1 3

DOI 10.1007/s00382-015-2573-6Clim Dyn (2016) 46:123–133

Zonal winds and southeast Australian rainfall in global and regional climate models

Acacia S. Pepler1 · Lisa V. Alexander1 · Jason P. Evans1 · Steven C. Sherwood1

Received: 15 October 2014 / Accepted: 16 March 2015 / Published online: 28 March 2015 © Springer-Verlag Berlin Heidelberg 2015

1 Introduction

The southeastern region of Australia is home to more than 70 % of Australia’s populace, including the major cities of Sydney, Melbourne, Adelaide and Brisbane. The Murray-Darling river system in inland southeast Australia is Aus-tralia’s largest river system, with the Murray-Darling Basin producing more than a third of Australia’s food and agri-cultural income. In this region, correlations between rain-fall and indices of tropical variability are large, with the El Niño-Southern Oscillation (ENSO) and the Indian Ocean Dipole contributing to large variability in rainfall, dam lev-els and agricultural output (McBride and Nicholls 1983; Meyers et al. 2007; Ummenhofer et al. 2009; Risbey et al. 2009b; Timbal and Hendon 2011). Variations in the latitude of the polar westerlies [Southern Annular Mode (SAM)] and the subtropical ridge are also tied to rainfall variabil-ity in the region, although these are not fully independent of the drivers of tropical variability (Meneghini et al. 2007; Williams and Stone 2009; Timbal and Drosdowsky 2013; Hendon et al. 2014b). These relationships exhibit substan-tial spatial variability, with a notable difference between correlations along the coastal strip and those elsewhere in southeast Australia, related to the effects of local topogra-phy (Pepler et al. 2014a, b; Timbal 2010).

The southern parts of southeast Australia suffered a prolonged and severe drought during the years 1997–2009 (Murphy and Timbal 2008; Nicholls 2009; Timbal and Drosdowsky 2013). While the drought was followed by record-breaking rainfall during the La Niña years of 2010–2012 (Hendon et al. 2014a; Imielska 2011; Webb 2012), a long-term drying trend remains apparent during the May–October season, which is the main growing sea-son in southeast Australia (e.g. Pepler 2013). This is con-sistent with a southward shift of the subtropical ridge and

Abstract Southeast Australia is a region of high rainfall variability related to major climate drivers, with a long-term declining trend in cool-season rainfall. Projections of future rainfall trends are uncertain in this region, despite projected southward shifts in the subtropical ridge and mid-latitude westerlies. This appears to be related to a poor representa-tion of the spatial relationships between rainfall variability and zonal wind patterns across southeast Australia in the latest Coupled Model Intercomparison Project ensemble, particularly in the areas where weather systems embedded in the mid-latitude westerlies are the main source of cool-season rainfall. Downscaling with regional climate models offers improvements in the mean rainfall climatology, and shows some ability to correct for poor modelled relation-ships between rainfall and zonal winds along the east coast of Australia. However, it provides only minor improve-ments to these relationships in southeast Australia, despite the improved representation of topographic features. These results suggest that both global and regional climate mod-els may fail to translate projected circulation changes into their likely rainfall impacts in southeast Australia.

Keywords Rainfall · Circulation · Models · Downscaling · Australia

* Acacia S. Pepler a.pepler@student.unsw.edu.au

1 The Australian Research Council Centre of Excellence for Climate System Science, and Climate Change Research Centre, University of New South Wales, Sydney, Australia

124 A. S. Pepler et al.

1 3

expansion of the tropics, as well as a strong positive trend in the SAM (Cai et al. 2005; Murphy and Timbal 2008). However, despite the long-term observed drying trends, there is little consistency between the rainfall projections of different Global Climate Models (GCMs) across southeast Australia (Collins et al. 2013).

Numerous studies have identified the mid-latitude west-erlies, with the associated cold fronts and cut-off lows, as the major source of cool-season rainfall in much of inland southeast Australia (Meneghini et al. 2007; Risbey et al. 2009a, b; Hendon et al. 2014b). In contrast, onshore easterly flow is a major contributor to rainfall on the east coast throughout the year (Rakich et al. 2008; Pepler et al. 2014b). This spatial variation in the influences of zonal wind is associated with local topography, and explains much of the spatial variation in seasonal rainfall and the impacts of major climate drivers. While the most recent suite of models from the Coupled Model Intercomparison Project (CMIP5; Taylor et al. 2012) have been assessed in terms of their ability to represent the relationships between southeast Australian rainfall and major climate drivers such as ENSO (King et al. 2014), their ability to represent the ties between zonal winds and Australian rainfall patterns is unknown. This is of particular relevance as model projec-tions favour a southward shift of the subtropical ridge and mid-latitude storm tracks under climate change (Bengtsson et al. 2006; Chang et al. 2012).

In this paper we seek to answer two questions. First, to what extent is the CMIP5 model ensemble able to repre-sent not just the seasonal patterns of rainfall and zonal wind across southeast Australia, but also the observed correlation patterns. Second, while regional downscaling has the abil-ity to substantially improve spatial rainfall patterns (Evans et al. 2011; Evans and McCabe 2013; Önol 2012; Tseli-oudis et al. 2012) when compared to GCMs, particularly

related to high or variable topography, can regional downs-caling also provide novel information as to the relationship between rainfall variability and zonal wind, or other broad-scale climate features?

Through answering these questions, we will improve our understanding of how well both the CMIP5 ensemble (Tay-lor et al. 2012) and regionally-downscaled projections such as the NARCliM project (Evans et al. 2014) can translate projected changes in broadscale climate features into their impacts on southeast Australian rainfall. This will provide additional guidance when interpreting projected rainfall trends, particularly in comparison to the trends from statis-tically downscaled models (Timbal et al. 2008).

2 Data

The main area of interest for this study is southeast Aus-tralia, which can be loosely defined as the mainland area south of 25°S and east of 135°E (Fig. 1). This definition is consistent with that used in a major research project, the South East Australian Climate Change Initiative (SEACI 2010), but extends further northward than some similar studies (Risbey et al. 2009a; Ummenhofer et al. 2009). The region is divided by an area of relatively high topography known as the Great Dividing Range (GDR), which runs roughly parallel to the east coast, resulting in distinctly different rainfall patterns east and west of the GDR (Dros-dowsky 1993; Pepler et al. 2014b; Timbal 2010).

This study focuses on one historical run from each of a suite of 37 CMIP5 GCMs (Table 1). As the spatial reso-lution varies between models, for ease of comparison the monthly rainfall accumulations and 850 hPa zonal winds during the 1950–2005 period were in each case regridded to a consistent 2.5° latitude–longitude grid.

Fig. 1 The NARCliM regional downscaling domain and the 50 km-resolution topography, with the inner domain marked in red (roughly corresponding to southeast Australia). The transect used to calculate the zonal wind anomalies is marked in magenta

125Zonal winds and southeast Australian rainfall in global and regional climate models

1 3

In addition to the GCM output, we also address the capacity of a regional climate model to improve the rep-resentation of rainfall patterns and variability. Regionally-downscaled precipitation and zonal wind data are obtained from the NARCliM regional downscaling project (Evans et al. 2014), which used the Weather Research and Fore-casting (WRF) model to downscale both reanalysis and

GCM data for the Australian domain. The model was run over two domains, with a larger 50 km domain extending well beyond the borders of the Australian region, while a 10 km domain was run over southeast Australia (Fig. 1). The project used three distinct sets of WRF parameters (Table 2) as well as four GCMs from the previous genera-tion of models (CMIP3; Meehl et al. 2007): CSIRO Mk3.5,

Table 1 The 37 CMIP5 models used during this study and their native spatial resolution, with model name italicised where a corresponding CMIP3-WRF simulation is used

The correlation for each model between cool season zonal windflow and rainfall averaged across the east coast and south coast regions (Fig. 4b) are also shown, with bold text indicating correlations that are statis-tically significant at the 5 % level

Model Resolution South coast correlation

East coast correlation

AWAP (observed) 0.05° × 0.05° +0.52 −0.38

ACCESS1-0 1.25° × 1.875° −0.07 −0.40

ACCESS1-3 1.25° × 1.875° 0.03 −0.50

bcc-csm1-1 2.8° × 2.8° 0.13 −0.35

bcc-csm1-1-m 2.8° × 2.8° 0.10 −0.36

BNU-ESM 2.8° × 2.8° −0.27 −0.47

CanESM2 2.8° × 2.8° 0.14 −0.02

CCSM4 1.25° × 0.9° 0.06 −0.39

CESM1-BGC 1.25° × 0.9° 0.18 −0.33

CESM1-CAM5 1.25° × 0.9° 0.21 −0.26

CESM1-FASTCHEM 1.25° × 0.9° 0.08 −0.11

CNRM-CM5-2 1.4° × 1.4° 0.19 −0.30

CNRM-CM5 1.4° × 1.4° 0.04 −0.18

CSIRO-Mk3-6-0 1.875° × 1.875° −0.15 −0.06

FIO-ESM 2.8° × 2.8° −0.03 −0.37

GFDL-CM2p1 2.0° × 2.5° −0.14 −0.51

GFDL-CM3 2.0° × 2.5° −0.15 −0.52

GFDL-ESM2G 2.0° × 2.5° −0.03 −0.55

GFDL-ESM2 M 2.0° × 2.5° 0.05 −0.55

GISS-E2-H-CC 2.0° × 2.5° 0.11 −0.33

GISS-E2-H 2.0° × 2.5° 0.18 −0.29

GISS-E2-R-CC 2.0° × 2.5° 0.15 −0.29

GISS-E2-R 2.0° × 2.5° 0.31 −0.39

HadGEM2-AO 1.25° × 1.875° 0.03 −0.47

HadGEM2-CC 1.25° × 1.875° 0.29 −0.42

HadGEM2-ES 1.25° × 1.875° 0.17 0.02

IPSL-CM5A-LR 1.25° × 2.5° 0.46 −0.44

IPSL-CM5A-MR 1.875° × 3.75° 0.32 −0.33

IPSL-CM5B-LR 1.25° × 2.5° 0.35 −0.44

MIROC5 1.4° × 1.4° −0.06 −0.21

MIROC-ESM-CHEM 2.8° × 2.8° 0.24 −0.18

MIROC-ESM 2.8° × 2.8° 0.13 −0.19

MPI-ESM-MR 1.875° × 1.875° 0.06 −0.24

MPI-ESM-P 1.875° × 1.875° 0.00 −0.31

MRI-CGCM3 1.125° × 1.125° 0.46 −0.51

MRI-ESM1 1.125° × 1.125° 0.28 −0.29

NorESM1-ME 1.875° × 2.5° 0.32 −0.07

NorESM1-M 1.875° × 2.5° 0.18 −0.26

126 A. S. Pepler et al.

1 3

CCCAM CGCM3.1, ECHAM5 and MIROC3.2, to produce 12 distinct regional climate projections. These combina-tions were chosen to maximise model independence as well as to span a broad range of potential future climate changes in southeast Australia. The NCEP1 reanalysis (Kalnay et al. 1996) has been downscaled over the period 1950–2010, while a combination of the historical simulations (1990–2005) and the projections using the A2 emissions scenario (2006–2009) provided boundary conditions for the down-scaled simulations of the CMIP3 models. While the par-ent CMIP3 models have also been regridded to the same 2.5° latitude-longitude grid used for the CMIP5 results, the 50 km Regional Climate Model (RCM) results have been regridded to a constant 0.5° latitude–longitude grid and the 10 km RCM to a 0.1° latitude–longitude grid.

Historical rainfall observations are obtained from the Australian Water Availability Project (AWAP) gridded

rainfall data (Jones et al. 2009), regridded to the same 2.5° resolution. While this product has some issues, particularly with representation of rainfall extremes (King et al. 2013) as well as rainfall in areas of lower station density in inland Western Australia, it is expected to perform well over southeast Australia where rainfall station density is high.

Observational studies of the relationship between zonal winds and southeast Australian rainfall have focussed on the Gayndah-Deniliquin Index (GDI), a station-based index of pressure differences between Gayndah (25.6°S, 151.6°E) and Deniliquin (35.6°S, 145.0°E) which approxi-mates the historical zonal geostrophic wind anomalies in this region (Pepler et al. 2014b; Rakich et al. 2008). As this was a proxy to accommodate a lack of wind data and station-based indices are more susceptible to small varia-tions in model climatology, for the purposes of this paper, we instead use the 850 hPa zonal winds along the 150°E transect between 25° and 35°S (marked in Fig. 1), which is easily obtained from the gridded wind fields.

Historical zonal wind speeds are obtained from the NCEP1 reanalysis, for consistency with its use for the NARCLIM regional downscaling project. NCEP1 is known to have some issues, particularly in the representation of extreme events, with anomalously large trends in extreme wind events (Smits et al. 2005) and clear inhomogeneities in extreme temperatures (Donat et al. 2014) when com-pared to observations and other reanalyses. However, dur-ing the overlap period 1979–2012 the correlation between zonal wind anomalies in the NCEP1 and ERA-Interim datasets (Dee et al. 2011) is greater than 0.99 in all seasons and differences in mean wind speeds are less than 1 m s−1, so the two datasets can be considered essentially identical. Additionally, during the full 1950–2005 period the corre-lation between the GDI and NCEP1-derived zonal wind is −0.84 for the cool season (May–October), giving con-fidence in derived relationships between rainfall and zonal wind derived from the NCEP1 reanalysis.

Table 2 The physics parameters used for the three WRF regionally downscaled models

The WRF Double Moment 5-class (WDM5) microphysics scheme is used in all cases, with cumulus physics using either the Kain-Fritsch (KF) or Betts-Miller-Janjic (BMJ) scheme. Planetary bound-ary layer (PBL) and surface physics use either the Yonsei University PBL scheme with the MM5 similarity theory surface layer scheme (YSU/MM5) or the Mellor-Yamada-Janjic PBL scheme with the Eta similarity theory surface layer (MYJ/Eta). The two radiation phys-ics schemes use either the NCAR Community Atmosphere Model scheme (CAM) for both longwave and shortwave radiation, or a com-bination of the Rapid Radiative Transfer Model for longwave radia-tion with the Dudhia shortwave scheme (RRTM/Dudhia). For more information on the specifics of the model parameters see Skamarock et al. (2008)

RCM name

Cumulus physics

PBL/surface physics

Radiation physics

R1 KF MYJ/Eta Dudhia/RRTM

R2 BMJ MYJ/Eta Dudhia/RRTM

R3 KF YSU/MM5 CAM/CAM

110 120 130 140 150 160

−45

−35

−25

−15

110 120 130 140 150 160 −3

−2

−1.5

−1

−0.5

0

0.5

1

1.5

2

3

Fig. 2 Difference between the mean annual rainfall across the CMIP5 ensemble and the 2.5° regridded AWAP averages in May–October (left) and November–April (right), 1950–2005. The average

difference is normalised by the standard deviation of seasonal rainfall in the (regridded) AWAP data

127Zonal winds and southeast Australian rainfall in global and regional climate models

1 3

3 Rainfall and zonal wind in the CMIP5 GCMs

Over the 56 years between 1950 and 2005, the CMIP5 ensemble mean overestimates rainfall across much of the mainland when compared to the observed AWAP data, par-ticularly during the warm season, while slightly underesti-mating cool-season rainfall in the southernmost regions of the mainland (Fig. 2). In southeast Australia, where rainfall is typically winter-dominated, this results in a weakening of the observed seasonal cycle, with too much rainfall dur-ing the warm season and not enough during the cool sea-son, although errors are lower in southeast Australia than

many other parts of the country. While the average zonal wind varies between the CMIP5 models, all models capture the seasonal pattern of enhanced westerly flow during win-ter and spring. Most models also identify the shift to east-erly prevailing flow during late summer and early autumn (Fig. 3), with negative zonal winds (easterly prevailing flow) during the warm season, peaking in March, and west-erly flow (positive values) during the cool season.

During the warm season (November–April), east-erly (negative) zonal flow anomalies are associated with enhanced rainfall across most of eastern Australia. This is well captured in the CMIP5 ensemble, although the

Fig. 3 Monthly mean 850 hPa zonal wind over the region 150°E, 25–35°S in the NCEP1 reanalysis (blue line) and 37 CMIP5 GCMs during 1950–2005, with the ensemble average shown (red line)

Fig. 4 (Top) Correlations between the AWAP rainfall and the NCEP1 zonal wind anomaly in November–April (left) and May–October (right) during 1950–2005. The grey (south coast) and black (east coast) boxes in b indicate the regions of statistically significant correlations used for calculating averages, with crosses indicating the corre-sponding grid points for the 2.5° resolution models. (Bottom) The corresponding median of the correlations between zonal wind and rainfall across each member of the CMIP5 ensemble, with black dots indicating where correlations are statistically significant at the 95 % level

−45

−35

−25

−15

(a) (b)

110 120 130 140 150 160−45

−35

−25

−15

(c) (d)

110 120 130 140 150 160

−0.7

−0.5

−0.3

−0.1

0.1

0.3

0.5

0.7

−0.7

−0.5

−0.3

−0.1

0.1

0.3

0.5

0.7

128 A. S. Pepler et al.

1 3

absolute magnitudes are lower along the east coast than observed (Fig. 4, left panels). During the cool season the subtropical ridge and mid-latitude westerlies move north-ward, with strong positive correlations observed between cool-season rainfall and zonal winds in southwestern parts of southeast Australia (Fig. 4b). The movement of the sub-tropical ridge is a major driver of the winter-dominated rainfall regime in this region, with substantially lower rain-fall during the summer months. In contrast, onshore east-erly flow remains the major source of rainfall on the east coast, resulting in large spatial variations in cool season correlations across southeast Australia.

While the CMIP5 models are generally able to represent the continued negative relationship between zonal wind and rainfall on the east coast during this season, across most of mainland and southeast Australia the ensemble mean cor-relation for CMIP5 fails to reach statistical significance (Fig. 4d), with weakly positive correlations in most mod-els. In particular, there is an inability to represent the strong positive correlations observed in southeast Australia. As westerly-driven systems are responsible for more than 80 % of rainfall across southwestern southeast Australia during this season (Risbey et al. 2009a; Pepler et al. 2014b), with changes in the position of the subtropical ridge and associ-ated westerly winds expected to result in cool season rain-fall declines, this may cause significant problems with accu-rately representing future climate change in this region.

The ability to represent the cool season correlation patterns across the suite of climate models can be tested

through comparing the mean correlations across two regions where correlations in AWAP are statistically signif-icant—these regions are indicated by the outlined regions in Fig. 4b. Correlations for these two regions across the CMIP5 ensemble are shown in Fig. 5, with values for each model in Table 1. More than two-thirds of the ensemble members show statistically significant positive correlations between zonal wind and rainfall along the east coast, with a mean correlation of −0.32, similar to that in the observa-tions (−0.38).

In comparison, there is a large variation in correlations over the south coast region, with only eight models pro-ducing statistically significant positive correlations in this region (Table 1). This group is dominated by three ver-sions of the IPSL model and two versions of MRI, with all but one model (NorESM1-ME) also exhibiting skill on the east coast. It is important to note that statistically signifi-cant correlations when averaged across the region of inter-est do not imply statistically significant correlations across the entire region—even in the subset of seven “best” mod-els (Fig. 6), ensemble mean correlations fail to reach sta-tistical significance in the northern part of the south coast region.

It is important to try to discern the features of individual models that aid their skill in representing the rainfall-zonal wind relationship. Surprisingly, skill had little relationship with the horizontal resolution of the model, with GISS-E2-R (2 × 2.5°) capturing the observed correlation patterns in both regions, while CESM1-FASTCHEM (0.9 × 1.25°) has correlations close to zero in both regions. There is also no relationship between correlation strength in either region and the mean rainfall or zonal wind. Notably, while the IPSL models capture the observed relationships between zonal wind anomalies and rain, the mean zonal wind is twice as strong as that in NCEP. Average rainfall on the east coast is less than half that in either the observations or the mul-timodel mean, with rainfall on the south coast also slightly

Fig. 5 Average correlations between the zonal windflow and rainfall in the east coast (black) and south coast (grey) regions indicated in Fig. 4b during 1950–2005. The 37 CMIP5 models are indicated in grey, with solid circles representing statistically significant negative correlations in the coastal region, and diamonds indicating those that also have statistically significant positive correlations in the inland region. A black circle indicates the CMIP5 ensemble mean, and a black cross the observed relationship from the AWAP rainfall and NCEP1 zonal winds. Statistical significance is tested at the 5 % level

110 120 130 140 150 160

−45

−35

−25

−15

−0.7

−0.5

−0.3

−0.1

0.1

0.3

0.5

0.7

Fig. 6 As in Fig. 4d, but for the seven CMIP5 ensemble members that capture the observed correlation patterns in both regions indi-cated in Fig. 4b

129Zonal winds and southeast Australian rainfall in global and regional climate models

1 3

below average; the IPSL-CM5A models and CSIRO Mk3.6 are the only CMIP5 GCMs to record lower cool-season aver-age rainfall on the east coast than the south. The models with good skill in representing the zonal wind-rainfall relationship in the cool season also bear little similarity to those that cap-ture warm season specific humidity and ENSO-rainfall rela-tionships in eastern Australia (King et al. 2014).

As the zonal wind anomaly is strongly correlated to both the intensity and position of the subtropical ridge, it is unsurprising that models which capture the observed zonal wind-rainfall relationship also capture the observed rela-tionships between subtropical ridge intensity and rainfall (Grose et al. 2014). The converse does not apply, perhaps related to representation of the subtropical ridge position. Importantly, correlations have no relationship with the abil-ity of models to represent the observed climatology of the subtropical ridge, with IPSL-CM5B-LR among the worst.

4 Rainfall and zonal wind in a regionally downscaled model

Before examining the influence of downscaling on global climate models, it is useful to compare the

regionally-downscaled NCEP1 data. During the 1950–2005 period, all three realisations of the WRF model reproduce the observed seasonality of zonal wind in the NCEP1 rea-nalysis. While values are more easterly in the downscaled data, at 0–0.5 m/s (1.25 m/s in NCEP1), correlations between the NCEP1 and downscaled zonal winds across all months were 0.92–0.93. As observed for the CMIP5 model ensemble in Sect. 3, all three WRF realisations underesti-mate the spatial extent of positive correlations in southern southeast Australia, while capturing the relationship on the east coast (Fig. 7). However, the higher-resolution model is able to better identify the spatial patterns of the corre-lations, particularly the area of enhanced positive correla-tions in the southeast, corresponding to the western flank of the Great Dividing Range. The high-resolution model also restricts the large negative correlations to the coastal strip east of the Great Dividing Range, instead of extending them into inland eastern Australia as in the CMIP5 models. The skill of the RCMs used in simulating observed climate when driven by reanalyses is encouraging for their use for downscaling of GCMs.

As the four CMIP3 models were chosen to span the large range of plausible future climate projections, and the equiv-alent CMIP5 models (italicised in Table 1) span the full range of outcomes, it is unsurprising that the median corre-lation patterns across the four CMIP3 models (Fig. 8a) are similar to those observed for the CMIP5 models (Fig. 4d). However, in the CMIP3 subset statistically significant negative correlations on the east coast extend substantially further into inland Australia. This potentially reflects lower model and topographic resolution in the older suite of mod-els, causing the Great Dividing Range to be poorly repre-sented and thus have reduced impact on rainfall patterns.

The twelve 50 km WRF runs driven by the CMIP3 GCMs have overall correlation patterns more similar to those in the driving CMIP3 models than those in the NCEP-WRF downscaled runs (Fig. 7) or the observa-tions (Fig. 4d), particularly with respect to the nega-tive correlations across inland Australia. However, they offer finer spatial detail than the GCMs, with a particular

110 120 130 140 150 160

−45

−35

−25

−15

−0.7

−0.5

−0.3

−0.1

0.1

0.3

0.5

0.7

Fig. 7 As in Fig. 4d, but the ensemble median of the three realisa-tions of the 50 km-downscaled NCEP data, 1950–2005

Fig. 8 As in Fig. 4d, but for the four CMIP3 models during 1950–2005 (left) and the twelve regionally-downscaled realisa-tions (1990–2009), right

110 120 130 140 150 160−45

−35

−25

−15

110 120 130 140 150 160−0.7

−0.5

−0.3

−0.1

0.1

0.3

0.5

0.7

130 A. S. Pepler et al.

1 3

improvement in simulating the spatial pattern of correla-tions over the Great Dividing Range when compared to the CMIP3 models. This includes a notable strengthen-ing of positive correlations along the southwest flank of the range, as well as restricting the area of statistically significant negative correlations in 25–35°S to the coast east of the ridgeline (north of this, the range decreases in height and turns inland, decreasing its influence on coastal rainfall patterns). The higher resolution (10 km) WRF runs further enhance the spatial detail (Fig. 9), but do not change the broad patterns of correlations. Similar to the parent models, the negative correlations between zonal winds and rainfall along the east coast extend well into central Australia, a sharp contrast to results for the NCEP-driven runs and the AWAP observations. The short time period of the downscaled models (20 years) means these correlations are not statistically significant, and while the long-term correlations in the AWAP database are generally weak east of the divide (Fig. 4d), over shorter time-periods the relationship between zonal winds and rainfall in this area may vary.

We then compare the correlations in the 12 CMIP3-WRF realisations with the driving CMIP3 models over the east coast and south coast regions. The medians of the CMIP3 and WRF realisations are similar, with the various WRF runs adding additional spread but no increased infor-mation. While the MIROC3.2 GCM failed to reproduce the negative correlations on the east coast, consistent with results for all MIROC versions in CMIP5, correlations in this region for all three downscaled versions of MIROC3.2 are indistinguishable from the other GCMs, resulting in a clear improvement in the ensemble mean correlation (Fig. 10). This suggests that using a regionally downscaled model such as WRF offers some ability to improve on rep-resentations of rainfall patterns associated with circulation anomalies, even where the source model simulated this rainfall relationship poorly. It is also worth noting that all but two CMIP3-WRF runs produce positive correlations on the south coast at a 50 km resolution, and all models at a 10 km resolution, an improvement over the CMIP5 ensemble.

There is some indication that the parameters chosen for the WRF model may influence its ability to represent these changes. Of the three WRF versions, the “R2” version pro-duced stronger correlations on the south coast for all four CMIP3 models, with an ensemble mean correlation of 0.25 (0.16 for R1 and R3). The R2 ensemble mean has statisti-cally significant positive correlations extending further into southeast Australia and a clear reduction in the proportion of inland southeast Australia with negative correlations, which is closer to the spatial patterns in both the observa-tions and the downscaled NCEP runs. The R2 parameters are particularly useful for downscaling the MIROC3.2 model, producing improved correlation patterns on both the north and south coast, while the average south coast cor-relation is not statistically significant for the R1 or R3 ver-sion, despite statistically significant correlations in the raw model data. This is consistent with previous studies, which found this parameter combination the most effective at rep-licating both average southeast Australian rainfall patterns

Fig. 9 As in Fig. 4d, but for the median of 12 GCM-driven WRF simulations at 50 (left) and 10 km (right) resolution. The region shown closely approximates the 10 km WRF domain

135 140 145 150 155

−35

−30

−25

135 140 145 150 155−0.7

−0.5

−0.3

−0.1

0.1

0.3

0.5

0.7

Fig. 10 As in Fig. 5, but for the four CMIP3 models (1950–2005) and twelve GCM-driven WRF simulations at both 50 and 10 km resolution. The ensemble means are shown by large open symbols, with the observed relationship from the AWAP rainfall and NCEP1 zonal winds indicated by a black cross. The CMIP5 ensemble mean is shown with a black circle for comparison

131Zonal winds and southeast Australian rainfall in global and regional climate models

1 3

and major rain events such as East Coast Lows (Evans et al. 2011; Ji et al. 2014). It is worth noting that the major dif-ference between the R2 run and the other two model reali-sations lies in the choice of cumulus paramaterisation, suggesting that this may play an important role in the simu-lation of rainfall variations in the mid-latitude storm tracks.

The inability of both GCMs and RCMs to capture the full spatial extent of positive correlations between zonal wind and rainfall in southern parts of southeast Australia suggests that this lack is not simply a consequence of poor topographic resolution, nor is it related to the ability of models to simulate either the average zonal wind or its seasonal cycle. This is an area in need of detailed further examination, but may reflect a southern bias in the fre-quency of synoptic systems that develop in the westerlies, such as cold fronts and cut-off lows, although Catto et al. (2014) observed that the annual frequency of fronts across 18 CMIP5 models in southeast Australia was very similar to that in reanalyses. Alternately, this may be associated with a bias in the rainfall pattern associated with these systems, with too few fronts producing rainfall in inland regions of southeast Australia, or too little rainfall occur-ring associated with each front.

5 Conclusions

Southeastern Australia features a number of distinct cli-mate regions, with cool-season rainfall patterns along the east coast different to those further inland and in the south-west. Despite a long-term drying trend in cool season rain-fall associated with a southward shift of the subtropical ridge and the associated mid-latitude westerlies, there is no clear consensus across global climate models as to the sign or magnitude of any future changes.

The spatial patterns of relationships between southeast Australian rainfall and onshore zonal wind are important to both the rainfall climatology and interannual variability across this region (Rakich et al. 2008; Risbey et al. 2009a, b; Pepler et al. 2014a, b). Both the CMIP5 ensemble and a set of regionally-downscaled CMIP3 models are able to capture both the seasonality of zonal wind flow and the strong relationship between easterly zonal flow anoma-lies and enhanced rainfall on the east coast. The improved topography of the WRF model at both 50 and 10 km reso-lutions also allows improved representation of spatial vari-ations along the Great Dividing Range, particularly the area of enhanced rainfall correlation on the southwest slopes. Importantly, the use of a regional climate model also sub-stantially improved the relationship between zonal wind and rainfall in models where this was poorly simulated.

Both the CMIP5 ensemble and the CMIP3-WRF ensem-ble have difficulties in replicating the spatial extent of

positive correlations between rainfall and zonal wind in southern southeast Australia, a region where westerly sys-tems are known to contribute more than 80 % of cool-sea-son rainfall. Only 8 of the 37 CMIP5 ensemble members were able to replicate the statistically significant positive correlations over a broad area of inland southeast Australia, and even the best models fail to capture the northern extent of this region.

Two of the four CMIP3 models, CCCMA and MIROC3.2, also produced statistically significant correla-tions along the south coast, with MIROC3.2 failing to cap-ture the observed negative correlations on the north coast, suggesting very different model behaviour. These posi-tive correlations were also observed in four of the twelve 10 km WRF models—all three downscaled versions of the CCCMA model, as well as one downscaled version of MIROC3.2—with downscaling generally degrading the average correlation on the south coast for MIROC3.2 but vastly improving correlations in the northeast. Best results were observed for the R2 parameter combination, which produced the strongest average correlations in the south coast region for all CMIP3 models, potentially related to its different cumulus scheme. These results indicate that regional climate models can offer significant advantages in improving upon global climate models for represent-ing some aspects of the large-scale circulation, but may not provide any additional value in other aspects, despite improvement in representation of the present-day mean precipitation.

Ability to represent this pattern has no apparent relation-ship to model resolution or the average rainfall, zonal wind, or representation of the subtropical ridge, and is an area in need of substantial further work. One possible cause is the representation of rainfall associated with westerly-driven synoptic systems such as cut-off lows and cold fronts, which may not be well captured by models.

As climate model projections consistently favour a pole-ward expansion of the tropics, and corresponding shifts in the mid-latitude westerlies (Bengtsson et al. 2006; Chang et al. 2012; Quan et al. 2014), this result raises impor-tant concerns about the ability of the current suite of cli-mate models to translate these projections to their impacts. Notably, improvement in mean present-day precipitation patterns using regional downscaling does not guarantee improved representation of the impacts on precipitation of changes in the large-scale circulation. This may have sig-nificant implications for our ability to project future rainfall trends in southeast Australia and other regions, and could explain the inconsistency between projections. Further work will address this question by assessing the relation-ship between the observed zonal wind-rainfall relationships in the CMIP5 model ensemble and projected rainfall trends for southeast Australia over the coming century.

132 A. S. Pepler et al.

1 3

Acknowledgments This research was supported by the Australian Research Council Centre of Excellence for Climate System Science, grant CE110001028, and Linkage project grant LP120200777. Jason Evans was supported by the Australian Research Council Future Fel-lowship FT110100576, with the WRF downscaled runs available thanks to the NSW Office of Environment and Heritage backed NSW/ACT Regional Climate Modelling (NARCliM) Project and the East-ern Seaboard Climate Change Initiative—East Coast Low Project (ESCCI-ECL). This research was undertaken with the assistance of resources provided at the NCI National Facility systems at the Aus-tralian National University through the National Computational Merit Allocation Scheme supported by the Australian Government. The authors also wish to thank two anonymous reviewers for their helpful comments.

References

Bengtsson L, Hodges KI, Roeckner E (2006) Storm tracks and cli-mate change. J Clim 19:3518–3543. doi:10.1175/JCLI3815.1

Cai W, Shi G, Cowan T, Bi D, Ribbe J (2005) The response of the Southern Annular Mode, the East Australian Current, and the southern mid-latitude ocean circulation to global warming. Geo-phys Res Lett 32:L23706. doi:10.1029/2005GL024701

Catto JL, Nicholls N, Jakob C, Shelton KL (2014) Atmospheric fronts in current and future climates. Geophys Res Lett 41:7642–7650. doi:10.1002/2014GL061943

Chang EKM, Guo Y, Xia X (2012) CMIP5 multimodel ensemble pro-jection of storm track change under global warming. J Geophys Res Atmos 117:D23118. doi:10.1029/2012JD018578

Collins M, Knutti R, Arblaster J, Dufresne JL, Fichefet T, Friedling-stein P, Gao X, Gutowski WJ, Johns T, Krinner G, Shongwe M, Tebaldi C, Weaver AJ, Wehner M (2013) Long-term climate change: projections, commitments and irreversibility. In: Climate Change 2013: The physical science basis. Contribution of Work-ing Group I to the Fifth Assessment Report of the Intergovern-mental Panel on Climate Change

Dee DP, Uppala SM, Simmons AJ et al (2011) The ERA-Interim reanalysis: configuration and performance of the data assimi-lation system. Q J R Meteorol Soc 137:553–597. doi:10.1002/qj.828

Donat MG, Sillmann J, Wild S, Alexander LV, Lippmann T, Zwi-ers FW (2014) Consistency of temperature and precipitation extremes across various global gridded in situ and reanalysis datasets. J Clim 27:5019–5035. doi:10.1175/JCLI-D-13-00405.1

Drosdowsky W (1993) An analysis of Australian seasonal rainfall anomalies: 1950–1987. I: spatial patterns. Int J Climatol 13:1–30. doi:10.1002/joc.3370130102

Evans JP, McCabe MF (2013) Effect of model resolution on a regional climate model simulation over southeast Australia. Clim Res 56:131–145. doi:10.3354/cr01151

Evans JP, Ekström M, Ji F (2011) Evaluating the performance of a WRF physics ensemble over South-East Australia. Clim Dyn 39:1241–1258. doi:10.1007/s00382-011-1244-5

Evans JP, Ji F, Lee C, Smith P, Argueeso D, Fita L (2014) Design of a regional climate modelling projection ensemble experi-ment–NARCliM. Geosci Model Dev 7:621–629. doi:10.5194/gmd-7-621-2014

Grose M, Timbal B, Wilson L, Bathols J, Kent D (2014) The subtropi-cal ridge in CMIP5, and implications for projections of rainfall in southeast Australia. Submitted to Int J Climatol

Hendon HH, Lim EP, Arblaster JM, Anderson DLT (2014a) Causes and predictability of the record wet east Australian spring 2010. Clim Dyn 42:1155–1174. doi:10.1007/s00382-013-1700-5

Hendon HH, Lim EP, Nguyen H (2014b) Seasonal variations of sub-tropical precipitation associated with the southern annular mode. J Clim 27:3446–3460. doi:10.1175/JCLI-D-13-00550.1

Imielska A (2011) Seasonal climate summary southern hemisphere (summer 2010–2011): second-wettest Australian summer on record and one of the strongest La Nina events on record. Aust Meteorol Oceanogr J 61:241–251

Ji F, Ekström M, Evans JP, Teng J (2014) Evaluating rainfall patterns using physics scheme ensembles from a regional atmospheric model. Theor Appl Climatol 115(1–2):297–304. doi:10.1007/s00704-013-0904-2

Jones DA, Wang W, Fawcett R (2009) High-quality spatial climate data-sets for Australia. Aust Meteorol Oceanogr J 58:233–248

Kalnay E, Kanamitsu M, Kistler R et al (1996) The NCEP/NCAR 40-year reanalysis project. Bull Am Meteorol Soc 77:437–471. doi:10.1175/1520-0477(1996)077<0437:TNYRP>2.0.CO;2

King AD, Alexander LV, Donat MG (2013) The efficacy of using gridded data to examine extreme rainfall characteristics: a case study for Australia. Int J Climatol 33:2376–2387. doi:10.1002/joc.3588

King AD, Donat MG, Alexander LV, Karoly DJ (2014) The ENSO-Australian rainfall teleconnection in reanalysis and CMIP5. Clim Dyn 44:2623–2635. doi:10.1007/s00382-014-2159-8

McBride JL, Nicholls N (1983) Seasonal relationships between Aus-tralian rainfall and the Southern Oscillation. Mon Weather Rev Boston 111:1998–2004

Meehl GA, Covey C, Taylor KE, Delworth T, Stouffer RJ, Latif M, McAvaney B, Mitchell JFB (2007) THE WCRP CMIP3 multi-model dataset: a new era in climate change research. Bull Am Meteorol Soc 88:1383–1394. doi:10.1175/BAMS-88-9-1383

Meneghini B, Simmonds I, Smith IN (2007) Association between Australian rainfall and the southern annular mode. Int J Climatol 27:109–121. doi:10.1002/joc.1370

Meyers G, McIntosh P, Pigot L, Pook M (2007) The years of El Niño, La Niña, and interactions with the tropical Indian Ocean. J Clim 20:2872–2880. doi:10.1175/JCLI4152.1

Murphy BF, Timbal B (2008) A review of recent climate variabil-ity and climate change in southeastern Australia. Int J Climatol 28:859–879. doi:10.1002/joc.1627

Nicholls N (2009) Local and remote causes of the southern Australian autumn-winter rainfall decline, 1958–2007. Clim Dyn 34:835–845. doi:10.1007/s00382-009-0527-6

Önol B (2012) Effects of coastal topography on climate: high-resolu-tion simulation with a regional climate model. Clim Res 52:159–174. doi:10.3354/cr01077

Pepler A (2013) Seasonal climate summary southern hemisphere (winter 2012): dry conditions return to Australia. Aust Meteorol Ocean J 63:339–349

Pepler A, Coutts-Smith A, Timbal B (2014a) The role of East Coast Lows on rainfall patterns and inter-annual variability across the East Coast of Australia. Int J Climatol 34:1011–1021. doi:10.1002/joc.3741

Pepler A, Timbal B, Rakich C, Coutts-Smith A (2014b) Indian Ocean Dipole overrides ENSO’s influence on cool season rainfall across the Eastern Seaboard of Australia. J Clim 27:3816–3826. doi:10.1175/JCLI-D-13-00554.1

Quan XW, Hoerling MP, Perlwitz J, Diaz HF, Xu T (2014) How fast are the tropics expanding? J Clim 27:1999–2013. doi:10.1175/JCLI-D-13-00287.1

Rakich CS, Holbrook NJ, Timbal B (2008) A pressure gradient metric capturing planetary-scale influences on eastern Australian rain-fall. Geophys Res Lett 35:L08713. doi:10.1029/2007GL032970

Risbey JS, Pook MJ, McIntosh PC, Ummenhofer CC, Meyers G (2009a) Characteristics and variability of synoptic features asso-ciated with cool season rainfall in southeastern Australia. Int J Climatol 29:1595–1613. doi:10.1002/joc.1775

133Zonal winds and southeast Australian rainfall in global and regional climate models

1 3

Risbey JS, Pook MJ, McIntosh PC, Wheeler MC, Hendon HH (2009b) On the remote drivers of rainfall variability in Aus-tralia. Mon Weather Rev 137:3233–3253. doi:10.1175/2009MWR2861.1

SEACI (2010) Climate variability and change in south-eastern Aus-tralia a synthesis of findings from phase 1 of the South Eastern Australian Climate Initiative (SEACI). CSIRO Australia

Skamarock WC et al (2008) A description of the advanced research WRF version 3. NCAR, Boulder

Smits A, Klein Tank AMG, Können GP (2005) Trends in stormi-ness over the Netherlands, 1962–2002. Int J Climatol 25(10):1331–1344

Taylor KE, Stouffer RJ, Meehl GA (2012) An overview of CMIP5 and the experiment design. Bull Am Meteorol Soc 93:485–498. doi:10.1175/BAMS-D-11-00094.1

Timbal B (2010) The climate of the Eastern Seaboard of Aus-tralia: a challenging entity now and for future projec-tions. IOP Conf Ser Earth Environ Sci 11:012013. doi:10.1088/1755-1315/11/1/012013

Timbal B, Drosdowsky W (2013) The relationship between the decline of Southeastern Australian rainfall and the strength-ening of the subtropical ridge. Int J Climatol 33:1021–1034. doi:10.1002/joc.3492

Timbal B, Hendon H (2011) The role of tropical modes of variability in recent rainfall deficits across the Murray-Darling Basin. Water Resour Res 47:W00G09. doi:10.1029/2010WR009834

Timbal B, Li Z, Fernandez E (2008) The Bureau of meteorology sta-tistical downscaling model graphical user interface: user manual and software documentation. CAWCR Technical Report No. 004

Tselioudis G, Douvis C, Zerefos C (2012) Does dynamical downs-caling introduce novel information in climate model simulations of precipitation change over a complex topography region? Int J Climatol 32:1572–1578. doi:10.1002/joc.2360

Ummenhofer CC, England MH, McIntosh PC, Meyers GA, Pook MJ, Risbey JS, Gupta AS, Taschetto AS (2009) What causes south-east Australia’s worst droughts? Geophys Res Lett 36:L04706. doi:10.1029/2008GL036801

Webb MA (2012) Seasonal climate summary southern hemisphere (summer 2011–12): a mature La Nina, strongly positive SAM and active MJO. Aust Meteorol Oceanogr J 62:335–349

Williams AAJ, Stone RC (2009) An assessment of relationships between the Australian subtropical ridge, rainfall variability, and high-latitude circulation patterns. Int J Climatol 29:691–709. doi:10.1002/joc.1732