PICO
The focus of this session will be on bringing the expertise from different fields together as well as exploring differences, similarities, problems and solutions between forecasting systems for varying hazards including climate emergency. Case studies of system implementations - configured at local, regional, national, continental and global scales - will be presented. An operational warning system can include monitoring of data, analysing data, making and visualizing forecasts, impact-based solutions, giving warning signals and suggesting early action and response measures.
Contributions are welcome from both scientists and practitioners who are involved in developing and using operational forecasting and/or management systems for climate and water-related hazards, such as flood, drought, tsunami, landslide, hurricane, hydropower etc. We also welcome contributions from early career practitioners and scientists, and those working in multi-disciplinary projects (e.g. EU Horizon Disaster Resilience Societies).
We particularly welcome contributions aligned with the objectives of the WMO World Weather Research Programme project InPHRA (Integration of Precipitation and Hydrology for Early Action). InPHRA aims to advance transdisciplinary knowledge and skills for the research and development of effective multi-hazard flood forecasting and early warning systems so that “no one is surprised by a flood.” This includes integrating meteorology, hydrology, and social science, together with local and Indigenous knowledge systems, to improve the value chain from forecasts to community action, with particular attention to vulnerable populations.
For local governments, it is essential to quickly and accurately understand the extent of flooding damage caused by typhoons, linear rainbands, or other heavy rainfall events in order to make critical decisions such as broadcasting evacuation notices or requesting emergency assistance to national government. In recent years, various systems have been developed to quickly predict and assess flood damage, but high implementation costs, computational demands, or operational complexity have become barriers to widespread adoption. Here, we develop a flood depth estimation system that keeps implementation as well as computational costs down while meeting practical needs of disaster management applications.
Using actual flood measurements obtained by low-cost water level sensors and digital elevation model (DEM), the system estimates flooded areas and depths in near real-time based on the sum of the measured flood depth and the ground elevation at each sensor location and visualize them quickly on the system. The system also includes features designed for convenience during imminent disasters, such as alerting every evacuation warning level, regularly saving and exporting flood depth maps and logs.
Additionally, estimating flood areas from past heavy rainfall events and validating these estimates, we assess the system accuracy. By involving disaster management personnels in using the system, we build a solution that is easy to operate even in the field during emergencies.
Figure 1. A schematic diagram of the system.
REFERENCES
- Idehara, A. and K. Hirano, 2020: Quick Estimation Method of Flood Inundation Mapping using Single Point Information, Report of the National Research Institute for Earth Science and Disaster Prevention (NIED), 85
(https://dil-opac.bosai.go.jp/publication/nied_report/PDF/85/85-1idehara.pdf, 2026.1.12). - NIED: https://midoplat.bosai.go.jp/web/shinsui/index.html (2026.1.12).
- ArcGIS Online: https://www.esri.com/en-us/arcgis/products/arcgis-online/overview (2026.1.12).
How to cite: Kitauchi, H., Nagao, A., Nakamura, M., and Igari, T.: A near real-time flood depth estimation system for practical disaster management applications, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-15958, https://doi.org/10.5194/egusphere-egu26-15958, 2026.
Extreme rainfall events can trigger flash floods that pose serious risks to communities, infrastructure, and critical services, particularly in arid and rapidly urbanizing environments. In the Kingdom of Saudi Arabia, short hydrological response times, strong spatial variability of precipitation, complex topography, and limited observational data significantly challenge flood early warning capabilities, which affect emergency management at the national scale. Addressing these challenges requires integrated and scalable hydro-meteorological forecasting systems capable of operating across large spatial domains while resolving convective weather events and associated localized flood impacts in urban/suburban areas.
This study presents a nationwide, operational flash flood early warning system developed for the Kingdom of Saudi Arabia. The system is designed to provide consistent coverage across the country while capturing fine-scale weather, hydrological and hydrodynamic processes relevant to flash flooding in arid environments. It operates over 137 hydrological domains, representing more than 6,000 outlets, delivering 2D flood simulations at a spatial resolution of 30 m nationwide, with enhanced resolution of up to 2.5 m in selected urban areas.
The forecasting framework is structured as an end-to-end modeling chain that links atmospheric forcing, hydrological response, hydraulic flood propagation, and infrastructure impacts. High-resolution numerical weather predictions generated by the Weather Research and Forecasting (WRF) model are combined with real-time radar and rain gauge observations to produce hourly ensemble weather and precipitation forecasts and hindcasts. These meteorological inputs drive a distributed hydrological model (CREST), which simulates runoff generation across arid catchments using spatially explicit information on topography, land cover, soil properties, and drainage networks. A reservoir management module is fully integrated within the modeling chain, allowing the system to account for reservoir storage dynamics, controlled releases, and spillway operations, and to assess the influence of dam infrastructure on downstream flood evolution.
Hydrological outputs are used as boundary conditions to a two-dimensional hydrodynamic model, which simulates floodplain dynamics, water depths, and inundation extents.
All model components are coupled within a WebGIS-based operational platform that displays deterministic and ensemble weather and hydrologic forecasts, probabilistic flood warnings, and real-time nowcasting products. Flood hazard information is delivered through interactive maps, warning levels, and time series, to support decision- making by civil protection authorities and emergency managers at national and local scales.
The functionality and operational performance of the system are demonstrated through its application on a recent extreme rainfall and flash flood events that affected the entire region of Saudi Arabia in the period of December 9-16, 2025. The system successfully captured the timing, spatial extent, and severity of flooding across multiple domains, providing useful lead times and high-resolution inundation maps. This case study highlights the robustness, scalability, and operational value of the framework, demonstrating its potential to enhance flood preparedness through early warning, and risk management across the Kingdom of Saudi Arabia under increasing hydro- meteorological extremes.
How to cite: Sofia, G., Anagnostou, E., Patlakas, P., Chaniotis, I., Christidis, Z., Kallos, A., Zaidi, S., Alzabari, F. M., and Alomary, M. A.: An Operational Flash Flood Early Warning System for the Kingdom of SaudiArabia, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-6332, https://doi.org/10.5194/egusphere-egu26-6332, 2026.
Since the launch of the national integrated natural disaster risk survey in 2022, followed by the flood risk mapping programme in 2025, an integrated basin-scale flood forecasting and risk analysis framework has been progressively developed to support operational flood warning and risk management in Sichuan Province. The framework integrates a multi-source refined database, basin-scale coupled hydrological–hydrodynamic models, and an operational dynamic flood simulation platform for major rivers.
At the data level, GIS and BIM technologies are integrated to construct L1–L3 refined three-dimensional databases for key basins, integrating fundamental geographic information, socio-economic and POI data, flood hazard investigation results, inundation extents for typical return periods, risk zoning products, as well as building distributions, oblique photography, BIM models, and near-real-time rainfall and hydrological observations. This results in a unified, updatable data foundation that supports operational flood simulation and loss assessment.
At the modelling level, an integrated basin-scale hydrological modelling system is coupled with one- and two-dimensional hydrodynamic models. By using meteorological forecasts as forcing, the system supports end-to-end simulation from forecast precipitation, through rainfall–runoff generation, to river flood routing, thereby enhancing temporal continuity and spatial accuracy for operational flood forecasting.
At the application level, an operational dynamic flood simulation and analysis platform has been developed for major rivers. Under operational conditions, the platform integrates real-time and forecast data to support multi-area and multi-scenario flood simulations, prediction of water levels and discharge at key cross-sections, and assessment of inundation extent and potential losses. The platform provides technical support for flood warning issuance, emergency evacuation, and risk management decision-making. It has been operationally deployed in the Minjiang, Dadu, Tuojiang, Fujiang, and Qujiang river basins, and is currently being extended to the Jialing and Qingyi river basins.
How to cite: Wang, T., Cao, D., Jiang, W., Wan, H., Wang, Z., and Qin, Q.: An operational basin-scale flood forecasting and dynamic risk analysis system: a case study from Sichuan Province, China, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-2058, https://doi.org/10.5194/egusphere-egu26-2058, 2026.
The construction of three separate national-scale flood forecast models for the operational flood forecast system for Denmark presents a unique opportunity to compare “top-down” vs. “bottom-up” modeling approaches and “process-based” vs. “data-driven” model types. To implement an operational flood forecast model for Denmark as quickly as possible, a “top-down” process-based hydrological model (E-HYPE DK) was first extracted from the pan-European E-HYPE model developed from European and global data sources. A separate process-based model, DK-HYPE, as well as a data-driven model, DK-LSTM, were developed for Denmark from the “bottom-up” using national data sources combined with high-resolution catchment delineations and more detailed model process representations.
Evaluation of the two modeling approaches showed a trade-off between time invested and societal benefit. Overall, the top-down E-HYPE DK model provided benefit early in the project by providing rapid access to model results which could be used to guide the development of the entire forecast chain and warning system. In contrast, the bottom-up DK-HYPE model developed later in the project, provided better model performance and higher-resolution outputs than the top-down model but required longer time to develop and deploy. While the addition of local high resolution forcing data and hydrological properties in DK-HYPE certainly contributed to the improved performance, changing the representation of groundwater process better captured the importance of surface water-groundwater interactions in Danish river systems.
Results from the project also highlighted trade-offs between the process-based and data-driven models. Compared to the process-based HYPE models, the data-driven DK-LSTM model required the shortest time for development and offered the best match between simulated and observed discharges. However, the data-driven model had difficulty in making predictions for events outside the training conditions (e.g. storms with unusually high precipitation) and did not provide information about internal variables that are provided by the process-based models (e.g. local runoff and soil moisture) which can be valuable for operational decision making.
The DK-HYPE model is now operational, providing public warnings for high river flows. The DK-LSTM is currently used as a supporting model during warning situations.
How to cite: Brendel, C., Martinsen, G., Payet-Burin, R., Dhaubanjar, S., Thrysøe, C., Dalgaard Jensen, L., Aarestrup, P., H. Madsen, M., W. Pedersen, J., A. Plum, C., D. Thomassen, E., Capell, R., Andersson, J., and Butts, M.: Building a National Operational Flood Forecast System for Denmark: Evaluating Top-Down vs. Bottom-Up and Process-Based vs. Data-Driven Modeling Strategies , EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-3982, https://doi.org/10.5194/egusphere-egu26-3982, 2026.
The INLINE project aims at advancing on EWS capabilities with tools to anticipate the impacts caused by storms, heavy rain and floods to support the decision-making workflows of various levels of Civil Protection Agencies (CPAs), including their coordination and cooperation.
To achieve this, the project is developing impact-forecasting products and functionalities with European coverage, which are being tested in real time over a 15-month demonstration period. Results are co-evaluated with the participation of a number of end-users (both partners and stakeholders integrated in the INLINE Community of Interest) to assess their operational value.
This study presents results form the first months of the demonstration (starting in September 2025) focusing on (i) the skill of the products to anticipate the occurrence of the most significant events, and the magnitude of the resulting impacts; and (ii) the first results obtained with end-users during recent high-impact events in their regions.
How to cite: Berenguer, M., Park, S., Baugh, C., O'Regan, K., Pulkkinen, S., and Myllykoski, H.: Co-evaluation of integrated pan-European rainfall and flood impact forecasts for cooperation in emergency management, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-21091, https://doi.org/10.5194/egusphere-egu26-21091, 2026.
In Navarra - a 10,000 km² region in northern Spain with 700,000 inhabitants - 50 municipalities are required to implement a local plan for flood risk management.
The Regional Flood Risk Management Plan of Navarra identifies these 50 municipalities based on their level of risk. It also establishes the structure of each local plan, which must follow four standardized documents.
Municipal plans include one pre-emergency level and four emergency levels: 0, 1, 2, and 3. The pre-emergency level does not necessarily need to be communicated to the public. The emergency levels are defined as follows:
- Level 0: Flooding has not yet begun, but streamflow has significantly increased.
- Level 1: Expected flooding will affect low-lying areas near riverbanks.
- Level 2: Severe damage is expected in urban areas.
- Level 3: The regional government assumes control of the local plan because the situation exceeds local capacity.
The activation of each emergency level of the plan has to be communicated to the population.
To prepare a plan, we visit each municipality and hold technical meetings with local authorities and staff, including the local police. We inspect strategic locations where local resources have historically acted to minimize flood damage. Typical actions include door-to-door warnings, street closures, and on-site alerts in public buildings such as schools or nursing homes.
The most critical task in drafting the plan is defining the thresholds that trigger each emergency level. These thresholds are based on historical rainfall and streamflow data within the river catchment. Usually, streamflow data from upstream measuring stations is used, while in small catchments, accumulated rainfall over a specific time period is also considered.
Once the paper version of the plan is complete, it is transferred to a digital platform that enables coordinated operations by local authorities (mayors and other officials) and staff. This platform includes both a mobile app and a web-based interface, offering:
- Real-time data updates every 10–15 minutes (from different observing networks: regional government, Spanish Meteorological Agency, Water Agencies, etc.).
- Easy activation of emergency levels.
- GIS maps showing the location of all planned actions.
- A mass SMS alert system for rapid communication with the population using predefined messages.
Since 2018, technicians from the Government of Navarra and Orekan have worked to implement these operational and consistent structures. They are based on local knowledge gathered from municipal staff, site visits, and collaborative planning. Information about the plans is shared with residents through detailed leaflets and public information sessions in each municipality.
How to cite: Loizu, J., Sanz, L., Varela, A., Zaragüeta, E., Castiella, A., and Ursua, A.: Flood risk management at municipality level in Navarra, northern Spain., EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-2210, https://doi.org/10.5194/egusphere-egu26-2210, 2026.
Impact-based forecasting (IBF) represents a significant advance in disaster risk management when considering the vulnerabilities of the population, livelihoods and exposed assets. In this paper we present the scaling up of PANDORA, an IBF tool developed for the Andean-Amazonian region of Peru, using the Madre de Dios River (MDR) as a case study. This initiative is in the framework of the BID Project: Hydrological and hydrodynamic monitoring and forecasting system for river floods in the Andean-Amazonian region of Peru, Ecuador and Bolivia.
PANDORA integrates a large-scale hydrological-hydrodynamic model (MGB) with precipitation forecasts, generating probabilistic flow projections with a five-day horizon. These forecasts are contrasted with flood thresholds associated with return periods of 2, 5 and 10 years, corresponding to moderate, severe and extreme levels of severity, respectively.
The intersection between the potentially flooded areas and the exposed elements (population, educational and health centres, transport routes and agricultural areas) allows us to estimate impacts at different political and administrative levels. Given the limited availability of hydrometeorological data in the MDR region, altimetry satellite information was incorporated to improve the performance and validation of the MGB model. The system was evaluated against the flood event recorded in February 2021, obtaining satisfactory results despite the limitations identified. Overall, PANDORA shows a high potential to support local decision-making in flood risk management using IBF.
How to cite: Lavado-Casimiro, W., Saavedra, D., Collado, R., Montesinos, C., Felipe, O., and Sanahuja, H.: Impact-based forecasting of river floods in a Peruvian Andean-Amazonian basin: first results in the Madre de Dios River, Peru., EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-4632, https://doi.org/10.5194/egusphere-egu26-4632, 2026.
The accuracy of flood warnings should ideally be evaluated on real impact data, although such data are often difficult to obtain and work with. The Swedish Meteorological and Hydrological Institute (SMHI) has recently acquired records of flood-affected roads and emergencies attended by fire and rescue services in Sweden over the last 25 years. We analysed whether the impacts were covered by flood-related warnings and whether they coincided with hydrometeorological conditions exceeding flood warning thresholds from hindcast data. Here we present our experiences on tackling challenges associated with using the impact dataset, insights into what type of flood events the warning system handles well and how it can be developed further.
The existing SMHI warning methodology explained only 26% of the reported flood impacts, although this proportion increased to 43% after filtering out minor and isolated impacts. Incorporating runoff data from a recently developed sub-daily hydrological model further increased the proportion of explained impacts to 54%. Sub-daily runoff was especially effective in explaining summer flood impacts from cloudbursts in small flashy streams, illustrated through a case study of the Västernorrland flood in September 2025. Notably, total runoff generated in subcatchments was a more important predictor of flood impacts than streamflow, while precipitation did not account for almost any impacts alone without coinciding hydrological causes. Nevertheless, impacts from winter processes, such as urban snowmelt and rain-on-snow floods, remain poorly represented in the warning system. Our findings highlight the importance of filtering impact records prior to evaluation and reveal the benefit of utilising high-resolution hydrological models with outputs beyond streamflow in operational flood warning systems.
How to cite: Svatoš, J., Karimi, S., van de Beek, R., Olsson, J., and Hjerdt, N.: Insights on using flood impact data for evaluating hydrometeorological warnings in Sweden, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-19110, https://doi.org/10.5194/egusphere-egu26-19110, 2026.
Real-time flood modeling is increasingly important given the increased frequency and intensity of severe storms and flood damage. Machine learning provides unique opportunities for improving modeling outcomes, adjusting model outputs in real-time which can then be used as input data to inform subsequent predictions. In this work, we focus on improving George Mason University’s (GMU) iFlood Integrated Flood Forecast System. iFlood currently provides high-accuracy flood forecasts from twice-daily runs over the tidal region of the Potomac River from Lesieta to Little Falls, covering the Washington Metropolitan region and including coastal areas of the National Capital, Alexandria, and Arlington. iFlood has been operating for multiple years and is currently included in local forecast ensembles used by local weather forecasters to make valuable flood assessments. Our results explore how various machine learning techniques can be used to alter flood predictions, assessing impacts on model outputs as well as changes to computational dependencies.
How to cite: Ruess, P. J., de Souza de Lima, A., and Ferreira, C.: Improvements to GMU iFlood using Machine Learning for Real-time Flood Modeling Corrections, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-22800, https://doi.org/10.5194/egusphere-egu26-22800, 2026.
Since the early 2020s, artificial intelligence (AI) has gained substantial attention across both industry and academia. In the field of water resources, AI-based approaches for improving the prediction of floods, water supply, droughts, and related hydrological phenomena have been actively explored. More recently, the emergence of agentic AI, in which large language models (LLMs) orchestrate multiple AI tools for analysis, prediction, and operational services, has attracted increasing attention.
Despite these advances, most research efforts remain focused on model development, while the establishment of sustainable operational systems, such as those enabled by machine learning operations (MLOps), remains limited. This gap is particularly evident in water resources applications, where continuous retraining, performance evaluation, and system-level reproducibility are critical for real-world deployment.
In this study, we propose NeuralRiverOps, an operational framework that integrates MLOps and agentic AI for multi-point flood prediction in large-scale river basins using long short-term memory (LSTM) networks. First, we design a workflow that supports sequential model development and prediction from upstream to downstream and from tributaries to main streams, leveraging the neuralhydrology Python library as the core modeling engine. Second, to enable systematic model retraining, storage, inference, and performance evaluation, we construct an MLOps pipeline based on MLflow. PostgreSQL is employed for structured time-series data management (e.g., rainfall, dam releases, and river water levels), while MinIO is used for scalable object storage, such as trained LSTM models. Furthermore, we develop an agentic AI system that allows users to interactively invoke the MLOps pipeline through a chat-based interface. This system is implemented using Ollama as an open-source LLM platform and OpenWebUI as the conversational interface. All components - including AI models, MLflow, PostgreSQL, MinIO, Ollama, and OpenWebUI - are containerized and orchestrated using Docker Compose to enhance computational reproducibility, scalability, and maintainability.
The proposed framework demonstrates a practical architecture for integrating agentic AI into analytical systems and highlights the essential role of MLOps in the sustainable operation of AI models for disaster preparedness, such as flood and drought forecasting. This study provides a pathway for future research to move beyond isolated model development toward robust, operational AI systems supported by MLOps and agentic AI.
How to cite: Choi, Y., Yang, H., Kim, S., and Ryu, J.: NeuralRiverOps: An Operational Framework for Implementing MLOps and Agentic AI in LSTM-based Flood Forecasting for Large-scale River Basins, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-6212, https://doi.org/10.5194/egusphere-egu26-6212, 2026.
Operational hydrological forecasting systems still suffer from uneven performance across regions, particularly in data-scarce environments, where errors in model states and parameters propagate rapidly and limit short- to medium-range forecast skill. Within the SEED-FD (Strengthening Extreme Events Detection for Floods and Droughts) project, we investigate how multi-source data assimilation strategies can be configured to improve the propagation of observational information into short- to medium-range hydrological forecasts under operational constraints.
We implement and evaluate data assimilation workflows based on an Ensemble Kalman Filter (EnKF) within the GLOFAS system, as used in the Copernicus Emergency Management Service (CEMS) Hydrological Forecast Modelling Chain. Multiple observation types are considered, including in-situ river discharge, satellite-derived discharge and water levels, and altimetric water level observations from Earth Observation (EO) missions. Assimilation experiments are conducted across several contrasted river basins representative of diverse hydro-climatic and socio-environmental conditions, including the Niger, Paraná, and Juba–Shebelle basins.
The analysis focuses on short- to medium-range streamflow forecasts and examines how different assimilation configurations influence the persistence and propagation of corrections beyond the assimilation window. In particular, we compare state-only approaches, including filtering and smoothing strategies, with exploratory joint state-parameter estimation experiments, with the aim of identifying configurations that maximize the temporal impact of observational information while remaining compatible with operational requirements. Ensemble-based methods are employed throughout the study to ensure consistency with probabilistic forecasting frameworks.
This work presents the results of these experiments and discusses key scientific aspects relevant to the design of data assimilation strategies for improving the propagation of corrections in large-scale operational flood and drought forecasting systems.
How to cite: Pedinotti, V., Sadki, M., Barresi, O. L., Martin, N., and Alim, Y.: Operational data assimilation of Earth observation hydrological data across contrasted river basins: insights from the SEED-FD project, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-11763, https://doi.org/10.5194/egusphere-egu26-11763, 2026.
Risk can be defined as the likelihood of a certain level of impact occurring. The Flood Forecasting Centre (FFC) communicate the flood risk in England and Wales using a risk matrix. This risk matrix compares the likelihood and impact to conclude the overall flood risk. This is communicated through the Flood Guidance Statement (FGS), which is issued daily. A similar risk matrix is used across the UK, including by the UK Met Office (UKMO) and the Scottish Environment Protection Agency (SEPA). However, currently there are differences in how the risk matrices are used and communicated. Recent storm events, such as Storm Bert, have highlighted the importance of clear and consistent messaging of risk across events and organisations to ensure that users of the risk matrix make appropriate decisions.
To address this issue, the FFC have been part of the Common Warnings Framework (CWF). This has included working alongside the UKMO and SEPA, as well as Environment Agency (EA), Natural Resources Wales (NRW) and Department of Infrastructure Northern Ireland (DfI). The research work, led by the UKMO, has been based around the Common Alerting Protocol (CAP). CAP has been used as a guide on how likelihood and impacts can be communicated. The main outcome of this work has been to agree a commonality in communicating flood risk. This will provide greater clarity, consistency and visibility around flood risk for emergency services, government and the public. The FFC have established a taskforce this year to deliver the changes to the flood risk matrix in time for winter 2026/2027.
Alongside the Common Warnings Framework, the FFC are exploring making more use of ensemble data. Working with the UKMO and EA, this has involved using meteorological ensemble data to drive hydrological ensemble output. A primary aim is to make the assessment of the likelihood of flooding impacts more objective and consistent during and between events. This is to improve flood incident management action. This approach has been trialled this winter with preliminary results expected during 2026.
This presentation will explain the upcoming changes to the FGS flood risk matrix. It will highlight how the flood risk matrix has evolved with time, the benefits the changes will make and how the changes link to the output from the ensemble trial. This includes looking at how useful ensemble based meteorological and hydrological summary tools may be for flood forecasters and decision makers, with the overall aim to improve the communication of risk. With more ensemble data becoming available this does create additional challenges in communicating risk. This presentation will also discuss the work the FFC has started in this area, looking at what AI can offer around impact assessments and communicating risk.
How to cite: Lattimore, C., Millard, J., Miller, C., Turner, R., Duke, A., and Fenwick, K.: Common Warnings Framework for flood risk in England and Wales – improving communication language for flood risk and how ensembles and AI may provide more objective risk assessment , EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-17584, https://doi.org/10.5194/egusphere-egu26-17584, 2026.
Climate change is intensifying short-duration heavy rainfall over Northwestern Europe, increasing the frequency of rapid hydrometeorological impacts. These events increase the probability of short-term bathing water (BW) pollution, especially in catchments affected by combined sewer overflows and agricultural runoff. In England, mandatory monitoring of all storm overflows has revealed 450,398 recorded spills in 2024, leaving bathers unacceptably exposed. Coinciding increases in self-reported illness following contact with polluted BW highlight the need to reconsider how BW quality is forecast in the context of increasing extreme rainfall regimes.
Operational BW forecasts in England currently combine radar nowcasts, deterministic (UKV) rainfall forecasts, wind and UV data in a multiple linear regression model. Crucially, the forecast is issued once in the morning and not revised later in the day, even if rainfall forecasts change, providing only static, same-day guidance and constraining bathers’ ability to make informed decisions. While improvements in numerical weather prediction and monitoring remain critical, recent UK bathing water regulatory reforms increase the operational value of anticipating sustained or clustered pollution episodes across the bathing season and beyond, rather than relying on single-day exceedances.
Here we explore the use of synoptic weather patterns as a complementary framework for anticipating multi-day bathing water pollution risk. Synoptic weather patterns describe persistent, physically coherent circulation regimes. They influence not only how much rain falls, but also the type of rainfall (frontal versus convective) and the accompanying conditions (wind, cloud cover and solar irradiance). Using the Met Office 30-class daily weather pattern (WP) catalogue, microbiological data and 1 km Nimrod radar composites for South West England (May–September 2012–2023), we derive daily rainfall depth, intensity and wet fraction and link these, together with WP, to the site-day intestinal enterococci exceedances (IE ≥ 63 cfu/100 mL) used to inform operational advice against bathing.
We collapse 30 synoptic weather patterns into four physically interpretable families: Cyclonic Atlantic (frontal), Showery maritime/unsettled, Convective extremes, and Settled anticyclonic quiet. In observed data, “advice against bathing” varies significantly by family; it is highest under Cyclonic Atlantic and elevated under Showery maritime/unsettled. We use these families to construct plausible bathing water season storylines (persistent wet, persistent dry, dry with storm outbreaks, and transition scenarios wet to dry and dry to wet). For each storyline, we simulate 5,000 May–September seasons by resampling historically observed, physically coherent daily driver “packages”.
Comparing rainfall-only and weather pattern-based statistical models under a fixed advisory frequency shows that pattern-based approaches identify fewer, longer advisory windows, while rainfall-only methods produce shorter, intermittent alerts. In practice, this would mean fewer stop-start bathing advisories and clearer identification of sustained periods when extra attention, sampling, or precautionary messaging is needed. Since weather patterns can often be forecast several days ahead, this suggests that synoptic-scale information can support more actionable multi-day guidance for bathing water management, monitoring, and public communication.
How to cite: Krupska, K., Speight, L., Robinson, J. S., and Cloke, H.: From weather patterns to warnings: supporting multi-day bathing water advisories using synoptic weather regimes , EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-13889, https://doi.org/10.5194/egusphere-egu26-13889, 2026.
Every week, somewhere on our planet, people die in a flood. We can now predict many types of floods well before any rain has even fallen, or the storm has even begun to form. We have spent billions of Euros setting up sophisticated flood prediction systems that undertake billions of calculations to predict when and where floodwaters will be. But what is the point of all of this if nobody can understand the danger that they are in, or imagine their homes and lives swept away? The floods in Germany in 2021 and in Valencia in September 2024 showed failures to prevent deaths. But was this a failure of forecasting science, or a failure of imagination?
How to cite: Cloke, H.: Preparing for floods in an uncertain future: forecasting, warning and imagination, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-2765, https://doi.org/10.5194/egusphere-egu26-2765, 2026.
The demand for seasonal hydrologic forecasts is significant and various applications for water resources management are increasing. Since some years, the lead time is going up to several months or a season. However, the uncertainty is also increasing with lead time.
We assess the potential of seasonal streamflow and sediment forecasting as a tool for management of water resources and sediment flow in the Upper Blue Nile Basin (UBNB) of Ethiopia, upstream the GERD (Great Ethiopian Renaissance Dam). A coupled hydro-meteorological seasonal forecasting system requires a performance evaluation of both numerical weather prediction (NWP) models and hydrological models to accurately represent atmospheric and hydrological conditions. We evaluate the ECMWF-SEAS5 precipitation product in conjunction with the large-scale process-oriented hydro-sedimentological model WASA-SED. The aim is to generate forecasts for streamflow and suspended sediment fluxes with a lead time of up to seven months for the UBNB.
Three different large-scale rainfall “products” were tested and compared ref. their representativity of observed rainfall. We show that such a rainfall evaluation is indispensable for hydrological simulation as well as for seasonal forecasting. We consider this step a “hydrological verification” of rainfall data.
Seasonal streamflow and sediment flux data were than forecasted for June to December of the year, based on the seasonal meteorological forecast in the preceding month. An ensemble of 51 regional meteorological forecast members in daily resolution and 7 months lead time, each initiating on the first day of each month, was used. A post-processing step with an autoregressive model was applied to adjust for forecast biases in seasonal streamflow predictions. Results indicate that the coupled meteorological/hydrological models skilfully predict rainfall and discharge on a seasonal scale for the Blue Nile Basin.
How to cite: Bronstert, A., Zargar, M., Francke, T., Kindie, W., Zimale, F., and Harald, K.: Seasonal forecast of streamflow and suspended sediment in the Blue Nile Basin, Ethiopia, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-7569, https://doi.org/10.5194/egusphere-egu26-7569, 2026.
Flood Early Warning Systems (FEWS) are among the most effective non-structural measures for reducing flood risk, particularly in data-scarce regions with rapid hydrological responses. However, designing efficient FEWS requires balancing forecasting skill with the economic costs of dense monitoring networks. Identifying the most influential observation points is therefore essential for reliable flood forecasting with minimal instrumentation.
In this study, we propose a data-driven framework to identify critical sub-basins whose monitoring provides the greatest benefit for flood early warning. The framework integrates long-term stochastic rainfall simulation, semi-distributed hydrological modeling, machine learning, and feature importance analysis. High-resolution synthetic rainfall time series are generated using a multifractal-based stochastic approach and used to drive a hydrological model, resulting in an extensive virtual database of flood events across multiple sub-basins. Simulated sub-basin discharges are then used as predictors in a Random Forest model to forecast outlet discharge at different lead times.
Feature Importance Measures (FIM) quantify the relative contribution of each sub-basin to flood forecasting performance, enabling identification of a reduced set of hydrologically dominant sub-basins. The methodology is demonstrated in the semi-arid, mountainous Torghabeh River Basin (northeastern Iran), where limited hydrometric infrastructure and short response times pose significant challenges for flood monitoring. Results show that only a subset of sub-basins exerts dominant control on outlet flood response, while many others contribute marginally. The identified influential sub-basins vary with the forecasting lead time, highlighting the importance of tailoring FEWS design to operational objectives.
Overall, the proposed framework offers a flexible approach for optimizing FEWS design, supporting evidence-based decisions on sensor placement and providing new insights into the internal organization of flood-generating processes.
How to cite: Helmi, M., Cappelli, F., Dastourani, M., Kleidorfer, M., and Grimaldi, S.: Machine-learning Identification of Critical Sub-Basins for Optimized FEWS Design, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-10194, https://doi.org/10.5194/egusphere-egu26-10194, 2026.
