Urban Flood Risk and Resilience: How Can We Protect Our Cities from Flooding?

Abstract

This article draws on over 40 years of the author’s experience with hydroinformatics tools for water and sustainability challenges, including flooding. It aims to spark discussion on urban flood risk and resilience rather than provide a literature review or definitive answers. Assessing urban flood risk and resilience is complex due to the spatio-temporal nature of rainfall, urban landscape features (e.g., buildings, roads, bridges and underpasses) and the interaction between aboveground and underground drainage systems. Flood simulation methods have evolved to analyse flood mitigation schemes, damage evaluation, flood risk mapping and green infrastructure impacts. Advances in terrain mapping technologies have improved flood analyses. Despite investments in flood management infrastructure, a residual flood risk remains, necessitating an understanding of the recovery and return to normality post-flood. Both risk and resilience approaches are essential for urban flood planning and management. Future challenges and opportunities include both technological and governance solutions, with artificial intelligence advancements offering the potential for digital twins to better protect urban communities and enhance the recovery from flood disasters.

1. Introduction

Flooding can result from various causes, including intense or prolonged precipitation, snowmelt, rain on snow or ice, flow obstructions, dam failures, storm surges and flooding due to insufficient urban drainage infrastructure systems in urban areas. Combined with factors such as unpredictable weather, limitations of infrastructure, growing urbanisation and geographical and topological factors, the overall result is that while we can influence those causes and significantly reduce the risk and impact of flooding through various measures, achieving absolute protection is not feasible. However, by understanding the risk and building resilience, communities can better protect themselves from potential disasters and recover more quickly when they occur.

Due to the spatio-temporal nature of rainfall over urban areas, landscape features (such as buildings, roads, bridges and underpasses) and the interaction between aboveground and underground drainage systems, the assessment of urban flood risk and resilience is a complex and somewhat intractable task. Therefore, we often rely on models to help us better understand urban flooding.

2. Urban Flood Modelling

Modelling urban flooding has traditionally been accomplished by treating the urban surface (major system) as a 1D or 2D system, i.e., as a 1D network of open channels and ponds or as a true 2D flow domain. A significant step forward in modelling urban flooding was the introduction of the dual-drainage concept, where the urban surface flow is combined with the sewer flow. This approach allows for dynamic flood simulation that can be used for analysing flood mitigation plans, damage evaluation, flood risk mapping, green infrastructure impact assessments, etc. Modern terrain mapping technologies, such as synthetic aperture radar, aerial digital photogrammetry and light direction and ranging (LiDAR), offer significant improvements in data availability and accuracy for flood modelling applications in urban areas. High-resolution digital elevation models (DEMs), which include buildings, trees and other surface features contributing to flood risk, have become more widely available.

Modelling tools range from highly accurate ones that use implicit finite-difference schemes and a double-sweep procedure to solve the full Saint-Venant equations [1] to simplified but computationally much more efficient models, such as those based on cellular automata [2,3], to provide flood simulation over a DEM. Both approaches require detailed rainfall and topographical data that are often not available. Furthermore, there are seldom data on flood extents and runoff in urban areas during flooding episodes to establish the ground truth for model calibration and validation [4]. Therefore, urban flood modelling should always be coupled with an uncertainty analysis as input data; model parameters, the model structure and observational data are the main sources of potentially significant uncertainty [5].

In addition to the complex characteristics of urban flooding and the complexities of its modelling, flood impacts are expected to increase drastically in the future due to climate change causing an increased frequency and intensity of storms and sea levels rising [6]. Therefore, even with advanced modelling, forecasting and a great investment in infrastructure, the risk of unexpected flooding still exists and is growing with time!

3. Flood Risk Management

Flood risk reduction has become increasingly common for managing flooding worldwide, following the realisation that even with flood control infrastructure implemented, disasters continue to occur. The risk-based approach to management traditionally focuses on identifying, assessing and mitigating potential threats before they occur [7]. Risk-based flood management relies on calculating the probabilities and consequences of specific flood events, agreeing upon the acceptable level of risk and building infrastructure to limit that risk. This involves a quantitative evaluation of the risks, costs and benefits of flood management activities. While knowing the risk is important for managing it, such as targeting resource allocation and improved decision-making, this approach has its disadvantages.

One of the key challenges is how to quantify risk. The most often used metric is the expected value of losses (often expressed as expected annual flood damage). This is expressed as the probabilities of a full range of flood events multiplied by the loss caused by them and then summed up. However, the expected-value operation equates flood events of high probability and low losses with those of a low probability and high losses [8]. For example, a flood event with a return period of 1 year (probability p₁ = 1 in Equation (1)) causing a GBP 1 million loss (cost c₁ = 1 × 10⁶ in Equation (1)) and another with a return period of 1000 years (probability p₂ = 0.001) causing a GBP 1 billion loss (cost c₂ = 1 × 10⁹) yield the same contribution to the overall expected value of flood loss:

$$E\left[loss\right]=p_1\times c_1+p_2\times c_2=1\times {10}^6+1\times {10}^6$$

(1)

To a decision maker, the two events cannot be equivalent as the potential consequences are so different, and even though there is a low probability that a catastrophic loss will occur (a ‘black swan’ event), the potential for it to happen must influence the decision maker. An additional problem with flood risk management is that flood risk continues to increase with climate change (i.e., a higher probability) and socio-economic development (i.e., higher losses).

4. Is Resilience the Answer?

While the focus of risk analysis is on the reduction of vulnerabilities before abnormal but expected disturbances occur, resilience, on the other hand, suggests the ability of something or someone to recover and return to normality after confronting often unexpected shocks. Therefore, it is commonly understood as a path for managing disruptions and adapting to changing conditions [9]. This seems like a different approach that explicitly acknowledges that unexpected flooding can and will happen and describes the ability for adaptation and recovery [10]. However, there is no single accepted definition of resilience or at least no agreed mathematical expression describing it [11].

To understand resilience, its four properties (or the 4Rs) can be considered (Figure 1): Robustness (considers reliability and the ability to absorb and withstand disturbances), Redundancy (involves having additional capacity and backup systems to enable core functionality during disturbances), Resourcefulness (measures the ability to mobilise resources in the case of disturbances) and Rapidity (measures the ability to quickly restore functionality). Quantifying these 4Rs into performance indicators leads to a better understanding of resilience to flooding.

Considering the different nature of the two approaches, risk and resilience analyses are complementary and should both be considered in the case of urban flood planning and management.

5. Future Challenges and Opportunities

Technological and governance solutions. Urban flood risk will increase with climate change, population growth and urbanisation. Flood management, including disaster prevention, mitigation, preparedness and vulnerability reduction, requires both technological and governance solutions to deal with this increased risk. Flood modelling tools, early warning systems and recovery decision-support systems are some examples of the technological solutions required to manage flood risks. On the other hand, flood risk governance needs to improve and involve the complex institutional arrangements that shape the behaviour of governing bodies (from local to country governments) and stakeholders concerned with flood risk management [12]. While technological solutions provide insights necessary to manage urban flood risks and improve resilience, moving from insights to action requires a coordinated strategy that aligns with other pressing urban challenges, such as the energy transition. Fostering collaboration between different sectors, including water and wastewater management, urban planning, energy and transportation, could create synergies and holistic solutions to many of the challenges, including urban flooding. Planning for the synergistic capacity expansion of urban water, sewer and road networks in Montreal (Canada) demonstrates that significant savings can be achieved when compared to uncoordinated master planning for each of the sectors [13].

Artificial intelligence. The proliferation of remotely sensed information from space- and ground-based sensors with increasing capabilities of spatial, temporal and spatial resolution brings new challenges and opportunities for flood risk management, including the ever-increasing use of artificial intelligence (AI). Information and knowledge gained from data will allow a more efficient and reliable monitoring, modelling and management of floods at global, regional and local scales. In addition to standard hydrodynamic models, data-driven modelling approaches, particularly those based on AI methods, have found their way into flood risk management research [14]. In practice, AI is changing our society, including banking, transport, entertainment and tourism, due to the big data revolution and Internet of Things (IoT) advances [15]. It is thus logical to expect a significant impact of AI on flood management practices in the future. There is also an increasingly urgent need to accelerate AI development and adoption, to provide tools for flood forecasting (e.g., digital twins), impact assessment and improved societal resilience.

Hybrid and ensemble modelling. Ensemble and hybrid AI models are the more advanced types used for flood risk management. Given their proven success in providing a higher accuracy and outperforming most of the conventional single AI models, they require further investigation and application in the field of flood management. Another promising avenue for improving AI models is blending data-driven and process-based models, where the advantages of individual models are taken fully within the hybrid [16].

Interconnected critical infrastructure. Flood management infrastructure systems are part of the much more complex critical infrastructure in urban areas. Critical infrastructure systems, such as energy, telecommunication and healthcare systems, face increasing risks due to flood hazards. The interconnectedness of critical infrastructure can lead to indirect effects of urban flooding that exacerbate the overall impact on city functionality. Additionally, flood risk disproportionately affects lower-income and minority communities, highlighting significant inequities in vulnerability and access to resources for disaster preparedness and recovery [17]. The resilience management of these systems is becoming increasingly challenging due to their high complexity and the existence of interdependencies between them. To effectively plan resilience enhancement measures given resource constraints, prioritization approaches at the system and an infrastructure asset level are thus needed.

Spatio-temporal nature. As flooding occurs in space and time, new approaches will need to be adapted to provide spatially accurate and timely model outputs for highly complex spatio-temporal dynamic systems. The field of Cellular Automata has already been explored for flood management; however, there is more that can be done to support decision-making at various spatial and temporal scales.

High dimensionality. Flood processes and defence systems can be complex, with a huge number of potential variables that interact with each other. Understanding the relationships among these multiple variables at fine resolutions of space and time requires flood data that are highly dimensional. Including all the relevant variables poses challenges to modelling algorithms, which results in the curse of dimensionality. Dimension reduction methods are required to be incorporated with simulation methods to better model flood processes.

Rare events. While climate change is advancing, there is an increase in the likelihood of flooding of all magnitudes [18]. However, we normally require large datasets of homogenous and error-free data. The requirement for such large volumes of data cannot always be met, and methods that can work with small datasets should be explored further. Even machine learning methods learn patterns in existing data locally, as they are not good at extrapolating. A different approach is required for providing predictions outside of the known conditions.

Multi-source multi-resolution data. Information about flooding can be collected via different data sources, which are provided at varying spatio-temporal scales. However, only if they are analysed together can the full picture of flood risks be obtained [19]. These datasets may exhibit varying characteristics, such as sampling rate, accuracy, and uncertainty. Methods that can deal better with a variety of sources and resolutions will be required more in the future.

Lack of ground truth data. There is a general paucity of flood data to establish the ground truth. This is because of limited observation networks or the inability to access the necessary data. Given that the manual collection of data is typically costly, the use of AI methods may be advantageous in cases with very few data points. For instance, a possible solution for classifying unlabelled samples is to first cluster them using unsupervised machine learning and then label representative samples from each cluster by domain specialists, which could be potentially useful for flood-related applications.

Deep uncertainty. Climate change leads to more extreme weather events, like heavy rainfall and storms that are less predictable. This creates deep uncertainty about when and where they will occur. To manage these impacts, advanced modelling techniques need to be combined with adaptive decision-making strategies to improve resilience and reduce risks of urban flooding.

Moonshot. Finally, while the moonshot for urban flood analysis may be seen as a timely and accurate forecast of the spatio-temporal dynamics of rainfall combined with a digital twin for flooding in urban areas, it is not yet immediately attainable. However, the advent of artificial intelligence models with improvements in collecting sensor data from various sources could lead to a global urban flood hub providing forecasting tools, similar to the one developed by Google Research for river flows [20].