Ecological restoration in Tibet optimises the cognitive structures of stakeholders on social-ecological systems

Abstract

Facing the dual threats of climate and socioeconomic changes, how Tibet can seize the opportunity for ecological restoration to enhance environmental quality while improving the relationship between humans and nature is of great significance for regional sustainable development.Situated in an ecologically vulnerable area, the cognitive structure of Tibetan residents regarding resources and policies is a crucial reference for evaluating whether environmental governance is sustainable. We aimed to detect what impact ecological restoration in Tibet had on the cognitive structures of stakeholders.We distributed questionnaires to Tibetan residents, distinguished experimental groups (EG, n = 325) and control groups (CG, n = 165) by the implementation of ecological restoration projects or not, used a network approach to construct indicators of overall connectivity and evenness for assessing the cognitive structure and explored the influencing factors on cognitive structures. We found that interviewees in areas where ecological restoration was implemented had more positive perceptions of the importance of ecosystem services, the relationship between ecological conservation and well-being, attitudes towards ecological restoration, and the impact of governance measures. The overall connectivity and evenness of the EG were significantly higher than those of the CG. It showed an increase in positive stakeholder perceptions about policy and nature and a greater emphasis on the balance between environmental protection and productive life. The implementation of ecological restoration enhanced the connection between local people and the environment. Elevation and annual precipitation were the geographic influencing factors that dominated overall connectivity.Based on the cognitive structure of Tibetan residents, the implementation of ecological restoration enhanced connections in social-ecological systems elements, played a positive role in the human–nature relationship and promoted collective governance of common pool resources. Safeguarding and improving residents' livelihoods and enhancing the regional low-connectivity cognitive structure under different geographical contexts are important for future governance.

Read the free Plain Language Summary for this article on the Journal blog.

INTRODUCTION

Climate change and human interventions have intensified conflicts between humans and nature, increasing the vulnerability of complex regional social-ecological systems (SES; Hopping et al., 2016; Wang et al., 2018). The issue of how to make the coupled human and natural system highly adaptive to disturbances is of high interest at global and local scales (Cinner & Barnes, 2019; Grafton et al., 2019; Salomon et al., 2019). Adaptive governance, with the core concept of participatory governance, is an effective way to address changes in SES by taking measures to increase resilience, which in turn makes it easier to adapt to disruptions without collapsing SES (Chaffin & Gunderson, 2016). Adaptive governance systems are socially constructed, and their design and function are influenced by social perceptions, which requires attention to the potential for stakeholder participation in governance, especially by the public (DeCaro et al., 2017). For governance to achieve a sustainable outcome, stakeholders are required to understand policy and ecologically relevant information and construct a close understanding of the human–nature connection (Plummer et al., 2013). The structure of stakeholders' cognition about policy and resources can reflect stakeholders' preferences for benefits and their understanding of and readiness to adapt to change (DeCaro et al., 2017; Garau et al., 2021). Understanding this cognitive structure can help policymakers better assess the bridges and barriers to adaptive governance and avoid developing overly simplistic, one-size-fits-all solutions (Whittaker, 2020). It is also the key to getting out of the top-down and bottom-up collaborative dilemma in adaptive governance.

Ecological restoration is a typical form of adaptive governance implemented in ecologically vulnerable and easily disturbed areas. Ecological restoration is defined as ‘the process of halting and reversing degradation, resulting in improved ecosystem services and recovered biodiversity (Cole, 2021; Guan et al., 2019; Perring et al., 2015). It provides a good experimental site to study the changes in cognitive structure in adaptive governance processing (Garnett et al., 2018). Governance measures such as ecological compensation and ranger employment set up in ecological restoration reflect the concept of shared governance and play the subjectivity of stakeholders in environmental governance (Addison & Greiner, 2016). The cognitive structure of stakeholders is an important social factor in providing the resilient function of SES (Cinner & Barnes, 2019). Whether ecological restoration can achieve long-term adaptation effects rather than short-term recovery and whether it can be sustainable rather than stimulate new conflicts depends on a range of stakeholder perceptions and trade-offs about policy and resources (Bain et al., 2019; Hicks et al., 2016).

The cognitive structure of stakeholders that we emphasise, including elements such as perceptions and understanding of environmental change, preferences for the importance of ecosystem services and well-being (Sun et al., 2020), and perceptions of ecological restoration and livelihood aspects affected (Josephs, 2018), essentially refers to their understanding of the coupled social-ecological connections in complex adaptive systems (Cinner & Barnes, 2019; DeCaro et al., 2017). Most studies on cognition have been conducted to investigate the attitudes and values of stakeholders, while the characteristics of the way their views on particular issues are organised remain inadequately studied (Carmenta et al., 2017; Poortinga et al., 2019). The emergence of methods such as mental maps and networks has provided tools to describe the complex connections between series of cognitions (Aminpour et al., 2020; Felipe-Lucia et al., 2020; Janssen et al., 2006).

At the same time, the complexity of cognitive structures requires us to find the theoretical basis for understanding and structuring the various perceptions of stakeholders about environmental governance. The SES framework provides good theoretical guidance for analysing the cognitive structure of the public and the geographic influencing factors that regulate it during the implementation of ecological restoration (Ostrom, 2009). This framework provides an important theoretical basis for studying the structural properties, functional dynamics and adaptation pathways of SESs in small-scale communities. Although there may be problems with too many variables and complex processes in practical applications, the SES framework has helped scholars clarify the logical relationships of many variables in the analysis of SES processes (e.g. ecological governance activities; Cao et al., 2018; Li et al., 2021). The elements of governance in ecological restoration are often nested in a multilevel structure. With the SES framework, we can construct individual perceptions of the extent and trade-offs between these elements, identify important elements in cognitive structures and observe whether there is significant fragmentation or bias in stakeholders' perceptions of ecosystems and policies (Brehony et al., 2020). Moreover, the characteristics of cognitive structures constructed based on the SES framework are influenced by natural attributes, social attributes and rules, which are considered conceptual variables of the framework (Ostrom, 2009). How differences in these influencing factors influence cognitive structures, particularly the role of geographic factors, rarely receives deep discussion.

As the main part of the ‘roof of the world’, the Tibet Autonomous Region (referred to as Tibet afterwards) has greater exposure and vulnerability to climate change and human activities. In the past 20 years, the government has invested a total of 1.42 billion dollars in ecological restoration (The State Council Information Office of the People's Republic of China, 2018), such as grassland protection and restoration and protective forest construction engineering, aiming to protect and restore the fragile ecological environment and to protect the livelihood of residents (Addison & Greiner, 2016). Ecological restoration in Tibet is an enduring and unique practice, and it remains unclear whether it will enhance the resilience of the local SES. A coupled cognitive structure contributes to the success of ecological restoration governance (Hopping et al., 2016). Nevertheless, studies on the treatment effects of ecological restoration are mostly reported in small-scale ecological observations (Yan & Lu, 2015). Investigations of residents' perceptions are mostly found at the village and county scales. Studies that have been conducted on a large scale are still limited (Bain et al., 2019; Hicks et al., 2016). Whether the implementation of ecological restoration can effectively contribute to a more coupled cognitive structure of stakeholders, where it exposes risks that hinder long-term adaptation, and what the underlying influencing factors are, are still questions that need to be answered (Byg & Salick, 2009; Josephs, 2018; Tian et al., 2018).

To assess the impact of ecological restoration projects on the cognitive structure of stakeholders based on the SES framework, we described the differences in the cognitive structures of residents through interview questionnaires and network modelling. Both ecological and social variables were selected to quantify the contribution of geographic influencing factors to cognitive structures with the help of random forest regression. We hope to develop a method for integrating social factors that may influence the resilience of SES and control contextual variables of the SES framework in the adaptive governance process.

METHOD Study area and data Study area

Tibet is located on the main part of the Tibetan Plateau, which is an important part of the wildlife habitat of China. The average elevation is approximately 4000 m, with large elevation differences within the region (Shu-Qin et al., 2021). Numerous glaciers and permafrost have developed here. The major vegetation types are alpine meadow, steppe, forest, and desert steppe (Tian et al., 2018). There are vast natural pastures, and pastoralism is still the main local industry. However, there was a trend of pasture degradation on the Tibetan Plateau (Harris, 2010). To protect the pastures and improve the livelihood of local residents, the government has carried out a series of ecological restoration efforts (Bardgett et al., 2021). Since the 1990s, the government has implemented much ecological construction, including the release of various types of ecological compensation funds, restoring grazing lands to grasslands, water erosion control, the construction of key shelter forests, the development of nature reserves, and pest and rodent prevention and control. The key ecological project areas include the Qilian Mountains, Qinghai Lake, Three Rivers Source, and the basins of the Yarlung Tsangpo, Nujiang, Lhasa, Nyangchu, Yalong, and Shiquan rivers. After long-term efforts, a series of ecological restoration measures implemented on the Tibetan Plateau have yielded positive results, to some extent containing the degradation of the ecosystem. A total of 9.9% of the grassland enhanced vegetation index (EVI) showed a significant increasing trend (Wang et al., 2021), and the trend of desertification was initially contained. The forest area increased by 20.47 million cubic meters. Wildlife habitats have been improved, and populations of rare and endangered species have expanded, indicating that biodiversity conservation has achieved better outcomes (The State Council Information Office of the People's Republic of China, 2018). Ecosystem services on the Tibetan Plateau are gradually improving, especially water yield services in the northeast and habitat quality in the northwest (Hou, 2021).

Data sources Questionnaire survey

We conducted an extensive household survey in Tibet between August 2019 and May 2020. Stratified random sampling was conducted to capture changes in local grassland cover and the implementation of ecological restoration projects. Based on the average area of grassland per capita and the scope of implementation of ecological projects, we determined that seven prefecture-level cities in Tibet were selected for random household questionnaire surveys (Figure 1). Due to accessibility and the lack of population, some points were discarded. For example, southeastern Tibet was not surveyed because it was near the national boundary and was difficult to access. Finally, 43 counties were identified for extraction, including 2 in Linzhi, 3 in the Ali region and Changdu, 6 in Lhasa, 8 in Nagqu, 10 in Shannan and 11 in Shigatse. Our sample did not include cases of relocation villages, but rather focussed on surveys of local residents who have resided in the area for a long time. The interviewees were mainly local residents who spoke the Tibetan language and were usually the head of the household. The purpose of the survey was explained to the respondents prior to the survey as ‘to understand the views on ecological conservation in Tibet’. After obtaining prior informed consent, the questionnaire was completed by the team's Tibetan-speaking guide in a face-to-face interview. A total of 601 questionnaires were obtained from the survey, of which 490 questionnaires were valid after deleting incomplete questionnaires.

FIGURE 1. (a, b) Distribution map of interviewees in the control group (CG) and experimental group (EG); (c–f) ecological restoration projects in Tibet; (g) conducting household interviews. The people in the photographs who are interviewing others are members of the survey team, all of whom consented to the use of their photographs in the study.

Using the village as the basic unit for implementing ecological projects, we asked the respondents whether they had experienced ecological compensation projects such as restoring grazing land to grassland. If more than 60% of interviewees in the same village had experienced these projects, the village was considered to have implemented ecological projects. The interviewees located in the village without the implementation of ecological restoration projects were divided into a control group (CG, n = 165), and others were divided into an experimental group (EG, n = 325). We conducted structured interviews with them.

The content of the interview involved the basic information of the interviewees, including age, annual income, livelihood sources and pasture area. Knowing the information can help us understand the interviewees' personal experiences and economic and pasture conditions. The foundational information was used in the subsequent calculation of social factors that influence the contextual variables of cognitive structure.

To quantify the cognitive structures of the interviewees, we investigated the following questions. We were interested in knowing whether the interviewees had observed changes in the surrounding environment to briefly assess whether the ecological restoration project had achieved some effect in the local area and therefore asked the interviewees if they had observed changes in grasslands, water, forests, wetlands, temperature and precipitation over the last 20 years. We also aim to understand interviewees' needs for ecosystem services and environmental quality by asking them to assess the importance of key ecosystem services on the Tibetan Plateau and to identify whether the various goals of improving the ecosystem would enhance their happiness. Finally, we also needed to know the interviewees' perceptions of ecological restoration projects and governing measures. We asked the interviewees who had experienced ecological projects about their perceptions of the impact of ecological projects and specific measures on their livelihood and environment, as well as their attitudes towards ecological projects. For respondents who did not experience ecological engineering, we set scenario questions asking them to answer specific ecological restoration measures, such as the impact of grazing exclusion on livelihoods and the surrounding environment. The subjective questions were coded using the Lister Five-Level Scale (Jamieson, 2004). The complete questionnaire can be found in Appendix S1. The ethical nature of the above survey protocol was approved by the Second Tibetan Plateau Scientific Expedition and Research office. We conducted anonymous interviews, where the interviewees were informed of the researchers' identities and were assured that the interview content would be solely used for scientific research purposes. Interviews regarding the environmental effects of ecological engineering interventions were carried out only with the consent of the interviewees. As anonymous interviews were conducted, to ensure the privacy of the interviewees, we did not ask them to sign a written informed consent form, but obtained their verbal consent before conducting the interviews.

Vegetation data

To calculate the influence of vegetation status on cognitive structures, we chose EVI as the contextual variable of cognitive structure to characterise the vegetation status of the villages where the interviewees were located. The EVI for 2000 and 2018 with a resolution of 1000 m was collected from the Moderate Resolution Imaging Spectroradiometer products (MOD13A1, Version 6). We aggregated max monthly values to obtain monthly resolution data by the maximum value compositing method (Holben, 1986). The mean values of the growing season were used to refer to the current year's vegetation status. The 2-year EVI difference indicated the change in vegetation. Meanwhile, the EVI within 2018 was used to indicate the current vegetation status.

Meteorological data

To calculate the influence of recent precipitation and temperature on the cognitive structure, annual precipitation and annual mean temperature from 2000 to 2017 with a resolution of 1000 m were selected to represent climate conditions for the contextual variables of cognitive structure. Both precipitation and temperature data were obtained from the monthly precipitation and temperature datasets over China during 1901–2017 (Peng, 2019).

Elevation data

Similarly, elevation was also selected as a geographic influencing factor or contextual variable of cognitive structure. We obtained the data with a resolution of 1000 m from the Global 30 Arc-Second Elevation DataSet (https://webmap.ornl.gov/ogc/dataset.jsp?ds_id=10003).

Evaluating cognitive structure Framework and definition

The SES framework proposed by Ostrom is a classic framework for diagnosing the sustainability of coupled human and natural systems, showing advantages in analysing complex, multilevel, multivariable, nonlinear interactions (Brehony et al., 2020; Ostrom, 2009). There are four core subsystems in the framework: resource system (such as grassland), resource unit (such as grass), governance system (such as governance measures) and users (such as farmers and pastoralists). We use the framework to organise the various components and variables to organise the network and structure residents' cognitions about policy and resources.

A network approach is a powerful tool for establishing complex connections among variables nested at different levels (Hicks et al., 2013). We can describe the characteristics of each interviewee's cognitive structure by calculating the indicators of network characteristics, such as node connectivity and evenness and overall connectivity. For ease of understanding, we display the relevant concepts involved in building indicators in Table 1. We used overall connectivity and evenness as structural indicators.

TABLE 1 Definitions and meanings of different concepts.

Type	Definition	Cognitive structure meaning
Node	Elements used to construct network structures, such as grassland management projects and water and soil conservation ecosystem services. Nodes can be a policy, a type of environmental change, and an ecological conservation goal, consistent with the core research content of Question Nos. 5–9 of the questionnaires.	The basic elements that make up the cognitive structure can be divided into four categories according to the SES framework: resource system, resource unit, user, and governance system. The details of the node setup are given in Table 2.
Edge	Connection between two nodes, obtained by questionnaire survey, such as the perception of water erosion	Connection between the different components; the size of the edge is the weight. Here, the score of the questionnaire item represents the weight.
Connectivity	Connection degree of the node in the connections, measured by weighted centrality	Degree of connection between an element and relevant nodes. High connectivity of nodes indicates a high number and strength of connections to surrounding nodes in the cognitive network. Nodes with high connectivity are important elements for building cognitive structures.
Evenness	Describes uniformity of the connection weight among nodes	Homogeneity of cognitive structure. High evenness indicates that interviewees' perceptions of the connection between policy and nature are not notably biased. Low evenness indicates that perceptions are biased in some aspects.
Overall connectivity	Mean weighted centrality of all nodes, calculating the closeness of all nodes	Overall connectivity of cognitive structure. High overall connectivity indicates that policy and nature are strongly linked in interviewees' perceptions. Interviewees are aware that the implementation of policies affects nature.

The construction of network

Based on the framework, we set the components of the analysis as the resource system, resource unit, governance system and user. According to Ostrom's list of secondary variables, the four types of factors were further subdivided into the variability of resource units, the existing conditions of common pool resources, governance rules, the importance of SES to users and knowledge about SES (Ostrom, 2009). The nodes of the network were set up according to the elements of the four categories of factors associated with Questions 5–9 in the questionnaire, with a total of 20 nodes. The details of node selection are shown in Table 2.

TABLE 2 Nodes selected when building indicators.

Component type	Framework meaning	Element	Specific node
Governance system	Operational rules	Policy	Grassland protection and restoration engineering (ENG)
			Protective forest construction engineering (ENF)
			Prevention and control of desertification engineering (END)
			Water erosion control (ENW)
		Governance Measure	Livestock reduction (ML)
			Defarming (MD)
			Enclosure (ME)
Resource unit	Variability of resource units	Variation	Land desertification (VD)
			Groundwater decrement (VG)
			Wildlife reduction (VW)
			Water erosion (VE)
			Pasture deterioration (VP)
Resource system	Existing conditions of common pool resources	Quality	Pasture quality (QP)
			Water quality and quantity (QW)
		Ecosystem service	Providing agriculture and pasture products (ESP)
			Water regulation (ESW)
Users	Importance of SES to users	Demand	Livelihood demand (DE)
		Feeling of happiness	Happiness brought about by obtaining resources (HB)
	Knowledge about SES	Cognition	Views on governance strategy (CO)
		Perception	Perception of environmental change (PE)

The edges are obtained by asking questions in the questionnaire. The weight of the edges is the score of the question item. Meanwhile, we tried to simplify the building process by setting the following rules to contribute to understanding. In our questionnaire set-up, the networks that experienced ecological projects will have more connections than networks that did not (dashed line in Figure 2a). However, to prevent the variation in indicators due to the difference in network construction, the network structures of the CG and the EG were consistent (solid line in Figure 2b). In the cognitive networks of the CG and EG, the number of nodes connecting the same node was the same. To ensure the robustness of the results, we computed the results for both the EG and CG networks with the same structure and the EG network with the addition of two kinds of edges (dashed lines in Figure 2a). We set five kinds of 15 edges in total to quantify the interaction between different elements (Figure 2c).

FIGURE 2. Constructing evaluation indicators based on the social-ecological systems framework. (a) The framework used for the study; (b) schematic diagram of the evaluating indicators; (c) Network constructed through Questions 5–9 of the questionnaire. ① Interviewees' attitudes towards ecological restoration (according to Question No. 8.3); ② Interviewees' perceptions of the impact of the governance measures on livelihoods (according to Question No. 9.1); ③ Interviewees' perceptions of the impact of governance measures on environmental change (according to Question No. 9.2); ④ Changes in residence perceived by interviewees (according to Question No. 5); ⑤ Interviewees' demand for ecosystem services and quality (according to Question Nos. 6 ~ 7); ⑥ Impact of implementing ecological restoration on interviewees' livelihoods (according to Question No. 8.1); and ⑦ Impact of the implementing ecological restoration on the interviewees' surroundings (according to Question No. 8.2).

Meanwhile, we tried to simplify the building process by setting the following rules to contribute to understanding. If the connectivity in the EG is higher than that in the CG, the score of the question item related to that node has increased after the implementation of the ecological restoration. When some edges are enhanced, the connections between the component types (according to Table 2, e.g. users) to which the nodes belong will be enhanced. For example, when the connection between node ENG and node DE is enhanced, then the connection between the resource unit and user is also enhanced. In addition, to simplify the analysis model, the interaction between the same component type of node would not be calculated, and the default would have a certain synergistic effect. For example, if a user's perception of environmental change was strengthened, his or her knowledge of the connection between humans and nature would also increase, which might promote the recognition of rules and eventually enhance the governance effect. Therefore, we calculate the connectivity of a node by observing the connection between nodes and then analysing the impact of the change in this node on the overall governance effect.

Network metrics

Connectivity, also known as strength, measures the importance of nodes in cognitive structures (Barrat et al., 2004). Overall connectivity is the mean connectivity of all nodes. We used the strength function of the igraph R package to calculate connectivity and overall connectivity as follows:[Image Omitted. See PDF]where $\mathrm{CO}{\mathrm{N}}_i$ is the connectivity of node i, $A_{\mathit{ij}}$ is the adjacency matrix, and $W_{\mathit{ij}}$ is the weight of the edge.

Evenness measures the evenness of the edges (Shannon, 1948). We used the Vegan R package for calculation. The formulas of evenness are as follows:[Image Omitted. See PDF][Image Omitted. See PDF]where EVEN is the evenness of the connections, s is the number of edges, and $p_i$ is the proportion of edge i.

The impact of geographic context on interviewees' cognitive structures

After evaluating the interviewees' cognitive structures, a series of variables that lead to interactions and choices by influencing the outcome of action situations were further calculated. We divided these external variables into natural attributes, social attributes and rules. We used EVI within 2018, annual precipitation, annual mean temperature and elevation at the interviewee villages as the natural attribute variables of geographic influencing factors. The social attributes were the annual income, livelihood sources and pasture area of the interviewees, which were key community characteristics used to describe them. The rule is whether there is an ecological restoration project implemented.

We used a random forest (RF) algorithm to analyse how the above variables affect the overall connectivity and evenness. RF is a machine learning algorithm based on a classification tree (Breiman, 2001). Each tree fitting in the forest uses a bootstrapped sample to train observations. Unused portions of the dataset can be used to test the tree prediction (Zhang et al., 2021). This helps RF to be fitted and validated when being trained without an extra independent validation dataset (Schwalm et al., 2017). RF can judge the importance of features through a fast training speed, which can also be used for regression. In our work, each random forest consisted of 450 regression trees with no less than four leaf nodes. We used the increase in mean squared error (IncMSE) to assess the importance of variables, that is, by randomly assigning values to each predictor variable; if the predictor variable was more important, then the error of the model prediction would increase after its value was randomly replaced. Therefore, a larger value indicates greater importance of the variable. The above calculations used the randomForest R package.

RESULTS Statistics of the questionnaire

We collected basic information about the interviewees and the vegetation status of the villages where they lived (Figure 3). The age distributions of the CG and EG were similar, with the largest proportions of interviewees aged 40–59 at 64.2% and 58.3%, respectively. The difference in annual income between the CG and EG was not significant (p = 0.539). However, EG has a higher income preference, with the largest percentage of people earning from 2966 to 7415 USD per year at 43.8%. Among CG, the greatest number of interviewees had three types of livelihood sources, at 35.3%. The greatest number of interviewees in the EG had four kinds of livelihood sources, at 44.4%. The area of pasture owned by 73.9% of CG interviewees was less than 13.3 ha. However, a larger proportion of EG interviewees had larger pastures, with 26.2% having more than 66.67 ha. The difference in EVI values between CG and EG inhabited villages was significant (p = 0.0035), and the EVI decreased more in CG than in EG.

FIGURE 3. Basic information of the interviewees and the situation of the control group (CG) and experimental group (EG) in the village. (a) Composition of respondents' ages. (b) Composition of respondents' total annual household income. (c) Quantitative statistics of respondents' livelihood types. (d) Size of respondents' household pasture. (e) Annual average enhanced vegetation index (EVI) of interviewees' residence. (f) Variation in EVI of interviewees' residence over years.

We showed questions with preference differences in perception and cognition in Figure 4 and Figure A1. Interviewees who did not experience ecological projects observed more land desertification, wildlife decline, and water erosion around their residences (Figure 4a–c). The reduction of groundwater is a common environmental problem (Figure A1A). When we asked about the environmental changes around their residence in the last 20 years, 15.1% of the EG interviewees felt that water erosion occurred less frequently, while 45.5% of the CG interviewees felt that water erosion occurred frequently (Figure 4c). A good sign is that respondents are generally aware of the importance of ecosystem services, such as the supply of agro-pastoral products and water regulation (Figure A1B,C). Additionally, interviewees considered the implementation of ecological projects on the Tibetan Plateau as helping enhance their happiness (Figure A1D). In our survey, we asked interviewees to think about the impact of ecological conservation on happiness for specific ecological conservation goals. The percentage of interviewees in the CG who considered that happiness had little to do with ecological conservation was 20.0%, but the percentage in the EG was 8.31% (Figure 4d; View 1 means that interviewees believe that happiness in life is very important; View 2 means that interviewees feel happy when there is no significant financial loss; View 3 means that interviewees feel a small amount to enhance happiness; View 4 means that interviewees consider that the realisation of ecological goals has little impact on happiness; and View 5 means that the realisation of ecological goals is not the source of happiness). Views on ecological restoration also differed among interviewees who had or had not experienced the implementation of ecological restoration. A higher proportion of interviewees who had not experienced ecological restoration projects believed that the measures of livestock reduction and defarming had a strong impact on production and livelihood (Figure 4e,f). As we asked the interviewees to judge the impact of defarming, 37.0% of the CG interviewees thought it would have a greater impact on their lives, yet this percentage was only 19.7% in the EG (Figure 4f). In general, most of the interviewees supported the implementation of ecological restoration (Figure A1E–G). However, those who were opposed to the prevention and control of desertification engineering were fewer in CG than in EG, with proportions of 7.9% and 13.5%, respectively (Figure 4g). Regarding whether the enclosure would improve the environment, 51.4% of interviewees in the EG thought it would, but 45.5% of CG interviewees thought it would not (Figure 4h).

FIGURE 4. Perception of interviewees in the control group (CG) and experimental group (EG) in the village. (a) Frequency of occurrence of land desertification in the last 20 years; (b) frequency of occurrence of wildlife decline in the last 20 years; (c) frequency of occurrence of water erosion in the last 20 years; (d) relationship between achieving the ecological goals of water quality and water quantity guarantee and happiness; (e) impact of the measure on livestock reduction; (f) impact of the measure on defarming; (g) attitude towards water erosion engineering; and (h) perceptions of whether measures improve the environment.

Differences in cognitive structure

All evaluating indicator values in the EG were higher than those in the CG (Figure 5). The mean value of overall connectivity in the EG was 4.72, while that in the CG was 4.61. Independent t-test analysis showed that the overall connectivity values were higher in the EG than in the CG (p = 0.062, Figure 5a). This indicated that the implementation of ecological restoration as a whole improved the interviewees' scoring on the relevant questions. The cognitive structure of interviewees was enhanced in terms of perceptions of ecosystems and policies. The comparison of the evenness also exhibited the same characteristics. The mean value of evenness in the EG was 0.96 and that in the CG was 0.97. The evenness values in the EG were significantly higher than those in the CG (p = 0.01, Figure 5b), indicating that the originally dominant connections were weakened. EG's cognitive structure was less biased, meaning that there was less bias towards the perception of policy versus nature. In addition, we calculated the evaluating indicators after adding edges ⑥ and ⑦ (Figure 1) and obtained consistent results (Figure A2). Both the overall connectivity and evenness in the EG were significantly higher than those in the CG (p < 0.001).

FIGURE 5. Evaluating indicators of the control group (CG) and experimental group (EG). (a) The overall connectivity of the cognitive network. (b) The evenness of the cognitive network.

To analyse how the implementation of ecological restoration enhanced overall connectivity and whether it exposed obstructions in governance, we then calculated the connectivity of specific nodes in the two groups. We found 11 nodes whose connectivity did not differ significantly between the two groups (Figure A3). By comparing nodes with significant differences in connectivity (p ≤ 0.05), we found interesting divergences (Figure 6). For nodes VD (land desertification), VG (groundwater decrement), VW (wildlife reduction), VE (water erosion), ESW (water regulation) and PE (perception of environmental change), their connectivity was significantly greater in the EG than in the CG, meaning that the implementation of ecological restoration strengthened their connectivity and elevated their importance. The connectivity of the EG's perceived environmental change type nodes increased significantly, indicating that interviewees were less likely to perceive phenomena such as land desertification. They positively perceived changes in the surrounding environment. Similarly, the enhanced connectivity of ESW and PE indicated that EG interviewees scored higher on the corresponding questions, placed more emphasis on ecosystem services such as water quality and quantity guarantee in Tibet and recognised a greater effect of ecological restoration on environmental improvement. We also noted that for nodes DE (livelihood demand), ML (livestock reduction) and MD (defarming), their connectivity was significantly smaller in the EG than in the CG, indicating that the ecological restoration implementation weakened their connectivity. This result showed that the expectations of ecological restoration for livelihood improvement have not been met, as the scores for questions involving the improvement of ecological restoration on livelihoods dropped, while interviewees who had experienced ecological restoration considered that livestock reduction and defarming had less impact on their livelihoods.

FIGURE 6. Connectivity of specific nodes of the control group (CG) and experimental group (EG). (VD means land desertification; VG means groundwater decrement; VW means wildlife reduction; VE means water erosion; ESW means water regulation; PE means perception of environmental change; DE means livelihood demand; ML means livestock reduction; and MD means defarming.)

Importance of geographical influencing factors

The results of the random forest regression indicated that geographic influencing factors dominated the overall connectivity in the CG and EG (Figure 7). Among them, the variable explanation rate of the regression model for the overall connectivity in the CG was 58.8%, while that in the EG was 51.7%. For evenness, the regression models for CG and EG had lower variable explanations at 28.57% and 11.95%, respectively (Figure A4). We considered that the regression model construction for evenness was poor, so further analysis of influencing factors affecting overall connectivity was conducted in the subsequent analysis. For CG, the effect of elevation on overall connectivity was the greatest, with an IncMSE of 27.8%, followed by annual precipitation (IncMSE = 24.9%) and pasture area (IncMSE = 22.2%). It was noted that the interviewees' annual income also had an effect on coupling, with an IncMSE of 17.6%. The overall connectivity of the EG was more influenced by geographical factors. Similarly, the effects of elevation and annual precipitation were the largest, with IncMSEs of 31.7% and 31.0%, respectively. The effects of social factors were relatively weak, such as annual income and sources of livelihood, with IncMSEs of only 10.4% and 13.5%, respectively.

FIGURE 7. Importance of the influencing factors, including elevation, precipitation, enhanced vegetation index (EVI), temperature, pasture area, income and livelihood type, of the connectivity in the control group (CG) and experimental group (EG).

DISCUSSION Governance performance and exposed issues

In general, the ecological restoration implementation led to a more pronounced perception of ecological improvement among EG interviewees, as well as greater support and acceptance of a range of restoration projects and treatment measures, enhancing the overall connectivity and evenness of cognitive structures. The increase in overall connectivity indicated that the average scores of EG interviewees were higher for perceptions of environmental changes and perceptions of the significance and effects of the project, and the increased evenness indicated that EG interviewees scored different questions more similarly. The residents more closely and roundly connected with the environment and recognised the importance both of ecosystem services on the Tibetan Plateau and the role of ecological restoration. There were 11 nodes with insignificant differences in connectivity, resulting in small differences in overall connectivity. By comparing the nodes with significant and nonsignificant differences, we found that the interviewees of CG and EG were approximately the same in some perceptions, such as those of grassland restoration projects, protective forest projects and desertification control projects. Local perceptions play an important role in promoting joint responses towards sustainable management of natural resources (Fernández-Llamazares et al., 2016). The perception of environmental changes can be considered a kind of tacit and situational knowledge (Pyhala et al., 2016). Based on such perceptual information, local resource users will evaluate whether changes warrant some kind of response, including a shift in management strategy. When communities and ecosystems are poorly connected, reduced natural resource availability may not be seen as a public problem, and the efficiency of collective action will be diminished (Jacob et al., 2021). Our findings suggested that the cognitive structures between Tibetan community residents and the environment were strengthened. This meant that local ecological knowledge was updated and expanded, which could facilitate the process of spontaneous ecological conservation by users, enhancing their ability to adapt to climate and socioeconomic changes (Klein et al., 2014). Therefore, the results provide a positive signal for evaluating the governance performance of ecological restoration.

Although we found some encouraging results in the elevation of the effectiveness of ecological restoration governance, we should still be wary of the governance risks exposed in the findings, as users' livelihoods remained somewhat vulnerable. The livelihood status of resource users is a critical aspect reflecting the socioeconomic benefits of governance policies and is vital to achieving ecological restoration goals, and usually vulnerable in the pasturing area of developing countries. For example, China's Grassland Ecological Compensation Policy increased pastoralists' incomes but increased income inequality among local pastoralists (L. Hou et al., 2021). A decline in livestock numbers and productivity affects pastoralists' livelihoods (Nyima & Hopping, 2019). As a reminder, the improvement of grassland conditions may carry a downside; while it improves the environment (Zhang et al., 2018), it still comes at the cost of the livelihoods and well-being of some pastoralists. Livelihood decline and environmental degradation are two interrelated and mutually reinforcing issues (Waldron et al., 2010; Wang, Liu, Liu, et al., 2023; Wang, Liu, Wu, et al., 2023). Low incomes, for example, tend to force resource users to make greater use of the environment, often leading to excessive grazing pressure and degraded grasslands. However, degraded grasslands struggle to provide adequate ecosystem services, increasing the cost of resource use to households (López-i-Gelats et al., 2016). This contradicts the original intent of ecological restoration and may exacerbate the negative feedback loop of livelihood decline and environmental degradation (Addison & Greiner, 2016). Current ecological governance in Tibet has achieved initial results, but sustainability has not met expectations. Therefore, it is still necessary to explore a more moderate and flexible governance approach, considering the actual needs of resource users.

Implications of dominant factors for optimisation

Overall connectivity indicates how well connected the residents are to the policies and environment in their cognitive structure, which means that users can more easily access information on the status of the resource system, the variability of resource units, and the rules of governance and adjust to them. Our study found that for the cognitive structure of Tibetan interviewees, overall connectivity was more influenced by natural factors. To further analyse the influence of geographical influencing factors on overall connectivity and the implications for the ongoing improvement of ecological restoration, we selected the two factors of elevation and precipitation in the top order of importance and used quartiles to make a split box display. As shown in Figure 8, the variation in CG connectivity with both elevation and annual precipitation decreased to quantile 3 (approximately 4100 m) and then increased. However, the overall connectivity in the EG varied more moderately between different gradients of elevation and annual precipitation.

FIGURE 8. Changes in overall connectivity with elevation (a, c) and precipitation (b, d) of the control group (CG) and experimental group (EG). The x-axis is the quartile of elevation and precipitation.

The use of resources by residents in Tibet varies with elevation. In the elevation range of 3000–4100 m (i.e. quantiles 1–3), the livelihood of residents is mainly grazing supplemented by farming. The proportion of grazing as a livelihood increased with elevation (Wang et al., 2016). In other words, residents are more dependent on local ecosystem services. Overall connectivity in CG also showed a decreasing trend with the increase in elevation in the region, suggesting that in ecologically fragile areas, the increasing degree of residents' resource utilisation and development may aggravate the opposition of resources and policy in stakeholder cognitive structures. Another study also observed that the situation of grassland protection at an elevation of approximately 4000 m was still dire (Wei et al., 2021). Overall connectivity in the EG changed more gently in this elevation interval, indicating that the implementation of ecological restoration effectively promoted collective awareness and rational management of natural resources (Hicks et al., 2016). The change in overall connectivity with annual precipitation also showed a similar pattern. Ecological restoration in Tibet is still ongoing, and our findings provide us with the directions for the future optimisation of governance. First, corresponding to the regions with low overall connectivity, the alpine steppes with elevations of approximately 4000 m and the semiarid regions with annual precipitation of 400–500 mm are still the regions that require optimisation. Second, attention also needs to be paid to improving management approaches (Adams et al., 2019), focussing more on the social and economic outcomes of restoration (Wortley et al., 2013) and raising the weak SES coupling relationships due to geographical contexts.

The effectiveness of ecological restoration projects aims to achieve harmony between humankind and nature, but the outcomes of China's restoration efforts may not always result in co-benefits (Liu et al., 2023). Cao et al. (2021) estimated that the investment return rate of China's Three North Shelter Forest System Project and the Grain-for-Green Project is below 100%, expressing concerns about the insufficient economic benefits of some of the projects. Li et al. (2015) describe a Relocation and Settlement Program in Shaanxi Province in which poor households are unable to participate in the benefits due to difficulties in paying significant upfront costs. Wang et al. (2020) found that ecological conservation policies in the Three River Region did not significantly improve the livelihood assets of local people. It has been a challenge for ecological restoration projects to incorporate the actual needs of local people and to realise ecological restoration while avoiding damage to the livelihoods of local people and benefiting as many stakeholders as possible. Therefore, the implementation of ecological restoration projects at the current stage needs to adopt a co-evolutionary view on SES to promote the translation of ecosystem management benefits into social, economic and policy incentives, so that local people benefit from restoration projects and are motivated to support bottom-up adaptive management rather than always accepting policies in a top-down and passive manner (Fu et al., 2023).

Our findings emphasise the significance of community-scale resident perceptions of the improvements in rural ecosystems achieved by ecological restoration projects in Tibet. Previous assessments of ecological restoration projects in China have paid limited attention to the perceptions of local residents and may have neglected their interests and well-being. Our survey also found that livelihood-related nodes such as DE, ML and MD had low connectivity in EG, as residents did not score questions about ecological restoration for improved livelihoods as highly as expected. This is a further indication when carrying out ecological projects in ethnic minority areas, where residents have unique living environments and cultural practices, it is necessary to strengthen community surveys. We emphasise the significance of community-scale resident perceptions on the improvement of rural ecosystems by ecological restoration projects in Tibet. This bottom-up surveys can be used as a complementary method for planning ecological restoration projects in Tibet.

According to research limitations, due to the complexity of Tibet's natural environment and the inaccessibility of some counties, our study has not yet been able to conduct long-term follow-up of the questionnaire. It remains unclear how the length of time that ecological restoration projects are implemented affects the cognitive network and the effectiveness of restoration. Also, the intensity of ecological restoration projects implemented in different regions varies, but the impact on the results is difficult to quantify.

CONCLUSION

Tibet faces the dual challenges of climate and socioeconomic changes, and the government has implemented a series of engineering projects to achieve ecological restoration. We found that the Tibetan interviewees in the EG had more positive perceptions of the importance of ecosystem service, the relationship between ecological conservation and well-being, attitudes towards ecological restoration and the impact of the governance measures. Overall connectivity and evenness were higher in the EG than in the CG. It was notable that the implementation of ecological initiatives did enhance the connection between local residents and the environment but also undermined some livelihoods. This reminds us that while the governance of ecological restoration has now reached a new stage, it still faces challenges in how to achieve long-term, sustainable governance, and there is a need to explore more flexible governance approaches that incorporate the actual needs of resource users. Elevation and precipitation are the geographical influencing factors that dominate the overall connectivity of cognitive structure in Tibet. It will be necessary to focus on low overall connectivity areas in alpine steppes with elevations of approximately 4000 m and semiarid areas with annual precipitation of approximately 400–500 mm.

Although we proposed a method to construct a cognitive structure and found that the implementation of ecological restoration in Tibet has optimised the cognitive structure of stakeholders, the portrayal of SES changes in ecological governance is still inadequate, and further research on the impact of ecological restoration projects on the resilience of SES on the Tibetan Plateau is still needed in future. The coupling of humans and nature in Tibet is strongly constrained by the control and adaptation of nature. This requires us to strictly implement the principle of local adaptation when formulating governance measures to avoid letting ecological restoration evolve into causing other kinds of damage to the natural and social environment. Additionally, the power of collective and policy governance should be combined to make ecological restoration a sustainable governance tool to enhance regional weak SES coupling under geographical contexts.

AUTHOR CONTRIBUTIONS

Yijia Wang and Yanxu Liu designed the study and planned the analysis. Yijia Wang and Ying Yao prepared the basic data. Yijia Wang and Xinsheng Wang performed the data analysis. Yijia Wang and Zhiwei Zhang conducted the investigation. Yijia Wang drafted the manuscript. Yanxu Liu, Xutong Wu and Bojie Fu revised the manuscript. All authors made important contributions to the manuscript.

ACKNOWLEDGEMENTS

This research is financially supported by the Strategic Priority Research Program of Chinese Academy of Sciences (XDA20020402), the Second Tibetan Plateau Scientific Expedition and Research Program (2019QZKK0405) and the Fundamental Research Funds for the Central Universities of China.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest. Yanxu Liu is an Associate Editor for People and Nature, but was not involved in the peer review and decision-making process.

DATA AVAILABILITY STATEMENT

The original transcripts of the interviews can be obtained from https://doi.org/10.6084/m9.figshare.23972607.v1. Both precipitation and temperature data were obtained from the monthly precipitation and temperature datasets over China during 1901–2017 (https://doi.org/10.5281/zenodo.3114194). The EVI for 2000 and 2018 with a resolution of 1000 m was collected from the Moderate Resolution Imaging Spectroradiometer products (MOD13A1, Version 6, MODIS Web (nasa.gov)). We obtained the data with a resolution of 1000 m from the Global 30 Arc-Second Elevation DataSet (https://webmap.ornl.gov/ogc/dataset.jsp?ds_id=10003).