1. Introduction

According to [1], human activities are assessed to have caused approximately 1°C of global warming above pre-industrial levels. If these levels do not stabilize at 1.5°C before 2100, there will be irreversible or lost lasting effects on a wide range of climatic factors [1]. These include loss of precious ecosystems, melting of glaciers, rise in global sea levels, an onslaught of natural disasters etc. [1].

[2] and [3] found that the increasing atmospheric CO₂ concentration has been universally accepted to be the main driver of global warming. He further noted that global CO₂ has steadily declined since 1990, however even at this rate of decrease, according to various scientists, the target of 1.5°C will be difficult to achieve given that we are already standing at a 1.1°C increase. [4] have sufficient evidence to state that among countries such as U.S., India, some nations of the EU, and China will very likely cause global warming to reach 5. 1°C. Also, according to the authors, the US and Australia, combined, are only slightly behind in hurling global temperature to rise over 4C above pre-industrial levels.

Australia is the world’s sixth largest country and among the top twenty largest global greenhouse gas (GHG) emitters [5]. [6] informs that the geographical set in Australia is comprised of the Commonwealth (federal) government, six states and two territories, and about 560 municipalities. Here, environmental responsibilities are shared [6]. It is observed that political cycles are relatively short which negatively affects long-term policy planning [6]. Since the 1992 Intergovernmental Agreement on the Environment, states and territories have had the principal role in environmental protection [6]. This defines responsibilities shared between the federal and subnational levels [5]. Moreover, Australia’s per capita CO₂ emissions tend to be the highest amongst OECD countries, making it difficult to stem back emissions, as per international standards unless states and territories, individually and collectively, take firm action [5].

According to the [7], the main contributor to Australia’s GHG emissions at the sector level, in the period 1990–2018, is the stationary energy sector, with a contribution that ranges from 300 Mt CO₂-e in 1990 to 420 Mt CO₂-e in 2018. The second largest contributor to Australia’s GHG emissions is the agriculture sector, with participation in national total emissions that ranged from 100 Mt CO₂-e in 1990 to 90 Mt Co₂-e in 2017 [7]. The third highest emitting sector in Australia was LULUCF, with a contribution range for the period 1990–2017 of 200 Mt CO₂-e and -20 Mt CO₂-e, respectively [7].

As shown in Table 1 (figures taken from [7]) at the state and territory level, the change in GHG emissions in the period 1990–2018 ranged between -817% in Tasmania (TAS) and 18.84% in the Northern Territory (NT). The economic development of all the states was roughly similar in the two decades under observation. Four states and territories showed a decrease in GHG emissions. These include -52.575 in South Australia (SA), -34.06% in New South Wales (NSW), -7.25% in Queensland (QLD) and -7.16 in Victoria (VIC). Further, the highest change in emissions was seen in the NT. The only other positive change in emissions was in Western Australia (WA) by 17.24%. The differences between states and territories in the GHG emissions change for the period 1990–2018 are explained by the diverse economic, political, demographic, climatic, and behavioural characteristics of each. This means that there are different drivers for GHG emissions growth in them. However, in certain states and territories, it is seen that none of the drivers seems to explain the change in GHG emissions. These could be attributed to the higher intensity and number of times climate change policies were instituted at the state or territory level. Most of the states and territories have reported a decrease in emissions over the years. Emission management actions and policies vary across sectors, the great decline in land use change emissions over the observed period and structural changes in the economy are gauged to be responsible for less emissions [8].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Australia’s GHG emission growth and percentage change 1990–2018.

https://doi.org/10.1371/journal.pclm.0000091.t001

The understanding of both the main factors determining GHG emissions growth and the differences in the impact of these factors at the state or territory level are of extreme importance. This is to enhance the design and implementation of state and territory driven appropriate mitigation to minimize emissions at the overall national level in Australia.

2. Literature review

The objective of this paper is to overcome the selection and modelling limitations of earlier works. The literature review below lists methods and determinants of GHG emissions in other data sets along with their limitations.

To select the explanatory factors, we considered statistical significance as in previous studies. A study of Australia’s greenhouse gas emissions was done by [9]. The author found that the current gulf between growth and exports was the key problem in need of addressing. But there was sparsity in the selection of variables which is not present in our research. Also, input-output models are severely restricted which casts doubt on the significance of the results.

Using Johansen Cointegration Technique [10] found no causal link between CO₂ emissions and economic growth in Australia (from 1965–2007). This is why we eliminated economic growth from the study.

Several studies have used regression analysis to determine drivers of GHG emissions. Regression analysis mathematically finds out which variables have an impact. The authors in [11] used 40 years (1971–2012) data to analyze changes in GHG emissions in ten countries. They used multiple linear regression model based on the Kaya identity. It was found that heterogeneous factors were critical for explaining GHG emission evolution in different countries (carbon intensity for China, GDP for US, Canada and India etc.). However, it is imperative to mention that the data was not taken from the same source, hence giving rise to the problem of the non-homogeneity of data. The heterogeneity of data gives rise to the possibility of errors in the data modelling.

A large number of studies have used different variations of decomposition analysis. It is a statistical technique in which predetermined variables are extrapolated from the economy and each of their impacts is gauged for GHG emissions growth. It also provides a better understanding of the impact of various variables on energy use. [12] and [13] improved the decomposition analysis by employing different regression techniques for assessing the heterogeneity of GHG emissions and emission intensity. Sector-specific CO₂ decomposition was done by [14]. The author used Logarithmic Mean Divisia Index (LMDI) method on electricity generation in seven Asia-Pacific and North American countries. They found that the production effect was the major factor responsible for rising in CO₂ emissions during the period 1990–2005 as well as the projected values till 2030. [15] used time-series and cross-sectional decomposition and found that growth in per capita GDP and population were the two chief drivers to the increase in CO₂ emissions in most cases in APEC countries. Structural or Index Decomposition Analysis (IDA) extended from energy consumption to energy-related CO₂ emission studies by research carried out since 1991 [16] and [17]. [18] also undertook decomposition analysis and applied it to a larger sample (1971–2010). They found that economic growth was the main driver for increasing CO₂ emissions among G-20 countries. [19] used decomposition analysis, Theil index, to study inequality in CO₂ emissions across different countries. They studied carbon intensity of electricity production, electricity intensity of GDP, economic growth in terms of labour productivity and employment rate. They found that economic growth in terms of labour productivity was the responsible variable for global inequality in CO₂ emissions. [9] using LMDI, found that over 1978–2010 energy efficiency played a leading role in a 17% reduction in CO₂ emissions aggregate intensity in Australia. More recently, [20] used LMDI decomposition and decoupling analysis methods. Their research implied that energy intensity and per capita GDP were the principal factors of CO₂ emissions in OECD countries. But decomposition analysis faces certain limitations, such as not allowing for hypothesis testing. Hence, empirically tested results cannot be derived. Moreover, these studies faced limited identification problem casting doubt over their results.

Our study does not suffer from limited identification problem in terms of comprehensiveness of variables. The data gathered is from homogenous sources. We do not have sparsity in the data, unlike previous studies. Also, we carry out regression analyses for hypothesis testing and gauging the significance of GHG determinants. A comprehensive list of the explanatory variables that we initially selected along with their causal theory, supporting literature and expected sign, are summarized in the table in Table 2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Causal relationship for each determinant of GHG emissions.

https://doi.org/10.1371/journal.pclm.0000091.t002

3. The methodological framework of the study

First, we conducted the methodological review to find out determinants applicable to the dataset. Next, multicollinearity, stationarity and heteroskedasticity tests were done on the data to find a suitable model. Multicollinearity test was done to determine whether several independent variables in a model were correlated.

After filtering out variables, time series data model (FGLS Regression) was used on the state and territory level and the panel data model was used on Australia for the time period 1990–2018. This method can estimate coefficients of a multiple linear regression model and their covariance matrix in the presence of nonspherical innovations with an unknown covariance matrix. We then carried out the climate change policy analysis both on the national level as well as on the state and territory level. Later, the panel data model was applied by decade (1990–99, 2000–09 and 2010–18) to check for robustness of results. Finally, the results of time series model and panel data models along with the crux from the discourse of climate change policies were combined to produce discussion and conclusions.

4. Results

By looking at iterative Cochrane-Orcutt estimates’ results, it becomes evident that only NSW, VIC and QLD show significance in certain variables. The results for SA, WA, TAS and NT show that none of the determinants were found to be significant in these states and territories.

Particularly, EI is significant in NSW, VIC and QLD. PE and Pop are significant in both NSW and QLD. Con is significant in only VIC. And SE is significant in only QLD.

In the Panel Data Model, Hausman test was used to discriminate between fixed and random effects. The value of the test was 2.844 with a p-value of 0.828. Hence, the null correlation among individual effects and regressors was accepted and decided that the random effects panel data was efficient. The results of the parameter estimators and their significance using the random effect panel data model show that all variables except Pop are significant.

5. Analysis

The first model tested shows various results for the explanatory power of the independent determinants in determining GHG emissions. The value of R² ranges from 85% in Victoria to 27% in NT. Overall, normality or heteroscedasticity was not found in the data. There might be model specification issues with SA, WA, TAS and NT because none of the independent variables was found to be significant in them.

The results for the random effect Panel Data Model show that all determinants are significant for explaining GHG emissions growth except population. This indicates that at the national level, except for population, all the determinants analysed explain changes in GHG emissions.

This discussion goes into deeper detail, as we tie statistical results with the development of Australia’s climate change policies on national as well as state and territory levels, over the years.

5.1. Australia’s national and state (or territory) climate change policies

In Australia, states and territories have their own policies and instruments. According to [39], during the mid-1990s, concurring with free market economic ideology, certain industries such as coal, oil, natural gas and other extractive industries, along with other multinational corporations, such as the energy-intensive aluminium smelting industry got together and influenced the Australian government [40]. These emissions were characterised by few intensive, geographical locations [41]. Also, the politicians in power were staunch supporters of stimulating economic growth and sceptical of climate change [42]. Although the CO₂ emission targets were instituted, former science minister Barry Jones in World Meteorological Day address (1992), in favour of combatting economic growth, was reported to withdraw support for climate change and put climate mitigation policies on the back burner. However, even with these forces, in December 1992, Australia was the ninth country to ratify the UNFCCC. In April, 1998 Australia signed the Kyoto Protocol, along with 20 other countries. However, Australia did not ratify the Kyoto protocol. This meant that the emission targets were not legally binding [41].

The initial lukewarm response to climate change in the field, as opposed to on paper, led Australia’s climate mitigation policies towards lower emission reduction as opposed to a more responsible climate change policy. This was especially visible in NSW for the first decade under consideration, where unlike other Australian jurisdictions, including VIC, TAS, SA and the ACT, there were no legal mechanisms or all-embracing climate action in place [43]. It was only later that the state government decided to introduce changes in its environmental policy. This is apparent from decreasing emissions from NSW since 2008 (Fig 1). Similarly in QLD, limited policy measures [44] translated into causing emissions to remain highest among all the other territories and states, and rising since 2016. The case is the opposite for TAS whose target to run on 100% renewable energy has been achieved [45]. Since the 1990s, Tasmania’s overall emissions have been constant, and since 2012 they have started decreasing.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. GHG emissions in Australia.

https://doi.org/10.1371/journal.pclm.0000091.g001

Australian Government’s Carbon Pollution Reduction Scheme (CPRS) that caused much stir during 2008–09 met its gloomy fate when lobbying caused changes to subsequent proposals, favouring carbon intensive industries. In June 2002, the Howard Government claimed to the Australian Parliament that ratifying the Kyoto Protocol was not in the national interest [46]. But in December 2007, Prime Minister Kevin Rudd ratified the Kyoto Protocol, as promised during the 2007 election campaign [47]. The on and off policy circus became visible in some states, for instance, VIC’s emissions increased till 2012, but then decreased. This is because policies that ran up to The Victorian Climate Change Act 2017 were not favourable, but the act replaced the earlier. Climate Change Act 2010 (VIC) and was a comprehensive legal framework unlike earlier policy activities (Project Victoria) that undid actions that reduced Australia’s emissions through energy-demand management strategies [48].

Some states such as SA, NT and WA have had stable emissions over the past two decades (Fig 1) although SA was the first Australian state to legislate targets to reduce greenhouse emissions [49]. But it seems that a blend of policies and a general stability in demand-led emissions seem to have not decreased emissions although no notable increase was seen in all three cases. Owing to rising emissions in some states and territories and non-changing emissions in others, Australia could not meet its Kyoto 2008–12 target [5]. In December 2015, at COP21, Australia announced it had a net zero emissions target by 2100 [50]. However, due to the political and economic factors embedded deep in the environmental landscape of Australia, a greener future remains uncertain. Thus, if Paris Agreement goals are to be met, the government needs to streamline its approach and clarify how existing and new instruments could be implemented [51] and [52].

5.2 Robustness of results

We subdivided the sample by decade; 1990–1999, 2000–2009 and 2010–2018 as shown below in Table 3. This was done to check robustness of results using the Panel Data Model. The analysis by states and territories using multiple linear regression models could not be performed as the degrees of freedom of the estimation would have lead to inconsistent results. This analysis is done to assess the dynamics of the parameters, i.e., to analyse if the effect of each determinant has changed during the given time period.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Analysis by decade- panel data model estimates where * and ** show significance at 5% and 1% respectively.

https://doi.org/10.1371/journal.pclm.0000091.t003

The results indicate that only in 1990–1999 the intercept was significant at 1% which means that it shows different results as compared with other decades. This decade was marked with overall skepticism in Australia regarding climate change [53]. Due to this reason, it seems that other than SE all other estimates were found to be significant in determining GHG emissions. This confirms the failing of climate change policy in Australia for the decade. The temporal evolution of estimates is such that all of them appear to be decreasing by the time that we reach the decade of 2010–2018. In 1999–2018, only Pop was not significant. During 1999–99 Pop and SE were insignificant at 5% while other variables, such as PE and AFF, exhibited 1% significance. In 2000–09, we saw a general weakening of the relationship between the determinants and GHG emissions. Lastly, in 2010–18 none of the determinants remained significant. This result aligns with the projections of a report by [5] that observed that Australia is on track to reach its 2020 emission reduction target although it is still too early to know whether it would achieve its goal of zero emissions by 2050.

6. Discussion

According to the literature, the determinants chosen in this study are among the most significant of all variables that can determine GHG emissions in a given data set. However, one limitation to be noted is that some unique variables, such as Savanna burning, for instance, which contributed heavily to the NT emission in 2009 [54] were not included due to non-availability of data. Also, some variables which have contributed heavily to emissions over the years have either limited or no data available to reflect their exclusive contribution to GHG emissions e.g. Transport and LULUCF.

Some variables were found to be chief contributors in only some, but not in the majority of states and territories, so they were not included in the statistical analysis. This is because time series regression does not take into account the unique characteristics of the data set. Hence, this seems to be the reason that in some states and territories (QLD, SA, TAS and WA), none of the given determinants was found to be significant.

Moreover, an analysis of variables significant from year to year shows huge disparities. For example, [54] compares emissions from different sectors in 2005 to 2017 levels. In NSW and WA, initially, transport and energy sector were notable contributors to GHG emissions, but they dropped to almost zero in 2017, while waste, which contributed practically zero became a chief variable later. In SA, LULUCF (Land Use, Land Use Change and Forestry) became a main contributor in 2017. Mn and Con which had zero contribution in 2005 became notable variables later in VIC and WA. This uneven contribution of variables over the years, further explains why in some states some or all of these variables were not found to be statistically significant. This is a limitation of FGLS regression.

Further, there is a significant probability that a variable might seem insignificant because of good policies that decoupled the link between GHG emissions and the given determinant rather than one or more of the reasons listed above. Different states and territories have different experiences in this regard as already discussed in section 5.1.

PE and EI have been significant determinants in the case of NSW and QLD. One unit change in energy intensity causes GHG emissions to change by a high elasticity. Both these states are two of the highest emitters of GHG emissions in Australia. This result shows that since the past two decades, even with the breadth and width of climate change policies introduced in these two states, the strength of their link with emissions could not be made tenuous.

Pop has been an important determinant in GHG emissions, meaning that demand led emissions are still strong in both NSW and QLD. Pop has a negative sign which means that with one unit increase in population, emissions decrease by a respective unit. It might mean that with increase in population, policies are getting more and more stringent. Also, a higher EI would mean it would be more expensive to convert energy into GSP; EI is found to be an important determinant for both states.

Con and EI seem to be significant variables in VIC. This is because there has been no policy that determined building and construction rules over the past few decades [48]. Although, since 2012 its emissions have been decreasing, no significant or effective renewable energy policies have been instituted as discussed in section 5.1.

NT, SA and TAS emitted lowest GHG emissions which have averaged over 1990s levels. Among these states, TAS has achieved 100% renewable energy target. It also seems that TAS has devised successful policies to maintain emissions in its highest emission producing sector, LULUCF, followed by emissions from Mn, AFF and Waste [45]. But it seems that if the state is aiming at zero emissions by 2050, sectors other than energy, need to be targeted as well. Other states could also use the renewable energy policy model employed in TAS. Transport was a chief contributor in SA over the sample period [55], but there was no data available to reflect its contribution. It might be that due to favourable policies, emissions were stable and so none of the determinants was found to be significant, but this cannot be said with certainty because neither full data for the chief contributors were available nor any record of matching policies could be found. Further, there have been no favourable actions by the NT government in the past years, and so, NT’s emissions have not declined from 1990s levels. We can cautiously infer that environmental policies were perhaps successful in weakening the link between emissions and sources in cases such as NT, TAS and SA, as discussed, but we cannot state this with complete certainty due to a lack of relevant policy data.

WA’s emissions are expected to rise rapidly during the short to medium term. Also, WA’s gas consumption is highest among all the states and territories in Australia [55]. Gas is consumed mostly by large industrial and mining users, the mineral processing sector, and electricity generation at high levels. This means that PE, Mn and SE should have been significant determinants for WA. However, this seems not to be the case, perhaps because more reflective variables such as quantified electricity supply should have been used, although there is limitation to the availability of data for this variable.

Mn and AFF were two variables that were not found to be significant in any of the states or territories. In section 5.1., we discussed that Australia was unable to become a forerunner of climate mitigation policies because of the problem of small firm lobbying. However, after the 1990s, the government steered its efforts towards environmentally friendly policies. Also, while it is complex to move towards 100% renewable energy, legislation can, with fuller force, target emissions produced by the manufacturing sector. It seems that with recent effective policies, SE, decoupled from growth in emissions. Regarding AFF, one plausible explanation of it being non-significant is that this variable counts both forestry and fishery in its composition. Agriculture is critical for a more holistic study because it is one of the top three contributors to GHG emissions in any state or territory [44]. And while we have seen agriculture to be a significant potential contributor to GHG emissions, forestry and fishery might have skewed statistical output towards rendering it insignificant. Further, no information regarding agriculture was available, so we used AFF to reflect it. Hence, a proper measure of Agriculture is required.

Lastly, when comparing robustness of results through Panel modelling, we find that during 1990–2018 all the determinants were found to be significant. Particularly, during 1990–99 most of the variables were significant. It reflects the national policy of Australia for not being sufficiently green in the 1990s. Although, this does not reflect favourable policies instituted on sub levels, such as in TAS, SA and ACT during the decade. Although by the decade of 2010–18, none of the variables remained significant. It seems that overall climate change policy has been effective, progressively, but [56] stated that emissions were reported to be highest in 2017 (in Australia) in the past two million years. This could be due to some outlier year or years, or due to certain states and territories. More research is needed in this regard. However, both our panel data model and an independent study [5] observations indicate that Australia has been positively moving towards more stringent climate change policies (in at least a few states).

7. Conclusion

We have explored factors affecting GHG emissions by taking into account state and territory heterogeneity. Normally, a good environment-friendly policy decouples the relationship of emissions with a particular variable. However, only the presence of a policy is not sufficient. There have been major policy changes and actions done in Australia both on a national level as well as in states and territories as per climate change, but lobbying by the industry and a general scepticism has led only some of the states or territories to report encouraging decrease in GHG emissions.

The conclusions can be summarised as follows:

1. The uneven impact of the main determinants of GHG emission growth suggests that a differentiated application of Australian policies at state and territory level will enhance the efficiency of mitigation efforts in Australia.
2. Improvements in renewable energy, such as in TAS, will be crucial for future reductions in GHG emissions moving forward and also to achieve the mitigation goals agreed in the Paris Agreement.
3. Moreover, it is also to be kept into consideration that without a state or territory analysis, the results can mislead us into thinking that the environmental policy is stronger than the case in reality. So, it should be cautioned that the idiosyncratic effects, as per state or territory, should be taken into account.
4. Lastly, imagining a 100% green future, as evidenced, by all the determinants chosen, seems to be a probable task, but the fact that some of them are still significant, on territory or state level, shows that the work of Australia in mitigating GHG emissions is far from done.

The study also had some limitations:

1. Certain variables were not included because they did not contribute majorly in all states and territories. Also, for some variables, limited or no data was available.
2. One limitation of time series regression is that it does not take into account significance of variables in specific time periods, hence, in many states, variables that were otherwise significant in certain time periods did not make it to the list of final significant variables to be studied. Future studies can employ more overarching statistical methods and data gathering techniques that address these limitations to present a more holistic picture.