Spatial Differentiation Characteristics, Driving Mechanisms, and Governance Strategies of Rural Poverty in Eastern Tibet

Abstract

Rural areas in Tibet, with its complex terrain, fragile ecology, and poor facilities, are subject to a combination of social–ecological system elements, facing the typical risks of recurrent, marginal, and potential poverty. At present, the spatial differentiation and driving mechanism of rural spatial poverty risk in Tibet are not clear, which adversely affects the formulation of differentiated and precise governance strategies. Thus, based on the social–ecological system perspective, 967 poor rural villages in eastern Tibet were taken as an example, using intelligent techniques such as random forest, geographic detector, and multi-scale geographically weighted regression to identify the spatial differentiation characteristics and the driving mechanism of poverty. The results indicated that (1) the high poverty incidence of rural areas in eastern Tibet showed a scattered block distribution, of which approximately 37% of the villages presented a spatial distribution characterised by a high degree of clustering of the high poverty incidence. (2) Topography and the level of public facilities were key factors influencing the poverty levels of rural areas in eastern Tibet, in which the coupling explanatory power between the construction land slope index (CLSI) and several poverty-causing factors was high. (3) Geological disaster, land surface temperature, CLSI, traffic accessibility, livestock resources, cropland per capita, and tourism resources differentially drove the poverty incidence of rural areas in eastern Tibet, forming spatial partitions dominated by the risks of potential, marginal, and recurrent poverty. For different partitions, differentiated governance strategies of upgrading ecological environments, optimising geographical locations, and revitalising social resources were proposed to provide references for solving the problem of relative poverty in the new period.

1. Introduction

‘Poverty Eradication’ is the first of the 17 Sustainable Development Goals that has been proposed in ‘Transforming our World: The 2030 Agenda for Sustainable Development’ in 2015 [1]. By 2020, the number of people living in poverty globally with incomes of <USD 3.2 (the poverty line for lower-middle-income countries) and USD 5.5 (the poverty line for upper-middle-income countries) per day were 1.81 billion and 3.27 billion, respectively, suggesting that the problem of global poverty remains serious [2]. The persistence of significant poverty gaps in different regions [3], and the existence of approximately 80% of the poor living in rural areas, makes rural poverty critical for global poverty governance [4].

In recent years, research on global poverty has unveiled the intricate connections between poverty and a variety of socioeconomic factors, which are interwoven with issues such as energy access, climate change, food security, disability, and natural disasters [5]. Furthermore, revisions to the estimates of purchasing power parity (PPP) have influenced global poverty assessments, highlighting that the multidimensionality of poverty means that a single economic indicator is not sufficient to fully reflect the actual situation. For instance, although the extreme poverty rate in Sub-Saharan Africa has declined, the issue of poverty remains severe [6]. Additionally, the links between vulnerability brought by globalization, poverty, and natural disasters have increasingly gained attention; these factors interact with each other, collectively shaping socioeconomic vulnerability [7]. Lastly, the concept of multidimensional poverty has been introduced to assess various aspects of poverty more comprehensively, including health, education, and living standards [8]. The study of global poverty reflects a multidimensional approach to understanding and resolving the issue, emphasising the interrelated nature of poverty with energy access, food insecurity, disability, and natural disasters, and underscores the necessity of exploring coordinated and sustainable poverty reduction strategies from a socio–ecological system perspective.

The Chinese government has always emphasised the issue of poverty governance and had achieved tremendous results and progress in poverty eradication by 2020 [9,10]. However, approximately 19% of China’s population (267 million people) was still below the World Bank’s poverty line for upper-middle-income countries in 2022 at USD 6.85 per person per day (2017 purchasing power parity). Moreover, due to the large number of poor people in China’s rural villages, the diversity and complexity of the topography and the uneven distribution of living and production resources [9], many villages will remain in a state of relative poverty for a long time [11]. Large-scale poverty alleviation and development have resulted in the current poor population being distributed mainly in ecologically fragile and geographically remote areas with poor infrastructure, such as the Tibet region. In addition, China’s short-term intensive support behaviours are hardly sustainable to consolidate the results of poverty eradication [12]. How to sustainably mitigate the risk of rural poverty in socially fragile areas has become the key to China’s rural poverty reduction efforts.

In existing studies on poverty, there has been a shift from an early focus on the single dimension of income poverty to multi-dimensional poverty, such as education, healthcare, housing, and employment. With the introduction of the geographic perspective, it gradually expanded to spatial poverty triggered by elements such as regional geographic conditions, ecological environment, resource endowment, and infrastructure [13,14,15]. The spatial poverty theory was developed from the earlier Spatial Economics theory [16] and the evidence of instances where natural geographic factors lead to spatial poverty traps [17], covering the research of geo-environmental and social–structural factors and emphasising the important role of spatial geographic elements in the formation of poverty [18]. Thus, rural spatial poverty can be understood as the lack of geographic capital (e.g., locational conditions, natural endowments, infrastructure, and cultural resources) in a particular region leading to the emergence of poverty traps. Its non-variable and complex characteristics lead to the vulnerability and imbalance of rural poverty, which manifests itself as poverty formed by the insufficiency of, or differences in, natural resources and socioeconomic development [19,20,21]. In terms of social systems, the spatial mismatch between agricultural production facilities and education, as well as healthcare facilities and the needs of the villagers, leads to a mismatch between supply and demand that exacerbates the risk of recurrent poverty in rural areas [22,23,24,25]. The qualitative and locational disadvantages of cropland, an important resource for supporting rural industrial development, and road transportation, an important link between urban and rural interactions, can lead to the risk of marginal poverty [26]. In terms of ecosystems, complex geography, a fragile ecological environment, and frequent natural disasters have intensified the vulnerability and instability of China’s localised rural development [27], bringing the risk of potential poverty.

It has been shown that rural poverty manifests itself as a state of factor shortage and structural imbalance in the social–ecological system [27], which provides the basis for the study of the spatial differentiation characteristics and driving mechanisms of rural poverty. However, established rural poverty studies based on social–ecological system perspectives have focused on vulnerability or resilience assessments of poor areas [28,29,30], but have paid insufficient attention to the mechanisms that lead to spatial differences in rural poverty. At the same time, rural spatial poverty is facing new challenges due to the spatial differentiation of natural resources and socioeconomic development in the new period, that is, potential poverty caused by a fragile ecological environment, marginal poverty caused by a poor geographical location, and recurrent poverty caused by insufficient endogenous motivation. Identifying the spatial heterogeneity of different poverty-causing factors has become the key to resolving the risk of rural spatial poverty in the new period. Therefore, this study explores the characteristics of spatial differentiation of poverty and its driving factors in rural areas of eastern Tibet from the perspective of social–ecological systems and proposes targeted poverty governance strategies. The following questions are specifically explored: (1) What are the spatial differentiation characteristics of poverty in rural areas of eastern Tibet? (2) What are the driving factors of spatial differentiation of rural poverty? Including the composition of the driving factors, the magnitude of independent, and interactive effects of different factors? (3) What are the strategies for poverty governance in villages with different poverty types? The results of this study can effectively mitigate the risks of poverty, such as recurrent, marginal, and potential, and can provide a reference for global sustainable poverty alleviation measures.

2. Research Perspectives and Technical Route

2.1. Research Perspectives

The global incidence of poverty has been on the rise in recent years, with absolute and relative poverty coexisting. In the case of China, relative poverty has become the dominant type, and the effects of poverty reduction show large variations across regions. Together, social system and ecological system factors influence the current relative poverty rate in China. Therefore, the social–ecological system research perspective is the key to analysing the current spatial differentiation mechanism of rural poverty and its governance strategy (Figure 1).

At present, there is significant spatial differentiation in China’s rural poverty areas, and localised villages have prominent relative poverty problems under the influence of social–ecological system elements involving insufficient endogenous motivation, fragile ecological environments, and unfavourable geographical locations. First, recurrent poverty is caused by poor organisational management, unsustainable industrial development, and unsuitable facilities, leading to insufficient endogenous motivation and causing rural areas that have been lifted out of poverty to fall back into poverty. Second, potential poverty, owing to the menace of geological hazards, paucity, or maldistribution of natural resources, and environmental contamination arising from both residential and productive activities, leads to a high poverty risk in localised rural areas. Third, marginal poverty, caused by the complex topography and remote geographical location, leads to the decentralised production and life of settlements and isolated transportation, generating the phenomenon of poverty spatial concentration (Figure 2). Therefore, focusing on relatively poverty-stricken rural areas from a social–ecological system perspective and identifying the heterogeneous driving mechanisms of spatial poverty can help target and address recurrent, potential, and marginal poverty, thereby narrowing the poverty gap between regions.

2.2. Technical Route

Poverty-related social–ecological system elements involve the ecological environment, geographical location, and social resources. In the aspect of the ecological environment, there is an interdependence between poverty and natural resources [31,32], and natural disasters are an important contributor to poverty [33]. In terms of social resources, industrial development and infrastructure can reflect the underlying economic development and potential of poor villages. In terms of geographical location, the life and production of villages are susceptible to the constraints of the topographic location [34,35], and the degree of difficulty in making external contacts is an important factor in poverty. Following the principles of scientificity, representativeness, and data availability, 33 variables such as the Ecological Remote Sensing Index were obtained through technical methods such as GEE, Python, FragStats, and data statistics, covering the three elements of ecological environment, geographical location, and social resources. First, the poverty incidence and degree of clustering of poor villages were calculated to explore the spatial clustering of poverty. Second, the effective social–ecological system poverty-causing factors based on intelligent techniques such as the multi-collinearity test, RF, and GeoDetector were filtered. Then, MGWR was used to analyse the spatial differentiation driving mechanism of the poverty-causing factors. Finally, the classified sustainable poverty governance policy recommendations were proposed (Figure 3).

3. Materials and Methods

3.1. Study Area

Eastern Tibet is located in the Hengduan mountainous region of the Qinghai–Tibet Plateau, which is China’s only provincial-level area of concentrated and contiguous poverty and is one of the regions in China with the widest scope of poverty, the deepest degree of poverty, and the greatest difficulty in eradicating poverty. The study area is predominantly mountainous, with an average elevation of more than 3500 m above sea level, presenting a ridge-and-valley landscape with three parallel mountains and three rivers (Figure 4). At the end of 2019, eastern Tibet entered a new phase of consolidating the results of poverty eradication and comprehensively promoting rural revitalisation. However, due to the complex geological environment, high incidence of disasters, difficulty of land development, backwardness of supporting facilities, dependence on agriculture and animal husbandry, and other practical difficulties, marginal poverty caused by poor geographical location, potential poverty caused by the fragile ecological environment, and recurrent poverty brought about by short-term social assistance were extremely typical. Therefore, this study analysed the spatial differentiation characteristics of poverty and its driving mechanism by studying 967 poor villages in eastern Tibet as examples.

3.2. Data Source and Processing

This study’s data mainly consisted of two parts: geographic environment and socioeconomics. Geographic environment data included land use, elevation, transportation facilities, educational and medical facilities, urban settlements, and the remote sensing ecological index (RSEI) (e.g., land surface temperature [LST], WET, NDBSI, and normalised difference vegetation index [NDVI]). Among them, land use data were obtained from the Third National Land Resource Survey of Chamdo City, TAR; elevation data were obtained from Geospatial Data Cloud (http://www.gscloud.cn, accessed on 1 March 2023); data for transportation facilities and urban settlements were obtained from the Geographic Information Public Service Platform of TAR (https://xizang.tianditu.gov.cn, accessed on 1 April 2023); RSEI data were obtained from the Remote Sensing Big Data Cloud Computing Service Platform (https://earthengine.google.com, accessed on 22 May 2023); and the data on educational and medical facilities were obtained from the Amap API Open Platform (https://lbs.amap.com, accessed on 26 May 2023). The second part consisted of economic statistics, in which the number of poor people in each village and the data of poor villages came from the 13th Five-Year Plan of Chamdo City for Accurate Poverty Alleviation in Tourism Industry; and that of geological disaster, population of villages, residents’ income, and cultural heritage sites came from the statistical yearbooks of districts and counties in Chamdo City and the data collected from field surveys, which were supplemented by checking the latest master plan and special plan.

3.3. Methods

3.3.1. Spatial Correlation Analysis of Poverty Incidence Based on Bivariate Spatial Autocorrelation

Bivariate spatial autocorrelation analysis reveals the correlation of spatial variables with other variables in neighbouring spaces [36]. In this study, with the help of GeoDa 1.18 software, a spatial weight matrix was established based on distance spatial weighting, and the spatial correlation between poverty incidence and the factors affecting poverty was analysed through bivariate spatial autocorrelation analysis. Characterisation was accomplished using the global Moran Index. The calculation formula is as follows:

$$I=\frac{\sum _{i=1}^n\sum _{j=1}^n{W}_{ij}(x_i-\overline{x})(y_i-\overline{y})}{S^2\sum _{i=1}^n\sum _{j=1}^n{W}_{ij}}$$

(1)

Bivariate local spatial autocorrelation was used to further reflect the clustering and differentiation characteristics of poverty incidence and poverty-causing factor variables in the local region, and local Moran’s I was used for characterisation. The calculation formula is as follows:

$$I_i=z_i\sum _{j=1}^n{W}_{ij}{z}_j$$

(2)

In the formula: I and I_i are bivariate global Moran Index and local Moran index, respectively. Moran’s I takes the value of [−1, 1], Moran’s I > 0 indicates the existence of spatial positive correlation, Moran’s I < 0 indicates the existence of a spatial negative correlation, and Moran’s I = 0 indicates the non-existence of spatial correlation, i.e., random distribution; W_ij is the spatial weight matrix of spatial units i and j; x_i and y_j are the values of the independent and dependent variables in spatial units i and j, respectively; S² is the variance of the variable values in all spatial units; z_i and z_j are the values of the variables in the spatial units standardised by the variance.

3.3.2. Screening for Poverty-Causing Factors Based on GeoDetector and RF

The factor detection model of the GeoDetector can identify the degree of influence of different poverty-causing factors and evaluate the degree of contribution of each factor to the poverty incidence [37]. This method has been used in the relationship between the environment and human beings to test the stratified heterogeneity of each factor and to solve the problem of ambiguity and uncertainty [31]. However, there are analytical results susceptible to factor diversity, data discretisation, and other factors [38]. RF manages high-dimensional data with multiple factors and maintains better accuracy; it discriminates the importance of the input variables through a multi-decision classification tree, and regularisation reduces overfitting, making it better than a single classifier. Therefore, this study combines GeoDetector and regularised random forest (RRF) to screen factor types to improve the precision of poverty-causing factors.

Factor and interaction detectors were mainly used in this study. The interaction detector is used to identify the interaction between different influencing factors, i.e., to assess whether the explanatory power of poverty incidence Y is enhanced or weakened when influencing factors X₁ and X₂ act together or whether the effects of these factors on poverty incidence Y are independent of each other. The factor detector is used to detect the spatial differentiation characteristics of the variables as well as the similarity of the spatial distribution of the independent variable X and the dependent variable Y. In addition, the explanatory power of independent variable X on dependent variable Y was measured through the statistic q. The larger the value of the statistic q, the stronger the explanatory power of the influencing factor on the incidence of poverty [26]. The calculation formula is as follows:

$$q=1-\frac{1}{N{\sigma }^2}\sum _{h=1}^L{N}_h{\sigma }_h^2$$

(3)

where L is the number of strata, i.e., classifications or partitions, of poverty incidence Y or influencing factors X, h = 1, ..., L, N_h and σ_h are the number and variance of layer h and N, and σ are the number and variance of the entire study area.

Based on RF, RRF achieves feature variable selection by adjusting the formula for measuring the importance score of feature variables, thus enhancing the predictive effect of the algorithm [39].

The method applies the Tree Regularisation framework to RF with the aim that a subset of important feature variables can be selected [40]. The formula for calculating the importance score is as follows:

$$GI\left(v\right)=\sum _{c=1}^C{\widehat{p}}_c^v$$

(4)

$$Gain\left(X_i,v\right)=\left\{\begin{array}{c}\lambda ·Gain\left(X_i,v\right) i\notin F\\ Gain\left(X_i,v\right) i\in F\end{array}\right.$$

(5)

$${VIM}_i^{\left(Gini\right)}=\frac{1}{n}\sum _{v\in S_{X_i}}Gain\left(X_i,v\right)$$

(6)

where GI(v) denotes the Gini index at node v, ${\widehat{P}}_c^v$ is the proportion of samples accounted for by category c in node v, and C denotes the number of sample categories; Gain(X_i, v) denotes the Gini information gain of feature variable X_i at split node v, F is the set of features used for node splitting in the previous nodes, and lambda and $\lambda \in \left(0\right.,\left.1\right]$ is the penalisation coefficient; when $\lambda \notin F$, the coefficients of the feature variable X_i at split node v are penalised, and the smaller λ is, the stronger the penalty is, and the stronger the feature variable X_i coefficients are compressed to 0; Sx_i is the combination of all node splits of feature variable X_i in n trees.

3.3.3. Analysis of the Mechanism of Spatial Differentiation of Poverty Based on MGWR

Geo-detector and RF explore linear or nonlinear relationships between poverty incidence and social–ecological system factors and explain the average relationship between poverty incidence and variables across the study area; however, it is difficult to parse the spatial heterogeneity of the drivers of poverty incidence. The Multiscale Geographically Weighted Regression (MGWR) model is a new version of software applications based on Microsoft Windows and MacOS (https://sgsup.asu.edu/sparc/multiscale-gwr, accessed on 8 September 2023), designed for calibrating multiscale geographically weighted regression (GWR) models. It is gradually being more widely used to explore the geographically varying relationships between the dependent variable/response variable and the independent variables/explanatory variables. Incorporating widely used methods for modelling spatial heterogeneity, it relaxes the assumption that all modelled processes operate at the same spatial scale. When considering the degree to which the influence of different independent variables on the dependent variable changes with the spatial scale, the main advantage of the MGWR model is that it not only allows for spatially varying parameter estimation but also generates a unique optimal bandwidth for the relationship between the dependent variable and each independent variable [41].

In this study, the administrative villages in the municipal area are the basic unit, and the study area has a complex topography, and the different independent variables are affected by the spatial scale. Thus, the MGWR model was used to detect spatial heterogeneity affecting rural spatial poverty drivers. It is possible to more precisely analyse and identify the magnitude of the effects of different poverty-causing influences in different geographic locations. The driving factor coefficients β of MGWR are all obtained based on data-differentiated bandwidths, which are better than the fixed bandwidths of classical geographically weighted regression, but still use the quadratic kernel function and AICc criterion of classical GWR [42]. The calculation formula is as follows:

$$y_i={\beta }_o\left(u_i{,v}_i\right)+\sum _{j=1}^k{\beta }_{bwj}\left(u_i,v_i\right)x_{ik}+{\epsilon }_i$$

(7)

where bwj represents the elastic bandwidth of the regression coefficient of the j-th variable; (μ_i,ν_i) are the centre coordinates of countryside i; and β_o(μ_i,ν_i) and ε_i denote the intercept term and error term of the model, respectively.

4. Results

4.1. Characteristics of Spatial Poverty Clustering in Rural Areas of Eastern Tibet

4.1.1. Spatial Poverty Differentiation in Rural Areas of Eastern Tibet

Based on the Natural Breakpoint Method, the poverty incidence in eastern Tibet is divided into five levels: low, lower, middle, high, and higher, as shown in Figure 5. Among them, Paksho, Dayak, Palbar, Gongjo, and Markham had a mean rural poverty incidence of not less than 0.3, indicating that the risk of poverty is generally high in eastern Tibet. As shown in Figure 6, 22.23% of poor villages in eastern Tibet were in high or higher poverty incidence, indicating that the effectiveness of poverty eradication showed great variations within eastern Tibet. Villages with high or higher poverty incidence were located mainly on the steep slopes of the mountain valleys.

Based on the kernel density analysis of the degree of spatial poverty incidence in rural spatial poverty in eastern Tibet, as shown in Figure 7, the poor villages in Dayak and Gongjo had the highest degree of clustering, and these villages were clustered in the Damara−Ningjing mountain zone, whereas the poor villages in the Nujiang river basin had a contiguous cluster distribution, indicating a correlation between poverty risk and geospatial environment.

4.1.2. Spatial Correlates of Rural Poverty in Eastern Tibet

The Kernel method of the GeoDa was used to establish the distance spatial weight matrix, calculate the degree of spatial autocorrelation of rural spatial poverty incidence, and obtain the Moran scatterplot and LISA agglomeration map of the clustering degree of poor villages. Figure 8 shows that there was a significant spatial autocorrelation in the poverty incidence in eastern Tibet, and the neighbouring space had an important effect on the poverty level. There was a bivariate spatial autocorrelation between the poverty incidence and the degree of spatial clustering of poor villages, indicating significant spatial clustering of rural spatial poverty.

From Figure 9a, there are 452 villages with significant spatial autocorrelation of poverty incidence and the areas with ‘High–High’ correlation were mainly in the northwestern part of Paksho and Markham, the central and southern part of Dayak, the southeastern part of Gongjo, and the northeastern part of Palbar, which indicated that the spatial clustering degree of poverty was larger in this area. As shown in Figure 9b, the number of villages with significant spatial correlation between poverty incidence and clustering of poor villages was 558, of which 143 villages with significant ‘High–High’ correlations were mainly located in Dayak and Gongjo, indicating that the higher the poverty level of the villages in the region, the more likely it is to result in the spatial concentration of poverty.

4.2. Influencing Factors of Poverty-Causing Factors in Rural Areas of Eastern Tibet

4.2.1. Identification and Spatial Mechanisms of Poverty-Causing Factors

In this study, the strength of multi-collinearity was analysed using the variance inflation factor (VIF) and tolerance value (TOL) to exclude the poverty-causing factors with strong autocorrelation to improve the accuracy of the model. Theoretically, VIF > 10 or TOL < 0.1 indicates that there exists a multi-collinearity problem in the factors [43]. The results of the multi-collinearity analysis after removing the highly autocorrelated factors are shown in

Table 1. Screening and classification of poverty-causing factors.

Dimension	Impact Factor	Multi-Collinearity Test		Factor Detector		Significance	Moran’s I
		TOL	VIF	q-Value	p-Value
ecological environment	disaster sensitivity index (DSI)	0.495	2.019	0.013	0.039	0.481	0.198
	biodiversity index (BI)	0.172	5.803	0.019	0.003	0.378	0.179
	normalised difference vegetation index (NDVI)	0.568	1.761	0.024	0.000	0.565	−0.170
	land surface temperature index (LST)	0.321	3.118	0.022	0.000	0.628	0.206
geographical location	relief degree of land surface (RDLS)	0.236	4.23	0.041	0.000	0.470	0.225
	construction land slope index (CLSI)	0.384	2.605	0.019	0.016	0.887	0.284
	cropland slope index (CSI)	0.387	2.584	0.051	0.000	0.473	0.217
	construction land elevation index (CLEI)	0.733	1.364	0.019	0.016	0.281	0.110
	the convenience of urban areas (CUA)	0.32	3.128	0.042	0.000	0.692	−0.200
	traffic accessibility (TA)	0.884	1.132	0.017	0.007	0.287	−0.203
social resources	industrial and commercial land resources index (ICLR)	0.773	1.293	0.018	0.010	0.314	−0.057
	cropland per capita index (CPCI)	0.702	1.425	0.037	0.000	0.733	0.223
	livestock resource index (LRI)	0.35	2.859	0.016	0.004	0.589	−0.223
	education service level (ESL)	0.338	2.955	0.044	0.000	0.816	−0.278
	medical service level (MSL)	0.431	2.318	0.024	0.000	0.538	−0.212
	tourism industry resource index (TIRI)	0.495	2.021	0.012	0.033	0.489	−0.112

. We first coupled spatial elements with the help of a GeoDetector factor detection model to identify poverty-causing factors with a stronger degree of explanation (p-value < 0.05), whose relative degree of explanation is expressed as a q-value. Then, RRF was used to rank the importance of the influencing factors of poverty incidence, and the top 20 influencing factors were selected. The RF model fit superiority R² was 75.32%, which was better. Finally, the integrated GeoDetector and RF screened 16 poverty-causing factors, among which the importance coefficients of construction land slope index (CLSI), education service level (ESL), and cropland per capita index (CPCI) were 0.887, 0.816, and 0.733, respectively, which indicated that all three were the key poverty-causing factors of the rural areas in eastern Tibet.

Poverty in rural areas of eastern Tibet was affected by factors such as ecological environment, geographical location, and social resources, showing spatial heterogeneity. As shown by the local Moran’s I index of the bivariate correlation between poverty incidence and poverty-causing factors in Table 1, in terms of the ecological environment, the local Moran’s I index of the disaster sensitivity index (DSI), biodiversity index (BI), and LST were all higher than 0, suggesting that the rural areas of eastern Tibet as a whole showed the ‘High-High’ clustering phenomenon. The local Moran’s I index of NDVI is −0.17, indicating that poverty incidence is lower in areas with high green cover. Therefore, LST, DSI, and BI were the main factors influencing the spatial clustering of potential poverty risk. In terms of geographical location, the local Moran’s I of the construction land slope index (CLSI), construction land elevation index, cropland slope index (CSI), and relief degree of land surface (RDLS) were all above 0, showing positive spatial correlation while showing negative spatial correlation with convenience of urban areas (CUA) and transportation accessibility (TA). Thus, topographical complexity and transportation inaccessibility were the main factors affecting the spatial clustering of marginal poverty risk. In terms of social resources, the local Moran’s I of ESL, the livestock resource index (LRI), medical service level (MSL), the tourism industry resource index (TIRI), and the industrial and commercial land resources index were less than 0, suggesting that deficiencies in these elements contribute to the spatial clustering of recurrent poverty risk. However, the local Moran index of CPCI was 0.223, showing the anomaly that the richer the cropland resources, the more concentrated the poverty risk, indicating that simply increasing the cropland resources could not alleviate the current recurrent poverty and could even increase the possibility of the risk of returning to poverty.

From local Moran’s I, the spatial autocorrelation strength is ranked as follows: CLSI, ESL, RDLS, CPCI, LRI, LST, TA, and DSI. Mainly influenced by the above eight poverty-causing factors, rural villages in eastern Tibet show spatial clustering of different degrees of potential poverty, marginal poverty, and recurrent poverty.

4.2.2. Mechanisms of Compound Effects of Poverty-Causing Factors

Further analysis of the interactions of poverty-causing factors in rural areas of eastern Tibet is conducted with the help of the GeoDetector interactive detector.

As shown in

Table 2. The interaction detector results for the influencing factors of rural poverty characteristics.

Rank	Interaction of Factors	q Statistics	Interaction Results
1	CLSI∩LST	0.1437	Enhance, nonlinear
2	ESL∩CUA	0.1385	Enhance, nonlinear
3	CLSI∩NDVI	0.1333	Enhance, nonlinear
4	CSI∩LST	0.1320	Enhance, nonlinear
5	CLSI∩ESL	0.1282	Enhance, nonlinear
6	CLSI∩MSL	0.1238	Enhance, nonlinear
7	CLSI∩CPCI	0.1208	Enhance, nonlinear
8	CLSI∩CUA	0.1188	Enhance, nonlinear
9	TIRI∩RDLS	0.1188	Enhance, nonlinear
10	CLSI∩BI	0.1187	Enhance, nonlinear
11	CLSI∩CLEI	0.1177	Enhance, nonlinear
12	MSL∩CUA	0.1156	Enhance, nonlinear
13	CLSI∩TIRI	0.1141	Enhance, nonlinear
14	CSI∩ESL	0.1125	Enhance, nonlinear
15	DSI∩ESL	0.1121	Enhance, nonlinear

, the interactions among the poverty-causing factors were significantly enhanced in a nonlinear manner, and there was no independent or weakened relationship, indicating that the interaction of any two factors would enhance the explanatory power of rural spatial poverty differentiation, i.e., the spatial differentiation of poverty in rural areas of eastern Tibet was subjected to the combined effect of multiple factors. Among them, the combined explanatory power of CLSI and several poverty-causing factors were high, further validating that the CLSI was a key factor in the spatial differentiation of rural poverty in eastern Tibet. In addition, the top three interactions were CLSI and LST (0.1437), ESL and CUA (0.1385), and CLSI and NDVI (0.1333). Thus, vegetation growth status, surface thermal conditions, level of educational facilities, and matching topographic conditions had a greater impact on the spatial pattern of rural spatial poverty in eastern Tibet.

Spatial poverty incidence in rural areas of eastern Tibet was affected by a combination of factors. As shown in Figure 10, in terms of ecological environment and geographical location, the interaction between CLSI and NDVI or LST was the most pronounced, reflecting the challenges faced by steep-slope areas between ecological conservation and economic development. For instance, the undertaking of construction and reclamation on more substantial gradients incurred not only significant costs but also engendered soil erosion and a diminution in vegetative cover, exacerbating the ecological fragility of steep-slope areas, making them more sensitive to climate change and anthropogenic disturbances, which increased the risk of potential poverty and marginal poverty in the countryside.

In terms of ecological environment and social resources, the interaction between DSI and ESL was the most pronounced, with geological disaster-sensitive areas causing difficulties in siting facilities, leading to an uneven distribution of educational resources and posing the risk of recurrent and potential poverty. Not only does this limit access to quality education for the local population and exacerbate educational inequalities, but since education is a key pathway to enhancing the ability of individuals to lift themselves out of poverty and to reduce poverty in a sustainable manner, uneven access to education further increases the risk of marginalisation and recurrent poverty in rural areas, creating a negative cycle.

In terms of geographical location and social resources, the interactions between CLSI and CPCI and ESL and MSL were the most obvious. Highlighting the deep-seated contradictions and challenges in rural areas across multiple dimensions such as geographical environment, land resources, social services, and economic development, the risk of persistent poverty in rural regions is exacerbated. Initially, this is manifested in the limitations of land use, where steep slopes restrict the area of arable land, directly impacting the scale and output of agricultural production, thereby weakening the self-sufficiency of the rural economy. Secondly, the reduction in land resources and the challenges posed by terrain intensify the dispersion of public service facilities, leading to an uneven distribution of educational and medical resources in rural areas, increasing the difficulty and cost for residents to access these essential services, and thereby aggravating social inequality and economic disparity. Moreover, the inequitable access to education and healthcare not only affects the quality of life and health standards of the residents but also limits talent cultivation and social mobility, making it difficult for rural areas to break free from the shackles of poverty.

4.2.3. Mechanisms Driving Spatial Differentiation of Poverty-Causing Factors

Based on the above analysis, in the current rural spatial poverty areas of eastern Tibet, the spatial distribution of potential poverty risk was mainly affected by the high frequency of geological disasters and unsuitable LST, the spatial distribution of marginal poverty risk was mainly affected by the large terrain slope and low TA, and the spatial distribution of recurrent poverty risk was mainly affected by the weak level of educational service level and the poor or insufficient quality of industrial resources.

Therefore, nine main indicators influencing the spatial distribution of poverty risk were selected to further analyse the characteristics of the spatial divergence drivers of poverty-causing factors based on the MGWR analysis. As the results in

Table 3. Comparison of the main parameters of the GWR model and the MGWR model.

Projects	GWR Model	MGWR Model
Residual sum of squares	707.389	432.491
AICc	2512.479	2243.426
R2	0.231	0.553

show, MGWR had a better fitting effect compared with the classical GWR model. The MGWR model test showed smaller values for the residual sum of squares, AICc, and larger R2 values for goodness of fit, and multi-scale bandwidth selection adaptation was somewhat elastic and more scientific.

As shown in Figure 11, the driving forces of DSI, LST, CLSI, TA, CPCI, LRI, and TIRI satisfied the p-value < 0.1, whereas ESL and RDLS did not satisfy the significance requirement. Combined with the findings of the MGWR model analysis results in

Table 4. Summary statistics for MGWR parameter estimates.

Variable	Mean	STD	Min	Median	Max	Adj t-val (95%)
intercept	0.025	0.496	−1.218	0.033	1.164	3.315
disaster sensitivity	0.054	0.007	0.046	0.05	0.068	2.005
land surface temperature	0.158	0.086	−0.065	0.199	0.268	2.606
construction land slope	0.135	0.002	0.128	0.134	0.143	2.092
road accessibility	0.009	0.074	−0.071	−0.04	0.147	2.277
cropland per capita	0.138	0.109	−0.038	0.142	0.369	2.671
livestock resource	0.353	0.002	0.348	0.352	0.363	2.037
tourism industry resource	0.049	0.11	−0.138	0.034	0.201	2.358

, the direction of the influencing factors and the magnitude of the role of different driving factors were not identical. Poverty-causing factors were ranked according to the size of the mean value of the regression coefficient: LRI, LST, CPCI, CLSI, DSI, TIRI, and TA, with the mean value of the coefficient of effect greater than 0, and all of them showed a significant positive effect as a whole. However, in terms of standard deviation, the coefficient of force between TIRI and CPCI fluctuated greatly, and the direction of the driving force changed with geospatial changes. In terms of spatial distribution, CLSI and LRI influenced the distribution of rural spatial poverty areas in eastern Tibet, with poor villages driven by LST, DSI, and TA concentrated in the northern high-altitude areas of Beshura Ridge, Nujiang River, and Tenasserim chain. The poor villages driven by CPCI were mainly located in the districts of Paksho, Zogong, Dayak, Gyamda, and Gongjo. The poor villages driven by TIRI were concentrated in the Damara−Ningjing mountain area. It can be seen that elements of ecological environment, geographical location, and social resources were all major driving forces of the spatial differentiation in poverty risk in rural areas of eastern Tibet.

4.3. Zonal Identification and Governance of the Dominant Type in Rural Poverty of Eastern Tibet

Combined with p-value and t-value significance tests, the regression coefficients of each influencing factor were spatially expressed using ArcGIS 10.7 to clarify the dominant roles of different poverty-causing factors in the local geospatial space and to identify the spatial agglomerations of the three major types of villages in terms of potential, marginal, and recurrent poverty, to put forward the differentiated poverty management strategy in a zoned and targeted manner.

4.3.1. Potential Poverty-Dominated Villages

As shown in Figure 12, the risk of potential poverty in rural spatial poverty in eastern Tibet was significantly influenced by DSI and LST and was based on the spatial differentiation characteristics of the driving factors, the potential poverty villages were categorised into DSI-driven, LST-driven, and LST–DSI co-driven types. Among them, a total of 73 villages with poverty risk aggravated by the frequent occurrence of geological disasters, which were mainly concentrated in the distribution of Palbar, Tengchen, and Lhorong, with the coefficient of poverty-causing risk showing a gradually increasing trend from the southeast to the northwest. A total of 469 villages affected by LST that increased the risk of poverty were mainly located in the eastern Damara−Ningjing mountain and northwest region, which were closely related to natural and human factors such as the growth of regional plants, the water cycle in the region, and the urbanisation process [44]. A total of 145 villages were affected by the combination of DSI and LST, mainly concentrated in the Bianba and Dinh Thanh districts. Hence, for villages dominated by potential poverty risk, on the one hand, when conducting the layout and construction of village settlements, it is necessary to identify areas with a high risk of geological disasters or areas with heat unsuitable for production and life, thereby constructing a new pattern for the safe and coordinated development of production, ecology, and life. In contrast, it is necessary to conduct work such as planting trees and combating soil erosion, to reduce the potential risk of poverty brought about by geological disasters and the unsuitable land surface temperature. Furthermore, for potential poverty-stricken villages, a government-led ‘county-town-village’ normalisation control model should be established. The local government should establish a disaster and ecological early warning and prevention department, which undertakes the main tasks of monitoring and prevention, emergency management, and guiding and encouraging social forces. Through emergency training, transfer payments, and transfer resettlement, a rapid relief mechanism should be constructed to prepare for risk prevention in advance and to build a strong defence against poverty caused by disasters.

4.3.2. Marginal Poverty-Dominated Villages

As shown in Figure 13, the risk of marginal poverty of poor villages in eastern Tibet was significantly influenced by the CLSI and TA and was based on the spatial differentiation characteristics of the driving factors; the marginal poverty villages were classified into CLSI-driven and CLSI−TA co-driven types. Poor villages in eastern Tibet were affected by the CLSI, and the coefficient of its role in poverty-causing risk showed a gradually increasing trend from east to west. Poor villages in Tengchen, Palbar, and Rioche, located on sloping terrain or at high altitudes, with difficult terrain and transportation, can hardly support the development of industries and the construction of supporting facilities, which in turn exacerbates the risk of poverty in villages in the region. Therefore, for poor villages of the CLSI-driven type, the efficient allocation of spatial resources is promoted through the implementation of relocation strategies for settlements with large and scattered CLSI. For poor villages of the CLSI–TA co-driven type, on the one hand, the layout of public transportation facilities linking existing villages and towns should be optimised following the conditions of the topographic slope. On the other hand, settlements with complex topographical conditions and poor transportation will be relocated to areas around towns and central villages that are easily accessible, alleviating the risk of marginal poverty brought about by such factors as complex topographical conditions, dispersed settlements, and closed internal and external transportation. Furthermore, for villages characterised by marginal poverty, it is essential to develop location-specific policies [45]. In areas where there is a concentration of population and economic activities or where population growth is observed, the focus should be on enhancing the supply and pre-emptive planning of land use and related services. For underdeveloped regions or rural areas with sparse or declining populations, the emphasis should be on ensuring basic public services, providing employment assistance, and facilitating income growth. These measures should be aligned with the demographic trends of the countryside to promote village development and minimize the waste of investment during the rural revitalisation process.

4.3.3. Recurrent Poverty-Dominated Villages

As shown in Figure 14, the risk of recurrent poverty in rural areas in eastern Tibet was significantly affected by the CPCI, LRI, and TIRI. According to the spatial differentiation characteristics of the driving factors, the recurrent poverty villages were classified as LRI-driven, agricultural and livestock resources index (ALRI)-driven, ALRI–TIRI co-driven, and LRI–TIRI co-driven types. The quantitative increase in resources such as arable land, animal husbandry, and tourism in eastern Tibet in the new period will lead to over-exploitation of resources and cause problems of ecological and environmental damage. At the same time, most villages depended on external support and lacked endogenous motivation, thus exacerbating the risk of recurrent poverty in rural areas. Thus, for poor villages of the LRI-driven type, it is necessary to conduct ecological restoration of degraded grasslands with the help of artificial intervention measures, to improve the productivity of the stock of grasslands and at the same time rationally plan the pastoral industrial parks, relying on the natural grassland resources and sustainably enhancing the production efficiency of the highland pastoral industry. Regarding poor villages of the ALRI-driven type, mainly in Paksho, Gongjo, and Dayak, respectively, following the conditions of topography and water resources, optimising the industrial structure of arable land and animal husbandry in the rural areas, and implementing a rational layout of highland pastoralism and river-valley agriculture is suggested. For poor villages of the ALRI–TIRI co-driven type, promoting the integrated development of the animal husbandry and tourism industries and constructing a two-way industrial pattern of ‘village-town-county’ multi-layer linkage is suggested. Concerning the poor villages of the LRI–TIRI co-driven type, mainly concentrated in Gongjo, proposing the strategy of taking agricultural and animal husbandry development as the root and promoting the development of the tourism industry according to local conditions, and implementing the cultural and tourism industry layout characterised by plateau agriculture and animal husbandry and Khamba culture, fostering the new kinetic energy of the rural industry, is suggested. Furthermore, for villages experiencing recurrent poverty, policies should be designed to facilitate an orderly and gradual exit, while coordinating with other policies to fill any gaps, thereby reducing or preventing regional poverty resurgence due to policy withdrawal. For instance, initially, by fostering collaboration between banks and insurance companies, a variety of agricultural insurance types should be introduced to provide basic development security for farmers. Subsequently, the state should formulate or optimize talent supply channels for impoverished areas, enabling the empowerment of rural revitalisation through diversified industrial development. Lastly, a robust evaluation mechanism should be established during the poverty reduction phase, with government departments at all levels summarizing experiences in poverty alleviation and governance. A normalised top-down and bottom-up information sharing channel should be established to facilitate the summarization of effective countermeasures and to modify or adjust ineffective response strategies.

5. Discussion

This study identified the spatial distribution pattern of poverty risk and analysed the driving mechanism of spatial differentiation for 967 poor villages in eastern Tibet. The results of this study show that rural spatial poverty in eastern Tibet is driven by the DSI, LST, CLSI, TA, LRI, CPCI, and TIRI, and faces potential, marginal, and recurrent poverty risks with significant spatial differentiation. This is consistent with the conclusion of the empirical study by Pan Jinghu et al. that deep poverty in eastern Tibet belongs to the type of terrain constraints and regional transportation constraints [21]. At the same time, it also proves that relative poverty in China’s Tibet region faces the real dilemmas of high vulnerability under geospatial constraints, high risk of poverty, and insufficient endogenous motivation [46]. Additionally, the dynamics of the social–ecological system behind land use change affect the spatial distribution of rural poverty. Specifically, in areas with complex terrain, the changes in the ecosystem are minimal, and the conversion of farmland back to forest shows a significant positive impact. However, for villages on the edge of cities, the negative impact of converting forest and grassland to arable land is evident. Urban expansion encroaches on forest and farmland, causing farmers to lose their land or be forced to change the way land is used. Moreover, different landscape pattern indices have a significant heterogeneity in their impact on ecological carrying capacity [47]. This also proves that the impact of social–ecological systems on poverty risk presents significant spatial heterogeneity. It is very necessary to explore the patterns and driving mechanisms of the spatial distribution of poverty risk.

Located on the Tibetan Plateau, which is the highest elevation in the world, the average annual temperature in eastern Tibet is increasing at a rate of 0.026 °C per year, which exceeds the rate of global temperature increase, and the unique and stable micro-environment is resulting in a significant warming of the soil, which has exacerbated the soil and environmental degradation in ecologically fragile areas of eastern Tibet [48], thus affecting the incomes and livelihoods of the farming households. In addition, eastern Tibet is in the transition zone between the first and second steps of the Chinese ladder and is prone to geological disasters [49]. Increasing surface soil temperatures will lead to an increased risk of geological disasters caused by freezing and thawing, and the geological disasters are mostly concentrated in the valley areas where human activities such as cities, towns, and roads are relatively intensive, which will easily result in significant property losses and casualties when geological disasters occur [50]. Thus, LST and the DSI as risk factors of geo-potentiality in eastern Tibet constrain the growth of vegetation, the development of agriculture and animal husbandry, and the construction of human habitat in poor areas [51,52], and are prone to triggering potential poverty and the risk of returning to poverty or causing poverty suddenly.

A total of 70% of China’s poverty areas are located in regions with average ground slopes of 10° or more, and the complex topographic location has a positive driving effect of considerable strength on the spatial distribution of poverty [53]. Eastern Tibet is located along the Hengduan Mountain Range, where the relative topographic conditions are harsh, and the available construction land and arable land show the characteristics of a small area and dispersed layout. Meanwhile, the risk of marginal poverty among indigenous villagers rises as people who have been lifted out of poverty spontaneously move to cities or areas with favourable geographical locations. In addition, many studies on poor areas in central and western China have concluded that transportation factors in poor areas have a larger and more significant role in promoting local economic growth [54,55,56], and in rural areas where urban and rural transportation is inconvenient and the land for construction is relatively gentle, strengthening road development and construction can promote urban and rural factor mobility and alleviate the risk of marginalised poverty in poor geographical locations.

Cultivating endogenous motivation in villages by effectively integrating industrial resources to form a blood-forming approach to alleviate poverty is a scientific means of governing the risk of recurrent poverty in villages. Research has shown that livestock resources, cropland resources, and tourism industry resources are key factors affecting recurrent poverty in rural areas of eastern Tibet and that, under the influence of geographical location and ecological environment, the quantitative abundance of cropland resources cannot sustainably alleviate rural poverty, whereas the abundance of tourism resources without efficiently activating them exacerbates the risk of recurrent poverty [57]. Established studies have also confirmed that poor quality and location of cropland can constrain the sustainable development of rural economies [58,59]. Furthermore, rural tourism has become a major alternative livelihood for farmers in the Tibetan Plateau [57], and the development of tourism by tapping into the resources of impoverished areas has a significant role to play in boosting the incomes of impoverished households. However, the effect of poverty reduction in rural tourism differs greatly across different traffic locations. Due to the weak transportation accessibility of villages with complex topographic locations and the lagging construction of transportation facilities in grass-roots villages, not only can they fail to take advantage of the tourism resources in eastern Tibet, but they can also severely constrain the enhancement of the reception capacity of rural tourism, which leads to fewer opportunities and benefits for participation in tourism development [60]. Hence, the mechanisms of rural poverty risk under different locational conditions are diverse, and the factors of transportation location lead to regional heterogeneity in the effectiveness of tourism in reducing poverty [61].

From the perspective of global poverty areas, particularly those in mountainous and watershed areas of Asia and Africa, they are significantly impacted by socio–ecological system factors. These include the complexity of the terrain, scarcity of resources, and inadequate infrastructure, which contribute to the risk of marginal, recurrent, and potential poverty, as well as spatial differentiation. For instance, in Nepal, the current trend of population migration from mountainous regions to urban areas has left behind a trapped poor population, leading to a spatial concentration of poverty and the emergence of marginal poverty [62]. In African nations such as Nigeria, Tanzania, Ethiopia, and the Democratic Republic of Congo, there is an imbalance in regional energy distribution due to energy scarcity and an unequal spatial and temporal allocation of electricity usage. This has triggered multidimensional poverty among marginalised communities [63]. The Purulia region, located in the arid eastern part of India, is at risk of latent or recurrent poverty. This risk is due to underdeveloped social, economic, and infrastructural facilities, along with a limited water supply, which hampers the region’s capacity for social development [64]. In the Nile Basin countries, poverty and inequality are widespread. The spatial and temporal patterns of GPI (Growth-Poverty-Inequality) evolution differ based on geographic positions upstream and downstream, with high inequality levels further aggravating poverty conditions [65]. Thus, globally typical impoverished areas are constrained by factors such as topographical location, infrastructure, and ecological resources and also face the three major risks of poverty: potentiality, recurrence, and marginality. This scenario bears resemblance to the spatial differentiation of rural poverty and its driving factors in Eastern Tibet, where, on one hand, there are communities with complex terrain and limited transportation, exhibiting a trend of migration towards towns with better accessibility or their surrounding areas, thereby revealing risks of marginal poverty. On the other hand, these areas are also confronted with inadequate infrastructure provision or a deficiency in ecological resources, which indicates risks of recurrent or potential poverty.

From the perspective of rural development in China, the evolution of Chinese rural areas has transformed from a top-down, policy-driven system before the digital era to a system in the digital era that is driven by both top-down and bottom-up approaches, with technology and policy working together [66]. Currently, the development of rural areas in China in the digital age is facing the challenge of the “digital divide”, especially in remote areas where insufficient infrastructure and high service costs lead to poverty risks that are diverse and spatially heterogeneous. Utilising diverse smart analysis technologies to explore the spatial differentiation characteristics of rural poverty risks, identifying the three main types of poverty risks—”potential, recurrent, and marginal”—and proposing targeted measures and policy recommendations is needed for the shift from post-event management to pre-event prevention of rural poverty in China. In addition, digitalization also provides opportunities to alleviate relative poverty. Through policy drive and technological innovation, it is possible to achieve urbanisation in place in rural areas and restructure the socioeconomic structure, promote the construction and application of information and communication technology infrastructure, and narrow the urban–rural digital gap. This can effectively alleviate the current complex and diverse poverty risks, especially those related to marginal poverty risks.

In the global context of the multi-dimensional poverty risk coexistence and prominent relative poverty, our study took a social–ecological system perspective, based on traditional data collection, compensating for the lack of geographic data by using Python, GEE, and other multi-source data acquisition methods, comprehensively utilising RF and geo-detectors to jointly identify the main poverty-causing factors, analysing the coupling-enhanced effects of different factors on poverty formation, and systematically analysing the spatial differentiation driving mechanisms of recurrent poverty, marginal poverty, and potential poverty with the help of MGWR. In this study, we have constructed a methodology to analyse the mechanism of spatial poverty differentiation in macro-regions using villages as data units and proposed targeted poverty governance strategies. However, due to the lack of grass-roots data in villages, we failed to further analyse the evolution of spatial poverty differentiation rules from a dynamic perspective, which should be the direction of our efforts in the future.

6. Conclusions

This study adopted a social–ecological system perspective and used spatial analysis, RF, GeoDetector models, and MGWR to analyse the characteristics of the spatial distribution of rural poverty and the driving mechanism of spatial differentiation in eastern Tibet, proposing differentiated governance strategies for the three typical distribution areas of potential, marginal, and recurrent poverty risk. The main conclusions were as follows: (1) the rural areas in eastern Tibet showed a spatial distribution of poverty in an agglomerative manner, with villages with a high poverty incidence distributed in the steep slopes of mountain valleys, and about 37% of the villages showed a spatial distribution characterised by a high degree of agglomeration of high poverty incidence, and the higher the degree of poverty, the easier it is to cause spatial concentration of poverty, with the most prominent spatial agglomeration characteristics of rural spatial poverty in the areas of Dayak and Gongjo. (2) Topographic location and the level of public facilities were key factors influencing the poverty level of rural areas in eastern Tibet. Different factors showed a coupling-enhanced effect on poverty incidence, and the combined explanatory power of the CLSI and multiple poverty-causing factors was high. The top three combinations of factors with coupling-enhanced effects were CLSI and LST, ESL and CUA, and CLSI and NDVI. (3) We identified village poverty-risk partitions driven by different factors by coupling the driving forces of spatial differentiation and proposing differentiated governance strategies. For example, coordinating the development of production–life–ecological spatial security to reduce the risk of potential poverty through the demolition and removal of mergers promoting efficient allocation of spatial resources and to alleviate the risk of marginal poverty, and, through the rational layout of plateau animal husbandry, river-valley agriculture, and the cultural tourism industry of Khamba, enhance the endogenous motivation of the rural industry and avoid the risk of recurrent poverty. The above strategies can provide references for accurately addressing the current risk of the relative poverty of villages in eastern Tibet.

In response to the global reality of diverse regional types, multifaceted risks of poverty recurrence, and the pursuit of high-quality governance objectives, this study leverages spatial poverty theory to analyse the mechanisms behind the formation of potential, recurrent, and marginal poverty risks from a socio–ecological systems perspective. Employing advanced multi-intelligence analysis techniques, it discerns the spatial differentiation and driving factors of these three major types of poverty risks. The research proposes targeted governance measures and policy recommendations, thereby enabling the precise implementation of rural poverty risk management strategies and the transformation of planning techniques. This has significant practical implications for advancing poverty alleviation in vulnerable and less-resilient rural areas globally. However, current research on the spatiotemporal dynamics of rural poverty risk necessitates further in-depth investigation. The significant challenges in acquiring regional data have impeded systematic longitudinal studies on the evolution of poverty risk within villages. Moving forward, we aim to leverage predictive methods such as the random forest algorithm and neural network models to address the issue of data scarcity. Additionally, we will employ geographically and temporally weighted regression to forecast the future trajectory of rural poverty risk and to devise effective long-term strategic planning.