IntroductionMountains have long been managed and shaped by human societies. They constitute an emblematic example of social-ecological systems (SESs)1, embodying the paradigmatic shift from viewing nature and society as separate entities to recognising their interconnectedness as parts of the same system2,3. Interactions among SES elements create feedbacks that influence the system’s resilience, adaptive capacity and transformability in sometimes unpredictable ways4,5. However, due to the complex nature of SESs, their outcomes cannot be fully understood by considering them just as the sum of their parts6,7. Accounting for the direct and indirect relationships among SES elements is thus necessary to understand not only their behaviour and persistence, but also their responses to different drivers of change. Mountain SESs’ biophysical and cultural complexity, cross-scale ecosystem services, physical isolation, and marginalisation from decision-making centres make them particularly vulnerable to climate change1. Their sustainability is weakened by changes occurring at several scales, from top-down markets and governance strategies or climate change, to sociocultural, demographic and political transformations8,9,10,11,12. Overcoming these challenges depends on our ability to understand mountain SESs’ dynamics and requires developing mitigation strategies involving stakeholders and decision-makers1.Mountain SESs are often subjected to unsuitable top-down management policies imposed by outsiders1, potentially giving rise to conflicts13 or maladaptation14 due to the diverse priorities among co-existing inhabitants. A key challenge in such contexts is people’s subjective, context-dependent and incomplete representations of reality15,16,17, or mental models, which are shaped by culture, education, values and experiences18,19,20, and in turn determine people’s behaviour and decisions21,22. This makes decision-making a complex process as stakeholders may have diverging and competing preferences regarding the management of their environment, e.g. ref. 23.The Pyrenean mountains, in the northern Iberian Peninsula, exemplify the socioeconomic challenges faced by most European and potentially worldwide mountain regions, characterized by high depopulation during the second half of the 20th century. They have transitioned from an economy based on agriculture and farming to one based on services24,25. In this context, divergent views have emerged among local inhabitants regarding the region’s future development26, raising potential challenges when defining adaptation measures aimed at mitigating social and environmental changes14.Clarifying trade-offs and areas of alignment between stakeholders’ objectives and incorporating local perspectives in decision-making is critical for inclusive regional policy planning27. By integrating social, ecological and economic processes and weighing competing stakeholders’ interests, SES modelling frameworks can help address complex social-ecological challenges. Such efforts can foster opportunities for compromise27,28, and may result in decisions that better fit local needs and priorities in contexts such as landscape planning29, reconciling development and biodiversity conservation30, or designing resilience actions to face climate change31.Models with high representational detail have the potential to closely match observable mechanisms of real-world systems and support decision-making, but they are often difficult to parametrize and to interpret, and may be challenging to use for broad scenario exploration32. Conversely, more theoretical or conceptual models are simpler and focus on generic processes rather than case-specific details. SES models with an intermediate level of abstraction and grounded on empirical data can represent selected aspects of the SES relevant to the model’s purpose, and can be used as “virtual laboratories” to test hypotheses or improve understanding of the system’s behaviour32,33. They may also serve as the baseline for developing higher-fidelity models32 and for identifying effective intervention points for governance33.In this paper, we aim to elucidate trade-offs and synergies between competing stakeholder priorities and potential intervention points in a mountain SES that could facilitate the development of sustainable management strategies capable of aligning the diverse local objectives across stakeholders. To achieve this, we propose a multi-stage modelling framework (Fig. 1) drawing from (1) a previously developed quantitative static network model formalizing interactions among 31 elements of a Pyrenean mountain SES34, including water resources, biodiversity, and socioeconomic, land-use, climatic and other environmental elements (Fig. 2a, b); and (2) local perspectives and priorities regarding rural development in the Pyrenees, as described by López-i-Gelats et al.26 (Fig. 2c). Using the SES representation from Zango-Palau et al.34 as a baseline model defining the relationships between SES components, we developed a dynamical model to simulate the dynamics of the system components through time. This dynamical model allows us to define and estimate from empirical data key parameters such as intrinsic growth rates (r), interaction coefficients (aij) and self-regulation coefficients (aii) (see Fig. 1 and “Methods” for parameter definition) that can be used to investigate the components’ contributions to system dynamics under different scenarios. We fixed interaction coefficients as these have been established in a previous study34 and explored the system’s possible trajectories under a wide range of varying combinations of r values to simulate a multitude of plausible development scenarios. We used evidence on local discourses and priorities26 to relate model scenarios to four mental frameworks: conservationist, entrepreneurial, agriculturalist and endogenous development. These differ notably in their values and priorities regarding the economy, nature conservation, tourism development, and traditional landscapes. We refer to the shared, group-level perspectives as “mental frameworks”, emphasizing their typological nature rather than individual mental models. We translated such priorities into satisfaction criteria applied to selected SES indicators in the model, which enabled us to assess which simulated development trajectories satisfy one or more of these mental frameworks. Finally, we used a classification tree model to identify the key drivers of the social-ecological system shaping its trajectory towards different types of satisfaction scenarios: “agriculturalist-only”, “conservationist-only”, “entrepreneur-only” and “endogenous development only” (i.e., scenarios where only one of the stakeholders’ mental frameworks is considered as satisfied, respectively), and “all inclusive” (i.e., scenarios where all four are satisfied) (Fig. 2c). We expected that scenarios where all mental frameworks are satisfied would emerge under conditions that balance diverging priorities (e.g. economic diversification vs. focus on tourism or agriculture) and/or fulfil shared priorities (e.g., regarding landscape composition).Fig. 1: A multi-stage modelling and analysis framework integrating stakeholder perspectives to inform sustainable social-ecological systems (SESs) management.In stage 1 a network comprising socioeconomic and ecological elements of the SES is defined based on expert knowledge on their relationships (1a and Fig. 2b). In parallel, satisfaction criteria defining the mental frameworks of the stakeholders involved in the system and used to classify model outputs into satisfaction scenarios are defined (1c) based on previous knowledge of the system (1a) and evidence on stakeholders’ values and priorities (1b and Fig. 2c). During stage 2, a quantitative mathematical representation of the system is defined based on the information gathered in stage 1a. Stage 3 involves running numerical simulations to explore the system’s behaviour under a variety of combinations of intrinsic growth rates (r) values, and classifying plausible simulation outcomes into actor satisfaction scenarios based on the criteria defined in stage 1c. In stage 4, additional exploratory simulations are run to balance the simulation outputs dataset across actor satisfaction scenarios. Stage 5 consists in performing a classification tree analysis to classify simulation outputs into homogenous subsets of data within which most of the observations belong to the same stakeholder satisfaction scenarios. The classification tree algorithm partitions the data based on the r values used in the simulation runs, and therefore enables the identification of the key SES elements whose r values are involved in differentiating the different stakeholder satisfaction scenarios. Lastly, stage 6 aims at analysing raw simulation outputs based on the classification tree predictions. Simulation outputs are divided into subsets following the splitting conditions predicted by the tree (e.g., Fig. 3), making it possible to quantify the relative change from the original r value for the key SES elements in the most relevant subsets for each stakeholder satisfaction scenario (e.g., Fig. 4). This also facilitates the evaluation of the percentage of simulations in every subset where each satisfaction criterium was met (e.g., Fig. 5). Arrows connecting the boxes indicate methodological flows between different stages. Full details on every stage and their relationships can be found in the “Methods” section. Icons for stages 1b, 4, 5 and 6 were designed by Freepik. The network figure illustrating stage 1a is adapted from Zango et al. (2024), licensed under CC BY 4.0.Full size imageFig. 2: Location of the mountain social-ecological system (SES) of the Pyrenees, foundational network model of the SES and mental frameworks included in the study to define stakeholder satisfaction scenarios.a Location of the study area in Europe. The map uses the LAEA Europe projection (ETRS89-extended/LAEA Europe, EPSG:3035). b Integrated network of the mountain social-ecological system used as foundation for the study, adapted from Zango-Palau et al.34 (licensed under CC BY 4.0.). Arrows indicate the relationships between 31 social-ecological variables, inferred through piecewise structural equation modelling. Dotted arrows indicate marginally significant relationships (p value