3. Latest Advances and Challenges in Extreme Flood 3D Simulation

Extremes, by definition, are very rare events. This makes them events that are difficult to anticipate or protect oneself from. They impact extensively, often over spatial and societal scales covering the entire countries, necessitating interventions that are instituted by governments responsible for ensuring society remains safe and its institutions operational. Flood extremes are a special type of naturally occurring extreme events that are modulated by the climate and also by the landscape they occur in. Wildfires are also modulated by both the climate and the landscape but evolve with greater randomness given their dynamic interaction with the elements over time. The occurrence of either floods or wildfires is a random extreme event characterisable by a suitable extreme value probability distribution. The evolution of such an extreme event, however, is deterministic. While both floods and wildfires are modulated by the same factors (climate and landscape), much of the discussion that follows focuses on floods, their prediction and the various factors that affect the extent of damages they result in.

Floods are the biggest and most severe natural disaster we face each year (Doocy et al., 2013). Apart from the loss of lives, floods disrupt daily activities, stop productivity and create a sense of helplessness that is socially and mentally disheartening. Technological advances have led to a doubling of life expectancy and per capita agricultural productivity in the last 100 years, yet flood deaths continue to rise (Doocy et al., 2013). The number of people affected will double within 20 years, due to rising populations, their concentration around floodplains and intensifying storm extremes resulting from planetary warming (Gassert et al., 2013).

Flood disasters are inevitably a result of an extreme storm event. However, if adequate flood protection infrastructure (such as reservoirs or levees) is in place, communities can remain largely unaffected. Enhanced atmospheric greenhouse gas concentrations and the increase in the moisture-holding capacity of a warmer atmosphere have resulted in a worldwide intensification of extreme storm events. As a result, extreme floods, especially over smaller catchments, are increasing and are expected to worsen with time (Sharma et al., 2018). Hence, the probability of failure for existing flood protection infrastructure is greater than when first designed. A consequence of this is an increasing frequency of flood disasters resulting from failure of existing flood infrastructure, such as failure of a dam upstream of a city, or rupture of a levee designed to keep flood waters at bay.

This chapter presents the rationale behind predicting flood impact in three different settings. First, we outline the factors that modulate a flood extreme, as well as the basis that is used to predict floods in advance. Next, we discuss a flood event that results in failure of the flood infrastructure, amplifying the damage caused, a scenario that is increasing in likelihood as a result of global warming. This is followed by a discussion of ‘compound flood extremes’, i.e. a flood event that is a result of not one but multiple extreme events occurring simultaneously. Such an event is difficult to plan for and is again increasing in likelihood as a result of global warming. We conclude by discussing how flood visualisation can assist with both real-time assistance and longer-term planning, along with the challenges that such visualisation systems typically face.

So, how are flood warnings issued, and what are the key factors that introduce uncertainty into the quality of predictions made? To answer this question, one must first list the key inputs needed to quantify the evolution of a flood over a catchment. As mentioned before, the two main factors that modulate floods and make them different from one location to another are the contributing area (or landscape) and the climatic event that results in the flood. While the contributing area can be defined with ease given a topographic map and a catchment outlet, there exists uncertainty in how the climate interacts with this area to create the flood wave. A significantly greater uncertainty exists in the climate, which changes for the same catchment from one event to the other.

A key uncertainty in flood prediction is the specification of the model that could accept the climate as an input and estimate the flood magnitude as a function of time at the location of interest. The translation of the rainfall sequence to the flood is a deterministic process. Yet, for a model to be accurate, there needs to exist measured streamflow at the location of interest to ensure model parameters can be calibrated and the model thereby used into the future. When flood data is not available—as is the case in the vast majority of catchments, especially in low- and middle-income countries or in remote settings worldwide—the uncertainty in the flood prediction model becomes significant. There is extensive evidence that although the total population exposure to natural hazards in low- and middle-income countries is similar to high-income countries, then fatalities are far higher in the former (Strömberg, 2007; Lindersson et al., 2023). Flooding constrains population development opportunities, and 89% of the world’s flood-exposed people live in low- and middle-income countries (Lindersson et al., 2023).

Another key uncertainty in flood prediction lies in the specification of the climate anomaly that translates into the flood. While the density of precipitation gauges is markedly greater than the corresponding streamflow measurements, these represent point precipitation and do not capture the significant variability across a catchment. Furthermore, precipitation gauges are few in remote areas, creating significant uncertainties in the depth and spatial distribution of recorded precipitation. Given floods are extreme events that are a result of extreme climatic anomalies, such observational uncertainties lead to large errors in ensuing flood warnings and predictions. When precipitation is not measured but predicted ahead of time, such uncertainties increase manifold (Ehrendorfer, 1997).

To address such uncertainties, modellers use surrogate data, representing both streamflow (to use directly for forecasting or to use as the basis for calibration) and the rainfall storm event (as an interpolated event based on point observations or using a weather forecasting approach that is based on satellite observations of storm trajectories and the dynamics of storm creation). An example of a surrogate-streamflow model developed using publicly available satellite remote sensing data is presented in Fig. 3.1, with procedural details described by Yoon et al. (2022, 2023). No ground streamflow data was used in predicting the shown flood anomaly. Instead, the researchers used satellite retrievals that served as a surrogate to the actual streamflow over time. The implications of that for ungauged catchments, especially in remote settings, are significant.

A double-line graph of Q and S R versus months. Both lines first remain flat, then rise or decline to reach a peak or a trough, then rise or fall to become flat again. 6 inset photos are of flooded areas with submerged and tilted vehicles. A few buildings and signage boards are partially submerged.

However, observational-based predictions, such as those using remotely sensed data, generally do not provide enough time for action by communities and emergency managers. Thus, accurate forecasts of extremes are vital to increase the lead time available for preparation. Numerical weather prediction (NWP) models are vital components of flood warning systems. Despite the substantial advances in NWP skill over the last 40 years, which have assisted with improving event preparation, forecasts of precipitation are generally biased. Biases in precipitation are one of the most challenging aspects of NWP models to correct due to the presence of zero amounts (dry hours/days) as well as the highly skewed distribution of rainfall, with many small rainfall amounts and few very large or extreme rainfall events, which are the events that lead to floods. These errors in NWP rainfall simulations come from improperly resolved topography, temperature biases leading to convection biases and biases in the location of storm tracks. These issues cause underestimated high rainfall intensities, too much drizzle (Huang & Luo, 2017) and under-dispersed ensemble predictions (uncertainty ranges that are too confident). Most research into NWP rainfall correction has used quantile mapping (Hamill & Scheuerer, 2018). Quantile mapping corrects the modelled rainfall to match the observed rainfall across the full range of rainfall values from small to large, but it cannot correct incorrect sequencing of rainfall, e.g. how often a dry day is followed by a wet day, known as persistence biases. Bennett et al. (2014) addressed this problem by incorporating a Schaake shuffle, which retrospectively addresses the lack of persistence. Recently, Jiang and Johnson (2023) proposed a new method for post-processing NWP forecasts using continuous wavelet transforms (CWT), coupled with quantile mapping. After decomposing the forecasts and observations using CWT, the amplitudes of the decomposed series are bias-corrected with quantile mapping, while the phases are randomised to correct timing errors. Spatial structure in the forecasts is improved by adopting the same phase randomisation at all gauge locations being corrected. A discrete wavelet-based approach combined with quantile mapping was used to bias-correct streamflow forecasts by Johnson (2023). It was found that calibrating the variance corrections over the full historical period provided marginally better forecast skill than a time-window-based approach (Fig. 3.2).

3 spatial distribution graphs of latitude versus longitude with gradient scales for rain. a. It plots decreasing trend data points with increasing rainfall values enclosed in an oval with a gauge point marked. b and c. Both plot decreasing trend ovals, with rainfall values on the rest of the grid.

Pappenberger et al. (2015) quantified the monetary value of the European Flood Awareness System (EFAS) at 400:1 in terms of the potentially avoided flood damages across Europe compared to the cost of developing and operating EFAS. However, of the total Australian spending on disaster relief, 97% goes towards post-disaster recovery and only 3% to improved preparation and resilience (Productivity Commission, 2014), so there is a clear need to improve the focus on preparation for events. There have been calls for almost 20 years on the need for people-centred approaches in the design and operation of early warning systems (Basher, 2006), with the Sendai Framework (United Nations Office for Disaster Risk Reduction, 2015) continuing to emphasise the importance of people-centred preventative approaches to disaster risk. Traditionally, early warning systems have been viewed as a warning chain (Basher, 2006), a linear set of actions from observation to warning generation to dissemination. More recently, there has been recognition that instead of focusing on dissemination as the ‘last mile’ in the warning system, the system should be designed based on who the community is, their degree of vulnerability and their capacity (Marchezini, 2020). New and creative visualisation methods are thus urgently required to allow communities to better understand the threats for natural hazards and co-design effective warning systems to improve preparedness.

Unlike the purely weather-driven flood events discussed in the previous section, a significantly worse disaster can occur if the flood infrastructure that exists to protect the community from flooding fails. An example of such a failure occurred in September 2023 in Derna, Libya (Saeed et al., 2023). This event was brought upon by anomalously warmer oceans and presence of high degrees of atmospheric moisture across the Mediterranean Sea and culminated in the collapse of two flood protection dams upstream of Derna. It resulted in a confirmed 4000 deaths with 10,000 missing, in a city with a total population of 90,000 residents.

Disasters such as that in Derna can be avoided if the flood protection infrastructure in place (here, the two dams that collapsed upstream of the city) operates as designed. In the design of such infrastructure, engineers typically derive an upper limit that a flood peak may assume, which is commonly referred to as the ‘probable maximum flood’ (PMF). The PMF is derived using an equivalent upper limit of the causative storm, referred to as the ‘probable maximum precipitation’ (PMP). Neither the PMF nor the PMP are absolute maximums—their estimation uses observations making the estimate a random variable with an extremely low probability of exceedance. Large dams around the world are often required to withstand a PMF, especially if communities live downstream.

The PMP is estimated using guidelines established by the World Meteorological Organization (WMO) and is proportional to the atmospheric moisture-carrying capacity of a warming atmosphere (2009). Figure 3.3 presents results from a recent study assessing change in the PMF over time, using both observational records on a surrogate of atmospheric moisture and climate model projections of the same surrogate up to the year 2100 (Visser et al., 2022). The change projected in the figure is surprisingly consistent across the landmass of Australia. Given the PMP represents an upper limit, it is prudent to assume that its projection for the year 2100 uses the more extreme SSP5–8.5 pathway, based on which one can expect the PMP will increase by 38% across existing reservoirs in Australia.

A scatterplot of the persisting 24-hour 1000-hectopascal dewpoint versus years from 1950 to 2100. It plots a gradually rising line for historical simulation, starting at 22 and continuing until 2024. Then, it plots 2 rising and 2 falling lines from the same point with a vertical dashed line at 2024.

The projections of the increase in PMP by 38% in Fig. 3.3 implies that (a) PMFs will be increasing at least at the same rate as the PMP, (b) this increase will be remarkably consistent across climate zones and driven solely by a warming atmosphere and (c) the risk of failure for existing dams is now much greater than what it was when the dam was first constructed—with this risk increasing year after year as warming continues. While a PMF is by definition an extremely rare flood event that would never typically be recorded during the life span of a reservoir, global warming has made the PMF more likely. This means that disasters, such as happened at Derna, are becoming much more likely as we progress into this century.

As outlined above, communities are safeguarded from the impact of floods by putting in place flood protection infrastructure, such as levees, dams or other barriers that stop or restrict the extent of damage that would otherwise occur. Such infrastructure is either designed on the basis of risk or designed to fail on average once in a certain number of years (Razavi et al., 2020). Typically, such design assumes the flood is a result of an extreme storm event that has the same risk of occurrence as the ensuing flood.

Of late, concerns have been growing that flood failure events are not a result of a single causative factor (i.e. an extreme storm) but an outcome that is due to two or more factors occurring simultaneously. An example of such an event may be two storm systems that occur simultaneously on two branches of a river system. This may result in extreme flood damage at their confluence, or the flooding of a coastal city that could be aggravated by extreme winds, resulting in a storm surge. Leonard et al. (2014) have presented a framework for assessing such compound events.

Moreover, there have been concerns that the risk of compound events is on the rise as a result of climate change. Two recent studies by Gu et al. (2022) and AghaKouchak et al. (2020) use observed data to establish that the risk of coastal floods, and the risk of high temperatures coupled with flood conditions, is rising across the world. How these pan out into the future is a question left largely untouched by the research community, although tools exist for simulating future compound extremes coupled to climate model simulations. These are downscaled to higher resolutions after correcting the simulations for systematic biases across all boundaries of the domain being downscaled (Kim et al., 2023). More work is needed on this front to better assess how society can be safeguarded from such future extremes as risks increase over time.

In preceding sections, we outlined three scenarios whereby flood disasters can impact communities downstream. The first scenario is the occurrence of a flood following an extreme storm event. The second scenario is a flood causing existing flood protection infrastructure to fail and result in magnified damage on communities downstream. The third scenario articulated the complexity in the flood generation mechanism, noting that floods may occur through a combination of causative events, such as a storm coupled with a coastal storm surge. All these scenarios can be represented through mathematical models of the climate, flood and landscape. Other data streams, such as satellite data or real-time rainfall and streamflow data, are also vital to understand the hazards. However, without effective visualisation of the data streams or model simulations, there is limited benefit in terms of facilitating preparedness. In this section, we review traditional and new developments in visualisation that can be leveraged for flood management and preparation.

Traditionally, flood visualisation has focused on mapping of flood extent, velocity and depth derived from hydrodynamic models. An excellent review on modelling and visualising flood inundation is provided by Teng et al. (2017). Flood mapping entails tracking the evolution of a flood wave in space and time as it follows the path of gravity and disperses over the downstream terrain. Such inundation modelling forms the basis for hazard mapping (a function of the maximum inundation depth), erosion modelling (based on flow velocity), flood insurance valuations (a function of the frequency and hazard) and a range of other actions. Flood inundation modelling utilises hydrodynamic modelling as the basis for tracking the evolution of the flood wave and offers a range of simplifications to reduce computational effort and produce inundation maps to varying degrees of accuracy. The resulting maps from these analyses are generally presented as part of a flood study, are static and represent averaged long-term risk of floods at any particular location.

Increasingly though, a range of new approaches are being investigated to better communicate the risk and impacts of flood events. Participatory practices for flood mapping are increasingly being considered (e.g. Auliagisni et al., 2022). For example, Disaster Relief Australia (2023) has recently implemented ‘Big Map Workshops’ to increase community resilience to flood events in order to improve post-event recovery for both floods and fires. Instead of the flood map being part of a report, it is a room-sized visualisation of the community and its risk. Such workshops allow a range of stakeholders to better understand the dynamic nature of such natural hazards and thereby prepare effectively as a community for the next event.

Similarly, Li et al. (2022) describe a three-dimensional storytelling approach to visualise the impacts of infrastructure failure, such as a dam-break flood discussed earlier in this chapter. Storytelling approaches have been used more widely in communicating climate change adaptation options but are yet to be routinely used in floodplain management. Such methods provide exciting promise for helping communities to understand the dynamic nature of flood events compared to the static flood maps discussed above.

Other methods have been developed using strength-based approaches for engaging stakeholders. For example, relatively simple methods such as schematic diagrams can be used to provide real-time updates from a range of data streams and forecasts. The advantage of these schematic diagrams is that they do not rely on high spatial awareness skills for users. Yasmin et al. (2023) point to the importance of such inclusive practices in the design of communications. Similarly, Auliagisni et al. (2022) argue for the importance of engagement with a wide range of stakeholders (Bakhtiari et al., 2024), such as insurance providers and policy makers, in the use of new visualisation approaches such as digital twins or augmented and virtual reality. One of the major open questions in visualisation approaches is how to usefully communicate uncertainty in predictions and forecasts. This is vital for improving the ability of emergency services to respond to increasing threats from compound events, which will stretch already limited resources.

The World Resources Institute publishes current and projected flood damages on a country-wide basis (Gassert et al., 2013). According to this, Australia incurs an annual flood-related damage of US$18.7 billion, a number that will increase to US$29.9 billion by 2030. This increase is small when compared to Australia’s neighbours in South-East Asia, with three-fold (Malaysia) to 15-fold (Myanmar, Cambodia) increases predicted to severely impact resident societies and economies. Much of this damage relates to urban flooding, which is expected to increase most significantly due to intensification of storms and lack of the significant antecedent storage, which dampens flow peaks in rural settings. Much of such damage can be contained if adequate flood protection infrastructure is put in place. However, the damage increases when existing infrastructure becomes inadequate, which we have argued in this chapter. This is indeed occurring as the world records higher temperatures each year in the recent past.

We outlined three scenarios that are important to study as the world prepares for flood extremes on a different scale. A significantly increased exposure to risk exists as populations concentrate in urban centres that often lie near a major water body that serves as the source of flood water supply. The first of these scenarios represented flood events that are caused solely by extreme storms and argued that more could be done to predict such events in advance through a mixture of modelling and weather forecasting. The second and third flood scenarios referred to design extremes that represent hypothetical events used to design flood protection infrastructure that is increasingly vulnerable under condition of climate change. In all of the scenarios discussed, there is a common need of methodologies to visualise flood damage, to assist in planning, risk/hazard quantification as well as devising methods for evacuation, all serving as important legs of a comprehensive damage control strategy. Such a strategy is sorely needed as flood damages rise with time (Wasko et al., 2021), a rise that will only get worse as temperatures increase and populations converge on concentrated urban centres across the world. The capability to viscerally preview the dynamic evolution and likely extent of future extreme flooding events can help protect and save lives as well as safeguard assets and infrastructure by enabling us to build effective preparedness for a hotter future that is inevitably coming to meet us.