Probabilistic Forecasts and Civil Protection

Weather forecasts are subject to inherent uncertainties due to the chaotic nature of the atmosphere and the limitations of weather prediction models. Ensemble prediction systems (EPSs) are widely used by the meteorological community to quantify these uncertainties, but these were not communicated to end-users until recently. However, such information on uncertainties has great potential to improve time-critical decision-making processes in civil protection agencies and by other stakeholders, especially if tailored to their needs and worded using their terminology. PROFORCE has delivered just that through multidisciplinary and cross-sector collaboration between meteorological services and end-users. It has enhanced awareness of the potential impacts of extreme events in civil protection.

A seamless probabilistic forecasting system

PROFORCE started in December 2013 and lasted two years. It was co-funded by the European Commission Humanitarian Aid and Civil Protection department (ECHO) and led by the Austrian weather service, ZAMG. Based on transnational cooperation between the Austrian and Hungarian national weather services (ZAMG and OMSZ, respectively) and civil protection partners in the districts of Lower Austria and Somogy County, PROFORCE aimed to improve preparedness and decision-making procedures in civil protection agencies by building an innovative seamless probabilistic forecast system.

The main characteristic of the PROFORCE seamless forecasting system is the probabilistic feature containing information on the uncertainty and predictability of severe weather events. Civil protection agencies should be able to use it to optimize their decision-making process in terms of preparedness and awareness and, therefore, better protect society and the environment from the impact of severe weather.

The PROFORCE seamless probabilistic forecasting merges four different systems:

• The European Centre for Medium-Range Weather Forecasts (ECMWF) EPS, representing the medium range and synoptic scale forecast;
• The Central European Limited Area Ensemble Forecasting system (ALADIN-LAEF), representing the region (and ALADIN-HUNEPS for the Hungarian domain);
• The French Application of Research to Operations at Mesoscale model (AROME) EPS, which provides forecasts up to 30 hours ahead with a focus on convection; and
• The Integrated Nowcasting through Comprehensive Analysis (INCA) Ensemble, which provides decision-making information to help civil protection agencies to make better judgment calls in face of imminent disaster risks.

Each system plays its own role in the final seamless product according to the nature of the predicted weather event (whether convective or large scale) and the forecast range. Under the assumption that each EPS produces the best available results for its designated timeframe, each individual time step in the forecast uses the data from the corresponding EPS, thereby making the forecast seamless. Thus, each day that an extreme event draws closer, the accuracy of the forecasts and warnings improves. The distinct gaps in the time steps, which result from the different background models, are reflected in the end product as the PROFORCE project specifically-trained civil protection staff to handle such uncertainties.

Seamless warnings → seamless actions 

The seamless information exchange and actions between meteorologists and civil protection staff follow this structure:

Pre-warning/response - ECMWF EPS shows a potential severe weather event with a relatively high probability and a first pre-warnings are given to civil protection agencies, which enter the “response” action phase and take early measures such as arranging duty rosters.

More specific warning/prepare - Two or three days before the potential event, the next EPS in the chain renders forecasts that are more precise as regards both space and time. Thus, it can give a more specific warning. Based on this, civil protection staff enter the “prepare” action phase with intensified activities like the provision of equipment and personnel. One day after this, the convection-permitting EPS provides more details, particularly in topographically complex regions such as the Alps.

Nowcasting/go-action - Finally, the nowcasting-range EPS material provides the most accurate picture of the severe weather situation and final decisions can be made. Civil protection staff enter the third, “go-action” phase where final plans – such as concentrating operational activities on the most affected regions – are put in place. This nowcasting approach is especially important for summer season convective events that are characterized by high spatial and temporal variability. For this reason, the nowcasting model INCA Ensemble is run every hour while other models – AROME EPS, ALADIN-LAEF and ECMWF EPS – are only updated every 12 hours; in the case of Hungary, the ALADIN-HUNEPS model is only updated once every 24 hours.

Setting up of web portals

Meteorologists and civil protection staff worked together to develop appropriate sets of weather data that illustrate forecasts clearly and simply, thereby promoting a fast decision-making process. First, the civil protection partners defined reasonable thresholds for the key meteorological parameters: wind speed, precipitation and temperature. These thresholds differed considerably in the two countries. For example, a 60 km/h wind gust is normal and occurs frequently in the mountainous regions of Austria, but poses a serious hazard to the extremely vulnerable Lake Balaton region in Hungary, the site of many water sport activities in summer. As a result, two separate web portals were set up, one for each country with different thresholds and visualizations.

On the sites, product illustrations are prepared in such a way that the forecast appears to be seamless, that is medium-range and lower resolution sources are automatically replaced by higher resolution forecasts at shorter ranges. Thus, users do not need to identify the EPS model in question. The individual EPS systems’ output is visualized both as probability maps, showing the likelihood of exceeding certain thresholds, and as point information for preselected locations in the form of meteograms or plumes. The probability maps in the Austrian portal are accompanied by image elements that illustrate the threat level in a manner similar to traffic light signals: green, yellow, orange and red as the threat increases.

Seamless probabilistic forecast information is processed, visualized, then posted on a purpose-built web portal that can be accessed by civil protection authorities and disaster managers (example from the Austrian web portal)

An overall threat index was created in order to provide civil protection authorities with general information immediately upon logging into the web portal. It takes into account the probability of an event, its riskiness (intensity) and the lead time, given that a longer forecast generally contains larger uncertainties. In other words, a “yellow” warning a week ahead should not cause panic, but a “red” warning in the next day’s forecast should ring some alarm bells.

Training and first hand experience

Interdisciplinary cooperation between meteorologists and civil protection agencies was a key to the success of PROFORCE.  Civil protection staff had to become familiar with probabilistic forecasts in order for them to derive maximum benefit from the web portal. Training sessions in both countries helped to strengthen transnational cooperation and the feedback from the civil protection agencies on the probabilistic information also permitted model developers to improve their EPSs.

Intense testing of the system was realized during severe weather events in the pilot regions of Lower Austria and Somogy County over the course of the project (December 2013 – November 2015). The feedback from end-users and beneficiaries was generally positive and the system’s applicability for operational use in civil protection agencies was confirmed. Prevention and preparedness actions could be launched much earlier and were more specific with the new system than with the classic deterministic weather forecast.

Storm Niklas (30 March – 2 April 2015) offers a good case in point. High probabilities of wind gust exceeding thresholds in large parts of Austria were signalled by ECMWF EPS over seven days in advance. Once a severe weather pattern shows up for the first time in global EPS, it is necessary to continuously check the consistency of its position, time and intensity from one forecast to the next. This led to the announcement of initial pre-warnings in the relevant regions five days before the event started. Three days before the event, the higher resolution EPS showed very high probabilities of gusts above 80 km/h (level 2) and high probabilities of gusts exceeding 100 km/h (level 3, the highest warning level). In the end, the maximum gust registered at low-altitude weather stations was 121 km/h in Enns, Upper Austria. The additional reliability and uncertainty information in the new system facilitated an improved assessment of the situation; however, the nowcasting EPS did not provide additional benefit. Performance was tested by checking the warnings against the civil protection agency’s deployment maps. There was a very high correlation between the density of civil protection actions and the highlighted areas in the EPS forecast products, especially for large-scale storms and flood events.

Combination of riskiness, probability and lead time to establish a common threat level.

At the Hungarian disaster management centre, the probabilistic forecasting system was used extensively in the preparations for several contests at and around Lake Balaton during the 2014 and 2015 summer seasons, including the Blue Ribbon Race and the Lake Balaton Cross-Swimming contest. For the latter event, the EPS information was helpful when it came to deciding whether or not to postpone the contest. Activities at Lake Balaton are vulnerable to sudden changes in weather, which can sometimes be regionally confined. The experience with PROFORCE demonstrated that while EPS forecasting techniques can improve the predictability of these changes, further development is still necessary, particularly in the very short nowcasting range.

To continue long into the future

PROFORCE was a challenging project that essentially had two goals. The first was to build a bridge between the forecasting community, which monitors the weather and knows the strengths and weaknesses of each individual model, and the civil protection community, which focuses on the impacts of severe weather. Building understanding between these two groups was challenging.

The second goal concerned the handling probabilities. Though familiar in weather and climate research, civil protection authorities and, to a lesser extent, forecasters are only just starting to learn to manage and trust probabilities. The civil protection staff must undergo regular training in order for them to become more confident in applying the concept of probabilistic forecasting when making everyday decisions concerning the deployment of aid or cancellation of mass events.

It is important to emphasize that the seamless EPS system does not completely replace the standard warnings issued by the forecasting office, which also takes other numerical weather prediction model results or observations into account. It is intended, above all, to provide additional information and details about spatial distribution and the intensity of the meteorological parameters.

Ultimately, everyone in the warning chain, from official authorities to local stakeholders and finally to the general public, can benefit from this project. Changing weather always presents new challenges for both meteorologists and civil protection experts. Although the PROFORCE project has already ended, the transregional and interdisciplinary cooperation it established will continue long into the future.

Meteosat (MSG) composite HRV and infrared image showing the situation during the Balaton Cross-Swimming contest on 19 July 2014.

ECMWF EPS probability forecast of the thunderstorm index (CAPE) exceeding 100 J/kg (weak shower or thunderstorm possible) and the 300 J/kg threshold (ordinary thunderstorm possible).

Acknowledgements

The authors thank all of the colleagues who contributed to the PROFORCE project. Special thanks go to the scientific advisory board: Fritz Neuwirth, Matthias Steiner, Siegfried Jachs, Jianjii Wang, Simon Jackson and Alice Soares. We are grateful for all materials and information provided by György  Heizler (DMDSC) and Johann Dantinger (NOEL-CP).The implementation of the PROFORCE project was 75% co-funded by the European Commission Humanitarian Aid and Civil Protection department (ECHO).

Authors

Clemens Wastl, Yong Wang, Martin Kulmer, Andrea Sigl, Department of Forecasting Models, Zentralanstalt fur Meterologie und Geodynamik (ZAMG), Austria

Andre Simon, Hungarian Meteorological Service (OMSZ), Hungary