From rural hollowing to smart shrinkage: zonal governance considering spatial non-stationary effects.

Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Sci Rep . 2026 Mar 3;16:11913. doi: 10.1038/s41598-026-41875-8 Search in PMC Search in PubMed View in NLM Catalog Add to search From rural hollowing to smart shrinkage: zonal governance considering spatial non-stationary effects Chi Chen Chi Chen 1 Institute of Land and Urban-Rural Planning, Huaiyin Normal University, Huai’an, 223300 China 2 Urban Design Analysis Lab, Graduate School of Urban Studies, Hanyang University, Seoul, 04763 Korea Find articles by Chi Chen 1, 2, # , Chengxiang Wang Chengxiang Wang 1 Institute of Land and Urban-Rural Planning, Huaiyin Normal University, Huai’an, 223300 China 2 Urban Design Analysis Lab, Graduate School of Urban Studies, Hanyang University, Seoul, 04763 Korea Find articles by Chengxiang Wang 1, 2, ✉, # , Lei Cao Lei Cao 1 Institute of Land and Urban-Rural Planning, Huaiyin Normal University, Huai’an, 223300 China Find articles by Lei Cao 1 , Chang Gyu Choi Chang Gyu Choi 2 Urban Design Analysis Lab, Graduate School of Urban Studies, Hanyang University, Seoul, 04763 Korea Find articles by Chang Gyu Choi 2, ✉ Author information Article notes Copyright and License information 1 Institute of Land and Urban-Rural Planning, Huaiyin Normal University, Huai’an, 223300 China 2 Urban Design Analysis Lab, Graduate School of Urban Studies, Hanyang University, Seoul, 04763 Korea ✉ Corresponding author. # Contributed equally. Received 2025 May 25; Accepted 2026 Feb 23; Collection date 2026. © The Author(s) 2026 Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/ . PMC Copyright notice PMCID: PMC13065826  PMID: 41776232 Abstract The phenomenon of rural depopulation poses a critical challenge to sustainable rural development and requires evaluation through the lens of smart shrinkage. Existing research has inadequately explored the formation of rural hollowing from the village environmental perspective, generally neglecting the spatial non-stationary effects. This study addresses this gap by constructing a theoretical model based on residential field and power theory using a full-sample survey of rural households. It innovatively applies the Geographically Weighted XGBoost (GWXGBoost) model combined with the SHAP method for analysis. The results demonstrate that GWXGBoost outperforms conventional methods in regression performance. Rural hollowing in the study area exhibits a ring-like pattern, with moderately and severely hollow villages accounting for 45% and 43% of all villages, respectively. Living and location fields significantly influence hollowing. In addition, the spatial non-stationarity of influencing factors also displays a ring-like differentiation. Customized management strategies are proposed for each area, including enhancing the economic momentum and promoting balanced urban-rural integration of suburban villages; improving accessibility for remote villages with high hollowing, building on existing development trajectories and supporting villagers’ self-directed initiatives for remote villages with low hollowing villages; for villages in the middle zone, strategies include encouraging the upgrade and reutilization of dilapidated housing, conducting land consolidation, and promoting agricultural modernization. Keywords: Rural hollowing, Village environment, Spatial field and power theory, Geographically weighted XGBoost model, Smart shrinkage, Zonal governance Subject terms: Socioeconomic scenarios, Sustainability Introduction World Bank data show that the global rural population’s share of the total population reduced from 66% in 1960 to 43% in 2023. This has caused the deterioration of living environment 1 , abandonment of cultivated land 2 , shortage of public infrastructure 3 , and pension issues for the elderly population 4 in the rural area. According to the China Statistical Yearbook 2023, the rural population in China at the end of 2022 was 491 million. This accounts for 34.78% of the total population and corresponds to a decrease by 39.26% compared with 2000. Although China’s urbanization process is advancing rapidly, the problem of uncoordinated development between urban and rural is becoming increasingly prominent and exacerbated. Rural areas are gradually declining and hollowing out 5 . Therefore, it is essential to investigate the drivers of rural hollowing to inform rural revitalization, particularly in the Chinese context. China’s rural hollowing is a special product of the urban–rural dualistic system. In this system, the non-agricultural out-migration of the rural population is decoupled from the contraction of rural settlements. This results in the undesirable evolutionary process of “external expansion and internal hollowing” 6 – 8 . Herein, the expansion of rural settlements is accompanied by the coexistence of vacant housing, which is different from the rural decline phenomenon in other countries 6 , 9 . The current research on rural hollowing focuses on its spatial characteristics 10 , 11 , spatiotemporal evolution 8 , measurements 12 , impact mechanisms 6 , and governance strategies 13 . Most studies focused on the macro level, such as the municipal 14 , county 15 , and township scales 11 , 16 . These used sampling statistics or estimated data that cannot accurately describe the real characteristics of rural hollowing. Although a few studies focused on the administrative village scale 12 , 17 , 18 , these tended to analyze only specific villages and did not provide general guidance. Scholars generally consider that variables such as geographical locations, natural conditions, socioeconomic characteristics of rural households, affect rural hollowing 6 , 18 , 19 , and usually use regression models to analyze their impact mechanisms. However, a few studies have shown that the same factor can exert varying effects on rural hollowing due to differences in its spatial distribution. For example, rural shrinkage is more significant when a village is closer to a large city 20 . Moreover, the higher the concentration of homesteads, the farther away these are from the village center 21 . Also, most of these studies assumed a linear relationship between the independent variables and the rural hollowing. However, the relationship between hollowing and its influencing factors is more complex. It involves the interaction of factors such as natural resources and the social economy. Some influencing factors may also exhibit effects of varying magnitude across different spatial scales 5 . Therefore, scholars have attempted to use machine learning algorithms to analyze the nonlinear paths between influencing factors 22 , 23 . These have been widely used in fields such as transportation 24 , 25 , human settlements 26 , 27 , and ecological environments 28 . Some studies also have attempted to use machine learning for analysis in the field of rural hollowing. For example, Guo et al. used a random forest model to identify the population hollowing in central China 29 , and Fu et al. used XGBoost to assess the importance of the rural hollowing characteristics in the Yellow River basin 30 . However, such studies remain scarce, and the evaluation of nonlinear relationships and spatial heterogeneity is insufficient. Only a few studies have attempted to evaluate this. For example, Shen et al. combined spatial weights and the XGBoost algorithm to study the impact of public service facilities on the rent levels 31 , Ye et al. combined GWR (Geographically Weighted Regression) and XGBoost to estimate soil pollution 32 , Li used a geographically weighted ensemble machine learning method to estimate wind speed 33 , and Hagenauer and Helbich combined geographically weighted and neural network methods 34 . Also, GWRF (Geographically Weighted Random Forest) is a more mature and widely used research method that combines GWR and random forest 23 , 27 , 35 . These approaches share a common feature, that they integrate GWR with another method to account for spatial heterogeneity and nonlinear relationships. In empirical applications, scholars have widely reported that incorporating geographically weighted effects can substantially improve model performance. However, these studies did not accurately quantify the spatial influence of nonlinearities or provide an in-depth explanation of the influence of factors in geographical space. Based on the research of influencing factors, scholars have proposed policies for the redevelopment of hollow villages. However, these are mostly top-down governance responses such as recommendations for optimizing the layout of residential areas and extending the reform of the homestead system 19 , 36 . However, the governance model should not be unified because of the differences between regions and individual households. A few studies have shown that a bottom-up governance approach that considers the actual situation in each village is more effective in addressing rural decline 37 . Therefore, the magnitude and direction of the influencing factors need to be quantified spatially to more accurately identify the main influencing factors in each region and thus, propose targeted policy recommendations. It is also noteworthy that the current research generally argues that hollow villages can lead to imbalances in the humans-land relationship 38 and the deterioration of urban-rural landscapes 39 . Accordingly, scholars intend to mitigating rural hollowing by encouraging population return to rural areas and facilitating two-way flows of resources between urban and rural areas 37 , 40 , 41 . However, with the continuous development of cities, rural decline is also a natural result of the urbanization process to a certain extent 42 . In recent years, a few scholars introduced the concept of “Smart Shrinkage” to attempt to optimize spatial resources through scientific planning and policy measures while conforming to the shrinkage of the rural population 20 , 43 . The concept of Smart Shrinkage originated in Germany. Subsequently, American scholars defined it as “planning for less—fewer individuals, fewer buildings, and fewer land uses 44 .” This concept requires planners and local governments to modify their mindset 45 and resolve the human-land imbalance during the shrinkage process 46 . Smart Shrinkage acknowledges the reality of population decline and advocates optimizing development by scaling down population targets and overall volume of construction. However, rural shrinkage does not imply a decline. It can develop in parallel with rural revitalization with the aim of achieving more efficient resource integration and common development 43 . It emphasizes the achievement of sustainable development in rural areas while adapting to the continuous reduction in the rural population and improving the quality of life in rural areas 43 , 47 . To summarize, this study attempted to establish a machine learning method that incorporates spatial heterogeneity analysis. While improving regression performance, it also examined the non-stationary impact of independent variables on rural hollowing and mapped these effects explicitly in geographical space. This enabled a more fine-grained understanding of region-specific influence pathways, supported more targeted policy recommendations for rural smart shrinkage, and offered a methodological contribution to hollow village governance research. Therefore, this study considered administrative villages as the research unit, combined the GWR and XGBoost methods, and used the SHAP interpretation model to evaluate spatial non-stationary effects of explanatory variables on rural hollowing. A clustering algorithm was used to divide the research area into subzones to provide specific policy support for the smart shrinkage of each zone. The remainder of this paper is structured as follows: “ Theoretical frame ” explains the theoretical framework and selects the independent variables. “ Data and methodology ” describes the research area, data, and methods. “ Results ” presents and discusses the analysis results. “ Discussion ” and “ Conclusion ” present the discussion and conclusion, respectively. Theoretical frame Rural hollowing and theoretical explanation of the spatial field and power theory Rural hollowing is an important manifestation of “rural disease” 48 . It is characterized mainly by the hollowing of population, land, housing, and industry 49 – 52 , with population hollowing often viewed as the primary driver 53 . In many contexts, hollowing out is triggered by the out-migration of younger residents to cities in search of better employment opportunities and living conditions. This process can result in widespread housing vacancy, the expansion of homestead land through “building new without demolishing old” 10 , the loss of young agricultural labor and associated stagnation in agricultural development 17 , and deficiencies in local infrastructure and public services 4 . Rural depopulation can be explained using the theory of spatial field and power 54 . Originating in physics 55 , the theory of spatial field and power focuses on physical elements in space and their influence on their surroundings as interactive forces 56 . It has been widely used in various disciplines such as sociology 57 and ecology 58 . In this theory, “field” refers to the spatial form of the main activities in a location, and “power” refers to the interactive force in space 55 . When residents select places to live, they tend to move to those with more “fields” and higher “powers.” Areas with more “powers” have better resources and environments, which makes these more attractive to villagers and causes the emergence of depopulation and hollow villages 54 . Classic “push–pull” theory in human geography and migration studies argues that population movement results from the joint effects of push forces in the place of origin and pull forces in the destination 59 . This logic is broadly consistent with field-and-power theory in that both emphasize interregional differentials: spatial units differ in resources and opportunities, and people tend to relocate toward places with greater relative advantages. In this study, however, we do not adopt the push–pull framework to interpret rural hollowing. The main reason is data availability. Our dataset contains village-level information only, while comparable city-side indicators are unavailable. As a result, we cannot operationalize both “push” and “pull” components in a symmetric manner. Instead, we employ “field and power” theory to focus on the endogenous attractiveness of rural villages. Based on the connotations of the residential field and power theory, a structural model of the spatial interaction between rural residential fields and powers can be constructed (Fig. 1 ). In general, villages with stronger residential field-and-power conditions tend to have relatively better living environments, more concentrated settlement patterns, and higher overall residential quality. Such villages therefore exert greater agglomeration forces and attractiveness for local residents 54 . Owing to the complexity of the rural environment and differences in the individual attributes of rural households, the rural residential field and power should be a multi-dimensional spatial system unit composed of multiple resource variables. Describing the rural “residential field and power” from a multi-dimensional space is more conducive to the accuracy of spatial analysis 55 . Fig. 1. Open in a new tab Structural model of the spatial interaction between rural residential fields and powers. Independent variable selection Previous research has shown that the location and the natural, economic, and social environments are the main factors affecting rural hollowing 6 , 18 , 40 , 54 . This study used administrative villages as the research unit, combined the theory of residential field and power, and selected 17 independent variables from six fields to construct an index system for rural hollowing (Table 1 ). Table 1. Description of independent variables. Categories Variables Description of variables Unit Max Min Mean Location field City_Access Driving travel time for the village committee to reach the center of Huai’an City min 113.77 8.46 49.64 County_Access Driving travel time for the village committee to reach the center of the county where each village is located min 53.51 1.53 23.85 Town_Access Driving travel time for the village committee to reach the center of the town where each village is located min 63.87 0.00 17.12 Ecological field Water_Density Area of water within the village/total area of the village % 99 0 10 Production field Farmland_pc Total cultivated land area/registered population within the village m 2 /person 74,480 0 1373 Agri_Modernization Mechanized planting area within the village/total agricultural land area % 99 0 6 Seclnd_FirmCount Number of secondary industry enterprises in the village – 256 0 9.03 Economic field Household_Income Total income of villagers/total registered residents in the village 10,000 yuan 43.93 1.05 5.21 Town_Fiscal_Revenue GDP of the town where the village is located 10,000 yuan 5437 1193 12,751 Poverty_Rate (Registered impoverished households + low-income households + rural individuals in extreme poverty receiving scattered support)/total registered population within village % 55 0 14 Social field Household_Size Total registered population in the village/total number of registered households in the village Person 6.03 2.10 4.27 Settlement_Dispersion (Number of households living scattered in the village area + number of households living alone)/total number of registered households in the village area % 129 0 6 New_Settlement_Count Number of new rural residential settlements built after 2006 within the village area – 31 0 0.83 Living field Commercial_POI_Density Number of commercial facilities in the village/total village area % 34 0 1 Kindergarten_Access Electric bicycle travel time from the village committee to the nearest kindergarten min 44.59 0.00 9.10 Offsite_Homeownership_Rate Number of households with property outside the village within the village area/total number of registered households in the village area % 99 0.00 46 MultiStorey_Dwelling_Rate Number of households with buildings in the village/total number of registered households % 97 0 42 Open in a new tab Given data availability constraints, the location field is operationalized by the accessibility of a village to external centers, such as the city, county seat, and township center. Greater external connectivity implies stronger field-and-power conditions. The ecological field is represented by water-network density as a proxy for the attractiveness of natural resources; villages with better ecological endowments are assumed to exhibit stronger field-and-power conditions. The production field captures productive capacity by measuring agricultural development via per capita cultivated-land area and the level of agricultural modernization, and by approximating industrial development using the number of secondary-industry enterprises within the village; better production conditions correspond to stronger field-and-power conditions. The economic field reflects local economic strength using average household income and township-level fiscal revenue, while the poverty rate is used to indicate relative affluence; villages with higher incomes and lower poverty rates are expected to have stronger field-and-power conditions. The social field characterizes household structure and settlement form using average household size, settlement dispersion, and the number of newly established rural residential settlements; in general, larger populations and more concentrated settlement patterns imply stronger field-and-power conditions. Finally, the living field evaluates the convenience of public services and housing conditions through commercial POI density, preschool education accessibility, the share of households living in multi-storey dwellings, and the rate of off-site homeownership; more convenient living conditions are associated with stronger field-and-power conditions. Data and methodology Study area This study selected Huai’an City in Jiangsu Province as a case study, as it is a typical representative of China’s plain agricultural regions (Fig. 2 ). Jiangsu Province is located on the Yangtze River Delta and is one of the most economically developed regions in China. Huai’an City is located on the north of Jiangsu Province. It has long been influenced by its southern neighbors, the provincial capital of Nanjing and the megacity of Shanghai. As a result, the rural population continues to decline, and urbanization is gradual. Huai’an City covers an area of 10,030 km 2 and comprises 7 administrative districts and 1511 administrative villages. In 2019, the registered population was 5.6047 million, the permanent population was 4.9326 million, and the urbanization rate of the permanent population was 63.50%. Fig. 2. Open in a new tab Study area. Data resource The dataset used in this study is based on a full-sample rural household survey released by the Huai’an Municipal Bureau of Statistics in 2019. Building on this baseline, our research team conducted follow-up investigations across all villages during 2020–2022 to supplement and verify key information. The dataset included basic village attributes and household-level socioeconomic and demographic characteristics. Additional city- and county/district-level statistics were obtained from the statistical yearbooks of Huai’an and its counties/districts. The geospatial and water network data at the village scale were obtained from the Third National Land Survey. The information on kindergartens, commercial facilities, and the locations of county and township government was obtained from POI data and then validated and corrected through field verification. After excluding villages with missing indicators and values, the final sample comprised spatial data for 1382 villages. Methodology Measurement of rural hollowing As mentioned above, rural hollowing manifests in four aspects: population, land, housing, and industry. Population hollowing refers to the depopulation of rural areas driven by large-scale out-migration 12 , 60 . The most direct manifestations of land hollowing include vacant and derelict housing as well as abandoned farmland 50 , 61 . Because it is difficult to obtain data on cultivated-land abandonment, studies usually use housing hollowing as a rough calculation, i.e., the proportion of vacant and abandoned homestead sites 18 , 39 , 52 , 62 . Industrial hollowing is reflected in sluggish development of non-agricultural sectors and the decline of agricultural industries 60 . However, owing to the difficulty in obtaining industrial data for administrative villages, this study only measured the level of rural hollowing from the two dimensions of population and housing. Population hollowing is expressed as the ratio of the number of households with all the members away from home to the total number of registered households in the village area. Housing hollowing is expressed as the ratio of the number of vacant houses in the village area to the total number of residential houses. Finally, the entropy method was used to comprehensively evaluate population hollowing ( ) and housing hollowing ( ), and the comprehensive rural hollowing level ( ) was calculated. The specific steps are as follows: Standardization. Because both population and housing hollowing are positive indicators, these are standardized using positive indicators. 1 In formula ( 1 ), is the standardized value of the indicator for sample , ( indicates the sample size); is the original value of the indicator for sample ; is the maximum value of the indicator; is the minimum value of the indicator; and is the shift range. (2) Calculate the weight. 2 3 4 In formula ( 4 ), is the weight of the indicator. is the difference coefficient of the indicator and is calculated as shown in Eq. ( 3 ). is the homogeneity quantification of the indicator and is calculated as shown in Eq. ( 2 ). (3) The level of rural hollowing was calculated. 5 In Eq. ( 5 ), is the level of rural hollowing for sample . Spatial autocorrelation The spatial autocorrelation method was used to study the degree of geographical spatial correlation of rural hollowing, including global and local autocorrelations. Moran’s I index is used to represent global autocorrelation. Its value is within the range [− 1,1]. The larger the magnitude, the stronger is the spatial correlation. Moran’ s I > 0 indicates that rural hollowing shows a positive spatial correlation with geographical distribution. The converse is true for a negative correlation. Hotspot analysis (Getis Ord Gi*) is a local autocorrelation analysis method used to identify the spatial differentiation patterns of rural hollowing and the distribution of cold and hot spots. Geographically weighted XGBoost (GWXGBoost) XGBoost is an extensible gradient-boosting machine learning system designed based on the concept of decision trees and is widely acknowledged 26 , 30 . This method has a high accuracy and strong interpretability. The introduction of a regularization term reduces the risk of overfitting and can achieve better regression results. However, similar to other machine learning algorithms, XGBoost does not consider spatial heterogeneity and cannot reveal the geographical variation of variables or provide more accurate governance strategies for rural smart shrinkage. Therefore, this study constructed a geographically weighted XGBoost (GWXGBoost) model. This model attempts to incorporate the concept of geographical weighting into the machine learning method and simultaneously consider non-stationary relationships and spatial heterogeneity to improve the accuracy of the model. GWR is a local regression method that is based on the first law of geography. This law states that “everything is related to everything else, but near things are more related than distant things 63 , 64 .” The GWR formula is expressed as 6 In formula ( 6 ), represents the spatial coordinates of sample , represents the intercept value, represents the regression parameter of sample , and is the residual. Among these, the regression parameter is calculated as 7 In formula ( 7 ), is the input matrix for all the samples, and are the corresponding spatial weight matrices. The spatial weights are calculated using the Gaussian function: 8 In formula ( 8 ), represents the bandwidth, and represents the Euclidean distance between samples and . After determining the spatial weights, is multiplied by to obtain a new data matrix. It is then used in XGBoost for calculations. By introducing spatial weights, the data points closer to the target sample location have a higher effect than those farther away. For a given sample , represents all its eigenvalues, and the final prediction of the XGBoost model is 9 In formula ( 9 ), represents the decision tree model in the iteration, represents the result of sample in the decision tree, and is the space of all decision trees. The objective function consists of a training loss function and a regularization term: 10 In formula ( 10 ), is the loss function, is the dependent variable of sample , is the predicted value, is the regularization term, is the functional expression of the decision tree, and is the total number of samples. The GWXGBoost computations in this study were conducted in Python. An adaptive bandwidth was adopted for the spatial component. For the machine-learning component, we selected the optimal hyperparameters using grid search with K-fold cross-validation to improve generalization performance and achieve a well-fitted model. Shapley additive explanation (SHAP) The SHAP model is based on game theory. It can overcome the low interpretability of machine learning models. At its core, it computes the marginal contribution of each feature to the model, which reflects the impact of each feature on the final predicted value and indicates the positivity or negativity of the impact. The SHAP value of feature is calculated as follows: 11 12 where is the interpretable model; = 0 or 1 ( indicating that the feature is observable, and indicates that the feature is missing); is the predicted mean of the predictive probability model; represents the contribution of feature , i.e., the SHAP value; is a set with features; and and are the results of the model with and without feature , respectively. Regionalization with dynamically constrained agglomerative and partitioning (REDCAP) To accurately identify the dominant influencing factors in each research area, we applied REDCAP to spatially cluster the SHAP values. REDCAP explicitly enforces spatial contiguity and regional integrity, producing geographically coherent zones that minimize within-region heterogeneity while maximizing between-region differences. The REDCAP method combines the SCHC (Spatially Constrained Hierarchical Clustering) and SKATER (Spatially ‘k’luster Analysis by Tree Edge Removal) algorithms and has the advantages of high flexibility and the capability to handle complex spatial relationships. Compared with other methods, the REDCAP method performs better in terms of the two indicators “within-cluster sum of squares” and “total sum of squares between clusters/total sum of squares”: REDCAP has a smaller within-cluster sum of squares, which indicates that the data points within the cluster are closer. Furthermore, it has a larger total sum of squares between clusters/total sum of squares, which indicates that the clusters are more separated. Therefore, the REDCAP method exhibits a better clustering effect. Model comparisons After the model was built, its effects were evaluated by R-square, mean square error (MSE), and mean absolute error (MAE). The fitting results for the GWXGBoost, GWR, and XGBoost models are listed in Table 2 . Compared with the other results, GWXGBoost has lower MSE and MAE values and higher R-squared values, which indicates a better fitting effect. Table 2. Results of model comparison. Model R -square MSE MAE GWR 0.494 0.0039 0.047 XGBoost 0.767 0.0081 0.033 GWXGBoost 0.791 0.0016 0.032 Open in a new tab Results Descriptive statistics and spatial characteristics of rural hollowing Descriptive statistics The rural hollowing level is generally high and varies substantially across villages (Table 3 ). The overall hollowing index ( ) ranged from 0% to 58.40%, with a median value of 10.46%. Housing hollowing and population hollowing exhibit similar minimum, mean, and median values; however, the maximum of housing hollowing is approximately 25% points higher than that of population hollowing, suggesting that housing vacancy constitutes a prominent manifestation of rural hollowing. In terms of quantity, 732 villages had values above the mean, whereas 480 villages had and 598 villages had above their respective means. Overall, the composite index (which is weighted by and ) provides a more comprehensive representation of rural hollowing conditions. Table 3. Description of dependent variables. Hollowing Max Min Mean Median (comprehensive rural hollowing) 58.40% 0.00% 12.05% 10.46% (population hollowing) 57.45% 0.00% 14.63% 12.93% (housing hollowing) 81.03% 0.00% 17.46% 15.35% Open in a new tab It should be clarified that, due to the unavailability of indicators for cultivated-land hollowing and industrial hollowing, our results represent a composite characterization based on population hollowing and housing hollowing. It is worth noting, however, that many existing studies operationalize rural hollowing primarily using population- or housing-based measures. Moreover, Qu et al. 52 , who constructed a multi-dimensional hollowing index using five indicators including population, housing, and industry, reported that single-dimension measures and composite indices can exhibit broadly similar spatial patterns, and that in some contexts the differences in land- and industry-related hollowing are not pronounced. This evidence suggests that, under certain research settings, population and housing dimensions can capture key features of overall rural hollowing to a meaningful extent. Nevertheless, we emphasize that the extent to which this holds is context-dependent. Accordingly, we explicitly acknowledge in the limitations section that the absence of cultivated-land and industrial hollowing indicators may introduce bias and may under-represent hollowing processes dominated by land abandonment or industrial land vacancy in some areas. Spatial characteristics of rural hollowing Overall, the spatial distribution of rural hollowing in Huai’an City shows a high degree of consistency in terms of population and housing (Fig. 3abc). Areas with low hollowing are mainly distributed along the outer periphery of the urban core and around the fringes of county/district-level centers. Meanwhile, areas with high hollowing are concentrated in the northern part of Lianshui County, most of Hongze District, and Xuyi County. A global Moran’s I test was used to examine the spatial autocorrelation of rural hollowing. The Moran’s I values for , , and were 0.2892, 0.3890, and 0.4008, respectively. At the 5% significance level, the Z-values of these three values were 16.9359, 22.7843, and 23.4531, respectively. This indicated that , , and were significantly positively correlated. The Getis-Ord Gi* values of , , and were calculated, and a local spatial cluster map was plotted (Fig. 3 d–f). These three still exhibited a strong spatial consistency. The hotspot areas are mainly distributed in the north and south of Huai’an City and concentrated in Shihu Town and Tangji Town of Lianshui County, Xuliu Town of Huaiyin District, Sanhe Town in the southern part of Hongze District, and Xuyi County. The cold spots are mainly concentrated around the urban area, with a small number distributed in Gaogou Town of Lianshui County and the surrounding areas of Hongze, Xuyi, and Jinhu counties. Fig. 3. Open in a new tab Spatial distribution of rural hollowing. Relative importance of variables The relative importance (RI) of the variables was measured using the SHAP method (Fig. 4 a). The global RI is the average magnitude of the feature over all the samples. It indicates the relative contribution of each variable 35 , 65 . The local interpretation plot (Fig. 4 b) visualizes the direction of the SHAP values for each variable. Here, red and blue indicate high and low eigenvalues, respectively. SHAP values above and below zero indicate the positive and negative effects of the variable on rural hollowing, respectively. Fig. 4. Open in a new tab Relative importance of variables. Overall, the factors related to the living, location, social, and economic fields contributed more to the rural hollowing level. This is consistent with the performance of Feature Importance in the GWXGBoost model (Fig. 4 c). In the local interpretation plot, “Offsite_Homeownership_Rate” contributes the most to and has a significant positive effect. “MultiStorey_Dwelling_Rate” has the second largest contribution to . However, it has a significant negative effect. That is, the higher the number of households with multi-storey dwellings in the village area, the lower is the level of hollowing out. Among the location field, “City_Access” and “County_Access” have positive effects on , whereas “Town_Access” has a negative effect. Among the other indicators, “New_Settlement_Count,” “Settlement_Dispersion,” and “Water_Density” have positive impacts on , whereas “Kindergarten_Access” and “Agri_Modernization” have negative impacts on . Spatial non-stationary effects of independent variables ArcGIS was used to spatially visualize the SHAP values of each variable (Fig. 5 ). The results were classified using the natural breakpoint method after positive and negative stratification. The impact patterns of City_Access, County_Access, Town_Access, Seclnd_FirmCount, Commercial_POI_Density, Kindergarten_Access, Offsite_Homeownership_Rate, and MultiStorey_Dwelling_Rate show clear spatial distribution characteristics. Fig. 5. Open in a new tab Spatial non-stationary effects of independent variables. Overall, certain variables exhibited spatial patterns of circular differentiation. For example, the spatial influence of the City_Access variable increases from the center of the urban area outward, and becomes a positive driver in the northeast of Lianshui, Xuyi, and Jinhu Counties. The influence structure of the County_Access variable is similar with City_Access, however, its negative driving circle is smaller than that of City_Access. The circular structure of Town_Access is weaker. Moreover, unlike the previous two variables, it mainly decreases outward from the town center. The spatial influence of Seclnd_FirmCount variable is mostly negative around urban and county areas and even around most towns, but the negative circle layer around the urban areas is larger. The influence of Commercial_POI_Density and Kindergarten_Access on space shows a structure of positive influence near the town area and negative influence in the outer areas of the town area. Also, the positive influence range in the three southern counties is generally smaller than that in the three northern counties. Offsite_Homeownership_Rate and MultiStorey_Dwelling_Rate also show a significant negative impact around urban areas. However, the overall range is small. In other areas, these two variables mainly show a positive impact. Notably, although Household_Size shows no overall spatial pattern, it has a significant negative impact in southern Jinhu County, which is the farthest area from the urban center. This is likely closely related to the local policies. Rural spatial zoning based on REDCAP To support precision-oriented rural governance and smart shrinkage, and to account for non-stationary relationships and spatially explicit influence pathways, we applied REDCAP to delineate zones within the study area. Specifically, we jointly considered the sign of each variable’s local relative importance (RI) and its non-stationary effects and ultimately selected the SHAP values of 13 key features as inputs to the REDCAP procedure. As shown in Fig. 6 , the within-cluster sum of squares levels off when the number of clusters reaches six, indicating that six is an appropriate and near-optimal choice. The resulting spatial zoning is presented in Fig. 7 . Fig. 6. Open in a new tab The optimal number of clusters determined by the REDCAP algorithm. Fig. 7. Open in a new tab Rural spatial zoning based on REDCAP ( a ) REDCAP clustering result; ( b ) Regression coefficient distribution across six zones. Zone I includes 545 villages, thereby accounting for 32.66% of the study area. These are mainly distributed in north-central Lianshui County, northern Huaiyin District, and eastern Huai’an District. These villages are located on the northernmost and easternmost parts of Huai’an City. The level of rural hollowing is relatively high, with an average of 16.06%. There are two rural hollowing hotspots and a cold spot in these villages. This area is mainly affected by the Offsite_Homeownership_Rate and MultiStorey_Dwelling_Rate variables in the living and location fields. The impact of the other variables is relatively small. This is an outer suburban population shrinking area dominated by the location and living fields. Zone II includes 214 villages, thereby accounting for 20.76% of the study area. It is mainly located in Hongze District, eastern Xuyi County, and western Jinhu County. These villages are distributed along major transportation corridors with a high level of hollowing (averaging 17.81%) and multiple hollowing hotspots. Mainly affected by the positive influence of Offsite_Homeownership_Rate and MultiStorey_Dwelling_Rate in the living field and significantly affected by the Household_Size and Settlement_Dispersion variables in the social field, it is an outer suburban population shrinking area dominated by living and social fields. Zone III includes 211 villages, thereby accounting for 11.49% of the study area. These are mainly distributed around the metropolitan area. The average hollowing-out level is 6.52%. This is the lowest level in the entire city and is consistent with the cold spot area of hollowing out. The area has convenient transportation and is mainly negatively affected by the variables of Offsite_Homeownership_Rate and MultiStorey_Dwelling_Rate in the living field. The influence of the other variables is weak. It is an inner suburban population shrinking area dominated by living field. Zone IV includes 181 villages, thereby accounting for 11.90% of the study area. It is mainly located in the western part of Huaiyin District, with an average hollowing level of 11.85%. This area is similar to Zone III, which is mainly affected by the Offsite_Homeownership_Rate and MultiStorey_Dwelling_Rate variables in the living field. However, Zone IV has both negative and weakly positive effects. In addition, the Poverty_Rate variable has a higher impact. It is an inner suburban population shrinking area dominated by economic and living fields. Zone V includes 166 villages, thereby accounting for 18.03% of the total study area. It is mainly located in the western part of Xuyi County, in the southwest corner of the study area. It is the farthest from the urban area, has the highest degree of hollowing (an average of 21.28%), and contains multiple hollowing hotspots. This area is mainly affected by the positive influence of City_Access. It is also influenced by multiple variables such as MultiStorey_Dwelling_Rate, Offsite_Homeownership_Rate, County_Access, Town_Fiscal_Revenue, and Water_Density. It is an outer suburban population shrinking area dominated by all the fields. Zone VI includes 65 villages, thereby accounting for 5.16% of the study area. It is mainly located in the eastern and southern parts of Jinhu County. The average level of hollowing in the villages is 9.24%, which is generally low. It is dominated by cold spots. This area is at a distance from the city and negatively affected by Household_Size in the social field. It is also significantly affected by Kindergarten_Access and is an outer suburban population shrinking area dominated by social and living fields. Discussion Spatial patterns of rural hollowing In this study, we referred to Li and Wang’s study to categorize hollow villages 42 . These can be classified into mildly hollow villages (≤ 5%), moderately hollow villages (between 5% and the mean value), and deeply hollow villages (≥ the mean value). The results show that 12% of the villages in the region were classified as mildly hollow, 45% as moderately hollow, and 43% as deeply hollow. This proportion is significantly higher than that of deeply hollow villages in Li and Wang’s study for the entire China (29.98%) and Jiangsu Province (19%) 42 . This indicates that with the development of urbanization and intensification of population exodus, a high level of rural hollowing occurs in economically developed areas. Spatially, villages hollowing level in the study area shows an increasing trend outward from the urban area. This is inconsistent with the performance of the classical push–pull theory, which indicates that the closer the population is to the economic center, the more likely it is to generate mobility 59 . This can be explained by the commuting distance. The urban area is the main place of employment for villagers. Villagers around the urban area with commuting distances less than the commuting limit distance can commute daily and have a low hollowing level 14 , 66 . Beyond the commuting limit distance, villagers tend to purchase houses at their workplace 67 . This results in a high hollowing level. This is most significant particularly in the villages in the northern part of Xuyi County and Lianshui County. The possible causes of rural hollowing In this study, we proposed to combine geographically weighted with XGBoost to construct the model and achieve a significant improvement in the overall performance of the model. GWXGBoost has lower MSE and MAE and a higher R-square. In addition, the use of SHAP to enhance the interpretability of the model that illustrates the positive, negative, and non-stationary characteristics of factors can better express the impacts of different variables on the geospatial space. This is also more conducive to the precise governance of rural spaces and the proposal of smart shrinkage policies. The effects of a few variables show evident spatial characteristics of a circular structure. This indicates that these variables differ significantly across geographical areas. Moreover, more circular structures are centered in urban areas. This indicates that urban areas are the most attractive to the entire city, which is in line with the actual perception. Specifically, in the location field, both the City_Access and the County_Access variables indicate that closer proximity to urban areas or county districts inhibits the development of rural hollowing. This is because areas closer to urban or county districts typically benefit from better infrastructure, public services, and employment opportunities, which increases the willingness of rural residents to remain in their areas and reduces population out-migration 68 , 69 . It is worth noting that the positive effect of areas farther from urban areas on rural hollowing is significantly higher than the negative effect of areas closer to urban areas, yet this is not reflected in the County_Access variable, which again suggests that distance from urban centers has a more significant impact on rural population relocation 70 . However, the Town_Access variable shows the opposite trend, with closer proximity to townships contributing to rural hollowing. This effect is exacerbated by the fact that townships tend to be more centralized in terms of economic development and resources 71 . As a result, rural populations are more likely to relocate to these townships. In addition, in this study area, the three northern counties exhibit a larger range of positive local effects, while the three southern counties show a greater range of negative local effects. This is because the three southern counties have a higher density of water networks, are predominantly mountainous, and have poorer transportation access compared to the northern region, resulting in relatively small positive local effects. In the production field, although the relative importance of Seclnd_FirmCount is the smallest, the positive local effect is the largest and spatially reflects the more obvious impact of the village’s location relative to urban areas and county districts on rural hollowing. In the living field, although the Offsite_Homeownership_Rate and MultiStorey_Dwelling_Rate variables have different directions of influence, their spatial non-stationary effects are similar, indicating that as proximity to urban areas increases, more farmers own multi-storey dwellings, fewer farmers buy houses in the town, and the effect on rural hollowing is more inhibitory. This is because these areas generally have a better economic level, and proximity to the city allows residents to enjoy the benefits of urban infrastructure without bearing the high cost of urban living, which improves the quality of life and increases the willingness of farmers to stay, reducing the need for migration to towns. It is also noteworthy that Xuyi County and Jinhu County in the study area were the farthest from the urban area. However, their hollowing levels differed significantly. Xuyi County has a higher hollowing level, which is mainly positively affected by Household_Size. Meanwhile, Jinhu County has a lower hollowing level, which is negatively affected by Household_Size. In addition to family responsibilities and educational considerations, policy differences may also contribute to this pattern. In Xuyi County, local practice has tended to prioritize the relocation of original residents and the introduction of outside business operators. Meanwhile, Jinhu County relies on local villagers to develop tourism. This has generated a difference in hollowing out between the two places. Strategies for counteracting rural hollowing based on smart shrinkage Building on the characteristic zoning derived from the key influencing factors, this study applies the smart shrinkage perspective to rural governance by explicitly accounting for the distinctive features of rural areas. The resulting policy recommendations are further aligned with China’s national rural revitalization agenda and relevant top-level policy documents, including the Rural Construction Action Implementation Plan (2022) 72 , the No. 1 Central Document for 2025 on deepening rural reform and advancing all-around rural revitalization 73 , and the Recommendations of the CPC Central Committee for Formulating the 15th Five-Year Plan for National Economic and Social Development (for the 2026–2030 period) 74 . In this way, we propose policy measures with clearer implementation relevance and pathways, ensuring that zone-specific strategies are both theoretically grounded and practically applicable within the current national policy framework. Zone Ⅰ is located at the edge of the city. It is a concentrated residential area, with a relatively poor economy and a high proportion of poor households. It is also experiencing severe population loss mainly owing to the influence of location and living fields. Therefore, it is recommended that the living environment in this area be improved, transportation accessibility be enhanced, attention be paid to the remaining poor population, the endogenous growth capacity of the rural economy be improved 40 , and skills training be promoted to increase the income of villagers. Zone Ⅱ is located along the lakeshore of Hongze Lake in the southern part of the urban area. It is dominated by agricultural land and is characterized by a relatively high per capita cultivated-land area and a high level of agricultural modernization. The proportion of multi-storey dwellings owned by farmers is the lowest in the city, whereas the number of home buyers is the highest. Despite these advantages, the zone has experienced substantial population out-migration. This is mainly owing to the influence of the social and living fields. The area has relatively good economic development and few poor households. However, the hollowing-out level is higher than that in Zone I. This shows that endogenous variables play a more decisive role 75 . Therefore, it is recommended to conduct a census of empty and abandoned houses, guide and encourage the adaptive reuse of vacant and dilapidated houses, improve villagers’ educational attainment, enhance the quality of the rural population, promote the voluntary and compensated exist from homesteads rights while respecting villagers’ preferences, conduct land consolidation, convert the characteristic of land use, develop agricultural industries and strengthen agricultural advancement to reduce the agricultural load on villagers. Zone Ⅲ is located closest to the urban area, yet it falls within the inner-suburban area. It hosts a relatively large number of industrial enterprises, which provide local employment opportunities. In contrast to Zone II, Zone Ⅲ has the highest citywide share of villagers owning multi-storey dwellings, whereas the share of home buyers is the lowest. Although this zone is primarily affected by the negative influence of the living field, it experiences the smallest population loss. Overall, Zone Ⅲ is relatively well developed and has strong potential for further growth. Policy priorities should therefore focus on enhancing the endogenous industrial growth and supporting enterprise development, in line with existing development trajectories, to encourage nearby employment among local residents. Zone IV is also located in the suburban area. Although its overall GDP is relatively high, it has the lowest average household income in the city. A relatively large share of villagers owns multi-storey dwellings, while the share of home buyers is low few. Meanwhile, the rate of hollowing-out remains relatively high, primarily influenced by the economic and living fields. In this type of area, out-migration is largely driven by the pursuit of higher incomes and a better quality of life 75 . Therefore, policy efforts should focus on improving the local economic base and raising household incomes. More broadly, better coordination of urban-rural development is essential for advancing regional urban–rural integration. Zone Ⅴ is located on the southwest of the city, largely within Xuyi County. It is characterized by a relatively high level of agricultural modernization and dispersed rural settlements. The southern part of this area borders Nanjing, the provincial capital, and benefits from strong external connectivity; nevertheless, it has experienced severe population out-migration. We recommend that this area utilize its location and transport advantages to strengthen outbound logistics for agricultural products and promote agricultural scaling-up and modernization. Meanwhile, as mentioned above, the tourism-oriented strategy that primarily encourages the relocation of local residents and relies on outside operators may be less conducive to long-term development. Instead, it is recommended to conduct an inventory of existing houses, support remaining residents in upgrading and rehabilitating usable dwellings, identify and leverage distinctive cultural assets, and develop community-based tourism with local residents as key participants. Tourism revenues can then serve as a supplementary income source, enhancing residents’ sense of participation and well-being. Zone Ⅵ is located at the interface between the two lakes, largely within Jinhu County. It is an important rural tourism area in the Yangtze River Delta and exhibits a relatively low rate of rural hollowing. Tourism development in this zone is largely community-based, with local residents playing a central role. Notably, many villagers live in the county seat while working in rural areas, which helps sustain local employment and realizes the economic value of the rural tourism economy. Overall, Zone VI is relatively well developed and has strong potential for further growth. Policy should therefore respect and support villagers’ initiatives, improve tourism facilities and services, strengthen destination branding and promotion, and broaden local participation so as to contribute to common prosperity. Limitations and future work This study has certain limitations. First, owing to the deficiency of village-level data on industries and abandoned cultivated land, it is infeasible to accurately measure land hollowing and industrial hollowing. Accordingly, “rural hollowing” in this paper primarily refers to population hollowing and housing hollowing, which may introduce measurement error and may under-represent hollowing processes dominated by cultivated-land abandonment and industrial land vacancy. Second, constrained by data availability, the ecological field is represented by a single indicator—water-network density—which may not fully capture the multi-dimensional ecological conditions of villages and could affect the robustness of ecological-field inferences. Third, our analysis relies on aggregated data at the administrative-village level and does not incorporate household- or individual-level information within villages. Fourth, although the combination of GWR and XGBoost considers spatial heterogeneity, the specific calculation process algorithm limits the calculation of spatial effects to a certain extent. Future work could further refine the method to better characterize spatial effects. In addition, the current study evaluates only one machine-learning algorithm within the spatial framework and does not benchmark other spatial machine-learning approaches (e.g., geographically weighted random forests, GWRF). Future research could compare multiple spatial ML models in terms of goodness-of-fit and explanatory performance to identify the most suitable approach. Finally, SHAP models can also be used to examine the interaction effects among variables, which merits deeper investigation in subsequent studies. Conclusion This study used a full-sample survey to examine rural hollowing, constructed a GWXGBoost model, revealed the spatial non-stationary effects between variables and rural hollowing. Spatial zoning was conducted using REDCAP clustering to identify the most important influencing factors in each region and propose policy recommendations based on Smart Shrinkage. The research results are summarized as follows: (1) Rural hollowing shows a spatial distribution, with higher levels in the outer areas of the city and lower levels in the surrounding urban areas and the southern part of Jinhu County. Except in the southern part of Jinhu County, rural hollowing generally shows a ring structure. The moderate and deep hollow villages account for 45% and 43% respectively. (2) From the global RI perspective, the factors of location and living fields have the highest impact on hollowing, with variables such as Offsite_Homeownership_Rate, MultiStorey_Dwelling_Rate, and City_Access contributing significantly. From the local RI, Offsite_Homeownership_Rate, City_Access, County_Access, New_Settlement_Count, Scattering, and Water_Density have a significant positive impact. MultiStorey_Dwelling_Rate, Town_Access, Kindergarten_Access, and Agri_Modernization have significant negative impacts. (3) In terms of spatial non-stationary effects, the influence of the location and living fields’ variables shows the spatial characteristics of circular differentiation. Among the other spatial fields, only Seclnd_FirmCount exhibits weak spatial characteristics. Specifically, the impact of City_Access, Offsite_Homeownership_Rate, and MultiStorey_Dwelling_Rate increase outward from the city center. The impact of County_Access increases outward from the city center and county center. The impact of Town_Access, Commercial_POI_Density, and Kindergarten_Access decrease outward from the town center. (4) To propose precise governance countermeasures for smart shrinkage of hollow villages with large development differences, based on that regions with similar local effects generally have similar hollowing levels and impact characteristics and considering the RI and spatial non-stationary, the study area is divided into six subzones using REDCAP clustering. For hollowed villages near the city, policies should prioritize enhancing the endogenous economic momentum and promoting balanced urban-rural integration. For villages far from the city and with relatively high rural hollowing, inventions should focus on improving accessibility, establishing an inventory of vacant housing, and facilitating the adaptive reuse or removal of vacant dwellings. For villages far from the city and with relatively low rural hollowing, a more flexible approach, building on existing development trajectories and supporting villagers’ self-directed initiatives, is advised. For the hollowed villages in the middle-zone area, policies should encourage the rehabilitation and upgrading of dilapidated houses, implement land consolidation, and advance agriculture modernization. Author contributions C.C. and C.W. designed the study jointly. C.W. collected the data. C.C. performed the analyses and collaborated with C.W. in interpreting the results and writing the manuscript. L.C. and C.G.C. reviewed the manuscript and provided valuable suggestions for revision. Funding This research was funded by the National Research Foundation of Korea grant, Grant number NRF-2020R1A2C1008509. Data availability The original data presented in the study are openly available in Mendeley Data at DOI: 10.17632/yrfntmjsyz.1. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Chi Chen and Chengxiang Wang contributed equally to this work. Contributor Information Chengxiang Wang, Email: [email protected]. Chang Gyu Choi, Email: [email protected]. References 1. Liu, Y. Research on the urban-rural integration and rural revitalization in the new era in China. Acta Geogr. Sin. 73 , 637–650 (2018). [ Google Scholar ] 2. Liu, Z. Rural population decline, cultivated land expansion, and the role of land transfers in the farming-pastoral ecotone: A case study of Taibus, China. Land 11 , 256 (2022). [ Google Scholar ] 3. Pontones-Rosa, C., Pérez-Morote, R. & Santos-Peñalver, J. F. ICT-based public policies and depopulation in hollowed-out Spain: A survey analysis on the digital divide and citizen satisfaction. Technol. Forecast. Soc. Chang. 169 , 120811 (2021). [ Google Scholar ] 4. Hu, X. Implementation of village-based policies: A trap of hollow rural population research on transformation of development path. J. Lanzhou University(Social Sciences) . 44 , 50–60 (2023). [ Google Scholar ] 5. Luo, X. et al. Revealing the formation mechanism of rural housing hollowing via considering the nonlinear effects of spatial factors and planning variables. Appl. Geogr. 155 , 102957 (2023). [ Google Scholar ] 6. Wang, L. & Wu, J. Study on the influencing factors of homestead hollowing in a households’ perspective. Geogr. Res. 38 , 2202–2211 (2019). [ Google Scholar ] 7. Liu, Y. & Liu, Y. Progress and prospect on the study of rural hollowing in China. Geogr. Res. 29 , 35–42 (2010). [ Google Scholar ] 8. Long, H., Li, Y. & Liu, Y. Analysis of evolutive characteristics and their driving mechanism of hollowing villages in China. Acta Geogr. Sin. 64 , 1203–1213 (2009). [ Google Scholar ] 9. De Montis, A., Ledda, A., Ganciu, A., Serra, V. & De Montis Recovery of rural centres and albergo diffuso: A case study in Sardinia, Italy. Land. Use Policy . 47 , 12–28 (2015). [ Google Scholar ] 10. Xu, A., Gao, X., Li, Q. & Xie, T. Characteristic analysis and type identification of village hollowing on the Chengdu Plain. Resour. Sci. 38 , 0196–0205 (2016). [ Google Scholar ] 11. Xiao, L., Hu, L., Dai, L. & Li, H. Characteristics and influencing factors of rural ‘hollow houses’ renovation reclamation types: A case study of Miluo City. Econ. Geogr. 41 , 173–180 (2021). [ Google Scholar ] 12. Zheng, D., Wen, Q., Wang, Y. & Mi, H. Differentiation mechanism and reconstruction strategy of rural population hollowing in China. Econ. Geogr. 39 , 161–168 (2019). [ Google Scholar ] 13. Yu, K., Wang, J. & Su, Z. Ecological management and spatial conduction mechanism of hollow villages based on the town scale: A case study from Hengkou District, A model area for Town-Level City reform in Shanxi Province. Ecol. Econ. 38 , 182–188 (2022). [ Google Scholar ] 14. Zhao, L. & Wang, X. Rural housing vacancy in metropolitan suburbs and its influencing factors: A case study of Nanjing, China. Sustainability 13 , 3783 (2021). [ Google Scholar ] 15. Gao, X. et al. Understanding rural housing abandonment in China’s rapid urbanization. Habitat Int. 67 , 13–21 (2017). [ Google Scholar ] 16. Chen, Y., Li, P., Ma, X. & Yang, W. The study of rural hollowing in Shandong based on the township geographical unit: A case study of six villages in Shandong. Morden Urban Res. 24–30 10.3969/j.issn.1009-6000.2018.03.004 (2018). 17. Long, H., Li, Y., Liu, Y., Woods, M. & Zou, J. Accelerated restructuring in rural China fueled by ‘increasing vs. decreasing balance’ land-use policy for dealing with hollowed villages. Land. Use Policy . 29 , 11–22 (2012). [ Google Scholar ] 18. Wang, D., Zhu, Y., Zhao, M. & Lv, Q. Multi-dimensional hollowing characteristics of traditional villages and its influence mechanism based on the micro-scale: A case study of Dongcun Village in Suzhou, China. Land. Use Policy . 101 , 105146 (2021). [ Google Scholar ] 19. Liu, Z., Wang, Q. & Du, G. Spatial differentiation and influencing factors of hollowing rural residential land in northeastern major grain production region: A case study of Keshan County in Heilongjiang Province. China Land. Sci. 34 , 94–102 (2020). [ Google Scholar ] 20. Liu, H., Zhang, H., Wang, Y., Xu, Y. & Dai, D. Types of shrinking villages and flexible planning strategies in the Jianghan Plain region: Based on an investigation of 606 villages in Xiantao City. Resour. Sci. 46 , 407–420 (2024). [ Google Scholar ] 21. Cui, J., Guo, G. & Zhang, H. Hollowing-out characteristics of rural residential intensive land use: Microscale analysis of a typical village in the North China Plain. J. Arid Land. Resour. Environ. 37 , 94–103 (2023). [ Google Scholar ] 22. Lu, D., Zhang, Y. & Ji, S. The nonlinear effect of urban family age structure on consumption: An empirical study based on machine learning. Northwest. Popul. J. 43 , 89–98 (2022). [ Google Scholar ] 23. Chen, Y., Su, Z., Chen, Y. & Lin, Z. Modeling of spatial effects of PM2.5 in the Yangtze River Delta based on geographically weighted random forest. China Environ. Sci. 1–9. 10.19674/j.cnki.issn1000-6923.20240326.004 (2024). 24. Wei, X., Zhen, F. & Xi, G. Characteristics and influencing factors of residents’ non-commuting travel in Nanjing metropolitan area. Scientia Geogr. Sinica . 43 , 1740–1750 (2023). [ Google Scholar ] 25. Zhang, X., Xu, Q., Song, L. & Deng, J. Classification of urban rail transit lines and determinants of the ridership from the perspective of land use. Progr. Geogr. 42 , 2337–2349 (2023). [ Google Scholar ] 26. Tang, G. & Chen, C. Analysis of factors influencing the evolution of rural settlements in major grain producing areas based on explainable machine learning: A case in central China. Sci. Technol. Eng. 23 , 9378–9387 (2023). [ Google Scholar ] 27. Zhao, H., Gu, T., Sun, D. & Miao, C. Dynamic evolution and influencing mechanism of urban human settlements in the Yellow River Basin from the perspective of ‘production-living-ecological’ function. Acta Geogr. Sin. 78 , 2973–2999 (2023). [ Google Scholar ] 28. Zhu, H. et al. Investigating the multiscale associations between urban landscape patterns and PM1 pollution in China using a new combined framework. J. Clean. Prod. 456 , 142306 (2024). [ Google Scholar ] 29. Guo, B. et al. Identifying population hollowing out regions and their dynamic characteristics across central China. Sustainability 14 , 9815 (2022). [ Google Scholar ] 30. Fu, Z. et al. Geographical types and driving mechanisms of rural hollowing-out in the Yellow River Basin. Agriculture 14 , 365 (2024). [ Google Scholar ] 31. Shen, L., Long, Y. & Hao, Y. Exploring the impact of public service facilities on housing rent price from the perspective of spatial heterogeneity: An empirical study of an improved machine learning method. China Land. Sci. 37 , 67–79 (2023). [ Google Scholar ] 32. Ye, M. et al. Estimation of the soil arsenic concentration using a geographically weighted XGBoost model based on hyperspectral data. Sci. Total Environ. 858 , 159798 (2023). [ DOI ] [ PubMed ] [ Google Scholar ] 33. Li, L. Geographically weighted machine learning and downscaling forhigh-resolution spatio-temporal estimations of wind Speed. Remote Sens. 10.3390/rs11111378 (2019). [ Google Scholar ] 34. Hagenauer, J. & Helbich, M. A geographically weighted artificial neural network. Int. J. Geogr. Inf. Sci. 36 , 215–235 (2022). [ Google Scholar ] 35. Yang, W., Li, Y., Liu, Y., Fan, P. & Yue, W. Environmental factors for outdoor jogging in Beijing: Insights from using explainable spatial machine learning and massive trajectory data. Landsc. Urban Plann. 243 , 104969 (2024). [ Google Scholar ] 36. Wen, Q., Li, J., Ding, J. & Wang, J. Evolutionary process and mechanism of population hollowing out in rural villages in the farming-pastoral ecotone of northern China: A case study of Yanchi County, Ningxia. Land. Use Policy . 125 , 106506 (2023). [ Google Scholar ] 37. Li, Y., Westlund, H., Zheng, X. & Liu, Y. Bottom-up initiatives and revival in the face of rural decline: Case studies from China and Sweden. J. Rural Stud. 47 , 506–513 (2016). [ Google Scholar ] 38. Zhu, S., Kong, X. & Jiang, P. Identification of the human-land relationship involved in the urbanization of rural settlements in Wuhan city circle, China. J. Rural Stud. 77 , 75–83 (2020). [ Google Scholar ] 39. Liu, Y., Liu, Y., Chen, Y. & Long, H. The process and driving forces of rural hollowing in China under rapid urbanization. J. Geogr. Sci. 20 , 876–888 (2010). [ Google Scholar ] 40. Wang, F. et al. A novel index for assessing the rural population hollowing at fine spatial resolutions based on Tencent social media big data: A case study in Guangdong Province, China. Land. Use Policy . 137 , 107028 (2024). [ Google Scholar ] 41. Zhang, R., Jiang, G. & Zhang, Q. Does urbanization always lead to rural hollowing? Assessing the spatio-temporal variations in this relationship at the county level in China 2000–2015. J. Clean. Prod. 220 , 9–22 (2019). [ Google Scholar ] 42. Li, Y. (ed, H.) Spatial distribution of rural population flow at the village level in China: Evidence from village samples in the Third National Agricultural Census. Chin. Rural Econ. 124–144. https://link.cnki.net/urlid/11.1262.F.20200427.1124.014 (2020). 43. Jiao, L. & Zhang, M. Vacant mechanism and smart shrinkage strategies for shrinking village: based on field studies in rural areas of eastern Henan. Econ. Geogr. 41 , 221–232 (2021). [ Google Scholar ] 44. Popper, D. E. & Popper, F. J. Small can be beautiful. Planning 68 , 20–23 (2002). [ Google Scholar ] 45. Reis, J. P., Silva, E. A. & Pinho, P. Spatial metrics to study urban patterns in growing and shrinking cities. Urban Geogr. 37 , 246–271 (2016). [ Google Scholar ] 46. Wiechmann, T. & Pallagst, K. M. Urban shrinkage in Germany and the USA: A comparison of transformation patterns and local strategies. Int. J. Urban Reg. Res. 36 , 261–280 (2012). [ DOI ] [ PubMed ] [ Google Scholar ] 47. Hu, Z., Li, Y., Long, H. & Kang, C. The evolution of China’s rural depopulation pattern and its influencing factors from 2000 to 2020. Appl. Geogr. 159 , 103089 (2023). [ Google Scholar ] 48. Jedwab, R., Christiaensen, L. & Gindelsky, M. Demography, urbanization and development:Rural push, urban pull and … urban push? J. Urban Econ. 98 , 6–16 (2017). [ Google Scholar ] 49. Zhang, T., Wang, Y., Liu, Y. & Zhao, M. Establishing an economic insurance system under a multiple dynamic evolution mechanism after rural hollowing renovation. Resour. Sci. 38 , 799–813 (2016). [ Google Scholar ] 50. Zhang, Z. A study on the formation and rejuvenation path of ‘hollow villages’ under the background of rural revitalization: Based on a case of Chencun in S County of Z Province. J. Grad. School Chin. Acad. Soc. Sci. 98–105 (2019). 51. Wang, L., Chen, K. & Li, N. The degree measure and spatial-temporal variation characteristics of rural population hollowing on county-level in China. Popul. J. 39 , 14–24 (2017). [ Google Scholar ] 52. Qu, Y. et al. Multidimensional form identification and targeted governance of hollow villages from the rural revitalization perspective: Taking Fangsi Town of Yucheng City in Shangdong Province as an example. Resour. Sci. 43 , 776–789 (2021). [ Google Scholar ] 53. Yang, R., Liu, Y., Guo, L. & Li, Y. Spatial-temporal characteristics for rural hollowing and cultivated land use intensive degree: Taking the Circum-Bohai Sea region in China as an example. Progr. Geogr. 32 , 181–190 (2013). [ Google Scholar ] 54. Li, J. & Chen, C. The research of the mountian households residential relocation on the basis of village residential field and power. Econ. Geogr. 30 , 1164–1169 (2010). [ Google Scholar ] 55. Wang, C., Pang, Z., Liu, C. & He, P. Inter-village differences and driving factors of rural households’ urban housing purchase in typical county: A case study of Lianshui County in Jiangsu Province. Econ. Geogr. 42 , 158–168 (2022). [ Google Scholar ] 56. Du, Y., Liang, Z. & Zhao, Q. Ontology and mechanism: Construction, evolution and application of field theory. Urban Plann. Int. 37 , 59–66 (2022). [ Google Scholar ] 57. Kluttz, D. N. & Fligstein, N. Varieties of sociological field theory. In Handbook of Contemporary Sociological Theory (ed Abrutyn, S.) 185–204. 10.1007/978-3-319-32250-6_10 (Springer International Publishing, 2016). [ Google Scholar ] 58. Kump, B. Lewin’s field theory as a lens for understanding incumbent actors’ agency in sustainability transitions. Environ. Innov. Soc. Trans. 46 , 100683 (2023). [ Google Scholar ] 59. Lee, E. S. A theory of migration. Demography . 3 , 47–57 (1966). 60. Li, Y., Liu, Y., Long, H. & Cui, W. Community-based rural residential land consolidation and allocation can help to revitalize hollowed villages in traditional agricultural areas of China: Evidence from Dancheng County, Henan Province. Land. Use Policy . 39 , 188–198 (2014). [ Google Scholar ] 61. Jiang, S. & Luo, P. A literature review on hollow villages in China. China Popul. Resour. Environ. 24 , 51–58 (2014). [ Google Scholar ] 62. Wang, J., Liu, Y. & Chen, Y. Empirical analysis on influencing factors of the hollowing village degree—Based on the survey data of sample villages in Shandong Province. J. Nat. Resour. 28 , 10–18 (2013). [ Google Scholar ] 63. Fotheringham, A. S., Charlton, M. E. & Brunsdon, C. Geographically weighted regression: A natural evolution of the expansion method for spatial data analysis. Environ. Plan. A . 30 , 1905–1927 (1998). [ Google Scholar ] 64. Tobler, W. R. A computer movie simulating urban growth in the detroit region. Econ. Geogr. 46 , 234 (1970). [ Google Scholar ] 65. Chen, E., Ye, Z. & Wu, H. Nonlinear effects of built environment on intermodal transit trips considering spatial heterogeneity. Transp. Res. Part. D Transp. Environ. 90 , Article102677 (2021). [ Google Scholar ] 66. Antipova, A. Analysis of commuting distances of low-income workers in Memphis metropolitan area, TN. Sustainability 12 , 1209 (2020). [ Google Scholar ] 67. Wang, X., Hu, P. & Zhu, Y. Location choice of Chinese urban fringe residents on employment, housing, and urban services: A case study of Nanjing. Front. Architect. Res. 5 , 27–38 (2016). [ Google Scholar ] 68. Hong, H., Liu, T., Liao, H., Cai, Z. & Wang, G. Analysis of the housing–jobs separation characteristics of different village types in the mountainous and hilly region of southwest China. Land 11 , 1990 (2022). [ Google Scholar ] 69. Zambon, I., Cerdà, A., Gambella, F., Egidi, G. & Salvati, L. Industrial sprawl and residential housing: exploring the interplay between local development and land-use change in the valencian community, Spain. Land 8 , 143 (2019). [ Google Scholar ] 70. Hidayat, A. R. T., Onitsuka, K., Sianipar, C. P. M. & Hoshino, S. Distance-dependent migration intention of villagers: comparative study of peri-urban and remote villages in Indonesia. Admin. Sci. 12 , 48 (2022). [ Google Scholar ] 71. Dou, R. et al. How does townships’ economic landscape affect urban–rural income inequality in China? Socio-Econ. Plann. Sci. 98 , 102150 (2025). [ Google Scholar ] 72. Xinhua News. General Office of the CPC Central Committee and General Office of the State Council Issued the Rural Construction. Action Implementation Plan. (2022). https://www.gov.cn/zhengce/2022-05/23/content_5691881.htm 73. Xinhua News. China outlines key tasks to deepen rural reforms, advance rural revitalization. (2025). https://english.www.gov.cn/policies/latestreleases/202502/23/content_WS67bb002bc6d0868f4e8efe69.html 74. Xinhua News. Full text: Recommendations of the Central Committee of the Communist Party of China for Formulating the 15th Five-Year Plan for National Economic and Social Development. (2025). https://english.www.gov.cn/news/202510/28/content_WS6900adb9c6d00ca5f9a07216.html 75. Liu, Y., Shu, L. & Peng, L. The hollowing process of rural communities in China: Considering the regional characteristic. Land 10 , 911 (2021). [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Data Availability Statement The original data presented in the study are openly available in Mendeley Data at DOI: 10.17632/yrfntmjsyz.1. Articles from Scientific Reports are provided here courtesy of Nature Publishing Group ACTIONS View on publisher site PDF (8.2 MB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top