Investigating spatial distribution characteristics and influencing factors of traditional villages in the Yellow River Basin from the perspective of human–land relationship

Abstract

Traditional villages record long-term interactions between settlement activities and environmental conditions. The Yellow River Basin-related regions provide a useful setting for examining these interactions because they include marked geomorphological gradients, river constraints, long settlement histories, and dense cultural heritage resources. Using 2,022 traditional village points from nine provincial-level regions, this study combined the Nearest Neighbor Index, DBSCAN, Kernel Density Estimation, cultural-type stratification, and Geodetector to examine spatial clustering and associated human–land factors. The villages showed a statistically significant clustered distribution, with a nearest neighbor ratio of 0.299161, a z-score of -61.162528, and p < 0.001 (reported as 0 by ArcGIS). DBSCAN identified seven regional-scale clusters and 166 noise points under MinPts = 24 and Eps = 100 km; relatively high-density areas were concentrated in the middle and lower Yellow River-related regions. Geodetector results showed that national key cultural relic protection units had the largest q-value among the selected variables (q = 0.1755), followed by slope (q = 0.0536), intangible cultural heritage items (q = 0.0320), transportation network density (q = 0.0238), and river network density (q = 0.0229). These findings provide spatial references for conservation prioritization and differentiated heritage governance. The statistical associations should not be interpreted as evidence of historical causality, settlement typology, or tourism-development effects.

Introduction

Traditional villages are spatial carriers of long-term interactions between human settlement activities and local environmental conditions. Their location, morphology, continuity, and preservation are related not only to architectural heritage and cultural memory, but also to terrain constraints, water accessibility, agricultural land use, transportation connections, social organization, and historical-cultural accumulation¹. From the perspective of coupled human and natural systems, rural settlements are formed and transformed through feedback between environmental processes and human activities^2–7. The spatial distribution of traditional villages therefore reflects the long-term adaptation of rural communities to local land systems, water systems, and social development processes. Identifying where traditional villages are concentrated or dispersed, and which factors are associated with their spatial differentiation, is necessary for interpreting them as both cultural heritage resources and outcomes of human–environment interaction.

The Yellow River Basin provides a suitable regional context for examining this relationship. As one of China’s major river-basin systems, it has been shaped by geomorphological gradients, river-course changes, periodic flooding, loess landforms, agricultural development, transportation corridors, and historical-cultural accumulation. River dynamics and flood disturbance have repeatedly affected land-use conditions, agricultural production, settlement security, and regional development^8,9. The middle reaches are closely associated with loess landforms, river-valley settlement corridors, and long-term agricultural settlement, whereas the lower reaches have been strongly affected by river-course instability, alluvial plains, and human intervention in river control. The Yellow River Basin is therefore not only a geographical setting for traditional village distribution, but also a dynamic human–water–land system in which environmental constraints and settlement activities have interacted over a long historical period^10–13.

Traditional villages in the Yellow River Basin-related regions are well suited for examining the relationship between settlement heritage and environmental adaptation. Their spatial distribution may reflect the combined effects of terrain suitability, hydrological accessibility, agricultural settlement continuity, transportation conditions, and historical-cultural accumulation. Their modern preservation is also shaped by administrative governance, infrastructure development, population change, heritage recognition systems, and conservation policy^14–17. These characteristics make the Yellow River Basin a meaningful study context for analyzing traditional village distribution from a human–land relationship perspective, especially when the analysis considers not only where villages are located, but also how land-system conditions, water-related constraints, accessibility, and heritage-resource accumulation correspond to village distribution.

Existing studies provide an important foundation for understanding the spatial distribution of traditional villages. National-scale research has shown that Chinese traditional villages are unevenly distributed and that their spatial patterns are associated with natural geographical conditions, socioeconomic development, transportation accessibility, and cultural resources^18,19. Provincial and regional studies have further examined how topography, hydrology, road networks, population, economy, and heritage resources influence traditional village distribution in different geographical contexts, including Fujian, Jiangsu, Inner Mongolia, Henan, Guizhou, and other regions^20–29. These studies show that traditional villages are shaped by multiple factors rather than by a single environmental or socioeconomic condition. They also show that spatial methods such as nearest-neighbor analysis, kernel density estimation, spatial autocorrelation, and Geodetector are now commonly used to identify spatial clustering and influencing factors in traditional village research.

Traditional village research has also developed across macro-, meso-, and micro-scale analyses. Macro-scale studies reveal national or regional spatial imbalances and broad influencing factors^18,19. Meso-scale studies, especially those based on watersheds or county-level administrative units, connect settlement distribution with hydrological systems, terrain conditions, transportation network density, and governance units^23,30. Micro-scale studies of village morphology, architectural adaptation, local construction traditions, and cultural landscapes provide detailed evidence of how settlements respond to local environmental and social conditions^1,31–34. However, these levels of analysis are not always well connected. Large-scale spatial studies can identify broad distribution patterns but often lack village-level mechanism explanation, whereas micro-scale studies can reveal local adaptation but cannot fully explain regional spatial differentiation. This gap calls for a multi-level interpretation that links regional clustering, county-level influencing factors, and human–land analytical logic.

Traditional village studies in the Yellow River Basin have also increased in recent years. Existing research has analyzed the spatiotemporal distribution characteristics, clustering patterns, and driving factors of traditional villages in the basin or in the provinces through which the Yellow River flows^35–37. Related studies on rural settlement and environmental change in the Yellow River region have shown that river-course changes, flooding, geomorphic evolution, and settlement distribution can be closely connected^8,9,38,39. These studies provide valuable empirical references for the present research. However, three issues remain. First, the Yellow River Basin is often treated as a spatial container for mapping and statistical analysis, while its distinctive human–land and human–water–land characteristics—especially river-course change, flood disturbance, terrain transition, hydrological accessibility, and settlement continuity—are not fully integrated into the analytical framework. Second, existing studies have paid limited attention to the internal cultural-spatial heterogeneity of traditional villages within the basin-related regions. Third, the relationship between statistical association and historical mechanism is not always clearly distinguished, which may lead to overinterpretation of spatial clustering results.

The human–land relationship perspective provides a framework for addressing these issues, but it needs to be operationalized clearly. In this study, “land” refers to the environmental background and physical constraints that affect settlement formation and persistence, including elevation, slope, precipitation, temperature, and river network density. “Human” refers to socioeconomic activities, accessibility conditions, population background, and heritage-resource recognition, including transportation network density, population size, GDP, national intangible cultural heritage items, national key cultural relic protection units, and minority population proportion. The interaction between “human” and “land” is not treated as a direct causal claim based on a single statistical model. Instead, it is examined as spatial correspondence among environmental conditions, hydrological background, settlement accessibility, development context, and historical-cultural accumulation. This perspective guides the study in three ways: it organizes variable selection into natural environmental, socioeconomic, and historical-cultural dimensions; it supports the use of county-level administrative units as the spatial scale for integrating multi-source data and governance-related interpretation; and it requires cautious interpretation of spatial clustering and factor detection results as evidence of spatial association rather than direct proof of historical causality.

This conceptualization distinguishes the present study from conventional spatial differentiation analysis. A conventional analysis may identify where traditional villages are concentrated and which variables have higher statistical explanatory power. This study further asks how these spatial patterns and variable associations can be interpreted within a river-basin human–land system. DBSCAN, Kernel Density Estimation, and Geodetector are therefore not treated as isolated technical tools. They are combined to examine overall clustering, regional density differentiation, cultural-spatial heterogeneity, factor explanatory power, and interaction effects within a single analytical framework. The methodological contribution does not lie in proposing new algorithms, but in integrating complementary spatial methods to connect distribution pattern identification with human–land interpretation.

GIS-based spatial methods are widely used in settlement and traditional village research because they can identify point-pattern clustering, density gradients, spatial autocorrelation, and factor-related spatial heterogeneity^40–44. DBSCAN can identify density-based clusters without predefining the number of clusters⁴¹. Kernel Density Estimation can reveal continuous density surfaces⁴². Geodetector can evaluate spatial stratified heterogeneity and interaction effects among explanatory variables⁴⁴. These methods are useful for large-scale regional analysis, but their limitations must be considered. DBSCAN and KDE are sensitive to parameter settings, including neighborhood distance, minimum points, bandwidth, and cell size. Geodetector reveals spatial association and stratified heterogeneity, but it does not directly estimate independent causal effects or fully eliminate multicollinearity among variables. Related spatial models, such as geographically weighted regression, generalized additive models, and variance inflation factor testing, can provide complementary tools for examining spatial nonstationarity, nonlinear relationships, and multicollinearity in future studies^45–47. The interpretation of GIS-based results must therefore remain within the evidential scope of spatial statistical analysis.

Based on these considerations, this study focuses on 2,022 traditional village points in nine provincial-level regions related to the Yellow River Basin: Shaanxi, Shanxi, Henan, Shandong, Inner Mongolia, Sichuan, Gansu, Ningxia, and Qinghai. The study addresses four research questions. First, do traditional villages in the study area show a statistically significant clustered distribution, and where are the main concentration areas located? Second, do different cultural types of traditional villages show internal spatial heterogeneity in clustering intensity and dispersion? Third, how do natural environmental, socioeconomic, and historical-cultural variables explain the county-level spatial differentiation of traditional villages? Fourth, how can the spatial pattern, factor detection, and interaction detection results be interpreted through a human–land relationship framework without overextending statistical association into unsupported causal claims?

The study makes three main contributions. First, it provides a multi-level spatial interpretation of traditional village distribution in the Yellow River Basin-related regions by combining the Nearest Neighbor Index, DBSCAN, Kernel Density Estimation, cultural-type stratification, and Geodetector. Second, it operationalizes the human–land relationship perspective by linking variable selection, spatial scale, and result interpretation to terrain, hydrology, accessibility, regional development, and historical-cultural accumulation. Third, it clarifies the evidential boundary of spatial analysis by distinguishing clustering patterns and factor associations from direct historical causality. The findings provide spatial references for conservation prioritization, heritage-resource monitoring, and differentiated governance, while avoiding unsupported claims about tourism development or rural revitalization effects that are not directly analyzed in this study.

Study area and data sources

Study area

The Yellow River Basin is one of China’s major river-basin regions in terms of geomorphological diversity, agricultural development, settlement history, and cultural accumulation. As a long-term human–land system, the basin includes high-relief mountains, loess landforms, alluvial plains, tributary networks, and areas historically affected by river-course changes and flood disturbance. Research on coupled human and natural systems shows that human settlement patterns are shaped by long-term feedbacks among environmental processes, social organization, and spatial development². In the Yellow River Basin, river dynamics, agricultural development, and social processes are closely connected because historical flooding and river-course changes have repeatedly influenced settlement environments and land-use patterns⁸. The region therefore provides an appropriate geographical context for examining the spatial distribution of traditional villages from a human–land relationship perspective.

The study area consists of nine provincial-level administrative regions associated with the Yellow River Basin: Shaanxi, Shanxi, Henan, Shandong, Inner Mongolia, Sichuan, Gansu, Ningxia, and Qinghai (Fig. 1). This spatial framework was adopted for three reasons. First, these provincial-level regions are closely related to the natural, historical, cultural, and settlement systems of the Yellow River Basin. Second, traditional village lists, socioeconomic statistics, population data, and heritage-resource data in China are generally organized and released by administrative unit. Using provincial-level administrative regions therefore improves consistency and comparability among multi-source datasets. Third, traditional village conservation and management are implemented mainly through administrative governance systems. An administrative regional framework is therefore suitable for linking spatial analysis with conservation and governance interpretation.

Fig. 1.

Study area, Yellow River-related river system, elevation background, and provincial distribution of traditional villages.(a) Location of the nine provincial-level study regions in China; (b) elevation pattern based on DEM data; (c) major river system and administrative boundary with DEM/hillshade background; (d) number of traditional villages in each province.

The study area is not equivalent to the strict hydrological boundary of the Yellow River Basin. The hydrological basin boundary represents the drainage area of the river system, whereas the nine provincial-level administrative regions provide the spatial framework for data integration, statistical analysis, village-point verification, and governance-oriented interpretation. In this study, the river system is used to represent the hydrological and environmental background, while administrative boundaries are used to organize the traditional village dataset and county-level explanatory variables.

The nine provincial-level regions contain substantial environmental and cultural variation. The western and northwestern parts of the study area, including Qinghai, Gansu, Ningxia, Sichuan, and western Inner Mongolia, are characterized by higher elevation, complex terrain, and relatively fragmented settlement environments. The middle reaches, especially Shaanxi and Shanxi, are closely associated with loess landforms, river-valley settlement corridors, and long-term agricultural and cultural development. The eastern regions, including Henan and Shandong, contain lower-elevation plains and historically intensive agricultural settlement areas. These environmental gradients and historical-cultural differences make the study area suitable for examining regional concentration, cultural-spatial heterogeneity, and the combined effects of natural, socioeconomic, and historical-cultural factors on traditional village distribution.

Recent studies have examined the spatial distribution and influencing factors of traditional villages at national and Yellow River Basin scales^18,35–37. These studies provide an important basis for analyzing traditional villages using GIS-based spatial methods. The present study combines traditional village point data, environmental variables, socioeconomic indicators, heritage-resource data, and cultural-type stratification to examine how traditional village distribution is associated with land-system conditions, human-system activities, and historical-cultural accumulation in the nine provincial-level regions related to the Yellow River Basin.

Data sources

This study used multi-source data, including traditional village point data, administrative boundary data, Yellow River Basin boundary and river-network data, DEM data, soil-type data, transportation network data, climate data, socioeconomic data, population data, and heritage-resource data. All datasets were accessed in April 2025. Table 1 summarizes the data categories, data contents, sources, access links, processing methods, analytical uses, and access dates.

Table 1.

Data sources and processing methods used in this study.

Data category	Data content	Source	Access link / URL	Processing and use	Access date
Traditional village data	List and coordinates of traditional villages	Official lists of Chinese traditional villages released by MOHURD and related national departments; Amap-derived geocoding information	Official MOHURD notices for the first to sixth batches were used. Sixth-batch notice: https://www.mohurd.gov.cn/gongkai/zhengce/zhengcefilelib/202303/20230320_770845.html; Chinese Traditional Village directory: https://www.dmctv.cn/directories.aspx; Amap Geocoding API: https://lbs.amap.com/api/webservice/guide/api/georegeo	Villages located in the nine provincial-level study regions were retained; coordinates were checked, converted from GCJ-02 to WGS84, and used to construct a 2,022-point village dataset	April 2025
Administrative boundary data	Provincial- and county-level administrative boundaries	Guihua Cloud	https://guihuayun.com/	Used to define the nine provincial-level study regions, verify village locations, and construct county-level spatial analysis units	April 2025
DEM data	Elevation data	European Space Agency (ESA), Copernicus DEM GLO-30 China dataset	https://dataspace.copernicus.eu/explore-data/data-collections/copernicus-contributing-missions/collections-description/COP-DEM	Used to extract average elevation and slope and support elevation-background mapping	April 2025
Basin boundary and river data	Yellow River Basin boundary and river network	National Basic Geographic Information Database	https://www.ngcc.cn/dlxxzy/	Used to represent the hydrological background, map the river system, and calculate river network density	April 2025
Transportation data	Road network	OpenStreetMap	https://www.openstreetmap.org/; https://wiki.openstreetmap.org/wiki/Downloading_data	Used to calculate transportation network density	April 2025
Climate data	Precipitation and temperature	National Tibetan Plateau Data Center	https://data.tpdc.ac.cn/en/; precipitation dataset: https://data.tpdc.ac.cn/en/data/faae7605-a0f2-4d18-b28f-5cee413766a2; mean temperature dataset: https://data.tpdc.ac.cn/en/data/71ab4677-b66c-4fd1-a004-b2a541c4d5bf	Used to calculate county-level precipitation and temperature indicators	April 2025
Socioeconomic data	GDP and related county-level socioeconomic indicators	China County Statistical Yearbook; National Bureau of Statistics of China; China Statistics Press	https://www.stats.gov.cn/zs/tjwh/tjkw/tjzl/202302/t20230215_1908004.html; https://www.stats.gov.cn/sj/ndsj/	Used to calculate county-level socioeconomic indicators	April 2025
Population data	Resident population and minority population proportion	Seventh National Population Census; National Bureau of Statistics of China	https://www.stats.gov.cn/sj/pcsj/rkpc/d7c/; https://www.stats.gov.cn/english/PressRelease/202105/t20210510_1817185.html	Used to calculate county-level population size and minority population proportion	April 2025
Heritage-resource data	National intangible cultural heritage items; national key cultural relic protection units	Official national cultural heritage and intangible cultural heritage lists	National intangible cultural heritage representative items: https://www.ihchina.cn/project.html; National Cultural Heritage Administration service window: https://app.gjzwfw.gov.cn/jmopen/webapp/html5/gjwwjzwfwck/index.html; eighth-batch national key cultural relic protection units: https://www.ncha.gov.cn/art/2019/10/18/art_2289_157100.html	Used to calculate county-level historical-cultural indicators	April 2025
Cultural-type data	Sanjin, Guanzhong, Central Plains, Qilu, Hehuang, Longyou, Longxi, Bashu, and Grassland cultural types	Historical context, regional cultural geography, dialectal divisions, and architectural traditions	Not applicable as an independent online dataset; the classification was constructed by the authors based on relevant literature and regional cultural-geographical interpretation	Used as exploratory stratification for cultural-spatial heterogeneity analysis	April 2025
Soil type data	Soil type	Resource and Environment Science and Data Center, Chinese Academy of Sciences; 1:1,000,000 Soil Map of the People’s Republic of China, compiled by the National Soil Survey Office	https://www.resdc.cn/data.aspx?DATAID=145	Used to describe the environmental background of the study area; auxiliary environmental data, not used as a Geodetector variable	April 2025

Note: Access links were added to improve the traceability and reproducibility of the study. For several administrative and geographic datasets, stable platform-entry links are provided where permanent dataset-specific download pages were not available. Cultural-type data were constructed by the authors based on historical context, regional cultural geography, dialectal divisions, and architectural traditions; no independent online dataset was used for this classification.

The traditional village dataset was constructed from the official lists of Chinese traditional villages released by the Ministry of Housing and Urban-Rural Development of the People’s Republic of China and related national departments⁴⁸. The first to sixth batches of Chinese traditional villages were included. Villages located within the nine provincial-level study regions were retained, resulting in a final dataset of 2,022 traditional village points. Village coordinates were obtained and checked using Amap-derived geocoding information. Amap-derived coordinates in mainland China are commonly provided in the GCJ-02 coordinate system; all village coordinates were converted to WGS84 before spatial analysis.

Administrative boundary data were obtained from Guihua Cloud⁴⁹ and used to define the nine provincial-level study regions, support spatial overlay, verify village locations, and construct county-level spatial analysis units. Yellow River Basin boundary and river-network data were obtained from the National Basic Geographic Information Database⁵⁰ and used to map the major river system, represent the hydrological background, and calculate river network density. DEM data were obtained from the European Space Agency Copernicus DEM GLO-30 China dataset⁵¹ and used to represent elevation and derive terrain-related variables, including elevation and slope. Soil-type data were obtained from the Resource and Environment Science and Data Center, Chinese Academy of Sciences⁵², and were used to describe the environmental background of the study area. Transportation network data were obtained from OpenStreetMap⁵³ and used to calculate transportation network density. Climate data, including precipitation and temperature, were obtained from the National Tibetan Plateau Data Center⁵⁴ and used to calculate county-level climate indicators. Socioeconomic and population data were obtained from the China County Statistical Yearbook⁵⁵ and the Seventh National Population Census⁵⁶. Heritage-resource data, including national key cultural relic protection units and national intangible cultural heritage items, were obtained from official national heritage lists^57,58. The cultural-type data were constructed by the authors according to historical context, regional cultural geography, dialectal divisions, and architectural traditions; therefore, no independent online dataset was used for this exploratory cultural classification.

Coordinate preprocessing and data quality control

Coordinate preprocessing and data quality control were conducted before spatial statistical analysis to improve the reliability of the traditional village point dataset. Village coordinates obtained from Amap-derived geocoding information were converted to WGS84. The coordinate reference systems of all spatial datasets were unified before spatial overlay, mapping, and spatial analysis.

Duplicate detection was conducted for village names and coordinate points. The final dataset contains no duplicate village names and no duplicate coordinate points. Spatial overlay analysis was then used to examine whether all village points were located within the nine provincial-level study regions. All 2,022 traditional village points fall within the defined study area. Potential abnormal points were checked by comparing village-point locations with their corresponding administrative attributes, reducing the possibility of coordinate mismatch and geocoding errors.

Boundary effects were considered when interpreting the spatial results. Because the study uses nine provincial-level administrative regions rather than the strict hydrological basin boundary, the spatial statistics represent the distribution pattern within the defined administrative study area. In the kernel density estimation and clustering interpretation, edge areas were interpreted cautiously, and the density surface was clipped to the nine provincial-level study regions. The results should therefore be understood as an analysis of the spatial distribution and influencing factors of traditional villages within the nine Yellow River Basin-related provincial-level regions, rather than as a strict hydrological-boundary analysis of the entire Yellow River Basin.

Human–land analytical framework and methodology

Traditional villages are spatial carriers of long-term interactions among environmental conditions, settlement activities, and historical-cultural accumulation. In the Yellow River Basin-related regions, village distribution is associated not only with natural background conditions, such as terrain, hydrology, climate, and elevation, but also with human-system factors, including transportation accessibility, population distribution, economic development, and the spatial accumulation of heritage resources. This study therefore constructs a human–land analytical framework to examine the spatial distribution and associated influencing factors of traditional villages.

In this framework, “land” refers to the environmental background and physical constraints that affect settlement formation and persistence, including elevation, slope, precipitation, temperature, and river network density. “Human” refers to socioeconomic activities, accessibility conditions, and heritage-resource recognition, including transportation network density, population size, GDP, national intangible cultural heritage items, national key cultural relic protection units, and minority population proportion. The human–land perspective guides the study in three ways. First, it organizes explanatory variables into natural environmental, socioeconomic, and historical-cultural dimensions. Second, it supports the use of county-level administrative units as the spatial analysis unit, because these units can integrate multi-source statistical data and correspond to governance-oriented heritage protection. Third, it requires spatial clustering and statistical association to be interpreted as evidence of spatial correspondence among human–land factors rather than as direct proof of historical causality.

To avoid treating spatial methods as isolated technical procedures, this study links the methodological workflow more explicitly with the spatial context of traditional villages in the Yellow River Basin-related regions. In this study, the relationship between the human–land framework and traditional village space is examined through five analytical dimensions: watershed environmental constraints, river–land spatial coupling, settlement accessibility and development context, historical-cultural accumulation, and cultural-spatial heterogeneity.

The watershed environmental constraint dimension focuses on elevation, slope, precipitation, temperature, and river-network density. These variables describe the terrain, climatic, and hydrological background that may correspond to settlement location, agricultural suitability, water accessibility, and long-term village persistence. The river–land spatial coupling dimension further considers whether village concentrations correspond to terrain-transition zones, river-valley environments, loess landforms, and lower-elevation alluvial areas within the Yellow River Basin-related regions. In this dimension, DBSCAN and Kernel Density Estimation are used to identify regional-scale clustering and continuous density gradients.

The settlement accessibility and development context dimension is represented by transportation network density, population size, and GDP. These indicators are used to describe present accessibility, demographic background, regional development context, and governance-related conditions at the county level. The historical-cultural accumulation dimension is represented by national intangible cultural heritage items, national key cultural relic protection units, and minority population proportion. These variables are used to examine whether the spatial concentration of traditional villages corresponds to recognized heritage resources and ethnic-cultural background.

The cultural-spatial heterogeneity dimension is examined through cultural-type stratification. Traditional villages were divided into Sanjin, Guanzhong, Central Plains, Qilu, Hehuang, Longyou, Longxi, Bashu, and Grassland cultural types. DBSCAN and KDE were applied to these cultural types to identify differences in clustering intensity and dispersion. In this way, NNI tests the overall clustering tendency, DBSCAN identifies regional-scale density clusters, KDE reveals continuous density differentiation, cultural-type stratification captures internal cultural-spatial heterogeneity, and Geodetector evaluates the spatial explanatory correspondence between county-level village counts and selected human–land variables. These methods are therefore used as complementary tools for linking spatial pattern identification with human–land interpretation, rather than as independent calculations detached from the Yellow River Basin context.

The methodological workflow consists of four steps. First, the Nearest Neighbor Index is used to test whether the overall distribution of traditional villages is clustered, random, or dispersed. Second, DBSCAN and kernel density estimation are used to identify regional-scale spatial clusters and continuous density differentiation. Third, cultural-type stratification is used to examine whether traditional villages belonging to different cultural types show different clustering characteristics. Fourth, Geodetector is used to assess the relative explanatory power and interaction effects of selected human–land variables at the county level. These methods are complementary: NNI provides a global statistical test, DBSCAN identifies density-based clusters, KDE visualizes continuous spatial density, and Geodetector evaluates spatial stratified heterogeneity and factor interactions (Fig. 2).

Fig. 2.

Research framework.

The methodological contribution of this framework lies in integrating complementary spatial methods within a human–land analytical logic. GIS-based methods are not treated as isolated technical tools. Instead, the framework links overall clustering, density differentiation, cultural-type heterogeneity, factor detection, and interaction analysis into a unified workflow. This integrated approach strengthens the logical connection among variable selection, spatial scale, parameter interpretation, and evidence boundaries. It is suitable for examining traditional village distribution in a large river-basin context where environmental constraints, settlement accessibility, and historical-cultural accumulation overlap spatially.

Indicator system and variable calculation

The dependent variable for Geodetector analysis is the number of traditional villages in each county-level administrative unit. Traditional village points were spatially joined with county-level administrative boundaries to calculate the village count for each county. This county-level count was used as the dependent variable Y.

The explanatory variables were selected according to the human–land analytical framework, previous studies on traditional village spatial distribution, data availability, and expert consultation. Previous studies have shown that the spatial distribution of traditional villages is closely associated with natural environmental conditions, socioeconomic context, and historical-cultural accumulation. At the national scale, traditional villages in China have been found to be related to terrain, altitude, slope, climatic background, population, economic development, transportation conditions, hydrological conditions, and historical-cultural resources^18,19,59,60. Regional studies further indicate that topography, river systems, road networks, socioeconomic conditions, and heritage resources jointly influence the spatial distribution of traditional villages in different geographical contexts, including Fujian Province, the Yangtze River Basin, the Yellow River Basin, the lower Yangtze River Basin, Hakka cultural regions, Jiangnan cultural regions, and Guizhou^{20,23,30,35,59–63}. These studies provide the literature basis for including elevation, slope, precipitation, temperature, river network density, transportation network density, population size, GDP, intangible cultural heritage items, national key cultural relic protection units, and minority population proportion in the present indicator system.

Within the human–land analytical framework of this study, the selected indicators were organized into three dimensions. The natural environmental dimension includes average elevation, average slope, annual precipitation, average annual temperature, and river network density, which represent terrain conditions, climatic background, and hydrological constraints related to settlement formation and persistence. The socioeconomic dimension includes transportation network density, population size, and GDP, which reflect accessibility conditions, demographic background, regional development context, and governance-related capacity. The historical-cultural accumulation dimension includes the number of national intangible cultural heritage items, the number of national key cultural relic protection units, and minority population proportion, which represent recognized cultural heritage resources and ethnic-cultural background associated with the continuity of traditional village landscapes.

Fifteen experts from heritage conservation, human geography, and urban-rural planning participated in a multi-round consultation process to further screen and refine the candidate indicators. Candidate indicators were assessed according to theoretical relevance, data reliability, spatial comparability, interpretability, and suitability for county-level analysis. Based on the literature review, human–land analytical framework, data availability, and expert consultation, the final indicator system includes 11 variables: average elevation, average slope, annual precipitation, average annual temperature, river network density, transportation network density, population size, GDP, number of national intangible cultural heritage items, number of national key cultural relic protection units, and minority population proportion (Table 2).

Table 2.

Indicator system and variable calculation.

Dimension	Code	Variable	Meaning and calculation
Natural environment	X1	Average elevation	Mean DEM elevation within each county-level administrative unit
Natural environment	X2	Average slope	Mean slope derived from DEM within each county-level administrative unit
Natural environment	X3	Annual precipitation	Mean annual precipitation within each county-level administrative unit
Natural environment	X4	Average annual temperature	Mean annual temperature within each county-level administrative unit
Natural environment	X5	River network density	Total river length within a county divided by county area
Socioeconomic condition	X6	Transportation network density	Total road length within a county divided by county area
Socioeconomic condition	X7	Population size	Total resident population of each county-level administrative unit
Socioeconomic condition	X8	GDP	Gross domestic product of each county-level administrative unit
Historical-cultural accumulation	X9	Number of national intangible cultural heritage items	Count of national intangible cultural heritage items within each county
Historical-cultural accumulation	X10	Number of national key cultural relic protection units	Count of national key cultural relic protection units within each county
Historical-cultural accumulation / ethnic-cultural background	X11	Minority population proportion	Proportion of minority population in each county-level administrative unit

Continuous variables were discretized using the Jenks natural breaks method before Geodetector analysis. Jenks classification was selected because it reduces within-class variance and increases between-class differences, making it suitable for identifying spatial stratified heterogeneity in continuous geographical variables. Quantile classification and equal interval classification were also used as alternative discretization methods in the sensitivity test to evaluate whether the main q-value ranking remained stable.

Nearest neighbor index analysis

The nearest neighbor index is used to assess the spatial arrangement of geographic features. By comparing observed distances to those expected under a random distribution, patterns can be classified as clustered, dispersed, or random.

The calculation formula is:

1

2

3

where R is the nearest neighbor ratio, is the observed mean nearest-neighbor distance, is the expected mean nearest-neighbor distance under complete spatial randomness, is the nearest-neighbor distance of point i, n is the number of traditional village points, and A is the area of the study region.

DBSCAN spatial clustering algorithm

DBSCAN was used to identify density-based spatial clusters of traditional villages. Unlike methods that require the number of clusters to be specified in advance, DBSCAN identifies clusters according to local point density and distinguishes clustered points from noise points. This feature makes it suitable for analyzing traditional village distributions with uneven density and irregular spatial forms.

DBSCAN is controlled mainly by two parameters: Eps and MinPts. Eps defines the neighborhood distance threshold, and MinPts defines the minimum number of points required to form a core point. A point is classified as a core point if at least MinPts points fall within its Eps-neighborhood. Points reachable from core points form clusters, while points that do not meet the density-reachability condition are classified as noise.

The criterion for identifying a core point is expressed as:

4

5

6

where is the ε-neighborhood of point p, D is the point dataset, dist(p, q) is the distance between points p and q, ε is the neighborhood radius, and MinPts is the minimum number of points required to define a core point.

The final DBSCAN parameters for the final 2,022-point dataset were set as MinPts = 24 and Eps = 100 km. The parameter selection was based on the spatial scale of the study area, the sample size of the village points, the K-distance curve, and comparison of clustering results under different parameter combinations. The purpose of using Eps = 100 km was to identify regional-scale density clusters rather than village-level settlement boundaries. Therefore, the DBSCAN results are interpreted as regional aggregation patterns of traditional villages, not as deterministic cultural or settlement boundaries.

Under the final 2,022-point dataset, the parameter setting MinPts = 24 and Eps = 100 km identifies 7 clusters and 166 noise points. Noise points represent villages that do not belong to regional-scale density clusters under the selected parameter setting. They should not be interpreted as unimportant villages; instead, they indicate spatially dispersed villages that may require site-specific protection and further field investigation.

For cultural-type analysis, DBSCAN was also applied to different cultural-type groups to examine internal spatial heterogeneity. Because different cultural types have unequal sample sizes and spatial ranges, the cultural-type DBSCAN results are used as exploratory stratified analysis rather than direct quantitative comparison among cultural regions. The results provide evidence of differences in clustering tendency and noise ratios among cultural-type groups, but they do not directly prove cultural mechanisms, migration routes, commercial networks, or settlement typologies.

Kernel density estimation

Kernel Density Estimation was used to visualize the continuous density surface of traditional villages and to identify relative high-density areas. KDE converts discrete village points into a continuous raster surface by calculating the density contribution of nearby points within a defined search radius. It is useful for revealing spatial density gradients that cannot be fully captured by point clustering alone.

The KDE function is defined as:

7

where f(x) denotes the estimated density at location x, n is the number of sample points, h represents the bandwidth, x_i corresponds to the position of each sample point, and K is the kernel function.

KDE was implemented in ArcGIS using the WGS84 coordinate system. The ArcGIS default kernel density algorithm was used, and the output raster was clipped to the nine provincial-level study regions. The KDE result is interpreted as a relative density surface for identifying areas with higher or lower concentration of traditional villages within the study area. It is not used for absolute quantitative comparison across different projections or external regions.

Edge effects were considered during interpretation. Because the study area is defined by nine provincial-level administrative regions rather than a strict hydrological basin boundary, density values near the boundary may be affected by clipping. Therefore, KDE results are used to support regional pattern interpretation and are interpreted together with DBSCAN and NNI results.

Geodetector model

Geodetector was used to examine the spatial stratified heterogeneity of traditional village distribution and to assess the explanatory power of selected human–land variables at the county level. Geodetector is suitable for identifying whether the spatial distribution of a dependent variable corresponds to the spatial stratification of explanatory variables. In this study, the dependent variable (Y) was the number of traditional villages in each county-level administrative unit. The explanatory variables included the 11 indicators selected in the human–land analytical framework: average elevation, average slope, annual precipitation, average annual temperature, river network density, transportation network density, population size, GDP, number of national intangible cultural heritage items, number of national key cultural relic protection units, and minority population proportion.

The factor detector was used to calculate the explanatory power of each variable. The q-statistic was calculated as follows:

8

where q measures the explanatory power of an explanatory variable for the spatial stratified heterogeneity of the dependent variable, h is the stratum of the explanatory variable, L is the number of strata, N_h and N are the numbers of units in stratum h and the whole study area, respectively, and are the variances of the dependent variable within stratum h and the whole study area, respectively.

Before Geodetector analysis, continuous explanatory variables were discretized using the Jenks natural breaks method. Jenks classification was selected because it reduces within-class variance and increases between-class differences, which is suitable for detecting spatial stratified heterogeneity in continuous geographical variables. The dependent variable (Y), namely the number of traditional villages in each county-level administrative unit, was calculated by spatially joining the traditional village point dataset with county-level administrative boundaries.

The interaction detector was used to examine whether pairs of explanatory variables jointly increased or reduced their explanatory power for the county-level distribution of traditional villages. The interaction q-value (q(X_i∩X_j)) was compared with the corresponding single-factor q-values (q(X_i)) and (q(X_j)). The interaction types were identified according to the criteria shown in Table 3.

Table 3.

Interaction types of Geodetector.

Interaction type	Basis for judgment
Nonlinear weakening	q(X1∩X2)< Min(q(X1),q(X2))
Single-factor weakening	Min(q(X1),q(X2))< q(X1∩X2)< Max(q(X1),q(X2))
Two-factor enhancement	q(X1∩X2)> Max(q(X1),q(X2))
Independent	q(X1∩X2) = q(X1) + q(X2)
Nonlinear enhancement	q(X1∩X2)> q(X1) + q(X2)

In this study, the Geodetector results were interpreted as evidence of spatial explanatory correspondence rather than direct causal effects. The factor detector identifies the degree to which the spatial stratification of a selected variable corresponds to the spatial differentiation of traditional village counts. The interaction detector identifies whether two variables provide stronger joint spatial explanatory power than either variable alone. These results do not prove direct historical causality, settlement formation mechanisms, or independent marginal effects. Therefore, Geodetector was used to support the interpretation of human–land spatial associations among natural environmental conditions, socioeconomic context, and historical-cultural accumulation.

Parameter sensitivity and robustness assessment

Parameter sensitivity and robustness assessment were conducted to improve methodological transparency. For DBSCAN, alternative Eps and MinPts settings were compared according to the number of clusters, number of noise points, noise ratio, and stability of the main regional clustering pattern. The final setting of MinPts = 24 and Eps = 100 km was selected because it provided a balanced regional-scale clustering structure for the 2,022-point dataset and avoided both excessive fragmentation and over-aggregation.

For KDE, the search radius, cell size, coordinate system, and boundary clipping were explicitly reported to improve reproducibility. For Geodetector, discretization sensitivity was assessed by comparing Jenks natural breaks with alternative classification strategies. The main result remained that historical-cultural accumulation, represented by national key cultural relic protection units, provides the most stable spatial signal among the selected variables, while GDP remains low in explanatory value. Detailed parameter comparisons can be reported in the supplementary materials.

Methodological applicability and limitations

The combination of NNI, DBSCAN, KDE, and Geodetector provides a multi-level spatial analytical framework. NNI identifies whether the overall village distribution is significantly clustered. DBSCAN identifies regional-scale density clusters and noise points. KDE reveals continuous density differentiation. Geodetector evaluates the relative explanatory power and interaction effects of selected human–land variables. Together, these methods support a structured interpretation of traditional village distribution from overall pattern identification to influencing-factor analysis.

The methods also have limitations. DBSCAN results are sensitive to Eps and MinPts, and different parameter combinations may produce different cluster boundaries and noise-point proportions. KDE results are affected by search radius, cell size, coordinate system, and boundary clipping. Geodetector identifies spatial stratified heterogeneity and interaction effects, but it does not estimate independent marginal effects in the same way as regression models and cannot fully address multicollinearity among explanatory variables. Therefore, the results are interpreted as spatial association and explanatory correspondence rather than as direct causal mechanisms.

The county-level analysis scale also affects interpretation. County-level units are suitable for integrating multi-source data and linking results with administrative governance, but they may obscure micro-scale village processes such as household livelihood, clan organization, architectural adaptation, land-use change, and local conservation practice. The cultural-type analysis is exploratory and is used to identify internal spatial heterogeneity, not to define fixed cultural boundaries or prove specific historical mechanisms.

This methodological framework is appropriate for identifying regional-scale spatial distribution patterns and associated human–land factors. Further village-level fieldwork, historical documentation, multi-period data, and advanced spatial statistical models are needed to explain local formation processes and causal mechanisms in greater detail.

Results

Overall spatial distribution characteristics of traditional villages

The Nearest Neighbor Index showed that traditional villages in the nine provincial-level regions related to the Yellow River Basin were significantly clustered. For the 2,022 traditional village points, the observed mean nearest-neighbor distance was 9872.183379 m, whereas the expected mean nearest-neighbor distance under random distribution was 32999.610193 m. The nearest neighbor ratio was 0.299161, with a z-score of -61.162528 and p < 0.001 (reported as 0 by ArcGIS). These results indicate that the village points were more clustered than expected under a random point pattern (Fig. 3).

Fig. 3.

Nearest Neighbor Index analysis.

DBSCAN further identified the regional-scale clustering structure of traditional villages. Under the parameter setting of MinPts = 24 and Eps = 100 km, seven spatial clusters and 166 noise points were identified. Clustered points accounted for most of the traditional village dataset, whereas noise points represented villages that were relatively dispersed under the selected regional-scale clustering threshold. The DBSCAN result shows that traditional villages were not evenly distributed across the study area, but formed several identifiable regional aggregation areas, especially in the middle and lower Yellow River-related regions.

Kernel Density Estimation provided a continuous density surface for interpreting the spatial intensity of traditional village distribution. The density surface showed that relatively high-density areas were mainly concentrated in the middle and lower Yellow River-related regions, while the western and northwestern parts of the study area had lower and more fragmented density values. The KDE result is interpreted as a relative density pattern within the nine provincial-level study regions, rather than as an absolute density comparison across different projection systems or external regions (Fig. 4).

Fig. 4.

Kernel density estimation results of traditional village distribution in the Yellow River Basin-related regions.

Taken together, the NNI, DBSCAN, and KDE results show that traditional villages in the study area were characterized by significant overall clustering, regional-scale aggregation, and differentiated density gradients. The density classes shown in the KDE map describe relative density gradients. They do not represent fixed settlement hierarchies or deterministic spatial structures. The main value of this result is to identify where traditional villages are more concentrated or more dispersed, providing a spatial basis for cultural-type stratification and county-level influencing-factor analysis.

Spatial differentiation among cultural types

The cultural-type stratified analysis showed that the overall clustering pattern contained internal heterogeneity. Different cultural-type groups contributed differently to the regional distribution of traditional villages. Under the cultural-type DBSCAN analysis, the Hehuang cultural type had the lowest noise ratio, at 1.2%, indicating the strongest within-group spatial concentration under the selected parameters (Fig. 5). Bashu, Guanzhong, Sanjin, and Central Plains cultural types also showed relatively low noise ratios, at 9.6%, 10.1%, 10.9%, and 11.3%, respectively. These results indicate that these cultural types formed relatively compact spatial concentrations.

Fig. 5.

DBSCAN clustering results of different types of traditional villages.

The Qilu and Longyou cultural types showed intermediate clustering characteristics. Their village distributions included both concentrated local groups and more dispersed points, indicating a transitional pattern between compact concentration and dispersed distribution. In contrast, the Longxi and Grassland cultural types had higher noise ratios, at 26.9% and 27.4%, respectively. This result indicates that traditional villages belonging to these two cultural types were more spatially dispersed under the selected clustering parameters (Table 4).

Table 4.

Noise ratios of different types of traditional villages.

Culture Type	V-BS	V-CY	V-GZ	V-SJ	V-ZY	V-LX	V-LY	V-QL	V-HH
Noise point rate	9.6%	27.4%	10.1%	10.9%	11.3%	26.9%	14.2%	16.3%	1.2%

These results show that the distribution of traditional villages was differentiated not only at the overall regional scale, but also among cultural types (Fig. 6). Cultural types with lower noise ratios tended to form more compact spatial groupings, whereas cultural types with higher noise ratios included more dispersed village points. The cultural-type analysis provides an exploratory stratified perspective for identifying cultural-spatial heterogeneity. It does not define fixed cultural boundaries, nor does it directly demonstrate specific historical routes, commercial networks, religious systems, military systems, migration processes, or settlement typologies. Its analytical value lies in showing measurable differences in clustering intensity and spatial dispersion among cultural-type groups.

Fig. 6.

Kernel density estimation results of traditional villages of different types.

Uneven sample sizes among cultural types should also be considered when interpreting these results. Cultural-type DBSCAN results are therefore used to describe within-group clustering characteristics and noise ratios, rather than to make deterministic comparisons among cultural regions. This interpretation keeps the cultural-type analysis within the evidential scope of spatial clustering results.

Influencing factors of traditional village distribution

The Geodetector factor detection results show that the spatial distribution of traditional villages is associated with multiple dimensions of the human–land system. The explanatory power of the selected variables differs substantially, and no single variable fully explains the spatial differentiation of traditional villages. This result indicates that traditional village distribution is a multi-factor spatial phenomenon shaped by the combined background of natural environment, socioeconomic conditions, and historical-cultural accumulation.

The dependent variable Y is the number of traditional villages in each county-level administrative unit. The explanatory variables were grouped into three dimensions: natural environmental factors, socioeconomic factors, and historical-cultural factors (Table 5). Continuous variables were discretized using the Jenks natural breaks method before Geodetector analysis.

Table 5.

Detection results of spatial heterogeneity factors for traditional villages in the study area.

Primary indicator	Secondary indicator	Evaluation indicator	q-value	p-value
Natural Environmental Factors	Topographic environment	Average elevationX1	0.0196	0.00078
		Average slopeX2	0.0536	0.00000
	Climatic factors	Annual precipitationX3	0.0210	0.00042
		Average annual temperatureX4	0.0202	0.00060
	Ecological Environment	River network densityX5	0.0229	0.00018
Socioeconomic factors	Transportation	Transportation network densityX6	0.0238	0.00012
	Demographic Factors	PopulationX7	0.0117	0.02314
	Economic Factors	GDPX8	0.0055	0.26069
Historical and Humanistic Factors	Cultural heritage	Number of national intangible cultural heritage itemsX9	0.0320	0.00000
		Number of national key cultural relics protection unitsX10	0.1755	0.00000
	Population share	Minority population proportionX11	0.0171	0.00240

Natural environmental factors

The scatterplots between county-level village count and natural environmental variables show weak bivariate linear relationships overall (Fig. 7). Among the five natural variables, average slope has the highest linear fit, with an (R²) value of 0.115. This indicates that slope has a stronger association with county-level village count than the other natural variables in the bivariate scatterplot analysis. However, the dispersion of points remains substantial, showing that slope alone cannot explain the spatial differentiation of traditional villages.

Fig. 7.

Bivariate fitted relationships between county-level village count and natural environmental variables.

Average annual temperature shows a weak linear relationship with village count (R² = 0.049), while average elevation (R² = 0.033), river network density (R² = 0.022), and annual precipitation (R² = 0.005) show lower bivariate explanatory values. In particular, precipitation shows almost no linear relationship with county-level village count in the scatterplot. The distribution of counties with zero or few traditional villages across a wide range of elevation, precipitation, temperature, and river network density indicates that no single natural environmental factor provides a sufficient explanation for traditional village distribution.

These results are consistent with the factor-detection result that slope is the most informative natural environmental variable among the selected indicators, while river network density provides supplementary hydrological background information. However, the low (R²) values in the scatterplots indicate that the effects of natural environmental factors should be interpreted as background constraints and spatial conditions rather than as direct determinants. Therefore, the role of natural factors is better understood in combination with socioeconomic accessibility and historical-cultural accumulation, rather than through single-factor linear relationships.

Socioeconomic factors

The scatterplots between county-level village count and socioeconomic variables also show weak bivariate linear relationships (Fig. 8). Transportation network density has the highest (R²) value among the three socioeconomic variables (R² = 0.024), but the value remains low. This indicates that transportation accessibility provides only limited explanatory information in the bivariate linear analysis. Counties with low or moderate road network density include both counties with few traditional villages and counties with relatively higher village counts, suggesting that road density alone does not explain the spatial concentration of traditional villages.

Fig. 8.

Bivariate fitted relationships between county-level village count and socioeconomic variables.

Population size shows the weakest linear relationship with village count (R² = 0.004). The scatterplot indicates that counties with similar population levels can have substantially different numbers of traditional villages. This result suggests that county-level population size is not a sufficient indicator for explaining the spatial differentiation of traditional village distribution.

GDP also shows a weak bivariate relationship with village count (R² = 0.014). Most counties are concentrated at relatively low village-count values across a wide range of GDP levels, and counties with higher GDP do not consistently contain more traditional villages. This result is consistent with the Geodetector result that GDP has a low explanatory value in the selected indicator system. Therefore, contemporary county-level economic scale should not be interpreted as a strong predictor of traditional village distribution.

Overall, the socioeconomic scatterplots show that transportation, population, and GDP provide limited single-factor linear explanations for traditional village distribution. Among them, transportation network density provides relatively more information than population and GDP, but its explanatory value remains weak. These results support a cautious interpretation: socioeconomic variables are better understood as contextual conditions related to accessibility, development pressure, and governance capacity, rather than as direct determinants of traditional village formation or concentration.

Historical-cultural factors

Compared with the natural environmental and socioeconomic variables, the historical-cultural variables show stronger bivariate associations with county-level village count (Fig. 9). The fitted relationship between the number of national key cultural relic protection units and village count has the highest (R²) value among the three historical-cultural variables (R² = 0.296). The number of national intangible cultural heritage items also shows a relatively higher fitted association with village count (R² = 0.253). These results indicate that counties with richer recognized heritage resources are more likely to contain higher numbers of traditional villages, although the scatterplots also show substantial dispersion.

Fig. 9.

Bivariate fitted relationships between county-level village count and historical-cultural variables.

Minority population proportion has a much lower fitted value (R² = 0.032), indicating a weak bivariate association with county-level village count. Counties with different minority population proportions contain both low and relatively high village counts, suggesting that minority population proportion alone does not provide a sufficient explanation for traditional village distribution.

These scatterplot results are consistent with the Geodetector factor-detection results, in which national key cultural relic protection units provide the strongest spatial explanatory signal among the selected variables. The results suggest that historical-cultural accumulation is more closely associated with traditional village distribution than contemporary socioeconomic scale or any single natural environmental variable. However, the fitted relationships should be interpreted as statistical associations between county-level heritage-resource indicators and village counts. They do not demonstrate that cultural relic protection units or intangible cultural heritage items directly caused the formation of traditional villages.

Overall, the historical-cultural factors provide the most evident single-dimension signal in the bivariate analysis. This supports the interpretation that traditional villages are embedded in broader heritage-resource landscapes. The spatial concentration of traditional villages is therefore better understood in relation to long-term historical-cultural accumulation, while further historical documentation and village-level field investigation would be needed to explain specific local formation processes.

Integrated interpretation of factor detection results

The factor detection results show clear differences among the three groups of explanatory variables. Historical-cultural accumulation provides the strongest single-factor spatial signal in the selected indicator system. Natural environmental variables, particularly slope and river network density, describe terrain and hydrological background conditions. Socioeconomic variables provide information on accessibility and development context, with transportation network density showing more explanatory information than population size and GDP. GDP shows limited explanatory value at the county scale.

The generally low q-values of most variables are also relevant to the interpretation. They indicate that the spatial differentiation of traditional villages cannot be explained by one dominant factor. Instead, the results point to multiple weak but measurable spatial associations at the county level. This pattern is consistent with the human–land analytical framework, which interprets traditional village distribution as the spatial overlap of land-system conditions, human-system activities, and historical-cultural accumulation.

Interaction effects among influencing factors

The Geodetector interaction detector was used to examine whether pairs of explanatory variables increased or reduced their joint explanatory power for the county-level distribution of traditional villages. Interaction types were determined by comparing the interaction q-value, (q(X_i∩X_j)), with the single-factor q-values, (q(X_i)) and (q(X_j)). When (q(X_i∩X_j)) is greater than both single-factor q-values but lower than their sum, the interaction is identified as two-factor enhancement. When (q(X_i∩X_j)) is greater than the sum of the two single-factor q-values, it is identified as nonlinear enhancement. When (q(X_i∩X_j)) falls between the two single-factor q-values, it is interpreted as single-factor weakening. The interaction between national key cultural relic protection units and transportation network density has an interaction q-value of approximately 0.18 and is identified as two-factor enhancement. This result indicates that the joint explanatory power of historical-cultural accumulation and transportation accessibility is higher than the explanatory power of either variable alone. Counties where heritage-resource accumulation overlaps with stronger accessibility therefore show a clearer spatial correspondence with traditional village concentration.

Interactions between national key cultural relic protection units and natural environmental variables, including slope, river network density, and elevation, further indicate that terrain and hydrological conditions add contextual information to the historical-cultural dimension. When these interaction results are read together with the factor detection results, traditional village distribution is better explained by the spatial overlap of heritage-resource accumulation, terrain background, hydrological conditions, and accessibility than by any single variable.

The interaction results also clarify the role of socioeconomic variables. Although GDP has a low single-factor q-value, interactions involving socioeconomic variables may still provide contextual information on the current preservation environment, accessibility, and governance conditions of traditional villages. Contemporary socioeconomic conditions are therefore not the primary explanatory dimension in this study, but they remain relevant to understanding the present management context of traditional villages.

Overall, the interaction detection results provide evidence of joint spatial explanatory power. They show how different variables correspond spatially with traditional village distribution when considered together. They do not directly demonstrate historical causality, cultural mechanisms, settlement typologies, or complete formation processes. Their main analytical value is to show that traditional village concentration is more clearly associated with combined human–land conditions than with isolated variables (Fig. 10).

Fig. 10.

Interaction q-value matrix of influencing factors for traditional village distribution in the Yellow River Basin-related regions.

Summary of results

The results show that traditional villages in the nine provincial-level regions related to the Yellow River Basin are significantly clustered. The NNI result confirms the overall clustering tendency. DBSCAN identifies seven regional-scale clusters and 166 noise points, while KDE shows that relative high-density areas are mainly located in the middle and lower Yellow River-related regions. Cultural-type stratification further shows that this clustering pattern contains internal heterogeneity. Hehuang, Bashu, Guanzhong, Sanjin, and Central Plains cultural types show stronger within-group concentration, whereas Longxi and Grassland cultural types show more dispersed distributions.

The factor detection results indicate that historical-cultural accumulation, represented by national key cultural relic protection units, provides the strongest spatial explanatory signal among the selected variables. Natural environmental variables describe terrain and hydrological background conditions; socioeconomic variables provide accessibility and development-context information; and historical-cultural variables provide the most evident heritage-resource signal. GDP has limited explanatory value at the county scale. The interaction detection results further show that the combination of historical-cultural accumulation and transportation accessibility enhances the spatial explanation of traditional village concentration. Together, these findings provide quantitative evidence for interpreting traditional village distribution through a human–land framework, while maintaining a clear distinction between spatial association and causal explanation.

Discussion

This section interprets the empirical results in relation to the research questions proposed in the Introduction. The discussion focuses on three issues. First, it examines what the identified spatial pattern indicates about the regional concentration and dispersion of traditional villages. Second, it discusses how natural environment, transportation accessibility, historical-cultural resources, and regional development correspond to the spatial differentiation of traditional villages. Third, it considers how these spatial findings can inform conservation-oriented governance. Throughout the discussion, spatial statistical associations are distinguished from direct historical causal explanations.

Spatial pattern and human–land interpretation of traditional villages

The results show that traditional villages in the nine provincial-level regions related to the Yellow River Basin are significantly clustered. The Nearest Neighbor Index confirms the overall clustering tendency, while DBSCAN and Kernel Density Estimation identify regional aggregation areas and continuous density gradients. These findings indicate that traditional villages in the study area are not randomly scattered heritage points, but spatially aggregated heritage resources with clear regional concentration and peripheral dispersion.

This pattern is closely related to the Yellow River Basin as a long-term human–land system. The basin is not only a hydrological region, but also a historical settlement region shaped by terrain gradients, river dynamics, agricultural development, settlement continuity, and cultural accumulation. Research on coupled human and natural systems shows that settlement patterns are often shaped by long-term interactions among environmental processes, social organization, and spatial development². In the Yellow River Basin, river-course changes, flood disturbance, loess landforms, alluvial plains, and tributary systems have historically produced both opportunities and constraints for settlement distribution³⁰. The clustered distribution of traditional villages can therefore be understood as a spatial association formed through the overlap of environmental background, settlement continuity, accessibility conditions, and heritage-resource recognition.

The most evident concentration areas are located in the middle and lower Yellow River-related regions. These areas generally combine relatively favorable terrain conditions, long-term agricultural settlement, denser historical settlement foundations, and more visible historical-cultural accumulation. By contrast, western plateau, mountainous, and grassland-related areas show more dispersed village distributions. This pattern corresponds to more complex terrain, lower settlement density, and more fragmented accessibility conditions. The east-west and middle-lower reach differentiation should therefore be interpreted as spatial unevenness in land-system conditions and human settlement accumulation within the broader Yellow River Basin context, rather than as a simple administrative pattern.

The human–land perspective clarifies the value of these spatial results. The main finding is not only that traditional villages are clustered, but that clustering is more evident where terrain suitability, hydrological background, accessibility, and historical-cultural resources overlap. In this sense, the spatial pattern is not merely a computational output of clustering algorithms. It provides a spatial basis for understanding traditional villages as heritage resources embedded in a river-basin human–land system.

The interpretation remains limited by the available data and methods. NNI, DBSCAN, and KDE identify aggregation, density differentiation, and relative concentration areas. They do not reconstruct historical settlement evolution, agricultural civilization processes, migration routes, religious networks, or commercial systems. Such mechanisms require additional evidence from historical documents, field investigation, local gazetteers, oral history, and village-level architectural surveys.

Cultural-spatial heterogeneity, comparison with previous studies, and research contribution

Cultural-type stratification provides an intermediate analytical layer between overall spatial clustering and county-level factor analysis. The results show that the overall clustering pattern contains clear internal cultural-spatial heterogeneity. Hehuang, Bashu, Guanzhong, Sanjin, and Central Plains cultural types show relatively stronger within-group concentration, whereas Longxi and Grassland cultural types show more dispersed distributions. These differences indicate that traditional villages should not be treated as a homogeneous heritage category at the regional scale.

The value of cultural-type stratification lies in its ability to describe measurable differences in clustering intensity and noise ratios among cultural-type groups. Cultural types with lower noise ratios form more compact heritage-resource zones and are more suitable for type-based survey and cluster-oriented conservation analysis. Cultural types with higher noise ratios require more attention to dispersed village points, site-specific documentation, and small-scale conservation support. The analysis therefore helps identify cultural-spatial heterogeneity within the overall regional distribution.

This stratification is exploratory and should not be interpreted as a deterministic cultural boundary. The results do not prove cultural mechanisms, migration routes, commercial networks, military systems, religious service areas, or settlement typologies. They show that different cultural-type groups have different degrees of aggregation and dispersion under the selected analytical framework. This interpretation avoids cultural determinism and keeps the analysis within the evidential scope of spatial clustering.

Existing studies have examined the spatial distribution and influencing factors of traditional villages at national, regional, and Yellow River Basin scales^18,35–37. National-scale studies have identified uneven traditional village distribution and the roles of natural, socioeconomic, and cultural factors⁴. Studies of the Yellow River Basin have examined spatiotemporal distribution, clustering patterns, and influencing factors of traditional villages^35–37. These studies provide important methodological and empirical foundations. The present study differs from previous work in three main respects.

First, it interprets traditional village distribution within a human–land framework, linking spatial patterns with terrain, hydrology, accessibility, and historical-cultural accumulation. Second, it introduces cultural-type stratification into the spatial analytical process, moving beyond overall clustering to examine internal cultural-spatial heterogeneity. Third, it separates spatial association from causal explanation. The results identify statistically meaningful spatial patterns and explanatory correspondences, but they do not extend clustering results into unsupported claims about settlement types, historical routes, tourism networks, or cultural mechanisms.

The contribution of this study is not the development of a new spatial algorithm, but the organization of existing spatial methods within a human–land analytical framework for traditional village research in the Yellow River Basin-related regions. Compared with studies that mainly report spatial distribution patterns or individual influencing factors, this study makes four advances. First, it links traditional village clustering with the river-basin human–land context, emphasizing the combined roles of terrain, hydrology, accessibility, and historical-cultural accumulation. Second, it uses cultural-type stratification as an exploratory layer to identify internal cultural-spatial heterogeneity. Third, it integrates NNI, DBSCAN, KDE, and Geodetector into a multi-level workflow, moving from overall clustering verification to density differentiation, cultural-type comparison, factor detection, and interaction interpretation. Fourth, it clarifies the evidential boundary of spatial statistical results by distinguishing spatial association from historical causality. This distinction is important because spatial clustering alone cannot prove settlement typologies, migration routes, commercial networks, religious systems, or tourism mechanisms.

The new findings can be summarized at three levels. At the spatial level, traditional villages in the nine Yellow River Basin-related provincial-level regions are significantly clustered and are most clearly concentrated in the middle and lower Yellow River-related regions. At the cultural-spatial level, different cultural-type groups show different degrees of concentration and dispersion. At the influencing-factor level, historical-cultural accumulation provides the strongest single-factor spatial signal, while terrain, hydrology, transportation accessibility, and intangible cultural heritage provide supplementary explanatory information. Together, these findings shift the interpretation of traditional village distribution from a single clustering pattern to a multi-level human–land spatial association.

Influencing factors and human–land coupling mechanism

The factor detection results indicate that the spatial differentiation of traditional villages is associated with natural environmental conditions, socioeconomic context, and historical-cultural accumulation. Most single-factor q-values are relatively low, suggesting that no single variable provides a sufficient explanation for the observed distribution. This pattern supports a multi-factor interpretation rather than a deterministic explanation based on one dominant factor.

Among the selected variables, the number of national key cultural relic protection units has the highest q-value (q = 0.1755). This indicates that historical-cultural accumulation provides the strongest spatial explanatory signal in the current indicator system. Counties with more recognized heritage resources tend to contain more traditional villages. This relationship is consistent with the role of traditional villages as carriers of cultural memory, construction traditions, regional social organization, and historical landscape continuity. The spatial correspondence between traditional villages and national key cultural relic protection units therefore suggests that village concentration is embedded in broader heritage-resource landscapes.

Natural environmental variables provide a second layer of interpretation. Slope has the highest q-value among the natural variables (q = 0.0536), while river network density provides additional hydrological information (q = 0.0229). Terrain conditions affect construction suitability, agricultural land availability, settlement accessibility, and landscape stability. River networks provide water accessibility and agricultural support, but they may also involve flood exposure, riverbank change, and environmental risk. Studies on traditional villages in river-basin contexts have also shown that water networks can structure village distribution and coordinated conservation strategies³⁰. In this study, terrain and hydrology are therefore interpreted as background constraints and enabling conditions rather than as independent deterministic causes.

Socioeconomic variables show more limited explanatory power at the county level. Transportation network density has a q-value of 0.0238, indicating that accessibility provides useful information for understanding the present distribution, visibility, and governance context of traditional villages. Better accessibility may support heritage identification, monitoring, and management, but it may also increase pressure from road expansion, construction, commercialization, and landscape disturbance. GDP has the lowest q-value (q = 0.0055) and is not statistically significant in the current model. Contemporary county-level economic scale alone therefore does not explain the distribution of traditional villages.

The interaction detection results further show that combined factor conditions provide clearer spatial explanatory power than many single factors. The interaction between national key cultural relic protection units and transportation network density has an interaction q-value of approximately 0.18 and is identified as two-factor enhancement. This indicates that traditional village concentration is more clearly associated with areas where historical-cultural accumulation overlaps with transportation accessibility. Interactions between historical-cultural variables and slope, river network density, elevation, or intangible cultural heritage further indicate that village distribution is better understood through the spatial overlap of heritage accumulation, environmental background, and accessibility.

This coupling relationship can be understood as a layered human–land process. Terrain and hydrology form the land-system background; transportation network density represents accessibility and present-day spatial connectivity; heritage-resource accumulation reflects long-term cultural and historical continuity; and regional development affects conservation pressure, management capacity, and public-service conditions. These dimensions do not operate separately. Their spatial overlap helps explain why traditional villages are concentrated in some areas and more dispersed in others.

The interaction results should be interpreted as evidence of joint spatial explanatory power rather than as direct proof of causal coupling mechanisms. Geodetector identifies spatial stratified heterogeneity and interaction enhancement⁴⁴, but it does not estimate independent marginal effects in the same way as regression models and cannot fully remove multicollinearity among explanatory variables. The results therefore support a human–land spatial association interpretation. Further mechanism-oriented research would require historical data, village-level evidence, and additional statistical models.

Conservation and governance implications

The spatial patterns and influencing-factor results provide a basis for differentiated conservation and protection-oriented governance of traditional villages in the Yellow River Basin-related regions. These implications should not be read as tourism-development conclusions. This study does not examine tourist flows, visitor behavior, tourism revenue, tourism carrying capacity, or tourism performance. The results are therefore more suitable for conservation prioritization, heritage-resource monitoring, risk-sensitive management, and governance zoning.

Traditional village conservation should not follow a uniform model. Recent studies have also emphasized that traditional village conservation is shifting from single-site protection toward coordinated protection, adaptive use, and regional heritage governance^64–67.Different areas require different conservation pathways according to clustering intensity, cultural-type characteristics, heritage-resource accumulation, accessibility conditions, and environmental sensitivity. Table 6 summarizes the conservation and governance implications derived from the empirical findings.

Table 6.

Differentiated conservation and governance strategies based on regional types and influencing factors.

Regional type	Empirical basis and key influencing factors	Main conservation issue	Targeted conservation and governance strategy
High-density concentration areas in the middle and lower Yellow River- related regions	Identified by NNI, DBSCAN, and KDE; overlap with historical-cultural accumulation and relatively better accessibility	Isolated site-based protection may fragment the integrity of traditional village landscapes	Establish cluster-based conservation zones; protect village groups together with cultural relics, historical roads, agricultural landscapes, river-related settings, and visual corridors; conduct regional heritage monitoring
Secondary clustering areas	Moderate spatial aggregation; weaker than the main concentration areas but still forming identifiable local clusters	Village resources may be fragmented across county or administrative boundaries	Build county-level or cross-county conservation networks; connect core villages and surrounding traditional villages through shared documentation, monitoring standards, infrastructure control, and coordinated landscape management
Cultural-type concentration areas, such as Hehuang, Bashu, Guanzhong, Sanjin, and Central Plains types	Lower noise ratios indicate stronger within- group concentration	Uniform policies may ignore differences in settlement form, architectural tradition, and cultural expression	Carry out type-based heritage surveys; document settlement layouts, architectural forms, local construction techniques, and intangible cultural practices; develop conservation guidelines adapted to different cultural types
Moderate transition areas, such as Qilu and Longyou cultural types	Intermediate noise ratios and mixed concentration- dispersion patterns	These areas may not fit either high-density cluster protection or dispersed-site protection	Adopt flexible network-based management; identify local core villages as conservation anchors; strengthen connections between core villages and surrounding dispersed villages
Dispersed areas, especially Longxi and Grassland cultural types	Higher noise ratios indicate weaker internal clustering and more dispersed spatial patterns	Villages may be overlooked because they do not belong to large clusters	Adopt site-specific conservation; conduct field verification and digital documentation; support small-scale repair and maintenance; protect local environmental adaptation, architectural features, and cultural continuity
Areas with high values of national key cultural relic protection units	X10 has the highest q-value and provides the strongest spatial signal in the selected indicator system	Traditional villages and other heritage resources may be managed separately	Promote integrated heritage-resource management; combine traditional village conservation with cultural relic protection, intangible heritage documentation, and historical landscape interpretation; establish county-level heritage inventories and monitoring platforms
Areas where heritage- resource accumulation overlaps with transportation accessibility	X10 ∩ X6 shows two-factor enhancement; accessibility helps identify village concentration areas	Better accessibility may bring road expansion, construction pressure, visual disturbance, and commercialization risks	Use accessibility for conservation management rather than uncontrolled development; locate parking and service facilities outside core protection areas; control new construction along access roads; assess carrying capacity and landscape impact before tourism use
Terrain- and water-sensitive areas	Slope and river network density provide environmental background information	Villages may face slope instability, flood exposure, riverbank erosion, drainage pressure, or construction constraints	Integrate conservation with disaster-risk monitoring, land-use control, drainage improvement, river-environment management, and ecological restoration; avoid large-scale construction in sensitive terrain or flood-prone areas
Areas with weak contemporary socioeconomic explanatory power but important heritage value	GDP and population have low q-values; current economic scale does not fully explain traditional village distribution	Heritage value may be underestimated in economically weaker or less populated areas	Use heritage value and cultural continuity rather than GDP level as the basis for conservation priority; provide technical guidance, maintenance subsidies, public-service improvement, and community participation mechanisms

The main message of Table 6 is that conservation strategies should be linked to empirical spatial evidence. High-density areas require cluster-based conservation and regional heritage monitoring. Secondary clustering areas require network-based coordination across counties or administrative units. Cultural-type concentration areas require type-based surveys and conservation guidelines, whereas dispersed cultural-type areas require site-specific documentation and small-scale conservation support. Areas with high values of national key cultural relic protection units require integrated heritage-resource management. Areas where heritage accumulation overlaps with transportation accessibility require control of development pressure rather than simple promotion of tourism use. Terrain- and water-sensitive areas require risk-sensitive conservation. Economically weaker areas with important heritage value require public support based on heritage significance rather than current economic scale.

This framework translates spatial findings into conservation and governance pathways without extending the results beyond the empirical evidence. The results can support conservation priority identification and differentiated management, but they do not directly justify conclusions about tourism performance, cultural-tourism integration, or tourism-driven rural revitalization.

Limitations and future research

Several limitations should be acknowledged. First, this study uses cross-sectional spatial data and county-level indicators. It can identify spatial patterns and associations, but it cannot reconstruct the historical processes through which river-course changes, flooding, migration, land-use transformation, agricultural development, and local governance shaped individual villages. Future research should integrate multi-period spatial data, historical hydrological records, archival materials, village-level socioeconomic data, and field-based heritage assessment to examine the temporal evolution of traditional village distribution.

Second, the human–land analytical framework provides a regional-scale explanatory structure, but it cannot fully explain the micro-scale formation mechanisms of individual villages. The formation and persistence of traditional villages may involve household livelihood strategies, lineage organization, land-use change, architectural adaptation, local governance, and community participation. Future studies should combine regional spatial analysis with village-level case studies, architectural surveys, oral history, local gazetteers, and community-based investigation.

Third, DBSCAN and KDE results are influenced by parameter settings. Different values of Eps, MinPts, search radius, and cell size may produce different cluster boundaries, noise-point proportions, or density surfaces. Parameter sensitivity tests improve robustness, but spatial clustering results still require cautious interpretation, especially when they are used to discuss regional aggregation or cultural-type differences.

Fourth, the cultural-type analysis is exploratory. It identifies measurable differences in clustering intensity, noise ratios, and spatial concentration among cultural-type groups, but it cannot directly prove cultural mechanisms, migration routes, commercial networks, religious service areas, military systems, or settlement typologies. Future research should verify cultural-type interpretations through historical documents, field investigation, local gazetteers, oral history, and village-level architectural evidence.

Fifth, the explanatory power of the statistical results is limited. Most single-factor q-values in the Geodetector analysis are relatively low, indicating that the selected county-level variables capture only part of the spatial differentiation of traditional villages. Geodetector is useful for identifying spatial stratified heterogeneity, but it does not estimate independent marginal effects in the same way as regression-based models and cannot fully address possible multicollinearity among environmental, socioeconomic, and historical-cultural variables⁴⁴. Future research could combine Geodetector with variance inflation factor testing, generalized additive models, geographically weighted regression, spatial regression, or multi-scale models to examine nonlinear relationships, multicollinearity, local spatial variation, and scale-dependent effects.

Finally, this study does not directly analyze tourism development. Tourism-related interpretations should therefore remain cautious. The results can provide spatial references for conservation prioritization and governance zoning, but they do not directly support conclusions about tourist flows, visitor behavior, tourism revenue, tourism performance, or tourism-driven rural revitalization. If tourism development becomes a central research question, future research should incorporate tourism-flow data, visitor behavior, carrying capacity, community participation, and heritage-impact assessment.

Overall, future studies of traditional village distribution in the Yellow River Basin should integrate multi-period data, historical hydrological information, village-level social data, field investigation, and advanced statistical models. Such work would help move the analysis from spatial association toward a fuller explanation of the long-term human–land processes underlying the formation, preservation, and transformation of traditional villages.

Conclusions

This study examined the spatial distribution and associated influencing factors of 2,022 traditional villages in nine provincial-level regions related to the Yellow River Basin. By integrating the Nearest Neighbor Index, DBSCAN, Kernel Density Estimation, cultural-type stratification, and Geodetector, the study identified the overall clustering pattern, regional density differentiation, cultural-spatial heterogeneity, and human–land factor associations of traditional villages within a large river-basin context.

The results show that traditional villages in the study area present a statistically significant clustered distribution. DBSCAN identified seven regional-scale clusters and 166 noise points, while KDE indicated that relatively high-density areas are mainly located in the middle and lower Yellow River-related regions. Cultural-type stratification further revealed internal heterogeneity within the overall clustering pattern. Some cultural types showed stronger within-group concentration, whereas Longxi and Grassland types showed more dispersed spatial patterns. These findings indicate that traditional villages in the Yellow River Basin-related regions are not evenly distributed heritage points, but spatially differentiated heritage resources shaped by regional concentration, peripheral dispersion, and cultural-spatial variation.

The Geodetector results show that traditional village distribution is associated with multiple human–land dimensions. Among the selected variables, national key cultural relic protection units provided the strongest spatial explanatory signal, followed by slope, national intangible cultural heritage items, transportation network density, and river network density. Interaction detection further indicated that combined conditions, especially the overlap between historical-cultural accumulation and accessibility, provided clearer spatial explanatory information than many single variables. These results suggest that the spatial distribution of traditional villages is better interpreted through the overlap of terrain background, hydrological conditions, accessibility, and historical-cultural accumulation, rather than through any single environmental or socioeconomic factor.

The main contribution of this study lies in providing a multi-level spatial interpretation of traditional village distribution in the Yellow River Basin-related regions from a human–land relationship perspective. The study connects overall clustering, density differentiation, cultural-type heterogeneity, factor detection, and interaction analysis within one analytical framework. It also clarifies the evidential boundary of spatial statistical analysis: the results reveal spatial associations and explanatory correspondences, but they should not be interpreted as direct evidence of historical causality, fixed settlement typologies, cultural mechanisms, tourism development, or rural revitalization effects.

These findings provide spatial references for conservation prioritization, heritage-resource monitoring, and differentiated governance of traditional villages. High-density and culturally concentrated areas may require regional-scale monitoring and type-sensitive conservation, whereas dispersed villages may require site-specific documentation and localized protection support. Future research should further integrate multi-period data, historical documents, hydrological records, village-level fieldwork, and more detailed socioeconomic information to explain local formation and preservation processes more fully.

Author contributions

Conceptualization, S.P. ; methodology, S.P.; software, S.P.; validation, S.P.; formal analysis, S.P.and Z.L.; investigation, S.P.; resources, Z.L.; data curation, S.P.and Z.L.; writing—original draft preparation, S.P.; writing—review and editing, Z.L.; visualization, S.P.; supervision, Z.L.; project administration, Z.L.; funding acquisition, Z.L. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Data availability

This study used multi-source data, including traditional village point data, administrative boundary data, Yellow River Basin boundary and river-network data, DEM data, soil-type data, Transportation network data, climate data, socioeconomic data, population data, and heritage-resource data. All datasets were accessed in April 2025. Table 1 summarizes the data categories, data contents, sources, processing methods, analytical uses, and access dates. The traditional village dataset was constructed from the official list of Chinese traditional villages released by the Ministry of Housing and Urban-Rural Development of the People’s Republic of China and related national departments. The first to sixth batches of Chinese traditional villages were included. Villages located within the nine provincial-level study regions were retained, resulting in a final dataset of 2,022 traditional village points. Village coordinates were obtained and checked using Amap-derived geocoding information. Amap-derived coordinates in mainland China are commonly provided in the GCJ-02 coordinate system; all village coordinates were converted to WGS84 before spatial analysis. Administrative boundary data were obtained from Guihua Cloud and used to define the nine provincial-level study regions, support spatial overlay, verify village locations, and construct county-level spatial analysis units. Yellow River Basin boundary and river-network data were obtained from the National Basic Geographic Information Database and used to map the major river system, represent the hydrological background, and calculate river network density. DEM data were obtained from the European Space Agency Copernicus DEM GLO-30 China dataset and used to represent elevation and derive terrain-related variables, including elevation and slope. Soil-type data were obtained from the Resource and Environment Science and Data Center, Chinese Academy of Sciences, and were used to describe the environmental background of the study area. Transportation network data were obtained from OpenStreetMap and used to calculate transportation network density. Climate data, including precipitation and temperature, were obtained from the National Tibetan Plateau Data Center and used to calculate county-level climate indicators. Socioeconomic and population data were obtained from the China County Statistical Yearbook and the Seventh National Population Census. Heritage-resource data, including national key cultural relic protection units and national intangible cultural heritage items, were obtained from official national heritage lists.

Competing interests

The authors declare no competing interests.

Declaration of AI-assisted language editing

During the preparation of this manuscript, the authors used ChatGPT for language editing, grammar checking, and improvement of clarity and readability. The tool was not used to generate original research data, conduct spatial or statistical analyses, create references, or draw scientific conclusions. All AI-assisted edits were critically reviewed, revised, and approved by the authors, who take full responsibility for the final content of the manuscript.