Driving factors and clustering analysis of expressway vehicular CO₂ emissions in Guizhou Province, China

Abstract

Expressways are essential for intercounty trips of passenger travel and freight mobility, which are also an important source of vehicular CO₂ emissions in transportation sector. This study takes the expressway system of Guizhou Province as the research objective, and establishes the multi-year expressway vehicular CO₂ emission inventories at the county level from 2011 to 2019. We employ the extended STIRPAT model incorporating ridge regression to identify driving factors from six different aspects, and then utilize the affinity propagation cluster method to conduct the differentiation research by dividing Guizhou’s counties into four clusters. Based upon clustering analysis, localized and targeted policies are formulated for each cluster to reduce expressway vehicular CO₂ emissions. The results indicate that generally: (1) Guizhou’s expressway vehicular CO₂ emissions manifest a continuously upward trend during 2011–2019. Small-duty passenger vehicle (SDV), light-duty truck (LDT), and heavy-duty truck (HDT) contribute to the largest CO₂ emissions in eight vehicle types. (2) GDP and population are the foremost two positive driving factors, followed by urbanization rate and expressway length. The proportion of secondary industry is also a positive driver, but that of tertiary industry exhibits an opposite effect. (3) Regional disparity exists in four county clusters of Guizhou Province. Efficient policies are proposed, such as improving the layout and infrastructure of transportation hubs, promoting multimodal integration, and implementing industrial upgrading as per regional advantages. Sustainable expressway vehicular CO₂ emission reduction is realized from both the source of industry and low-carbon modes of transport.

Appendix

Appendix 1. Structure of the extended STIRPAT model

The STIRPAT model stems from the IPAT (Impact=Population×Affluence×Technology) model established by Ehrlich and Holdren (1971). It was first developed by Dietz and Rosa (1997), which addressed some limitations existed in IPAT. Currently, the STIRPAT model has been widely employed to identify the driving factors with the following formula:

$$I=\alpha {P}^{\beta }{A}^{\gamma }{T}^{\varphi}\varepsilon$$

(12)

where I is environmental impact (denoted by the CO₂ emissions in this study); P, A and T represent population, wealth, and technology respectively; α is the constant term; β, γ and φ are the coefficients of P, A and T, respectively; and ε is the error term. By logarithmic conversion on both sides of Eq. (12), the STIRPAT model is given by:

$$\ln I=\ln \alpha +\beta \ln P+\gamma \ln A+\varphi \ln T+\ln \varepsilon$$

(13)

The STIRPAT model allows additional factors to be added, and in this study, we select ten potential influencing factors from six aspects, which include population, economy, urbanization, technology, industrial structure, and transportation based upon research findings of extant literature (e.g., Xiao et al. 2017; Tian et al. 2022; Li et al. 2023). Among them, the aspect of economy is expanded into GDP, total exports and imports, and total fixed assets investment. The aspect of urbanization subsumes urbanization rate. The aspect of industrial structure includes the proportion of primary industry, secondary industry, and tertiary industry. The aspect of transportation is the length of expressway. Then, the formula is expressed as below:

$${\displaystyle \begin{array}{c}\ln CE=\ln \alpha +{\theta}_1\ln TP+{\theta}_2\ln GDP+{\theta}_3\ln EI+{\theta}_4\ln FA+{\theta}_5\ln UR+{\theta}_6\ln RD\\ {}+{\theta}_7\ln PI+{\theta}_8\ln SI+{\theta}_9\ln TI+{\theta}_{10}\ln EK+\ln \varepsilon, \end{array}}$$

(14)

where CE shows the expressway vehicular CO₂ emission, and θ₁, θ₂, …, θ₁₀ denote the model coefficients. The detailed description of all the potential influencing factors and abbreviations in Eq. (14) is shown in Table 1 in Appendix 1.

Table 1 Description of potential influencing factors

The STIRPAT model is prone to be impacted by the multicollinearity existed in multiple influencing factors. In this study, we utilize the extended STIRPAT model incorporating ridge regression, which has the capability of surmounting such drawback. The detailed statement of ridge regression is provided as follows.

The ridge regression is modified from the standard model of multiple linear regression, where the unbiased estimation of multiple linear regression is given as:

$$\varphi ={\left({X}^{\top }X\right)}^{-1}{X}^{\top }Y,$$

(15)

where X denotes the independent variables and φ represents the model parameter.

If the multicollinearity exists in the variables, then the matrix is ill conditioned, namely ∣X^⊤X ∣  ≈ 0. Therefore, in order to eliminate this case, ridge regression is applied, and a non-negative term uI is introduced to modify the term X^⊤X, which is given as follows:

$$\varphi ={\left({X}^{\top }X+ uI\right)}^{-1}{X}^{\top }Y,$$

(16)

where u ∈ (0, 1) denotes the ridge coefficient.

Appendix 2. Results of the extended STIRPAT model

This study uses R software to solve the extended STIRPAT model fitted by ridge regression, which is specifically dedicated to multicollinearity analysis (Wang et al. 2013). The ridge regression parameter u in Eq. (16) was tested in the interval (0, 1) with step size setting as 0.01. Experiments indicate that when the ridge regression parameter u equals 0.16, the regression coefficients of influencing factors stabilized and the corresponding model results are provided in Fig. 9 and Table 2 in Appendix 2.

Fig. 9

The ridge traces of the significant independent variables

Table 2 Results of ridge regression

The model results indicate that there are 6 influencing factors significant at the 10% level, in which factors TP, GDP, UR, SI, and EK exert a positive influence on increasing the expressway vehicular CO₂ emissions of Guizhou Province, whereas factor TI is conducive to decreasing the CO₂ emissions. According to the value of regression coefficients, the order of positive factors is GDP > TP > UR > EK > SI.

Among all these driving factors, GDP is the primary driver of vehicular CO₂ emissions of the expressway system, which is followed by TP. Both GDP and TP have the direct effect on vehicle population and associated travel demand using expressways. The urbanization rate also exerts a positive influence on expressway vehicular CO₂ emissions in Guizhou Province. As a high proportion of Guizhou’s counties are still at the early stage of urbanization, the continuous agglomeration of economic activities between different counties, and intercounty production and consumption behaviors result in the growth of expressway vehicular CO₂ emissions. EK is an indicator that reflects the convenience of using the expressway service. The higher the expressway length, the more accessible counties that have the potential to reach by using expressways, resulting more expressway CO₂ emissions. Furthermore, the development of many Guizhou’s counties is highly reliant on resource-based secondary industry, which is at the lower end of industrial chain. It signifies that more freight vehicles are needed to complete such travel demand, thereby generating more expressway CO₂ emissions. In contrast, tertiary industry has a relatively low dependence on vehicles with high CO₂ emissions, which exhibits a negative influence on the expressway vehicular CO₂ emissions of Guizhou Province.