Flash floods have the potential to damage infrastructure and buildings, and pose a considerable threat to human life. The short lag times associated with flash floods (typically a few hours) present a significant challenge to existing flood warning systems. These systems currently rely on rainfall and runoff measurements, as well as hydrological models. Consequently, they are of limited applicability in ungauged catchments. This is a critical issue given that climate change is intensifying extreme rainfall and thereby increasing the potential for flash flooding.

This study investigates the detection of flash floods based solely on rainfall characteristics, thus eliminating dependency on runoff measurements and hydrological infrastructure. Using high-resolution radar rainfall data and 15-minute runoff observations, 1,330 extreme rainfall-runoff events are selected across 147 German catchments with an area of up to 100 km². These events are subsequently classified as either flash or non-flash floods using a rainfall-runoff-based classification scheme as a reference, with 103 of the selected events identified as flash floods. For each event, various space-time rainfall characteristics are quantified. A random forest model for flash flood detection is then trained using only the rainfall metrics and static catchment attributes. The main aim is to assess the potential for rainfall-driven flash flood detection without relying on runoff. In addition, the most relevant rainfall characteristics associated with flash floods are identified, thereby improving our understanding of the underlying drivers.

In future work, the developed detection approach will be combined with real-time rainfall nowcasting to enable earlier prediction and warning of flash floods.

How to cite: Brandt, A. and Haberlandt, U.: Rainfall-Driven Flash Flood Detection: A Framework for Ungauged Basins, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-9484, https://doi.org/10.5194/egusphere-egu26-9484, 2026.

Flash floods represent a major hydrogeological hazard in fast-responding coastal catchments of southern Italy, where intense and short-duration rainfall can lead to sudden increases in stream water levels. In this context, the Amalfitan coast area constitutes a particularly relevant case study, as it has been historically affected by extreme meteorological events, flash floods, and hydrogeological instability, resulting in significant impacts on urban areas and infrastructure. This study presents the development and evaluation of a Random Forest algorithm, aimed at predicting stream water levels and analyzing conditions likely to trigger flash floods.

The model relies exclusively on dynamic data continuously acquired through an IoT-based monitoring network deployed within the study basin, installed in the municipality of Amalfi. The network includes soil water content and soil suction sensors installed at shallow depths, allowing the characterization of hydrological conditions within the topmost soil layers. These measurements are complemented by a stream level sensor and rain gauges distributed across the basin. The integration of these variables enables the definition of relationships between weather forcing and hydrogeological response of the catchment.

The available dataset was split into training and testing subsets to evaluate model performance. The Random Forest model predicted stream water level dynamics and identified potential flash flood conditions, with accuracy assessed using established performance metrics. The integration of in-situ IoT monitoring data and Machine Learning provides a powerful approach for flash flood prediction, as continuous environmental measurements can be automatically analyzed to identify early-warning signals, capture complex interactions between rainfall and stream water level, and support real-time decision-making in highly dynamic catchments. The future integration of the model into an operational early warning system is considered as a potential advancement, with the aim of enhancing flood risk management and mitigation strategies in Amalfi and similar high-risk catchments.

How to cite: Menichini, R., Pecoraro, G., and Calvello, M.: Integrating IoT Monitoring Data and Machine Learning for Flash Flood Forecasting: A Case Study in Amalfi, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-9497, https://doi.org/10.5194/egusphere-egu26-9497, 2026.

From September 12-20, 2024 heavy rainfall events and massive flooding occurred in Austria due to the low-pressure system "Boris". Lower Austria and Vienna were severely affected by flooding and associated on- and off-site impacts. Flash flood discharge values in Vienna's torrent systems were particularly extreme – some experiencing a 1000-year event magnitude. It is evident that heavy precipitation events have shown a clear increasing trend in recent decades due to climate change, and these events are highly likely to continue to increase in the future, in some cases with still unforeseeable consequences. The project FlaMoVie (Flash Flood Monitoring Vienna, 2024–2026) is dedicated to (i) investigate the causes, course, and potential consequences of flash flood events in Viennese torrent catchments, and (ii) assess the determination of associated hydrological and geomorphological effects (incl. different climate change scenarios) using a hydro-geomorphological monitoring and modelling approach. In this contribution, we will highlight some monitoring results derived from multi-temporal Terrestrial Laser Scanning, field mapping/measurements, and hydrometeorological gauge data of the Alsbach system, i.e. a ca. 2 km² large flash-flood-prone, densely forested, torrential catchment in the northwest of Vienna, Austria. Field data is further integrated in CAESAR-Lisflood landscape evolution modelling using yielding high-resolution rates of erosion and sedimentation across the catchment. The investigations at the Alsbach allow to deduce important implications for the hydro-geomorphological response to torrential rainfall in forested small-scale headwaters in Viennese torrent catchments.

How to cite: Pöppl, R., Walk, J., and Marr, P.: Flash Flood Monitoring in Vienna, Austria (FlaMoVie), EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-12033, https://doi.org/10.5194/egusphere-egu26-12033, 2026.

Rapid urbanization and climate change are increasing the frequency and severity of floods, posing significant threats to both lives and infrastructure. To mitigate these risks, it is essential to enhance the alerting and communication mechanisms for pluvial flash floods, thereby improving community resilience and reducing losses.

This study proposes an effective, citizen-oriented Early Warning System (EWS) implemented and tested in the heritage city of Matera, Italy. This EWS aims to empower citizens by improving their understanding of local flood risks, enabling them to assess their personal exposure and the potential characteristics of floods that may affect them. This knowledge allows individuals to make informed decisions about when to act and which life-saving measures to take.

The system integrates Artificial Intelligence (AI) for flood monitoring, flood modeling, and risk communication.  Internet of Thing (IoT)-based cameras combined with deep learning algorithms, specifically the You Only Look Once (YOLO) model, estimate flood water depth and car submergence levels. Additionally, flood surface velocity can be computed using the Fudaa-LSPIV (Large-Scale Particle Image Velocimetry) method. A deep convolutional neural network (CNN) model has been developed for rapid and accurate real-time prediction of water depth and flow velocity of forecasted urban flash flood scenarios. The EWS includes threshold-based alerts concerning flood instability for pedestrians and vehicles, accompanied by signals and designed symbols for communicating risk and self-protection measures to enhance citizen resilience.

 Overall, the proposed citizen-oriented EWS is not intended to replace existing systems from competent authorities but to complement existing systems by fostering "flood literacy" among citizens. Furthermore, this research can assist municipal authorities in emergency management by providing reliable information about the timing of flood recession, which is crucial for prioritizing the accessibility of affected areas and determining which roads should be restored for traffic in the short term.

How to cite: Albano, R., Asif, M., Mishra, M., Ermini, R., and Sole, A.: Artificial Intelligence-driven early warning system for flood risk management in urban areas, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-12724, https://doi.org/10.5194/egusphere-egu26-12724, 2026.

Flash floods are among the most destructive and unpredictable hydrometeorological hazards, frequently causing severe economic damage and loss of life. In Germany, the intensity and frequency of such events are projected to increase under ongoing climate change, underscoring the need for robust flood risk management supported by comprehensive spatial data and hazard information. However, a simple, homogeneous, and nationwide model for flash flood hazard assessment is still lacking.

This study presents the development and testing of an uncalibrated, index-based approach for flash flood hazard assessment in Germany, utilizing exclusively freely available and nationwide homogeneous geospatial datasets. Based on an extensive literature and data review, the Flash Flood Potential Index (FFPI) was identified as a suitable indicator for estimating flash flood susceptibility. A Python-based model was developed to calculate the FFPI using four key parameters—slope, land use, tree density, and soil type—derived from open national geodata. The relative weighting of these parameters was determined using the Analytic Hierarchy Process (AHP) method. The model was applied to three study areas in Germany representing diverse topographic and land cover conditions, and tested with varying parameter weightings and digital elevation model (DEM) resolutions.

In addition, a novel, supplementary module was implemented to compute FFPI-weighted flow accumulation, enabling the identification of downstream areas potentially affected by flash flood propagation. Test results indicate that the proposed modelling framework and additional module are suitable for flash flood hazard assessment across Germany, with four out of five predefined model expectations satisfactorily fulfilled. With further calibration and refinement, the model is expected to provide a cost-effective, transferable, and operationally simple tool for nationwide flash flood hazard estimation, contributing to improved risk management and early warning capacities under changing climatic conditions.

How to cite: Geiß, C., Raven, S., Aravena Pelizari, P., Taubenböck, H., and Greve, K.: An Index-Based Approach for Flash Flood Hazard Assessment in Germany Using Freely Available Geospatial Data, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-18013, https://doi.org/10.5194/egusphere-egu26-18013, 2026.

We present probabilistic warning tools for flood hazards and risks based on deep learning (DL) techniques. The scope is on nowcasting heavy rainfall and associated flash floods in short time ranges (0-3 hours) and at high spatial and temporal resolutions (2 km and 15 minutes). Rainfall nowcasts are produced from pan-European OPERA radar composites by using a convolutional neural network based on the SimVP architecture. Two post-processing techniques are applied to enhance the utility of the DL-based nowcasts. First, underestimation of heavy rainfall is reduced by applying quantile mapping. Second, ensembles that provide realistic estimates of forecast uncertainty are generated by utilizing a stochastic technique. With these enhancements, the DL-based nowcast is shown to outperform the traditional extrapolation-based nowcasting techniques. The rainfall nowcasts are translated into color-coded hazard levels by using user-specified thresholds and statistically optimized probability thresholds that maximize hits and minimize false alarms. These are further translated into flood risk levels by using exposure information. Real-time feed of the warning products is displayed in a web platform developed in the EU-funded INLINE project. Demonstrations of the proposed methodology are given using major flood events during the years 2024 and 2025 that affected multiple European countries.

How to cite: Pulkkinen, S., Myllykoski, H., Baugh, C., and Berenguer, M.: Probabilistic rainfall and flash flood nowcasting on pan-European scale, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-20130, https://doi.org/10.5194/egusphere-egu26-20130, 2026.

Antecedent precipitation estimates and precipitation forecasts are critical inputs for flood forecasting systems, particularly in basins where flood response is strongly controlled by short-term rainfall variability rather than by slowly evolving catchment states. In regions with sparse rain gauge networks, satellite-based precipitation products and numerical weather prediction (NWP) models are therefore frequently relied upon, despite their known uncertainties.

This study presents a post-event evaluation of rainfall estimates and forecasts during four recent major flood events in the state of Rio Grande do Sul, southern Brazil — a region that has experienced recurrent and increasingly severe flooding in recent years. The analysis considers both simulation and operational forecasting contexts, assessing the performance of near–real-time satellite precipitation products (IMERG Early Run and GSMaP Near Real-Time) and short-range precipitation forecasts from two widely used NWP systems: ECMWF and NCEP.

Satellite products were evaluated against telemetric rain gauge observations for historical flood events (2023–2025) as well as for a longer reference period (2018–2025), using standard performance metrics. Results indicate that IMERG Early Run outperformed GSMaP Near Real-Time in terms of bias and overall representativeness, particularly under data-scarce conditions.

The analysis of NWP forecasts during the extreme April–May 2024 flood revealed substantial limitations even at a 24-hour lead time. Rainfall underestimations of up to 60 mm (basin-average) were identified in the Guaíba basin during peak impact periods, while spatial displacement of rainfall maxima further reduced forecast usability.

These results highlight that improvements in hydrological modeling alone are insufficient to enhance flood forecasting reliability. Advancing rainfall estimation and predictability — through improved satellite products, enhanced data merging strategies, and more accurate meteorological forecasts — remains a critical challenge for flood early warning systems in Southern Brazil and similar hydroclimatic regions.

How to cite: Cristina de Oliveira, R., Petry, I., Mainardi fan, F., and Sampaio Medeiros, M.: Post-Event Evaluation of Rainfall Estimates and Forecasts for Major Floods in Southern Brazil, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-20249, https://doi.org/10.5194/egusphere-egu26-20249, 2026.

Rainfall-induced landslides pose a major threat to communities across Europe. This study analyses the performance of a prototype Landslide Early Warning System developed within the GOBEYOND project (Project: 101121135) that uses datasets (landslide susceptibility and precipitation) available across Europe.

The prototype is driven by gauge-adjusted precipitation composites from EUMETNET OPERA and the ELSUS v2 susceptibility map (Wilde et al., 2018). The system is designed to provide real-time landslide warnings in regions where local models are not available. Rather than replacing existing systems, the prototype aims to fill gaps in continental-scale monitoring.

The system’s performance is evaluated in two distinct contexts. In Catalonia (NE Spain), the framework is applied to a continuous inventory (2024–2025) to systematically benchmark the European prototype against the local system, which utilizes high- resolution susceptibility data and radar-gauge QPE. Conversely, in the Alpes-Maritimes (SE France), the analysis adopts a targeted approach, focusing on the validation of specific recent events using inventory data.

To account for uncertainties in inventory data (affected by location inaccuracies or imprecise reported dates), the study combines EDuMaP with a fuzzy approach. Preliminary results demonstrate that performance is sensitive to the temporal evaluation window. Extending the window from 24 to 72 hours more than doubled the proportion of correctly detected landslide events, suggesting that the prototype successfully identifies the hazardous conditions when accounting for reporting delays.

The European prototype effectively captures widespread triggering conditions, but its current calibration results in excessive over-forecasting. Therefore, it serves as a valuable baseline hazard indicator for regions with limited data, establishing a homogeneous observation standard across the continent.

How to cite: Tapia Hurtado, L. A., Berenguer, M., Bernardie, S., Park, S., and Sempere-Torres, D.: Evaluating a Prototype Europe-Wide Landslide Early Warning System: High-resolution comparative analysis in Catalonia and Alpes Maritimes, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-20477, https://doi.org/10.5194/egusphere-egu26-20477, 2026.

Mountain regions exhibit complex topography and climate patterns, leading to highly variable meteorological conditions that complicate the prediction of heavy rainfall, flash floods, and landslides. Gridded radar rainfall data are therefore essential for monitoring and forecasting intense storms in mountainous catchments where rain gauge observations are limited. The reliability of near-real-time radar rainfall products depends on individual radar data quality, radar compositing techniques, and bias correction procedures. Hydrological modelling is a key component of operational Early Warning Systems (EWS) for monitoring and forecasting runoff conditions; however, model accuracy remains sensitive to uncertainties in measurement instruments, rating curves, and rainfall inputs. This study aims to investigate the impact of rainfall input quality on the accuracy of flood simulations in a mountainous catchment in northern Thailand, namely the Klong Suan Mak basin. Radar compositing was performed using data from three weather radar stations: Omkoi (approximately 180 km northwest), Takhli (approximately 170 km southeast), and Chainat (approximately 167 km southeast) relative to the Klong Suan Mak basin, with a quality-index-based approach applied to rainfall estimation. A spatially distributed, physically based hydrological model for flash flood simulation was driven by three rainfall inputs: (i) rain gauge observations, (ii) an event-based bias-corrected radar composite, and (iii) an hourly Kalman filter–based bias-corrected radar composite. Model calibration was performed using the Flow Duration Curve (FDC) approach in logarithmic scale, based on discharge observations and spatial rainfall inputs during the 2022 flood events.

Results clearly indicate that the sparse rain gauge network in the mountainous region yields the poorest flood simulation performance. In contrast, radar-based rainfall products exhibit more complex behavior: although the hourly Kalman filter–based product provides the highest rainfall data quality, variability in its bias factor increases uncertainty in model parameter estimation. By comparison, the event-based bias-corrected radar composite with a single bias factor yields more stable model parameters, making it more suitable for both calibration and validation across multiple events. Consequently, the event-based radar rainfall product was adopted as the baseline input to stabilize model parameters and subsequently integrated with the dynamic Kalman filter–based product, leading to a significant enhancement in model performance and improving prediction accuracy by up to 32%, particularly during high-discharge periods where the dynamic Kalman filter–based radar rainfall data exhibited a significant improvement in efficiency. This integrated approach has strong potential to support multi-hazard mitigation, including landslides and soil erosion. When combined with short-term radar rainfall nowcasting, it could provide critical lead time for disaster preparedness and national early warning systems.

How to cite: Bogaard, T., Puttaraksa Mapiam, P., and Budrach, T.: Influence of High-Resolution Radar Rainfall Data Quality on Flash Flood Simulation Performance for Operational EWS in Mountainous Thailand, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-21318, https://doi.org/10.5194/egusphere-egu26-21318, 2026.

Flooding is one of the costliest natural disasters; from 1980 to 2024 it has cost the United States over $200 billion cumulatively according to NOAA. Texas has been impacted by extreme flood events, including Hurricane Harvey (2017), Tropical Storm Imelda (2019), and the July 2025 floods which tragically caused over 100 fatalities. We focus on Southeast Texas, where low-relief terrain contributes to compound floods driven by fluvial and pluvial forcings. This work is part of the Southeast Texas Urban Integrated Field Laboratory (SETx-UIFL), where we collaborate with regional stakeholders—representing local government agencies, practitioners, community organizations, and industry partners—through task forces with whom we identify areas of concern, share scientific findings, and co-create actionable flood information. Compound flooding is computationally expensive to model at high resolution because coupled physical models are necessary to accurately capture compound flood processes and feedbacks. We downscale coarser results from the Advanced Terrestrial Simulator (ATS), a fully distributed surface-subsurface hydrologic model that includes compound fluvial-pluvial flood processes, to map flood inundation at residential block scale (1 to 3 m). We force ATS with ensembles of synthetic storm events generated using flood frequency analysis and stochastic storm transposition. We develop and apply a volume-conservative downscaling technique to the ensembles of ATS flood output, increasing resolution from an unstructured mesh with element edge length O(100 m) to a regular grid with element edge length O(1 m). We compute probabilistic flood maps by calculating the annual recurrence interval (ARI) at each pixel in our downscaled product, and validate against an extensive local gage network and FEMA's 100-year ARI floodplain. To translate probabilistic flood maps into actionable information co-developed with stakeholders, we perform impact analysis on urban centers within our study area: we intersect downscaled inundation maps with population data, building footprints, and transportation infrastructure. We also classify flood depths using human-meaningful thresholds to communicate flood impacts intuitively. This approach provides a nuanced understanding of flood risk by illustrating spatial variations in flood probability and quantifying impacts on people and infrastructure. 

How to cite: Wang, M., Passalacqua, P., Coon, E., Rathore, S., and Perez, G.: Probabilistic Flood Mapping and Impact Analysis from Downscaled Compound Flood Ensembles in Southeast Texas, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-2858, https://doi.org/10.5194/egusphere-egu26-2858, 2026.

The UK Hydrological Outlook (UKHO) provides sub-seasonal to seasonal forecasts of river flows and groundwater levels, providing insight into possible future hydrological conditions over the UK (https://ukho.ceh.ac.uk/). In 2012 the extreme drought–flood transition highlighted the need for a proactive anticipatory system to support better water resource management, and led to development of the UKHO. Since then the UKHO has evolved to provide ensemble forecasts encompassing multiple methods, including Ensemble Streamflow Prediction (ESP) and Historical Weather Analogues (HWA), applied at a daily time-step for multiple lead-times using the catchment-based airGR model (GR6J) and the grid-based Grid-to-Grid/Water Balance Model (G2G-WBM). Work is ongoing to extend the suite of models to include additional catchment models that have previously been successfully applied to UK catchments (e.g., Hydrologiska Byråns Vattenbalansavdelning, HBV and the Probability Distributed Model, PDM). However, use of multiple ensemble methods and models can make it challenging for users and decision-makers to interpret their probabilistic forecasts effectively, especially when compared to the simplicity of a single deterministic forecast. To address this challenge, it is essential to integrate these diverse procedures to deliver skilful, standardized, and easy-to-interpret forecasts. Here, we aim to advance the UKHO by first applying bias correction and then blending ensemble forecasts based on the skill of each method and model for individual catchments at different lead times, to produce consolidated probabilistic predictions that can be utilised more simply. We evaluate several blending techniques designed for probabilistic forecasts, combining the individual strengths of different methods and models while preserving the ensemble spread, which is essential for representing forecast uncertainty. This approach will inform water resource management and support hydrological hazard mitigation by delivering forecasts that are both comprehensive and easy to understand and use for operational decision-making.

How to cite: Bulut, B., Chan, W., Chevuturi, A., Facer-Childs, K., Rhodes-Smith, M., Davis, H., Bell, V., Wallbank, J., Wells, S., Moore, R. J., and Cole, S. J.: Advancing the UK Hydrological Outlook using skill-based forecast blending, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-5012, https://doi.org/10.5194/egusphere-egu26-5012, 2026.

Accurately predicting hydrological extremes is critical for effective flood risk and water resources management, yet it remains a major scientific and operational challenge. A wide range of loss functions is available for evaluating predictive performance, ranging from well-established hydrological metrics to less known alternatives drawn from the broader statistical literature. These loss functions differ in their mathematical properties and implicit assumptions, which might lead to substantially different model behaviour and predictive skills when they are used for model calibration.

We compile and systematically evaluate a comprehensive suite of loss functions for calibrating hydrological models, with a particular emphasis on the prediction of streamflow extremes. By comparing their performance across a range of conditions, we highlight how the choice of calibration objective influences model sensitivity to high and extreme flows. Our findings provide practical guidance for selecting appropriate loss functions in hydrological modelling applications, with the aim of improving the reliability and robustness of predictions for high-impact hydrological events.

Acknowledgements: This work was funded by the Research Center on Climate Change Impacts - University of Padova, Rovigo Campus - supported by Fondazione Cassa di Risparmio di Padova e Rovigo.

How to cite: Papacharalampous, G., Marra, F., Dallan, E., and Borga, M.: Evaluating loss functions for extreme streamflow predictions, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-5477, https://doi.org/10.5194/egusphere-egu26-5477, 2026.

The WMO Research Board Task Team on Hydrology Research (TT-HydroResearch) operated from September 2023 to December 2025 with the objective of strengthening coordination, coherence, and strategic direction of hydrological research within the World Meteorological Organization (WMO). The Task Team supported integration across WMO research programmes, including the World Weather Research Programme and the World Climate Research Programme, while fostering strong links with external scientific and operational communities such as UNESCO-IHP, IAHS, HEPEX, EGU, and AGU. A central mandate of TT-HydroResearch was the review and update of the WMO Hydrology Research Strategy 2022-2030 (known now as WMO Operational Hydrology Research Strategy 2030) and the Plan of Action for Hydrology, ensuring alignment with emerging scientific challenges, operational priorities, and societal needs.

This contribution presents the rationale, objectives, and key outcomes of the Task Team’s work, with a particular focus on advancing research-to-operations (R2O) and operations-to-research (O2R) pathways in global hydrological monitoring, process understanding, and prediction. By identifying priority research gaps, promoting interdisciplinary collaboration, and strengthening the interface between science and services, TT-HydroResearch contributes to enhanced predictive capabilities and more effective hydrological services under conditions of climate change, increasing extremes, and growing water-related risks.

How to cite: Pechlivanidis, I. and Uhlenbrook, S. and the WMO TT-HydroResearch members: From Strategy to Action: Strengthening Global Hydrological Research within WMO, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-7321, https://doi.org/10.5194/egusphere-egu26-7321, 2026.

Probabilistic forecasts, when optimally calibrated, have the ability to encode both the current state of knowledge and the uncertainty about the future value of a variable. Forecasting systems developed by experts and trained on data, and optimized using proper scoring rules, are therefore well positioned to produce calibrated probabilistic forecasts. Conversely, prediction markets capture the collective state of knowledge of their participants in the form of market-implied probabilities.

If participants in such markets have access to a forecasting system, as well as additional information (e.g. local or contextual knowledge), they can trade against the forecast and thereby recalibrate the implied probabilities. This interaction has the potential either to improve forecast quality or, alternatively, to increase confidence in the forecasting system among the participants it outperforms. An additional benefit is that participation in prediction markets can help users and stakeholders develop better intuition for probabilistic forecasts and uncertainty.

In this presentation, we discuss several potential set-ups for connecting forecasting models, users, local experts, and armchair hydrologists through prediction and betting markets. We highlight theoretical connections to proper scoring rules and information-theoretic forecast evaluation, as well as practical considerations related to implementation using publicly available platforms, including their promises and limitations. Finally, the presentation will include an opportunity for the audience to put their (play-)money where their skill is and take a chance.

How to cite: Weijs, S.: Prediction markets as a bridge between probabilistic hydrological forecasting and user beliefs, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-16576, https://doi.org/10.5194/egusphere-egu26-16576, 2026.

Reliable flood forecasting systems are an important prerequisite for local authorities and flood defence units to prepare for potential flooding at an early stage and to initiate the required measures. Small catchments in mountain ranges pose particular challenges in this regard, as they respond very quickly to rainfall. Furthermore, forecasts of rainfall in terms of their spatial and temporal extent and the associated impact on the areas are subject to a high degree of uncertainty. Ensemble rainfall forecasts as input data for rainfall-runoff (RR) models allow for the evaluation of the uncertainty of the resulting runoff. However, this requires fast-computing RR models to cope with the simulation effort, for which artificial intelligence methods are being used increasingly. Against this background, two hydrologic ensemble forecasting systems (EFS) are compared and evaluated that have been in operational use for small catchments in Saxony, Germany, for two years.

System 1, called EFS-howa, was developed in the HoWa-PRO research project, and its predictions can be tracked via the warning platform https://howapro.de/. As an RR model, it includes the event-based, deterministic hydrological model DeHM. DeHM covers the hydrologic processes for runoff formation and concentration, channel routing, and the simulation of flood retention dams. Measured discharge and water level data are assimilated within the forecasting process. For the hydrological ensemble forecast, rainfall data for observation and prediction from established products of the German Weather Service are used (radar-based QPE: RADOLAN-RW, radar-based nowcasting: RADOLAN-RV, ensemble QPF: ICON-D2-EPS). The runoff forecast lead time is 48 hours, and new forecasts are released every half hour if the QPF indicates a potential flood threat.

System 2, called EFS-kiwa, was developed in the KIWA research project, and its predictions can be tracked via the web demonstrator http://howa-innovativ.hydro.tu-dresden.de/WebDemoKiwa/. The RR model is a regional AI model (based on LSTMs) that was developed using measured RR data and hydrologic characteristics from 52 small and medium-sized catchments in Saxony, Germany. The current setup of the regional AI-RR model is based on hourly measurements of rainfall (using RADOLAN-RW), runoff, and rainfall forecasts. Thus, it achieves runoff forecasts with a lead time of up to 24 hours. The regional AI-RR model also allows for the fast and robust processing of ensemble rainfall forecasts from ICON-D2-EPS, enabling runoff forecasts with uncertainty/reliability information.

Both systems are evaluated in terms of their performance. Across various forecast lead times, different metrics such as KGE or percentage peak error, as well as threshold-based metrics such as false alarm ratio or area under the ROC curve (AUC), are calculated to explore the quality of both forecasting systems. The differences between the two demonstrators are highlighted by means of the selected metrics and specific simulation results. The associated benefits, advantages, and disadvantages for flood early warning are discussed.

How to cite: Grundmann, J., Wagner, M., Morgenstern, T., Mietrach, R., and Schütze, N.: Ensemble flood forecasting in small catchment using AI-based and deterministic rainfall runoff models – a performance comparison, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-13490, https://doi.org/10.5194/egusphere-egu26-13490, 2026.

Numerical Weather Prediction (NWP) models have become an integral part of hydrological forecasting and research. Despite their advances and applications in hydrology, these models exhibit significant inherent errors due to imperfect representations of atmospheric physics, inaccuracies in initial conditions, and limitations in parameterisation schemes. Consequently, accurate quantitative precipitation forecasts (QPFs), which are critical for flood forecasting and early warning systems, remain challenging to obtain. Considering these limitations, it is essential to systematically evaluate available global QPFs derived from various NWP models before employing them as forcing inputs for hydrological models. This study evaluates the skill and reliability of three global quantitative precipitation forecast products archived in the TIGGE database - ECMWF, NCEP and NCMRWF, over the Chaliyar river basin, Kerala, against the gauge-based observation data for the Indian summer monsoon season from 2018 to 2023. The forecasts are evaluated at multiple lead times using a comprehensive set of deterministic and probabilistic metrics. The skill of the control forecasts is quantified by the correlation coefficient and root-mean-square error (RMSE), whereas perturbed forecasts were assessed using the mean Continuous Ranked Probability Score (CRPS). The analysis indicated that the NCMRWF model achieved the highest correlation skill, with values of 0.64 and 0.41 at 1-day and 2-day lead times, respectively, outperforming both ECMWF and NCEP. In terms of forecast errors, RMSE values indicated that ECMWF produced lower errors than NCMRWF and NCEP at both 1-day and 2-day lead times. In terms of probabilistic performance, NCMRWF achieved the lowest mean CRPS at a 1-day lead time, followed by ECMWF and NCEP. However, its probabilistic skill declined at the 2-day lead time, as indicated by higher CRPS values. Overall, both deterministic and probabilistic evaluations indicated that NCMRWF outperforms the other two models for the study area. As envisaged, forecasting skill significantly declined with increasing lead time across all models. These results highlight the need for further improvements, such as ensemble post-processing, to enhance the reliability of operational forecasting applications.

How to cite: H Babu, S. and Kumar D, S.: Evaluation of quantitative precipitation forecasts over a monsoon-dominated catchment in Kerala, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-16015, https://doi.org/10.5194/egusphere-egu26-16015, 2026.