Assessment of resilience and key drivers of Tibetan villages in Western Sichuan

Villages in the Tibetan regions of western Sichuan face multiple challenges due to their unique geographical environment and cultural-ecological significance. These challenges include extreme plateau climates, a single-industry structure, and external pressures such as urbanization and policy changes^1,2. Most of these villages are in high-altitude mountainous regions characterized by fragile ecosystems, limited arable land, and a high risk of natural disasters. Meanwhile, Tibetan ancestors accumulated extensive experience in ecological protection and resource management by adopting a livelihood model that combined nomadism and farming, leveraging community autonomy and religious cultural practices³.

In the context of sustainable development, these villages must tackle environmental and climatic challenges, and the social and economic impacts of population decline and modernization. Given their ecological vulnerability and cultural uniqueness, it is essential to systematically assess these villages’ resilience and capacity for adaptive governance. This assessment is crucial for informing sustainable development strategies in ecologically fragile regions⁴. Assessing village resilience from a multi-scalar perspective has become increasingly essential considering the complex social, economic, and ecological interactions. This process includes identifying key drivers, revealing spatial variations and temporal trends, and exploring effective models for adaptive governance, issues that have emerged as significant areas of current research^5,6. This study applies the Social-Ecological System (SES) theory and the Driver–Pressure–State–Impact–Response (DPSIR) framework^7,8, to investigate six representative villages from the Kham Tibetan, Jiarong Tibetan, Amdo Tibetan, Ersu Tibetan, and Baima Tibetan communities. Through quantitative analysis methods, this study evaluates the impacts of social, economic, environmental, and policy factors on village resilience. The findings seek to deepen the understanding of resilience mechanisms in Tibetan villages and provide empirical insights for the sustainable development and governance of other ecologically sensitive plateau regions.

Research flowchart(self-illustration). Note: The Flowchart(see Fig. 1) shows the overall research framework and design, covering the theoretical foundations, data collection, analysis methods, and anticipated outcomes. The illustration shows the typical functional areas of Tibetan villages (residential areas, public service areas, cultural activities, etc.), which provide a visual reference for the socio-ecological environment under discussion.

Full size image

Achieving balanced economic and social development in ecologically fragile regions is a key challenge in the context of global sustainable development⁹. Since C.S. Holling (1973) introduced the Social-Ecological System (SES) concept, this framework has been extensively adopted to investigate the interactions between human and environmental systems¹⁰. The SES framework emphasizes the dynamic interactions between social systems, such as population, economy, culture, and governance, and ecological systems, including land, water resources, climate, and biodiversity. Together, these systems form a complex adaptive system through multi-level feedback mechanisms^11,12. Recent studies have extended the SES and DPSIR framework to better analyze the relationship between social dynamics and environmental pressures in rural areas¹³.

The Driver–Pressure–State–Impact–Response (DPSIR) framework, endorsed by organizations such as the OECD, is vital for environmental management and policymaking. This framework enables researchers to systematically assess the impacts of human activities on the environment and provide insights into policy responses and the role of social capital in resilience^14,15. The traditional DPSIR framework, while useful, has limitations in rural settings. It has been suggested that combining SES with DPSIR gives a more valid interpretation¹⁵. In China, the ongoing promotion of ecological civilization has led to the progressive application of these two frameworks in studies of ecologically sensitive high-altitude regions in the western part of the country, demonstrating significant potential for practical implementation¹⁶.

Resilience refers to a system’s capacity to maintain its functions, absorb shocks, and adapt to external disturbances¹⁷. In rural contexts, resilience is often conceptualized across multiple dimensions: ecological resilience (e.g., soil and water conservation, ecosystem health), social resilience (community networks, governance systems, policy support), and economic resilience (diversified livelihoods, market connectivity). Collectively, these dimensions determine the capacity of rural communities to adapt to both sudden disruptions and long-term pressures^{15,18,19,20,21}. Unlike sustainability, which focuses on balancing resource use between the present and the future^22,23,24, resilience emphasizes a system’s ability to preserve its core functions in the face of both short-term fluctuations and long-term changes^13,25.

The Social-Ecological System (SES) theory emphasizes the significance of coupling, adaptive capacity, and multi-scale interactions in understanding rural resilience. The interdependence between social and ecological subsystems plays a crucial role in shaping resource management and environmental governance²⁶. From this perspective, scholars have investigated how communities respond to external shocks by analyzing factors such as social networks, policy interventions, and technological innovations²⁷. However, research on Tibetan regions remains relatively limited. Most existing studies focus primarily on the unidirectional impacts of economic pressures or environmental changes on villages^15,28. Few studies have adopted data-driven and comprehensive approaches to dynamically assess resilience mechanisms in these regions. This underscores the need for further research to explore how Tibetan villages manage environmental, economic, and cultural interactions within an integrated social–ecological framework.

The Driver–Pressure–State–Impact–Response (DPSIR) framework was introduced by the Organization for Economic Co-operation and Development (OECD) and later expanded by the European Environment Agency (EEA). It is widely used for causal analysis in environmental management and policymaking¹⁵. This framework categorizes the effects of human activities on the environment into five interconnected components:

• Driving forces: Factors such as population growth, economic development, and policy changes that initiate pressures on the environment^29,30.

• Pressure: The direct stresses exerted on environmental systems, including land degradation, climate change, inadequate infrastructure, and energy shortages^31,32.

• State: The current conditions of ecological and social systems, encompassing resource availability, environmental quality, and ecosystem health³³.

• Impact: The direct and indirect consequences for communities and ecosystems, such as increased livelihood risks and declining agricultural productivity²⁶.

• Response: Adaptive strategies and measures implemented by governments, communities, or markets to mitigate or manage these impacts³⁴.

The DPSIR framework is widely used in environmental management due to its effectiveness in clarifying causal relationships between policy interventions and environmental changes^8,35. However, applying the DPSIR framework to assess resilience in Tibetan villages reveals several limitations.

First, the DPSIR framework is primarily based on generalized environmental management models and lacks sufficient consideration for the high-altitude geographical conditions and unique cultural values of Tibetan regions³⁶. Second, the framework’s emphasis on a “top-down” policy approach may neglect the crucial role of community self-governance and traditional governance practices in strengthening village resilience³⁷. In particular, the framework fails to capture how geographic remoteness, ecological vulnerability, and cultural embeddedness interact in shaping localized adaptive mechanisms. For example, high-altitude environments pose physical challenges (e.g., harsh climate, difficult access) and reinforce tightly knit social structures and spiritual worldviews, which together inform community responses to environmental change.

Notably, Tibetan religious and customary systems foster resilience through deeply embedded social-ecological norms. Tibetan Buddhism incorporates ecological ethics, such as reverence for sacred mountains and rivers, which promote conservation behaviors like seasonal hunting bans and grazing restrictions. Monasteries, serving as both sacred spaces and informal governance centers, facilitate conflict resolution and communal coordination. These culturally embedded, bottom-up mechanisms act as adaptive institutions that complement formal policy interventions, especially in remote and infrastructure-limited settings^5,19.These limit the DPSIR framework’s capacity to capture culturally embedded and community-led resilience pathways that are central to Tibetan rural contexts.

In addition, the DPSIR framework is mainly designed for static assessments and lacks a strong emphasis on capturing the dynamic evolution of resilience over tim³⁸. Overcoming these limitations is crucial for accurately assessing the resilience of Tibetan villages and for developing more context-sensitive and adaptive management strategies. A more comprehensive framework is needed to incorporate spatiotemporal dynamic observation and grassroots adaptation simultaneously.

To achieve this goal, it is essential to consider the strengths and limitations of alternative conceptual frameworks. While theories such as Social Capital and Institutional Analysis emphasize collective norms and institutional arrangements, they often overlook ecological feedback³⁹. In contrast, integrating the Social-Ecological Systems (SES) theory with the DPSIR model provides a dual-lens perspective: SES captures cross-scale feedback loops and adaptive capacity, whereas DPSIR offers a structured representation of the relationships between drivers, pressures, and institutional responses. This hybrid framework enables the analysis of both bottom-up resilience mechanisms and top-down policy impacts. It is particularly well-suited to Tibetan villages, where ecological fragility, cultural embeddedness, and governance complexity converge.

In recent years, scholars have aimed to develop a more comprehensive approach to resilience assessment by integrating the Social-Ecological System (SES) theory with the Driver–Pressure–State–Impact–Response (DPSIR) framework. This integration embeds multi-scale social–ecological interactions into the DPSIR causal chain, thereby incorporating social, cultural, and policy dimensions into the analysis^40,41. The SES theory provides insights into the interactions between social and ecological systems through feedback mechanisms, forming complex adaptive systems. In contrast, the DPSIR framework offers a systematic approach to analyzing the sequence of “Driving Forces–Pressure–State–Impact–Response,” highlighting the impact of policy interventions on environmental change and community development⁴².

This integrated approach enables a more nuanced understanding of resilience by concurrently highlighting the roles of social drivers and ecological pressures. Additionally, it recognizes the significance of local traditional governance and community self-organization, thereby addressing the limitations of approaches that focus exclusively on macro-level policies or individual ecological indicators³³. The integration of these two frameworks enables researchers to gain a deeper understanding of the complex and dynamic interactions that shape resilience, offering a more balanced perspective for managing social–ecological systems.

Although previous studies have undertaken similar explorations, a significant gap remains in systematic, quantitative analyses that encompass multiple time scales and facilitate cross-village comparisons in Tibetan regions. Some case studies indicate that, even without external interventions, Tibetan villages can preserve or restore their ecological environments through self-initiated actions by individuals and communities (Ecological Awareness in a Tibetan Village)⁴³. This “community-led” approach to resilience is frequently underrepresented in the traditional DPSIR framework and has not yet been adequately integrated into the SES–DPSIR model. Furthermore, achieving a balance between government policies, local self-governance, and cultural heritage within the context of fragile plateau ecosystems remains a critical area for further research.

To better assess the complex interactions and resilience of rural systems, scholars have developed a range of methods, including:

• Qualitative research methods: Approaches such as literature-based knowledge mapping, field interviews, and participatory research offer in-depth insights into the internal governance mechanisms and adaptive strategies of communities^27,44. However, qualitative methods often have limitations in enabling cross-case comparisons and are susceptible to research bias.

• Spatial analysis methods: Remote sensing and GIS technologies are effective tools for monitoring land-use changes and ecosystem dynamics, providing objective environmental data^45,46. However, these methods have limitations in interpreting social and economic dimensions, including the influence of community networks and policy interventions.

• Ecosystem service assessment models: Models such as RUSLE provide quantitative assessments of ecological restoration, supporting environmental planning and policymaking^47,48. However, these models primarily focus on ecosystem health and pay limited attention to social adaptation and economic resilience, making it difficult to capture the multidimensional nature of rural resilience.

Addressing these methodological limitations is crucial to advancing a more comprehensive understanding of the resilience of Tibetan villages, particularly in terms of the dynamic interactions among society, ecology, and policy.

Although existing research methods have offered valuable insights into rural resilience, three major limitations remain. First, most studies lack a comprehensive perspective that integrates mutual feedback between social and ecological systems. Second, there is a lack of longitudinal analyses that capture changes across multiple time scales, with most research relying on cross-sectional data^49,50. Third, the linkage mechanisms between policies and community practices have not been thoroughly examined, especially concerning the roles of traditional governance and community self-organization in Tibetan villages^51,52 .

A review of existing literature reveals several key research gaps:

• Theoretical level: There is a lack of systematic research on the interaction mechanisms between social and ecological systems in Tibetan rural areas, as most studies focus on either economic or environmental dimensions independently.

• Methodological level: The integration of the SES theory and the DPSIR framework is insufficient, and comprehensive quantitative models capable of addressing complex social–ecological interactions across multiple time scales are lacking.

• Content level: Limited attention has been given to community-led resilience strategies and traditional governance practices.

• Dynamic assessment: The lack of multi-temporal dynamic assessments hinders the ability to capture the long-term evolution of resilience and to identify the key drivers behind these changes.

This paper proposes the following key research questions (RQs) to guide the subsequent analysis:

• RQ1: What are the key social, economic, environmental, and policy factors influencing the resilience of Tibetan villages?

• RQ2: What are the differences in resilience among different Tibetan ethnic groups (Kham, Jiarong, Amdo, Ersu, Baima)?

• RQ3: How has the resilience of Tibetan villages evolved between 2015 and 2022?

In response to these research questions, this paper integrates the Social-Ecological System (SES) theory with the DPSIR framework to construct a resilience assessment system for Tibetan villages. This approach integrates cross-village comparisons with longitudinal analyses to address the limitations of traditional cross-sectional studies, thereby revealing the spatial heterogeneity of resilience and its long-term evolutionary patterns. The findings of this study provide empirical support for formulating targeted policies, optimizing ecological compensation mechanisms, and enhancing sustainable community governance in Tibetan villages.

Location of sample village

Research ___location(self-illustration). Note: Fig. 2b, c, and d were created using ArcMap 10.8, QGIS 3.22.5, and CorelDRAW 2019. The base map in Fig. 2a is sourced from OpenStreetMap contributors and used under the Open Database License (ODbL), available at https://www.openstreetmap.org/copyright.

Full size image

The western Sichuan region features a high-altitude ecological environment. The selected villages—Moluo, Kegeyi, Xisuo, Shenzuo, Minzu, and Jiangba—are situated in Zhonglu Township and Suopo Township of Ganzi Tibetan Autonomous Prefecture, Zhuokeji Town and Chali Township of Aba Tibetan and Qiang Autonomous Prefecture, Muzuo Tibetan Township of Mianyang City, and Xieluo Tibetan Township of Ya’an City, respectively. The elevations of these villages range from 1200 to over 4000 m (see Fig. 2).

These villages are listed in the Catalogue of Traditional Chinese Villages and represent various elevation gradients, ecological environments, economic structures, and ethnic and cultural backgrounds, reflecting the diverse characteristics of different Tibetan village communities.

Research data resource and processing

This study systematically evaluates the socio-economic and ecological transformations of Tibetan villages in western Sichuan, utilizing both primary and secondary data collected in 2015 and 2022. Comprehensive field surveys were conducted across all selected villages to ensure data accuracy and enhance the feasibility of the research, thereby providing a robust empirical foundation for the subsequent analytical framework.

Primary data were collected through field surveys, focusing on the systematic observation and recording of key quantitative indicators⁵³. Specifically, an “observation indicator checklist” was pre-developed to record information on village infrastructure, land-use types, main industrial sectors, public service facilities, and environmental conditions. Additionally, continuous observations were conducted to assess the layout of village buildings, potential geological hazards in high-altitude areas, and sanitary conditions, providing direct insights into the interactions between ecological and social aspects of the villages. These quantitative records provide a crucial foundation for comparing and integrating primary and secondary data in the subsequent analysis⁵⁴.

Secondary data was mainly obtained from the Catalogue of Traditional Chinese Villages (2023), the Sichuan Statistical Yearbook (https://www.zgtjnj.org/index.aspx), open data platforms (www.gotohui.com/user/login.html), regional administrative boundary POI data (http://guihuayun.com/), government development reports, and ecological and environmental monitoring data (https://sthjt.sc.gov.cn/). These sources provided key indicators, including population size, vegetation coverage, economic growth rate, and road density, which were primarily used for modeling.

To minimize reliance on any single source and mitigate issues with missing data, a cross-validation strategy was adopted across multiple data sources. For variables that were partially missing at the village level, such as specific ecological indicators and urbanization rates, county-level or township-level means were used as proxy estimates, based on spatial and administrative proximity. These values were imputed using a mean substitution method, rather than model-based estimation techniques (e.g., regression imputation), due to the relatively small proportion of missing data. To assess the robustness of these imputations, a sensitivity analysis was conducted by re-running key regression and clustering models with and without the imputed values. Results showed only negligible variation in classification patterns and regression coefficients, indicating that the imputations had a minimal effect on the overall findings.

Due to incomplete statistical data for some villages in 2023 or 2024, the year 2022 was selected as the most recent reliable benchmark. To ensure temporal comparability and data integrity, this study focused on 2015 and 2022 as the two anchor years. But it’s worth noting that the fixed time horizon limits the ability to capture long-term resilience trends, especially those related to slowly changing ecological, social, and economic factors. Future studies should consider extending the time horizon to gain a deeper understanding of resilience dynamics over an extended period.

Next, a purposive sampling strategy was employed to ensure that the selected villages reflected diversity in elevation, socio-economic status, and ecological⁵⁵. To calculate the composite scores for each village, the SUMPRODUCT function in Excel was utilized, enabling the weighted summation of key indicators. The scores were subsequently ranked and filtered using the SORT & RANK functions and conditional statements (IF functions) to identify villages that met the selection criteria^56,57.

By integrating primary and secondary data, this study constructed a comprehensive dataset encompassing 35 core indicators across six Tibetan villages—Moluo, Kegeyi, Xisuo, Shenzuo, Minzu, and Jiangba (see Table 1). These variables spanned social, economic, and ecological dimensions. To address differences in measurement units and value ranges, min–max normalization was applied using the formula: \(\left({X}_{norm}\right)\)\(\:=\;\frac{{X}_{i}-{X}_{min}}{{X}_{max\:}-{X}_{min}}.\) with normalized values ranging between [0, 1]⁵⁸, ensuring comparability and providing a reliable foundation for downstream analysis(see Appendix Tables 1 and 2).

Research approach

This study developed a resilience assessment model for Tibetan villages by integrating the Social-Ecological System (SES) perspective with the Driver-Pressure-State-Impact-Response (DPSIR) framework. The entropy weight method was employed to objectively assign weights to each indicator. Subsequently, multi-level analysis and validation were conducted using the Generalized Additive Model (GAM), Lasso regression, and shallow neural networks. These models were chosen for their ability to capture the complex nonlinear relationships between the social, economic, and ecological factors influencing village resilience. GAMs effectively manage nonlinear relationships, while shallow neural networks excel at identifying complex patterns, particularly when working with small sample sizes. To reduce overfitting, Lasso regression was used for regularization, ensuring simplicity and addressing multicollinearity, while cross-validation ensured model robustness.

First, data integration was performed, followed by the development of the DPSIR indicator system⁵⁹. This study integrated field survey data, which provided micro-level insights, with statistical literature data. Village-level survey data were harmonized with county-level or township-level statistics, and variance analysis was conducted when necessary to assess the reliability of proxy data. Following the causal chain of the DPSIR framework (D → P → S → I → R), the initial 35 indicators, encompassing aspects such as economic development, social structure, resource environment, and policy support, were systematically categorized (see Table 1). Additionally, the SES theory was utilized to emphasize the self-organizing and adaptive capacities of villages.

Secondly, the entropy weight method (EWM) was applied to determine the weights of each indicator within the five levels⁶⁰. Assuming there are m indicators and n villages, the entropy value \(\:{E}_{j}\), was first calculated for each indicator. The weights \(\:{W}_{j}\) were then determined using the following equation:

Here, \({W}_{j}\)represents the weight of indicator j, and \({E}_{j}\) is its entropy value. A higher value indicates that the indicator has lower variability and thus a smaller contribution to the overall assessment. By using this objective weighting method, the study ensured that the comparability of multi-source data was maintained⁶¹.

Next, a stratified scoring approach was applied. The weights \({w}_{i}\) derived from the entropy weight method were assigned to each indicator within the five levels (D, P, S, I, R). The overall score L for each level was then calculated using the weighted average method as follows:

Here, L represents the aggregate score for each level, \(\:{S}_{i}\) is the score of the ith indicator, reflecting its significance in the overall assessment, and n is the total number of indicators⁶². By employing this method, the study ensured that indicators with higher weights contributed more substantially to the overall score, providing a balanced and objective assessment.

Subsequently, resilience measurement. To assess changes in village resilience between 2015 and 2022, we compared the comprehensive scores of the five levels. “D, P, S, I, R” represent the scores for the Driving, Pressure, State, Impact, and Response levels, respectively. The change in resilience, denoted as ΔR, is calculated as follows:

Here, wD, wP, ws, wI, wR represent the weights assigned to each level based on the entropy weight method, which serve to quantify the contribution of each level to the overall change in resilience. This approach allows for a balanced assessment of how different factors influence resilience over time.

We utilized the Generalized Additive Model (GAM) to examine the nonlinear effects of social, economic, environmental, and policy factors on the resilience index ΔR^63,64. The model is specified as follows:

Here, \(\:{\beta\:}_{0}\) is the intercept term, \(\epsilon\) represents the error term, and s(⋅) denotes the smoothing function, which is employed to capture the potential nonlinear relationships between the independent variables and the resilience index ΔR.

After identifying significant variables, we proceeded to quantify the impact of key driving factors on ΔR.To overcome the challenges posed by small sample sizes, we applied the Lasso regression method⁶⁵. This approach incorporates regularization to alleviate multicollinearity issues and enhance the simplicity and robustness of the model⁶⁶. Lasso regression prevents overfitting and enhances both the predictive accuracy and interpretability of the model by introducing a penalty term that shrinks some coefficients to zero. The objective function of Lasso regression is defined as follows:

Where \({y}_{i}\) represents the observed value. \({x}_{ij}\) is the jth predictor for the ith observation. \({\beta}_{j}\) denotes the coefficient of the jth predictor. λ is the regularization parameter that controls the strength of the penalty term \({\uplambda}\sum\:_{j=1}^{p}\left|{\beta\:}_{j}\right|,\) which controls the strength of regularization to achieve variable selection and prevent overfitting; \(\frac{1}{2n}\sum\:_{i=1}^{n}({y}_{i}-{\beta\:}_{0}-\sum\:_{j=1}^{p}{\beta\:}_{j}{x}_{ij})^2\) is the least squares error term used to ensure the accuracy of the model fit.

Finally, to assess the robustness of the research model, a shallow Backpropagation (BP) neural network was applied for validation⁶⁷. Two key considerations guided the choice of this approach. On the one hand, given the relatively small sample size, the use of deep neural networks or ensemble models (e.g., Random Forest) could easily lead to overfitting or require extensive parameter tuning. Second, shallow BP networks, especially those with one or two hidden layers, have demonstrated strong performance in small-sample contexts, particularly when the modeling objective involves capturing nonlinear relationships.

During the modeling process, to solve multicollinearity, we integrate the key variables determined by Lasso regression and GAM model analysis, ensuring the independence of features and the robustness of the model. Considering the small sample size, the model was structured with one input layer, two hidden layers, and one output layer, employing the Sigmoid activation function to enhance its capacity for capturing nonlinear relationships. The dataset was divided into 70% for training, 15% for validation, and 15% for testing, ensuring effective control over overfitting. Additionally, the learning rate and regularization parameters were carefully tuned to improve model generalizability⁶⁸. If the predictions for ΔR generated by this model demonstrate high consistency with the observed values, as indicated by Mean Squared Error (MSE) and R² metrics, this would imply that the resilience assessment proposed in this study is both accurate and reliable (see the next section for details). A Shallow BP neural network has achieved an effective balance between model simplicity, nonlinear expression ability, and robustness under small sample conditions, making it a more suitable method for current analysis.

Village overview

Due to the diverse topography and complex climatic conditions, most of these villages are situated along rivers or adjacent to forests and grasslands, reflecting their long-term adaptation to the natural environment. This phenomenon exemplifies the coexistence of multiple ecological types within a relatively small area^69,70. These villages have a long history, with their architectural styles and living environments preserving distinct ethnic and cultural characteristics. As a result, they have been included in the National Catalogue of Traditional Villages, highlighting their cultural and historical significance (see Fig. 3).

Field survey. Note: The photographs in Fig. 3 were taken by the authors during fieldwork conducted in Garzê County, Aba County, Pingwu County, and Shimian County, Sichuan Province, China, between 2022 and 2023. All images are original and are licensed under CC BY 4.0.

Full size image

Figure 3a–f presents real-world images of six representative Tibetan villages captured during field surveys. By leveraging information recorded through direct observations, including architectural styles, vegetation conditions, infrastructure, education levels, and ecological management practices, we conducted a more in-depth investigation into the core characteristics of each village’s natural environment and social structure. The variations among these villages are influenced not only by their geographical and ecological conditions but also by a combination of external policies (such as infrastructure investments and environmental regulations) and internal cultural factors (including religious practices and community self-organization)⁷¹. Collectively, these factors contribute to the development of unique approaches to resource management and social governance in each village.

Environmental, social, and cultural characteristics observed on-site were cross-referenced with quantitative indicators, such as road density, architectural layout, forest coverage, per capita income, and community participation. Subsequently, the DPSIR framework was applied to evaluate the adaptability and resilience of each village across multiple dimensions⁷². This integrated approach enables both a macro-level quantitative analysis of the villages and the incorporation of micro-level insights from field observations, thereby elucidating the social and cultural mechanisms underlying the quantitative results. As a result, this approach offers a more comprehensive understanding of human–environment interactions and potential development pathways for Tibetan villages in western Sichuan⁷³.

Land use in the township where the village is located, 2015–2022.

Full size image

To gain a deeper understanding of the dynamics of the ecological environment and resource utilization in the townships where the villages are located, this study utilized multi-source data from 2015 to 2022 to develop Fig. 4a–f. This figure illustrates the distribution and spatiotemporal changes of forest land, water bodies, arable land, and construction land at a macro level, providing a solid basis for assessing the “State (S)” and “Pressure (P)” dimensions within the DPSIR framework.

The results indicate that, although overall land use types have remained relatively stable, significant local changes have occurred, primarily driven by local governance measures and policy directives. For instance, in the areas of ecological protection and smart agriculture, Muzuo Township (Minzu Village) and Xieluo Township (Jiangba Village) have experienced an increase in forest coverage due to logging bans and ecological management initiatives. Meanwhile, Shenzuo Village in Chali Township has improved the quality of arable land and enhanced agricultural productivity by establishing a river management organization focused on water resource monitoring⁷⁴.

Additionally, the progress of urbanization and infrastructure development is evident in certain regions, as construction land around Moluo and Kegeyi has expanded slightly. Field surveys also reported significant improvements in road networks, public facilities, traffic accessibility, and living conditions⁷⁵. According to government reports, the forest coverage in Moluo Village increased from 36 to 49%, suggesting that ecological protection and economic growth can coexist. In Kegeyi Village, the urbanization rate rose from 29 to 32%, agricultural output grew from 4.51 to 7%, and per capita disposable income increased from 9,623.96 CNY to 18,446.92 CNY (Sichuan Statistical Yearbook), highlighting the positive impacts of external investments and industrial upgrades on social development.

Figure 4 serves as a valuable visual reference for understanding land use changes at a regional scale, providing a foundation for the subsequent resilience analysis focused on environmental dimensions. By integrating micro-level field surveys with quantitative indicators, this study aims to investigate how these environmental changes interact with village policies, industrial structures, and community participation. This approach seeks to offer a more comprehensive assessment of the adaptive capacity and resilience of Tibetan villages.

Villages function as complex social–ecological systems, where human–environment interactions are influenced not only by agriculture, animal husbandry, and forestry but also by the advancing processes of urbanization and industrial diversification^7,73. As previously discussed, factors such as agricultural output, per capita arable land, and infrastructure development are subject to fluctuations driven by policy support, external investments, and community self-organization. The implementation of the Rural Revitalization Strategy and the Ecological Civilization Construction Plan has positively influenced human–environment dynamics in these regions, significantly mitigating challenges such as population decline, natural disasters, and resource overexploitation^76,77.

However, these factors are interdependent, requiring a balance between economic growth and environmental conservation, as well as addressing the diverse demands of social culture and community governance. To systematically examine how villages adapt and respond to environmental and social changes, the next section applies the DPSIR model to analyze the social–ecological systems of each village across five dimensions: Driving Forces (D), Pressure (P), State (S), Impact (I), and Response (R). This approach seeks to quantify the key challenges and opportunities facing the villages, providing a more targeted basis for planning and decision-making.

Assessment and validation

Evaluation based on the DPSIR model

This section provides a quantitative analysis and comparison of the development trends of the villages between 2015 and 2022 (See Fig. 5a, b). Using the DPSIR model, we calculated the scores for each indicator and level based on Eqs. (1) and (2). Subsequently, the comprehensive scores and their changes across the five levels for the six sample villages were compared (see Tables 2 and 3 and Appendix).

Composite score, 2015 & 2022. Note: The p value is 0.0058, indicating that the scores from 2015 to 2022 are statistically significant. The 95% confidence interval (CI) is [−0.0672, −0.0269], which suggests that the estimated change in score falls within this range.

Full size image

Figure 5(a) illustrates that in 2015, the scores of the villages across the DPSIR model dimensions—Driving Forces (D), Pressure (P), State (S), Impact (I), and Response (R) were relatively balanced. The scores for the Driving Forces (D) and Pressure (P) dimensions were moderate, suggesting that the villages managed to maintain a balance between resource use, environmental capacity, and economic development.

However, notable differences emerged at the State (S) and Impact (I) levels among the villages. Shenzuo and Minzu recorded the highest scores in these dimensions, likely due to their favorable natural conditions, abundant ecological resources, and relatively advanced infrastructure. Additionally, Shenzuo and Jiangba exhibited higher scores at the Impact (I) level, indicating that these villages were more significantly affected by external policies, market dynamics, and environmental changes.

Field surveys indicate that, in response to geological hazards and post-disaster reconstruction, local governments in Shenzuo and Jiangba actively implemented national policies to improve the living environment and ecological management. The adoption of the “Three-Life Space” strategy (production, living, and ecological spaces) enhanced spatial optimization and improved village conditions, contributing to higher State (S) and Impact (I) scores. However, the Response (R) scores were relatively low, particularly in Moluo and Minzu, indicating deficiencies in policy implementation, environmental management, and social adaptation, reflecting weak community governance⁷⁸.

From 2015 to 2022 (see Fig. 5b), the increase in Pressure (P) and State (S) scores suggests intensified resource and environmental pressures due to population growth, infrastructure expansion, and higher economic demands. This also reflects moderate improvements in the villages’ social, economic, and environmental conditions.

However, the decline in Impact (I) scores despite the rise in State (S) scores suggests a reduced influence of external policies and market factors, possibly due to diminished policy support, lower market volatility, or more stable internal development. Additionally, the slower progress in governance and social adaptation compared to economic growth and environmental changes has widened the gap between State (S) and Impact (I) levels, underscoring the need to strengthen governance capacities.

The decline in Driving Forces (D) scores, particularly in Xisuo, Kegeyi, and Minzu, suggests stagnation in industrial development and a weakening of economic vitality, likely due to population outflow and resource shortages. Meanwhile, the slight increase in Response (R) scores, though still low, indicates that policy implementation and governance systems remain insufficient, with incomplete mechanisms for absorbing and responding to external policies.

The rise in Pressure (P) and State (S) scores from 2015 to 2022 indicates improved adaptive capacities in social, economic, and environmental aspects but also reflects intensified resource pressures. This highlights the need for more effective governance strategies to manage these challenges⁷⁹. The decline in Impact (I) scores despite higher State (S) scores suggests that external policies and market factors have had a reduced influence on village development. This pattern implies that economic and social progress has not been fully translated into effective impacts, revealing a need to further enhance adaptive capacity.

The decrease in Driving Forces (D) scores implies a weakening of sustainable development momentum, underscoring the importance of industrial revitalization, population return, and optimized resource allocation. The consistently low Response (R) scores highlight deficiencies in policy implementation, environmental management, and social adaptation, emphasizing the urgent need to strengthen governance systems significantly⁸⁰.

Relying solely on static scores is insufficient for assessing long-term development potential. There is an urgent need for a dynamic resilience assessment to evaluate the recovery, adaptation, and transformation capacities of villages under external shocks and pressures⁵⁰. A resilience assessment can integrate ecological, economic, social, and governance dimensions to identify key factors influencing village resilience and to optimize sustainable development pathways.

Calculation of resilience indicators

Evaluating the resilience of villages is crucial for understanding their capacity to sustain development under complex and evolving conditions⁸¹. Given that different villages face distinct challenges, it is imperative to implement targeted adaptive strategies to enhance their resilience and stability⁸².

Changes in resilience in villages. Note: The p value is 0.0004, indicating that the scores from 2015 to 2022 are statistically significant. The 95% confidence interval (CI) is [−0.7129, −0.5951], which suggests that the estimated change in score falls within this range.

Full size image

Resilience comparison of different levels. Note: “D”, p = 0.0019, 95% CI [−0.0672, −0.0269]; “P”, p = 0.0000561, 95% CI [−0.1230, −0.0813]; “S”, p = 0.0000108, 95% CI [−0.2060, −0.1535]; “I”, p = 0.00000154, 95% CI [−0.2013, −0.1652]; “R”, p = 0.000122, 95% CI [−0.1318, −0.0808]. All p values < 0.05, significant changes occurred in all strata.

Full size image

Based on Eq. 3, we calculated the changes in Resilience (ΔR) for each village between 2015 and 2022 (see Fig. 6a, b). The results indicate that Resilience has declined across all villages, suggesting a weakening of adaptive capacity and implying that development efforts have been insufficient to counteract external shocks⁸³. In 2015, the Resilience scores were relatively balanced across the villages; however, by 2022, a general decline was observed.

Among the villages, Xisuo (−0.710) and Jiangba (−0.705) experienced the most substantial declines, whereas Minzu (−0.592) and Moluo (−0.595) showed comparatively smaller reductions. Overall, the rate of decline in Resilience exhibited spatial heterogeneity, likely influenced by a combination of factors such as economic structure, resource management, and policy interventions.

Further analysis based on the DPSIR framework (see Fig. 7) revealed that different villages exhibited distinct patterns of Resilience decline. Although the overall trend was downward, the Response (R) level demonstrated a modest increase in all villages, indicating improvements in policy governance and social adaptation capacity. However, this improvement in the Response (R) level was insufficient to counterbalance the declining trends observed in the Driving Forces (D), Pressure (P), and State (S) levels⁸⁴. Consequently, many villages continued to face challenges such as insufficient development momentum, industrial decline, and increasing pressures on resources and the environment. This imbalance ultimately prevented Resilience from maintaining a positive trajectory.

The following sections will examine the changes at each level in more detail:

• Xisuo Village (−0.710) and Jiangba Village (−0.705) exhibited the lowest resilience scores, suggesting that their economic adaptability and ecological carrying capacities are the weakest. This could be attributed to factors such as industrial decline, insufficient infrastructure, and limited resource utilization.

• In contrast, Shenzuo Village (−0.697) and Minzu Village (−0.592) experienced the most pronounced declines at the D–S levels (Driving Forces and State), indicating a lack of economic growth momentum and a reduction in resource and environmental adaptability. To address these challenges, it will be essential to prioritize industrial upgrading, ecological management, and policy support.

• Moluo Village (−0.595) showed a smaller decline at the P–S levels (Pressure and State), suggesting a relatively stable socio-economic foundation. However, the significant decrease at the D (Driving Forces) level may indicate issues such as population outflow, challenges in industrial restructuring, or limited economic growth potential. This highlights the need to strengthen industrial support and economic development strategies.

• Meanwhile, Kegeyi Village (−0.624) experienced a more substantial decline at the P–S levels, indicating bottlenecks in resource and environmental adaptability and economic development. Further improvement in policy implementation and governance systems will be necessary to enhance its long-term development potential.

Although the overall Resilience of all villages has shown a downward trend, this does not imply that their adaptive capacities are uniformly declining. The patterns of resilience observed within the DPSIR framework exhibit significant heterogeneity, suggesting that more complex nonlinear mechanisms may influence resilience. The formation process is influenced by a combination of policy, social, economic, and environmental factors, and the interactions among these variables and their threshold effects. This indicates that simple linear models may be insufficient for explaining these patterns.

Therefore, in subsequent research, we will apply the Generalized Additive Model (GAM) to examine the nonlinear impacts of multiple factors on Resilience change. This approach aims to identify key variables and their mechanisms of influence and to further quantify the contributions of each variable to Resilience change^85,86. By doing so, we seek to gain a deeper understanding of the mechanisms underlying the evolution of rural adaptive capacity.

Key factor identification

In this section, we integrated the datasets from 2015 to 2022, defining ΔR as the dependent variable and ΔX as the independent variable. Following the Generalized Additive Model (GAM) procedure, we performed variable selection, nonlinear modeling, and variable importance analysis in sequential stages. This approach is particularly practical for addressing complex nonlinear relationships and helps mitigate interpretation biases arising from linear regression methods⁸⁷. Consequently, it enhances both the scientific rigor and the interpretability of the analysis⁸⁸.

Nonlinear analysis process. Note: Variables with low correlation were excluded to reduce noise. Weakly correlated variables were filtered out. (List of explanation of variable names: X1 Gross regional product per capita(%), X2 Growth rate of total agricultural output(%), X3 Forestry coverage(%), X4 Percentage production value of agriculture(%), X5 Growth rate of agricultural output(%), X6 Urbanization rate(%), X7 Rural population, X8 Disposable income per capita, X9 Consumption level, X10 Arable Land Area per capita(hm²), X11 Food Possession per capita(%), X12 Rural Resident Population, X13 Rural Labor Force to Total Rural Population(%), X14 Cultural and educational level, X15 Infrastructure level, X16 Elevation and topography, X17 Land use and type, X18 Plow land cover(%), X19 Natural Resource Status Assessment, X20 Biodiversity, X21 Environmental quality, X22 Pollution Indicators, X23 Population growth rate(%), X24 Economic Indicators, X25 Social Development Indicators, X26 Quality of external investments(score), X27 Employment and Income level, X28 Social Service Facilities Level, X29 Employment Structure Environmental, X30 Management and policy support, X31 Environmental Impact Assessment, X32 Policy formulation, X33 Community participation, X34 Environment quality assessment, X35 Shaping the multifunctionality of rural space level)

Full size image

We first calculated the Pearson correlation coefficient (r) to assess the linear relationships between each independent variable (ΔX) and Resilience (R) (See Fig. 8a). Variables with significant correlations are indicated as either positive (red) or negative (blue). A threshold of |r| > 0.3 was set to identify key variables closely related to resilience (see Fig. 8b). While the Pearson correlation effectively captures linear relationships, it cannot detect nonlinear effects. Therefore, we applied the Generalized Additive Model (GAM) to explore nonlinear dynamics and identify the key drivers of Resilience⁸⁹.

Based on the Pearson correlation analysis, we selected 10 key variables for the GAM model (see Fig. 8c). Among them, positively correlated variables (r > 0.3), such as X8 (per capita disposable income), X27 (employment and income levels), and X9 (consumption level), are linked to economic adaptability and market vitality, suggesting that income growth and employment are crucial for enhancing Resilience⁹⁰. Additionally, X13 (proportion of rural labor) suggests that an adequate labor force can further boost economic resilience.

In contrast, negatively correlated variables (r < −0.3) such as X31 (environmental impact assessment), X5 (agricultural output growth rate), and X30 (policy management support) indicate that overuse of resources, strict environmental regulations, and excessive policy dependence may constrain Resilience. For instance, X6 (urbanization rate) suggests that accelerated urbanization could lead to population outflow, thereby weakening Resilience⁹¹.

In summary, economic growth, higher incomes, and stable employment emerge as the key drivers of Resilience, whereas environmental regulations, policy dependence, and urbanization may, under certain conditions, restrict Resilience. To further investigate these dynamics, we applied GAM modeling to explore nonlinear effects and identify potential tipping points, saturation effects, or thresholds.

For the GAM analysis, we employed the smoothing parameter (λ:0.5–4.2) using thin-plate regression spline as the smoothing basis. In addition, we employed the Akaike Information Criterion (AIC) to optimize the degrees of freedom (df: 1.3–5.7) of smoothing terms to prevent overfitting. To assess the robustness of the model, we performed a residual analysis and evaluated the goodness of fit (R²=0.91). This approach ensures that the model effectively captures nonlinear dynamics without compromising predictive accuracy⁹².

Pearson correlation analysis indicates that X8 (per capita income) and X27 (employment level) have the strongest positive correlations with Resilience (r > 0.8), identifying them as key drivers. In contrast, X31 (environmental impact assessment) and X5 (agricultural growth) exhibit significant negative correlations (r < –0.6), suggesting a long-term inhibitory effect on Resilience⁹³. Nonlinear trends (Fig. 8d) reveal the following patterns:

• Positive driving factors: For X8 (per capita income), Resilience increases with income but declines beyond an optimal range, indicating that both very high and very low incomes may weaken Resilience. For X27, Resilience rises rapidly when employment exceeds −2.5, showing a threshold effect, suggesting that adequate employment is critical.

• Nonlinear inflection effects (complex mechanisms): The impact of X9 (consumption level) on Resilience fluctuates significantly, implying potential interactions among multiple factors. The effects of X13 (labor force) and X11 (food supply) vary across different intervals, indicating that their impacts are not purely positive and may be influenced by socio-economic dynamics.

• Long-term inhibitory factors: X31 shows a consistently declining trend, suggesting that strict environmental policies might limit economic adaptability, especially in villages with fragile industries. The impact of X5 suggests that agricultural expansion may lead to overexploitation of resources, thereby weakening Resilience.

Overall, X8 (per capita income) and X27 (employment level) are identified as core variables for enhancing Resilience, yet their impacts exhibit threshold effects, both excessively high and low levels can lead to different outcomes. The effects of X9 (consumption level), X13, and X11 display multiple nonlinear inflection points, possibly due to interactions among various factors. In contrast, X31 and X5 appear to function as long-term inhibitory factors, suggesting a need for a precise balance between environmental policies and development^94,95.

The observed nonlinear trends in Resilience imply that some variables have threshold effects or nonlinear inflection points, indicating that traditional linear models may underestimate the complex impacts of these variables. However, correlation analysis and nonlinear trends alone are insufficient to confirm the significance of these variables in the model. To validate the results, we applied Lasso regression analysis to quantify the effects of key variables on Resilience and performed cross-validation to ensure the statistical reliability of the findings^96,97.

Results of Lasso regression analyses.

Full size image

Figure 9a presents the model coefficient plot, illustrating the strength and direction of each variable’s influence on Resilience (R) while also displaying the coefficient trends under both standardized and non-standardized frameworks. Lasso regression, employing an L1 norm penalty term, was used to shrink some coefficients toward zero, thereby achieving feature selection. We applied 5-fold cross-validation to determine the optimal regularization parameter (λ = 0.02) by minimizing validation error, thus balancing model complexity and fit quality. Additionally, Fig. 9b shows that the model’s overall explanatory power (R²= 0.947) suggests a good fit and a strong association between the variables and Resilience. However, further validation is necessary to ensure the robustness of the coefficient selection.

To validate the model further, we utilized the Neural Network Toolbox in MATLAB R2022b to evaluate each village’s resilience performance. This tool is particularly well-suited for managing complex nonlinear relationships and enhancing the accuracy of data predictions. Neural networks can identify intricate patterns within the data, capturing nonlinear interactions among key variables that influence resilience, which a single statistical model may not fully represent.

To improve the generalization ability and interpretability of the network model, we constructed a dataset based on the key variables, the comprehensive scores of the D-P-S-I-R framework, and the resilience scores and ΔR of each village for 2015 and 2022. The dataset was divided into a training set, validation set, and test set to facilitate model training, parameter tuning, and performance evaluation, respectively⁹⁸. This approach ensures that the model effectively captures complex nonlinear interactions without overfitting and provides a robust framework for resilience assessment⁹⁹. Then, we employ cross-validation and early stopping techniques to prevent overfitting to adjust the model’s hyperparameters, such as the number of hidden nodes and the learning rate, during the training process.

As shown in Fig. 10a, we initially developed a shallow BP neural network with an input layer (\(\left({x}_{1} 1 \ldots {x}_{n}\right),\) a hidden layer \({u}_{1}\ldots {u}_{m},\) and an output layer (y). The number of nodes in the input layer was aligned with the number of data features, while the output layer nodes were determined based on the prediction objectives¹⁰⁰.

Neural network simulation and validation.

Full size image

We selected the 10 key factors from the dataset as the input layer and used the scores for “D, P, S, I, R” and the Resilience (R) values for 2015 and 2022 as the hidden layer. ΔR was set as the output layer (see Fig. 10b). The dataset was randomly divided into three parts: 70% for the training set to train the model, 15% for the validation set to fine-tune model parameters and prevent overfitting, and the remaining 15% for the test set to assess model performance.

During model training, we employed the backpropagation algorithm to iteratively adjust the weights and biases of the neural network, aiming to minimize errors. By systematically adjusting hyperparameters such as the number of hidden nodes and the learning rate, we optimized the model’s performance. After repeated rounds of training and validation, the final Mean Squared Error (MSE) of the BP network was 1.9495, and the coefficient of determination (R²) reached 0.992 (see Fig. 10c–d). These results indicate that the model achieves high validation accuracy and exhibits good fit quality.

The neural network model in this study can effectively analyze and confirm the relationships among influencing factors, level scores, and Resilience values, providing a reliable tool for future research and practical applications¹⁰¹. In addition, in rural areas where the interactions among socio-economic, environmental, and policy factors are intricate and non-linear, this pooled approach is particularly well-suited for exploring complex, multidimensional resilience factors.

Resilience strategies and community governance

Based on the quantitative results of the model, the following table provides a summary of the key positive driving factors, negative driving factors, and overall trends for the six villages, as well as their similarities, differences, and distinct characteristics in terms of Resilience.

In summary, the findings in Table 4 indicate that multiple factors shape the Resilience of Tibetan villages. Positive influences such as ecotourism and income growth enhance Resilience, while negative factors like industrial monotony, resource pressures, and policy dependence pose significant challenges¹⁰². Although some villages have made progress at the Response (R) level, the declining trends at the Driving Forces (D) and State (S) levels underscore ongoing challenges related to industrial stagnation and resource management¹⁰³. The marked decline in Resilience observed in Xisuo and Jiangba highlights the limitations imposed by traditional agriculture and underdeveloped infrastructure on their economic adaptability.

The key challenge is to strengthen Resilience by optimizing resource use and industrial structures through a combination of community self-organization and collaborative governance, supported by effective policy measures. The following sections will explore pathways for self-organized governance models, focusing on geographical conditions, industrial structure, population and income, and cultural strategies.

As complex socio-ecological systems, Tibetan villages depend on the sustainability of their economies and environments, self-organizational capacity, and the effectiveness of local governance mechanisms¹⁰⁴. Previous assessments have shown that factors such as geographical conditions, industrial structure, population, income, and governance strategies have complex and layered impacts on Resilience. Addressing these challenges requires a targeted approach that integrates“government management + village self-organization,” tailored to the regional characteristics and ethnic cultural contexts. The following sections will detail these strategies.

1. (1)

   Adaptation strategies based on geographic conditions.

   The geographic conditions of Tibetan villages vary significantly, with altitude, topography, and climate directly influencing resource use and ecological carrying capacity. For high-altitude areas, enhancing ecological compensation and grazing subsidies is recommended, while river valleys should focus on advancing agricultural technology and ecotourism^105,106. Additionally, empowering communities to establish forest protection teams and ecological management groups can improve resource management and self-organization capacity. For instance, eco-management practices in Kamiza and Kegeyi villages have successfully preserved forest land and water systems. The synergy between supportive policies and community self-organization has significantly strengthened village resilience.

2. (2)

   Industrial structure and economic resilience.

   The industrial structure in Tibetan villages is relatively limited, constraining economic resilience. Xisuo Village and Minzu Village rely heavily on traditional agriculture, limiting adaptability. In contrast, Moluo Village and Kegeyi Village have diversified their economies by promoting ecotourism and handicrafts, resulting in a more resilient income structure.

   To address these challenges, providing targeted subsidies and tax incentives for ecotourism, handicrafts, and smart agriculture, along with promoting infrastructure development and market expansion, is advisable. Encouraging cooperative models and establishing benefit-sharing mechanisms through agricultural processing and e-commerce can further enhance economic resilience¹⁰⁷. For example, Jiangba Village has effectively boosted income and resilience through cooperative models. Integrating policy support with collective community economies can help diversify risks and strengthen resilience¹⁰⁸.

3. (3)

   Population return and community co-construction.

   Population outflow and income disparities have significantly undermined the resilience of Tibetan villages. Xisuo Village and Jiangba Village have experienced a substantial loss of young residents, which has weakened their economic adaptability. In contrast, Kegeyi Village has successfully improved income levels through the return of entrepreneurship and rural tourism.

   To counter these challenges, attracting returning entrepreneurs with subsidies, skills training, and tax incentives is recommended, focusing on specialty agricultural processing and digital village development¹⁰⁹. Additionally, encouraging communities to establish returnee funds could support specialty industries and skills development¹¹⁰. Combining returnee incentives with community funds can help mitigate population outflow, reduce income disparities, and enhance village resilience³⁰.

4. (4)

   Cultural integration and policy inclusiveness.

   Religious culture and traditional governance systems play a significant role in resource management in Tibetan villages. Although these factors did not score highly in the resilience assessment, their potential influence should not be underestimated¹¹¹. To enhance policy implementation, adopting a strategy of cultural identity + policy inclusiveness is advisable. Involving temples and village representatives in resource management can help build a collaborative governance model that integrates government + community + temples¹¹².

   Moreover, integrating religious culture with rural tourism can support specialized tourism industries. For example, Minzu Village has successfully merged traditional festivals with tourism, significantly boosting villagers’ incomes. Combining cultural governance with policy inclusiveness can strengthen cultural identity and improve policy effectiveness¹¹³.

   In summary, enhancing the resilience of Tibetan villages requires an integrated approach that combines “top-down policy support” with “self-organized community governance,” tailored to each village’s geographical environment, industrial structure, population, and cultural traditions. Key strategies include:

• Targeted ecological compensation and industrial support policies to improve policy precision¹¹⁴ .

• Community-based resource management and economic cooperatives to strengthen economic and social resilience¹¹⁵.

• A collaborative governance model that integrates cultural identity with policy inclusiveness to support both cultural preservation and modern development¹¹⁶.

Future research should focus on integrating dynamic data and conducting long-term tracking to assess the sustainability of different governance models. Additionally, exploring mechanisms for incorporating the perspectives of community and cultural leaders into policy implementation is crucial. Establishing a coordinated governance mechanism that blends top-down and bottom-up approaches would significantly support the sustainable development of Tibetan villages.

This study, using the SES-DPSIR framework, systematically assesses the impact of socioeconomic, environmental, and policy factors on the resilience of Tibetan villages in western Sichuan, revealing the underlying mechanisms of their resilience. By examining six representative villages, Guoluo, Keji, Xisuo, Shencuo, Minzu, and Jiangba, the analysis highlights how disparities in income, employment, environmental policies, and governance strategies significantly influence community resilience.

The main findings indicate that: (1) Per capita disposable income and employment levels are key positive factors for enhancing resilience. Economic stability and resource investments significantly strengthen communities’ adaptive capacities. This aligns with previous studies suggesting that a strong economic base directly improves a rural community’s ability to withstand shocks and invest in adaptive measures²⁹. (2) Over-reliance on top-down policies may be one of the constraints, e.g., subsidies and stringent regulations, which promote dependency and weaken a community’s ability to self-organize. The dual role of environmental policies—promoting ecological conservation while limiting economic activities, highlights the need for adaptive frameworks that balance environmental sustainability with economic growth²¹. However, while Lasso regressions identify policy-related variables (X31, environmental impact assessment) as inhibitors, it’s important to note that the causal relationship between policy dependence and local self-organization remains speculative. The current analysis is based on correlation rather than causation, and more empirical research is needed to clarify these relationships. (3) Furthermore, inclusive governance significantly enhances resilience^52,73,105. Villages with higher levels of rural community participation demonstrate stronger adaptive capacities, underscoring the importance of integrating local knowledge into governance processes¹⁰⁷. However, existing governance models struggle to balance policy support with local autonomy.

The findings suggest that a multidimensional policy framework is essential for coordinating economic, environmental, and governance strategies to enhance the resilience of Tibetan villages. To strengthen resilience, policies should address both external pressures (such as economic instability and environmental degradation) and internal capacities (such as community self-organization and local governance). The following specific policy recommendations are proposed:

• Economic diversification: Promote eco-tourism, handicrafts, and digital agriculture to create alternative income sources, mitigating economic risks and enhancing community resilience^74,111and helps communities adapt to the pressures of urbanization and environmental change.

• Adaptive environmental policies: Develop flexible regulations that protect ecosystems without excessively constraining local economic activities. These policies should be adapted to the local context, balancing ecological protection with sustainable economic development.

• Inclusive and adaptive governance: Strengthen community-based governance by establishing community funds and cooperatives to reduce reliance on external policies and enhance local self-governance.

• Targeted support strategies: Design differentiated policy support for villages in high-altitude areas and river valleys, adapting interventions to local environmental and socio-economic conditions to improve the accuracy and effectiveness of support measures.

This study demonstrates the effectiveness of the SES-DPSIR framework in capturing the multi-scale and dynamic characteristics of resilience, offering a comprehensive approach to assess the interactions between socio-economic, environmental, and governance factors. The findings emphasize the importance of integrating community governance with adaptive policies to strengthen resilience, providing valuable insights for policymakers in village contexts. However, while demonstrating the applicability of the SES-DPSIR framework, this study also highlights the need for further refinement of the framework to capture nonlinear interactions between factors that affect resilience over time.

Although this study provides valuable insights into the resilience mechanisms of Tibetan villages, it also has certain limitations. First, the analysis relies solely on cross-sectional data from 2015 to 2022, which constrains the assessment of long-term resilience dynamics and may affect the accuracy of trend analyses. Fixed time frame provides a snapshot of resilience dynamics but may not fully capture long-term changes or the cumulative effects of certain drivers. Future research should consider using longitudinal methods to monitor resilience changes over extended periods^13,49,63. Secondly, the linear assumptions of the SES-DPSIR framework may not adequately capture the nonlinear interactions that characterize complex systems, especially those between resource utilization and policy interventions¹¹⁷. This limitation may reduce the framework’s explanatory power in ecologically fragile regions, where interactions between ecological, social, and economic systems are more dynamic and complex¹⁵.

Additionally, while this study employs methods such as Lasso regression and Generalized Additive Models (GAM) to analyze the correlations between environmental, economic, and policy support factors and resilience, its ability to infer causality remains limited. The current analysis is predominantly correlation-based, making distinguishing between primary drivers of resilience and accompanying factors challenging^118,119. Consequently, future studies should consider incorporating mediator and moderator variables to clarify the causal pathways underlying resilience mechanisms. For example, there is a correlation between policy variables such as ecological compensation and the level of community self-organization. Future research should introduce causal mediation analysis methods, such as causal graph modeling, to clarify the direct and indirect action paths of policies in the resilience mechanism¹²⁰.

Notably, some short-term policy interventions, such as ecological compensation, can initially aid ecological restoration and income stability. However, in the long run, they might weaken the community’s endogenous adaptation ability, resulting in path and even institutional dependence. These potential consequences should be identified and mitigated through empirical research over a longer period. Conversely, economic diversification policies, such as the tourism cooperatives in Kegeyi Village, which boost villagers’ incomes, have the potential to enhance resilience without compromising local autonomy¹¹¹. Future research should differentiate between “hard” (regulatory/financial) and “soft” (capacity-building) policies to quantify their varying impacts on resilience pathways, particularly those exerted via cultural institutions like monasteries and art and handicraft workshops.

Moreover, future research should focus on accumulating data and conducting long-term monitoring, integrating dynamic data and longitudinal tracking to assess the sustained impacts of policy interventions. Using complex network analysis and machine learning techniques can help identify nonlinear relationships among key variables, thereby improving the accuracy of resilience assessments³. Expanding the geographical scope to include more Tibetan villages across different ecological regions could facilitate cross-regional comparisons and provide deeper insights into resilience patterns and key drivers.

It is also essential to consider cultural factors and self-organizing capacities in resilience mechanisms. Greater attention should be given to the roles of religious culture⁷⁸ and community self-organization⁵²which play a crucial role in local adaptive strategies. These factors may help to balance the current research’s emphasis on economic and environmental drivers by highlighting the importance of social capital and territoriality in increasing resilience.

Finally, adopting causal inference methods such as Propensity Score Matching (PSM) or Instrumental Variable (IV) approaches could enhance the accuracy of identifying causal links between influencing factors and resilience, helping to develop more targeted strategies for enhancing resilience in Tibetan villages^28,121.

In summary, this study elucidates the resilience mechanisms of Tibetan villages in western Sichuan, which are shaped by the combined effects of economic, environmental, and policy factors. It provides empirical support for more targeted and sustainable development strategies. However, the reliance on the SES-DPSIR framework’s linear assumptions and the use of short-term data limit the generalizability of the findings. Future studies should focus on longitudinal data collection and adopt nonlinear analytical methods to deepen the understanding of resilience mechanisms in ecologically fragile plateau regions¹²². Additionally, further research is needed to explore the integration of community governance with adaptive policy measures to enhance resilience.

The resilience of Tibetan villages in western Sichuan hinges on striking a balance among socioeconomic stability, environmental management, and adaptive governance. Insights derived from the SES-DPSIR framework suggest that economic development, policy support, and community self-organization are pivotal determinants of resilience. Nevertheless, challenges such as waning economic driving forces and escalating resource pressures highlight issues including population outflow and industrial monotony. Although policy responses have led to some improvements, they still fall short of effectively alleviating resource pressures.

Merely relying on economic growth is insufficient for substantially enhancing resilience. It is imperative to cultivate synergy between ecological conservation and social development. Promoting diversified industries, such as ecological agriculture and cultural tourism, can boost resilience by generating alternative income streams and mitigating economic risks. Strengthening community participation is also crucial for amplifying their role in environmental governance and policy implementation. Efforts should focus on refining ecological compensation mechanisms, fostering collaboration between local governments and communities, strengthening policy enforcement, and developing sustainable development strategies tailored to local circumstances.

Furthermore, long-term dynamic monitoring is advisable to assess the lasting effects of policy interventions. Incorporating social network analysis can offer valuable insights into how community relationships impact resilience. Cross-regional comparisons can help identify key driving factors that are both universal and context-specific. These measures can deepen our understanding of the spatiotemporal heterogeneity of resilience mechanisms, thereby providing a robust scientific foundation for more targeted policy-making.

In conclusion, this study enriches our understanding of resilience mechanisms in ecologically fragile plateau regions. It validates the efficacy of the SES-DPSIR framework in revealing the influence of multidimensional factors. Moreover, it presents empirical evidence and policy recommendations to underpin the sustainable development of Tibetan villages.