Open research questions in Flood Risk Assessment and Management

What the literature leaves open

• Results indicate that Random Forest and Support Vector Machines achieved high predictive accuracy but showed tendencies toward overfitting, whereas CART, MaxEnt and MARS demonstrated more stable generalization under limited data conditions.

  Comparative assessment of ensemble machine learning and Google Earth Engine for flood susceptibility mapping in the Dipota River Basin, Assam, India · 2026 · DOI

• The relevant questions are not only whether AI can contribute, but under what conditions it contributes usefully: how it should be constrained, how it should be validated, what parts of the workflow it should accelerate, and how it can support research, operations, and training without weakening scientific or institutional discipline.

  Operational SAR flood mapping as a full-stack systems problem: an AI-enabled perspective · 2026 · DOI

• These results suggest the feasibility of AI‐agent‐assisted hydrologic modeling in the tested case, while robustness and generalization across broader basins and events remain to be established through systematic validation, probabilistic meteorological forcing, and expert review of automated outputs.

  AI Agent for Hydrologic Modeling: Definition, Development, and Application · 2026 · DOI

• The integration of synthetic storm modeling, hydrody- namic simulation, HAZUS-based loss estimation, and screening-level regression provides a repeatable approach for evaluating natural-infrastructure benefits, especially where empiri- cal storm-damage data are limited.

  Assessing the role of the North Carolina outer banks as a nature-based feature in reducing coastal flood damages · 2026 · DOI

• Abstract Despite increasing recognition of the severe and compounding risks posed by post-wildfire flooding and debris flows, little is known about available post-wildfire flood and debris flow decision-making frameworks and their effectiveness in disaster response, recovery, and resilience efforts.

  Decision making tools for post-wildfire flood response and resilience: a systematic literature review · 2026 · DOI

• Urban pluvial flood risk in industrial zones is intensifying under climate change, yet the joint influence of digital elevation model (DEM) resolution, surface roughness heterogeneity, and infiltration capacity on simulation accuracy remains insufficiently characterized.

  Balancing Accuracy and Efficiency for Sustainable Flood Adaptation: Multi-Resolution LiDAR DEM Sensitivity Analysis of Urban Pluvial Flooding in the Gumi Industrial Complex · 2026 · DOI

• While this study provides evidence linking LULC changes to surface runoff intensifica- tion in the FCC, several limitations should be acknowledged: 1. The study assumed spatially uniform rainfall distribution due to limited access to a dense network of ground-based rain gauges within the FCC. Only three NIMET stations provided long-term records, requiring IDW interpolation that may smooth localized convective storm cells common in tropical climates. 2. Landsat 30 m resolution may fail to capture fine-scale urban features (< 30 m) such as individual drainage channels, permeable pavements, small detention ponds, or informal settlement patterns that influence localized runoff. Odiji et al. Discover Cities (2026) 3:103 Page 23 of 26 3. The FCC lacks stream gauging stations for direct calibration of the SCS-CN model. Validation relied on proxy data (SAR flood extents, field-observed flood points) rather than direct runoff measurements. 4. The study did not incorporate drainage network capacity, maintenance frequency, building density, or population distribution—factors that mediate actual flood impacts even when runoff is accurately modeled. 5. Long-term rainfall trends (1990–2020) were assumed stationary. Future work should integrate downscaled climate projections (RCP/SSP scenarios) to assess non- stationary runoff responses. Despite these limitations, the methodology remains sound, and the findings are con- sistent with other urban hydrology studies globally. The integration of multi-temporal remote sensing data with GIS-based hydrological modeling provides a useful framework for assessing surface runoff in data-scarce environments like Abuja. The study clearly demonstrates that LULC change particularly the expansion of built- up areas and concurrent loss of vegetated surfaces is a major driver of increased surface runoff in Abuja. These changes are occurring in a context of limited drainage capac- ity and hydrologically unfavorable soils, compounding flood risks across the city. The results offer valuable guidance for land use planning and urban infrastructure design, emphasizing the urgent need for integrating green infrastructure, strengthening policy enforcement on urban development, and enhancing stormwater management systems. As urbanization accelerates, such evidence-based planning tools will be critical for fos- tering resilience and achieving sustainable urban futures. Additionally, they reinforce the need for integrated urban planning frameworks that incorporate green infrastructure, enforce zoning regulations, and prioritize natural landscape preservation. These mea- sures are essential for mitigating flood hazards and achieving the sustainability targets outlined in SDG 11: making cities inclusive, safe, resilient, and sustainable.

  Assessment of urban surface runoff potential in response to land use land cover changes in the Federal Capital City, Abuja, Nigeria · 2026 · DOI

• focus on quantitative performance evaluation, including latency measurement, false alarm rate, and long-term reliability testing, as well as the implementation of multi-sensor configurations and more improve scalability and system resilience. Overall, the proposed system demonstrates the feasibility of a low-cost and practical IoT-based solution for flood early warning applications at the community level.

  Design and Evaluation of an IoT-Based Flood Early Warning System Using Conductive Water Level Sensor · 2026 · DOI

• In this study, the flood susceptibility of the Putna River basin was assessed using advanced machine learning models, namely MLP and three optimized hybrid approaches: ABC–MLP, EHO–MLP and DE–MLP, integrated in a GIS environment. The combination of these mod- Natural Hazards (2026) 122:4061 3 406 Page 28 of 32 els for the first time in the literature for flood susceptibility estimation and their integration in the GIS environment represents the main novelty of the present research work. The comparative analysis clearly demonstrated the superiority of the hybrid models over the classic MLP neural network, highlighting the essential role of optimization algorithms in increasing predictive performance. Among the tested models, DE–MLP achieved the best results, recording an accuracy of 0.982 and an AUC-ROC value of 0.985 on the test set, indicating an excellent discrimination capacity between flood-prone and unaffected areas. Also, the high values of precision, recall and F1 score (all 0.982) confirm the balanced and robust nature of this model. The EHO–MLP and ABC–MLP models performed very well, with accuracies of 0.964 and 0.946, respectively, and AUC-ROC values of 0.970 and 0.975, but lower than the DE–MLP model. The simple MLP model performed the worst (AUC-ROC = 0.945), highlighting the clear advantage of hybrid approaches. The implications of this work also explain the better flood management with the optimized models produce flood susceptibility maps that can be used for better flood management. High risk areas can be targeted for flood mitigation, reducing impact on communities and infrastructure. And finally, accurate flood susceptibility assessment means better resource allocation, focusing on high risk areas. The analysis of the importance of variables indicated slope (0.334), altitude (0.249), distance from the river (0.244) and rainfall (0.183) as the dominant factors in controlling the flooding processes. The final susceptibility maps highlight the high-risk areas, providing valuable support for territorial planning, flood risk management and decision-making at local and regional levels. The results confirm that the DE–MLP model represents an efficient and reliable solution for assessing flood susceptibility in regions with similar characteristics. Based on the new findings of this work, as well as its limitations, the future works can consider other areas to explore with different geography and climate to test the models. In addition, real time data such as rainfall and river flow for dynamic flood susceptibility assessment is an interesting problem and they may assist improve the flood susceptibility mapping works. Combine multiple machine learning algorithms to create more accurate models is also a great work that can build on this research and improve flood risk management.

  Evaluation of flood susceptibility through an artificial neural network-based differential evolution optimization algorithms and GIS techniques · 2026 · DOI

• Although the method for estimating the number of people affected by a flood is promising, it was limited by the currency of the input data and could therefore have been overstated or underestimated. For example, the SPOT Building Count (SBC) data is incomplete and outdated, with the latest update in 2017, while the average household size was based on the 2016 census. This limits the study’s accuracy in quantify- ing the number of people affected. The SBC layer had 577,459 buildings in eThekwini, while Johan- nesburg had 705,579. These numbers are below the official statistics reported by Stats SA, indicating that there were 963,011 in eThekwini and 1,434,856 in Johannesburg in 2011 (Mbambo & Agbola, 2020). This discrepancy is even larger compared to the 2016 statistics of 1,125,767 buildings in eThekwini and 1,853,371 in Johannesburg. Moreover, other types of settlements, such as informal settlements, are mapped using generalized polygons since individual build- ings cannot be detected with 2.5 m spatial resolution SPOT imagery (Kemper et al., 2015). The lack of data on informal settlements reduces the accuracy of the results, as most informal settlements are near riv- ers where most flooding was detected. This is evident with the April 2022 flood incident, where our results show about 20,000 people being affected, while newspaper reports indicated close to 40,000 (Grab & Nash, 2023). However, the results for the April 2019 flood incident in eThekwini indicated the highest number of people affected by the floods, consistent with the previous declaration of this flood being the most catastrophic in history (Olanrewaju & Reddy, 2022). Comparatively, the October 2017 floods had the second-highest number of people affected, despite relatively low rainfall, and the locations of the floods in relation to buildings. Indeed, the affected popula- tion estimates varied between Sentinel-1 and Senti- nel-2, reflecting differences in the flooded areas esti- mated by the two sensors (see Table 3). It is also worth noting that validating satellite- derived flood inundation is difficult without coin- cident, very-high-resolution data. Other sources of under- and overestimation are related to assumptions made in change-detection techniques, such as the assumption of stable vegetation conditions between the pre-flood and during-flood images. In reality, such changes will affect SAR backscatter, leading to Environ Monit Assess (2026) 198:495 false flood detections (Hess et al., 1995; Tran et al., 2022). Moreover, changes in incidence angles of the Sentinel-1 imaging system between the pre-flood and during-flood images may also lead to incorrect inter- pretations in the selection of optimal thresholds (Pul- virenti et al., 2016). Therefore, the correct choice of preprocessing is important for good results (Conde & Muñoz, 2019). In this study, we incorporated the scene-specific orbit file, slope masking, and ter- rain correction to avert such errors. Nonetheless, the results in the current study are promising and were consistent with weather station rainfall measure- ments (Fig. 8). Although the extent of the flooded area between Sentinel-1 and Sentinel-2 could not be validated due to a lack of reliable validation data, this study used recommended methods that were vali- dated in previous studies (Adiba & Bioresita, 2023; Clement et al., 2018; Notti et al., 2018; Risling et al., 2024). Therefore, similar accuracy is expected in this study.

  Geospatial analysis of flooding events using Sentinel-1 and Sentinel-2 data: a tale of two South African cities · 2026 · DOI

• Frontiers in Environmental Science 17 frontiersin.org Paramanik et al. 10.3389/fenvs.2026.1729457 More dynamic models that account for changes in land use patterns could provide more accurate predictions. Additionally, this study does not critically incorporate socioeconomic factors in flood risk management. Factors such as population growth, urbanization, and infrastructure development could significantly influence flood susceptibility but were not explicitly considered this assessment. Future studies should aim to integrate socioeconomic projections with climate and environmental data to offer a more holistic view of flood risks. in 4.4 Limitations in machine learning-based flood risk evaluation and policy consequences Although there is considerable progress, machine learning algorithms for flood as well as landslide vulnerability continue to encounter substantial limits concerning data quality along with model generalizability. Class imbalance as well as biased training datasets frequently result in skewed predictions, wherein non-event conditions predominate, thereby diminishing model reliability and obstructing precise risk identification in future climate scenarios, particularly when conditioning parameters are selected inconsistently across areas (e.g., topography, climate variables), which limits transferability and extrapolation efficacy. The dependency on historical as well as remote sensing datasets, devoid of substantial real-time inputs, constrains operational prediction capabilities, as numerous places lack the continuous high-resolution imagery and hydrological observation infrastructure necessary for dynamic prediction (Han and Semnani, 2025). Third, the ability to interpret of intricate deep learning models poses a barrier; although achieving great accuracy, their “black-box” characteristics diminish transparency and hinder stakeholders’ comprehension of the physical links between predictors and dangers. Ultimately, the majority of current assessments incidents or certain study regions, concentrate on singular prompting apprehensions regarding the ability of trained models to consistently generalize to novel flood or landslide scenarios across varied geographies or in response to changing climate conditions. among emphasize framework.

  Impact of climate change on future flood susceptibility using different climatic parameters and deep learning algorithms in eastern Himalayan region · 2026 · DOI

• We need to acknowledge several limitations. First, the ML models rely on historical flood damage data as the target variable, which may not fully capture future dynamics under climate change. In addition, incorporating more flood damage locations from a larger number of flood events spanning broader spatial and temporal extents would further improve ML model training and robustness. Second, al- though ML models have achieved strong predictive performances, this data- driven approach, without considering the underlying com- plex physical process, is highly subject to the quality of data and se- lection of risk factors. Third, the analysis is made on two scenarios (GFD and EFD), which does not account for a spectrum of scenarios such as the cascading impacts of compound flooding. Overall, the framework’s scalability beyond the USGAC requires careful adapta- tion to local contexts. Future research should incorporate additional flood damage lo- cations from a larger number of events across broader spatial and temporal scales. Integrating climate projections can help capture the future dynamics of flood risk, whereas including compound flood events can provide a more comprehensive assessment. In addition, complementing data- driven ML models with physics- based models can enhance the robustness and reliability of predictions. MATERIALS AND METHODS This study primarily required two key elements: flood damage data and FRFs, both of which were integrated into the ML models (Fig. 2). Here, we used flood damage data as target variables and FRFs as pre- dictor variables. The methodological process of this study is briefly described below.

  A tale of two coasts: Unveiling US Gulf and Atlantic coastal cities at high flood risk · 2026 · DOI

• The study employs NFIP claims data which may underestimate actual damages due to uninsured properties and underinsurance. A comparison of damage estimates between NFIP-insured properties and uninsured or self-insured properties under the same simulated storm scenarios would validate the representativeness of the empirical damage functions across the full population.

  Valuing Salt Marshes as Nature-based Infrastructure for Coastal Flood Mitigation: A Case Study of Chatham County, GA · 2026 · DOI

• The distributional analysis shows substantial variation in salt marsh mitigation benefits across poverty quartiles and racial composition of census tracts, with lower-income and higher-minority areas receiving proportionally fewer benefits. The mechanisms driving this spatial inequity in nature-based infrastructure protection and strategies to equalize benefit distribution require investigation.

  Valuing Salt Marshes as Nature-based Infrastructure for Coastal Flood Mitigation: A Case Study of Chatham County, GA · 2026 · DOI

• The analysis restricts the damage function estimation sample to water depths not exceeding eight feet and single-family residential buildings with up to three stories. The effectiveness of salt marsh flood mitigation for deeper water depths, multi-family residential structures, and commercial buildings has not been quantified.

  Valuing Salt Marshes as Nature-based Infrastructure for Coastal Flood Mitigation: A Case Study of Chatham County, GA · 2026 · DOI

• The integrated hydrodynamic modeling and empirically estimated damage functions approach was applied only to Chatham County, Georgia. The transferability of this methodology to other coastal regions with different marsh morphologies, bathymetries, soil types, and building stock characteristics remains unvalidated.

  Valuing Salt Marshes as Nature-based Infrastructure for Coastal Flood Mitigation: A Case Study of Chatham County, GA · 2026 · DOI

• Historical damage functions estimated from past NFIP claims may not fully capture damage processes under future climate conditions, particularly in the presence of compound flooding where storm surge and heavy rainfall occur simultaneously. The paper acknowledges this limitation but does not specify validation protocols for testing salt marsh mitigation effectiveness under compound flood scenarios.

  Valuing Salt Marshes as Nature-based Infrastructure for Coastal Flood Mitigation: A Case Study of Chatham County, GA · 2026 · DOI

• Despite the large costs of covering flood losses, little is known about whether the National Flood Insurance Program (NFIP) affects households’ decisions to sort into more flood-prone locations.

  Does the National Flood Insurance Program Drive Migration to Higher Risk Areas? · 2023 · DOI

• Moreover, the study highlighted the benefits of cloud‐computing platforms like GEE in addressing challenges associated with big data filtering, processing and analytics, thereby enhancing environmental monitoring and assessments, which may have been limited by the unavailability of advanced processing tools and seamless cloud‐free data.

  Available satellite data for monitoring small and seasonally flooded wetlands in semi‐arid environments of southern Africa · 2023 · DOI

• Abstract Google Trends (GT) offers a historical database of global internet searches with the potential to complement conventional records of environmental hazards, especially in regions where formal hydrometeorological data are scarce.

  The utility of Google Trends as a tool for evaluating flooding in data‐scarce places · 2021 · DOI

• Finally, we argue that remote sensing techniques, specifically, the combined use of unmanned aerial vehicles and structure from motion photogrammetry, are key to bridging gaps in understanding and meeting the challenges of managing Mediterranean IRES.

  Mediterranean intermittent rivers and ephemeral streams: Challenges in monitoring complexity · 2019 · DOI

• Abstract Flood risk in semi-arid, snow-fed basins is increasingly influenced by both land-use and climate change, yet their combined future effects remain poorly quantified.

  Predicting the flood susceptibility under land use and climate change scenarios using deep learning algorithms · 2026 · DOI

• Despite their critical importance and high risk of vulnerability to floods, limited research has examined the failure mechanisms and their contributing factors on steel truss bridges, as one of the most widely used types of bridges.

  Advanced framework for post-flood assessment of steel truss bridges under data-constrained conditions: integrating engineering insights and empirical fragility models · 2026 · DOI

• A limitation of this research is that it does not account for seasonal vegetation changes that significantly affect infiltration and runoff rates in the Tyler-Flint region.

  Predictive Modeling of Flash Floods: Investigating Hydrology and Land Cover Dynamics through Remote Sensing Data · 2026 · DOI

• Approaches to make cities more resilient to floods are emerging, notably with the design of flood-resilient structures, but relatively little is known about the role of urban form and its complexity in the concentration of flooding.

  How urban form impacts flooding · 2024 · DOI

Most-cited papers in Flood Risk Assessment and Management

• The Surface Water and Ocean Topography Mission: A Breakthrough in Radar Remote Sensing of the Ocean and Land Surface Water · Geophysical Research Letters · 2024 · 295 citations
• Integrating spatial continuous wavelet transform and kernel density estimation to identify ecological corridors in megacities · Landscape and Urban Planning · 2020 · 176 citations
• The extraordinary Zhengzhou flood of 7/20, 2021: How extreme weather and human response compounding to the disaster · Cities · 2022 · 149 citations
• Managing flood risk in the UK: towards an integration of social and technical perspectives · Transactions of the Institute of British Geographers · 2002 · 134 citations
• Urban flood risk and green infrastructure: Who is exposed to risk and who benefits from investment? A case study of three U.S. Cities · Landscape and Urban Planning · 2022 · 124 citations
• Machine Learning-Based Wetland Vulnerability Assessment in the Sindh Province Ramsar Site Using Remote Sensing Data · Remote Sensing · 2024 · 118 citations
• Increasing urban flood risk in China over recent 40 years induced by LUCC · Landscape and Urban Planning · 2021 · 116 citations
• Vertical accuracy assessment of freely available global DEMs (FABDEM, Copernicus DEM, NASADEM, AW3D30 and SRTM) in flood-prone environments · International Journal of Digital Earth · 2024 · 113 citations
• Enhancing human resilience against climate change: Assessment of hydroclimatic extremes and sea level rise impacts on the Eastern Shore of Virginia, United States · The Science of The Total Environment · 2024 · 113 citations
• Urban flood risk management needs nature-based solutions: a coupled social-ecological system perspective · npj Urban Sustainability · 2024 · 112 citations

Most recent work

• Impacts of climate change and urbanization on soil moisture dynamics have reduced regional flood resilience · Landscape and Urban Planning · 2026
• Flood Risk Assessment Using Multitemporal Satellite Data and Explainable Artificial Intelligence (XAI): Analyzing Susceptibility and Driving Factors · Water Resources Management · 2026
• Understanding process differences in the impact of built–natural environments on compound heat–flood risks through urban physical characteristics · Landscape and Urban Planning · 2026
• Balancing Accuracy and Efficiency for Sustainable Flood Adaptation: Multi-Resolution LiDAR DEM Sensitivity Analysis of Urban Pluvial Flooding in the Gumi Industrial Complex · Sustainability · 2026
• Deconstructing urban flooding from the perspective of vulnerable communities in Northwestern Pakistan · Frontiers in Human Dynamics · 2026
• Foundational Models and Intelligent Agents for Urban Drainage Systems: The Road Ahead · 2026
• Valuing Salt Marshes as Nature-based Infrastructure for Coastal Flood Mitigation: A Case Study of Chatham County, GA · 2026
• Evaluation of Flood Protection Infrastructure Safety through Satellite Image Analysis with Artificial Intelligence Methods · Communications of the IBIMA · 2026
• Data-driven flood susceptibility assessment using hybrid machine learning and optimization techniques: case of the Sedrata Watershed, NE Algeria · Scientific Reports · 2026
• Flood-Resilient Road Design Standards for the Sudd Wetland Region of South Sudan · Zenodo (CERN European Organization for Nuclear Research) · 2026

Related topics in Environmental Science