It welcomes, without being restricted to, presentations on:
• Advances in sub-seasonal, seasonal and decadal hydrological predictions;
• Process-based, data-driven, AI, machine learning, and hybrid methods;
• Seamless forecasting techniques and applications;
• Hydro-climate forecasts and scenario-based projections of water availability and hydrological extremes (floods, droughts, compound events);
• RTFF modelling including physically/processed/conceptually/experimentally based or data-driven modelling such as artificial Intelligence (AI), machine learning (ML), Data mining (DM), and deep learning (DL) or hybrid modelling including citizen-knowledge-informed and physics-informed modelling
• Application RTFF for flood alleviation or engagement with the public and authorities, such as early warning and early action systems, citizen-knowledge based modelling, digital twins (DT), augmented reality (AR), virtual reality (VR), and mobile apps.
• RTFF: impact on flood risk management, insurance, capacity building, vulnerability assessment, and community resilience.
• Methods for post-processing and refining the hydro-climate information (e.g., downscaling, bias correction, temporal disaggregation, spatial interpolation).
• From (near) real-time monitoring to predicting water availability;
• Impact-based assessments of forecasts for decision-making
• Co-development of forecasts between scientists and service providers;
• Operational hydro-meteorological forecasting systems and hydro-climate services;
• Forecast verification, sensitivity analysis and tools; and
• Perspectives on forecast value for end users.
The session will bring together research scientists and operational managers in hydrology, meteorology and climate, with the aim of sharing experiences and foster discussions on this momentous topic. We encourage presentations with implications for early warning, water resources management, drinking water supply, transport, energy production, agriculture, disaster risk reduction, forestry, health, insurance, tourism and infrastructure.
Orals: Mon, 4 May, 14:00–18:00 | Room 2.31
The Copernicus Climate Change Service (C3S) Water Service is an operational hydrological information system delivering authoritative water and hydroclimate information at European and global scales, spanning current conditions, seasonal forecasts, and long-term climate projections. Commissioned by ECMWF on behalf of the European Commission, the C3S Water Service supports climate change adaptation and resilient water management across sectors by delivering data via interactive web applications and the Copernicus Climate Data Store (CDS). The service is delivered by a consortium of European research centres, universities and operational hydrometeorological organisations, ensuring rigorous scientific underpinning while promoting operational sustainability and user-oriented design.
The operationalisation of the C3S Water Service advances beyond experimental prototypes, engaging users to co-develop and deliver timely, quality-controlled hydrological information at global and regional scales. Meeting universal user needs, it provides variables such as river discharge, soil moisture, runoff, and snow water equivalent, alongside specific indicators defined by users, tailored to their sectors. This supports operational decision-making for flood and drought risk, water resources planning, and climate risk assessments.
The C3S Water Service adopts a multi-model ensemble framework, enabling a seamless service from seasonal outlooks to multi-decadal assessments and climate projections. Consistent re-gridding and bias adjustment are applied to the climate input data for both the seasonal forecasting system and the hydroclimate projections production. The seasonal forecast models currently used are the ECMWF SEAS5 model and CMCC-SPS3.5, with the inclusion of more systems planned over the coming years. For climate projections, a selection of members from the CMIP6 ensemble is used for the Global domain, and CMIP6 EURO-CORDEX members for the European domain. The hydrological models ECLand, HYPE, JULES, LISFLOOD, mHM, PCR-GLOBWB and VIC-WUR are applied consistently across both time horizons at a resolution of 5 km over Europe, and 0.1 deg. (approximately 10 km) globally. In addition, the service provides the processed meteorological forcing data, allowing future inclusion of additional hydrological models with minimal effort.
Critically, the C3S Water Service is founded on an iterative development and community engagement strategy to refine product scope, improve usability, and co-develop applications that meet evolving scientific and operational needs. Through ongoing workshops, user forums, and collaborative research activities, we invite hydrologists, hydroclimate modellers, data scientists, water practitioners, and policy makers to contribute insights, test emerging products, and shape the future trajectory of the service.
How to cite: Facer-Childs, K., Barker, L., Thober, S., Berg, P., de Graaf, I., Kelbling, M., Klehmet, K., Macleod, K., Samaniego, L., Vignola, R., Sutanudjaja, E., Wanders, N., Harrigan, S., and Cagnazzo, C. and the C3S Water Service Team: The C3S Water Service: Operational seasonal forecasts and climate change projections co-developed for the water sector , EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-7573, https://doi.org/10.5194/egusphere-egu26-7573, 2026.
The services provided by streams and rivers are conditioned by water quantity as well as quality. Streamflow and water temperature are important for aquatic biodiversity, drinking water provision, electricity production, agriculture, and recreation. We inform users about the ongoing hydro-meteorological conditions in Switzerland through the platform www.drought.ch/de, with a focus on drought-related hazards. Here we provide operational probabilistic forecasts of daily runoff anomalies and maximum water temperature for the next 32 days. These forecasts are generated twice per week using ensemble meteorological forecasts as input to the process-based PREVAH hydrological model (runoff) and to a Deep Learning Temporal Fusion Transformer (TFT) model (water temperature).
In part one, we present the setup of our TFT model and assess its predictive skill. The continuous rank probability score (CRPS) is 0.70 °C averaged over all 32 lead times, 54 stations, and 90 forecasts distributed over 1 year. It degrades from 0.38 °C at a lead time of 1 day to 0.90 °C at a lead time of 32 days, largely driven by the uncertainty of the meteorological ensemble forecasts.
In part two, we use Swiss climate projections to obtain future scenarios of streamflow (PREVAH model) and stream water temperature (TFT model). Our results highlight a projected intensification of combined drought-heat conditions under further warming. Events with an average occurrence probaility of 2% over the last 30 years are expected with 26% probability under an additional 2 ºC of warming in Switzerland. This steep increase illustrates the challenges that lay ahead to maintain the services that rivers provide today.
How to cite: Padrón, R. S., Zappa, M., Bernhard, L., and Bogner, K.: Drought-heat conditions in Swiss rivers: Operational forecasts and future projections, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-15417, https://doi.org/10.5194/egusphere-egu26-15417, 2026.
Discharge prediction is essential for water resource management, enabling a better understanding of hydrological variability and its response to environmental and human factors. We present a Spatio-Temporal Graph Geural Network (STGNN) model predicting hourly river discharge for the upper Weiße Elster, flowing into the Saale river a tributary of the Elbe. The STGNN is a novel graph-based Long Short-Term Memory (LSTM) neural network jointly modeling spatial connectivity and temporal dynamics. It uses elevation, land‑use and soil data as well as approximately one year of temporally aggregated weather variables, but no previous discharge to generate one prediction at the 8793 nodes of this catchment. It was evaluated against a traditional random forest regression model and a graph neural network without explicit temporal structure, outperforming them across multiple metrics, achieving a $R^2$ of 0.763, a Root Mean Square Error (RMSE) of 9.54e-3 mm/h and a Kling–Gupta Efficiency (KGE) of 0.753. The trained STGNN was applied to the nearby Schwarzwasser catchment and achieved a KGE of 0.618, highlighting is ability to generalize. These results show that data-driven modeling can profit from physical realism and offer an adaptable framework combining predictive accuracy and generalizability to support water resource management.
How to cite: Ninnemann, L. and Schanze, J.: Spatiotemporal Graph Neural Network for River Discharge Prediction, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-8946, https://doi.org/10.5194/egusphere-egu26-8946, 2026.
Seasonal hydro-meteorological forecasts are increasingly crucial for water resource management and risk mitigation in Mediterranean coastal regions, where climate variability and anthropogenic pressures create complex challenges. However, the application of global seasonal prediction systems at local scales remains limited by systematic biases and coarse spatial resolution, which hinder their operational use in decision-making processes. This contribution presents a two-level BCSD (Bias Correction Spatial Disaggregation) seasonal forecasting system applied to a complex orographic Mediterranean area.
The system is based on the application of EQM (Empirical Quantile Mapping), as described by Lorenz et al. (2021), and focuses on precipitation and 2m air temperature variables from 1981 to 2024. The original product is the SEAS5 ensemble forecast released by ECMWF, with a horizontal resolution of approximately 36 km. The first level covers all of southern Italy and uses ERA5-Land as the reference dataset, yielding a horizontal resolution of approximately 9 km, whereas the second level, which can be seen as a refinement, focuses on the Calabria region and uses an observed, high-quality dataset, achieving a horizontal resolution of 5 km. Finally, a water balance model, calibrated over the Crati catchment, the main Calabrian river (southern Italy), has been intensively tested, focusing on the streamflow prediction. The performance has been evaluated using bias, spatial correlation, and the CRPSS (Continuous Ranked Probability Skill Score) metrics.
The results reveal systematic biases in raw SEAS5 predictions, with precipitation consistently underestimated by up to 20 mm/month, particularly during transitional months (May, June, and September), whereas 2m air temperature exhibits a persistent warm bias of approximately +1°C. The first-level BCSD correction substantially reduces these errors across southern Italy, yielding positive CRPSS values for precipitation, especially during the summer season (JJA), and marked improvements in temperature predictions (CRPSS > 0.40). The spatial correlation increases, with precipitation average increasing by 19% and temperature by 16%. However, compared with observation, residual biases persist at the catchment scale, with winter precipitation (NDJ) remaining underestimated by more than 60 mm/month over the Crati, and autumn temperatures (SON) slightly overestimated. The implementation of the second-level BCSD effectively addresses these local-scale discrepancies, enhancing the spatial correlation and CRPSS skill scores and ensuring that the hydrological model receives minimally biased forcings, predicting realistic streamflow.
This two-stage correction framework demonstrates the system's capability to preserve probabilistic forecast skill while enabling reliable impact assessments through the hydrological modeling chain, thereby bridging the gap between global seasonal predictions and local water resource management applications. As a further step, the first-level BCSD has been operationally implemented and is freely available at https://cesmma.unical.it/cwfv2/seasonal.html, as requested by several local stakeholders. In the future, the second-level BCSD and the hydrological impacts evaluation will be operationally implemented.
Reference: Lorenz, C., et al. Bias-corrected and spatially disaggregated seasonal forecasts: a long-term reference forecast product for the water sector in semi-arid regions, Earth Syst. Sci. Data, 13, 2701–2722, https://doi.org/10.5194/essd-13-2701-2021, 2021.
How to cite: Furnari, L., Cortale, F., Lorenz, C., Kunstmann, H., Mendicino, G., and Senatore, A.: Seasonal meteo-hydrological forecasts: a two-level bias-corrected high-resolution modelling chain in a coastal Mediterranean area, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-13909, https://doi.org/10.5194/egusphere-egu26-13909, 2026.
Long-term precipitation forecasting is a critical issue for water resource management, disaster mitigation, and agricultural production. However, due to the uncertainty and dependency on global cycle events and various impacts, seasonal precipitation forecasting remains a challenge in terms of both lead time and accuracy. To address the problem, a Local-Remote AutoGRU model was developed based on the background of the Asian Monsoon Region (AMR). The structure of the model consists of four key components: screening and selection prediction factor from local and global cycle events; decomposition and extraction prediction features, including trends, periodicities, and correlations; employing the Gated Recurrent Unit (GRU) machine learning algorithm to explore the relationship between monthly precipitation and input series; and finally, composing and reconstructing the prediction series. This integrated strategy enables the prediction of monthly precipitation by leveraging local precipitation periodicity and tendency, global cycle events and grid location information. The results across 6.33 million km2 Asian Monsoon Region demonstrated the proposed model’s remarkable performance. It achieved an overall NSE of 0.816 in the total area and all 12 lead months, representing a 21.69% accuracy improvement over baseline models. Additionally, the study revealed that the ENSO-related global cycle events play the primary drivers in the AMR, contributing 25.86–33.47% impacts to monthly precipitation in 7-10 months in advance, only next to the local precipitation periodicity. This study provides an effective approach for long-term monthly precipitation forecasting, particularly for the AMR.
How to cite: Yang, Y., Li, C., Wang, Z., Chen, X., Liu, D., and Kang, B.: A Local-Remote AutoGRU Model for Long-Term Monthly Precipitation Forecasting in Asian Monsoon Region, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-2739, https://doi.org/10.5194/egusphere-egu26-2739, 2026.
The focus of this activity is on seasonal meteorological forecasts (time horizon greater than one month), which are crucial for implementing strategies to manage hydroelectric reservoirs, especially for large-capacity plants where water resources can be stored during wetter seasons to ensure availability during dry periods.
Seasonal forecasts are a cutting-edge product, consisting of a statistical synthesis of meteorological information up to seven months ahead. Since the atmosphere is a chaotic system, forecast evolution is sensitive to errors in initial conditions, limiting the ability to predict weather variations beyond 15 days. However, longer-term forecasts are possible by considering components of the Earth system that evolve more slowly than the atmosphere, such as the oceans. Seasonal forecasts rely on an ensemble of different atmospheric evolutions, from which average values and expected anomalies for upcoming seasons can be derived, compared to historical periods with available reference climatology.
Currently, there is limited evidence in the literature of using seasonal forecasts in the hydroelectric context. In this activity, their performance is assessed at the national scale of Italy for the ensemble mean of key meteorological variables: 2-meter temperature and precipitation. For temperature, results show a fair level of performance in the early months, suggesting added value compared to simple climatology, with degradation as the forecast horizon extends. For precipitation, performance is generally lower than for temperature, and the model struggles to provide more useful information than climatology at longer horizons. Snow is also considered, revealing a significant underestimation by the forecast model, and a simple correction method is proposed.
How to cite: Sperati, S.: Preliminary Assessment of Seasonal Weather Forecasts for Water Resources Management, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-9554, https://doi.org/10.5194/egusphere-egu26-9554, 2026.
Devastating large-scale droughts are increasing in frequency and severity under climate change, posing major challenges for preparedness and mitigation. Reliable information on the timing, extent, and intensity of droughts is therefore crucial. Seasonal forecasts with lead times of up to twelve months offer potential for early drought warning, but raw model output is often affected by substantial biases and temporal drifts relative to reanalysis products such as ERA5, limiting its direct applicability.
Here, we assess the skill of a global bias-corrected ECMWF SEAS5 seasonal forecast dataset (SEAS5-BCSD, DOI: in preparation), processed using the Bias Correction and Spatial Disaggregation (BCSD) method, for predicting extreme drought events at multiple time scales. For the period 1981–2024, we analyze 36 major droughts selected based on spatial extent and mean Standardized Precipitation Evapotranspiration Index (SPEI), representing the two most severe events per continent (excluding Antarctica) and accumulation period (1-, 3-, and 6-month SPEI).
Forecast performance is evaluated using probabilistic skill metrics including the Continuous Ranked Probability Skill Score (CRPSS) and the Brier Skill Score (BSS). Results show positive CRPSS skill relative to climatology for all analyzed droughts, with SEAS5-BCSD consistently outperforming uncorrected forecasts across all metrics. One-month droughts exhibit the highest predictability, while three- and six-month droughts show comparable but slightly reduced skill. Predictability varies regionally, with African droughts showing the highest skill and North American droughts the lowest. Forecast skill is highest for moderate drought thresholds (SPEI < −1) and decreases for more severe events (SPEI < −1.5 and −2), though remaining superior to climatology in most cases.
Overall, the results demonstrate that bias-corrected seasonal forecasts substantially enhance the predictability of extreme large-scale droughts and provide clear added value over both climatology and uncorrected seasonal forecasts.
How to cite: Weber, J. N., Lorenz, C., Schober, T., Dehn, H., and Kunstmann, H.: Predictability and skill of large-scale extreme droughts using global bias-corrected seasonal forecasts (SEAS5-BCSD), EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-2999, https://doi.org/10.5194/egusphere-egu26-2999, 2026.
Flood forecasting and early-warning systems have become a central non-structural strategy to mitigate the impacts of increasingly frequent hydrometeorological extremes. These challenges have become particularly critical in southern Brazil, where unprecedented flood events in 2023 and especially in 2024 resulted in record river and lagoon water levels, widespread inundation, and severe social and economic impacts. In response to these events, the state of Rio Grande do Sul has fostered scientific and technological initiatives focused on real-time hydrological forecasting systems. Within this region, the Piratini River watershed has experienced recurrent and severe urban flooding along its main urban reach, particularly affecting the municipalities of Pedro Osório and Cerrito, with historical events and more recent episodes in 2023–2024 that resulted in extensive urban inundation and persistent social and economic impacts, reinforcing its strategic relevance for hydrometeorological monitoring and early-warning actions. This watershed drains approximately 4,700 km² upstream of the main urbanized river reach, representing an intermediate-scale watershed that remains underrepresented in event-based hydrological modeling studies, particularly under conditions of limited real-time hydrometeorological monitoring, which predominantly focus on smaller catchments. Since 2023, this context has marked the beginning of structured hydrological forecasting activities, developed in direct collaboration with municipal governments and Civil Defense agencies to support decision-making during flood emergencies. Within the real-time, hourly hydrological forecasting framework being developed for this watershed, event-based hydrological modeling using the Hydrologic Engineering Center – Hydrologic Modeling System (HEC-HMS) has been adopted to represent rainfall–runoff processes. The objective of this study is to assess the robustness and structural sensitivity of different event-based conceptual hydrological model configurations implemented in HEC-HMS, addressing key operational questions related to the suitability of different combinations of loss, rainfall–runoff transformation, and routing methods, the impact of spatial discretization on model robustness under limited rainfall and streamflow monitoring, and the role of antecedent hydrological conditions and rainfall temporal concentration in controlling flood generation. Five recent extreme rainfall–runoff events were analyzed using multiple combinations of loss methods, rainfall–runoff transformation methods, baseflow representation, and channel routing schemes, as well as two spatial discretization thresholds, based on rainfall inputs from automatic rain gauges with poor spatial coverage across the watershed and on water level and streamflow data from the Pedro Osório non-automatic gauging station. These data were used for model calibration and validation, and model behavior was subsequently assessed using standard goodness-of-fit and error metrics. Results indicate that model robustness is strongly influenced by the combination of hydrological methods adopted, with configurations including explicit channel routing providing a more realistic representation of flood wave routing. Coarser spatial discretization produced more stable and robust simulations, suggesting reduced parameter uncertainty under limited rainfall station density. Finally, antecedent hydrological conditions and rainfall temporal concentration were identified as critical constraints on flood generation and forecasting reliability. The findings enhance the understanding of flood response mechanisms in subtropical lowland watersheds and provide technical guidance for the design of parsimonious and reliable event-based hydrological models to support operational flood forecasting, highlighting their relevance for climate risk adaptation in developing countries.
How to cite: Araujo Rodrigues, A., Navarini Kurz, M., Beskow, S., Leitzke Caldeira, T., Fuchs Bueno Repinaldo, H., and da Silva Teixeira, M.: Event-based hydrological modeling for real-time flood forecasting under data-scarce conditions: insights from a subtropical watershed in Brazil, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-15051, https://doi.org/10.5194/egusphere-egu26-15051, 2026.
Real-time flood risk assessment requires integrated frameworks that not only forecast flood dynamics accurately [1] but also capture how people respond to rapidly changing hazard conditions [2,3]. This study develops a novel real-time flood impact assessment framework that couples AI-driven flood forecasting (Long Short-Term Memory) with an agent-based model (ABM) to evaluate human mobility disruption and behavioural adaptation during flood events.
The outputs of the flood forecasting model are dynamically transferred to an ABM that represents urban populations, daily activity schedules, and transport networks. Agent (i.e., road users with different demographic attributes) decision-making is governed by a novel risk priority index (RPI), which integrates direct flood exposure, official communication, and demographic vulnerability of agents. Watts-Strogatz small-world network was used to consider the interaction of agents and realistically represent information diffusion, allowing the dissemination of risk awareness.
Results reveal substantial real-time impacts on urban mobility, with significant increases in travel times, particularly during peak hours. Moreover, incorporating behavioural adaptation through the RPI in agent-based modelling highlights the critical role of flood-risk information sharing among. A balanced combination of personal flood experience (Individual RPI) and socially shared information (average RPI of neighbouring agents) leads to faster and more effective formation of risk awareness across the population. The proposed AI-driven, agent-based framework enables real-time evaluation of flood impacts on population groups and transport systems, offering a powerful tool for emergency response planning and operational flood risk mitigation by local authorities.
References
[1] Piadeh F., Bakhtiari, V., Piadeh, F. (2026). Automated novel real-time framework for rainfall data imputation in flood early warning systems, Engineering Applications of Artificial Intelligence, 1164(B), p.113348. https://doi.org/10.1016/j.engappai.2025.113348
[2] Qin, H., Liang, Q., Chen, H., De Silva, V. (2024). A Coupled Human and Natural Systems (CHANS) framework integrated with reinforcement learning for urban flood mitigation. Journal of Hydrology, 643, 131918. https://doi.org/10.1016/j.jhydrol.2024.131918
[3] Bakhtiari, V., Piadeh, F., Chen, A.S., Behzadian, K. (2024). Stakeholder analysis in the application of cutting-edge digital visualisation technologies for urban flood risk management: A critical review. Expert Systems with Applications, 236, 121426. https://doi.org/10.1016/j.eswa.2023.121426
How to cite: Najjar-Ghabel, S., Behzadian, K., Piadeh, F., and Ardakanian, A.: Real-Time Flood Risk Assessment Using Coupled Agent-Based and AI-Driven Flood Forecasting Models, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-14951, https://doi.org/10.5194/egusphere-egu26-14951, 2026.
Reliable early warning is crucial for the operational stability of urban wastewater treatment infrastructure in response to emergencies that might threaten the environment and human health. In sewage treatment works (STWs), process anomalies can be driven by extreme influent loads, episodic operational interventions, or sensor faults. Therefore, under these multivariate, non-stationary conditions, it is challenging to ensure timely and robust practical response based on manual supervision alone. This study presents an AI-driven analytical framework for real-time early warning of process anomalies using multi-source sensor monitoring data. The framework offers a pipeline to transform continuous, real-time monitored sensor data into fixed-length time-series windows suitable for deep learning and real-time inference. A deep unsupervised learning model is innovatively introduced to learn multivariate dynamics and cross-variable dependencies of normal operation modes, and then to generate window-level scoring and map it to early warning alerts. To improve the interpretability, window-level alerts are aligned with timestamped, manually recorded event management logs to distinguish between sensor malfunctions and process disturbances. The framework is demonstrated on a multi-year real-world urban STW’s dataset, and evaluated based on detection timeliness, false alarm behaviour, and consistency with logged operational events. Results indicated that the proposed framework is a feasible approach to integrate contextual evidence with AI-driven early warning alarms. It also offers a promising powerful tool to support real-time anomaly diagnosis and decision-making for STWs’ operators.
How to cite: Yang, S., Behzadian, K., Coleman, C., Holloway, T. G., and Campos, L.: AI-driven Early Warning of Process Anomalies in Wastewater Treatment Plants Using Real-time Monitoring Data, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-13096, https://doi.org/10.5194/egusphere-egu26-13096, 2026.
Agent-based modelling (ABM) is increasingly recognised as an essential tool for urban flood risk assessment to analyse the response of people and transport systems under dynamically evolving flood conditions [1,2]. However, the dynamic evolution of risk awareness and the exchange of risk-related information during flood events remain inadequately represented in many existing modelling approaches [3, 4]. This study presents an integrated flood-impact assessment framework that couples a hydrodynamic flood model with a risk-based ABM to evaluate the impact of flooding on travelling population groups and road network performance.
Flood modelling is simulated using a hydrodynamic model calibrated through the sequential uncertainty-fitting algorithm, providing reliable spatio-temporal flood characteristics. These hydraulic outputs are dynamically linked to an ABM representing urban populations with realistic daily activity. People’s behavioural adaptation is governed by a novel risk priority index, which evolves based on direct flood exposure, institutional communication, and risk-information exchange. People are assumed to interact together through Watts-Strogatz small-world network, enabling realistic diffusion of risk awareness across the population.
Results show that flood-induced road closures trigger sharp increases in travel times. Agent-based analysis revealed that, among population groups, adults experienced the highest total flood exposure, followed by seniors and children. Moreover, travel mode strongly influences vulnerability, with cycling users experiencing the highest exposure levels, followed by public transit, walking, and driving users. The proposed framework provides a robust decision-support tool for evaluating how risk awareness and social interaction through an agent-based model influence road users and road network performance.
References
[1] Bakhtiari, V., Piadeh, F., Chen, A. S., & Behzadian, K. (2024). Stakeholder analysis in the application of cutting-edge digital visualisation technologies for urban flood risk management: A critical review. Expert Systems with Applications, 236, 121426. https://doi.org/10.1016/j.eswa.2023.121426
[2] Najjar Ghabel, S., Zarghami, M., Akhbari, M., & Nadiri A.A. (2019). Groundwater Management in Ardabil Plain Using Agent-Based Modeling, Iran-Water Resources Research 15, 1–16.
[3] Kunreuther, H., & Pauly, M. (2006). Rules rather than discretion: Lessons from Hurricane Katrina. Journal of Risk and Uncertainty, 33(1–2), 101–116. https://doi.org/10.1007/s11166-006-0173-x
[4] Lo, A. Y. (2013). The role of social norms in climate adaptation: Mediating risk perception and flood insurance purchase. Global Environmental Change, 23(5), 1249–1257. https://doi.org/10.1016/j.gloenvcha.2013.07.019
How to cite: Najjar-Ghabel, S., Piadeh, F., Behzadian, K., and Ardakanian, A.: Coupled Risk-aware Agent-Based Framework and Hydrodynamic Modelling for Urban Flood Impact Assessment, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-3462, https://doi.org/10.5194/egusphere-egu26-3462, 2026.
Soil freezing is a major cold region process that influences the hydrological behavior of northern catchments during winter rainfall or snowmelt events. Growing ice within the soil matrix reduces the pore space available for water to infiltrate, while the presence of soil macropores in structured soils maintains rapid water percolation even in frozen conditions. Representing the complex effect of soil freezing on water infiltration in land surface models is therefore a challenging task. This is especially the case for operational systems where the integration of a physical process must improve or maintain reasonable model performance and minimize the increase in complexity and computational cost. In this study, we propose a conceptual approach to represent the effect of macropores on frozen soil infiltration into the Soil, Vegetation and Snow (SVS) land-surface scheme used within the operational prediction systems of Environment and Climate Change Canada (ECCC). In this approach, the macropores are activated when soil moisture exceeds 55% of the available pore space. We assessed the impact of this new approach on streamflow simulations at more than 580 hydrometric stations located in the Great Lakes-St. Lawrence domain over a five-year period. The conceptual representation of macropores results in a major upgrade to the soil freezing scheme of SVS with an improvement of the Kling-Gupta Efficiency (KGE) at 88% of the stations (KGEmed = 0.55; KGEmed = 0.28) as it better captures the timing and amplitude of peak flows. Detailed analysis of a decomposed hydrograph shows that the macropore configuration increases SVS soil drainage (slow response) and reduces surface runoff and lateral flow (quick response). The SVS experiment with macropores also results in accurate simulations of freezing depth and surface meteorological variables (i.e. air and dew point temperature) which paves the way for an operational implementation of this new configuration in the numerical weather and hydrologic prediction systems at ECCC.
How to cite: Bouchard, B., Vionnet, V., Gaborit, É., and Fortin, V.: A non-explicit representation of macropores in the SVS land surface scheme improves streamflow simulations under frozen soil conditions, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-5615, https://doi.org/10.5194/egusphere-egu26-5615, 2026.
The Destination Earth (DestinE) programme of the European Commission is developing high-resolution digital twins of the Earth system to improve the simulation of extreme weather events and their impacts. Within this initiative, the global Extremes Digital Twin (G-EDT) provides meteorological simulations at 4.4 km resolution that are coupled to ECMWF’s Land Surface Modelling System (ecLand) and the CaMa-Flood routing model to produce global river discharge forecasts. As digital twins move towards operational use, there is a growing need for automated approaches to assess forecast performance as events unfold, rather than relying solely on manual, delayed, aggregated evaluations. In this context, continuous verification becomes an integral part of operational monitoring, supporting both scientific development and system oversight.
This contribution presents the methodology behind an automated framework for weekly post-event flood analysis, designed to support near-real-time monitoring of flood activity and forecast skill. The system runs on a fixed weekly cycle and analyses recent hydrological conditions using river discharge reanalysis as a proxy for observations. Flood events are identified based on exceedance of return-period thresholds.
Individual station exceedances are grouped into spatially coherent flood events using a density-based clustering approach. For each detected event, river discharge forecasts are evaluated using event-based metrics that target key flood characteristics, including peak timing error, peak magnitude error and flood duration error, assessed across lead times and affected locations. In addition, contingency-table-based verification is applied to threshold exceedances, enabling assessment of forecast event detection and discrimination using metrics derived from hits, misses, false alarms and correct negatives, such as the equitable threat score (ETS) and critical success index (CSI).
Beyond weekly reporting, the framework supports temporal analysis of forecast performance, allowing changes in skill to be tracked over time and across evolving model configurations. While the current implementation focuses on DestinE flood predictions, the methodology is generic and extensible, with planned integration of additional systems and observational datasets.
How to cite: Tanguy, M., Matthews, G., Denissen, J. M. C., Zsoter, E., Wortmann, M., Mazzetti, C., Prudhomme, C., Haiden, T., Vannière, B., Sandu, I., and Rüdiger, C.: A framework for automated event-based monitoring of flood forecast performance, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-13978, https://doi.org/10.5194/egusphere-egu26-13978, 2026.
Recent increases in urban flooding necessities the development of real-time autonomous prediction models to support forecasting and decision-making. Urban flooding is caused by the combined effects of multiple factors, such as the expansion of impervious surfaces and limitations in drainage pipe capacity. Due to this complexity, predicting urban flooding only with rainfall–runoff is insufficient. Physics–data-based hybrid approaches have been proposed to improve prediction reliability by compensating individual’s limitations. However, hybrid models that require repetitive physics-based simulations face limitations in real-time applications due to high computational costs and long execution times. To overcome these limitations, this study applies an agent-based approach that integrates and coordinates the urban flood hazard estimation process within hybrid modeling frameworks. The proposed framework consists of three agents interconnected through a graph-based orchestration structure to form an iterative analytical workflow of execution, validation, and improvement. The first agent performs physics-based hydrological modeling to reproduce rainfall–runoff processes and the temporal response of urban drainage systems. It also automates model calibration, validation, and optimal model selection. The second agent spatially predicts flood susceptibility using machine learning models based on topography, land use, soil characteristics, drainage infrastructure, and historical flood data. The models applied in this process include random forest, extreme gradient boost, artificial neural networks, long short-term memory, and tabular data–oriented foundation models (TabPFN). The final agent integrates the results of the first two agents to conduct a hazard assessment that simultaneously reflects the probability of urban flooding and its potential intensity. The integrated flood hazard modeling framework enables automated, near-real-time prediction of urban flood hazards. It can serve as a foundational dataset for advancing future urban inundation forecasting and warning systems and decision-support frameworks.
How to cite: Lee, Y., Lee, Y., Kim, M., and Lee, S.: Development of an LLM-agent-based physics–AI hybrid Model for urban flood hazard prediction, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-16226, https://doi.org/10.5194/egusphere-egu26-16226, 2026.
Urban and pluvial flooding increasingly threaten green infrastructure, particularly trees, which play a critical role in urban resilience, drainage, and ecosystem services [1]. Despite their importance, the physical mechanisms governing tree uprooting under pluvial flood conditions remain poorly quantified, especially with respect to flow characteristics, soil properties, and planting design [2].
This study presents a comprehensive experimental investigation into the impact of pluvial flooding on tree stability, with a specific focus on uprooting processes under controlled hydraulic conditions. More than 200 flume experiments were conducted using bonsai trees as scaled physical analogues of urban trees. The experiments were designed to systematically examine the influence of flow intensity, planting configuration, and soil depth on tree uprooting. Tests were carried out with both single-tree and two-tree arrangements to assess the effects of interaction between neighbouring trees. A wide range of flow conditions was imposed, representing low, medium, and high-intensity pluvial flooding scenarios, while soil depth was varied to simulate different urban planting constraints.
The results demonstrate that tree uprooting is strongly governed by design layout parameters, particularly tree height and the spacing between trees. Closely spaced trees exhibited altered flow patterns and load distributions, leading to either increased stability due to flow shielding or enhanced vulnerability due to soil disturbance, depending on the configuration. Soil depth was found to be a critical controlling factor, with shallower soils significantly reducing root anchorage capacity and increasing the likelihood of uprooting under flood conditions.
Analysis of flow intensity revealed the existence of a threshold behaviour. While low-intensity flows generally resulted in negligible structural response, medium- and high-intensity flows produced comparable levels of hydrodynamic loading, with no substantial increase in uprooting probability beyond a critical flow threshold. This indicates that once a certain hydraulic forcing is exceeded, additional increases in flow intensity do not proportionally amplify adverse impacts. Instead, the transition across this threshold marks the onset of significant instability and uprooting risk.
These findings highlight the non-linear nature of tree–flow–soil interactions during pluvial flooding and underscore the importance of considering layout design and subsurface constraints in urban tree planting strategies. The identification of critical thresholds for adverse impacts has practical implications for flood-resilient urban planning, suggesting that appropriate spacing, height selection, and soil depth provision can substantially enhance tree stability under extreme rainfall events.
[1] Piadeh, F., Bakhtiari, V., Piadeh, F. (2026). Automated novel real-time framework for rainfall data imputation in flood early warning systems, Engineering Applications of Artificial Intelligence, 1164(B), p.113348. https://doi.org/10.1016/j.engappai.2025.113348
[2] Défossez, P., Veylon, G., Yang, M., Bonnefond, J.M., Garrigou, D., Trichet, P., Danjon, F. (2021). Impact of soil water content on the overturning resistance of young Pinus Pinaster in sandy soil, Forest Ecology and Management, 480, p.118614.
How to cite: Piadeh, F.: Threshold Behaviour and Design Controls on Tree Uprooting during Pluvial Flood Events: A Hybrid AI and Experimental Flume-Based Study , EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-5199, https://doi.org/10.5194/egusphere-egu26-5199, 2026.
The growing reliance on early flood warning systems reflects broader advances in real-time flood forecasting, where increasing data availability, modelling complexity, and the integration of data-driven and AI-based approaches have been identified as critical enablers for effective urban flood risk management [1]. To investigate if flood warning distribution is affected by hydrological variables, socioeconomic vulnerabilities or physical exposure, this study evaluates the spatial and temporal trends of the Environment Agency (EA)’s flood warnings between 2006-2024 within London boroughs.
By applying a comprehensive methodology incorporating GIS analysis, socioeconomic correlation analysis and time-series integration, Flood Alert (FA), Flood Warning (FW) and Severe Flood Warning (SFW) data were analysed using EA’s open-access data. Significant spatial variations were identified within the analysis. Riparian boroughs along the River Thames corridor experienced significantly higher FW rates (>293 alerts per borough approximately per year) compared to non-riparian boroughs (<80 alerts per borough), implying that hydraulic exposure was the primary source of flooding. Temporal analysis presented significant winter seasonality, with approximately 55% of total FW distributed between 2006-2024 confirming limited soil moisture conditions and Atlantic-driven storm activity. Summer FA’s decreased within recent years despite climate change predictions estimating increased convective rainfall, indicating either potential inadequate warning levels of pluvial risks or enhanced operational discrimination.
Regardless of population and exposure controls, distribution analysis presented severe equality implications. London boroughs with increased deprivation projected disproportionately higher FW effects per capita. The trend remained consistent throughout the study period, demonstrating compound vulnerability across the intersection of reduced adaptive capacity and physical hazard exposure. Identifying specific city hotspots (Newham, Tower Hamlets, Southwark) involving higher FA, FW, SFW frequencies occurring simultaneously with socio-economic disadvantages, aging drainage systems and limited green infrastructure.
Findings suggest the requirement for change in policy and procedures for flood risk management, including the development of locally adapted, topography-specific operational guidelines for hydrologically maintained confluence zones; improved accountability via open reporting of operational metrics for lead times and false alarm rates; increased socioeconomically inclusive flood communication strategies, prioritising the engagement of vulnerable communities; improved monitoring design to mitigate coverage gaps; and adoption of comprehensive flood risk frameworks, integrating socioeconomic vulnerability risk assessments with physical exposure controls.
This study enrichens the understanding of operations of urban flood warning systems with complex socioeconomic and technical environments. The methodology and research findings demonstrate reproducible procedures for the analysis of operational flood warning systems within local and regional scales, promoting evidence-based advancements in urban resilience planning in accordance with UK Flood and Water Management Act 2010 and EU Floods Directive (2007/60/EC).
How to cite: Cekalska, N. and Behzadian, K.: Flood Warnings Risks: Spatial and Temporal Distribution of Flood Warning Alerts in London, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-20102, https://doi.org/10.5194/egusphere-egu26-20102, 2026.
Urban flood risks are characterized by substantial uncertainties, which pose challenges for deterministic inundation predictions in risk-informed applications. Consequently, ensemble-based probabilistic flood forecasting has gained increasing attention for its ability to indicate the likelihood of extreme inundation events and associated damage risks. This study presents an efficient and generalizable diffusion-based super-resolution (SR) framework for rapid, ensemble-based, high-resolution urban flood forecasting. The framework first employs a two-dimensional hydrodynamic model to simulate flood dynamics over extensive urban areas at a coarse spatial resolution (100 m). The resulting simulations are subsequently downscaled to a fine spatial resolution (5 m) using a conditional diffusion model that performs single-step, distillation-free super-resolution. By leveraging the inherent stochasticity of diffusion models, the framework naturally supports ensemble generation, allowing for uncertainty quantification in high-resolution inundation predictions. Applied to the Beijing Sub-Center, the model efficiently simulates the spatiotemporal flood dynamics of a 24-hour rainfall event in less than 10 minutes, producing high-fidelity, fine-scale flood inundation maps at substantially reduced computational cost. The integrated framework provides a scalable and uncertainty-aware pathway for real-time, high-resolution urban flood forecasting and ensemble-based scenario analysis in large metropolitan regions.
How to cite: Yin, B. and Li, R.: Efficient and Generalizable Ensemble Urban Inundation Forecasting with Diffusion-Based Super-Resolution, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-2411, https://doi.org/10.5194/egusphere-egu26-2411, 2026.
Reliable medium- and long-term streamflow forecasting is critical for water resources management and hydropower generation. This study proposes a 60-day streamflow forecasting framework that systematically integrates a convolutional neural network (CNN) for bias correction of precipitation forecasts from the UK Met Office (UKMO) numerical weather prediction model, the Geomorphology-Based Eco-Hydrological Model (GBEHM) for streamflow simulation, and an autoregressive with exogenous input (ARX) model for statistical post-processing. Applying the proposed framework to the Upper Yangtze River Basin, results indicate that the CNN model reduces the areal-averaged precipitation root mean square error (RMSE) by around 35% and elevates the temporal correlation coefficient (TCC) from 0.62 to 0.74 against raw UKMO forecasts across the 60-day horizon, with performance gains amplifying at longer lead times. Subsequently, when driving the GBEHM with corrected precipitation and applying ARX post-processing, the streamflow forecasts exhibit substantial enhancements with a reduction in RMSE of 36%, a decrease in relative error (RE) from 48.2% to 17.4%, and an increase in Nash–Sutcliffe efficiency (NSE) from 0.33 to 0.72 compared to those driven by raw forecasts in terms of 60-day mean performance. Error decomposition identifies precipitation forecast errors which intensify with lead time as the dominant source of uncertainty for medium- and long-term streamflow forecasting, while confirming that hydrological model uncertainty remains a significant component, highlighting that the selection of a robust hydrological model is crucial for enhancing the reliability and predictive skill of the streamflow forecasts. By systematically leveraging the CNN to mitigate drifting meteorological biases, the GBEHM to capture physical catchment dynamics, and the ARX to minimize residual errors, the proposed framework yields volumetrically accurate and temporally consistent forecasts across an extended 60-day horizon, providing valuable decision support and sufficient lead time for regional water management.
How to cite: Liu, Z., Yang, H., and Yang, D.: A 60-day streamflow forecasting framework coupling deep learning bias correction with process-based hydrological modeling in the Upper Yangtze River Basin, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-3301, https://doi.org/10.5194/egusphere-egu26-3301, 2026.
The spatial discretization of hydrological sub units (HSU) is an inevitable and effective way to achieve refined distributed simulation. It can not only strengthen the distributed characteristics of the model, but also help simulate the runoff process on a small-scale watershed. However, the spatial variability of the runoff characteristics of HSU divided on a smaller spatial scale increases, and the hydrological process mechanism shows different characteristics under the refined spatio-temporal scale, which then forms the spatio-temporal scale effect. Therefore, it is of great significance to study the theory of spatial discretization of basin hydrological simulation at different spatio-temporal scales, construct a division method for runoff simulation HSUs, quantitatively describe and calculate the corresponding runoff characteristics, develop corresponding runoff simulation methods, and construct a refined distributed basin hydrological model.
In this paper, We studies the influence of the spatial scale of HSU on the runoff simulation of hydrological models, and proposes an optimization method based on the division of HSU scales, the matching of input data time scales, the quantitative calculation of model parameters, and the evaluation of basin applicability. With the refinement of the HSU scales, the spatio-temporal resolution of precipitation and evaporation input data should be improved accordingly to ensure the precise matching of rainfall evaporation process with hydrological response. Specifically, the calculation units were first divided into different scales, gradually refined from large scale to small scale, and the time scale changes of precipitation and evaporation data were simulated, using time steps of 3 hours, 2 hours and 1 hour respectively to improve the resolution of the hydrological response of the basin. Secondly, the input parameters of the Xin'anjiang (XAJ) model were optimized based on the quantitative calculation of the basin's underlying surface characteristics (such as area, morphological factors, river network density, slope, etc.). By combining quantitative calculations and empirical derivations at different scales, the effects of confluence parameters at different scales were analyzed. Finally, this paper verifies the rationality of the basin division, especially by evaluating the closure of the calculation unit based on DEM and bedrock depth data to ensure that each calculation unit has the hydrological mechanism characteristics required for hydrological model. In order to optimize the spatial scale of the calculation unit, a multi-objective optimization algorithm (Pareto frontier optimization) was used, and the rationality of the basin selection was verified through empirical research, thus ensuring the the model’s reliability. The results show that with the refinement of the calculation unit scale, the temporal and spatial scales of the precipitation evaporation input data and the hydrological response are better matched, but if the unit scale is too small, it may not meet the requirements of basin closure and hydrological mechanism. Therefore, the selection of the calculation unit scale should comprehensively consider the basin characteristics, the temporal and spatial scales of the data and the model mechanism. The reasonable calculation unit scale should usually not be less than 100 km² to ensure the accuracy of the model and the reliability of the mechanism.
How to cite: qian, R., Cao, Y., Fang, Y., and Zhang, X.: Scale Effects of Distributed Hydrological Simulation: Forcing, Structure and Mechanism, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-4354, https://doi.org/10.5194/egusphere-egu26-4354, 2026.
Reliable subseasonal streamflow forecasting is essential for flood risk management, reservoir operation, and water allocation in large monsoon-dominated river basins such as the Godavari River Basin, India. However, forecast skill remains constrained by uncertainties in meteorological forcing and hydrological model structure, particularly at longer lead times. This study evaluates the relative contributions of precipitation input uncertainty and hydrological model uncertainty within a 1–4 week (up to 30-day) streamflow forecasting framework by integrating subseasonal-to-seasonal (S2S) precipitation forecasts with a physics-based distributed hydrological model. Deterministic and ensemble precipitation forecasts from the S2S Hydrological Simulation System are used to drive the Soil and Water Assessment Tool (SWAT), with precipitation bias correction implemented through empirical quantile mapping using the India Meteorological Department (IMD) 0.25° × 0.25° gridded rainfall dataset. The Godavari basin is discretized into headwater, midstream, and large downstream sub-basins, and simulated streamflow forecasts are evaluated against Central Water Commission (CWC) daily discharge observations. Forecast performance is assessed across lead times using both deterministic and probabilistic skill metrics, including coefficient of determination (R²), Nash–Sutcliffe efficiency (NSE), root mean square error (RMSE), percent bias (PBIAS), and flow-regime-specific diagnostics for high-flow (≥90th percentile) and low-flow (≤10th percentile) conditions. Results show a systematic decline in forecast skill with increasing lead time, with substantial variability across precipitation products, basin scales, and flow regimes. Bias correction of S2S precipitation significantly reduces systematic discharge errors and enhances forecast skill up to 3–4 weeks, particularly for low-flow conditions and larger downstream sub-basins. While hydrological model structure dominates forecast uncertainty at shorter lead times, precipitation forcing uncertainty becomes the primary source of error at longer lead times. Overall, the study demonstrates the value of jointly evaluating meteorological and hydrological uncertainties and highlights the potential of subseasonal hydrological forecasting to support operational flood early warning and water management decisions in large, regulated, monsoon-driven river basins.
How to cite: Singh, A. and Rathinasamy, M.: Subseasonal Streamflow Forecasting in the Godavari River Basin: Assessing Meteorological and Hydrological Uncertainties under Monsoon Conditions, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-10524, https://doi.org/10.5194/egusphere-egu26-10524, 2026.
Spring water is a vital resource and a cornerstone of Austria’s drinking water supply, providing roughly half of the national demand and serving as the sole source in some regions. Ongoing climate change is rapidly modifying hydrological conditions, particularly in the Alpine region. Understanding the future evolution of spring discharge dynamics is therefore essential for the sustainable management of Austria’s water resources. This study investigates the impact of climate change on 76 alpine springs distributed across Austria. For each spring, several potential catchments are delineated and a rainfall–runoff model incorporating a snow module is used to determine the most plausible catchment. Subsequently, the model is driven by climate input data from three RCP scenarios (2.6, 4.5, and 8.5) to assess climate change impacts up until the year 2100. Comparing three model periods (historical reference, near future, and far future) enables a systematic assessment of temporal changes and climate change impacts. Analyses are carried out for individual springs and for hydrologically classified groups with specific discharge characteristics. The results reveal a pronounced shift in seasonal discharge especially for fast-responding and snow-dominated springs, characterized by a strong increase in discharge during spring months for snow-dominated springs and a marked decrease in summer discharge for both fast-responding and snow-dominated springs. Furthermore, the timing of the 7-day minimum flow shifts into the summer season for all spring groups. Examining the spatial patterns of individual springs across Austria reveals that, in the near future, a decrease in total discharge is projected in the southwest of Austria under all RCP scenarios. In contrast, the far future shows an improvement for alpine springs under RCP 2.6 and 4.5, whereas under RCP 8.5, decreases in total discharge are projected to become more widespread across Austria. These findings offer a meaningful reference for future water management planning in Austria, highlighting potential trends while acknowledging scenario-based uncertainties.
How to cite: Masten, M., Seelig, M., Seelig, S., Vremec, M., Wagner, T., and Winkler, G.: Climate Change Impacts on Alpine Springs: Shifts in Discharge Seasonality and Water Availability, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-10919, https://doi.org/10.5194/egusphere-egu26-10919, 2026.
Accurate flood forecasting in complex river basins depends on the effective use of high-resolution hydro-meteorological information, physically based hydrological models, and appropriate calibration procedures. This work describes the development of an event-based flood modelling framework for the Arno River basin (Italy), designed to enhance the simulation of flood hydrographs and support operational flood forecasting activities.
Spatially distributed rainfall data were obtained from raster-based precipitation products and transformed into event-specific time series suitable for use within the distributed hydrological model MOBIDIC. Observed discharge records from several gauging stations were retrieved from raw monitoring archives and reorganized into event-based datasets, allowing a coherent and consistent comparison between simulated and observed hydrographs. A unified processing workflow was established to ensure proper temporal synchronization among rainfall inputs, model outputs, and discharge observations.
The proposed framework was tested on major flood events that occurred in November 2023. Model performance was assessed using standard evaluation metrics, including the Nash–Sutcliffe Efficiency (NSE), Root Mean Square Error (RMSE), correlation measures, and time-lag analysis. Results from the initial simulations show that the model is able to capture flood timing satisfactorily, while differences in peak discharge magnitude and recession dynamics indicate the necessity for targeted parameter calibration.
A preliminary manual sensitivity analysis was carried out to identify key soil and hydraulic parameters influencing runoff generation and channel routing processes. Building on these results, an automated calibration approach based on PEST++ is currently being developed to systematically optimize the most sensitive parameters and improve model performance across multiple flood events.
Overall, the presented framework offers a reproducible and scalable methodology for event-based flood modelling and calibration in complex catchments. It provides a solid basis for multi-event analyses, automated calibration, and the future incorporation of data assimilation and artificial intelligence techniques into operational flood forecasting systems.
How to cite: Abbasi, H. K. J., Castelli, F., and Masi, M.: Event-Based Calibration of a Physically-Based Hydrological Model for Flood Simulation in the Arno River Basin Tuscany Region, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-14773, https://doi.org/10.5194/egusphere-egu26-14773, 2026.
Quantifying the impacts of adaptation measures against climate change is essential for water management authorities, farmers, and climate change research teams. There are many approaches to adaptation, ranging from the overall organisation of farming methods and field layouts to specific interventions targeted at particular problems. This study focuses on addressing the scarcity of water resources under pressure from human needs, particularly for agriculture. This is done through simulating hydrological balance under future climate change scenarios (CMIP6) using hydrological model BILAN that is calibrated with respect to both runoff and observed (satellite-derived) evapotranspiration. The impact of adaptation measures on the climate is reflected in changes to evapotranspiration, which can be measured. Therefore, evapotranspiration is the second calibrating variable. Daily evapotranspiration data were derived from MODIS land surface temperature (LST) data using the DisALEXI model at a spatial resolution of 500 metres. The study was performed on catchments in the Czech Republic and Austria, ranging in size from small (~10 km²) to large (~10,000 km²), in the Danube River tributary area. The total domain was approximately 45 000 km². Results show that multi-objective calibration has no significant negative effect on model performance in runoff and evapotranspiration generation, with NSE values of around 0.65 being calculated for runoff and evapotranspiration, respectively. A set of hydrological balance scenarios was developed and analysed from two complementary perspectives. The scenarios determine information on potential water shortages within a given catchment, quantifying the need for compensation through adaptation measures. And the assessment of the ability of the model to simulate compensation of runoff shortages within other hydrological balance components by calibrating the model using the historical observed meteorological data and scenario runoff data.
Acknowledgement: This work originated in the Centrum Voda project, funded by the Technology Agency of the Czech Republic (project no. SS02030027)
How to cite: Píša, K., Pavlík, P., Vizina, A., Hanel, M., and Ghisi, T.: Assessment of water shortages in catchment areas under climate change: a conceptual modelling approach to quantifying the need for adaptation measures, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-18681, https://doi.org/10.5194/egusphere-egu26-18681, 2026.
Decadal hydrological prediction underpins strategic decisions in water security, infrastructure investment, and disaster risk reduction by extending actionable guidance beyond seasonal horizons. Large-scale climate indices often exhibit longer predictability than regional precipitation itself and leveraging them into multiyear hydrological prediction can be more effective than relying directly on raw decadal forecasts. Still, decadal predictability is constrained by two barriers: (1) scarce training samples in decadal climate prediction project (DCPP) and (2) spectral mismatch between climate predictors and hydrological responses. Although machine learning (ML) models have high capacity, independent grid-lead training routinely overfits DCPP forecasts and therefore fails to outperform regression model baselines. We overcome these limits by coupling spectral alignment using wavelet prediction system (WASP) with a spatiotemporal merging architecture. WASP decomposes relevant climate predictors into multi-scale components and learns frequency-targeted weights to align predictor spectra with local hydrological responses. Spatiotemporal merging then pools information across space and leads, expanding the effective sample size, stabilizing complex learners, and promoting spatiotemporally coherent outlooks.
Applied to Australian drought forecasting, the framework systematically shows an increasing prediction skill in 87% of grids with a mean gain of 0.16 in correlation relative to the regression model. Event-based diagnostics show more faithful results of extreme events, including the 2002 Millennium Drought and wet spells around 2001. This method also skilfully forecasts the prolonged 2018–2020 Australian drought.
Our results elucidate the critical dependency of decadal drought prediction skill on the interplay between model complexity and predictor quality: Spatiotemporal pooling stabilizes training in complex models and improves generalization instead of overfitting when trained independently. Crucially, we identify a predictability horizon beyond 36 months where skill peaks and the advantage of WASP over raw predictors vanishes, indicating that decadal forecast quality is limited by the performance of the underlying dynamical climate models rather than by post-processing techniques. These advances provide practical value for agencies such as WaterNSW in Australia, offering scientific guidance for reservoir operation, integrated water resources planning and climate resilient adaptation strategies for national communities and ecosystems.
How to cite: Zhang, L., Thapa, S., Sharma, A., and Jiang, Z.: Spectrum Transformation Enhanced Spatiotemporal Learning for Decadal Hydrological Forecasts, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-16447, https://doi.org/10.5194/egusphere-egu26-16447, 2026.
Global Climate Models (GCMs) are essential for simulating past and future climates but suffer from systematic biases and coarse resolution, limiting direct applications. Bias correction (BC) and downscaling, using dynamical or statistical methods, address these issues. Quantile mapping (QM)-based BC is widely used, yet it distorts dependencies, prompting multivariate approaches whose assumptions remain unclear and results inconsistent. This study evaluates four BC techniques, including one univariate (QM) and three multivariate (dOTC, R2D2, MBCn), in correcting univariate, multivariate, and temporal features of daily precipitation and temperature over India during Indian Summer Monsoon (ISM). For univariate metrics, dOTC effectively corrected temperature mean, variance, and extremes, while QM and dOTC best addressed precipitation variance. Further, R2D2 was most effective for mean correction, and MBCn for dry days and P90. Among multivariate methods, R2D2 best preserved inter-variable dependencies, whereas MBCn better captured temporal features, especially precipitation autocorrelation. Additionally, the study evaluates the effectiveness of BC techniques to preserve intervariable dependence, focusing on the Pacific Walker circulation constructed using causal network, crucial for capturing complex climate signals. None of the techniques, however, reproduced the observed network across all GCMs. The overall performance of BC methods was evaluated by averaging ranks across categories since no single approach consistently excelled across all metrics. Among the techniques, dOTC showed the best overall performance, while R2D2 achieved the highest ranks in multivariate evaluations. The findings offer practical insights and highlight challenges in selecting appropriate BC methods for climate applications.
How to cite: Sharma, S., Singh Raghuvanshi, A., and Agarwal, A.: Evaluating the Performance of Uni- and Multivariate Bias Correction Techniques: Challenges in Preserving Temporal and Dependence Structures, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-18964, https://doi.org/10.5194/egusphere-egu26-18964, 2026.
Reservoirs play a vital role in water resources management by supporting irrigation, hydropower generation, urban water supply, flood mitigation, drought preparedness, and food security, particularly in monsoon-dominated regions. However, accurate forecasting of reservoir storage remains challenging due to the combined influences of climate variability and anthropogenic regulation, which limit the reliability of traditional hydrological and statistical models. In this study, we systematically evaluate four deep learning approaches Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Convolutional LSTM (CNN–LSTM), and the Temporal Fusion Transformer (TFT) for reservoir storage prediction across 79 major reservoirs in India. The models are trained using multivariate inputs that comprise historical storage, precipitation, and temperature data spanning the years 2000–2023. Our results demonstrate that the TFT consistently outperforms the recurrent and convolutional baselines, achieving testing coefficient of determination (R²) values exceeding 0.95 across most reservoirs and reducing prediction errors by approximately 20–30% relative to LSTM- and GRU-based models. Building on this superior performance, we conduct sub-seasonal to seasonal forecasts with lead times of up to three months. The TFT exhibits strong drought detection capability, with Probability of Detection values exceeding 0.8 for 1–3-month lead times. Furthermore, purely data-driven TFT forecasts outperform simulations that incorporate external precipitation and temperature forecasts from the Climate Forecast System Version 2 (CFSv2), highlighting the robustness of the learned temporal representations. Overall, this study demonstrates the potential of transformer-based deep learning models to enhance reservoir storage forecasting and early warning capabilities, offering a promising pathway for improving adaptive reservoir operations and water resources management under hydroclimatic variability.
How to cite: Parida, D., Aadhar, S., Tiwari, A. D., and Shukla, A.: Sub-Seasonal to Seasonal Reservoir Storage Forecasting Using an Attention-Based Temporal Fusion Transformer, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-19119, https://doi.org/10.5194/egusphere-egu26-19119, 2026.
How to cite: Martins Bueno, L., Luceiro Santana, E., Beskow, S., Leitzke Caldeira, T., Araujo Rodrigues, A., Galski Bonczynski, R., Leal Teixeira, D., Adolfo Karow Weber, G., and Ribas Alves, A.: River channel geometry representation from ADCP bathymetry complemented by UAV-based LiDAR for flood modelling in a data-scarce river watershed, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-19226, https://doi.org/10.5194/egusphere-egu26-19226, 2026.
Real-time flood forecasting and early warning systems (RTFF–EWS) have become central to contemporary flood risk management, particularly as climate change intensifies the frequency, magnitude, and spatial complexity of flood events. Recent advances demonstrate a clear shift from conventional physics-based forecasting towards integrated digital ecosystems that combine multi-source data acquisition, advanced analytics, artificial intelligence, and immersive decision-support interfaces. This study synthesises state-of-the-art tools and emerging research directions shaping next-generation RTFF–EWS.
On the data side, dense Internet of Things (IoT) sensor networks, remote sensing (radar, LiDAR, satellites, drones), and crowd-sourced information now enable near–real-time monitoring of hydrological and hydraulic states across urban and catchment scales. These heterogeneous data streams are increasingly fused through machine learning (ML) and deep learning (DL) frameworks, including recurrent neural networks, long short-term memory models, and hybrid physics-informed approaches, to enhance forecast lead time, accuracy, and robustness under data scarcity and uncertainty. Natural language processing (NLP) and large language–based pipelines (LLP) further extend RTFF capabilities by extracting actionable intelligence from unstructured data such as social media, emergency reports, and textual observations, improving situational awareness during rapidly evolving flood events.
Beyond forecasting, digital visualisation technologies are redefining how flood information is communicated and operationalised. Virtual reality (VR), augmented reality (AR), mixed reality (MR), and digital twins (DT) provide immersive and interactive representations of flood dynamics, impacts, and response options. These tools enable scenario testing, stakeholder engagement, and decision rehearsal across the full flood risk management cycle, from preparedness and response to recovery. Digital twins, in particular, are emerging as integrative platforms that couple real-time sensor data, predictive models, and visual interfaces into living representations of urban water systems. Despite these advances, key challenges remain, including data reliability, computational demands, interoperability across platforms, and the translation of complex model outputs into inclusive, actionable warnings for diverse stakeholders. This study also shows plausible deployment pathways for contemporary digital technologies, particularly NLP/LLP-enabled information extraction and AI-driven multi-agent frameworks, and demonstrates how their integration with RTFF–EWS can support adaptive, interpretable, and decision-centred flood risk management under real-world operational constraints.
How to cite: Behzadian, K., Piadeh, F., and Razavi, S.: Advances in Real-Time Flood Forecasting and Early Warning Systems: Integrating Artificial Intelligence, Digital Twins, and Immersive Visualisation Technologies, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-19746, https://doi.org/10.5194/egusphere-egu26-19746, 2026.
Over the past thirty years, floods have been the most frequent natural disaster globally, affecting billions of people and causing trillions of dollars in losses. Fast, accurate, high-resolution forward and inverse flood modeling is urgently required to mitigate the socioeconomic effects. Although machine learning offers fast flood simulations, high-resolution spatiotemporal modeling is fundamentally constrained by data scarcity and physical inconsistencies. Here we present a generative physical distillation neural network (GPDNN) that distills physical laws into neural networks via multi-path parallel generation, and we theoretically prove that it can approximate universal flood dynamic systems without observations. GPDNN is the first unified flood modeling network that supports forward forecasting (i.e., seen event extrapolation and unseen event generalization) and inverse parameter estimation for common flood types. Specifically, GPDNN is the first observation-free machine learning method that provides near-instantaneous and globally physically consistent flood dynamics forward forecast up to 24 h ahead at high spatiotemporal resolution. Simulations of dam-break floods, riverine floods, and urban inundation are found to incur errors that are one to two orders of magnitude lower than existing methods based on dense observations. Furthermore, GPDNN has high accuracy and robustness in inverse analyses of hydraulic parameters under both spatially homogeneous and heterogeneous conditions with sparse or even noisy observations. Our work has knock-on benefits for flood risk assessment, forecasting, and management.
How to cite: Cao, X., Yao, Y., and Qin, H.: Unified flood modeling with generative physical distillation neural network under data scarcity, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-21913, https://doi.org/10.5194/egusphere-egu26-21913, 2026.
