Projections of climate change-driven variations in the offshore wave climate off south eastern Australia


M. A. Hemer, CSIRO-CMAR/CAWCR, GPO Box 1538, Hobart, TAS 7000, Australia. E-mail:


This study presents an analysis of the wind-wave climate on the south-east Australian continental margin and its potential changes under the effects of projected climate change. The wind-wave climate is modelled for three 20 year time-slices (1981–2000, 2031–2051, and 2081–2100) using a dynamical wave modelling approach in which the WaveWatch III spectral wave model, with a nested domain SWAN model, is the region of interest. Surface wind forcing is obtained from an ensemble of three global climate model runs (CSIRO Mk3.5, GFDLcm2.0, and GFDLcm2.1) dynamically downscaled using CSIRO's Cubic conformal atmospheric model (CCAM), under two greenhouse gas emission scenarios (a high emission SRES A2 and low emission SRES B1 scenario). A third level of uncertainty associated with adjusting climate model derived winds for climate and variability bias is found to be the largest source of uncertainty among the limited range of sources considered in the twenty-eight 20 year wave model simulations undertaken herein. The ensemble of wave model runs for the 2081–2100 time-slice project a robust decrease in mean significant wave height along the south-east Australian coast relative to present (1981–2000) conditions. The magnitude of projected change in mean annual significant wave height is less than 0.2 m, with larger values in the north of the region of interest, and is associated with a projected decrease in storm wave energy in the region. An anticlockwise rotation in mean wave direction of approximately 5° is also projected over the same period. Copyright © 2012 Royal Meteorological Society

1. Introduction

Climate change has the potential to lead to coastal erosion via a number of processes. Increases in mean sea level and the frequency and/or intensity of storms can increase the cross-shore transfer of sediment, while changes in wave direction in response to shifting atmospheric circulation patterns can alter the along-shore transfer of sediment, which will all result in horizontal shifts in coastline position. Sea level rise and, to a lesser extent, the contribution of storm surges associated with projected climate change have received increased attention along the Australian coast over recent years (Church et al., 2006; McInnes et al., 2009a). However, recent studies have shown that along some coastlines, the potential changes in the wave climate, which is a principal driver of the coastal sediment budget, may dominate the effects of sea level rise (Cowell et al., 1995; Coelho et al., 2009). To date, only minimal consideration has been given to how the offshore wave climate along the eastern Australian coast will vary under projected climate conditions (McInnes et al., 2007). Nicholls et al. (2007) noted that there has been a great emphasis on the impact of sea level rise on the coastal zone and that impact studies needed to consider a broader range of coastal drivers. It was noted, in particular, that the effects of climate change on coastal erosion had been largely ignored. Furthermore, projected wave and near-coastal current conditions under future climate scenarios was identified by Christensen et al. (2007) as being critical for assessing the likely effects of climate change on coastal erosion. In order to support sustainable planning and adaptation studies in the coastal zone in the face of changing climate, a thorough understanding of the present day wave climate, and its potential future changes, is required. The aim of this study is to undertake a rigorous assessment of the impact of projected climate change on the offshore wave climate of Australia's most populated eastern continental margin. It is anticipated that these projections will aid the quantification of coastal impacts of climate change assessments and support coastal planning and adaptation strategies.

The wave climate along Australia's eastern margin is one of the best observed wave climates in the world. Over 20 years of non-directional wave-rider buoy data and 15 years of directional data are available from a network of six wave-rider buoys along the approximately 1200 km of coastline. Waves observed in the region are generated by a range of meteorological systems: tropical cyclones that form in the Coral Sea to the northeast, east-coast cyclones, and zonal anticyclonic highs that form to the east, intensified extra-tropical cyclones that form in the southern Tasman Sea, and local summer sea breezes (Short and Treneman, 1992). These systems drive wave events of similar magnitude but have significantly different directional components, which result in a strong seasonal cycle. How the projected changes in atmospheric circulation patterns will influence the eastern Australian wave climate is unclear.

Several recent studies have developed wave climate projections under future climate scenarios. These have been based on both statistical [where the statistical relationship between atmospheric variables—mean sea level pressure or surface wind speed—and wave height is exploited (Caires et al., 2006; Wang and Swail, 2006; Wang et al., 2009)] and dynamical [where high temporal and spatial resolution output from a climate model is used to force a spectral wind-wave model (Debernard et al., 2002; Andrade et al., 2007; Debernard and Roed, 2008; Grabemann and Weisse, 2008; Lionello et al., 2008; Wang et al., 2009)] approaches. Wang et al. (2009) compared the ability of dynamical and statistical downscaling approaches to reproduce climatological wave characteristics and variability and seasonal statistics of significant wave height in the North Atlantic and concluded that the statistical approach outperformed the dynamical approach. It should be noted, however, that Wang et al. (2009) adjusted the climate model wind dataset for mean wind climate biases only and not for climate variability biases. Hemer et al. (2012a) presented a more rigorous bias-adjustment method via bivariate quantile adjustment of climate model winds. In this approach, they used winds derived from the US National Centre for Environmental Prediction (NCEP) and Department of Environment (DOE) Reanalysis (NRA-2; Kanamitsu et al., 2002) to represent the present climate wind field. Hemer et al. (2012a) found that when forced with climate and variability bias-adjusted climate model derived winds, the modelled mean east Australian wave climate was well described.

A conscientious study on climate change projections should include an assessment of the inherent uncertainties (Hemer et al., 2010). There are multiple levels of uncertainty that are likely to be particularly relevant for wind-wave projections. These include uncertainties associated with forcing (i.e. future greenhouse gas emissions), uncertainties associated with the limitations of global climate models (GCMs) and the uncertainties associated with the downscaling methods used to produce wave climate projections. The Coupled Model Intercomparison Project (CMIP; Meehl et al., 2007) generates a number of GCM runs that aim to investigate the first two levels of uncertainty using an ensemble approach. To address uncertainty associated with forcing, GCMs are forced using a number of future emission scenarios taken from the IPCC Special Report on Emissions Scenarios (SRES). Uncertainty within GCMs is accounted for by quantifying both the mean and spread of the multi-model ensemble available from a large number of GCMs that are run by the World's leading climate research centres. However, these GCMs are typically not of sufficient spatial resolution and do not archive output at sufficiently high temporal resolution to adequately describe the atmospheric storm systems that need to be resolved if these data are to be used to generate a representative wind-wave field. Thus, downscaling (typically to about 50 km resolution) is required. In this study, downscaled projected climate conditions are obtained from a regionally applied climate model that dynamically downscales a number of CMIP GCMs. Other sources of uncertainty associated with the development of wave climate projections could include differences between statistical and dynamical projections, between different dynamical models, and between different approaches to generating the forcing datasets. Hemer et al. (2012a) investigated the latter and noted that the magnitude of the bias correction for the regional climate model winds was considerably larger than the range of uncertainty within the three-member ensemble associated with choice of GCM. We have been opportunistic in our use of downscaled GCM data to provide climate projections under future emission scenarios, and consequently, we are not able to present an exhaustive investigation of uncertainty associated with wave climate projections. For example, using projections from just one regional climate model, we are not able to quantify the variance associated with introducing a regional downscaling step to the modelling methodology. Similarly, our study is limited in that the size of the forcing (2), and GCM (3) ensembles limits our ability to fully quantify the variance associated with each of these sources of uncertainty. The main objectives of this study are to determine the projected east Australian wave climate under future climate change scenarios and investigate three levels of uncertainty associated with forcing, a range of downscaled GCMs, and adjustment of the model winds for dynamical wave model forcing.

2. Methodology

2.1. Regional downscaling of the Global climate change projections

While most GCMs provide relatively consistent projections of global mean parameters, at a regional scale, they provide highly variable results that may be due to the coarse resolution at which these models operate, which in turn leads to differences in the representation of regional flow patterns. Wave models require surface winds as forcing, and so it is important that these flow patterns are well represented. The approach that is commonly adopted is to dynamically downscale results from the GCMs by nesting a limited area, high-resolution regional atmospheric climate model (RCM), into a subdomain of the GCM. As an alternative, we use output from a variable resolution GCM (Conformal Cubic Atmospheric Model; CCAM, McGregor and Dix, 2008) in which the global domain is stretched using the Schmidt (1977) transformation to a maximum resolution of approximately 60 km over Australia (Corney et al., 2010) and close to this resolution over the oceans relevant to this study. This model offers the advantage of avoiding lateral boundary reflections that can cause spurious results within the RCM (McGregor, 2005). The CCAM model simulations were originally undertaken and carried out for the Climate Futures for Tasmania (CFT) project (Corney et al., 2010) and dynamically downscale several GCMs using bias-adjusted sea surface temperatures (SSTs) and sea ice as forcing (i.e. no atmospheric forcing; Katzfey et al., 2009).

This study presents results from an ensemble of six 60 km resolution CCAM runs, in which three GCMs (CSIRO Mk3.5, GFDLCM2.0, and GFDLCM2.1) were dynamically downscaled, representing multi-model ensembles to quantify uncertainties within the GCMs, under two different future emission scenarios (IPCC SRES A2 and B1), to quantify uncertainty associated with the unknown future global storylines. Modelled near-surface marine wind fields at 10 m height were extracted and archived at 6-h intervals, from three 20 year time-slices (1981–2000, 2031–2050, and 2081–2100). These wind fields were interpolated onto a regular 0.5° (approximately 60 km) latitude–longitude grid, to force a spectral wave model for the region of interest.

Hemer et al. (2012a) presented the performance of the CCAM derived winds and their usefulness as a forcing dataset to describe the present day wave climate. In order to improve the ability of climate model winds to generate a representative present day wave climate, they remapped the CCAM-derived bivariate wind distribution onto the NCEP-DOE Reanalysis (NRA-2) wind distribution, to correct winds for both climate and variability bias. They found that when the wave model was forced with bias-adjusted winds, considerable improvement in the model representation of the mean wave climate was achieved.

Defining the present climate winds from CCAM and NRA-2 as X0, e and Y0, respectively, and the projected climate winds from CCAM as X1, e, where e denotes that it is a member of an ensemble and subscripts 0 and 1 represent the present and projected future climate conditions, respectively; these three approaches can be taken to use the wind forcing to develop wave climate projections. Noting that a dash line represents a wind field derived after adjustment or perturbation, then these approaches can be described as follows:

  1. Unadjusted projections (NBA):
    equation image

Here, forcing winds for the wave model are taken directly from the CCAM model runs. For each downscaled climate model considered, the projected change in wave climate is the difference between the wave climate simulations for the present and the future time-slice.

  1. Bias-adjusted projections (BA):
    equation image

where {.}BQA denotes that the difference is determined using the bivariate quantile procedure outlined by Hemer et al. (2012a). The bias adjustment is conditional on each bivariate quantile category (ij) of X, from which the bias-adjusted field is obtained. In this case, it denotes a difference between two datasets in present climate conditions. Using this approach, we make use of the three sets of adjustment matrices (MCSIROMk3.5, MGFDLCM2.0, and MGFDLCM2.1) generated by Hemer et al. (2012a), which correspond to each of the three previously described multi-model ensembles for which 1981–2000 winds are available. Projected future adjusted winds are compiled by applying the appropriate adjustment matrices to future unadjusted winds (i.e. we assume that the bias adjustment is time in-variable), for each ensemble (e.g. the 2081–2100 CCAMCSIROMk3.5 A2 adjusted winds are determined by applying MCSIROMk3.5 to the 2081–2100 CCAMCSIROMk3.5 A2 unadjusted winds).

  1. Perturbed projections (PP):
    equation image

where {.}BQA denotes, as above, that the difference is determined using the bivariate quantile procedure outlined by Hemer et al. (2012a), but in this case, it denotes a difference between two time periods in the same dataset. In this approach, we argue that if approach 2 is valid, projected winds may be equally as well described as a perturbation of present day winds. Thus, an ensemble of projected wind conditions was obtained by determining the difference between projected and present winds in the CCAM ensemble members and perturbing the present observed winds (represented by NRA-2) by this difference. The bivariate quantile adjustment procedure described by Hemer et al. (2012a) is used to determine the difference between present NRA-2 and CCAM-derived winds, and hence estimate model biases was also used to determine the difference between present and projected CCAM-derived winds. Thus, for scenario A2, three sets of perturbation matrices (PCSIROMk3.5, A2, PGFDLCM2.0, A2, and PGFDLCM2.1, A2) were generated, corresponding to the three multi-model ensemble members.

Given that the adjustment of the surface winds is required to obtain a suitable wind field to force a dynamic wave model, each of these approaches are equally valid. We are therefore interested in the robustness of projected wave conditions when using these three sets of wind fields to force a wave model, and we use this to quantify a third level of uncertainty (bias adjustment).

2.2. Wave modelling

The response of the wave climate to projected climate change scenarios is investigated via the implementation of WaveWatch III (version 2.2; Tolman, 2002) and SWAN (Booij et al., 1999; Ris et al., 1999) wave models (Hemer et al., 2012a). The WaveWatch III model was implemented over a larger domain (90°—240°E, 65°—0°S) at 0.5° resolution, in order to encompass all of the generation regions for waves that influence the New South Wales (NSW) coast on the east Australian margin. The SWAN model was implemented over a smaller (150°—155°E, 38°—25°S) 0.1° resolution domain that was nested within the WaveWatch III domain over the region of interest along the NSW coast (Figure 1). The wave climate over the 1981–2000 period is represented by the results of seven wave model runs described by Hemer et al. (2012a), which include the NBA1990 ensemble, the BA1990 ensemble, and run CNRA1990 for the NBAP, BAP, and PP approaches, respectively (Table 1, column 1).

Figure 1.

Wave model grids with 200 m depth contour shown. (a) 0.5° resolution WaveWatch III domain. Box shows nested SWAN model domain, as shown in more detail in plot (b). (b) 0.1° resolution nested SWAN domain. Western boundary is at 150°E. Black dots indicate location of waverider buoys used for model validation, and at which projections are given

Table 1. Summary of the 28 wave model simulations (making up nine ensembles and the NRA-2 hindcast)
  1. Three CCAM-GCM ensemble indicates three wave model simulations, forced with CCAM downscaled CSIRO Mk3.5 GCM, GFDLcm2.0 GCM, and GFDLcm2.1 GCM, respectively.

CNRA98 P2090A2
NRA-23 CCAM-GCM ensemble
  Scenario: SRESA2
3 GCM ensemble3 CCAM-GCM ensemble3 CCAM-GCM ensemble
 Scenario: SRES A2Scenario: SRES A2
 3 CCAM-GCM ensemble3 CCAM-GCM ensemble
 Scenario: SRES A2Scenario: SRES A2
3 GCM ensemble3 CCAM-GCM ensemble3 CCAM-GCM ensemble
 Scenario: SRES B1Scenario: SRES B1

To investigate projected changes in wave climate, a further 21 wave model runs were carried out (Table 1) made up of seven 3-member ensembles. Three-member ensembles were defined to estimate uncertainty associated with GCM forcing, derived from the CCAMGFDLCM2.0, CCAMCSIROMk3.5, and CCAMGFDLCM2.1 datasets. The seven ensemble sets were established with three aims: (1) Investigate the sensitivity to forcing scenario, comparing projections from the high emission SRES A2 scenario and low emission SRES B1 scenario, (2) Investigate the sensitivity of projections to NBAP, BAP, and PP approaches to derive forcing winds, and (3) Determine whether projected changes are linear in time, by projecting wave conditions for two future time slices (2031–2050 and 2081–2100).

Analysis of the 20-year long model runs focused on two key aspects of the wave climate: temporally averaged conditions (20-year annual mean and mean annual cycle) and values during storm events. Storm events for the NSW coast are defined as waves exceeding 3 m significant wave height and account for approximately 6% of all wave measurements (Lord and Kulmar, 2000). At the location of each of the waverider buoys, the 50th and 99th percentiles (long-term percentiles) of wind speed and significant wave height, and mean peak wave direction, were calculated. The 99th percentile was considered as a robust measure of extreme conditions and calculated in order to determine whether changes in the upper tail had occurred. A more rigorous analysis of storm events was carried out by analysing the exceedance of 3 m wave heights, so that comparisons with prior wave studies on the NSW coast could be carried out (Harley et al., 2010). A declustering procedure was used such that exceedance events within 3 d of each other were merged into a single storm event. This analysis defined Nex as the frequency of storm events exceeding the 3 m threshold in the 20-year time series, Dur as the mean duration of the storm events [i.e. the time for which the significant wave height (or wind speed) were above the threshold], DirE as the mean direction over the period that waves exceeded the 3 m threshold, and Etot as the total cumulative storm energy over the 20-year period. The latter quantity represents the erosive potential of the storm record, following Mendoza and Jimenez (2006), where the cumulative energy of a single storm is defined by

equation image

where N is the number of 6-hourly data points i in the storm event (HS, i > 3 m), ρ is the mass density of sea water (1025 kg m−3), g is the acceleration due to gravity (9.8 m s−2), and Δt is the temporal resolution of the dataset (6 h), such that NΔt represents the duration of the storm. Units are expressed in MJh m−2. Etot is the 20-year sum of E for all storms in the record.

Each of the parameters defined above were determined from each of the 28 model runs (7 present climate runs and 21 present future climate runs), at the location of six waverider buoys along the NSW coast (Table 2). For each of the nine 3-member ensembles, the ensemble mean, minimum and maximum values within the ensemble were determined to provide an indication of the magnitude of uncertainty associated with climate sensitivity. Projected changes in wave climate were determined from the differences between the future time-slice ensembles and the respective NBA1990, BA1990, or CNRA1990 present climate run ensembles at these sites.

Table 2. Summary of buoys used for model validation
Site nameLatitude (°S)Longitude (°E)Water depthDate of first observationDate of final observation
  1. Data obtained from the Manly Hydraulics Laboratory. NSW State Govt. A 20-year mean is determined where possible, regardless of dates.

  2. a

    Denotes directional waverider buoy.

  3. b

    Sydney waverider was directional from 03-Mar-1992.

Byron Baya28.82153.737101-Jan-198131-Dec-2000
Coffs Harbour30.36153.277201-Jan-198131-Dec-2000
Crowdy Head31.83152.867901-Jan-198631-Dec-2005
Port Kembla34.47151.037801-Jan-198131-Dec-2000
Batemans Baya35.71150.347301-Jan-198731-Dec-2005

3. Results

3.1. Surface winds

Figure 2 shows the percentage change in projected ensemble mean wind speed between the 2081 and 2100 and the present climate between 1981 and 2000 wind field for the A2 scenario. The three maps correspond to the three approaches used to produce the projected fields (NBA, BA, and PP), and the ensemble mean is determined from the three-member ensemble of CCAM-downscaled GCMs. The approaches display a consistent projected response of a southward shift in the position of the subtropical ridge (STR). The east-west line of discontinuity shows the mean position of the STR for the present climate. The CCAM-derived wind field (NBA, Figure 2(a)) shows very little latitudinal variability in the position of the STR, and this is consistent with the speculation by Hemer et al. (2012a) that the climate model circulation has an exaggerated zonal flow. Figure 2(b) and (c), indicating projected changes for the BA and PP fields, respectively (which are both dependent on NRA-2 wind fields), displays greater latitudinal variability in the position of the present STR. Each approach shows a strengthening of the easterly winds north of the position of the STR, and a weakening of the westerly winds south of the STR. At the south of 45°S, a strengthening of the westerly winds is observed. Marked differences between the NBA projected changes and the BA and PP projected changes are observed adjacent to the NSW coast. NBA fields suggest a stronger increase of wind speeds north of the STR, and a relatively small decrease in wind speeds south of the STR. In contrast, the BA and PP fields show weaker increase in wind speeds north of the STR and a stronger decrease in wind speeds south of the STR.

Figure 2.

Fraction change (i.e. 0.5 is 50%) in projected ensemble mean wind speed between the 2081 and 2100 and the present climate between 1981 and 2000 wind field, for the A2 scenario. The three maps correspond to the three approaches used to produce the projected fields [(a) NBA, (b) BA, and (c) PP], and the ensemble mean is determined from the three-member ensemble of GCMs downscaled using the CCAM model

Figure 3 shows bias-adjusted wind speed and direction (mean and storm wind conditions) along the 155°E meridian for the 1981–2000 and 2081–2100 time-slices. The 2081–2100 conditions are shown for the SRES A2 scenario. The ensemble spread for the 1981–2000 time-slice is very narrow, as a result of the bias-adjustment procedure aligning all ensemble members to the NRA-2 wind distribution. Winds for the 2081–2100 time-slice show a range of uncertainty dependent on which GCM is downscaled by CCAM. This ensemble range is, however, small relative to the range observed between original GCM winds (not shown). The differences between the present and projected wind conditions for the A2 scenario are similar regardless of whether the NBAP, BAP, or PP approach is used (not shown). Projected changes in mean wind speed are relatively small (much smaller than the magnitude of adjustment from the unadjusted winds to NRA-2 winds, which was presented by Hemer et al., 2012a). An increase in the speed of the westerlies south of 40°S is projected. Otherwise changes in mean wind speed are minimal. The average position of the subtropical high is defined in the mean direction plot as the transition from easterlies to westerlies. A projected southward shift of approximately 3° in the position of the subtropical high is seen, and a much stronger zonal component to the winds in the adjacent latitudes, suggesting that this ridge of high pressure is consistent in its position during the future time-slice. Projected winds for the 2081–2100 time-slice for the SRES B1 scenario (not shown) have very similar characteristics, except that the magnitude of projected change is approximately one-half of that for the A2 scenario.

Figure 3.

Bias-adjusted surface winds along 155°E for 1981–2000 time-slice (dark shading) and projected winds for the 2081–2100 period under an SRES A2 scenario (light shading). The range of magnitudes represents the ensemble range from CCAM that has downscaled three GCM solutions (CSIRO Mk3.5, GFDLCM2.0, and GFDLCM2.1). The ensemble mean is represented by the solid line. The dotted line indicates present climate QuikSCAT winds, and the dashed line indicates 1981–2000 NRA-2 winds. (a) Mean wind speed, (b) mean wind direction, (c) 99th percentile of wind speed, and (d) mean wind directions of winds stronger than the 99th percentile

The annual cycle of bias-adjusted (BA) surface wind conditions for location (155°E, 30°S) for 1981–2000 and 2081–2100 are shown in Figure 4. The projected winds are shown for the SRES A2 scenario. The SRES B1 scenario projected winds show similar characteristics but with smaller magnitudes of change. We see that although mean wind speed shows very little projected change at this location (Figure 3), a shift in the annual cycle of winds can be seen. The annual minimum wind speed that occurs in October in the present climate occurs in August in the projected climate. A mean wind speed maxima that presently occurs in July (seen in observed datasets also) is projected to be significantly reduced in the future climate, suggesting a reduction of winter wave activity. Projected mean wind speeds are anticipated to be stronger in spring but weaker in summer and winter. Storm wind speeds, represented by the 99th percentile show relatively consistent projected decrease throughout the year of approximately 0.5 ms−1. Projected changes in the seasonality of mean wave direction are also seen by the projected short period of westerly winds at this latitude (July and August only), in contrast to the period June through September in the present climate.

Figure 4.

20-year mean annual cycle of bias-adjusted surface winds at 155°E, 30°S for 1981–2000 time-slice (dark shading) and projected winds for the 2081–2100 period under an SRES A2 scenario (light shading). The range of magnitudes represents the ensemble range from CCAM that has downscaled three GCM solutions (CSIRO Mk3.5, GFDLCM2.0, and GFDLCM2.1). The ensemble mean is represented by the solid line. The dotted line indicates present climate QuikSCAT winds, and the dashed line indicates 1981–2000 NRA-2 winds. (a) Mean wind speed, (b) mean wind direction, (c) 99th percentile of wind speed, and (d) mean wind directions of winds stronger than the 99th percentile

Figure 5 shows changes in the mean annual cycle of mean wind speed and direction for present and projected (2081–2100) time-slices at location (155°E, 30°S) for each of the three approaches (NBA, BA, and PP). The projected changes in wind speed are relatively consistent for all three methods, but there is considerable variability in the wind directions. Given that each ensemble consists of only three members, estimates of statistical significance are severely hampered. Thus, we define a ‘notable’ change as one where the change between current and future climate is large enough that there is no overlap between the ensemble ranges of the present and future projected conditions. Mean wind speed shows a projected decrease throughout the year. The magnitude of this projected decrease is notable for February to July in the NBA winds. BA winds indicate notable change during August and November also, and PP winds show notable change for all months except December. The decrease in wind speed is greatest during the winter months, reaching a peak of approximately 1 m s−1 decrease in June in all records. It is during this period where winds are predominantly from the south. There is a distinct change in wind direction during the Austral winter and spring in the NBA winds with a large anticlockwise rotation of the southerly winds from a July mean direction of approximately 180°N in the present climate to approximately 100°N in the project climate. A similar change is projected in the BA winds. In the present climate, the mean wind direction in June to September is from the south-west (approximately 220°N) and returns to easterly in October. In the projected climate, the period of south-westerly winds is shortened considerably (June only), and an extended period of easterlies commence in September. This projected change is consistent with the southward shift in the position of the STR shown in Figures 2 and 3.

Figure 5.

Mean annual cycle of mean wind speed (a,c,e) and mean wind direction (b,d,f) at 155°E, 30°S. Datasets shown are NBA (a,b); BA (c,d); and PP (e,f). Present climate (1981–2000) winds are shown in dark shading, projected winds (2081–2100) for the A2 scenario are shown in light shading. The ensemble mean and range are shown. In the PP dataset, present winds are represented by the NRA-2 winds, and no ensemble range is shown. Subplots (g) and (h) display projected change in wind speed and direction, respectively. The bold line indicates change between means, and the maximum range of uncertainty is also shown. Darkest shading correspond to NBA winds, lightest shading corresponds to PP winds, and mid-shading corresponds to BA winds

3.2. Surface waves

The winds derived in the previous section (Section '3.1. Surface winds') are used in this section to obtain projections of wave climate. Figure 6 shows the present (1981–2000) and projected (2081–2100) climate ensemble range of Sydney wave conditions for the SRES A2 scenario, as determined using the NBA, BA, and PP projection methods. The three methods show qualitative agreement in the direction of the change with mean significant wave height projected to decrease (Figure 6(a)). However, there is a large range in the projected future changes arising from different methods used to apply the wind forcing to the wave models. The magnitude of this decrease ranges from small (less than 2 cm decrease projected from the NBA runs) to considerable and notable by our definition above (a 13 cm decrease in mean HS projected from the PP runs). Lines on the plot correspond to the different GCMs that CCAM was forced by. The maximum decrease is projected to occur when CCAM is forced by CSIRO Mk3.5 conditions. All methods project a notable change in mean wave direction at the location of the Sydney waverider buoy (Figure 5(e)). However, the magnitude of the projected change of less than 5° at this location in approximately 100 m water depth is relatively small.

Figure 6.

Projected changes in wave properties at the location of the Sydney Waverider buoy under the SRES A2 scenario X indicates results from CNRA1990 run. Solid circle indicates buoy-derived value for the present climate. The ensemble mean (black line) and range are shown for each runset. Results from CCAM CSIRO Mk3.5 runs are shown with solid lines, CCAM GFDLcm2.1 with dash lines, and CCAM GFDLcm2.0 with dash-dot lines. Wave properties shown are (a) mean significant wave height, (b) 99th percentile of HS, (c) number of exceedances of HS = 3 m, (d) mean duration of exceedances of HS = 3 m, (e) mean wave direction, (f) mean wave direction of waves larger than 3 m, and (g) the mean annual storm energy

Notable decreases in storm wave heights (represented by HS99, Figure 6(b)) are projected using both BA and PP methods. The NBA projects decreasing HS99, but overlap is observed between the present and projected ensemble ranges. A notable clockwise rotation of storm wave directions of approximately 5° is projected using PP winds, but NBA and BA winds show no projected change. The ensemble range of the number of storm events (occurrences of HS > 3 m; Figure 6(c)) shows no consistent increase or decrease between ensemble members, and so we consider that changes to this quantity are less certain. The duration of storm events (Figure 6(d)) is projected to decrease notably by BA and PP methods. A non-notable decrease in duration is projected using NBA winds. These projected changes in duration of storm events lead to a projected decrease in mean annual storm energy (Figure 6(f)). The projected decrease in storm energy ranges from relatively small (less than 5% decrease using NBA winds) to large and notable (approximately 40% decrease using PP winds).

Figure 7 shows the mean annual cycle of mean HS and mean wave direction at the location of the Sydney buoy, for the present and projected future climates, from the three projection methods (NBA, BA, and PP) as well as the differences in the average annual cycle between the two time periods. The changes in the NBA and BA runs are similar with both indicating a decrease in mean HS in winter, particularly in June and a projected increase in spring, in November in both run sets, and in October in the NBA runs. There is also a projected decrease in mean HS during February, which is notable in the NBA runs. The perturbed runs display a contrasting signal, where mean HS is projected to decrease uniformly throughout the year by approximately 10–20 cm. The BA B1 scenario runs project changes in HS and direction (not shown), which are similar to the A2 scenario runs (Figure 7, second row). However, one notable difference between the B1 and A2 simulations is a projected decrease in mean HS in the B1 run during February. All other projected changes in HS are qualitatively consistent with the A2 scenario but are not notable in magnitude. An anticlockwise rotation of mean wave direction is projected predominantly during the Austral winter months, from all runsets. The change is notable in all runsets for months May and June. NBA and BA runs show notable change in October. NBA and PP runs project notable change in August and September.

Figure 7.

Mean annual cycle of mean HS (a,c,e) and mean wave direction (b,d,f) for present (1981–2000; dark shading) and projected (2081–2100; light shading) time-slices at Sydney. (a) and (b) show NBA runs; (c) and (d) show BA runs and E and F show PP runs, for SRES-A2 scenario. Ensemble range representing GCM uncertainty is shown. (g) and (h) show projected change between ensemble mean HS and Dir, respectively

Table 3 shows the projected change in wave conditions between the present (1981–2000) and projected (2081–2100) climates, for the SRES A2 scenario, at buoy locations along the NSW coast (see Figure 1 for buoy locations). Projected decrease in mean HS is larger further northwards. At Batemans Bay, the NBA runs show a projected increase in mean HS while PP runs project a decrease. Further northwards, all run sets project decreasing HS. This decrease in mean HS is notable at the two northern sites (Byron Bay and Coffs Harbour) in the NBA runs, and at all sites in the PP runs. (Note: our definition of ‘notable’ may be misleading in relation to the PP runs, as there is no ensemble range in the present climate, and the possibility of ensemble overlap is reduced). The projected decrease is not notable in any of the BA runs. The B1 scenario BA runs show non-notable decreases of HS that are smaller than those observed for the A2 scenario. A decrease in storm wave heights HS99, is also projected. The PP runs show notable decrease at all sites except Port Kembla. The BA runs show significant decrease from Sydney to Coffs Harbour, and the NBA runs project no consistent change.

Table 3. Projected change in wave properties between present (1981–2000) and projected (2081–2100) time-slices
  1. SRES A2 scenario (except last column that represents B1 scenario). Results from the three methods (NBA, BA, and PP) are shown. Uncertainty range indicates maximum differences between three-member ensemble ranges in the two time-slices. Significant changes are shown in bold. Projected changes that are inconsistent (i.e. increase in HS, or clockwise rotation in Dir) are in italics.

Byron Bay− 0.13(−0.15, − 0.10)− 0.06(−0.11, − 0.02)− 0.23(−0.27, − 0.21)− 0.06(−0.08, − 0.02)
Coffs Harbour− 0.08(−0.09, − 0.06)− 0.06(−0.11, − 0.01)− 0.18(−0.22, − 0.16)− 0.05(−0.07, − 0.01)
Crowdy Head0.01 (−0.04, 0.04)− 0.04(−0.11, 0.00)− 0.14(−0.20, − 0.11)− 0.04 (−0.07, 0.00)
Sydney0.00 (−0.04, 0.04)− 0.04(−0.12, 0.00)− 0.14(−0.20, − 0.10)− 0.04 (−0.10, 0.01)
Port Kembla− 0.03 (−0.08, 0.02)− 0.03 (−0.11, 0.02)− 0.07(−0.13, − 0.02)− 0.04 (−0.10, 0.01)
Batemans Bay0.06 (0.00, 0.12)− 0.02 (−0.10, 0.02)− 0.16(−0.22, − 0.12)− 0.04 (−0.10, 0.01)
Byron Bay− 0.01 (−0.19, 0.12)− 0.17(−0.26, − 0.03)− 0.79(−0.93, − 0.71)− 0.07 (−0.14, 0.01)
Coffs Harbour− 0.06 (−0.24, 0.05)− 0.20(−0.32, − 0.10)− 0.58(−0.72, − 0.45)− 0.09 (−0.24, 0.02)
Crowdy Head0.06 (0.07, 0.15)− 0.20(−0.29, − 0.10)− 0.44(−0.56, − 0.30)− 0.11(−0.18, − 0.03)
Sydney− 0.08 (−0.34, 0.13)− 0.23(−0.48, − 0.04)− 0.36(−0.60, − 0.20)− 0.14 (−0.39, 0.02)
Port Kembla− 0.16 (−0.52, 0.11)− 0.19 (−0.41, 0.07)− 0.06 (−0.26, 0.14)− 0.12 (−0.37, 0.06)
Batemans Bay0.01 (0.47, 0.41)− 0.13 (−0.46, 0.23)− 0.51(−0.74, − 0.34)− 0.09 (−0.36, 0.11)
Byron Bay− 1.8 (−8.7, 3.3)− 3.5 (−6.2, 0.2)− 3.9(−5.4, − 1.5)− 2.0 (−4.4, 0.8)
Coffs Harbour− 5.5(−11.7, − 0.8)− 3.0 (−5.2, 0.1)− 4.0(−5.3, − 1.5)− 2.3 (−4.2, 0.0)
Crowdy Head− 3.9 (−9.4, 0.4)− 3.0(−4.6, − 0.9)− 4.4(−5.7, − 2.4)− 1.9(−3.1, − 0.5)
Sydney− 6.0(−7.5, − 4.4)− 3.8(−5.5, − 0.7)− 3.1(−4.9, − 0.9)− 2.7(−4.0, − 0.9)
Port Kembla− 5.8(−7.6, − 3.8)− 2.9(−4.6, − 0.4)− 3.1(−4.4, − 1.5)− 1.8(−3.0, − 0.5)
Batemans Bay− 1.8 (−4.9, 3.0)− 3.9(−5.5, − 2.1)− 6.6(−7.9, − 4.8)− 2.6(−3.4, − 1.5)
Byron Bay16.9 (11.8, 21.4)− 1.0 (−5.5, 6.3)− 9.2(−12.5, − 5.4)1.9 (1.4, 6.4)
Coffs Harbour12.3 (6.7, 21.0)− 2.0(−4.2, − 0.6)1.6 (−1.5, 4.0)1.2 (− 0.3, 2.7)
Crowdy Head7.3 (1.0, 13.9)− 0.4 (−4.2, 2.8)1.0 (0.2, 1.7)1.5 (1.0, 4.1)
Sydney0.4 (−2.1, 3.9)0.5 (1.0, 2.5)7.4 (6.8, 7.9)1.9 (2.7, 8.1)
Port Kembla− 0.7 (−6.5, 5.2)1.8 (1.3, 5.5)9.7 (8.2, 11.2)2.7 (3.0, 9.0)
Batemans Bay− 1.8 (−8.3, 7.4)− 1.1 (−5.2, 2.1)− 1.8 (−5.3, 1.2)2.0 (4.2, 8.7)

An anticlockwise rotation of mean wave direction is projected at all sites along the NSW coast (Table 3). This rotation is predominantly notable along the southern coast using all approaches, but the magnitude (approximately 5°) is small. Storm wave direction displays a large range of uncertainty, with each of the three methods showing contrasting projected conditions. Projected changes are greater at the northern buoy sites in all runs. The PP runs project an anticlockwise rotation of 9.2° at Byron Bay, whereas the NBA runs project a clockwise rotation of almost 16.9° in storm wave direction at Byron Bay. The BA runs project little to no change in storm wave direction.

Figure 8 shows the ensemble range of mean HS across all NSW buoys, for NBA and BA runs, for time-slices 1981–2000, 2031–2050, and 2081–2100. The projected change in wave height is nearly linear in time, with the projected changes for 2031–2050 displaying similar characteristics as for the 2081–2100 time-slice but of smaller magnitude.

Figure 8.

Summary projected change in mean significant wave height (HS; in metres) from all 28 wave model simulations (nine ensembles) at buoy locations on the NSW coast. (a) Byron Bay; (b) Coffs Harbour; (c) Crowdy Head; (d) Sydney; (e) Port Kembla; and (f) Batemans Bay. Black circle represents buoy measured mean HS; Cross represents results of model forced with NRA-2 winds (CNRA98 thick black lines represent ensemble means. Results for individual runs are denoted for the NBA-A2 (black lines, joining darkest boxes) and BA-A2 (grey lines, joining dark boxes) runs. Solid lines represent runs forced with CCAM CSIRO Mk3.5; dash lines—GFDLcm2.1 runs; and dash-dot lines—GFDLcm2.0 runs

4. Discussion

Dynamical wave projections for the NSW coast under a range of forcing conditions were undertaken to investigate three different sources of uncertainty. The first two sources, those of climate sensitivity and emissions scenarios, have been widely considered, being associated with GCMs and forcing scenarios. We consider two emission scenarios (SRES A2 and B1) and three GCMs (from two modelling centres). While two of the GCMs considered (GFDLcm2.0 and GFDLcm2.1) are from the same modelling centres, and have some similarities, the dynamical core of these models are different from one another, along with other differences (e.g. cloud tuning, ocean and land surface configurations), which lead to significant differences in model projection properties (Delworth et al., 2006). Hemer et al. (2012a) show dynamical downscaling of the GCMs using the CCAM model reducing the range of uncertainty in surface wind speeds within the considered multi-model ensemble. The third source of uncertainty relates to the correction of climate model winds to produce a wind field that is capable of generating a reliable wave climate. The effectiveness of the wind corrections was investigated by Hemer et al. (2012a), who found that using bias-corrected climate model surface winds to force a wave model over the 1981–2000 period significantly improved the representation of the modelled present day wave climate. Having demonstrated how bias-correction of the surface winds leads to improved representation wave climate simulations in that study, this study goes on to investigate the sensitivity of different assumptions required to develop future climate projections of wind and waves using the bias-correction method. Three approaches have been compared: (1) The projected winds are represented by uncorrected winds, (2) Projected winds are represented by bias-adjusted winds, assuming that the corrections for the projected climate are the same as for the present climate, and (3) Projected winds are represented by perturbations of the present wind fields, where the perturbations applied are determined by the difference between climate model simulated future and present-day uncorrected wind fields. At a qualitative level, we see that the projected wave climate shows a relatively robust change, despite the range of forcing conditions that have been considered. The projected changes consist of a relatively small decrease in wave heights along the NSW coast, which is larger further northwards, accompanied by a small anticlockwise rotation of wave direction. The magnitude is, however, small, and the consequent coastal impacts because of these projected changes in wave climate through the twenty-first century are expected to be minimal.

The large-scale climatic feature, which is driving the projected changes in wind and wave climate reported in this study, is the position and intensity of the subtropical ridge (STR). The associated pattern of wind speed change is a common feature of present GCMs as shown by McInnes et al. (2011). The position of the STR in the present climate can be inferred from the discontinuity of change in Figure 2. The difference in shape of the STR across the Tasman Sea was analysed by Grose et al. (2010) in MSLP fields derived from the downscaled CCAM model runs. They found a reduced curvature in the isobars in the region in comparison to those derived from NRA-2 and noted that the reduced curvature is a common deficiency in GCMs, which was not improved by the downscaling procedure. In assessing the usefulness of the climate model winds for dynamical wave studies, Hemer et al. (2012a) identified that storm systems may be propagating across the Tasman Sea too quickly to adequately resolve storm wave systems. This accentuated propagation speed is consistent with the increased zonal flow, which is characterised by the uniform latitude of the position of the STR in the region. The projected southward shift in the position of the STR drives the projected change in surface wind conditions (Figure 3). Drosdowsky (2005) assessed historical trends in the position of the STR in the Australian region and reported no trend during the latter half of the twentieth century. Despite the larger change in position occurring during summer months, the proximity of the position of the STR to the NSW coast means that the winter wind climate change is more sensitive to the projected changes than the summer wind climate. Each of the methods used to derive projected winds (NBA, BA, and PP) result in notable changes in wind speed and direction during the winter months. In the NBA and BA methods, relatively consistent decreases in wind speed, and a clockwise rotation of wind direction, during Austral winter months occurs. In the PP method, there is less annual variation in the projected change in wind conditions. This seems to be a consequence of the way that the PP method was applied here whereby annual changes to winds were evaluated and so the NRA-2 wind field was perturbed by a more or less constant value throughout the year. In future applications, the PP method may be applied by determining perturbation matrices for each month or season of the annual cycle, and thus account for annually varying drivers of change in the projections.

Projected changes in the wind-wave climate also appear to be driven by the southward shift in position of the STR. As with the winds, the PP method, as applied here, lacks the ability to resolve the seasonal variability of projected change (Figure 8). However, the NBA and BA methods show relatively consistent features, which indicates that the winter wave climate (characterised by more southerly waves) spans a shorter period in future scenarios. In the present Sydney wave climate (Figure 8), we see that the easterly wave climate associated with summer conditions extends from November to March. In the projected wave climate, these easterly conditions extend from October to March, and mean wave directions have a more eastward component than the present climate for the remaining period of the year. We attribute this rotation to be associated with the STR not extending as far northwards during the winter months, thus reducing the influence of southerly storm systems that dominate the NSW wave climate during this period. It is difficult to resolve these changes in the annual cycle of HS at Sydney, largely because of the relatively flat annual cycle that is observed there. However, the most significant changes in wave heights between present and projected future wave climates are observed in June (Figure 8).

The largest future decrease in HS is projected to occur at the northern end of the NSW coast. The reduced northwards extent of the position of the STR during winter months leads to a greater decrease in the influence of the southerly wave systems at the more northerly sites. There is no evidence for any significant change in the circulation patterns that drive the north-easterly wave events in this region. The trade winds remain relatively steady near the Australian coast, and when not assessed in detail, there is no evidence to suggest changes in frequency, intensity, or tracks of tropical cyclone systems.

Of the three sources of uncertainty that have been considered in this study, we find that the method of adjustment of surface winds (i.e. the NBA, BA, and PP methods) leads to the largest variation in projected future wave climate compared to the climate sensitivity and emissions scenarios that were also considered. We found that the CCAM model, with the improved spatial resolution, reduces the climate sensitivity and leads to a narrower range of projected changes in the future climate compared to that seen in the three GCMS (CSIRO Mk3.5, GFDLcm2.0, and GFDLcm2.1). For example, the projected change in HS at Sydney for the A2 scenario shows an ensemble range of at most 0.12 m between the GCM ensemble (using BA winds). The method used to produce wind projections (NBA, BA, and PP) produces a range of projected changes of 0.14 m. In total, we see a range of projected change in HS at Sydney from − 0.2 m to + 0.04 m. The uncertainty associated with emission scenarios, in this study considered a high-emission SRES A2 and low emission SRES B1 scenario, leads to a relatively narrow range of response. Again for Sydney, the projected change in HS at Sydney of − 0.04 m is the same regardless of the emission scenario (using BA winds). Considering all three sources of uncertainty significant projected changes in wave climate amount to a decrease in mean HS of only 0.12 and 0.09 m at Byron Bay and Coffs Harbour, respectively, and an anticlockwise rotation of mean wave direction of only 3.9° and 3.4° at Sydney and Port Kembla, respectively.

A previous study applied a simple fetch modelling technique to estimate the projected change in wave conditions under future climate conditions (McInnes et al., 2007). They estimated wave conditions from surface winds derived from the CCAM model, which downscaled the results of the CSIRO Mark 2 (CCM2) and CSIRO Mark 3 (CCM3) GCM simulations, for the SRES A2 scenario. Using this approach, McInnes et al. projected increasing mean HS of approximately 0.1 m from CCM3 (associated with increasing winds) and decreasing mean HS of approximately 0.1 m from CCM2 (associated with decreasing winds). Changes in mean wave direction were less than 1° for both models. The results of McInnes et al. contrast with this study, where a relatively robust decrease in mean HS, and anticlockwise rotation of wave direction is projected.

The mean magnitude of projected change in wave climate is relatively small. The projected changes in wave direction of less than 5° at the location of the buoys, which are in water depths of approximately 100 m are unlikely to have influence on coastal littoral transports, once the wave refraction across the continental shelf has been taken into consideration. The inter-annual variability of Sydney wave direction that has been reasoned to drive rotation of embayed beaches along the NSW coast is of order 10° (Ranasinghe et al., 2004; Goodwin, 2005). However, there is a range of uncertainty around these projected changes that should be considered. The total range of projected changes in mean wave direction is an anticlockwise rotation of 11.7° at Coffs Harbour to a clockwise rotation of 3.3° at Byron Bay. This range of uncertainty is relatively small, and even at the outer limits of this range, the consequent coastal impacts of such changes could be expected to be relatively minor. The total range of projected changes in mean HS is a decrease of 0.27 m at Byron Bay to an increase of 0.12 m at Batemans Bay. There is lower confidence in the projected storm wave climate. This arises because of the inadequacies in the representation of storm systems in the CCAM simulations over the Tasman Sea, which were identified by Hemer et al. (2012a). These problems are either related to poor spatial resolution of storm systems or the transience of storm systems across the Tasman Sea. Large notable changes in the storm wave climate are projected, particularly when using the PP approach to generate a projected future wind field. In these runs, we see a decrease in HS99 of nearly 1 m projected at Byron Bay. The PP runset (for SRES A2 scenario) projects a decrease in the annual mean storm wave energy at the location of the Sydney waverider buoy of approximately 40% (and a range including over 50% decrease).

Global wave climate projections have been generated using a statistical approach by Wang and Swail (2006) and a dynamical approach by Mori et al. (2010). Qualitative comparisons of the results from this study with these available global projections indicate agreement of a small decrease in HS along the east Australian coast. Only qualitative comparisons are sensible, as each of these studies has generated projections for slightly different periods, for different emission scenarios, using different methods. In the present study, considerable computing resources have been required to undertake the regional climate model downscaling (Six - 2 scenarios, three GCMs—140-year runs), and the wave modelling (Twenty-eight 20-year wave model runs, which include modelling with a coarse resolution WW3 model and a nested fine resolution SWAN model). Thus, the advantages of undertaking global studies as advocated by Hemer et al. (2010) become evident. We therefore reiterate the proposal put forward by Hemer et al. (2010) that modelling effort is better focussed on generating global wave projections under a larger range of emission and GCM forcing, using several wave models (statistical or dynamical), within an internationally coordinated framework. Such a program will enable projection results consistent with regional studies over the global domain, but importantly will also enable the multiple sources of uncertainty in the wave climate projections to be quantified.

5. Conclusions

The wind-wave climate on the east Australian continental margin and their potential changes under the effects of projected climate change have been investigated using state-of-the art numerical modelling techniques. The wind-wave climate is modelled for three 20-year time-slices (1981–2000, 2031–2051, and 2081–2100) using a dynamical wave modelling approach using the WaveWatch III spectral wave model, with a nested SWAN domain in the region of interest. The main conclusions arising from this study are listed as follows:

  1. The uncertainty associated with the three methods used to derive projected wind conditions (unadjusted, bias-adjusted, and perturbed approaches) is greater than the uncertainty associated with the choice of CCAM downscaled GCM that considered the CSIRO Mk3.5, GFDLcm2.0, and GFDLcm2.1 GCMS, or the emission scenario with which the GCM was forced, which considered the high-emission SRES A2 and the low-emission SRES B1 scenarios. However, we note that a larger range of climate sensitivity may have resulted if wind forcing was obtained directly from a range of GCMs rather than the CCAM model used here. Furthermore, we note that our multi-model ensemble (three GCMs from two modelling centres) is limited, and a greater range of multi-model uncertainty might be expected if a larger ensemble was considered with the outlined dynamical approach.

  2. Projected changes in wave climate over the twenty-first century are relatively small but significant. A relatively robust decrease in mean wave heights of approximately 0.05–0.1 m is projected between the present time slice (1981–2000) and the end of the century (2081–2100) along the NSW coast. Projected changes are larger (and significant on the northern coast). A relatively small (approximately 5°) anticlockwise rotation in mean wave direction is also projected over the same period and region.

  3. The bias-adjustment procedure detailed by Hemer et al. (2012a) indicates reduced confidence in the climate model forced storm wave climate. Projected changes in storm wave climate display large ranges of uncertainty. While projected changes are typically small, in some circumstances, we project a large reduction in storm wave energy along the NSW coast (of up to 50% decrease).

  4. The results of this regional study are qualitatively consistent with available global wave projections. While this study used available CCAM simulations because the coarse spatial and temporal resolution output archived from previous CMIP experiments has been too limited for use in wave climate modelling, this will not be the case with the CMIP5 model simulations. Therefore, we advocate that wave climate projections effort be placed onto global studies within an internationally coordinated framework as proposed by Hemer et al. (2010, 2012b). Such an initiative will generate a larger ensemble of global wave projections so that a more complete estimate of the statistical confidence levels of global wave projections can be determined, including a larger range of GCMs, emission scenarios, and wave downscaling approaches. It will also ensure that wave climate projections are available for nations at most risk to potential change.


This research is a contribution from the Climate Variability and Change Program of the Centre for Australian Weather and Climate Research: A Partnership between the CSIRO and the Bureau of Meteorology and the CSIRO Climate Adaptation Flagship. The project is supported by research funds from the Australian Government Department of Climate Change. The NSW State Government Department of Environment and Climate Change provided in-kind support to this project. The surface wind projections were made available by the Antarctic Climate and Ecosystems Co-operative Research Centre's Climate Futures for Tasmania project, which was supported by the Australian Government through the Commonwealth Environmental Research Fund. We thank Jack Katzfey for useful discussions and comments, and an anonymous reviewer for their comments on an earlier version of this manuscript.