Site Selection for Wind Turbine Recycling Center Based on GIS and DEA

Abstract

Wind power development is accelerating globally, leading to an imminent large-scale retirement of wind turbines and increasing the need for recycling infrastructure. This study proposes an integrated framework for recycling center site selection by combining Geographic Information System (GIS), AHP–CRITIC weighting, and Super-Efficiency Data Envelopment Analysis (DEA). A hierarchical GIS indicator system is constructed by incorporating environmental, locational, and social compatibility factors, including elevation, slope, land use, transportation accessibility, proximity to wind farms, and population-related constraints. GIS performs Euclidean distance, kernel density, and weighted overlay analyses to identify suitable areas, while indicator weights are determined through a hybrid subjective–objective approach. A Super-Efficiency DEA model is then applied, using labor and land costs as inputs and annual decommissioning quantities as output, to evaluate and rank candidate sites, with higher-ranked sites regarded as reliable locations. A case study in Xilingol, Inner Mongolia, verifies the method’s effectiveness. The proposed framework supports scientific planning for wind turbine recycling and promotes sustainable wind energy development.

1. Introduction

With the implementation of the dual carbon strategy in China, new opportunities emerge for the green transition. Among them, green power is an important part of the dual carbon strategy. According to the Statistics Bureau of the Inner Mongolia, China, wind power generation in Inner Mongolia reached 164.3 billion kilowatt-hours in 2024, accounting for as much as 17.55% of the national total. Although the wind power sector in Inner Mongolia develops rapidly, many wind turbines that were put into operation around the year 2000 are about to reach their designed lifespan of 20 to 25 years, which will lead to the first large-scale retirement wave of wind turbines in 2025 across the region and the country (Figure 1). The treatment of wind turbine blades and the recycling of wind turbine units are matters of urgent concern.

Figure 1. The cumulative number of decommissioned wind turbine units.

From the perspective of decommissioned wind farm treatment, there are three main approaches: continuing operation after renovation, extending service life by turbine replacement, and site closure [1,2,3]. Most wind farms in China use imported turbines from Europe and the United States. Due to technical incompatibility with domestic systems, renovation is often unfeasible. Therefore, operation with replacement and site closure are more consistent with the context of China. For the disposal of decommissioned wind turbines, there are three main methods: mechanical crushing, chemical recycling, and thermal recycling [4]. End-of-life wind turbines contain recyclable metals, concrete, cables, and composite blades, among which composite blade materials are relatively difficult to recycle. Improper treatment may cause land occupation, secondary waste, and additional environmental pressure. Therefore, the site selection of wind turbine recycling centers should consider not only economic efficiency, but also transportation accessibility, environmental constraints, land-use compatibility, and recycling demand. Processed waste can facilitate the recycling of rare earth elements [5,6] and the reuse of materials, such as in cement production [7]. Chinese regulations explicitly prohibit the illegal disposal of waste through landfilling or random dumping. Against this backdrop, establishing wind turbine recycling centers becomes a critical initiative. The site selection of these centers requires scientific and rational planning, as it directly impacts operational efficiency and cost-effectiveness. Moreover, these centers perform a variety of core functions, including multi-mode blade processing such as mechanical, chemical and thermal recycling methods, component maintenance and remanufacturing, and material circulation. Such functional diversity significantly increases the complexity of site selection.

At present, the problem of decommissioned wind turbines is becoming increasingly urgent. Existing studies mostly focus on the recycling and utilization of wind turbine resources, the optimization of wind turbine recycling processes [8,9,10], and the circular economy of decommissioned wind turbines [11]. However, there remain significant research gaps in the infrastructure planning involved in recycling. An unreasonable geographical layout of recycling centers may lead to several risks. First, due to the extremely large size of decommissioned wind turbine blades, transportation over long distances in non-standard networks can lead to significantly higher logistics costs. Second, recycling processes such as chemical treatment and pyrolysis may generate high thermal loads and environmental emissions, which should be considered in recycling-center planning. Third, the downstream transportation cost from the recycling center to the customer site needs to be reasonably controlled when the recyclable resources generated from the recycling of wind turbines are resold. Therefore, the scientific planning of the location of the wind turbine recycling center plays a crucial role in the entire wind turbine recycling process.

To address this issue, this study proposes an integrated site-selection framework for wind turbine recycling centers, aiming to identify suitable candidate regions and further support recycling infrastructure planning in Xilingol, Inner Mongolia.

2. Literature Review

2.1. Wind Turbine Recycling

With the rapid expansion of global wind power capacity, the large-scale decommissioning of wind turbines becomes an emerging environmental and technical challenge. Current research on wind turbine recycling mainly concentrates on three areas: recycling technologies, material reuse, and policy design. In terms of technology, mechanical, thermal, and chemical recycling methods are extensively studied [12,13,14,15]. Mechanical recycling is regarded as a cost-effective and energy-efficient approach suitable for large-scale glass fiber reinforced plastic (GFRP) treatment [16,17]. Zhang et al. [11] proposed a mechanical recycling strategy that uses water jet cutting and roller grinding to convert GFRP waste into high-aspect-ratio rebars. Cheng et al. [18] and Hao et al. [19] further optimized physical and chemical recycling processes to enhance efficiency, energy performance, and product value. From the perspective of material reuse, Mao et al. [20] established a multi-stage model showing that recycled fiber-reinforced polymer (FRP) fibers can significantly improve the tensile and bending properties of concrete, providing theoretical support for their large-scale application in construction. From a policy perspective, Zhao et al. [21] employed game-theoretic and circular economy models to quantify the effects of subsidies and technological investment on industrial coordination and recycling profitability. In addition to recycling technologies and material reuse, some studies have begun to examine the spatial distribution and reverse logistics of end-of-life wind turbine blades. Delaney et al. [22] proposed an integrated geospatial approach to assess the spatiotemporal distribution of decommissioned wind turbine blades and support repurposing decisions. Rentizelas et al. [23] optimized the reverse supply network design for circular economy pathways of wind turbine blades in Europe, considering waste suppliers, processing facilities, and downstream reuse. Liu et al. [24] compared the economic feasibility of different end-of-life options for wind turbine blades and highlighted the cost differences among recycling pathways. Trivyza et al. [25] further investigated reverse supply network design for end-of-life wind turbine blades under different policy scenarios, involving material flows, recycling facility locations, and capacities.

2.2. Geographic Information System Site Selection

Geographic Information System (GIS) has been widely applied in facility site selection because it can integrate multi-source spatial data and evaluate the spatial suitability of candidate areas [26]. In GIS-based decision-making, spatial factors such as land use, transportation accessibility, topography, environmental constraints, and proximity to sensitive areas are commonly used for suitability evaluation [27]. By converting different spatial factors into comparable suitability layers, GIS improves the transparency and operability of site-selection decisions. GIS-based methods have been widely used in waste-treatment, recycling, landfill, and infrastructure facility site selection. Aghad et al. [28] integrated fuzzy-AHP and GIS to evaluate solid waste disposal sites by considering spatial and environmental constraints. Hosseinzadeh et al. [29] applied GIS-AHP to municipal landfill site selection and considered land-use conditions, accessibility, and environmental sensitivity. Jonuzi et al. [30] used GIS-supported MCDA for urban infrastructure location planning, showing its applicability to facility-location problems involving accessibility and land-use compatibility. GIS-based methods have also been applied to renewable-energy facility site selection. Umoh et al. [31] used AHP-GIS for floating offshore wind site selection, while Shi et al. [32] combined GIS with AHP-TOPSIS for photovoltaic power plant site selection. Existing studies show that GIS combined with weighting or ranking methods has formed a relatively mature spatial suitability evaluation framework for facility site selection [33].

2.3. Data Envelopment Analysis Site Selection

Data Envelopment Analysis (DEA) is widely applied in facility-location studies as a non-parametric efficiency evaluation technique that measures the relative performance of decision-making units under multiple inputs and outputs [34,35,36]. In renewable energy planning, DEA is integrated with other MCDM techniques to enhance the robustness of site evaluation. Pambudi & Nananukul [37] applied a hierarchical fuzzy DEA to determine the most efficient wind farm sites in Indonesia. Wang et al. [38] combined DEA with grey MCDM to evaluate photovoltaic sites in Vietnam, while Saraji et al. [39] proposed a two-stage DEA–CRITIC–TOPSIS approach for solar farm selection in the Baltic region. Similarly, Shafii et al. [40] developed a GIS-based hybrid framework integrating DEA with BWM, TOPSIS, and GRA for nuclear power plant site selection in Indonesia. To further improve accuracy, several studies introduced multi-phase or hybrid DEA structures. Wang et al. [41] integrated DEA with a fuzzy MCDM framework (FAHP–F-CoCoSo) for wind energy planning in Indonesia. Raad and Rajendran [42] applied an SBM-DEA combined with regression and GIS for airport site selection, and Wang et al. [43] used a composite CI-DEA–IDOCRIW–MABAC model for optimizing logistics terminal locations. Extensions of DEA also emerged in uncertain or behavioral contexts. Memari et al. [44] introduced a fuzzy DEMATEL–ANP–ZDEA model, Karuppiah et al. [45] employed a grey DEA–AHP–COPRAS method, and Wang & Chao [46] incorporated bounded rationality into geothermal site assessment. In addition, Wang et al. [47] applied prospect theory to DEA for offshore wave–wind selection, while Sharma and Sarin [48] and Sharma et al. [49] combined DEA with TOPSIS and clustering to handle large-scale solar PV decisions.

Taken together, existing studies provide useful support for this research from three aspects: wind turbine recycling studies clarify recycling technologies, material reuse, and reverse logistics; GIS-based site-selection studies provide tools for spatial suitability screening under land-use, accessibility, and environmental constraints; and DEA-based studies offer methods for evaluating the input–output efficiency of candidate alternatives. However, these streams are usually discussed separately. For the site selection of wind turbine recycling centers, both spatial suitability and operational efficiency need to be considered because decommissioned components are large, difficult to transport, and closely related to wind farm distribution, transport accessibility, land cost, labor cost, and downstream demand. Therefore, this study integrates GIS-based candidate-area screening with Super-Efficiency DEA-based efficiency evaluation to support the selection of wind turbine recycling center locations.

3. Methodology

The proposed study follows an integrated GIS–DEA framework. In the GIS stage, spatial indicators are processed and standardized to generate suitability layers, and AHP and CRITIC are used only for weight determination within the weighted overlay analysis. Specifically, AHP is used to determine the weights of the target-layer categories, while CRITIC is used to calculate the weights of the measurable indicator layer. Based on the weighted overlay results, highly suitable candidate regions are extracted. In the DEA stage, the Super-Efficiency DEA model is used to evaluate and rank the candidate regions according to input–output efficiency. The overall technical roadmap of the proposed GIS–DEA framework is shown in Figure 2. All GIS operations were performed using ArcGIS 10.8 (Esri, Redlands, CA, USA), and the CRITIC weighting and Super-Efficiency DEA calculations were implemented in MATLAB R2023a (MathWorks, Natick, MA, USA).

Figure 2. Technical Roadmap.

3.1. GIS Analysis Methods

GIS analysis is used to process spatial data, generate indicator layers, and screen candidate sites for subsequent efficiency evaluation. Considering the characteristics of wind turbine recycling center site selection, this study incorporates natural, locational, and social factors such as wind farm distribution, topography, road accessibility, downstream customers, and land use. Multi-source spatial data are first unified through coordinate system transformation, data clipping, and format conversion, and are then converted into comparable raster layers, as shown in Table 1.

Table 1. GIS Spatial Processing Methods.

Indicator Category	Specific Indicator	Data Type	GIS Processing Method	Output
Natural Factors	Elevation	Raster data	Raster processing, Fuzzy Membership	Elevation raster
Natural Factors	Slope	Raster data	Raster processing, Fuzzy Membership	Slope raster
Natural Factors	Aspect	Raster data	Raster processing, Fuzzy Membership	Aspect raster
Natural Factors	Annual mean temperature	Raster data	Mean value calculation, Fuzzy Membership	Annual mean temperature raster
Natural Factors	Land use type	Polygon data	Reclassification, Mask processing	Land use type raster
Natural Factors	Nature reserves	Polygon data	Reclassification, Mask extraction	Nature reserve constraint layer
Natural Factors	Distance from the river	Line data	Euclidean Distance, Raster Calculator	Distance to river raster
Natural Factors	Distance from the lake	Polygon data	Euclidean Distance, Raster Calculator	Distance to lake raster
Location Factors	Distance from the road	Line data	Euclidean Distance, Fuzzy Membership	Distance to road raster
Location Factors	Distance from the railway	Line data	Euclidean Distance, Fuzzy Membership	Distance to railway raster
Location Factors	Distance from the customer point	Point data	Euclidean Distance, Fuzzy Membership	Distance to customer point raster
Location Factors	Distance from the wind farm	Point data	Euclidean Distance, Fuzzy Membership	Distance to wind farm raster
Location Factors	Distance from the airport	Point data	Euclidean Distance, Raster Calculator	Distance to airport raster
Social Factors	Land prices	Polygon data	Kernel Density, Raster Calculator	Land price density raster
Social Factors	Population density	Polygon data	Kernel Density, Raster Calculator	Population density raster

Different GIS spatial analysis methods are applied according to the data type of each indicator. For point and linear features such as wind farms, downstream customer points, roads, and airports, Euclidean distance analysis is used to generate distance rasters and measure the spatial proximity between each location and the target facilities.

$$D_i=\sqrt{{(x-x_i)}^2+{(y-y_i)}^2}$$

(1)

where $D_i$ denotes the Euclidean distance from raster cell $(x,y)$ to target feature $i$, and $(x_i,y_i)$ denotes the location coordinates of the target feature.

For clustered point or polygon data, such as densely populated areas and land prices, kernel density estimation is used to generate a spatial density surface, and its general form can be expressed as:

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

(2)

where $f(x)$ denotes the kernel density estimate at location $x$, $n$ is the number of samples, $h$ is the bandwidth, $K\left(u\right)$ is the kernel function, and $x_i$ is the location of the $i\text{-}\mathrm{th}$ sample point.

Since the selected indicators differ in units, value ranges, and suitability directions, all raster indicators are standardized before weighted overlay analysis. In this study, three types of processing are adopted. First, for positive indicators, where larger values or higher grades indicate higher suitability, Fuzzy Membership is used to convert the original raster values into standardized suitability scores ranging from 0 to 1 according to Equation (3). Second, for negative indicators, where larger values indicate lower suitability, inverse linear normalization is applied according to Equation (4), and the calculation is implemented in ArcGIS 10.8 using the Raster Calculator. Third, constraint indicators, such as nature reserves and unsuitable land-use types, are processed through reclassification and mask extraction before overlay analysis.

$$R_i={\displaystyle \frac{X_i-X_{\min }}{X_{\max }-X_{\min }}}$$

(3)

$$R_i={\displaystyle \frac{X_{\max }-X_i}{X_{\max }-X_{\min }}}$$

(4)

where $R_i$ denotes the standardized value of the indicator, $X$ denotes the original indicator value, and $X_{\max }$ and $X_{\min }$ denote the maximum and minimum values of the indicator, respectively.

After obtaining the standardized results of all indicators, weighted overlay is performed by combining them with their corresponding weights to generate a comprehensive suitability map of the study area. The calculation is as follows:

$$S={\displaystyle \sum _{i=1}^m{w}_i}{R}_i$$

(5)

where $S$ denotes the comprehensive suitability value, $w_i$ denotes the weight of the $i\text{-}\mathrm{th}$ indicator, $R_i$ denotes the standardized raster value of the $i\text{-}\mathrm{th}$ indicator, and $m$ denotes the total number of indicators.

Based on the comprehensive suitability results, highly suitable areas are further reclassified and extracted, and an initial set of candidate sites is generated through raster-to-point conversion, as shown in Table 2.

Table 2. GIS Analysis and Candidate Site Screening.

Content	Processing Object	Method	Output
Primary overlay analysis	Overlay of indicators within natural factors, location factors, and social factors	Weighted overlay based on indicator-layer weights	Natural-factor suitability map, location-factor suitability map, and social-factor suitability map
Secondary overlay analysis	Overlay of the three suitability maps	Weighted overlay based on target-layer weights	Comprehensive suitability distribution map
Candidate area extraction	Comprehensive suitability distribution map	Reclassification, extraction of highly suitable areas	Candidate area map
Candidate site generation	Candidate area map	Raster-to-point conversion, spatial screening	Candidate sites

Through this process, the site selection problem in continuous space is transformed into a screening problem over a finite set of candidate regions, thereby providing input for the subsequent Super-Efficiency DEA evaluation.

3.2. Weight Determination Method

The weighting process is conducted at two levels. First, AHP is used to determine the weights of the target layer, including natural factors, location factors, and social factors. Second, CRITIC is used to determine the objective weights of the indicator layer within each target-layer category. In the GIS overlay process, the CRITIC weights are first used to generate the suitability map of each target-layer category, and the AHP weights are then used to combine the three target-layer suitability maps into the final comprehensive suitability map.

3.2.1. AHP

AHP is used to determine the relative importance of the target-layer categories based on expert judgment [50]. In this study, the target layer includes natural factors, location factors, and social factors. The judgment matrix is constructed according to expert scoring, and the consistency test is conducted to ensure the reliability of the obtained target-layer weights.

3.2.2. CRITIC

CRITIC is used to determine the objective weights of the indicator layer within each target-layer category [51]. After the spatial indicator raster data are obtained through GIS processing, they are imported into MATLAB R2023a for CRITIC calculation. CRITIC evaluates indicator importance according to data variation and conflict among indicators, thereby improving the objectivity of indicator-layer weight determination.

The CRITIC method first standardizes the original indicator data to eliminate the influence of dimensional differences. For a positive indicator, the standardized value is calculated as

$$x_{ij}^{\prime }={\displaystyle \frac{x_{ij}-min\left(x_j\right)}{max\left(x_j\right)-min\left(x_j\right)}}$$

(6)

For a negative indicator, the standardized value is calculated as

$$x_{ij}^{\prime }={\displaystyle \frac{max\left(x_j\right)-x_{ij}}{max\left(x_j\right)-min\left(x_j\right)}}$$

(7)

where $x_{ij}$ is the original value of the $j\text{-}\mathrm{th}$ indicator for the i-th sample, and $max\left(x_j\right)$ and $min\left(x_j\right)$ are the extreme values of the $j\text{-}\mathrm{th}$ indicator across all samples.

After standardization, the contrast intensity of each indicator is measured by its standard deviation. A larger standard deviation indicates that the indicator has stronger discriminating ability across the evaluated samples and contains more information. The standard deviation of indicator $j$ is expressed as

$${\sigma }_j=\sqrt{{\displaystyle \frac{1}{d}}{\displaystyle \sum _{i=1}^d}{\left(x_{ij}^{\prime }-{\overline{x}}_j^{\prime }\right)}^2}$$

(8)

where $d$ denotes the sample size, and ${\overline{x}}_j^{\prime }$ represents the mean of the standardized values of indicator $j$.

To reflect the degree of conflict among indicators, the Pearson correlation coefficient is introduced. If an indicator has a lower correlation with the others, it provides more independent information and should therefore be assigned a higher weight. The correlation coefficient between indicators $j$ and $k$ is given as

$$r_{jk}={\displaystyle \frac{{\displaystyle \sum _{i=1}^d}\left(x_{ij}^{\prime }-{\overline{x}}_j^{\prime }\right)\left(x_{ik}^{\prime }-{\overline{x}}_k^{\prime }\right)}{\sqrt{{\displaystyle \sum _{i=1}^d}{\left(x_{ij}^{\prime }-{\overline{x}}_j^{\prime }\right)}^2}\sqrt{{\displaystyle \sum _{i=1}^d}{\left(x_{ik}^{\prime }-{\overline{x}}_k^{\prime }\right)}^2}}}$$

(9)

Based on the contrast intensity and conflict degree, the information content of indicator $j$ is calculated as

$$C_j={\sigma }_j{\displaystyle \sum _{k=1}^m}\left(1-\left|r_{jk}\right|\right)$$

(10)

where $m$ denotes the total number of indicators.

Finally, the information content is normalized to obtain the objective weight of each indicator:

$$w_j={\displaystyle \frac{C_j}{{\displaystyle \sum _{k=1}^m}{C}_k}}$$

(11)

where $w_j$ denotes the weight of indicator $j$. Through this procedure, indicators with greater variation and lower redundancy are assigned relatively higher weights. In this study, the CRITIC-derived indicator-layer weights are further combined with the target-layer weights obtained from the subjective weighting method, so as to form the complete weighting system for subsequent GIS weighted overlay analysis. The resulting weights are then used to generate the comprehensive suitability evaluation results of candidate regions.

3.3. Super-Efficiency DEA

After the GIS-based suitability analysis, highly suitable areas are extracted and treated as candidate regions for further evaluation. Although GIS can identify spatially favorable areas, it cannot determine the relative efficiency of candidate regions in terms of input–output performance. Therefore, a Super-Efficiency DEA model is introduced to rank the candidate regions and identify the optimal location for the wind turbine recycling center. In this study, each candidate region is regarded as a decision-making unit (DMU), with land cost and labor cost taken as input variables and annual decommissioning quantity taken as the output variable.

The Super-Efficiency DEA model is an extension of the traditional DEA framework. Its main advantage is that the evaluated DMU is excluded from the reference set, allowing efficient units to obtain scores greater than 1 and thus enabling further ranking among efficient candidates. This feature is particularly important in the present study, because multiple candidate regions may all appear efficient after the initial GIS screening [52].

For DMU $k$, the input-oriented Super-Efficiency DEA model can be expressed as follows:

$$max\varphi$$

(12)

$${\displaystyle \sum _{\begin{array}{c}j=1\\ j\ne j_0\end{array}}^n}{\lambda }_j{z}_{ij}\le z_{i{j}_0},i=1,2,\dots ,m$$

(13)

$${\displaystyle \sum _{\begin{array}{c}j=1\\ j\ne j_0\end{array}}^n}{\lambda }_j{y}_{rj}\ge \varphi y_{r{j}_0},r=1,2,\dots ,s$$

(14)

$${\lambda }_j\ge 0,j=1,2,\dots ,n,j\ne j_0$$

(15)

where $n$ represents the number of DMUs, each of which has $m$ types of inputs and $s$ types of outputs, $z_{ij}$ denotes the quantity of the $i\text{-}\mathrm{th}$ input for the $j\text{-}\mathrm{th}$ DMU $(i=1,2,\dots ,m,j=1,2,\dots ,n)$, $y_{rj}$ refers to the quantity of the $r\text{-}\mathrm{th}$ output for the $j\text{-}\mathrm{th}$ DMU $(r=1,2,\dots ,s,j=1,2,\dots ,n)$, and the efficiency of the $j_0\text{-}\mathrm{th}$ DMU is to be evaluated.

If $\varphi >1$, the DMU is super-efficient and can be further ranked; if $\varphi =1$, the DMU lies on the efficiency frontier; if $\varphi <1$, the DMU is relatively inefficient. Based on the calculated Super-Efficiency values, the candidate regions are ranked, and the region with the highest value is identified as the optimal location for the wind turbine recycling center. This step further refines the GIS screening results from the perspective of input–output efficiency.

4. Case Study and Results

4.1. Overview of the Study Area

Xilingol is located in central Inner Mongolia, with an average annual wind speed of 3.2 m/s. It has wind energy reserves of more than 50 million kW, accounting for approximately 50% of the total in Inner Mongolia and 20% of the national total. Its exploitable wind energy capacity also exceeds 50 million kW (Figure 3). Figure 3 is displayed using the GCS_WGS_1984 geographic coordinate system.

Figure 3. Location Map of the Study.

Xilingol is rich in wind and solar resources and is one of the few Class I high-quality wind and solar resource areas in China. However, with the coming large-scale retirement of wind turbines, Xilingol still lacks specialized and localized recycling and processing centers for decommissioned wind turbines. Therefore, selecting an appropriate location for a wind turbine recycling center in Xilingol has important practical significance.

4.2. Data and Indicators

4.2.1. Source and Processing of Data

A large number of raster and vector data are involved in the GIS-based location analysis. Among them, digital elevation data (e.g., slope and aspect) are from the Geospatial Data Cloud; the annual mean temperature data were obtained from WorldClim at approximately 1 km resolution and represent long-term monthly averages for 1970–2000. The annual mean temperature layer was calculated by averaging the twelve monthly rasters; administrative area planning data are from the National Geographic Information Public Service Platform; land use types, rivers, and lakes data are from the National Earth System Science Data Center; road network data are from OpenStreetMap; and coordinate information of airports, wind farms, and customer points are from the Amap Open Platform and Baidu Map Open Platform. All spatial data were unified under GCS_WGS_1984 and projected into WGS 1984 UTM Zone 50N for spatial analysis (Table 3).

Table 3. Data Sources.

Data	Source	Year/Period	Spatial Resolution/Scale
Elevation, Slope, Aspect	Geospatial data cloud	Static topographic data	30 m
Annual mean temperature	WorldClim	1970–2000	Approx. 1 km
Administrative District Planning	National Geographic Information Public Service Platform	2020	Vector data
Land use types	GlobeLand30	2020	30 m
Rivers and lakes	National Earth System Science Data Center	2020	Vector data
Road Network Data	OpenStreetMap	Accessed in 2025	Vector data
Airports, wind farms, customer points	Amap, Baidu Map Open Platform	Collected in 2025	Point data

All raster layers were unified to a common spatial resolution of 1 km for the weighted overlay analysis.

4.2.2. Indicator System Construction

The site selection of wind turbine recycling centers is similar to the site selection of solid waste recycling and treatment facilities, as both require consideration of construction conditions, transportation accessibility, environmental constraints, and social compatibility. Based on the literature review and the characteristics of wind turbine recycling activities, this study constructs a hierarchical indicator system for GIS-based suitability analysis. The Triple Bottom Line (TBL) principle is used only as an auxiliary basis for organizing the target layer and classifying the corresponding indicator layer [53,54]. The target layer includes natural factors, location factors, and social factors, and the indicator layer translates these categories into measurable spatial variables for GIS processing. Table 4 summarizes the selected indicators and their rationale.

Table 4. Wind Turbine Recycling Center Location Indicators.

Target Layer	Indicator Layer	Rationale	Suitability Direction
Natural Factors	a. Elevation (m)	Elevation affects construction difficulty, equipment transportation, and operational conditions; excessively high elevation may reduce construction suitability.	Negative
	b. Slope (°)	Slope affects site leveling, construction stability, and the transportation of large wind turbine components.	Negative
Natural Factors	c. Aspect (°)	Aspect affects sunlight exposure, drainage, wind exposure, and local microclimate conditions related to site construction and operation.	Negative
	d. Annual mean temperature (°C)	Temperature is used as an auxiliary environmental suitability indicator. Relatively lower temperature conditions can help reduce heat-related operational pressure during cutting, crushing, storage, and equipment operation.	Negative
	e. Land use type	Land use type determines industrial construction compatibility and helps avoid conflicts with cultivated land, forest land, residential land, and water bodies.	Constraint
	f. Nature reserves	Nature reserves should be excluded to avoid ecological disturbance and meet environmental protection requirements.	Constraint
	g. Distance from the river (km)	Greater river distance reduces potential risks of wastewater leakage, runoff pollution, and surface-water impacts.	Positive
	h. Distance from the lake (km)	Greater lake distance reduces ecological disturbance and potential water-pollution risks.	Positive
Location Factors	i. Distance from the road (km)	Shorter road distance improves transportation accessibility and reduces logistics costs.	Negative
	j. Distance from the railway (km)	Railway accessibility supports medium- and long-distance transport of heavy or oversized components.	Negative
	k. Distance from the customer point (km)	Customer-point distance reflects the outbound transport cost and market response efficiency of recycled products.	Negative
	l. Distance from the wind farm (km)	Wind farms are the main source of decommissioned wind turbine components. Shorter distance can reduce inbound collection and transportation costs.	Negative
	m. Distance from the airport (km)	Recycling centers should avoid airport-adjacent areas because of airspace control, land-use restrictions, and aviation safety requirements. Greater airport distance improves spatial compatibility.	Positive
Social Factors	n. Land prices (CNY/m2)	Land price affects land acquisition and initial construction investment. Lower land price can improve the economic feasibility of recycling-center development.	Negative
	o. Population density (persons/km2)	Population density reflects surrounding human-activity intensity. Lower population density can reduce potential conflicts related to noise, heavy-vehicle traffic, safety risk, and environmental exposure.	Negative

The suitability direction of each indicator is defined according to its expected influence on the GIS-based suitability evaluation. Positive indicators indicate that larger values or more favorable standardized conditions improve site suitability, whereas negative indicators indicate that larger values reduce site suitability or increase construction, operational, environmental, or social constraints. Constraint indicators, such as land use type and nature reserves, are used to restrict or exclude unsuitable areas before weighted overlay analysis. Through standardization, all indicators are converted into comparable suitability layers, and areas with higher comprehensive suitability values are retained as candidate areas for subsequent evaluation.

4.3. GIS-Based Suitability Analysis Results

4.3.1. Spatial Distribution of Original Indicators

The collected metric data are processed and analyzed in ArcGIS. First, the coordinate system of the distance type indicators is converted to a projection coordinate system. The longitude and latitude coordinates of major wind farms in Xilingol, downstream customer points for wind turbine product recycling, and the airport are collected, and Euclidean distance analysis is performed to obtain corresponding indicator maps. Then, the locations, population and land prices of each city in Xilingol are collected. Kernel density analysis is conducted to generate population density and land price heat maps. Finally, the processed indicator maps are compiled to prepare for the next step. As shown in Figure 4a–h, the natural-factor maps include elevation, slope, aspect, annual mean temperature, land use type, nature reserves, distance from the river, and distance from the lake. Figure 4i–m present the location-factor maps, including distance from the road, railway, customer point, wind farm, and airport, while Figure 4n–o show the social-factor maps of land price and population density.

Figure 4. Spatial Information Map of Indicators.

4.3.2. Weight Determination Results

The weighting process is conducted at two levels. First, AHP is used to determine the weights of the target layer, including natural factors, location factors, and social factors. Second, the CRITIC method is used to determine the objective weights of the indicator layer within each target-layer category. In this way, expert judgment is used to reflect the relative importance of broad site-selection dimensions, while data-driven weighting is used to measure the information content and differentiation ability of measurable indicators.

Subjective Weighting Method (Analytic Hierarchy Process)

AHP is used to calculate the weights of the target layer, including natural factors, location factors, and social factors. Experts were selected based on their academic background and professional experience in logistics systems, GIS spatial analysis, environmental assessment, circular economy, industrial resource management, wind energy technology, and sustainable logistics. Their profiles, including gender, degree, and professional field, are presented in Table 5.

Table 5. Profile of experts.

Expert	Gender	Degree	Field
1	Female	Ph.D.	Logistics Systems Engineering
2	Female	Ph.D.	Circular Economy & Logistics
3	Male	Ph.D.	GIS & Spatial Analysis
4	Female	Ph.D.	Environmental Impact Assessment
5	Male	Ph.D.	Industrial Resource Management
6	Male	M.Sc.	Wind Energy Technology & Maintenance
7	Female	Ph.D.	Sustainable Logistics Assessment

The judgment matrix constructed based on the scores from invited experts is shown in Table 6. The 1–9 scale method was used for scoring. The normalized weight table is calculated as presented in Table 7, and the results of the consistency test are given in Table 8.

Table 6. Judgment Matrices.

Factors	Natural Factors	Location Factors	Social Factors
Natural Factors	1	2	4
Location Factors	1/2	1	2
Social factors	1/4	1/2	1

Table 7. Normalized Weights.

Factors	Weight
Natural Factors	0.571
Location Factors	0.286
Social Factors	0.143

Table 8. Consistency Test.

Factors	Weight
$\mathrm{Maximum}\mathrm{eigenvalue}({\lambda }_{\mathrm{m}\mathrm{a}\mathrm{x}}$)	3.006
Consistency index (CI)	0.003
Random consistency indicator (RI)	0.580
Consistency ratio (CR)	0.005

The consistency ratio of 0.005, which is less than 0.1, passes the consistency test.

The CRITIC Method Determines the Weights

Based on the spatial information raster map of indicators obtained in Section 4.3.1, the raster data are imported into MATLAB R2023a and the CRITIC method is implemented. Figure 5a–c show the weight distributions of natural factors, location factors, and social factors, respectively, while Figure 5d presents the correlation coefficient heatmap. CRITIC determines the indicator-layer weights, which are combined with the AHP-derived target-layer weights to obtain the final comprehensive weights shown in Table 9.

Figure 5. Plot of Weights and Correlation Coefficients.

Table 9. Weight Table.

Target Layer	Weights	Indicator Layer	Weights
Natural Factors	0.571	Elevation (m)	0.2346
		Slope (°)	0.1833
		Aspect (°)	0.1417
		Annual mean temperature (°C)	0.1475
		Distance from the river (km)	0.1176
		Distance from the lake (km)	0.1754
Location Factors	0.286	Distance from the road (km)	0.2874
		Distance from the railway (km)	0.1246
		Distance from the customer point (km)	0.1398
		Distance from the wind farm (km)	0.2390
		Distance from the airport (km)	0.2091
Social Factors	0.143	Land price (CNY/m2)	0.5079
		Population density (persons/km2)	0.4921

The correlation coefficient heatmap shows that most selected indicators have low correlations, while only a few indicators show relatively high correlations. However, these indicators represent different spatial constraints and operational considerations in the site-selection process. Therefore, they were retained to preserve the completeness of the indicator system.

4.3.3. Normalization and Weighted Overlay Results

Because the indicators differ in dimensions and suitability directions, all indicator layers were standardized in ArcGIS before weighted overlay analysis. Positive indicators were standardized using Fuzzy Membership, while negative indicators were standardized by inverse linear normalization implemented through the Raster Calculator. The normalized indicator maps are provided in Appendix A, Figure A1. Normalization only rescales the indicators to comparable 0–1 suitability scores without changing their spatial patterns.

The weighted overlay results of the social, location, and natural factors, together with the final weighted superposition graph, are shown in Figure 6a–d, respectively.

Figure 6. Weighted overlay results of target layers and final suitability map.

4.3.4. Extraction Results of Highly Suitable Areas

Extraction of Highly Suitable Areas

The final suitability map was reclassified into nine grades using the equal-interval classification method, with higher grades representing higher suitability. Since the GIS stage is used for candidate-area screening rather than final site determination, only retaining the highest grade may exclude potentially feasible areas too early. Therefore, grades 8 and 9 were retained as highly suitable areas. This threshold ensures that the extracted areas have high spatial suitability while maintaining a sufficient number of candidate regions for the subsequent Super-Efficiency DEA evaluation, as shown in Figure 7.

Figure 7. Extraction of highly suitable areas.

Generation and Screening of Final Candidate Points

The extracted highly suitable raster regions were converted into point features and overlaid with the county-level administrative boundary, nature reserve, and land-use layers. Points located within strictly protected nature reserves and unsuitable land-use areas, such as residential zones and water bodies, were excluded. The remaining points were identified as the final candidate points in highly suitable areas (Figure 8).

Figure 8. Final candidate points in highly suitable areas.

4.4. Super-Efficiency DEA Results

Based on the distribution map of highly suitable areas obtained in the previous section and the map of Xilingol, it is found that the wind turbine recycling centers can be established in East Ujimqin, West Ujimqin, Erenhot, Sonid Right, Sonid Left, Abag, and other areas. Labor cost and land cost are selected as inputs, and the annual wind turbine retirement volume is selected as the output for DEA efficiency evaluation. Finally, the most suitable area for establishing a wind turbine recycling center is obtained.

Potential Site Efficiency Assessment

The prices of Class II industrial land are collected for DMUs including East Ujimqin, West Ujimqin, Erenhot, Sonid Right, Sonid Left, and Abag, as the establishment of the wind turbine recycling centers falls under Class II industrial land. The monthly minimum labor price and the annual decommissioning quantity of wind turbines (calculated based on the number of wind turbines in each region) serve as input data (Table 10), and the Super-Efficiency DEA analysis is implemented using MATLAB R2023a, and the final efficiency values are shown in Figure 9.

Table 10. Input–Output Table.

Decision Unit	Inputs		Output
Region	Land Costs (CNY/m2)	Labor Costs (CNY/month)	Annual Decommissioning Quantity of Wind Turbines (MW)
East Ujimqin	95	1910	857
West Ujimqin	125	1910	583
Erenhot	288	2270	3426
Sonid Right	120	1910	1363
Sonid Left	123	1910	2043
Abag	116	1910	1731

Figure 9. Alternative Site Efficiency Values.

The Super-Efficiency values of each region are ranked as shown in Table 11. The results indicate that West Ujimqin achieves the highest Super-Efficiency score (3.528), followed by East Ujimqin (1.841), Sonid Right (1.462), Abag (1.113), Sonid Left (0.879), and Erenhot (0.709). In the Super-Efficiency DEA model, values greater than 1 denote that a DMU surpasses the efficiency frontier and can be ranked, whereas values below 1 indicate relative inefficiency.

Table 11. Ranking of Super-Efficiency Values.

Regions	Efficiency Value
West Ujimqin	3.528
East Ujimqin	1.841
Sonid Right	1.462
Abag	1.113
Sonid Left	0.879
Erenhot	0.709

4.5. Sensitivity Analysis

To evaluate the robustness of the weighting results, a sensitivity analysis was conducted by adjusting the target-layer weights. The original weights were used as the baseline scenario. In each alternative scenario, the weight of one target-layer category was increased or decreased by 10%, while the remaining weights were adjusted proportionally to keep the total weight equal to 1.

For each scenario, the weighted overlay analysis was repeated in ArcGIS, and highly suitable areas were extracted using the same reclassification criteria. The initial candidate points generated from these areas were compared before spatial infeasibility removal. The number of overlapping initial candidate points and the spatial distribution of highly suitable areas were used to evaluate the stability of the results.

As shown in Table 12, the number of initial candidate points ranged from 41 to 54 under different weighting scenarios, while the overlap rate with the baseline scenario remained between 86.96% and 100.00%. This indicates that moderate changes in the target-layer weights did not substantially affect the spatial distribution of highly suitable areas, suggesting acceptable robustness of the GIS-based suitability results.

Table 12. Sensitivity scenarios of target-layer weights.

Scenario	Adjustment	NF	LF	SF	ICP	OP	OR (%)
S0	Baseline	0.571	0.286	0.143	46	46	100.00
S1	Natural +10%	0.628	0.248	0.124	51	44	95.65
S2	Natural −10%	0.514	0.324	0.162	41	40	86.96
S3	Location +10%	0.548	0.315	0.137	42	41	89.13
S4	Location −10%	0.594	0.257	0.149	54	46	100.00
S5	Social +10%	0.561	0.281	0.157	47	45	97.83
S6	Social −10%	0.581	0.291	0.129	46	45	97.83

Note: NF, LF, and SF denote natural factors, location factors, and social factors, respectively. ICP denotes the number of initial candidate points. OP denotes the number of overlapping initial candidate points with the baseline scenario. OR denotes the overlap rate with the baseline scenario.

4.6. Discussion

In the previous section, the efficiency ranking of each region is determined using the Super-Efficiency DEA model. Among all regions, West Ujimqin achieves the highest efficiency value (3.528), indicating the greatest potential for the construction of a wind turbine recycling center. Therefore, West Ujimqin is selected as the candidate region for establishing the recycling center.

Several potential sites are identified in West Ujimqin (Figure 10), and their latitude and longitude coordinates are determined using the ArcGIS Calculate Geometry tool (Table 13). The final decision on the number of recycling centers to be constructed in this region should further consider regional policy guidelines and practical development needs.

Figure 10. Optimal Candidate Region.

Table 13. Candidate Sites’ Coordinates.

Regions	Candidate Sites’ Longitude	Candidate Sites’ Latitude
West Ujimqin	118.550679	44.906181
	118.749583	45.105085

West Ujimqin is identified as the preferred region because it is included in the highly suitable areas extracted through GIS screening and obtains the highest Super-Efficiency DEA value among the candidate regions. The weighting results show that natural factors receive the highest target-layer weight, followed by location factors and social factors. At the indicator layer, elevation, slope, distance from road, distance from wind farm, and land price receive relatively high weights within their respective categories. This indicates that the final result is consistent with the spatial suitability logic of the weighting process and is further supported by the DEA efficiency ranking. The government can select one of the two candidate sites within West Ujimqin as the construction location based on local conditions (Table 13).

5. Conclusions

This study proposed an integrated site-selection method for wind turbine recycling centers by combining GIS-based spatial suitability analysis, AHP–CRITIC weighting, and Super-Efficiency DEA.

(1) In the initial GIS-based site selection stage, a hierarchical indicator system was constructed using natural factors, location factors, and social factors. Euclidean distance analysis, kernel density analysis, weighted overlay, reclassification, and raster conversion were then used to identify highly suitable candidate regions in Xilingol. (2) In weight calculation for weighted overlay, AHP and the CRITIC methods were applied to the target layer and indicator layer, respectively. AHP reflects expert judgment at the target-layer level, while CRITIC captures data variation and conflict at the indicator-layer level, thereby improving the objectivity of the weighting process. (3) For the candidate sites shortlisted in the initial screening, the Super-Efficiency DEA model, which addressed the limitation of traditional DEA models where efficiency was capped at one, was employed to rank their operational efficiency. From the ranking results, West Ujimqin emerges as the optimal choice, with an efficiency score of 3.528, due to its favorable input–output efficiency under the Super-Efficiency DEA model.

Furthermore, to facilitate the practical implementation of wind turbine recycling centers, the study accurately obtained the coordinate information of the optimal site by using the Calculate Geometry tool in GIS. This provided a clear geographical reference for subsequent engineering planning and construction. Theoretically, this study constructed a multi-method integrated scientific site selection system, offering a methodological paradigm for similar infrastructure site selection research. Practically, the model effectively filled the gap in site selection critical infrastructure within decommissioned wind turbine recycling networks, providing solid support for future wind turbine recycling.

Although this study provides a systematic framework for wind turbine recycling center site selection, it does not fully consider the temporal dynamics of decommissioning volumes. Future research will incorporate time-series prediction for dynamic planning and offer policy recommendations for sustainable recycling development.