Construction feasibility evaluation for potential ecological corridors under different widths: a case study of Chengdu in China

Accelerating urbanization has led to the shrinkage of natural habitats, intensification of landscape fragmentation, reduction of cropland and forest areas, and degradation of biodiversity. Against this background, considering ecological corridors are important linear structures connected to habitat patches, helping enhance urban ecological resilience and protect biodiversity. A growing number of studies focus on identifying potential ecological corridors and evaluating the relative importance of such corridors on the effects of connected habitat patches. However, in the evaluation process, previous studies ignored the linear structures of corridors, and did not explore the internal characteristics differences under different corridor widths. Accordingly, this study used the minimum cumulative resistance model (MCRM) to identify potential ecological corridors. After that, we designed the construction feasibility evaluation system for evaluating the relative importance of such corridors with different widths, and applied it to the rapidly expanding city of Chengdu in western China. The results show that internal characteristics of potential ecological corridors vary significantly under different widths. Using the construction feasibility evaluation system, we found that more than 75% of the identified potential ecological corridors are difficult to construct. The construction sequence has obvious differences under different widths, besides providing a reference for construction and selecting widths for potential corridors in Chengdu. This research highlights the importance of width selection in corridor evaluation. The construction feasibility evaluation provided reference information on identification, analysis, construction sequence, and construction practices for urban ecological network spatial planning.


Introduction
Since the early twenty-first century, urban expansion has been particularly prominent (Angel et al. 2011;Connop et al. 2016). Urban expansion is the continuous depletion of land resources by intense human activity (Li et al. 2015a, b;Pauleit et al. 2005), which results in the gradual replacement of the original landscape dominated by natural vegetation with buildings and concrete surfaces (Zhao et al. 2012). This phenomenon is particularly prevalent in developing countries where urbanization rates are soaring, such as low-and middle-income nations in Asia, Africa, and Latin America (Li et al. 2015a, b;Lorenz and Lal 2009;Pauleit et al. 2005). Due to high density, rapid growth in urban populations, and constantly rising consumption levels, remaining urban green spaces (UGS) such as cropland, forest, and grassland are at greater risk of erosion and fragmentation (Kong et al. 2010;Li et al. 2015a, b).
The ecological network concept is often used in UGS conservation, focusing on ecology and providing UGS with a connected and networked structure (Harrison and Bruna 1999;Mackovcˇin 2000). The main elements of ecological networks are ecological corridors and core areas (Mackovcˇin 2000). Core areas are distributed within urban areas as habitat patches, which provide resources and sites for various ecological processes and species (Peng et al. 2018). As cities expanded, habitat patches have become relatively isolated, weakening the flow of information between species (Kong et al. 2010;Huang et al. 2021). Ecological corridors serve as links between habitat patches, facilitating animal migration, enhancing the flow of ecological information between habitat patches, and enhancing urban ecological resilience (Li et al. 2015a). In contrast to habitat patches, corridors do not have obvious boundaries or characteristics in cities. Meanwhile, linear corridor protection, construction, and evaluation are further complicated due to uncertainties regarding ideal length and width and the influence of resources on development and planning (Kong et al. 2010;Dong et al. 2020). It is, therefore, complex, but necessary, to research the development of urban ecological corridors.
In the 1980s and 1990s, a number of large-scale ecological corridor projects were pioneered in the United States and European countries (Peng et al. 2017). The construction in both countries targets achieving harmonious and sustainable development of humans and nature in cities (Peng et al. 2017). Recently, ecological corridor construction has been successively carried out and has proved its utility in both developed and developing countries (Teng et al. 2011). However, compared to American and European cities, relevant research in other regions, such as Asian cities, is still inadequate (Teng et al. 2011;Peng et al. 2017). As one of the most rapidly urbanizing countries, ecological corridor construction has received extensive attention in Chinese cities (Zhang and Wang 2006;Peng et al. 2017), like Beijing, Chengdu, and Shanghai, of which the master planning all highlight urban ecological corridor construction and protection (Beijing Municipal Commission of Planning and Natural Resources 2022; Chengdu Municipal Bureau of Planning and Natural Resources 2022; Shanghai Municipal People's Government 2022), and the construction of urban ecological corridors has become an important method to enhancing urban ecosystem services and mitigating natural disasters (Dong et al. 2015;Peng et al. 2017).
Currently, ecological corridor construction methods have been transformed from qualitative to quantitative and to spatial analysis (Peng et al. 2017). Minimum cumulative resistance models (MCRM) are often used to identify critical nodes and corridors in recent studies (Knaapen et al. 1992;McRae et al. 2008;Cushman et al. 2009;Li et al. 2015a). The MCRM model can identify the geological distribution of potential ecological corridors (PEC), whilst few studies have evaluated the feasibility of corridor construction in conjunction with the internal characteristics of PEC. In order to determine the current characteristics of PEC, the width of corridor need to be considered.
While urban ecological corridors have multiple functions (e.g., habitat, conduit, filter, barrier, source, and sink) (Hess and Fischer 2001), the width of ecological corridor is currently discussed more frequently in the field of biological conservation (Bolger et al. 2001;Hilty and Merenlender 2004;Červinka et al. 2013). For example, Ford et al. (2020) tested effective corridor width in Bow Valley (Canada), which concluded that people would negatively impact the movement of most species in corridors less than 400 m wide. Ives et al. (2011) investigated riparian corridor width influences on the diversity and community structure of ants in northern Sydney. They suggested that environmental managers should seek to retain riparian corridors wider than 50 m to minimize the impact of deleterious edge effects. Sieving et al. (2000) surveyed 24 forested corridors in Chile and found that birds were rarely found in corridors 10 m wide but were always present in corridors 25-50 m wide. Different species require different corridor widths to move and migrate (Zhu et al. 2005), which provides a reference for selecting ecological corridor widths for this study.
Considering the above, to explore the importance of corridor width in corridor evaluation, this study used MCRM to identify PECs. After that, we designed a construction feasibility evaluation system for evaluating the relative importance of corridors with different widths. The objectives of this study were: (1) to explore the differences in internal characteristics for PECs under different widths, (2) to construct an evaluation system for the feasibility of PEC construction based on the linear characteristics of corridors, and to discuss the dependence of relative importance for PECs on width, (3) to propose a corridor width selection method combining biological needs and construction characteristics based on PEC, (4) to analyze the construction sequence for PECs in multi-scenario, and to discuss the differences of construction measures under different widths.

Core patches identification
In landscape ecology theory, patch area and landscape connectivity index are important factors in maintaining landscape ecological function, with the former directly determining the latter and related characteristics (Fu et al. 2001). Landscape connectivity is also an important indicator in evaluating landscape patterns and functions (Saura and Pascual-Hortal 2007;Saura et al. 2011). The probability of connectivity (PC) is considered an important index reflecting the ecological function of habitat patches. It facilitates the evaluation of landscape connectivity levels between patches (Saura et al. 2011;Saura and Pascual-Hortal 2007). Accordingly, the patch area and the PC index were selected as the basis for patch importance classification. The importance of each patch's connectivity can be determined by the delta 1 3 of PC (dPC) value, which is the percentage of the variation in PC(dPC remove ) caused by the removal of each individual patch from the landscape (Formula 2). Patches with a larger dPC index and area were selected (Saura and Rubio 2010) (http:www. conef or. org/).
Here, PC represents the connection index value of all patches in the landscape, n represents the total number of patches in the landscape, a i and a j represent the areas of patch i and patch j, p ij represents the number of connections between patch i and patch j, and A L represents the size of the study area.
Here, dPC represents the change in landscape connectivity after a patch is removed, and is used to evaluate the importance of the patch for maintaining landscape connectivity. PC remove represents the connectivity index value of the remaining patches after one is removed.

Resistance surface construction
"Landscape resistance" refers to the ease with which species migrate between different landscape units . Landscape resistance surface construction requires consideration of the type composition, the spatial configuration of the landscape unit, and its ecosystem function. Accordingly, ( it is necessary to evaluate the ecological function values of landscape type, landscape pattern, built-up area, and road distribution. Based on previous studies (Schadt et al. 2002;Gurrutxaga et al. 2010;Yang et al. 2018;Xie et al. 2021), the five resistance factors selected in this study were land use type, slope, elevation distance from the main road and distance from the main town. Land use weights were calculated using landscape pattern index values and the equivalent value of ecosystem service (Dong et al. 2015;Li et al. 2015a, b;McGarigal and Marks 1995). Landscape pattern index values were selected based on the principle of non-duplication of information with reference to index classification and the correlation coefficient matrix method of factor analysis. Percentage of landscape (%PLAND), patch density (PD), edge density (ED), landscape shape index (LSI), shannon's diversity index (SHDI), and contagion index (CONTAG) proposed by McGarigal and Marks (1995) were used ( Table 1). The equivalent value of ecosystem services was determined using the Costanza et al. (1997) method for estimation of ecosystem service value based on to the Chinese ecosystem service value table, with values per unit area compiled by Xie et al. (2015). Using the entropy method, land use type resistance weights were determined (Yang et al. 2018). The other indices were treated with Natural Breaks Classification. Different levels were assigned integers ranging from 1 to 5, with larger values representing greater landscape ecological resistance. After several data simulations, and with reference to previous studies, further, through ArcGIS superposition analysis, five single-layer indicators (land use type, elevation, slope, distance from main road, and distance from main town) are superimposed to generate ecological resistance surfaces Largest patch index (LPI) LPI Approaches 0 when the largest patch of the corresponding patch type becomes increasingly smaller. LPI = 100 when the entire landscape consists of a single patch of the corresponding patch type Percentage of landscape (%PLAND) %PLAND Approaches 0 when the corresponding patch type becomes increasingly rare in the landscape. = 100 when the entire landscape consists of a single patch type Patch density Patch density (PD) The number of patches of the corresponding patch type divided by total landscape area Edge Edge density (ED) ED = 0 when the entire landscape and border, if present, consists of the corresponding patch type and the user specifies that none of the landscape boundary and background edge be treated as edge Shape Landscape shape index (LSI) LSI = 1 when the landscape consists of a single patch of the corresponding type and is square; LSI increases without limit as landscape shape becomes more irregular or as the length of edge within the landscape of the corresponding patch type increases, or both Diversity Shannon's diversity index (SHDI) SHDI = 0 when the landscape contains only 1 patch. SHDI increases as the number of different patch types increase or the proportional distribution of area among patch types becomes more equitable, or both Contagion and interspersion Contagion index (CONTAG) CONTAG Approaches 0 when the distribution of adjacencies among unique patch types becomes increasingly uneven. CONTAG = 100 when all patch types are equally adjacent to all other patch types according to the weight values of 0.47, 0.08, 0.11, 0.13 and 0.21 (Xie et al. 2021;Yang et al. 2018).

Corridor identification based on the MCRM
The MCRM fully considers source point, spatial distance and resistance surface, reflecting the cost of movement between different core patches (Dong et al. 2015;Costanza et al. 1997). It stimulates the optimal path for organisms to cross different landscape patches while effectively avoiding various external disturbances. The results simulate the potential for and trends of species movement (Dai et al. 2021). The formula is: Here, f is a function of the positive correlation, min denotes the minimum value of cumulative resistance produced in different processes of the landscape unit i transforming into a different source unit j, D ij is the spatial distance between landscape unit i and source unit j, and R i denotes the resistance coefficient in transition from landscape unit i to source unit j.

Corridor width determination
Currently, discussions on ecological corridor width often focus on specific functions, such as windbreak, biodiversity conservation, and recreation (Zhao et al. 2008;Ives et al. 2011;Peng et al. 2017). Based on different species require different corridor widths, and the purposes of the study and the principle of operability, from the perspective of biodiversity conservation, we discussed the width of ecological corridors. At the same time, considering the differences in the needs of multiple species in different regions, with reference to the width of the biodiversity conservation corridor summarized by Zhu et al (2005), a PEC was assigned a value based on four different widths with a variety of ecological services and benefits. (Falcy and Estades 2007;Zhu et al. 2005) ( Table 2).

Index selection
The establishment of PECs requires gradual implementation with comprehensive consideration of various factors, such as the necessity of corridor development, construction difficulty, maintenance, and ecological benefits. Accordingly, based on the four widths of potential ecological corridors, we proposed a corridor construction feasibility evaluation system for consideration of the corridor's internal composition in combination with the characteristics of patches connected at both ends of the corridor (Table 3).

Internal composition characteristics
(1) Length and internal land use area Mutual overlap occurs in PEC locations, but this study aimed to explore differences between single corridors and overlapping areas also need to be constructed separately in practice. Accordingly, PECs were extracted one by one, and to facilitate data statistics, corridors No. 1-2 (between patch 1 and patch 2), No. 1-3, and No. 1-4 were renumbered as 1, 2, 3, and 4. And then, the single corridor was imported into ArcGIS one by one and given widths of 30, 60, 200, and 600 m using the buffer tool. Extracting and calculating length, built-up area, and water area in internal land use types for PECs. To determine the variation between the built-up area and the waterbody under each corridor width, R 4.0.3 was used to calculate the coefficient of variation (CV) at each width. CV represents the ratio of the standard deviation to the mean. The higher the CV, the greater the dispersion.
(2) Gravitational value The gravity model originates from Newton's Law of Gravity, with the calculation based on the mutual attraction of corridors and the resistance of ecological sources (Geymen and Baz 2007). Here, pairs of nodes with higher habitat quality and lower impedance have greater interaction. The equations of the gravity model were adapted from Linehan et al. (1995) and Rudd et al. (2002) as follows: G ab is the interaction between a and b, N a and N b are the corresponding weights, and D ab is the normalized cumulative impedance. P i is the node weight, and S i is the Establish suitable width to protect biodiversity 600 Create natural species-rich landscape structures normalized patch size of node i. L ab is the cumulative impedance of corridor L between nodes a and b. L max is the maximum value of calculated impedance.
(3) Patch integration difficulty coefficient index (IDCI) Some authors applied the moving window algorithm in Fragstats (Liu and Guo 2009;Zhang et al. 2014) to clarify the graphical representation of the landscape pattern index. Combining index meanings and differences of the index, to this end, ED, PD, CONTAG, LPI, LSI, and SHDI were selected (Table 3). In the moving window tool, the window radius selection is an important factor determining the accuracy of landscape spatial pattern information (Abdullah and Nakagoshi 2006; Liu and Guo 2009). To avoid data processing problems caused by non-integer pixels, an odd multiple of the minimum grid precision for the remote sensing grid precision was adopted as the radius size of the moving window. The landscape pattern index was calculated by moving windows with different side lengths. Using ArcGIS data management tools to construct fishnet map and regional analysis to calculate each grid's average landscape pattern index, horizontal and vertical sections of the net were selected as transects for gradient analysis, and the radius of the moving window was determined from the results. Once the window radius had been determined, a spatial visualization map of six landscape indicators in the study area was created. Weight values for land use types were calculated using the entropy method, and an evaluation distribution map for IDCI was created via superposition-extraction.

Corridor evaluation
The indicators in the feasibility evaluation system have different promotion and inhibition effects on corridor construction. In this study, we used the IEW-TOPSIS Table 3 Feasibility evaluation system for potential ecological corridors * "+" and "−" indicate positive and negative correlation, respectively

Category Description
Internal composition Length (−)* A basic characteristic. Greater lengths are associated with a greater number of contradictions between land use types requiring coordination, a higher likelihood of surrounding human influence, and greater difficulty in maintaining corridor continuity Built-up area (−)* These are the most direct cause of resistance to corridor construction (Li et al. 2015a, b). Creating corridors through road green belts, parks and roof gardens, and eliminating/ restoring green spaces incurs massive cost. Larger built-up areas are associated with greater resistance to construction Waterbody area (−)* Green corridor discussions have highlighted the positive ecological benefits (Xie et al. 2021;Yang et al. 2018) of water areas as well as the negative aspects of related passage limitation for terrestrial species. Accordingly, corridors are more difficult to establish with larger water areas Gravitational value (+)* This is calculated in consideration of resistance values and ecological source areas ( (Information Entropy Weight-Technique for Order Preference by Similarity to an Ideal Solution) method to evaluate the importance of PECs. The IEW was defined and constructed based on information entropy and the practical background of raw data. It can be used to transform information on various indicators into data and realize the possibility of mutual comparison and can be effectively applied to the weight assignment of this study (Zhang et al. 2011). TOPSIS is a technique for ranking and selecting some possible alternatives based on the measurement of Euclidean distances. Its working principle is based on the fact that the chosen alternative should have the shortest distance from the positive ideal solution (PIS) and the farthest from the negative ideal solution (NIS) (Zhang et al. 2011). TOPSIS is widely used in many disciplines (Lin et al. 2020;Vavrek and Chovancová, 2019), but its application in ecological planning is rare. Methods such as analytical hierarchy processing and principal component analysis focus on the relative differences between schemes. TOPSIS sets the PIS and the NIS, and defines the schemes' level via comparison (Behzadian et al. 2012). Here, it was used to select a corridor construction scheme by identifying gaps between potential and optimal corridors. This method can be used to quantitatively evaluate the construction level of each potential corridor while assessing their relative importance.
Processing requires transforming different scales and units among various indices into common measurable units to allow index comparison. Here, X ij was the evaluation matrix X for the alternative i with evaluation index j. Dimensionless processing was adopted to obtain different index values as follows: x ′ ij is the matrix after dimensionless processing. The IEW was then used to determine weight objectively. In information theory, entropy is a measure of the uncertainty associated with a random variable. IEW was calculated using: The following was used to avoid the insignificance of ln f ij : x ij (Generate variables−Reverse NMMS).
(7) H j = − m ∑ i f ij ln f ij ;i = 1, 2, … , m;j = 1, 2, … , n G j represents deviation in the coefficients of indices j. Generally speaking, higher degrees of deviation in j equate to lower values of information entropy H j . This indicates that the greater amounts of information provided by index j equate to greater weights in the index. The weight W j is defined as: X + indicates the most preferable alternative and X − the least preferable. The formulas are: To calculate separation, n-index evaluation distance was used. This represents the separation from PIS and NIS for each alternative.
The relative closeness of the ith alternative with respect to the ideal solution X + is defined as c i .
If the value of c i is closer to 1, the alternative i will be closer to the PIS.

Study area
We demonstrated the application of the construction feasibility evaluation system for selecting corridor widths and determining construction sequence for PECs as a case study in the rapidly expanding city of Chengdu, Sichuan, China (103°01′E-104°53′E，30°05′-31°26′N) (Fig. 1). Our The urban expansion phenomenon is also prominent. In recent years, to protect the ecological foundation and promote the development of "Park City", the construction of ecological corridor has received a positive response in Chengdu. It is of great significance to take Chengdu as a study area for identifying PEC areas. In addition, Chengdu is a typical basin topography with diverse landscape features, such as Linpan, a peculiar landscape component in Sichuan, and contributes to multi-services (Liu et al. 2019). Combining characteristic local landscapes for the construction of PEC is equally noteworthy.

Data collection
The land use data were obtained by using Landsat satellite remote-sensor images in 2019, and supervised classification was then applied to divide the land use into the categories of forest, cropland, grassland, water, built-up area, and unused land (Fig. 1). Finally, high-precision satellite images and 2019 land use data were combined for revision to optimize the accuracy and authenticity of remote-sensor interpretation results (http:// www. resdc. cn/). The KAPPA coefficient was calculated as 80.29% (Geymen and Baz 2007). The accuracy of this land use data is suitable for research applications.
Other data include digital elevation model (DEM) (http:// www. gsclo ud. cn/), slope (generated by the DEM), distance to town, and distance to road (extracted from land use data obtained by Euclidean Distance Analysis).

Patch extraction
Morphological spatial pattern analysis (MSPA) was used to divide the binary raster imagery into seven MSPA classes based on the Euclidean distance threshold between raster cells (Soille and Vogt 2009;Vogt et al. 2007) (Table 4), as MSPA analysis can be used to extract UGS patches from the perspective of mathematical morphology. Considering that MSPA is sensitive to edge width, the same area with four widths was intercepted for comparative analysis enabling the selection of appropriate boundary widths (Fig. 2). The selected zone is in southeastern Chengdu, which is characterized by urban wetland parks, mountain forests and small  (Table.S1). According to the element form of MSPA classes, core, islet, edge, and perforation are equivalent to patch, and bridge, loop, and branch are equivalent to corridor (Dytham and Forman 1996). ArcGIS was used to reclassify the MSPA map, with core, islet, edge, and perforation identified as patch.

Core patches and resistance surface
Based on patch dPC and area data (Table S2), habitat patches in Chengdu were divided into 7 levels (Fig. S1), and 27 of these from levels 1-3 (Table S3, Fig. 3) were selected as core patches. The results indicate that large areas of ecological sources are mainly distributed in the southern part of the city. In the east-west direction, landscape connectivity between patches could have improved. The landscape pattern index and the equivalent value of ecosystem service for each land use type are shown in Table 5, and using the entropy method, the resistance weights of forest, cropland, waterbody, grassland, Foreground pixels surrounded on all sides by foreground pixels and greater than the specified edge width distance from background Islet Foreground pixels that do not contain core. Islet is the only unconnected class. Edges and perforations surround core, and loops, bridges and branches are connected to core Edge Pixels that form the transition zone between foreground and background Perforation Pixels that form the transition zone between foreground and background for interior regions of foreground. Consider a group of foreground pixels in the shape of a doughnut. The pixels forming the inner edge would be classified as perforations, whereas those forming the outer edge would be classified as edge Bridge Foreground pixels that connect two or more disjunct areas of core Loop Foreground pixels that connect an area of core to itself Branch Foreground pixels that extend from an area of core, but do not connect to another area of core built-up and unused land were determined to be 1. 711, 3.736, 4.732, 4.825, 5.000 and 4.996, respectively. The above data, as well as results of elevation, slope, distance from the main road, and distance from the main town, were integrated to obtain a grading standard for the ecological resistance surface in Chengdu (Table S4). Further, through ArcGIS superposition analysis, single-layer indicators were superimposed to generate ecological resistance surfaces (Fig. S2, Fig. 3). The results are in line with the current situation in which Chengdu plays a central-city role as a development core, and peripheral development is relatively slow.

PECs identification results
Using the MCRM, 351 PECs were identified (Fig. 4a). Widths of 30, 60, 200, and 600 m were assigned, and the land use types were counted. The results show that a large number of PECs cross the central urban area in Chengdu, in the construction process, facing many problems like high cost and low operability. In addition, the results also showed that there are differences in the basement covered by the PEC under different widths (Fig. 4b), which also means that the different widths in the construction process will lead to differences in the construction methods. The result of land use types of PECs (Table 6) shows that cropland and forest area with different widths were the largest, followed by built-up area. The significance of forest and cropland in the interior contributes to corridor construction, while the builtup area is the main land use type hindering construction.

Internal composition and connected patches characteristics
(1) Length and internal land use area results  The results show that PEC lengths span various in a wide range (Fig. 5) from 3.529 km (11)(12), with an average of 59.480 km. Through box plot analysis for internal land use area, internal built-up, and waterbody area of the PEC in Chengdu, the area increased with the increase of width, and the box area (25-75%) became larger. Additionally, the CV in the internal builtup is 90.29, 87.04, 80.56, and 76.81% from 30 m width to 600 m width, respectively. This study shows that internal built-up within corridor differences are obvious at the same width but weaken as the width increase. As the width of the waterbody area, CV in the internal waterbody is 98.38, 99.89, 104.60, and 106.20% from 30 m width to 600 m width, respectively. The CV increase, and the uncertainty becomes high. The CVs are all greater than 0.98 for the waterbody area, and there is a large difference between them.
(2) Gravitational value The gravity model results (Tab. S5) show a range from  to 44, with a large interval span and obvious differences among gravitational values. For example, the gravitational values of corridors 11-12 and 2-3 are all above 4,000, followed by 17-18 at only 17,470. Only eight corridors have gravitational values greater than 10,000. Gravitational values between 100 and 1,000 were seen for 232 corridors, accounting for 66.1% of the total, and with large internal differences. The span and difference in gravitational values show the necessity of ongoing evaluation in PEC construction.
(3) Patch IDCI result In window radius determination, the degree of stability and horizontal and vertical fluctuation trends were evaluated by using gradient analysis on strips with 16 horizontal number and 17 vertical number (Fig. 1). In consideration of the small curve spacing of the index after this radius, along with the fact that the fluctuation direction is basically the same, the radius of 390 is considered to be the ideal radius (Fig.  S3). The selected landscape index supports a smoother visualization effect and the creation of landscape index spatial  1 3 maps (Fig. S4). Via the entropy method, the proportions of ED, PD, CONTAG, LPI, LSI, and SHDI are calculated as 0.23, 0.13, 0.04, 0.35, 0.04, and 0.21, respectively. The IDCI results are obtained as shown in Fig. 6. Patch IDCI reflects morphological differences between internal patches of the PEC, with greater values indicating more obvious differences. According to the box plot analysis (Fig. 6), with the four widths (30, 60, 200, and 600 m), corridor IDCI value ranges are 1.658-3.346, 1.663-3.334, 1.668-3.267 and 1.725-3.163, respectively, and the numerical range is slightly narrowed. In addition, the CV results in the four widths are 10.35, 10.57, 10.38, and 9.42%, respectively. It shows that the internal difference between PECs under each width is small, and with the increase in width, the difference is not obvious. After the width exceeds 60 m, the larger the width is, the smaller the box range is, and the overall integration difficulty is reduced.
(4) Results for connected patches The dPC values and proportions of forest/grassland exhibit significant differences for habitat patches 1-3. The maximum is 57.4366 for patch No. 25, while the value for patch No. 5 is only 0.580739 (Table S3). The range among the 27 patches is large and reflects obvious differences in patch connectivity. Patch No. 25 has the largest proportion of forest and grassland at 73.77%, while patch No. 4 is only 0.19%. There are also significant differences in ecological benefits among the patches. An area-weighted approach was used to integrate the dPC and the proportion of forest and grassland for patches connected by PECs. The range of areaweighted dPC is from 0.0457 to 0.7353, and the proportion of area-weighted forest and grassland is from 2.895 to 52.3702%. This reflects obvious differences in the connectivity and ecology of patches at both ends of different PECs.

Feasibility evaluation results
Using IEW-TOPSIS, the c i index of 351 potential ecological corridors under four widths were calculated. Values of c i with the four widths are in the ranges of 0.3023-0.7037, 0.2960-0.6621, 0.2578-0.6582 and 0.2883-0.7001, respectively. And then, the above c i were processed and analyzed through box plot (Fig. 7). The results show that more than 75% simulated potential ecological corridors in Chengdu lower than 0.5. Based on the construction feasibility evaluation, the centralized distribution range of c i result is similar and presents a similar normal distribution, all around 0.40-0.50. Under different widths, the order of the same PEC has changed, although the c i distribution has not changed significantly.

Corridor construction sequence with different widths
The c i reflects the gap between the PEC and the PIS. Under the four widths, although c i has some similar characteristics, such as the number of PECs with feasibility evaluation results lower than 0.5, it is necessary to clarify the specific corridor location in the corridor construction scheme. For further development of the construction scheme, PECs with c i values greater than 0.55 were identified as having good current conditions, and graphic representation was completed. With corridor widths of 30, 60, 200, and 600 m, 50, 46, 43, and 47 corridors were extracted, respectively. These were further divided into preferential construction corridors ( c i greater than 0.650), easy-to-construct corridors ( c i is 0.610-0.650), standard construction corridors ( c i is 0.580-0.610), and constructible corridors ( c i is 0.550-0.580). The results (Fig. 8) show that: (1) Under different widths, the construction sequence is different. There are many PECs in southwestern Chengdu, with a relatively large number of the preferential type. However, none of these have a width of 60 m. Among them, PECs connecting to ecological source area No. 25 represent the majority and have a high level of importance. are all lower than 0.550, which poses a challenge to corridor network formation. (4) With the four widths, fewer high-level corridors cross the central urban area, which suits the actual situation of difficulty in changing land use in built-up areas of mature cities and towns. Although PECs do not cross the central urban area, those with the four widths connect a large green space area in central urban Chengdu. (5) PEC construction can be integrated. For example,

Comparison of internal characteristics of PECs with different widths
Building on previous studies regarding evaluation for PECs (Knaapen et al. 1992;Li et al. 2015a, b), this study also involved evaluation analysis from a new perspective. Interacting gravity between habitat patches is the main index for evaluating the relative importance of corridors (Kong et al. 2010;Wanghe et al. 2020). However, the gravity model neglected corridor width and did not consider the internal characteristics of PECs (Kong et al. 2010;Wanghe et al. 2020), which lacks certain completeness. This study designed the construction feasibility evaluation system for evaluating the relative importance of PECs in UGS and explored the differences in internal characteristics for PECs under different widths. We found that the patch fragmentation in the PEC will not change significantly with width ( Fig. 6). In addition, the areas of built-up and waterbody vary greatly under different widths (Fig. 5). From the perspective of each PEC, the internal characteristics of the same PECs are quite different under different widths, which differs from the single result obtained by the interacting gravity (Wanghe et al. 2020;Li et al. 2015a). It is therefore valuable and necessary to evaluate the internal characteristics of each PEC under different widths.

Construction differences under different widths
Previous studies based on gravitational values between habitat patches, uniformly assigned width to PECs, ignoring the construction methods of every single corridor is different under different widths (Knaapen et al. 1992;McRae et al. 2008;Dong et al. 2015;Li et al. 2015a, b;Costanza et al.1997). In addition, ecological corridor width also depends on urban expansions and surrounding environments (Ford et al. 2020). Previous studies lack discussion on corridor widths in a specific environment (Dong et al. 2015;Huang et al. 2021;Li et al. 2015a, b), which makes it very difficult to specific implementations. Accordingly, after evaluating each corridor, we conducted a field investigation on the location status of some PECs and discussed the corridor construction form in combination with the surrounding environment (Gregory et al. 2021; Clevenger and Huijser 2011) (Fig. 9). The results show that there are obvious construction differences under different widths. For example, a continuity of 30 m PEC can be constructed through an underground culvert (Fig. 9a). However, for 200 m PEC, an underground culvert may lead to hidden danger to the urban bridge, and continuity of 200 m PEC can be achieved through ecological corridor bridges (Fig. 9b). Moreover, we proposed that PECs construction needs to fully consider the surrounding environmental characteristics, for example, by using urban parks, urban roads, Riverside parks, and greenways to connect corridors and meet the requirements of corridor connectivity. Based on the characteristics of Chengdu, we also proposed to combine Wetland Park (Fig. 9e), Western Sichuan Linpan (A rural residential form in which the farmyard is integrated with the surrounding natural and agriculture environment in Sichuan, China) (f) and some characteristic local landscapes to realize corridors continuity. Accordingly, the specific construction form is closely related to the corridor width and surrounding environment in addition to ecological functions, and the results also indicated the importance of evaluation for corridors with a certain width.

Width selection reference for ecological corridors
From the perspective of biological conservation, corridor width can be determined by observing the characteristics of biological migration and plant growth inside corridors with different widths (Falcy and Estades 2007). Meanwhile, the corridor width of PEC is also necessary to consider the specific implementation difficulty and construction measures. We proposed the evaluation of differences in a relative corridor at different widths to explore more comprehensive and refined approaches to the corridor evaluation method.
We consider the construction of ecological networks to be a step-by-step process. When the decision-maker determines that a PEC needs to be built, corridor width can be selected by evaluating the corridor's biological functions, construction sequence, and value under different widths. For example, in this study, the first step of PEC construction is to basically determine the corridor width range by the wildlife conservation needs, and then, the width can be further determined by using construction feasibility evaluation at this width range. Like the PEC between patch No. 20 and No. 25 in Chengdu, the service target may include large mammals such as pandas and leopards (Chengdu Park City Construction Bureau 2022). For biological conservation, the corridor width should be greater than 200 m, while No. 20-25 has better feasibility evaluation results under 600 m (Fig. 8), so 600 m width needs to be considered. In addition, a large number of PECs in Chengdu are in built-up areas, and the biological protection in the metropolitan area is mainly for birds and reptiles as small animals (Chengdu Park City Construction Bureau 2022). Therefore, the evaluation results under 30 m and 60 m can provide a reference for corridor width selection. Accordingly, this study proposed a corridor width selection method based on the premise of considering current characteristics and important characteristics of the corridor, which can be combined with a comprehensive selection of width based on ecological functional needs.

Limitations and future research
This study focused on differences in the importance of PECs with certain widths and explored approaches to width selection. Certain shortcomings in the study should be acknowledged.
First, in terms of scale selection for corridor identification and discussion of differences in importance, this study was based on a city scale to allow verification of the evaluation system. Because of the resolution of the remote sensing image and the choice of research scale, most of the PECs identified are linked to large ecological core areas. The corridors identified were mostly connected with large ecological core areas. Accordingly, small-scale green spaces within the city were neglected. The suitability and breadth of the evaluation system in different scales need to be further researched.
Second, this study highlights the importance of corridor width in MCRM studies, providing methods for evaluating different widths based on the internal characteristics of PEC. However, in actual situations, PEC construction needs to involve economics, cultural characteristics, and local living conditions. Therefore, it is necessary to evaluate the feasibility of corridor construction in consideration of multiple factors in future research.
Finally, on the issue of width selection, although four different ecological function widths were selected, they are only used as the basis for width selection in verifying the construction feasibility of PEC. And the corridor width discussion should be more specific. For example, to meet the migration of large carnivores, its corridor width has a certain range (Červinka et al. 2013), and the construction feasibility evaluation is able to determine the most suitable width in

Conclusion
In this study, four different PEC widths were selected to explore the importance of corridor width in PEC evaluation. And we took the core plain city of Chengdu in western China as a research object to verify the method. The results show that the c i index of 75% of simulated potential corridors is lower than 0.5, and most of them are difficult to construct. The corridor construction sequence, according to c i index can screen out the corridor suitable for construction. Moreover, under four widths, there are obvious differences in the construction sequence. The feasibility evaluation provides reference information on identification, analysis, construction sequence, and construction practices for corridors in the city.
The study's exploration of ideas on corridor width selection highlighted differences in the relative importance of the same corridor at different widths. Accordingly, a method was proposed to support width choice by contrasting levels of difference in the importance of a single corridor combined with its ecological functional needs.

Data availability
The data for this project are confidential but may be obtained through data use agreements with Kyoto University and Sichuan University. Researchers interested in access to the data may contact yus_xie@outlook.com.