The initial green urbanization evaluation system for China was established based on the scientific evaluation principles of comprehensiveness, objectivity, comparability and operability (Pérez-Alonso et al., 2011; Zhou et al., 2018). Three green urbanization perspectives—level, quality and degree—were integrated in the index system to evaluate the green urbanization efficiency.

Five indices were selected to measure the urbanization level: urbanization rate; the ratio of rural employed labor; urban population density; per capita GDP; and the proportion of primary industry in GDP.

Social services, resource utilization and environmental impact indices were selected to measure the green urbanization quality. The social service index included education, science, health care, and employment, all of which promote a robust and sustainable economy. The following seven indices included four factors: living standards, education, health care and technology. Living standards were determined using Engel’s coefficient of urban residents and the funds for urban residents per capita under basic provision protection, education level was determined based on the number of high school enrollments and the education spending per capita, health care was determined by the hospital beds per thousand people and medical funds per capita, and due to limited data collection, technology spending per capita was chosen to represent the technology level.

As the main aim of green urbanization is reducing energy consumption, resource utilization was determined from the energy consumption of a unit of added value in industry and the urban daily water consumption per capita, the environmental impact was determined from the proportion of environmental investment in GDP, which directly reflected government efforts towards environmental improvements, and waste water discharge rate, which indirectly influences the environmental quality in a region.

The green urbanization degree was then employed to evaluate the impact on the environment, energy use, social stability and development in cities, with the specific environmental indices being the proportion of green areas in developed areas and the standard air quality proportion, the specific energy indices being energy consumption of unit GDP, which reflected the economic structure and resource utilization changes, and the specific social stability and development indices being the average per capita net income in rural areas.

The proposed green urbanization index system including 20 indices is shown in Table 1, therefore had a complicated structure, as 20 indices needed to be considered. After analyzing the characteristics of each index, the outputs were determined to be urbanization rate, the proportion of green area in developed areas, the standard air quality proportion, and the energy consumption per unit of GDP, which were respectively denoted y ₁, y ₂, y _3, y ₄, with the energy consumption per unit of GDP (y ₄) being the undesirable output. The principle of efficiency evaluation models is to maximize outputs. Therefore, the reciprocal y ₄ = 1/y ₄ was employed in the SBM models to ensure accurate results. The remaining 16 indices were the inputs, which were denoted x ₁ to x ₁₆.

(1)Principal component analysis (PCA)

Generally, the number of DMUs should be more than twice the number of inputs and outputs. As some of the selected indices had tight positive correlations, the information overlaps needed to be eliminated to ensure an accurate efficiency evaluation. Therefore, PCA, a powerful tool to reduce the number of variables (Põldaru and Roots, 2014; Azadeh, et al., 2013), was applied to remove the overlapping data from which the final reduced index system was determined to assess the green urbanization efficiency.

(2)Slacks-based measurement model

After the PCA index system reduction, the evaluation models were established.

The DEA model, first proposed in 1978 (Charnes et al., 1978), has been widely used to evaluate efficiency, with two main DEA models, the CCR and the BBC models, being the most well-known, with the CCR measuring overall efficiency under a constant return to scale assumption and the BBC measuring technical efficiency under a variable return to scale assumption. However, both have an obvious shortcoming as they presume that the inputs behave proportionally, which is unrealistic.

A second drawback of these radial models is that they neglect the slacks when reporting the efficiency. Therefore, of the DEA models, the slacks-based measure (SBM) model was selected (Tone, 2001) as it deals with the slacks directly and does not assume proportionate changes in the inputs and outputs (Chen and Jia, 2016). SBM models are more in line with reality, and have been used to measure energy efficiencies in Brazil, Russia, India and China (Song et al., 2013).

The efficiency was analyzed from technical, scale and general perspectives, for which two different models were employed: the SBM–CCR model which measures overall efficiency; and the SBM-BCC model which measures technical efficiency.

The respective equations are as follows: min h 0 = 1 − 1 m ( ∑ i = 1 m s i0 − x i0 ) 1 + 1 r ( ∑ i = 1 r s r0 + y r0 ) s . t . ∑ j = 1 n x ij λ j + s i − = x i0 ,     i = 1,2 , … , m ; ∑ j = 1 n y r j λ j − s r + = y r0 ,     r = 1,2 , … , s ; λ j ≥ 0 ,     j = 1,2 , … , n (1) and min h 0 = 1 − 1 m ( ∑ i = 1 m s i0 − x i0 ) 1 + 1 r ( ∑ i = 1 r s r0 + y r0 ) − ε ( ∑ i = 1 m s i − + ∑ r = 1 s s r + ) s . t . ∑ j = 1 n x ij λ j + s i − = x i0 ,     i = 1,2 , … , m ; ∑ j = 1 n y rj λ j − s r + = y r0 ,     r = 1,2 , … , s ; ∑ j = 1 n λ j = 1 λ j ≥ 0 ,     j = 1,2 , … , n (2) where ε is less than any positive number and greater than zero, and h 0 ∗ represents the optimal efficiency based on slacks; when h 0 = 1 , then the DMU is efficient, otherwise, the DMU is inefficient.

(3)Super-efficiency-SBM-C model (SE-SBM-CCR)

As traditional DEA models are unable to differentiate and rank the individual DMU efficiencies, the SE-SBM-CCR model, first proposed by Andersen and Petersen (1993), was applied for further comparisons and to rank the individual DMU efficiencies. The SE-SBM model principle is that for each efficient DMU, the model moves the production Frontier back and recalculates the efficiencies (Zhang et al., 2016). SE-SBM-CCR also eliminates the upper bound on the technical efficiency score and provides additional information regarding the relative performance of the efficient unit. Therefore, the SE-SBM-CCR model was applied in this paper to assess overall efficiency.

The constraints were as shown in Eq. (3), min h 0 = 1 − 1 m ( ∑ i = 1 m s r0 − x i0 ) 1 + 1 r ( ∑ i = 1 r s r0 + y r0 ) s . t . ∑ j = 1 , j ≠ k n x ij λ j + s i − = x ik ,     i = 1,2 , … , m ; ∑ j = 1 , j ≠ k n y rj λ j − s r + = y rk ,     r = 1,2 , … , s ; λ j ≥ 0 , j = 1,2 , … , n (3)

(4)Projection and improved analysis

The results from the DMU efficiency evaluations allow for projections and improvement analyses to assist the inefficient DMUs and provide quantitative and qualitative management suggestions. Because there are no uncontrollable variables, such as environmental variables, the projection model could be simplified to Eq. (4). Redundancy on inputs:  Δ x 0 = ( 1 − θ ∗ ) x i 0 + s i − (4) Shortfalls on outputs     Δ y 0 = s r + where Δ x 0 and Δ y 0 were the improvements in inputs and outputs. θ ∗ is the evaluated efficiency.

Table 4 shows the green urbanization efficiency trends in the 21 cities.

First, the overall efficiency (SBM-I-CCR) revealed that 13 cities were inefficient, ranging from relatively developed cities such as Zigong and Panzhihua to smaller autonomous prefecture cities such as Aba and Liangshan. As the average overall efficiency was 81.54%, an 18% improvement was needed. At the outset of the 13th Five-Year Plan in China in 2016, the Sichuan provincial government needed to focus on green urbanization as well as new urban construction. Aba ranked last of these 21 cities, with an SBM-I-CCR value of only 39.7%, which was far less than the 81.55% average, meaning that 60.29% improvements were needed. The efficiencies in both Aba and Liangshan were low because of poor natural conditions, underdeveloped economies, a lack of resources and backward technologies; however, these factors were not included in the environmental factors because of the difficulties in measuring qualitative factors, which are further discussed in the final section of this paper. Ganzi, however, was shown to be efficient, and therefore, the Aba and Liangshan governments should determine if Ganzi’s model would suit their local conditions.

Second, the average technical efficiency as assessed by the SBM-I-BBC was nearly 90%. Eight cities were found to be inefficient, with Guangyuan being the worst at only 58.46% efficiency, leaving a large space for technical improvements. Careful resource allocation is also critical as reducing resource waste could maximize the desirable outputs.

Finally, the scale efficiency average was 93.16%, with 13 of the 21 cities being found to be scale inefficient, all of which had decreasing returns to scale, which indicated that the marginal benefits were smaller than the marginal costs, resulting in a waste of inputs. Therefore, the scale would need to proportionally shrink to maximize the scale-efficiency.

The integration of the three models revealed the following. Only eight of the 21 cities were efficient. The technical and scale efficiency analysis showed that of these eight cities, Zigong and Luzhou were found to be inefficient by both the SBM-I-BBC and the scale-efficiency models, which meant that the inefficiencies were the result of both technical and scale issues. Of the other inefficient cities, five needed to only address scale problems.

As shown in Table 5 and Figure 4, the green urbanization efficiencies varied widely from 3.4022 to 0.5821, which revealed the serious unbalanced development in Sichuan Province. Most efficient cities were located in Chengdu, the capital city, and the northeast economic zones, with the efficiencies in the south, northwest and Panxi economic zones being sup-optimal. Therefore, the Sichuan government needs to adjust its development strategies and intensify investment in technology, infrastructure, and the provision of expertise in the southern and western areas.

A projection analysis was performed on the 13 inefficient cities, the results of which are shown in Figures 5, 6.

Figures 5, 6 identify the trends or features in the 21 cities and their green urbanization efficiencies.

First, with the exception of Ya’an, Aba and Liangshan, the inefficient cities had no basic development construction and most had basic urbanization input redundancies. Therefore, it was concluded that while the urbanization process in Sichuan was developing well, it was in a critical transition stage from basic urbanization to green urbanization, which means that the government needs to pay greater attention to improving the technical efficiencies and allocating redundant resources to improvements in the environment and energy.

Urban development in Aba and Liangshan has lagged far behind other cities, with the most serious problem being a poor utilization of the urbanization inputs. While there has been significant government investment, there are still large output shortfalls. Therefore, the respective city governments and the provincial government need to focus on greener urban development in these two cities by focusing on the possible shortfalls, which could be the poor natural conditions (poor resources, location or a shortage of qualified workers) or the lack of qualified experts and appropriate technical development.

Energy conservation appeared to be a difficult challenge for the inefficient cities, with almost every city (expect for Zigong) having a low energy utilization problem. For example, Panzhihua, the improvement ratio for which was greater than 100%, requires more resources to improve their energy utilization, which would require increased inputs and technical efficiency improvements.

Overall, as the greening construction in Sichuan was relatively good with more than 60% of the inefficient cities having no shortfalls, the main challenge is to reduce the pollutant discharges by updating technologies and moving toward industrial transition. However, the ratio of improvements for greening in both Aba and Liangshan exceeded 200%, which indicated that there may be an unwillingness by local governments to invest in the environment because of the unique and superior air quality in these cities. As both cities have low urbanization, the government needs to place its focus on urban construction rather than the environment; however, during the construction process, sound environmental protection methods should be used.