Improving Ecosystem Services Production Efficiency by Optimizing Resource Allocation in 130 Cities of the Yangtze River Economic Belt, China

Abstract

China has adopted extensive restoration practices to improve ecosystem function. The efficiency of these restoration efforts remains unclear, which may hinder the supply of ecosystem services (ESs). In this context, this study first employed InVEST models to clarify spatio-temporal changes in five key ESs. The static and dynamic efficiencies of ecosystem service production in 130 cities from 2015 to 2021 in the Yangtze River Economic Belt (YREB) were then measured using the Super-SBM-Malmquist model, with ESs considered as outputs. The results indicated that water conservation (WC), water purification (WP), and soil retention (SR) exhibited overall declining trends, decreasing by 28.32%, 3.22%, and 10.00%, respectively, while carbon storage (CS) and habitat quality (HQ) remained steady. More than 70% of studied cities exhibited static efficiency levels below 50%, which were attributed to inefficient utilization of labor, capital, and technology. Significant spatial heterogeneity was observed, with high-efficiency cities mainly located in mountainous areas and low-efficiency cities concentrated in flat regions. The downward trend in dynamic efficiency has been reversed from a 39.02% decline in 2015–2018 to a 38.31% increase in 2018–2021, despite being adversely affected by technological regression. Finally, several policy implications are proposed, including optimizing resource allocation, introducing advanced technology and setting the intercity cooperation and complementarity mechanisms.

1. Introduction

The remarkable economic growth in China over the past four decades has intensified pressure on natural ecosystems [1], resulting in serious ecosystem degradation [2]. For instance, water quality deterioration, soil erosion, and biodiversity loss have significant socioeconomic consequences [3]. In response, the Chinese government has launched a series of large-scale ecological restoration policies and programs [4,5]. One representative initiative is the Natural Forest Conservation Program (NFCP), which aims to halt commercial logging in natural forests across the upper Yangtze Basin [6]. The program has significantly increased forest cover, reduced soil erosion, and enhanced ecosystem services, such as carbon sequestration and soil retention [7,8,9]. As the third largest river in the world and the longest river in China, the Yangtze River plays an important role in both ecological security and economic development [10]. However, intensive urbanization and economic activities in the Yangtze River Basin over the past 30 years have severely degraded its ecosystems [11]. Given its ecological significance, the Yangtze River Basin has become a national priority area for restoration efforts [12,13]. Within the Yangtze River Economic Belt (YREB), ecological restoration investment exceeded 500 billion yuan between 2016 and 2022, involving more than 850,000 forest rangers in various conservation and rehabilitation tasks [14].

However, the National Audit Office of China [15] revealed that several municipal authorities exaggerated afforestation and rehabilitation targets to secure more ecological compensation funds from the central ecological compensation programme, leading to a spatial misallocation of capital, resources, technology, and labor [16]. Moreover, the allocation of production factors is often influenced by subjective elements such as individual or institutional preferences, risk perceptions, and willingness to invest in ecological restoration [17,18]. These issues often result in inefficiencies or even failure of ecological restoration [19]. Therefore, it is crucial to optimize the allocation of production factors to achieve the maximum ecological restoration benefits.

Data Envelopment Analysis (DEA) is a non-parametric method used to estimate production frontiers by assessing the relative efficiency of decision-making units (DMUs) based on multiple input and output indicators [20,21]. It has been applied to measure efficiency across various sectors, including food production [22,23], cultivated land use [24], carbon emissions [25,26], and energy use in the manufacturing industry [27]. Recently, DEA has been extended to evaluate ecological efficiency using ecosystem services (ESs) as key output indicators [28,29]. ESs, defined as the benefits that humans obtain directly or indirectly from ecosystem structures, functions, and processes [30], have received increasing attention in environmental policy, financial decision-making, and ecosystem management [31,32,33,34,35,36]. Accordingly, a growing number of studies have employed ESs in DEA-based frameworks to evaluate ecological performance, commonly referred to as ecosystem service production efficiency (ESPE) [18,37,38,39]. However, a key limitation of the traditional DEA model is its inability to differentiate DMUs that achieve an efficiency score of 1 [40]. To overcome this drawback, the Super Slack-Based Measure (Super-SBM) model, which is a non-radial extension of the DEA framework that incorporates input and output slacks directly into the efficiency evaluation, has been developed as an advanced non-radial extension of DEA, incorporating slack variables directly into the objective function and accommodating both desirable and undesirable outputs through self-weighted mechanisms [17,41]. This approach enables a more precise efficiency assessment and effective ranking of DMUs, even when their efficiency scores are greater than or equal to 1. Most existing studies have concentrated on assessing how natural factors (e.g., climate change, terrain conditions) [42,43] and human activities (e.g., land use change, urban expansion) [44,45] influence the production efficiency of ESs. Nevertheless, most studies overlook how direct production inputs—financial capital, labor, resource allocation, and technology—affect restoration outcomes and the resulting supply of ecosystem services. Thus, clarifying the spatial and temporal links between these inputs and ecosystem service outputs is essential for guiding efficient resource allocation and supporting sustainable ecological development.

To fill this gap, we propose an evaluation framework for ESPE based on the Super-SBM Malmquist model (Figure 1). In this framework, five quantified ecosystem service outputs—carbon storage, soil retention, water conservation, habitat quality, and water purification—are used as measurable indicators to represent the ecological benefits of restoration projects, enabling direct comparisons between resource inputs and ecological outcomes. First, we quantify ESs at the pixel level for 130 cities in the YREB using the InVEST model. Second, we assess the static and dynamic ESPE of these cities using the Super-SBM and Malmquist models. Third, this study sheds light on management practices that can optimize the allocation of production factors for ESs output. The remainder of this paper is organized as follows. Section 2 describes the study area, methods, and data sources. Section 3 presents the main research results. Section 4 presents the discussion and policy implications of the study. Finally, we present our concluding remarks.

2. Materials and Methods

2.1. Study Area

The YREB spans 11 provinces and 130 cities from west to east, including Yunnan, Sichuan, Guizhou, Chongqing, Hunan, Hubei, Jiangxi, Anhui, Zhejiang, Jiangsu, and Shanghai, and is located between 21°8′45″ N–34°56′47″ N and 97°31′50″ E–121°53′23″ E (Figure 2 and Figure S1). It is the largest economic belt in China, accounting for 21.4% of the total land area and more than 40% of the population and GDP. The YREB is an ecological conservation area featuring diverse ecosystem types, thousands of key protected species, and a high level of biodiversity [46]. In 2021, the urbanization rate and GDP reached 64.90% and 53,075 billion yuan, which increased by about 28% and elevenfold compared with 2000, respectively [47]. To mitigate the trend of ecological degradation in the YREB, more than 582.7 billion yuan was invested between 2015 and 2021 [48]. Among these, investment in afforestation and forest tending, grassland restoration, wetland conservation and restoration, and desertification prevention accounted for approximately 89.10%, 1.28%, 6.27%, and 0.43%, respectively [49].

2.2. Ecosystem Services Assessment

In the Super-SBM framework, outputs are categorized as either desirable or undesirable. Desirable outputs are ecosystem services for which higher values reflect stronger ecological benefits, whereas undesirable outputs are environmental disservices that should be reduced as much as possible [50]. Water conservation (WC), soil retention (SR), carbon storage (CS), and habitat quality (HQ) were therefore treated as desirable outputs. Water purification (WP), in contrast, was represented by the annual mass of nitrogen exported to surface waters, simulated with the Nutrient Delivery Ratio (NDR) module of InVEST [51]. A higher NDR-derived export value indicates a weaker in-situ purification capacity and a greater risk of downstream eutrophication. Therefore, WP was modelled as an undesirable output in the Super-SBM analysis. The definitions, calculation methods, and data sources for all ES indicators are summarized in

Table 1. Definitions, methods, and data requirements for the studied ecosystem services.

ES	Definition	Unit	Method	Data Required
WC	WC refers to the ability of an ecosystem to intercept or store water resources from rainfall, which is equal to water yield minus surface runoff [1].	mm	Water yield model of InVEST	Precipitation, Potential evapotranspiration, surface runoff, soil data and LULC
SR	SR is defined as the soil retained by the ecosystem, which is expressed as the gap between potential and actual soil erosion [54].	t·hm−2	Sediment retention model of InVEST	Rainfall erosion factor, soil erodibility factor, DEM, and LULC
CS	CS refers to the quantity of carbon storage and sequestration in an ecosystem, which includes aboveground carbon, belowground carbon, dead organic carbon, and soil organic carbon [55].	t·hm−2	Carbon Storage and Sequestration model of InVEST	LULC and carbon pools data
HQ	HQ is an important indicator that reflects biodiversity by estimating the extent of habitat and vegetation types across a landscape, and their state of degradation [56].	Index	Habitat quality model of InVEST	LULC, threat data, and sensitive data
WP	WP is defined as the retention capacity of nitrogen by the ecosystem, which is expressed by total nitrogen delivered to a water body. A larger nitrogen delivery value indicates weaker WP capacity [57].	kg·km−2	Nutrient delivery ratio model of InVEST	Precipitation, nutrient load, retention efficiency, length, DEM, and LULC

, and full modelling details are provided in the Supplementary Information [1,52,53].

2.3. Quantifying the Ecosystem Services Production Efficiency

Here, we use the Super-SBM-Malmquist model to evaluate the static and dynamic efficiencies of ESs output of YREB. The input variables affected by management decisions include capital, labor, resources, and technology, the details of which are listed in

Table 2. Input and output indicators for evaluating ESPE in the YREB.

Category	Indicator	Variable	Variable Description	Unit
Input	Capital	Financial investment	Annual investment of local government in ecological restoration and management	Billion yuan
	Labor	Employees	Number of employees, including researchers, administrators, and forest rangers in the forestry and grassland system	Persons
	Resource	Land	Afforestation area and grass planting area	km−2
	Technology	Green innovation	The annual grant quantity of green invention patents and green new utility patents	Count
Output	Desired outputs	Water conservation (WC)	Details are shown in Table 1	Billion m3
		Soil retention (SR)		Billion tons
		Carbon storage (CS)		Billion tons
		Habitat quality (HQ)		Index (0–1)
	Undesired output	Water purification (WP)		Million tons

. The output variables include desirable and undesirable outputs (Table 1). WP, measured by the total nitrogen delivered to water bodies, is designated as an undesired output in this study. This study evaluated dynamic efficiency at the city level from 2015 to 2021, as most ecosystem restoration programs were implemented during this period.

2.3.1. Super Slacks-Based Measure Model

The Super-SBM model allows for the ranking of DMUs, even when multiple units achieve an efficiency score of 1, and captures input redundancies and output shortfalls more accurately than traditional DEA models [58]. In this study, cities are treated as DMUs, and the Super-SBM model is employed to evaluate the static efficiency of ESPE of 130 cities in the YREB. The model is formulated as follows:

$${\theta }^{\mathrm{*}}=\underset{\lambda ,s^{-},s^{+}}{\min }{\displaystyle \frac{1-\frac{1}{m}\sum _{i=1}^m{s}_i^{-}/x_{i0}}{1+\frac{1}{q+u}\left(\sum _{r=1}^u\frac{s_r^{+}}{y_{r0}}+\sum _{k=1}^h\frac{s_k^{-}}{b_{k0}}\right)}}$$

(1)

subject to:

$$\left\{\begin{array}{c}{x}_{i0}={\displaystyle \sum _{j=1}^n}{\lambda }_j{x}_{ij}+s_i^{-},\hspace{1em}\hspace{1em}i=1,2,\dots ,m;\\ y_{r0}={\displaystyle \sum _{j=1}^n}{\lambda }_j{y}_{rj}-s_r^{+},\hspace{1em}\hspace{1em}r=1,2,\dots ,u;\\ b_{k0}={\displaystyle \sum _{j=1}^n}{\lambda }_j{b}_{kj}+s_k^{-},\hspace{1em}\hspace{1em}k=1,2,\dots ,q;\\ {\lambda }_j\ge 0\left(\forall j\right),s_i^{-}\ge 0\left(\forall i\right),s_r^{+}\ge 0\left(\forall r\right),s_k^{-}\ge 0\left(\forall k\right){\displaystyle \sum _{j=1}^n}{\lambda }_j=1\end{array}\right.$$

(2)

where ${\theta }^{\mathrm{*}}$ represents the ESPE value. $n$ denotes the number of DMUs, each of which has $m$ inputs, $u$ desired outputs, and $q$ undesirable outputs. $s_i^{-}$, $s_r^{+}$ and $s_k^{-}$ are the slack variables of inputs, desirable outputs, and undesired outputs, respectively. $\lambda$ denotes the weight vector. $x$, $y,$ and $b$ are the vectors of inputs, desirable outputs, and undesirable outputs, respectively. When θ* = 1, the DMU is in a fully efficient status, indicated by $s_i^{-}$ = $s_r^{+}$ = $s_k^{-}$ = 0, signifying no redundancy in inputs and undesirable outputs, and no deficiency in desirable outputs. If ${\theta }^{\mathrm{*}}<1$, then $s_i^{-}>0$ indicates the amount of input redundancy; $s_r^{+}>0$ indicates the extent of desirable output deficiency; and $s_k^{-}>0$ represents the reduction required in undesirable outputs.

2.3.2. Malmquist Index Model

The Malmquist Index model (MI), originally proposed by Malmquist in 1953 [59], is used to assess changes in efficiency between two consecutive time periods, t and t + 1 [60]. In this study, the MI is applied to evaluate the dynamic evolution of ESPE, capturing temporal changes in the efficiency of input allocation relative to ecosystem service outputs. The MI is calculated as follows:

$$\begin{array}{c}{M}_i\left(X_{t+1},Y_{t+1},X_t,Y_t\right)={\left[{\displaystyle \frac{d_s^t\left(x_{t+1},y_{t+1}\right)}{d_s^t\left(x_t,y_t\right)}}\times {\displaystyle \frac{d_s^{t+1}\left(x_{t+1},y_{t+1}\right)}{d_s^{t+1}\left(x_t,y_t\right)}}\right]}^{{\displaystyle \frac{1}{2}}}\\ =TEC\times TPC=PEC\times SEC\times TPC\end{array}$$

(3)

where $M_i\left(X_{t+1},Y_{t+1},X_t,Y_t\right)$ represents the change in efficiency between periods t and t + 1; the higher the value, the more efficient the utilization and allocation of input indicators. When it >1, =1, or <1, denoting the efficiency improves, remains, or declines, respectively. $x_t$ and $x_{t+1}$ stand for the input indicator vectors between periods t and t + 1, respectively. Similarly, $y_t$ and $y_{t+1}$ stand for the output indicator vectors between periods t and t + 1, respectively. $d_s^t\left(x_t,y_t\right)$ and $d_s^{t+1}\left(x_{t+1},y_{t+1}\right)$ represent the distance functions of the inputs and outputs between periods t and t + 1, respectively.

TEC measures the “catch-up” effect, namely the movement of the production frontier from period t to t + 1, which can reveal the impact of the utilization and allocation of input indicators on efficiency at the current technical level. TEC > (or <1) indicates a gain (or loss) in the technical efficiency. TEC can be further decomposed into pure efficiency change (PEC) and scale efficiency change (SEC), which reflect changes in efficiency due to optimal resource utilization and deviations from the optimal scale of allocation, respectively. TPC is defined as the “frontier shift” effect, namely whether the production technology of a specific DMU is progress or regression between periods t and t + 1, which reflects the impact of technological improvement and innovation on efficiency. TPC > 1 (or <1) indicates a gain (or loss) in technological progress.

2.4. Data Sources and Processing

The data sources used to quantify the ESs are presented in Tables S1 and S2. Financial investment and employee data at the city level are derived from the China Statistical Yearbooks and Statistical Bulletins [47,48]. Green innovation data are obtained from the Chinese Research Data Services Platform [61]. Land data is derived from Yang and Huang [62], who published 30 m annual land cover raster data for China. To obtain panel data on forest and grassland areas at the city level, land cover raster data were extracted from the field and then summarized by zone using ArcGIS 10.6 (Esri, Redlands, CA, USA).

3. Results

3.1. Spatiotemporal Patterns of Multiple Ecosystem Services

To reflect the spatiotemporal heterogeneity of ESs, the results of city-scale zoning statistics are shown in Figure 3, while the spatial distribution of ESs in the YREB from 2015 to 2021 is shown in Figure S1. Overall, WC, WP, and SR exhibited declining trends, whereas CS and HQ remained relatively stable in the YREB between 2015 and 2021. Figure 3a–c show that WC in YREB decreased from 845.46 mm in 2015 to 606.02 mm in 2021, a decrease of 28.32%. Spatially, WC shows a decreasing trend from southeast to northwest, western cities exhibit an upward trend, and eastern and central cities show a downward trend. The spatiotemporal distribution of WP in YREB is illustrated in Figure 3m–o. From 2015 to 2021, the mean WP value decreased from 306.74 kg·km⁻² to 297.18 kg·km⁻², indicating that the WP capacity in YREB increased by 3.22%. There is significant heterogeneity in the spatial distribution of ESs, with high values mainly distributed in eastern Hubei, Jiangxi, Anhui, Jiangsu, and the Chengdu Plain, and low values concentrated in the mountainous regions of Sichuan, Yunnan, and western Hubei. The most significant decline occurred in the central cities of the YREB, followed by those along the eastern coast.

As shown in Figure 3d–f, per hectare of SR in the YREB slightly decreased from 198.23 t/ha in 2015 to 178.61 t/ha in 2021, representing a 10.00% decline. Spatial heterogeneity is also observed, with low values (less than 50 t/ha) mainly located in flat regions, such as the Sichuan Basin, Jianghan Plain, and Yangtze Delta Plain. In contrast, high values are mainly distributed in the mountainous regions, including the Hengduan, Qinling-Daba, and Wuyi mountains, most of which exceed 300 t/ha. Except for 15 cities in Sichuan and Yunnan, SR across the YREB showed a downward trend from 2015 to 2021, with the most significant declines observed in the central region. The spatiotemporal patterns of CS and HQ in the YREB are presented in Figure 3g–i and j–l, respectively. CS and HQ remained nearly unchanged, with their average values fluctuating from 117.89 t/ha and 0.655 t/ha in 2015 to 118.43 t/ha and 0.651 t/ha in 2021, respectively. High values of CS and HQ are primarily observed in mountainous cities. In contrast, low values of CS and HQ are concentrated in flat cities. Notably, a downward trend in CS and HQ is predominantly observed in cities located downstream of the YREB. These consistent declines suggest that restoration efforts may be insufficient to offset ongoing degradation trends in urban and lowland areas.

3.2. Ecosystem Services Production Efficiency Analysis

3.2.1. Descriptive Statistics of Indicators

Descriptive statistics of the input and output indicators are presented in

Table 3. Descriptive statistics of input and output indicators for ESPE.

Indicator	Sample Quantity	Unit	Mean	Std. Dev	Max	Min
Financial investment	390	Billion yuan	6.51	6.85	50.98	0.62
Employees	390	Persons	5420	8082	63,968	26
Land	390	km−2	170.53	285.13	2900.02	0.27
Green innovation	390	Count	603	1430	13,320	2
Water conservation (WC)	390	Billion m3	10.35	9.38	78.07	0.26
Soil retention (SR)	390	Billion tons	6.72	17.57	173.93	0.00035
Carbon storage (CS)	390	Billion tons	0.20	0.22	1.64	0.01
Habitat quality (HQ)	390	Index	0.65	0.13	0.89	0.32
Water purification (WP)	390	Million tons	3.80	2.85	27.36	0.28

. The dataset includes 390 observations covering 130 cities in 2015, 2018, and 2021. The results show that for all input indicators, the standard deviations exceed the mean values, and the ranges between the maximum and minimum values are substantial. These patterns indicate a high degree of regional heterogeneity in input factor allocation.

3.2.2. Spatiotemporal Patterns of Static Efficiency

As shown in Figure 4e, the static ESPE of the 130 cities studied ranged from 0.03% to 192.91% over the study period, with a mean value of 33.19%. More than 70% of the studied cities had efficiency values below 50% during the study period. The spatial distribution of the static ESPE at city scale is shown in Figure 4a–c. Significant spatial heterogeneity in efficiency is observed. Cities with efficiency values greater than one are mainly distributed in the mountainous areas, including Garze Tibetan Autonomous Prefecture (160.66%, mean over the study period), Xishuangbanna Dai Autonomous Prefecture (128.39%), Shennongjia Forestry District (126.83%), Ngawa Tibetan and Qiang Autonomous Prefecture (123.65%), and Nujiang Lisu Autonomous Prefecture (115.22%). In contrast, lower efficiency values were mainly concentrated in flat regions, such as the Jianghan Plain and Yangtze River Delta Plain. These results indicate that the allocation of production inputs, such as capital and labor, is more efficient in cities with higher ES output levels than in those with lower ES outputs.

The annual mean ESPE of the studied cities initially decreased from 29.57% in 2015 to 22.65% in 2018, and then rebounded to 47.34% in 2021. Concurrently, the number of cities with efficiency values greater than one increased from 21 in 2015 to 37 in 2021. Notably, Xiantao, Jiaxing, and Bazhong experienced the most significant declines in efficiency, decreasing by 92.89%, 85.32%, and 70.47%, respectively. In contrast, Yancheng, Nantong, and Qianjiang experienced the most significant increases, with efficiency values increasing by approximately 16, 15, and 9 times, respectively.

Figure 4d illustrates the temporal evolution of ESPE across 11 provincial-level regions in the YREB from 2015 to 2021. Overall, ESPE exhibits an upward trend, indicating gradual improvements in the efficiency of ecosystem restoration efforts. Notably, the provinces in the upper reaches of the Yangtze River, such as Guizhou and Yunnan, experienced substantial increases in ESPE. Specifically, the average ESPE in Guizhou increased from 11.42% in 2015 to 69.82% in 2021, while Yunnan’s ESPE rose from 55.93% to 75.18% over the same period. In contrast, Zhejiang Province, located in the downstream region, showed a slight decline in ESPE, decreasing from 79.38% in 2015 to 71.38% in 2021. These changes highlight regional differences in restoration outcomes and underscore the importance of region-specific policy interventions.

3.2.3. Redundancy Analysis

The redundancy rates of the Super-SBM model are expressed as the ratio of slack variables to inputs, providing insights into the gap between the current and optimal ESPE [17]. Figure 5 shows the redundancy rates of all input and undesirable output indicators, including capital, labor, land, and green innovation, in the studied cities between 2015 and 2021. The mean redundancy rate of green innovation was the highest during the study period at approximately 65.05%, with high values primarily observed in the central and eastern regions of the YREB. These results suggest a significant underutilization of green innovation technologies and regional disparities in resource allocation. In contrast, capital, labor, and land showed relatively lower redundancy rates, averaging 35.64%, 37.35%, and 40.60%, respectively. This indicates that these inputs are utilized more efficiently than green innovation. However, their spatial distribution remains unbalanced, suggesting that capital, labor, and land are not optimally allocated across regions.

From a temporal perspective, the mean redundancy rates of capital, labor, and land resources increased from 25.81%, 45.84%, and 25.57% in 2015 to 53.68%, 46.75%, and 59.78% in 2018, and then declined to 27.59%, 18.97%, and 36.93% by 2021, respectively. The number of cities with zero input redundancy rates for capital, labor, and green innovation increased from 45, 27, and 22 in 2015 to 51, 73, and 40 in 2021, respectively. In contrast, the number of cities with zero input redundancy in land resources decreased from 54 in 2015 to 42 in 2021. These trends suggest divergent dynamics in input utilization efficiency across different production factors over time.

Moreover, cities with zero input redundancy typically exhibit higher levels of ESPE, suggesting that capital, labor, land, and green innovation are consistently and efficiently utilized in these cities. The redundancy rate of the undesirable output variable WP is also relatively low, decreasing from 11.87% in 2015 to 5.75% in 2021, with an average of 8.46%. This trend indicates that total nitrogen discharge into water bodies was effectively controlled during the study period.

From a spatial perspective, several cities, including Dazhou (from 0% in 2015 to 89.69% in 2021), Chuzhou (0% to 83.47%), Chizhou (0% to 67.92%), Fuyang (0% to 51.75%), and Jiaxing (0% to 51.63%)—experienced a shift from zero to high redundancy in capital inputs. For labor inputs, redundancy rates in Guangyuan, Dazhou, Bazhong, and Enshi Tujia and Miao Autonomous Prefecture increased from zero in 2015 to 60.87%, 58.77%, 48.92%, and 17.21%, respectively. In terms of land inputs, Anqing, Jingzhou, Nanchang, Fuyang, and Bazhong saw redundancy rates rise from zero in 2015 to 85.29%, 77.18%, 72.92%, 67.11%, and 63.58% by 2021. Similarly, green innovation input redundancy rates increased from zero in 2015 to 91.43% in Yibin, 90.85% in Mianyang, 90.62% in Yuxi, 89.94% in Jiaxing, and 71.79% in Taizhou. These sharp increases in input redundancy reflect a shift from efficient to inefficient resource use, which likely contributed to the observed decline in ESPE in these cities.

In contrast, capital redundancy rates in Jinhua, Qiannan Buyi and Miao Autonomous Prefecture, Yangzhou, Xinyu, and Qiandongnan Miao and Dong Autonomous Prefecture decreased from 82.29%, 68.76%, 61.86%, 59.13%, and 58.79% in 2015 to zero in 2021, respectively. For labor inputs, redundancy rates in Bijie, Anshun, Qiannan Buyi and Miao Autonomous Prefecture, Chuxiong Yi Autonomous Prefecture, and Qiandongnan Miao and Dong Autonomous Prefecture declined from 89.91%, 88.60%, 85.30%, 83.10%, and 79.43% to zero over the same period. Regarding land inputs, Qianxinan Buyi and Miao Autonomous Prefecture, Qiannan Buyi and Miao Autonomous Prefecture, Chuxiong Yi Autonomous Prefecture, Zhangjiajie, and Qiandongnan Miao and Dong Autonomous Prefecture saw redundancy rates fall from 79.53%, 77.22%, 73.01%, 66.50%, and 63.75% in 2015 to zero in 2021. Similarly, the green innovation input redundancy rates in Ngawa Tibetan and Qiang Autonomous Prefecture, Bengbu, Changzhou, Chuxiong Yi Autonomous Prefecture, and Dali Bai Autonomous Prefecture dropped from 99.83%, 99.65%, 99.30%, 99.04%, and 97.65% to zero by 2021. These results suggest a marked improvement in input allocation efficiency, particularly in cities with high vegetation cover, indicating a transition from inefficient to efficient resource utilization.

3.2.4. Dynamic Efficiency and Its Decomposition

The dynamic efficiency (Malmquist Index, MI) and its decomposition for all the cities in YREB from 2015 to 2021 are shown in Figure 6. Compared with static efficiency, dynamic efficiency can further analyze the dynamic evolution process and contributing sources of ESPE [63], including TEC (divided into PEC and SEC) and TPC [25]. The results indicate that the mean value of MI in YREB from 2015 to 2021 was 0.823, indicating that the ESPE of the YREB maintained an overall declining trend, decreasing by approximately 17.73%. However, MI fluctuated during the study period. It first decreased by 39.02% from 2015 to 2018, and then increased by 38.31% from 2018 to 2021. This fluctuation was mainly due to the excessive redundancy of the input from 2015 to 2018. The results indicate that the ESPE in the YREB was unstable. While the downtrend of MI has been reversed, it reflects an improvement in the efficiency of resource allocation since 2018. Both PEC and SEC showed an increasing trend in the YREB from 2015 to 2021, with average increases of 17.53% and 1.31%, respectively. However, TPC declined by 30.11%, indicating that technological regression constrained the improvement of ESPE in the YREB.

The spatial distributions of the MI, PEC, SEC, and TPC in the cities during the study periods are shown in Figure 6a–c, d–f, g–i, and j–l, respectively. The four indices showed significant improvement in 2018–2020 compared to 2015–2018 and displayed spatial heterogeneity. As shown in Figure 6c, the cities in the YREB with improved MI are mainly distributed in the Guizhou, Jiangsu and Anhui provinces from 2015 to 2021, and the top five cities with the highest MI values were Nantong (8.092), Huainan (3.933), Huaibei (3.359), Yancheng (3.266) and Suqian (2.561). In Guizhou Province, TEC was the main contributor to the improved MI, whereas in Jiangsu and Anhui Provinces, TPC was the dominant contributor. These results indicate that the improvement in MI in the underdeveloped and developed areas of YREB mainly benefited from the effects of TEC and TPC, respectively. Additionally, although the overall declining trend of MI in YREB was reversed from 2018 to 2021, as shown in Figure 6b, the MI of some cities continued to decline. A total of 68, 93, and 32 cities had average MI, TEC, and TPC values greater than 1, respectively, in the YREB from 2018 to 2021. However, only 22 cities have an average value greater than 1 for all three indices, and they are mainly distributed in the western (including Sichuan, Yunnan, and Guizhou provinces) and eastern (including Jiangsu and Anhui) regions of YREB. These results indicate that ESPE in the YREB still has significant potential for improvement, especially in central cities. Notably, in most of the cities examined, technological progress impeded efficiency improvement, suggesting a potential regression or a mismatch in innovation-driven inputs. Although technical efficiency contributed positively to overall efficiency growth, it remains constrained by pronounced spatial imbalances.

4. Discussion

4.1. Optimization of Ecosystem Services Production Efficiency

Although the implementation of ecological restoration policies in the YREB can improve ecosystem service supply [13,64,65], their efficiency remains insufficiently understood. Based on dynamic DEA and ESs accounting for 130 cities in YREB, the results indicate that ESPE did not show significant improvement from 2015 to 2021, although the downward trend was reversed after 2018. One possible explanation is the long-term nature of ecological improvement and time lag in ecological effects [4]. However, the spatial mismatch in production factor inputs, driven by the absence of targeted policy measures, should also be regarded as a critical concern. Therefore, optimizing the spatial and structural allocation of capital, labor, and land resources is crucial. The findings of this study offer valuable guidance for minimizing input redundancy and designing more differentiated, region-specific ecological restoration policies to enhance resource efficiency and ecological outcomes.

The static ESPE shows pronounced spatial heterogeneity, with high values mainly concentrated in mountainous cities, consistent with the findings of Zhang et al. [66]. Mountainous cities with high vegetation cover not only play a crucial role in enhancing ESs but are also primary targets for implementing ecological restoration policies, such as forest and grassland restoration and the establishment of protected areas [67,68]. This suggests that regions with high vegetation-based ESPE are ecologically important areas that closely align with China’s Ecological Redline strategy. The Ecological Redline strategy designates critical ecological zones, such as mountainous and forest-dominated regions, as strictly protected, which reinforces the need to prioritize ecological inputs and avoid resource redundancy in these areas. In addition, differences in redundancy further reflect spatial disparities in ESPE. The implementation of ecological restoration policies relies heavily on financial support from both national and local governments [69,70]. In the study area, economically developed cities often exhibit weaker ecological function. In these cities, redundant inputs of production factors combined with insufficient ecosystem service outputs lead to low efficiencies. Conversely, ESPE in mountainous cities can be enhanced through increased investment in capital, labor, resources, and technology. Therefore, our results can provide scientific support for identifying and managing high-efficiency ecological zones under the Ecological Redline strategy framework and offer a reference for incorporating efficiency indicators into national spatial planning. These findings suggest that strengthening inter-city complementarities and promoting the optimal allocation of production factor inputs are critical for improving overall ESPE across the region.

The results of this study indicate that the improvement of technical efficiency—which reflects the optimization of factor combinations and management levels—is a key driving force in reversing the decline in management inefficiency, a finding consistent with Liu et al. [58]. These results suggest that the allocation and management of ecological restoration inputs in the YREB have been increasingly optimized, reflecting a shift toward integrated planning and coordinated development following the release of the Development and Planning Outline of the YREB. Nevertheless, technological progress remains insufficient to effectively enhance ESPE, highlighting the urgent need to strengthen innovation-driven approaches and increase investments in ecological restoration technologies.

4.2. Policy Implications

This section presents key implications for enhancing ESPE in the YREB based on our findings. Firstly, both this study and previous studies confirm that the input redundancy and output insufficiency are primary drivers of inefficiency. In particular, the insufficient output of ESs in this study is largely attributed to low forest and grass vegetation cover. Therefore, direct ecological restoration measures, such as afforestation, grassland establishment, and the rehabilitation of degraded forest and grassland ecosystems, should be further promoted in the forestry and grassland sectors [44,71]. Meanwhile, erosion control projects can also be widely implemented. Previous studies have shown that such ecological restoration efforts can enhance key ESs, including CS, WC, SR, and WP [2,72,73]. In addition, sustainable spatial planning measures, such as the ecological redline policy and the designation of protected areas, can further support the improvement of ES supply [74,75].

Secondly, the results of this study indicate that cities located in mountainous areas tend to exhibit higher ESPE, primarily due to lower input redundancy and higher ecosystem service outputs. In contrast, more economically developed cities located in relatively flat areas tend to exhibit lower ESPE due to higher input redundancy and insufficient ES output. To address this spatial mismatch, integrated policy mechanisms and market-oriented coordination strategies should be established to promote the optimal allocation of production factors. In addition, when the ES supply reached a certain threshold, the input of production factors for ecological restoration would exceed urban needs, leading to low efficiency [64]. Therefore, rather than implementing rigid ecological policies, it is essential for national and provincial governments to adopt top-level policy designs that allow timely adjustments tailored to regional conditions. Finally, a dynamic monitoring and evaluation mechanism for ESPE should be established to ensure adaptive management and continuous improvement.

Thirdly, although declining technological progress negatively impacts ESPE, this can be offset by improvements in technical efficiency. Therefore, it is necessary to promote innovation-driven development by introducing and applying advanced ecological restoration technologies. Such technologies include slope-to-terrace transformation [76], construction of ecological buffer zones [77,78], and establishment of ecological corridors [79]. In addition, regional technological cooperation should be enhanced to promote balanced ecological restoration across the YREB. For instance, economically developed provinces with advanced technological capabilities, such as Jiangsu, Anhui, and Shanghai, should provide financial and technical support to less developed regions to improve their ecological restoration efficiency.

4.3. Limitations and Prospects

This study has several limitations. Firstly, this study only assessed five key ESs in the YREB. This deficiency may hinder a more comprehensive understanding of the dynamic changes in ecological conditions. Future studies can be improved by considering other ESs, such as food and timber supply, flood mitigation, sandstorm prevention, and biodiversity conservation. Although region-specific parameters were applied, the InVEST model remains sensitive to land cover accuracy and biophysical inputs and lacks the capacity to simulate process-based feedback, such as nutrient cycling and biodiversity thresholds [51,80]. These limitations may affect the reliability of ES estimates, particularly in complex ecological conditions. Future studies should consider coupling InVEST with more process-based models or using ensemble approaches to improve robustness. Secondly, this study was conducted only at the city level for ESPE evaluation due to the unavailability of statistical data on production factor inputs at the county level. However, the allocation, redundancy, and efficiency of input indicators may vary with spatial scale [81]. Future research could obtain more detailed input data through county-level investigations, thereby yielding more accurate and spatially refined results. Thirdly, this study does not explicitly account for the time lag [82,83] between phased ecological investments and delayed ecosystem service responses, which may lead to an overestimation of short-term efficiency gains. Future research could incorporate lagged response models or temporal matching approaches to better align restoration inputs and ecosystem service dynamics. Finally, although this study utilized a dynamic model to obtain a more effective assessment, the indicator selection of the inputs and outputs may not include all natural and human factors that can also influence local ESs supply [18,84]. Future research should consider incorporating a broader set of variables, such as climate variability, land-use intensity, ecological policy interventions, and socio-economic drivers, to more comprehensively capture the multi-dimensional influences on ESs supply and better reflect the local ESPE.

5. Conclusions

Ecological restoration has been given high priority to maintain ecosystem functions in the face of rapid economic development and urbanization. Assessing the efficiency of these ecological interventions can inform more systematic solutions for sustainable management. This study proposes a static and dynamic efficiency integrated evaluation framework based on the Super-SBM-Malmquist model, with ESs as output indicators. A total of 130 cities from 2015 to 2021 in YREB were evaluated.

From 2015 to 2021, WC, WP, and SR showed a general decline, while CS and HQ remained steady. More than 70% of the studied cities had static efficiency lower than 50%, reflecting a significant spatial mismatch in input allocation. This highlights the pronounced spatial heterogeneity of static efficiency across regions. High ESPE values are mainly distributed in mountainous cities, and low values are concentrated in flat cities. Furthermore, the downward trend in dynamic efficiency has been reversed, and the negative impacts of technological regression on efficiency can be offset by increased technical efficiency. This suggests that enhancing technological progress is key to improving dynamic ESPE.

In summary, this study presents an assessment framework for ecological restoration that can be applied globally to regions facing similar challenges in balancing ecological restoration and economic development.