This study first compares two methods by which the global, variable resolution Cubic Conformal Atmospheric Model (CCAM) is forced by reanalysis over Australia. The methods are the spectral nudging and bias-corrected sea surface temperature (SST) forcing. Surface winds and sea level pressure are compared since these influence coastal sea levels. SST forcing was found to better preserve the mean and standard deviation of these quantities. Second, a hydrodynamic model is used to model sea levels over southern Australia over 1980–1999 and 2080–2099 to investigate how changes in weather patterns affect extreme sea levels. Forcing from one Global Climate Model (GCM) and two CCAM simulations in which CCAM was used to downscale two GCMs over Australia with bias-corrected SST forcing (including the GCM considered in this study) were used. While there are differences in the spatial patterns of change between seasons over the modeled coastline between the three models, extreme sea levels were mostly lower in the future period over the southern mainland coastline from autumn to spring due to reduced westerlies in the climate models. The sea level changes around Tasmania varied from positive to negative depending on the model and season. The projected extreme sea level changes were within 10 cm of current climate values. This suggests that over southern Australia extreme sea level changes will be dominated by changes in mean sea level due to thermal expansion and ice sheet and glacier melt rather than changes in weather patterns.
 Extreme sea level events have had enormous social and economic impacts on low-lying coastal communities often resulting in significant loss of lives and property damage [e.g., Gerritsen, 2005; Webster, 2008; Paul, 2009]. Globally, many of the so-called mega cities are located in close proximity to the sea and, furthermore, population numbers are increasing rapidly in these locations. In Australia, it is estimated that roughly 50% of the population lives within 7 km of the coast [Chen and McAneny, 2006].
 Observations from tide gauges indicate that sea levels rose at a rate of around 1.7 mm/yr globally over the 20th century and since 1993 this rate has increased to around 2.8–3.2 mm/year based on tide gauges and satellite altimetry [Church and White, 2011]. Intergovernmental Panel for Climate Change (IPCC) estimates of future increases in mean sea level due to thermal expansion and melting ice caps, are projected to lie in the range of 18–79 cm by 2090–2099 relative to 1980–1999 [Bindoff et al., 2007] based on climate modeling and the inclusion of an additional 10 to 20 cm allowance for a possible rapid dynamic response of the Greenland and West Antarctic ice sheets. Even higher estimates of sea level rise by 2100 have been suggested using statistical methods that relate sea level changes to temperature changes [e.g., Vermeer and Rahmstorf, 2009; Grinsted et al., 2010]. Concerns about the impacts of rising sea levels have prompted several recent studies into observed changes in extreme sea levels [e.g., Menéndez and Woodworth, 2010] and the implications of projected future sea level rise on extreme sea levels [e.g., Bernier and Thompson, 2006; Hunter, 2010; Brown et al., 2010; McInnes et al., 2009, 2012b]. Building on these, studies carried out nationally [e.g., Department of Climate Change, 2009] and regionally [e.g., Bernier et al., 2007; McInnes et al., 2012b] have combined extreme sea level information with coastal elevation data to investigate potential future inundation.
 In addition to projected rises in mean sea level, storm events that generate extreme sea levels may also undergo changes in frequency and severity as a result of global warming leading to future changes in the frequency and severity of storm surges. Along the European and UK coastline, hydrodynamical downscaling of climate models has been used to investigate these changes [Lowe and Gregory, 2005; Debernard and Roed, 2008; Wang et al., 2008; Sterl et al., 2009]. These studies highlight that differences in the projected changes in weather conditions between different General Circulation Models (GCMs) and Regional Climate Models (RCMs) lead to large regional differences in the magnitude and even the sign of changes in sea level extremes. For example, Lowe and Gregory  used the Hadley Centre regional atmospheric model (HadRM3H), which downscaled the global atmospheric model (HadAM3H) under the Special Report on Emissions Scenarios (SRES) A2 and B2 scenarios [Nakićenović and Swart, 2000], to provide atmospheric forcing for a hydrodynamic model around the UK and North Sea. They found that increases in storm surge height due to changes in winter storms were most pronounced on the European North Sea coast, the English Channel and the southeast and northwest English coast while decreases in storm surge height were found on the western Irish coast. However, in comparison to previous studies with the same hydrodynamic model but using different climate models for forcing, large differences in the spatial patterns of change were noted. A similar finding was also noted in Debernard and Roed  who used a hydrodynamic model to investigate storm surge changes over northern Europe in four regionally downscaled GCMs. This points to the need to consider a range of climate model simulations to explore the uncertainties in developing future climate projections.
 In Australia, the potential for changes in weather patterns to bring about changes in storm surges, assessed by forcing a hydrodynamical model directly with climate model output, has not been previously attempted mainly due to the previously limited availability of climate model wind and pressure data at sufficient spatial and temporal resolution. However, simple wind scaling assumptions that incorporated the range of climate model wind speed changes were applied to investigate the relative influences of sea level and wind changes over southeastern Australia in McInnes et al. . A recent study of wind speed change in Coupled Model Intercomparison Project (CMIP3) climate models [McInnes et al., 2011] indicates a robust response (i.e., more than 66% of models agree on the direction of change in a 19-member multimodel ensemble) of average and 99th percentile wind speed decline in winter over Australia between 30 and 40°S and wind speed increase south of 40°S.
 Recently, regional climate model simulations for Australia have become available as part of the Climate Futures for Tasmania (CFT) project. Using the Conformal Cubic Atmospheric Model (CCAM) [McGregor and Dix, 2008], six of the CMIP3 [Meehl et al., 2007] simulations under two IPCC [Nakićenović and Swart, 2000] emission scenarios (A2 and B1) were dynamically downscaled to approximately 60 and 15 km resolution to cover Australia and Tasmania respectively. These experiments provide a major resource for a range of climate change impact studies beyond those undertaken directly as part of the CFT project. For example, three of these simulations were used to investigate future wave climate change on Australia's east coast [Hemer et al., 2012].
 This study investigates how two different methods of applying boundary conditions to the CCAM model affect the weather patterns produced by the model, particularly in the context of how they influence coastal sea levels. Following this, the present and future climate output of two CCAM simulations and one GCM simulation are used to force a hydrodynamic model. The selected GCM also provided forcing for one of the two CCAM simulations and so its use in the present study provides the opportunity of assessing how the climate produced by the CCAM simulation compares to that of its forcing GCM. An investigation of how changes in the future climates, as simulated by the atmospheric models, affect coastal sea levels is then undertaken. In the following section an overview of the ocean and atmospheric models used in this study is given. In section 3, the atmospheric model experiments and the effect of different methods of dynamical downscaling of atmospheric conditions in regional climate models are presented and the ocean model's ability to simulate sea levels is investigated. section 4 presents future climate extreme sea level results while section 5 provides a concluding discussion.
2. Model Setup
 This section describes the hydrodynamic ocean model used to calculate maximum sea surface height (SSH) and ocean currents. The atmospheric forcing fields for present and future climates are discussed followed by an overview of the designed experiments.
2.1. Ocean Model
 We use the Rutgers version of the Regional Ocean Modeling System (ROMS) [e.g., Shchepetkin and McWilliams, 2005]. ROMS is a 3-dimensional (3D), hydrostatic, primitive equations model, featuring a nonlinear free-surface in the barotropic mode. It uses terrain following stretched-coordinates in the vertical and an orthogonal curvilinear grid transformation in the horizontal. For simplicity and efficiency we use the barotropic (i.e., 2D) model formulation only. Results from the 2D formulation have been found to not differ significantly from results of a full 3D simulation (not shown) but the 2D formulation is more computationally efficient [see also Kauker and Langenberg, 2000]. The reason for the similarity in model results is because SSH is a function of the continental shelf width and depth rather than of baroclinicity. This simplified approach has been successfully used in other modeling studies concerned with sea level extremes in this region [e.g., McInnes and Hubbert, 2003; McInnes et al., 2009] and is therefore appropriate for the investigation of short-lived extreme SSH events.
 Two model domains outlined in Figure 1 have been used. The decision to move to the larger grid occurred after the completion of initial model runs but time limitations prevented the rerunning of all simulations on the larger grid. The smaller domain stretches from 120 to 152°E and from 45 to 30°S including South Australia (SA), Victoria (VIC), Tasmania (TAS) and parts of New South Wales (NSW), the large domain extends further to the west and east including large areas of Western Australia (WA) and NSW. Both domains ensure that Kelvin waves generated by atmospheric disturbances along the south Australian coast are captured. These have been shown to exert a significant effect on maximum SSH along the south Australian coast [e.g., McInnes and Hubbert, 2003, and references within]. Both model domains have a horizontal resolution of about 5 km. The bottom topography is taken from the 1′ resolution General Bathymetric Chart of the Oceans (GEBCO) data set provided by the British Oceanographic Data Centre (BODC). The interaction of tidal currents with storm surges has been found to have a negligible effect on sea level heights along the southern coastline, although the phase of the tide has been found to be modified during episodes of strong westerly winds in Bass Strait [McInnes and Hubbert, 2003; McInnes et al., 2012a] and is thus neglected in our model simulations.
2.2. Atmospheric Forcing and Experimental Design
 Atmospheric forcing for the ROMS hydrodynamic model was obtained from various sources. For the purpose of validating ROMS, 4 times daily sea level pressure (SLP) and 10 m winds from NCEP reanalysis [Kalnay et al., 1996] were used. To investigate the different methods of forcing, as well as to benchmark the CCAM model's ability to represent observed climate conditions, two CCAM simulations that apply NCEP forcing via a spectral nudging (SN) and a bias-corrected SST forcing (SST) method were considered. Future changes in extreme sea levels were investigated by forcing the ROMS model with atmospheric forcing from one GCM and two regionally downscaled climate model simulations using CCAM. These atmospheric sources are summarized in Table 1 and are discussed in turn below.
Table 1. Summary of Ocean Model Simulations Undertaken in the Study, Their Associated Atmospheric Forcing, Domain and Length of Each Simulation
CCAM forced by NCEP SSTs
Spectral nudging used to apply NCEP temperature, winds and pressure to CCAM
CCAM forced by GFDL SSTs
CCAM forced by Mk3.5 SSTs
 The ability of the ROMS model to represent sea level variations arising from weather forcing was assessed using 10 m winds and pressure from the NCEP reanalyses. Although a range of other reanalysis products are available, NCEP was chosen to ensure consistency when later assessing the performance of the CCAM model over the observational time period, since CCAM had been run using NCEP forcing. However, to assess the quality of the NCEP data for the purposes of forcing the ROMS model, we also compared the wind speeds from this and other reanalysis products to wind speeds from anemometers at several tide gauge locations on the southern Australian coastline. The observational data was obtained from the National Tidal Centre (NTC) of Australia (http://www.bom.gov.au/oceanography/projects/abslmp/data/index.shtml). The tested reanalysis products were the National Centre for Atmospheric Research products: NCEP, NCEP 2 and the NCEP Climate Forecast System Reanalyses (CFSR) [Saha et al., 2010], the Common Ocean Ice Reference Experiment (CORE) data set [Large and Yeager, 2009] and the European Centre for Medium-Range Weather Forecasts ECMWF ERA40 reanalysis [Uppala et al., 2005]. Three statistics; correlation, standard deviation (normalized by the observed correlations) and the root mean square (RMS) difference, were prepared between time series of anemometer winds (adjusted to 10 m height using the logarithmic wind profile relationship) and respective reanalysis data products over the 1992–1999 period (the time interval was selected because anemometer time series commenced in 1992). The results are summarized in Figure 2 using a Taylor diagram [Taylor, 2001]. The results indicate that the standard deviation (STD) varies considerably across data products and station locations. Most reanalysis products have a tendency to overestimate the variability rather than to underestimate it (indicated by most STD values being greater than 1). NCEP 2 performs most poorly in this regard while CFSR winds show the smallest spread in STD with values closest to 1. NCEP, CORE, and ERA40 exhibit a similar spread in STD values. Correlations across the reanalysis products range from 0.5 to 0.7 for all stations with generally lower values for Burnie and higher values for Esperance. CFSR winds exhibit the highest correlation values, above 0.6 for all locations. RMS differences are also lowest for CFSR winds (indicated by the least distance of the points to the value of 1 on the horizontal axis). This analysis suggests that CORE, NCEP and ERA40 products exhibit similar biases with regards to representing southern Australian wind speed whereas NCEP 2, somewhat surprisingly, exhibits larger differences with respect to winds across this region. We conclude that NCEP forcing is of similar quality to products that are comparable in terms of available spatial and temporal resolution (e.g., CORE and ERA40) over the region of interest considered in this study and is therefore suitable for the purpose of forcing the ROMS model over the period of 1980–1999. However, we note that ocean model simulations forced with CFSR winds for the present climate may better compare with station data.
 A key requirement for the accurate modeling of storm surges is sufficient temporal resolution in the atmospheric forcing fields. Wind fields from the CMIP3 climate model simulations were available from the Program for Climate Model Diagnosis and Intercomparison (PCMDI) only once-daily and so were not suitable for our modeling needs. Therefore we were limited to using the Commonwealth Scientific and Industrial Research Organisation's (CSIRO) Mark 3.5 (Mk3.5) GCM output, which was available to us with 4 times daily temporal resolution.
 Our sample of future climate simulations was supplemented by using dynamical downscaled model simulations performed using the Conformal Cubic Atmosphere Model (CCAM) [McGregor and Dix, 2008]. CCAM is a fully global atmospheric model, which implements a stretched grid using a Schmidt  transformation allowing for higher grid resolution in any particular area of interest. Typically, forcing for the CCAM model has been applied in one of two ways depending on availability of forcing data. Where atmospheric fields of sufficient temporal resolution are available, a spectral nudging (SN) technique [Thatcher and McGregor, 2009] is used. This technique specifies the state of the regional atmosphere by perturbing surface pressure, air temperature and winds and moisture fields at model layers above 850 hPa at large spatial scales (typically 3600 km and above) using a specified source of atmospheric data. The atmosphere at small spatial scales is allowed to evolve freely within CCAM. In the absence of atmospheric model data of sufficient temporal resolution, forcing for CCAM is achieved by applying bias-adjusted sea-surface temperatures and sea-ice forcing as a surface boundary condition [Katzfey et al., 2009]. This approach is referred to here as the SST approach.
 The CCAM simulations selected for use in this study were undertaken for the Climate Futures for Tasmania (CFT) project. In that project six CMIP3 GCMs were downscaled over the Australian region at approximately 60 km horizontal resolution each under an A2 and B1 emission scenario [Nakićenović and Swart, 2000] (see McGregor and Dix , Corney et al. , and Nguyen and McGregor  for more details). Due to the aforementioned problem of limited temporal resolution of the archived output of the CMIP3 climate models at the PCMDI, the SST method of forcing was used by CCAM to downscale the selected CMIP3 GCMs, i.e., sea-surface temperatures and sea-ice obtained from the GCM were bias-adjusted and applied as a surface boundary condition in CCAM. In addition, radiative forcing, consistent with the emission scenario used by the GCM was also applied to CCAM. Additionally, CCAM was used to downscale the NCEP reanalyses using both SN and SST approaches over the period 1980 to 1999 to enable a comparison of the downscaling techniques.
 Due to limited resources in the present study, only two of the available CCAM downscaled climate model simulations from the CFT project were selected to force ROMS. The selection of models was made on the basis of Hemer et al. , who investigated the wind changes in the six CCAM A2 simulations and found they exhibited a high degree of similarity but selected a subset of three of the simulations on the basis that the changes were larger and would therefore elicit a larger response in wave model simulations carried out over eastern Australia. Here, we select two of those three models simulations to force the ROMS model; one that downscaled the Geophysical Fluid Dynamics Laboratory model (GFDL2.0), hereafter referred to as the CCAM-GFDL2.0 simulation and the other that downscaled the Mk3.5 model hereafter referred to as CCAM-Mk3.5. The use of these two experiments together with the CSIRO Mark 3.5 GCM (referred to as Mk3.5 hereafter) enables an exploration of the uncertainty associated with different future climate realizations. It also provides an opportunity to investigate the differences between the future climate of the Mk3.5 GCM and the CCAM-Mk3.5 model and hence the additional uncertainty introduced through the regional downscaling step. We also select the two experiments in which CCAM downscaled the NCEP reanalyses using both SN and SST approaches over the period 1980 to 1999 to enable a comparison of the downscaling techniques prior to investigating the future climate simulations.
 The set of ROMS model simulations are summarized as follows. The ROMS simulation forced by 6-hourly SLP and 10 m wind-forcing from NCEP over the 1980–1999 period (hereafter referred to as R_NCEP) is to assess how ROMS performs in simulating daily maximum sea surface height (SSH) through comparison with tide gauge observations. The ROMS simulations forced by the two CCAM simulations that downscale NCEP via SN and SST approaches over the 1980–1999 period, (hereafter referred to as R_CCAM-NCEP-SN and R_CCAM-NCEP-SST respectively) are to investigate the effect on modeled daily maximum SSH of the two methods for downscaling atmospheric forcing fields in CCAM. The ROMS simulation forced by 6-hourly SLP and 10 m winds from the CCAM-GFDL2.0 and CCAM-Mk3.5 and Mk3.5 climate simulations are to investigate future changes in SSH between the years 1980–1999 and 2080–2099 (see Table 1 for a summary).
3. Model Performance
 In this section the mean atmospheric conditions and variability for each of the models used to force the ocean model over the period 1980–1999 are discussed. This is followed by an investigation of the ocean's response to the different atmospheric forcing.
3.1. Atmospheric Conditions
3.1.1. Sea Level Pressure
Figure 3 displays seasonal SLP over the period 1980–1999 for the NCEP reanalysis and the 5 sources of atmospheric forcing indicated in Table 1. Comparing Figures 3a and 3b, fairly good agreement can be seen between NCEP and CCAM-NCEP-SN in both spatial pattern and magnitude. The seasonal migration of the subtropical high pressure systems over the Indian and South Pacific Oceans, leads to large SLP gradients between 35 and 40°S in austral winter (JJA) and this is well represented in the CCAM-NCEP-SN simulation. During austral summer a high pressure ridge centered at 37°S stretching along the southern Australian coastline is evident and produces easterly winds near the coast (not shown), which can trigger upwelling over the shelf. This feature is more pronounced in the CCAM-NCEP-SN simulation. Figures 3c, 3d, and 3e agree similarly well in SLP for austral summer and winter months in terms of spatial pattern. However, it appears that in each simulation the magnitude of SLP is overestimated for each season when compared to NCEP. Mk3.5 (Figure 3f) shows the strongest deviation from the NCEP data. This is particularly the case in JJA and leads to an overestimation of winds associated with the westerly storm band over Tasmania (not shown).
3.1.2. Standard Deviation
Figure 4 displays the mean standard deviation (STD) for 10 m wind speeds over the period 1980–1999 for each source of atmospheric data. CCAM-NCEP-SN (Figure 4b) displays generally weaker variability compared to NCEP (Figure 4a), particularly around Tasmania and along the southern Australian coastline, where values differ by up to a factor of 2. This may suggest an inability of CCAM to model storm tracks with a far northern extension. In comparison to CCAM-NCEP-SN in which variability was underestimated, the STD in CCAM-NCEP-SST, CCAM-GFDL2.0 and CCAM-Mk3.5 (Figures 4c, 4d, and 4e) simulations all exhibit higher values particularly over the southern ocean south of 30°S. The largest variability occurs in the Mk3.5 (Figure 4f) model, which is up to 25% higher than NCEP (Figure 4a). Higher variability in CCAM-NCEP-SST (Figure 4c) appears to be the result of stronger modeled winds at the 10 m level compared to CCAM-NCEP-SN (Figure 4b) 10 m winds (not shown).
3.2. Oceanic Conditions
 In this section the modeled daily maximum SSH simulated in the R_NCEP and R_CCAM-NCEP-SN runs are compared with observations from tide gauges in the region as well as with the R_CCAM-NCEP-SST experiment.
3.2.1. Modeled SSH for Selected Locations
Figure 5 compares time series of the daily maximum SSH from the ocean model simulations R_NCEP and R_CCAM-NCEP-SN with detided sea level data obtained from the NTC, Australia for 1994 at selected locations. This particular year was chosen because it contained several extreme sea level events throughout the year. Note that the R_CCAM-NCEP-SST modeled sea levels cannot be compared in this way because the CCAM-NCEP-SST simulation is forced only by SSTs and the internal dynamics of CCAM generate the weather conditions on a day to day basis. Therefore there is no correspondence between the daily weather events in this run and the NCEP reanalyses although agreement between the climatological average of the daily weather events between this CCAM simulation and NCEP is expected.
 In general, the modeled maximum SSH of R_NCEP agrees well with detided sea level observations for Thevenard and Portland but less so for Spring Bay and Sydney (Port Kembla). There is a tendency for ROMS to underestimate the higher sea levels in summer months, but to overestimate sea levels in winter. For southern Australia, the weather conditions that are commonly responsible for high sea levels are low pressure systems emanating from the midlatitudes often with cold fronts that travel from west to east along the coastline. Two examples are evident in May and June 1994. Since the associated meteorological forcing is generally well captured by NCEP, there is reasonable agreement in the timing of the modeled and observed sea level residuals.
 For the east coast locations of Spring Bay and Sydney, the poorer agreement may relate to poorer representation in NCEP and CCAM_NCEP-SN of the weather systems that cause extreme sea levels on the east coast. On this coastline extreme sea levels are commonly caused by east coast low pressure systems, [McInnes and Hubbert, 2001], which can sometimes undergo rapid mesoscale development over the ocean areas to the east of Australia [Holland et al., 1987]. This means that the intensity of the systems may be underestimated in some cases by NCEP.
 In the case of the R_CCAM-NCEP-SN simulation, agreement with observations is considerably poorer than the R_NCEP run. This is characterized mostly by reduced amplitudes of SSH signals, consistent with the reduced variability in SLP discussed in section 3.1.2. In some cases the SSH signals present in the R_NCEP run are absent in the R_CCAM-NCEP-SN also suggesting some atmospheric conditions may be poorly resolved in the CCAM-NCEP-SN run. This will be examined further in the next section.
 The ROMS model simulations of observed SSH are compared to the detided residuals over the 1992–1999 period for selected locations and the results are summarized in Figure 6. This shows that in general higher agreement between model and observations is seen in R_NCEP compared to R_CCAM-NCEP-SN as indicated by lower RMSE values and normalized standard deviations closer to 1. Locations on the southern coastline such as Thevenard, Adelaide (Port Stanvac) and Portland show higher correlations than locations on the east coast such as Spring Bay and Sydney (Fort Denison).
 We also compared Probability Distribution Functions of observed SSH with those simulated by R_NCEP, R_CCAM-NCEP-SN and R_CCAM-NCEP-SST (not shown). Closest agreement was seen between R_NCEP and observations. Both SN and SST forced simulations underestimate the occurrence of high SSH events and overestimate the occurrence of low SSHs. However, the SST forced simulation better represents the distribution of higher SSH values than the SN simulation.
3.2.2. Daily Maximum SSH Along the 25 m Isobath
 Here we compare Hovmoeller diagrams of daily maximum SSH calculated along the 25 m isobath in the R_NCEP and R_CCAM-NCEP-SN simulations for the years 1994 (Figure 7) and 1997 (Figure 8). The Hovmoeller diagrams illustrate the degree of coherence between extreme sea levels along different stretches of coastline and allow elevated sea levels due to individual synoptic events to be identified.
 Comparing Figures 7a with 7b and 8a with 8b reveals that there is reasonable correspondence in the timing of daily maximum SSH suggesting that both model simulations capture similar SSH events along the 25 m isobath. However, daily maximum SSH in R_CCAM-NCEP-SN are much lower (by up to a factor of 2–3) of maximum SSH in R_NCEP. This reduction in amplitude is most noticeable along the south coast. We examine the difference in atmospheric forcing between the two sources for two particular examples of strong SSH that occur in the R_NCEP simulation centered around day 140 and 170 in 1994 (Figure 7a). The first maximum SSH event at day 140 is reasonably well captured by R_CCAM-NCEP-SN whereas the second maximum SSH event around day 170 is hardly discernible. Maps of SLP around day 170 are shown in Figure 9 for both NCEP and CCAM-NCEP-SN. The NCEP fields (Figure 9a) display a distinct low pressure system extending from the Southern Ocean to the Australian coast leading to a strong signature in SSH. The CCAM-NCEP-SN fields for the same event (Figure 9b) fail to reproduce the northward extension of the low pressure system south of 40°S. Consequently the winds are weaker and produce lower SSH. This synoptic situation is consistent with the difference in STD of wind speed shown in Figure 4 in which the STD in R_CCAM-NCEP-SN (Figure 4b) is much weaker than R_NCEP (Figure 4a). Therefore it is likely that this example may be typical of the differences in CCAM-NCEP-SN compared with NCEP where CCAM may consistently under-represent major low pressure systems.
 In 1997, following the passage of a frontal trough in the preceding week, a strong east coast SSH event developed on the east coast on May 9th (Figure 8a, day 130). This event was associated with the synoptic situation shown in Figure 10a with a low pressure system ‘cut-off’ from the westerlies by a ridge of high pressure [e.g., Holland et al., 1987]. The resulting southeasterly winds pile up water along the eastern coastline and result in higher sea levels there [McInnes et al., 1992]. From Figure 10b it is clear that although CCAM reproduces the high pressure system south of southern Australia, it does not reproduce the low pressure system and hence fails to produce sufficiently strong winds (not shown) that lead to extreme sea levels along the east Australian coastline. As a consequence the modeled maximum SSH for the simulation using CCAM-NCEP-SN forcing are lower than those using NCEP forcing. We also note that the lower sea levels on the east coast in the R_CCAM-NCEP-SN simulation in the previous week (Figure 8a) appear to be due to the absence of the frontal trough passage in CCAM-NCEP-SN (not shown) and therefore the persistence of the high pressure over this period leading to a depression in sea levels on the east coast.
 Both of these examples demonstrate that although CCAM uses spectral nudging to transfer large scale forcing to the CCAM model, this forcing does not necessarily lead to the development of small scale systems to sufficient intensity within CCAM and this in turn reduces the height of the associated SSH elevations modeled by ROMS near the Australian coast.
3.2.3. Standard Deviation of Maximum SSH Along the 25 m Isobath
 As already indicated, maximum SSH events modeled in R_CCAM-NCEP-SN are generally smaller in magnitude than observations, suggesting differences in the variability for the ocean simulation. Figures 8c and 8d compare the seasonal STD along the 25 m isobath in both experiments. Both model simulations display the highest (lowest) variability in austral winter (summer). Furthermore, there is a good agreement between the relative magnitudes of SSH along the coastline. The southward facing coastline generally experiences stronger variability, which is related to its exposure to the southern ocean and its generally wide continental shelf that favors storm surge amplification (except along the coast near the SA and VIC border). The west and east facing coastlines have lower variability which is related to the narrow continental shelves on these coastlines that are not conducive to storm surge amplification. However, maximum STD for R_CCAM-NCEP-SN is only about 50% of that shown for R-NCEP, which is consistent with results in the previous section.
 A comparison of SLP from the reanalyses and various model simulations over the 1980–1999 period has been undertaken. With respect to NCEP data, the seasonal cycle is well represented in the two CCAM simulations with NCEP forcing via the SN and SST approaches and the three climate model simulations (two CCAM and one GCM) under present conditions. Best agreement is apparent for the CCAM-NCEP-SN data. This is not surprising since the low frequency variability in this CCAM simulation is being specified according to NCEP fields. In the other simulations, the magnitude of SLP is overestimated with the largest difference from NCEP data being for the Mk3.5 GCM in JJA.
 Variability, expressed in terms of the standard deviation for the downscaled NCEP wind fields, is lower in the CCAM-NCEP-SN simulation in many parts of the area of interest with maximum reduction of 50% at times. This has consequences for the modeled maximum SSH, which are reduced by about 50% when compared to ocean model simulations forced with NCEP data. Furthermore, case studies for selected high SSH elevation events suggest that when CCAM downscales NCEP atmospheric forcing using the SN method, high wind events associated with certain types of low pressure systems are underestimated, leading to reduced variability in atmospheric fields, and this adversely affects the simulation of daily maximum SSH.
4. Present Versus Future Conditions
 This section places the focus on change in atmospheric conditions and how this may affect the modeled oceanic circulation and daily maximum sea surface height. Change is assessed by evaluating the difference between future (2080–2099) and present (1980–1999) conditions for the variables of interest.
4.1. Changes in Atmospheric Conditions
Figure 11 displays the change in wind speed for CCAM-GFDL2.0, CCAM-Mk3.5 and Mk3.5. The vectors represent each model's seasonal mean climatology over 1980–1999. All models are similar in terms of their mean conditions and changes for DJF and MAM. The changes for DJF indicate enhanced winds north of 40°S and somewhat reduced winds in the far northeast for the CCAM-GFDL2.0 and CCAM-Mk3.5 models (Figures 11a and 11b). This results in more frequent and stronger easterlies along southern Australia, stronger southeasterlies along the west coast and enhanced onshore winds for most parts of NSW. South of 40°S westerlies are reduced for all models with larger reductions for CCAM-GFDL2.0 and CCAM-Mk3.5 and smaller reductions for the Mk3.5 model. For MAM there are smaller reductions to the west of Tasmania for CCAM-GFDL2.0 and greater reductions for Mk3.5.
 Changes in the wind speed pattern during austral winter (JJA) are more substantial than the previous two seasons. The areas of greatest difference coincide with areas of strong westerly winds from 30°S southward which are mostly enhanced for CCAM-GFDL2.0 and particularly for CCAM-Mk3.5 over Tasmania and further to the east. Some reduction in westerly winds along the southern Australian coast is apparent particularly for CCAM-GFDL2.0. Offshore winds are also reduced in both the CCAM simulations (Figures 11a and 11b) along the NSW coast. In contrast, the Mk3.5 model (Figure 11c) displays smaller reductions over most areas under consideration with enhanced winds only over parts of eastern Tasmania. In SON both CCAM models suggest enhanced winds over all of the displayed area, while Mk3.5 displays a reduction in winds in a band between 30 and 40°S stretching from 110 to 160°E.
4.2. Changes in Maximum SSH
 Here we assess changes in modeled daily maximum SSH and variability arising from the forcing from the climate models listed in Table 1. Spatial maps of change in the daily maximum SSH are examined and discussed for selected locations. Last we discuss the changes in height along the 25 m isobath for mainland Australia and Tasmania.
4.2.1. The 95th Percentile of Maximum SSH
Figure 12 displays spatial maps of changes in the 95th percentile of daily maximum SSH for each of the model simulations. In DJF all model simulations agree on an increase in sea levels along the east coast and a reduction of sea levels in the southernmost part of the model domain. The main areas of disagreement between the model simulations occur close to the southern mainland coast where two models (R_CCAM-GFDL2.0 and R_Mk3.5) indicate declines in SSH and one model (R_CCAM-Mk3.5) indicates an increase.
 In MAM the spatial pattern of change is similar between the atmospheric models and this is reflected in the change in SSH. The strongest reductions in SSH of up to 0.10 m are confined to the near coastal region east of 130°E between 30 and 40°S and around Tasmania and Bass Strait. Weaker positive anomalies are apparent along the NSW coast. Away from the Australian continent the anomalies are mostly negative but differ strongly in magnitude with largest negative anomalies for R_Mk3.5. The ocean's response is consistent with the observed change in wind speed (Figure 11) for this season. For example, the enhanced easterly winds near the southern Australian coast are unfavorable for the occurrence of positive SSH and the reduced westerlies south of 40°S also lead to reduced values of SSH.
 A similar pattern of reduced daily maximum SSH along the southern Australian coastline is evident through JJA with some variation in intensity apparent across the model simulations. However, south of 40° and including much of the TAS coastline, the three simulations show large areas of increase in SSH although the magnitudes are small. There are marked differences between R_CCAM-Mk3.5 and R_Mk3.5 with the former showing mostly declines on the east and west TAS coasts and increases on the south TAS coast and the latter showing increases on the east TAS coast and declines on the west TAS coast.
 The differences between R_CCAM-Mk3.5 and R_Mk3.5 continue in SON with the former showing weak increases over the east and west coasts of TAS while the latter shows weak increases to the east of TAS and declines elsewhere. R_CCAM-GFDL2.0 shows most declines over the southeastern Australian coastline.
 For selected coastal locations across the model domain the relationship between the changes to the 95th percentile SSH and the changes in higher percentiles was also investigated. It was found that there was a high degree of consistency between changes at the 95th and 99th percentile across all seasons and locations (not shown), which suggests that the changes found here for 95th percentile SSH along the coast are also broadly applicable to 99th percentile SSH.
4.2.2. SSH Change Along the 25 m Isobath
Figures 13 and 14 display the change along the 25 m isobath between 2080–2099 and 1980–1999 of the average of the two highest values of SSH per season averaged across the 20 years considered, (hereafter referred to as the block maxima) for the mainland coast and TAS respectively. The graphs for each model simulation agree well with the spatial pattern displaying the 95th percentile of maximum SSH (Figure 12) indicating consistency between the two different extremes metrics. SSH is mostly reduced all along the coastline with the largest reduction between SA and VIC. For DJF change is generally small with largest negative anomalies between SA and VIC in the order of 2–4 cm. Positive anomalies are evident between VIC and NSW for R_CCAM-Mk3.5. During MAM negative anomalies between SA and VIC become larger in magnitude (up to 8 cm) with strongest values for R_Mk3.5. The other two experiments show large negative anomalies of similar size over Bass Strait. Changes in SSH over WA and NSW are small and mostly negative with some positive anomalies apparent for R_Mk3.5 over NSW. In JJA all experiments show similar strong negative anomalies along the entire coastline with strongest values along the SA and VIC coastlines for R_CCAM-GFDL2.0 and R_CCAM-Mk3.5. In SON anomalies are similar to MAM with R_Mk3.5 displaying maximum negative anomalies between SA and VIC.
 Over Tasmania (Figure 14) SSH changes are mostly negative for R_Mk3.5 with the strongest reduction in SSH during MAM. Model differences become apparent especially in SON where R_Mk3.5 show reduced SSH, while R_CCAM-Mk3.5 and R_CCAM-GFDL2.0 show an increase in SSH. This difference is also seen in Figure 12 although numbers are small. It is likely that the cause for the model differences lies in the modeled position and strength of the westerly storm track crossing Tasmania, which appears to be further to the north for Mk3.5 (Figure 3f) when compared to CCAM-Mk3.5 and CCAM-GFDL2.0 (Figures 3d and 3e).
 Associated variability changes are shown in Figure 15 for mainland Australia and in Figure 16 for Tasmania. The displayed graphs in Figure 15 correlate well with those displaying the block maxima for mainland Australia (Figure 12), hence STD is mainly reduced for all simulations for all seasons, thereby suggesting that enhanced/reduced SSH are related to enhanced/reduced variability for each model simulation. The comparison does not hold so well for Tasmania. STD is mostly reduced (Figure 16) over all seasons. The only exception is seen in Figure 16c for R_CCAM-GFDL2.0, which displays enhanced variability for southwestern Tasmania.
 In this section the effect of climate change on atmospheric circulation and extreme SSH events was investigated. The atmospheric fields across the three models show similar signs of change for austral summer DJF and austral autumn MAM with enhanced easterly winds along the southern Australian coast and reduced westerlies over the Southern Ocean. The oceanic response to this climate change signal is strongest in MAM and similar for all three model simulations. It exhibits a reduction of daily maximum SSH all along the south Australian coast and also TAS and an increase in SSH over the east coast, consistent with the change in atmospheric forcing. However, strong inter-model variability is apparent with reduced SSH ranging from 1 to 5 cm.
5. Discussion and Conclusions
 This study commenced with an investigation into two methods (the so-called SN and SST methods) that can be applied to the global, variable resolution CCAM model when used to dynamically downscale atmospheric conditions over a region of interest. In the particular runs available to this study, CCAM had been used to downscale NCEP reanalyses over the Australian region over 1980 to 1999. While the SN method preserves the temporally varying features of the forcing fields, allowing the direct comparison of weather events, the SST approach allows CCAM to spin-up its own weather conditions and so a direct comparison of daily weather patterns with the source of the forcing conditions is not possible. In the comparison undertaken here, the experiments were evaluated in terms of SLP and 10 m wind variability, two variables that influence coastal sea levels. In this regard, the present study complements the findings of other CCAM assessment studies such as Nguyen et al.  and Grose et al.  that focus more on rainfall and broad circulation features. This study found that the CCAM simulation with the SN approach well represented seasonal SLP patterns compared with the SST approach. However, it also underestimated wind speed variability and this resulted in considerable underestimation of extreme sea level magnitudes. It is possible that a more realistic CCAM simulation using the SN approach could be achieved by adjusting the characteristics of the filter to nudge toward smaller scales of variability. In the SST approach on the other hand, slightly stronger than observed pressure gradients occurred between the subtropical ridge at around 30–35°S to 50°S which was associated with stronger mean winds and slightly greater variability occurring in the wind fields. However, SSH's produced by forcing ROMS with CCAM simulations using the SST approach was found to better represent the distribution of higher SSH values than the ROMS simulation in which forcing was provided by CCAM using the SN approach.
 The second part of the study investigated the effect of climate change on extreme sea levels around southern Australia and Tasmania. The simulations included a GCM experiment and two regional climate model experiments in which CCAM downscaled two GCM experiments at 60 km resolution over Australia. All experiments incorporated radiative forcing under future climate conditions from the SRES A2 emission scenario.
 The two CCAM simulations that downscaled GCMs using SST forcing exhibited similar variability in 10 m winds when compared to the CCAM simulation that downscaled NCEP reanalyses over the 1980–1999 period using the SST approach. This suggests that the internal dynamics of CCAM when forced with SSTs are strongly affecting the day to day variability. Furthermore, the variability in 10 m winds of the Mk3.5 GCM was higher than that of the CCAM simulation that downscaled Mk3.5, again indicating the strong influence of the internal dynamics of the downscaling model in determining the characteristics of the resulting climate.
 Regarding changes to the wind fields between 2080 and 2099 and 1980–1999, there are marked differences in the pattern of changes across the seasons. The changes are generally characterized by increasing easterly winds between around 30–35°S and weakening westerlies south of around 40°S in DJF and MAM (summer and autumn) and strengthening westerlies south of around 40°S in JJA and SON (winter and spring).
 The two CCAM simulations exhibit a high degree of similarity in the spatial patterns of wind speed increase and decrease across the seasons. Indeed, these two simulations are more similar to each other than the wind speed changes between the Mk3.5 GCM and the CCAM simulation that downscales the Mk3.5 GCM. This suggests that the uncertainty that is purportedly sampled in an ensemble of downscaled simulations undertaken with a single regional climate model may in fact be suppressed, although this will depend on the method by which the regional climate model is being forced by the climate model. A similar finding was also reported by Hemer et al. .
 The effect of the changes in atmospheric circulation patterns on daily maximum SSH leads to mostly a reduction in SSH along the mainland coast which is strongest from MAM to SON, an increase in SSH along the NSW coast in DJF and MAM and along parts of the TAS coast in JJA and SON. Other locations such as Bass Strait indicated less robust changes with the results from the different models and seasons reflecting possible increases or decreases.
 However the results arising from projected circulation changes must be seen in the context of the future projected sea level rise, which has not been considered in this study. For the A2 emission scenario that is the basis of the future climate forcing used in the climate simulations in this study, the projected sea level rise for 1990 is in the range of 0.20–0.59 m [Hunter, 2010]. Clearly the largest declines in sea level arising from projected circulation changes, which were up to 0.10 m along the SA coast in JJA, would not negate the effects of even the low-end projected sea level rise.
 This work was undertaken as part of the Australian Climate Change Science Program, funded jointly by the Department of Climate Change and Energy Efficiency, the Bureau of Meteorology and CSIRO and the Victorian Department of Sustainability and Environment. The authors are grateful to Mark Hemer, Jack Katzfey and Marcus Thatcher for useful discussions. They acknowledge the modeling groups, the Program for Climate Model Diagnosis and Intercomparison (PCMDI) and the WCRP's Working Group on Coupled Modeling (WGCM) for their roles in making available the WCRP CMIP3 multimodel data set. Support of this data set is provided by the Office of Science, U.S. Department of Energy. Finally the authors wish to thank the two anonymous reviewers for their valuable comments on the manuscript.