Urban Pluvial Flooding Research Articles

Influence of lunar phases and meteorological factors on rainfall in Karachi City, Pakistan

Climate change is responsible for unpredictable weather patterns in Southeast Asia, especially in Pakistan. To better understand the influence of lunar phases and meteorological factors on extreme rainfall events in Karachi City, 42 years of rainfall data and 12 years of sea level data were analyzed. The study used wavelet transform coherence to assess the influence of lunar phases and sea level fluctuations on extreme rainfall. Additionally, three machine learning algorithms - random forest, decision tree, and extreme gradient boosting - were applied first to meteorological data (Case A) and then to combined meteorological and lunar data (Case B) to evaluate their influence on extreme rainfall. Humidity in case A, and cloud cover in case B were the most influential meteorological factors along with west winds, and north winds in both cases, while the waning crescent, waning gibbous, and waxing gibbous lunar phases were the most dominant lunar phases in Case B. The coefficient of determination (R2) in random forest increased from 0.871 in Case A, to 0.893 in Case B and proved to be the most accurate algorithm in both cases. A rainfall threshold of 50 mm, above which urban pluvial flooding occurred, was consistently observed during the waxing gibbous, waning crescent, and waning gibbous lunar phases. The study improves our understanding of lunar influenced extreme rainfall temporal variations values of machine learning models for rainfall analysis and extreme events projection and provide essential information to develop efficient adaptation and mitigation policies in pluvial flood-prone urban areas.

Compilation of Benchmark Pluvial Flood Datasets

Urban pluvial flooding is a dangerous natural hazard that has become more common in recent years, making it a global threat to metropolitan areas. Surface water floods and flash floods are two common types of pluvial flooding. The second one is extremely hazardous and damaging due to both the power of the water and the debris that is frequently swept up in the flow. Buildings and their contents are damaged by pluvial floods, which also disrupt stormwater drainage, transportation, and electrical supply. Future flood risks are anticipated to rise because of urbanization, climate change, and global warming. Many essential features of urban flood hazards are still poorly understood due to the complex nature of the relevant processes and the paucity of long-term field observations. Therefore, in this paper, we have collected different sources of data that are used for flood forecasting. Satellite data, gauge measurements, and citizen observations are the most important sources of precipitation and flood data. Conducting a global case study review also showed that the Digital elevation model (DEM) and rainfall are the most crucial variables when it comes to floods, and they are nearly universally employed in all models. The discussion suggests that soil moisture conditions are likely to be the predominant mechanism causing the observed flood trends.

Efficacy of rain barrels and rain gardens to reduce urban pluvial flooding in densely built-up residential areas: a case study on Miyahara-Cho in Saitama City, Japan

In recent years, flooding in densely built-up residential areas has tended to increase because of the increased frequency of heavy rain and climate change. Urban pluvial flooding mitigation is thus needed by reducing the runoff of sewer systems using green infrastructure solutions. In this regard, rain barrels (RBs) and rain gardens (RGs) are considered effective stormwater management methods. This study aimed to verify the effectiveness of RBs and RGs and their combined use in reducing urban pluvial flooding in densely built-up residential areas. This research undertook a case study in Miyahara-Cho, Saitama City, Japan. It involved the analysis of the total runoff of sewer systems and the flood condition in the sewer to verify the rainwater runoff control effect of RBs and RGs through simulations using the US Environmental Protection Agency’s stormwater management model. This study simulates two main types of rainfall events: short duration and long duration. The results of the short duration rainfall showed that the combination of RBs and RGs is highly effective and can reduce the impact of excess runoff. In the case of the long duration rainfall, the volume and speed of stormwater runoff of the first peak rainfall were reduced.

A comparison of machine learning surrogate models of street-scale flooding in Norfolk, Virginia

Low-lying coastal cities, exemplified by Norfolk, Virginia, face the challenge of street flooding caused by rainfall and tides, which strain transportation and sewer systems and can lead to personal and property damage. While high-fidelity, physics-based simulations provide accurate predictions of urban pluvial flooding, their computational complexity renders them unsuitable for real-time applications. Using data from Norfolk rainfall events between 2016 and 2018, this study compares the performance of a previous surrogate model based on a random forest algorithm with two deep learning models: Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU). The comparison of deep learning to the random forest algorithm is motivated by the desire to utilize a machine learning architecture that allows for the future inclusion of common uncertainty quantification techniques and the effective integration of relevant, multi-modal features.

Mapping impervious surface area increase and urban pluvial flooding using Sentinel Application Platform (SNAP) and remote sensing data.

Expansion of urban impervious surface (UIA) and increased urban pluvial flooding (UPF) have an impact on urban dynamics, socioeconomic activities, and our environment. Therefore, monitoring the increase in UIS and its effect on UPF is essential. The notion of this research is based on the mapping of impervious surface area increase in three major cities of Pakistan. There were two key objectives: (i) Mapping impervious surface area growth using the global impervious surface area index (GISAI) on Google Earth Engine from 1992 to 2022 and (ii) mapping the pluvial flood extent in selected urban areas using Sentinel-1 Ground Range Detected (GRD) data. Thus, we have utilized the GISAI for mapping urban impervious surface area (UISA) using Landsat time-series data on GEE. Our research findings revealed that about 16.8%, 23.5%, and 16.4% of the impervious surface have been increased in Islamabad, Lahore, and Karachi, respectively. Also, Lahore city has the highest overall accuracy, aiming at the GISAI of 93%, followed by Karachi and Islamabad with an overall accuracy of 86% and 85%, respectively. The results indicated that urban flooding has occurred in those areas where the ISA has grown during the last three decades. It shows significant changes in the impervious surface area that cause enhanced urban pluvial flooding in major cities of Pakistan. Also, Sentinel-1 data and the SNAP tool significantly mapped flooded areas in the selected zones. So, providing cities and local governments with increased quick flood detection capabilities is essential. It can also provide feasible policy recommendations for Pakistan decision-makers in city management. Therefore, we suggest a modeling-based solution to identify high-risk locations in major cities for upcoming UPF events.

Public preferences and willingness to pay for low impact development: a comparative case study of pilot sponge cities in China

Abstract Increased urbanization and climate change globally have increased the frequency of extreme weather, especially rainstorms and flooding. Low impact development (LID) has been adopted for several decades to reduce urban pluvial flooding and cope with urban climate change. However, LID hasn't been widely used due to the lack of sustainable funding. In this article, a discrete choice experiment method was used to understand public preferences and willingness to pay (WTP) for the functions of LID facilities. Four attributes were selected, and after that, 958 and 646 valid questionnaires were collected in the pilot sponge cities of Zhengzhou and Hebi, respectively. The results showed that respondents showed a significant WTP for LID facilities, with a desire to support rainwater drainage, recreation and entertainment, and the landscape environment. Experience of flooding had a significant effect on Zhengzhou respondents, but not on Hebi respondents. We have concluded that rainwater drainage should still be considered the most important function of LID facilities. At the same time, educational level and flood experiences may not affect the public's WTP. This article can provide a reference for cities that would like to implement LID and create a stormwater fee system to adapt to global change.

A drainage network-based impact matrix to support targeted blue-green-grey stormwater management solutions

Urban floods will continue to be an alarming issue worldwide due to climate change and urban expansion. The costly and less environmentally friendly grey infrastructure is not always the most adequate solution to resolve urban pluvial flooding issues. The combination of grey and blue-green infrastructures, also called hybrid infrastructure, has been considered a promising solution for urban stormwater management. Existing approaches for identifying suitable hybrid solutions frequently rely on global multi-objective optimization algorithms. We developed a pre-screening method that decomposes a drainage network into clusters of pipes connected to sub-catchments, based on pipe hydraulic characteristic that allows for the impact of infrastructure combinations (blue-green and grey) to be mapped. Four impact matrices are proposed to map the total, local, upstream, and downstream flood reduction of all possible blue-green, grey, and hybrid solutions. Using an urban catchment in Guangzhou (China) as a case study, results showed that such an exercise could identify prime candidate locations for blue-green and grey infrastructure while filtering out ineffective locations for flood reduction. Furthermore, the impact matrices enabled the identification of flood zones where blue-green infrastructure could handle flood mitigation without the need of local grey infrastructure upgrades. As such, they are not only useful for quick screening of suitable interventions for each flooded zone, but can also potentially serve as a priori knowledge before diving into the data and computationally expensive process of finding the most effective flood mitigation solutions.

A novel and efficient method for real-time simulating spatial and temporal evolution of coastal urban pluvial flood without drainage network

With increasing urban pluvial flood risks, proposing a real-time simulation method is essential. However, accurate simulation of spatiotemporal flood evolution is often impeded by incomplete or missing drainage data. This study proposes a hybrid method where a machine learning module is applied to generate point waterlogging depth for immediate calibration of equivalent infiltration and flood maps in the equivalent drainage module to address this issue. The accuracy and efficiency of hybrid method in flood real-time simulation under missing drainage data are highlighted by comparing with two hydrodynamic models. The outcomes evince that the waterlogging simulation deviation of the hybrid method is less than 0.1 m during design storms, while the computational efficiency can ideally reach up to 5 times of the traditional 1D/2D coupled hydrodynamic model. Overall, the hybrid method offers a promising solution for early warning and mitigation of urban pluvial floods, especially for cities lacking drainage data.

Matrix scenario-based urban flooding damage prediction via convolutional neural network

This study introduces a cutting-edge, high-resolution tool leveraging the predictive prowess of convolutional neural networks to advance the field of hazard assessment in urban pluvial flooding scenarios. The tool uniquely accounts for the high heterogeneity of urban space and the potential impact of complex climate scenarios, which are often underestimated by traditional data-reliant methods. Employing Shenzhen as a case study, the model showcased superior accuracy, resilience, and interpretability, illuminating potential flood hazards. The performance analysis shows that the model can accurately predict the vast majority of urban flood depths, but has errors in extreme flood predictions (depths greater than 35 cm). Findings underscore escalating flood impacts under enhanced scenario loads, with western and central Shenzhen—regions rife with construction—highlighted as particularly vulnerable. Under the most severe matrix scenario (Scenario 25), economic losses are estimated to be about $25,484 million. These commercial and residential hotspots are anticipated to suffer maximum economic loss, with these two areas accounting for 39.6% and 25.1% of the total losses, necessitating reinforced mitigation efforts, especially during extreme rainfall events and high soil saturation levels. In addition, the flooding control strategies should prioritize the reduction of flood inundation areas and integrate functionally oriented land use characteristics in their development. By aiding in the precise identification of flood-prone areas, this research expedites the development of efficient evacuation plans, bolsters urban sustainability, and augments climate resilience, ultimately mitigating flood-induced economic tolls.

Discharge coefficients to be used in inlet hydraulics

Urban pluvial floods can be produced or exacerbated by an insufficient density of inlets or by their poor hydraulic efficiency. Therefore, proper consideration of the hydraulic performance of inlets is essential to guarantee the correct functioning of urban drainage systems during heavy storm events. Recent advances in computational analysis in the field of hydrodynamics modelling allow the use of the well-known concept of dual drainage for design and planning purposes by simulating flow transfers between the surface layer (streets) and underground layers (sewers) through a proper hydraulic characterisation of inlet efficiency. Powerful commercial software packages allow simulations of flow transfer and use different approaches and formulas. Many of these approaches include the possibility of treating a sewer inlet as an orifice. In this context, this paper presents a methodology to obtain orifice discharge coefficients for three inlets previously tested at the hydraulics laboratory of the Technical University of Catalonia. Discharge coefficients of 0.18–0.58 were obtained for Froude numbers of 1.12–4.40, quite far from the usual recommended values. The proposed procedure can also be applied to non-tested grates.

Developing the Actual Precipitation Probability Distribution Based on the Complete Daily Series

The defense against urban pluvial flooding relies on the prediction of rainfall frequency, intensity, and long-term trends. The influence of the choice of the complete time series or the wet-day series on the rain analyses remains unclear, which affects the adaptive strategies for the old industrial cities such as Changchun in Northeastern China, with the outdated combined sewer systems. Based on the data from the two separate weather stations, four types of distributions were compared for analyzing the complete daily precipitation series, and their fitting accuracy was found in decreasing order of Pearson III, Pareto–Burr–Feller distribution (PBF), generalized extreme value (GEV), and Weibull. The Pearson III and the PBF probability distribution functions established based on the complete time series were found to be at least 458% and 227%, respectively, more accurate in fitting with the consecutive observations than those built from the wet-day-only series, which did not take account of the probability of the dry periods between the rain events. The rain depths of the return periods determined from the wet-day-only series might be over-predicted by at least 76% if the complete daily series were regarded as being more closely representative of the real condition. A clear threshold of 137 days was found in this study to divide the persistent or autocorrelated time series from the antipersistent or independent time series based on the climacogram analysis, which provided a practical way for independence determination. Due to the significant difference in the rain analyses established from the two time series, this work argued that the complete daily series better represented the real condition and, therefore, should be used for the frequency analysis for flood planning and infrastructure designs.

Pluvial flood adaptation using nature-based solutions: An integrated biophysical-economic assessment

Globally, flood events are considered the costliest natural hazard. Changes in precipitation patterns and large areas of impervious surfaces in urban environments are increasing the sensitivity of these systems to runoff production. At the same time, projected global sea-level rise may further increase the frequency of compound flooding due to simultaneous storm surge, sea-level rise and pluvial runoff that cause vast socio-economic and ecological impacts to coastal cities. In this context, over the last decade, the role of Nature-Based Solutions (NBS) has been recognised to support climate change adaptation by addressing ideas of multi-functionality, non-linearity and heterogeneity in urban design. Thus, increasing awareness about NBS benefits increases the willingness to accept these solutions. However, empirical evidence of NBS effectiveness at the urban catchment scale is still subject to debate. This study develops a spatial biophysical-economic framework that allows for the integrated assessment of NBS flood risk mitigation impacts, costs and benefits in the face of climate change, combining the Integrated Valuation of Ecosystem Services and Tradeoffs (InVEST) model, benefit transfer methods and Geographic Information System (GIS) tools. Specifically, the InVEST Urban Flood Risk Mitigation model was used to assess the biophysical impacts of NBS on urban pluvial flood risk, benefit-transfer methods were used to evaluate the economic implications of such solutions, and GIS was used to integrate and map biophysical impacts and economic implications. For the case of the coastal lagoon city of Aveiro (Portugal), NBS scenarios of green roofs and bioswales under current and future climate conditions were assessed. The main findings of this study show that green roofs scenarios would save 32 % of the flood damages to buildings and infrastructures every year, while bioswales help save only 0.1 %. Moreover, green roofs implementation provides larger benefits in the future climate scenario (representative concentration pathway – RCP – 4.5). The findings confirm the extent to which knowledge on NBS benefits and costs is partial and uncertain, thus requiring constant progress through biophysical-economic assessment to support an evolutive decision making process in climate adaptation planning.

Fast simulation and prediction of urban pluvial floods using a deep convolutional neural network model

Urban pluvial floods induced by rainstorms can cause severe losses to human lives and property. Fast and accurate simulation and prediction of urban pluvial flood are of significance for disaster prevention and mitigation. However, physics-based models still experience excessive computational time when used for flood simulation and prediction. In this study, we explore a deep learning (DL) approach employing convolutional neural networks (CNN) as an alternative model to achieve fast prediction of urban floods. We first cluster a rainstorm-inundation database generated using a physics-based coupled model, then we develop a CNN model for predicting the spatial and temporal evolution of rainstorm inundation, and finally we compare the effectiveness of the CNN model with that of the coupled model and three other classical machine learning (ML) models. The results show that: 1) The inundation water depths predicted using the CNN model are close to those predicted using the coupled model, and the average PCC, MAE and RMSE metrics under a test rainstorm reach 0.983, 0.020 m and 0.086 m, respectively. 2) The CNN model reproduces well the trend of water depth in each model grid cell over time, especially for a heavily inundated grid. 3) The predicted effectiveness of the CNN model outperforms an extreme gradient boosting (XGBoost) model, followed by a multi-objective random forest (MORF) model and K-nearest neighbor (KNN) model. 4) The computational speed of the CNN model is extremely fast. The model can simulate inundation water depth with a spatial resolution of 8 m by 8 m (about 74 km2) and a temporal resolution of 30 min for a 6-hour lead time within 12 s, which is 600 times faster than that of a coupled model. We confirm that the CNN model with clustering method is a powerful surrogate model for fast simulation of urban pluvial flood, providing an important reference for the use of DL in early warning and mitigation in relation to urban flood disasters.

A Study on the Accessibility of the Emergency Medical Services for Urban Kindergartens and Nursing Homes Based on Urban Pluvial Flooding Scenarios

Vulnerable groups such as children and the elderly are the focus of emergency medical rescue during urban pluvial floods. Taking the Erqi District of Zhengzhou City as an example, the SCS-CN model and Chicago rainfall model are used to simulate pluvial flooding based on the comprehensive consideration of urban rainfall, runoff, topography, and drainage. Additionally, the accessibility of emergency medical aid for kindergartens and nursing homes is evaluated in the Erqi District of Zhengzhou under different pluvial flooding scenarios using GIS network analysis technology. The results showed that the number of kindergartens and nursing homes without timely access to emergency medical rescue increased with the increase in precipitation return periods. Under the 500-year and 1000-year pluvial flooding scenarios, kindergartens and nursing homes that can obtain emergency medical rescue had delayed response times. Furthermore, with the increase in the precipitation return periods, both the number and delay time of kindergartens and nursing homes significantly increased. The accessibility and delay time of emergency medical rescue in kindergartens and nursing homes were determined by the intensity of pluvial flooding (including inundation area and depth), road traffic conditions, and the number and location of medical institutions, nursing homes, and kindergartens. The research results can provide a scientific basis for improving the refinement level of urban flood disaster management and emergency response services.

Deep reinforcement learning for optimal rescue path planning in uncertain and complex urban pluvial flood scenarios

An urban pluvial flood is a devastating, costly natural disaster requiring effective rescue path planning to mitigate the loss of lives and property. The inherent uncertainty and complexity of the risks associated with urban flooding limit the ability to plan optimal rescue paths that prioritize both timeliness and safety. This study addresses the challenge by proposing an innovative assessment methodology to output risk values and probability representing safety and timeliness in each passable area while simulating real-world flood scenarios. Furthermore, the paper develops a pioneering path-planning algorithm based on deep reinforcement learning, incorporating improved stochastic reward exploitation and heterogeneous reward exploration mechanisms to function in a simulation rescue path-planning scenario with uncertainty and complexity. According to the findings, the proposed algorithm outperforms current state-of-the-art algorithms in converging to the optimal path, fully sampling, and running efficiency. The study contributes to theoretical progress on urban pluvial flood rescue, deep reinforcement learning, risk assessment, and decision intelligence while offering practical implications for smart cities, emergency management, and optimizing real-world problems by employing artificial intelligence.

Influence of Terrain Factors on Urban Pluvial Flooding Characteristics: A Case Study of a Small Watershed in Guangzhou, China

Urban roads in China, particularly low-lying areas such as underpasses, tunnels, and culverts, are highly vulnerable to the dangers of urban pluvial flooding. We used spatial interpolation methods and limited measured data to assign elevation values to the road surface. The road network was divided into tiny squares, enabling us to calculate each square’s elevation, slope, and curvature. Statistical analysis was then employed to evaluate the impact of terrain on flood characteristics in urban road systems. Our analysis reveals a strong spatial correspondence between the distribution of flood-prone points and the curvature parameters of the terrain. The spatial coincidence rate can reach 100% when an appropriate sampling scale is chosen. The presence of depressions is necessary but insufficient for forming flood-prone points. In lowland/gentle slope (LL/GS) areas with higher drainage pressure, we observe a significant negative correlation between flood-prone points and terrain curvature (Spearman’s r = 0.205, p < 0.01). However, in highland/steep slope (HL/SS) areas, we find no significant correlation between them. Notably, terrain matters, but effective drainage is more influential in flood-prone areas. The maximum flood depth (MFD), submerged area, and ponding volume during urban pluvial flooding are constrained by depression topography, while the characteristics of the upstream catchment area also play a role in determining the MFD and flood peak lag time(FPLT). Larger upstream catchment areas and longer flow paths normally result in greater MFD and longer emergency response times/FPLT. Additionally, a higher flow path gradient will directly contribute to an increased flood risk (greater MFD and shorter FPLT). These findings have important implications for flood risk identification and the development of effective flood mitigation strategies.

Modelling urban stormwater management changes using SWMM and convection-permitting climate simulations in cold areas

Urbanization coupled with climate change is expected to put pressure on the urban stormwater network. To predict and mitigate the effects of these trends, accurate modeling of urban stormwater changes is required at scales and resolutions meaningful to stormwater management. Although numerous studies have analyzed the effect of climate change on urban flooding risk using the event-based approach, none have incorporated the continuous modelling approach to investigate the whole spectrum of changes in an urban catchment. This study analyzes seasonal changes in future urban hydrological behavior using a mini-ensemble of six state-of-the-art climate model projections and a calibrated hydrological-hydraulic Storm Water Management Model (SWMM). The modelling results show future changes in seasonal and monthly hydrological behavior. The notable winter warming is the major driver in the future snow processes, resulting in considerably less snow days and an increase in the flow events frequency during the winter months. The modelling results also suggest an increase in the annual maximum hourly flow in all seasons, with the clearest trend modelled in winter. Monthly average runoff during the cold period is modelled to increase, while no clear trends are detected for the rest of the year. There is a clear added benefit in using convection-permitting regional climate models throughout the year. Overall, the climate change mitigation and adaption strategies in urban catchments should focus more on the whole spectrum of changes rather than only on urban pluvial flooding risk. These findings call for a transition from traditional to a more advanced stormwater management.

Rainfall-driven machine learning models for accurate flood inundation mapping in Karachi, Pakistan

Urban pluvial flooding (UPF) has emerged as a serious natural hazard, especially in recent years. Previous research on UPF prediction has mainly focused on hydrological models, which required a large amount of data. However, a data-driven method can significantly reduce the computational cost by using rainfall amounts to predict pluvial flooding. Intensity-duration-frequency (IDF) curves using the Gumbel method can provide a better interpretation of the correlation between rainfall intensity, duration, and probability of occurrence of a given rainfall amount. In this study, machine learning models (ML) for rainfall amounts were used to identify flood points in a case study conducted in Karachi, Pakistan. Thirteen inundation factors were used for the ML models, including a new factor, curve number. Ten ML models were applied first on training and then on validation data, yielding the inundation points. The training and validation process of the model included 384 flood points. Several statistics were used to verify the performance and accuracy of the model. We found that the Light Gradient Boost Machine and Random Forest Classifier models were the most accurate in training and validating the model, while the Decision Tree and K-Nearest Neighbor models were the least accurate in training and validating the model. The study provides valuable information for decision makers to protect communities from flood hazards by incorporating the likely intensity and duration of rainfall events and carefully selecting influencing factors into flood event prediction models.

Water Sensitive Urban Design (WSUD) Spatial Prioritisation through Global Sensitivity Analysis for Effective Urban Pluvial Flood Mitigation

Water Sensitive Urban Design (WSUD) has attracted growing attention as a sustainable approach for mitigating pluvial flooding (also known as flash flooding), which is expected to increase in frequency and intensity under the impacts of climate change and urbanisation. However, spatial planning of WSUD is not an easy task, not only due to the complex urban environment, but also the fact that not all locations in the catchment are equally effective for flood mitigation. In this study, we developed a new WSUD spatial prioritisation framework that applies global sensitivity analysis (GSA) to identify priority subcatchments where WSUD implementation will be most effective for flood mitigation. For the first time, the complex impact of WSUD locations on catchment flood volume can be assessed, and the GSA in hydrological modelling is adopted for applications in WSUD spatial planning. The framework uses a spatial WSUD planning model, the Urban Biophysical Environments and Technologies Simulator (UrbanBEATS), to generate a grid-based spatial representation of catchment, and an urban drainage model, the U.S. EPA Storm Water Management Model (SWMM), to simulate catchment flooding. The effective imperviousness of all subcatchments was varied simultaneously in the GSA to mimic the effect of WSUD implementation and future developments. Priority subcatchments were identified based on their influence on catchment flooding computed through the GSA. The method was tested for an urbanised catchment in Sydney, Australia. We found that high priority subcatchments were clustering in the upstream and midstream of the main drainage network, with a few distributed close to the catchment outlets. Rainfall frequency, subcatchment characteristics, and pipe network configuration were found to be important factors determining the influence of changes in different subcatchments on catchment flooding. The effectiveness of the framework in identifying influential subcatchments was validated by comparing the effect of removing 6% of the Sydney catchment's effective impervious area under four WSUD spatial distribution scenarios. Our results showed that WSUD implementation in high priority subcatchments consistently achieved the largest flood volume reduction (3.5-31.3% for 1% AEP to 50% AEP storms), followed by medium priority subcatchments (3.1-21.3%) and catchment-wide implementation (2.9-22.1%) under most design storms. Overall, we have demonstrated that the proposed method can be useful for maximising WSUD flood mitigation potential through identifying and targeting the most effective locations.

Transferability of data-driven models to predict urban pluvial flood water depth in Berlin, Germany

Abstract. Data-driven models have been recently suggested to surrogate computationally expensive hydrodynamic models to map flood hazards. However, most studies focused on developing models for the same area or the same precipitation event. It is thus not obvious how transferable the models are in space. This study evaluates the performance of a convolutional neural network (CNN) based on the U-Net architecture and the random forest (RF) algorithm to predict flood water depth, the models' transferability in space and performance improvement using transfer learning techniques. We used three study areas in Berlin to train, validate and test the models. The results showed that (1) the RF models outperformed the CNN models for predictions within the training domain, presumable at the cost of overfitting; (2) the CNN models had significantly higher potential than the RF models to generalize beyond the training domain; and (3) the CNN models could better benefit from transfer learning technique to boost their performance outside training domains than RF models.