Modelling the effects of climate change on urban coastal-fluvial flooding

Flooding is the most frequent and hazardous of all natural disasters (ICHARM 2009), with the most extreme flooding generally occurring in estuaries as a result of a combination of high river flows and high coastal water levels. Globally, many urban areas have been developed in coastal plains along the banks of large rivers, and these are subject to flooding due to both fluvial and coastal drivers. The most severe coastal floods are those driven by a combination of high river discharges, astronomical tides, storm surges and/or waves acting simultaneously.

Extensive research exists demonstrating that flooding has increased in frequency in recent decades (Kundzewicz 2012; Hall et al. 2014), fuelling concern that climate change is influencing flood regimes. Predicted increases in sea level, rainfall and storm winds are likely to escalate the risk of flooding in the future (Purvis et al. 2008). This will have significant socio-economic consequences, compounded by the fact that, over recent decades, coastal populations have continued to grow much more rapidly than the global mean population (e.g. McGranahan et al. 2007). Currently, over 600 million people worldwide live along the coast (<10 m elevation) and it is predicted that this number will increase to more than 1 billion by 2050 (Merkens et al. 2016). Human activities in coastal regions, including land reclamation and infrastructure development, also alter the natural behaviour of the coastal zone and impact on the nature of flooding.

With projected changes in climate and predicted increases in coastal development (associated with upwardly trending coastal populations), it is envisaged that flood risk and associated costs will increase in the future. The average annual cost of flood damage in coastal cities is projected to rise globally from US$6 billion in 2005 to US$1 trillion by 2050, under present protection levels (Hallegate et al. 2013). Taking into account an infrastructure-based adaptation, the global annual flood losses are still projected to exceed US$60 billion by 2050. Consequently, an in-depth understanding of flood mechanisms and the effects of climate change will provide a significant support for the decision making in flood defence design and flood risk management.

Coastal-fluvial flooding is a complex compound event as it results from a combination of interacting drivers, such as stochastic meteorological conditions and deterministic local sea levels. Additionally, the already hydraulically complicated hydrodynamics of natural floodplains is exacerbated by complex urban developments. These generally consist of dense street networks and extensive buildings, which control the routing of water across the floodplain. Even if flood flow patterns are well understood for the current climate, flood characteristics are likely to alter in future in response to climate change, continuing urbanisation and flood defence adaptation. These changes may affect not only flood water levels and associated flood extents but also the pattern of inundation due to a shift in flood mechanisms. Understanding hydraulic conditions on existing floodplains is not trivial, and projecting into the future is extremely difficult. However, when the mechanisms of flooding are identified for a particular floodplain and potential climate-driven changes estimated, the hazard and impact of future river discharges and coastal water levels can be determined using flood inundation models. Currently, the most common models utilised in flood assessments are hydrodynamic models, which solve equations of fluid motion (Teng et al. 2017). These are particularly useful as boundary conditions can be modified to investigate inundation in response to different future scenarios.

Hydrodynamic modelling can provide an assessment of the degree of flooding and related impacts in response to particular flood mechanisms, and so be utilised to reduce and manage flood risk (Parry et al. 2009). This is essential as changes in inundation do not necessarily exhibit a linear relationship with changes in water volumes (Veijalainen et al. 2010). Accurate modelling of complex coastal-fluvial flood dynamics and interactions between multiple drivers is critical for realistic simulation of inundation; however, this is not a trivial task to accomplish. In recent years, the amount of hydrodynamic modelling of floods due to fluvial and coastal mechanisms has risen dramatically and so has the number of numerical models in various combinations of setups. Nonetheless, a model accuracy and computational cost remain the issues to be addressed. Some simple modelling approaches treat fluvial and coastal drivers separately without considering the compound effect of both signals and possible interactions between them (e.g. De Angeli et al. 2018); this may lead to a significant misrepresentation of flood depths and extends. More advanced approaches link 1-D/2-D hydraulic models with coastal models (e.g. Yin et al. 2015; Pasquier et al. 2019). However, when linking different models with different dimensions, significant numerical errors may be introduced resulting from poor conservation of mass and/or momentum (Yang et al. 2006). For these reasons, two-way dynamically linked models would be generally more accurate than the externally linked models; this, however, increases the computation cost dramatically. An alternative solution is to use one model across the fluvial and coastal domain to incorporate both mechanisms for a compound event simulation (e.g. Lopes et al. 2017; Kumbier et al. 2018). The models of this group, regardless of their mesh structure, often encounter a problem of insufficient spatial resolution. As the coastal-fluvial flooding of urban floodplains is a multi-scale problem, the accurate solution is required at various scales ranging from coastal sea or estuary scale down to a dense street network of the inundated urban area (e.g. O'Neill et al. 2018; Barnard et al. 2019). This problem of spatial resolution may be overcome by multi-scale model grid nesting which involves embedding higher-resolution grids within a lower-resolution global large-scale grid model (e.g. Nash & Hartnett 2010). Such a solution allows users to specify high resolution in a subregion of the model domain without incurring the computational expense of fine resolution over the entire domain.

The computational effort is of paramount importance in climate change studies where modelling work involves multi-decadal simulations for a range of future climate scenarios. While recent advances in computational resources through numerical domain decomposition and multi-core architecture allow us to decrease model runtime significantly, the computational effort still limits the amount of scenarios to be examined. As such, the selection of climate change projections and combination of boundary conditions is a compromise between two competing requirements: manageable computational time and informative dataset of model outputs. Recently, some efforts have been made to link statistical modelling with physical modelling to limit a number of simulation combinations; such hybrid systems show that misrepresentation of the dependence between multiple drivers may lead to an underestimation of flood risk (Serafin et al. 2019).

Considering the complexity of compound flood events due to multiple drivers that interact and may behave non-stationary in future climate, the selection of frameworks for high-resolution urban flood modelling is extremely difficult. In this study, the MSN_Flood multi-scale flood model is selected for high-resolution modelling of urban flooding; the model has been successfully used in recently completed flood research for Cork City and was found capable of resolving the complex hydrodynamics of Cork floodplains (Comer et al. 2017; Olbert et al. 2017). Through a cascade of nested models, this modelling system allows simulation of the propagation of open-sea conditions up to the tidally active river upstream as well as rural and urban floodplains. The model has flooding and drying routine and so-called moving boundary, so flooding and drying may occur both within the domain and along boundaries. As such, the model is ideally suited for flood modelling while maintaining accuracy and computational efficiency.

In this context, the primary objective of this paper is to provide a methodology for comprehensive forecasting and assessment of urban flooding in consideration of climate change. The methodology is illustrated with a case study of Cork Harbour, for which a numerical model is used to investigate (1) conditions under which coastal-fluvial flooding may occur under current and future climate, and (2) potential effects of climate change on shifts in pattern and impact of coastal and fluvial mechanisms.

In this paper, high-resolution grid scale modelling of a dense urban area under various projected climate scenarios has been conducted and various combinations of extreme floods analysed. A range of future climatic conditions for different flood drivers employed to provide a comprehensive forecast of potential future flood scenarios. The urban area of Cork City, Ireland – frequently subject to coastal and fluvial flooding – was used to investigate changes in flood mechanisms, dynamics and extents due to climate change. Extensive fluvial-coastal flooding of Cork City in November 2009, resulting in damage of over €100 million, prompted the current investigation. The development and implementation of techniques that enhance confidence in projections of change in flooding are essential to aid climate adaptation decisions and facilitate effective foreshore management.

Cork's low average elevation and extensive coastline make the city especially susceptible to coastal hazards. With sea level rise and projected climate change the flood risk is likely to increase significantly.

Cork City is situated on the mouth of the River Lee, on the south-west coast of Ireland (Figure 1). The Lee drains an area of approximately 1,253 square kilometres (km²) (Appendix A, Figure A.1) and discharges to Cork Harbour, a 350 km² tidal estuary feeding to the the Celtic Sea and wider North-East (NE) Atlantic. The river is approximately 115 km long and is sourced in the Shehy Mountains on the western border of County Cork. The Electricity Supply Board (ESB) operates two hydro-electric dams on this river, forming the Innishcarra and Carrigadrohid reservoirs, approximately 13 km and 27 km west of Cork City respectively. These provide up to 35 × 10⁶ cubic metres (m³) of storage, and controlled discharge, of floodwaters (Halcrow 2014). Discharges from these dams, combined with inflows from the downstream Shournagh, Bride and Curragheen tributaries, control the water volumes entering Cork City. Seawater intrusion from Cork Harbour also influences fluvial water levels in the River Lee, although this is constrained by the waterworks weir approximately 8 km upstream of the harbour.

Increases in mean temperatures and rising global sea levels are projected for the 21st Century by the Intergovernmental Panel on Climate Change (IPCC) Fifth Assessment Report (IPCC 2013). Elevated temperatures increase the amount of water vapour in the atmosphere, thereby increasing mean precipitation and the frequency of extreme precipitation events. Alongside elevated sea levels, these patterns of change are likely to exacerbate flooding in north-western Europe (Lehner et al. 2006; Murphy & Charlton 2008). However, the exact impact on flood mechanisms, particularly on a catchment scale, is difficult to ascertain.

Fluvial flooding is generally caused by intense or prolonged precipitation, associated flood waters are controlled not only by the nature of the precipitation event but also by catchment characteristics and antecedent weather conditions. In Ireland, climate change is expected to alter evaporation and precipitation patterns and so significantly affect the hydrological cycle (Murphy 2013). Temperatures are projected to increase by up to 2.9 °C (Joint Research Centre 2014), with winters becoming wetter and summers drier (Dunne et al. 2008). However, there is substantial uncertainty in estimates of climate change impacts on peak fluvial flows (Bastola et al. 2011a). A number of different studies have assessed these impacts for Irish catchments for the 21st Century, using a range of different climate models, emission scenarios and hydrological models, and these are summarised in Table 1. Although no studies were identified relating directly to the Lee Catchment, research does exist encompassing the Blackwater and Bandon Catchments, immediately north and south of the Lee (e.g. Steele-Dunne et al. 2007; Bastola et al. 2011b). These consistently predict an increase in peak flows, largely due to projected increases in winter precipitation and in extreme events. However, the magnitude of change varies between catchments and simulations, with greater uncertainty associated with larger flood events (Bastola et al. 2011b).

Along the majority of northern European coasts, mean sea level rise (SLR) is projected to be the main driver of change in coastal flooding, enhanced by changes in storm surges and waves (Vousdoukas et al. 2017). Identified projections of future changes in coastal flood mechanisms are summarised in Table 1, for different possible future climate scenarios. This includes potential changes in MSL, land elevation, tidal amplitude and maximum surge height.

Eustatic SLR is expected to continue throughout the coming century due to increasing global temperatures and an associated influx of meltwater from glaciers and ice sheets (Marzeion et al. 2012; Fettweis et al. 2013; Levermann et al. 2014). This has the potential to be enhanced by steric SLR caused by increasing global average sea surface temperature (SST), driving salinity variations and thermal expansion of the oceans (Olbert et al. 2012). Projected SLR for the 21st Century varies depending on the climate model utilised and emissions scenario modelled (Table 1). Notably global SLR will not be uniform, due to the complex interrelationships controlling impacts on a local scale (Olbert et al. 2012). However, global projections (IPCC 2013) are found to be reasonably comparable to projected changes for the Irish coast for mid-range future emissions scenario (Olbert et al. 2012). Along the southern coast of Ireland relative sea level rise (RSLR) is also enhanced by land subsidence – driven by ongoing isostatic adjustment (Bradley et al. 2009), which commenced following the retreat of the British-Irish Ice Sheet after the Last Glacial Maximum (Chiverrell & Thomas 2010).

Sea level can also be enhanced locally by tides, waves and storm surges. While astronomical tides are expected to remain constant over time, it should be recognised that tidal amplitude can be affected by the bathymetry of the ocean (Green 2010) and so may be altered by SLR (Woodworth 2010). Where SLR floods low-lying land, friction and other shallow water effects can increase tidal dissipation, whilst an increase in overall depth within an ocean basin can alter tidal resonance (Pelling & Mattias Green 2014). Despite this potential, it is generally expected that only excessive SLR (>2.0 m) will impact existing tidal patterns (Pickering et al. 2012; Vousdoukas et al. 2017) (Table 1).

Ocean waves result from strong winds blowing over adjacent seas (Gill 1982), and extreme wave heights occur in response to intense weather systems. These are expected to be altered by climate change (Gallagher et al. 2016), however, as Cork City is sheltered within the extensive Cork Harbour, it is considered that there is a negligible effect of waves on flooding in the study area (Olbert et al. 2017).

Storm surges generated by low-pressure systems (cyclones) and/or strong winds (Wells 1997), are of relatively low magnitude in the Celtic Sea and are understood to be generated primarily by low-pressure systems in this region (Olbert & Hartnett 2010). A potential increase in the frequency of low-pressure events (IPCC 2013), combined with projected SLR, which will exacerbate surge peaks, has resulted in a global concern that climate change will result in an increase in the frequency of occurrence and/or elevation of storm surges (e.g. Woth et al. 2006; Wang et al. 2008; Brown et al. 2010). There is, however, substantial uncertainty in projected climate-driven changes in these systems for the North Atlantic region (IPCC 2013). Research is often contradictory, with some predicting an increase in cyclone intensity and frequency (e.g. Dunne et al. 2008; Haarsma et al. 2013) and others predicting a decrease (e.g. Eichler et al. 2013). A decrease in maximum surge height has been estimated for Cork Harbour (Wang et al. 2008) (Table 1), although the same research also projects an increase in the frequency of surges of the magnitude typically associated with coastal flooding. This will potentially increase the probability of a major surge occurring contemporaneously with a high tide, although consideration must also be given to tide-surge interactions (Olbert et al. 2013; Arns et al. 2015).

There have been a number of attempts to model Cork City flooding (e.g Halcrow 2009, 2014); however, these studies treated coastal and fluvial mechanisms disjointly and without considering dependencies and interactions between individual drivers. Zscheischler et al. (2018) showed that weather and climate events due to a combination of multiple drivers may lead to compound effects. Hence, in the context of flood events, modelling individual flood drivers separately can lead to inaccurate characterisation of flooding (Wahl et al. 2015; Moftakhari et al. 2017). More recently, the high-resolution multi-nested MSN_Flood model was applied to coastal-fluvial flooding in Cork City (Comer et al. 2017) and the multivariate capabilities of the system found to allow realistic modelling of the compound effects of multiple flood drivers in response to given boundary conditions (Olbert et al. 2017). Robust representation of wetting and drying within MSN_Flood results in a geographically unconstrained model, adaptable to model domains of any complexity and to multi-open boundary problems. The ability of this model to accurately simulate flood dynamics and quantify the key attributes associated with flood risk – flood wave heights, speeds, propagation patterns and inundation extent – mean that it is highly suited to predicting flood responses to climate change.

The MSN_Flood model comprises a cascade of 90, 30, 6 and 2 metre resolution nested grids (Figure 1). The lower River Lee is contained fully within the 6 m child grid (CG06), which includes an embedded 2 m grid (CG02) capable of resolving the complex hydraulics of Cork city centre. The model integrates the continuity and momentum equations in order to simulate water elevations and velocities. Full details of the hydrodynamics, nesting structure, calibration and performance of the model are available in Comer et al. (2017). Some critical features of the model include the wetting and drying routine, computational efficiency, and accuracy of simulated water elevations and velocity fields – as tested in Olbert et al. (2017). These characteristics make the model particularly applicable to this study.

Due to the computational power required for the high-resolution CG02, this embedded domain was neglected and city centre inundation was resolved purely by the 6 m domain (CG06) of the MSN_Flood model. The model was initially run for a variety of scenarios under present climatic conditions; this provided a basis for the quantification of impacts of climate change.

The western boundary of the CG06 model is forced by fluvial inflows largely controlled by discharges from the upstream Inniscarra Dam, as well as the Shournagh tributary. Due to a lack of river gauge records, a synthetic hydrograph representative of a 100-year event was produced using flood frequency analysis, and quantification of the relationships between physical catchment descriptors, flood magnitudes and hydrograph shape of a hydrologically similar pivot site containing a sufficient record of large flood events. This provided an adjusted QMED of 286 m³/s and a flow of 512 m³/s associated with a 100-year return period, comparable to flows predicted by Olbert et al. (2015) and to the peak recorded at 19012 during the 2009 flood event (560 m³/s). A river flow of 75 m³/s is considered a reasonable estimate of model inputs from the River Lee during normal (baseflow) conditions. Figure 2 shows the synthetic curve of river flows used as boundary conditions for current and future climate scenarios.

Published mean spring and neap tide ranges are typically 3.6 m and 2.0 m respectively in Cork Harbour (Hewitt & Lees-Spalding 1982). The average of these values was taken as a representative mean tidal range (2.8 m), resulting in a mean high water tide (HWT) of 1.4 m above mean sea level (MSL) (Figure 3(a)). A maximum high water spring tidal range of 4.53 m above chart datum (CD) has been determined for a 1,000-year return period (Olbert et al. 2013), from which a maximum HWT of 2.23 m above MSL was estimated (Figure 3(b)).

Olbert et al. (2013) carried out extreme value analysis of surges based on field records and numerical model outputs for 48 historical surge events. In order to consider the worst case scenario, the maximum surge residual (0.97 m) associated with the largest available return period (1,000 years) was used to represent the impact of surges on tidal signal in this study. Surge peaks in Cork Harbour are associated with the tidal phase between mid-flood and high water (Olbert et al. 2013), with the occurrence of the peak surge on rising as opposed to peak tide causing a significant difference in inundation extent (Olbert et al. 2015). For the purpose of incorporating surge impact on coastal water levels, the peak surge residual was applied as a constant across the entire tidal signal, as illustrated in Figure 3(c) and 3(d). This allowed consideration of the worst-case scenario. When considering fluvial mechanisms or MSLR impacts individually, it was assumed the surge residual is equal to zero.

Following identification of average and peak fluvial and coastal model inputs for current conditions, the predictions for climate-driven changes in these boundary conditions considered most pertinent for an MRFS and HEFS were applied to peak discharges and water levels based on the collated literature (Table 1). These are summarised in Table 2. A range of model input files were subsequently generated for both present and future climatic conditions (Figures 2(b) and 3). The model is driven by these boundary conditions and was set-up to compute solutions to the continuity and momentum equations across the model domain at 0.3-second timesteps over a 50-hour duration. Computation time for each model run was 13.4 hours.

The model was validated (Section 4.1) and subsequently used to present a range of flood events, for current and potential future climate scenarios (Sections 4.2, 4.3 and 4.4). A total of 96 flood scenarios (eight current and 88 climate-driven future scenarios) were simulated for Cork City, based on estimates of current, MRFS and HEFS extreme river flows and sea levels. These scenarios, and associated maximum areal extent of inundation and volume of floodwater, are detailed in Table 3. Identified runs consider changes to both individual flood mechanisms and combinations of different flood mechanisms.

Olbert et al. (2017) calibrated and validated the nested CG06 and CG02 model using flood extent mapping and water level marks from 38 locations in Cork City, collated by the OPW following the November 2009 flood event. The CG06 model alone was utilised here to simulate the 2009 flood and validated against the same observations. The simulation was carried out using boundary conditions defined by water levels recorded in the River Lee and Cork Harbour during the flood event (Figure 4).

A good general spatial match was found between observed flood extent and the CG06 simulation (Figure 5). However, the OPW flood extent polygon is based on a rough field assessment and can be used to provide only a rudimentary assessment of model performance. Additionally, this polygon does not extend across the entire CG06 domain, and so comparisons can be made only of simulated inundation within the vicinity of available data. Marginal differences in simulated and observed extent are detected where data is available, however, these are considered primarily a result of the poorly refined observations rather than inaccuracies within the model. Notably, CG06 was observed to overestimate the extent of flooding in the city centre, with floodwaters extending further east than observations.

Watermarks infer maximum water depth at the time of peak inundation. Time series of water elevations were computed at the locations of watermarks to determine maximum simulated water levels, and a linear regression (Figure 6) illustrates the relationship between maximum modelled and observed water depths. Agreement is good, with the majority of points lying approximately on the 45° line and the coefficient of determination (r²) tending towards one. There is, however, a tendency towards overestimation of water elevation, correlating with the overestimation of extent.

Minor deviations in spatial extent and water elevation from observed values are likely a result of the coarser resolution of the CG06 model, in comparison to the CG02 model for which roughness coefficients were calibrated (Olbert et al. 2017). The increased propagation of floodwaters into the city centre is expected to be a result of a failure of the 6 m bathymetry to fully resolve the dense street network, and associated constraints on flows within this region. Additionally, the coarser representation of topography affects the gradient of the water surface, which can influence flow dynamics and explain further deviations in depths and propagations patterns. Some small portion of the error (∼0.1 m RMSE) may also be attributed to errors in LiDAR data, which served to construct the model bathymetry (Bates et al. 2010). Overall, regardless of deviations, the CG06 flood model was found capable of reproducing the 2009 flood event and considered suitable to utilise as a predictive tool for characterising potential future inundation.

Autonomous modelling of fluvial flooding has shown that fluvial events are the primary contributor to extreme flood events under the current climate scenario (Table 3–C5). The flood wave associated with a 100-year return period discharge affects large areas of rural land to the west of the city before propagating eastwards into suburban areas and channelling through the city street network, causing severe flooding of the western city centre (Figure 7). Climate-driven increases in fluvial flows exacerbate inundation extent, water depths and velocities, with floodwaters extending into the eastern reaches of the city centre for MRFS and HEFS river discharges (Figure 7). Simulation of flooding in response to a MRFS increase in the 100-year return period flow (593 m³/s) results in 208 ha of inundation (Table 3– M21), a 12% increase over the current scenario. Simulation of flooding in response to an HEFS increase in fluvial flows (728 m³/s) results in 242 ha of inundation (Table 3–H21), a 30% increase over the current scenario. Under this scenario, the majority of streets within the city centre are inundated, with only the far eastern reaches of the central island remaining dry.

The maximum water depths associated with fluvial flooding are detailed in Table 4 and illustrated in Appendix A, Figure A. 2; under MRFS and HEFS river discharges water depths increase significantly. Floodwaters are deepest in rural areas in the west of the model domain, reaching up to 3.52 m under a HEFS. Under all scenarios the greatest water depths in the city centre are located in the low-lying areas adjacent to the north channel, exceeding 2 m in places for a HEFS. Water depths are generally much shallower within the street network, generally <0.5 m under the current scenario. The maximum total velocity from model snapshots was extracted for each cell within the model domain, for each simulated scenario (Appendix A, Figure A. 3). Under all fluvial scenarios this illustrates patterns of higher peak velocities immediately adjacent to the river channel and in areas where obstacles to flow are minimal, including the rural areas to the west of the city and the plains represented by recreational areas in the city centre. Velocities reduce in the flooded street network where complex topography and narrow flow pathways result in greater bottom friction, and at more distal locations from the river channels as floodwaters lose momentum, becoming increasingly stagnant. The increased volumes of water in the MRFS and HEFS result in an alteration in the wave velocity. Changes in propagation patterns can result in an increase in velocity in some locations and a decrease in others; on average floodplain velocities increase by 0.07 m/s in the MRFS and 0.10 m/s in the HEFS, with maximum increases of 0.12 and 0.32 m/s respectively.

Coastal mechanisms are also capable of triggering inundation, shown by simulations incorporating MSLR and the exacerbating influence of storm surge residuals on water levels at the downstream boundary. Relative increases in coastal inundation under future climate scenarios, in comparison to the current climate, are much greater than those modelled for fluvial events.

Simulations suggest that existing defences will be capable of preventing flooding under average tides for both an MRFS and HEFS MSLR (Figure 8(a)), and under spring tidal conditions for an MRFS MSLR; with the exception of minor flooding of low-lying land near the river mouth (Figure 8(b)). However, 37 ha of inundation is simulated for an HEFS MSLR under spring tides (Figure 8(b), Table 3– H2). This is critical as, should such sea-level rise occur, coastal inundation would occur on a bi-monthly basis during each spring tide, regardless of surge conditions. Associated flooding is focused in eastern portions of the city centre, with waters primarily overtopping the south channel and propagating north. Water depths are generally <0.5 m in areas of inundation, with discrete areas reaching depths of up to 1 m; associated velocities are low (<0.25 m/s).

Consideration must also be given to the potential impact of surge residuals on water levels. When applied to average tides for present-day sea levels, neither the current, MRFS or HEFS peak surge residuals cause flooding (Figure 9). However, a combination of MSLR and either current or HEFS surge residuals can initiate coastal inundation under average tides (Figure 9(a) and 9(c)). Associated water depths are generally shallow (<0.5 m) and velocities slow, however, these marginally increase under the HEFS surge/HEFS MSLR scenarios (Appendix A, Figure A. 4).

When applied to spring tides, the current peak surge residual causes notable flooding (38 ha) regardless of climate change (Figure 10(a), Table 3–C4). Associated water depths are again generally <0.5 m (Appendix A, Figure A. 5a) and velocities <0.25 m/s (Appendix A, Figure A. 6a), although these increase in a discrete area to the south of the river. Under an MRFS and HEFS sea level, the co-occurrence of the current peak surge residual with spring tides has the potential to cause significantly greater flooding – 168 ha and 182 ha respectively (Figure 10(a), Table 3– M5, H5). Under these scenarios, the majority of the city centre becomes inundated, and large areas of flooding occur in the Marina industrial area to the south of the river, with significant increases in water depths and velocities (Appendix A, Figure A. 5 and Figure A. 6).

Significant flooding continues under all sea level scenarios for the MRFS reduction in surge height, although this results in a marginal decrease in affected area (Figure 10(b)). Under an HEFS projection of stronger storm surges, inundation extent increases for all sea level scenarios when compared to the current surge scenario (Figure 10(c)), with a maximum inundated area of 191 ha resulting from the co-occurrence of an HEFS surge event with an HEFS MSLR (Table 3–H4). This scenario causes flooding over a greater extent than the current 100-year fluvial flood event, impacting the majority of the city centre and eastern industrial areas (Figure 10(c)). Water depths and velocities also increase for each associated MSLR scenario (Appendix A, Figure A. 7 and Figure A. 8), with depths of up to 2 m simulated in the city centre.

As well as consideration of autonomous flood drivers, the interaction of sea levels and fluvial flow is an important consideration for effective flood prediction and management (Moftakhari et al. 2017).

High coastal water levels are observed to exacerbate fluvial flood extent, with the coincidence of a peak fluvial event with spring rather than average tides increasing city centre inundation (Figure 11, Table 3–C7). This is expected to be a result of high coastal waters restricting fluvial discharge and causing a stacking of water along the riverbanks (Hoitink & Jay 2016). Similarly, simulated medium-range and high-end MSLR, as well as the co-occurrence of surge conditions during a fluvial flood event, have the ability to increase the extent of inundation associated with peak fluvial discharges, under both average and spring tidal conditions (Table 3). Where coastal water levels sufficient to cause inundation independently are combined with peak fluvial flows, flooding becomes exacerbated both by coastal inundation and by the influence of coastal waters on fluvial flows.

Combined fluvial and coastal events are further complicated by the influence of extreme fluvial discharges on coastal inundation. Comparison of inundation extent associated with an HEFS MSL under mean fluvial flows (H2) and under current (H12), MRFS (H39) and HEFS (H24) peak fluvial flows suggests that coastal inundation is reduced by increased river discharges (Appendix A, Figure A. 9). This is expected to be a result of increased volumes and velocities of freshwater inflows restricting the propagation of tides upstream, a process illustrated by Leonardi et al. (2015).

Despite a complex relationship between different flood drivers, the most extreme flood events result from the compound impact of high river flows and high coastal water levels. When combined with spring tides under a HEFS MSL, the HEFS peak fluvial event results in 265 ha of inundation (Table 3–H24). The entirety of the city centre becomes inundated and fluvial flooding of the city downtown is exacerbated by coastal flooding (Figure 12). When combined with an HEFS surge event, 387 ha of inundation is simulated (Table 3–H26). River discharges are insufficient to notably reduce tidal propagation resulting from the associated extreme sea levels, and extensive coastal inundation occurs to the east of the city centre and in urban areas to the south (Figure 12).

Associated water depths and velocity are illustrated in (Appendix A, Figure A. 10 and Figure A. 11). Under this extreme event inundation to the west of the city remains solely fluvial driven and water extent, depth and velocity remains comparable to that resulting exclusively from an HEFS fluvial flood event. To the east of the city centre water depths and velocities are generally comparable to those resulting exclusively from coastal inundation. However, the city centre is impacted by both fluvial and coastal inundation, resulting in an increase in water depths compared to those experienced under autonomous drivers. Water velocities are primarily fluvial driven; however, high coastal water levels have the potential to reduce velocities by reducing the gradient of flow. A reduction in maximum water velocity is observed within the eastern reaches of the Lee channel for the combined future event, when compared to the scenario forced exclusively by an HEFS fluvial discharge. Despite this, water velocities in the city centre floodplains are generally unaffected, with the additional overtopping of waters forced by coastal mechanisms increasing velocities in the eastern reaches of the city centre.

This study aims to understand the conditions under which urban flooding will occur under current and future climatic conditions, with the overall goal of providing a methodology for comprehensive forecasting and assessment of flooding under changing climate. The urban area of Cork City frequently impacted by both fluvial and coastal flood mechanisms occurring individually or as compound events was used to explore climate driven changes in flood mechanisms, dynamics and extents. The MSN_Flood hydrodynamic model has been used to provide a range of estimates of future inundation. The modelling system was found to be capable of accurately resolving the complex hydrodynamics of the domain at scales commensurate with flow features. Inundation in both the upstream rural floodplains and in the downstream network of dense streets was accurately reproduced by the 6 m urban flood model. The range of widely varying inundation patterns generated within this research based on different estimates and combinations of current, medium-range and high-end projections of extreme river flows and sea level indicate that the most pertinent inferences on potential future inundation patterns must be drawn from an ensemble of model simulations. A low computational effort of the nested modelling system enabled multiple climate change scenarios to run in a relatively short timeframe, and so facilitated a methodology for accurately stress testing climate change effects on complex coastal-fluvial flooding in urban areas.

The research clearly shows that while the fluvial signal will remain a primary driver of flooding and responsible for the greatest inundation in future climate (30% increase), the climate-driven changes in both MSL and storm surges could result in a staggering 400% increase in coastal inundation (when compared to current climate) and therefore a significant shift in contribution towards the coastal mechanism. Rising MSL will have the potential in the future to overwhelm existing defences and inundate parts of the city centre during spring tides, regardless of surge conditions. As such, without appropriately adapted flood defence systems, coastal inundation could occur on a bi-monthly basis during each spring tide. Moreover, increased coastal water levels are shown to impact the conveyance of fluvial flows to the ocean and extreme fluvial discharges to impact the propagation of coastal waters inland; this illustrates the requirement for accurate modelling of the dynamics of compound events due to coastal and fluvial signals. Climate-driven increases in both fluvial and coastal mechanisms have the potential to cause up to 387 ha of inundation across the model domain for the worst-case scenario coastal-fluvial flood event. This is an increase in inundation of 76% compared to the 100-year flood event of November 2009. Water depths and velocities, exacerbated by both fluvial and coastal mechanisms, will create a significant threat to the populace across the majority of the city.

While this research focuses on coastal-fluvial drivers and flood hazard, the impact of such a compound event is another important aspect to consider. As coastal flooding prevails in highly urbanised areas where a significant portion of socio-economic wealth is accumulated, future coastal flooding will thus have a major socio-economic impact.

Overall, this research demonstrates that the adopted methodology can be successfully used to understand the major effects that climate change may have on future flooding. As the potential exposure of communities to flooding is a critical task for long-term planning and risk assessment, the methodology utilised within this research can help establish the most effective adaptation plan and as such facilitate decision making in flood defence design and flood risk management.

The authors would like to thank OPW, Ireland for hydrological data and Dr Stephen Nash for making the MSN_Flood model available. Comments from anonymous reviewers were much appreciated.

The Supplementary Material for this paper is available online at https://dx.doi.org/10.2166/wcc.2020.166.