Behavioral feedbacks to climate mitigation under intensifying disasters

Summary

The current IPCC assessment outlines idealized pathways to limit warming below 2°C, yet record-high emissions in 2024 breached the 1.5°C threshold, underscoring the persistent gap between aspirational targets and real-world progress. Meanwhile, frequent extreme weather events increasingly trigger complex human responses, influencing public beliefs, risk perceptions, and support for climate policies, factors not captured by existing integrated assessment models underpinning IPCC reports. Here, we explore how key behavioral feedbacks to mitigation—cognition, communication, and conflict—might evolve under escalating climate disruptions, potentially undermining global climate mitigation efforts. Given the profound implications of neglecting psychological and societal dynamics, and the paucity of research on futures with severe climate impacts, we propose a targeted research agenda for rigorously integrating these behavioral feedbacks into IAM-based scenarios, enhancing the realism and policy relevance of upcoming IPCC assessments.

Graphical abstract

Global change; Social sciences

Introduction

The IPCC states that limiting global warming below 2°C requires rapid GHG reductions—50% by 2030 and net-zero by 2050.¹ Yet, emissions hit record highs in 2024,² as US and Argentina withdrew from agreements and geopolitical instability further weakened globally coordinated mitigation.³ This divergence between necessity and reality highlights a stark gap between human aspirations and actions. These actions are further evolving against a backdrop of escalating climate extremes⁴: In 2023, over 90% of the globe experienced above-average temperatures; deadly heatwaves affected more than 2 billion people; floods displaced 30 million; sea levels rose at an average of 4.8 mm yr⁻¹ (2013–2023); global temperature rose at its fastest; and glaciers recorded unprecedented mass loss.

Extreme climate events can trigger diverse human responses. Empirical studies suggest that direct experience of climate-related disasters can trigger elevated climate concern and stronger subjective attribution of climate change to human causes, but emerging research shows that such effects are often short-lived, hazard-specific, and frequently constrained by economic and infrastructural barriers that inhibit sustained behavioral shifts.⁵ For instance, hurricanes may temporarily affect risk perceptions and policy preferences and show stronger impacts than floods or wildfires.⁶ Experimental and survey data show that the underlying responses are often shaped by political orientation, cultural context, and individual beliefs and coping mechanisms. Psychological research further highlights that emotions and cognitive processes shape climate engagement in heterogeneous and sometimes contradictory ways: Climate anxiety has been linked to increased activism, yet also to diminished wellbeing,^7,8 whereas hope can foster constructive engagement but may also coincide with inaction when based on complacency or externalized expectations of technological rescue. Collectively, these findings indicate that human responses to climate change are highly variable, context-dependent, and often difficult to sustain, complicating efforts to project behavioral trajectories and to design durable intervention strategies.

Existing frameworks for projecting and managing climate futures, such as those used in the IPCC scenarios, have underrepresented these human dimensions.⁹ Integrated assessment models (IAMs) that underpin these scenarios often assume idealized conditions such as seamless technology deployment, uniform policy implementation and perfectly rational, cooperative human behavior¹⁰ while inadequately accounting for humans acting irrationally and inconsistently, societies not supporting change, and geopolitics marred by conflict. This omission constitutes a major gap in climate scenario research, as severe climate impacts could trigger substantial cultural, economic, ecological, and technological shifts that fundamentally alter societal exposures, vulnerabilities, and ultimately global temperature change itself. Earth system modeling alone shows that accounting for such dynamics could widen projected end-of-century warming to 3.4°C–6.3°C, compared to a static baseline of 4.9°C.¹¹ As a result, existing climate mitigation pathways that emphasize emissions reductions and technological innovation but neglect these dynamics risk misrepresenting plausible futures and undermining the robustness of policy decisions.

In this work, we examine this critical but largely overlooked the dimension of climate mitigation pathways: behavioral feedbacks that could significantly delay, dilute, or derail global efforts to act. We use the term behavioral feedbacks deliberately to reflect that such dynamics are sign-ambiguous: they may either accelerate or impede mitigation depending on institutional, political, and climatic contexts. However, in our review of feedback mechanisms, and to spell out the implications for IAM-based IPCC scenarios, we focus on the risk of mitigation-impeding feedbacks, as current modeling frameworks often embed implicitly optimistic assumptions about behavioral adjustment and policy uptake. If behavioral responses were uniformly virtuous, their inclusion would only reinforce confidence in existing overly optimistic trajectories. By contrast, identifying conditions under which feedback impede mitigation is necessary from a precautionary, risk-management perspective. Having said this, the technical research agenda we offer here would detect both mitigation-impeding and -accelerating feedbacks, and is thus not based on optimistic/mono-directional assumptions. In addition, we use “behavior” in a generalized sense to include also institutional actors such as members of governments and firms. Behavioral feedback thus operates through institutional mediation rather than merely through the aggregation of individual responses.

We firstly group these feedbacks into three interconnected domains: cognition, communication, and conflict (Figure 1) - capturing psychological, societal, and geopolitical processes that can collectively disrupt climate initiatives. We then propose a targeted, novel research agenda: First, rather than relying on stylized or purely theoretical parameterizations of social dynamics, we outline an empirical identification agenda that uses VR experiments, AI-enhanced surveys, and event-linked social media analysis to estimate behavioral responses relevant to climate mitigation. Second, we articulate a micro-macro integration strategy that links heterogeneous micro-level models (MLMs) (e.g., agent-based models, ABMs) to macro-scale IAMs via meso-level economic structures such as multi-region input-output (MRIO) systems. Third, we design the agenda so as to explicitly dovetail with IPCC-style mitigation scenarios. Our contribution is therefore to operationalize behavioral feedbacks within policy-relevant long-term scenario architectures. Recognizing behavioral dimensions is timely and essential, as rising emissions and intensifying climate extremes expose the limits of models that still assume rational and cooperative human behavior. Incorporating such insights in IPCC scenario design could significantly change mitigation outcomes,¹¹ and are, therefore, of particular importance to the upcoming IPCC assessment cycle.

Figure 1.

Conceptual framework illustrates key human-related feedbacks (green) to climate mitigation, integration strategies (red), and the proposed research agenda (blue)

Framework and research gap

Behavioral feedbacks to climate action under severe future climate impacts

Human and societal dimensions that can constrain mitigation are largely overlooked in IPCC mitigation pathways that predominantly depend on technological and fiscal levers.¹² The global IAMs¹ underpinning these pathways typically assume idealized behaviors, such as 100% vegetarian diets in 2050 or uniform adoption of clean technologies, and rarely capture dynamic feedback between human behavior and climate change. Yet, a growing body of research^{3,5,6,7,13,14,15,16,17,18,19,20,21,22} from cognitive psychology, behavioral economics, sociology, and social psychology shows that cognitive biases, social denial, misinformation, and geopolitical instability and conflict significantly affect the modeled mitigation outcomes. Here, we synthesize these influences into three interacting behavioral feedbacks: cognition, communication, and conflict (green blocks in Figure 1). We use this tripartite scheme as a heuristic rather than a set of hard boundaries; mechanisms frequently co-occur and can cascade across levels. Operationally, we classify mechanisms by their primary locus and effect: Cognition denotes intra-individual perception and decision processes (e.g., risk appraisal, psychological distance, heuristics, perceived efficacy, and temporal discounting); Communication captures inter-individual and system-level information flows that govern exposure, credibility, belief updating, and social-norm signals; Conflict concerns inter-group and institutional incentives and constraints that shape coordination, policy feasibility, and resource allocation. Together, these three domains form a potential self-reinforcing feedback loop that can erode collective capacity for mitigation. Biased cognition limits the uptake of scientific evidence, which weakens public discourse and trust in institutions; misinformation and social fragmentation then amplify these distortions, while geopolitical and societal conflicts divert attention and resources away from mitigation. This sequence can entrench inaction, as denial fuels misinformation, misinformation deepens division, and division fosters further conflict, ultimately undermining global cooperation. A topical example is Germany’s recent policy reversal amid rising energy costs and the increased influence of the Alternative für Deutschland (AfD), where skepticism toward green reforms, polarized discourse, and institutional resistance converge as the loop unfolds.²³ Recognizing and quantifying these feedbacks is essential for embedding realistic human dynamics into future climate scenarios.

Cognition shapes climate-related perceptions and behaviors and often acts as a barrier to mitigation.^19,20,21 Cognitive biases contribute to climate risks underestimation, their perception as temporally or geographically distant, and the prioritization of immediate economic or personal concerns. The complexity of climate projections can reinforce skepticism, while confirmation bias strengthens pre-existing beliefs, leading to selective exposure to information. Social denial further compounds this effect, reducing emotional engagement and preventing the translation of expressed concern into concrete behavioral change.²⁴ Beyond individual cognition, collective identity and moral frameworks critically shape climate behavior: Cultural worldviews determine whether climate responsibility is viewed as a moral duty, social norm or economic cost, and identity-protective cognition helps explain why skepticism persists within ideological groups even when scientific consensus is accepted.²⁵

Communication dynamics amplify these cognitive barriers by translating individual biases into system-level distortions, especially in the post-truth era. The proliferation of misinformation and fake news, often strategically disseminated by political and economic actors, undermines scientific consensus and delays policy action. The rise of digital media accelerates echo-chamber effects, where individuals are mainly exposed to information that confirms their existing views, reinforcing ideological divides and weakening collective mobilization.²²

Conflict, a relatively underexplored feedback, both arises from and reinforces cognitive and communicative barriers. Geopolitical tensions stall negotiations, increase reliance on fossil fuels, and divert financial resources from mitigation initiatives. Political disputes obstruct agreements, while armed conflicts exacerbate emissions through military activities and economic instability.³ Trade disruptions, sanctions, and unfulfilled climate finance pledges further erode trust, undermining prospects for a cohesive global mitigation strategy.

Limited representation of behavioral and societal feedback in IAMs

Psychological and sociological research projecting behavioral response relationships into futures with severe climate impacts is fraught with complications and uncertainties. Will escalating disasters reduce psychological distance and denial, promote urgency and cooperation, or instead trigger short-term reactivity, deepen ideological divisions, and intensify conflict? Could post-truth dynamics collapse under undeniable human-made climate disasters, or will selective interpretation and entrenched polarization continue to obstruct climate action?

Prior technical work: While evidence remains limited and uneven across contexts, prior work has attempted to model human behavior more realistically: Nielsen et al.²⁶ outline a six-part research agenda to optimize the contributions of behavioral science toward climate action (see also SI 1): The authors propose six strategies: (1) expanding beyond consumer behavior to policy and collective action, (2) identifying leverage points through individual, social, and systemic interactions, (3) prioritizing scalable interventions, (4) using advanced mixed-method research, (5) ensuring global relevance through diverse contexts, and (6) integrating disciplines for cohesive climate strategies. A recent study by van den Berg et al.¹¹ integrates insights from psychology, sociology, and systems science to build scenario narratives for lifestyle-induced reductions of transport and residential emissions and feeds these narratives into the IMAGE IAM.²⁷ Their approach aims at enhancing the representational ability of IAMs by reflecting real-world challenges in achieving sustainable lifestyle changes. While the work emphasizes the importance of behavior for future scenarios, it does not delve into specific modeling methodologies. More importantly, neither Nielsen et al. nor Berg et al. touch on behavioral responses varying with increasing severity of climate impacts. A recent study³ suggests a potential vicious feedback loop in which increasing future global fragmentation and conflict could increasingly obstruct mitigation. Such work, though still emerging, underscores the need for the systematic integration of behavioral feedbacks into modeling frameworks. Depending on how psychological and social feedbacks evolve, markedly different emissions and climate futures could emerge, underscoring the urgency of revising IAMs to capture these dynamics.

Cognitive biases and behavioral uncertainty: While increasing exposure to climate events is often assumed to reduce psychological distance, empirical findings suggest that heightened risk perception does not always translate into sustained behavioral change.^28,29 Psychological resilience,³⁰ social norms, and policy constraints are often responsible for this inconsistency, since they prevent individuals from adopting sustainable behaviors.^31,32,33 These dynamics underscore the persistence of cognitive biases and the necessity of integrating such behavioral processes into models and policy frameworks to more accurately anticipate climate trajectories.³⁴ Capturing these dimensions in IAMs is essential because they govern perceptions of fairness and legitimacy in mitigation policies. For IAMs, this implies bounded elasticity of behavioral change and group-specific parameters (e.g., adoption thresholds, discount rates, and efficacy priors) rather than a single representative agent.

Climate change communication: Case studies on climate change communication³⁵ indicate that trust appears to be a pivotal factor in shaping climate policy decisions: Without efforts to rebuild trust, even well-designed narratives in support of interventions may fail to shift attitudes toward climate action. Given the broader societal trend of declining institutional trust, overcoming skepticism remains a critical barrier to effective climate change mitigation efforts.³⁵ If post-truth is seen as enabled by societal complexity and growth,³⁶ then a future decline of misinformation cannot be expected. Together, these cases illustrate trust-dependent diffusion, compliance fatigue, and culturally grounded leadership - features that can be represented via diffusion rates, compliance persistence, and context-specific norm priors in scenario design.

Conflict and climate change: Among the three feedbacks, the climate => conflict pathway is possibly best understood.^37,38,39 Studies link climate change to conflict, involving civil wars,⁴⁰ resource scarcity,⁴¹ migration,⁴² and political violence.^43,44,45 Indirect pathways, such as infrastructure failures from extreme weather events and/or heightened conflict risks in politically fragile states,⁴⁶ further illustrate the diversity of these interactions.⁴⁷ A relatively new area of study deals with conflict impeding climate change mitigation (conflict => climate). The two opposing causal directions could therefore form in the future a climate $\to$ conflict$\to$ climate self-reinforcing feedback cycle, in which climate change leads to conflict that in turn hampers mitigation, leading to additional climate change and conflict.³ To represent such dynamics, scenario families could vary institutional cohesion, cooperation, and fragmentation parameters.

Taken together, these three feedbacks reveal substantial limitations in how IAMs currently represent human and societal processes. At the behavioral scale, cognitive biases and bounded rationality challenge the assumption of optimizing, forward-looking agents; at the communicative scale, misinformation and declining trust disrupt the smooth policy transmission IAMs typically assume; and at the institutional scale, conflict and fragmentation undermine the global coordination embedded in most model architectures. The cumulative effect is to widen the uncertainty range of mitigation trajectories - small behavioral deviations can propagate into large macroeconomic and climatic differences. Addressing these challenges will require extending IAMs to capture feedback across these cognitive, communicative, and conflict-related dimensions, a task explored in the next section through targeted strategies for embedding behavioral dynamics into scenario design.

A research agenda

Integrating human behavior into future scenarios: A conceptual framework

Building on the above three interacting behavioral feedbacks, the development of socio-behavioral dynamics in future climate scenarios is necessary and requires systematically linking empirical findings across psychology, behavioral economics, sociology, and political science to the modeling of climate mitigation pathways. Such an integration would allow the construction of a coherent body of knowledge capable of explaining how behavioral drivers, e.g., cognitive biases, communication dynamics, and conflict processes, shape mitigation and adaptation outcomes, while also capturing feedback between climate impacts, societal responses, and policy effectiveness.

Crucially, while IAMs remain the backbone of scenario design, they are not designed to represent the full spectrum of human and organizational behavior. Their optimization structures and representative-agent assumptions make it difficult to capture evolving social processes such as trust erosion, misinformation diffusion, or collective coordination breakdown. Therefore, improving behavioral realism in climate scenarios requires connecting IAMs with complementary models, such as agent-based, system-dynamics, or multi-regional input-output (MRIO) systems that can represent the heterogeneity and feedbacks absent from standard frameworks.

We therefore propose a two-step integrative framework to translate climate-event exposure into model-ready, parameterized human responses for scenarios modeling: 1) consolidating cross-disciplinary findings on behavioral dispositions across the social sciences to anticipate responses under severe climate futures, and 2) embedding these responses within multi-model architectures, including, but not limited to, IAMs, to generate more realistic and policy-relevant projections. (green and red boxes in Figure 1).

Step i): Anticipating behavioral responses to severe climate impacts

Research in psychology, sociology, and political science has produced rich but non-integrated knowledge on how risk perception, social norms, trust, and conflict shape responses to climate stress.^5,6,7,8 These literatures offer measurable constructs, such as risk weighting, discounting, psychological distance, norm sensitivity, trust, and threshold effects that can be elicited experimentally, estimated from surveys, and inferred from digital traces. The challenge lies in translating these into model-ready functions that capture both individual cognition and system-level adaptation. Climate-related disasters play a central role in this framework as behavioral shock events that modify risk perception, social dynamics, and economic decision-making. Here we highlight three complementary methods to explore behavioral responses to such climate shocks under more severe future climate conditions (Figure 2): virtual reality (VR), AI-enhanced survey analytics, and social media data mining. This selection is not meant to be exhaustive, but we consider it a valuable starting point to provide hypothetical, experimental, and real-world behavioral data that populate behavioral response functions directly usable in Step ii.

Figure 2.

Anticipating behavioral responses to severe climate impacts

Virtual Reality for reducing cognition bias: VR offers a powerful tool for reducing psychological distance and enhancing public engagement with climate change by immersing individuals in vivid, emotionally salient simulations of future climate-related disasters.⁴⁸ Such experiences can transform abstract risks into immediate, personal realities, thereby influencing attitudes, countering misinformation, and fostering pro-environmental behavior.⁴⁹ Beyond those pro-environment engagement, VR generates rich data on behavioral and emotional management, such as responses of fear or grief,⁵⁰ that can be systematically analyzed across varying scenarios to better estimate how people might respond to different climate futures. In experimental settings, VR may provide a controlled environment for exploring short-run behavioral responses, for example, by examining how changes in perceived severity, proximity, or attribution are associated with shifts in policy support or protective intentions, conditional on moderating factors. By systematically varying scenario characteristics (e.g., disaster intensity or framing), researchers can estimate and test behavioral response functions¹¹ that could inform decision rules in agent-based or hybrid modeling architectures. At the same time, because VR simulations cannot fully reproduce the emotional intensity, material stakes, and social complexity of real-world disaster experience, such estimates should be interpreted cautiously as approximations of short-term reactions rather than definitive representations of lived behavior.

AI-enhanced survey analysis for behavioral response modeling: AI methods can leverage structured, self-reported behavioral and attitudinal data, scaling them with machine learning to detect hidden patterns and nonlinear dynamics.⁵¹ Applied to time-series data on climate-related disasters (hurricanes, wildfires, floods, and droughts), in relation to the frequency and severity of such disasters (explanatory variables), combined with survey data on socio-demographics, cognitive biases, psychological distance, miscommunication, conflict and mitigation efforts (explanatory variables), they can aid in establishing response functions for IAMs. Time-series data on climate-related disasters provide the foundation for analyzing how increasing frequency and severity of climatic events influence behaviors.^6,52,53 These datasets document historical trends in disaster occurrence, intensity, and impacts, allowing researchers to model the evolving nature of climate risks. Survey data on cognitive biases and psychological distance complement this information, and track socio-political instability linked to environmental stress. Additionally, machine-learning techniques can enhance analysis by identifying nonlinear relationships and high-dimensional interactions. Random forests and gradient boosting models can detect complex patterns in behavioral responses,⁵⁴ while neural networks can model dependencies in climate perception shifts. These models can reveal culture-conditioned response patterns⁵⁵ and provide estimates for empirical priors for behavioral parameters, such as adaptation thresholds or risk discounting that can inform scenario calibration across modeling frameworks.

Social media analysis for behavioral insights: Complementing traditional surveys that are often constrained by recall bias and low participation,⁸ social media, as one emerging approach, provides a unique, real-time large-scale data on reactions to extreme climatic events.⁴ By applying natural language processing and sentiment analysis to posts from directly affected individuals, filtered through geolocation, references to personal impact, and community-level engagement, researchers can capture behavioral shifts in risk perception, adaptive actions, policy demands, and economic decisions.⁵⁶ By examining behavioral patterns following discrete events, researchers may gain insights into how societies respond to climate-related disruptions, while recognizing that such responses are shaped by specific institutional, cultural, and socio-economic conditions. For example, coping strategies or shifts in public discourse observed after a hurricane-induced power outage in Florida cannot be assumed to arise in the same way elsewhere. However, in the absence of systematically comparable data across regions and event types, such cases provide one of the few empirical reference points available. Systematic, event-linked analysis of social media data can therefore contribute evidence for informing behavioral response functions in IAMs, albeit with uneven demographic coverage and strong context dependence.

Together, these three methods convert the behavioral feedback into model-relevant quantities: cognition informs decision bias and risk-perception parameters; communication shapes trust and misinformation dynamics; and conflict translates into cooperation, coordination, and fragmentation coefficients. The resulting behavioral evidence base provides an empirical foundation for integrating social processes into scenario frameworks, allowing climate-related shock events to be translated into behavioral response functions that can be embedded in modeling architectures, enabling IAMs and complementary models to more realistically represent human and organizational behavior under climate stress.

Step ii): Embedding behavioral dynamics into IAMs

Insights from Step i must be systematically translated into modeling architectures that can represent social heterogeneity, behavioral feedbacks, and institutional fragility. While IAMs remain the foundation of global scenario design, and are therefore chosen as the starting point here, their techno-economic orientation and representative-agent assumptions limit their ability to capture social processes such as trust erosion, misinformation, and cooperation breakdowns. One promising avenue is to embed behavioral dynamics into IAMs, and link IAMs with complementary MLMs, such as ABMs or system dynamics to represent heterogeneous households, firms, and institutions, whose choices are shaped by bounded rationality, norms, and political constraints. These micro-level responses can then be aggregated into sectoral demand, investment, and technology-adoption patterns and scaled to the macro-economy via MRIO linkages that trace sectoral interdependencies and supply-chain effects (blue box in Figure 1). In practical terms, coupling across modeling layers requires explicit translation rules between behavioral, economic, and scenario variables. Behavioral response functions derived in Step i can be implemented in MLMs as agent decision rules governing consumption, investment, mobility, or technology adoption. These decisions evolve through interactions among heterogeneous agents and through exposure to climate-related shocks such as floods, heatwaves, or wildfires. The resulting behavioral outcomes can then be aggregated into sectoral demand, production, and trade patterns using MRIO accounting structures, which translate agent-level behavior into economy-wide flows. IAMs subsequently use these aggregated quantities to simulate long-term energy demand, technology deployment, and emissions pathways. Macro-level outputs from IAMs, such as energy prices, policy constraints, or technology costs, can then feedback into agent decision environments in subsequent iterations, allowing behavioral dynamics and macroeconomic trajectories to co-evolve across modeling scales.

We stress upfront that such coupling is not presented as a solution to structural uncertainty (that is, uncertainty about how behavioral processes and feedbacks should be fundamentally represented, or which mechanisms are included or omitted), nor as a guarantee that all relevant feedbacks are captured. Rather, it reflects the practical recognition that no single modeling framework can represent behavioral heterogeneity, economic interdependence, and technological-emissions pathways simultaneously, and that an integrated architecture is therefore required to explore alternative behavioral structures.

We identify three complementary directions for advancing this agenda (Figure 3).

• (1)Enhancing IAMs’ social process representation with MLMs: New developments for MLMs are essential for capturing the diverse social processes underlying behavioral change, technology adoption, and collective action. Their high-resolution detail can allow for nuanced simulations of societal dynamics, enabling IAMs to better account for heterogeneity in decision-making. While a few studies have integrated IAMs with agent-based models,^57,58 these focus mainly on behavioral change related to energy systems (see SI 2). These studies overlook key aspects such as cognitive biases and emotional norms, how they shape social interactions such as communication and conflicts, which are essential for capturing the complexity of human behavior and social systems. Future research could focus on developing hybrid MLM-IAM frameworks where behavioral interventions can be explicitly tested under varying socio-economic and policy pathways.
• (2)Bridging IAMs’ resolution gaps with MRIO frameworks: The integration of MRIO frameworks further addresses the resolution gap between high-detail MLMs and economy-wide IAMs. Prior work integrates the MESSAGEix-Australia IAM with an IO database to enhance energy demand and policy impact analysis. By incorporating sectoral interdependencies and supply-chain details, this approach improves the assessment of indirect energy implications. It is applied to Australian decarbonization pathways and self-healing roads, demonstrating its effectiveness in evaluating socio-economic scenarios.⁵⁹ Similarly, the MESSAGEix IAM and the EXIOBASE MRIO database have been coupled for policy evaluation of emissions mitigation options.⁶⁰ MRIO frameworks provide a means to aggregate agent-level data from ABMs into regional and sectoral insights compatible with IAM structures. This ensures that localized behaviors and regional economic activities can influence and inform global climate-economic projections.
• (3)Micro-to-macro mapping: Integrating MLMs into IAMs represents a promising frontier in capturing the diversity of human behaviors and their interactions within climate and energy systems. First, it enables simultaneous high agent- and economic-sector resolution, capturing both micro-level behaviors and macro-level systemic effects. Second, this approach facilitates the creation of scenarios that align with IPCC guidelines, enhancing the capacity of IAMs to contribute to the IPCC’s scenario database and reports. Such hybridization also enables uncertainty propagation from micro-behavioral variability to macro-economic outcomes, improving scenario transparency. Specifically, ABMs offer bottom-up representation of agent decisions and interactions; MRIO frameworks translate these into sectoral and regional dynamics; and IAMs provide the global context linking policies, technology, and emissions. Embedding these models sequentially - ABM → MRIO → IAM - supports consistent multi-scale feedbacks, where individual and collective behaviors aggregate into macro-economic trajectories: agent-level decisions shape sectoral demand and technology uptake; MRIO structures translate these into regional economic activity and supply-chain effects; and IAMs simulate the resulting long-term energy, technology, and emissions pathways. In turn, IAM outputs such as prices, policy constraints, or technology deployment can be fed back into agent decision environments in subsequent iterations, allowing behavioral and macroeconomic dynamics to co-evolve across modeling scales. Advances in computational capacity and open-source model coupling now make such hybridization feasible, supporting scenario generation that aligns with IPCC standards while approximating the diversity of human behavior and organizational responses.

Figure 3.

Conceptual framework for model integration

These directions form an iterative cycle rather than a linear pipeline: empirical behavioral evidence refines model design, and model outcomes identify priority gaps for further behavioral research. Embedding this feedback loop within IAM development can bridge the long-standing divide between behavioral and systems modeling communities. Importantly, the link between Step i and Step ii thus becomes reciprocal: empirical findings refine behavioral assumptions in models, and scenario outcomes identify priority areas for new data collection.

Moving Forward: Achieving the Paris Agreement’s climate goals requires rapid global emissions reductions, yet real-world progress remains insufficient, partially due to behavioral, societal, and geopolitical barriers. IAMs, which underpin IPCC mitigation pathways, largely assume rational decision-making, seamless policy implementation, and technological adoption. However, behavioral feedbacks such as cognitive biases, social denial, misinformation, geopolitical instability, and conflict significantly affect climate action. Existing IAMs inadequately capture these dynamics, limiting their ability to produce realistic mitigation trajectories. We present here key behavioral barriers categorized into three domains: cognition, communication, and conflict. Cognitive biases shape risk perception, leading to inaction despite growing concern. Misinformation and ideological polarization undermine public and political support for mitigation. Geopolitical instabilities and conflicts disrupt cooperation, increase fossil fuel dependency, and divert resources from climate action.

A key rationale for this work is its recognition that existing research on cognition, communication, and conflict is based solely on behavioral responses observed under current conditions. However, these responses may differ significantly under severe climate impacts, fundamentally altering global mitigation efforts. To estimate behavioral responses to severe future climate impacts, we propose the use of: (1) VR, (2) social media analysis, and (3) statistical modeling. VR creates immersive experiences of climate disasters, reducing psychological distance and improving climate risk perception. Social media analysis provides real-time, large-scale insights into public reactions to extreme weather events. Statistical methods, including time-series analysis, machine learning, and causal inference, use historical climate data and behavioral surveys to infer evolving societal responses to increasing climate risks. In unison, these approaches enable the extrapolation of behavioral response functions to future climate conditions. To integrate these behavioral insights into IAMs, we suggest using ABMs and MRIO frameworks. The proposed ABM-MRIO-IAM triad bridges micro-level decision-making with macro-level climate projections, improving scenario development and enhancing IAMs’ ability to inform IPCC climate assessments.

Having said this, we make no claim that improving behavioral realism in models will necessarily generate the political impetus required for mitigation. However, even if modeling cannot directly alter policy outcomes, there remains a scientific and normative responsibility to articulate clearly the conditions under which mitigation pathways are feasible or fragile. If psychologically and socially informed scenarios indicate that mitigation is more constrained or unstable than commonly assumed, then this must be stated explicitly. This is because, ultimately, if psychologically and socially informed IAM scenarios indicated a more pessimistic future climate trajectory, decision-makers would be compelled to rethink current mitigation and adaptation strategies. Traditional policy mechanisms relying on gradual public buy-in may prove ineffective where attitude does not translate into sustained behavioral change.^61,62 Instead, governments may have to implement stronger regulatory measures, pre-emptively address misinformation, and invest in social resilience strategies to counter polarization and climate apathy.⁶³ Crucially, adaptation strategies must account not only for biophysical impacts but also for social instability and systemic disruptions, requiring enhanced crisis preparedness at both national and global levels.

Uncertainties and limitations

First, incorporating behavior into IAMs introduces uncertainties: Behavioral responses are nonlinear, context-dependent, and sensitive to social tipping points, making extrapolation from current observations inherently fragile. Translating behavioral constructs (e.g., risk weighting and norm sensitivity) into model-ready parameters requires assumptions about stability across cultures, generations, and time that may not hold. Coupling different model types (e.g., ABMs, MRIOs, and IAMs) cannot be expected to eliminate structural uncertainty, since each framework is itself subject to uncertainty about how behavioral processes and feedbacks are formally specified, and their combination will inevitably carry these structural ambiguities across scales.

Second, while present-day individual local fires, floods, or heatwaves may match the intensity of some events expected in the future, the anticipated frequency, simultaneity, and geographic spread of such events under higher warming levels are not realized in the empirical record. Consequently, any estimation of future behavioral adjustments necessarily involves extrapolation beyond the observed sample range. In addition, responses that can be observed are short-term; neither of the three approaches listed here can be used to elicit potential long-term adaptations of behavior, as long-term records either do not exist or follow-ups are impractical. These limitations are not unique to our proposed agenda but are inherent to all forward-looking climate projections, including IAM-based scenarios underpinning IPCC assessments. Estimates derived from the research agenda outlined here must therefore likewise be understood as projections that infer future dynamics from a necessarily more limited set of present-day occurrences and contexts.

Third, and more generally, the limitation of the research proposed here is both the need to extrapolate spatially (across regions) in addition to temporally (into the future). Again, this spatial limitation is pervasive, e.g., in economic (CGE) models in which behavioral parameters such as own- and cross-price elasticities are routinely transferred across regions and time periods because comprehensive, context-specific data are absent. Specifically, embedding behavioral dynamics in climate scenarios based on evidence may implicitly universalize Western cognitive patterns and political norms. Extrapolating such biases can misrepresent behavioral trajectories in the Global South or Indigenous communities, where worldviews, risk perceptions, and collective decision-making differ markedly. Behavioral feedbacks must also be assessed for their unequal power and impacts: First, people can behave differently when elevated to power; and second, misinformation, institutional fragility, and climate stress tend to disproportionately harm marginalized groups, reinforcing historical inequities. Neither of those aspects is covered in this agenda.

Behavioral integration thus introduces parametric, structural, and extrapolative uncertainties, relating respectively to response magnitudes, model representation of behavioral processes, and projection beyond observed temporal, spatial, and cultural contexts. These uncertainties can be explored through sensitivity analysis, alternative structural formulations, and robustness testing across scenario ensembles that vary climatic, institutional, and cultural conditions. Consequently, results must therefore be communicated as ranges rather than point projections.

Acknowledgments

This work was financially supported by the University of Sydney through its USYD-Fudan Ignition grant (My.L., Yt.W, Ma.L., B.C.), Horizon Fellowships (My.L.); by the Hanse-Wissenschaftskolleg in Delmenhorst, Germany through its HWK Fellowships (Ma.L., K.B.); by the Deutsche Forschungsgemeinschaft (grant no. 524984234, K.B.); by the Australian Research Council through its Discovery Project grants DP0985522, DP130101293, DP190102277, DP200103005, DP200102585, DP250100514 (Ma.L. and My.L.) and Discovery Early Career Researcher Award (DE250101250) (My.L); as well as through its Large Infrastructure, Equipment and Facilities (LIEF) grant LE160100066 (Ma L.); and through the National eResearch Collaboration Tools and Resources project (NeCTAR) through its Industrial Ecology Virtual Laboratory VR201 (Ma L. and My.L.). NeCTAR is an Australian Government project conducted as part of the Super Science initiative and financed by the Education Investment Fund.

Author contributions

My.L. and Ma.L. developed the initial idea and drafted the manuscript. All authors contributed to the conceptual development of the work, participated in discussions, and were involved in editing and revising the manuscript.

Declaration of interests

The authors declare no competing interests.