17th EFAS Annual Meeting - banner
17th EFAS Annual Meeting - banner

17th EFAS Annual Meeting

After two years of online meetings, the 17th EFAS Annual Meeting was held as a hybrid meeting on 27-28 September 2022. The meeting was organized by the Joint Research Centre (JRC) with the support of all EFAS operational centres. The physical meeting was held in Ispra (Italy) at the European Commission's Joint Research Centre to which 76 participants from the EFAS partners, third parties and operational centres joined in person and another 47 followed online.

Participants of the 17th EFAS Annual Meeting at the JRC Conference Centre

Summary of the meeting

Status of EFAS operational – experiences, issues, challenges

The CEMS Hydrological Forecast Centre - Analytics and Dissemination (DISS) gave an overview on the various tasks and services provided. Apart from the daily analysis of the EFAS-IS and sending out notifications, DISS conducts overall analyses of the service through the annual survey, the detailed assessment report of one distinct event and the analysis of the feedback received on the notifications. In addition, DISS conducts webinars and partners can request trainings. The new EFAS User Guide is another source of information for the users. The 3 new full partners and 4 new third party partners that have joined EFAS since the last annual meeting were presented. In 2021, 147 formal, 169 informal and 1540 flash flood notifications were issued. In the period from January till August 2022, 78 formal, 87 informal and 420 flash flood notifications were sent to EFAS partners. Since April 2022, three pre-taskings of satellite images were conducted, which is a new task for DISS since the current Framework Contract. Earlier, JRC had the lead in this task. All EFAS partners were reminded to check the contact information that is stored on EFAS homepage, to make sure that the communication from the operational centres reaches the right recipients.

The Hydrological Data Collection Centre (HDCC) started with introducing its new branding: recently Soologic became part of Ghenova Digital. The domain name and email-addresses are going to change soon, and all data providers will be contacted in advance. Other than that, not much will change for the data providers. The current data transfer server will operate in parallel to the new one and all changes will be made in a controlled manner. During 2021, the total number of stations and the number of active ones have increased by 418 and 377, respectively. A new feature is that since this current contract, HDCC is going to collect data on reservoirs as well. Other main current activities of HDCC are the review of metadata and an enhanced quality control protocol, which was presented in detail under the section “What’s new in EFAS”. Results from the annual report 2021 were presented, in which 2.386 active stations were selected for the analysis. The report will be available on EFAS homepage soon.

The Meteorological Data Collection Centre (MDCC) introduced the team and the system in general and gave an overview on the data flow and processing. MDCC currently collects data from over 293000 sensors and stations, does quality control checks on these data, runs automatic data aggregation and validation calculations to prepare data for the generation of gridded spatial information that is used as input for COMP. Ongoing and planned developments include improvements of the data validation, which was presented in detail under the section “What’s new in EFAS”. It was furthermore explained, how partners can contribute to the METEO database: in general, as many variables as possible are collected in near real-time but also for the historical period 1970-present. The temporal resolution should be at least at a 6-hourly time step, and if available, also hourly data is collected. In general, the provided data from the partners (both hydrological and meteorological) significantly improves the system.

Also the CEMS Hydrological Forecast Centre – Computation (COMP) introduced their large team and gave an overview on the various activities that COMP conducts for the EFAS service: among others running 24/7 operations of both EFAS and GloFAS as well as development, operation and maintenance of products, webservices and interfaces including extensive documentation of all components. COMP collaborates with all EFAS centres since they are located in the middle of the chain between data collection and dissemination. Recent releases include adjusted ERIC reporting point criteria and visualisation, reporting points upgrade with a new algorithm, the 6-hourly post-processing, improved legends and introducing the social media layer to the EFAS-IS. COMP furthermore informed about the move of the ECMWF data tapes to the new HPC (High Performance Computing) in Bologna that started three weeks ago. There are some updates necessary to be able to run the service on the new HPC, for example a recent version of the programming language and general IT language harmonisations. A quick outlook on EFAS Next was given as well, which was presented more in detail under the section “What’s next for EFAS” on the second day of the meeting.

What is new in EFAS: Innovations implemented since the 16th EFAS Annual Meeting

Under the section What’s new in EFAS?, HDCC and MDCC presented their respective new data validation rules. The near-real time data that HDCC receives is considered raw data on which a series of general data checks are conducted to flag and filter out erroneous values. For this, HDCC needs rating curves from the data providers. The current data validation applied by MDCC is tolerant to avoid real extremes, but main shortcomings are the lack of geographical and seasonal/climatological contexts when evaluating the data. To account for the seasonal and/or climatic variation, MDCC implemented climate zones to group stations for threshold calculation as well as seasonally varying thresholds. In addition, neighboring stations are considered to detect possible erroneous stations.

The Joint Research Centre (JRC) presented a new development recently implemented in EFAS-IS that visualizes social media information. The information is automatically collected and displayed during events. These layers are available for the public. The regions are colored depending on the intensity of the conversation, not the actual impact. The algorithm leverages on AI models to listen to flood-related conversations on social media as soon there is a flood peak forecasted within the next 48 hours, but the scanning can also be triggered manually. Planned future developments include image classification and impacts assessment.

COMP presented three recent additions to EFAS, (1) the new Flash Flood layers based on the TAMIR products, (2) the upgraded reporting points layer and (3) the new CEMS Floods User Guide corner. The new Flash Flood layers are a blending of radar data corrected for underestimation bias and NWP data. These new products were collaboratively developed with a number of colleagues and institutions. The spatial resolution is 1 km for the complete EFAS domain (unlike the ERIC product based on COSMO-LEPS) and the time step is 1h for the first 6 hours and a 6h-time step afterwards. The forecast is updated every hour. A significant difference to ERIC is that also the impact is calculated using pan-European exposure data and the product package incorporates also several animated products. It is suggested to first load the catchment summary layer and then zoom in to the area of interest. The recently upgraded Reporting point layer was explained in detail as well. This interactive layer received an upgrade in terms of rules and thresholds for creating the reporting points. The Reporting point layer is used by the Dissemination centre in their daily duty; the upgraded version facilitates the analysis. Moreover, a new layer named Flood probability persistence layer was added, this layer shows the flood probability persistence to clarify how the reporting points are created. Extensive documentation is available on the EFAS wiki and the upgrades are covered in a recent webinar that will soon be available on EFAS homepage. Finally, the CEMS floods user guide corner was presented, which collects all user guide information in one space on the Wiki. DISS, HDCC and MDCC have their own pages, where they can contribute with data and COMP added the CEMS-Floods Data User Guide, which contains detailed technical instructions on how to download data. Besides this new user guide corner, the documentation on the Wiki underwent a general big update with the help of all EFAS operational centres to make it more intuitive and self-explaining. New documentation on reporting points, the flood probability persistence layer, the flash flood products including the TAMIR products and the post-processing was added. Any feedback on the new pages is highly welcome!

Demonstration of the Global Flood Monitoring:  layers & functionalities soon to be available from EFAS-IS

The Global Flood Monitoring layers (GFM), that are already integrated in GloFAS, were presented and will be integrated into EFAS in the coming weeks as well. They are based on a constant analysis of Sentinel-1 data to automatically detect water surfaces in the retrieved satellite image scenes and visualize flooded area. The output is an ensemble product of three state-of-the-art flood mapping algorithms. Ten output layers show among others the flood extent, likelihoods and the exclusion mask (containing natural water surfaces like e.g. oceans and lakes). The data can be accessed via GloFAS (and soon also EFAS) and through many dissemination systems. There are tools available to, for example, define an area of interest and download processed data for this selection only. For the area of interest, the user can also subscribe to notifications. Data acquisition does not need to be triggered manually, it is automatically processed all the time. Processing time is currently 5-6 hours. The processed images are available within 8 hours since the Sentinel-1 acquisition.

New developments from EFAS partners

Three of the many EFAS partners explained in interesting presentations how they are working with different aspects of the forecasting chain.

A presentation on Hydrowarnings and CAP gave a short overview on the Common Alerting Protocol (CAP), a standard, XML-based text message, which is easy to disseminate in real-time to end-users worldwide. The members are actively sending CAP files to Meteolarm, including hydrological warnings for rainfall, flooding and rain + flood. Re-users of Meteolarm in 37 countries need both impact descriptions and advice on how to act in case of a certain warning. It is required to have the message in all official national languages plus English. Benefits of becoming a Meteoalarm partner include that warnings are re-distributed via re-users such as Google, Apple etc., reaching hundreds of millions of end-users worldwide.
The challenges and solutions for Operational developments of the Surface Water Flooding Hazard Impact Model (SWFHIM) for England and Wales were presented. More specifically, the hazard impact framework incorporates danger to life, denial of access and damage to physical environment and infrastructures. The service is operational since 2020 and provides forecasts of flood risk maps at the county scale using a pre-calculated impact library. As input, short-range ensemble precipitation data and nowcasts are used consisting of 24 members at a 2.2 km resolution. To calculate the risk of flooding, 9 effective rainfall scenarios (3 durations, 3 return periods) are applied. This product is used rather as a tool in flood forecasting then as a distinct warning.

The third contribution was an overview of the Development of a national flood forecasting and early warning system in Georgia. There is a long history of hydro-meteorological observations in Georgia with currently 150 hydro-meteorological stations, most of them being automatic. A significant increase in flood events since 1950 was observed and this highlighted a need for better preparedness and therefore a need for forecasting. Currently, there is a large project on risk reduction of climate-driven disasters, within which a number of observational stations were set up. Both hydrological (HEC-HMS) and hydraulic models (MIKE 11, HEC-RAS) are applied. Although there are station data available, the total number of stations for calibration is still limited, which challenges the accuracy of the model simulations.

Future of EFAS

General overview

The JRC gave a general overview on the ideas and current planned developments for the EFAS service, where the main new feature will be the higher spatial model resolution that is going to significantly affect all users. The many developments during the past years might also impact the criteria for issuing formal and informal flood notifications, which is another aspect of the service that is going to be assessed in the near future. Questions like: “Is the 30% threshold still the best choice?”, “Should we use total probability of all forecasts instead of a combination of separate forecasts (1 deterministic and 1 probabilistic)?”, “Can the minimum catchment area be revised with the higher spatial resolution?”, “Is the persistence rule still adding value to the forecast?”. These are some of the questions that were raised as a consequence of the extensive development that the whole EFAS service has gone through. In the light of the higher spatial resolution, even the flash flood layers will be examined more in detail, whether they still add value in their current form based on ERIC. The GFM layers are another important addition to the existing product portfolio available to the EFAS users. But also for non EFAS users, there will be more data made available: both static maps within EFAS and GloFAS, but also the hydrological model, LISFLOOD, is available fully open source including many tools for example for calibration. A more long-term development could be a new model calibration using a deep parameter learning algorithm. There is still a lot to come!

EFAS-Next: 1 arcmin resolution

The more general overview was followed by a more detailed presentation on EFAS-Next, the next major version of the system, whose most significant feature is the increased spatial resolution from 5 to 1 arcmin (approximately 1,3 km). But this is not the only new addition, all the input maps have been completely revamped which incorporates changes in values and spatial variability for the new maps. The higher spatial resolution implies 14 times more grid cells making model calibration quite a challenge. At the same time the number of calibration points was doubled, now including a lot of new stations, whereas the calibration area is extended only to a lesser extent (i.e. the new stations were not situated in new regions). The smallest catchments calibrated have a catchment area of around 150 km2. Where possible, the calibration period was extended: all stations have a minimum of 4 years calibration period, which can be extended to 8 years or half of the total length of the record period, data availability permitted. A dual calibration was run considering both 6-hourly and daily stations. Significant effort has been put on the soil moisture routine that should prevent the soil from getting too dry through enhancing the redistribution of soil moisture in the model. In addition to the model parameters, a lot of work has been done on the IT side as well: all tools are open source now and the code was optimized so that run time decreases significantly – a mandatory requirement to be able to run such a data intense modelling system. The performance of the calibration (measured with the KGE) shows an overall quite good fit, but there are some general issues mostly in Spain due to extensive regulations and in Scandinavia mostly related to soil moisture representation that was improved in EFAS Next, but that impacted the model performance in turn. Compared to EFAS 4.0 about 2/3 of the calibration points show the same or a better performance, with particularly large improvements in the lower KGE (meaning a large effect) and largest improvements for smaller catchments (< 500km2). It was also obvious that with the increased model resolution, the precipitation (intensity, location) becomes much more important for the modelling results. Currently, early summer 2023 (June) is targeted for the launch of EFAS Next.

The Global Flood Monitoring

The scientific background of the GFM layers was presented in detail: the Gobal Flood Monitoring ensemble flood products. GFM is a global near real-time (NRT) flood monitoring system based on the Copernicus Sentinel-1 SAR mission. The GFM provides an estimate of the flooded area in an acquired satellite image scene within 8 hours after a Sentinel-1 data acquisition. The approach relies on flood mapping taking into consideration three state of the art flood mapping algorithms from three institutions (DLR, TU, LIST). The system is designed to prevent false positive (water looking areas) and false negatives (existing water bodies). Flood and likelihood results are going into a flood ensemble allowing for uncertainty assessment of flood/no flood areas. The ensemble approach balances out the limitations of the three different algorithms. However, if the algorithms differ a lot, the system relies on a consensus approach. Different products are available: a) observed flood extent (=flooded area), b) reference water mask (=existing water bodies) and c) water extent (= combination of a and b), but also likelihood layers and a layer with advisory flags are available. A manual and product definition document are available in the https://extwiki.eodc.eu/GFM/PUM.

Newly published EFAS datasets

The Verification of the European flood hazard maps was presented by the JRC. The EU Flood Directive requires all Member States to develop flood hazard (FH) maps, but national FH maps can differ significantly for different countries with limited information and availability outside of each respective country and non-EU countries might not have national FH maps at all. EFAS, however, needs continental scale maps for impact-based forecasts and continental scale analysis of flood risk and therefore calculated flood extent and water depth maps. This motivated the production of a catalogue of river flood hazard maps for Europe and the Mediterranean Basin. The EFAS climatology (i.e. long-term LISFLOOD runs) as input to the 2-dimennsional hydraulic model LISFLOOD-FP to produce a 100m spatial resolution catalogue of flood extent and water depth maps. The catalogue includes six discharge return periods (10 to 500 yrs) and is available for catchments > 500km2. A comparison (validation) with official and openly available flood hazard maps showed that 2/3 of the flood prone areas were correctly identified, however modelled maps generally overestimate flood-prone areas. The overestimation is generally caused by data limitations (flood defenses, topography, river channels); higher skill was detected for high return period events (> 100 yr RP) compared to low return period events. The EFAS flood hazard maps have limitations (e.g. only available for larger rivers), and therefore the question arises whether the flood hazard maps can be complemented with national available information. Preliminary tests were conducted in Spain, indicating a high potential of improving the EFAS flood hazard maps. However, since access to national data is limited, this task is highly dependent on the help of the local partners.

The JRC also presented EMO-5: Publication of the CEMS meteorological gridded dataset, a high-resolution, multi-variable gridded meteorological dataset for Europe from 1990 to 2019. The EMO-5 v1 data is available on a 5x5 km raster grid and includes daily variables (precipitation, minimum and maximum temperature, wind speed, solar radiation, water vapor pressure), as well as 6-hourly variables (precipitation, and mean temperature).  As input data, roughly ~ 19000 ground weather stations were used. Where ground weather stations coverage is sparse, four high resolution regional observational grids and ERA Interim/land were used. Quality control of the input data is applied which includes availability checks, monthly statistics, cross validation, minimum/maximum validation, rate of change validation.  The dataset was evaluated for two variables (precipitation and temperature). For precipitation, a comparison with two regional high-resolution products from Norway; an analysis with 15 heavy precipitation events was done, where 13 of the events showed greater precipitation amounts in EMO-5. The spatial resolution of the EMO dataset has been recently increased from 5km to 1arcmin (approximately 1.3 km). Version 2 of the dataset, namely EMO-1arcmin is available starting from 1990, with real time releases. The data is freely downloadable via the JRC data catalogue or from the CEMS-EFAS ftp service.

The importance of user feedback

The talk Process modelling - Lesson's learned from studying various flood emergency management chains presented the summary of a recent master thesis. While EFAS is available for a large number of different partners, no evaluation of the use for these partners has been done since the pre-operational phase, so there is a need to re-evaluate the service. The model that was used in this study was developed based on a limited number of interviews. The aim was to assess the design objectives of the service, roles and responsibilities of the different parties, communication flows as well as decisions and procedures. One conclusion was that the use of the service for pre-warnings is generally appreciated but often only somewhat informal, conversely, the partners could consider utilizing the products more, as confidence in EFAS improves. Another outcome of the study was the relationship between forecasting agencies and Civil Protection agencies that have slightly different preferences, e.g. for threshold-setting: Forecasters emphasize accuracy, whereas Civil Protection agencies emphasize the need for preparedness.

Tools and protocols for feedback collection are widely in place in EFAS and a talk by DISS once more emphasized the value of providing feedback (on notifications and missed events). All feedback that the users send to the Dissemination centre – be it through the annual EFAS partners survey sent out at the beginning of each year or through constant provision of feedback on the received notifications – is eventually feeding into the development and improvement of the system as a whole. Obviously, not every single suggestion for improvement can be followed up, but the sum of feedback received is analysed through DISS in a number of annual reports and is used by different internal working groups with participants form COMP, JRC and DISS. Examples of integration of partners’ feedback are the upgraded reporting point and flash flood layers. Moreover, the relatively new feedback on missed events can reveal insight into processes not captured adequately in the current setup of the system. Finally, the presentation showed how well the partners performed in providing feedback; two clear winners for the feedback provided throughout the years could be found: a special thanks goes to the Spanish and Greek partners that showed a constant high performance over the years with at the same time high numbers of notifications received!

Facilitating communication during flood events

Partners’ experience: DWD activities after the 2021 extreme floods in Germany

In the final presentation of the EFAS AM, the Activities in Germany after the 2021 extreme flood event from DWD's perspective were presented by the DWD. In the aftermath of this severe flooding one immediate measure considered is to strengthen the Communication and Advisory between different (federal) levels of responsibility in case of an event through a networking platform that can be used before and during an event. There are ongoing discussions on forming a common expert group from national and federal authorities as a competency centre for decision making and information. In general, the communication between the DWD and regional centres has been intensified through regular visits, trainings and video conferences both before and during events. Another lesson learned is that there are too many applications and tools available that make choosing the “right one” difficult. Instead, the focus is now put on co-design activities between the DWD and regional flood forecasting centres for a more user-oriented product development. In addition, DWD’s warning system is being adapted to the needs of the forecasting centres.

EFAS Conditions of Access

The 17th EFAS Annual Meeting concluded with a discussion on the proposed update of the Conditions of Access (CoA). More specifically, the proposal suggested reducing the embargo period for the public availability of EFAS products from 30 days to 15 days. The foreseen adjustment of the CoA was triggered by the extreme floods in central Europe during summer 2021 during which it was practically impossible for EFAS centres to respond to requests from the media due to the embargo period of 30 days, as currently defined in the CoA. The JRC first presented a list of pros and cons. The partners were then invited to share their views. The details of this discussion can be found in the attachment “Discussion_on_the_CoA” at the bottom of the page. No decision was taken during the 17th EFAS Annual Meeting. The JRC will follow up in individual discussions with the EFAS partners to see whether a compromise solution can be found or the raised concerns can be addressed.  The discussion will be resumed in a future meeting.

Workshops

Workshop I: What is EFAS PP, and how to access forecasts (including SOS)

The method and aim of EFAS PP were explained in detail, which is a bias correction to reduce errors and uncertainty in the forecast so that it more accurately predicts the observed discharge. The workshop participants discussed in smaller groups the pros and cons of the product and what possibly could be improved.

Workshop II: Flash flood forecasts

In the workshop on the flash flood forecasts, the participants were divided into smaller groups to be able to discuss a number of questions on the existing product (mainly ERIC) and to come up with suggestions for improvements. There are some known shortcomings of the ERIC products, and the participants were generally interested and saw a significant potential in the new developments for the flash flood layers, i.e. the TAMIR products that were presented earlier this day.

Workshop III: LISFLOOD OS (EFAS and GloFAS hydrological model is open source!)

In the second workshop slot, LISFLOOD Open Source was presented, that is available on Github and free to use (and collaborate) for everyone. The modelling approach of LISFLOOD OS was presented in detail including the necessary input files. A considerable part of the workshop was dedicated to very hands-on instructions on how to get started, for example how to install the code on a user’s own PC and what the necessary steps to prepare own simulations are. Apart from the available user guide, it is possible to get in touch with the developers to get help, but it is also highly welcome to collaborate.

Workshop IV: The cartography game

The second workshop on day two of the annual meeting was a very EFAS user-oriented session with the aim to understand the various EFAS partners better and – more importantly – 1) to quantify the added value of the EFAS products in comparison with the services that are already used/provided at the partner organisations, and 2) to understand their needs for complementing products from EFAS. Both the online participants and in-presence participants were actively engaged in the workshop and had to discuss and answer four sets of questions on what kind of forecasts/services the users are currently providing and which components they are missing (that are available within EFAS) as well as what their future developments are. Many of the respondents listed flash flood forecasts, extended time horizons and/or impact assessments/flood mapping as missing components, clearly showing the added value provided within EFAS. A next step would be to assess if the data provided within EFAS really fill these gaps at the EFAS partner organisations with the appropriate information.

Posters

The poster session invited to a more direct interaction between the presenter and the audience and was used to exchange experience and views on the presented topics. The presented posters were:

  • Croatia - EFAS partner feedback for 2021 and short overview of recent development
  • Nowcasting of rainfall return periods by municipality, a pragmatic tool for warning of rainfall flooding
  • Impact-based flood warnings in Sweden - Development and first experiences

 

Conclusions

After two days of interesting and engaging talks it can be concluded that it is valuable to meet face to face, because it makes the interaction between the participants much easier and the informal and spontaneous chats and discussions over a coffee or around a poster really make a difference. The hybrid form of the meeting is a good opportunity for those who cannot make it to the venue.  Based on this experience, the JRC and the EFAS centers will continue their efforts to further encourage and strengthen the interactions with all the participants during the next EFAS annual meeting.

The next EFAS annual meeting is taking place on 28-29 September 2023 in Offenbach, Germany.

 


Presentations

Below you can find the presentations given at the 17th EFAS Annual meeting:

Status of EFAS operational – experiences, issues, challenges:

What’s new in EFAS?​

Demonstration of the GFM (layers & functionality) - Tobias Stachl (GFM)

Workshops

Posters

New developments from EFAS partners

What’s next for EFAS?

Verification of the European flood hazard maps – Francesco Dottori (JRC)

EMO-5: Publication of the CEMS meteorological gridded dataset - Vera Thiemig (JRC)

Process modelling - Lesson's learned from studying various flood emergency management chains - David Novak (University Muenster)

The value of providing feedback (on notifications and missed events) - Nina Bosshard (DISS)

Facilitating communication during flood events by Peter Salamon

 

Discussion on the CoA.pdf (Not available to non-registered users)