A numerical modelling study of coastal flooding - Springer Link

Meteorol. Atmos. Phys. 80, 217±233 (2002)

1 2

CSIRO Atmospheric Research, PMB#1 Aspendale, Australia Global Environmental Modelling Systems, Warrandyte, Australia

A numerical modelling study of coastal ¯ooding Kathleen L. McInnes1 , Graeme D. Hubbert2 , Debbie J. Abbs1 , and Steve E. Oliver2 With 15 Figures Received October 16, 2001 Revised December 28, 2001 Summary

1. Introduction

A coastal ocean model capable of modelling tides, storm surge and the overland ¯ow of ¯oodwaters has been further developed to include the ¯ux of water from tributaries and the forcing from wave breaking that leads to wave setup in the nearshore zone. The model is set up over the Gold Coast Broadwater on the east coast of Australia. This complex region features a coastal lagoon into which ®ve tributaries ¯ow and is subject to ¯ooding from extreme oceanic conditions such as storm surge and wave setup as well as terrestrial runoff. Weather conditions responsible for storm surge, waves and ¯ooding include cyclones of both tropical and mid-latitude origin. Two events are modelled. The ®rst is an east coast low event that occurred in April 1989. This event veri®ed well against available observations and analysis of the model simulations revealed that wave setup produced a greater contribution to the elevated water levels than the storm surge. The second case to be modelled was tropical cyclone Wanda, responsible for the 1974 ¯oods. Modelled water levels in the Broadwater were reasonably well captured. Sensitivity experiments showed that storm surge and wave setup were only minor contributors to the elevated sea levels and their contribution was con®ned to the earlier stage of the event before the runoff reached its peak. The contribution due solely to runoff exhibited a tidal-like oscillation that was 180 out-of-phase with the tide and this was attributed to the greater hydraulic resistance that occurs at high tide. A simulation of this event with present day bathymetry at the Seaway produced sea levels that were 0.3±0.4 m lower than the simulation with 1974 bathymetry highlighting the effectiveness of deepened Seaway channel to reduce the impact of severe runoff events in the Broadwater.

Over 80% of the Australian population resides within 50 km of the coast and this ®gure is increasing. Coastal areas can be vulnerable to both terrestrial and oceanic ¯ooding particularly if they are low-lying. The Insurance Council of Australia estimates that ¯ood damage costs at least $4±500 million a year in Australia in total (including uninsured losses) and with an increasing demographic shift to coastal regions, this ®gure can only increase. Furthermore, as the effects of human induced climate change are experienced, it is likely that the extreme weather events that cause ¯ooding and severe ocean conditions will become more severe thereby increasing the vulnerability of coastal inhabitants to the risk of ¯oods and storm surges. There is therefore an urgent need to develop tools that can be used to model the complex processes leading to coastal ¯ooding and assess the risks involved for coastal communities. The coastal zone can experience ¯ooding from extreme rainfall, tides, storm surge and high waves. The weather events that produce signi®cant rainfall such as tropical or mid-latitude cyclones can also generate severe winds over the coastal ocean leading to storm surges and high waves. The present modelling study focuses on

218

K. McInnes et al.

Fig. 1. The map of the region showing places mentioned in the text

the Gold Coast Broadwater on the east coast of Australia which experiences storms of both tropical and mid-latitude origin. The Gold Coast Broadwater is a large coastal lagoon situated to the north of the Gold Coast (see Fig. 1). It is connected to the Paci®c Ocean via the Seaway at the southern end of South Stradbroke Island and via a semi-permanent channel at the northern end. A complex network of channels to the west of North Stradbroke Island also connects the Broadwater to the southern end of Moreton Bay. The Broadwater is an out¯ow region for a number of major rivers and streams including the Nerang, Coomera, Pimpama and Logan. The ¯oodplains of the Broadwater are extensively developed and a large number of dwellings are at risk of ¯ooding during major ¯ood events. Furthermore, increasing pressure to develop remaining ¯oodplain areas will reduce the ¯oodwater storage capacity during severe ¯ood events. Storm tides within the Broadwater can exacerbate ¯ood events by elevating sea levels at the out¯ow regions of rivers and streams thereby reducing their ¯ow rates. In many situations, the weather conditions that

cause storm tide events are also accompanied by severe rainfall. The establishment of ¯ood levels for planning and development purposes is of critical importance to minimize the risk of damage to infrastructure during such events. Previous studies conducted in this region have modelled the storm surges resulting from tropical cyclones (Harper et al., 1977), and determined return periods for storm tides (Blain et al., 1985), on the open coastline of the Gold Coast. Hydrological modelling of the Coomera and Nerang river systems that drain into the Broadwater has also been undertaken to establish return periods for peak ¯ow events. No studies to date have attempted to model the complex ¯ows within the Broadwater itself. Design storm tide levels within the Broadwater, up to now, have been derived from those calculated on the open coast by Blain et al. (1985) with an additional contribution due to wave set-up added to these levels. In the present study, Global Environmental Modelling Systems (GEMS) storm surge and inundation model, GCOM2D (Hubbert and McInnes, 1999a, b) has been set up over the entire

Numerical modelling study of coastal ¯ooding

Broadwater region to model the coastal ¯ooding caused by two severe storms. GCOM2D, which is capable of modelling the overland ¯ow of ¯ood waters due to tides and storm surge, has been further developed to incorporate contributions from riverine runoff and wave setup. A threedimensional, non-hydrostatic model (RAMS) is used to model the atmospheric conditions required for the simulation of waves and storm surge. This study is the ®rst step towards developing an integrated modelling system capable of modelling the various physical processes that contribute to coastal ¯ooding. In the ®rst event to be modelled, which occurred in April 1989, high coastal sea levels occur as a result of storm surge and waves. The second event to be modelled is tropical cyclone Wanda which occurred in January 1974 and caused severe ¯ooding throughout the region. In this event, runoff is the most signi®cant contributor to the coastal ¯ooding. The remainder of this paper is set out as follows: in Sect. 2, a brief discussion is given of the oceanic contributions to extreme sea levels. The weather events that affect the central east coast of Australia are described. Section 3 describes the various models used in the study. The results of the model simulations are presented in Sect. 4 and conclusions are presented in Sect. 5. 2. Meteorology and coastal impacts 2.1 Meteorology Southeastern Queensland experiences severe storms from both tropical and sub-tropical regions. Tropical cyclones are potentially the most severe class of storm to affect the region although they are relatively infrequent. For example, between 1953

219

and 1997 twenty-one tropical cyclones occurred between 26 S and 30 S and within 2 of the coast which translates to a rate of approximately one every two years. Four of the cyclones attained minima of less than 980 hPa. Depressions of subtropical origin, such as east coast lows, are more frequent but generally less intense than tropical cyclones. In the twenty years from 1977 to 1996, twenty-seven sub-tropical cyclones occurred in the region. However, their larger spatial scale and longer duration suggests that they are likely to be the most frequent contributor to elevated sea-level events in the region. Both storm systems are capable of generating severe rainfall, which if coincident with elevated coastal sea levels, can further increase the likelihood of coastal ¯ooding. 2.2 Coastal effects Increases in coastal sea levels during severe storm conditions are due to storm surges and breaking waves superimposed on the astronomical tides, as illustrated in Fig. 2. Storm surges are generated by wind stresses and, to a lesser extent, falling atmospheric pressure that produces a rise in water level at the rate of approximately 1 cm per hPa fall in pressure (the so-called inverse barometer effect). Wind stresses induce ocean currents that, if blocked by a coastal barrier, pile water against the coast to produce elevated sea levels. This process is commonly referred to as wind set-up. Alternatively, wind stresses acting in the longshore direction with suf®cient fetch and duration produce longshore currents that eventually become de¯ected to the left of the current direction in the Southern Hemisphere due to Coriolis effects. If a coastal barrier blocks the path of the de¯ected ¯ow, elevated sea levels can also occur due to a process referred to as current set-up or Ekman drift.

Fig. 2. Contributions to extreme sea-levels at the coast

220

K. McInnes et al.

Coastal bathymetry can also affect storm surge intensity. Shallow coastal bathymetry tends to amplify the storm surge while topographic features can channel the coastal currents. Along the southern Queensland coast, the narrow continental shelf is not conducive to large storm surge generation. Breaking surface waves in the coastal zone cause a change in momentum ¯ux that is balanced by an increase in the still water level and is referred to as wave set-up. Wave set-up at the coast is a function of wave height and direction at breaking which is dependent upon the wind direction, strength, duration and fetch as well as the bathymetry in shallow water. The contribution to extreme sea levels along the southern Queensland coast due to wave set-up is generally likely to be comparable to the storm surge effect in many circumstances. Astronomical tides, storm surges and wave set-up interact nonlinearly to produce total sea levels that are lower than the sum of the individual contributions. This is due to the friction of the ocean ¯oor that has a retarding effect on the ¯ow proportional to the square of the coastal current. For this reason, it is preferable to integrate the effects of all contributions to coastal currents and sea levels in a dynamical model simulation rather than adding the effects of some components separately at a later stage. 3. Models The models used in this study include the coastal ocean model, GCOM2D, the atmospheric model, RAMs for generating the meteorological conditions accompanying the storm events, and the third generation spectral wave model WAM. The interaction of the various components is illustrated in Fig. 3. 3.1 Coastal ocean model The coastal ocean model, GCOM2D (Hubbert and McInnes, 1999a, b) solves the depth-averaged hydrodynamic equations over a region de®ned by detailed topographic and bathymetric information to provide currents, sea level heights and overland ¯ow of water due to tidal and meteorological conditions. The inundation algorithm

Fig. 3. Diagram illustrating the modelling system used in the study

¯oods and drains at rates that are dependent on the modelled currents. It has been shown that sea levels simulated at the coast are more realistic when a moveable coastal boundary is used, compared to the simpler to implement, ®xed-coast storm surge models that tend to overestimate the coastal sea levels (Yeh and Chou, 1979; Hubbert and McInnes, 1999a, b). In the present study, GCOM2D is run over the two regions illustrated in Fig. 4. The larger, lower resolution, outer grid has a horizontal grid spacing of 1.0 km, while the smaller inner grid has a grid resolution of 100 m. Atmospheric forcing in the form of 10 m winds and surface pressure are interpolated both spatially and temporally from the atmospheric model grid to the coarse and ®ne mesh ocean model grids. Tidal constituents (tidal phases and amplitudes) for the M2, N2, K2, S2, K1, Q1 and O1 tides are obtained from a 30 minute global tidal model which models tidal currents and assimilates satellite measurements of ocean surface heights from the TOPEX=POSEIDON altimeters (Le Provost et al., 1995). These data are accurate to within 0.2±0.3 m in waters deeper than 200 m but are generally inaccurate in shallow coastal waters. As a result, GCOM2D was used to generate an improved set of tidal constituents along the Queensland coast prior to the modelling carried out in this study. A model simulation was carried out at 15 km resolution from 11 S to 30 S


221

Fig. 4a. Coarse 1500 m resolution (140 188 gridpoints), and b ®ne 100 m resolution (230 440 gridpoints) topographical grid used by GCOM2D. Bathymetric contours on a are shown every 50 m up to 500 m and then 500 m intervals thereafter, and on b every 5 m. Locations used throughout the report are marked, as are the river in¯ow points for the 1974 ¯ood event. (Note: The wave model was run on a larger open ocean grid)

for several months using tidal heights from the global tidal model as deep water boundary conditions. The results were then subjected to Fourier analysis at each model grid point to obtain an improved set of tidal phases and constituents in the shallow coastal waters. These values were then used to augment the global tidal data set. The tides are applied as an amplitude change in water levels on the lateral boundaries of the outer model grid. The action of breaking waves in the surf zone causes a shoreward momentum ¯ux which leads to a mean increase in water level known as wave setup. As shown in Battjes and Janssen (1978), wave setup can be related to the signi®cant wave heights of deep water waves offshore via the calculation of horizontal gradients of the wave radiation stress. These are incorporated as an additional forcing term analogous to the wind stress term in the horizontal momentum equations of GCOM2D (see, for example Mastenbroek et al., 1993). Deep water waves are calculated using a wave model. The algorithm used to calculate the proportion of waves that break as they approach the shore and the wave radiation stress is described in McInnes et al. (2000). The effects of freshwater runoff can also be modelled by GCOM2D by specifying a ¯ow hydrograph (a time series of the ¯ux of water) at a particular location along a river channel. For

reasons of computational stability, each channel in the model domain is widened to four gridpoints. The water ¯ux is converted to a current directed parallel to the channel in a downstream direction. This is achieved by multiplying the ¯ux by the cross-sectional area of the channel and dividing by the number of gridpoints across the channel. Mass balance is monitored throughout the simulation to ensure that the mass added to the system via ¯uxes at the boundaries (open ocean boundaries as well as tributaries) agrees with the incremental mass change integrated across the computational domain. 3.2 Wave model The wave parameters (height, direction and period) required for the radiation stress calculation are simulated using the third generation spectral ocean wave model WAM (The WAMDI group, 1988). The wave model is run over a region spanning 26 S to 29.25 S and 152.8 E to 156 E at a resolution of 0.05 (approximately 5 km). Time series of the wave parameters are then interpolated to the GCOM2D ®ne resolution grid. 3.3 Atmospheric model The numerical model used in this study is the Colorado State University Regional Atmospheric

222

K. McInnes et al.

Modeling System (CSU RAMS) described in detail by Tripoli and Cotton (1982), Cotton et al. (1982, 1986). This is a quasi-Boussinesq model with predictive equations for the three velocity components, the Exner function, the ice-liquid water potential temperature (see Tripoli and Cotton, 1981), the total water mixing ratio and the mixing ratios of rain droplets, pristine ice crystals, snow, graupel and aggregate particles. Diagnosed quantities are potential temperature, temperature, pressure and the mixing ratios of water vapor and cloud droplets. The equations are solved on the staggered grid described by Tripoli and Cotton (1982). A terrain-following sigma-z coordinate (Gal-Chen and Somerville, 1975a; 1975b) is employed for the vertical direction. In our study, the non-hydrostatic, compressible version of the modelling system is used. This uses a time-splitting scheme in which the acoustic terms are integrated on a short time step using the Crank-Nicholson scheme. All other terms are integrated on a long time step using the leap frog procedure. Long and short wave radiation effects are parameterised using the scheme described by Chen and Cotton (1983b, 1983a). The surface ¯uxes of heat, water vapor and momentum were calculated using the scheme of Louis (1979) and mixing is parameterized using a modi®ed form of the eddy viscosity type scheme. The soil model of McCumber and Pielke (1981) was coupled to the atmosphere using the surface energy balance model of Tremback and Kessler (1985). Cloud microphysics are parameterised using the scheme described by Flateau et al. (1989) in which ice crystals are initiated whenever the cloud becomes water saturated and the temperature is below 0 C. Activated crystals have an initial diameter of 12.9 mm corresponding to an initial mass of 10 9 g. It was necessary to couple RAMS to a convective parameterisation scheme that was suitable for use at resolutions of the order of 10 km. The convective parameterisation scheme chosen was that of Frank and Cohen (1985, 1987), and Chen and Frank (1993). This scheme has been chosen as it has the advantage of coupling the convective parameterisation scheme with the explicit microphysics parameterisation. This is accomplished by detraining ``convective'' hydrometeors to the grid scale and in doing so allows for the subsequent phase changes of these hydrometeors.

Fig. 5. The grid domains used by the atmospheric model

In the simulations discussed here, the numerical model has been initialized using analyses from the U.S. National Centers for Environmental Prediction (NCEP). These analyses are interpolated horizontally and vertically to the outer model grid that has a resolution of 45 km. The analyses also provide the temporal forcing on the lateral boundaries of the outer model grid. Three levels of interactive grid nesting were used, the middle and ®nest resolution grids having a horizontal grid spacing of 15 km and 5 km respectively. The grid domains are shown in Fig. 5. The terrain used on all model grids was interpolated from a 1=40th degree data set. The sea-surface temperatures (SST) were obtained from the NCEP analyses. 4. Model results In this section, two severe events are modelled in detail and the results are presented and discussed. The ®rst event is studied to illustrate the running and the performance of the modelling system used in this study. It was chosen because it produced high sea levels that led to localised ¯ooding in the study region that was not attributed to severe rainfall runoff. An investigation of the relative sources of storm tide and ¯ood waters in the Broadwater is undertaken and the key sources of error are identi®ed. The second event involves modelling tropical cyclone Wanda to reconstruct the processes contributing to the 1974 ¯oods in


the Broadwater. The ¯oods resulting from this event are the worst to affect southeastern Queensland during the twentieth century. 4.1 Case 1: 25 April 1989 This event commenced as a tropical low some ®ve hundred kms north of the Gold Coast on 24 April. The low subsequently travelled south along the coast ®nally to be located to the east of the Gold Coast early on 26 April. Gale to storm

223

force winds caused severe erosion along the Sunshine coast to the north of Moreton Bay while heavy rainfall produced widespread ¯ooding. Sea levels peaked at 0.43 m and 0.5 m above predicted tidal levels in the Broadwater and at Brisbane Bar, respectively. 4.1.1 Atmospheric model results A three-day simulation of this event commencing at 0000 UTC 23 April 1989 was performed using the RAMS atmospheric model. Figure 6 shows

Fig. 6a, b. Manually drawn Bureau of Meteorology analyses at 0000 UTC and 1200 UTC 25 April 1989, respectively, and c and d RAMS model simulation conducted at 15 km resolution at the corresponding times

224

K. McInnes et al.

the MSLP during the ®nal 24 hours of the simulation along with verifying analyses from the Bureau of Meteorology. At 0000 UTC 25 April (Fig. 6) the simulated low is deeper and situated slightly further to the south than indicated in the analyses although the pressure gradient along the Gold Coast is generally well captured. At 1200 UTC there is closer agreement in the location and intensity of the low in the vicinity of the Gold Coast. Modelled wind speed and direction are compared with three hourly observations at Cape Moreton and the Gold Coast in Fig. 7. Modelled winds at Cape Moreton are weaker than observations particularly in the latter 24 hours of the simulation. This is consistent with the modelled depression being located further to the south at around this time. Closer agreement can be seen at the Gold Coast. Wind direction indicates sustained easterly ¯ow for the ®rst 60 hours at

both locations and this is well captured by the model. 4.1.2 Tidal model results Prior to running GCOM2D with atmospheric forcing, a tides-only simulation was carried out over the coarse resolution model domain to ascertain how well tidal variations are represented. These are compared with the tides predicted using the known tidal phases and amplitudes. Results for the three-day interval commencing 0000 UTC 23 April 1989 are shown for the Brisbane Bar and the Seaway in Fig. 8. Also shown are the observed sea levels indicating the degree of meteorological in¯uence on the coastal ocean during this period. Tides are predominantly semi-diurnal in this region and the tidal range at Brisbane Bar is approximately 0.6 m greater than that at the Seaway. This may

Fig. 7. Modelled and observed wind speed and direction at Cape Moreton a and b, and the Gold Coast c and d


225

Fig. 8. Predicted and modelled tides at a Brisbane Bar, and b the Gold Coast Seaway. For comparison, the observed sea levels (also containing contributions from storm surge and wave setup) are shown. The simulation commences at 0000 UTC 23 April 1989

be due to the hydraulic resistance created by the channel islands at the southern end of the Bay. Model simulations conducted using GCOM2D underestimate the higher high tide by approximately 0.1 m at Brisbane Bar. However values at low tide and lower high tide are well captured. At the Gold Coast Seaway, much closer agreement is seen between the modelled and predicted tides at high tide whereas the lower high tide is overestimated by GCOM2D by about 0.1 m. There is a slight phase error between the modelled and predicted tides with the modelled tide leading the predicted tide by up to an hour. The level of agreement between the GCOM2D tides and those predicted at the tide gauge, is noteworthy considering the errors in the global tidal constituent data used on the open boundaries of CGOM2D. It indicates ®rstly, the value in using GCOM2D to generate improved tidal phase and constituent data in the coastal region, and secondly, its ability to accurately model the tidal dynamics in the shallow coastal regions when nested within these improved tidal boundary conditions. 4.1.3 Wave model results The incorporation of radiation stress forcing in GCOM2D requires the simulation of wave heights, directions and periods. For this event, a 72-hour simulation of the wave conditions for the April 1989 event was conducted using the third generation wave model WAM. Figure 9 shows the simulated wave heights, directions and periods at

0000 UTC 25 April 1989. Wave heights exceed 6 m over a region immediately to the east of Moreton Island extending southwards to the Gold Coast and are directed onshore and the wave periods in this region are in excess of 8 seconds. These compare with signi®cant wave heights of up to 6.5 m recorded off Moreton Island with wave periods of 10.6 s (Allan and Callaghan, 1998). 4.1.4 Storm surge and wave setup results Three ®ne mesh model simulations were conducted. The ®rst incorporated tidal forcing only, the second had tides and wind forcing and the third also contained wave radiation stresses. Time series for the third simulation at the Seaway and the northern and southern arm of the Coomera River are shown in Fig. 10 along with the observed sea level at the Seaway. Note that only the last 24 hours of the simulation, when the highest sea levels occurred are shown. Comparing the modelled and observed sea levels at the Seaway reveals that a slight phase difference is evident and is consistent with that seen on the coarse resolution tides-only simulation. The modelled sea level heights are also underestimated by about 0.1 m. The underestimation of sea levels by the model, particularly in the last 12 hours, may be due to an underestimation of the wind strength along with a shift to northeasterlies in the atmospheric model simulation that was not seen in the observations.

226

K. McInnes et al.

Fig. 9a. Wave height (in metres) and direction, and b wave period (in seconds) at 0000 UTC 25 April 1989 as simulated by the WAM model

Fig. 10. Modelled and observed sea levels at the locations indicated. The storm surge and wave setup components of the total sea level signal at the Seaway are also shown. The time series commences on 0000 UTC 25 April 1989

The sea levels at the northern and southern arms of the Coomera River show lower sea-level peaks occurring approximately one hour later. This phase delay is consistent with the observed tidal signals at the Seaway and Paradise Point at the southern arm of the Coomera (not shown). To investigate the contribution due to wave setup, the sea level results from the tides and surge simulation at the Seaway were subtracted from the full simulation that also included wave radiation stresses. The resulting residual is shown in Fig. 10 and indicates that the contribution

from wave setup was relatively uniform during the 24-hour period and accounted for approximately 0.25±0.30 m of the total sea level. The contribution due to the storm surge was obtained by subtracting the sea levels due to tides-only from the tides and surge simulation. The result indicates that the storm surge contributed between 0.10±0.15 m to the total sea level. This ®nding is consistent with the results of a study into the relative contributions of storm surge and wave setup to the sea levels recorded along the NSW coast during severe east coast low events (McInnes and Hubbert, 2001). Figure 11a shows the current vectors at 1200 UTC on April 1989 corresponding to high tide. A northerly current exceeding 0.5 m s 1 is evident along the east coast while in southern Moreton Bay, it is southerly. Flow at Jumpinpin and the Seaway is into the Broadwater. Currents shown six hours later (Fig. 11b), indicate that the tides have turned and ¯ow is to the north in Moreton Bay. Flow at Jumpinpin and the Seaway are directed towards the east. Height contour charts (not shown) indicate that there was only minimal inundation of low-lying land within the Broadwater. 4.2 Case 2: Tropical cyclone Wanda Tropical cyclone Wanda was responsible for major ¯ooding from Brisbane to the Gold Coast. Severe beach erosion also occurred along the


227

Fig. 11. Modelled currents at a 1200 UTC 25 April, and b 1800 UTC 25 April. Grid labels are in kms from origin

south Queensland coast. Freshwater runoff was a major factor in the ¯ooding within the Broadwater. As a result, the fresh water ¯ux derived previously from various hydraulic modelling studies (Betts, private communication) is also incorporated into GCOM2D. Hydraulic models contain detailed information of the crosssectional shape and hydraulic features such as bridges and culverts present along a length of a river. Using physical and empirical relationships, the characteristics of the river ¯ow are determined at points along the river. The input to a hydraulic model may be the water ¯uxes obtained by running a hydrological model that converts rainfall measurements over a catchment into a water ¯ux within a river. The ¯ow hydrographs for the ®ve major rivers that ¯ow into the Broadwater are shown in Fig. 12 and indicate that the combined peak discharge from the rivers was around 11000 m3 s 1. The locations at which these ¯uxes are incorporated into GCOM2D is indicated in Fig. 4b (note that the Coomera ¯ux is divided across the two branches). In 1974 the Gold Coast Seaway consisted of a much shallower channel situated to the north of its present day location. The bathymetric data used in GCOM2D was modi®ed to incorporate this difference although the event is also simulated with the present day bathymetry for comparison purposes.

Fig. 12. Volume of water ¯ow along the major rivers that ¯ow into the Broadwater during the 1974 ¯ood

4.2.1 Atmospheric model results At 2300 UTC 23 January 1974 (Fig. 13a) a tropical depression embedded in a strong monsoon trough approached the Queensland coast to the north of Fraser Island at around 25 S. By 1100 UTC 24 January 1974, the depression had intensi®ed to 1000 hPa as it made landfall producing strong winds to the south of the low centre (Fig. 13b). A 6.5 day simulation of the event commencing at 1200 UTC 23 January 1974 was performed using the RAMS atmospheric model. At 12 hours into the model simulation (Fig. 13c), the trough of low pressure has deepened from 1010 hPa at the initialization time to 1006 hPa. This is slightly

228

K. McInnes et al.

Fig. 13a, b. Manual analyses of mean sea-level pressure at the times indicated with observed surface wind vectors also shown; c,d RAMS model simulations of mean sea-level pressure and 10 m wind vectors at the corresponding times. (Note that only a subset of the RAMS model grid corresponding to the area covered by the analyses is shown)

weaker than the observed low of 1004 hPa although the location of the low is well captured. At 1100 UTC 24 January, the model has captured the broad monsoon trough but has failed to capture the development of the low-pressure system. The observed winds associated with the low pressure system in Fig. 13a are from east to south-east and have strengths between 10 and 18 m s 1. The strongest winds along the Queensland coastline lie between the northern tip of Fraser Island and Moreton Bay. The model also predicts the strongest winds to occur in this

region with wind speeds between 12 and 16 m s 1 (Fig. 13c). Both the model and analyzed winds are much weaker inland with speeds less than 10 m s 1. The region with winds greater than 12 m s 1 has moved towards the coast since the initial time. By 1100 UTC 24 January wind speeds along the Gold Coast are between 12± 17 m s 1, having strengthened from 10 m s 1 observed 12 hours earlier. The model also moves the region of strong winds to the south with a broad region of winds in excess of 12 m s 1 occurring between 25 S and 29 S. Due to the


failure of the model to simulate the development of the closed low, the winds along the Gold Coast are from the ENE rather than the ESE or E, as indicated in the observations. At later times, manual analyses were not available for comparison with the model simulation. However, comparison of measured wind speed and direction at Brisbane airport with model simulated winds (not shown) indicates that the modelled winds increasingly gained a NE component whereas the observations indicated winds uniformly from the E or ESE up to about 0000 UTC 26 January. The greatest impact of the increasing northerly component in the modelled winds would be a reduction in the storm surge along the open coast compared with that generated by winds directed onshore or from the south. Recent simulations of this case have shown the importance of using ``bogus'' tropical cyclone winds to initialize the atmospheric model. In these simulations the model simulated a closed low, resulting in E to ESE winds along the Gold Coast. The modelled daily rainfall for the three days of most intense rain as well as the 5-day total for the period 25±29 January has been compared with the observations (not shown). This comparison indicates a number of similarities between the observed and modelled rainfall but also reveals some important differences. The model reproduced the two main regions of highest rainfall, one in the mountainous region north of the Gold Coast and the second in the coastal ranges inland from the Gold Coast. The evolution of the modelled rainfall is similar to that of the observed. In both cases, extreme rainfall occurs during the 24 hours ending 25 January, intensi®es to produce the most extreme rainfall during the 24 hours ending 26 January and then slightly weakens to produce lighter, but still extreme rainfall on the 27th. The model also produced close to the highest observed rainfall with rainfall in excess of 390 mm predicted on the 26th. For the 5-day period, the model predicted in excess of 1100 mm, compared with an observed total of 1200 mm. The main difference between the modelled and observed rainfall is in the areal extent of the region of heaviest rainfall. For each of the days simulated, the model concentrated the extreme rainfall over a much smaller area than observed. The area of predicted highest rainfall,

229

inland from the Gold Coast, occurred further south than observed. These model results highlight some of the dif®culties that are encountered when trying to use even high-resolution atmospheric model rainfall predictions to force hydrological models. Despite the qualitative agreement between the predictions and the observations, these rainfall predictions were unable to produce any run-off in the catchment of interest. Work is continuing to determine if the rainfall predictions can be improved for this and other high rainfall events. 4.2.2 Ocean response Three simulations of GCOM2D were carried out over 72 hours commencing 0000 UTC 24 January 1974 using bathymetry modi®ed in the vicinity of the Seaway to approximate 1974

Fig. 14a. Observed sea levels inside the Seaway and corresponding modelled sea levels using 1974 bathymetry at the Seaway (note: there is missing data at around 48 hours). Also shown are contributions due to freshwater runoff and saltwater (storm surge and wave setup); b as for a except using present day bathymetry

230

K. McInnes et al.

conditions. The ®rst was a tides-only simulation, the second incorporated tides, storm surge and wave setup (referred to here as the saltwater simulation) and the third was a full simulation including freshwater runoff from the key river systems ¯owing into the Broadwater as well as the saltwater contribution. Figure 14a compares the water levels from the full simulation of GCOM2D with those measured in the southern Broadwater. Modelled water levels show reasonable agreement with observations although low tides and lower high tides are slightly overestimated. The higher high tide at 24 hours into the simulation is underestimated by about 0.2 m. Examination of the relative contributions to the total water level are also examined. The contribution due to storm surge and wave setup, calculated by subtracting the time series of the tides-only simulation from the saltwater simulation, is shown in Fig. 14a. The maximum contribution of about 0.4 m occurs at around 12 hours into the simulation and decreases thereafter. A further breakdown of this time series to determine the storm surge component (not shown) indicates that the storm surge reaches a maximum of 0.1 m at about 12 hours and decreases to zero by 24 hours. The lack of a storm surge signal from 24 hours onwards may re¯ect the northerly bias in the modelled winds at around this time compared

with observations as discussed previously, and may account for the underestimation of the total sea levels at around this time. The wave setup may also be underestimated slightly. The ¯ux hydrographs shown in Fig. 12 indicate that the latter 36 hours of the simulation is when the greatest contribution to Broadwater sea levels due to freshwater in¯ux may be expected. The modelled freshwater contribution, calculated by subtracting the time series of the saltwater simulation from the full simulation, is also shown in Fig. 14a and indicates that indeed the largest freshwater contribution occurs in the latter 36 hours. It is noteworthy that the freshwater time series exhibits a tidal-like oscillation that is 180 out-of-phase with the tide. This is presumably a response to the greater hydraulic resistance that occurs at high tide and causes the water to back up and ¯ow onto adjacent ¯oodplains. Figure 14b shows the water levels attained in a model simulation that utilizes present day bathymetry at the Seaway. Water levels are around 0.3 m lower at high tide and 0.4 m lower at low tide. This relative difference may re¯ect more effective drainage occurring at low tide as a result of the deeper Seaway. Figure 15a shows the current vectors at 1800 UTC on 25 January corresponding to low tide in the present day bathymetry simulation. A

Fig. 15. Modelled currents at a 1800 UTC 25 January, and b 0000 UTC 26 January. The contour de®ning the ¯ooded land area is also shown. Grid labels are in kms from origin


southerly current of around 0.5 m s 1 is evident along the east coast while in southern Moreton Bay, ¯ow is northward. At Jumpinpin and the Seaway, easterly ¯ow of about 1.0 m s 1 is evident. Currents shown six hours later (Fig. 15b) just prior to high tide, indicate dramatically reduced in¯ow in southern Moreton Bay, and weaker easterly ¯ow at Jumpinpin and the Seaway. Clearly the high water levels within the Broadwater are driving currents in these regions that counter the tidal currents. The extent of the ¯ooded land is also shown in Fig. 15 covers an area of approximately 120 square km. A comparison of modelled water levels in the Broadwater and on the open coast at high tide is made in the latter half of the simulation when the freshwater ¯ux is at a maximum. This indicates that the Broadwater levels are around 0.5 m higher than water levels on the open coast. This is the opposite tendency to the saltwater events in which the stormtide level on the open coast is typically about 0.1 m higher that the Broadwater levels. Clearly, the source of the coastal ¯ooding has implications for the severity of the ¯ooding experienced within the Broadwater.

5. Summary and conclusions In this study, the coastal ocean and inundation model GCOM2D, capable of modelling tides, storm surge and overland ¯ooding has been further developed to incorporate two additional physical contributions to coastal ¯ooding. The ®rst of these is the forcing created by breaking waves in the surfzone (the so-called wave radiation stress) that leads to wave setup. The second is the contribution due to rainfall run-off via a ¯ux imposed near the downstream end of a river channel. Wave conditions required as input were simulated using a third generation wave model while ¯ow hydrographs were obtained from previously run hydraulic models. Tidal constituent data were obtained from a global tidal model. However, since the tidal data is known to contain large errors in shallow coastal regions, improved constituent data along the east coast of Australia were generated using GCOM2D. This was achieved by simulating several months of tides over a large portion of the east coast and fourier

231

analysing the sea levels at each model grid point to generate an improved set of constituents. Detailed simulations were carried out of two cases: the April 1989 east coast low and tropical cyclone Wanda responsible for the 1974 ¯oods. The modelling system was able to reproduce the April event with a high degree of accuracy. Simulations suggested that the greatest contribution to sea levels during the east coast low event was due to the wave setup rather than the storm surge. To simulate the 1974 ¯ood event, data from ¯ow hydrographs for the ®ve main river systems ¯owing into the Broadwater were also incorporated as boundary conditions to GCOM2D along with the tidal, wind and wave forcing. In addition, the bathymetry in the vicinity of the Seaway was modi®ed to re¯ect the shallower channel that existed at that time. The model simulations of the water levels were reasonably well captured. The storm surge and wave setup made their largest contribution to the water levels during the ®rst half of the simulation. It is likely that the storm surge and to a lesser extent, the wave setup were underestimated due to errors in the modelled wind ®eld. Nesting conditions for the atmospheric model were based on NCEP analyses that are created by assimilating available meteorological observations into a forecast model to produce more detailed meteorological analyses. Limited open ocean data available in 1974 would lead to less accurate analyses over ocean regions. This conclusion is further supported by the fact that the wind directions in the atmospheric model showed a bias to the northeast. Other factors that could provide a minor contribution to differences between the model and observations in the present study may include omission of runoff from minor tributaries and rainfall directly over the catchment. During the latter half of the simulation, freshwater runoff was the main contributor to the sea levels in the Broadwater. This particular event highlights the fact that rainfall has the capacity to considerably worsen the impact of extreme storm tide events in this region. As a consequence of this, future efforts to determine the risk of severe ¯oods in the Broadwater should incorporate freshwater runoff in combination with wave setup and storm surge to be inclusive

232

K. McInnes et al.

of all the major physical processes. A simulation of the 1974 ¯ood event with present day bathymetry produced sea levels that were between about 0.3 and 0.4 m lower than those attained with the 1974 bathymetry. This illustrates the effectiveness of deepening the Seaway channel to minimise the ¯ooding from future ¯ood events in which severe runoff is the major contributor. This study represents the ®rst step towards the development of an integrated modelling system that links together a range of models representing the various contributions to coastal ¯ooding. This study has focussed on the ability of the coastal ocean model GCOM2D to model severe ¯ood events with contributions due to wave setup and rainfall runoff. In ongoing work, a hydrological model will be linked to the system so that modelled rainfall can be routed through the catchment to provide ¯uxes for GCOM2D. Acknowledgments The authors wish to acknowledge valuable information provided by Mr. Haydn Betts of the Gold Coast City Council, Lawson and Treloar (Brisbane), The Queensland Department of Transport and Mr. Jeff Callaghan of the Bureau of Meteorology (Brisbane). They are also grateful for valuable discussions with Mr. Haydn Betts, Dr. Tom Beer and Mr. Jeff Callaghan.

References Allan M, Callaghan J (1998) Extreme wave conditions for the south Queensland coastal region. Environment Technical Report, 32, Environment Protection Agency, 28 pp Battjes JA, Janssen JPFM (1978) Energy loss and setup due to breaking in random waves. Proc 16th Int Conf on Coastal Eng, Hamburg, pp 569±587 Blain, Bremner, Williams (1985) Storm tide statistics ± surfers paradise region. Report prepared for the Beach Protection Authority, Queensland, January 1985, 15pp Chen C, Cotton WR (1983a) Numerical experiments with a one-dimensional higher order turbulence model: Simulation of the Wangara-Day 33 Case. Bound Layer Meteor 25: 375±404 Ð (1983b) A one-dimensional simulation of the stratocumulus-capped mixed layer. Bound Layer Meteor 25: 289±321 Chen SS, Frank WM (1993) A numerical study of the genesis of extratropical convective mesovortices. Part I: Evolution and dynamics. J Atmos Sci 50: 2401±2426

Cotton WR, Stephens MA, Nehrkorn T, Tripoli GJ (1982) The Colorado State University three-dimensional cloud= mesoscale model ± 1982. Part 2: An ice parameterization. J Rech Atmos 16: 295±320 Cotton WR, Tripoli GJ, Rauber RM, Mulvihill EA (1986) Numerical simulation of the effects of varying ice crystal nucleation rates and aggregation processes on orographic snowfall. J Clim Appl Meteor 25: 1658±1680 Flatau PJ, Tripoli GJ, Verlinde J, Cotton WR (1989) The CSU-RAMS cloud microphysics module: General theory and code documentation. Colorado State University, Dept. of Atmospheric Science, Paper No. 451 Frank WL, Cohen C (1985) Properties of tropical cloud ensembles estimated using a cloud model and an observed updraft population. J Atmos Sci 42: 1911±1928 Ð (1987) Simulation of tropical convective systems. Part 1: A cumulus parameterization. J Atmos Sci 44: 3787±3799 Gal-Chen T, Somerville RCJ (1975a) On the use of a coordinate transformation for the solution of the NavierStokes equations. J Comp Phys 17: 209±228 Ð (1975b) Numerical solution of the Navier-Stokes equations with topography. J Comp Phys 17: 276±310 Harper BA, Sobey RJ, Stark KP (1977) Numerical simulation of tropical cyclone storm surge along the Queensland coast, Part X ± Gold Coast. Department of Civil and Systems Engineering, James Cook University, Townsville, 15pp Hubbert GD, Leslie LM, Manton MJ (1990) A storm surge model for the Australian region. Quart J Roy Met Soc 116: 1005±1020 Hubbert GD, McInnes KL (1999a) A storm surge inundation model for coastal planning and impact studies. J Coastal Research 15: 168±185 Hubbert GD, McInnes KL (1999b) Modelling storm surges and coastal ocean ¯ooding. In: Modelling coastal sea processes (Noye, BJ, ed) World Scienti®c Publishing, pp 159±187 Kain JS, Fritsch JM (1998) Multiscale convective overturning in mesoscale convective systems: reconciling observations, simulations and theory. Mon Wea Rev 126: 2254±2273 Le Provost C, Bennett AF, Cartwright DE (1995) Ocean tides for and from TOPEX=POSEIDON. Science 267: 639±642 Mastenbroek C, Burgers G, Janssen PAEM (1993) The dynamical coupling of a wave model and a storm surge model through the atmospheric boundary layer. J Phys Oceang 23: 1856±1866 McCumber MC, Pielke RA (1981) Simulation of the effects of surface ¯uxes of heat and moisture in a mesoscale numerical model-Part 1: Soil layer. J Geophys Res 86: 9929±9938 McInnes KL, Hubbert GD (2001) The impact of eastern Australian cut-off lows on coastal sea levels. Meteorol Appl 8: 229±243 McInnes KL, Hubbert GD, Oliver SD, Abbs DJ (2000) Gold Coast Broadwater Study: Storm tide return periods and

Numerical modelling study of coastal ¯ooding 1974 ¯oodwater modelling. Report to Gold Coast City Council. May 2000, 58 pp Tremback CJ, Kessler R (1985) A surface temperature and moisture parameterization for use in mesoscale numerical models. Preprints, 7th Conf on Num Wea Pred, Montreal, pp 17±20 Tripoli GJ, Cotton WR (1981) The use of ice-liquid water potential temperature as a thermodynamic variable in deep atmospheric models. Mon Wea Rev 109: 1094±1102 Ð (1982) The Colorado State University three-dimensional cloud=mesoscale model ± 1982. Part 1: General theoretical framework and sensitivity experiments. J Rech Atmos 16: 185±220

233

The WAMDI Group (1988) The WAM Model ± A Third Generation Ocean Wave Prediction Model. J Phys Oceang 18: 1775±1810 Yeh G-T, Chou F-K (1979) Moving boundary numerical surge model. J Waterw Port Coastal and Ocean Div 105: 247±263 Authors' addresses: Kathleen McInnes and Debbie Abbs, CSIRO Atmospheric Research, PMB#1 Aspendale, Victoria, Australia 3195 (E-mail: [email protected]); Graeme Hubbert and Steve Oliver, Global Environmental Modelling Systems, PO Box 149, Warrandyte, 3113, Australia

Verleger: Springer-Verlag KG, Sachsenplatz 4±6, A-1201 Wien. ± Herausgeber: Prof. Dr. Reinhold Steinacker, Institut fuÈr Meteorologie und Geophysik, UniversitaÈt Wien, Althanstraûe 14, A-1090 Wien. ± Redaktion: Innrain 52, A-6020 Innsbruck. ± Satz und Umbruch: Thomson Press (India) Ltd., Chennai. ± Druck und Bindung: Grasl Druck&Neue Medien, A-2540 Bad VoÈslau. ± Verlagsort: Wien. ± Herstellungsort: Bad VoÈslau. ± Printed in Austria.