Real-Time Flood Risk Assessment Using Coupled Agent-Based and AI-Driven Flood Forecasting Models

Estimating the geographical flood extent is a key element in impact-based flood forecasting and crucial for countries with long coastlines, and places where storm surges pose a significant risk, such as Denmark. For local flood mitigation measures and climate adaptation strategies, inundation mapping is often performed using physical models. However, in the context of flood forecasting and early warning, these are computationally demanding and therefore may not be able to provide timely forecasts and effective warnings.The Danish Meteorological Institute (DMI) has developed a real-time flood forecasting system for storm surge events in Denmark together with the company SCALGO. This system couples the HBM regional oceanic storm surge forecasting model, developed by DMI, with a rapid inundation mapping, developed by SCALGO, using a 0.4 m resolution Digital Elevation Model (DEM). All inland pixels in the DEM are connected to a coastline pixel through pre-computed hydrological flow paths. The predicted water level from the storm surge model at each coastline pixel is then instantaneously projected inland through the pre-mapped flow paths. This study evaluates the performance of the flood forecasting system on the Oct. 20-21 (2023) storm surge event, with an estimated return period of over 100 years and affecting large parts of southern Denmark (and northern Germany).This flood forecasting system creates a simple inundation mapping based on forecasted sea levels based on a high-resolution DEM modified to account for hydrological flow processes. This real-time flood mapping allows for a visualization of full five-day ocean model forecasts updated continuously at 6h intervals and has been operational for flood warning since October 2022, to supplement DMIs operational ocean forecasting system [1]. The evaluation is performed by comparing the inundation map from the flood forecasting system with media reports, photographs, and other data sources to get an overview of spatial and temporal accuracy and accuracy of the severity of the event. We see a large overlap between areas with forecasted flood risks and actual flooded areas. In some cases, the extent of the flooding differs from the area at risk due to errors in the DEM or local emergency services mitigation strategies.We conclude that the flood forecasting system is useful for identifying coastal areas at risk. While it does not account for detailed physics of flow on land, it is able to reflect the effects of, even very local, geographical variations in sea level that determine the distributions of local-scale flood risk. The current inundation mapping does not currently include the impact of waves, which resulted in larger differences between predictions and actual flooded areas, for easterly-facing locations exposed to large waves. Proposed activities to include the effect of waves will therefore improve the flood forecasting system. [1] Andrée, E., Su, J., Larsen, M. A. D., Madsen, K. S., & Drews, M. (2021). Simulating major storm surge events in a complex coastal region. Ocean Modelling, 162, 101802.

Frequent flooding in the central part of the Hindu Kush region, Nepal, results in substantial damage to lives and property, particularly in the plains. This study explores the catchment-scale flood hazard and vulnerability, generating flood risks under climate change scenario. Flood model was developed using HEC-RAS 2D rain-on-grid approach and CMIP6 projected climate data for SSP245 and SSP585 scenarios was used. Flood hazard, vulnerability and risk maps were generated delineating potential inundation scenarios for both 50-year, and 100-year return periods, under present and future climate scenarios. Hazard maps revealed a probable increase in flooding under future climate scenarios. Vulnerability assessments also demonstrated an anticipated escalation in future. Cropland dominates the land-use type in the study area, with agriculture vulnerability increasing by 22.14% (SSP245) and 29.322% (SSP585). The number of inundated buildings increases from 2967 in the existing scenario to 4601 (SSP245) and 5274 (SSP585). Similarly, road vulnerability areas are increased by 28.06% (SSP245) and 37.86% (SSP585). The impact of flooding is higher in SSP585 scenario as compared to SSP245 scenario. The catchment level outcomes have diverse applications, particularly in decision-making for flood management, integrating risk maps for effective strategies, and providing valuable insights into shape climate change adaptation policies. Keywords: Catchment scale, Climate change, Flood vulnerability, Flood risk, HEC-RAS 2D.

Floods cause substantial losses to life and property, especially in flood-prone regions like northwestern Bangladesh. Timely and precise evaluation of flood impacts is critical for effective flood management and decision-making. This research demonstrates an integrated approach utilizing machine learning and Google Earth Engine to enable real-time flood assessment. Synthetic aperture radar (SAR) data from Sentinel-1 and the Google Earth Engine platform were employed to generate near real-time flood maps of the 2020 flood in Kurigram and Lalmonirhat. An automatic thresholding technique quantified flooded areas. For land use/land cover (LULC) analysis, Sentinel-2's high resolution and machine learning models like artificial neural networks (ANN), random forests (RF) and support vector machines (SVM) were leveraged. ANN delivered the best LULC mapping with 0.94 accuracy based on metrics like accuracy, kappa, mean F1 score, mean sensitivity, mean specificity, mean positive predictive value, mean negative value, mean precision, mean recall, mean detection rate and mean balanced accuracy. Results showed over 600,000 people exposed at peak inundation in July-about 17% of the population. The machine learning-enabled LULC maps reliably identified vulnerable areas to prioritize flood management. Over half of croplands flooded in July. This research demonstrates the potential of integrating SAR, machine learning and cloud computing to empower authorities through real-time monitoring and accurate LULC mapping essential for effective flood response. The proposed comprehensive methodology can assist stakeholders in developing data-driven flood management strategies to reduce impacts.

The socioeconomic impacts of flooding are huge. Between 1980 and 2013, flood losses exceeded $1 trillion globally, and resulted in approximately 220,000 fatalities. To reduce these negative impacts of floods, effective flood risk management is required. Reducing risk globally is at the heart of two recent international agreements: the Sendai Framework for Disaster Risk Reduction and the Warsaw International Mechanism for Loss and Damage Associated with Climate Change Impacts. Prerequisites for effective risk reduction are accurate methods to assess hazard and risk, based on a thorough understanding of underlying processes. Due to the paucity of local scale hazard and risk data in many regions, several global flood hazard and flood risk models have been developed in recent years. More and more, these global models are being used in practice by an ever‐increasing range of users and practitioners. In this chapter, we provide an overview of recent advances in global flood hazard and risk modeling. We then discuss applications of the models in high‐level advocacy in disaster risk management activities, international development organizations, the reinsurance industry, and flood forecasting and early warning. The chapter concludes with several remarks on limitations in global flood risk models and the way forward for the future.

A novel framework for urban flood risk assessment: Multiple perspectives and causal analysis

Effective flood forecasting and warning systems depend on robust hydrometric networks tailored to meet the functional, geographical, and operational demands of real-time monitoring. This study outlines the ideal requirements for hydrometric data collection and transmission systems to support the National Flood Forecasting and Warning Service (NFFWS). Key considerations include spatial and temporal coverage, gauge placement in flood-prone areas, and sufficient resolution to align with hydrological models.Recommendations emphasize the need for durable, reliable infrastructure designed for extreme conditions, adherence to international standards, and regular maintenance protocols. Priority gauges—those critical for public safety, hydrological model calibration, and flood risk management—should be safeguarded, with enhancements in accuracy, redundancy, and telemetry for real-time data transmission. Network resilience is bolstered by backup power, dual communication systems, and fault alerts.Another critical component for flood forecasting is the sub-daily rainfall network. High-resolution, high-quality rainfall data at sub-daily intervals are essential for capturing short-duration, intense rainfall events that can lead to flash flooding. The design of a sub-daily rainfall network should prioritize strategic placement of gauges in areas with high rainfall variability and known flood risk areas. Real-time data collection and transmission are paramount to ensure timely updates for flood forecasting models. Integration of automated quality control processes can further enhance the reliability of sub-daily rainfall data.We propose systematic reviews of network performance, integration of historical and real-time data archives, and automated quality control to improve data reliability. Advanced metadata management and API-based data dissemination enhance usability for stakeholders. These standards ensure hydrometric networks remain integral to flood forecasting, minimizing flood risks and improving public safety. The findings provide a blueprint for developing resilient, effective hydrometric networks that address the evolving needs of flood forecasting and warning services.

This paper aims to evaluate the impacts of the economic context on traffic congestion and its consequential effects on private vehicle accessibility. We conduct a long-term analysis of spatiotemporal traffic congestion patterns in Madrid (Spain), comparing two urban realms: the 2008 economic crisis and the following post-crisis situation. We apply TomTom Speed Profiles data to assess daily variations in traffic congestion and their changes between both periods, and Twitter data to capture spatial patterns of the daily pulse of the city. Increased traffic, a by-product of economic recovery, resulted in higher congestion, particularly during peak hours. Nevertheless, these changes are spatially uneven. In the city core, an increase in congestion is relatively temporally homogeneous, while in the peripheral suburban zones, there has been only a marginal increase in travel times. On the other hand, in the urban outskirts, increased traffic congestion is more severe but visibly different between north and south. These differences have strong social connotations: over 40% of the population experienced a dramatic increase in travel times (more than 25%) during peak hours. Moreover, low-income groups are more likely to live in the more affected southern districts, suffering most the negative consequences of increased congestion.

This study uses GIS spatial analysis and open-source data based on flood hazards formulas to provide a thorough analysis and forecast of floods using Geographic Information Systems (GIS) and Remote Sensing (RS). The Wadi Baysh Dam overflowed with water in 2016 as a result of the torrential downpour that hit the dam, its streams, and the surrounding surroundings results the increasing the storage capacity. In terms of water reserves, it is the dam that is thought to be the biggest in the Kingdom. Some settlements in Baysh, which are under siege since their highways are blocked off, suffered greatly as a result of flooding. However the storage capacity of Wadi Baysh Dam around 200 million cubic meters of water. In order to identify the actual Wadi Baysh flood-vulnerable area, this study combined data from remote sensing, the Geographic Information System (GIS) technique, and an Analytical Hierarchy Process (AHP) computation. The investigation was carried out to determine the viability of using remote sensing open-source data,basemap of the study area andArcGIS10.4 software in Wadi Baysh risk management. Slope (10%), height (15%), drainage density (20%), distance to stream (20%), rainfall intensity (25%), and land cover (10%) all have a proportion of relative weight that determines how flooding is analyzed. As a result of the most recent GIS analysis (weighted overlay approach), five zones are identified within the extracted flood region based on the categories of hazard. There are ten blocks in all—five in a high-risk area and five in a moderately dangerous one. The paper looks at the flood risk in Wadi Baysh using a combination and integration betweenremote sensing opensources data and GIS models. High precipitation and runoff in the dam’s downstream sections will be the main cause of difficulties, causing a great deal of death and destruction to agricultural and urban areas. Eventually, a control room connected to the GIS model and remote sensing devices will be needed in this valley to implement an efficient warning system and flood hazard management system.

A flood risk decision making approach for Mediterranean tree crops using GIS; climate change effects and flood-tolerant species

Abstract. Urban areas are increasingly experiencing more frequent and intense pluvial flooding due to the combined effects of climate change and rapid urbanization – a trend expected to continue in the coming decades. This highlights the growing need for effective flood forecasting and disaster management systems. While recent advances in GPU computing have made high-resolution hydrodynamic modeling feasible at the urban scale, operational use remains limited, particularly for large domains where single-GPU processing falls short in terms of memory and performance. This study demonstrates the capabilities of the hydrodynamic model RIM2D (Rapid Inundation Model 2D), enhanced with multi-GPU processing, to perform high-resolution pluvial flood simulations across large urban domains such as the whole state of Berlin (891.8 km2) within operationally relevant timeframes. We evaluate RIM2D’s performance across spatial resolutions of 2, 5, and 10 m using GPU configurations ranging from 1 to 8 units. Two flood scenarios are analyzed: the real-world pluvial flood of June 2017 and a standardized 100-year return period (HQ100) event used for official hazard mapping. Results show that RIM2D can deliver detailed flood extents, flow characteristics, and impact estimates for the 48 h 2017 event in 8 min at 10 m resolution, 34 min at 5 m, and approximately 5.5 h at 2 m using 8 A100 GPUs – fast enough to be integrated into real-time early warning systems. Multi-GPU processing proves essential not only for enabling high-resolution simulations (e.g., dx= 2 m or finer), but also for making simulations at resolutions finer than 5 m computationally feasible for flood forecasting and early warning applications. Additionally, we find that beyond 4 GPUs, runtime improvements become marginal for 5 and 10 m resolutions, and similarly, more than 6 GPUs offer limited benefit at dx= 2 m resolution, illustrating the balance between computational nodes of the used GPUs and number of raster cells of the model. Moreover, simulations at a finer dx= 1 m resolution demand more than 8 GPUs to be run. Overall, this work demonstrates that large-scale, high-resolution flood simulations can now be executed rapidly enough to support operational early warning and impact-based forecasting. With models like RIM2D and the continued advancement of GPU hardware, the integration of detailed, real-time flood forecasting into urban flood risk management is both technically feasible and urgently needed.

The past few years have seen increasingly frequent and intense floods, culminating in 2024 with a year characterized by widespread and devastating inundations worldwide. In Europe despite major advancements in flood forecasting and flood protection measures undertaken in the past, in2024 heavy rainfall events resulted in severe flood impacts and massive socio-economic losses, claiming over three hundred lives. Extreme precipitation, breaking observed records, is expected to have an increased likelihood under global warming all the more we should question and scrutinize our knowledge about the frequency and severity of floods. Reliable estimates of these are challenging even for the current climate conditions, and disaster gaps can lead us to an underestimation of the risks. The use of the UNSEEN method (Thompson et al. 2017) has been proved to be very valuable in estimating both, unprecedented but plausible extreme floods and droughts.Here we present a simple method to expand the UNSEEN method to develop storylines under various global warning levels. We selected precipitation scenarios with different spatial patterns for estimated return periods between 100 and 1000 years from pooled re-forecasts from ECMWF (ENSext and SEAS5), providing 8400 years of plausible weather sequences. The selected climate scenarios are perturbed by increasing the precipitation intensity according to the Clausius-Clapeyron relation for five different global warming levels, and used to run coupled hydrologic-hydraulic simulations. The results show that record-breaking, high-impact river floods are possible under the current atmospheric conditions, and climate change substantially aggravates flood impacts, as the relative increase in peak discharge can be significantly larger than the increase in precipitation, leading to a disproportionally high flood impact increase. The development of storylines of extreme flood events with a high spatial and temporal resolution are a valuable tool to explore, describe, and communicate extreme events and their dynamics. Such instruments are key for developing an informed vision and comprehensive protective measures in terms of flood risk management and emergency response. ReferencesThompson, V., Dunstone, N. J., Scaife, A.A., Smith, D. M., Slingo, J. M., Brown, S. and Belcher, S.E.: High risk of unprecedented UK rainfall in the current climate. Nat Commun 8, 107, https://doi.org/10.1038/s41467-017-00275-3, 2017.

In the context of rapid global population growth and extensive economic development, urbanization is expanding rapidly. The expansion of urbanization brings about increasingly complex challenges for cities, and flooding is one of the disasters faced. Climate change has led to a significant increase in extreme hydrological events, particularly a sharp rise in rainfall intensity, further elevating the risk of flooding in low-lying urban areas. The study area is located in Taipei City, characterized by low-lying terrain surrounded by mountains, and is influenced by subtropical climate. The frequent occurrence of heavy rainfall during the monsoon season and typhoons contributes to frequent flooding events, with the additional impact of climate change increasing the risk of intense rainfall. Therefore, the real-time prediction of regional flooding and its application in urban management becomes an imperative task, aiding in early warning, effective flood risk response, and ensuring sustainable urban development. This study utilizes the Double-Encoder Transformer model for real-time flood forecasting leveraging dual-encoder architecture to process and analyze diverse data types relevant predicting floods. One encoder could be dedicated to interpreting meteorological data, such as rainfall spatial distribution. This encoder focuses on extracting and understanding the complex patterns in weather-related data, which are crucial for predicting the likelihood of flooding. The second encoder, on the other hand, could handle geographical and environmental data, including terrain topology, and land use patterns. This encoder is adept at understanding how environmental factors contribute to flood risk in specific areas. By concurrently processing these two streams of information, the Double-Encoder Transformer can create a more comprehensive prediction model. It can identify correlations between meteorological conditions and environmental responses, leading to more accurate and timely flood forecasts. This approach enhances the model's ability to predict not only when and where floods might occur but also their potential severity, aiding in disaster preparedness and resource allocation. Overall, the application of the Double-Encoder Transformer in flood forecasting represents a significant advancement in disaster management, leveraging AI's power to integrate and analyze complex, multi-faceted data for better, more informed decision-making in critical situations.

A comparative study of Machine Learning and Deep Learning methods for flood forecasting in the Far-North region, Cameroon

Flooding imposes considerable property risk, and flood maps and flood insurance help prospective and existing property owners assess the potential risk. The US Federal Emergency Management Agency (FEMA) works with local and state officials to produce flood maps. Using these flood maps and demographic attributes, prior research has demonstrated correlations between the percent of a tract identified as disadvantaged and the percent of the tract covered by flood zones. Until recently, FEMA flood maps were the primary assessment tool for flood risk, but First Street Foundation (FSF) has developed its own flood risk tools. This paper compares these alternative flood risk measures as a percent of census tracts in the Southeastern US states and assesses models of the risk measures with demographic, housing, policy and control variables. The main results are first that the FEMA and FSF maps often reveal diverging levels of risk per tract. Second, the demographics correlating with tract-level risk differ markedly for the two risk measures. Third, the results vary considerably by state with more divergence in some states than others, and who is at risk of flooding across the states varies between the FEMA and FSF measures.

Considering the intensified floods in the tropical river basins, this study assesses the climate change‐induced flood risk accounting for the reservoir releases associated with crop damage and the monetary benefits from paddy cultivation. Using the three best‐performing Global Climate Models (GCMs) out of nine, it analyzes the flood risk for agricultural production associated with a design flood in the typical Mahanadi River delta under a retrospective and six (2010s, 2040s and 2070s; and Representative Concentration Pathways (RCP): RCP4.5 and RCP8.5) projected scenarios. Subsequently, a probable flood adaptation strategy employing alternate rice planning in the near‐future period and a selected GCM is proposed. It reveals an increased flood hazard in the mid‐century followed by a decline towards the end of the century. Prolonged duration of low flood depth leads to very high crop vulnerability to floods in the projected periods under both RCP4.5 and RCP8.5 scenarios. While a little to moderate increase in overall flood risk is depicted by the MIROC‐ESM‐CHEM model‐based predictions, the HadGEM2‐AO and IPSL‐CM5A‐MR models indicate high flood risk in the future projected periods as compared to the historical episode. An alternate agricultural planning considering the net return from crop production can serve as a potential adaptation strategy dealing with the harsh effects of extreme floods in the river delta. The advocated methodology of flood risk assessment and adaptation using hydrodynamic flood modelling with consideration of reservoir releases is one of its kind and can be implemented in any world river basin.