Spatiotemporal Evolution and Suitability Evaluation of Rural Settlements in the Typical Mountainous Area of the Upper Minjiang River: A Case Study of Lixian County, Sichuan Province, China

Abstract

Under the framework of the Rural Revitalization Strategy, optimizing the layout of rural settlements in mountainous areas and guiding their sustainable development must be based on a deep understanding of the evolution characteristics of rural settlements and suitability evaluations. This study focuses on Lixian County, located in the southwestern part of China, Sichuan Province, as the research area and employs methods such as the average nearest neighbor index, kernel density analysis, and landscape pattern index to analyze the spatiotemporal evolution characteristics of rural settlements in 2000, 2010, and 2020. Additionally, the Maxent model, based on ecological niche theory, is applied to evaluate the suitability of rural settlements. The results reveal the following: (1) Rural settlements in Lixian County exhibit a spatial distribution characterized by “sparser in the west, denser in the east, and a belt-like pattern”, with a clustered distribution trend. The number and area of settlement patches increased, with settlement distribution becoming more centralized, shapes becoming more complex, and connectivity between settlements improving. (2) The area of highly suitable land for rural settlements has decreased annually, with over 85% of the land classified as unsuitable for rural settlement layout. Suitability transitions mostly occur between adjacent levels, and it is difficult for unsuitable land to become suitable. (3) In earlier years, settlement suitability was significantly influenced by the distance to cultivated land, slope, and distance to geological hazard sites. By 2020, however, the distance to roads had become the second most important environmental factor, following the distance to cultivated land. Natural environmental factors, particularly topographic features such as elevation and slope, were found to exert a greater influence than socioeconomic factors in evaluating the suitability of rural settlements in Lixian County. These findings provide a scientific foundation for optimizing rural settlement layouts in mountainous regions, offering valuable insights into rural transformation and sustainable development not only in the upper Minjiang River area but also for reference in other similar mountainous regions.

1. Introduction

Rural settlements serve as vital spaces for the production and daily activities of rural populations [1]. Mountainous rural settlements represent a form of population agglomeration within distinct mountainous geographical environments. Compared to rural settlements in plains, those in mountainous areas are more significantly influenced by natural factors, such as mountain gradient energy, spatial heterogeneity, and surface fragmentation [2]. These settlements often face challenges, including small scale, large quantity, scattered layouts, and low land use efficiency, which severely constrain new urbanization efforts, rural transformation, and coordinated urban-rural development [3,4]. In response to these challenges, the Rural Revitalization Strategy has been proposed, aiming to promote the comprehensive development of rural settlements. The connotation of the Rural Revitalization Strategy is summarized as “prosperous industries, ecological livability, rural cultural civilization, effective governance, and affluent living.” Among these, ecological livability is a key component of the strategy, is closely related to rural settlements, and aims to build a livable rural living space that is environmentally sustainable and resource-efficient. China is characterized by extensive mountainous areas, with mountains covering more than two-thirds of its territory. The spatial patterns of mountainous rural settlements continuously evolve under the influence of natural, economic, and social environmental factors [5]. Moreover, the spatial distribution, infrastructure conditions, and other characteristics of these settlements reflect the level of local socioeconomic development. Therefore, investigating the spatiotemporal evolution of mountainous rural settlements is essential for promoting sustainable development and achieving comprehensive rural revitalization in mountainous regions [2,6]. Within the framework of the Rural Revitalization Strategy, which prioritizes ecological livability, improving the living environment and enhancing the well-being of mountainous rural residents must be grounded in a comprehensive understanding of the spatiotemporal evolution of mountainous rural settlements and suitability evaluations of these settlements.

In recent years, research on rural settlements in China’s mountainous areas has gradually increased. Studies on these settlements, based on topographic characteristics, can be broadly categorized into those focusing on the western [7], central [8], and eastern [9] mountainous regions. Due to the significant differences in topography and landforms in the western mountainous areas, particularly the southwestern mountainous regions, rural settlements in the southwestern mountainous areas have become a major research focus. The research focuses on various aspects, including the distribution [10,11,12], spatiotemporal evolution [13,14,15], driving mechanisms [16,17,18], spatial optimization and reconstruction [19,20], and suitability evaluation [21,22] of rural settlements in mountainous areas.

In terms of spatiotemporal evolution, studies often focus on the distribution, morphology, and scale of rural settlements, employing methods such as spatial autocorrelation analysis, landscape pattern index, kernel density analysis, and standard deviational ellipse [10,23,24,25,26]. Jin et al. [27] conducted a study on the Wuhan Metropolitan Area, employing a comprehensive approach that included landscape pattern index, average nearest neighbor index, and kernel density analysis to quantitatively identify the spatial distribution characteristics and evolutionary patterns of rural settlements in the region from 2009 to 2017. Ji et al. [17] employed spatial landscape pattern index, GIS spatial analysis, and kernel density methods to investigate the spatiotemporal dynamics of the spatial distribution, aggregation, and distribution characteristics of rural settlements in Danfeng County. These methods have been proven to be highly effective in summarizing the characteristics and spatial configuration of rural settlement spatiotemporal evolution [28], making it a widely used tool among scholars.

For suitability evaluation, many scholars tend to select evaluation factors based on aspects such as location and socioeconomic conditions or construct rural settlement suitability evaluation systems based on the “production-living-ecology” spatial framework. Hong et al. [29] applied theories related to landscape ecological patterns and utilized the minimum cumulative resistance (MCR) model to assess the ecological suitability of rural settlement land in Dujiangyan City. Liu et al. [30] developed a rural settlement suitability evaluation system based on “production-living-ecology” space, introducing accessibility to interpret the matching degree between residents’ needs and surrounding production and living resources in Tingzu Town, Hubei Province. This approach provided a more scientific and accurate evaluation method. A review of previous studies reveals that most suitability evaluations of rural settlements have relied on multi-factor evaluation systems [31,32,33]. The determination of evaluation factor weights often relies on subjective weighting methods, such as the Delphi method, the Analytic Hierarchy Process (AHP), and fuzzy evaluation methods [34,35]. These studies typically quantified individual factors and then performed spatial overlay analysis to classify suitable areas for rural settlement. This “layer-cake” overlay mapping method has been widely used [22]; however, it is influenced by subjective experience, which limits its application. In contrast, machine learning methods, which do not depend on subjective prior knowledge, have shown significant advantages [36,37,38].

Mountainous rural settlements are complex human–environment systems shaped by the interaction of local society, economy, culture, and natural environment while also being influenced by the mountain ecosystem. The ecological niche of these settlements is seen as a reflection of their geographic location and the available resource space within the mountain ecosystem, which represents the settlement’s adaptation to environmental conditions [22,39]. Machine learning-based evaluation methods are particularly effective in revealing the spatial relationships between settlements and their environmental context. The Maxent model is a typical data-driven method that calculates the potential distribution of a species in a study area based on known species geographic distribution data and corresponding environmental factors by maximizing entropy, which represents the most ideal distribution under environmental constraints. The model requires only presence data for modeling and is characterized by high predictive accuracy, effective modeling performance, and compatibility with both continuous and categorical environmental factors [40]. The Maxent model enhances the robustness of modeling by incorporating both natural and socioeconomic factors [41], making it highly applicable for land use suitability evaluation and prediction of rural settlements [42,43,44,45]. This method is equally applicable in the suitability evaluation of rural settlements. For example, Cao et al. [46] evaluated the suitability of rural settlements in Shengzhou City. The evaluation results delineated the optimal spatial range required for the sustainable development of rural settlements and, from an ecological niche perspective, re-examined rural structural adjustments. This evaluation method also provided valuable insights for rural reconstruction. Zhou et al. [47] applied the Maxent model to evaluate the suitability of rural settlements in the agro-pastoral ecotone. The results demonstrated that the Maxent model can quantitatively assess the contribution and importance of each factor, as well as their temporal variations. This evaluation provided a scientific basis for decision-making related to the siting, spatial planning, and suitability assessment of rural settlements in the agro-pastoral ecotone.

The mountainous rural areas in the upper reaches of the Minjiang River, characterized by fragile natural ecosystems and underdeveloped socioeconomic conditions, present a significant challenge for promoting rural revitalization in Sichuan Province [48]. Under the Rural Revitalization Strategy, optimizing mountainous rural settlements and guiding their sustainable development are key challenges currently faced. In light of this, the study selects Lixian County, a typical mountainous area in the upper reaches of the Minjiang River, as the research region. It reveals the spatiotemporal evolution characteristics of rural settlements from 2000 to 2020. Using the data-driven Maxent model, the study dynamically evaluates the suitability of rural settlements in Lixian County and explores the main environmental factors influencing their suitability. The aim of this study is to provide references for rural transformation and sustainable development in the upper Minjiang River region and other similar mountainous areas.

2. Study Area and Selection of Research Nodes

2.1. Study Area

Lixian County is located in the Aba Tibetan and Qiang Autonomous Prefecture of Sichuan Province, in the southwest mountainous region of China, at the eastern edge of the Qinghai–Tibet Plateau and the northwest of the Sichuan Basin. It lies in a mountainous canyon area where the Chengdu Plain transitions into high mountains and plateau (Figure 1). The county is situated on the western side of the upper reaches of the Minjiang River, along the Zagu’nan River, a primary tributary of the Minjiang River, which flows from northwest to southeast, running through the entire county. Lixian County is located between 30°54′ N to 31°12′ N and 102°32′ E to 103°30′ E, with an average altitude of 2700 m. The lowest point is 1422 m, while the highest reaches 5922 m, resulting in substantial elevation variations. Geologically, it belongs to the middle section of the Longmen Mountains fault zone, characterized by complex topography and landforms. The area is seismically active and frequently experiences geological hazards such as landslides, mudslides, and rockfalls. Lixian County spans 4318 square kilometers, measuring 83 km east-west and 78.2 km north-south. After administrative adjustments in 2019, the county now governs six towns and five townships, with a total of 63 administrative villages and a population of about 46,000. Lixian County is a multi-ethnic region primarily inhabited by Tibetans, Qiang, and Han people.

2.2. Criteria for Selecting Research Nodes

The study spans 20 years, selecting time nodes—2000, 2010, and 2020—based on both the national strategic level and the development trajectory of rural settlements in Lixian County, ensuring a certain level of scientific validity. Since the development of mountainous rural settlements should also be assessed from the perspectives of living conditions and social infrastructure in mountainous areas [49], the selection of research nodes is grounded in the characteristics of rural settlements in Lixian County at three different years, as follows:

• Since the reform and opening-up, China has rapidly advanced urbanization due to urban-biased policy factors. However, due to the absence of an effective system for integrated urban-rural development, rural areas have lagged behind and faced numerous challenges. The “Western Development Strategy” and the concept of rural transformation proposed in 2000 provided new opportunities for the development of regions such as Lixian County. As 2000 marks the starting point for this study, it serves as a reference for comparing the evolution of rural settlements in subsequent years.
• Lixian County’s rural settlements underwent significant changes after the 2008 Wenchuan Earthquake. By 2010, post-disaster reconstruction had largely been completed, and rural settlements had been revitalized. At the same time, transportation infrastructure development was comprehensively promoted, with over 80% of roads being paved. This period saw considerable progress in the living environment and transportation infrastructure of rural settlements in Lixian County, making 2010 an important time node for the study.
• At the end of 2019, Lixian County underwent an administrative restructuring of townships, which had significant implications for the spatial layout planning of towns, resource integration, and the promotion of rational resource allocation. With the achievement of building a moderately prosperous society in all respects by 2020, the Lixian County government placed greater emphasis on agricultural modernization and rural living environment improvements, which greatly influenced the evolution of rural settlements.

3. Methodology and Data Sources

3.1. Data Sources

The research data are primarily divided into land use data, topographic data, remote sensing imagery, POI data, and other statistical data (

Table 1. Description of the data used in this study.

Data	Year	Data Sources
Land use database	2000, 2010, 2020	Lixian County Natural Resources Bureau
Geological hazard point data	2019
Administrative boundary data	2019
Landsat 5 TM/8 OLI imagery	2000, 2010, 2020	U.S. Geological Survey website (https://www.usgs.gov/ (accessed on 12 April 2024))
DEM	/
Slope	/	Extracted from DEM
Aspect	/
POI data	2000, 2010, 2020	Baidu Maps open platform (https://lbsyun.baidu.com/ (accessed on 16 June 2024))
Population statistical data	2000, 2010, 2020	Aba Tibetan and Qiang Autonomous Prefecture Bureau of Statistics (https://tjj.abazhou.gov.cn/ (accessed on 3 June 2024))

). The land use database was provided by the Lixian County Natural Resources Bureau to extract rural settlements, rivers, roads, and townships. Due to the 20-year span of the study, there are significant differences in the accuracy and classification standards of the land use database. Therefore, it is essential to calibrate the rural settlement patch data, particularly for the years 2000 and 2010. To ensure data accuracy, rural settlement patches were corrected using historical imagery from Google Earth. Geological hazard point data and administrative boundary data were also sourced from the Lixian County Natural Resources Bureau. The DEM data were obtained from the SRTM elevation dataset, which can be downloaded from the U.S. Geological Survey (USGS) website (https://www.usgs.gov/ (accessed on 12 April 2024)). Slope and aspect were extracted from DEM data through surface analysis. The remote sensing images were acquired from the USGS website (https://www.usgs.gov/ (accessed on 12 April 2024)). The images for 2000 and 2010 were obtained from Landsat 5 TM, while the image for 2020 was obtained from Landsat 8 OLI. To account for vegetation growth conditions, images from July to August, the peak growing season, were selected. Detailed information about the images is provided in

Table 2. Sources of the remote sensing images.

Year	Data Sources	Image ID
2000	Landsat 5 TM	LT05_L1TP_130038_20000813_20200906_02_T1
2010	Landsat 5 TM	LT05_L1TP_130038_20100809_20200823_02_T1
2020	Landsat 8 OLI	LC08_L1TP_130038_20200719_20200911_02_T1

. Population data were sourced from the Aba Prefecture National Census Statistical Bulletin (https://tjj.abazhou.gov.cn/ (accessed on 3 June 2024)). Data for tourist attractions, schools, and hospitals were obtained from the Baidu Maps open platform (https://lbsyun.baidu.com/ (accessed on 16 June 2024)). Since Lixian County underwent administrative adjustments in 2019, the most recent administrative boundaries after the adjustment are used to ensure data comparability. To ensure consistency in data resolution and coordinate systems, the spatial resolution is set to 30 m × 30 m, with the coordinate system defined as the CGCS2000 geographic coordinate system and the projected coordinate system as the CGCS2000 3 Degree GK Zone 34.

3.2. Methods

This study analyzes the spatiotemporal evolution characteristics of rural settlement patterns in Lixian County using the average nearest neighbor index, kernel density analysis, and landscape pattern index. Based on the rural settlement patch vector data extracted from the land use database, the data were imported into Google Earth Pro 7.3.6. By comparing the patch locations and shapes with historical imagery, manual adjustments were made to the boundaries of patches that exhibited inconsistencies. The corrected patches were then used for subsequent analysis. Additionally, a suitability evaluation system for rural settlements in Lixian County was constructed from the perspectives of natural environment and socioeconomic factors. The Maxent model was employed to evaluate the suitability of rural settlements.

3.2.1. Remote Sensing Images Processing

The remote sensing images are preprocessed to ensure data quality and consistency. Landsat 5 TM images are obtained at the Level-1T (L1T) processing level, while Landsat 8 OLI images are available at the Level-1 (L1) processing level, which is already orthorectified by the USGS. All images are further processed for radiometric and atmospheric corrections using the FLAASH algorithm. The Normalized Difference Vegetation Index (NDVI) for the three time periods is then calculated and used as one of the key indicators in the suitability evaluation of rural settlements. The formula is as follows:

$$NDVI={\displaystyle \frac{(NIR-Red)}{(NIR+Red)}}$$

(1)

where NIR refers to the reflectance of the near-infrared band (Landsat 5 TM Band 4, Landsat 8 OLI Band 5), and Red refers to the reflectance of the red band (Landsat 5 TM Band 3, Landsat 8 OLI Band 4).

3.2.2. Average Nearest Neighbor Analysis

The average nearest neighbor (ANN) index measures the average distance between the centroids of each settlement patch and its nearest neighboring settlement patch centroid. By comparing the measured distance value with the expected average distance under a hypothetical random distribution, it can determine whether rural settlements exhibit a clustered distribution pattern in space. The formula is as follows:

$$\mathrm{A}\mathrm{N}\mathrm{N}={\displaystyle \frac{\overline{{\mathrm{D}}_0}}{\overline{{\mathrm{D}}_{\mathrm{e}}}}}={\displaystyle \frac{{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{d}}_{\mathrm{i}}/\mathrm{n}}{\sqrt{\mathrm{n}/\mathrm{A}/2}}}$$

(2)

where $\overline{{\mathrm{D}}_0}$ is the observed average distance between each rural settlement and its nearest neighbor; $\overline{{\mathrm{D}}_{\mathrm{e}}}$ is the expected average distance under the assumption of a random distribution; $n$ is the total number of rural settlements; $\mathrm{d}\mathrm{ᵢ}$ is the distance between settlement $\mathrm{i}$ and its nearest neighboring settlement; and $\mathrm{A}$ refers to the total area of the study area.

3.2.3. Kernel Density Analysis

Kernel density analysis is a nonparametric density estimation method that has been widely applied in studies of rural settlement spatial distribution. This method calculates the density of rural settlements within their surrounding neighborhoods, primarily using kernel density values and kernel density maps to reflect the spatial distribution characteristics of rural settlements. The formula is as follows:

$$\mathrm{f}\left(\mathrm{x},\mathrm{y}\right)={\displaystyle \frac{1}{\mathrm{n}{\mathrm{h}}^2}}\sum _{\mathrm{i}=1}^{\mathrm{n}}\mathrm{K}\left({\displaystyle \frac{{\mathrm{d}}_{\mathrm{i}}}{\mathrm{n}}}\right)$$

(3)

where $\mathrm{f}(\mathrm{x},\mathrm{y})$ represents the kernel density value of a rural settlement point $(\mathrm{x},\mathrm{y})$; $\mathrm{n}$ denotes the total number of rural settlements in the study area; h is the bandwidth; K is the kernel function; and $\mathrm{d}\mathrm{ᵢ}$ is the distance between the rural settlement point (x, y) and the i-th point.

To assess the spatiotemporal changes in the distribution of rural settlements, we compute the difference in kernel density values at two distinct time points, $t_1$ and $t_2$. The change in kernel density at any given point $(\mathrm{x},\mathrm{y})$ is expressed as follows:

$$\Delta \mathrm{f}\left(\mathrm{x},\mathrm{y}\right)=\mathrm{f}\left(\mathrm{x},\mathrm{y},t_2\right)-\mathrm{f}\left(\mathrm{x},\mathrm{y},t_1\right)$$

(4)

where $\mathrm{f}\left(\mathrm{x},\mathrm{y},t_1\right)$ and $\mathrm{f}\left(\mathrm{x},\mathrm{y},t_2\right)$ represent the kernel density values at times $t_1$ and $t_2$, respectively. The kernel density change $\Delta \mathrm{f}\left(\mathrm{x},\mathrm{y}\right)$ reflects the variation in the density of rural settlements over the time period, allowing for the identification of areas where significant changes in settlement patterns have occurred. When $\Delta \mathrm{f}\left(\mathrm{x},\mathrm{y}\right)$> 0, it indicates a reduction in rural settlement density, whereas $\Delta \mathrm{f}\left(\mathrm{x},\mathrm{y}\right)$< 0 signifies an increase in rural settlement density.

3.2.4. Landscape Pattern Index

Rural settlements, as part of the rural landscape, can effectively summarize the landscape characteristics of rural settlements using the landscape pattern index. By employing eight landscape pattern indices—Total Class Area (CA), Number of Patches (NP), Patch Density (PD), Mean Patch Area (AREA_MN), Largest Patch Index (LPI), Landscape Shape Index (LSI), Cohesion Index (COHESION), and Landscape Aggregation Index (AI)—this study characterizes the changes in the scale, area, and shape of rural settlements in Lixian County. The formulas and their meanings are shown in

Table 3. Landscape pattern index.

Category	Formula	Meaning
Scale Characteristics	$NP=n$	The number of rural settlement patches represents the degree of landscape fragmentation
	$PD={\displaystyle \frac{NP}{A}}$	The patch density per unit area reflects the degree of spatial distribution sparsity of the patches
Area Characteristics	$CA={\displaystyle \sum _{i=1}^n}{A}_i$	The total area of rural settlement patches reflects the overall scale of the rural settlement
	$AREA\_MN={\displaystyle \frac{CA}{NP}}$	The mean patch area of rural settlements measures the typical size of landscape patches
Shape and Aggregation Characteristics	$LPI={\displaystyle \frac{A_{max}}{CA}}\times 100$	The proportion of the largest rural settlement patch to the total number of patches indicates that a higher value reflects a stronger dominant role of a single rural settlement patch in the overall landscape
	$LSI={\displaystyle \frac{P}{\sqrt{A}}}\times 100$	Measures the complexity of patch shape, with a higher value indicating more complex shapes
	$AI={\displaystyle \frac{{\sum }_{i=1}^n{\left(\frac{A_i}{A}\right)}^2}{NP}}$	The degree of patch aggregation in the landscape, with a higher value indicating more concentrated patch distribution
	$COHESION={\displaystyle \frac{{\sum }_{i=1}^n{A_i}^2}{A^2}}$	The degree of patch connectivity, with a higher value indicating stronger landscape connectivity

$n$ is the number of rural settlement patches; $A$ is the area of the study area; $A_i$ is the area of each patch in class i; $A_{max}$ is the area of the largest patch; and $P$ is the perimeter of all rural settlement patches.

.

3.2.5. Suitability Evaluation Method

The suitability evaluation modeling of rural settlements in the mountainous areas of Lixian County is conducted using Maxent 3.4.4 [50]. Its working principle is based on the maximum entropy theory, which models the relationship between environmental factors and known data to predict the spatial distribution of the target variable. The core objective is to infer the most probable probability distribution by minimizing the uncertainty of the known information, in other words, by maximizing entropy.

To evaluate the suitability using the Maxent model, both rural settlement location data and environmental factors data are required. The rural settlement patch data used in the spatiotemporal evolution analysis of rural settlements were converted into point data using the ‘Feature to Point’ tool, which then served as the sample points for the Maxent model. Considering the availability of data, 12 environmental factors were selected from both the natural environment and socioeconomic aspects (

Table 4. Rural settlement suitability evaluation system.

Target Layer	Criterion Layer	Indicator Layer
Rural Settlement Suitability Evaluation	Natural environmental factors	Elevation
		Slope
		Aspect
		NDVI
		Distance to geological hazard sites
		Distance to rivers
	Socioeconomic factors	Distance to cultivated land
		Distance to roads
		Distance to township centers
		Population density
		Distance to tourist attractions
		Distance to schools and hospitals

).

Among these, slope and aspect were derived from DEM data, and NDVI data, representing vegetation coverage, were calculated from preprocessed Landsat imagery for three time periods. These datasets were resampled to a consistent resolution and clipped to the boundaries of Lixian County. The river, cultivated land, road, and town patches for three time periods were extracted from the Lixian County land use database. Euclidean distances were then calculated for each of these factors, and all datasets were resampled to a pixel size of 30 m × 30 m within the administrative boundaries of Lixian County. The point data for tourist attractions, schools, and hospitals were obtained from Baidu Maps, and similar processing was applied to calculate Euclidean distances and output the data at the same resolution. The processed environmental factors are shown in Figure 2; only the 2020 factors are presented here, while the factors for 2000 and 2010 are similar. All environmental factors were converted to ASCII format for input into the Maxent model.

To examine whether there is a significant correlation between the environmental factors, a Pearson correlation analysis was used to calculate the correlation coefficients. The results showed that the absolute values of the correlation coefficients were all less than 0.75, indicating that there is no significant correlation between the factors. The parameter settings for suitability modeling were referenced from previous studies [47,51]. In the Maxent3.4.4 software, the data were randomly split into a training set and a validation set according to a certain proportion. We selected 70% of the rural settlement location data as the training set to drive the model, with the remaining 30% used as the test set for accuracy verification. The model generated a receiver operating characteristic (ROC) curve during prediction, and model accuracy was evaluated using the area under the curve (AUC). The model was iterated 10 times, and the final output was presented in logistic format.

The suitability evaluation results ranged from 0 to 1. Suitability levels were classified into four levels using the natural breaks method, and a suitability map was generated. The suitability results for each year were compared to analyze the trends in the suitability of rural settlements in the mountainous areas of Lixian County. The model quantitatively assesses the contribution of each environmental factor to the model’s predictions during the process, using percent contribution and permutation importance for evaluation. Additionally, under the condition of fixing other environmental factors, the impact of a single environmental factor on the suitability of rural settlements is assessed. The response curve is derived by averaging the results of 10 iterations of the model.

4. Results and Analysis

4.1. Evolution of the Spatiotemporal Pattern of Rural Settlements

4.1.1. Spatial Distribution and Scale Characteristics of Rural Settlements

The corrected rural settlement patches, extracted from the land use survey database, were organized to generate the dynamic changes in rural settlements (Figure 3). Overall, the rural settlement patches in Lixian County exhibit a spatial distribution characterized by a pattern of “sparse in the west and dense in the east, with a banded distribution”, primarily along river valley areas. Considering the small scale and concentrated distribution of rural settlements, representative small areas within the study region were selected to illustrate the patterns of expansion and decline in rural settlements from 2000 to 2020. In these small areas, the expansion of rural settlements primarily occurs on the basis of existing settlements, particularly in Area A and Area D, where settlements have expanded in contiguous patches. Additionally, some settlements have expanded independently, such as in Area F. The decline in settlements is mainly observed in small settlements located away from the valleys, such as in Area H.

To quantify the dynamic changes in rural settlements and understand the characteristics of their area, number, and density, the landscape pattern index was used for representation (

Table 5. Landscape pattern index representing scale and area characteristics.

Year	CA (ha)	NP	PD (Patches/ha)	AREA_MN (ha)
2000	454.7	4108	396.4621	0.2522
2010	469.6	4334	390.0367	0.2564
2020	488.0	4367	381.2962	0.2623

). From 2000 to 2020, both CA and NP exhibited a gradual increasing trend. CA increased from 454.7 hectares in 2000 to 488 hectares in 2020, with a total increase of 33.3 hectares over 20 years and a more significant increase observed from 2010 to 2020. Regarding NP, there was a substantial increase of 224 patches from 2000 to 2010, reflecting a notable change in the number of rural settlements during this period, whereas NP increased by only 33 patches from 2010 to 2020. Over the 20 years, PD decreased gradually from 396.4621 in 2000 to 381.2962 in 2020, indicating a declining trend in the density of rural settlement patches. Meanwhile, AREA_MN increased from 0.2522 in 2000 to 0.2623 in 2020, suggesting that some smaller rural settlement patches have gradually merged into larger patches.

4.1.2. Agglomeration Characteristics of Rural Settlements

As shown in

Table 6. ANN and related indicators.

	2000	2010	2020
Average observation distance (m)	81.297260	80.084153	86.325613
Expected average distance (m)	516.827399	507.153047	509.336976
ANN	0.157301	0.157909	0.169486
z-score	−103.328181	−106.055754	−104.995167
p-value	0.000000	0.000000	0.000000

Note: ANN > 1 indicates dispersion, ANN ≈ 1 indicates random, ANN < 1 indicates clustering; z ≤ 1.645 indicates aggregated distribution, and z > 1.645 indicates dispersion distribution.

, the average observation distances for the years 2000, 2010, and 2020 were considerably lower than the expected average distances, suggesting a non-random spatial pattern. Correspondingly, the ANN of rural settlements in Lixian County was 0.157301, 0.157909, and 0.169486, respectively. As all values were less than 1, these results indicate that the spatial distribution of rural settlements consistently exhibited a clustered pattern throughout the study period. Furthermore, the z-scores were all below the critical value, and the p-values were statistically significant at p < 0.01, further validating that the clustered distribution pattern of rural settlements in Lixian County was highly significant.

The kernel density values were classified into different levels using the natural breaks method, and the difference in density between adjacent years was calculated to illustrate the spatial changes in rural settlement density (Figure 4).

Temporally, the overall kernel density distribution pattern remained relatively stable, while the maximum kernel density value of rural settlements in Lixian County showed an increasing trend. The change in kernel density between 2010 and 2020 was more significant compared to the period between 2000 and 2010. Spatially, the density distribution of rural settlements exhibited significant differences between the eastern and western regions. High-density areas were primarily concentrated in the eastern part of Lixian County, where kernel density values underwent noticeable changes. Particularly in the southeastern towns, the kernel density of rural settlements has continuously increased over the past 20 years. In contrast, the western region was predominantly characterized by low-density areas, with only a few medium-density zones, indicating that the growth potential of rural settlements in the western region was weaker than that in the eastern region.

4.1.3. Evolution of Spatial Morphology in Rural Settlements

Table 7. Landscape pattern index representing shape and aggregation characteristics.

Year	LPI	LSI	COHESION	AI
2000	0.9353	48.5775	55.2772	31.8127
2010	0.8689	49.2847	55.9064	31.9268
2020	1.0269	49.6994	57.1034	32.2480

presents the landscape pattern index used to describe the morphology of rural settlements. LPI decreased initially, from 0.9353 in 2000 to 0.8689 in 2010, but then rose to 1.0269 in 2020. LSI showed a continuous upward trend, increasing from 48.5775 in 2000 to 49.6939 in 2020, indicating an increase in the complexity of rural settlement shapes. COHESION steadily increased from 55.2772 in 2000 to 57.1034 in 2020, while AI rose from 31.8127 in 2000 to 32.248 in 2020. These trends reflect an improvement in the spatial continuity of rural settlements in Lixian County over time, with enhanced connectivity between settlements and a tendency toward a more concentrated distribution.

4.2. Suitability Evaluation Results

4.2.1. Assessment of Maxent Model Accuracy

The model’s accuracy is evaluated using the AUC value, which quantifies the area under the ROC curve. This metric represents the model’s ability to distinguish between positive and negative class samples across all possible thresholds. In this study, the positive class refers to regions predicted as suitable for rural settlement layout, while the negative class denotes unsuitable regions. The vertical axis, sensitivity, indicates the proportion of suitable rural settlements correctly predicted by the model. The horizontal axis, 1-specificity, represents the proportion of unsuitable rural settlements incorrectly predicted as suitable by the model. A higher sensitivity value indicates a greater probability that the model correctly classifies the suitability of rural settlements, reflecting better model performance [52]. Therefore, the ideal model curve should be as close as possible to the top-left corner, where sensitivity approaches 1 and 1-specificity approaches 0, resulting in an AUC value that is closest to 1. The ROC curves for the three periods are shown in Figure 5.

The ROC curves show that the AUC values for both the training and testing sets in three periods are between 0.9 and 1. The proximity of the model curves to the upper-left corner suggests that the Maxent model exhibits high accuracy in predicting rural settlement suitability, confirming that the results are non-random. The significant overlap between the training and testing set curves in three periods further indicates that the rural settlement distribution in Lixian County is concentrated in the river valley areas, with similar distribution characteristics among settlements. The performance of the model on the testing set reflects its learning from the training set. Overall, the Maxent model demonstrates strong generalization ability and stability, making it a reliable tool for evaluating rural settlement suitability.

4.2.2. Evaluation Results of Rural Settlement Suitability

The suitability evaluation results for the three periods were reclassified using the natural breaks method. The results were divided into four levels: unsuitable areas, less suitable areas, moderately suitable areas, and highly suitable areas (Figure 6). The area and proportion of each level were calculated (

Table 8. Area and proportion statistics for different suitability levels.

Year	Levels	Area (km2)	Percent (%)
2000	Unsuitable	3745.7775	86.95%
	Less suitable	313.7112	7.28%
	Moderately suitable	128.3877	2.98%
	Highly suitable	120.1302	2.79%
2010	Unsuitable	3686.9697	85.58%
	Less suitable	368.3889	8.55%
	Moderately suitable	137.4777	3.19%
	Highly suitable	115.1694	2.67%
2020	Unsuitable	3723.7239	86.44%
	Less suitable	331.8346	7.70%
	Moderately suitable	140.2497	3.26%
	Highly suitable	112.2021	2.60%

).

The results indicate that the majority of Lixian County is unsuitable for rural settlement layout, with unsuitable and less suitable areas accounting for over 90% of the total area. In contrast, only 5–6% of the area is classified as moderately or highly suitable for rural settlements.

Regarding the proportion of suitable areas in each period, the highly suitable and moderately suitable areas in 2000 were 120.13 km² (2.79%) and 128.39 km² (2.98%), respectively. By 2010, these values changed to 115.17 km² (2.67%) and 137.48 km² (3.19%) and further shifted to 112.20 km² (2.60%) and 140.25 km² (3.26%) in 2020. These trends indicate a slight decline in highly suitable areas over time, accompanied by a gradual increase in moderately suitable areas, suggesting potential spatial adjustments in land suitability patterns during the study period. Overall, from 2000 to 2020, moderately suitable areas exhibited an increasing trend, while highly suitable areas experienced a slight decline, and less suitable and unsuitable areas remained relatively unchanged, indicating a tendency toward spatial reorganization in land suitability distribution.

Based on the suitability evaluation results, a suitability transition matrix for rural settlements in Lixian County was created (Figure 7).

The results reveal notable dynamic transitions between different suitability levels. Transitions between adjacent levels are predominant, such as between highly suitable and moderately suitable areas or between less suitable and unsuitable areas. However, direct transitions between highly suitable and unsuitable areas are rare, with almost no transitions from unsuitable to highly suitable areas. This pattern highlights the dominant role of the high mountain and canyon terrain in Lixian County, which imposes severe physical constraints on improving the suitability of previously unsuitable land for settlement development.

4.2.3. The Relationship Between Suitability and Environmental Factors

To analyze the influence of environmental factors on the suitability of rural settlements in Lixian County, the percentage contribution and permutation importance of each factor in the Maxent model are presented in

Table 9. Percent contribution and permutation importance of the environmental factors.

Environmental Factors	Percent Contribution (%)			Permutation Importance (%)
	2000	2010	2020	2000	2010	2020
Distance to cultivated land	80.7	79.4	72.1	10.4	12.6	9.6
Slope	8	6.6	6.1	22.7	14.8	15.1
Distance to geological hazard sites	5.6	6.1	4.2	8.4	3	6.1
NDVI	2.4	3.1	2.2	2.5	2.2	1.8
Elevation	1.2	3	3.7	52.9	65.9	63.5
Distance to schools and hospitals	0.8	0.7	0.2	0	0.3	0.5
Distance to rivers	0.4	0.3	0.2	0.4	0.2	0.5
Population density	0.3	0.3	0.3	2	0.7	1.4
Distance to tourist attractions	0.3	0.1	0.8	0.3	0	0.1
Aspect	0.1	0.2	0.1	0.1	0.1	0.1
Distance to roads	0.1	0.1	9.8	0.1	0	1.1
Distance to township centers	0.1	0.2	0.3	0.1	0.1	0.2

. Based on the data in Table 9, radar charts were generated to visually illustrate the results, as shown in Figure 8. Percentage contributions reflect the impact of environmental factors on evaluation outcomes, where higher contribution rates indicate a greater influence on settlement suitability. Permutation importance is assessed by measuring the decline in model prediction accuracy after randomly shuffling the values of the environmental factors, thereby explaining the independent contribution of each factor.

According to the percentage contribution, the top three environmental factors with high percentage contributions in 2000 and 2010 were distance to cultivated land (80.7%, 79.4%), slope (8%, 6.6%), and distance to geological hazard sites (5.6%, 6.1%). By 2020, however, distance to roads emerged as the second most influential factor, increasing its contribution to 9.8%. The top three contributors in 2020 were distance to cultivated land (72.1%), distance to roads (9.8%), and slope (6.1%). Overall, although the percentage contribution of distance to cultivated land decreased over time, it remained the most significant factor influencing settlement suitability. Meanwhile, the influence of distance to roads has increased in recent years, whereas the percentage contribution of distance to geological hazard sites has shown a declining trend.

In terms of permutation importance, elevation and slope remained the most critical factors affecting the suitability of rural settlements in Lixian County, highlighting the dominant influence of topography. The importance of elevation was 52.9%, 65.9%, and 63.5% in 2000, 2010, and 2020, respectively, while that of slope declined from 22.7% in 2000 to 14.8% in 2010 before slightly increasing to 15.1% in 2020. Factors closely related to topography, such as distance to geological hazard sites, also demonstrated relatively high importance, reflecting persistent geophysical constraints in Lixian County. Among socioeconomic factors, distance to cultivated land remained the most significant, with permutation importance values of 10.4%, 12.6%, and 9.6% in 2000, 2010, and 2020, respectively. The slight decline in recent years may be attributed to advances in transportation infrastructure, reducing reliance on cultivated land. Population density showed lower importance, with values of 2%, 0.7%, and 1.4% across the three periods, reflecting its limited impact on rural settlement patterns. Except for the factor of distance to roads, whose importance increased in 2020, the importance of other socioeconomic factors remained relatively low and had a limited impact on the suitability of rural settlements.

Among the environmental factors with low percent contribution and permutation importance, the influence of distance to schools and hospitals on rural settlement suitability diminishes, while the impact of distance to tourist attractions and township centers becomes more pronounced. However, the overall effect remains relatively minor. Rivers do not emerge as a significant determinant of settlement suitability in high mountain canyon regions. Overall, natural environmental factors exhibited greater and more consistent importance compared to socioeconomic factors [53], emphasizing that topographic conditions played a dominant role in shaping the spatial patterns of rural settlements in Lixian County.

The response curve shown in Figure 9 represents the relationship between the environmental factor and the suitability for rural settlements. The X-axis represents the range of values for the environmental factor, while the Y-axis indicates the suitability for rural settlements.

As shown in the figure, areas below an elevation of 2500 m and with slopes less than 25° are more suitable for rural settlement layouts. Suitability decreases rapidly when the elevation exceeds 2500 m or the slopes surpass 25° (Figure 9a,b). Settlement suitability is lower on eastern and western slopes but higher on southern and southwestern slopes. By 2020, the impact of aspect on suitability had diminished (Figure 9c). During the three periods, rural settlements were most suitable in areas where the NDVI value ranged from 0.5 to 0.8, with suitability declining sharply after reaching its peak. In 2020, compared to 2000 and 2010, suitability in areas with high NDVI values improved, indicating that improved vegetation quality over time enhanced rural settlement suitability (Figure 9d). Areas close to geological hazard sites were generally more suitable for rural settlements. While geological hazards pose negative impacts, the ultimate distribution of rural settlements reflects a trade-off between risks and benefits (Figure 9e). Suitability for rural settlements is highest within approximately 300 m of rivers (Figure 9f). Within a 1000 m range, closer proximity to cultivated land correlates with higher suitability. In 2020, the response curve for the distance to cultivated land was smoother compared to 2000 and 2010 (Figure 9g). Suitability decreases rapidly with increasing distance from roads, and after 2020, areas within 1000 m of roads were more suitable, indicating a growing reliance on roads (Figure 9h). In 2000 and 2010, rural settlement suitability decreased as the distance to township centers increased. By 2020, the influence of township centers on rural settlements extended over a broader range (Figure 9i). Suitability increased with population density, stabilizing after reaching a certain threshold (Figure 9j). Areas within approximately 5000 m of tourist attractions, schools, and hospitals were generally more suitable for rural settlement layouts (Figure 9k,l).

5. Discussion

5.1. Spatial Distribution and Suitability Evolution of Rural Settlements

Lixian County, located at the junction of the eastern edge of the Tibetan Plateau and the Sichuan Basin, is characterized by a high mountain and canyon terrain. Due to topographic constraints, rural settlements are primarily distributed in river valleys and relatively flat areas, exhibiting a general spatial pattern of “dense in the east, sparse in the west, and linear distribution.” This conclusion aligns with studies on rural settlements in the mountainous areas of the upper Minjiang River [16]. In canyon regions, significant elevation differences and steep slopes hinder the contiguous expansion of settlements. These settlements are predominantly small to medium-sized and exhibit clear clustering characteristics, consistent with the findings of Tan et al. [54]. Lixian County is prone to geological hazards, with frequent occurrences of debris flows and landslides, which significantly impact the spatial distribution of rural settlements [55]. Affected by geological hazards, particularly the 2008 Wenchuan Earthquake, many rural settlements were damaged and subsequently reconstructed in flat, contiguous river valley areas. These reconstructed villages were uniformly planned and spatially organized, leading to the emergence of larger-scale settlements around 2010. Over the study period, rural settlements in Lixian County gradually evolved from small-scale settlements into larger units with a more concentrated spatial distribution. Overall, the settlement morphology became increasingly complex [56], spatial continuity improved, and connectivity between settlements strengthened, reflecting a trend toward more intensive land use for rural settlements.

The suitability of rural settlements is influenced by natural resource conditions and planning management factors. For example, rural settlements in the southeastern region, near the lower reaches of the Zagu’nan River, have favorable resource endowments and are key areas for county development, thus showing a large area of highly suitable land. In contrast, towns with less suitable and unsuitable areas are typically found in key ecological protection zones [21]. The areas classified as highly suitable and moderately suitable for rural settlements showed an increasing trend between 2000 and 2010. During this period, settlement suitability was influenced by land development and natural disasters. On the one hand, the 2008 Wenchuan Earthquake caused substantial damage to rural settlements in Lixian County, necessitating the relocation of some settlements to new sites. On the other hand, in response to post-earthquake economic and social development needs, intensified land use development in rural areas led to the transformation of some less suitable areas into moderately suitable areas. However, between 2010 and 2020, as a large number of rural settlements were rebuilt post-earthquake, land development for settlements reached saturation. Suitable land resources became increasingly limited, and policy priorities shifted toward sustainable development and ecological conservation, resulting in a stabilization of the total area of highly suitable and moderately suitable zones.

5.2. Environmental Factors Analysis and Optimization Suggestions

One of the key strengths of the Maxent model is its ability to evaluate the importance of environmental factors [57]. Among these, distance to cultivated land, slope, distance to geological hazard sites, and distance to roads were identified as the most important factors influencing rural settlement suitability in Lixian County. Due to the reliance on agriculture for livelihood during the early stages of settlement formation and the fact that Lixian County is still primarily focused on the development of the primary industry, the distance to cultivated land becomes a crucial factor affecting rural settlement suitability [58]. Topographic conditions often determine the development space for rural settlements, with suitable land for settlement layout decreasing as altitude and slope increase [59].

Over time, the influence of slope and distance to geological hazard sites on suitability weakened, while the impact of distance to roads significantly increased. During the period from 2010 to 2020, the construction of National Highway 317 and rural roads significantly improved transportation accessibility, expanding the reach of the road network [60]. Roads serve as vital links for rural residents to transport goods and access materials and information. The construction of national highways has connected more rural settlements, enhancing their suitability, which aligns with findings from other related studies on settlements [61]. In the future, rural settlement planning in Lixian County should place greater emphasis on investment in transportation infrastructure to promote sustainable rural development. The distance to township centers had a relatively minor impact on settlement suitability. This is largely due to the scattered distribution of remote rural settlements and their poor infrastructure, which makes it difficult for residents to access employment, education, and healthcare resources in townships [47]. Similarly, factors such as distance to schools and hospitals also had limited influence. The township mergers and administrative boundary adjustments implemented in Lixian County around 2018 extended the influence of township centers over rural settlements, as reflected in the response curves for the distance to township centers factor. This reform is expected to enhance the suitability of rural areas in Lixian in the future.

To optimize rural settlement layouts in Lixian County, efforts should focus on improving the land-use efficiency of highly suitable and moderately suitable areas while encouraging settlement relocation to river valleys. Enhancing the resilience of rural settlements through cultural preservation, village construction, and improvements in public services can help mitigate the constraints imposed by the natural environment. Policies and management measures should be implemented to strengthen ecological conservation in unsuitable areas and enhance disaster prevention measures to mitigate risks associated with geological hazards, thereby achieving sustainable development of rural settlements in Lixian County.

5.3. Contributions, Limitations, and Future Perspectives

This study evaluates suitability based on natural environmental and socioeconomic factors. Compared to previous research, a significant advantage of the Maxent model is its ability to quantitatively assess environmental factors and determine how suitability changes with individual environmental variables [47]. This provides valuable insights for optimizing and adjusting rural settlements in Lixian. However, due to the overwhelming constraints imposed by the natural environment in Lixian, the influence of socioeconomic factors on rural settlement suitability appears relatively minor, potentially obscuring their underlying impact. In the future, we need to conduct a more detailed exploration of socioeconomic factors. Moreover, the distribution and suitability of rural settlements are influenced by a complex interplay of various factors. For instance, this study did not consider the influence of ethnic cultural factors despite the fact that mountainous regions often foster unique and diverse cultures. Although Lixian is a typical Tibetan and Qiang cultural area, future research should focus more on sociocultural influences, adopting more specific themes and conducting studies at smaller spatial scales. For example, field surveys or interviews could be conducted to quantify the living habits and cultural traditions of rural residents, incorporating these factors into model analyses to more comprehensively assess their impact on the livability of rural settlements.

6. Conclusions

To achieve the sustainable development of rural settlements in the upper reaches of the Minjiang River, this study selected Lixian County, a typical mountainous area in the region, as the research area. Using methods such as the landscape pattern index, the average nearest neighbor index, and kernel density analysis, the study analyzed the distribution, aggregation, and morphological characteristics of rural settlements in Lixian County. Furthermore, the Maxent model was used to evaluate the suitability of rural settlements, focusing on the suitability distribution and its changes, as well as the impact of environmental factors on rural settlement suitability in Lixian County. This study expands the applicability of the Maxent model. The main conclusions are as follows:

• Rural settlements in Lixian County are constrained by an east-high, west-low elevation gradient, exhibiting a linear distribution along river valleys. In the three periods, rural settlements displayed a clustered distribution pattern. The density of settlements revealed significant east-west differences, with denser distributions in the eastern regions and sparse distributions in the west. Over the 20 years, the scale and morphology of rural settlements became more complex, and spatial continuity and connectivity improved, reflecting a trend toward land use intensification.
• The area of highly suitable land for rural settlements has shown a decreasing trend year by year, while over 85% of the land remains unsuitable for settlement distribution. The four suitability levels for rural settlements remained relatively stable, with transitions occurring mainly between adjacent levels. Transitions between unsuitable and highly suitable levels were almost non-existent. Natural environmental factors, such as elevation and slope, largely determined the suitability levels, while socioeconomic factors, such as cultivated land and roads, contributed to improvements in settlement suitability.
• The suitability of rural settlements is significantly influenced by both natural environmental and socioeconomic factors. High suitability was observed in areas below 2500 m in elevation, with slopes less than 25° and within approximately 1000 m of cultivated land and roads. During the period from 2010 to 2020, the importance of roads increased significantly due to upgrades in the road network, confirming the critical role of transportation infrastructure in determining settlement suitability.