OSA1.1 | Forecasting, nowcasting and warning systems
Forecasting, nowcasting and warning systems
Conveners: Timothy Hewson, Yong Wang | Co-conveners: Bernhard Reichert, Fulvio Stel
| Tue, 05 Sep, 09:00–16:00 (CEST)|Lecture room B1.05, Wed, 06 Sep, 09:00–10:30 (CEST)|Lecture room B1.05
| Attendance Tue, 05 Sep, 16:00–17:15 (CEST) | Display Mon, 04 Sep, 09:00–Wed, 06 Sep, 09:00|Poster area 'Day room'
Orals |
Tue, 09:00
Tue, 16:00
This session presents and explores the increasingly sophisticated systems developed to aid, and often automate, the forecasting and warning process. The rapid proliferation of data available, including probabilistic and rapidly-updating NWP as well as a plethora of observations, combined with a growing appreciation of user needs and the importance of timely and relevant forecasts, has brought the development of these systems to the fore. The opportunities afforded by the WMO's HiWeather programme will also be discussed in this session.

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
• Use of machine learning and other advanced analytic techniques

Orals: Tue, 5 Sep | Lecture room B1.05

Chairperson: Yong Wang
Onsite presentation
Seppo Pulkkinen and Heikki Myllykoski

Convective storms and the associated heavy rainfall, flooding, hail, wind gusts and lightning can result in significant damage to property and loss of lives. Thus, there is a need for accurate prediction of the future location and severity of such storms (i.e. in the sub-kilometer resolution for the next hour) to assist the decision making of civil protection authorities. The typical approach to produce such nowcasts is to identify storm cells as separate entities from radar images, which provides a natural way for associating the storm attributes with its severity. Following this approach, we have implemented a set of pan-European nowcast products in the TAMIR and EDERA projects funded by the EU Civil Protection Mechanism. This has been done by combining a cell tracking method with a random forest-based machine learning (ML) model and a Kalman filter model. In the nowcast products, storm cells are identified from the OPERA radar composites. The ML model is used for predicting the storm severity level. It is trained using a large database of meteorological features and weather hazard reports during summer months (May-September) between 2018-2022. The storm features include basic cell and track properties (i.e. area and age), lightning flash and wind observations, and also indicators of convective potential from ERA5 reanalyses. The target variable for training the ML model is the storm hazard level. The hazard level estimation is done based on the distances and time delays between the storms and the associated weather hazard reports obtained from the European Severe Weather Database (ESWD). We have trained different ML models against the hazard levels estimated for each event type (e.g. heavy rain, lightning and severe wind gusts). The above models are combined with a Kalman filter-based methodology to produce probabilistic nowcasts of future storm locations together with their severity level. The added value of such nowcasts for decision making is demonstrated with case studies and relevant verification metrics. Finally, we demonstrate how the nowcasts can be combined with different exposure layers to translate them into predictions of actual storm impacts.

How to cite: Pulkkinen, S. and Myllykoski, H.: Machine Learning-Based Nowcasting of Convective Storm Impacts on a Pan-European Scale, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-375, https://doi.org/10.5194/ems2023-375, 2023.

Onsite presentation
Robert Warren, Harald Richter, Ivor Blockley, and Dean Sgarbossa

Operational guidance for thunderstorms and severe convective hazards at the Australian Bureau of Meteorology has undergone a major uplift in recent years, with the introduction of new NWP models, post-processing systems, and convection diagnostics. The latest guidance is based on the third generation of the Australian Community Climate and Earth-System Simulator (ACCESS) suite, which includes global and convection-allowing models run in both deterministic and ensemble configurations. In this presentation, we will introduce the ACCESS suite and discuss three sources of convective guidance based on its outputs. The first is the new ConvParams post-processing suite, which ingests model-level data from the ACCESS Global and Global Ensemble models and computes a wide array of convective parameters for use in ingredients-based forecasting of thunderstorms and associated hazards. The second is the Bureau’s lightning prediction system, Calibrated Thunder, which combines ACCESS Global Ensemble forecasts and recent lightning observations to produce calibrated probabilistic forecasts of lightning within a 10-km radius across Australia and surrounding coastal waters. The third comprises storm attributes from the convection-allowing ACCESS City and ACCESS City Ensemble models, including simulated reflectivity, updraft helicity, and parameterised lightning flash rates. For the ensemble, these diagnostics are post-processed to obtain the ensemble maximum and neighbourhood-maximum ensemble probabilities (NMEPs), and interrogated using a variety of novel visualisation strategies. Our presentation will provide an overview of each of these guidance streams, describe how they are used in operations, and assess their strengths and limitations. We will also highlight promising avenues for future guidance developments at the Bureau of Meteorology.

How to cite: Warren, R., Richter, H., Blockley, I., and Sgarbossa, D.: New convective guidance at the Australian Bureau of Meteorology, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-178, https://doi.org/10.5194/ems2023-178, 2023.

Online presentation
Yu-Shen Chen and Li-Pen Wang

Object-based radar rainfall nowcasting is a widely-used technique for convective storm prediction. Due to the data and algorithmic limitations, most existing object-based nowcasting methods focus on predicting the movements of each rain object (or cell). The evolution of rain cells’ properties (e.g. cell size, shape and intensity) themselves is often neglected. It is however critical to account for the temporal changes in cells’ properties in order to improve the predictability for convective storms. 


In the literature, three-dimensional (3D) radar images have been used for observing the vertical feature changes through the formation process of convective rain cells. This shows the potential of extracting useful information from 3D images to facilitate characterising the life cycle of rain cells. Most of these works however focused on analysing or reconstructing the life cycles of individual convective rain cells or storm events. It remains an open challenge to incorporate 3D radar rainfall information into object-based radar rainfall nowcasting. 


In this research, we would like to explore the use of deep learning techniques to predict the evolution of convective rain cells. The proposed work comprises two main parts. The first part is rain cell data preparation. An enhanced TITAN storm tracking algorithm proposed by Muñoz et al. (2018) is employed to identify 2D rain cells and their temporal associations (or tracks) across successive time steps. The information of 2D cells are then used to extract cell properties from 3D radar images. These include mean reflectivity, area, major and minor axis lengths and the convective core altitude of each rain cell. In the second part of the work, a LSTM-Encoder-Decoder model is developed, which uses cells’ properties from the past 15 min to predict the evolution of these properties in the next 15 min. 


A total of 4708 lifespans of rain cells extracted from high-resolution (5-min, 1 km, 24 levels) 3D radar images are used to train the model, and a total of 1177 extracted lifespans are used to validate the prediction result. The result suggests that the proposed LSTM-Encoder-Decoder model can well predict the evolution of cells’ properties, and, with the employed 3D information (core altitude), the prediction errors of mean reflectivity can be further reduced by 20-25% at 15-min forecast lead time.  



Muñoz, C., Wang, L.-P., and Willems, P. (2018). Enhanced object-based tracking algorithm for convective rain storms and cells. Atmospheric Research, 201:144–158.


How to cite: Chen, Y.-S. and Wang, L.-P.: Exploring the use of 3D radar measurements in predicting the evolution of convective rain cells, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-399, https://doi.org/10.5194/ems2023-399, 2023.

Onsite presentation
Nico Bader, Hartwig Deneke, Matthias Tesche, and Andreas Macke

Thunderstorms pose a large risk for human safety and can strongly affect various economic sectors by their ground effects such as flash floods, hail, and lightning occurring during severe thunderstorm events. With a reliable thunderstorm forecast, people and vulnerable sectors can be warned of this danger in advance through warning systems.

A thunderstorm forecast based on NWP models is not sufficient since convection develops quickly and on a subgrid-scale and cannot be fully resolved. To achieve a more reliable warning system for the next few hours, nowcasting applications based on observations are of particular interest. State-of-the-art nowcasting systems based on radar data have their limitations as convective cells cannot be detected before the onset of precipitation.

An improvement of lead time for convective cell detection can potentially be achieved through geostationary satellite information, which is the focus of the present study.

Satellite-derived cloud properties are calculated along storm tracks, and their correlation with the subsequently derived convective severity is analyzed. 28 convective cells over Germany in 2021 are examined and used to conduct a correlation analysis focusing on the strength and significance of the correlations. Based on these correlations, linear regression models are trained and verified by using an additional validation set consisting of eight cases. Using the multi-spectral radiance data from the Spinning Enhanced Visible and Infrared Imager (SEVIRI) aboard the geostationary Meteosat Second Generation (MSG) satellite, dynamical and microphysical cloud-top properties are derived to define the convective state during convective initiation (CI). Convective severity is assessed by precipitation properties determined from radar data from the Radar Climatology (RADKLIM) of the German Weather Service (DWD), as well as lightning properties derived from lightning data from Vaisala.

The lead time of the cell detection based on these geostationary satellite data is found to depend on the threshold for the detection process. An improvement against radar data can be seen for a brightness temperature threshold of 260 K. Maximum 5 minute precipitation intensity shows significant correlations with CI conditions, and can be well predicted with a mean absolute error (MAE) of 0.24 mm. With a significant Pearson-R of -0.88, the maximum lightning amplitude also shows good predictive skill with a MAE of 44 kA. The predictions of the time of maximum precipitation intensity, time of maximum lightning frequency, and time of first lightning are associated with high uncertainties and no significant correlation.

These results show that geostationary satellite data can provide an earlier detection of convective cells. Furthermore, it has the potential of providing information on the future development of convective cells. All in all, the integration of satellite-based information in nowcasting systems can enable more reliable early warnings and an enhanced assessment of the severe weather potential of thunderstorms.

How to cite: Bader, N., Deneke, H., Tesche, M., and Macke, A.: Can Convective Initiation Provide Indicators for Convective Severity?, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-522, https://doi.org/10.5194/ems2023-522, 2023.

Onsite presentation
Manuel Baumgartner, Guido Schröder, and Cristina Primo

Forecasting the occurrence of thunderstorms is a well-known challenge in weather prediction. Since a thunderstorm is by definition accompanied by at least one lightning, we aim to forecast the occurrence of at least one lightning within a pre-defined area and a pre-defined time-interval. Numerous prior research studies on forecasting and detecting lightnings provide a rich database of model diagnostics and observational data. One example of such a model diagnostic is the so-called (subgrid-scale) Lightning Potential Index (LPI), that was recently implemented in the operational ICON-model (1) and is now available operationally, even in the ICON-EU ensemble. On the other hand, there are extensive observation networks that provide lightning observations, such as the Linet-network (2) that provides lightning observations over Europe. 

In our work, we use lightning observations from the Linet-network as ground truth and establish a translation from LPI to lightning probabilities. For this task, we trained neural networks to predict the desired lightning probabilities in a "global postprocessing mode", i.e. using the same network for the forecasts on the whole domain of ICON-EU, which is significantly larger than the domain of the Linet-network. We present the setup of the postprocessing method together with details of the training of the neural networks and show first results from its forecasts and their evaluation. In particular, these lightning probabilities outperform raw model probabilities as derived by counting the exceedance of a LPI-threshold and thereby avoid the need to even define such a threshold. We will also address aspects of producing stable operational forecasts by applying an ensemble of neural networks.



(1) Schröder et al., 2022: Subgrid scale Lightning Potential Index for ICON with parameterized convection. Reports on ICON (10), DOI: 10.5676/DWD_pub/nwv/icon_010

(2) Betz et al., 2009: LINET - an international lightning detection network in europe. Atmos. Res., 91, 564–573, DOI: 10.1016/j.atmosres.2008.06.012

How to cite: Baumgartner, M., Schröder, G., and Primo, C.: Forecasting lightning probabilities derived from the Lightning Potential Index using neural networks, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-34, https://doi.org/10.5194/ems2023-34, 2023.

Online presentation
Jun Xu and Kan Dai

Heavy rainfall, generated by multi-scale processes and characterized by small spatial scale, sudden occurrence, low predictability and strong disaster-causing, is the bottleneck of short-time forecast in China. At present, the grid resolution of quantitative forecast on precipitation in national quantitative precipitation forecast operation is only 5km, which could not meet the requirements of fine forecast for disaster prevention and reduction. Recent study in Europe and the United States shows that the high resolution short-time probabilistic forecast with advantages in heavy rainfall forecast could be generated by using the deep generative models to correct the error and improve the resolution of the low resolution numerical model forecast. In view of the above facts, based on the European Centre for Medium-Range Weather Forecasts Integrated Forecasting System (IFS) forecast data and China Meteorological Administration Land Surface Data Assimilation System (CLDAS) observed precipitation gridded data in 1km resolution, the short-time probabilistic precipitation forecast was studied in 1km resolution forecast by using the deep generative model in Eastern China. Results show that the short-time probabilistic forecast could be effectively generated by the deep generative model with the forecast feature inputs associated to the formation of heavy rainfall. Meanwhile, the critical success index and fractions skill score of 3 hours accumulation precipitation above 20mm could be greater than the model forecast. Case studies showed that the forecast generated by the deep generative model could effectively improve the forecast of the location and intensity of heavy rainfall, which showed promising application in the future in forecast operation in China.

How to cite: Xu, J. and Dai, K.: Short-time forecast on heavy rainfall in Eastern China using the deep generative models, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-597, https://doi.org/10.5194/ems2023-597, 2023.

Coffee break
Chairperson: Timothy Hewson
Blending / Modelling
Onsite presentation
Gavin Evans, Bruce Wright, Stephen Moseley, and Benjamin Ayliffe

Blending weather forecasts from different sources aims to generate a single forecast which is temporally seamless, spatially consistent and more skilful than the individual inputs. This forecast can capitalise on all available forecast sources throughout the lead time range, such as extrapolation-based nowcasts and convection-permitting ensemble models at shorter lead times, and coarser ensemble models at longer range. Optimising the individual models contributing to the blend using post-processing techniques like neighbourhood processing and calibration helps to optimise overall forecast skill.

A multi-model blend can be created either in physical space or in probability space depending upon the desired output format and the interaction of multi-model blending with other processing steps. The IMPROVER codebase (https://github.com/metoppv/improver) utilises multi-model blending in probability space, which results in spatially and temporally smooth probability and percentile forecasts that are ideal for some use cases. Blending in probability space can avoid artefacts that can be difficult to overcome when blending in physical space. The aim for generating physically consistent realizations from the multi-model blend in IMPROVER is therefore to retain the benefit of the forecast source-specific processing and the relative ease of the probability space blending. Physically consistent realizations are desired by multiple users, including as inputs to hydrological models that require physical consistency including between diagnostics, such as, precipitation and snow melt.

This presentation outlines the requirements for physically consistent realizations, including as input for hydrological models. Prior work focusing on the usage of Ensemble Copula Coupling (ECC) and Schaake Shuffle will be reviewed and challenges with implementing ECC where the choice of dependence template is not obvious will be discussed. Proposals and initial findings will be presented for creating realizations that are self-consistent spatially, across lead times and across diagnostics following multi-model blending in probability space. 

How to cite: Evans, G., Wright, B., Moseley, S., and Ayliffe, B.: Strategies for generating physically consistent realizations from a multi-model blend, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-194, https://doi.org/10.5194/ems2023-194, 2023.

Onsite presentation
Gavin Evans, Fiona Rust, Benjamin Ayliffe, and Ben Hooper

Creating a forecast that is seamless across time yet is optimal at each forecast validity time is often achieved by blending forecasts from multiple Numerical Weather Prediction models (or using other forecast sources, such as an extrapolation nowcast). With the increasing usage of convection-permitting ensemble models at shorter lead times, the blending of these forecasts with longer range ensemble models with parameterised convection can lead to a clear transition from one forecast source to another. This is particularly noticeable when visualising the evolution of the gridded forecast. Calibrating the forecast sources with a common truth prior to blending provides a method of improving forecast skill whilst also unifying the characteristics of the forecasts to create a smoother blend throughout the evolution of the forecast.

This presentation aims to describe a non-parametric method, utilising tools from the Met Office’s IMPROVER codebase (https://github.com/metoppv/improver), for calibrating the reliability of the forecast without degrading the forecast resolution. This approach is assessed for its usability for gridded precipitation rate and total cloud amount forecasts. Reliability is markedly improved resulting in similar skill between forecast sources during the blending period and therefore extends the lead time range at which the forecast is more skilful than climatology. This approach is also presented as a step within a series of steps to improve forecast skill therefore highlighting that this approach can be complementary to other techniques without significant tuning. Further refinements to the Reliability Calibration technique removed artefacts in the gridded forecasts. Caveats, including a reduction in sharpness following calibration, are also presented.

How to cite: Evans, G., Rust, F., Ayliffe, B., and Hooper, B.: Improving the blend of multiple weather forecast sources by Reliability Calibration, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-224, https://doi.org/10.5194/ems2023-224, 2023.

Onsite presentation
Lionel Moret, Christoph Spirig, Colombe Siegenthaler, Jonas Bhend, Daniele Nerini, Matteo Buzzi, Mathieu Schaer, and Mark Liniger

On a daily basis, MeteoSwiss provides a wide range of automatic weather forecasts to the general public, the aviation and to private customers. These data are provided by an ensemble of heterogeneous individual system wich forces the end-user to choose between different sources and sometimes to combine them despite a limited knowledge of their quality and shortcomings. In addition, the forecasts of the different systems are provided in different formats and through different channels, making combined use even more difficult.

Furthermore, from a scientific point of view, combining nowcasting and post-processing approaches in a single step using statistical and/or machine learning methods has been shown to give the best forecast performance for twelve hour precipitation forecasts Deep learning for twelve-hour precipitation forecasts. Nat Commun 13, 5145 (2022)). In such an approach, there is no need for a separate system for the nowcasting range, and therefore no need to combine different forecasts a posteriori, which requires further assumptions and introduces further source of errors and inconsistencies.

MeteoSwiss has therefore initiated a project to build a system that will integrate data from different weather observation and weather forecasts data (ECMWF, MeteoSwiss regional ICON implementation) in order to provide a consolidated and easily accessible weather forecast dataset. We aim as well for probabilistic gridded forecasts that are seamless in space and time which can be used by a wide range of application, like hydrological models, automatic generation of warning proposals for forecasters, probabilistic animation of precipitation for the MeteoSwiss App. This project is very much oritented toward end-users and a large effort will be made to assess and meet their needs.

How to cite: Moret, L., Spirig, C., Siegenthaler, C., Bhend, J., Nerini, D., Buzzi, M., Schaer, M., and Liniger, M.: Toward seamless weather forecasts., EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-109, https://doi.org/10.5194/ems2023-109, 2023.

Onsite presentation
Julie Thérèse Pasquier, Johannes Rausch, Lukas Umek, Christian Schluchter, Alexander Schlauch, and Martin Fengler

Accurate and precise weather forecasting is essential for a wide range of applications and industries, from agriculture to transportation to renewable energy. However, current weather models often struggle to represent atmospheric processes accurately due to limitations in spatial resolution. To fill this gap, Meteomatics has developed the EURO1k model, the first pan-European weather model with a spatial resolution of 1 km.

The EURO1k model grid covers 4250 x 4580 km and is run in a rapid-update setup with 24 initializations per day, each run with a lead time of 24 hours. It is based on the Weather Research and Forecasting (WRF) model and uses global ECMWF-IFS model data for boundary conditions. In addition to standard data sources such as weather stations, radar and satellite data, and radiosondes, the EURO1k model also assimilates data from a network of Meteodrones, unmanned aircraft systems (UAS) developed by Meteomatics which collect vertical atmospheric profiles up to 6000m in altitude.

The high spatial and temporal resolution of the EURO1k model allows to accurately represent small-scale weather patterns, resulting in highly accurate and precise forecasts. Indeed, statistical analyses of EURO1k model output against observations from 5000 weather stations in Europe demonstrate increased accuracy compared to other global and regional models. This has important implications for industry and the public. The EURO1k model improves the forecasting of extreme weather events, allowing for better preparation and response. It also enhances the prediction of renewable energy production, thereby improving their cost efficiency and accelerating the energy transition. And, most importantly, it provides a more accurate and reliable weather forecast for communities across Europe. Overall, the EURO1k model represents a major advance in numerical weather prediction, bringing improved understanding and forecasting of the weather to a wide range of users.

How to cite: Pasquier, J. T., Rausch, J., Umek, L., Schluchter, C., Schlauch, A., and Fengler, M.: EURO1k: A high-resolution European weather model developed by Meteomatics, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-486, https://doi.org/10.5194/ems2023-486, 2023.

Online presentation
Joseba Egaña, Jaione Munarriz, Jon Ander Arrillaga, Antonio Castaño, and Santiago Gaztelumendi

A wind event is analysed during the passage of a depression over the Bay of Biscay. From the northwest of the Iberian Peninsula, it moves in a northeasterly direction until it finally reaches the south of the British Isles. In the upper layers, the flow is strong from the southwest. This configuration generates intense southwesterly winds, especially in central hours and especially in Alava, with Vitoria-Gasteiz being one of the towns most affected. Wind gusts of more than 120 km/h are recorded in exposed areas and more than 100 km/h in non-exposed areas, for example at the Gasteiz station a wind gust of 124 km/h is recorded (third highest record in the Gasteiz station data series). In addition, the high temperatures, the strong wind and the state of the vegetation mean that the risk of forest fires is high and, in fact, a major fire takes place in the Balmaseda area (the most significant fire in Bizkaia in the last 30 years). Due to the wind, there were numerous incidents caused by falling trees, obstacles on the road, falling electricity and telephone poles, flight diversions, etc.

The wind gusts initially forecasted were exceeded. Wind gusts between 80 and 100 km/h were expected in non-exposed areas, but these values are exceeded punctually in many stations. The data from the stations are analysed and compared, and the factors that could have caused the wind to increase over than expected are studied.

The synoptic and mesoscale factors of this episode are analysed. The information from Basque Country Automatic Weather Station Mesonetwork is reviewed, analysing the spatial-temporal evolution of the wind at different locations. After analysing the situation, different conclusions and key factors to be taken into account from the point of view of forecasting are extracted.

How to cite: Egaña, J., Munarriz, J., Arrillaga, J. A., Castaño, A., and Gaztelumendi, S.: An analysis of a extreme wind event in the Basque Country : the 23 october 2022 case, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-262, https://doi.org/10.5194/ems2023-262, 2023.

Onsite presentation
Iris Odak Plenkovic, Ena Kozul, and Ivan Vujec

The error reduction of NWP wind speed forecast is a rather expected effect when an analog-based statistical post-processing method (ABM) is applied. The efficiency of ABM, naturally, directly depends on the exact setup. For instance, the wind direction NWP is proven to be the most important predictor variable in the analog search process, in addition to the wind speed itself. The properly weighted additional predictor variables improve results even further, especially in complex terrain. Additionally, since the post-processing methods can be prone to underestimate the intensity of extreme events, the correction for high wind speed is implemented as a valuable tool, affecting the ABM when the raw NWP forecast exceeds a certain value.

After several upgrades were recently made in the operational DHMZ suite, these upgraded ABM forecasts are thoroughly evaluated, taking into consideration the overall performance. Since wind speed often reaches gale values during the bora and jugo wind episodes in the coastal region of Croatia, the effect on rare event forecasts is also considered, especially in complex terrain prone to high wind speed (e.g., bora wind).

Results show that the error, in general, is larger in complex terrain prone to high-wind speed than in less complex continental areas, and the reduction of error by ABM is thus more pronounced. In addition to improvement for the climatologically more common events, it can also be achieved for rare events in some cases as well. This is especially the case after applying the statistical correction which enlarges the frequency of such forecasts. The large errors occurred mainly due to short-term weakening of the wind speed within the strong wind episode, or because of the time delay of the beginning or the ending of the forecasted strong wind episode.

How to cite: Odak Plenkovic, I., Kozul, E., and Vujec, I.: Evaluation of analog-based post-processing with a special focus on strong wind episodes, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-238, https://doi.org/10.5194/ems2023-238, 2023.

Online presentation
Statistical post-processing of Wind Speed Gridded Forecasting Based on Deep Learning
Yang Xuan, Dai Kan, and Zhu Yuejian
Online presentation
Ruixia Zhao, Kan Dai, Yong Cao, and Yong Wang

Improved Surface Wind Speed Forecasts over Beijing-tianjin-hebei Region of China during Spring by Random forests Approach with Sliding-Time-Window and Region Regression[1]

Ruixia Zhao1, Yong1, Yong Cao1, Yong Wang2

1 National Meteorological Centre, Beijing, China

2 Nanjing University of Information Science and Technology, Nanjing, China

Strong winds are among the most significant natural hazards, posing great threats to transportation, construction, agriculture, and even the safety of people's lives. Therefore, the accuracy of wind speed forecasting is concerned very much. In our study, machine learning (ML)-based solutions are developed to reduce forecast errors of 10m wind speed produced by the ECMWF’s Integrated Forecasting System (IFS). Two ML approaches, namely decaying averaging method (DAM) and random forest decision trees (RF), are tested at 1985 stations in Beijing-Tianjin-Hebei (BTH) region during the spring of 2021. Considering the importance of computation efficiency in daily operation and the increasing demand for guidance forecast at any specific locations, an so-called sliding-time-window and region regression RF (SR-RF) is designed by pooling the data from sliding-time-windows in recent two years and the whole target region to train the regression models which can be applied at any points within the region. The different models for different lead times of 3h intervals within 72h and 6h intervals within 240h are designed to be daily updated. SR-RF method shows a significant excellent ability to capture the characteristics of IFS errors with 25% to 46% performance improvements in terms of average absolute error (MAE) for all lead times, which is much higher than 7% to 20% improvements of DAM. In particular, the SR-RF method demonstrates its outstanding performance advantages significantly outperforming DAM by dramatically reducing the large forecast errors of IFS over the high-terrain areas in western and northern region, as well as the eastern coastal regions, and overcoming the weakness of excessive strong wind prediction over BTH region in IFS. Furthermore, taking into account of the importance of elevation, latitude, and longitude predictors second only to 10m wind speed in feature importance analysis, and the excellent performance of SR-RF forecast, it is demonstrated that the SR-RF method designed in this study can learn a good knowledge of error distribution characteristics of numerical weather prediction models under different locations and terrain environments, and can learn the downscaling law well while improving the accuracy of prediction.

[1]Supported projects: National Key R&D Program Project (2021YFC3000903), China Meteorological Administration Key Innovation Team Project (CMA2022ZD04)

Author Introduction: Ruixia Zhao, mainly engaged in statistical post-processing of numerical weather prediction model and developing the operational objective global weather forecasting system (CMA-GOWFS).

E-mail: zhaorx@cma.gov.cn / 122323497@qq.com


How to cite: Zhao, R., Dai, K., Cao, Y., and Wang, Y.: Improved Surface Wind Speed Forecasts over Beijing-tianjin-hebei Region of China during Spring by Random forests Approach with Sliding-Time-Window and Region Regression, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-537, https://doi.org/10.5194/ems2023-537, 2023.

Lunch break
Chairpersons: Bernhard Reichert, Timothy Hewson
Warning Systems
Onsite presentation
Irina Mahlstein, Lea Beusch, Lionel Moret, Saskia Willemse, and Mark Liniger

In the framework of the renewal of the warning system at MeteoSwiss, we use the possibility to redesign the software as well as the scientific approach of generating warnings. The production chain is designed such that the data is guided and refined along the pathway. Here, we present the operational production process currently in development for the next generation warning system. The first step is the combination of all available data into one data stream (seamless weather). The second stage prepares warning proposals for the forecasters. The only human interaction with the system (third step) is the one of the forecaster in case of extreme weather. However, the automatically generated warning proposals aim to minimize the time needed by the forecaster to issue a warning. In a last step at the end of the chain, the warning products are customized and distributed to our customers. Furthermore, the warnings will also be verified automatically to monitor the quality of the system.

Along the chain, we run into a number of challenges, which ask for clever solutions: How do we group grid cells with similar extreme weather information into meaningful warning polygons? How do we facilitate the interaction of the system with the forecaster? Which aspects of the warning do we verify? How do we judge the quality of the warning? Finally yet importantly, what does the warning that we issue actually mean? If we do not have a common understanding about what we are expecting to happen within a warning polygon a clear communication of uncertainties and measuring the quality of our warnings is impossible.

Hence, the system involves new applications based on specific developments aiming at generating the greatest value for public warnings. Most of the warning chain is purely machine driven; nonetheless, the human interaction remains a key aspect of the new warning system.

How to cite: Mahlstein, I., Beusch, L., Moret, L., Willemse, S., and Liniger, M.: Next generation warning production system at MeteoSwiss, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-190, https://doi.org/10.5194/ems2023-190, 2023.

Onsite presentation
Lea Beusch, Irina Mahlstein, Daniele Nerini, Urs Graf, Jonas Bhend, Saskia Willemse, and Mark A. Liniger

At MeteoSwiss, we are currently developing a modular framework to automatically identify extreme weather events that will allow us to generate a broad range of user-tailored warning products. Here, we introduce this Extreme Weather Identifier (EWI) framework in the context of public warnings, namely warning polygon proposals for forecasters to support them in issuing public warnings of high quality in a timely manner.

The EWI translates NWP ensemble forecasts into warning products – in our case, warning polygon proposals – by employing a pre-defined sequence of processing steps with configurable parameters. By applying grid-point-specific warning thresholds to each NWP ensemble member, it obtains local-scale warning information and by aggregating the underlying information in space and time, it accounts for spatio-temporal representativeness issues and facilitates communication. In a first aggregation step, the spatio-temporal representativeness issues are addressed by employing neighborhood approaches to detect extreme weather in each ensemble member individually. Subsequently, all members are evaluated jointly in order to assess the probability that the extreme weather actually takes place. Afterwards, areas exceeding a minimal probability threshold are grouped together into individual regionally-valid warning polygons. At this point, the communication-motivated aggregation to the visual scale of interest starts and all further changes simply serve the goal to produce warning products that can be easily communicated to their target audience without any additional physical justifications.

To allow the forecasters to obtain an in-depth understanding of the EWI’s proposals and thoroughly assess their quality, not only the proposals themselves will be distributed but also outcomes of key intermediate steps of the EWI’s processing sequence. Products covering the EWI’s processing steps until the communication-motivated aggregation starts are intended to be made available to forecasters in real-time towards the end of this year and we will illustrate them in this contribution with examples from past warning events.

How to cite: Beusch, L., Mahlstein, I., Nerini, D., Graf, U., Bhend, J., Willemse, S., and Liniger, M. A.: An extreme weather identifier for severe weather warnings: general concept and first results, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-317, https://doi.org/10.5194/ems2023-317, 2023.

Onsite presentation
Kathrin Feige, Bodo Erhardt, and Renate Hagedorn

One of the most important tasks of the German Meteorological Service (DWD) is to issue weather warnings. In the current operational warning system, forecasters interpret a multitude of meteorological data with respect to a set of threshold-based warning criteria. Warning regions are then manually identified and sent to an automatic post-processing chain, which generates the final warning products to be distributed in relevant communication channels.

Even though DWD’s weather warnings are mostly perceived well by its end-users, there are some drawbacks to the current system. They include a short lead time of warnings, a complex catalog of warning criteria, and missing flexibility towards specialized user requirements. To tackle these shortcomings, DWD launched a program called RainBoW (“Risikobasierte, anwendungsorientierte, indiviualisierbare Bereitstellung optimierter Warninformationen” or “Risk-based, application-oriented and individualizable delivery of optimized weather warnings” in English) to optimize its warning system.

RainBoW focusses on three fields of action. First, the forecast horizon of warnings will be extended up to 7 days into the future to inform users early on, while also communicating the uncertainties resulting from larger lead times. Second, the comprehensibility of warnings will be enhanced by reducing the complexity of the warning criteria catalog and by taking weather impacts into account. Third, warnings will be made individualizable. This means, that users with specific requirements, e.g. in terms of warning thresholds and/or considered areas, will get the possibility to configure individual warnings matching their particular use case and their individual meteorological thresholds. These individualized warnings will be generated automatically based on user-created warning profiles. 

The three fields of action serve RainBoW’s overarching goal to tailor warnings more strongly towards the needs of end-users, such that they are enabled to take appropriate action in case of significant and extreme weather.

This contribution will describe the conceptual ideas behind RainBoW along with some first results.

How to cite: Feige, K., Erhardt, B., and Hagedorn, R.: Developing a new warning system at the German Meteorological Service, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-675, https://doi.org/10.5194/ems2023-675, 2023.

Onsite presentation
Ilian Gospodinov and the Teams from the National Institute of Meteorology and Hydrology of Bulgaria

The National Institute of Meteorology and Hydrology is the national weather service of Bulgaria. It has been operating a national weather early warning system for more than 10 years. The national system has been developed in accordance with the principles of the European system METEOALARM built within the international cooperation facilitated by the EUMETNET – a network of European national hydro-meteorological services. The color warning scale of METEOALARM gained popularity in the country and is currently easily recognizable by the wider public. Originally the national weather early warning system was develop to provide warnings with coarser resolution at the level of administrative region – 28. The advantage is that such warnings can be directly used for decision making by the civil protection authorities that often have territorial branches with zones of responsibility matching the bigger administrative regions. The disadvantage is that the natural extreme weather events are of scale smaller than that of an administrative region or occur in a part of it. This is a source of potential misunderstanding that requires detailing by text or specific consultation with a forecaster if useful decision is to be made upon. The national weather early warning system consists of three steps: 1. Automatic extraction of weather warnings by administrative regions from the regional numerical weather prediction model of NIMH - the result is offered to the weather forecaster in duty as first guess; 2. Evaluation of the first guess and and weather warning issuance by the forecaster in duty; 3. Dissemination to authorities and the general public by website, email, radio. The most recent development of the system was to go to finer resolution and provide weather warning for municipalities – 265. The advantage is that it allows for better match in scale between natural phenomena and the area of the municipality. The disadvantage is that it is much harder to produce by the forecaster. This is where the automatic first guess from model becomes essential. Sea storm early warnings for the Black sea coast for dangerous wave height has also been added during the last years.

How to cite: Gospodinov, I. and the Teams from the National Institute of Meteorology and Hydrology of Bulgaria: Development of the Bulgarian national weather early warning system, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-606, https://doi.org/10.5194/ems2023-606, 2023.

Onsite presentation
Matti Kämäräinen, Ari-Juhani Punkka, and Ilona Láng-Ritter

Accurately predicting the impacts of severe weather events is crucial for improving meteorological warnings and aiding weather-vulnerable organizations in decision making. However, achieving a high-quality estimate of impacts is difficult because of complex and detail-sensitive interactions between the weather and the impacted quantity. For instance, forest damages caused by a storm are influenced not only by the strength of gusts but also by factors such as ground frost, tree leaf maturity, and wind direction. 


To address this challenge, the SILVA project developed a nation-wide weather-impact database with over 10 individual impact datasets, which were then utilized as targets for machine learning modeling. Storm damage clearance tasks, wildfire fighting tasks, traffic accidents, and pedestrian slipping accidents stored in the database were aggregated to the counties of Finland, and a separate gradient boosting model was fitted between the historical ECMWF HRES weather forecasts and the impact data in each county.


Subsequently, a reliable production system was set up to produce forecasts automatically twice per day. Four risk classes were determined based on the extremeness of the forecasted incidents, taking into account the seasonal differences, and the Meteoalarm-like four step color scheme was used to visualize the forecasts as traditional warning maps and as time series. Two of the products primarily describe weather risks in summer (storm damage clearance tasks and wildfire fighting tasks), while the other two are more useful for warning the risks in winter (traffic accidents and pedestrian slipping accidents). 


The impact forecasts were tested by numerous end-users during a seven-month pilot phase. Both the feedback from the pilot participants and the numerical validation results clearly indicate the value of the products. The selected modeling method, gradient boosting, was found to be effective in taking into account the nonlinear and complex interactions when explaining the variability of the impact data. 


The forecasting system is currently undergoing further development in the Europe Horizon CREXDATA project, with efforts focused on exploring possibilities for expanding the geographical areas of application and incorporating new impact datasets.

How to cite: Kämäräinen, M., Punkka, A.-J., and Láng-Ritter, I.: Machine learning forecasts of storm damages, forest fires, pedestrian slippings, and vehicle accidents for operational early warnings of weather impacts, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-539, https://doi.org/10.5194/ems2023-539, 2023.

Onsite presentation
Gabriele Messori, Stephen Jewson, and Sebastian Scher

Ensembles have revolutionized weather forecasts, by providing a probabilistic view of future weather. However, ensembles contain a large amount of information, and using this to understand the potential impacts of the forecasted weather requires a skilled user with extensive computational resources. This challenge is particularly acute for users considering the weather conditions over a geographical region, rather than at a single location. In these cases, considering every single ensemble member and its impacts may be practically unfeasible. Many such users thus simply consider the ensemble mean and some measure of ensemble spread, such as the ensemble’s standard deviation. While this facilitates the use of ensemble forecasts, it does not explore the range of possible impacts of the forecasted weather. Here, we propose a framework facilitating the use of ensemble forecasts for weather impacts. We specifically represent the ensemble by the mean and a single deviation from the mean. This deviation is defined so as to both be representative of the variability in the ensemble, and have a significant impact according to some impact metric. We determine such a deviation using a statistical method known as Directional Component Analysis, which is based on linearizing an impact metric around the ensemble mean. We provide a concrete example using 2-m temperature forecasts for continental Europe from ECMWF, and show that this approach is more robust than considering the single worst (in terms of impacts) ensemble member. This same approach can be applied to ensembles of projections of future climates. We illustrate this by deriving representative deviations for the UKCP18 and EURO-CORDEX projections of future precipitation in Europe. We conclude that the mean and representative deviation method we propose may both contribute to automated early warnings for weather impacts and support users who wish to explore the implications of longer-term climate impacts in a resource-effective fashion.

How to cite: Messori, G., Jewson, S., and Scher, S.: Representative impact scenarios from weather and climate ensembles, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-2, https://doi.org/10.5194/ems2023-2, 2023.

Onsite presentation
Angelos Chasiotis, Nastos Panagiotis, and Elissavet Feloni

Climate change affects the severity and frequency of natural disasters, mainly because the positive trends in global surface temperatures increase the possibility of more droughts and also the increased intensity of rainfall leads to intense flood phenomena.

The Municipality of Ermionida, located in the Argolis regional unit (Peloponnese, Greece), is an area that only slightly affected by the climate crisis. The area is particularly dry, and suffers from several flash flood events especially during autumn and early winter months, thus, there is a need for a forecasting system operation at local scale.

In the frame of this work, SMILE project’s objectives (which is funded by Greek Government) are presented, as well as, the general scheme of a proposed tool that is equipped with a monitoring system is described in detail. SMILE system for the Municipality of Ermionida is a user friendly online tool, designed with the scope of monitoring and processing data from connected sensors, i.e., stage records from hydrometric stations installed on torrents and meteorological parameters from stations installed on several areas in the watershed. This on-line tool allows the user to access data from a central screen, and to create specific diagrams per parameter and per station. The system-involved sensors are connected to a datalogger with internal 4G modem, for the real-time monitoring and inter-operability with an additional 1D/2D hydraulic model that will be created in the frame of the same research project for critical areas, with the scope of issuing warnings.

Key Words: Civil protection, Meteorological phenomena, Flooding, flashfloods, early warning system

How to cite: Chasiotis, A., Panagiotis, N., and Feloni, E.: Designing a smart early warning hydrometeorological system for Greek Municipality of Ermionida, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-607, https://doi.org/10.5194/ems2023-607, 2023.


Orals: Wed, 6 Sep | Lecture room B1.05

Chairpersons: Fulvio Stel, Bernhard Reichert
Short Range Forecasting / Other Topics
Online presentation
Alexander Kann and the WP4-Team of Destination Earth On-Demand Extreme (DE_330)

In phase 1 of Destination Earth On-demand Extremes, the capabilities of an on-demand digital twin designed for extreme weather phenomena is currently implemented. The DT workflow consists of an automatic event detection module which triggers a hyperresolution NWP model run on a flexible domain and resolution (depending on the phenomena), followed by error-correction and uncertainty estimation with post-processing techniques and impact models for different applications (renewables, air quality, hydrology). The post-processing framework, as part of the production chain, exploits existing event detection methods by searching for precursors of expected extremes, e.g. applying EFI and SOT index or storm tracking methods and investigates their applicability for triggering the on-demand workflow. Downstream the on-demand hyperresolution NWP model run, post-processing procedures are tailored to specific applications, event type, region and data availability, and complement the hyperresolution model simulations by error correction and probability estimation. These method are flexibly implemented, depending on the availability of NWP data and observational data for training purposes and will include baseline algorithms like neighborhood approach for areas with sparse observations as well as deep learning methods for regions with large samples of observation data. The presentation gives an overview of the current implementation status and its methods and techniques. First demonstrators will be discussed on behalf of selected events and use cases, including technical and scientific challenges. The usability of the hyperresolution NWP model and downstream applications for key users in Civil Protection, Marine Safety and Agriculture will be highlighted and perspectives for future achievements will be outlined.

How to cite: Kann, A. and the WP4-Team of Destination Earth On-Demand Extreme (DE_330): Post-processing activities in Destination Earth On-demand Extremes, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-519, https://doi.org/10.5194/ems2023-519, 2023.

Onsite presentation
Jan Bondy, Christian Berndt, Markus Schultze, and Ulrich Blahak

Small-scale, convective heavy rainfall events present a major challenge in flood forecasting, primarily affecting smaller catchments. At such scales, flood forecasts are challenging due to short response times, a lack of stream gauges and limitations of operationally used hydrological models. These models are typically designed for larger catchments, therefore we explore alternative strategies to support the German flood forecasters.

The AREA product (Areal Rainfall Extremity Assessment) aims at complementing the flood forecasters’ workflow based on hydrological modeling by rapidly identifying catchments affected by strong rainfall. We intend to provide a novel post-processing product containing information about the extremity of catchment-specific areal rainfall. First, a nationwide catchment delineation is performed for each pixel of a 50 m x 50 m digital elevation model, selecting only catchments smaller with an area between 10 and 500 km². The catchments and stream network are subsequently upscaled to the operational radar grid with a resolution of 1 km. Finally, areal rainfall is computed for each pixel using the underlying catchment geometry from rain gauge-adjusted radar observations as well as seamless rainfall forecasts resulting from the SINFONY project with lead times of up to 12h.

In order to estimate the extremity of a catchment rainfall event in real-time, we derive return periods based on extreme value statistics. To that end, we consider various accumulation durations and perform a recalculation of areal rainfall for all catchments using a 20-year dataset of radar-derived rainfall. Given the short and limiting observation period of radar data, we attempt to combine the obtained extreme value distributions with existing, regionalized long-term rain gauge statistics (DWD-KOSTRA), in order to estimate longer return periods.

The resulting spatial distribution of areal rainfall with corresponding return periods based on an exhaustive collection of catchments appears to be a useful visualization technique to identify small catchments affected by heavy rainfall. The product will be illustrated based on the analysis of specific case studies.

How to cite: Bondy, J., Berndt, C., Schultze, M., and Blahak, U.: Real-time extremity assessment of rainfall observations and SINFONY forecasts for small hydrological catchments, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-269, https://doi.org/10.5194/ems2023-269, 2023.

Onsite presentation
Arne Spitzer, Harald Kempf, Matthias Jerg, Manuel Werner, and Ulrich Blahak

Since July 2020 the DWD WarnWetter-App comprises the Crowdsourcing module “User Reports”. This module provides users the functionality to report observations about current weather conditions and severe weather to DWD and other users.

The user reports represent the current meteorological conditions at a certain place at a certain point of time. The Crowdsourcing module provides 10 different meteorological categories (lightning, wind, hail, rain, wet icy conditions, snowfall, snow cover, cloudiness, fog, tornado), each of which contains specific characteristic levels and optionally additional attributes. In addition, the user has the option of setting the location and time of the event manually.

The benefit of the data is that meteorological information at ground level is collected at places where no weather station is located in the immediate vicinity. The dataset is able to complement the existing synoptic station network. Forecasters from DWD already benefit from user-based observations that are available in near real-time.

In recent years, a new nowcasting algorithm has been developed at DWD, called KONRAD3D. The algorithm aims to automatically detect, track, and nowcast convective cells in order to support DWD’s warning management.

KONRAD3D uses three-dimensional radar reflectivity data as main input. In addition, also lightning data and information about hydrometeor types based on polarimetric radar data is regarded. In particular, in the latest version KONRAD3D features the new hail flag - a warning parameter that assesses a cell’s threat of hail. The new parameter rests upon the hydrometeor data and should roughly estimate the expectable near-ground hail size. Other features of KONRAD3D are the gust flag – a warning parameter that estimates the maximum speed of wind gusts - and the heavy rain flag which assesses the potential of heavy rain.

This is where the crowdsourcing data comes into play. Observations from app users are able to confirm expected hail sizes on the ground and provide promptly information about wind gusts and rain intensity. Preliminary results show that KONRAD3D tends to overestimate hail and underestimate gusts and heavy rain. Our analyses will show, in which cases the warning parameter estimates were reasonable and at which point the user reports could complement the real-time operation of KONRAD3D.

How to cite: Spitzer, A., Kempf, H., Jerg, M., Werner, M., and Blahak, U.: DWD-Crowdsourcing: Are User Reports beneficial for Object-based Nowcasting?, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-273, https://doi.org/10.5194/ems2023-273, 2023.

Online presentation
Marc Rautenhaus, Andreas Beckert, Kameswarrao Modali, and Thorwin Vogt

Visualization is an important and ubiquitous tool in the daily work of atmospheric researchers and weather forecasters to analyse data from simulations and observations. Visualization research has made much progress in recent years, for instance, with respect to techniques for ensemble data, interactivity, 3-D depiction, and feature-detection. Met.3D is an open-source research software aiming at making novel interactive, 3-D, feature-based, and ensemble visualization techniques accessible to the meteorological community (code repository, conda package, and documentation available at https://met3d.wavestoweather.de). Since its first public release in 2015, Met.3D has been used in multiple visualization research projects targeted at atmospheric science applications, and has evolved into a feature-rich visual analysis tool facilitating rapid exploration of atmospheric simulation data. The software is based on the concept of “building a bridge” between “traditional” 2-D visual analysis techniques and interactive 3-D techniques powered by modern graphics hardware. It allows users to analyse data using combinations of feature-based displays (e.g., atmospheric fronts and jet streams), “traditional” 2-D maps and cross-sections, meteorological diagrams, ensemble displays, and 3-D visualization including direct volume rendering, isosurfaces and trajectories, all combined in an interactive 3-D context.

Met.3D has in recent years been advanced within the German research projects “Waves to Weather (W2W)” and “Climate, Climatic Change, and Society (CLICCS)”. In this presentation, we will present the current state of the Met.3D software and discuss recent updates we consider beneficial for the weather forecasting community, including use of open forecast data and interactive visual analysis of forecast 3-D cloud fields and other volumetric data.

How to cite: Rautenhaus, M., Beckert, A., Modali, K., and Vogt, T.: Interactive 3-D visualization for rapid exploration of numerical weather prediction data – recent updates to Met.3D, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-599, https://doi.org/10.5194/ems2023-599, 2023.

Onsite presentation
Irene Schicker, Markus Dabernig, Alexander Kann, and Maria Meingast

Road weather conditions, especially road temperature, have a major impact on road safety even more so in case of unusual early/late snow events leading to damages to logistic infrastructure, road infrastructure, and fatalities. Road maintenance services use meteorological forecasts as well as targeted road temperature and precipitation forecast to estimate when, where, and how often they have to treat roads before and during such events happen and rely, thus, on accurate and targeted predictions.

Numerical weather prediction models are able to provide a good guesstimate but still lack the detail, temporal and spatial resolution, which is needed especially in regions with rugged terrain. Here, we take a two-fold approach in (i) implementing different post-processing methods, deterministic and probabilistic, and (ii) evaluate the skills of the separate models and a combined multi-model ensemble approach. Furthermore, simplistic transfer learning approaches are implemented to test the models’ skills in unobserved areas. For step (i) the following methods are used: (a) multilinear regression, (b) Kalman filtering, (c) random forest, (d) a feed forward neural network, (e) a transformer neural network, (f) the EMOS model, and (g) the Metro model.

Two Austrian regions are considered here, namely Tyrol and the state of Salzburg. Results for these regions show that the simplistic transfer learning approaches are only in a few cases better than a temperature height correction approach. The Kalman filter results show that a careful parameter selection is needed to achieve good results. They rely, too, on onsite measurements and cannot be applied to regions with no measurements. All other methods are able to improve the raw NWP forecasts indicating that overall a mix of method is better suitable than relying on one single method (deterministic/probabilistic).

How to cite: Schicker, I., Dabernig, M., Kann, A., and Meingast, M.: Probabilisitic road temperature forecasting using machine learning techniques – comparison of a multi-model approach in the Austrian Alps, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-471, https://doi.org/10.5194/ems2023-471, 2023.

Online presentation
Joseba Egaña, Virginia Palacio, Jon Ander Arrillaga, and Santiago Gaztelumendi

In this work we analyse one of the most intense heat episodes ever happened in the Basque Country. This historic event produces numerous absolute record. with the highest maximum and minimum temperatures, as well as the highest maximum and minimum values for a month of July. In addition, numerous fires are produced during this period. A red alarm for extreme high temperatures has been set for the first time (note that colour-coded warnings have been implemented in Euskalmet case in 2009).

The worst is between the 16th and 18th, although the period of high temperatures lasts from the 11th to the 18th. From the 16th to the 18th many stations exceed 40ºC. On the 16th, the highest temperatures are recorded in the south part, specifically at the Zambrana station with 41.7 ºC. On the 17th the intense heat is repeated, with temperatures rising in several areas, especially in the Cantabrian area of Alava. During this day, values above 42 ºC are recorded at several stations and above 40 ºC at a large number of them. Gardea and Saratxo reach 42.9 ºC and Ordunte 42.4 ºC respectively. The maximum temperatures are above 40 ºC in many stations in Alava, including Abetxuko in Vitoria (40.1 ºC). The 18th is another day of suffocating heat. With a prevailing southerly wind, temperatures rise on the Cantabrian slope, especially in Gipuzkoa, and 11 stations reach or exceed 42 ºC. In Gardea, 43.6 ºC is recorded, the highest temperature value recorded during the whole episode. In Arrasate and Saratxo the temperatures are also around 43 ºC, with 42.9 ºC and 42.7 ºC respectively.

Due to the heat, there are numerous incidents. One person dies of heatstroke. According to the Basque Government, health services treat a total of 184 people for conditions related to the high temperatures, 83 of whom are taken to hospital.

In this work, the synoptic and mesoscale factors of this episode are analysed. The information from Basque Country Automatic Weather Station Mesonetwork is reviewed, analysing the spatial-temporal evolution of the temperatures at different locations. Finally we include some conclusions taking into account other previous persistent and extreme temperatures episodes and some impact information.

How to cite: Egaña, J., Palacio, V., Arrillaga, J. A., and Gaztelumendi, S.: Analysis of July 2022 persistent and extreme high temperatures episode in the Basque Country, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-267, https://doi.org/10.5194/ems2023-267, 2023.

Posters: Tue, 5 Sep, 16:00–17:15 | Poster area 'Day room'

Display time: Mon, 4 Sep 09:00–Wed, 6 Sep 09:00
Chairperson: Timothy Hewson
Ziqiang Huo, Pu Liu, and Yong Wang

Over recent decades the deterministic and probabilistic NWPs have been improved significantly. It becomes the essential toll for the meteorological operation and applications. It is very often that there are several deterministic NWPs and EPSs with different resolution available for meteorological operation and applications. Those forecasts are with different characteristics of systematic bias and dispersion errors. Many statistical calibration methods have been proposed and been implemented in the operation, for example, ensemble model output statistics (EMOS) and standardized anomaly model output statistics (SAMOS). Further, Artificial intelligence (AI) based method has been used in different way for calibration.  In this study we applied EMOS and SAMOS to calibrate multi-scale deterministic and probabilistic forecasts. In the frame of SAMOS/EMOS we have introduced AI based methods for selecting the important variables and building the non-linearity for calibration. The CMA(China meteorological Administration) NWP model chain, a convection permitting NWP (3km resolution), a regional NWP (9km) and a global NWP (25km), a regional EPS (10km) and a global EPS (50km) have been used for the calibration. Two years observation and NWP data over Beijing region was selected for training the EMOS/SAMOS method.  EMOS and SAMOS, AI based variable selection and Boosting method etc. have been compared. 2m temperature, 10m Wind and precipitation forecasts have been calibrated and verified with statistical scores such as, root mean square error of ensemble mean, continuous ranked probability score(CRPS)and so on. The results of calibrated ensemble mean and ensemble spread are quite encouraging, which will be presented at the conference.

How to cite: Huo, Z., Liu, P., and Wang, Y.: Calibrating multi-scale deterministic and probabilistic forecasts, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-524, https://doi.org/10.5194/ems2023-524, 2023.

Lukas Josipovic, Gregor Pante, Andreas Brechtel, and Ulrich Blahak

The precise forecast of convective cells is essential for meteorological services as they can be accompanied by life-threatening severe hail, wind gusts, or heavy rain.

However, state-of-the-art NWP models usually possess update frequencies of several hours so that forecasters must use predictions that are outdated when new thunderstorm cells develop. NWP models do often accurately simulate the intensity of convective cells, but with shifts in space and time.

Object-based nowcasting algorithms with higher update frequencies became necessary to deliver information on the evolution of convective storms for the first two hours since observation. Furthermore, the combination of nowcasting and model data enables the relocation of simulated cells towards observed cells.


Many deterministic object-based nowcasting tools as DWD’s KONRAD3D algorithm assume that detected cells will have persistent intensity.

Within the SINFONY (Seamless INtegrated FOrecastiNg sYstem) project at DWD, we aim at modelling the life-cycles of storm cells in a truthful way and capturing the uncertainties of object-based nowcasts. Hence, we extended our nowcasting algorithm towards an ensemble prediction system called KONRAD3D-EPS. Each ensemble member is initialized by drawing from parameterized distributions of storm lifetime and maximum severity. Inspired by previous studies, e.g. Wapler (2021), KONRAD3D-EPS uses a set of horizontally flipped parabolas to model the life-cycle of convective cells in terms of their severity. In case of redetection of a convective cell, the algorithm corrects the previously estimated lifetime and severity maxima. Thus, the parabolas can be adapted individually for any convective storm in any weather condition.


Besides life-cycle predictions, KONRAD3D-EPS delivers information on the probability of thunderstorm occurrence for the next 2 hours depending on detected cells and their severity. In order to condense the ensemble data, we also provide the representative member for each convective cell. This is done by applying the pseudomember algorithm by Johnson et al. (2020) to the ensemble data.


We will give an overview of our probabilistic object-based nowcasting algorithm KONRAD3D-EPS and present its predictions for prominent example cases. Moreover, we will show first verification results.

How to cite: Josipovic, L., Pante, G., Brechtel, A., and Blahak, U.: Object-based Ensemble Prediction System KONRAD3D-EPS, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-52, https://doi.org/10.5194/ems2023-52, 2023.

Robert Warren, Ivor Blockley, Dean Sgarbossa, and Harald Richter

The Australian Bureau of Meteorology (BoM) has recently operationalized a post-processing suite called ConvParams, which computes a wide array of convective parameters using output from the BoM’s global deterministic and ensemble NWP models. Outputs from the suite include parcel parameters such as CAPE and CIN (computed for a range of different initial parcels), kinematic diagnostics such as bulk wind difference and storm-relative helicity (computed for a range of different atmospheric layers), and composite indices such as the supercell composite parameter and significant tornado parameter. In addition, the suite also identifies important features in the atmospheric profile such as capping inversions and elevated mixed layers. A unique feature of ConvParams, compared to other similar codebases (NSHARP/SHARPpy, MetPy) is its use of high-order polynomials to approximate pseudoadiabatic processes, which permits parcel calculations that are both fast and highly accurate. Significant computational advantages also come from the use of an ahead-of-time compiler (Pythran), which “transpiles” the native Python code into fast C++ code. As well as being used in operations, ConvParams is being run as part of the second-generation BoM Atmospheric Regional Reanalysis for Australia (BARRA2), a regional downscaling of the ERA5 reanalysis, and the BoM Atmospheric Regional Projections for Australia (BARPA), a regional downscaling of CMIP6 climate projections. Once complete, these simulations will provide the most comprehensive picture of historical and future convective environments in Australia to date, supporting major research in this space over the coming years. This presentation will provide an overview of the ConvParams suite and highlight its applications in both operational forecasting and future research endeavours.  

How to cite: Warren, R., Blockley, I., Sgarbossa, D., and Richter, H.: Convective parameters for severe weather forecasting and research in Australia, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-179, https://doi.org/10.5194/ems2023-179, 2023.

Rui Wang, Jimmy Chi-Hung Fung, and Alexis Kai-Hon Lau

Accurate and timely rainfall nowcasting is important for protecting the public from heavy rainfall-induced disasters. In recent years, deep-learning models have been demonstrated to significantly outperform traditional methods in heavy rainfall nowcasting. However, the performance of existing deep-learning-based nowcasting models is still limited by short effective prediction time (< 3 hours), insufficient training data, and the rapid growth of blurriness increases in forecast time. In this work, we propose a novel heavy rainfall nowcasting model based on an innovative task-segmented architecture, namely the TS-RainGAN, consisting of two modules: the MaskPredNet predicts the spatial coverage of different rainfall categories to provide bounding for rainfall with various intensities, and the IntensityGAN predicts the intensity of rainfall based on the rainfall coverage produced by the MaskPredNet. To overcome the data scarcity, we develop another novel model called RainMaker which can generate huge amounts of new radar data based on limited observed radar data. In addition, in order to improve the typhoon-induced precipitation nowcasting, we design a new typhoon structure segmentation method, which distinguishes the typhoon structure into typhoon eye, principal rainband, and outer rainband. The TS-RainGAN training with new AI-synthetic radar data produced by RainMaker can accurately capture the spatiotemporal features and evolutions of rainfall systems and provide skillful precipitation prediction with high skill scores compared with the results of the widely used baseline models. The performance of typhoon-induced precipitation nowcasting shows significant improvement by our innovative typhoon structure segmentation method. Meanwhile, the blurriness of the predicted images is significantly reduced. This enables district-level heavy rainfall nowcasting with competitive forecast skills for up o 6 hours.

How to cite: Wang, R., Chi-Hung Fung, J., and Kai-Hon Lau, A.: Skillful deep learning-based precipitation nowcasting based on new AI-synthetic radar data, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-55, https://doi.org/10.5194/ems2023-55, 2023.

youngsu lee

Korea Institute of Atmospheric Prediction Systems(KIAPS) is currently developing a numerical weather prediction model, including a data assimilation system, to replace the Unified Model(UM). The KIAPS Integrated Model(KIM) consists of a spectral element-based non-hydrostatic dynamical core using a finite-volume method and physics packages. The data assimilation system adopted a hybrid 4D-EnVAR. 4D-EnVAR means that combined KIM VARiational data assimilation system(KVAR) and Local Ensemble Transform Kalman Filter(LETKF) data assimilation technique. Ensemble members currently uses 50 members. To evaluate the performance of the KIM, it is one of the important factors to understand the performance of system by operating and combining the individually developed systems. KIM Operational System(KOS) constructed a cycle experiments using the cylc meta-scheduler, which is widely used by various operational agencies and research laboratories. The cyclical experiments involves a data assimilation process every 6 hours, including KIM Package for Observation Processing(KPOP). The cyclical experiments were performed DA systems at 4 times a day, and 10-day forecasts are execute at 00 and 12 UTC due to model verification. Each task was written by python scripts and was configured to efficiently parallelize by using the cylc meta scheduler. Each task was configured pre- and post-processing progress and can be executed independently. Post processing contains visualization and remap from cubed-sphere grid to lat/lon grid. In addition, the grid remapped lat/lon grid data is used for visualization and displayed on the monitoring system for performance verification and stable operation of the numerical forecasting system. All of procedure are carried out automatically without any operation by user.

How to cite: lee, Y.: Current status and future plan for KIM operational system, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-183, https://doi.org/10.5194/ems2023-183, 2023.

Dayoung Choi, Adam Clayton, and In-Hyuk Kwon

   The renewed KIAPS (Korea Institute of Atmospheric Prediction Systems) aims at developing a unified framework for seamless prediction from the very short range (~6 hours) to the extended medium range (~30 days), including coupling to various Earth system components, such as the land surface, oceans, and sea ice. For very short range forecasts, KIAPS is currently developing a high-resolution (1-5km) model targeting storm-scale high-impact weather over East Asia, centered on the Korean peninsula. In order to start developing methods for assimilating Korean radar reflectivity observations, we have developed a “testbed” system based on a high-resolution, limited-area version of WRF, configured by the Korea Meteorological Administration (KMA) to use the same physics packages as the Korean Integrated Model (KIM) global model. The data assimilation (DA) system is based on the Local Ensemble Transform Kalman Filter (LETKF), and currently only includes support for assimilation of 3D synthesis data derived from the radar reflectivity observation.
   In order to get reasonable ensemble analysis fields at the convective scale through the LETKF-based radar DA system, we have been evaluating and tuning the reflectivity assimilation method. In order to improve the analysis of synoptic scales, we are now also adding support for the assimilation of conventional observations (Sonde, Surface, Aircraft, etc.). To allow this, the KIM Package for Observation Processing (KPOP), which has so far only been able to use background fields from the KIM global model, is being extended to support use WRF limited-area background fields. When this work is completed, the LETKF system will be extended to assimilate the conventional observations provided by KPOP in addition to radar reflectivity observations. We will then continue evaluation and develop a full cycling system. The final goal will then be to replace WRF with the new high-resolution KIM model.

How to cite: Choi, D., Clayton, A., and Kwon, I.-H.: Plans for the evaluation of the KIAPS LETKF-based radar reflectiviity DA system, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-372, https://doi.org/10.5194/ems2023-372, 2023.

Nikolaos Antonoglou, Ulrich Blahak, Manuel Werner, and Kathleen Helmert

The combination of C-band and X-band radar networks has become increasingly popular in recent years, as it offers significant benefits for a range of applications. C-band and X-band radars use different frequencies, with the prior operating at around 5.6 GHz and the latter operating at around 9.3 GHz, leading to wavelengths of 5.3 and 3.2 cm, respectively. This means that the two types of radar have different characteristics and are suited to different applications.

When combined, the two radar types complement each other's strengths and weaknesses. C-band radar can cover a larger area and provide a general overview of the environment being monitored, while X-band radar can provide detailed information about specific targets within that area. By combining these two types of radar, it is possible to obtain a more accurate and complete picture of the environment being monitored, with both high resolution and extended range. On the other hand, the combination of different-frequency radars in one processing chain brings several challenges. The backscattering profiles are different in each system, resulting in distinct observation patterns. Moreover, each frequency is subject to different attenuation rates, according to which, the appropriate corrections need to be applied.

The German Weather Service (Deutscher Wetterdienst – DWD) operates a network of 17 C-band radars and in the following months will start to install four additional X-band systems in the urban areas of Karlsruhe, Nürnberg, Halle, and Bremen. The goal is to extend the coverage of the network and improve the early detection of thunderstorms that could potentially cause flash floods. As an initial step, we aim at quantifying the specific attenuation for all types of precipitation and the two frequencies. This process is fundamental for the realistic correction of the observations, particularly for higher-frequency signals that are more heavily impacted by attenuation.

How to cite: Antonoglou, N., Blahak, U., Werner, M., and Helmert, K.: Integration of four X-band radars into the existing C-band network: focus on the correction for specific attenuation, EMS Annual Meeting 2023, Bratislava, Slovakia, 4–8 Sep 2023, EMS2023-275, https://doi.org/10.5194/ems2023-275, 2023.