Flooding is increasingly exacerbated by cascading and compounding risks, necessitating a systematic understanding of natural hazard interrelationships and the influence of anthropogenic processes on these dynamics. Integrating multi-hazard scenarios into flood hazard, impact, and risk models provides a more comprehensive understanding of flood risks by considering interactions such as earthquake-triggered landslides and rainfall-induced flooding, alongside the impacts of urbanisation, land-use change, and climate change. This approach enhances decision-making frameworks, enabling more effective preparation and response. Here, we first illustrate hazard interaction matrices as a methodology to systematically identify potential triggers and secondary hazards by synthesising evidence from local and global literature. We then highlight multi-hazard scenarios through examples from Nairobi (Kenya), İstanbul (Türkiye), the Kathmandu Valley (Nepal), and the Philippines, where frequent and high-magnitude hazards underscore the critical importance of preparedness and mitigation. Effective flood risk management also requires interdisciplinary collaboration among researchers, policymakers, and practitioners, leveraging disaster science methodologies to assess hazard relationships, enhance response strategies, and build resilient communities.

How to cite: Malamud, B. D.: Integrating Multi-Hazard Scenarios for Enhanced Flood Risk Management, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-21109, https://doi.org/10.5194/egusphere-egu25-21109, 2025.

Efficient evacuation during storm floods remains a critical challenge for coastal cities, primarily due to its dynamic and complex nature involving flood progression, human behavioral uncertainties, and emergency resource constraints.  Current evacuation models inadequately capture these multifaceted dynamics, limiting effective emergency planning. This study introduces an Agent-based Dynamic Coastal Flood Evacuation (DCFE) model that comprehensively simulates the interactions among flood dynamics, human behavioral responses, GIS-based transportation networks, and shelter systems.  Using Shanghai as a case study, we evaluate city-scale evacuations during a 1,000-year return period storm event. Our analysis shows that issuing warnings 12 hours before flood peak reduces casualty rates by over 25% compared to scenarios without early warning, while optimized decision-making can double evacuation efficiency. The results further reveal critical spatial disparities in evacuation performance due to inequitable shelter distribution. This integrated approach provides practical guidelines for enhancing evacuation strategies in coastal megacities worldwide.

How to cite: Yang, Y.: Dynamic Flood Evacuation Modelling for Coastal Cities: A Case Study of Shanghai, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1829, https://doi.org/10.5194/egusphere-egu25-1829, 2025.

Fast flood modelling for pluvial flood management in Innsbruck, Austria

The increasing frequency and intensity of precipitation events leads to urban flooding, often causing significant damage to infrastructure and property. This phenomenon, known as pluvial flooding, arises when heavy precipitation exceeds the capacity of urban drainage systems, leading to surface water accumulation. Climate change is expected to exacerbate this issue, emphasizing the urgent need for efficient predictive models to mitigate the associated risks and impacts.

Traditional hydrodynamic models, such as coupled 1D-2D simulations, offer highly detailed flood assessments by simulating both surface runoff and sewer network interactions. However, these models are computationally demanding, requiring significant resources and time, making them unsuitable for real-time flood forecasting and decision-making during extreme weather events.

To address these limitations, fast flood models like the dynamic CA-ffé model, based on Jamali et al. (2019) and further developed by Gholami Korzani and Deletic (2023), provide a practical alternative. These models efficiently integrate surface flow and sewer network dynamics, enabling accurate flood forecasting at a much lower computational cost. Previously validated in smaller Australian catchments, the dynamic CA-ffé model has demonstrated its ability to provide timely and accurate urban flood simulations, significantly improving flood forecasting and risk management.

To address flooding challenges in Innsbruck, a larger, mountainous catchment area (~50 km²), the dynamic CA-ffé model was adapted based on the model approach of Gholami Korzani and Deletic (2023). This model approach combines a cellular automata-based 2D simulation with a 1D sewer network model using SWMM (Stormwater Management Model). By synchronizing data exchanges between surface runoff and sewer discharge at regular intervals, the model achieves faster and more accurate flood predictions, enabling high-resolution urban flood forecasting.

Adapting the model to Innsbruck required adjustments to account for the city's complex mountainous terrain and boundary conditions. Additional case-specific modifications were implemented to ensure compatibility with the larger and more challenging catchment area. The model was tested using historical flood events and validated against fire brigade records and photo documentations, as no prior citywide flood models were available for comparison.

The model's fast computation times allow the simulation of different flood scenarios, including assessments of the effects of climate change. These simulations will help to identify flood risks and inform heavy rainfall management strategies. Initial results confirm the model's ability to simulate urban-scale flooding, while highlighting challenges in adapting the approach to larger and more topographically complex study areas, such as land-use based runoff coefficients and the use of multiple rain gauges for precipitation data.

Funding:

BlueGreenCities (project No. KR21KB0K00001), funded by the Austrian Climate and Energy Fund from October 2022 until September 2025

Early Stage Funding (project: FFMFF) funded by the Vice-Rectorate for Research of the University of Innsbruck from November 2023 until October 2024.

References:

Gholami Korzani, M., Deletic, A., 2023. Dynamic CA-ffé: a hybrid 1D/2D fast flood evaluation model for urban floods. Sydney.

Jamali, B., Bach, P.M., Cunningham, L., Deletic, A., 2019. A Cellular Automata Fast Flood Evaluation (CA-ffé) Model. Water Resources Research 55, 4936–4953. https://doi.org/10.1029/2018WR023679

How to cite: Hauser, M., Rauch, N., Gholami Korzani, M., Deletic, A., and Kleidorfer, M.: Fast flood modelling for pluvial flood management in Innsbruck, Austria, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6551, https://doi.org/10.5194/egusphere-egu25-6551, 2025.

This study investigates flood-prone areas in the Loukous Basin, Northern Morocco, utilizing machine learning and remote sensing techniques to analyze and map zones at risk. Renowned for its agricultural significance, the region features low-lying, flat terrain, an active river system, and oceanic influences, all of which exacerbate flooding risks. Additionally, climate change poses increasing challenges, intensifying extreme weather events and altering precipitation patterns that further threaten this vulnerable region. The research aims to enhance flood prevention strategies and mitigate economic and human losses by identifying and prioritizing highly vulnerable zones. Results consistently highlight significant flood susceptibility along the Loukous River and its adjacent plains, areas characterized by lowland topography, high drainage density, proximity to canals, and intensive agricultural activity. While spatial variations exist among the models, a strong consensus emerges regarding zones of low and very high vulnerability, emphasizing the need for tailored interventions. These findings provide critical insights for integrating agricultural development planning with flood risk management in the Basin of Loukous. They underscore the importance of adaptive strategies that consider the compounded effects of climate change, such as improved land-use practices, enhanced drainage systems, and sustainable water management. This study establishes a robust scientific foundation for implementing targeted measures to reduce flood impacts, safeguard livelihoods, and build resilience in this economically vital and environmentally sensitive region, with broader implications for similar flood-prone areas worldwide.

How to cite: Mekkaoui, O., Morarech, M., En-negady, I., Bouramtane, T., and Akka, H.: Evaluating Flood Vulnerability in the Loukous Basin, Northern Morocco: Integrating Machine Learning, Remote Sensing, and Climate Change Impacts, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12481, https://doi.org/10.5194/egusphere-egu25-12481, 2025.

Bridges worldwide face significant risks from flood-related hazards, with hydraulic actions being the primary cause of structural damage, service disruption, and catastrophic failure. Bridge owners and regulatory agencies must assess multiple hydraulic risks, including scour, uplift forces, drag effects, debris impact, and deck displacement, all of which can compromise the load-carrying capacity of a bridge. The assessment of hydraulic actions on highway bridges in many parts of the world still relies on simplified or qualitative methods, whereby hydraulic models are yet to be embedded within guidance.

In the United Kingdom, the CS469 standard governs the assessment of hydraulic actions on highway bridges, providing guidance for risk evaluation and management strategies. The CS469 methodology calculates hydraulic flow characteristics at critical cross-sections within channels and bridge crossings utilising Bernoulli theorem and specific energy. While computationally efficient, this simplified approach relies on non-physical approximations. This fundamental limitation introduces substantial uncertainty into risk and vulnerability assessments, potentially compromising the reliability of management decisions.

This study presents an alternative approach utilising 2D HEC-RAS hydraulic models with bridges modelled as 1D elements within flow areas. The proposed methodology crucially maintains compatibility with existing data requirements from CS469 while adhering to open-source principles, requiring only publicly available data or information from existing assessments. This approach ensures cost-effectiveness and accessibility for bridge management teams while providing significantly improved accuracy.

Comparative analyses between the 2D HEC-RAS model and traditional CS469 calculations for six case study bridges revealed substantial differences in hydraulic response. The 2D model showed water depths up to 138% higher and flow velocities 64% lower than CS469 estimates. These differences significantly impact scour risk assessments, with HEC-RAS models typically predicting scour depths up to 2.3m lower (averaging 1.5m reduction) compared to simplified equations, resulting in more realistic risk classifications.

While hydraulic vulnerability assessments showed limited variation, the CS469 approach only considers threshold values without quantifying effects. Our findings demonstrate that numerical hydraulic simulations provide more accurate risk estimations with comparable resource requirements, suggesting that future revisions of risk assessment guidelines should prioritise this methodology. This advancement would enhance the accuracy and reliability of bridge risk assessments, ultimately improving infrastructure resilience and safety management strategies.

How to cite: Panici, D., Kripakaran, P., and Brazier, R.: A comparative analysis for assessment of hydrodynamic actions at bridges of 2D hydraulic models and traditional highway standards, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13884, https://doi.org/10.5194/egusphere-egu25-13884, 2025.

The risk associated with river floods has escalated significantly due to the increasing frequency and intensity of extreme precipitation events, compounded by the complex interplay of flood-generating mechanisms under a changing climate. Accurately estimating flood risks poses a formidable challenge, as these mechanisms often interact and exhibit varying influences across spatial and temporal scales. An in-depth understanding of flood-generating processes is critical for improving hydrological modeling, flood frequency analysis, and risk management strategies in diverse climatic regions. Despite India being one of the most flood-prone countries globally, a systematic classification of hydrometeorological flood-generating processes remains largely absent. Understanding the role of catchment and climate attributes in flood generation is crucial for advancing our ability to predict and manage flood risks. This study proposes a robust framework to classify flood-generating processes into three primary categories: long rainfall floods, short rainfall floods, and excess rain floods. The analysis focuses on major river basins across India, offering insights into region-specific flood dynamics. To achieve this, we leverage the CAMELS-IND dataset, a comprehensive repository of hydrological and meteorological data, covering 471 catchments across India from 1980 to 2020. Using a peaks-over-threshold (POT) approach, we identify significant flood events and assess their characteristics. We then employ a Light Gradient Boosting Machine (LightGBM) model, an advanced machine learning algorithm known for its efficiency and accuracy, to evaluate the contribution of various climate and catchment attributes in triggering these floods. To enhance interpretability, we integrate Shapley additive explanations (SHAP), which provide a localized and global understanding of the model's predictions, highlighting the relative importance of each attribute. Our findings underscore the dominant role of climate attributes, such as precipitation intensity, antecedent soil moisture, and temperature, in determining the spatial distribution of flood-generating processes across diverse climatic zones. Catchment attributes, including soil type, slope, and land use, also contribute but to a lesser extent. These insights have significant implications for flood risk management, particularly in ungauged catchments, and can enhance the accuracy of hydrological and hydrodynamic models under changing climatic conditions.

How to cite: Tripathi, V., Deopa, R., and Mohanty, M.: Unraveling Hydrometeorological Drivers of Floods: A Data-Driven Analysis Across India's River Catchments, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15096, https://doi.org/10.5194/egusphere-egu25-15096, 2025.

Floods are among the most impactful natural phenomena affecting society. The associated risks are influenced by factors such as urban expansion into high-risk areas, modifications to rivers and watersheds (e.g., artificial channels, flow redirection, and structures like dams and dikes), and the effects of climate change, including more frequent and severe events. In recent years, the application of Artificial Intelligence (AI) in climate and weather risk assessment has gained increased attention due to its ability to handle numerical and categorical variables, uncover nonlinear relationships, and achieve high performance.
In this study, we introduce FloodGuard, an AI-powered tool for flood vulnerability and risk mapping. FloodGuard employs the concept of regionalization in ungauged basins and leverages a flood inventory derived from satellite imagery (e.g., Copernicus Emergency Mapping Service) over extensive areas (e.g., national or continental scales). The methodology selects the most relevant historical flood events and transfers this information to train a Random Forest machine learning model for estimating flood extent and producing a flood exposure map. Inputs to the model include the Geomorphic Flood Index (GFI), the Elevation, the Horizontal Distance to the Nearest River, Precipitation, the NDVI, and information on Land Use and Lithology. Flood prediction map is evaluated using maps generated from hydrological and hydraulic models. To assess vulnerability, we apply a geomorphic approach proposed by Manfreda and Samela (2019). This approach estimates flood depth, which is useful for estimating fast vulnerability levels. Finally flood risk is estimates with a GIS-based model.
The primary objective of this study is to provide a preliminary simple tool to estimate a flood risk and provide risk maps. At the same time, this study evaluates evaluate the transferability of machine learning models from regions with flood records to ungauged areas using satellite observations. Limitations include uncertainties inherent to machine learning models and the lack of association with specific return periods. Preliminary results across Italy demonstrate that the Random Forest model achieves high performance (AUC > 0.9) and exhibits robust generalization capabilities (e.g., combined error (rfp + (1-rtp)) of 0.58).

Keywords: Artificial Intelligence, Machine Learning, Flood risk, Flood vulnerability, GFI. 

This study was carried out within the RETURN Extended Partnership and received funding from the European Union Next-Generation EU (National Recovery and Resilience Plan -NRRP, Mission 4, Component 2, Investment 1.3 -D. D. 1243 2/8/2022, PE0000005). 

How to cite: Saavedra Navarro, J., Zhuang, R., Albertini, C., Samela, C., and Manfreda, S.: FloodGuard: An AI-Powered Tool for Flood Risk and Vulnerability Mapping in Ungauged Basins., EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19363, https://doi.org/10.5194/egusphere-egu25-19363, 2025.

How to cite: Pellerin, J., Hayes, S., Chastko, K., Ballard, M., Bourke, R., and Van de Valk, J.: Canada-wide Modelling – Analysis of Model Accuracy to Drive Appropriate Use and Risk Reduction Program Development, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7327, https://doi.org/10.5194/egusphere-egu25-7327, 2025.

As climate change drives sea level rise and raises critical questions about the design and efficacy of flood defences, there is an increasing demand for high-resolution coastal flood data. As a commercial provider of flood risk data to the insurance and banking sector, we have users that emphasise accuracy. For instance, these users seek not only to determine if floodwaters might reach a property but also whether they will exceed the height of the doorstep. While advancements in flood modelling continue to meet these growing needs, a persistent challenge remains: how can we balance the need for high accuracy with the practical constraints of production costs and the need for timely data delivery?

Focussing on coastal flood mapping examples in the UK, our presentation will compare outputs from two ends of the modelling spectrum: complex full hydrodynamic modelling and a simplified projection approach. Simplified approaches often face criticism for their limitations, but we will argue that they can provide valuable – as well as timely and efficient – outputs, when applied appropriately and with a clear understanding of their constraints.

This work aims to explore the importance of balancing advanced and simplified techniques, offering insights into the factors that most significantly influence flood modelling outputs. For example, we examine whether the choice of input data (e.g., the terrain data, the extreme sea levels) has a greater impact on results than the modelling approach itself (hydrodynamic compared to projection modelling techniques). By highlighting the trade-offs and opportunities, we aim to contribute to a more nuanced understanding of how to optimise coastal flood risk data production to meet user needs.

How to cite: Raven, E., Charge, J., Liam, H., James, S., and Rhiannon, W.: Balancing complexity and efficiency in coastal flood risk modelling, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8901, https://doi.org/10.5194/egusphere-egu25-8901, 2025.

Calibration and validation are necessary steps in the field application of physically-based models for floods and landslides. These processes validate the model assumptions and adjust the parameterization to align with historical observations. However, calibration remains a time-consuming task due to the lack of globally available datasets directly linked to the calibration schemes of physically-based models.

As part of the FastFlood.org and FastSlide.org rapid modeling platforms, we have developed a built-in calibration scheme that leverages global observational datasets and data-driven approaches. This approach links the results of global calibration datasets to a standardized calibration scheme for flood and landslide models, enabling automatic calibration globally based on these datasets.

In this paper, we outline the setup, global dataset processing workflows, and the initial outcomes of this research. The global datasets are derived from multiple observation types. For example, discharge data for return period events in rivers is based on GloFAS reanalysis data, bias-corrected using an extreme-value analysis performed globally on GRDC discharge station time series. Flood extents from the Global Surface Water Explorer are included but corrected for the underrepresentation of flash flood events in the observational data. For landslide processes, landslide inventory collections and data-driven global landslide susceptibility maps are utilized to provide built-in calibration functionality.

To streamline the calibration process, specific automated schemes have been developed to preprocess these global datasets, enabling direct comparison with model outputs without user intervention. Furthermore, we explore some initial results of this calibration scheme, comparing its accuracy and relative performance against other calibration methods.

How to cite: van den Bout, B., Kolaparambil, F., Katarya, S., van Westen, C., and Meijvogel, D.: Global automated calibration procedures for the FastFlood and FastSlide rapid hazard models, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9325, https://doi.org/10.5194/egusphere-egu25-9325, 2025.

Floods are considered one of the most damaging natural disasters known to humankind, and their severity has increased significantly due to the impacts of climate change and global warming in recent decades. Floods have become more unpredictable and erratic due to the influence of extreme hydroclimatic events. Therefore, obtaining near-real-time and accurate flood inundation maps of such events is essential for effective flood emergency response. These can be achieved easily by leveraging multi-source satellite imagery and remotely sensed data. The freely accessible satellite imagery and remotely sensed products can provide essential information that can significantly reduce the resources needed to create flood inundation maps and improve the accuracy of mapping and monitoring systems. This study integrated high-resolution satellite imagery and multiple remote sensing data to improve the flood inundation mapping technique in a data-scare South Asian watershed. The study considered the 2008 Bihar flood caused by the embankment breach of the Koshi River as a case study. This study used Landsat satellite's surface reflectance data to map the flood inundation using the commonly used water index known as the Modified Normalized Difference Water Index (MNDWI). MNDWI is a commonly used water classification technique to detect open surface water features using surface reflectance data sensed by the satellite. Further, the Normalized Difference Vegetation Index (NDVI), permanent water bodies, and Height Above the Nearest Drainage (HAND) datasets are used to mask the MNDWI map (initial flood inundation map) and improve the accuracy of the inundation map. In addition, different thresholding values of the final masked MNDWI map are applied to obtain more accurate and robust flood inundation maps with fewer false positive and false negative pixels.

How to cite: Aryal, A., Sinha, R., and Lakshmi, V.: Improving the Flood Inundation Mapping Technique using Satellite Imagery and Remote Sensing Data: A Case Study of the Bihar Flood Caused by the Koshi Embankment Breach, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12576, https://doi.org/10.5194/egusphere-egu25-12576, 2025.

Floods are the most frequent disasters, affecting the largest number of people. In 2023, 164 floods were recorded worldwide, killing 7763 people, affecting 32.4 million people, and resulting in 20.4 billion USD losses (CRED 2023 Disasters in Number report). To mitigate flood risk, decision makers and civil protection authorities need reliable information on flood risk assessment, covering all disaster management stages.

In the framework of a Programming Agreement with the Prefecture of Attica, NOA/IAASARS/BEYOND, in cooperation with NTUA/ITIA, developed a holistic multiparameter methodology that was implemented in five flood-stricken river basins at high spatial resolution. The research teams collected all available Earth Observation data, spatial data and technical studies; conducted detailed field visits; and modified the DEM and land cover accordingly. Following rainfall-runoff modeling and hydraulic modeling, the flood hazard was assessed for different scenarios. Vulnerability was considered a weighted estimation of population density, population age, and building characteristics on the basis of the population-housing census at the building block level. Exposure was based on the land value. Flood risk was eventually assessed on the basis of the combination of flood hazard, vulnerability, and exposure. Moreover, critical points, which were identified from the field visits, were also crosschecked with the flood inundation maps. Finally, refuge areas and escape routes were proposed for the worst-case flood scenario. This innovative methodology was applied, among other methods, in the Mandra river basin and was validated with the results of the urban flash flood, which took place in 2017, the deadliest flood in Greece in the last 40 years. BEYOND developed a user-friendly web GIS platform in which all the collected and produced data, including flood risk maps, critical points, refuge areas and escape routes, are made available.

This flood risk assessment methodology was applied, following adaptation, in the Garyllis river basin in Cyprus, within the framework of the EXCELSIOR project, as part of the collaborative activities between ECoE and BEYOND. Data were collected from multiple sources, including satellite missions, governmental portals, in situ measurements, and historical records, at different resolutions. The collected data were calibrated via onsite visits and discussions with relevant actors, harmonized in terms of spatial and temporal resolution and used as inputs to estimate the flood hazard for different return periods. The vulnerability levels of the study area were quantified via the weighted linear combination of population density, population age, and building characteristics at the road level. The exposure levels were quantified in terms of the land value. Flood risk levels were estimated as a product of hazard, vulnerability and exposure levels. The validity of the proposed methodology was evaluated by comparing the critical points identified during the field visits with the estimates of the flood risk levels. Consequently, escape routes and refuge regions are recommended for the most extreme scenario.

This work supports relevant authorities in improving disaster resilience and in implementing the EU Floods Directive 2007/60/EC, the Sendai Framework for Disaster Risk Reduction, the UN SDGs, and the UN Early Warnings for All initiative.

How to cite: Tsouni, A., Panagiotou, C., Sigourou, S., Kountouri, J., Pagana, V., Dimitriadis, P., Iliopoulou, T., Sargentis, G.-F., Mettas, C., Evagorou, E., Ioannidis, R., Chardavellas, E., Dimitrakopoulou, D., Alexopoulos, M. J., Mamasis, N., Koutsoyiannis, D., Hadjimitsis, D., and Kontoes, C. (.: A transferable multicriteria methodology for flood risk assessment: two case studies in Greece and Cyprus, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15732, https://doi.org/10.5194/egusphere-egu25-15732, 2025.

Understanding flood hazard at scale requires consistently generated flood maps. To build flood maps at scale, automated frameworks must be used that can take input data, transform this into hydrodynamic simulations and process it into output flood maps.

This transformation from inputs into simulations can be influenced by the design of the framework and the choices made by modellers as to how the transformations are made. These choices can be influenced by quality and availability of input data, availability of computational resources and desired outputs. There may not be clear objectively better options, so subjective choices have to be made. Understanding the sensitivity of these choices is key in both making them, and rating the uncertainty of the outputs.

Here we present the outputs from some of the sensitivity testing undertaken when developing a global model framework including both their influence on flood hazard and their influence on other deciding factors (such as computational cost) made when building a global flood map. This presentation includes results from tests that were shown to develop key or interesting results, such as:

• How rivers are split into separate simulations
• Minimum size of fluvial river assessed
• Length of pluvial storm assessed

How to cite: Cooper, A. and Savage, J.: Sensitivity of global flood modelling frameworks to input parameter choices, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17337, https://doi.org/10.5194/egusphere-egu25-17337, 2025.

Flooding is a common and devastating disaster that happens recurrently in Bangladesh. Climate change, rapid urbanization, and inadequate drainage systems have exacerbated this event in recent years. In 2023, Satkania upazila in Chattogram district was severely affected by flood. Additionally, communities weren’t prepared for this flood, as they hadn’t faced any kind of flood in recent years, as reported by field observation. That severity helped to select Satkania upazila as a study area for this research. This study focuses on developing a GIS-based evacuation route plan to mitigate the adverse impacts of flooding by ensuring safe and efficient evacuation during emergencies. To identify high-risk and low-risk zones, Sentinel-1 SAR imagery was used to create an inundation map. Frequent and severe flooding criteria were used to find the high-risk zones. This area is designated as an assembly point where people can gather before moving to safer places. Low-risk zones were identified as suitable locations for emergency shelters to ensure people's safety during disaster events. ArcGIS network analyst tools are used to perform closest facility analysis and service area analysis. For this analysis, road network, elevation data, and building density were integrated with ArcGIS network analysis. Closest facility analysis helped to optimize the evacuation routes based on minimum travel time and shortest distance. To determine zones to locate emergency shelters, a service area analysis was performed to improve accessibility for vulnerable populations. The inundated map showed that 11.78% area was inundated during the 2023 flood. According to the network analysis, the assembly point that is closest to the emergency shelter is more accessible in both time and distance cases. A total of 3 out of 14 suggested emergency shelters were within 30 min, 7 within 60 min, 3 within 90 min, and 1 within 90 min walk from the assembly point. In terms of distance from the assembly site, there are 3 emergency shelters within 1000 m, 7 within 2000m, 3 within 3000 m, and 1 within 4000 m distance. The result showed the effectiveness of GIS-based route planning to minimize flood causalities and enhance disaster preparedness. This study provides actionable insights to design targeted interventions and response strategies for local governments and disaster management authorities.

How to cite: Ahmed, M., Anny, N. S., Khadem, S., and Baten, T. A.: A GIS-Based Evacuation Route Planning in Flood Inundated Area of Satkania, Chattogram, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20503, https://doi.org/10.5194/egusphere-egu25-20503, 2025.

Floods are globally catastrophic, with profound impacts on India’s environment, agriculture, and infrastructure. Between 1953 and 2010, floods annually affected 7.2 million hectares and 3.2 million people. Odisha, particularly the Mahanadi basin, ranks seventh in flood vulnerability, with over 90\% of its annual rainfall occurring during the monsoon. Synoptic systems from the Bay of Bengal exacerbate rainfall, causing frequent and severe floods. Despite existing forecasting systems, downstream regions remain inadequately protected, highlighting the need for more accurate predictive mechanisms.

Recent advancements in Machine Learning (ML) and Deep Learning (DL) offer new opportunities for flood forecasting. Long Short-Term Memory (LSTM) models excel at capturing intricate temporal patterns in rainfall and streamflow data, outperforming conventional methods. Innovations like hybrid LSTMs and Spatio-Temporal Attention (STA) mechanisms enhance their performance, while novel architectures such as Temporal Convolutional Networks (TCNs) and Bi-LSTMs improve long-term predictions.

This study introduces a Kolmogorov-Arnold Network (KAN)-enhanced LSTM model for five-day-ahead flood prediction in the Mahanadi basin. KAN leverages learnable activation functions and spline representations, improving accuracy, interpretability, and computational efficiency. Evaluated against traditional LSTMs using metrics such as Nash-Sutcliffe Efficiency (NSE), time-to-peak prediction, and convergence speed, the KAN-enhanced model consistently outperformed standard LSTMs. It demonstrated a 12\% improvement in NSE, superior peak timing, and 20\% faster convergence, offering crucial advancements for early warning systems.

By integrating KAN's ability to model non-linear relationships with LSTM’s strength in sequential data analysis, this framework addresses complex hydrological dynamics, providing reliable flood forecasts. These findings underscore the potential of KAN-enhanced architectures to revolutionize flood prediction, offering scalable and interpretable solutions for flood-prone regions.

How to cite: Sarkar, S., Dey, A., Chatterjee, C., and Mitra, P.: KAN-Enhanced LSTM for Accurate and Scalable Flood Forecasting: A Case Study of the Mahanadi Basin, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8031, https://doi.org/10.5194/egusphere-egu25-8031, 2025.

Hydrodynamic models solving various forms of the shallow water equations are commonly used to assess flow depths and velocities during floods. Recent advances have extended these models to incorporate elements of the catchment such as sewer systems, culverts, bridges, and weirs. However, many flood models still represent buildings as raised elevations or with increased roughness coefficients, blocking water from entering building areas. This can lead to an overestimation of flood depths and extents, as these storage areas are not considered. Until now, the cumulative effects of flow into buildings on city-scale inundation have not been simulated using physically based models. This study develops and implements a building indoor-inundation model, comparing it with simulations that neglect this effect (storage and retention inside buildings) to quantify the differences in flood depth and extent.

A hydrodynamic model was coupled with the indoor-inundation model to estimate flow into and out of buildings (through windows, doors, etc.). This model incorporates building geometry, including walls, doors, and stairwells, to determine flow throughout the building. It also accounts for the transition between pressurized and non-pressurized flow, allowing water to move from lower to upper floors. The building indoor-inundation model was coupled with the 2-D diffusive wave model P-DWave, which simulates surface flood inundation outside the buildings. Water is exchanged between the two models at run-time in a bi-directional manner, with water flowing both into and out of the buildings.

Flood simulations were conducted for the city of Baiersdorf in northern Bavaria, which experienced flash floods in 2007 caused by heavy rainfall, resulting in over 86 million euros in damages. The event was recreated using the dual-drainage model PD-Wave/SWMM to simulate interactions between overland flow and the sewer system. Three test cases were modeled: buildings represented with increased roughness coefficients, buildings raised above the floodplain, and buildings modeled using the presented hydrodynamic approach. The results show that modeling the flow into and out of buildings has a moderate to significant impact on both flood depths and extents, highlighting the importance of including building inundation in urban flood modeling.

How to cite: Johnson, T. G. and Leandro, J.: Assessing the Impact of Building Indoor-Inundation on Flood Depth and Extent at City Scale: A Novel 2-D Coupled Hydrodynamic and Building Inundation Model for Urban Flood Modeling, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10398, https://doi.org/10.5194/egusphere-egu25-10398, 2025.

Traditional flood modelling approaches, which rely on deterministic methods, often fail to account for the inherent uncertainties of flood events, such as discharge estimates or flood defence parameters. Probabilistic flood modelling addresses this gap by quantifying the likelihood of various scenarios, based on the probability of occurrence of its inputs. However, the large number of required numerical simulations makes this framework computationally expensive. In this study, we explore the use of a multi-scale graph neural networks inspired by finite volume methods to accelerate probabilistic flood simulations. This approach is applied to several dike rings in the Netherlands - regions enclosed by levees - considering uncertainties in dike breach locations and inflow discharges. To improve the reliability of the model, we select among the output simulations only the ones that approximately preserve the total flood volumes over time, as calculated from the inflow boundary conditions. Our model generates thousands of flood scenarios with orders-of-magnitude speedups compared to traditional methods. The resulting output maps provide the expected frequencies of inundation extents for specific water depths, offering a robust tool for efficient and comprehensive flood risk assessment.

How to cite: Bentivoglio, R., Isufi, E., Jonkman, S. N., and Taormina, R.: Probabilistic flood modelling with multi-scale hydraulic-based graph neural networks, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10951, https://doi.org/10.5194/egusphere-egu25-10951, 2025.