Abstract: Encouraging farmers to shift from the traditional “high-input, high-land-consumption” farming model to low-carbon farming has become an important strategy for mitigating agricultural emissions worldwide. As a successful low-carbon alternative in China, rice-crayfish farming, which has raised soil organic carbon storage and reduce annual CH₄ emissions by 6.4% and 2.4% respectively, relates to food security and farmers’ income growth, has becoming an effective pathway for the transformation of China’s agricultural development. Despite the proven benefits of rice-crayfish farming, such as economic and ecological benefits, farmers’ adoption decisions are complex and highly dynamic due to technical barriers, production costs, risks and community social factors. Agent- Based Modelling (ABM) has been applied to explore farmers’ adoption and diffusion of low-carbon agriculture, however, studies that link empirical data on social/spatial networks, types of interactions (informative and normative), and personality characteristics (cognition and capabilities) and integrate them into the integrated HUMAT framework (Jager et al., 2025) remain limited. Taking Longkanhu Village in China as the case, this study aims to develop a robust and behaviorally rich agent-based model for exploring the dynamic processes of innovation diffusion in farming communities on the basis of HUMAT framework and to uncover the key mechanisms underlying the success of collective action in rice-crayfish farming, which to get a better understanding of adoption dynamics, potential practical bottlenecks and effective strategies to support the transition to low-carbon agriculture.

Specifically, the HUMAT framework, which provides an integrated dynamic framework connecting cognition including human needs, values and cognitive dissonance with social persuasion and social networks, is employed to guide the modeling of farmers’ low-carbon agricultural transition decisions. We adapt the HUMAT framework to incorporate four core motivations, including economic, social, ecological and policy motivations, which are derived from farmers’ three fundamental needs of experiential, social and value needs, and also considers two types of social interactions among farmers, including inquiring and signaling. Based on this rich data set covering farmers’ general properties, year of adoption, knowledge cognition, and social networks, we construct a micro-simulation of the farmer community to define the initial population at t=1. The HUMAT framework is implemented in this micro-simulation to examine the social dynamics of adoption behavior over time and generate simulated adoption patterns. These simulated patterns are then compared with empirically observed adoption patterns, allowing us to verify in detail whether the relationships between farmer characteristics and adoption behavior are consistent with real-world data. Further, To precisely reveal the dynamic patterns of collective transformation, this study will explore the role of leadership, social norms and social learning (endogenous and exogenous social learning) by using a set of counterfactual scenarios. This study not only provide a reference framework for agent-based modeling research on farmers’ decision-making, but also offer complementary decision support for the promotion and application of low-carbon agricultural technologies and the formulation of relevant policies.

Acknowledgements

This work is supported by the key project of National Social Science Foundation of China (No.25&ZD171), the program of China Scholarship Council (Grant No:202506760057).

Households and their assets are increasingly exposed to multiple hazards, complicating risk assessment and the design of effective risk-reducing strategies. This challenge is amplified when considering how resulting impacts and uncertainty can place additional pressure on households and institutions. Insurance can provide a buffer against financial losses and support recovery; however, with climate change, the pressure on affordability and solvency is increasing. This highlights the need to understand how insurance structures can incentivize household-level risk reduction, particularly in multi-risk settings.

This study evaluates three insurance structures under multi-hazard risk: (1) a mandatory public-private system, characterized by stable premiums and a strong solidarity component; (2) a voluntary private market with dynamic risk-based premiums, where adaptation influences pricing; and (3) a hybrid reform combining elements from both. The model is applied to France, a country exposed to riverine and coastal flooding and windstorms. Here, the CatNat system provides a relevant baseline as a public-private insurance structure for policy comparison. The analysis focuses on how these three alternative structures influence household adaptation, insurance uptake, and risk over time.

To explore these dynamics, we develop an agent-based model of household and insurer behavior that integrates the Dynamic Integrated Flood and Insurance (DIFI) model in the Geographical Environmental and Behavioral model (GEB). Both modeling frameworks are expanded to a multi-hazard context. First, a probabilistic risk module simulates flood and windstorm scenarios using historical data. This information feeds into the behavioral model, where households make decisions based on Subjective Expected Utility Theory, choosing whether to invest in structural adaptation, purchase insurance, combine both, or take no action. Depending on the insurance structure, these decisions can change expected damages and influence the premium pricing over time.

By simulating interactions between households and insurance systems under multi-hazard conditions, the model provides insights into how policy design can support preparedness, reduce vulnerability, and manage the increasing pressure on the insurance systems. The results contribute to evidence-based policymaking by assessing how insurance can function not only as a compensation mechanism but also as a tool to promote risk-reducing behavior, which in turn strengthens community resilience in multi-hazard contexts.

North African farming systems become increasingly vulnerable to climate change issues including drought, irregular rainfall and biodiversity loss, that undermine their long-term viability (Lange, 2019). It is essential to strengthen the resilience and sustainability of agricultural systems in the face of increasingly challenging environmental conditions in order to ensure their long-term endurance (Darnhofer, 2014). Transition to agroecology in North Africa could be one of the most effective solutions to the current and future challenges facing the farming systems (Wezel et al., 2014).

This study adopts a systematic and transdisciplinary approach to understanding farmers’ adoption to agroecology, addressing social, economic and ecological dimensions in a holistic manner. We first apply the Social Ecological System Framework (McGinnis & Ostrom, 2014) to conceptually model farming systems within a living lab context in North Africa, identifying the key social and ecological components and their interaction to identify factors that shape farmers' adoption to agroecology. After this conceptual assessment, we utilize the DAHBSIM bioeconomic model, in a participatory approach where the agroecological transitions in the living labs co-designed with farmers and representative board. DAHBSim offers a robust simulation structure that comprehensively represents both ecological components such as water balance, the nitrate cycle, and soil dynamics, and economic components, including farm income, input costs and utility. However, the social dimension of farmers’ decision-making processes is largely missing from the model, and this limits its ability to reflect the behavioral and institutional realities that determine whether agroecological practices can be actually adopted in the field. Through DAHBSIM model we simulate agroecological transition pathways through different locally selected scenarios and their combinations over a 15-year period, accounting the future climate change trends. Finally, the social factors identified by the SESF including the influence of neighboring farmers’ choices, access to cooperative networks, and the degree of trust farmers place in agricultural institutions are incorporated into the DAHBSIM model to ensure that institutional and behavioral dimensions are properly taken into account, in parallel to existing biophysical and economic variables. By introducing social factors into the model’s structure, this study goes beyond purely economic optimization to reflect the behavioral and institutional realities of smallholder agriculture in north african living labs.

DAHBSIM scenario simulations demonstrated that certain agroecological practices can help maintain or improve farm incomes and ecological resilience while reducing the risks that climate change poses to agricultural production. Social factors identified by a context specific focus, such as cooperatives and market access, are likely to play a decisive role in the farmers’ adoption of agroecology. These findings are expected to contribute to a more socially grounded approach to bioeconomic modeling and to have a direct impact on a realistic design of agroecological transitions in North Africa.

Abstract: Climate change is affecting cities and regions worldwide through floods, heatwaves and extreme weather events (IPCC, 2023). In response, many authorities have developed climate change adaptation (CCA) plans outlining measures to protect societies (Aguiar et al., 2018; Buzási et al., 2024; Reckien et al., 2015). However, while regional and local adaptation plans are proliferating, their translation into effective, context-specific actions remains slow and uneven. This persistent shortfall is often described as the adaptation gap (Fünfgeld et al., 2023; Rogers et al., 2025; Storbjork et al., 2024; UNEP, 2025). Especially the implementation of CCA policies continues to lag (Roest et al., 2025; Rogers et al., 2025; Wang et al., 2025), due to multi-level governance complexity, implementation barriers, and justice issues (Aguiar et al., 2018; Brink et al., 2023; Fünfgeld et al., 2023; Otto et al., 2025). To address these challenges, recent literature highlights the potential role of intermediaries, such as NGOs, consultants and peer networks (Bower et al., 2024; Dąbrowski, 2018; Ziervogel, 2019). Intermediaries can support implementation, for example, by providing technical expertise (Karhinen et al., 2021; Lau et al., 2025), connecting communities and formal adaptation actors (Bower et al., 2024) and shaping processes with attention to social, political, and economic inequalities (Soanes et al., 2021). Yet, existing research primarily focuses on what intermediaries do, offering limited insight into how they influence implementation processes and outcomes (Tosun et al., 2023). This study aims to address this gap by analysing how, and under what conditions, intermediaries can influence behavioural dynamics in regional and local CCA policy implementation, to reduce the adaptation gap and make the process more just.

The study develops an Agent-Based Model (ABM) to simulate interactions between policymakers, intermediaries, and citizens. It builds on empirical insights from five European pilot regions with diverse governance structures and climate risks: Zagreb, Athens, Lake Constance, Nouvelle-Aquitaine, and Belfast. First, semi-structured interviews with decision-makers examine how CCA policies are implemented in practice, focusing on decision-making processes, informal institutions, and the role of intermediaries. Second, a citizen questionnaire explores how residents perceive intermediary actors, and whether these actors influence their attitudes or behaviours regarding adaptation measures. Building on these insights, an ABM will be developed. The ABM will be used to explore scenarios in which intermediaries assume different roles. This enables a systematic analysis of their impact on implementation dynamics, adaptation outcomes, and procedural justice. By integrating empirical data in the ABM, this research aims to advance the understanding of the behavioural mechanisms underlying the adaptation gap. Ultimately, it seeks to identify pathways for effective and just implementation of climate change adaptation policies.

Keywords: Climate adaptation, Justice, Intermediaries, Agent-based modelling

Acknowledgements:This research is funded by the JustREACH project, funded by the European Union’s HORIZON Research and Innovation Actions under the grant agreement No 101214666.

Disclosure of Interests: The authors have no competing interests to declare that are relevant to the content of this article.

References

Aguiar, F. C., Bentz, J., Silva, J. M. N., Fonseca, A. L., Swart, R., Santos, F. D., & Penha-Lopes, G. (2018). Adaptation to climate change at local level in Europe: An overview. Environmental Science & Policy, 86, 38–63. https://doi.org/10.1016/j.envsci.2018.04.010

Bower, E., Harrington-Abrams, R., & Priem, B. (2024). Complicating ‘community’ engagement: Reckoning with an elusive concept in climate-related planned relocation. Global Environmental Change-Human and Policy Dimensions, 88, 102913. https://doi.org/10.1016/j.gloenvcha.2024.102913

Brink, E., Falla, A. M. V., & Boyd, E. (2023). Weapons of the vulnerable? A review of popular resistance to climate adaptation. Global Environmental Change, 80, 102656. https://doi.org/10.1016/j.gloenvcha.2023.102656

Buzási, A., Simoes, S. G., Salvia, M., Eckersley, P., Geneletti, D., Pietrapertosa, F., Olazabal, M., Wejs, A., De Gregorio Hurtado, S., Spyridaki, N.-A., Szalmáné Csete, M., Torres, E. F., Rižnar, K., Heidrich, O., Grafakos, S., & Reckien, D. (2024). European patterns of local adaptation planning—A regional analysis. Regional Environmental Change, 24(2), 59. https://doi.org/10.1007/s10113-024-02211-w

Dąbrowski, M. (2018). Boundary spanning for governance of climate change adaptation in cities: Insights from a Dutch urban region. Environment and Planning C: Politics and Space, 36(5), 837–855. https://doi.org/10.1177/2399654417725077

Fünfgeld, H., Fila, D., & Dahlmann, H. (2023). Upscaling climate change adaptation in small- and medium-sized municipalities: Current barriers and future potentials. Current Opinion in Environmental Sustainability, 61, 101263. https://doi.org/10.1016/j.cosust.2023.101263.

IPCC. (2023). Climate Change 2022 – Impacts, Adaptation and Vulnerability: Working Group II Contribution to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change (1st edn). Cambridge University Press. https://doi.org/10.1017/9781009325844

Karhinen, S., Peltomaa, J., Riekkinen, V., & Saikku, L. (2021). Impact of a climate network: The role of intermediaries in local level climate action. Global Environmental Change, 67, 102225. https://doi.org/10.1016/j.gloenvcha.2021.102225

Lau, K., Yuan, C., & Ng, E. (2025). Urban heat island adaptation and mitigation in practice: Lessons from policy implementation in five cities. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 383(2308), 20240581. https://doi.org/10.1098/rsta.2024.0581

Otto, A., Miechielsen, M., Schmidt, K., & Thieken, A. H. (2025). Drivers and barriers to municipal climate change adaptation: A comparative analysis of selected measures and different implementation stages. Journal of Environmental Studies and Sciences. https://doi.org/10.1007/s13412-025-01049-w

Reckien, D., Flacke, J., Olazabal, M., & Heidrich, O. (2015). The Influence of Drivers and Barriers on Urban Adaptation and Mitigation Plans—An Empirical Analysis of European Cities. PLOS ONE, 10(8), e0135597. https://doi.org/10.1371/journal.pone.0135597

Roest, A. H., Restemeyer, B., Van Den Brink, M., Horlings, I., & Boogaard, F. C. (2025). Unravelling the Policy‐to‐Implementation Gap: Climate Adaptation Governance Arrangements in Medium‐Sized Cities. Environmental Policy and Governance, eet.70042. https://doi.org/10.1002/eet.70042

Rogers, N. J. L., Adams, V. M., & Byrne, J. A. (2025). Moving beyond the plan: Exploring the opportunities to accelerate the implementation of municipal climate change adaptation policies and plans. Environmental Policy and Governance, 35(2), 276–291. https://doi.org/10.1002/eet.2142

Soanes, M., Bahadur, A., Shakya, C., Smith, B., Patel, S., Coger, T., Dinshaw, A., Patel, S., Huq, S., Musa, M., Rahman, F., Gupta, S., Dolcemascolo, G., & Mann, T. (2021). Principles for locally led adaptation. IIED, Londen.

Storbjork, S., Hjerpe, M., & Glaas, E. (2024). The necessity of pragmatic muddling. Ten Swedish early adopter cities navigating climate adaptation policy-implementation in the urban built environment. Environmental Science & Policy, 160, 103842. https://doi.org/10.1016/j.envsci.2024.103842

Tosun, J., Tobin, P., & Farstad, F. M. (2023). Intermediating climate change: Conclusions and new research directions. Policy Studies, 44(5), 687–701. https://doi.org/10.1080/01442872.2023.2230900

UNEP. (2025). Adaptation Gap Report 2025: Running on Empty - The World is Gearing up for Climate Resilience — without the Money to get there. United Nations Environment Programme. https://doi.org/10.59117/20.500.11822/48798

Wang, T., Zhang, K., Chen, D., & Wang, X. (2025). Big ambitions and insufficient actions: A decade of climate adaptation progress. Advances in Climate Change Research, 16(5), 1087–1099. https://doi.org/10.1016/j.accre.2025.08.002

Ziervogel, G. (2019). Building transformative capacity for adaptation planning and implementation that works for the urban poor: Insights from South Africa. Ambio, 48(5), 494–506. https://doi.org/10.1007/s13280-018-1141-9

The transition toward a sustainable economy requires profound changes in consumption behaviour. Yet despite widespread environmental concern, the adoption of green products remains persistently limited. The existing literature has largely explained this gap through the lens of barriers such as high prices, limited infrastructure, and behavioural inertia. While these factors are undeniably relevant, I argue that this framing could be incomplete. Purchasing decisions are not merely the residual of removed obstacles, but the outcome of active social and economic forces that may independently drive or suppress green consumption regardless of price and infrastructure conditions (Elliot, 2013).

This paper proposes a reframing of the problem, investigating the socio-economic drivers that lead agents to purchase, or deliberately avoid, sustainable goods. I focus on two interrelated forces: the propensity of agents to compete for social status through consumption, and the dynamics of conformism and distinction that such competition generates. To this end, I build on two foundational theoretical traditions. The first is Veblen's (1899) theory of conspicuous consumption, which posits that agents derive utility not only from the intrinsic qualities of goods but from the social signals their purchase conveys. The second is Bourdieu's (1984) sociology of distinction, which extends this logic to cultural capital: agents seek differentiation through consumption choices that mark their membership in, or distance from, particular social groups.

The paper advances a cautionary argument. While social competition can under certain conditions promote green consumption (Griskevicius et al., 2010), I contend that it may equally function as a structural impediment to the green transition. Three mechanisms can be identified. First, when status is attached to high-visibility brown goods, Veblenian dynamics actively suppress sustainable choices among high-income agents. Second, when green consumption becomes mainstream, Bourdieusian distinction may lead culturally motivated agents to abandon it precisely because it has lost its signalling value. Third, when conformist pressures dominate, the system may become locked into a brown equilibrium that is self-reinforcing and resistant to change. In short, the same social forces that can drive the green transition can, under different configurations of preferences and signals, act as a cap on it.

To formalise these mechanisms, I develop an agent-based model in which heterogeneous agents are embedded in a social network and update their consumption choices over time in response to the evolving behaviour of their peers (Ghiglino & Tabasso, 2024). This framework allows for the endogenous emergence of social norms and the simulation of sustainable consumption diffusion under alternative policy scenarios. The findings carry direct implications for the design of demand-side climate policies, suggesting that interventions targeting social norms and identity may complement conventional price-based instruments.

REFERENCES:

Bourdieu, P. (1984). Distinction: A Social Critique of the Judgement of Taste (R. Nice, Trans.). Harvard University Press.

Elliott, R. (2013). The taste for green: The possibilities and dynamics of status differentiation through green consumption. Poetics, 41(3), 294–322.

Ghiglino, C., & Tabasso, N. (2024). Endogenous identity in a social network. arXiv preprint arXiv:2406.10972.

Griskevicius, V., Tybur, J. M., & Van den Bergh, B. (2010). Going green to be seen: Status, reputation, and conspicuous consumption. Journal of Personality and Social Psychology, 98(3), 392–404.

Veblen, T. (1899). The Theory of the Leisure Class: An Economic Study of Institutions. Macmillan.

Agent-based models (ABMs) are recognised as powerful tools for simulating the emergent dynamics of complex systems and are increasingly used to explore policy and intervention effects. However, there are persistent methodological debates regarding their empirical grounding. Agents may follow rules driven by mechanistic parameters calibrated e.g. through approximate Bayesian computation, but such models may struggle with complex systems because of the “curse of dimensionality”. Regression-based methods fitting to observed data can effectively address this issue but can make it harder to transparently represent causal structures or fuse multiple evidence streams when the data are irregular, heterogeneous, or partly qualitative.

This paper argues instead for the integration of Bayesian Belief Networks (BBNs) for ABM parameterisation, positioning them as probabilistic graphical models that offer a natural layer between empirical evidence and simulation. BBNs encode conditional dependency structures and enable probabilistic inference under uncertainty. Their inferential architecture further supports adaptive agent behaviour: as agents accumulate experience and encounter changing conditions, they update their behavioural propensities through evidence propagation across the network’s causal structure. This facilitates a form of principled, dynamic adaptation that static regression-based approaches do not readily accommodate.

We present an end-to-end workflow examining resilience and vulnerability in maternal and child health systems in rural Zambia under climate-related shocks, particularly floods and extreme heat. This work forms part of the REACH project (Building Resilience to Floods and Heat in Maternal and Child Health Systems in Brazil and Zambia). To ground the ABM within communities’ lived experiences and local epistemologies, we employed participatory approaches, specifically Group Model Building (GMB) sessions and Causal Loop Diagrams (CLDs), co-developed with community members and health workers. The Bayesian Belief Network (BBN), whose structure mirrors the key causal pathways identified in the CLDs, provides a robust bridge between community-derived insights and the formal parameterisation of the ABM.

We leverage the BBN to integrate multiple data sources: the Zambia Demographic and Health Survey (ZDHS), which supplies nationally representative quantitative evidence on MCH determinants and indicators (including antenatal care utilisation, skilled birth attendance, and child immunization coverage), and a community survey, which provides locally contextualised data on sociodemographic characteristics and other resilience and vulnerability indicators and their relation to MCH outcomes during and after climate extreme events. The resulting BBN-informed ABM simulates how individual household agents, health worker agents, and community-level structures interact under varying conditions of vulnerability and resilience, producing emergent patterns of MCH outcomes across different intervention scenarios.

The paper outlines a workflow for converting CLDs into BBN structures, using survey data to generate probabilistic decision rules for agents. It highlights the framework’s strength in enabling sensitivity analysis and scenario testing, and argues that combining Bayesian methods, empirical data, and community co-production is essential for developing policy-relevant ABMs in low- and middle-income contexts.

Driving in a city is a highly social endeavour. Every decision when braking, accelerating, or changing lanes is dependent on the behaviour of all other drivers in the vicinity. Even long-distance driving depends on the behaviour of drivers across the whole system. The social network of driving has over time produced informal and formal systems of habits, expectations, and rules.

Autonomous Vehicles (AVs) are a new type of participant that is gradually being introduced into this social network. Unlike their human counterparts, they do not rely only on observations of the surrounding drivers. They have the ability to communicate directly with each other. Using built-in Cellular Vehicle-to-Everything (C-V2X) communication and the novel Manoeuvre Coordination Messaging (MCM) service, AVs can instantly share information with each other, coordinate actions, and plan group driving behaviour.

At lower levels of AV adoption, the human driver network dominates the system. AVs adapt to human unpredictability and lack the numbers for efficient cooperation. As more AVs are introduced into the system, cooperative manoeuvres like space negotiation, traffic light negotiation, platooning, and stop-and-go wave management become possible and can change the nature of vehicle traffic.

This project explores that transition. Using agent-based modelling to first simulate various cooperative driving manoeuvres separately, then to simulate the different manoeuvres together in one model. We will also gradually introduce AV agents with the capacity for cooperative driving into a city model to understand how different percentages of AVs can leverage cooperative driving in a human-driver system.

This project is important as it aims to give us an understanding of the potential benefits of collective intelligence in driving. How smart cities might leverage the communication between vehicles as well as infrastructure to produce more efficient cities. And what are the unseen consequences for governance priorities when allowing AVs to enter a city.

The effects of climate change are being felt worldwide, making the need for adaptation across scales – from public to private – urgent. Though households are increasingly encouraged to take autonomous adaptations, like flood-proofing homes, such private adaptation actions are inevitably influenced by what governments do on public adaptation. Traditionally, government-led public adaptation has dominated, taking the forms of either large-scale infrastructure protection, governmental post-disaster refunds, or issuing policies to support individuals’ private actions. Incorporating institutional Policy Agents remains a key challenge in all decision support tools, including agent-based models. To date, such formal institutions typically enter simulation models either in the form of (i) external targets against which a model optimizes, (ii) exogeneous ‘what-if’ policy scenarios, (iii) a simple cost-benefit analysis agent, or (iv) a hard-coded fixed policy shift that gets triggered either at a specific time step or after specific response thresholds are reached. Yet, neither of these approaches resolves the key issues: policy institutions remain static, oversimplified and disconnected from reality. Recently, large language models (LLMs) have entered this discourse, offering the means to represent dynamic, adaptive, structurally rich, diverse set of policy institutions, and different stakeholders that can negotiate and explain their decision logic. Policymaking agents driven by LLMs could be used within ABMs: given the system state and a set of policy options, they can reason for different pathways, while interacting and reacting to other agents’ decisions. This is especially relevant for understanding the interplay between private and public climate change adaptation, both because climate risks are geographically uneven and drastically changing over time, and because public policy may trigger unintended consequences, demanding constantly updating tailored solutions.

To address this science and policy gap, our poster presents a methodological workflow and the results from applying LLMs to represent dynamic policy-making within the FAST-ABM (Flood Adaptation via Social Transformation agent-based model) for climate-induced risks, testing the interplay between public and private climate change adaptation. Specifically, building up on the newly proposed Agentic World Analysis (AWA) approach, we specify a Policy Agent that adaptively decides whether and what policy to introduce as the cumulative climate damages evolve in a region, both driven by the physical climate risks (here conceptualized as floods) and the climate adaptation actions of households (e.g. wet- and dry-proofing their property). The spatial FAST-ABM we employ here was developed to study private household adaptation to climate-induced floods. The decisions of household agents in FAST-ABM are grounded in the Theory of Planned Behavior from psychology, simulating dynamic social norms explicitly, and relying on tailored survey microdata. Using the power of LLMs and AWA, we expand the model to incorporate policy agents like a regional government and an environmental agency that are partly informed by an LLM reasoning process. We present the results of the simulation coupling adaptive Policy Agents with the households’ agents simulated using ABM. While reporting content results, climate-induced risk, cumulative private adaptation uptake, under different public adaptation actions, and potential inequality outcomes, we also critically reflect on our methods and future research.

Developing empirically grounded agent-based models (ABMs) that integrate environmental hazards with social dynamics remains a central challenge. This is especially acute in multi-site, multi-disciplinary settings where heterogeneous data sources, divergent geographic contexts, and competing modelling objectives must be reconciled into a coherent framework. We present methodological solutions developed for an ABM simulating the impacts of flood and heat hazards on maternal and child health and healthcare access across three contrasting sites: urban Recife (Brazil), and two rural Zambian districts (Senanga and Sinazongwe),each facing distinct flood regimes and social contexts. We argue that empirical grounding must extend beyond environmental inputs to shape what agents experience, how they make decisions, and which social dynamics the model can plausibly represent.

We advance three empirical grounding strategies, each translating environmental heterogeneity into meaningful variation in agents' exposure, mobility, and health-seeking decisions; grounding social dynamics in local realities.

First, rather than fitting parametric distributions to climate variables to reproduce discrete extreme events, we adopt a time-series approach, drawing from downscaled CMIP6 models across multiple SSPs and time horizons. Sampling individual years from thousands of potential yearly climates preserves inter-variable physical consistency, accommodates non-stationarity, and sidesteps the validation burden that distribution-fitting imposes, since physical plausibility and cross-variable correlations are inherited directly from the climate model outputs. Moving from an event-based to a time-series approach offers the additional benefits of greater flexibility in defining context-sensitive indicators, and capturing both acute shocks and cumulative stress, which can trigger distinct health-seeking behaviours and household decisions.

Second, we address flood mapping under data scarcity through a tiered approach. Floods are highly localised and their impact are highly dependent on social contexts: who can move, by what means, and through which routes. We combined local precipitation percentiles with available flood products, supplemented by local ground-truthing where remotely sensed and modelled products fail to adequately capture lived realities, such as in Sinazongwe, where no existing product records locally reported flash floods. Without this level of fidelity, agents would be exposed to an environment that misrepresents what communities actually experience, undermining the model’s validity against stakeholder data. We are currently exploring local hydrological modelling and participatory mapping as solutions, directly embedding local knowledge into the model's environment.

Third, we enable context-specific mobility environmnent and behaviours, such as road-based travel in urban Recife; road, off-road, and boat travel in Senanga, by offloading healthcare access calculations to a GIS pre-processing pipeline using friction maps. These contextual differences shape agents access time, and, thus, their health-seeking behaviours. As added benefits, this approach reduces the computational and memory constraints of real-time routing, and facilitates this component independent validation by isolating it, reducing the ABM validation burden

Together, these solutions demonstrate that empirical grounding in distinct social-environmental settings requires actively “embracing the social: surfacing local mobility patterns, infrastructural inequality, and community knowledge as determinants of what agents experience, decide, and do. We offer these approaches as transferable building blocks for multi-disciplinary ABMs in climate-health and other social-related fields.

Designing institutions and artificial systems that sustain cooperation effectively requires knowing which mechanisms actually drive the emergence of social norms. Theory offers multiple candidates e.g.: deterrence, prosocial values, reputation - but empirically disentangling them in human experiments is difficult, since participants cannot be observed before socialisation has shaped their behaviour. This poster presents both an experiment and a general-purpose computational framework for investigating which factors impact the emergence of social norms in multi-agent populations.

The framework is built on multi-agent reinforcement learning (MARL) in configurable social dilemma environments. Its core design principle is modularity: motivators that may drive norm emergence are implemented as interchangeable treatments applied to otherwise identical agent populations. The environment simulations of different social dilemmas can be interchanged without modifying the treatment logic. Treatments can also be combined - for instance, agents could receive both a sanctioning tool and prosocial reward shaping simultaneously - enabling factorial designs that test for interaction effects. The framework's modularity allows for generalisation beyond the specific mechanisms studied here - reputation systems or institutional rules can be added as additional treatments in future work.

As a first application of this framework, we compare two canonical motivator types: an external motivator - peer-to-peer punishment, where agents can directly sanction others at a cost - and an internal motivator - prosocial preferences, implemented through Social Value Orientation (SVO) reward shaping that makes agents value others' welfare. Three cohorts (control, punishment, prosocial preferences) are trained in two Melting Pot 2.0 environments: Commons Harvest, a tragedy-of-the-commons setting where the emergent norm is resource restraint, and Clean Up, a free-rider setting where the emergent norm is contributing fairly to a public good. Each environment hosts seven concurrently learning agents over identical training conditions.

Norm emergence is measured through cooperation levels, wealth inequality, free-riding intensity, and behavioural dynamics such as punishment frequency evolution and convergence to prosocial actions. Statistical analysis combines non-parametric distribution tests, permutation robustness checks, and effect-size estimation. Preliminary results suggest that the two motivator types produce qualitatively different pathways to cooperation: punishment stabilises moderate cooperation with constrained inequality, while prosocial preferences yield initially volatile but ultimately high cooperation. Feedback from the SSC2026 attendees will form a basis for refinement of the framework design and generalisation to additional social settings.

Regional land use models are powerful tools for simulating physical land use change, but they typically treat the population as a static, homogeneous mass. They model where people live, but not who they are, why they move, or how different household groups behave differently. Agent-based models can address this gap, yet most existing ABMs in the residential mobility domain remain focused on single cities, rely on synthetic populations or assumed decision rules, and are rarely connected to operational planning tools. This research presents the design and first results of an empirically grounded ABM that extends an existing cellular automata based land use model with a behavioural population module, simulating residential mobility and household dynamics across an entire region at 100m resolution.

The model is built around a modular architecture with two core components. A population dynamics module handles demographic evolution, including ageing, births, deaths, household transitions, and migration flows. A relocation module determines whether and where households move. Relocation is operationalised through a trigger-based mechanism: rather than assigning households fixed annual move probabilities, the model waits for life course events such as the birth of a child, retirement, or household dissolution to occur before evaluating whether and where the household relocates. Different trigger types produce different behavioural responses, enabling the model to distinguish between forced and voluntary moves. The relocation module consists of two submodels: a move propensity model determining who moves, and a destination choice model determining where they go.

A central design principle is that all agent decision rules are empirically grounded rather than assumed. Move propensity scores are derived from machine learning and logistic regression models estimated on linked census microdata from 2011 and 2021, covering over 7 million individuals at 100m grid resolution. A household survey provides complementary behavioural inputs that cannot be derived from administrative data alone, including residential preference weights and location satisfaction ratings that directly calibrate the displacement and utility functions governing agent relocation decisions. Together, observed mobility behaviour from the census and stated preferences from the survey form the empirical backbone of the simulation.

This research presents the model architecture, the trigger-based relocation concept, first empirical results from the move propensity model, and the planned integration pathway with the land use model. By coupling an empirically parameterised ABM with an operational land use simulation at regional scale, this work contributes to bridging the gap between behavioural social simulation and applied spatial planning practice.

Background
The climate crisis represents one of the most urgent and complex challenges of the twenty-first century. Its effects unfold across multiple temporal and spatial scales, and its communication involves translating scientific models and global processes into forms that can resonate with diverse publics. Traditional modes of science communication, which emphasise facts, data, and policy implications, often struggle to reach audiences on an emotional and experiential level. As a result, communicators increasingly turn to storytelling as a means of bridging the gap between scientific knowledge and everyday experience.
Project idea
The project investigates how cultural contexts shape the relationship between scientific knowledge and personal experience in climate-crisis storytelling. Communicating the climate crisis poses a dual challenge: translating complex scientific knowledge while evoking emotional and moral engagement. In Germany, science communication has historically relied on rational, expert-based explanations but now increasingly employs affective and experiential storytelling to connect with audiences. In India, climate narratives emerging from cli-fi, graphic novels, and social-media campaigns are grounded in cultural symbolism and lived experience yet increasingly draw on scientific references to enhance credibility. This cross-cultural paradox - each context integrating its “opposite” strategy - forms the starting point of the proposed research: German communicators use personal experiences and charismatic presentation styles to connect to audiences on an emotional level, while Indian cli-fi novels use scientific knowledge to increase the impact and realism of their experiential exposure stories.
Agent-based modelling
AI FORA introduced a participatory, model-based approach to designing future AI systems for public services. The AI FORA research strategy is presented as an iterative process that combines agent-based modelling, serious games, synthetic data generation, and policy workshops. This strategy creates a safe innovation space where stakeholders can make decision processes explicit, explore alternative rulesets, and collaboratively prototype AI systems aligned with societal values. The chapter discusses three key roles of modelling in this process - representing current welfare practices, enabling experimentation with new rules, and bridging human deliberation and machine learning to build transparent AI prototypes. It explains why ABM is uniquely suited for capturing the complexity and heterogeneity of real-world welfare systems and supporting democratic innovation cycles.
ABM in the project
In the project an interpretive agent-based model (ABM) will simulate narrative dynamics and audience responses, allowing systematic exploration of how cultural factors and communicative strategies interact to shape story resonance. A final phase of comparative synthesis and reflexive evaluation will integrate empirical and modelling results, combining sequence and policy analysis with participatory workshops involving science communicators and policymakers.
Expected outcome
By linking narrative hermeneutics with computational social-science methods, the project advances theoretical and methodological innovation in cross-cultural communication research. It contributes new understanding of how stories mediate between science and experience, and how culturally specific repertoires influence the public resonance of climate communication. Beyond its academic significance, the project will produce empirically grounded insights and conceptual tools for designing culturally adaptive and socially engaging forms of climate communication, thereby strengthening the science-society interface in the context of global climate challenges.

A mobility hub is a transportation node that integrates multiple modes of transport—such as trains, light rail transit (LRT), route buses, on-demand mobility services, ridesharing, and shared bicycles—into a single location, enabling seamless transfers and addressing last-mile transportation challenges.

In the context of declining birthrates, population aging, and overall population decline, there is an increasing need for urban environments and local public transportation systems that enhance residents’ health, well-being, and sense of security. This study develops and validates a simulation model to evaluate the effectiveness of mobility hub policies in promoting residents’ daily physical activity and community engagement.

As populations age, facilitating healthy lifestyles within local communities can yield significant benefits for both individuals and society, including improved quality of life, increased life satisfaction, and reduced healthcare costs.

To address these challenges, this study adopts the concept of the “mobility hub” as a framework for promoting residents’ health through integrated digital and spatial interventions.

Beyond serving as transportation nodes, mobility hubs have the potential to provide a wide range of functions, including local events that utilize transfer time, community-oriented spaces such as cafes, opportunities for social interaction over meals, basic health checkups and lifestyle consultations, economic activities such as shopping, and integration with MaaS (Mobility as a Service).

This study develops a simulation platform for healthy city design using synthetic population data, mobile spatial data capturing residents’ mobility behavior, local facility data (e.g., shops and community centers), transportation network data, and social network structures. The simulator evaluates the impact of policy scenarios aimed at increasing daily physical activity and reducing social isolation within local communities.

The simulation results indicate that, in addition to the effectiveness of mobility hubs themselves, complementary initiatives—such as health ambassadors engaging with residents, participation in local events, and mechanisms that encourage outings for dining and shopping—also play an important role in promoting physical activity and social interaction.

How do collective narratives emerge and change over time? This question is key to better grasp the complex dynamics of phenomena such as polarization on social issues, like climate change, global pandemics, or inter-group conflicts [1]. As social media is a key place of polarization [2,3], studying the dynamics of collective narratives in these spaces appears to be of high importance to generate theoretical explanations of their consequences on group behavior. Following this direction, we developed a Twitter/X model with a scale-free, directed network of individuals. The inter-individual connections are dynamic and change with a probability related to homophily. Agents see posts on their feed, mostly coming from their followees. The tweets can be categorized into different topics, for which individuals hold stances (a continuum between strong agreement and disagreement). For each post seen, an individual can decide between two choices: do nothing or react. This choice process is represented by a phenomenological model, widely used for perceptual decision-making, known as the drift-diffusion model [4]. This cognitive model’s behavioral outputs -- reaction time and accuracy -- reproduce the empirical observations. The choice dynamics are modulated by the initial bias to react and the stance towards a topic. These effects vary in importance, depending on the tweets seen and the choices made. When agents choose to react, they can further select from several actions with different probabilities: retweet, tweet with the same topic, or tweet with a new topic. We observed that some topics are shared much more frequently than others during a limited period and are subsequently replaced by other topics. These topics, when they are popular, are often strongly polarizing. This model is designed for application to Twitter/X data, utilizing available metrics such as reaction time distributions, weekly tweet and retweet numbers, and topic similarity. Using Bayesian inference, we aim to identify which underlying mechanisms -- intra- and inter-individual -- among the ones described above are the most likely to explain the data, leading to a better mechanistic understanding of the processes at hand.

References:

[1] Bliuc, A.-M., & Chidley, A. (2022). From cooperation to conflict: The role of collective narratives in shaping group behaviour. Social and Personality Psychology Compass, 16(7), e12670.

[2] Kubin, E., & Von Sikorski, C. (2021). The role of (social) media in political polarization: a systematic review. Annals of the International Communication Association, 45(3), 188-206.

[3] Van Bavel, J. J., Rathje, S., Harris, E., Robertson, C., & Sternisko, A. (2021). How social media shapes polarization. Trends in cognitive sciences, 25(11), 913-916.

[4] Ratcliff, R., Smith, P. L., Brown, S. D., & McKoon, G. (2016). Diffusion decision model: Current issues and history. Trends in cognitive sciences, 20(4), 260-281.

Reproducibility is a cornerstone of trustworthy science, yet capturing the full provenance of complex computational experiments remains a practical challenge. We present RECORD (Reproducible Execution, Collection, and Ontological Representation of Data), a provenance framework applied to the replication of a large-scale agent-based social simulation — FEARLUS-SPOMM — which investigated nonlinearities in biodiversity incentive schemes across over 20,000 simulation runs. RECORD captures coarse-grained provenance at the "file surface" level, recording inputs, outputs, software versions, and execution metadata without requiring modification of the underlying model code. The resulting provenance graph comprises approximately 700,000 vertices and 1.8 million edges, stored in a TinkerPop Gremlin graph database, enabling intuitive lineage queries including backward slicing, input tracing, and chain depth measurement. Visual inspection of reproduced outputs confirms successful replication of the original analysis. Beyond replication, the framework provides a foundation for reproducibility and a step towards robustness — the capacity to reproduce results independently of previously used resources. This work demonstrates that systematic provenance capture is both practically achievable and scientifically valuable for complex simulation workflows.

Extreme weather events increasingly threaten macroeconomic stability, yet their aggregate consequences depend critically on how climate damages are distributed across space and sectors. Existing agent-based integrated assessment models such as the DSK framework (Lamperti et al., 2017) and its stock–flow consistent extension (Reissl et al., 2025) endogenies climate-economy feedback but largely abstract from spatial heterogeneity in climate exposure. As a result, they may underestimate the role of regional asymmetries in shaping macroeconomic dynamics.

This paper develops a regionalised extension of the DSK stock–flow consistent agent- based integrated assessment model. The national economy is disaggregated into multiple regions that differ in their economic size and exposure to stochastic climate shocks while remaining integrated through common goods, labour, credit, and energy markets. The model allows to test the macroeconomic and regional effects of heterogeneous climate shocks, and the role of regional as well as macroeconomic policy to dampen their effects.

Preliminary results show that national aggregates can mask substantial regional divergence. While output, unemployment, and emissions at the country level may appear similar under alternative exposure configurations, regional trajectories display persistent dispersion in income, capital accumulation, and fiscal positions. Shocks concentrated in larger regions tend to induce systemic restructuring dynamics, whereas shocks concentrated in smaller regions generate localized collapses with limited aggregate visibility but strong distributional consequences. These dynamics generate path dependency through investment responses, credit conditions, and regional fiscal capacity.

Physical climate risks (PCRs) damage not only the physical and economic value of assets, but also the fiscal survival of governments across scales. However, macroeconomic assessments of PCRs narrowly focus on marginal analysis and perfect foresight, assuming that systems will price unprecedented risks using historical events [1, 2]. This approach consistently underestimates the actual climate damage because it overlooks how localized fiscal stress scales up to systemic instability. Specifically, local damage to household or firm assets can collapse regional tax bases, trap citizens in impaired balance sheets, and force massive public bailouts. Furthermore, this sets off a self-perpetuating cycle where bailouts trigger sovereign downgrades, raising government borrowing costs and restricting future public adaptation [3, 4]. Earlier models often ignore the endogeneity of government decisions, which are shaped by fiscal and adaptation strategies and, in turn, affect the economic baseline [5]. This includes issues like the "safe development paradox," where new defenses unintentionally promote investment in hazard-prone areas [6], increasing vulnerability before a disaster.
To address this gap, this poster presents a conceptual framework for an Agent-Based Model (ABM) that simulates the dynamic interplay among PCR, taxpayer behavior, and multilevel public finance, grounded in European regional archetypes. The framework features a multilevel government structure alongside a heterogeneous taxpayer agent. Local government agents aim to balance PCR safety with the provision of other public goods and competition for mobile taxpayers, under a balance-sheet constraint [5, 7]. Higher-tier government acts as a fiscal backstop, managing debt markets and transfers to cover local fiscal gaps [1, 3]. Taxpayer agents (households and firms) seek to maximize utility from PCR safety, public goods, and tax rates, with limited foresight and varying mobility [8, 9].
The simulation bridges micro-behaviors and macroeconomic dynamics through a nonlinear feedback loop in which rising adaptation costs trigger high-wealth capital flight, thereby hollowing out the tax base through Tiebout-style sorting [8, 9]. This leaves the immobile, financially strained demographic to bear the burden of rising costs, caught in "sticky" debt and devalued assets when disaster occurs [10]. These emergent micro-insolvencies cascade through a macro-engine in the model, leading to systemic sovereign risk and forcing trade-offs between national solvency and disaster relief [3].
This framework illustrates how feedback between taxpayers and public finance can transform fragmented public adaptation into systemic fiscal traps. Higher governments must choose between under-resilience for tax competitiveness or defenses that inadvertently trigger the safe development paradox [6]. By exposing these trade-offs, this research highlights the need to shift from fragmented municipal defenses to macro-fiscally sustainable, cross-scale policies [1, 3].
References
1. Bolton P, Despres M, Silva LAP da, et al (2020) The green swan: central banking and financial stability in the age of climate change. Bank for International Settlements, Basel, Switzerland
2. Farmer JD, Hepburn C, Mealy P, Teytelboym A (2015) A Third Wave in the Economics of Climate Change. Environ Resour Econ 62:329–357. https://doi.org/10.1007/s10640-015- 9965-2
3. Zenios SA (2024) The climate-sovereign debt doom loop: what does the literature suggest? Curr Opin Environ Sustain 67:101414. https://doi.org/10.1016/j.cosust.2024.101414
4. Beirne J, Renzhi N, Volz U (2021) Feeling the heat: Climate risks and the cost of sovereign borrowing. Int Rev Econ Finance 76:920–936. https://doi.org/10.1016/j.iref.2021.06.019
5. Preinfalk E, Knittel N, Bednar-Friedl B, et al (2026) Fiscal implications of public climate change adaptation: An analysis of three European countries. Ecol Econ 243:108915.
https://doi.org/10.1016/j.ecolecon.2025.108915
6. Di Baldassarre G, Viglione A, Carr G, et al (2013) Socio-hydrology: conceptualising humanflood interactions. Hydrol Earth Syst Sci 17:3295–3303. https://doi.org/10.5194/hess-17-3295-2013
7. Jerch R, Kahn ME, Lin GC (2023) Local public finance dynamics and hurricane shocks. J Urban Econ 134:103516. https://doi.org/10.1016/j.jue.2022.103516
8. Tiebout CM (1956) A Pure Theory of Local Expenditures. J Polit Econ 64:416–424.
https://doi.org/10.1086/257839
9. Kollman K, Miller JH, Page SE (1997) Political Institutions and Sorting in a Tiebout Model. Am Econ Rev 87:977–992
10. Koo RC (2013) Balance sheet recession as the ‘other half’ of macroeconomics. Eur J Econ Econ Policies Interv 10:136–157. https://doi.org/10.4337/ejeep.2013.02.01

Research focus

The use of artificial intelligence (AI)-generated outputs by governments and public institutions to support regulatory and administrative decision-making has been conceptualised as algorithmic governance [1]. Focusing on the integration of AI-based tools into water management, this PhD project examines how algorithmic governance, which relies on processes of categorization [2], reshapes collective meanings and everyday practices, thereby constituting “a form of social ordering” [3].

Building on a case study in Catalonia [4], the project analyses empirical data on AI-enabled smart metersinstalled in households to monitor water use during drought periods. These devices collect near real-time data on water usage and can infer not only quantities of water used but also types of use based on detected patterns. During drought periods, when authorities restrict both water volumes and permitted uses, such systems can signal potentially non-compliant practices. In doing so, they distinguish between citizens who use water “properly” and those who do not, thereby enacting categorization processes [2] and contributing to the reconfiguration of social norms around water use, where “social norms” are understood as “neither mere sanctioning behaviour, nor behaviour which just satisfies others’ expectations, but both” [5]. At the same time, translating these assessments into quantified indicators transforms pre-existing social meanings. Water, traditionally embedded in local environments and social practices, is increasingly reframed as a discrete and manageable resource that can be counted, allocated, and assigned a monetary value; and AI systems further reinforce this shift by treating water primarily as a resource to be optimized and managed [6].

ABM contribution

The project adopts a mixed-methods approach, combining documentary and media analysis, actor mapping, and semi-structured interviews to enable methodological and data triangulation [7]. Building on these empirical insights, an agent-based model (ABM) is being developed to explore the emergence and stabilisation of social norms related to water use. The model aims to explore a range of counterfactual (“what-if”) scenarios by combining, to varying degrees, the role of institutional communication (coming from human institutions) and algorithmic governance systems (operationalised through smart meters) in monitoring and regulating behaviour. In particular, the ABM enables comparison across scenarios ranging from those where institutional communication occurs without explicit sanctioning mechanisms, to contexts in which behavioural control is delegated to algorithmic systems that classify and sanction users based on their consumption patterns, as well as hybrid configurations. The ABM thus complements qualitative analysis by allowing the exploration of causal mechanisms through controlled parameter variation [8]. The model integrates both quantitative information (e.g. water consumption data) and qualitative insights derived from the analysis of institutional discourses and social practices, thereby capturing both behavioural dynamics and meaning-related dimensions within the simulation.

Literature

[1] Blankenship, J. (2020). Algorithmic governance. In A. Kobayashi (Ed.), International Encyclopedia of Human Geography (2nd ed., Vol. 1, pp. 105–109). Elsevier. https://doi.org/10.1016/B978-0-08-102295-5.10509-8

[2] Fourcade, M. (2016) Ordinalization. Lewis A. Coser memorial award for theoretical agenda setting. Sociological Theory. vol. 34, no. 3, pp. 175–195. https://doi.org/10.1177/0735275116665876

[3] Katzenbach, C., & Ulbricht, L. (2019). Algorithmic governance. Internet Policy Review, 8(4). https://doi.org/10.14763/2019.4.1424

[4] Catalonia Law 16/2017, of August 1, on climate change. DOGC no. 7426, 03/08/2017. Consolidated text.

[5] Linares Martínez, F. (2024). Una simulación multi-agente del mecanismo de generalización de una norma social. Revista Española De Investigaciones Sociológicas, (138), 19–40. https://doi.org/10.5477/cis/reis.138.19

[6] Bakker, K. (2005). Neoliberalizing Nature? Market Environmentalism in Water Supply in England and Wales. Annals of the Association of American Geographers, 95(3), 542–565. https://doi.org/10.1111/j.1467-8306.2005.00474.x

[7] Betrián Villas, E., Galitó Gispert, N., García Merino, N., Monclús, G. J., & Macarulla Garcia, M. (2016). La Triangulación Múltiple como Estrategia Metodológica. REICE. Revista Iberoamericana Sobre Calidad, Eficacia Y Cambio En Educación, 11(4). https://doi.org/10.15366/reice2013.11.4.001

[8] Edmonds, B., & Hales, D. (2005). Computational Simulation as Theoretical Experiment. The Journal of Mathematical Sociology, 29(3), 209–232. https://doi.org/10.1080/00222500590921283

There have been increasing calls for the use of agent-based models (ABMs) to inform public health policy in recent years [1]. Dietary behaviour is well suited to ABM approaches, as it arises from complex, interacting processes that are difficult to capture using traditional models. Obesity and food insecurity are interrelated challenges, both influenced by food environments including food access and availability. Understanding the potential impact of place-based interventions that change food environments requires models that can represent these interdependencies and underlying mechanisms.

This poster presents the design of NOURISH (Nutrition, Obesity, and Unhealthy food Relationships in food Insecure populations: Systems modelling of Health-related policies and outcomes), a mechanism-based agent-based model (ABM) developed to represent how local food environment policies influence dietary behaviour, food insecurity, and health inequalities.

The model is designed to initially appraise two place-based interventions: (1) restrictions on out-of-home advertising of high fat, sugar, and salt (HFSS) foods, and (2) business rates relief for fruit and vegetable retailers in deprived areas with low supermarket access, although other place-based interventions could be evaluated in future work. The model design follows the mechanism-based social systems modelling (MBSSM) architecture [2] and adopts a macro-micro-macro structure, where heterogeneous individuals are embedded within Lower Super Output Area (LSOA) local food environments.

The model is grounded in stakeholder-informed behavioural systems mapping, integrating evidence from academic, public health, and third sector stakeholders to identify key pathways linking policy, food environments, food insecurity, and dietary behaviour. These are formalised using the COM-B framework, with behaviour arising from capability, opportunity and motivation. A synthetic population will inform key model variables using statistically merged national and local datasets to represent key sociodemographic, behavioural and environmental variables. Structural equation modelling is used to estimate initial relationships between observed variables and composite COM-B constructs. The model will be calibrated to reproduce population-level patterns in obesity, diet and food insecurity.

The model design specifies how policy-driven changes to LSOA-level food environments (e.g., advertising exposure) are perceived by individuals and influence dietary behaviours (HFSS and fruit and vegetable consumption) at weekly time steps. These behaviours determine individuals’ calorie intake and body weight and are designed to link to a health economic microsimulation model [3] to estimate long-term health outcomes.

The NOURISH model will provide a theoretically grounded and empirically informed framework for simulating the estimated impact of food environment policies. By integrating stakeholder input, behavioural theory, and population data, the model will enable policy appraisal and scenario analysis, alongside exploring mechanisms driving dietary behaviour change.

[1] M. Tracy, M. Cerdá, and K. M. Keyes, ‘Agent-Based Modeling in Public Health: Current Applications and Future Directions’, Annual Review of Public Health, vol. 39, no. Volume 39, 2018. Annual Reviews, pp. 77–94, 2018. doi: https://doi.org/10.1146/annurev-publhealth-040617-014317.

[2] T. M. Vu et al., ‘A software architecture for mechanism-based social systems modelling in agent-based simulation models’, J. Artif. Soc. Soc. Simul. JASSS, vol. 23, no. 3, 2020.

[3] P. R. Breeze et al., ‘SPHR Diabetes Prevention Model: Detailed description of model background, methods, assumptions and parameters’, 2015.

Among all modeling methods, agent-based models (ABM) are the most advanced in systematically incorporating social science insights to develop behaviorally realistic representations of human behavior [1, 2, 3]. Yet, implementing behavioral processes in an ABM remains difficult, both because there are many theories across the social sciences that describe similar behaviors [4, 5], and because often, these theories operate with vague, qualitative, and partially ambiguous concepts that need explicit quantitative formalization before they are ready to use in computational models [6, 7, 8]. To address these issues, better integration is needed between theory, empirical, and modelling research [6, 8, 9, 10, 11]. Importantly, this integration is a two-way street, with simulation models increasingly adding value for theory development, especially for exploring dynamic feedbacks that are otherwise underrepresented in social science theories [1, 9, 10]. Our current work aims to add to this goal by studying temporal dynamics of decision-making processes in the context of households’ climate change adaptation (CCA) behavior.

The current empirical understanding of CCA behavior builds on Protection Motivation Theory (PMT) as a theoretical basis [12, 13]. However, these findings are largely based on cross-sectional studies, which leads to several issues in translating this knowledge into models. First, little is known about possible feedback loops between adaptation behavior and its drivers [14, 15], leaving open whether and how such feedback loops should be modeled. Second, to avoid measuring changes in drivers resulting from such feedback loops, it is common practice to measure intentions rather than actual behavior [16]. Often, however, there is a gap between them (the so-called intention-behavior gap, or: IBG) [17, 18], and how to best represent this in ABMs remains largely unknown. Finally, cross-sectional studies provide limited information on interactions between different drivers and how they may change over time [14], leaving open questions on how to connect these static factors to the dynamically changing context.

In our current work, we utilize previously collected longitudinal survey data [19, 20] to study these temporal aspects of the households’ CCA decision-making process. In our ongoing study of the IBG and the extent to which previous PMT-based work on intentions generalizes to explain actions we find that, for most types of CCA behavior, intentions form a crucial step in the behavioral process, as most individuals do not act without previously stated intent. However, we also find that there is still a significant IBG that remains largely unexplained by PMT, confirming its importance in CCA decision-making.

At SSC 2026, we plan to show how our empirical analysis of the IBG can improve the representation of individual decision-making processes in ABMs. We will also discuss next steps to utilize this longitudinal dataset to study anticipated feedback loops (e.g. the effects of hazard events on dynamics of psychological drivers) both empirically and within ABMs, to gain insight into these behaviors, and how this could contribute to fundamental knowledge useful for further theory development of these behavioral processes.

References:

1. Verburg, Peter H., John A. Dearing, James G. Dyke, Sander van der Leeuw, Sybil Seitzinger, Will Steffen, and James Syvitski. 2016. “Methods and Approaches to Modelling the Anthropocene.” Global Environmental Change 39:328–40. doi:10.1016/j.gloenvcha.2015.08.007

2. Schill, Caroline, John M. Anderies, Therese Lindahl, Carl Folke, Stephen Polasky, Juan Camilo Cárdenas, Anne-Sophie Crépin, Marco A. Janssen, Jon Norberg, and Maja Schlüter. 2019. “A More Dynamic Understanding of Human Behaviour for the Anthropocene.” Nature Sustainability 2(12):1075–82. doi:10.1038/s41893-019-0419-7.

3. An, Li. 2012. “Modeling Human Decisions in Coupled Human and Natural Systems: Review of Agent-Based Models.” Ecological Modelling 229:25–36. doi:10.1016/j.ecolmodel.2011.07.010.
4. Schlüter, Maja, Andres Baeza, Gunnar Dressler, Karin Frank, Jürgen Groeneveld, Wander Jager, Marco A. Janssen, Ryan R. J. McAllister, Birgit Müller, Kirill Orach, Nina Schwarz, and Nanda Wijermans. 2017. “A Framework for Mapping and Comparing Behavioural Theories in Models of Social-Ecological Systems.” Ecological Economics 131:21–35. doi:10.1016/j.ecolecon.2016.08.008.
5. Wijermans, Nanda, Geeske Scholz, Émile Chappin, Alison Heppenstall, Tatiana Filatova, J. Gareth Polhill, Christina Semeniuk, and Frithjof Stöppler. 2023. “Agent Decision-Making: The Elephant in the Room - Enabling the Justification of Decision Model Fit in Social-Ecological Models.” Environmental Modelling & Software 170:105850. doi:10.1016/j.envsoft.2023.105850.
6. Muelder, Hannah, and Tatiana Filatova. 2018. “One Theory - Many Formalizations: Testing Different Code Implementations of the Theory of Planned Behaviour in Energy Agent-Based Models.” Journal of Artificial Societies and Social Simulation 21(4):5. doi:https://doi.org/10.18564/jasss.3855.
7. Jager, Wander. 2017. “Enhancing the Realism of Simulation (EROS): On Implementing and Developing Psychological Theory in Social Simulation.” Journal of Artificial Societies and Social Simulation 20(3):14.
8. Schwarz, Nina, Gunnar Dressler, Karin Frank, Wander Jager, Marco Janssen, Birgit Muller, Maja Schluter, Nanda Wijermans, and Jurgen Groeneveld. 2020. “Formalising Theories of Human Decision-Making for Agent-Based Modelling of Social-Ecological Systems: Practical Lessons Learned and Ways Forward.” Socio-Environmental Systems Modelling 2:16340–16340. doi:10.18174/sesmo.2020a16340.
9. Elsawah, Sondoss, Tatiana Filatova, Anthony J. Jakeman, Albert J. Kettner, Moira L. Zellner, Ioannis N. Athanasiadis, Serena H. Hamilton, Robert L. Axtell, Daniel G. Brown, Jonathan M. Gilligan, Marco A. Janssen, Derek T. Robinson, Julie Rozenberg, Isaac I. T. Ullah, and Steve J. Lade. 2020. “Eight Grand Challenges in Socio-Environmental Systems Modeling.” Socio-Environmental Systems Modelling 2:16226–16226. doi:10.18174/sesmo.2020a16226.
10. Schlüter, Maja, Christa Brelsford, Paul J. Ferraro, Kirill Orach, Minghao Qiu, and Martin D. Smith. 2023. “Unraveling Complex Causal Processes That Affect Sustainability Requires More Integration between Empirical and Modeling Approaches.” Proceedings of the National Academy of Sciences 120(41):e2215676120. doi:10.1073/pnas.2215676120.
11. Grimm, Volker, Uta Berger, Matthias Meyer, and Iris Lorscheid. 2024. “Theory for and from Agent-Based Modelling: Insights from a Virtual Special Issue and a Vision.” Environmental Modelling & Software 178:106088. doi:10.1016/j.envsoft.2024.106088.
12. van Valkengoed, Anne M., and Linda Steg. 2019. “Meta-Analyses of Factors Motivating Climate Change Adaptation Behaviour.” Nature Climate Change 9(2):158–63. doi:10.1038/s41558-018-0371-y.
13. Bamberg, Sebastian, Torsten Masson, Katrin Brewitt, and Natascha Nemetschek. 2017. “Threat, Coping and Flood Prevention – A Meta-Analysis.” Journal of Environmental Psychology 54:116–26. doi:10.1016/j.jenvp.2017.08.001.
14. Hudson, Paul, Annegret H. Thieken, and Philip Bubeck. 2019. “The Challenges of Longitudinal Surveys in the Flood Risk Domain.” Journal of Risk Research 23(5):642–63. doi:10.1080/13669877.2019.1617339.
15. Siegrist, Michael. 2013. “The Necessity for Longitudinal Studies in Risk Perception Research.” Risk Analysis 33(1):50–51. doi:10.1111/j.1539-6924.2012.01941.x.
16. Bubeck, Philip, and W. J. Wouter Botzen. 2013. “Response to ‘The Necessity for Longitudinal Studies in Risk Perception Research.’” Risk Analysis 33(5):760–62. doi:10.1111/risa.12028.
17. Sheeran, Paschal. 2002. “Intention—Behavior Relations: A Conceptual and Empirical Review.” European Review of Social Psychology 12(1):1–36. doi:10.1080/14792772143000003.
18. Osberghaus, Daniel, W. J. Wouter Botzen, and Martin Kesternich. 2025. “The Intention-Behavior Gap in Climate Change Adaptation: Evidence from Longitudinal Survey Data.” Ecological Economics 231:108543. doi:10.1016/j.ecolecon.2025.108543.
19. Filatova, T., Noll, B., Need, A., & Wagenblast, T. (2022). SCALAR household climate change adaptation and resilience survey (Version v6) [Data set]. DANS Data Station Social Sciences and Humanities. https://doi.org/10.17026/DANS-X9H-NJ3W
20. Noll, Brayton, Tatiana Filatova, Ariana Need, and Alessandro Taberna. 2022. “Contextualizing Cross-National Patterns in Household Climate Change Adaptation.” Nature Climate Change 12(1):30–35. doi:10.1038/s41558-021-01222-3.

Background
Large Language Models (LLMs) have rapidly advanced across various domains. In transport research, LLMs support data management, mobility forecast, and network analysis, demonstrating strong potential in interpreting complex transport systems. However, limited research has applied LLMs to understand the transport-related information embedded in social contexts. The labour market is a particularly relevant case, where transport requirements, advantages and benefits influence access to employment. This study aims to evaluate the performance of different LLMs and task-adaption strategies in identifying transport-related factors within job advertisements and to develop an optimized model for this task.

Methods
This study employs four locally deployed LLMs (LLaMA-3.1-8B, Mistral-7B, Qwen-2.5-7B, and Qwen-3.5-9B) alongside a fine-tuned RoBERTa classifier as a traditional NLP baseline. The models are tasked to identify transport-related information in job advertisements and classify it across three themes (transport requirements, transport advantages, and employer-provided transport benefits) and 40 sub-categories. The analysis draws on the Adzuna-v2 dataset, covering over 400 million UK job advertisements from 2016 to 2025. Using a human-annotated subset, multiple task-adaptation strategies are applied in sequence: all LLMs are first evaluated under zero-shot conditions to establish baselines, followed by iterative testing of two prompts with varying levels of instruction complexity. The two best-performing models from the prompt engineering stage are then fine-tuned on 1,500 annotated job listings and benchmarked against both prompt-engineered baselines and the RoBERTa classifier. Model performance is evaluated through standard classification metrics and qualitative examination of the explainability on selected postings.

Results
Results show that LLMs outperformed the trained NLP classifier in complex classification tasks. While the RoBERTa classifier achieved acceptable performance at the theme level, its accuracy declined substantially at the transport type level, suggesting the limitations of conventional approaches when handling fine-grained classification with constrained annotated training data. Among LLMs, Qwen 3.5 achieved the strongest zero-shot and few-shot performance. Prompt engineering with detailed categorical definitions, exclusion examples, and explicit boundary conditions further improved classification performance for indirect transport types. For example, the designed prompt instructions enable the model to distinguish administrative travel-booking responsibilities from actual work travel requirements. The fine-tuned models further outperformed the prompt engineering condition in type-level classification, with a significant improvement on boundary cases where requires contextual interpretation. Overall, the findings demonstrate a cross-model comparison based on different task adaptation conditions. The final optimized model achieved an F1-score of 0.87 at the theme level and above 0.80 for the majority of transport types, showing that fine-tuned LLMs offer a robust solution for identifying transport-related factors in unstructured labour market texts.

Contribution
By proposing a methodology framework for detecting transport-related factors in labour-market texts, this study offers a new perspective on transport-employment relationships from the employer-side evidence. The proposed framework enables large-scale extraction of transport information from job advertisements, supporting new data source for transport and employment studies. Beyond this case, the framework is transferable to other corpora containing transport-related information, offering broader methodological insights for adapting LLMs to text-based tasks in transport research.

Keywords:
Transport, Employment, Large Language Models, Labour markets

In a world of tightening border controls, it is imperative for firms and governments to understand the impacts of changes to short-term international mobility. Long-run trade effects of mobility frictions are generally simulated using structural, representative agent models at the macro level. This paper instead applies a complexity economics framework, through a bottom-up agent-based modelling (ABM) approach, to assess the impacts of changes in visa policy.

We develop an ABM of trade that captures how micro-level mobility changes affect macro-level trade outcomes using ‘zero-intelligence’ (ZI) agents informed using UK business microdata. Simulating trade outcomes at the agent-level distinctly permits modulation of market participation. Furthermore, it highlights distributional impacts of border-policy shocks, revealing ‘winners and losers’ across countries, sectors and agent-characteristics.

This ABM innovates by modelling trade from a complexity-economics perspective, leveraging UK business microdata to shock firm-level network structures and, applying this agent-based approach to assessing how visa policy reshapes interactions and generates new trade outcomes within regions and specific sectors.

We leverage UK business microdata to inform agent interactions by introducing empirically grounded network structures into the matching process, consistent with the view that trade is shaped by persistent buyer-seller relationships and search frictions rather than fully random encounters. Specifically, agents are embedded in interaction networks informed by firm characteristics and geography, which constrains the set of feasible trading partners and anchors simulated trade in observed economic relationships. Mobility‑policy shocks then operate by influencing specific cross‑border links, so their effects depend on where international ties are concentrated: disruptions can therefore generate sharply sectoral and regionally uneven consequences via the network. We find that aggregate trade responses can be decomposed into heterogeneous and path‑dependent impacts across sectors, regions, and firm types, allowing policy impacts to be traced to specific network structures rather than absorbed into averages.

In our model, agents are either buyers or sellers, and expected trade is a function of matching and bidding between them based on their network positioning and random valuations (hence, ZI). Trade emerges from two auctions: consumer-to-firm trades in one, and firm-to-firm trades in another. We model two countries directly affected by a border-policy shock, and focus on the first-order impact of the border-policy shock, thereby implicating only a relatively small subset of agents which match and bid internationally.

Model performance is validated by comparing model predictions with empirical outcomes, bootstrapping simulations to construct confidence intervals, and comparing outputs with benchmark methods. Initial, aggregate results suggest that the imposition of restrictions on short term mobility may reduce exports by the imposing country and can be decomposed into reductions in market thickness and lower in matching rates.

Climate change is increasing the frequency and intensity of heatwaves, posing growing risks to urban populations. Population ageing amplifies the proportion of vulnerable individuals, amplifying population-level exposure to extreme heat. At the same time, people develop social practices to adapt to increasing heat. Understanding how environmental and social changes interact is thus essential for assessing heat-related risks. However, cross-sectional and static empirical methods (e.g., survey-based analyses) are limited in their ability to capture the dynamic and nonlinear nature of adaptive behaviour. While modelling offers tools to represent such dynamics, existing approaches often rely on aggregate indicators (e.g., mortality and morbidity), obscuring behavioural processes underlying heat exposure and the lived reality of individuals. This creates a need for modelling approaches that explicitly represent individual behaviour, which is heterogeneous in exposure, individual characteristics and heat-related adaptation practices.

The aim of this study is to examine how environmental conditions, population ageing, evolving social practices, and urban changes jointly shape individuals’ ability to maintain thermal comfort during heatwaves, and to evaluate the effectiveness of both private behavioural and public urban adaptation strategies.

To address this, we develop an Agent-Based Model (ABM) simulating how individuals adapt their behaviour under varying environmental, social, and urban conditions. Agents represent heterogeneous individuals characterised by sociodemographic attributes, health, lifestyle, and housing conditions. Their objective is to maintain thermal comfort under changing summer conditions by adjusting to perceived environmental conditions. Individual behavioural adaptations of agents are grounded in Social Practice Theory [1], which frames behaviour as socially structured practices shaped by individual lived experiences and material contexts. As the behavioural empirical foundation, we employ longitudinal data collected in summer 2025 in Warsaw (N=312). Data collection began with a face-to-face questionnaire survey, followed by a 3-week Experience Sampling Method study capturing participants’ behaviour and sensations, along with measurements of ambient temperature and humidity. The obtained data enable the development of heterogeneous and behaviourally-rich agents, facilitating a better understanding of within-person variability in thermal conditions. The spatial environment in our ABM is modelled to capture key features of urban design, such as urban greenery and building density, allowing for spatial variation in heat exposure.

Simulations are conducted across multiple summer seasons under stochastic conditions to explore variation in individual behaviour, leading to differing emergent patterns of individual thermal comfort and evolving social practices in the population. Our design of experiments systematically explores cumulative and distributional effects of public adaptation policies to heat, ranging from changes to urban design (e.g., increasing urban greenery) to measures promoting private adaptation practices (e.g., improving adaptation capacity by means of cooling devices).

To address our main research goal, we report the ABM results in terms of aggregate thermal comfort patterns and their distributions across modelled populations, providing insight into heterogeneity in heat exposure and adaptive capacity. By integrating behavioural dynamics with the study of social and environmental processes, we aspire to deliver insights into patterns of adaptive capacity and thermal comfort, informing policies that shape heat adaptation, especially given future climate uncertainty.

References:
1. Blue, S.: Practice Theory and Process Philosophy: Towards a Sociology of Becoming. Taylor & Francis, London (2025)

As the effects of climate change intensify across Europe, local authorities (LAs) have emerged as frontline actors in the development and implementation of adaptation strategies. Urban areas, home to over 75% of the European Union’s population, face heightened exposure to a range of climate risks—from coastal and pluvial flooding to intensified heat waves exacerbated by the urban heat island effect. Consequently, local authorities are called to put in place mechanisms to manage the context-specific risks climate change bears on citizens, businesses and infrastructures. However, cities’ ability to plan and coordinate adaptation varies considerably from place to place, with wealthier municipalities generally performing better than their less-resourced counterparts. In this context, the transnational exchange of knowledge and experience between cities is critical, as it may bridge the capacity gap through diffusion of best practices.

This article investigates the ways in which cities use existing networks to advance their agenda in terms of climate adaptation policy. Two main types of networks currently exist: i) Transnational Municipal Networks (TMNs), which are non-hierarchical networks where municipalities exchange knowledge and experiences; and ii) research and innovation projects connecting knowledge creation on the part of Universities and Institutes, to knowledge circulation with and among cities. Research on transnational cooperation among European cities has left two research gaps that this article aims to bridge. First, while significant research has taken TMNs as their object, studies rarely adopt the cities’ perspective—yet the constitutive part of these networks—or explore the usages they have of these networks. Second, research eludes other forms of cooperation beyond TMNs, leaving research and innovation projects out of their scope, despite their fundamental importance in terms of knowledge creation and diffusion.

This article investigates cities’ strategic usages of networked governance in the transnational knowledge ecosystem with a view to simulate alternative scenarios for the betterment of knowledge circulation in Europe. This article draws on original data, Social Network Analysis, semi-structured interviews, and inferential statistics to inform an Agent Based Model that simulates the circulation of knowledge as a result of cities’ pursuance of their strategic interests.

Interventions that leverage social networks typically aim to maximise the adoption of new ideas, practices, or health behaviours across as many individuals as possible. However, widespread adoption does not necessarily imply equitable outcomes, understood here as inequalities in access to resources across individuals or groups within the network depending on their specific needs. To assess the trade-off between adoption and inequity, we introduce a network inequity metric that measures the extent to which people have comparable weighted access to resources via their social relationships. Using agent-based modelling, we simulate adoption via complex contagion across several ideal social network structures and compare network inequity to the proportion of the network adopting. The results show that the trade-off between inequity and adoption is not straightforward, as this depends on the underlying social network structure. Moreover, we show that depending on how the specific needs of people in the network are formalized, the network inequity shows markedly different results whilst holding the proportion of the network adopting constant. Together, the results highlight gaps in our current thinking about social network interventions and guide future efforts in monitoring and anticipating inequity.

Extended Abstract:

1 Introduction

With roughly 50% of forest-covered land in Switzerland privately owned, understanding how small private forest owners (SFOs) make decisions about forest biodiversity and ecosystem service (BES) provision is essential for reaching Swiss policy targets (FOEN., 2025). However, effective outreach and engagement of SFOs in active forest management proves difficult for policy actors (Stockmann et al., 2024). Issues arise from the high number of forest owners, their heterogeneity and fragmentation (Schmithüsen & Hirsch, 2010; Tiebel et al., 2024; Wiersum et al., 2005). Professional forestry actors such as rangers serve as key intermediaries in the engagement with private forest owners, but their resources are often not sufficient for the high administrative costs the management of non-coordinated small forest plots produce (UNECE, 2020).

To address this gap, we develop an agent-based model (ABM) to simulate the spread of management practices through SFO social networks over several forest management cycles. The simulations build on empirical data of social networks, intentions of collaboration and personal utility of forest BES, enabling to test how incentive structures and social network configurations shape SFO decision-making and, in turn, landscape-level BES provisioning outcomes.

2 Theory

The empirical data collected and the processes in the ABM are guided by three main theories, already well embedded in agent-based modelling. Firstly, social network theory (Fuhse, 2018) defines the channels through which management offers and BES provision options reach SFOs and allows us to evaluate which actor characteristics hold important nodes to influence management adoption and BES provision. We assume that available management propositions are affected by network position and that this position also influences their BES provision goals. Rangers occupy structurally central bridging positions, initiating offers to owners within their professional reach, while peer owners reinforce adoption through observation. SFOs who see neighbours participating in coordinated management experience positive norm reinforcement, which incrementally shifts their own willingness to engage. Social network theory also provides a framework of S-shaped adoption patterns (Rogers, 2003) against which the incentive practices for each BES can be compared, establishing which SFOs are early adopters and which are laggards.

The decision processes of private forest owners are separated into two main steps. Using theory of planned behaviour (TPB; Ajzen, 1991), the cost-intensive practice of mobilizing SFOs to engage in cross-border collaborations is approached. Following the theory, we include social network influences from stated social norms, defining which actors are perceived as important by SFOs and which actions are supported by those actors. An SFO's willingness to collaborate is shaped by attitudes toward collective BES delivery, subjective norms peers and professional advisors, and perceived behavioural control over participation. The TPB module produces an intention score that acts as a threshold gate. Owners whose intention falls below the threshold do not evaluate incoming offers, regardless of contract terms.

For owners who pass this intention gate, a second step draws on random utility theory (RUT; Manski, 1977), operationalized through a discrete choice experiment (DCE). This theory assumes that individuals select the option with the highest personal utility, which comprises both observed and unobserved components rather than a single measure. Each contract offer is characterized by attributes including management aims, financial compensation, contract duration and flexibility clauses. The DCE yields part-worth utilities that allow the model to predict whether an owner accepts a specific contract. This two-stage architecture separates the question of whether an owner is willing to participate in principle from whether a specific offer is attractive enough to accept, navigating the constraints of pure rationality assumptions that are disputed for SFO land-use decisions (Domínguez & Shannon, 2011; Ficko et al., 2019).

By grounding the intention gate in empirical TPB survey scores and the contract evaluation in DCE-derived utilities, we reduce reliance on assumed behavioural parameters while maintaining an interpretable theoretical structure. This structure adds to the ongoing discussion on how to operationalize decision theories in agent based models (Muelder & Filatova, 2018; Taraghi & Yoder, 2025).

Two hypotheses structure the analysis. First, we expect that BES provision aim adoption follows an S-curve dynamic driven primarily by the network position of early adopters rather than the incentive level. This implies that targeting structurally central owners yields faster diffusion than increasing payments across the board. Second, we hypothesize that scenarios maximizing total SFO participation do not necessarily maximize spatially contiguous participation or a diverse set of produces BES. The incentive designs that perform best on overall enrolment and those that produce the most spatially clustered uptake may diverge, revealing a coordination-configuration trade-off with implications for landscape-level BES provisioning effectiveness.

3 Method

The model is defined as a landscape in the Basel region with three agent types. SFOs are the primary decision-making agents, each carrying an individual TPB intention score and DCE-derived part-worth utilities reflecting their heterogeneous preferences. Rangers are intermediary agents parameterized from semi-structured interview data capturing outreach strategies, network reach and institutional constraints. Forest plots as spatially explicit units initialized from national forest maps with ownership attribution.

The ABM is calibrated and initialized with the findings on the social networks (Will et al., 2020) and a utility function based on the DCE (Holm et al., 2016). The model simulates the spread of BES provision aims through the social networks over several management cycles. Two crossed scenario dimensions are explored: the first changes the incentive levels for specific BES provisions or support for BES engagement. The second varies network structure reflecting policy strategies for SFO management support. Comparing simulation outputs provides insights on which BES provision aims spread more readily through the social network, and which incentives are most effective.

4 Contributions

This model aims to contribute to three aspects: methodologically it demonstrates how to formally integrate TPB intention formation with DCE-parameterized contract evaluation in a two-stage decision architecture, extending the DCE-ABM approach from market transactions (Holm et al., 2016) and technology adoption (Chappin et al., 2022) to collaborative forest governance. From a policy perspective, the crossed scenario design provides evidence on whether public investment in forest coordination is more effectively channelled through owner incentives or ranger capacity and allows decision-makers to evaluate trade-offs between participation rates and spatial configuration of BES provision uptake. The model's modular architecture is designed to accommodate future extension with an ecological submodule coupling owner decisions to BES outcomes, enabling analysis of governance-ecology trade-offs in a socio-ecological systems framework.

Residential segregation remains one of the most persistent and widely studied phenomena in urban sociology. From classical urban theory to contemporary debates on inequality and social cohesion, segregation is understood as both a structural outcome and a driver of social stratification. A central insight of research in this area is that even relatively weak individual preferences can give rise to strong aggregate patterns of separation. Agent-based models, most prominently those based on the assumptions of Schelling (1978, 2006), have played a crucial role in demonstrating how local interactions generate macro-level segregation. These models show that even mild preferences for similar neighbours can lead to highly segregated spatial configurations. At the same time, however, they rely on a key simplifying assumption: preferences are typically treated as static. This assumption stands in contrast to empirical observations. Social preferences, including tolerance toward other groups, are not fixed but may evolve through interaction. In particular, intergroup contact has been shown to influence attitudes and reduce prejudice under certain conditions (Allport, 1991). From this perspective, segregation is not only an outcome of preferences but also a context in which these preferences are formed and transformed. Against this background, the present contribution situates agent-based simulation within a broader understanding of sociology as an experimental and exploratory science of possible futures. Rather than merely reproducing known dynamics, simulations can be used to systematically investigate how alternative assumptions about social behaviour, such as adaptive tolerance, shape the evolution of social systems.

This paper proposes an exploratory agent-based model to analyse how digital interaction practices generate and transform social structure in sports clubs. The model integrates structuration theory, activity theory, and social network analysis into a unified multi-level framework linking decision-making, interaction dynamics, and structural emergence. Social structure, understood as domination, legitimation,
and signification, emerges endogenously from repeated interactions and recursively constrains agent behaviour.
The modelling approach combines explicit formalization of perception, decision, and interaction mechanisms with a three-stage calibration procedure [6].
Results from 230 simulations show that imposed structural configurations do not persist, as local interactions rebalance structural dimensions. Network indicators remain stable, suggesting that relational topology acts as a structural substrate, while transformations occur at the level of practices.
The paper contributes to social simulation methodology [5] and outlines a neuro-symbolic extension integrating ontology-based representations and empirical data from 25 sports clubs.

The rapidly evolving digital political information environment—the ecosystem of online platforms, actors, content, and algorithmic processes through which citizens encounter and exchange politically relevant information—offers expanded avenues for political participation. At the same time, it also introduces significant challenges to democratic citizenship by enabling the spread of harmful and manipulative information. In this study, we build on existing simulations of social media phenomena to develop an agent-based model that aims to more realistically represent online public discourse. The model integrates widely shared social media features and affordances, while drawing on theoretical and empirical insights from communication research and neighboring fields. The purpose of the model is to develop hypotheses about attitude change among ordinary citizens exposed to political information from different sources and under varying conditions.

People need context to understand a situation. In Europe, ICLEI connects researchers, planners, facilitators, implementers, and politicians working on climate change adaptation. For knowledge to be implemented in a municipality, everyone involved needs a shared understanding, which is not always the case. Using frames as contextual descriptions, drawn from in-depth interviews with ICLEI participants, reveals how opportunities and challenges emerge. Frames highlight bottlenecks in knowledge flow and climate adaptation, not just from lack of resources or political support. This paper analyses 20 interviews with researchers, facilitators, and implementers to show how these frames, seen as assemblages, reveal that different contexts create both obstacles and opportunities.

While looking at climate adaptation through frames provides input to the field, this paper takes a second angle by examining their usefulness in social simulations. The frame adds a filter of possible actions and capacities to human agency. Frames are contextual wrappers that allow artificial humans to select slices of their environment, modify them, and create new pieces. Frames created from empirical data, bridging a cognitive agent to small-scale, contextualised social events, pave the way for mimicking social and experiential learning that can change environments by evoking new, recognised individual and collective capabilities in multi-agent social simulations.

Urban planning in Japan faces mounting pressure from the simultaneous advance of demographic decline, rapid population ageing, and the deterioration of infrastructure built during the high-growth era. In many local cities, compound disaster risk persists, including the interaction of earthquake, flood, and severe winter conditions, while fiscal capacity weakens. As a result, maintaining existing settlements, facilities, and infrastructures in their current form is becoming increasingly difficult, calling for anticipatory approaches to spatial adaptation rather than post-disaster response alone.

Existing studies have identified hazardous and relatively safer areas and have discussed relocation as a planning option. Yet many remain static or primarily spatial, with limited attention to how relocation unfolds as a social and policy process over time. The present study therefore develops an agent-based model to capture relocation as a dynamic process shaped by household decisions, local demographic constraints, and public facility reorganization.

This poster presents a NetLogo-based agent-based model designed to examine how public facility provision in safer areas — such as school consolidation — can induce voluntary household relocation, thereby reducing long-term exposure in high-risk zones. Households are represented as heterogeneous agents whose attributes — location, demographic composition, and household type — are derived from synthetic population microdata. Agents interact within a space structured by compound hazard conditions, safer receiving areas, and key public facilities, and update their relocation tendency through an ε-greedy reinforcement learning mechanism that captures adaptive behaviour under repeated risk exposure.

The model draws on a prior spatial analysis of Aomori City examining multi-hazard conditions, evacuation accessibility, and school-district-scale relocation potential. Because Aomori is a heavy-snow region, the model reflects how multi-hazard exposure is further conditioned by seasonal differences in evacuation time and route availability. By integrating geospatial hazard screening with individual-level synthetic population data, this study provides an empirical basis for the model's initial conditions and for representing the socio-spatial constraints specific to Aomori.

The poster compares stylized scenarios under compound disaster conditions, including no anticipatory relocation, voluntary relocation, incentivized relocation, and relocation linked to school or shelter reorganization. Scenarios are compared using indicators of compound exposure, seasonal access to safer shelters or schools, the share of households remaining in high-risk areas, and pressure on receiving areas. The evaluation additionally tracks community-level consequences: continuity of school-district catchments, disruption of established social ties, and changes in the spatial concentration of elderly residents in receiving zones.

Although still under development, the model helps clarify how relocation policies interact with local social structure, public facilities, and capacity constraints. The findings will inform anticipatory planning practice in demographically declining cities facing spatially uneven compound disaster risk.