Temporal–Spatial Variations and Convergence Analysis of Land Use Eco-Efficiency in the Urban Agglomerations of the Yellow River Basin in China

Abstract

Achieving synergistic development of efficient urban land use and the natural environment is crucial in promoting green urbanization. The assessment of land use eco-efficiency (LUEE) and its temporal–spatial changes provides an effective means of quantifying the relationship between the urban ecological environment and land use. Targeting 55 selected cities in the Yellow River Basin (YRB), in this study, we utilize the Super-EBM method to gauge the LUEE. We explore the temporal patterns and the spatial convergence of LUEE utilizing kernel density estimation and spatial econometric methods. Considering the resource and environmental costs of land use, we assumed the industrial pollutant emissions generated during urban land use as the undesired outputs and designed a framework for measuring the level of LUEE under double constraints, which theoretically revealed the formation process and spatial convergence mechanism of LUEE. The results show the following: (1) Throughout the sample period, the LUEE of the YRB urban agglomeration decreased from 0.158 in 2009 to 0.094 in 2020, indicating a decreasing spatial disparity in LUEE over time. Notably, the Lanxi urban cluster exhibited the largest gap in LUEE, whereas the Guanzhong Plain urban agglomeration displayed the smallest gap. The hyper-variable density exceeded the inter-group gap as the main factor leading to the difference in LUEE. (2) Although the LUEE of urban agglomerations has increased, there still exists a noticeable polarization phenomenon. (3) The LUEE of YRB demonstrates a pattern of conditional convergence and exerts a significant spatial spillover effect. Over time, the LUEE of YRB will tend towards an individual steady state. The findings have implications for strengthening linkage and synergy among cities in YRB, promoting factor integration across administrative regions, and formulating heterogeneous policies.

1. Introduction

Urbanization is a result of the progress of human civilization. China is currently experiencing an urbanization phase [1], and this continuing process is dramatically increasing the demand for urban land. In 2021, the urbanization rate in China reached 64.7%, contributing significantly to economic growth and increasing farmers’ incomes [2]. The Yellow River Basin (YRB), spanning eastern, central, and western China, is an area where issues concerning economic and social development, as well as the ecosystem, are particularly prominent. In 2019, the value of urbanization in the YRB was 3.7 percentage points lower than the national average [3]. In the past, extensive land use practices tended to ignore the eco-efficiency of land use itself, resulting in issues such as low land use efficiency, sloppy use, and ecological degradation in various cities within the YRB. Statistics from the Ministry of Natural Resources show that in the YRB, 14.47 hectares of new construction land is consumed for every unit of GDP growth, which is about 1.60 times the national average [3,4,5]. The land average GDP and investment intensity of fixed assets in building land are relatively lower than the Chinese average, showing the characteristics of rapid expansion of building land but a low land use efficiency. Furthermore, the proportion of building land in cities and rural areas in the YRB is about 2.32%, the proportion of arable land area is about 18.64%, and the proportion of ecological conservation land area is about 79.04%. The intensity of land development in the YRB (urban area) is 8.53%, higher than the nationwide average of 7.02%, and as high as 16.04% and 18.01% in the downstream areas of Henan and Shandong [6]. The intensity of land resource use in certain areas has exceeded the eco-environmental carrying capacity, threatening the ecological and environmental security of the watershed. “Eco-efficiency” also refers to the environmental management of nature [7]. The unilateral pursuit of a high expected output of land in the urbanization process, which reduces our ability to be in contact with nature [8,9], is no longer compatible with the current ecological philosophy.

The Central Committee of the Communist Party of China took the modern route of prioritizing ecology and green development by proposing the promotion of comprehensive ecological improvement of beach areas. With a focus on intensive land use and environmental protection [9], their strategy was to obtain the maximum economic, social, and eco-efficiency with the least land investment—crucial to achieving regional green development [10]. How to enhance the LUEE and establish the rational urban planning and governance framework [11,12,13], especially the coordinated development of land use and ecology, are fundamental points of common concern in current academic circles. The study of the ecological efficiency of land use can provide decision-making support for optimizing ecological conservation projects and achieving green urban development [14,15,16].

The available research literature on land use has focused on efficiency measures and their influencing factors. On the one hand, the measurement methods of land-use efficiency are the entropy method [17], the DEA method [18,19], stochastic frontier analysis (SFA) [20], the Cobb–Douglas production function, and the Malmquist index [21,22]. Chen et al. (2016) assessed the built-up land efficiency of urban areas utilizing the DEA model and investigated the influence of urban structures on land use efficiency [23]. Xie and Wang (2015) [24] and Jiang (2021) [25] employed the SBM model to measure the land use efficiency of Chinese cities. On the other hand, some studies have identified factors such as industrial structures, population density, GDP per capita, foreign direct investment, and degree of marketization as influencing land use efficiency [24,26]. Zhou et al. (2022) concluded that per capita food ownership was the major element of agricultural land use efficiency in most areas of the YRB [27]. Xue et al. (2022) found that population density, foreign investment, level of innovation, and transportation facilities contributed to urban land use efficiency in the YRB [28]. In addition, several scholars have also started to focus on the relationship between land use and the ecological environment. Huang et al. (2017) found that the FDI and firm heterogeneity significantly affect land use efficiency in economic development zones [29]. Zhu et al. (2019) found that economic development, infrastructure, and markets significantly impact land use efficiency [30]. Li et al. (2022) concluded that concentrated urban land use in China had differential beneficial effects on lifestyle and the eco-environment [31]. In summary, existing studies provide a good research basis for this study, but further expansion and improvement are still needed. This study takes a total of 54 prefecture-level cities in five urban agglomerations in the YRB as the research objects, measuring the urban land use eco-efficiency of these urban agglomerations for the 2009–2020 period and analyzing the convergence characteristics.

The marginal contributions of this study are as follows. First, the remaining literature mainly focuses on the measurement and drivers of land use efficiency [32], while research on the ecological efficiency of land use is still lacking. This study systematically explores the formation process of the “resource–economic–ecological” system in land use and the convergence mechanism, which can provide theoretical support for analyzing the distributional evolution of LUEE. Second, the land use efficiency in the existing literature reflects the connection between land use and economic output, which have not integrated ecological indicators that encompass non-expected outputs. We assumed industrial pollutant emissions as the undesired outputs and designed a framework for measuring the level of LUEE under the two constraints, which helped us understand the relation between urban land use and ecology. Third, although certain researchers have studied the relationship between land use and ecology [33], they have not yet measured LUEE. The Super-EBM model and spatial econometric methods are applied to measure the LUEE and reveal its spatial and temporal evolutionary characteristics, which can bridge this shortcoming.

The remainder of the paper is arranged as follows. Section 2 presents the theoretical framework. Section 3 describes the materials and methodology. Section 4 presents the results. Section 5 presents the discussion. Section 6 contains conclusions and implications.

2. Analysis Framework

Applicable to agricultural land, the urban land use eco-efficiency can reveal the interaction between the environment and the economy in the land use process [34,35,36,37]. During China’s urbanization process, many rural laborers flocked to the cities, which increased the need for land [38,39]. As this urbanization continues, the area of land used for urban construction is rapidly expanding [40,41]. The increasing regional imbalances and disparities in urban development directly hinder the harmonious relationship between urban development and the eco-environment [42,43]. First, as shown in Figure 1, in terms of ecological resource inputs, the process of land use alters the utilization levels of various land resources, thereby promoting the establishment of the urban infrastructure [44,45]. Each type of land resource exhibits different levels of economically efficient outputs. A high level of economically efficient land output stimulates the flow of production factors towards sectors with higher productivity or growth rates [46]. In addition, the rapid expansion of cities poses a substantial risk to sustainable urban development [47]. Unreasonable urban land use results in a substantial waste of land area [48,49,50] and consumes a considerable amount of arable land [51,52,53], leading to the inevitable generation of large quantities of pollutants and emissions, which negatively impact the eco-efficiency [54].

Secondly, land use exerts an impact on eco-efficiency through its influence on economic outputs. On one hand, agricultural land users strive to maximize economic output per unit of land area by implementing agricultural mechanization and modernization [55,56]. Urban land users utilize funding to introduce new technological innovations [57,58] and optimize the allocation of land resources [59], aiming to maximize economic output with minimal land inputs and industrial pollutant emissions [46]. Additionally, the Chinese government, as a key participant in land resource development and utilization [60], can reallocate and manage land resources to maximize the financial returns derived from land development. The health of regional biodiversity and ecosystem services will deteriorate if an increased land use efficiency is pursued at the expense of neglecting the ecosystem bearing capacity [55,61], impacting the ecological benefits of land use [62,63]. Conversely, if the development of land resources is constrained to protect the health of the ecosystem, the phase urbanization process will slow down [64].

Accompanying the acceleration of new urbanization, production factors such as capital and labor will over-concentrate in a specific geographic space, resulting in a “crowding effect”, which would lead to the transfer of some factors to neighboring regions. In addition, technological spillover has an evident controlling effect on pollutant emissions from land use, so that the LUEE of neighboring cities tends to increase simultaneously. At the macro level, in the context of the increasing the dependence of economic development on technological progress, secondary and tertiary industries are closely linked in terms of scale, structure, and spatial layout. Inter-industry synergistic development and technology diffusion promote the upgrading of industrial structure in the direction of low-carbonization and low-pollution, and the LUEE of cities tends to be spatially convergent. In addition, national support policies promote the concentration of knowledge and technology factors in less developed areas, thus increasing the LUEE of cities in such areas. The Chinese central government issued the Outline of the Plan for the Ecological Protection and High-Quality Development of the YRB, which requires cities to make rational use of land resources in beach areas and to implement the differentiated use control of land space in such areas.

In the long run, with the application of green technology, the upgrading of the industrial structure, and the implementation of favorable government policies, the economic output of land use and industrial pollutant emissions on the constraints of LUEE will be weakened, and the theoretical phenomenon of spatial convergence should occur. Therefore, achieving the optimal exploitation of land resources and improving land use eco-efficiency are essential strategies to promote green urban development [65]. Examining the land use eco-efficiency helps us understand the relationship between ecological resources and land use and is conducive to facilitating the coordinated development of a natural resource–economy–ecological system (Figure 1).

3. Materials and Methods

3.1. Sample Selection and Data Source

Considering the administrative divisions, 55 cities were selected as the sample for this study, covering five major urban agglomerations in the YRB. We used ArcGIS 10.2 software to map the urban agglomerations in the YRB. Figure 2 depicts the spread of the five major urban agglomerations in the YRB, including the Central Plains (CP), Lanxi (LX), Guanzhong Plain (GP), Shandong Peninsula (SP), and Hubao–Egyu (HE). The data used were from the China Urban Statistical Yearbook and the statistical yearbooks of each province (city) in 2009–2020. The specific city groups in the YRB are listed in Appendix A. The relevant descriptions of the data sources used in this study are shown in

Table 1. Data information and sources.

Data	Unit	Time Interval	Source
Land cover data	km2	Year	http://www.stats.gov.cn/, accessed on 20 October 2022
Demographic data	person	Year	http://www.stats.gov.cn/, accessed on 8 November 2022
Energy input data	-	Year	https://www.epsnet.com/, accessed on 24 November 2022
Industrial emissions	-	Year	https://www.epsnet.com/, accessed on 17 December 2022
Economic data	CNY	Year	http://www.stats.gov.cn/, accessed on 9 January 2023
Number of hospital beds	sheet	Year	https://www.epsnet.com/, accessed on 30 January 2023
Number of cell phones	unit	Year	https://www.epsnet.com/, accessed on 7 February2023

.

3.2. Methods

3.2.1. Super-EBM Model

To avoid the limitations of the DEA method, which often relies on strict assumptions, Tone et al. proposed an EBM model that incorporates radial and non-radial efficiency measures [66]. However, a challenge remains in analyzing the efficiency differences among the evaluated units (DMUs). To address this issue, Andersen et al. proposed a Super-DEA model that can differentiate between efficient DMUs [67]. In this study, to ensure the comparability of efficiency values, we constructed a non-desired output, non-oriented, and scaled payoff-invariant Super-EBM model to measure land use eco-efficiency (LUEE) in the YRB. The formulas for the model are as follows.

$${\gamma }^{\ast }=min\frac{\theta -{\epsilon }^{-}{\sum }_{i=1}^m\frac{w_i^{-}{s}_i^{-}}{x_{ik}}}{\phi +{\epsilon }^{+}({\sum }_{r=1}^s\frac{w_r^{+}{s}_r^{+}}{y_{rk}}+{\sum }_{p=1}^q\frac{w_p^{-}{s}_p^{-}}{z_{pk}})}$$

(1)

$$S.T.\left\{\begin{array}{c}{\displaystyle \sum _j^n}{\lambda }_j{x}_{ij}+s_i^{-}=\theta x_{ik} (i=\mathrm{1,2},\dots ,m)\\ {\displaystyle \sum _j^n}{\lambda }_j{y}_{rj}-s_r^{+}=\phi y_{rk} (r=\mathrm{1,2},\dots ,s)\\ {\displaystyle \sum _j^n}{\lambda }_j{z}_{kj}+s_p^{-}=\phi z_{pk} \left(p=\mathrm{1,2},\dots ,q\right)\\ {\displaystyle \sum _j^n}{\lambda }_j=1,{\lambda }_j\ge 0 \\ s_i^{-},s_r^{+},s_p^{-}\ge 0\\ \theta \le 1,\phi \ge 1\end{array}\right.$$

(2)

where ${\gamma }^{\ast }$ is the integrated efficiency value, $m$ denotes the number of inputs, $n$ denotes the number of DMUs, $s$ denotes the number of desirable outputs, $q$ denotes the number of undesirable outputs, $x$ denotes the inputs, $y$ denotes the desirable outputs, $z$ denotes the undesirable outputs, $w_i^{-}$, $w_r^{+}$, and $w_p^{-}$ are the weights of the corresponding output variables, $\theta$ is the calculated radial efficiency, and $s_r^{+}$ and $s_p^{-}$ are slack variables.

3.2.2. The Dagum Gini Coefficient

In this study, the Dagum Gini coefficient and the subgroup decomposition method were chosen to analyze spatial differences in the LUEE in urban agglomerations [68]. The Dagum Gini coefficient takes into account the crossover between samples and the distribution of sub-samples, allowing for more accurate tracking of the sources of spatial variation [69,70,71]. Before decomposing the Dagum Gini coefficient, each urban cluster should be ranked according to its average level of LUEE. Formulas (A1) and (A2) are shown in Appendix B. The Dagum Gini coefficient is decomposed into three components, including the contribution of intra-regional disparity $G_w$, inter-regional disparity $G_{nb},$ and hyper-variance density $G_t$. Here, $G_w$ denotes the distribution gap of LUEE within city group $j\left(h\right)$, $G_{nb}$ denotes the distribution gap of LUEE between city groups $j$ and $h$, and $G_t$ denotes the residual term of cross-influence of LUEE among the five major city groups. In measuring $G_w \mathrm{a}\mathrm{n}\mathrm{d} G_{nb}$, the Gini coefficients $G_{jj}$ within $j$ urban clusters and $G_{jh}$ between $j$ and $h$ urban clusters are calculated. Formulas (A3)–(A10) are shown in Appendix B.

3.2.3. The Kernel Density Estimation

In this section, the kernel density estimation was chosen to profile the dynamic evolution of LUEE in urban agglomerations. A widely adopted non-parametric estimation method with an emphasis on focusing on the actual data, it has been commonly applied to research inhomogeneous distributions of geospatial things [72,73,74,75]. We assume that $f\left(x\right)$ is the density function reflecting the LUEE in urban agglomerations and $\{x_1,x_2,\dots ,x_n\}$ is an independent identically distributed sample.

$$f\left(x\right)=\frac{1}{nh}\sum _{i=1}^{n}K\left(\frac{X_i-x}{h}\right)$$

(3)

where $K\left(·\right) \mathrm{r}\mathrm{e}\mathrm{p}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{s} \mathrm{t}\mathrm{h}\mathrm{e} \mathrm{k}\mathrm{e}\mathrm{r}\mathrm{n}\mathrm{e}\mathrm{l} \mathrm{f}\mathrm{u}\mathrm{n}\mathrm{c}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n},$ n$\mathrm{r}\mathrm{e}\mathrm{p}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{s} \mathrm{t}\mathrm{h}\mathrm{e} \mathrm{w}\mathrm{i}\mathrm{n}\mathrm{d}\mathrm{o}\mathrm{w} \mathrm{w}\mathrm{i}\mathrm{d}\mathrm{t}\mathrm{h}, \mathrm{a}\mathrm{n}\mathrm{d} h$ represents the number of samples. The Gauss kernel function, the most widely used, is given in Formula (4).

$$f\left(x\right)=\frac{1}{\sqrt{2\pi }}\mathit{exp}\left(-\frac{x^2}{2}\right)$$

(4)

3.2.4. Convergence Analysis

$\sigma$ convergence implies that the dispersion of LUEE in different regions decreases over time. The coefficient of variation was chosen to measure the dispersion of the data in this study [76,77] and is calculated as follows.

$$\sigma =\frac{\sqrt{{\sum }_i^{n_j}{\left(y_{ij}-{\overline{y}}_j\right)}^2/n_j}}{{\overline{y}}_j}$$

(5)

where $y_{ij}$ is the LUEE of city $i$ within city group $j$ in period $t$. $n_j$ represents the number of cities within city group $j$. ${\overline{y}}_j$ represents the average level of LUEE in city group $j$ in period $t$.

Absolute $\beta$ convergence assumes that the LUEEs of the different cities in the YRB have the same external conditions [78,79]. This study tests whether the LUEEs of different urban agglomerations converge to the same steady state. The formula is given below.

$$ln\left(\frac{y_{i,t+1}}{y_{i,t}}\right)=\alpha +\beta \mathit{ln}\left(y_{i,t}\right)+{ϵ}_{it}$$

(6)

In order to reduce the bias, and considering the role of spatial agglomeration or correlation in the convergence of the LUEE in urban agglomerations [80], spatial effects are included in the absolute $\beta$ convergence model.

$$ln\left(\frac{y_{i,t+1}}{y_{i,t}}\right)=\alpha +\beta \mathit{ln}\left(y_{i,t}\right)+\rho {\omega }_{ij}ln\left(\frac{y_{i,t+1}}{y_{i,t}}\right)+{ϵ}_{it}$$

(7)

where ${\omega }_{ij}$ is the spatial weight matrix and $y_{i,t}$ and $y_{i,t+1}$ denote the LUEE of urban agglomeration $i$ in periods $t$ and $t+1$. If $\beta <0$ and is significant, this implies that urban agglomerations with a low LUEE have a tendency to catch up with urban agglomerations with a higher LUEE.

Drawing on the relevant studies [81,82,83], we considered that the convergence of LUEE in urban agglomerations may be affected by other factors, so control variables ($X$) were introduced to construct a conditional $\beta$ convergence model. The formulas are as follows.

$$ln\left(\frac{y_{i,t+1}}{y_{i,t}}\right)=\alpha +\beta \mathit{ln}\left(y_{i,t}\right)+BX+{ϵ}_{it}$$

(8)

$$ln\left(\frac{y_{i,t+1}}{y_{i,t}}\right)=\alpha +\beta \mathit{ln}\left(y_{i,t}\right)+\rho {\omega }_{ij}ln\left(\frac{y_{i,t+1}}{y_{i,t}}\right)+BX+{ϵ}_{it}$$

(9)

3.3. Indicator System

Combining the connotation of LUEE [84,85], this study proposes an evaluation system of LUEE for urban agglomerations in terms of both input and output dimensions. As shown in

Table 2. Evaluation system of LUEE.

Elements	Primary Indicators	Secondary Indicators
Input indicators	Labor input	Ground average practitioners
	Energy input	Total land average water supply; Local average total social electricity consumption
	Natural factor input	Built-up area
Output indicators	Expected output	GDP per land
		Per capita disposable income of urban residents
	Unexpected output	Land average industrial wastewater discharge; Land average industrial soot emissions; Land average industrial SO2 emissions

, input indicators include the average number of employees, the average total water supply, the average total social electricity consumption, and the built-up area. The expected output indicator includes urban GDP per unit of land area and urban disposable income per capita. The non-expected output indicator includes industrial sulfur dioxide emissions, industrial soot emissions, and average and industrial wastewater emissions. In this study, we calculated the LUEE based on DEA-SOLVER Pro 15 software (v15).

Drawing on the relevant literature [85,86], this study introduces six control variables for the convergence analysis: the level of urban transportation infrastructure ($utf$), the level of urban public service facilities ($psf$), population density ($pd$), the level of information technology ($tech$), the degree of external openness ($open$), and government behavior ($gov$), as shown in

Table 3. Selection and measurement of variables.

Elements	Variable Name	Variable Representation
Explanatory variable	LUEE	Land use eco-efficiency
Control variable	utf	Road area per capita
	psf	Number of hospital beds per 10,000 people
	pd	Population density
	tech	Number of cell phones per 10,000 people at the end of the year
	open	Ratio of actual foreign capital utilization to GDP
	gov	Ratio of general budget fiscal expenditure to GDP

.

4. Results

4.1. Analysis of Measurement Results

4.1.1. Spatial Gap Analysis of LUEE in the YRB Urban Agglomeration

In Table 3, except for small fluctuations in individual years, the overall spatial disparity of LUEE in the YRB shows a decreasing trend during the period from 2009 to 2020, indicating that the spatial gap in LUEE of urban agglomerations in the YRB decreased over time. The sample period can be divided into two phases according to the characteristics of the changes in the values in the table. The first stage was from 2009 to 2012, during which the spatial gap decreased from a maximum of 0.158 in 2009 to 0.138 in 2012, with an average annual decrease of 3.3%. The second stage was from 2013 to 2020, during which the value shows a decreasing trend, with an average decrease of 5.4%. Compared with the first stage, the decrease in spatial gap in the second stage increased.

In

Table 4. Spatial Gini coefficient and its decomposition.

Year	G	Intra-Regional Gini Coefficient					Gw	Gn	Gt	Contribution Rate
		CP	LX	GP	SP	HE				Gw	Gn	Gt
2009	0.158	0.149	0.241	0.084	0.094	0.085	0.044	0.072	0.042	27.88	45.64	26.88
2010	0.152	0.154	0.207	0.086	0.100	0.110	0.046	0.057	0.049	30.55	37.37	32.08
2011	0.131	0.138	0.188	0.056	0.112	0.086	0.042	0.043	0.046	32.06	32.72	32.22
2012	0.138	0.143	0.204	0.054	0.102	0.068	0.042	0.057	0.039	30.50	41.28	28.22
2013	0.147	0.156	0.179	0.069	0.102	0.107	0.046	0.054	0.047	31.33	36.97	31.7
2014	0.140	0.145	0.168	0.082	0.109	0.085	0.044	0.053	0.043	31.07	38.00	30.94
2015	0.129	0.122	0.130	0.052	0.123	0.082	0.037	0.060	0.032	28.40	46.42	25.18
2016	0.121	0.133	0.096	0.063	0.111	0.068	0.040	0.044	0.037	33.15	36.11	30.74
2017	0.114	0.117	0.114	0.078	0.105	0.076	0.037	0.036	0.041	32.22	31.57	36.21
2018	0.103	0.102	0.136	0.060	0.080	0.088	0.032	0.033	0.038	31.12	31.95	36.93
2019	0.090	0.094	0.095	0.049	0.087	0.083	0.030	0.023	0.037	33.05	25.83	41.12
2020	0.094	0.092	0.118	0.044	0.109	0.083	0.030	0.023	0.042	31.37	24.42	44.21

, the LUEE of CP, LX, and GZ all showed decreasing trends. Unlike the characteristics of the above urban agglomerations, the LUEE of SP and HE did not change significantly over the sample period, indicating that the spatial disparity in LUEE of these two urban agglomerations did not improve significantly. From the mean values, it can be seen that the spatial differences are the largest in LX, followed by CP, and the smallest in GZ. Thus, it can be seen that the imbalance in LUEE between LX and CP is higher than that for other regions.

In addition, the factors leading to spatial differences in the LUEE of the YRB have the characteristics of stage changes, and comprise two stages. During the period from 2009 to 2016, the contribution rates of inter-regional disparity exceeded those of intra-regional disparity and hypervariable density, indicating that the main factor leading to the difference in LUEE is inter-regional disparity. The second stage was 2016 to 2020, during which hypervariable density became the main factor leading to differences in LUEE. Overall, the changes in the contribution rate of intra-regional disparity stabilized in the range of 27–34%. The contribution rate of inter-regional disparity generally exhibited a decreasing trend, from 45.64% in 2009 to 24.42% in 2020. The contribution rate of hypervariable density brought by inter-group overlap tended to increase, ranging from 26.88% in 2009 to 44.21% in 2020, which indicates that the influence of the cross-overlap component on the overall gap generation gradually became larger.

Table 5. Inter-regional Gini coefficient.

Year	Inter-Regional Gini Coefficient Gap
	CP-LX	CP-GP	CP-SP	CP-HE	LX-GP	LX-SP	LX-HE	GP-SP	GP-HE	SP-HE
2009	0.165	0.152	0.153	0.157	0.152	0.166	0.196	0.090	0.087	0.093
2010	0.163	0.147	0.151	0.159	0.141	0.155	0.185	0.093	0.098	0.108
2011	0.147	0.125	0.137	0.138	0.105	0.145	0.150	0.085	0.069	0.106
2012	0.154	0.134	0.142	0.146	0.109	0.145	0.152	0.080	0.063	0.096
2013	0.162	0.147	0.152	0.160	0.112	0.136	0.153	0.086	0.084	0.109
2014	0.151	0.140	0.145	0.146	0.116	0.136	0.138	0.096	0.084	0.103
2015	0.128	0.127	0.128	0.129	0.083	0.128	0.113	0.092	0.063	0.115
2016	0.134	0.125	0.132	0.130	0.074	0.109	0.086	0.087	0.066	0.101
2017	0.119	0.115	0.118	0.117	0.093	0.110	0.100	0.092	0.079	0.096
2018	0.108	0.100	0.101	0.105	0.090	0.105	0.119	0.071	0.069	0.085
2019	0.096	0.087	0.094	0.094	0.065	0.092	0.090	0.068	0.060	0.087
2020	0.097	0.087	0.097	0.093	0.071	0.114	0.105	0.080	0.057	0.103

shows that the spatial gap between urban agglomerations tended to decrease over the period. Except for the spatial gap between SP-HE and GP-SP, the gap in LUEE between the urban agglomerations gradually diminished over time.

4.1.2. Kernel Density Analysis of LUEE in Urban Agglomerations in the YRB

We found the relative trajectories of LUEE for urban agglomerations in the YRB but could not reveal the dynamic distribution characteristics of LUEE in each urban agglomeration. Accordingly, this study adopted the kernel density estimation method to depict the changes in LUEE for each region. MATLAB 2022b software was used to plot the kernel density estimation results shown in Figure 3. Regarding the location of the distribution, Figure 3 shows that, in general, the center of gravity of nuclear density distribution curves in the YRB shifted to the right, suggesting that the LUEE in the YRB improved over time. In terms of main peak distribution, it appears that the height of the kernel density curve increased and its width shrank for the urban agglomerations, suggesting that the absolute differences in LUEE among cities decreased during the sample period. However, it is noteworthy that the main peak of the kernel density curve in the Lanxi urban agglomeration shows a brief downward shift in height, implying that there was a greater pressure to improve LUEE in this region during the sample period. In terms of extensibility, there is a left-trailing phenomenon in the distribution curve of nuclear density in the YRB, which means that the LUEE of certain cities in the region is apparently lower than that of others in the same region. In terms of number of peaks, there are double peaks, and even triple peaks, in the YRB in the sample period. This indicates a polarization of land use eco-efficiency between cities. For example, the distance from the main peak to the side peaks of the Hubao–Egyu urban agglomeration is farther, indicating the existence of a relatively significant polarization phenomenon among cities.

The LUEE greatly spatially varies in different regions, and this spatial variation depends to some extent on urban industrial structure, land use patterns, and environmental protection measures. For this reason, this study conducted a convergence analysis of LUEE in the YRB so as to further examine its spatial and temporal evolution characteristics.

4.2. Spatial Convergence Analysis

4.2.1. σ Convergence Analysis

Using Equation (5), we calculated the variation coefficients of LUEE for the five urban agglomerations in the YRB from 2009 to 2020. In Figure 4, during the sample period, the variation coefficients of LUEE show an overall decreasing trend with σ convergence characteristics.

The Central Plains urban agglomeration (CP) maintained a gentle decrease during the sample period. The Lanxi urban agglomeration (LX) remained on a downward trend from 2009 to 2016, and then experienced a large fluctuation. The Hubao –Egyu urban agglomeration (HE) and the Guanzhong Plain urban agglomeration (GP) do not show a significant downward trend. The Shandong Peninsula urban agglomeration experienced less fluctuations in the early stage, with the phenomenon of a “tail” appearing after 2019. The LUEE of the YRB shows a clear gradient trend and regional imbalances. Except for HE and GP, the development of LUEE in all urban agglomerations tends towards σ convergence.

4.2.2. 𝛽 Convergence Analysis

The spatial lag model (SAR) and spatial error model (SEM) were used to study the convergence of the combined LUEE panel data for the five urban agglomerations in the YRB from 2009 to 2020. Estimates of the econometric regression model were calculated using Stata 17 software. If the results indicate the presence of spatial interactions in the model, a spatial econometric model that incorporates spatial effects should be constructed. In

Table 6. Results of the model selection test.

Models	(Robust) LM Test
	LM Lag	Robust LM Lag	LM Error	Robust LM Error
Absolute $\mathsf{\beta }$-convergence	28.94 *** (4.674)	3.54 * (1.860)	26.99 *** (4.506)	1.59 (1.324)
Conditional $\mathsf{\beta }$-convergence	65.92 *** (5.781)	68.66 *** (6.243)	12.14 *** (3.911)	14.89 *** (4.180)

Note: “*” and “***” refer to the 10% and 1% significance levels, respectively. t values in parentheses.

, the LM lag and LM error statistics both passed the 1% level of significance test. The value of the LM Lag statistic is larger than that of the LM Error statistic, and the robust LM Lag is statistically more significant. For this study, we selected the fixed-effects spatial lag model for convergence analysis.

The estimation results for absolute β convergence are presented in

Table 7. Estimation results of absolute $\beta$ convergence.

OLS	SAR
$\mathit{\beta }$	$\mathit{\beta }$	$\mathit{\rho }$
−0.075 ** (−2.01)	−0.064 (−1.38)	0.410 *** (3.79)

Note: “**” and “***” refer to the 5%, and 1% significance levels, respectively. t values in parentheses.

. The $\beta$ convergence coefficients are less than zero in both models. The coefficients do not pass the significance test in the spatial lag model, indicating that the overall absolute convergence trend of the LUEE for the five urban agglomerations in the YRB is not obvious. This indicates that the LUEE in the YRB urban agglomerations has no significant absolute β convergence. When possible influences such as city economic level, market environment, and infrastructure are omitted, the likelihood of the LUEE of individual cities in the YRB eventually converging to an identical stable level over time is very low. The spatial lag coefficient $\rho$ is significant, indicating that, in general, the convergence trend for LUEE in urban agglomerations has a positive spatial correlation.

Table 8. Estimation results of conditional $\beta$ convergence.

Variable	The Yellow River Basin
	OLS	SAR
$lny$	−0.101 ** (−2.42)	−0.150 *** (−3.25)
$utf$	0.120 *** (4.90)	0.555 *** (12.25)
$psf$	−0.001 ** (−2.31)	−0.000 (−0.02)
$pd$	−0.123 ** (−2.07)	0.003 (0.01)
$tech$	−0.037 (−1.60)	−0.247 *** (−6.43)
$open$	0.037 (1.59)	0.062 (1.58)
$gov$	0.112 ** (2.03)	0.079 (1.17)
$\rho$	-	0.313 *** (3.489)

Note: “**” and “***” represent the 5% and 1% significance levels, respectively. t values in parentheses.

presents the results for the conditional β convergence. As can be seen, the coefficients of β are significantly smaller than zero for both the general regression model and the spatial lag model, suggesting that, in general, the LUEE tends towards conditional convergence after the incorporation of control variables. Considering the control variables, the LUEE of each urban agglomeration gradually evolves over time towards the steady-state level of the respective region. Moreover, the coefficient $\rho$ in the spatial lag model is significantly greater than zero, suggesting that the spatial spillover effects contribute positively to the convergence of LUEE in the YRB. Among the control variables, the improvement of the urban transportation infrastructure level can promote the convergence of LUEE.

5. Discussion

5.1. Analysis of Spatio-Temporal Characteristics of LUEE in the YRB

Urbanization exerts a significant impact on LUEE. Measuring LUEE provides an effective quantitative approach to studying the relation between the ecological environment and land use. While existing studies have incorporated expected outputs in the evaluation of land use efficiency [23,24,25], which partially reflect the connection between land use and economic output, they have not integrated land use efficiency with ecological indicators that encompass non-expected outputs. In contrast to land use efficiency evaluations, the LUEE emphasizes the integrated balance between environmental and economic output benefits of land use [87]. Surprisingly, there is a scarcity of literature that considers undesirable outputs and examines the spatial and temporal aspects of LUEE in the YRB. This study systematically explored the concept of the “resource–economic–ecological” system in land use from the point of view of ecological green development. The EBM model was utilized to measure LUEE in 55 cities within the YRB, providing insights into the spatio-temporal evolution of LUEE. This approach enhanced our understanding regarding the relation between urban land use and ecology in a more comprehensive manner. The findings indicate that during the sample period, the spatial disparity in LUEE among urban agglomerations in the YRB decreased over time, aligning with the conclusions of Sun et al. [88].

The LX urban agglomeration exhibited the largest gap in LUEE, followed by CP, suggesting a significantly higher degree of LUEE imbalance between cities within these two urban agglomerations than in other regions. The factors contributing to the spatial differences in LUEE within the YRB demonstrated a phased pattern of change [3]. Before 2016, interregional disparities made a substantial contribution compared to other factors. However, after 2016, hyper-variable density became the primary factor driving differences in LUEE, indicating a gradual increase in the impact of overlapping factors on spatial variation. The core density curve of LUEE in the YRB shifted to the right, displaying a bimodal or even trimodal distribution. Over time, the LUEE in the YRB improved consistently, leading to a reduction in spatial disparities. The urban agglomerations within the YRB exhibit distinct variations in terms of resource endowment, geographical location, and development conditions [85,86]. Notably, the limited availability of land in the middle and upper reaches of the YRB exacerbates the conflict between ecological preservation and economic development, resulting in a multipolarization phenomenon in the LUEE within the urban agglomerations of the YRB.

5.2. Analysis of Spatial Convergence of LUEE in the YRB

There is a relatively limited amount of convergence analyses of LUEE in the literature. However, studying convergence is an essential aspect of land use research as it reflects whether there is a catch-up effect in LUEE among regional cities; it also reflects the robustness of LUEE spatio-temporal evolution characteristics [80]. In this study, it was observed that the spatial spillover effect contributed positively to the conditional convergence process of LUEE in the YRB. Cities with initially low LUEEs in the basin exhibited a “catch-up effect”, where in the LUEE of each city converges towards a stable state. This finding is similar to that of Song et al. [3] and different from that of Ruan et al. [87]. Unlike the axial development pattern observed in the Yangtze River Economic Belt [87], the various urban agglomerations along the YRB, influenced by factors such as location endowment, demonstrated room for improvement in terms of economic connectivity, quality of labor division and collaboration, and the level of coordinated progress. This study analyzed the convergence of urban LUEE in the YRB by constructing a convergence model. The findings can help identify shortcomings in urban development and eco-environmental protection within the YRB and provide reasonable solutions for improving urban LUEE [31,83]. It is noted that blindly implementing policies that aim to reduce LUEE differences between cities may ignore the diversity of land needs of different regions in terms of industry, ecology, population, and energy. In 2020, the nine provinces and regions along the Yellow River will account for 80 percent, 58 percent, 44 percent, and 52 percent of the national production of raw coal, coke, raw salt, and caustic soda products, respectively. The standardization of planning and development has potential negative impacts on the industrial structure of cities, the industrial base, and the lives of residents. Combined with the basic conditions and the carrying capacity of the resources and environment of the cities, the evaluation of the suitability of urban land space development in the YRB is a prerequisite for the implementation of planning policies.

5.3. Recommendations

To enhance land use eco-efficiency (LUEE) in the YRB, it is crucial to strengthen the synergy among urban clusters and promote the integration of elements across administrative regions. Firstly, cities should establish cross-regional cooperation organizations to unify the deployment of land use, industrial development, infrastructure construction, and ecological protection strategies. This will facilitate the establishment of efficient and intensive industrial chains and clusters, leading to an efficient land use pattern. Secondly, cities should accelerate the market-based reform of land resources and explore institutional, technological, and management innovations to enhance the efficiency of land resource allocation and stimulate economic benefits and ecological utilization efficiency in the YRB.

Given the heterogeneity of LUEE in different urban agglomerations, it is imperative to adopt tailored approaches to enhance LUEE. Consequently, a standardized set of policies cannot be universally applied across regions to effectively synchronize urbanization and the ecological efficiency of arable land use. First, in the upper reaches of the YRB, the imperative lies in fortifying the ecological service of land and enhancing the scale efficiency of land utilization. This entails the implementation of a protection-oriented point development mode, augmented investment in factors, promotion of novel technologies and concepts, judicious avoidance of ecologically sensitive and fragile areas, and further augmentation of scale efficiency in land utilization. Second, the middle reaches of the YRB predominantly revolve around resource-based industries, characterized by a significantly higher land urbanization rate than the rate of population urbanization. In this context, the efficiency of land use and the level of land marketization are at their nadir, consequently leading to a pronounced redundancy of land input factors. To address these challenges, optimizing the input structure of factors is paramount, with a focal point on the optimization of extant land resources. This encompasses bolstering the oversight of high-pollution emissions during land utilization processes and propelling the adjustment and upgrading of the energy industry structure through technological advancements and innovations. Third, the lower reaches of the YRB exhibit particularly high levels of economic development, thereby exacerbating the conflict between land development and the preservation of arable land, along with the general phenomenon of diminishing land scale compensation. In this regard, cities within the lower reaches ought to concentrate their efforts on agglomeration and intensive development. This involves augmenting the allocation efficiency of capital and labor across diverse industries and fostering the enhancement of comprehensive land efficiency through the augmentation of technical efficiency measures.

6. Conclusions

6.1. Conclusions

This study utilized the Super-EBM model to assess the LUEE of 55 cities within the Yellow River Basin. Moreover, the Dagum Gini coefficient and kernel density estimation techniques were employed to investigate the spatial variation characteristics and dynamic evolutionary patterns of LUEE, followed by an examination of spatial convergence.

During the sample period, with minor fluctuations in certain years, the LUEE in the YRB generally exhibited a decreasing trend. This trend suggests a reduction in the spatial disparity of LUEE among urban agglomerations over time. Furthermore, except for SP-HE and GP-SP, the gaps between the city groups narrowed progressively over time. The contribution rate of excess density resulting from intergroup overlap showed an overall increasing trend. Kernel density estimation results revealed that the center of gravity of the LUEE for the YRB urban agglomeration shifted to the right, with a reduced width and the presence of double or even triple peaks. These findings indicate an improvement in LUEE over time and a decrease in spatial disparities, albeit with discernible polarization. A conditional β convergence analysis, after introducing control variables, revealed a trend of conditional convergence in the YRB urban agglomeration. The spatial spillover effect made a positive contribution to the conditional convergence of LUEE in the YRB, suggesting that cities with a lower LUEE tend to experience a “catch-up effect” with cities with a higher LUEE.

6.2. Limitations

This study has certain limitations. The evaluation of LUEE in this study only includes specific non-desired output indicators such as land average industrial wastewater emissions, land average industrial soot emissions, and land average industrial SO₂ emissions. Data on pollutants such as nitrogen oxides, volatile organic compounds (VOCs), and particulate matter (PM) are significantly missing in most cities in the statistical yearbooks, so these pollutants were not taken into account in this study. Other important pollutants and environmental changes were not considered due to the availability of relevant data. In addition, the positive eco-efficiency of the land use process has not been included. Therefore, on one hand, with the help of text mining methods or big data technology to collect data on land pollution, nitrogen oxides, and other non-expected output indicators, such as the amount of media coverage of urban land pollution events and carbon emission data, will make up for the limitations of the lack of non-expected output indicators in this study, which is one of the directions of future research. On the other hand, searching for suitable indicators to measure the positive eco-benefits arising from the urban land use process and incorporating them into the system of desired output indicators is another one of the future research directions.