Threshold Effects of Water Use Efficiency in Urbanization and Industrial Growth

Abstract

Based on panel data from 14 prefectures in Xinjiang from 2004 to 2022, this study employs the Super-SBM model and panel threshold regression to assess how urbanization and industrial growth influence industrial water resource utilization efficiency (IWRUE). Xinjiang exhibits a distinct “high-north–low-south” spatial pattern: Urumqi and other northern regions show continuous improvement and Tacheng maintains long-term superiority, while southern areas such as Kizilsu and Hotan remain persistently low. Although IWRUE increases overall, regional trajectories diverge considerably. Two significant thresholds are identified—industrial output value and urbanization rate. Below these thresholds, water consumption strongly suppresses IWRUE, industrial employment exerts a negative effect, and investment plays a positive role. Once the thresholds are exceeded, the negative effect of water consumption weakens, industrial employment turns positive, and investment becomes insignificant. Policy implications suggest that regions below the thresholds should strengthen investment in water-saving technologies and productive capital, whereas regions beyond the thresholds should focus on enhancing labor quality, promoting green innovation and improving refined management to stabilize IWRUE and foster coordinated regional development.

1. Introduction

Urbanization and industrialization are the core driving forces of regional development [1,2]. In resource-dependent arid regions, the structural contradiction between industrial expansion and rigid water resource constraints has become increasingly pronounced [3,4]. As a core node of the Silk Road Economic Belt [5], Xinjiang faces dual challenges of an extremely arid climate and leapfrog industrial development. Under such circumstances, reliance solely on increasing factor inputs is no longer sufficient to sustain industrial growth. Improving industrial water resource utilization efficiency (IWRUE) has therefore become critical to resolving the regional “water–production” dilemma. Previous studies have provided multidimensional perspectives for understanding the impacts of urbanization and industrialization on water resources. At the macro level, research on the spatial structure and determinants of water use efficiency in China indicates IWRUE exhibits significant spatial heterogeneity due to differences in resource endowments and industrial structures [6], implying the necessity of differentiated policy strategies across regions. At the urban scale, profound influence of urbanization was revealed on the water–energy–carbon nexus [7], suggesting that urbanization affects resource use efficiency through changes in consumption patterns and system coupling relationships. However, although these studies address spatial heterogeneity and system coupling, whether urbanization and industrial growth generate nonlinear “structural turning points” in IWRUE under the extremely stringent arid constraints of resource-based regions such as Xinjiang remains insufficiently explored. Compared with humid regions or national averages, Xinjiang’s industrialization process is more vulnerable to rigid water constraints, which may give rise to distinct threshold characteristics in IWRUE.

Extensive research has been conducted on the measurement of IWRUE, employing methods such as EIO-LCA [8], SDA [9], and the Cobb–Douglas production function [10,11]. Although these approaches provide valuable insights, they are largely based on linear or static assumptions [12,13], which limit their capacity to capture “structural breakpoints” during the industrialization process. Theoretically, the effects of industrial growth and urbanization on IWRUE are unlikely to follow a monotonic linear pattern [14,15]. In the early stages of development, scale expansion is often accompanied by extensive resource use (crowding effects), whereas once economic development surpasses a critical threshold, technological spillovers and intensive management begin to dominate (economies of scale) [16]. This nonlinear transition implies the existence of threshold effects. Therefore, the application of the Super-SBM model, which accommodates undesirable outputs and combines with a panel threshold model capable of endogenously identifying turning points, provides an important theoretical basis for revealing the dynamic evolution of water use efficiency in Xinjiang [17].

The novelty of this study is reflected in three aspects. First, in terms of research perspective, it focuses on the stage-specific characteristics of leapfrog development in arid regions and examines how IWRUE breaks out of a “low-level trap.” Second, in terms of methodology, it integrates super-efficiency measurement with nonlinear threshold analysis to identify directional changes in driving factors across different intervals. Third, in terms of policy implications, it provides a quantitative basis for differentiating water-saving pathways between northern and southern Xinjiang based on identified threshold values.

Accordingly, this study seeks to address the following scientific questions: (1) Does the growth of industrial output consistently inhibit IWRUE? (2) To what extent can the process of urbanization promote industrial water conservation and emission reduction? (3) Do the driving effects of key factors such as labor and investment on IWRUE differ fundamentally across threshold intervals? Based on these questions, we propose the following hypotheses: Hypothesis H1 (Scale Effect Hypothesis): the impact of industrial output on IWRUE exhibits nonlinear characteristics, and only after surpassing a specific output threshold does the marginal efficiency of factor inputs shift from negative to positive; Hypothesis H2 (Urbanization Dividend Hypothesis): the promoting effect of urbanization on IWRUE is subject to threshold effects and requires a certain level of urbanization before the benefits of intensive management and environmental regulation can offset the water pressure caused by population agglomeration; Hypothesis H3 (Factor Heterogeneity Hypothesis): across different threshold intervals, the driving directions of labor and investment on IWRUE change significantly.

Bordering Central Asia and Russia [18], the Xinjiang Uygur Autonomous Region serves as a key node along the Silk Road Economic Belt [19]. Since the implementation of the 10th Five-Year Plan in 2001 [19], Xinjiang’s industrial sector has expanded rapidly with steady GDP growth. In 2023, China’s Xinjiang Pilot Free Trade Zone covers an area of 179.66 square kilometers, encompassing three distinct zones, including the Urumqi zone of 134.6 square kilometers, the Kashgar zone of 28.48 square kilometers, and the Khorgos zone of 16.58 square kilometers. Economic transformation and industrial restructuring continue to advance, with the dual drivers of industrialization and urbanization posing severe challenges to water utilization and environmental sustainability. However, limited research is available on the impact of urbanization and industrialization on IWRUE in this region. This study analyzed the current status of industrial water utilization efficiency in Xinjiang by examining the threshold effects of urbanization and industrial growth and identifying the key factors influencing IWRUE. The aim was to provide scientific references to optimize regional economic development and promote sustainability.

2. Study Area and Its Characteristics

The Xinjiang Uygur Autonomous Region is located in China’s northwestern frontier, the hinterland of the Eurasian continent, and constitutes the largest provincial-level administrative region in China by land area. The region comprises 14 prefecture-level administrative units, including Urumqi, Changji Prefecture, Ili Prefecture, Kashgar Prefecture, and Hotan Prefecture (Figure 1a). Situated in the core area of the Belt and Road Initiative and bordering eight neighboring countries, Xinjiang occupies a strategically significant geographical position. Figure 1b shows that Xinjiang exhibits a typical geomorphological pattern described as “three mountain ranges enclosing two basins.” The Altai Mountains in the north, the Tianshan Mountains in the center, and the Kunlun Mountains in the south form the structural framework of the region, surrounding the Junggar Basin in the north and the Tarim Basin in the south. The region displays diverse landforms, dominated by high mountain systems, piedmont plains, plateaus, and extensive aeolian landscapes (deserts) in the basin centers. Glaciers and lakes are mainly distributed in high-altitude mountainous areas, functioning as natural “water towers” that sustain the ecological security of oases in arid zones. The land use/land cover (LULC) pattern is strongly constrained by the arid climate and the spatial distribution of water resources (Figure 1c). Due to the extremely arid environment, unused land occupies the majority of Xinjiang’s total area. Cultivated land and construction land exhibit distinct “water-oriented” and “oasis-dependent” characteristics, which are highly concentrated in piedmont alluvial plains and river valleys, such as the Tarim River and Ili River basins. Grasslands are mainly distributed along the slopes of the Tianshan and Altai Mountains, while forest land is relatively scarce and primarily appears in strip-like formations in mountainous areas or along riverbanks in arid regions.

3. Materials and Methods

3.1. Variable Selection and Data Sources

Data Sources

For IWRUE measurement, this paper comprehensively considered Xinjiang’s geographical characteristics, water resource endowment status, and the relevant literature. Ultimately, 14 prefecture-level cities in Xinjiang were selected and data from 2004 to 2022 were collected to construct an SBM efficiency evaluation system comprising three types of indicators, i.e., inputs, outputs, and environmental constraints. Regarding data sources, all variables primarily originated from authoritative statistical departments such as the Xinjiang Statistical Yearbook, the Xinjiang Department of Water Resources, and the Department of Ecology and Environment. For variables with missing values in certain years, data gaps were supplemented by consulting the Xinjiang Water Resources Bulletin. Linear interpolation was applied to smooth individual missing data points, ensuring the integrity and continuity of the data series.

Table 1. Variable description of SBM model.

Variable Type	Specific Indicator	Unit	Description
Input variables	Industrial water consumption	100 million cubic meters	Refers to the total water used by industrial enterprises during production, measuring the intensity of natural water resource consumption
	Industrial employees	Persons	The number of workers engaged in industrial production, reflecting the supporting role of labor in resource conversion efficiency
	Industrial construction investment	10,000 yuan	Investment in industrial fixed assets, representing capital input that influences technological progress and improvements in water-use processes
Output variables	Total industrial output value	10,000 yuan	The total value of final industrial products and industrial service activities produced by enterprises during the reporting period
	Ecological impact	100 million tons	Environmental disutility (wastewater discharge) generated alongside industrial output, reflecting pressure on the ecological environment
Social and population data	Population size	10,000 persons	Total resident population of the region, representing the macro-social background influencing resource allocation efficiency and environmental carrying capacity
	Urbanization rate	%	The proportion of urban population in the total population, reflecting industrial upgrading and agglomeration effects

systematically categorizes and defines the aforementioned input, output, and environmental impact variables.

This study selected total industrial output value and urbanization rate as threshold variables. This selection was based not only on empirical observation but also on strong theoretical foundations. The final explanatory variable system is presented in

Table 2. Explanation of factors affecting IWRUE.

Variable Type	Specific Indicator	Indicator Explanation	Unit
Industrial	IWRUE	Water consumption per unit of output during production	/
	Industrial water consumption	Water consumption during industrial production	108 m3
	Industrial workers	Persons engaged in industrial production	persons
	Industrial Investment	Amount invested in the industrial sector	104 Yuan
	Industrial output value	Industrial output value	104 Yuan
Social and Demographic Data	Population	Population of the entire region	104 persons
	Ecological impact	Wastewater discharged during production	108 t
	Urbanization rate	Urban population as a proportion of the total population of the region	%

Note: The inclusion of industrial wastewater discharge in the analytical framework follows a dual rationale. First, in the efficiency measurement stage, it is treated as an undesirable output to construct a “green water use efficiency” indicator, thereby avoiding overestimation of IWRUE caused by ignoring environmental costs. Second, in the regression analysis stage, it is treated as an ecological constraint indicator to capture the forcing mechanism of environmental regulation on technological improvement.

.

(1)

The theoretical basis for total industrial output value originated from agglomeration economies theory. Theoretically, the influence of industrial scale on IWRUE is expected to exhibit nonlinear characteristics. At early stages, when industrial scale is relatively small, industrial expansion may generate a “crowding effect” due to inadequate infrastructure, thereby inhibiting IWRUE. After surpassing a certain threshold, output growth may trigger scale effects, making research, development, and large-scale application of advanced water-saving technologies economically feasible, thereby improving IWRUE.

(2)

The theoretical basis for urbanization rate was derived from the Environmental Kuznets Curve (EKC) and structural transformation theory. Urbanization represents not only population agglomeration but also the modernization of management systems and technological innovation. Theoretically, only when urbanization reaches a certain critical level can the benefits of centralized wastewater treatment facilities, refined water resource management systems, and the regulatory “forcing effect” of environmental policies effectively offset the resource consumption pressures that arise in the early stages of urbanization.

To clarify the formation and influencing mechanisms of IWRUE, this study constructed a conceptual framework (Figure 2). The framework illustrates the process through which industrial factor inputs are transformed into economic outputs within a production system under ecological environmental constraints. Urbanization and the industrial growth scale serve as key exogenous regulatory factors, and through threshold effects, they determine the critical pathway by which factor allocation shifts from “resource dependence” to “intensive-driven” development.

3.2. Model Selection

3.2.1. Super-SBM Efficiency Measure Model

Data Envelopment Analysis (DEA) is a commonly used method for measuring efficiency. However, traditional DEA models cannot calculate efficiency values for fully efficient units. Furthermore, traditional DEA models are based on radial methods and do not account for slack variables in inputs and outputs.

In the super-efficient SBM model, total industrial output value is treated as a desirable output, while industrial wastewater discharge is treated as an undesirable output. The Super-SBM model is non-radial and non-angular, enabling it to directly handle negative externalities such as wastewater discharge. Importantly, it removes the upper limit of 1 for efficiency scores, allowing for a more refined ranking and comparison of IWRUE among multiple prefectures operating on the production frontier [20,21].

This study adopted the super-efficient SBM model as a theoretical tool to measure regional water resource utilization performance. Briefly, for the evaluation of n DMUs, each unit’s input vector was denoted as x = (x₁, x₂, …, x_m) and output vector as y = (y₁, y₂, …, y_s). During evaluation, the DMU being evaluated was excluded from the sample set when being evaluated to prevent its self-influence on the construction of the reference boundary. The reference set was then formed by the remaining n − 1 DMUs.

The super-efficient SBM model constructed the following objective functions based on this framework, jointly characterizing efficiency through input redundancy and output deficiency:

$$\min \mathsf{\theta } = {\displaystyle \frac{1-\frac{1}{\mathrm{m}}{\sum }_{\mathrm{j}=1}^{\mathrm{m}} \frac{{\mathrm{s}}_{\mathrm{j}}^{-}}{{\mathrm{x}}_{\mathrm{jo}}}}{1 + \frac{1}{\mathrm{s}}{\sum }_{\mathrm{r}=1}^{\mathrm{s}} \frac{{\mathrm{s}}_{\mathrm{r}}^{+}}{{\mathrm{y}}_{\mathrm{ro}}}}}$$

(1)

where ${\mathrm{s}}_{\mathrm{j}}^{-}$ denotes the slack variable for input j, representing input redundancy; ${\mathrm{s}}_{\mathrm{r}}^{+}$ denotes the slack variable for the r-th output item, indicating underproduction; and ${\mathrm{x}}_{\mathrm{jo}}$ and ${\mathrm{y}}_{\mathrm{ro}}$ represent the jth input and the r-th output of the o-th DMU, respectively. The model constraints are as follows:

$$\begin{array}{rr}& {\displaystyle \sum _{\mathrm{i}\ne \mathrm{o}}^{\mathrm{n}}} {\mathsf{\lambda }}_{\mathrm{i}}{\mathrm{x}}_{\mathrm{ji}} + {\mathrm{s}}_{\mathrm{j}}^{-} = {\mathrm{x}}_{\mathrm{jo}}, \mathrm{j} = 1,2,\dots ,\mathrm{m}\\ & {\displaystyle \sum _{\mathrm{i}\ne \mathrm{o}}^{\mathrm{n}}} {\mathsf{\lambda }}_{\mathrm{i}}{\mathrm{y}}_{\mathrm{ri}}-{\mathrm{s}}_{\mathrm{r}}^{+} = {\mathrm{y}}_{\mathrm{ro}}, \mathrm{r} = 1,2,\dots ,\mathrm{s}\\ & {\mathsf{\lambda }}_{\mathrm{i}} \ge 0, \forall \mathrm{i} \ne \mathrm{o}; {\mathrm{s}}_{\mathrm{j}}^{-},{\mathrm{s}}_{\mathrm{r}}^{+} \ge 0\end{array}$$

(2)

The weight ${\mathsf{\lambda }}_{\mathrm{i}}$ represents the weight of the i-th DMU in the reference set, reflecting its contribution to the construction of the optimal reference combination. The efficiency output by the model typically falls within the range of 0 to 1, with values closer to 1 indicating higher efficiency. If the value exceeds 1, it signifies that the decision unit possesses a relative competitive advantage among its peers.

3.2.2. Threshold Regression Analysis Model

The effects of industrialization and urbanization on water resources are not characterized by a simple linear relationship [22]. Traditional linear regression models, such as Ordinary Least Squares (OLS), essentially capture only conditional mean effects [23], which often obscure differences across development stages. Moreover, OLS assumes parameter stability and is therefore unsuitable for identifying structural breaks or threshold effects that commonly arise during economic transformation processes [24,25]. Unlike conventional exogenous partitioning methods, the threshold model proposed by Hansen (1999) [24] adopts a data-driven approach to endogenously identify precise threshold values for variables such as industrial output and urbanization rate. This method objectively detects structural changes in the relationship between economic growth and IWRUE [26], ensuring that identified “critical points” are statistically rigorous rather than arbitrarily determined and providing a scientific basis for identifying the transitional pathway of Xinjiang’s IWRUE from “extensive growth” to “intensive-driven” development. The basic form of the single-threshold regression model can be expressed as

$${\mathrm{Efficiency}}_{\mathrm{it}}={\mathsf{\mu }}_{\mathrm{i}}+{ \mathsf{\gamma }}_{\mathrm{t}}+\{\begin{array}{ll}{\mathsf{\beta }}_1{\mathrm{X}}_{\mathrm{it}}+{\mathsf{ϵ}}_{\mathrm{it}},& {\mathrm{q}}_{\mathrm{it}} \le \mathsf{\tau }\\ {\mathsf{\beta }}_2{\mathrm{X}}_{\mathrm{it}}+{\mathsf{ϵ}}_{\mathrm{it}},& {\mathrm{q}}_{\mathrm{it}} > \mathsf{\tau }\end{array}$$

(3)

${\mathrm{Efficiency}}_{\mathrm{it}}$ represents the IWRUE of the i-th prefecture-level city in year t, which is calculated using the SBM model; ${\mathrm{X}}_{\mathrm{it}}$ represents the core explanatory variables (industrial output value and urbanization rate); ${\mathrm{q}}_{\mathrm{it}}$ is the threshold variable (of the same origin as ${\mathrm{X}}_{\mathrm{it}}$ or selected from it); $\tau$ is the threshold value to be determined; and ${\mathsf{\mu }}_{\mathrm{i}}$ and ${\mathsf{\gamma }}_{\mathrm{t}}$ are the region- and time-fixed control effects respectively. Prior to the regression analysis, we conducted a preliminary test for spatial autocorrelation. Throughout most of the study period, the Global Moran’s I results for IWRUE across the 14 prefectures were not statistically significant (p > 0.10). This weak spatial dependence may be ascribed to Xinjiang’s unique “oasis economy” structure, where industrial hubs are geographically partitioned by vast deserts and mountain ranges, thereby constraining direct spatial spillovers. Consequently, a non-spatial panel threshold model is well-suited for identifying these nonlinear structures.

Compared with methods commonly applied in water resource studies, such as EIO-LCA [8] and the Cobb–Douglas production function [10,11], the Super-SBM model [27,28] better balances economic benefits and environmental costs under multiple input variables, while the panel threshold model [24,26] effectively overcomes the limitations of linear models in explaining leapfrog development stages.

From a research logic perspective, this study constructs a two-stage analytical framework described as “measurement–diagnosis.” In the first stage, the Super-SBM model functions as a measurement tool. Its core role is to integrate multiple industrial inputs (water, labor, and capital) and dual outputs (desirable industrial output and undesirable wastewater discharge) into a standardized efficiency score, thereby addressing the measurement issue of IWRUE. In the second stage, the threshold regression model functions as a diagnostic tool. It takes the efficiency scores generated in the first stage as the dependent variable and aims to diagnose, through endogenous identification, the levels of urbanization and industrial growth at which qualitative changes in IWRUE occur. This coordinated analytical approach avoids the one-sidedness of traditional linear analysis and provides a more intuitive understanding of the stage-specific evolution of IWRUE.

Given Xinjiang’s vast geographic expanse and the latent economic ties between its regions, spatial econometric approaches are theoretically applicable. However, preliminary diagnostics indicated weak spatial autocorrelation for IWRUE during the observed period. Since our primary objective was to pinpoint nonlinear structural shifts across different development stages, we utilized the panel threshold model for its superior robustness in capturing phase-specific effects. Future research could incorporate spatial weight matrices to further examine how geographic proximity moderates the accelerating economic integration within the region.

Before conducting empirical estimation, all original variables were transformed using natural logarithms. The primary purposes of the logarithmic transformation were to (1) alleviate heteroscedasticity among variables and make data distributions closer to normality and (2) eliminate dimensional inconsistencies caused by differences in measurement units (such as “10,000 persons,” “100 million yuan,” and “100 million m³”). (Note: Detailed descriptive statistics of all prefectures, including means, standard deviations, and extreme values, are provided in Appendix A,

Table A1. Descriptive statistical table of industrial raw data.

	Observation Value	Mean Value	Standard Deviation	Minimum Value	Maximum Value
Industrial efficiency	266	0.574	1.780	0.100	1.318
Industrial water consumption	266	0.488	2.570	0.050	2.718
Industrial employees	266	23,270	3.028	403.430	162,755
Industrial investment	266	904,300	5.018	2981	8,886,000
Total industrial output value	266	1,693,000	4.889	8103	24,150,000
Population	266	115	2.245	20.090	1097
Ecological impact	266	0.085	4.302	0.000336	1.000
Urbanization rate	266	1.701	1.229	1.234	2.718

).

4. Results and Analysis

4.1. Trends in IWRUE Across Regions

Between 2004 and 2022, IWRUE in Xinjiang exhibited pronounced spatial differentiation and stage-based fluctuations, as depicted in Figure 3. From a spatial perspective, the region consistently displayed a clear “higher in the north and lower in the south” pattern, primarily attributable to the stronger industrial base and relatively advanced water-saving technological spillovers in northern Xinjiang (e.g., Urumqi, Changji, and Tacheng) [17].

From a temporal evolution perspective, fluctuations in IWRUE across Xinjiang closely corresponded with national and regional water resource management policies. The evolution of IWRUE in Xinjiang captured a complex interplay between industrial expansion and the tightening of water resource management. Prior to 2012, efficiency experienced a steady ascent, propelled by the implementation of the “Most Stringent Water Resources Management System” (the “Three Red Lines” policy), which accelerated the elimination of obsolete production capacities. However, the period of 2013 to 2020 saw marked volatility. For instance, the Ili region experienced efficiency fluctuations due to the expansion of water-intensive coal chemical industries, while southern prefectures like Kashgar achieved efficiency gains through the development of modern industrial parks under the Belt and Road Initiative. Despite these advancements, regions such as Kizilsu and Hotan consistently struggled with low efficiency scores (below 0.30), signaling a persistent “efficiency trap” characterized by distorted factor allocation and a dearth of advanced water-saving technologies.

Overall, the spatiotemporal variation in Xinjiang’s IWRUE is not random but results from the combined effects of natural resource endowments, industrial structural adjustments, and the implementation of the strictest water resource management policies. Mere output growth, if it does not cross the threshold of technological transformation, is often accompanied by temporary declines in IWRUE.

4.2. Descriptive Statistics and Correlation Analysis

Xinjiang’s industrial system is acutely sensitive to water availability given its arid “oasis economy” structure. As depicted in Figure 4, precipitation during the study period was characterized by both low volume and high volatility (exemplified by the severe drought in 2019). Such climatic swings directly constrained initial water allocation and exerted periodic pressure on industrial water supplies. This 19-year dataset accounts for this environmental variability, ensuring that the measured IWRUE reflects enduring structural patterns rather than transient climatic anomalies.

In Figure 5 (vertical axis: dimensionless relative values after natural logarithmic transformation), the “dimensionless values” represent relative levels following logarithmic processing, providing a unified baseline for observing the distribution characteristics and dispersion of variables. Negative values do not indicate negative production efficiency; rather, they result from original scores below 1 after logarithmic transformation and indicate that the region has not yet reached the production frontier and remains in an inefficient state. Larger the absolute value of a negative number corresponds to greater the gap between the prefecture and the optimal water resource allocation target and more severe degrees of resource misallocation. The average IWRUE for Xinjiang stands at −0.556, with negative values indicating that most regions have yet to reach the production frontier. While industrial water consumption exhibits significant spatial heterogeneity, other indicators—such as industrial employment, investment, and output—remain relatively stable across the region. These statistical distributions serve as a baseline for identifying the effects of environmental constraints and population agglomeration in the subsequent regression analysis.

Correlation analysis (Figure 6) reveals potential statistical associations among variables. It should be emphasized that these statistical associations do not imply causal driving mechanisms; rigorous causal identification will be conducted in the subsequent threshold regression analysis. The following observations are made based on the correlation matrix. Association between IWRUE and output-related factors: IWRUE is significantly positively correlated with total industrial output value (r = 0.48; p < 0.001), industrial employment (r = 0.44; p < 0.001), and industrial investment (r = 0.25; p < 0.001). This suggests that industrial expansion and increased factor inputs are often accompanied by synchronous changes in water resource utilization performance. Synergy between urbanization and the industrial system: urbanization rate is significantly positively correlated with IWRUE (r = 0.37), industrial output value (r = 0.37), and industrial employment (r = 0.50) at the 0.001 significance level. This reflects the strong spatial consistency between urbanization and industrialization processes and suggests the potential influence of population agglomeration on industrial production patterns. Ecological environmental pressure: ecological impact (wastewater discharge) shows strong positive correlations with industrial output value (r = 0.53) and industrial employment (r = 0.51), reflecting the environmental burden accompanying industrial expansion.

4.3. Threshold Regression Model Estimation

4.3.1. Preliminary Regression Analysis

As depicted in Figure 7, the coefficients for industrial output, investment, number of employees, and water consumption were all positive (0.0625, 0.0315, 0.0111, and 0.0090, respectively). Regional development variables exhibited positive interdependence with 0.586 for the coefficient of urbanization rate, 0.123 for the coefficient of population, and −0.014 for the coefficient of ecological impact.

4.3.2. Testing and Determination of the Threshold Effect

Through threshold tests of industrial output value and urbanization rate, this study identified two key structural turning points (see

Table 3. Summary of threshold regression results for IWRUE.

Threshold Variable	Driving Factor	Interval I (Below Threshold)	Interval II (Above Threshold)	Change in Marginal Effect (Δ)
Industrial Output Value	Industrial water consumption	−0.110 (0.0644) *	−0.0378 (0.0563) (not significant)	0.0722
	Employment	−0.134 (0.0581) **	0.145 (0.0565) **	0.279
	Industrial investment	0.214 (0.0368) ***	−0.0129 (0.0292) (not significant)	−0.2269
	Population	0.121 (0.0630) *		—
	Ecological impact	−0.00087975 (not significant)		—
	Urbanization rate	0.522 (0.324) (not significant)		—
Urbanization Rate	Industrial water consumption	−0.316 (0.135) **	−0.0055284 (not significant)	0.214
	Employment	−0.161 (0.222) (not significant)	0.0876 (0.0519) *	0.2486
	Industrial investment	1.871 (0.404) ***	0.00872 (0.0262) (not significant)	−1.8623
	Industrial output value	−1.777 (0.548) ***	0.0897 (0.0546) (not significant)	1.8667
	Population	0.140 (0.0608) **		—
	Ecological impact	0.00226 (0.0233) (not significant)		—

* Note: Values in parentheses are standard errors. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. The change in marginal effect (Δ) reflects the direction and magnitude of changes in factor influence across development stages.

). The results indicate that the effects of various driving factors on IWRUE exhibit clear stage-specific characteristics.

(1)

Analysis of the industrial output threshold effect: When industrial output is below 13.234, the elasticities of industrial water consumption and industrial employment are negative (−0.110 and −0.134, respectively). This reflects that, in the early stage of industrialization, economic growth is characterized by extensive development [27], in which output expansion is primarily driven by large-scale accumulation of resource inputs [28]. During this period, the absence of economies of scale and advanced management means with increased output are often offset by rapid growth in resource consumption, thereby reducing IWRUE [29,30]. However, once this threshold is crossed, a qualitative shift occurs. The inhibitory effect of water consumption weakens significantly and loses statistical significance, while the effect of employment changes from negative to positive (r = 0.145). This indicates that, with industrial expansion, the “human capital dividend” and “technology spillover effect” begin to emerge. High-quality labor enhances IWRUE by optimizing production processes and applying water-saving technologies. As the industrial scale expands, the emergence of human capital dividends and technological spillovers becomes a key driver of IWRUE improvement [31,32]. High-quality labor facilitates process optimization and the rapid adoption of water-saving technologies, enabling enterprises to shift from extensive resource consumption toward intensive management [33,34].

(2)

Analysis of the urbanization rate threshold effect: As a typical representative of China’s arid northwest, Xinjiang has experienced leapfrog urbanization and industrialization over the past two decades. The urbanization threshold identified in this study (0.2301) is relatively low, mainly due to the unique starting point of oasis city development in arid regions and the imbalance between early industrialization and urbanization. This low threshold clearly captures the sensitivity of IWRUE to the early stages of urbanization in arid regions.

When the urbanization rate stays below the 23.01% mark, there is a significant negative correlation between industrial output and industrial water use efficiency (IWRUE), reflected in an elasticity coefficient of −1.777. While the sheer scale of this number is significant, it actually captures the sensitivity of water resource systems to industrial expansion during Xinjiang’s early stages of urbanization. This stems from the fact that industrialization at this embryonic stage is typically characterized by extensive growth; here, even a modest uptick in production scale can trigger a disproportionate spike in water consumption relative to a low initial efficiency baseline [35]. During this period, expansion of the production scale is accompanied by disproportionate increases in water consumption, producing severe resource crowding effects on fragile local ecosystems [7]. At this stage, the negative impact of industrial water consumption is most pronounced (r = −0.316), while industrial investment shows a strong marginal contribution (r = 1.871). This suggests that, during the initial phase of urbanization, capital deepening and the construction of water-saving infrastructure constitute the primary forces for alleviating rigid water constraints and improving management efficiency [36,37,38]. Once this threshold is crossed, the marginal effect of investment declines significantly, indicating that, as urban scale expands in arid regions, reliance solely on capital investment enters a stage of diminishing marginal returns. This transition signals a fundamental shift in the driving logic of IWRUE in Xinjiang. At higher levels of urbanization, improvements in IWRUE no longer depend solely on output expansion or capital accumulation but increasingly rely on refined urban management and the green transformation of industrial structures [39,40].

In both models, the coefficients for employment are positive (+0.279 and +0.2486). This indicates that, whether measured by industrial output or urbanization level, “crossing the threshold” is a prerequisite for realizing the human capital dividend.

4.3.3. Model Robustness and Technical Notes

To address the relatively large elasticity coefficients observed in the model (e.g., output coefficient −1.777), this study conducted targeted technical evaluations.

• Interpretation of elasticity coefficients and their magnitude: Within the double-logarithmic framework, a coefficient of −1.777 indicates that a 1% increase in industrial output corresponds to a 1.777% decline in IWRUE. From an economic perspective, such a high elasticity requires a nuanced interpretation. It likely captures the “low-base effect” characteristic of Xinjiang’s early industrial development, where even modest absolute shifts in output can trigger disproportionate percentage swings in efficiency scores. Moreover, in threshold modeling, coefficients of this magnitude often emerge near “structural breakpoints,” representing the volatile fluctuations inherent in the transition from resource-dependent extensive growth to more intensive modes. Consequently, this value should be viewed more as a barometer of structural imbalance during a specific phase than as a constant, long-term elasticity.
• Multicollinearity test: Variance inflation factor (VIF) testing (

  Table 4. Multicollinearity test of industrial indicators.

  Variable	VIF	1/VIF
  Industrial employment	6.04	0.16546
  Industrial output value	5.8	0.172378
  Industrial investment	3.68	0.271593
  Industrial water consumption	3.41	0.29335
  Urbanization rate	2.94	0.339621
  Population	2.16	0.462421
  Ecological impact	1.57	0.6381

  ) shows that, except for industrial employment and industrial output value, the average VIF values of all variables are below 5, thereby excluding serious multicollinearity concerns.
• Stability: Robustness tests (see Figure 8) confirm that, after removing extreme values and adjusting control variables, the core thresholds and coefficient signs remain highly consistent, indicating strong statistical support for the model results.

4.3.4. Discussion of Endogeneity and Robustness

Considering the potential bidirectional causality among industrial growth, urbanization level, and IWRUE, this study adopted the following measures to mitigate endogeneity bias. First, panel fixed effects were employed to effectively control for time-invariant omitted variables across prefectures and cities, such as geographic characteristics and resource endowments. Second, the Hansen threshold regression method used in this study determines threshold values through a grid-search procedure, which provides strong robustness in identifying nonlinear relationships among variables. Finally, in the robustness tests, explanatory variables were lagged by one period. The results show that the core conclusions and threshold intervals remain consistent, indicating that the findings are minimally affected by endogeneity arising from reverse causality and that the model identification results are highly reliable.

5. Discussion

5.1. Internal Mechanisms of the Threshold Effect on IWRUE

This study finds that IWRUE in Xinjiang does not increase linearly with urbanization and industrial growth but instead exhibits significant stage-specific transitions. The results reveal a “threshold logic” in water resource allocation in arid regions: before industrial scale and urbanization rate exceed the identified thresholds (13.234 and 0.2301), the system shows clear characteristics of factor misallocation. At this stage, industrial expansion relies heavily on extensive inputs of water resources and low-skilled labor, resulting in marginal output contributions that fail to offset resource consumption and thus suppress efficiency. The significant negative elasticity observed within the sub-threshold interval underscores the inherent fragility of the “water–industry” synergy across Xinjiang’s oases. It is crucial to note, however, that these high-intensity effects are distinctly phase-specific; as the region advances toward the identified threshold, the system’s sensitivity to output expansion gradually stabilizes, thereby facilitating a transition toward a more resilient, efficiency-driven paradigm. Once the threshold is crossed, the system enters a stage characterized by economies of scale and technological spillovers. Improvements in labor quality begin to offset the negative effects of resource consumption, enabling IWRUE to undergo a qualitative transformation from a “resource-driven” model to an “efficiency-driven” model.

Moreover, the relatively low urbanization threshold (0.2301) identified in this study reflects the stage-specific characteristics of Xinjiang’s regional development rather than model bias. First, from a temporal perspective, at the beginning of the study period (2004), the urbanization levels of many prefectures and cities in Xinjiang (average 23.01%) were still at an early stage. These regions were at a critical transitional point from traditional agricultural societies toward industrialization, with production factors beginning to concentrate in urban areas. Second, Xinjiang’s distinctive “oasis economy” pattern results in highly concentrated urbanization. When the urbanization rate reaches approximately 23%, industrial production can initially realize scale effects, making the construction of centralized water supply and water-saving infrastructure feasible. This threshold essentially captures the “early dividend” of urbanization in improving IWRUE, whereby even small increments in urbanization at relatively low development levels can release substantial potential for resource allocation efficiency.

5.2. Regional Heterogeneity

Using a super-efficiency SBM model and 3E system, Rui Zhang calculated green water resource utilization efficiency in the Yangtze River Economic Belt from 2000 to 2018 and found that the phenomenon of urban priority implies imbalanced urbanization, which may lead to urban diseases and pollution transfer, with effects depending on the level of urbanization. However, IWRUE does not automatically improve with urbanization, highlighting the need for a scientifically planned urbanization scale and pace [41]. In contrast, this study finds that, in Xinjiang, urbanization in its early stage exerts a more severe inhibitory effect on IWRUE. This difference reflects fundamental divergences in development pathways between arid and humid regions. The Yangtze River Economic Belt benefits from mature industrial clusters that facilitate rapid diffusion of water-saving technologies [17,42]. By contrast, Xinjiang, as a typical resource-based economy, is highly susceptible in its early development stage to the “resource curse,” whereby low water resource costs and a single industrial structure lead to substantial losses in IWRUE.

Furthermore, Masayuki Morikawa’s empirical analysis, based on firm-level microdata from Japan (2007–2008), found that high-density industrial cities generate significant positive spillover effects on service-sector energy intensity, whereas low-density population areas tend to experience efficiency drag [43]. Unlike these findings, the present study shows that, during the low-urbanization stage, industrial employment does not exert a significant effect on IWRUE. However, as urbanization levels increase, this factor becomes significantly positive. This suggests that IWRUE improves through the combined effects of rising labor productivity, technological innovation, and management improvements [44]. At higher levels of urbanization, increases in industrial employment are typically accompanied by higher production efficiency and more refined water resource management, thereby promoting improvements in IWRUE [45,46]. Meanwhile, when industrial output remains relatively low, increases in industrial employment exert a significant negative effect on water resource efficiency because early-stage industrial development often entails increased resource consumption resulting from workforce expansion [47]. However, as industrial output increases, the expansion of industrial employment begins to exert a positive effect on IWRUE, reflecting improvements in labor productivity [48], technological progress, and management optimization. At this stage, additional employment contributes to the application of advanced technologies and the refinement of water resource management, thereby facilitating more efficient water use and enhancing IWRUE [17,48].

5.3. Threshold-Based Pathways for Sustainable Development

The quantitative thresholds derived in this study provide a scientific benchmark for implementing differentiated water resource management strategies in Xinjiang. The findings indicate that, for regions that have not yet reached the identified thresholds (such as certain prefectures in southern Xinjiang), policy orientation should shift from the mere pursuit of output scale expansion toward deep optimization of resource allocation. Specifically, strengthening investment in water-saving infrastructure and reinforcing administrative regulatory measures are necessary to correct distortions in factor allocation. These efforts aim to help such regions effectively offset the “efficiency trap” and achieve a smooth transition toward an intensive utilization model.

In contrast, for regions that have surpassed the threshold (such as Urumqi and Changji), the observed diminishing marginal returns on investment—as evidenced by the declining investment coefficients in Table 3—suggest that these areas might consider a strategic transition from factor-driven to innovation-led development, potentially shifting their focus toward refined management and green technological innovation. By enhancing the quality of human capital and establishing endogenous growth mechanisms, these regions can sustain continuous improvements in IWRUE.

Using panel data from Xinjiang covering the period 2004–2022, this study systematically examines the spatiotemporal evolution of IWRUE and its nonlinear driving mechanisms. The results show that IWRUE in Xinjiang exhibits a pronounced spatial pattern of “higher in the north and lower in the south,” with an overall fluctuating upward trend. However, significant disparities in evolutionary trajectories persist across regions, and stability remains to be strengthened. Further threshold effect analysis reveals structural shifts in the impacts of industrial growth and urbanization on IWRUE. When industrial output and urbanization rates exceed the critical thresholds of 13.234 and 0.2301, respectively, the negative association between industrial scale expansion and water consumption growth weakens significantly, and the marginal contribution of labor input shifts from negative to positive. The system thereby transitions from a “factor-driven” model to an “efficiency-driven” model. Based on these findings, this study proposes differentiated policy pathways. Our analysis indicates that for regions trailing the threshold, a strategy centered on correcting factor allocation distortions, bolstering water-saving infrastructure, and optimizing scale might be the most effective way to break the “low-efficiency lock-in.” For mature regions past the threshold, the results suggest a policy shift toward upgrading labor quality, driving green tech innovation, and refining management to sustain efficiency gains.

6. Summary

Although this study provides empirical evidence and policy implications for improving IWRUE in Xinjiang, several limitations remain. Due to data constraints, continuous indicators of water resource endowment could not be directly incorporated into the model, nor could heterogeneity at the micro-enterprise level be examined to explain the mechanisms underlying macro-level patterns. In terms of the indicator system, environmental constraints are represented solely by wastewater discharge, without incorporating additional green transformation indicators such as water governance performance. Moreover, industry categories were not disaggregated to identify sectoral heterogeneity in threshold effects. From a methodological standpoint, this analysis operates on the assumption of regional independence. While initial tests suggested weak spatial dependence—rooted in the scattered and isolated nature of Xinjiang’s oasis industrial zones—this remains a limitation of the current work. Subsequent studies should account for the evolving spatial interactions and spillover effects that arise as regional connectivity strengthens, driven by intensifying interregional resource allocation, industrial linkages, and technological diffusion. These limitations could be addressed in future investigations by incorporating continuous environmental variables, firm-level microdata, industry-specific indicators, and spatial econometric models.