Behind uncertainties in projections of Australian tropical climate: Analysis of 19 CMIP3 models



[1] A systematic study is undertaken for projected changes in tropical Australian climate in 19 CMIP3 Coupled Models for the A2 scenario over the 21st century. While equatorial regions to the north of Australia are projected to have increased precipitation during austral summer (December to February) by the end of the 21st century, there is no significant change over northern Australia based on the model ensemble mean. There is a large spread in model simulations of precipitation change, with both large positive and negative anomalies. The ensemble mean change in the seasonal cycle of precipitation over tropical Australia is small, with precipitation increase during March and April, suggesting a prolonged Australian wet season. There is no model consensus on how interannual variability of tropical Australian precipitation will change in future climate, although more models simulate increased variability than decreased. Correlations between full wet season (October to April) precipitation and austral spring (September to November) NINO3.4 sea surface temperature anomalies show a slight weakening. The spread in projected precipitation seasonal cycle changes between simulations from the same model is larger than the inter-model range, indicating that there is large internal or natural variability in tropical Australian precipitation relative to the climate change signal. Zonal wind changes indicate an intensification of austral summer low level westerlies although combined with a weakening of upper easterlies. Low level westerlies also persist for longer, consistent with a delay in the monsoon retreat. All models simulate an increase in the land-ocean temperature contrast in austral summer, with a significant correlation between changes in land-ocean temperature contrast in the pre-monsoon (austral spring) and summer precipitation changes. Analysis of precipitation changes using regime-sorting techniques shows offsetting tendencies from thermodynamic changes associated with enhanced atmospheric moisture and dynamic changes associated with a weakened atmospheric circulation.

1. Introduction

[2] Precipitation over tropical Australia and in the region to the immediate north to the equator is dominated by the Australian component of the Asian-Australian monsoon system [e.g., McBride, 1998]. Although populations affected by the monsoon are much smaller than for South or East Asia, regional impacts of any changes in this system are nevertheless likely to be large, particularly on vulnerable indigenous populations and on ecosystems [Intergovernmental Panel on Climate Change (IPCC), 2007b]. Green et al. [2010] have clearly highlighted the exposure risks to climate change of indigenous communities on low-lying island such as the Torres Strait Islands in Northern Australia. This then provides motivation for the present study.

[3] Despite broad scale consensus on climate changes induced by increasing greenhouse gases, large uncertainties remain at regional levels [IPCC, 2007a], particularly for precipitation projection. For most regions around the globe precipitation intensity is projected to increase, but its frequency to decrease [e.g., Sun et al., 2007]. In marked contrast to the Asian region, however, changes to the Australian component of monsoon have been little studied. Uncertainties in possible changes to the mean circulation and monsoon variability remain key challenges for Australian regional climate change projection [IPCC, 2007a, 2007b]. In the tropical Australian region, uncertainties in precipitation projections are particularly large [IPCC, 2007a; CSIRO and Bureau of Meteorology, 2007], with model disagreement as to the projected sign of the change over northern Australia and adjacent oceanic regions [see IPCC, 2007a, chapter 11]. While some of this disagreement disguises general model consensus for relatively small changes [Power et al., 2012], most does not, meaning that projected model changes are relatively large (compared with natural variability) but with large model disagreement over much of this area [Power et al., 2012; Tebaldi et al., 2011].

[4] For the Asian component of the monsoon, most studies have found a slight increase in precipitation, but with a reduction in overall circulation strength [e.g., IPCC, 2007a, chapter 10], consistent with mechanisms proposed by Held and Soden [2006], Vecchi et al. [2006] and Vecchi and Soden [2007] and Lu et al. [2007]. Zhang et al. [2012] found that changes in the Asian monsoon circulation seen in CMIP3 models are mainly related to the model response to changes in either Pacific equatorial sea surface temperatures (SSTs) or zonal winds with these patterns of anomalous SST and wind conditions likely linked to the weakening and westward shift of Walker circulation in the warm pool and maritime continent region. Increases in precipitable water associated with global (atmospheric) warming do not change Asian monsoon precipitation and circulation seasonality much but they can result in increased precipitation intensity once the summer monsoon is established.

[5] For the Western Pacific monsoon region (between 120°E and 180°E), Smith et al. [2012] found that the predominant signal of climate change was an increase in precipitation during the summer season with a weakening or negligible change to the strength of the low level monsoon winds. The authors concluded that these results indicated that the Western Pacific monsoon climate will respond more to global scale drivers such as an increase in atmospheric water vapor content and a weakening of the global atmospheric overturning circulation, rather than to more regional changes such as an increase in the land/ocean temperature contrast.

[6] In an earlier study, projected changes specifically to the Australian monsoon were examined by Suppiah [1995] for a single model. Under a 2 × CO2 scenario, the simulated Australian monsoon circulation was strengthened and precipitation was increased by about 20%, but the interannual variability of precipitation remained unchanged in north Australia. Also, the study found that on average the low level monsoon shear line moved south under 2 × CO2 conditions, although such a movement was not statistically significant and also showed little variation on interannual time scales. The monsoon shear line did not show any changes under 2 × CO2 conditions in its location over oceanic regions off the northwest and northeast coasts of Australia. Moise et al. [2005], using the earlier generation of CMIP2 [Meehl et al., 2000] models found ensemble mean decreases in winter precipitation across southern Australia and over northwestern Australia during summer (December–February). Increased precipitation was simulated over parts of eastern Australia during winter, extending further north during summer. Projected changes to Australian precipitation generally were examined in the CMIP3 models by Suppiah et al. [2007] for three different emission scenarios, but did not show a conclusive result across tropical Australia. None of these studies however performed a systematic study of projected changes of broad-scale climate features in the tropical Australian region, nor did they elucidate processes responsible for precipitation changes in particular. This study will consider and analyze projected changes to large-scale climate features in the region, including the Australian component of the Asian-Australian monsoon.

[7] An earlier study [Colman et al., 2011] examined how well tropical Australian and particularly monsoon features for the current climate are simulated by state-of-the-art Coupled Global Climate Models (CGCMs), as represented by the models that took part in the World Climate Research Program (WCRP) Coupled Model Intercomparison Project phase 3 (CMIP3) [Meehl et al., 2007]. In this follow-up study we consider projected tropical Australian climate, with a focus on Australian monsoon changes for the coming century.

[8] This paper is organized as follows. Section 2 presents the models and validation data sets used, and describes the methodology used for regime sorting of precipitation, the models as well as validation data sets. Section 3 discusses simulated changes to tropical Australian climate with an emphasis on the Australian monsoon for both mean climate and inter-annual variability along with associated correlations and “regime based” analysis of precipitation changes. The final section contains discussion and conclusions.

2. Methodology

2.1. CMIP3 Models

[9] Models considered (listed in Table 1) were those submitted to the World Climate Research Program (WCRP) Coupled Model Intercomparison Project phase 3 (CMIP3) [Meehl et al., 2007]. Monthly results were extracted from the 20th century coupled runs (20C3M) for the period 1980–1999 and from the SRES-A2 scenario runs for the period 2080–2099 for analysis of mean fields. This allows for comparison with previous studies that also analyzed simulations over similar length [e.g., Colman et al., 2011]. The 20C3M simulations included forcings from anthropogenic greenhouse gases and sulphates, but differ in detail between models. In particular, forcings from volcanic aerosols, ozone changes and solar variability are not included in all model simulations (see AchutaRao et al. [2007] for a summary of model forcing). The SRES-A2 simulations for future climate change provide a set of future climates associated with higher levels of CO2 concentration by the end of the 21st century. Sensitivity of changes to different scenarios are beyond the scope of the current paper, although CSIRO and Bureau of Meteorology [2007] suggests that overall patterns of temperature and precipitation changes in this region are not sensitive to scenario choice. Monthly fields of precipitation, surface air temperature, vertical pressure velocity at 500 hPa (ω500), mean sea level pressure (MSLP), and zonal winds were analyzed. Not all 24 CMIP3 models provided these fields for the SRES-A2 scenario simulations resulting in a smaller ensemble of 19 models. Therefore it should be noted that the multimodel mean presented in Colman et al. [2011] differs slightly from that of the present paper.

Table 1. Modeling Groups That Provided SRES-A2 Simulation Data to the CMIP3 Archive and Their Acronyms as Used in This Paper
Group NumberModeling CenterModel Acronym
1Bjerknes Centre for Climate Researchbccr_bcm2_0
2Canadian Centre for Climate Modeling & Analysiscccma_cgcm3_1
3Météo-France/Centre National de Recherches Météorologiquescnrm_cm3
4CSIRO Atmospheric Researchcsiro_mk3_0
5CSIRO Atmospheric Researchcsiro_mk3_5
6U.S. Dept. of Commerce/NOAA/Geophysical Fluid Dynamics Laboratorygfdl_cm2_0
7U.S. Dept. of Commerce/NOAA/Geophysical Fluid Dynamics Laboratorygfdl_cm2_1
8NASA/Goddard Institute for Space Studiesgiss_model_e_r
9Instituto Nazionale di Geofisica e Vulcanologiaingv_echam4
10Institute for Numerical Mathematicsinm_cm3_0
11Institut Pierre Simon Laplaceipsl_cm4
13Meteorological Institute of the University of Bonn, Meteorological Research Institute of KMA, and Model and Data group.miub_echo_g
14Max Planck Institute for Meteorologympi_echam5
15Meteorological Research Institutemri_cgcm2_3_2a
16National Center for Atmospheric Researchncar_ccsm3_0
17National Center for Atmospheric Researchncar_pcm1
18Hadley Centre for Climate Prediction and Research/Met Officeukmo_hadcm3
19Hadley Centre for Climate Prediction and Research/Met Officeukmo_hadgem1

[10] For model inter-comparisons, a single realization (“run1”) was selected for each model where multiple realizations were available. The issue of within-model variability versus between-model variability for individual models was also investigated and is addressed below. We have excluded one model (bcc_cm1) entirely, due to errors in specified fields as detailed in the accompanying CMIP3 documentation (

[11] Model resolution varied considerably (from ∼1° × 1° to 5° × 4° for the atmosphere) so, for consistency of analysis, all model data were first interpolated to a common 2.5° × 2.5° grid before averaging. Where appropriate to the analysis, the model specific land-sea mask was applied before the interpolation to the common grid.

2.2. Validation Data Sets

[12] While this study mainly analyses data from CGCMs, where appropriate we have also displayed selected observations and reanalysis results for comparison with model results. Winds and MSLP were taken from monthly means from the European Centre for Medium Range Weather Forecasting (ECMWF) 40 year reanalysis (ERA40) [Uppala et al., 2005]. Sea surface temperature derived indices were obtained from the U.S. National Weather Service Climate Prediction Center (CPC) ( All observational data sets were first interpolated to the common 2.5° × 2.5° grid, before comparison with model results.

2.3. Regime Sorting of Precipitation

[13] Bony et al. [2004] showed that a useful means for quantitatively relating precipitation changes to monsoon dynamics is by the use of 500 hPa vertical motion (ω500) “regimes.” This approach seeks to describe the relative contributions to precipitation changes due to by both large-scale dynamics and the “thermodynamics” within an increasingly moist atmosphere under global warming. Variants of this approach have provided useful insights into the processes associated with both convective precipitation biases and projected changes in the tropical Pacific [e.g., Bellucci et al., 2010; Allan, 2012] and also changes in precipitation extremes [Emori and Brown, 2005].

[14] In the regime-sorting approach, regimes of monthly mean ω500 are created by identifying, at each grid point, vertical motion belonging to certain bins of ω500 (here we have used a bin width of 10 hPa/d). Convectively active regions are represented by large negative values of ω500 while strong subsidence shows positive values [see Bony et al., 1997]. Within the region of interest, the relative occurrence of convective regimes is represented by the probability density function (PDF) of ω500 [see, e.g., Colman et al., 2011, Figure 8a]. In order to calculate how much precipitation occurs within each vertical motion regime, composite precipitation amounts are added within each bin, thereby stratifying precipitation along convective strength. The total precipitation that occurs as a function of the vertical motion regime (and incorporating the distribution of these regimes) is then the product of these two terms [e.g., Colman et al., 2011, Figure 8c].

[15] Following Colman et al. [2011], this analysis was undertaken separately over land points over northern Australia (120°E–150°E, 10°S–20°S), as well as in the region incorporating both land and ocean extending to the equator (120°E–150°E, 0°S–20°S). This ensured that results for Australian monsoonal precipitation would be put in context with the broader monsoonal circulation. It also allows comparisons of changes over land only and over the larger domain.

[16] It has been shown [Bony et al., 2004; Emori and Brown, 2005] that it is possible to decompose changes in tropical precipitation (ΔPR) associated with future climate change into three components:

display math

where PRω is the regime-sorted precipitation in the current climate; Pω the probability of vertical pressure velocity and Δ indicates the difference between future (2080–2099) and current (1980–1999) climate periods. Note that the equation is discretized as a sum over a finite number of bins. The first component is linked to precipitation changes associated with changes in the PDF of vertical pressure velocity, i.e., changes to the atmospheric circulation. Therefore this component is considered the dynamic component of precipitation changes. The second component arises from changes in regime-sorted precipitation under given dynamical conditions. These changes arise from changes in the thermodynamical structure of the atmosphere, linked to the increase in moisture content due to (generally) higher atmospheric temperatures. Therefore this component may be termed the thermodynamic component of precipitation changes. The last term originates from the co-variation of dynamical and thermodynamical precipitation changes. This separation allows assessment of the nature of precipitation changes within the CMIP3 models for the Australian monsoon, linking regional changes in precipitation to changes in the larger circulation as well as changes in the thermodynamic structure of the atmosphere.

[17] While Bony et al. [2004] binned over an area (i.e., the tropics), Emori and Brown [2005] analyzed model output at each grid point to provide spatial patterns of the decomposition. Here we used both methods, first looking at binning summed over the Australian monsoon region and then comparing the results to the spatial pattern obtained using individual grid point binning.

[18] Other studies have also investigated the changes in dynamical versus thermodynamical components of precipitation under enhanced greenhouse conditions using different methods. Chou et al. [2009] argued that with increased warming increased atmospheric moisture content is expected to result in wet regions getting wetter [see also Chou and Neelin, 2004; Held and Soden, 2006; Chou and Chen, 2010], leading to enhanced precipitation in tropical monsoon regions. This was further investigated by Seager et al. [2010] who found that existing patterns of precipitation minus evaporation (P-E) are enhanced under future climate change conditions.

3. Model Simulated Climate Change

[19] Throughout this study results are presented for single models as well as the ensemble mean model. Many variants exist on methods for deriving projections from a multimodel ensemble [e.g., Tebaldi and Knutti, 2007]. A commonly used, straightforward and pragmatic approach consists of simply averaging models from the full ensemble, however more sophisticated methods are appropriate in some circumstances. Weigel et al. [2010] introduced a conceptual model for projections and examined the effect of weighting schemes based on assessed model “skill.” They concluded that although in principle projections could be improved by optimal weighting, in practice it was problematic to derive such weights, and there was risk of degradation of projections by incorrect weighting. They recommended equal weighting, if necessary combined with elimination of models found not to represent key physical processes, an approach recommended by a number of other studies [e.g., Stainforth et al., 2007; Räisänen, 2007; Knutti, 2010; Knutti et al., 2010a, 2010b; Weigel et al., 2010; Perkins et al., 2012]. This methodology will be followed here. Equal weighting for the full ensemble will be examined for most fields, but the impact of elimination of “weaker” models will also be considered in particular cases.

3.1. Change in Mean Climate Over Tropical Australia

[20] Figure 1 shows the December to February (DJF) changes (from 1980 to 1999 to 2080–2099) in temperature and precipitation for the ensemble mean of the models as well as a corresponding measure of variability for this change. Strong warming of greater than 4°C is seen over the interior of the Australian continent with values of around 3°C across the tropical North. Weaker warming is found over the surrounding ocean and also over New Guinea (although note that the topography of New Guinea is poorly represented in most CMIP3 models). Warming over the Australian continent is greatest over the tropics (i.e., north of 20°S) in winter (not shown), whereas in summer it is centered further south over the middle of the continent. The large-scale impact of the warming pattern is to increase land-sea contrast by around 1°C in DJF. Implications of this enhanced land–sea contrast for on the monsoon are discussed below.

Figure 1.

Ensemble mean changes in (a) surface air temperature (°C) and (c) precipitation (%) for December–February seasonal means. Compared are averages from the 2080–2099 period from the SRES-A2 scenario versus 1980–1999 period from the 20c3m runs. Also shown are measures for inter-model variation: standard deviation of changes in (b) surface air temperature (°C) and (d) precipitation (in % of 2080–99 average). The stippling in Figure 1d occurs where more than 66% of the models agree on the direction of the change.

[21] While equatorial regions to the north of Australia are, on average, projected to have increased precipitation during summer (DJF) by the end of the 21st century [e.g., IPCC, 2007a; Australian Bureau of Meteorology and CSIRO, 2011], tropical Australia shows no statistically significant increases in DJF precipitation. Fractional precipitation changes over this region are greatest in winter (an increase of around 10%, not shown) although climatological totals are small in this season, so the net change is small. The inter-model range of DJF precipitation projections over tropical Australia is far greater than for surface air temperature. The standard deviation of temperature change (Figure 1b) is smaller than the mean change. For precipitation projections, the inter-model range is mostly larger than the mean change (in %) with some regions showing good model agreement on the sign of the change (mostly for precipitation increases, see stippling in Figure 1d). Large inter-model variability is also seen in regions where models do not agree on the sign of the change, such as the decreases along the Western Australian coast line (no stippling in Figure 1d).

3.2. Seasonal Cycle Changes

[22] The seasonal cycle of precipitation, wind and temperature are defining features of monsoonal climate, and projected changes to this cycle are examined here.

[23] Figure 2 shows the seasonal migration of monthly mean precipitation for the end of the 21st century (Figure 2, top) for the model ensemble, averaged across the Australian region (100° to 150°E). The peak in Australian continental precipitation over the summer monsoon is clearly apparent, although it is notable that continental latitudes fall on the periphery of the main precipitation centers, which are located to the north. Agreement with observations in pattern and intensity is good (not shown), although this skill in the ensemble mean disguises a broad range of skill in individual models [Colman et al., 2011]. The change (in %) in seasonal migration is shown in Figure 2 (bottom). Clearly visible is a signal of increased precipitation (+10% to +15%) in the deep tropics near the equator, and decreased precipitation outside this belt, in the subtropics. This pattern in precipitation change is one of the robust results that has emerged from the IPCC Fourth Assessment Report [IPCC, 2007a]. Tropical sections of the Australian continent lie close to the boundary of these changes (indicated by green box in Figure 2), consistent with only small changes south of 10°S during the peak of the Australian wet season (DJF). Other large scale convergence zones, such as the Inter-Tropical Convergence Zone and South Pacific Convergence Zone are projected to have increased precipitation due to enhanced moisture convergence [IPCC, 2007a; Australian Bureau of Meteorology and CSIRO, 2011; Brown et al., 2012], consistent with earlier studies by Held and Soden [2006] and Vecchi and Soden [2007].

Figure 2.

(top) Mean model seasonal precipitation averaged over 100°E to 150°E (in mm per month) for period 2080–2099 and (bottom) changes (in %) relative to 1980–1999 period. The green box encompasses tropical Australia during DJF.

[24] Figure 3 shows the projected changes in seasonal cycles of precipitation, surface air temperature and MSLP for each model as well as the ensemble mean, averaged across the Australian region (100° to 150°E). The ensemble mean changes in the seasonal cycles of precipitation (Figure 3a) across tropical Australia are weak although there is an indication of an enhancement in precipitation during March and April. This suggests a delay in the monsoon retreat. Very large inter-model variation is found, with peak changes from individual models in excess of ±2 mm/day. To investigate the impact of model skill on the simulated change in precipitation seasonal cycle, CMIP3 models were grouped into three categories depending upon the total wet season (October to April) precipitation over Australia north of 20°S (with classifications delimited at ±25% of the observed total): “low,” “medium,” “high” (following Colman et al. [2011]) as simulated for the current (1980–1999) climate. Figure 3b shows this stratification applied to projections of future changes to the seasonal cycle of precipitation. Models with below observed precipitation simulate an increase in precipitation across the entire wet season which is not the case for models that have a near-observed or higher precipitation. These two groups show a decrease during some months of the wet season, suggesting a possible shift in the precipitation seasonality. Common to all categories, however is the increase at the end of the wet season suggesting a prolonged Australian monsoon. Some of this increase could also be affected by the small number of models within the categories (4–5) and the disparity in their projections.

Figure 3.

Changes in seasonal cycle of precipitation (top) across tropical Australia (land areas north of 20° South) between periods 2080–2099 and 1980–1999 (left) for all CMIP3 models compared to (right) a subset of models showing greatest skill in reproducing the current seasonal precipitation cycle. For explanation see text. The changes in (bottom left) seasonal cycle for surface air temperature and (bottom right) MSLP are for the same tropical Australian region (all models). AVG represents the mean of the appropriate ensemble.

[25] The changes in seasonal cycle of surface air temperature (Figure 3c) and MSLP (Figure 3d) are shown for the same region. While there is no statistically significant change in the surface air temperature seasonal cycle, the between-model spread is quite large (2.5°C and 5°C) – representing the variability in the strength in the greenhouse gas response across the CMIP3 ensemble. Again, some individual models show large deviations from the mean. In particular the cccma_cgcm3_1 and ukmo_hadgem1 models indicate very large warming: during spring for the former (over 5°C) and similar during the entire dry season for the latter. Model cccma_cgcm3_1 also shows large changes in MSLP, with strong decreases throughout most of the year. Interestingly, the model with the least warming across the seasonal cycle (ncar_pcm1) has very little change in MSLP.

[26] Overall the ensemble mean of the models shows little change in the seasonal cycle of temperature. MSLP on average shows a small increase throughout most of the year, with this being a minimum (indeed a small decrease) in March–April, corresponding with the period of (small) precipitation increases. Results from only the skilful precipitation models do not significantly change the conclusions for temperature or MSLP (not shown).

3.3. Interannual Variability

[27] The model simulated inter-annual variability in precipitation over tropical Australia during the late 20th century was found to be somewhat too small for most models [Colman et al., 2011], and Figure 4 shows there is a weak model consensus on how this variability would change under enhanced greenhouse conditions. There are fewer models simulating enhanced variability than models simulating reduced variability, but the ensemble average change is a decrease of about −0.15 mm/day. While individual changes in variability were found to be significant in several models, the overall mean change was strongly driven by two models with very strong decreases. These two models (gfdl_cm2_0 and gfdl_cm2_1) are those with strongest 20th century inter-annual variability [Colman et al., 2011]. In fact, all models that simulated near-observed or stronger precipitation variability in the 20th century period show reduced variability by the end of the 21st century. Also the group of models that show skilful precipitation amounts in the current climate also generally show a decrease in variability.

Figure 4.

Change in inter-annual standard deviation of precipitation (STD, in mm/day between wet season (October to April, inclusive) from 1980 to 1999 to 2080–2099 under the SRES-A2 scenario for 19 CMIP3 models. Model mean change is shown in dark blue on the right (20). Numbers are identified in Table 1. Colors represent models which are more than one standard deviation below (blue), above (pink) total wet season observed precipitation for the period 1980–1999. Grey bars indicate models which are within one standard deviation of the observations (refer text). The asterisk denotes that these changes are significant (p = 0.05) using a t-test.

[28] Tropical Australian precipitation is strongly correlated with El Niño Southern Oscillation (ENSO) variability [e.g., Power et al., 2006], reflected in the strong observed correlation (r = −0.54 for the period 1950–1999) between entire wet season precipitation and September to November (SON) NINO3.4 sea surface temperature (SST) anomalies [Holland, 1986]. Changes in ENSO variability or the strength of the ENSO-monsoon relationship would be of great importance to the Australian monsoon system and associated precipitation. Earlier results for the 1980–1999 period showed that models with weak/strong NINO3.4 variability tend to show weak/strong correlations [Colman et al., 2011]. Some CMIP3 models failed to simulate ENSO altogether, while most models indicated a stronger than observed correlation (even though their interannual variability in both precipitation and NINO3.4 SSTs were weaker than observed).

[29] Figure 5 shows the change in correlation between SON NINO3.4 SSTs and total wet season (October to April) precipitation across tropical Australia from 1980 to 1999 to 2080–2099 periods for the SRES-A2 scenario. Also shown are the 2080–2099 period correlations in the insert. Out of the 23 models studied in the Colman et al. [2011] paper, three had either no correlation or the wrong sign of the correlation (GISS models). Of these, only giss_model_e_r is included in this study (the other two did not supply SRES-A2 data).

Figure 5.

Change in correlation between SON nino34 SST's and wet season (October to April, inclusive) precipitation across tropical Australia from (1980–1999) to (2080–2099) periods for the SRES-A2 scenario. Also shown are the (2080–2099) period correlations themselves in the insert. Colors represent models which are more than one standard deviation below (blue), above (pink) total wet season observed precipitation for the period 1980–1999. Grey bars indicate models which are within one standard deviation of the observations. Identifying numbers for models are specified in Table 1. The asterisk denotes significant changes and 95% confidence intervals are shown when the edge of the interval does not cross the zero-change mark.

[30] While the overall mean change indicates a slight weakening of the ENSO–precipitation correlation, the individual model results are spread very widely: from a reversal of the correlation (model 2) to a strong intensification (model 9) to no change (models 1, 7, 17) to weakening (model 11, 12). The very strong weakening seen in model 13 represents a loss of the strength of the correlation from the 20th century results altogether, while model 18 halves its much stronger than observed correlation from the 20th century.

[31] The different colors of the bars refer to the grouping of models into groups with different total wet season precipitation skill in their 20th century simulations as described above. There does not seem to be a relationship between such skill in the 20th century and changes to ENSO correlations with tropical Australian precipitation in the 21st century. Only a few changes are statistically significant (shown as an asterisk in Figure 5), and even fewer (3 models) of these appear to have 95% confidence intervals that do not cross the zero-correlation line. Two out of these three models are simulating either a reversal (model 2) or a strong weakening (model 13) of the correlation. Part of the overall result in changes in inter-annual variability of precipitation (Figure 4) could be caused by the large spread in the changes in the amplitude of ENSO correlations to northern Australian precipitation (Figure 5).

3.4. Intramodel Variability

[32] How representative is the 20 year averaging period used here for characterizing regional precipitation changes? Colman et al. [2011] found that inter model spread in Australian tropical precipitation was large compared with the intra (ensemble) model spread for individual models. Figure 6 illustrates this point, showing the seasonal cycle of 1980–1999 averaged precipitation for three CMIP3 models providing multiple ensemble members (top) and the spread within each model ensemble (thin lines, representing one standard deviation either side of the mean). The arrows indicate the mean difference between pairs of the models and the bars indicate the size of the within-model spread. For precipitation across tropical Australia, the differences between models can be much larger than the spread within multiple realizations from one model.

Figure 6.

(top) Seasonal cycle of 1980–1999 averaged precipitation for three CMIP3 models showing the spread within each model (thin lines representing one standard deviation). The number in brackets, n, represents the number in the ensemble. The arrow indicates the mean difference between two of the models during DJF and the bars indicate the size of the within-model spread. (bottom) The corresponding plot for changes in the seasonal cycle and the within model spread. The arrow indicates the mean difference between the models and the bars show the within model spread over the (boxed) period during DJF (thin lines representing one standard deviation of all possible combinations of realizations within each model).

[33] This situation is reversed, however, when considering the corresponding figure for changes in the seasonal cycle under the SRES-A2 scenario. Figure 6 (bottom) shows that the within model spread of anomalies can be larger than the mean difference between two different models (thin lines show one standard deviation either side of the mean considering all combinations of realizations). Choosing between two models would be similar to choosing between two simulations from the same model. This suggests that multiple realizations, or much longer averaging periods may be required for robust regional precipitation projections. The larger number of models ensembles available in the upcoming CMIP5 [Taylor et al., 2011] experiments will provide an opportunity for such comparison.

3.5. Changes in Winds and Monsoon Duration

[34] Results from the ensemble mean model of CMIP3 simulations show no significant changes in Australian tropical precipitation during the summer and only slightly enhanced inter-annual variability, but with a suggestion of enhanced precipitation during the monsoon retreat phase. How do large-scale wind fields change and what do they tell us about the onset and retreat of the monsoon over Australia?

[35] Note that there are many possible definitions of monsoon onset, retreat and duration [e.g., Drosdowsky, 1996; Zhang et al., 2012; Kajikawa et al., 2010]. The purpose of the current paper is not to investigate this aspect in detail, as only monthly mean changes are considered. Nevertheless, even at this coarse temporal resolution some large-scale changes are apparent. Considering the transition to/from a low-level westerly wind regime averaged over tropical Australia as defining monsoon onset/retreat, there are large differences between models in their depiction of both onset month and monsoon duration, and models are consistently delayed in onset compared with the observation [Colman et al., 2011]. We present the ensemble mean zonal wind versus height annual cycles over tropical Australia (Figure 7) for the end of the 20th century (Figure 7a), the 21st century under the SRES-A2 scenario (Figure 7b) and their difference (Figure 7c). The overall changes with regard to the monsoon season over Australia are clearly visible: a moderate intensification of the westerlies at low levels combined with a weakening of the upper level easterlies. Furthermore, while the onset month (defined as the low level wind reversal) seems to stay the same (mid December on average), the low level westerlies are enhanced toward the end of the monsoon period and the retreat month is somewhat extended toward the middle of March by the end of the 21st century. Changes to upper level easterlies, another important feature of the monsoon circulation [e.g., Drosdowsky, 1996], are less clear. For the ensemble mean, easterlies are weakened through a deep layer at the peak of the monsoon, following a pattern of general weakening of the mid to upper tropospheric circulation in all seasons [Vecchi et al., 2006]. Daily analysis would be required to provide further details of onset changes, and is to be the subject of future analysis.

Figure 7.

Zonal wind versus pressure annual cycles over tropical Australia (land areas north of 20° South) from the ensemble mean model for period (a) (1980–1999), (b) (2080–2099) from the SRES-A2 scenario and (c) the difference. Unit, m/s.

[36] This late season delay in westerly wind retreat is also evident in the analysis of the monsoon “shear line” (defined as the change from low level easterlies to westerlies). Note that the importance of changes to this feature is not only in delineating the extent of low level westerlies, but also it is known that a large fraction of tropical cyclones in the Australian region develop along the shear line within several hundred kilometers of land [McBride and Keenan, 1982]. Figure 8 shows the position of the 925hPa shear line for ERA40, simulated 20th century and 21st century for the months January (close to onset over northern Australia [Drosdowsky, 1996]) and March (close to retreat). There is a large spread between the different model simulations (gray lines for the 20th century in Figure 8), however the position of the ensemble mean model’s shear line (light blue) is quite close to the ERA40 re-analysis data for the same time period (black). Under enhanced greenhouse gas conditions (dark blue), the ensemble mean model shear line seem to only move slightly south over Northwest Western Australia during the onset period, but more markedly so during the retreat period. The statistical significance of this change is not assessed, but it is consistent with a somewhat later retreat of the Australian monsoon over the tropical Northwest region of Australia, and the (single model) finding of Suppiah [1992].

Figure 8.

Monsoon shear line for (top) January and (bottom) March as simulated by CMIP3 models for the period 1980–1999 (light blue), 2080–2099 (dark blue) and ERA40 for 1980–1999 (black). Individual CMIP3 models for the 20th century are shown in gray.

[37] Zhang et al. [2012] investigated changes in the Asian monsoon using a precipitable water based definition of monsoon onset/retreat. They found that if the model-simulated monsoon onset/retreat is correlated to the central and eastern Pacific warming and at the same time the model simulates much larger warming due to climate change of the central and eastern Pacific Ocean, then it is very likely that these models will show significant delay of south Asian monsoon onset and shortened duration. Furthermore, if the model simulates more reduction of westerlies in the west of the warm pool region, then these models are likely to predict even larger delay of summer monsoon onset in the Asian monsoon region. Zhang et al. [2012] conclude that simulated future changes to the monsoon circulation depend on the sensitivity of the GCM to either equatorial warming or changes in the zonal winds.

[38] Large-scale 850hPa wind vectors in late 21st century, and late 21st century changes for the model mean are shown in Figure 9, for January and March. Overall changes are relatively modest with weakened easterlies found across northern Australia, most notably in January. More broadly, an overall strengthening of westerlies is found north of Australia, combined with a steady cross equatorial flow west of New Guinea. Equatorial westerly flow over Indonesia is weakened which is consistent with the overall weakening of tropical circulation, already apparent in observations [Vecchi and Soden, 2007], and predicted from models [Vecchi and Soden, 2007] and theory [Held and Soden, 2006]. Projected changes to the Walker and Hadley circulations however are currently unclear [Held and Soden, 2006; Vecchi and Soden, 2007], as is their interaction with greenhouse gases under aerosols in the region under investigation [Bollasina et al., 2011]. Note also that model biases in general and the over extent of the Pacific “cold tongue” in particular are likely to affect changes especially in the eastern region [Cai et al., 2009; Perkins et al., 2012]. Specifically, mean model westerlies in the late 20th-century are too weak compared with observations in a broad region between 120° and 160°E [Colman et al., 2011; Irving et al., 2011].

Figure 9.

(left) Change in 850 hPa winds between SRES-A2 (2080–2099) and 20c3m (1980–1999) ensemble mean model simulation for (top) January and (bottom) March and (right) the respective SRES-A2 winds. Shown are the magnitude (shading) and directions (vectors).

3.6. Changes in Land-Ocean Temperature Contrast

[39] A common feature in this region in models is an increase in the summer land/sea temperature contrast under climate warming (Figure 1), consistent with changes in other regions [e.g., Boer, 2011]. A warmer land is a necessary precursor for monsoon-like circulations [e.g., Kawamura et al., 2002], but its relationship with monsoon strength is complex, and indeed models that have larger summer land-sea contrast in the late 20th-century have less contemporaneous precipitation [Colman et al., 2011], as precipitation and land temperatures are anti-correlated across a range of timescales including inter-annual variability [Power et al., 1998], and under forced climate change [Wardle and Smith, 2004]. Figure 10 shows DJF and SON land-sea temperature contrast changes against fractional DJF precipitation changes over Australia. It can be seen that while there is no significant relationship between DJF changes, there is a statistically significant relationship between land-sea-temperature changes in the pre-monsoon season (SON) and DJF precipitation changes. Models with increased strength in land-sea-contrast tend to show higher (fractional) precipitation changes. With an increase in land-ocean contrast under enhanced greenhouse conditions seen in most models (14 out of 19), this pre-conditioning of the monsoon might play a more important role. Plotting zonal wind changes against land-sea temperature changes reveals no significant relationship (not shown).

Figure 10.

Scatterplot of changes in land-ocean-contrast (in °C) and precipitation changes (in %) for (a) DJF season and (b) the preceding SON season in land-ocean-contrast versus DJF rainfall.

3.7. Changes in “Regime-Sorted” Precipitation

[40] Figure 11 shows the CMIP3 ensemble mean results for the various terms in the decomposition of precipitation changes. Note that because not all models supplied the necessary fields (in particular ω500) the ensemble incorporates only 15 of the 19 models. The regime sorted precipitation regimes for the 1980–1999 and 2080–2099 periods are shown in Figure 11a and the difference in Figure 11b. There is a small decrease in the subsidence-related precipitation but the strongest change is the increase in precipitation associated with deep convective regimes, i.e., large negative values of ω500. The changes shown in Figure 11b are stratified into models with high, medium and low precipitation totals (which also corresponds with regime separation [see Colman et al., 2011, Figure 8]). The agreement over most of the convective (and suppressed) range is striking. Only for the very strongest convection is there a difference. Strongest precipitation changes in these deeper convective regimes are simulated by models that showed above observed convective strength in the 20th century.

Figure 11.

(top left) The ensemble mean DJF precipitation for given ω500 from CMIP3 simulations for 1980–1999 (20c3m) (black line) and 2080–2099 (A2) (purple line). (top right) The corresponding changes, stratified into groups of models which are more than one standard deviation below (blue), above (orange) total wet season observed precipitation for the period 1980–1999. The green line indicates models which are within one standard deviation of the observations. (bottom left) The ensemble mean PDF for ω500 and (bottom right) its changes, with the same colors for 1980–1999, 2080–2099 and the corresponding stratification for the changes.

[41] At the same time, the changes in the distribution of convective regime types are such that the deeper convective regimes decline (i.e., declining in frequency) while the frequency of weaker convective regimes increases (Figures 11c and 11d), consistent with the slow down in the tropical circulation noted elsewhere by [Held and Soden, 2006; Vecchi and Soden, 2007]. The differences in regime frequency (PDF) are again shown stratified into models with high, medium and low precipitation totals (and regime separation). The overall structure of the frequency changes seen is similar between the groups, with some subtle differences. Models that showed near-observed precipitation totals simulate an increase in both medium-strength convective regimes as well as weak subsidence regimes, whereas models with above/below observed precipitation totals show an increase in the number of weak/medium strength convective regimes. All models indicate the decrease in number of strong convective regimes. This analysis suggests that the changes to regimes are robust: warmer climates induce decreased deep convection but increased precipitation for such convection irrespective of model control climate biases.

[42] The area-averaged contributions of the three different components (dynamic, thermodynamic and covariant) to the overall change in precipitation under the SRES-A2 scenario for tropical Australia (land-only) and the wider region (north of 20°S) are shown in Figure 12. When considering the wider region (which includes the maritime continent north of Australia) the situation is relatively coherent across all CMIP3 models: all thermodynamic contributions are positive, suggesting a strong influence of the increase in moisture content in the atmosphere toward increased precipitation. Both the dynamical and covariant components are usually smaller and negative, suggesting a drying associated with a slow down of the larger circulation. This is in line with other studies such as Held and Soden [2006] and Vecchi et al. [2006] which also reported on a slowing down of the tropical atmospheric circulation. The total overall precipitation change is clearly positive which has also been suggested as a common model response for the deeper tropical regions [e.g., IPCC, 2007a].

Figure 12.

Contributions from the different components to the overall change in DJF precipitation for (top) wider region including tropical Australia and (bottom) land-only tropical Australia.

[43] Over the Australian continent (land only) the results are less uniform (Figure 12, bottom). Total change is small on average, but varies widely between models. The thermodynamic term is positive in all cases, but ranges in strength from very small (e.g., ukmo_hadcm3) to being the dominant component (e.g., giss_model_e_r). By contrast, this term is dominant for all models over the broader region. The dynamic component over Australia is generally negative, although two models have small positive contributions. The covariant term is quite large over the broader region for a number of models, indicating that simultaneous changes in both moisture content and circulation play an important role in precipitation anomalies (e.g., changes in the location of the edges of convective regions).

[44] The multimodel mean geographic distribution of regime-stratified precipitation change for DJF, based on grid point binning, is shown in Figure 13. The spatial pattern of dynamic and thermodynamic changes is consistent with the results for individual models discussed above.

Figure 13.

(a) DJF precipitation change (mm) under SRES-A2 scenario: 2080–2099 minus 1980–1999 averages, as simulated by the CMIP3 ensemble mean model and the three components contributing to the overall change: (b) dynamic, (c) thermodynamic and (d) covariant. Refer to equation (1) and text for details.

[45] The thermodynamic component of precipitation change (Figure 13c) is positive virtually everywhere, most strongly so broadly over the maritime continent, but extending over northern Australia. The dynamic term (Figure 13b) is negative over most of the region (with the exception of the ocean north of New Guinea), with strongest contributions to the immediate north and northwest of Australia. The covariant term is weaker and broadly follows the pattern of the dynamic term. The total change in precipitation (Figure 13a) results from the sum of increases due to thermodynamic processes, and decreases due to dynamic processes, with large precipitation increases north of 10°S and smaller changes over the Australia continent, and precipitation decreases to the west.

4. Summary and Conclusions

[46] This paper has considered changes to Australian tropical and adjacent region climate as modeled by CMIP3 climate models under the A2 scenario. Particular motivations for the study were that projected precipitation changes in this region are particularly uncertain [IPCC, 2007a], and potential impacts large [IPCC, 2007b]. Furthermore, there are large known biases in regional climate [Irving et al., 2011; Australian Bureau of Meteorology and CSIRO, 2011] and most notably in tropical precipitation processes [e.g., Dai, 2006]. In light of these factors it was of importance to document and understand projected changes and uncertainties in tropical Australian regional climate. Further scientific motivation came from the relative lack of analysis of the Australian branch of the Asian-Australian monsoon and its associated processes, and in particular those operating under climate change. The study also aimed to form a benchmark for comparison for future (CMIP5) model analyses.

[47] The main findings in the present study are as follows.

[48] 1. CMIP3 models project an overall warming over northern Australia of around +2.5°C to +5°C, of similar size and range to their global projections. Little change is seen to the annual temperature range, although summer land/sea contrast is enhanced by about 1°C on average.

[49] 2. Total summer precipitation in tropical Australia shows little change for model ensemble average, largely because the models are nearly equally divided into positive and negative projected changes. Greater agreement is found in a slight increase in the length of the wet season, characterized by increased precipitation in March/April, and associated with increases in low-level westerlies and decrease in MSLP. Upper level winds, both easterlies and westerlies weaken throughout most of the year. Stratifying models by their skill in representing the current climate does not change any of these conclusions.

[50] 3. Over the broader scale, significant precipitation increases are found to the north of the continent. Decomposing precipitation changes into “thermodynamic” and “dynamic” components shows broad-scale increases in precipitation from thermodynamic processes, particularly marked over the deep tropics. Offsetting this are decreases, principally associated with reductions in deep convection. These are strongest over oceans to the immediate north and west of the Australian continent. Over Australia the small change in precipitation is found to be the result of near balancing (negative) dynamic and (positive) thermodynamic contributions.

[51] 4. Inter-annual variability is found to significantly change only moderately on average, with this conclusion not being sensitive to model representation of the current climate. Correlations with ENSO are also not found to substantially change for the model mean.

[52] 5. Broad-scale wind changes indicate a strengthening of low level westerlies during the summer season and a weakening of high level easterlies resulting from changes to the overall tropical circulation.

[53] 6. Land/sea contrast changes point toward a significant relationship between pre-monsoon season (SON) land-sea contrast and DJF precipitation changes. Models with increased strength of land-sea-contrast tend to show larger precipitation changes.

[54] In conclusion, significant warming of the Australian tropical region is projected by the late 21st-century, but significant uncertainties exist in overall wind and precipitation changes in Australian monsoon and tropical Australian climate. These changes are not found to be strongly dependent on skill of models in representing the current climate. This is in accordance with earlier results [Whetton et al., 2007] where the relationship between inter-model similarity in patterns of current regional climate and inter-model similarity in patterns of regional enhanced greenhouse response was found to vary significantly regionally, and to be especially weak in the tropics.

[55] The small changes in precipitation, however, are found to be the result of offsetting thermodynamic (wetter) and dynamic (drier) contributions. In some models these components are very large, indicating the possibility that the small net change may be the result of offsetting errors. The models also have substantial biases in the tropical Pacific, particularly in the eastern warm pool [Australian Bureau of Meteorology and CSIRO, 2011] and the impact of these on mean climate and climate variability (particularly associated with ENSO) is unknown. To better understand these processes and uncertainties, it would be useful if this analysis were repeated with the new set of models available from the CMIP5 intercomparison [Taylor et al., 2011].


[56] We would like to thank Huqiang Zhang and Ian Smith for constructive comments on this paper. We acknowledge the modeling groups for making their simulations available for analysis, the Program for Climate Model Diagnosis and Intercomparison (PCMDI) for collecting and archiving the CMIP3 model output, and the WCRP's Working Group on Coupled Modeling (WGCM) for organizing the model data analysis activity. The WCRP CMIP3 multimodel data set is supported by the Office of Science, U.S. Department of Energy. The present work was partially supported by the Australian Climate Change Science Program, funded jointly by the Department of Climate Change and Energy Efficiency, the Bureau of Meteorology and CSIRO.