High-resolution flood modeling of urban areas using MSN_Flood

Michael Hartnett*,Stephen Nash

Department of Civil Engineering,National University of Ireland,Galway,Ireland Received 12 December 2016;accepted 23 June 2017 Available online 18 October 2017

High-resolution flood modeling of urban areas using MSN_Flood

Michael Hartnett*,Stephen Nash

Department of Civil Engineering,National University of Ireland,Galway,Ireland Received 12 December 2016;accepted 23 June 2017 Available online 18 October 2017

Although existing hydraulic models have been used to simulate and predict urban flooding,most of these models are inadequate due to the high spatial resolution required to simulate flows in urban floodplains.Nesting high-resolution subdomains within coarser-resolution models is an ef ficient solution for enabling simultaneous calculation of flooding due to tides,surges,and high river flows.MSN_Flood has been developed to incorporate moving boundaries around nested domains,permitting alternate flooding and drying along the boundary and in the interior of the domain.Ghost cells adjacent to open boundary cells convert open boundaries,in effect,into internal boundaries.The moving boundary may be multi-segmented and non-continuous,with recirculating flow across the boundary.When combined with a bespoke adaptive interpolation scheme,this approach facilitates a dynamic internal boundary.Based on an alternating-direction semi-implicit finite difference scheme,MSN_Flood was used to hindcast a major flood event in Cork City resulting from the combined pressures of fluvial,tidal,and storm surge processes.The results show that the model is computationally ef ficient,as the 2-m high-resolution nest is used only in the urban flooded region.Elsewhere,lower-resolution nests are used.The results also show that the model is highly accurate when compared with measured data.The model is capable of incorporating nested sub-domains when the nested boundary is multi-segmented and highly complex with lateral gradients of elevation and velocities.This is a major bene fit when modelling urban floodplains at very high resolution.

?2017 Hohai University.Production and hosting by Elsevier B.V.This is an open access article under the CC BY-NC-ND license(http://creativecommons.org/licenses/by-nc-nd/4.0/).

Multi-scale nesting;Surge- fluvial flooding;Urban flooding;Multi-segmented boundary;Moving boundary

1.Introduction

Globally,large urban areas have developed in coastal plains along the banks of large rivers.Many of these conurbations are subject to severe flooding arising from complex interactions between tides,storm surges,and fluvial events.The hydraulics of flooding of coastal floodplains is one of the most dif ficult hydraulic issues to understand.Storm surges occurring at the same time as high volumes of fluvial water result in complex and highly dynamic boundary conditions.Urban developments builtonthese floodplains,whereurbanstreetsbecomerelatively narrow hydraulic channels,further complicate the hydraulic conditions.Urban floodplains are usually the main focus ofattention during large flood events,as loss of human life and economic value often occurs.It is estimated that the annual economic flood damage for coastal cities amounted to 6 billion US dollars in 2005.Future developments,such as sea level rise due to climate change and coastal migration,will result in increasingly costly flooding.It is projected that,due to these developments,thetotalglobal flooddamageby2050willriseto approximately 60 billionUS dollars(Muisetal.,2015).TheUS cities of Miami,New York-Newark,New Orleans,Tampa-St.Petersburg,and Boston rank among the 20 cities with the largest expected flood losses in 2050,in the order of 12 billion US dollars annually.Nearly all of these 20 cities are located on coastal floodplains.It is projected that the world's 136 largest coastal cities could experience flood damage costs in the order of 1 trillion US dollars annually by 2050.

Two particular characteristics of urban coastal floodplains make the application of hydraulic models dif ficult:(1)the large lateral extent of the fluvial floodplains upstream of urban areas,and(2)the requirement of high spatial resolution in order to simulate the hydraulics in the narrow urban streets.This paper addresses the development of a model and its application to situations in which large rivers flow through urban coastal floodplains also subject to tides and storm surges.

It is desirable to have one integrated modeling system that can perform hydraulic simulations from the open sea to the upper reaches of rivers flowing into urban floodplains.Previous researchers have developed some models to tackle this problem.Liang et al.(2007)developed a coupled onedimensional(1D)and two-dimensional(2D)system,where the 1D river model(FASTER)is dynamically coupled with the 2D coastal model(DIVAST).Although this system has merits,it does not fully resolve the domain in a hydraulically consistent manner.Bates et al.(2010)developed the LISFlood model for application to the urban floodplain.Again,this model has great merit(particularly computational ef ficiency).However,some components of the model only conserve mass,with a major physical process ignored.Other more recent models have looked at effects of spatial resolution and hydraulic modeling(Nguyen et al.,2016;Altenau et al.,2017).However,they have not developed important functionality with respect to locating nested boundaries on complex floodplains,nor have they considered flooding from the perspectives both of the river and of tides/surges.

In order to accurately and completely address this hydraulic problem,the authors developed the MSN_Flood system.This is a multi-scale nested modeling system,developed through the novel treatment of nested boundaries as internal computational boundaries.The main focus of the development of MSN_Flood was to develop a highly accurate model that would be computationally ef ficient and had the potential to be applied to any urban coastal floodplain.When developing a nested modeling system,the boundary treatment is obviously most important.In addition,it is required that the extent of the high resolution of the urban area must be as small as possible for computational ef ficiency.For this reason,MSN_Flood was developed so that flooding and drying are permitted along the nested boundary:when the upstream river is not flooding,the boundary width is small,whereas during flooding the boundary width increases.State of the art models,such as Mike Flood,which is a coupled 1D-2D model,are unable to allow flooding and drying along a boundary(DHI,2007).Also, flow across a nested boundary does not have to be normal to the boundary.Thus,recirculating flows are permitted along the boundary of MSN_Flood,allowing the solution of realistic and complex hydraulic conditions.Some of the basic principles about this development are presented in Nash(2010).

MSN_Flood,developed by the authors,solves the full 2D,depth-integrated Navier-Stokes equations.In the initial test stage,MSN_Flood was tested against data from a scale-model rectangular harbor in a tidal basin where the model was validated against a range of velocities from controlled experiments(Nash,2010).The model was applied to the coastal water bodies of Galway Bay and sub-sections of Cork Harbor to simulate tidal flows.Although not presented here,the model also incorporated an adaptive meshing ability,and this was applied to Galway Bay.It was subsequently necessary to assess model performance when it was applied to large-scale flood events.Cork City,Ireland was chosen as the test case to simulate a major historical flood event.This case study is the main focus,as it shows for the first time that MSN_Flood is capable of solving complex,real-world hydraulic problems.One of the novel aspects of the application of MSN_Flood to Cork City is that it demonstrates that the moving boundary located on a floodplain is robust and working correctly.

An overview of the nested modeling system is given in the next section,where the treatment of the nested boundary is summarized,particularly the use of ghost cells.In Section 3,the application of MSN_Flood to a complex fluvial-tidal-surge flooding event in Cork City is described in detail,with emphasis on the behavior of the novel moving boundary.

2.Methodology

MSN_Flood was developed in a general-purpose manner to allow multiple levels of nesting between a spatially coarse grid(parent grid)and higher-resolution grids(child grids)nested within the parent grid.The model uses an alternating-direction semi-implicit finite difference scheme for all hydraulic calculations.There is no effective limit to the degree of subnesting;the authors have tested MSN_Flood using four levels of nesting and have obtained excellent results with a nesting ratio of 9:1 in tidally dominated flows(Nash,2010).MSN_Flood incorporates a moving boundary at the nest interface;this allows the width of a boundary to change depending on the stage of flooding,and therefore alternate flooding and drying is allowed along the boundary.The flooding and drying routine used in MSN_Flood was originally developed by Falconer and Chen(1991).Although this type of boundary itself was numerically stable,in order to ensure model accuracy at a boundary,due to non-uniform drying of the coarse domain,an adaptive linear interpolation scheme needed to be developed.When this was implemented,the model results became highly accurate.

Conservation of mass and momentum between the parent grid(PG)and child grid(CG)of a nested modeling system is crucial to modeling accuracy.Conservation of incoming fluxes at child grid boundaries is of greatest importance,as boundary data have the most in fluence on the interior solution during the period of inward propagating fluxes.A novel approach to boundary formulation was developed,which reduced boundary formulation errors and thus ensured high levels of conservation of incoming fluxes of mass and momentum to the child grids.The approach internalizes the open boundaries of a child grid by incorporating ghost cell data in a modi fied boundary formulation.The approach is general in nature and could therefore be applied to any other finite difference hydraulic modeling system.Boundary data are interpolated from the parent grid to the child grid,and thus the interpolation schemes used must conserve mass and momentum.Standard linear interpolation was used for temporal processes in MSN_Flood(Miyakoda and Rosati,1977;Pullen and Allen,2001;Korres and Lascaratos,2003).Four schemes were tested for spatial processes:zero-order interpolation,linear interpolation,quadratic interpolation,and inverse distance weighted interpolation.Based on these tests,linear interpolation was adopted within MSN_Flood.

Flooding and drying of the nearshore region is a prevalent processinmanycoastalzones.Theinternalboundaryapproach,together with the implementation of a tailored adaptive linear interpolation scheme,enables the child grid boundary to move laterally.Thedynamicinternalboundaryfacilitatesthe flooding and drying of boundary grid cells without adversely affecting modelperformance.Importantly,thistreatmentoftheboundary does not constrain the boundary conditions of a child grid to flow that is only perpendicular or normal to the boundary;the nested boundaries now facilitate recirculating flows across the boundary.The nested boundary may also be multi-segmented and discontinuous,with different water levels along different segments;this is a signi ficant development over previous flood models.Furthermore,a nested domain may have two nested boundaries meeting at a corner of the nested domain and retain numerical stability and model accuracy.These features arevery important when modeling inter-tidal zones in real-world applications.MSN_Flood has previously been extensively tested forperformanceandaccuracypriortoreal-worldapplicationfor urban flood hindcasting.It was applied to an idealized rectangular harbor,an experimental harbor in a tidal basin,and a real coastal system,namely Cork Harbor.Those tests are described in Nash(2010).

It hasbeen generallyconsidered(Koch and McQueen,1987)that there are two types of errors at the nested boundary:speci ficationerrorsandformulationerrors.Theauthorsidentify a third source of error in the formulations of the conservation equations on the boundary itself.The authors suggest the followingthreeerrorsources:(1)boundaryspeci fication:dueto poor data on the boundary;(2)boundary operators:due to improperly designed boundary operators;and(3)boundary formulation:due to overly simpli fied boundary formulations.

Our primary objective was to develop a nested model that would signi ficantly reduce these three types of errors.Different types of boundary conditions were tested during the development,following work from Blayo and Debreu(2005),and Marchesiello et al.(2003).In particular we investigated how the following boundary conditions behaved:(1)Dirichlet,(2) flow relaxation,and(3)radiation.The Dirichlet condition proved to be stable and the most accurate,so it was adopted.Results of the application of relaxation and radiation boundary conditions can be found in Roed and Cooper(1987),Palma and Matano(1998),and Nycander and Doos(2003).

One of the problems with previous attempts at developing nested modeling solutions is that researchers have ignored momentum flux gradients across a boundary.In the development of MSN_Flood,the authors set up ghost cells outside the actual nested boundary.Results from a coarse grid model along theghostcellsandtheboundarycellswerethenusedtocompute all of the terms in the momentum equation along the nested boundary.This approach allows MSN_Flood to conserve momentum across the nested boundary,whereas other nested modeling systems do not allow for this.This is in keeping with the suggestion by Zhang et al.(1986)that boundary operators should propagate resolvable waves smoothly across boundaries and that mass and momentum must be conserved between parent and child grids.Fig.1 shows the boundary con figuration used within MSN_Flood.Both the ghost cells and the internal boundary are shown.The CG domain(thin gridlines and small symbols)overlaps the PG domain(thick gridlines and small symbols).

3.Case study:Cork City flooding

MSN_Flood was applied to a highly complex,real-world problem.Cork City is a low-lying,coastal conurbation with relatively high-frequency floods.At the upstream limits of the city a large weir controls the water level of an urban floodplain,and just downstream of the weir the main Lee River bifurcates,encompassing the main downtown area until the two channels join about 5 km downstream,at the entrance to Cork Harbor.

During winter,Ireland is subject to low-pressure storm systems tracking northeast across the Atlantic Ocean.These storm systems induce storm surges around the Irish coast and they also cause large amounts of precipitation.During large storm events,water enters the urban floodplains of Cork City from two directions:(1)from upstream river flood waters,and(2)from Atlantictidesandstormsurges.InNovember2009,amajor flood eventoccurredinCorkCityanditshinterlands.Duringthestorm,river discharges peaked at over 500m3/s(normal values of discharge in the Lee River are around 40m3/s).Also,a surge of about0.5mpropagatedintoCorkHarborfromtheAtlanticOcean during a period of high astronomical tidal range.The storm resulted in signi ficant flooding throughout the city and caused approximately 100 million euros worth of damage to property.

Fig.1.Example portion of model grid con figuration for a 3:1 nesting ratio.

A modeling system comprised of two externally linked models was developed to hindcast the flood event.An ocean model of the northeast Atlantic Ocean(approximately 5-km grid cells),using the Princeton ocean model(POM),provided boundary conditions for the MSN_Flood model of Cork Harbor that,through nesting,resolved hydrodynamics of the region at four scales of spatial resolution:90 m,30 m,6m,and 2m.Water levels consisting of joint tides and surge signals generated by the ocean model were spatially and temporally interpolated and prescribed to the MSN_Flood model of Cork Harbor at each model time step.At the first level,the water elevations,velocities,and velocity gradients of the coarse parent grid were interpolated in space to provide boundary data for the finer child grid.Subsequently,each child grid became a parent grid for its own higher-resolution child grid.The structure of nesting cascades and the extent of each model are shown in Fig.2.The water depth in the figure means the depth below the mean sea level.The 6-m gird model(CG06 model)was driven by hybrid boundary conditions:(1)the eastern boundary was constructed as a nested boundary receiving its data from the 30-m grid model(CG30 model),and(2)the western boundary was a true open boundary receiving data from observations at a river gauging station.

The parent grid model resolves the hydrodynamics of the entire domain of Cork Harbor at a grid spacing of 90 m and a time step of 18 s(PG90 model).The first-level nested model embedded within the PG90 model,the CG30 model,downscales the area of interest at a 3:1 nesting ratio and computes hydrodynamics at a 30-m grid spacing and a 6-s time step.At a nesting ratio of 5:1,the CG30 model provides boundary conditions for the east boundary of the CG06 model at a 6-m grid spacing,which narrows down the area of interest to the Lee River and its estuary;the CG06 model has a time step of 0.6 s.The discharge data from a gauging station(station number 19011)in the Lee River were provided by the Irish Of fice of Public Works(OPW),and are speci fied at the western boundary of the CG06 domain.The hydrograph of discharge at this western boundary is presented in Fig.3.

Finally,the highest spatially resolved model was the CG02 model with a grid spacing of 2m.It was fully embedded within the CG06 model at a 3:1 nesting ratio.This model resolves the hydrodynamics of the Lower Lee River and its floodplains,covering the urban area of Cork City.

The urban topography was constructed from 2-m-resolution light detection and ranging(LIDAR)data provided by the OPW.The data consisted of both a digital surface model(DSM)that included buildings and a digital terrain model(DTM),which represented the ground surface only.Postprocessing merged the DTM and DSM to include buildings and structures that would greatly impede the flow of water.These data were combined and interpolated onto 6-m and 2-m regular Cartesian grids for the area of interest to produce the floodplain topography shown in Fig.4.

3.1.Model validation

The POM surge model of the Northeast Atlantic Ocean has been well validated against observed surges.Details of the model setup and validation can be found in Olbert and Hartnett(2010).The PG90 and CG30 models of Cork Harbor have also been validated for tidal flows against measurements of water levels and current velocities.Details can be found in Nash and Hartnett(2014).This paper focuses on aspects of the validation and application of the nested models to the case study of rural and urban floodplains.It can be very dif ficult to validate an urban flood model for an extreme flood event due to a lack of a complete data set.The authors were fortunate that all main boundary conditions were available and good records of maximum water elevations had been kept.An assessment of MSN_Flood application to Cork City comprises an extensive validation of the CG06 model and the ultra- fineresolution CG02 urban flood model against available data for the flood event.High water marks after the flood event were collected by the OPW at 45 survey points across the flooded area de fined jointly by the CG06 and CG02 models(Fig.5).

The high water marks surveyed at the locations shown in Fig.5 were used to validate the models.During validation,model hindcast flood levels were compared with flood level records.The surveyed points were relatively uniformly distributed;for analysis purposes they were grouped into five locations:complete model domain/region,upper floodplain,river channel,south channel,and city center.Table 1 presents a statistical analysis of comparison between the model hindcasts and observations of water levels in various groups.

Fig.2.Four-level nesting structure of coastal model.

Fig.3.Discharge data at station 19011 of Lee River over November 19 and 20,2009.

This analysis considers correlations between the observed data and hindcasts for the full CG06 domain and four subdivisions of the region as presented in Table 1.The results clearly show that the model is highly accurate throughout the domain.The average water level differences for all 45 points is 0.060 m.It is seen,signi ficantly,that these differences are strongly in fluenced by the upper rural floodplain results rather than those of the lower urban floodplains.This can be attributed to the fact that hydrodynamics in the upper floodplains are solved by the lower-resolution CG06 model,whereas flows in the urban floodplains are simulated using the CG02 model.

Fig.6 shows the modeled and observed maximum water level pro files along the river channel of the CG06 domain.It illustrates excellent agreement throughout the model domain.

Fig.7 presents a plot of the comparison of the hindcast and observed data.Note that the R2value between the data sets is 0.9.Figs.6 and 7 both illustrate the high degree of MSN_Flood accuracy when simulating maximum water levels during complex flood events.

While correlations between data and hindcasts are very high,indicating similar patterns in the model and observations,root mean square errors(RMSEs)exceeding 0.15 m in most locations result from a slightly over-predictive(conservative)tendency of the model.Interestingly,a lower RMSE and hence better agreement was obtained at urban floodplain locations as opposed to locations adjacent to the river bank.This outcome is signi ficant as it implies that MSN_Flood accurately represents the dif ficult mechanics of flood wave propagation into topographically and hydrodynamically complex urban floodplains due to higher model resolution in the urban area.

3.2.Evolution of flood inundation and flood risk to people

MSN_Flood was used to develop a hindcast of a storm in November 2009 and its inundation of the rural floodplains and the urban floodplains of Cork City.The flood was triggered by both fluvial and coastal mechanisms when extremely high river discharges coincided with high water levels due to spring tides and moderate storm surges propagating into Cork Harbor from the Atlantic Ocean.The western boundary of the 2-m ultrahigh-resolution MSN_Flood grid is located at the highly complex interface between fluvial and tide/surge interactions(Fig.4).During normal flow conditions in the Lee River the river is fully contained within the main channel.Fig.8 shows a cross-section of the model along this boundary.This demonstrates that the cross-section at this location is quite complex,with buildings separating the main and side channels generating a multi-segment,non-continuous nested boundary.In Fig.8,4hrefers toatime early during the flood event.Notethat the water is contained within the main channel of the Lee River at this time.Water levels are shown for different stages of the flood;the time progression of these water levels illustrates how the boundary here propagates laterally.

Fig.4.LIDAR DTM data of Cork City combined with seabed bathymetry of Lee River.

Fig.5.Survey locations of high water marks.

Table 1 Comparisons between observed data and model hindcasts.

As the flood waters rise at the western boundary,the channel width increases.Fig.9 shows how the water levels in different segments of the boundary(A,B,C,and D)rise and fall over the duration of the flood event.A particularly important feature of the boundary dynamics is the signi ficant gradient of the water level across the nested boundary,especially the water level gradient between segment D and other segments.It can also be seen that the gradient changes rapidly with time during phases of the flood.MSN_Flood boundary treatment ensures that the boundary is both stable and accurate during this complex event.The ability of MSN_Flood to represent the nested boundary dynamics in this manner illustrates the necessity of using a 2D model for the entire region and nesting where necessary;a linked 1D-2D modeling system is not capable of representing the dynamics of the boundary,as shown in Fig.9,at the interface between the 1D and 2D models.

Initially the authors performed a hindacast analysis using the CG06 model.Inundation extents at the start and end of the event over the upper rural and lower urban floodplains are shown in Fig.10,as simulated by the CG06 model.

Fig.10(a)shows that the model extends from the eastern boundary located in Cork Harbor.Westward from this boundary,the estuary narrows until downtown Cork City,where the channel bifurcates into a north and south channel surrounding the city.West of the city the channels rejoin just downstream of a large weir.The weir is the upper limit of tidal flows.However,during this flood event, fluvial,tidal,and surge actions caused the weir to be partially or fully submerged,resulting in complex flows and backwater effects.Westward of the weir,the Lee River is contained within its banks during normal flow conditions,as shown in Fig.10(a).Fig.10(b)shows the degree of inundation at the end of the event.It is clear that in the upper reaches the river has burst its banks and flooded large areas of urban floodplains.In the downtown area between the two channels,signi ficant amounts of flooding are observed;most of the economic damage was caused in this downtown region.

Fig.6.Modeled and observed maximum water level pro files along river channel between station 19011 and Tivoli(Tivoli is the downstream location of the study domain).

Fig.7.Comparison of all modeled and observed maximum water levels.

Fig.8.Cross-sections and water levels at western boundary of CG02 domain.

A more in-depth assessment of the downtown region is provided by analysis of the results of the CG02 model,described in inundation maps of the beginning and end of the event as simulated by the model,which are presented in Fig.11.Fig.11(a)is an inundation map during the early part of the flood event when most of the flow is contained within the north and south channels of the Lee River.A detailed,high-resolution map of the final inundation in the downtown area is presented in Fig.11(b).

Fig.9.Development of western boundary elevation over time.

Fig.10.Progression of flood inundation over upper and lower floodplains of Cork City.

It is clear that greater water depths are found in the main river channels,but water depths of over 2m are observed on the urban floodplain.Fortunately,there was no loss of life due to this flood event.It is also clear that a large proportion of the urban floodplain,particularly to the west of the city,is submerged.Figs.10 and 11 illustrate how MSN_Flood realistically propagates flood waves into both rural and urban floodplains.Further analysis by the authors is examining relative contributions to this flooding from tides and surges and from river flows,separately.

Fig.11.Progression of flood inundation in urban area of Cork City.

Flooding can obviously have substantial economic and social impacts in affected regions.Overall loss due to flooding can be estimated simply from flooding depth-damage curves,though such an approach has an inherent high degree of uncertainty.Fewtreel et al.(2011)note that there are additional important factors including flood wave velocity or duration of inundation.Flood wave velocity poses a signi ficant risk to people.A wide range of studies have been undertaken considering the safety of people in floodwaters(e.g.,Abt et al.,1989;Apel et al.,2009;Jonkman and Penning-Rowsell,2008).The authors adopted the method of Keller and Mitsch(1993),relating the physical stability of people to flow velocity and water depth.This method invokes functional relationships between critical velocity Ucand water depth h in the form of instability curves estimated separately for children and adults(Fig.12).Although this method provides only an approximation,as it does not consider crucial aspects such as local flow pattern,terrain,and personal wellbeing(Fewtreel et al.,2011),it is a good estimator of the general risk to people's safety.

MSN_Flood solves the full 2D Navier-Stokes equations.Hence,the model can produce maps of maximum water depths and velocities like those shown in Fig.13.

Fig.12.Relationship between water depth and critical velocity for stability of children and adults in floodplains(source:Keller and Mitsch,1993).

Fig.13.Contours of maximum water depth and velocity during simulation period.

Fig.14.Distribution of hazard degrees to children and adults.

Using water depths and velocities,and the functional relationships shown in Fig.12,hazard risk maps were developed for both children and adults for Cork City.With the application of instability curves for the water depths and velocities described above,the degree of hazard for people(HD)was quanti fied using the formula of Xia et al.(2011):where U is the flood water velocity(m/s)and Ucis a critical velocity(m/s)for the given water depth obtained from Fig.12.

For the flood of November 2009 the hazard risk maps are presented in Fig.14.These maps are invaluable for development of evacuation plans for the city and assessment of the risk associated with first responders driving rescue vehicles into the flooded city streets.

A considerable portion of the inundated area poses a risk to children(HD close to 1.0),including green areas adjacent to the river bank as well as some sections of main roads and residential estates.Risk to adults is less pronounced and is con fined mostly to recreational areas adjacent to the river channels.

4.Conclusions

The key conclusions and implications from this research are as follows:

(1)The POM-MSN_Flood modeling system was found to be capable of resolving hydrodynamics at scales commensurate with flow features.The spatial extent of the ocean model was large enough to allow dynamics of the Northeast Atlantic Ocean and evolution of a storm to be resolved,while resolutions of the nested models within MSN_Flood were suf ficient to adequately resolve dynamics at scales of nested domains.Such a setup enabled externally-generated surge and tidal waves to propagate from the ocean inshore through the semienclosed coastal embayment of Cork Harbor to the Upper Lee River.The model is also capable of simulating strong flow conditions in upper sections of the Lee River due to abrupt releases of high volumes of water from Inniscarra Reservoir.

(2)One of the most signi ficant aspects of this research is that MSN_Flood allows the boundary of a nested domain to be located on a floodplain with flooding and drying.This laterally moving boundary allows one to locate a nested subdomain almost anywhere in the region of interest,thereby reducing computational costs.This is a novel development that can be incorporated into other modeling systems.

(3)The flood inundation over both the upstream rural floodplains and downstream network of dense streets was accurately reproduced by the 2-m urban flood model.

(4)High-resolution LIDAR terrain data is crucial for accurate assessment of inundation.However,post-processing of the dataset was required to correct presence or absence of some surface objects such as trees and hedges,which were found to have an effect on flow fields due to misrepresentation of through flow and over flow.

(5)Analysis of fluvial and coastal flood mechanisms clearly demonstrates that river discharges were largely responsible for the November 2009 flooding in Cork City.Coastal mechanisms did not pose a threat to coastal flood defense structures.However,when combined with river flows,they contributed to flooding,particularly in the dense narrow streets of the downtown district.

(6)ManyengineersmodelingdomainssuchasCorkCityand its hinterlands would use dynamically linked 1D-2D models.However,the nested boundary at the western side of the 2-m Cork City model is quite complex.The water level boundary sometimes has a relatively signi ficant gradient across it.Also,thewaterlevelboundaryismulti-segmentedanddiscontinuous.MSN_Flood represents these processes accurately,whereas dynamically linked 1D-2D models do not incorporate the fluid mechanics necessary for this accurate representation.

In this study it was demonstrated how a numerical model can facilitate an understanding of dynamics of flood wave inundation through detailed temporal and spatial analysis of flood propagation.This can be further used for flood management through identi fication of future flood-prone areas,flood risk timeframes,inundation extents,and flood water heights.This knowledge can also serve rescue and relief operations after flooding.

Acknowledgements

The authors would like to thank Dr.Lei Ren and Dr.I.Olbert,post-doctoral researchers at National University of Ireland Galway,for input to analysis for this paper.

Abt,S.R.,Wittler,R.J.,Taylor,A.,Love,D.J.,1989.Human stability in a high flood hazard zone.Water Resour.Bull.25(4),881-890.https://doi.org/10.1111/j.1752-1688.1989.tb05404.x.

Altenau,E.H.,Pavelsky,T.M.,Bates,P.D.,Neal,J.C.,2017.The effects of spatialresolution and dimensionality on modeling regional-scale hydraulics in a multichannel river.Water Resour.Res.53(2),1683-1701.https://doi.org/10.1002/2016WR019396.

Apel,H.,Aronica,G.T.,Kreibich,H.,Thieken,A.H.,2009.Flood risk analyses:How detailed do we need to be?Nat.Hazards 49(1),79-98.https://doi.org/10.1007/s11069-008-9277-8.

Bates,P.D.,Horritt,M.S.,Fewtrell,T.J.,2010.A simple inertial formulation of the shallow water equations for ef ficient two-dimensional flood inundation modelling.J.Hydrol.387(1-2),33-45.https://doi.org/10.1016/j.jhydrol.2010.03.027.

Blayo,E.,Debreu,L.,2005.Revisiting open boundary conditions from the point of view of characteristic variables.Ocean Model.9(3),231-252.https://doi.org/10.1016/j.ocemod.2004.07.001.

DHI,2007.Mike 21 Flow Model:Hydrodynamic Module User Guide.DHI Water and Environment.

Falconer,R.A.,Chen,Y.P.,1991.An improved representation of flooding and drying and wind stress effects in a 2-D tidal numerical model.In:Proceedings of Institution of Civil Engineers,Part 2:Research and Theory 91.Institution of Civil Engineers,London,pp.659-678.

Fewtreel,T.J.,Duncan,A.,Sampson,C.C.,Neal,J.C.,Bates,P.D.,2011.Benchmarking urban flood models of varying complexity and scale using high resolution terrestrial lidar data.Phys.Chem.Earth 36(7-8),281-291.https://doi.org/10.1016/j.pce.2010.12.011.

Jonkman,S.N.,Penning-Rowsell,E.,2008.Human instability in flood flows.J.Am.Water Resour.Assoc.44(5),1208-1218.https://doi.org/10.1111/j.1752-1688.2008.00217.x.

Keller,R.J.,Mitsch,B.,1993.Safety Aspects of Design Roadways and Floodways.Urban Water Research Association of Australia,Melbourne.

Koch,S.E.,McQueen,J.T.,1987.A survey of nested grid techniques and their potential for use within the mass weather prediction model.In:NASATechnical Memorandum 87808.National Aeronautics and Space Administration.

Korres,G.,Lascaratos,A.,2003.A one-way nested eddy resolving model of the Aegean and Levantine basins:Implementation and climatological runs.Ann.Geophys.21,205-220.https://doi.org/10.5194/angeo-21-205-2003.

Liang,D.,Lin,B.,Falconer,R.A.,2007.Linking one-and two-dimensional models for free surface flows.Proc.Inst.Civ.Eng.Water Manag.160(3),145-151.https://doi.org/10.1680/wama.2007.160.3.145.

Marchesiello,P.,McWilliams,J.,Shchepetkin,A.,2003.Open boundary conditions for long-term integration of regional oceanic models.Ocean Model.3(1-2),1-20.https://doi.org/10.1016/S1463-5003(00)00013-5.

Miyakoda,K.,Rosati,A.,1977.One-way nested grid models:The interface conditionsand the numerical accuracy.Mon.WeatherRev.105,1092-1107.

Muis,S.,Gu¨neralp,B.,Jongman,B.,Aerts,J.C.,Ward,P.J.,2015.Flood risk and adaptation strategies under climate change and urban expansion:A probabilistic analysis using global data.Sci.Total Environ.538,445-457.https://doi.org/10.1016/j.scitotenv.2015.08.068.

Nash,S.,2010.Development of an Adaptive Mesh Inter-Tidal Circulation Model.Ph.D.Dissertation.National University of Ireland Galway,Galway.

Nash,S.,Hartnett,M.,2014.Development of a nested coastal circulation model:Boundary error reduction.Environ.Model.Softw.53,65-80.https://doi.org/10.1016/j.envsoft.2013.11.007.

Nguyen,P.,Thorstensen,A.,Sorooshian,S.,Hsu,K.,AghaKouchak,A.,Sanders,B.,Koren,V.,Cui,Z.T.,Smith,M.,2016.Ahighresolutioncoupled hydrologic-hydraulic model(HiResFlood-UCI)for flash flood modeling.J.Hydrol.541,401-420.https://doi.org/10.1016/j.jhydrol.2015.10.047.

Nycander,J.,Doos,K.,2003.Open boundary conditions for barotropic waves.J.Geophys.Res.108(C5),3168-3187.https://doi.org/10.1029/2002JC001529.

Olbert,A.I.,Hartnett,M.,2010.StormsandsurgesinIrishcoastalwaters.Ocean.Model.34(1-2),50-62.https://doi.org/10.1016/j.ocemod.2010.04.004.

Palma,E.D.,Matano,R.P.,1998.On the implementation of passive open boundary conditions for a general circulation model:The barotropic mode.J.Geophys.Res.103(C1),1319-1341.https://doi.org/10.1029/97JC02721.

Pullen,J.,Allen,J.S.,2001.Modeling studies of the coastal circulation off the northern coast of California:Statistics and patterns of wintertime flow.J.Geophys.Res.106(C11),26959-26984.https://doi.org/10.1029/2000JC000548.

Roed,L.P.,Cooper,C.,1987.A study of various open boundary conditions for wind-forced barotropic numerical ocean models.Elsevier Oceanography Series 45,305-335.https://doi.org/10.1016/S0422-9894(08)70454-9.

Xia,J.,Falconer,R.A.,Lin,B.,Tan,G.,2011.Numerical assessment of flood hazard risk to people and vehicles in flash floods.Environ.Model.Softw.26(8),987-998.https://doi.org/10.1016/j.envsoft.2011.02.017.

Zhang,D.L.,Chang,H.R.,Seaman,N.L.,Wamer,T.T.,Fritsch,J.M.,1986.A two-wayinteractivenestingprocedurewithvariableterrainresolution.Mon.Weather Rev.114,1330-1339.https://doi.org/10.1175/1520-0493(1986)114＜1330:ATWINP＞2.0.CO;2.

*Corresponding author.

E-mail address:Michael.Hartnett@nuigalway.ie(Michael Hartnett).

Peer review under responsibility of Hohai University.