OSA1.1

The main goal of the South-East Multi-Hazard Early Warning Advisory System project (SEE-MHEWS-A) is to provide support for the National Meteorological and Hydrological Services in Southeast Europe to produce timely and accurate warnings of hazardous weather and hydrological events. The reliability of such a system largely depends on the adequate performance of numerical weather predictions (NWP). Therefore, an adequate verification procedure applied to several NWPs region-wide is a necessary component of the process for building such a system.

The verification methodology consists of several building blocks, starting with the analysis of the missing observations and climatology analysis. After the preparatory steps, the methodology includes the verification of a continuous predictand, engaging several conventional verification measures such as Pearson correlation coefficient, systematic error, mean absolute error and root mean square error. Additionally, the verification of a categorical predictand is performed, using several thresholds to describe different precipitation intensities. For climatologically common events the measures such as frequency bias, hit rate,  false alarm ratio and equitable threat score are used to evaluate the forecast quality. The extremal dependence index is used to access the performance for rare events. Finally, the single observation - neighborhood forecast (SO-NF) verification approach is engaged. This approach allows for a more fair comparison between models of different resolutions by comparing results for similar spatial scales.

The five numerical modeling systems available for verification include ALADIN-ALARO (Aire Limitée Adaptation dynamique Développement InterNational), COSMO (Consortium for Small-scale Modeling), ECMWF-IFS (Integrated Forecast System), ICON (Icosahedral Nonhydrostatic) and NMM-B (Nonhydrostatic Multi-scale Model) models. The verification analysis is done on a domain that includes several countries in southeast Europe for a 24-h cumulative precipitation variable. The area tested includes a diversity in orography, which contributes to a better assessment of the performance of the NWP models being evaluated. 

Results often indicate moderately or strongly correlated data and bias mostly pronounced in the areas of the complex orography. The error increases from more flatter areas towards complex ones. The RMSE decomposition shows that dispersion error is the predominant source of error, while systematic sources of error are considerably smaller for all forecasts tested. Regarding categorical verification, all models produce excellent results for the dry day category, while there is usually lower quality for the remaining precipitation events. The SO-NF approach indicates that there are useful additional forecasts present in the proximity of the exact location, even though they are slightly displaced, showing the benefits in terms of better assessment of forecast quality for a rare event.

Finally, since all models do show specific benefits and limitations, verification results suggest that a multi-model ensemble might be a step further for the exploiting full predictive potential of these systems.

How to cite: Kozić, K., Odak Plenković, I., Keresturi, E., and Horvath, K.: Precipitation verification as an important task of South-East Multi-Hazard Early Warning Advisory System project, EMS Annual Meeting 2021, online, 6–10 Sep 2021, EMS2021-139, https://doi.org/10.5194/ems2021-139, 2021.

Over the last decade statistical postprocessing has become a standard tool to reduce biases and dispersion errors of probabilistic numerical weather prediction (NWP) ensemble forecasts. Most established postprocessing approaches train a statistical model using raw ensemble statistics on a typically small set of stations.  While raw ensemble statistics are available from high resolution NWP grid data, observations are missing at most grid points. Hence, the generation of spatial fields of forecast scenarios requires both some kind of interpolation and reshuffling of forecast quantiles based on a dependence template. The most widely used reshuffling approach, ensemble copula coupling (ECC), applies a reordering based on the raw ensemble rank order structure. ECC relies on the assumption that the spatial dependence structure of the raw ensemble is spatially consistent with the observed fields. This assumption may not always hold for hourly precipitation in particular over complex topography, since even high resolution models do not achieve a perfect representation of the real topography.

In this study, hourly CombiPrecip fields, which are a blend of precipitation observations from station and radar data, at a spatial resolution of 1 km over Switzerland serve as observations. Hourly precipitation raw ensemble forecast fields covering lead times up to 120 hours with a spatial resolution of 2 km are provided by COSMO-E. This enables us to postprocess hourly  COSMO-E ensemble precipitation forecasts over Switzerland at different spatial scales, from a single global ensemble model output statistics type model, over regional quantile regression  models up to grid point-wise local analog models. The mismatch in spatial resolution between COSMO-E and CombiPrecip as well as  the general issue of non-representative model topography over Switzerland’s complex topography may affect the spatial consistency of the (postprocessed) forecast fields. Starting with an analysis of systematic errors and spatial consistency of COSMO-E precipitation forecasts , we assess the potential for spatially multivariate postprocessing approaches, which are able to incorporate the spatial information from CombiPrecip and are yet simple and computationally efficient. To this end, we analyse the effects of using standard and new postprocessing model designs that vary in the (analog-based) selection of training data, spatial aggregation, postprocessing model parametrizations, and methods to obtain physically realistic forecast scenarios in space. 

How to cite: Hemri, S., Bhend, J., Spirig, C., Furrer, R., Moret, L., and Liniger, M. A.: Spatially consistent postprocessing of precipitation over complex topography, EMS Annual Meeting 2021, online, 6–10 Sep 2021, EMS2021-129, https://doi.org/10.5194/ems2021-129, 2021.

Localized heavy rainfall, which can be associated with flash floods, is difficult to predict accurately: both the predicted location and the intensity can exhibit large errors. Moreover, weather forecasts should be provided for points and not for the large regions represented by global model grid boxes. This mismatch can in principle be addressed using high-resolution limited-area models, or by applying some post-processing to global forecast models, as used in “ecPoint-rainfall”, a new ECMWF probabilistic post-processing technique to improve precipitation forecasts. One novel premise of ecPoint, which has a major positive impact on the calibration, is that the forecast-versus-point-observation relationship depends on “gridbox weather types” that could potentially occur in many parts of the world.

The MISTRAL (Meteo Italian SupercompuTing PoRtAL) project, funded under the Connecting Europe Facility (CEF) – Telecommunication Sector Programme of the European Union came to its end in January 2021. The main project goal was to facilitate and foster the re-use of datasets by weather-dependent communities, to provide added value services using HPC resources. ECMWF participated in the project with the goal of improving probabilistic 6-h rainfall forecast products, to improve the prediction of flash floods in Italy and nearby Mediterranean regions. One of the objectives was to exploit the CINECA supercomputer facilities in Bologna to extract maximum benefit from ecPoint-Rainfall and from a 2.2km resolution COSMO limited area ensemble. To address that, we applied a new and innovative scale-selective neighbourhood post-processing technique to the COSMOS output, which, on the one hand, identifies and preserves the most reliable heavy rainfall signals and, on the other, spreads out those signals which are less consistently handled. Then, it is blended with a new 6h ecPoint-Rainfall product in order to leverage the most skilful aspects of the two systems. The 6-h ecPoint Rainfall forecasts were also developed during the project, building on the pre-existing ecPoint-Rainfall 12h product (already delivered to ECMWF customers in real-time). The final blended product includes, for lead times of 1-10 days, 6-h accumulated rainfall for each COSMO gridbox in percentiles (1, 2,..99) and probabilities of exceeding certain thresholds.

The main objective of this work was to improve forecasts and support weather-alert decisions for flash flood prediction. As a legacy of the project, we are now providing forecast data for Italy and nearby regions with a higher level of quality and resolution than has hitherto been possible,  and we are also delivering a robust gateway to products for the European community within the MISTRAL portal (https://meteohub.hpc.cineca.it/app/maps/flashflood). The principles could also be usefully applied in other parts of Europe, or indeed the world, where limited area ensembles are running operationally.

In this presentation, we will introduce the methodologies, the verification results and will illustrate with forecast examples.

How to cite: Gascón, E., Montani, A., and Hewson, T.: The MISTRAL Project provides a new tool for Flash Flood Forecasting in Italy, EMS Annual Meeting 2021, online, 6–10 Sep 2021, EMS2021-87, https://doi.org/10.5194/ems2021-87, 2021.

In this talk we present a new statistical method for the seamless combination of two different ensemble precipitation forecasts (Nowcasting and NWP) using neural networks (NNs), see [1]. The method generates probabilistic forecasts for the exceedance of a set of predetermined thresholds (from 0.1mm up to 5mm). The aim of the combination model is to produce seamless and calibrated forecasts which outperform both input forecasts for all lead times and which are consistent regarding the considered thresholds. First, the hyper-parameters of the NNs are chosen according to a certain hyper-parameter optimization algorithm (not to be confused with the training of the NNs itself) on a 3-month dataset (dataset A). Then, the resulting NNs are tested via a rolling origin validation scheme on two 3-month datasets (datasets B & C) with different input forecasts each. Datasets A & B contain forecasts of DWD's RadVOR, a radar-based nowcasting system, and Ensemble-MOS, a post-processing system of NWP ensembles made by COSMO-DE-EPS, with a horizontal resolution of 20km, which is a predecessor of ICON-D2-EPS. Ensemble-MOS forecasts were provided for up to +6h, while RadVOR forecasts were available up to +2h. For dataset C, forecasts with a grid size of 2.2km are used from STEPS-DWD, a new implementation of the Short-term Ensemble Prediction System (STEPS) by  DWD, and ICON-D2-EPS as a NWP ensemble system. Forecasts were made up to +6h. In both validation datasets (B & C), the forecasts show the well-known behavior that the nowcasting systems RadVOR & STEPS are superior for short lead times, while NWP forecasts (Ensemble-MOS & ICON-D2-EPS) outperform these systems for later lead times. Based on the comparison of several validation scores (bias, Brier skill score, reliability and reliability diagram) we can show that the combination is indeed calibrated, consistent and outperforms both input forecasts for all lead times. It should be noted that the combination works on dataset C, although the hyper-parameters were chosen based on dataset A, which contains different forecasts for a different grid size.

[1] P. Schaumann, R. Hess, M. Rempel, U. Blahak and V. Schmidt, A calibrated and consistent combination of probabilistic forecasts for the exceedance of several precipitation thresholds using neural networks. Weather and Forecasting (in print)

How to cite: Schaumann, P., Hess, R., Rempel, M., Blahak, U., and Schmidt, V.: A calibrated and consistent combination of probabilistic forecasts for the exceedance of several precipitation thresholds using neural networks., EMS Annual Meeting 2021, online, 6–10 Sep 2021, EMS2021-220, https://doi.org/10.5194/ems2021-220, 2021.

In the framework of LEXIS (Large-scale EXecution for Industry & Society) H2020 project, CIMA Research Foundation is running a 3 nested domain WRF (Weather Research and Forecasting) model with European coverage and weather radar data assimilation over Italy. Forecasts up to 48 hours characterized by a 7.5 km resolution are then processed by ITHACA ERDS (Extreme Rainfall Detection System), an early warning system for the heavy rainfall monitoring and forecasting. This type of information is currently managed by ERDS together with two global-scale datasets. The first one is provided by NASA/JAXA GPM (Global Precipitation Measurement) Mission through the IMERG (Integrated Multi-satellitE Retrievals for GPM) Early run data, a near real-time rainfall information with hourly updates, 0.1° spatial resolution and a 4 hours latency. The second one is instead provided by GFS (Global Forecast System) at a 0.25° spatial resolution.
The entire WRF-ERDS workflow has been tested and validated on the heavy rainfall event that affected the Sardinia region between 27 and 29 November 2020. This convective event significantly impacted the southern and eastern areas of the island, with a daily rainfall depth of 500.6 mm recorded at Oliena and 328.6 mm recorded at Bitti. During the 28th, the town of Bitti (Nuoro province) was hit by a severe flood event.
Near real-time information provided by GPM data allowed us to issue alerts starting from the late morning of the 28th. The first alert over Sardinia based on GFS data was provided in the late afternoon of the 27th, about 40 km far from Bitti. In the early morning of the 28th, a new and more precise alert was issued over Bitti. The first alert based on WRF data was instead provided in the morning of the 27th and the system continued to issue alerts until the evening of the 29th, confirming that, for this type of event, precise forecasts are needed to provide timely alerts.
Obtained results show how, taking advantage of HPC resources to perform finer weather forecast experiments, it is possible to significantly improve the capabilities of early warning systems. By using WRF data, ERDS was able to provide heavy rainfall alerts one day before than with the other data.
The integration within the LEXIS platform will help with the automatization by data-aware orchestration of our workflow together with easy control of data and workflow steps through a user-friendly web interface.

How to cite: Mazzoglio, P., Pasquali, P., Parodi, A., and Parodi, A.: Improving weather forecasts by means of HPC solutions: the LEXIS approach in the 2020 Bitti flood event, EMS Annual Meeting 2021, online, 6–10 Sep 2021, EMS2021-125, https://doi.org/10.5194/ems2021-125, 2021.

MeteoSwiss has developed and is currently implementing a NWP postprocessing suite for providing  automated weather forecasts at any location in Switzerland. The aim is a combined postprocessing of high resolution limited area and global model ensembles with different forecast horizons to enable seamless probabilistic forecasts over two weeks leadtime. Further, the output should be coherent in space and provide predictions at any location of interest, including sites without observations. We use the full archive of MeteoSwiss’ operational local area models (COSMO-1 and COSMO-E) over the past four years and the corresponding IFS-ENS medium range predictions of ECMWF to develop postprocessing routines for temperature, precipitation, cloud cover and wind. Here we present selected key results on the performance of various postprocessing methods we applied but also on practical aspects of their implementation into operational production.

Both ensemble model output statistics (EMOS) and machine learning (ML) approaches are able to improve the forecasts in terms of CRPS by up to 30% as compared to the direct output of the local area model. The skill increase obtained by postprocessing varies depending on the parameter, region and season, with best results for temperature and wind in areas of complex orography and only marginal improvements for precipitation during seasons with a high fraction of convective situations. Particularly for temperature, the combined postprocessing of COSMO and IFS-ENS resulted in a skill benefit over postprocessing the COSMO models alone. Locally optimized postprocessing would allow further skill improvements, but only at sites where observations are available. However, the ability of non-local postprocessing approaches to provide calibrated forecast at any point in space is a key advantage for providing automated forecasts to the general public via the internet and smartphone app. Furthermore, the computational efficiency of these non-local approaches makes them attractive for operationalization in a realtime context. 

How to cite: Spirig, C., Bhend, J., Hemri, S., Rajczak, J., Nerini, D., Keller, R., Cattani, D., Schaer, M., Moret, L., and Liniger, M.: The new MeteoSwiss postprocessing scheme for medium-range surface weather forecasts: multi-model, probabilistic, seamless, and at any arbitrary location, EMS Annual Meeting 2021, online, 6–10 Sep 2021, EMS2021-288, https://doi.org/10.5194/ems2021-288, 2021.

There are different "optimal" forecast methods for different forecast lead times and different weather phenomena. Focusing on precipitation and convective events up to some hours ahead, radar extrapolation techniques (Nowcasting) show good skill up to about 2 h ahead (depending on the situation), while numerical weather prediction (NWP) outperforms Nowcasting only at later hours. Ensembles of both Nowcasting and NWP help to assess forecast uncertainties.

DWD's new Seamless INtegrated FOrecastiNg sYstem (SINFONY) combines forecast information from Nowcasting and NWP in an optimized way and as a function of lead time to generate seamless probabilistic precipitation forecasts from minutes to 12 hours. After four years of research and development, SINFONY is about to come to life in the upcoming two years, with an initial focus on the prediction of severe convective events.

For the development of SINFONY, different interdisciplinary teams work closely together in developing

• Radar Nowcasting ensembles for precipitation, reflectivity and convective cell objects
• Hourly SINFONY-RUC-EPS NWP on the km-scale with extensive data assimilation of high-resolution remote sensing (radial wind, reflectivity and cell objects from volume radar scans; Meteosat VIS channels; lightning)
• Optimal combination of Nowcasting and NWP ensemble forecasts in observation space (precipitation, radar reflectivity and cell objects)
• Systems for common Nowcasting and NWP verification of precipitation, reflectivity and objects.

For the SINFONY-RUC-EPS, new innovative and efficient forward operators for volume radar scans and visible satellite data enable direct operational assimilation of these data in an LETKF framework. Advanced model physics (stochastic PBL scheme, 2-moment bulk cloud mircophysics) contribute to an improved forecast of convective clouds.

As input for the combination of NWP and Nowcasting information, SINFONY-RUC-EPS generates simulated reflectivity volume scan ensembles of the entire German radar network every 5 min online during its forecast runs. Ensembles of composites and cell object tracks are generated by the same compositing and cell detection- and tracking methods/software packages, which are applied to generate the Nowcasting information.

To help evolve DWD's warning process for convective events towards a flexible "warn-on-objects", our Nowcasting- and NWP cell object ensemble forecasts are then blended into a seamless forecast ("probability objects") in a pragmatic way. Gridded combined precipitation and reflectivity ensembles are also under development, targeted towards hydrological warnings.

In addition to the development of SINFONY itself, focus is also put on the interaction with users (e.g. from flood forecasting centres) along the weather information value chain for co-designing the development of new forecast products and approaches to improve the prediction and warning process.

This presentation will introduce the goal and the concept of SINFONY and provide an overview on the ongoing developments as well as on the incipient interaction with users.

How to cite: Blahak, U. and Keller, J. and the SINFONY-Team: SINFONY - the combination of Nowcasting and Numerical Weather Prediction at the convective scale at DWD, EMS Annual Meeting 2021, online, 6–10 Sep 2021, EMS2021-284, https://doi.org/10.5194/ems2021-284, 2021.

In the framework of the SINFONY project at Deutscher Wetterdienst (DWD) we have developed data assimilation of visible satellite reflectances of the SEVIRI instrument (MSG) and radar observations in a rapid update cycle (ICON-D2-KENDA-RUC) which will be running in a first 24/7-testsuite starting in spring of this year. Our major goal related to the assimilation of these new observation systems is to improve the positioning of cloud and precipitation systems and their intensities, needed for the seamless transition of radar nowcasting to numerical weather prediction (NWP) in our SINFONY system. We give an overview of the steps undertaken in the course of developing the data assimilation of visible satellite reflectances. This includes quality control, observation error modelling, data reduction and bias correction of the reflectances. Further development and enhancement of the forward operator MFASIS is still ongoing. A major step to allow for a successful assimilation has been the improvement of microphysical consistency between the NWP model and MFASIS both with 1-moment and 2-moment microphysics to reduce the bias of first-guess departures. To further enhance and stabilize the agreement of observations and model climatologies over the course of the year and different weather regimes, an innovative histogram-based bias correction has been developed. We show results of data assimilation experiments combining visible reflectances and radar data in the ICON-D2-KENDA-Rapid Update Cycle using 2-moment microphysics. Further, we discuss the improvement of forecast skill from both observing systems and the way they complement each other – putting special emphasis to the key variable of interest in the SINFONY system, namely radar reflectivity.

How to cite: Bach, L., Deppisch, T., Scheck, L., de Lozar, A., Welzbacher, C., Khoshravian, K., Blahak, U., Stephan, K., Schraff, C., Ulbrich, S., and Potthast, R.: Assimilating visible satellite reflectances in combination with radar data in a pre-operational convective-scale seamless prediction system, EMS Annual Meeting 2021, online, 6–10 Sep 2021, EMS2021-333, https://doi.org/10.5194/ems2021-333, 2021.

In the Basque Meteorology Agency (EUSKALMET), numerical weather prediction (NWP) models, adapted to the particular characteristics of the territory, are executed daily for many different purposes. In order to improve nowcasting and forecasting tasks, a WRFDA base data assimilation tool, was implemented. Assimilation of meteorological data combines the information provided by measured data with the information coming from numerical models, supplying the numerical representation more consistent with observations.

Working with continuous assimilation-forecast cycles of the assimilation system allows constant updating of limited area forecasts, improving nowcasting tasks, especially severe weather events. Nowadays, the tool is being executed routinely in operational basis. The assimilation system includes several datasets from different sources (surface and upper air data), available in the forecast domains: RAOB soundings, SYNOP, Buoy, METAR, Automatic weather stations and Radar. The Basque Country Weather Mesonet, managed by EUSKALMET, is a high-density network with more than 100 Automatic Weather Stations (AWS), representative of a territory of complex orography such as the Basque Country. Some observations registered in this network (ten-minute data) are included on the Data assimilation system. Euskalmet Radar is a METEOR 1500 Doppler Weather Radar with Dual polarization capabilities located on Kapildui mountain top (1174 m). Two volumetric scan are available each 10 minutes (range 300 km in reflectivity mode, range 150km in Doppler/Reflectivity mode). Reflectivity data is included in assimilation cycles.

The objective of this paper is to present the assimilation system included in the tool and to explain the results of some sensitivity experiments during high-impact weather events, to test the system's skill nowcasting extreme weather events. We present different validation analysis based on punctual and areal approaches. With a special focus on the use of datasets from the Basque Country Automatic Weather Station Mesonetwork and the available radar data.

How to cite: R. Gelpi, I., Diaz de Arcaya, A., Pedruzo, X., and Gaztelumendi, S.: Data assimilation for an operational nowcasting tool, EMS Annual Meeting 2021, online, 6–10 Sep 2021, EMS2021-233, https://doi.org/10.5194/ems2021-233, 2021.

Modern automobiles are becoming more and more “computers on the wheels” having lots of digital equipment on board. Such equipment is both for the comfort and entertainment of the passengers and for their safety. Sensors play a key role in measuring several parameters of the car performance (e.g., traction control, anti-lock breaking system) and also environmental  parameters are observed directly (e.g., air temperature) or can be somehow inferred (e.g., precipitation via windscreen wipers activity/speed).

KNMI has been provided air temperature recorded every 10 minutes by thousands of vehicles driving in the Netherlands for the period January-October 2020. We have performed an initial exploratory temporal and spatial analysis to understand the most promising periods of the day and areas where sufficient data is available to perform a more thorough data analysis in the future. Furthermore, we have performed a correlation analysis between the outside temperature measured by cars and air and ground temperature observed by official weather station sensors placed at one location on the Dutch highways. The correlation results for three randomly selected days (with different weather conditions) show a good positive correlation coefficient ranging from 0.93 to 0.76 for car and station air temperature and from 0.91 to 0.67 for car temperature and station ground temperature.

This initial exploration paves the way to the use of (OEM) car data as (mobile) weather stations. We foresee in the future to use a combination of sensed variables from cars such as air temperature, traction control, windscreen wipers activity for example to improve observations of road slipperiness and related warning systems that are not restricted to Dutch highways only.

How to cite: Pagani, G. A., Molendijk, M., and Noteboom, J. W.: Cars as next generation mobile weather stations: an initial investigation, EMS Annual Meeting 2021, online, 6–10 Sep 2021, EMS2021-290, https://doi.org/10.5194/ems2021-290, 2021.

The most important question a national weather service should ask itself in connection with its warning task is: "Do our warnings contribute to reducing the impact of extreme weather events?". A perfect impact forecast of an extreme weather event does not necessarily contribute to reduce the impact of the event. Even the most perfect warning, whether based on physical thresholds or on potential impact, is not a guarantee for a reduction of the impact of the warned extreme event. Only If the warning reaches the recipient in time, is understood and action is taken, is there a chance that the impact can be reduced, which means that the warning unfolds an impact. Therefore, if we want the recipient to understand the warnings and to know what action to take, we have to know what his needs are.

In this contribution we describe a method (“Jobs to be done”) with which we investigated the needs of the authorities in terms of severe weather warnings in Switzerland and we will present the results of this investigation. This method focuses our attention on those processes that are important to the authorities but unsatisfactorily fulfilled. Once isolated, we engage our experts in cooperation with the authorities to find optimal and innovative solutions through design thinking workshops. In the Swiss federal structure, the warning chain extends over all levels of the governance structure: the severe weather warnings are issued at federal level and transmitted to the Cantons, these can decide to add local information, particularly concerning impact, and transmit them to the communities and the population. In our investigation, we concentrated on the administrative authorities and on the cantonal coordination bodies of the fire brigades. The aim of this study is to find indications for optimising the warnings, in terms of content, representation and also distribution.

The investigation started in January 2021 with a series of interviews with seven natural hazard experts and six fire inspectors of different Cantons. Currently (April 2021) we are running two surveys in all Cantons and in June we plan two workshops with representatives of the Cantons and of the fire brigades together with collaborators of the National Weather Service MeteoSwiss (forecasters, developers and key accounts). 

How to cite: Willemse, S., Popovic, N., Fuduric, N., Bisang, L., and Zachlod, C.: Severe Weather Warnings: impact of an event vs. impact of a warning., EMS Annual Meeting 2021, online, 6–10 Sep 2021, EMS2021-240, https://doi.org/10.5194/ems2021-240, 2021.

Development, verification and feedback of impact-based weather warnings require novel data and methods. Unlike meteorological data, impact information is often qualitative and subjective, and therefore needs some sort of quantification and objectivation. It is also inherently incomplete: an absence of reporting does not automatically imply an absence of impacts.
The reconciliation of impact information with conventional meteorological data demands a paradigm change. We designed and implemented a verification scheme around a backbone of weather-related fire brigade operations and eye-witness reports at ZAMG, the national meteorological service of Austria. Meteorological stations, radar and derived gridded data are conceptualized as a backstop to mitigate impact voids (possibly arising from a lack of vulnerability, exposure or simply a lack of reporting), but are not the primary basis anymore.
Operation data from fire brigade units across Austria are stored at civil protection authorities at federal state level and copied to ZAMG servers in real-time. Their crucial information is condensed into a few components: time, place, a keyword (from a predefined list of operations) and an optional free text field. This compact information is cross-checked with meteorological data to single out weather-related operations, which are then assigned to event types (rain, wind, snow, ice, or thunderstorm) and categorized into three different intensity levels („remarkable”, „severe” and „extreme”) according to an elaborated criteria catalogue. This quality management and refinement is performed in a three-stage procedure to utilize the dataset for different time scales and applications:
 „First guess” based on automatic filtering: available in real-time and used for an immediate adjustment of active warnings, if necessary;
 „Educated guess” based on a semi-manual plausibility check: timely available (ideally within a day) and used for an evaluation of latest warnings (including possible implications for follow-up warnings);
 Final classification based on a thorough manual quality control: available some days to weeks later and used for objective verification.
Eye-witnesses can report weather events and their impacts in real-time via a reporting app implemented at ZAMG (wettermelden.at). Reports from different sources and trustworthiness are funneled into a standardized API. Observations from the general public are treated like a „first guess”, those from trained observers like an „educated guess”, and are merged with the refined fire brigade data at the corresponding stages.
The weather event types are synchronized with our warning parameters to allow an objective verification of impact-based warnings. We illustrate our measures to convert these point-wise impact data into spatial impact information, to circumvent artifacts due to varying population density and to include the “safety net” of conventional meteorological data. Yellow, orange and red warnings are thereby translated into probabilities for certain scenarios, which are meaningful and intuitive for the general public and for civil protection authorities.

How to cite: Pistotnik, G., Rieder, H., Hölzl, S., Kaltenberger, R., Krennert, T., and Zingerle, C.: Verification of impact-based operational weather warnings at ZAMG using real-time fire brigade and eye-witness data, EMS Annual Meeting 2021, online, 6–10 Sep 2021, EMS2021-503, https://doi.org/10.5194/ems2021-503, 2021.

The Basque Country is periodically affected by severe coastal-maritime episodes which, depending on their severity, can significantly alter human activities on the coastal strip, cause considerable material damage or even directly or indirectly result in personal injury.

In particular, in the field of coastal-maritime impact, three types of risk are currently considered in the warning/alert/alarm system operated by the Emergencies and Meteorology Directorate. The first one is associated with wind reversals along the coastline ("galernas") with a particular impact on users of beaches and coastline during the summer season. The second one, associated with bad sea conditions, with an impact on navigation in coastal sea waters (2 miles). The third one, associated with high sea-wave and tide conditions that favour overtopping and flooding in the most exposed areas of the coast.

The process of determining and communicating warnings/warnings/alarms is a complex decision-making operation involving multiple actors analysing different types of information based on a variety of available tools.  In this contribution we include a description of the warning/alert/alarm system, some aspects related to communication and dissemination including an analysis of the warnings issued during the operation of the system. We also provide a brief description of the hazard indicators and the early warning system (EWS) currently in operation at the Basque Meteorological Agency (Euskalmet), which allows monitoring and predicting severe situations and their potential impact in advance.

With regard to the warnings issued, we will present the main characteristics of the warning/alert/alarm system for maritime-coastal risk, including a "historical" perspective and comparing it with the previous warning system. We will analyse the monthly, seasonal and annual distribution of the warnings/alerts/alarms issued in recent years. We will also present the results of the validation process of this system during these last years of operation.

With regard to the early warning system (EWS), a description of the current system operating in Euskalmet is presented, covering the very short, short, medium and long term. Describing its main components that allow estimating the precursor variables of impact in each case. Sharp wind-reversals with wind intensification and propagation along the coast in the case of "Galernas". Wave height, periods and sea state in the case of Navigation, and overtopping indexes in the case of coastal impact. Finally, some conclusions are included regarding its operational performance and future work to be carried out to improve some operational aspects of the system.

How to cite: Gaztelumendi, S., Egaña, J., R. Gelpi, I., Gomez de Segura, J. D., and Aranda, J. A.: Coastal-maritime risk and early detection in Basque Country, EMS Annual Meeting 2021, online, 6–10 Sep 2021, EMS2021-250, https://doi.org/10.5194/ems2021-250, 2021.

At Meteo Expert, a Italian private organization providing weather and climate services and formerly known as Epson Meteo Centre, we are using the Self Organizing Map (SOM) algorithm to study synoptic circulation over Southern Europe, evaluating the capability of five NWP global models and one multi-model ensemble to predict its variability in order to relate synoptic circulation patterns to temperature and precipitation forecast’s quality over Italy. SOM is an iterative algorithm that ‘learns’ the patterns of the input data vectors and organizes them into nodes within the SOM space, arranging like patterns in neighboring nodes and the most unlike patterns in nodes farthest from each other. Daily observed and predicted weather types from the five NWP global models and the multi-model ensemble were recognized and classified by the SOM. The SOM-based classification built for our purposes produces a 12-weather-type set using daily 500 hPa and 700 hPa geopotential, sea level pressure, 850 hPa temperature and 700 hPa specific humidity. The five global models are GFS from National Centers for Environmental Prediction, IFS from European Centre for Medium-Range Weather (ECMWF), Arpege from Meteo France, GEM from Canadian Meteorological Centre, ICON from Deutscher Wetterdienst, together with MIX, our multi-model ensemble. Here we would like to present some examples of this operational activity in the one-year-period, also showing how much the source of forecast errors may depend on large-scale dynamics rather than model's physical parameterisations. A quality index has been used to quantify the overall ability of models in predicting the circulation patterns, showing that MIX and ECMWF reached the best performance within 96 hours of forecast.

How to cite: Salerno, R. and Bertolani, L.: Application to NWP Models Verification of an Atmospheric Circulation Patterns Classification, EMS Annual Meeting 2021, online, 6–10 Sep 2021, EMS2021-36, https://doi.org/10.5194/ems2021-36, 2021.

The potential goal of weather forecast verification is to quantify the quality of given forecasts to determine the best possible setup available for the predictand tested in a certain area. Such procedure is performed to adequately select the wind gust forecasts at a 10 m height for different geographical and topographical areas of the Republic of Croatia using data from 61 stations in 2018. In addition to the raw ALADIN numerical model forecast, 3 additional forecasts based on the analog post-processing method are verified: a simple AnEn forecast, an AnEn forecast with predictor weight optimization, AnEnT, and a forecast with an additional correction for the high wind speed AnEnK.

In the analysis of wind gust as a continuous variable, the raw ALADIN forecast is the least successful in the coastal group of stations, while the analysis of wind as a categorical variable shows the deficiencies of the raw forecast in the continental group of stations. It is shown that predictions obtained by the analog method improve the predictions overall, compared to the raw ALADIN forecast. The largest improvements are achieved in the coastal group of stations, while the improvements in the continental group of stations are not as emphasized. For the climatologically more common wind gust speeds, there is a noticeable improvement made with the analog-based post-processing over the raw ALADIN NWP. The same result is shown even for the extreme events, except for the continental group of stations.

The results presented in this work suggest that all the forecasts considered are more suited to predict the bora than the sirocco wind. Overall, better post-processing results are achieved in the northern Adriatic than in the southern Adriatic. However, the relative improvement gained by the analog method is more pronounced in the southern Adriatic, where the occurrence of the sirocco wind is more frequent.

Among the forecasts obtained by the analog method, the AnEnK variant could be singled out as the best one, especially for high wind gust speed. However, the differences in performance between the three variants of the analog method are often very small.

How to cite: Vujec, I. and Odak Plenković, I.: Evaluation of analog-based post-processing in Croatia for the wind gust NWP, EMS Annual Meeting 2021, online, 6–10 Sep 2021, EMS2021-74, https://doi.org/10.5194/ems2021-74, 2021.