Drivers of Population Dynamics in High-Altitude Counties of Sichuan Province, China

Abstract

The population dynamics of high-altitude mountainous areas are shaped by a complex interplay of socioeconomic and environmental drivers. Despite their significance, such regions have received limited scholarly attention. This research identifies and examines the principal determinants of population changes in the high-altitude mountainous zones of Sichuan Province, China. Utilizing a robust quantitative framework, we introduce the Sustainable Population Migration Index (SPMI) to systematically analyze the migration potential over two decades. The findings indicate healthcare accessibility as the most significant determinant influencing resident and rural population changes, while economic factors notably impact urban populations. The SPMI reveals a pronounced deterioration in migration attractiveness, decreasing by 0.27 units on average from 2010 to 2020. Furthermore, a fixed-effects panel regression confirmed the predictive capability of SPMI regarding population trends, emphasizing its value for demographic forecasting. We also develop a Digital Twin-based Simulation and Decision-support Platform (DTSDP) to visualize policy impacts effectively. Scenario simulations suggest that targeted enhancements in healthcare and infrastructure could significantly alleviate demographic pressures. This research contributes critical insights for sustainable regional development strategies and provides an effective tool for informed policymaking.

1. Introduction

Following the initiation of reform and opening-up policies in the 1980s, rapid economic development and urban expansion have contributed to growing regional disparities in both population distribution and economic outcomes. These imbalances are primarily shaped by heterogeneous natural and geographical environments. These imbalances have further intensified the complex interdependence between demographic patterns and economic development, resulting in notable shifts in population structure and spatial distribution [1,2]. The escalation of intercity mobility has been largely fueled by rising incomes, infrastructure improvements, and the expansion of the consumer economy, alongside increasing convenience and changing consumption behaviors. Beyond traditional economic disparities across cities, population mobility is increasingly shaped by the distribution of educational opportunities, the quality of public services, and the overall quality of life. This trend indicates an evolution toward more complex and multidimensional determinants of migration [3,4].

Contemporary studies on population change have examined multiple aspects of population mobility, such as spatial distribution patterns [5,6], transformation modes [7,8], underlying drivers [9,10], and associated effects. Globally, growing scholarly interest has been directed toward the internalization and regionalization trends of demographic shifts [11]. In the Chinese context—marked by a large-scale migrant population—research efforts have primarily concentrated on interprovincial migration and the dynamics of economically advanced urban agglomerations [12]. According to the Seventh National Population Census, the new urbanization strategy, implemented since 2010, has spurred increased population movement, particularly with a growing share of interprovincial migration. However, uncertainties persist concerning the key driving forces, varying migration trends, and the underlying social dynamics [13].

Existing research has predominantly examined well-established urban regions, including the Yangtze River Delta, Pearl River Delta, and the Beijing-Tianjin-Hebei area, with less emphasis on the patterns of interprovincial migration in high-altitude zones and urban clusters in central and western China [14]. Supported by national strategies and local governmental initiatives, the attractiveness of central-western urban clusters to migrants has been gradually rising. To better understand regional demographic dynamics, future studies should investigate these diverse geographical settings—particularly those with a complex topography—to provide nuanced insights that can inform region-specific development policies [15].

Population change is commonly explained through the push-pull theory, where factors pushing people from their place of origin and those attracting them to a destination generate migration patterns. Migration motivation research is commonly divided into macro- and micro-level perspectives. Macro-level analyses focus on the interplay between population dynamics and regional economic development, incorporating factors such as spatial proximity, economic growth, and policy frameworks [16]. While these studies emphasize urban and lowland migration drivers, they overlook geographic constraints specific to mountainous contexts [14,15,16]. For instance, Liu et al. demonstrated that altitude-induced healthcare inaccessibility reduces population retention by 23% in Tibetan counties [17], while Liu and Li quantified a 40% higher infrastructure vulnerability in slopes > 25° [18]. Such terrain-dependent barriers fundamentally reshape migration mechanics—whereas lowland models prioritize economic gradients [19], which are central to our case. In contrast, micro-level studies center on individual attributes and decision-making rationales, exploring elements such as access to public services, infrastructure quality, family considerations, and life course stages [20,21].

Existing research on population dynamics has frequently integrated both qualitative and quantitative methodologies. Qualitative approaches—such as interviews, surveys, and case studies—draw upon established theoretical frameworks to explore change motivations at both individual and collective levels. Quantitative methods, including statistical analysis, spatial panel models, and econometric techniques, often harness big data to identify and measure key influencing factors. For example, Kraft examined demographic trajectories using mobile communication records and transportation network data [22], whereas Zhang utilized location-based big data to analyze population mobility patterns across 328 cities [23]. Liu applied spatial analysis and the geodetector method to investigate the drivers and spatial characteristics of population decline in mountainous areas. Moreover, mixed-methods research, which combines qualitative depth with quantitative rigor, provides a more comprehensive understanding of the complex mechanisms driving population movements [24].

This research investigates the high-altitude mountainous areas of Sichuan Province to identify the primary factors influencing population change, aiming to fill existing gaps in the scholarly literature. While the classical push–pull framework provides a foundational lens for understanding migration, its application in high-altitude mountainous contexts requires contextual adaptation. In such regions, environmental constraints—such as elevation-induced infrastructure costs, service delivery bottlenecks, and occupational rigidities in the primary sector—amplify both push pressures and barriers to in-migration. As a result, population decisions are shaped not only by conventional economic gradients but also by region-specific physical and institutional vulnerabilities. Recent empirical studies in China have underscored that mountainous zones exhibit structurally distinct migration drivers, where the loss of young labor is due more to deficits in healthcare access and educational continuity than to income alone. Therefore, this study reconceptualizes the push–pull logic to accommodate the dual role of terrain-dependent infrastructural disadvantage and policy-sensitive public service differentials. While prior research has considered the effects of economic, infrastructural, and healthcare factors, the unique geographical and socioeconomic characteristics of high-altitude areas remain underexplored. Pronounced elevation gradients and rugged topography—especially in areas exceeding 2000 m—introduce significant complexities into analyzing demographic dynamics. To uncover the mechanisms driving population shifts in these regions, this study applies the geodetector model based on 14 indicators related to the economy, infrastructure, and healthcare, derived from the Sichuan Statistical Yearbook. In addition, a novel Sustainable Population Migration Index (SPMI) is proposed to quantify migration potential by integrating attractive and push-pressure factors. A Digital Twin-based Simulation and Decision-support Platform (DTSDP) is also developed to visualize and simulate the spatial impacts of policy interventions. Together, these methods establish a comprehensive analytical framework for evaluating demographic changes in high-altitude regions and provide empirical insights for regional development and policy design.

2. Materials and Methods

2.1. Study Area

This study focuses on 42 non-key high-altitude counties in Sichuan Province, China, each with an average elevation above 2000 m (Figure 1). The region includes sub-high and high-altitude zones. Sichuan features diverse geomorphological regions—such as the Qinghai-Tibet Plateau, Hengduan Mountains, and Sichuan Basin—with terrain descending from west to east. The average elevation is 2598 m, ranging from Mount Gongga’s 7526-m peak to a low of 188 m in Yulin Town. These varied geographical conditions provide the essential context for the analysis.

2.2. Indicator System and Data Source

2.2.1. Dependent Variables: Resident Population, Urban Population, Rural Population

This study examines three distinct population categories: resident, urban, and rural populations. According to the Seventh National Population Census, the resident population encompasses individuals residing in a given township or sub-district, regardless of local hukou registration, including those temporarily abroad or absent from their registered location for more than six months. It reflects the actual number of people present and is key to assessing population distribution, labor supply, and resource allocation. According to national statistical standards, the urban population refers to individuals residing in cities or towns, while the rural population includes those living outside urban boundaries. This classification is based on residence and occupation. Urban population data help gauge urbanization, economic growth, and social change, whereas rural population figures are essential for rural planning, agricultural policy, and public service delivery. Together, these three metrics offer a comprehensive basis for evaluating demographic structure and guiding development policy.

2.2.2. Independent Variables: Economy, Infrastructure, Healthcare

As shown in

Table 1. Variable definitions and data sources. (Data frequency: annual).

Indicator Type	Indicator	Units	Data Sources	Variable ID
Economy	Gross Regional Domestic Product	Ten thousand RMB	Provincial, city and county statistical yearbooks	E1
	Primary Industry Production Value	Ten thousand RMB	Provincial, city and county statistical yearbooks	E2
	Secondary Industry Production Value	Ten thousand RMB	Provincial, city and county statistical yearbooks	E3
	Tertiary Industry Production Value	Ten thousand RMB	Provincial, city and county statistical yearbooks	E4
	Local Public Finance Revenue	Ten thousand RMB	Provincial, city and county statistical yearbooks	E5
	Total Retail Sales of Consumer Goods	Ten thousand RMB	Provincial, city and county statistical yearbooks	E6
	Urban Residents Average Disposable Income	RMB	Provincial, city and county statistical yearbooks	E7
	Farmers and Herdsmen Average Disposable Income	RMB	Provincial, city and county statistical yearbooks	E8
Infrastructure	Investment in Fixed Assets of the Whole Society	Ten thousand RMB	Provincial, city and county statistical yearbooks	I1
	Number of Primary Schools	Number of schools	Provincial, city and county statistical yearbooks	I2
	Number of General Secondary Schools	Number of schools	Provincial, city and county statistical yearbooks	I3
Healthcare	Number of Hospitals and Health Centers	Number of Hospitals and Health Centers	Sichuan Provincial Health and Wellness Statistical Yearbook	H1
	Number of Beds in Hospitals and Health Centers	Number of Beds	Sichuan Provincial Health and Wellness Statistical Yearbook	H2
	Number of Health Personnel in Health Facilities	Person	Sichuan Provincial Health and Wellness Statistical Yearbook	H3

Note: This table presents descriptive indicator definitions only. No inferential statistics or significance testing were applied. Variable IDs (e.g., E1–E8, I1–I3, H1–H3) are assigned based on the initial letter of the indicator type (Economy, Infrastructure, Healthcare) followed by a numeric sequence for clarity and consistency with prior studies.

, this study evaluates how the economy, infrastructure, and healthcare influence population dynamics. Based on a literature review, 14 key indicators were identified across these dimensions, reflecting the multifactorial drivers of demographic change in high-altitude mountainous regions. To ensure study integrity and isolate intrinsic drivers, we selected non-key counties. Al Abbasi et al. showed targeted poverty alleviation investments in “key counties” artificially inflate healthcare (e.g., hospital beds, doctor-patient ratio) and economic indicators (e.g., GDP per capita) by 18–35% based on 2015–2020 data, which could distort our analysis [25]. Similarly, research focusing on the healthcare workforce in Western China by Duan et al. quantified a 22% overestimation in health personnel density attributable to temporary poverty alleviation subsidies in supported counties [26]. Furthermore, analyses of economic data, such as those by Kang and Li examining counties in Yunnan province, reported GDP growth rates in neighboring key poverty counties that were inflated by approximately 18–23% due to concentrated policy interventions, potentially masking underlying economic fundamentals [27]. Including these designated counties would thus introduce substantial measurement bias and confound our analysis of intrinsic drivers.

To ensure a comprehensive and systematic selection of independent variables, this study adopted a two-step screening process. Initially, potential determinants of population change were identified based on a thorough review of the relevant literature [28,29,30,31] and the official evaluation framework for urban development in China [31]. Subsequently, the variable set was refined using criteria of scientific robustness and data availability, resulting in 14 measurable and policy-relevant indicators. Guided by the classical push–pull migration framework [32], this study classified the 14 explanatory indicators into pull factors and push factors. Pull factors denote favorable conditions that augment a region’s capacity to attract and retain residents [33], such as high-value-added industrial output, comprehensive public services, and high-quality healthcare resources. Push factors, in contrast, capture adverse conditions that erode population-retention capacity and propel labor out-migration [34]. Although the output value of the primary sector registers local economic activity, it is deemed a push factor for high-altitude counties. Heavy reliance on primary production typically signals limited economic diversification and short, fragile value chains that are highly vulnerable to market volatility and natural shocks. Such structural vulnerability constrains the availability of diverse, high-quality employment opportunities, diminishes the region’s appeal to young labor, and ultimately generates net out-migration pressure in interregional competition. These indicators were classified into three categories: economic factors, including gross regional domestic product, output values of the primary, secondary, and tertiary industries, local public finance revenue, total retail sales of consumer goods, and average disposable incomes of both urban residents and rural populations (farmers and herdsmen); infrastructure factors, comprising total investment in fixed assets, number of primary schools and general secondary schools; and healthcare factors, including the number of hospitals and health centers, hospital beds, and health personnel. While statistical yearbooks provide standardized, county-level data, we acknowledged potential underreporting in remote areas. As mitigation, we cross-validated demographic trends with census data and provincial statistical bulletins where available. Furthermore, qualitative drivers such as cultural attachment and place identity, though beyond the scope of this study’s quantitative framework, were recognized as significant and warrant integration in future mixed-methods research. Furthermore, we recognize the potential endogeneity concerns associated with urban residents’ average disposable income (E7) and farmers’ and herdsmen’s average disposable income (E8) [35]. This primarily stems from the well-established bidirectional relationship between income levels and migration patterns: higher income can attract migrants, while migration inflows (or outflows) can subsequently impact local income levels, as theorized by Todaro and supported by subsequent empirical work [36]. The Geographical Detector method (specifically the factor detector) [37,38], which confirmed that the explanatory power (q-statistic) of E7 and E8 did not exhibit significant overlap or dependence with other key drivers in the model. The results of these independence tests assured us that E7 and E8 could be meaningfully isolated and removed for robustness checks without fundamentally altering the structure of the remaining explanatory framework (

Table A2. The q-values of economic, infrastructure, and healthcare factor tests.

Period	q Value	E1	E2	E3	E4	E5	E6	E7	E8	I1	I2	I3	H1	H2	H3
10h	q_full	0.544	0.789	0.350	0.519	0.502	0.554	0.032	0.322	0.160	0.460	0.544	0.779	0.611	0.601
	q_noE	-	-	-	-	-	-	-	-	0.160	0.460	0.544	0.779	0.611	0.601
11h	q_full	0.617	0.819	0.477	0.614	0.480	0.316	0.061	0.283	0.051	0.497	0.686	0.612	0.813	0.759
	q_noE	-	-	-	-	-	-	-	-	0.051	0.497	0.686	0.612	0.813	0.759
12h	q_full	0.661	0.880	0.568	0.565	0.530	0.688	0.071	0.023	0.119	0.592	0.738	0.481	0.774	0.459
	q_noH	0.661	0.880	0.568	0.565	0.530	0.688	0.071	0.023	0.119	0.592	0.738	-	-	-
13h	q_full	0.619	0.805	0.524	0.488	0.540	0.675	0.156	0.082	0.338	0.692	0.695	0.827	0.900	0.863
	q_noH	0.619	0.805	0.524	0.488	0.540	0.675	0.156	0.082	0.338	0.692	0.695	-	-	-
14h	q_full	0.696	0.742	0.518	0.736	0.387	0.616	0.401	0.414	0.306	0.771	0.698	0.719	0.937	0.869
	q_noH	0.696	0.742	0.518	0.736	0.387	0.616	0.401	0.414	0.306	0.771	0.698	-	-	-

). These steps collectively ensured that any potential inflation of explanatory power caused by income-migration feedback loops was effectively addressed.

2.2.3. Temporal Harmonization of the Observation Framework

In China, macro-level statistical data are typically organized and published according to Five-Year Plan (FYP) cycles. Accordingly, this study aligned all annual observations with the relevant FYP periods: the 10th FYP (2001–2005), 11th FYP (2006–2010), 12th FYP (2011–2015), 13th FYP (2016–2020), and the currently ongoing 14th FYP. As of this study, only data from 2021 to 2022 are available for the 14th FYP, meaning that this window covers two years rather than a full five-year cycle. To ensure data continuity and comparability across counties and years, a small number of missing values—mainly due to rare omissions in individual yearbooks—were filled using mean values within the respective Five-Year Plan (FYP) windows. Notably, the proportion of missing values is extremely low, accounting for less than 1% of the total dataset. Given the window-based aggregation, this imputation had minimal impact on the robustness of the analysis.

2.3. Methods

2.3.1. Geodetector Method

Spatial autocorrelation is a statistical technique used to assess the spatial distribution patterns and relationships within geographic data [39]. It is based on the premise that spatially adjacent or nearby locations may exhibit similar or correlated values and that this similarity tends to diminish or disappear as the distance between locations increases [40]. Spatial autocorrelation is generally categorized into two types: global and local.

Global spatial autocorrelation provides an overall evaluation of spatial patterns within the entire study area. It helps determine whether the spatial data tend to cluster or disperse, as well as the intensity and statistical significance of such patterns [41]. Among the various global measures, Moran’s I is the most widely applied index [42], and its formula is as follows:

$$I={\displaystyle \frac{{\sum }_{i=1}^n{\sum }_{j=1}^n{W}_{ij\left(X_i-\overline{X}\right)\left(X_J-\overline{X}\right)}}{S^2{\sum }_{i=1}^n{\sum }_{j=1}^n{W}_{ij}}}.$$

(1)

In this equation, n denotes the total number of spatial units, while $\overline{X}$ represents the mean value of all observations. X_i and X_j refer to the observed values for spatial units i and j, respectively. W_ij corresponds to the spatial weight matrix, which characterizes the spatial relationship between units i and j. A Moran’s I value greater than 0 signifies positive spatial autocorrelation, meaning nearby areas tend to show similar values—either high-high or low-low clustering. A value less than 0 suggests negative spatial autocorrelation, where high values are surrounded by low values or vice versa. When Moran’s I approaches 0, it implies a random spatial pattern with no discernible correlation.

In contrast, local spatial autocorrelation emphasizes the spatial characteristics of individual units within the study area. It assesses the degree and statistical significance of spatial heterogeneity by examining the relationship between each unit and its immediate neighbors. The most commonly employed metric for this purpose is the Local Moran’s I index [40], which is defined by the following formula:

$$I_i={\displaystyle \frac{\left(X_i-\overline{X}\right)}{S^2}}{\sum }_j{W}_{ij}\left(X_j-\overline{X}\right)$$

(2)

It serves as a tool for detecting various spatial clustering or dispersion patterns within a study area, including clusters of high values, clusters of low values, high-value outliers surrounded by low values, and low-value outliers surrounded by high values.

Spatial autocorrelation has been extensively applied to investigate spatial distribution patterns across various domains, including road traffic [43] and disease incidence [44]. For example, Zhu Mingfei [45] employed spatial autocorrelation analysis to reveal the spatial clustering patterns and temporal evolution of COD (chemical oxygen demand) emissions across provinces in China. The findings provided important insights into the spatial distribution and underlying drivers of COD emissions.

The geodetector (Geodetector Software in Excel), developed by Jinfeng Wang and collaborators, is a statistical method designed to detect spatially stratified heterogeneity and to identify the dominant factors contributing to its formation [46]. One key advantage is its ability to analyze two core types of spatial variation: inter-factor interactions and intra-factor heterogeneity. Through this analysis, the geodetector provides a quantitative assessment of how well each factor explains variations in a target variable within spatial data.

The method includes four primary components: the factor detector, the interaction detector, the risk detector, and the ecological detector [46]. By computing and comparing the q-values of individual factors, the method evaluates their explanatory strength, where higher q-values indicate a stronger influence on spatial variation. The q-value is calculated using the following equation [37]:

$$Q=1-{\displaystyle \frac{\sum _{h=1}^L{N}_h{\sigma }_h^2}{N{\sigma }^2}}=1-{\displaystyle \frac{SSW}{SST}}$$

(3)

$$SSW={\sum }_{h=1}^L{N}_h{\sigma }_h^2, SST=N{\sigma }^2$$

(4)

In this formula, h = 1, …, L indicates the stratification of variable Y or factor X, referring to the classification or segmentation of data. $N_h$ and N represent the number of units in stratum h and the total number of units in the entire study area, respectively. ${\sigma }_h^2$ and σ² denote the variance of Y within stratum h and across the entire region. SSW (sum of squares within groups) and SST (total sum of squares) are used to measure internal and overall variability.

At present, the geodetector is widely applied across disciplines such as geography, environmental science, ecology, and social economics. It has become a key component of contemporary geographic information analysis [47,48]. For instance, Zhu Yaoyao utilized the geodetector to examine and quantify the multiple drivers influencing the transformation of urban open space in Shanghai, highlighting the predominant influence of socioeconomic factors [49].

2.3.2. Construction of a Sustainable Population Migration Model

To enhance prediction accuracy and deepen understanding of population migration dynamics in mountainous regions, this study built on key driving factors identified via the geodetector model. Specifically, we reconstructed the Sustainable Population Migration Index (SPMI) using a bidirectional framework incorporating attractive and push–pressure dynamics; the SPMI quantifies the net migration potential of each spatial unit by integrating positive and negative drivers within a weighted structure. The index is calculated using the following formula:

$${\mathrm{S}\mathrm{P}\mathrm{M}\mathrm{I}}_{\mathrm{i}}=\sum {{\mathrm{w}}_{\mathrm{j}}}^{\left(\mathrm{A}\mathrm{I}\right)}·{\mathrm{Z}}_{\mathrm{j}}-\sum {{\mathrm{w}}_{\mathrm{k}}}^{\left(\mathrm{PPI}\right)}·{\mathrm{Z}}_{\mathrm{k}}$$

(5)

where Z_j is the standardized score (Z-score) of the j-th Attractive Indicator (AI) in county i; Z_k is the standardized score of the k-th Push-Pressure Indicator (PPI) in county i; w_j^(AI) is the weight of the j-th AI, derived by normalizing its geodetector q-statistic: ${\sum }_{{\mathrm{j}=1}^{\mathrm{nAI}}}{{\mathrm{w}}_{\mathrm{j}}}^{\left(\mathrm{A}\mathrm{I}\right)}=1$; ${{\mathrm{w}}_{\mathrm{k}}}^{\left(\mathrm{PPI}\right)}$ is the weight of the k-th PPI, similarly based on its q-statistic and normalized as: ${\sum }_{{\mathrm{k}=1}^{\mathrm{nPPI}}}{{\mathrm{w}}_{\mathrm{k}}}^{\left(\mathrm{PPI}\right)}=1$; n^AI and n^PPI denote the numbers of AI and PPI variables, respectively.

The Z-score Z_j is computed as $({\mathrm{x}}_{\mathrm{j}}-{\overline{\mathrm{x}}}_{\mathrm{j}})/{\sigma }_{\mathrm{j}}$, where x_j is the raw indicator value, ${\overline{\mathrm{x}}}_{\mathrm{j}}$ is the county-wise mean of the indicator, and σ_j is its standard deviation. Positive SPMI values imply net pull effects (attraction > pressure), while negative values indicate net push effects (pressure > attraction), characterizing spatial migration dynamics.

We used geodetector-derived q-values as indicator weights in constructing the SPMI, as they directly reflect each variable’s explanatory power for spatial population differences. Compared with entropy (unsupervised) or AHP (subjective) weighting, this approach is data-driven and outcome-specific. To test robustness, we recalculated SPMI values using equal and entropy weights, based on data from the 13th Five-Year Plan (2016–2020). The rankings remained largely consistent (Spearman ρ = 0.70 vs. equal-weight; ρ = 0.53 vs. entropy), confirming the stability of our method.

The push-pull model, originally proposed by Lee (1966), is a widely recognized theoretical framework for analyzing the motivations behind population migration [50]. First, a city’s population size and economic scale (GDP) reflect its scale attractiveness [51]. Second, the level of public services, represented by the proportion of the tertiary sector and education level, also significantly influences its advantages [52]. Furthermore, average and minimum wage levels, as well as the unemployment rate, are important indicators of a city’s employment environment, directly affecting labor mobility [53]. Feng et al. focused on the population migration between Chinese cities, conducted an in-depth analysis using the push-pull theory, and clearly included the proportion of the primary industry and the income of rural residents in the empirical research of the “push” force factors, revealing the nonlinear and asymmetric characteristics of the push-pull factors in China’s population migration [54]. In addition, based on the push-pull theory, LI et al. empirically found with inter-provincial data from 1998 to 2012 that the output value of the primary industry and the average disposable income of farmers and herdsmen, which are indirectly related to agricultural technological progress, constitute a “push force” [55]. Based on the above theoretical description, the selected variables were classified into two categories: Attractive Indicators (AIs), which reflect factors that potentially draw population inflows, such as Gross Regional Domestic Product, output values of the secondary and tertiary industries, the number of primary and general secondary schools, and the number of hospitals and health centers; and Push-Pressure Indicators (PPIs), which represent factors that may contribute to population outflows, including primary industry production value and the average disposable income of farmers and herdsmen.

All indicators were normalized using Z-score standardization to ensure comparability, with values reversed for those negatively correlated with migration potential prior to normalization.

Specifically, the geodetector method provided factor significance measures (q-values), reflecting the explanatory power of each indicator regarding spatial variation in population change. We standardized these q-values to compute weights ($W_j$), ensuring their sum equaled 1, as follows:

$$W_j={\displaystyle \frac{q_j}{{\sum }_{j=1}^n{q}_j}}$$

(6)

The calculation of the SPMI involves the following steps: (1) standardizing all indicators using Z-score normalization; (2) deriving indicator weights directly from the q-values obtained via the geodetector model; (3) computing the SPMI score for each administrative unit based on the weighted combination of attractive and push-pressure factors; and (4) visualizing and analyzing the spatial distribution of SPMI results using Geographic Information System (GIS) software such as ArcGIS Pro3.5.

2.3.3. Model Validation Strategy

To evaluate the robustness and explanatory power of the Sustainable Population Migration Index (SPMI) in modeling resident population dynamics, a two-step validation strategy was employed. First, Pearson correlation analysis was conducted to examine the bivariate association between SPMI values and annual population growth rates across all counties and years. Second, to account for potential time-lagged effects and unobserved heterogeneity, a fixed-effects panel regression model was constructed using lagged SPMI variables as predictors. The specification is as follows:

$${PG}_{i,t}=\alpha +{\beta }_1{SPMI}_{i,t-1}+{\beta }_2{SPMI}_{i,t-2}+{\delta }_i+{\gamma }_t+{\epsilon }_{i,t}$$

(7)

where ${PG}_{i,t}$ denotes the annual population growth rate of the county $i$ in year $t$; ${SPMI}_{i,t-1}$ and ${SPMI}_{i,t-2}$ represent one- and two-year lagged values of SPMI, respectively; ${\delta }_i$ and ${\gamma }_t$ capture county and year fixed effects; and ${\epsilon }_{i,t}$ is the error term. Robust standard errors were clustered at the county level. Model performance was evaluated based on the signs and statistical significance of the estimated coefficients ${\beta }_1$ and ${\beta }_2$, as well as within-$R^2$ values, which indicate the proportion of variance explained. This approach enables a rigorous assessment of the SPMI’s predictive power while addressing potential spatial and temporal confounding factors.

To enhance model validation, we adopted a back-casting approach by retrospectively comparing SPMI trends with observed demographic and sectoral changes during a known policy period. This enabled us to assess whether the model could reflect real-world population responses following policy-driven shifts in migration-related conditions.

2.3.4. Digital Twin-Based Simulation and Decision-Support Platform

To integrate and visualize the complex spatiotemporal dynamics of mountainous population migration, we developed a Digital Twin-based Simulation and Decision-support Platform (DTSDP). Combining digital twin technology, GIS, and scenario simulation, the DTSDP enables real-time modeling, visualization, and interactive decision-making. It functions as a virtual counterpart to real-world mountainous regions, dynamically reflecting migration drivers and outcomes based on socioeconomic, infrastructure, and healthcare factors. The platform bridges theoretical modeling and practical policymaking, allowing stakeholders to explore the spatial effects of various intervention scenarios.

The DTSDP features a three-tier architecture (data–model–interaction): (1) the data layer integrates standardized geospatial/statistical indicators (economy, healthcare, infrastructure); (2) the model layer computes migration potentials via the Sustainable Population Migration Index (SPMI) with geodetector-derived weights; and (3) the interaction layer enables real-time parameter adjustment and spatial visualization through an ArcGIS Pro interface.

The Digital Twin-based Simulation and Decision-support Platform (DTSDP) facilitates scenario-based analysis through a structured four-step methodology. First, a baseline scenario is established by calculating the Sustainable Population Migration Index (SPMI_baseline) without policy intervention, serving as a reference for comparison. Next, scenario design and parameter adjustment are conducted by defining hypothetical policy scenarios—such as increasing healthcare accessibility or infrastructure investment—and interactively modifying relevant input parameters within the platform. In the third step, the platform performs scenario simulations, recalculating migration potentials (SPMI_scenario). The significance of scenario-specific SPMI gains was assessed using paired one-sample t-tests on county-level Δ values (n = 10 per scenario).

3. Results

3.1. Identification of Driving Factors

3.1.1. The Spatial Distribution Analysis of Population

As illustrated in Figure 2, the county-level population distribution in Sichuan Province in 2023 exhibited significant spatial disparity, with densely populated areas concentrated in the low-altitude central and eastern regions, while the high-altitude western regions remained sparsely inhabited. Central lowland areas had the highest populations (275,400–1,136,000), while northeastern and southern sub-high areas showed moderate levels (190,000–1,136,000). High-altitude zones had the lowest populations (25,000–190,000), highlighting altitude as a key factor influencing population distribution.

A one-way analysis of variance (ANOVA) was conducted to examine the differences in average population growth rates between non-key counties located in high-altitude regions and those in middle- and low-altitude regions of Sichuan Province over the period 2012–2022. As shown in

Table 2. Statistical significance of migration-related indicators.

Test Item	Comparative Group (Mean Difference ± Standard Deviation)		F- Value	p- Value
	Non-Key High-Altitude Counties (n = 42)	Non-Key Middle- and Low-Altitude Counties (n = 68)
Population growth rate	0.01 ± 0.01	−0.01 ± 0.01	62.904	0.000 **

Note: ** p < 0.01 (two-tailed t-test).

, the results revealed a statistically significant difference between the two groups, F (1, 108) = 62.904, p < 0.001, η² = 0.368. The mean growth rate for non-key high-altitude counties was 0.01 ± 0.01, while that for middle- and low-altitude counterparts was–0.01 ± 0.01. These results indicate a substantially higher population growth rate in high-altitude areas compared to lower-altitude regions.

3.1.2. Spatial Autocorrelation Analysis of Population

Spatial analysis of the 2023 county-level population distribution in Sichuan Province revealed a significant positive global spatial autocorrelation, with Moran’s I = 0.535 (p < 0.001). As illustrated in Figure 3, the LISA cluster map demonstrates a distinct spatial structure: high–high clusters are predominantly located in the eastern region, whereas low–low clusters are concentrated in the western high-altitude areas. Additionally, high–low spatial outliers were identified in parts of the central and eastern regions, while low–high outliers appeared in the southern counties, indicating localized spatial heterogeneity. The analysis suggests that high–high clusters are located in the high-altitude western region, while low–low clusters are situated in the low-altitude eastern region of Sichuan.

3.1.3. Identification of Driving Forces

As shown in Figure 4, in mountainous regions above 2000 m, healthcare is the dominant driver of resident and rural populations, while urban populations are influenced by both healthcare and economic factors. During the Tenth to Fourteenth Five-Year Plans, medical facilities consistently had the highest impact on resident populations (peak q = 0.86) and rural populations (peak q = 0.81). For urban populations, the economy was the main driver in early periods (peak q = 0.59), but medical facilities took precedence in later periods (average q = 0.79). These findings highlight healthcare and economic conditions as key determinants of population trends in high-altitude areas.

The geodetector factor detection results indicate that all 14 selected indicators had a statistically significant impact on population distribution in counties situated above 2000 m (p < 0.05), with notable variation across different Five-Year Plan periods (see Figure 5 and

Table 3. The values of factor detection for economy, infrastructure, and healthcare (2001–2022).

Five Year Plan	Population	Result	Economy								Infrastructure			Healthcare
			E1	E2	E3	E4	E5	E6	E7	E8	I1	I2	I3	H1	H2	H3
10th	Resident population	q	0.544	0.789	0.350	0.519	0.502	0.554	0.032	0.322	0.160	0.460	0.544	0.779	0.611	0.601
		p	0.000	0.000	0.000	0.000	0.000	0.000	0.172	0.000	0.000	0.000	0.000	0.000	0.000	0.000
	Rural population	q	0.455	0.768	0.270	0.434	0.407	0.461	0.021	0.254	0.121	0.433	0.458	0.804	0.535	0.533
		p	0.000	0.000	0.000	0.000	0.000	0.000	0.372	0.000	0.000	0.000	0.000	0.000	0.000	0.000
	Urban population	q	0.726	0.596	0.605	0.712	0.717	0.751	0.101	0.514	0.364	0.435	0.698	0.419	0.674	0.631
		p	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000
11th	Resident population	q	0.617	0.819	0.477	0.614	0.480	0.316	0.061	0.283	0.051	0.497	0.686	0.612	0.813	0.759
		p	0.000	0.000	0.000	0.000	0.000	0.000	0.014	0.000	0.234	0.000	0.000	0.000	0.000	0.000
	Rural population	q	0.525	0.804	0.379	0.510	0.394	0.264	0.064	0.204	0.029	0.500	0.606	0.686	0.769	0.691
		p	0.000	0.000	0.000	0.000	0.000	0.000	0.010	0.000	0.488	0.000	0.000	0.000	0.000	0.000
	Urban population	q	0.749	0.648	0.672	0.785	0.617	0.413	0.060	0.493	0.177	0.400	0.729	0.268	0.762	0.735
		p	0.000	0.000	0.000	0.000	0.000	0.000	0.015	0.000	0.000	0.000	0.000	0.000	0.000	0.000
12th	Resident population	q	0.661	0.880	0.568	0.565	0.530	0.688	0.071	0.023	0.119	0.592	0.738	0.481	0.774	0.459
		p	0.000	0.000	0.000	0.000	0.000	0.000	0.009	0.371	0.003	0.000	0.000	0.000	0.000	0.000
	Rural population	q	0.555	0.844	0.467	0.440	0.426	0.548	0.102	0.004	0.070	0.625	0.663	0.484	0.708	0.380
		p	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.950	0.058	0.000	0.000	0.000	0.000	0.000
	Urban population	q	0.775	0.813	0.695	0.761	0.653	0.872	0.027	0.139	0.269	0.393	0.769	0.354	0.756	0.536
		p	0.000	0.000	0.000	0.000	0.000	0.000	0.252	0.000	0.000	0.000	0.000	0.000	0.000	0.000
13th	Resident population	q	0.619	0.805	0.524	0.488	0.540	0.675	0.156	0.082	0.338	0.692	0.695	0.827	0.900	0.863
		p	0.000	0.000	0.000	0.000	0.000	0.000	0.009	0.006	0.000	0.000	0.000	0.000	0.000	0.000
	Rural population	q	0.497	0.744	0.408	0.358	0.407	0.509	0.208	0.107	0.278	0.714	0.610	0.826	0.861	0.757
		p	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000
	Urban Population	q	0.737	0.800	0.658	0.656	0.683	0.877	0.060	0.125	0.367	0.519	0.688	0.618	0.831	0.877
		p	0.000	0.000	0.000	0.000	0.000	0.000	0.020	0.000	0.000	0.000	0.000	0.000	0.000	0.000
14th	Resident population	q	0.696	0.742	0.518	0.736	0.387	0.616	0.401	0.414	0.306	0.771	0.698	0.719	0.937	0.869
		p	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000
	Rural population	q	0.611	0.651	0.415	0.610	0.312	0.463	0.507	0.404	0.280	0.818	0.629	0.700	0.896	0.780
		p	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000
	Urban Population	q	0.753	0.799	0.626	0.861	0.486	0.817	0.192	0.456	0.295	0.581	0.680	0.591	0.870	0.919
		p	0.000	0.000	0.000	0.000	0.000	0.000	0.007	0.000	0.000	0.000	0.000	0.000	0.000	0.000

Note: q-values represent the explanatory power of each variable for population outcomes, with associated p-values indicating statistical significance (tested at α = 0.05). Units of the original variables are detailed in Table 1.

). In addition, based on the original p-values corresponding to 14 indicators in each period, totaling 70, corrections were made using the Benjamini-Hochberg FDR. Among the 14 indicators, 11 of their adjusted p-values remained below 0.05; see

Table A1. Summary table of Benjamini-Hochberg FDR correction results (FDR = 0.05).

Population Type	Independent Variable	Five-Year Plan	p-Value	Significance	The Adjusted p-Value	Note
Permanent resident population	E7	10th	0.1724	No	0.1775
	E8	12th	0.3711	No	0.3711
	E9	11th	0.2339	No	0.2373
	other	All the others	0.000–0.0136	Yes	≤0.01423	67 significant results
Urban population	E7	12th	0.2516	No	0.2516
	other	All the others	0.000–0.0069	Yes	≤0.0072	67 significant results
Rural population	E7	10th	0.3720	No	0.383
	E8	12th	0.9503	No	0.9503
	E9	11th	0.4882	No	0.495
	E9	12th	0.0583	No	0.0609
	other	All the others	0.000–0.0105	Yes	≤0.0110	66 significant results

. During the Tenth to Twelfth Five-Year Plans, the primary industry production value emerged as the dominant driving factor for both resident and rural populations, with peak explanatory power reflected by q-values of 0.880 and 0.844, respectively. In the Thirteenth and Fourteenth Plans, this shifted to the number of hospital beds (resident q = 0.937; rural q = 0.896).

For rural populations, early drivers included health facility numbers (q = 0.804). Urban population drivers varied over time: retail sales led during the Tenth and Thirteenth Plans (peak q = 0.877), tertiary industry value in the Eleventh (q = 0.785), and health personnel in the Thirteenth and Fourteenth (peak q = 0.819). These results reflect the evolving population dynamics in high-altitude counties, driven by changes in economic and healthcare conditions (Table 3, Figure 6).

3.2. Application of Sustainable Population Migration Index

We applied the previously established SPMI to quantitatively analyze and visualize the spatial patterns of migration potential across the study area (Sichuan Province). Utilizing standardized indicator data and factor weights (

Table 4. Weights of AI and PPI indicators.

Population	Result	AI												PPI
		E1	E3	E4	E5	E6	E7	I1	I2	I3	H1	H2	H3	E2	E8
Resident population	Mean q-value	0.627	0.487	0.584	0.488	0.570	0.172	0.231	0.602	0.672	0.684	0.807	0.710	0.807	0.275
	Weight	0.095	0.074	0.088	0.074	0.086	0.026	0.035	0.091	0.101	0.103	0.122	0.107	0.746	0.254

) derived from geodetector analyses, we computed the SPMI and conducted spatial analyses to identify areas with high and low population migration potentials.

Figure 7 presents the spatial distribution of population migration potential, as indicated by the calculated SPMI values across all county-level administrative units within the study area. Results show clear spatial disparities in migration potential.

Across the 42 ≥ 2000 m counties, the Sustainable Population Migration Index (SPMI) showed a clear, stepwise decline over the five evaluation periods. (1) Early 2000s strength: During 2001–2005, the mean SPMI was modestly positive (0.210 ± 0.193), and 95% of counties still offered net attractive conditions (SPMI > 0). The index spanned a narrow range (–0.034 to 0.662), indicating limited inter-county disparity. (2) Gradual erosion (2006–2015): The average potential fell to 0.104 in 2006–2010 and 0.067 in 2011–2015. Although most counties remained above zero (69% and 57%, respectively), growing variance (SD rising to 0.302) and deeper negative outliers (minimum–0.716) signaled emerging heterogeneity. (3) Inversion and acceleration (2016–2020): The index turned negative for the first time (mean–0.167), with barely 29% of counties retaining a positive SPMI. The mean year-on-year drop (–0.234 versus the previous interval) was more than six-fold that of 2006–2015, underscoring an inflection in migration dynamics. (5) Recent contraction (2021–2022): The latest window registered the sharpest deterioration (mean–0.532, SD 0.621). Only 9% of counties still displayed net pull effects, while the gap between the most and least attractive areas widened four-fold since 2001–2005 (range –3.001 to 0.227).

Overall, the results reveal a transition from broadly favorable to predominantly adverse migration potentials within two decades and an escalating spatial inequality whereby a shrinking cluster of resilient counties contrasts with a growing cohort experiencing pronounced push pressures. These findings highlight the urgency of targeted interventions—particularly in the post–2015 period—to counteract the accelerating loss of demographic attractiveness in high-altitude regions.

3.3. Validation Outcomes: Lagged Effects and Back-Casting Evidence

To quantify the temporal influence of migration potential on population dynamics, we estimated a two-lag fixed-effects panel model. The results (

Table 5. Lagged-effects panel regression between SPMI and population growth.

Variable	Coefficient (β)	Robust s.e.	t	95% CI	p-Value
SMPIt−1	0.0024	0.0012	2.02	0.0001–0.0047	0.04
SMPIt−2	0.0022	0.0013	1.70	−0.0003–0.0047	0.09

) indicate that the one-year lagged SPMI exhibited a positive and statistically significant association with population growth (β = 0.0024, p = 0.044). Substantively, a one-standard-deviation increase in SPMI in year t − 1 was associated with an average 0.24 percentage point rise in the population growth rate in year t. The two-year lag remained positive (β = 0.0022) but was only marginally significant at the 10% level (p = 0.090), suggesting that the pull effect of SPMI attenuates over time yet persists beyond a single year.

The model explained roughly one-third of the within-county variation in demographic change (within-R² = 0.33) after controlling for unobserved heterogeneity. A joint Wald test confirmed that the two lagged terms are jointly significant (F = 5.61, p < 0.001). These findings corroborate the hypothesis that counties with stronger short-term migration potential experience faster subsequent population gains, whereas the longer-term effect weakens but remains directionally consistent.

Sensitivity checks with alternative weighting schemes and balanced-panel specifications yielded comparable effect sizes, reinforcing the robustness of the estimated relationship. Although the within-R² (0.33) may seem modest, it is comparable to values reported in recent migration-related studies (e.g., 0.16 in Tong & Lo 2021 [56]; 0.38–0.43 for conventional gravity models in Wang et al., 2023 [57]), underscoring the inherent stochasticity of population dynamics in complex mountain settings.

We conducted a historical backcasting analysis using county-level data from 2017 to 2020. This period was selected because it immediately followed the implementation of the “universal health insurance coverage” policy launched by Sichuan Province in 2016, which was expected to drive substantial improvements in local healthcare infrastructure and population retention. Specifically, we selected three representative high-altitude counties—Dechang, Zhaojue, and Puge—where the SPMI values exhibited substantial year-on-year increases. In Dechang County, for instance, the SPMI increased from −0.32 (2017) to +0.22 (2018), accompanied by a population growth from 221,700 to 225,000 and an increase of 216 in medical personnel. In Zhaojue County, the SPMI peaked at 0.97 in 2018, while the number of hospital beds rose by 91 and the population grew by 6.3%. Puge County also saw a moderate SPMI increase (+0.097) and a 1.18% population rise in 2019, with parallel improvements in healthcare staffing (Figure 8).

These cases demonstrate that the SPMI trends derived from our model can meaningfully align with real-world policy interventions and subsequent demographic shifts, suggesting the digital twin platform has practical utility in reconstructing historical population responses to spatial policy stimuli.

3.4. Implementation of the Digital Twin Platform

To quantify how concrete policy packages reshape migration dynamics, we ran the Digital Twin-based Simulation and Decision-support Platform (DTSDP) under a baseline (no intervention) and two contrasted policy scenarios. The baseline (SPMI_baseline) reproduced the status quo spatial heterogeneity of the Sustainable Population Migration Index (SPMI) across all 2022 counties and therefore served as the reference surface.

When the supply of basic healthcare was uniformly increased by 20% (healthcare centers, hospital beds, and medical staff), the DTSDP calculated a mean SPMI uplift of +0.11 units across the ten randomly sampled counties (

Table 6. Scenario A—Effect of a 20% boost on the county-level SPMI (2022).

Name	SPMI_base	SPMI_scenA	Delta	Percentage (%)
Yanyuan	−3.340	−3.071	0.269	8.05
Yuexi	−0.080	0.141	0.221	276.25
Muli	−0.199	−0.082	0.118	59.30
Songpan	−0.367	−0.281	0.086	23.43
Luhuo	−0.318	−0.235	0.083	26.10
Jinchuan	−0.349	−0.267	0.082	23.50
Ruoergai	−1.308	−1.234	0.074	5.66
Batang	−0.491	−0.434	0.057	11.61
Li	−0.515	−0.470	0.046	8.93
Xiangcheng	−0.540	−0.500	0.040	7.41
Average value	−0.751	−0.643	0.108	14.33

Note: Δ = Scenario minus baseline SPMI. t values from paired one-sample tests against zero. Bold Δ values are significant at p < 0.05.

). The improvement ranged from +0.27 (salt-poor Yanyuan County) to +0.02 (already well-served Meitan County), translating into an average ≈6% gain relative to the counties’ own baseline scores. Counties starting with the deepest medical deficits benefited the most, some turning their net “push-pressure” status into a slightly positive migration potential. Raising fixed-asset investment, primary-school capacity, and general secondary-school capacity by 20% produced a smaller but still positive response, with a mean SPMI increase of +0.05 units (max +0.09 in Jinyang County; min +0.02 in Dege County). The relative gain (~3%) confirmed that education-cum-investment upgrades enhance migration attractiveness, yet their marginal effect is moderated when the county already possesses a basic educational and economic platform (

Table 7. Scenario B—Effect of a 20% increase on the county-level SPMI (2022).

Name	SPMI_base	SPMI_scenB	Delta	Percentage (%)
Jinyang	−0.014	0.078	0.092	657.14
Ganluo	0.187	0.264	0.077	41.18
Pingwu	0.107	0.181	0.074	69.16
Songpan	−0.367	−0.308	0.060	16.35
Aba	−0.665	−0.623	0.042	6.32
Mao	−0.566	−0.531	0.035	6.18
Rangtang	−0.606	−0.578	0.029	4.79
Heishui	−0.554	−0.534	0.020	3.61
Daocheng	−0.437	−0.418	0.018	4.12
Xiangcheng	−0.540	−0.529	0.012	2.22
Average value	−0.346	−0.300	0.047	13.29

Note: Δ = Scenario minus baseline SPMI. t values from paired one-sample tests against zero. Bold Δ values are significant at p < 0.05.

). A paired two-tailed t-test confirmed that both policy scenarios produced statistically significant improvements in the SPMI (Scenario A: t = 4.44, p = 0.0016; Scenario B: t = 5.17, p = 0.0006).

The 20% increment assumption in healthcare and infrastructure capacity aligns with recent pilot upgrades in western China (e.g., the “Healthy Sichuan Initiative”), which targets annual growth rates of 15–20% in medical resources for mountainous counties [58]. This magnitude is technically feasible, as evidenced by its effectiveness in reducing migration pressure through marginal improvements in public service accessibility [59]. We acknowledge the fiscal constraints, and future iterations of the DTSDP will incorporate budget-sensitive scenarios based on county-level fiscal capacity data, following the phased implementation strategy recommended for underdeveloped regions [60].

Interactive rasters rendered by the DTSDP immediately highlighted these numerical shifts. Health-care strengthening (Scenario A) generated the most pronounced hot spots in previously underserved western mountain corridors, whereas infrastructure spending (Scenario B) produced gradual, garden-wall improvements clustered around existing road axes and market towns. These layered visualizations allow planners to pinpoint where each policy lever delivers the highest marginal return, facilitating a mix-and-match approach.

4. Discussion

4.1. Drivers of Migration Dynamics

This study investigated the primary drivers of population dynamics in high-altitude mountainous counties within Sichuan Province, China. Healthcare accessibility emerged as the most critical determinant of population changes, particularly for resident and rural populations. This aligns with prior findings that underscore the rising importance of quality-of-life factors in migration decisions [9,24,60,61,62]. Economic conditions played a secondary yet notable role, especially influencing urban migration. During earlier periods, tertiary industry production and retail sales drove urban inflows, consistent with the literature on industrial clustering and regional competitiveness [12,16,63].

4.2. Predictive Validity of SPMI

The Sustainable Population Migration Index (SPMI) proposed herein provides a robust, composite framework for tracking migration potential over time. Its consistent decline across the Five-Year Plan windows reflects the growing demographic pressures in highland regions. Furthermore, the fixed-effects panel model confirmed that lagged SPMI scores are significant predictors of subsequent population changes, validating its utility in longitudinal migration forecasting [4,64]. The ANOVA results indicating higher population growth in high-altitude counties reflect a horizontal comparison with lowland regions, many of which have faced sharper declines due to urban outmigration or demographic aging. In contrast, the SPMI measures longitudinal changes in migration potential within high-altitude counties. Its decline over time reveals a weakening structural capacity to attract or retain population, even as some counties still recorded short-term growth. This is further supported by our panel regression, which determined that population change often lags behind migration potential. Thus, the two findings reflect different dimensions and are not contradictory.

4.3. Policy Simulation Insights

The simulation results obtained from the Digital Twin-based Simulation and Decision-support Platform (DTSDP) highlighted the effectiveness of the targeted interventions. Scenario A (healthcare upgrade) achieved a mean uplift of +0.12 in the SPMI, particularly benefiting the most underserved counties. Scenario B (education and investment boost) produced a more modest gain (+0.05), with limited effect in structurally weaker areas. These results suggest that foundational healthcare access remains the bottleneck for population retention in many counties. Accordingly, tiered healthcare reforms and region-specific economic diversification programs are essential. A high-altitude demographic sustainability transfer payment, benchmarked against SPMI performance, could further institutionalize equity and accountability in resource allocation.

4.4. Relative Impact and Cost Implications

Although fiscal costs vary across counties, our indicative calculations indicate that a 20% healthcare upgrade yields over twice the SPMI improvement per investment unit compared to infrastructure interventions—approximately 0.006 versus 0.0025. This finding underscores the strategic value of healthcare investment in constrained regions. Zhu and Fang (2018) [30] demonstrated that economic development in high-altitude areas is often hindered by environmental and service accessibility barriers, rendering basic healthcare a key determinant of settlement viability. Similarly, Calero et al. (2009) found that, in low-income contexts [65], household-level investments in health and education are more effective than large-scale infrastructure in mitigating migration pressures. These studies support our conclusion that healthcare interventions offer higher marginal returns and may outperform relocation or infrastructure projects in moderately underperforming counties.

4.5. Theoretical Implications and Broader Significance

These findings extend classical migration theory by explicitly integrating both institutional and service-based constraints into the high-altitude context. Guo et al. (2007) highlighted that the out-migration of rural labor in China is largely driven by underdeveloped educational and healthcare systems [66], with basic service deficiencies functioning as critical “push” factors. Similarly, Fan and Ma (2011) emphasized that structural inequalities in public service provision—particularly in education and healthcare—have been central to China’s large-scale interregional migration flows [67]. Building on this, the present study advances the literature by quantifying the structural impact of basic healthcare provision on population retention through the SPMI metric, thereby empirically validating the claim that foundational service gaps undermine local settlement attractiveness. When essential needs such as healthcare remain unmet, traditional “pull” interventions—such as investment or education expansion—yield diminishing marginal returns. This underscores the theoretical imperative that sustainable migration governance must first address push-driven vulnerabilities. For peripheral counties in western China, these insights provide a policy rationale for prioritizing the closure of healthcare and basic service deficits, rather than relying solely on capital investment or educational improvements to curb out-migration.

4.6. Methodological Limitations and Future Directions

Despite the explanatory value of the fixed-effects panel model, the within-R² of 0.33 suggests that additional migration determinants remain overlooked. These may include social networks and kinship ties in origin communities, which significantly shape migration decisions and reduce mobility risks through information dissemination and mutual support mechanisms [68]. Ethnic and religious affiliations also influence migration through shared community norms and spatial clustering patterns, as shown in recent studies on spatial heterogeneity in mobility outcomes [69]. Furthermore, land tenure security and rural land policy reforms have long been recognized as institutional drivers affecting rural residents’ willingness and ability to migrate [70]. State-led resettlement programs, such as the Grain-to-Green ecological migration scheme, and temporally concentrated interventions like targeted poverty alleviation initiatives can create “policy window” effects [71]. In addition, environmental shocks such as landslides or droughts may trigger involuntary migration. Owing to data limitations, such factors were not quantitatively incorporated, yet they likely contribute to regional migration heterogeneity. Moreover, given the spatial nature of the data, future research may benefit from applying spatial panel regression techniques (e.g., Spatial Durbin Model) to better capture interregional spillover effects and improve model robustness.

5. Conclusions

This study systematically explored the determinants of population dynamics in the high-altitude mountainous regions of Sichuan Province, China, through a combination of quantitative modeling, spatial analysis, and policy simulation. By integrating the Sustainable Population Migration Index (SPMI) and a Digital Twin-based Simulation and Decision-support Platform (DTSDP), it supports a multidimensional understanding of how socioeconomic and environmental factors shape demographic changes in these vulnerable areas.

5.1. Core Findings in a Global Context

The analysis identified healthcare accessibility as the most critical driver of population changes, particularly for resident and rural populations, while economic factors exert a more pronounced influence on urban populations. This pattern aligns with broader global trends in high-altitude regions, where basic public services and quality-of-life factors increasingly outweigh traditional economic drivers in migration decisions. However, cross-regional comparisons reveal nuanced differences: unlike the Andes Mountains, where population dynamics are often dominated by extractive industry opportunities [72] or climate-induced displacement, or the Himalayan foothills, where environmental fragility (e.g., glacial melt) is a primary push factor [73], Sichuan’s high-altitude counties exhibit a unique interplay of state-led public service provision (e.g., healthcare infrastructure) and localized economic transitions. This highlights how institutional contexts—such as China’s rural revitalization policies and regional development strategies—mediate the relative importance of demographic drivers in high-altitude settings.

Regarding transferability, the SPMI framework is structurally adaptable. Although this study focused on high-altitude counties in Sichuan, the underlying methodology—based on attractive and push–pressure dimensions, geodetector-derived weighting, and standardized scoring—can be recalibrated to fit other regions with distinct socio-geographic profiles. We encourage future researchers to apply and validate the SPMI in varied settings such as the Qinghai–Tibet Plateau, the Andes, or the Himalayas, where topographic constraints and public service disparities similarly influence migration potential.

5.2. Theoretical and Practical Contributions

Theoretically, this study advances understanding of population dynamics in high-altitude regions by introducing the SPMI: a framework that quantifies migration potential through integrating push–pull factors with spatially weighted indicators. While the fixed-effects panel model confirms that the SPMI exhibits a predictive value for short-term population trends (one-year lag: β = 0.0024, p = 0.044), its explanatory power (within-R² = 0.33) and weaker two-year lag effect (p = 0.090) also highlight the complexity of demographic processes, emphasizing that migration decisions are shaped by both measurable factors and unobserved dynamics (e.g., cultural attachments or informal social networks). This modesty in predictive strength aligns with the inherently multifaceted nature of population mobility, reinforcing the need for complementary qualitative research to unpack residual variance.

Practically, the DTSDP demonstrated how digital twin technology can bridge theoretical analysis and policy action. Scenario simulations revealed that targeted investments in healthcare (20% enhancement) and infrastructure (20% increase in fixed-asset investment and educational facilities) can significantly mitigate migration pressures, with the former yielding larger gains—particularly in counties with historically inadequate medical resources. This tool offers policymakers a visualized, evidence-based platform to prioritize interventions, addressing the longstanding challenge of translating demographic insights into actionable strategies in remote high-altitude regions.

5.3. Implications for Sustainable Development

High-altitude regions worldwide face the shared challenges of maintaining population stability while balancing ecological preservation and socioeconomic development. This study emphasizes that sustainable development in such areas requires tailored strategies: for rural and resident populations, prioritizing healthcare accessibility can strengthen community resilience; for urban centers, fostering economic opportunities—particularly in tertiary industries—remains critical to retaining talent. The SPMI, as a dynamic monitoring tool, can help in tracking the effectiveness of these strategies over time, while the DTSDP enables proactive scenario planning to avoid reactive policymaking.

Illuminating the interplay of healthcare, economy, and infrastructure in shaping population dynamics, this research not only enriches the literature on high-altitude demography but also provides a replicable analytical framework for studying other mountainous regions. Future work can extend this model by integrating climate change impacts and cross-border migration flows, further enhancing its utility for global sustainable development agendas in fragile high-altitude ecosystems.