast by means of hybrid deterministic and AI-based tools

Daniele Carnevale, Federico Cassola, Mattia Cavaiola, and Andrea Mazzino

Mediterranean coastal regions are regularly affected by sudden heavy precipitation events leading to very dangerous flash floods. Severe convection prediction, being the result of many mutually interacting multiscale processes, not yet completely understood and modeled, is still a major challenge for numerical weather prediction (NWP) systems. In recent times, artificial intelligence (AI) emerged as a powerful tool for handling vast amounts of data and extracting patterns and relationships that might be challenging to identify through traditional fully-deterministic algorithms. In the framework of the AIxtreme (Physics-based AI for predicting extreme weather and space weather events) project, a suite of AI-based techniques is being developed to calibrate numerical models based on the physics of the atmosphere, with the aim of anticipating the occurrence of extreme weather events and supporting decisions of civil protection agencies.

A first significant result of the project is the development of a deep learning framework, named FlashNet, able to forecast lightning flashes up to 48 h ahead in terms of probability of occurrence. FlashNet is capable to find an optimal mapping of meteorological features predicted two days ahead by the state-of-the-art numerical weather prediction model by the European Centre for Medium-range Weather Forecasts (ECMWF) into lightning flash occurrence. The prediction skill of the resulting AI-enhanced algorithm turns out to be significantly higher than that of the fully deterministic algorithm employed in the ECMWF model. A remarkable Recall peak of about 95% within the 0-24 h forecast interval is obtained. This performance surpasses the 85% achieved by the ECMWF model at the same precision of the AI algorithm.

A second tool, in an advanced stage of development, is designed for forecasting extreme precipitation events. A neural network is trained with features from a ECMWF large-scale model and observations from the rainfall network, to predict the occurrence of an extreme precipitation event and its cumulative within 3 hours up to 48 h ahead. The network thus designed is able to improve the prediction of the raw model at any point of the ECMWF model grid, surpassing, at the level of classification and regression indices, the local-scale non-hydrostatic model MOLOCH.

How to cite: Carnevale, D., Cassola, F., Cavaiola, M., and Mazzino, A.: Lightning and precipitation forecast by means of hybrid deterministic and AI-based tools, 12th European Conference on Severe Storms, Utrecht, The Netherlands, 17–21 Nov 2025, ECSS2025-155, https://doi.org/10.5194/ecss2025-155, 2025.

P60

ECSS2025-188

The Sensitivity of Severe Convective Storm Mode to Updraft Initiation Morphology Across Large-scale Environments

Kristen Axon, Daniel Dawson II, Richard Thompson, Andy Dean, and Edward Mansell

Severe convective storm modes (SCSMs) are responsible for the majority of severe hazards such as tornadoes, hail, and damaging straight-line winds. This work aims to distinguish the role that the initial morphology of updrafts may play in determining SCSM and investigate sensitivities across various large-scale environments (LSEs). Supercells, quasi-linear convective systems (QLCSs), and mixed modes (supercells and QLCSs occurring concurrently in the same region), are examined using idealized simulations in Cloud Model 1. With 250-m grid spacing, ensembles are developed that use different initial updraft patterns that represent real-world setups (e.g. isolated updrafts, scattered clusters of updrafts, broken linear updrafts, and hybrid (broken linear updrafts with scattered clustered updrafts to the east)). LSEs representative of each storm mode are found using the Smith et al. (2012) severe convective mode database that cover a range of convective available potential energy and deep-layer shear regimes. Updraft initiation is accomplished through multiple ellipsoidal regions of upward acceleration centered at 1.5 km above the lowest model level with vertical radii of 1.5 km. The horizontal shape, location, maximum magnitudes in upward acceleration, and both the start time and duration of the forcing regions are randomly varied to mimic real world variability in convective initiation. Preliminary results indicate that the dominant SCSM is not considerably sensitive to initial updraft morphology; however, sensitivities occur with storm longevity and intensity likely due to the nature of cell interactions. These sensitivities will be further explored through additional experiments that modify the magnitude of the deep-layer shear.

How to cite: Axon, K., Dawson II, D., Thompson, R., Dean, A., and Mansell, E.: The Sensitivity of Severe Convective Storm Mode to Updraft Initiation Morphology Across Large-scale Environments, 12th European Conference on Severe Storms, Utrecht, The Netherlands, 17–21 Nov 2025, ECSS2025-188, https://doi.org/10.5194/ecss2025-188, 2025.

Discussion Post a comment

P61

ECSS2025-200

Using an OSSE to Estimate How Hundreds to Thousands of UAS Can Improve NWP Forecasts of Severe Storm Environments

Shawn Murdzek, Therese Ladwig, Adam Houston, and Eric James

Uncrewed aircraft systems (UAS) have been shown to be useful in severe storm research field campaigns. Given the success of UAS observations in these more targeted deployments, a possible next step is to explore the utility of routine UAS observations in severe storm forecasting. In an attempt to estimate the impact of routine UAS observations on severe storm NWP forecasts, we designed an observing system simulation experiment (OSSE). The benefit of the OSSE approach is that a model run is used as the truth rather than the real atmosphere, which allows us to test different configurations of routine UAS observations without having to deploy hundreds to thousands of UAS. Our OSSE uses a 1-km WRF run over CONUS that loosely follows a week in the spring of 2022 as the nature run, which is our surrogate for reality. Synthetic observations are then sampled from the nature run and assimilated into a variant of the prototype Rapid Refresh Forecast System (RRFS). Using this setup, a variety of UAS network configurations are tested. The primary experiments consist of UAS flying vertical profiles every hour up to 2 km AGL with average spacings between UAS sites of 150, 100, 75, and 35 km (which ranges from 347 to 6335 UAS). We find that for severe storm environments (simply defined here as nature run grid points with MUCAPE > 50 J kg-1), assimilating UAS observations improves the bulk verification statistics for various severe weather environmental parameters, such as MLCAPE, MLCIN, and 0–1 km SRH, with skill increasing as more UAS observations are assimilated. For the experiment with the most UAS observations, reductions in root-mean-squared errors (RMSE) for severe weather environmental parameters can exceed 25% for the analysis time for certain parameters. We also find that bulk verification statistics do not change linearly as more UAS are included. For example, the curve of the number of UAS versus RMSE tends to “flatten out” as more UAS are added, indicating that individual UAS have a smaller impact on the forecast as more UAS are assimilated. Altogether, these results suggest that a routine network of vertically profiling UAS can substantially improve forecasts of severe weather by improving the representation of severe storm environments.

How to cite: Murdzek, S., Ladwig, T., Houston, A., and James, E.: Using an OSSE to Estimate How Hundreds to Thousands of UAS Can Improve NWP Forecasts of Severe Storm Environments, 12th European Conference on Severe Storms, Utrecht, The Netherlands, 17–21 Nov 2025, ECSS2025-200, https://doi.org/10.5194/ecss2025-200, 2025.

P62

ECSS2025-205

Evaluating sub-kilometric simulations in the HARMONIE-AROME model on high-impact convective storms

Juan Jesús González-Alemán, Carlos Calvo-Sancho, Pablo Fernández-Castillo, Daniel Martín-Pérez, Samuel Viana, Javier Calvo, David Suárez, and César Azorín-Molina

Very high-resolution and sub-kilometric numerical weather simulations are becoming increasingly common, driven by advances in computational power and atmospheric modeling techniques. These finer-scale models offer the potential to more accurately represent small-scale processes, such as deep convection and storm dynamics, which are often poorly resolved in coarser operational models. As this modelling capability grows, it is crucial to rigorously evaluate the performance of these high-resolution simulations, particularly in reproducing high-impact weather events.

In this study, we assess the capability of sub-kilometric simulations to realistically represent severe convective storms over different regions of Spain. We focus on different selected high-impact weather cases where the operational HARMONIE-AROME model, running at a horizontal resolution of 2.5 km, failed to adequately simulate convective storm development. In contrast, the sub-kilometric simulations successfully captured key storm features, offering improved representation of storm intensity, structure, and evolution. The selected cases span a diverse range of convective phenomena, including supercells, mesoscale convective systems, and ordinary thunderstorms.

Additionally, we aim to support the development of robust conceptual models for forecasting severe convection. These models are grounded in the detailed output from high-resolution simulations and are intended to enhance the forecaster’s ability to anticipate and interpret complex storm behavior in real-world scenarios.

How to cite: González-Alemán, J. J., Calvo-Sancho, C., Fernández-Castillo, P., Martín-Pérez, D., Viana, S., Calvo, J., Suárez, D., and Azorín-Molina, C.: Evaluating sub-kilometric simulations in the HARMONIE-AROME model on high-impact convective storms, 12th European Conference on Severe Storms, Utrecht, The Netherlands, 17–21 Nov 2025, ECSS2025-205, https://doi.org/10.5194/ecss2025-205, 2025.

P63

ECSS2025-206

Cloud-radiation coupling over the lifetime of deep convective storms

Wouter Mol, Blaž Gasparini, and Aiko Voigt

Deep convective storms have a large influence on both solar and thermal radiation due to their initially optically thick updrafts and anvils, and later, in their mature and dissipating stage, thin but large anvil coverage. A storm's impact on the top of atmosphere net radiation is relevant for Earth's climate. More directly, radiation also plays a role on further storm evolution and later development by surface heating and affecting the thermal structure of the atmosphere. Our question is whether mid-level clouds and cloud overlap play a role in storm evolution and total cloud radiative effect over the storm's lifetime. For example, there are regions, such as (sub)tropical West Africa, where mid-level clouds near the freezing level are frequently observed and thought to originate from congestus detrainment or as remnants of previous deep convection. One hypothesis is that cloudiness around the freezing level affects anvil cirrus lifetime and local stability sufficiently to influence a convective storm's net radiative effects as a whole. We use high resolution modelling to investigate whether mid-level cloudiness and overlap with cirrus is sufficiently resolved compared to observations. Secondly, we test our hypothesis with a mechanism denial experiment. If confirmed, this cloud-radiation coupling needs to be well represented in models for weather forecasting (storm evolution, next-day storm environment) and climate simulations (net radiation balance).

How to cite: Mol, W., Gasparini, B., and Voigt, A.: Cloud-radiation coupling over the lifetime of deep convective storms, 12th European Conference on Severe Storms, Utrecht, The Netherlands, 17–21 Nov 2025, ECSS2025-206, https://doi.org/10.5194/ecss2025-206, 2025.

P64

ECSS2025-218

A novel Deep Learning framework for lightning probabilistic prediction based on ERA5 reanalysis data and lightning observations.

Adrien Burq, Greta Cazzaniga, Davide Faranda, Mathieu Vrac, Victor Xing, and Jean Jouhaud

Traditional methods for studying thunderstorm climatology and associated hazards often rely on analyzing long-term trends of key environmental variables. These variables are derived from low-resolution reanalysis datasets like ERA5 and are aggregated over extended time periods (Taszarek et al., 2021). The methods also derive probability of occurrence of convective events, providing useful insights into large-scale climatological trends (Battaglioli et al., 2023). However, they struggle to capture the fine-scale characteristics of individual events. Their representations tend to suffer from probabilistic smoothing effects—such as overestimating hazard occurrence in regions without activity and underestimating it in areas of high activity—leading to unrealistic distributions of hazards in space, time, and intensity.

Machine learning approaches, including random forests, gradient boosting and more recently deep learning models, have improved short-term lightning nowcasting (McGovern et al., 2023). However, they rely on high-resolution inputs such as radar and satellite data which constrains their usage because of the spatio-temporal limitation of such data. To address this limitation, we develop a deep learning model tailored for reanalysis data which enables us to apply the model on a global scale and over a much larger period of time to study the evolution of thunderstorm activity.

Given a state of the atmosphere, we generate an ensemble forecast of lightnings at an hourly time resolution and 0.25° spatial resolution. Compared to previous models (Battaglioli et al., 2023, Saha et al., 2025), we use 3D inputs in our model to directly output a map of lightning with statistically coherent spatial structures. Additionally, we make ensemble predictions to capture a wide range of possible realistic scenarios for a given set of thermodynamic and dynamic variables.

Technically, we develop a guided diffusion model that learns to generate lightning maps at an hourly timestep over western Europe. Given 3 maps of environmental conditions (representing instability, humidity and wind-shear) derived from the ERA5 reanalysis and an ensemble of 2D noise data, we generate an ensemble of possible lightning maps.

Because of the large temporal availability of ERA5 data, our model will enable us to detect changes in past thunderstorm activity.

How to cite: Burq, A., Cazzaniga, G., Faranda, D., Vrac, M., Xing, V., and Jouhaud, J.: A novel Deep Learning framework for lightning probabilistic prediction based on ERA5 reanalysis data and lightning observations., 12th European Conference on Severe Storms, Utrecht, The Netherlands, 17–21 Nov 2025, ECSS2025-218, https://doi.org/10.5194/ecss2025-218, 2025.

Discussion Post a comment

P65

ECSS2025-221

Severe storm research with thundeR package - improvements in calculation procedures of convective parameters

Mateusz Taszarek, Cameron Nixon, Tomas Pucik, Piotr Szuster, Pieter Groenemeijer, Francesco Battaglioli, and Bartosz Czernecki

ThundeR rawinsonde processing package has been under development since 2017. It is a freeware R language package for sounding and hodograph visualization, and rapid computation of convective parameters commonly used in the research and operational prediction of severe convective storms. Ability of the package to calculate more than 300 parameters in ~1 centisecond enables rapid processing of large numerical datasets. Over the recent years thundeR has been applied on global reanalysis datasets, operational numerical weather prediction models, and used to study environments associated with global lightning observations and severe weather reports from Europe, North America, South America and Australia. ThundeR also contributed to development of ESSL’s AR-CHaMo models and was applied by severe storm scientists in several countries, including national hydrometeorological institutes. Construction of environmental datasets collocated with severe storms observations from different parts of the world served as a platform to evaluate the skill of hundreds of convective parameters in predicting specific convective hazards, and allowed to test new parameter ideas through an iterative development process. In this work we will discuss modifications in calculation procedure of existing parameters, which led to more skillful identification of environments supportive of lightning, large hail, tornadoes and severe wind of both severe and significant severe intensity. Examples concern modifications in the calculation procedure of convective inhibition, lifted index or storm-relative helicity, and introduce new ventilation parameter along with two new parcel lifting types: most-unstable-mean-layer (MUML) and most-unstable-above-500m (MU5). Authors will also present examples of how obtained results can be applied in the operational prediction of severe convective storms and modeling their climatology on the global scale.

How to cite: Taszarek, M., Nixon, C., Pucik, T., Szuster, P., Groenemeijer, P., Battaglioli, F., and Czernecki, B.: Severe storm research with thundeR package - improvements in calculation procedures of convective parameters, 12th European Conference on Severe Storms, Utrecht, The Netherlands, 17–21 Nov 2025, ECSS2025-221, https://doi.org/10.5194/ecss2025-221, 2025.

P66

ECSS2025-235

Machine Learning Methods for the Postprocessing and Seamless Blending of Ensemble Forecasts

Fabian Schubert and Cristina Primo

Numerical weather prediction (NWP) models can deliver high-resolution forecasts, but achieving optimal performance requires the fine-tuning of model parameters and adjusting the physical models for different spatial, temporal, and seasonal scopes. Models differ in forecast accuracy for specific lead times and seasonal environments, offering a variety of fragmented forecasts exhibiting temporal discontinuities. This happens with the two regional ICON variants for Germany, ICON-D2 and ICON-RUC, which are operational at DWD and differ in their physical models and forecast range. Although both operate on a 2 km scale, ICON-RUC focuses on convection, is more computationally expensive, and it provides shorter lead times (+14 hours for ICON-RUC vs. +48 hours for ICON-D2). Therefore, blending is necessary for a seamless forecast for up to 48 hours. Moreover, NWP systems are biased and postprocessing is required for calibration and error correction.
As part of the SINFONY 3.0 project at DWD, our goal is to use machine learning methods to develop a postprocessing framework that delivers an improved combined probabilistic forecast that is calibrated and seamlessly blends the output data from both NWP models, with a focus on hourly precipitation. DWD’s radar network measurements is used as the ground truth for training.
We build on previous work that utilizes ML to improve probabilistic forecasts: Grönquist et al. [1] used deep U-Nets for bias and spread estimation. Rempel et al. [2] focus on the blending of ensemble nowcasting and ensemble NWP, and also show that their network can improve calibration. Primo et al. [3] generate calibrated probabilistic distributions using a neural network and include additional contextual data features such as lead time, seasonal, and orographic parameters to improve predictions. This work presents a neural network that blends both NWP systems and provides a calibrated probability.

[1] Peter Grönquist, Chengyuan Yao, Tal Ben-Nun, Nikoli Dryden, Peter Dueben, Shigang Li, and Torsten Hoefler. Deep learning for post-processing ensemble weather forecasts. Philos. Trans. A Math. Phys. Eng. Sci., 379(2194):20200092, April 2021.
[2] Martin Rempel, Peter Schaumann, Reinhold Hess, Volker Schmidt, and Ulrich Blahak. Adaptive blending of probabilistic precipitation forecasts with emphasis on calibration and temporal forecast consistency. Artificial Intelligence for the Earth Systems, 1(4), October 2022.
[3] Cristina Primo, Benedikt Schulz, Sebastian Lerch, and Reinhold Hess. Comparison of model output statistics and neural networks to postprocess wind gusts. arXiv [stat.AP], January 2024.

How to cite: Schubert, F. and Primo, C.: Machine Learning Methods for the Postprocessing and Seamless Blending of Ensemble Forecasts, 12th European Conference on Severe Storms, Utrecht, The Netherlands, 17–21 Nov 2025, ECSS2025-235, https://doi.org/10.5194/ecss2025-235, 2025.

P67

ECSS2025-254

Fewer but More Intense: Changes in Extreme Precipitation Cells from Global Kilometer-Scale Climate Modeling

Fabian Senf, Leonie Hartog, and William Jones

Earth system modeling is currently undergoing an exciting transformation, thanks to new technical capabilities that allow for significant spatial refinement. For the first time, these capabilities allow us to explicitly simulate extreme precipitation and its effects on climate-relevant timescales on a global scale. Thus, new Earth system data from high-resolution modeling approaches offer an exciting foundation for new analyses and research. In our study, we examine the distribution and changes in extreme precipitation from global simulations. We obtained this data from the ICON Earth system model simulations conducted within the nextGEMS project, which aims to create future projections up to the year 2050 with a grid spacing of approximately 5 km. Our analysis focuses on the portion of precipitation contributing to the top ten percent of globally accumulated precipitation. Using the open-source tool tobac we identify and track the resulting precipitation cells over time. Our analysis reveals that warming causes the most extreme precipitation cells to become more intense. At the same time, the data shows a significant decrease in the total number of cells, resulting in fewer, more intense extremes. Finally, we discuss these findings in relation to changes in the spatial distribution of the cells and changed environmental conditions.

How to cite: Senf, F., Hartog, L., and Jones, W.: Fewer but More Intense: Changes in Extreme Precipitation Cells from Global Kilometer-Scale Climate Modeling, 12th European Conference on Severe Storms, Utrecht, The Netherlands, 17–21 Nov 2025, ECSS2025-254, https://doi.org/10.5194/ecss2025-254, 2025.

P68

ECSS2025-260

Combining NWP and Observations with AI

Sarah Heibutzki, Thomas Deppisch, Ulrich Blahak, Jan Keller, Stefanie Hollborn, and Roland Potthast

The SINFONY Project (seamless integrated forecasting system) aims to enable seamless forecast between the Nowcasting (NWC) on the one hand and numerical weather prediction (NWP) on the other. The advantages of both systems are that NWP is reliable for longer lead times and NWC is a fast product. Whereas the disadvantages are that NWP is computational expensive and NWC is unreliable for longer lead times.

For example, in convective processes, NWP may not capture convective features on small scales and NWC may not capture the evolution of convective dynamics. A solution to this problem would be to combine the information provided by NWP with the recent data from observational systems and NWC. This is where the new methods of AI in weather forecasting and data assimilation can help us.

In our work we examine the application of the AI-Var algorithm (proposed by J. Keller and R. Potthast, arXiv:2406.00390) to convective scale weather forecasting. This algorithm allows for a fast calculation of the analysis state given a forecast and (nowcasted) observations. For this reason, we investigate how the temporal evolution of uncertainties can be included in the AI-Var algorithm.

More precisely, we show first conceptual results of a reformulation of the AI-Var algorithm. In this approach we are able to include time dependent background error correlations (“error of the day”). For example, we apply the algorithm to 2m-temperature and precipitation. In the future we plan to further include observations such as radiation, wind gusts, visibility, ceiling (clouds) and others.

How to cite: Heibutzki, S., Deppisch, T., Blahak, U., Keller, J., Hollborn, S., and Potthast, R.: Combining NWP and Observations with AI, 12th European Conference on Severe Storms, Utrecht, The Netherlands, 17–21 Nov 2025, ECSS2025-260, https://doi.org/10.5194/ecss2025-260, 2025.

P69

ECSS2025-270

The Microphysics of Severe Storms as Seen by a Lagrangian Super-Particle Model: ICON/McSnow

(withdrawn)

Fabian Jakub, Christoph Siewert, and Axel Seifert

P70

ECSS2025-276

An Idealized Parameter Study of Instability Maxima on the Cool Side of Airmass Boundaries

Jason Keeler and Adam Houston

The cool side of an airmass boundary can, paradoxically, consist of greater surface-based CAPE than the warm side of the boundary. This condition arises when the cool-side airmass is sufficiently moist relative to the warm-side airmass, with this phenomenon referred to as a Mesoscale Airmass with High Theta-E (MAHTE). This terminology is used since MAHTEs can also be identified using equivalent potential temperature and have a typical cross-frontal dimension of ~15 km. MAHTEs have the potential to enhance the severity of thunderstorms through their interaction with this region of increased CAPE, lower LFC, and greater low-level vertical wind shear. Recent climatological analysis determined that MAHTEs frequently occur in both coastal and continental regions, motivating an idealized parameter study of MAHTE development and characteristics near the coast and in areas with a uniform land surface type. While this climatological analysis was performed for the United States, characteristics of regions where MAHTEs are most common there suggests that MAHTEs likely occur frequently within Europe as well, particularly in southern coastal regions and in the lee of mountains.

Analysis of the coastal simulations indicates that the primary mechanism for MAHTE development is a relative decrease in moisture within the warm-side airmass through entrainment of dry air from above the boundary layer. MAHTE development is also supported through increased radiative forcing, which enhances the vigor of thermals, and increased water surface temperature, which increases moisture content in the cool-side airmass. This talk will expand on these earlier results through discussion of a series of CM1 large-eddy simulations whose domains are set in a continental location, with a broader parameter space including the initial temperature difference across the airmass boundary, radiative forcing, surface fluxes, moisture content above the boundary layer, and land surface type. Ultimately, the process-based understanding achieved through this study will improve our ability to forecast MAHTE development and characteristics over a broad parameter space.

How to cite: Keeler, J. and Houston, A.: An Idealized Parameter Study of Instability Maxima on the Cool Side of Airmass Boundaries, 12th European Conference on Severe Storms, Utrecht, The Netherlands, 17–21 Nov 2025, ECSS2025-276, https://doi.org/10.5194/ecss2025-276, 2025.

P71

ECSS2025-306

Simulation of Canadian severe weather events using Cloud Model 1

Lisa Schielicke, Luna Awad, and Keya Raval

We present a high-resolution modeling study of Canadian severe storms using Cloud Model 1 (CM1), with a focus on tornado- and hail-producing convective events. A curated dataset of 20 severe weather cases, 10 tornado and 10 hail events, across Canada, drawn from the Northern Tornadoes and Northern Hail Projects, forms the basis for the convection-resolving simulations. Initial conditions are constructed using thermodynamic profiles from ERA5 global reanalysis data. An ensemble of simulations is performed to evaluate the representativeness of the ERA5 vertical profiles and to identify potential limitations in using reanalysis data as ground truth. The ensemble includes the original ERA5 profiles and modified versions, as well as variations in convective initiation mechanisms and different microphysics schemes. Simulations are conducted on the high-performance computing system of the Shared Hierarchical Academic Research Computing Network (SHARCNET)/Digital Research Alliance of Canada. Key outputs include an integrated analysis of storm structure, dynamics, intensity, and evolution, along with comparisons to observed data. Results are visualized and made available through an interactive dashboard. This work provides new insight into the meteorological conditions of Canadian severe weather events and establishes a reproducible framework for convection-resolving ensemble storm simulations tailored to regional observational data. We plan to extend this work to more cases in the future.

How to cite: Schielicke, L., Awad, L., and Raval, K.: Simulation of Canadian severe weather events using Cloud Model 1, 12th European Conference on Severe Storms, Utrecht, The Netherlands, 17–21 Nov 2025, ECSS2025-306, https://doi.org/10.5194/ecss2025-306, 2025.

P72

ECSS2025-310

An open educational approach to teaching convection-resolving modeling with Cloud Model 1 (CM1)

Lisa Schielicke, Rayhan Abdul Rahman, Luna Awad, Oliver Heuser, Yidan Li, Keya Raval, Jerome Schyns, Jose Pablo Solano Marchini, Aaron Sperschneider, Patrick Zobec, and Christoph Gatzen

We present our experiences from a two-week educational block course, first conducted at the University of Bonn during the 2023 winter semester, which introduced students to the nonhydrostatic, time-dependent, convection-resolving Cloud Model 1 (CM1). The course provided hands-on training in configuring and running CM1 simulations on a high-performance computing cluster, offering participants practical experience in the numerical modeling of moist convection. An introduction to three-dimensional visualization tools enabled students to transform simulation output into graphical representations, facilitating the interpretation of cloud dynamics.

Pre- and post-course surveys demonstrated significant gains in students’ understanding of atmospheric processes and in transferable skills such as high-performance computing and data visualization. The course was structured in two parts: the first covered core concepts, while the second allowed students to apply their knowledge to independent research projects. Initially designed for meteorology students with a strong background in atmospheric science, the course is now being adapted for physics students at Western University. As part of this transition, new learning materials are being developed, and preliminary outcomes will be presented.

The course has already led to several bachelor’s and master’s thesis projects, as well as undergraduate research experiences focused on severe convective storms in Canada. Many of the students involved in these projects have contributed as co-authors to the present work. All course materials are available as open educational resources (Schielicke, L., January 2024: Cloud Model 1 & Visualization-A Block course. ResearchGate, http://dx.doi.org/10.13140/RG.2.2.30017.12642).

How to cite: Schielicke, L., Abdul Rahman, R., Awad, L., Heuser, O., Li, Y., Raval, K., Schyns, J., Solano Marchini, J. P., Sperschneider, A., Zobec, P., and Gatzen, C.: An open educational approach to teaching convection-resolving modeling with Cloud Model 1 (CM1), 12th European Conference on Severe Storms, Utrecht, The Netherlands, 17–21 Nov 2025, ECSS2025-310, https://doi.org/10.5194/ecss2025-310, 2025.