How can social–ecological system models simulate the emergence of social–ecological crises?

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2020 The Authors. People and Nature published by John Wiley & Sons Ltd on behalf of British Ecological Society 1Institute of Meteorology and Climate Research, Atmospheric Environmental Research (IMK-IFU), Department of Geo-Ecology (IFGG), Karlsruhe Institute of Technology, Garmisch-Partenkirchen, Germany 2School of Geosciences, University of Edinburgh, Edinburgh, UK


| INTRODUC TI ON
While global attention has been focused on the COVID-19 pandemic, rapid climatic change and biodiversity loss threaten further 'crises' or 'breakdowns' in which social-ecological dynamics undermine established human and natural systems (IPBES, 2019;Masson-Delmotte et al., 2018). Such events are increasingly referenced in scientific and popular media, and actions to reduce their likelihood are the subject of intense debate (Hagedorn et al., 2019). Computational models provide valuable information to these debates because they can synthesise, quantify and extrapolate from large bodies of evidence, making them core elements of science-policy interface programmes such as those of the IPCC and IPBES (Nicholson et al., 2019;Rogelj et al., 2018). Models that deal with the social-ecological interactions that can cause or potentially avert crises are especially pertinent, but most have been developed to reproduce relatively stable historical dynamics (Filatova et al., 2016;Ripple et al., 2020).
In principle, models can fulfil a wide range of roles, from developing and testing theory to analysing data and exploring system dynamics including breakdown or, conversely, recovery (Epstein, 2008). What they cannot do, in the context of complex social-ecological systems, is predict when, where and how specific future events will occur (Brown et al., 2016;de Matos Fernandes & Keijzer, 2020). This limit is significant, and essential to recognise, but within it exist a range of useful contributions that models can make. In this article, we consider whether and how models can be used to fulfil more of their realisable potential for simulating social-ecological crises or breakdowns as tools for informing societal and political responses. In doing so, we use the terms crisis and breakdown to refer to broad types of event in which socialecological dynamics have a destructive effect on (some of) the social or ecological properties of the system in question (see Box 1).
There are three main reasons to take stock of social-ecological modelling at this point in time. First, there is an ongoing increase in attention given to potential (and current) crises such as the 'climate crisis' (United Nations, 2019), 'biodiversity crisis' (Driscoll et al., 2018) and more specific 'crises' relating, for instance, to the management of water resources for ecological and societal sustainability (Srinivasan et al., 2012) or the decline of pollinating insects (Levy, 2011). There is also an increasing use of computational models to identify ways to avoid such crises (Rogelj et al., 2018). Second, despite this attention and modelling, efforts to prevent expected crises in climatic and ecological systems have been largely unsuccessful, with progress towards major international agreements such as the Paris climate targets and Sustainable Development Goals being inadequate at best (Brown, Alexander, et al., 2019;Xu et al., 2020).
Third, there has been a proliferation in the number, scope and type of available modelling tools and supporting data, suggesting that necessary new approaches may now be feasible. It is this third reason that we focus on here.
In the remainder of this article, we briefly outline aspects of current social-ecological systems modelling that make the simulation of crisis dynamics particularly challenging. We then go on to consider the nature of required improvements, before suggesting promising new approaches, resources and precedents. We finally identify potential contributions that new models can make and some main constraints that they will face.

| 'Crisis-blind' modelling
Social-ecological systems models have developed rapidly in number and scope, covering more and larger systems in increasing detail (Hamilton et al., 2015;Harrison et al., 2016). While they have numerous valid uses (including the simulation of social-ecological

BOX 1 Terminology
Social-ecological systems modelling involves concepts and terms from different scientific fields, applied to problems that are widely discussed across society as a whole.
Terminology is often unclear as a result, with many overlapping terms being defined differently by different groups (Colding & Barthel, 2019;Filatova et al., 2016;Herrero-Jáuregui et al., 2018). We do not attempt to select or develop definitions here, but provide this brief glossary to explain our own use of terms.
We use 'social-ecological' to refer to interacting social and ecological processes or systems, where those interactions determine outcomes to a substantial extent. We use 'socialecological' rather than 'socio-ecological' to formally give equal weight to both sides (Colding & Barthel, 2019).
We use crisis and breakdown as two words commonly used in popular discourse to refer to climatic, environmental and associated societal problems at large scales, with breakdown relating specifically to altered dynamics in the system in question. We also use related (but not synonymous) terms that are more widespread in the scientific literature (e.g. 'tipping points', 'thresholds' and 'regime shifts'; see e.g. Filatova et al., 2016), but as occurrences from which crises or breakdowns can emerge, or which can occur in their absence (see e.g. Figure 1). We also note that both 'crisis' and 'breakdown' imply a value judgement, and that many relevant changes in social-ecological systems can have both positive and negative impacts, depending on perspective.
While our usage of the terms reflects their broad and subjective meanings, it is also consistent with more formal (if disputed) definitions such as that of Craig (2017), in which crisis is the state in which societies' ecological 'conditions of possibility' are no longer secure. The term 'collapse', as used and reviewed by Cumming and Peterson (2017), is largely equivalent, and literature that uses converse concepts such as 'resilience' is also highly relevant (associated terminology is outlined by Egli et al., 2019). 'recovery'), their development to date does not necessarily make them suitable for simulating social-ecological breakdown, especially where those breakdowns are large-scale in nature. In fact, many of the models that operate over the large geographical extents relevant for climatic, environmental or other social-ecological crises contain basic assumptions that preclude breakdowns from emerging.
In ecology, models dealing with species, community or ecosystem dynamics across large geographical extents have often been correlational, statistical or pattern based. For instance, the great majority of studies of species extinction risks rely on Species Distribution Models (Urban, 2015(Urban, , 2019. The correlations that underpin these models are usually robust in observed conditions, but will not necessarily hold as those conditions change in the future. Indeed, rapid and substantial changes in climate and human activity are likely to fundamentally alter the basic processes behind those correlations, and are already affecting most core ecological processes in terrestrial and marine systems (Scheffers et al., 2016).
The correlational approach also makes it difficult to account for links to social system dynamics. Studies that seek to identify impacts of human land use on ecological communities (including around half of meta-analyses) do so using simple metrics such as total species richness and abundance that carry little information about ecosystem composition, function or stability (Blüthgen et al., 2016;Hekkala & Roberge, 2018). In omitting many of the characteristics that actually determine ecological responses (positive and negative) to landuse change, these studies can provide little guidance about potential future changes (Urban, 2019), and established models can become unreliable as a result (e.g. Williams et al., 2015).
Similarly, most models of large-scale human systems (e.g. economic or land-use systems) have adopted simplifying assumptions that allow general trends to be extrapolated into the future without accounting for underlying processes. Land-use models covering large areas tend to rely on assumptions that land management is optimised to meet demand for food or economic returns, and not the result of the social, cultural and behavioural processes that shape land managers' decisions (Huber et al., 2018;Rosa et al., 2014;Stehfest et al., 2019). Land uses such as forestry, which do not contribute to food production, are often left as 'residual' land covers on areas not assigned to agriculture (Brown et al., 2017;Rosa et al., 2014). As with ecology, the relevance of these equilibrium-based approaches to novel and variable future conditions is doubtful, and makes the simulation of crises particularly challenging.
Another consequence of such approaches is a disconnect between models and the stakeholders who use them. Stakeholders can find it hard to relate to models' primarily biophysical parameters, correlational structures and abstract output indicators, instead preferring models that represent recognisable processes and decision-relevant outcomes (e.g. Borsuk et al., 2001;Hunka et al., 2013;Jönsson et al., 2015;Scown et al., 2019). Quite apart from the basic need to improve model accuracy, this preference justifies improved representation of breakdown dynamics as a topic of great current interest to stakeholders and society at large (Holtz et al., 2015; Millington & Wainwright, 2017).

| New challenges: Seeing the wood for the trees
Closing the blind spots identified above is a major challenge for socialecological modelling. It requires the development of models that can simulate the emergence of unprecedented dynamics and impacts, including the feedbacks, regime shifts and thresholds involved in pushing system behaviour away from that observed in the past (Filatova et al., 2016;Synes et al., 2019;Wisz et al., 2013).
Inevitably, this forces modellers to grapple with highly complex relationships.
For example, the ongoing loss of tropical forests-a frequent subject of modelling studies-involves numerous social and ecological processes. Approximately 50% of tropical forests have already been partially or completely cleared (Asner et al., 2009;FAO, 2016) and the remainder fragmented into more than 130 million patches (Taubert et al., 2018). Forty percent of the world's population currently lives in the tropics (FAO, 2016;Harding et al., 2014), and land-use change is expected to have greater effects there than in any other biome this century (Sala, 2000).  (Schwartzman et al., 2000). Both positive and negative effects may be self-reinforcing, with negative impacts of forest clearance on local climate (e.g. rainfall and temperature) being particularly likely to undermine ecosystem dynamics and agricultural production (Lovejoy & Nobre, 2018;Oliveira et al., 2013). Interactions can therefore ripple out through space and time, producing successive waves of fragmentation, degradation and ultimate deforestation that may soon exceed a critical threshold beyond which runaway collapses in stocks of carbon and biodiversity occur as ecosystem functions break down (Laurance et al., 2011;Taubert et al., 2018).
The complexity of such systems represents a real problem for models intended to generate meaningful outcomes without themselves being excessively complex. Potential solutions might exist among the diversity of models used to simulate social-ecological dynamics across small geographical extents, but these models are usually impossible to apply over large extents because their data or resource requirements become impracticable (Elsawah et al., 2020). Where upscaling is possible, difficult choices must be made about which simplifications are achievable without undermining model utility or the coherence of the overall system representation. These choices are complicated by recent evidence, of the kind outlined above, that even very small-scale processes can have system-level impacts under some circumstances. Without novel ways of handling complexity at large scales, any insight gained through social-ecological modelling may therefore be limited.

| Ways out of 'the mess'
Simulating complex, unpredictable systems to understand and avoid damaging outcomes is not a problem faced by social-ecological modelling alone, and much can be learned from other disciplines . Lawton (1999) pithily summarised community ecology as 'a mess' when identifying a need for similarly novel approaches. Urban (2019) argued that biological modelling should follow the example of climate modelling and 'improve in accuracy by incorporating mechanistic understanding, employing multimodel ensemble approaches, coordinating efforts worldwide, and validating projections against records from a well-designed network of [observational] stations'. Achieving this vision might require considerably larger resources than are currently available to biological modelling, but certain methods may facilitate the incorporation of mechanistic understanding without incurring excessive costs or model complexity. We suggest that one approach in particular has promise in this regard: the simulation of 'high-level' social-ecological processes that play key roles in system dynamics.
Focusing on a few key fundamental processes that span scales and contexts, and omitting other less general or important dynamics, is an approach that has had significant success in community ecology. This approach emphasises understanding the nature and effects of these fundamental processes, in isolation if necessary, as sufficient for capturing the effects of numerous sub-processes without accounting for them directly (Rapacciuolo & Blois, 2019). In community ecology, Vellend (2010) argued that myriad processes belong to one of four fundamental 'high-level' processes-drift, speciation, dispersal and selection-that together produce observed patterns in the composition and diversity of species across timescales.
Identifying such processes of course requires a strong conceptual basis and evidence of the importance and generality of the processes included (Steel, 2007). Although the resultant processes are broad, this approach has been successful in prompting discussion, modelling and new results in community ecology, obviating the need for complex models and precise parameterisation to some extent. We suggest that it is possible in principle to identify a similar group of processes from literature on social and ecological systems as being key in contributing to or averting breakdown dynamics. We do not perform a systematic search for such processes here, but offer tentative suggestions based on our interpretation of earlier reviews and categorisations (Brown et al., 2017;Urban et al., 2016) and a nonsystematic review undertaken for this article (Table 1). We do so with the aim of prompting discussion, in the hope that debate, testing and refinement of these suggestions will benefit future modelling efforts.
For this purpose, we propose that high-level processes in socialecological systems could include adaptation, interaction, dispersal or movement, demographic change, and intervention by institutional and governance actors ( Table 1). Each of these processes has been identified as important in the literature (Table 1), although cases could certainly be made (and, we hope, will be) for additional or alternative processes. Notably, despite being broad in nature, these processes are currently not widely included in social-ecological systems models (e.g. Brown et al., 2017;Urban et al., 2016).
As a result, the appropriate inclusion of these or similar processes has the potential to increase the scope and utility of large-scale modelling.
These processes are analogous to classifications such as the mechanisms of collapse identified by Cumming and Peterson (2017), the 'action-situation' processes identified by Schlüter, Haider, et al. (2019) or the more empirical or theory-oriented classifications of review papers (Brown et al., 2017;Groeneveld et al., 2017;Huber et al., 2018;Meyfroidt et al., 2018;van Vliet et al., 2015). They differ in being specifically intended to contribute to modelling rather than theory, and in particular the modelling of dynamics that are outside the range observed in the recent past. They are intended to do so by focusing on general processes from which changes emerge (Figure 1), rather than providing a description of variation at a certain time, or specific effects or situations that lead to breakdowns.
Research that highlights observed similarities among social-ecological contexts such as land system archetypes (Václavík et al., 2013) or decision-making types , can also aid identification of shared major processes (Rocha et al., 2019) or drivers of change . A separate benefit of adopting a 'high-level' process approach is that the resulting models are likely to be widely understandable because the processes accord with those experienced by actors and stakeholders in any given system. This extra interpretability can be a benefit in itself, partially independent of ultimate model quality, because it allows models to enable informed dialogue by illuminating differing perspectives (Holtz et al., 2015;Millington & Wainwright, 2017;Parrott, 2017). Evidence suggests that models incorporating key, recognisable processes would be welcomed by many stakeholders who are uneasy about more 'black box' statistical models (Borsuk et al., 2001;Hunka et al., 2013). The shared understandings that can be developed in this way may additionally help to overcome social science's 'incoherency problem' by revealing-or generating-links between apparently contradictory perspectives on the basis of fundamental processes common to all (Grimm & Berger, 2016;Watts, 2017). If so, models can play a highly beneficial role in capitalising on a diversity of perspectives to generate improved understandings and responses to social-ecological problems (Page, 2014). Of course, high-level processes modelling represents just one possible approach in a field of great theoretical and practical diversity. One notable alternative could be to represent only the most fundamental of processes, and allow all others to emerge from these. For example, it has been suggested that social system dy- Richerson & Boyd, 2020). In this case, all other outcomes, including the processes we identify above (with the exception of interaction itself), can be seen as emergent phenomena. This argument is well grounded in theory and offers an elegant basis for limiting model complexity, although may not in itself ensure model versatility, for instance if only specific, limited forms of cognition and interaction are represented (Page, 2014). The high-level, action-situation or functional processes identified above, among others, may therefore remain useful in informing model design even under this more fundamental approach.

| Suitable models and data
Whatever the theoretical advantages of high-level and other forms of 'key' process modelling, their practicability cannot be taken for TA B L E 1 Potential 'high-level' processes relevant to breakdown dynamics in social-ecological systems. References are provided to give examples of the importance of these processes in social and/or ecological systems, but are not exhaustive

Institutional & governance interventions
Interventions associated with informal and formal groupings and entities. Includes institutional learning, architecture and adaptability, among others. Inherently related to social systems, even when operating on ecological systems Brooks et al., 2005;Butzer, 2012;Cote & Nightingale, 2012;Dryzek & Stevenson, 2011;Grove, 2014;Jedd & Bixler, 2015;Juhola, 2016;Preston et al., 2013;Spies et al., 2010;Young, 2010 granted. Currently, relatively few models incorporate the processes suggested above ( Correlational statistical and pattern-based models can include greater process accuracy to some extent, as demonstrated, for example, by 'joint attribute' models that represent entire ecological communities and their internal interactions (Clark et al., 2017).
System dynamics modelling can also incorporate some socialecological process accuracy (Elsawah et al., 2017), as can Earth System and marine 'whole ecosystem' models (Donges et al., 2020;Fulton et al., 2011;Pongratz et al., 2018). Network modelling has recently been used to explore the ways in which individual and social charac- F I G U R E 1 A stylised representation of social-ecological system dynamics relating 'high-level' processes to more fundamental processes, system properties and outcomes. Examples are illustrative, while the representation as a whole is presented as one possible conceptualisation that may support social-ecological modelling. The fundamental processes shown generalise the role of individual cognition and interaction in social systems described in the text. High-level processes can be seen as intermediate emergent phenomena, or as an alternative framework from which to construct models. Similarly, system properties and outcomes may emerge from either set of processes, but the highlevel processes listed may be especially pertinent to the simulation of breakdowns and crises Groeneveld et al., 2017;Lippe et al., 2019;Schulze et al., 2017) including in the contexts of breakdowns in large-scale food systems (Brown, Seo, et al., 2019) and marine fisheries management (Gao & Hailu, 2012). It has also been used to some extent to identify highlevel or cross-context processes (Parker et al., 2008) and their relative impacts on social-ecological change (Brown, Holzhauer, et al., 2018).
Such models have also been developed using functional accounts that are compatible across social and ecological sub-systems, and can similarly incorporate biologically and socially meaningful metrics re-  modelling has also been proposed for 'World-Earth' models that embed process-based social simulation within an Earth system model framework (Donges et al., 2020). Even more feasible from a technical perspective is the combination of a series of single human and natural process models to explore the dynamics that emerge from their interaction (e.g. Ullah, 2013) or, to avoid unwieldy combinations of different models, a targeted approach that focuses on specific social-ecological interactions (e.g. Sarjoughian et al., 2015).
Encouragingly, models that have adopted some of these methods to simulate fundamental social-ecological processes have repeatedly generated emergent dynamics that differ substantially from those produced by correlative or single-sector models, including dynamics that produce potential crises related to climatic, environmental and social change (Brown, Seo, et al., 2019;Bury et al., 2019;Lade et al., 2013;Synes et al., 2019;Ullah, 2013). Upscaling these approaches to make them operate over the large geographical extents relevant to major crises currently remains an acknowledged challenge, but one for which progress is being made (Elsawah et al., 2020;Robinson et al., 2018). Given that, the above precedents suggest that modelling high-level processes is both possible and profitable, with scope for rapid knowledge gains to be made.
While models may well be capable of modelling the roles of high-level processes in social-ecological crises, they will undeniably require substantial data inputs. In fact, data requirements have often proved an insurmountable barrier to process-based models of social-ecological systems even at relatively small spatial scales . Focusing on a few key, transferrable processes limits data requirements to some extent, particularly where it is used in hybrid modelling that seeks to remove unnecessary detail.
Furthermore, considerable advances have recently been made in data resolution and availability, with a number of datasets and repositories capable of supporting modelling of this kind (see e.g. Elsawah et al., 2020;Magliocca et al., 2018;Willcock et al., 2018). An illustrative selection of these datasets is provided in Table 2 While it is not yet possible to be certain where these avenues will lead, positive progress is likely for our ability to simulate and understand social-ecological crises. The correlational modelling approaches that are currently best-placed to simulate large-scale changes are notably constrained in their ability to simulate the emergence of unprecedented dynamics, and the potential for models to produce novel, unexpected results is the clearest advance that new forms of modelling can make.
Many of the major challenges that human societies now face are a product of local processes operating in a global context. Thanks to increasing computational and data resources, modelling these processes and context is now technically feasible. The exploration of social-ecological process interactions across scales that this permits is another clear and achievable objective for new models (Elsawah et al., 2020;Lippe et al., 2019;Robinson et al., 2018).
Nevertheless, the construction of more detailed process-based models is of limited utility in itself. Such models can quickly become excessively complex, and are in any case unlikely to have greater predictive accuracy than far simpler models (Grimm & Berger, 2016;Salganik et al., 2020). To maximise their contribution, models must be designed to focus in on the processes that are most relevant to the issues being studied, and most representative of social-ecological systems at large scales. The identification of such processes is a substantial challenge that requires engagement from a range of perspectives. Our suggested 'high-level' processes (Table 1) are examples only, inspired by similar and successful categories used in community ecology (Vellend, 2010), and presented here with the sole aim of prompting discussion, testing and refinement. To the extent that modelling can contribute to these aims, it may meaningfully contribute to socialecological research as a whole simply by supporting conceptual discussions. In linking research models to the concepts through which people experience social-ecological change, high-level processes may also be able to support greater social engagement and knowledge exchange (Holtz et al., 2015;Millington & Wainwright, 2017).
While developments in other disciplines suggest that highlevel and analogous approaches are useful, recently developed TA B L E 2 Examples of models and data that cover key social-ecological processes. Single examples are chosen to represent feasibility of incorporating processes in models, and do not provide a summary of existing approaches (relevant reviews include Brown et al., 2017;Egli et al., 2019;Fisher et al., 2018;Groeneveld et al., 2017;Huber et al., 2018;Leroux et al., 2013;Maguire et al., 2015;Robinson et al., 2018;Urban, 2019) Processes

Models Data
Adaptation Social: Agent-based modelling of water usage decisions under environmental variability (Arnold et al., 2015) Social: Mobile phone usage and social media data for assessing vulnerability and adaptation (Ford et al., 2016) Ecological: Loci-based modelling of genomic adaptation of poplar tree species to environmental gradient at community-level (Fitzpatrick & Keller, 2015) Ecological: Long-term tropical forest censuses from permanent plots around the world, incorporating environmental change and human disturbance Ecological: Stochastic population modelling of emperor penguin responses to climate change (Jenouvrier et al., 2009) Ecological: Bayesian modelling to extend species demography data coverage to under-studied species (Kindsvater et al., 2018) Social-ecological: Agent-based modelling of demographic change in indigenous hunting communities and their prey species (Iwamura et al., 2014) Social-ecological: Long-term data records covering changes in social and ecological communities as, e.g., road network develops in Amazon (Klarenberg et al., 2019) Institutional & governance interventions Social: Agent-based modelling of individual and institutional activities in land system (Holzhauer et al., 2019) Global/regional databases of policies and impacts relating to, for example, environment or climate (New Climate Institute, 2019; OECD, 2019) Ecological: Multi-model framework to identify pathways and policies to reverse biodiversity loss trends (Leclère et al., 2020) Social-ecological: Economic-environmental modelling to explore effects of different policies on land use and biodiversity (Bryan et al., 2016), and network modelling of the effects of social institutions on ecological conditions, for example, of coral reefs (Barnes et al., 2019) social-ecological models and datasets suggest that they are feasible (Table 2). If this approach is successfully developed and applied, it could have a number of other benefits. Most obviously, it could extend the ability of process-based models to simulate the emergence of breakdowns and crises (e.g. Brown, Seo, et al., 2019;Ullah, 2013) to other contexts and scales. Such simulation is strictly distinct from prediction (see below), but it can reveal situations and dynamics from which crises can emerge that might otherwise not be recognised, opening up potential opportunities for designing new interventions. Less tangible, but possibly more fundamental, is the scope for such models to allow exploration of the ways in which social-ecological systems diverge from anticipated behaviour; the nonlinearities, thresholds and regime shifts that characterise complex system dynamics (Filatova et al., 2016;Synes et al., 2019). In particular, the roles of social processes in prompting such events remain poorly understood, but amenable to modelling of this kind (Barceló & Del Castillo, 2016;Lade et al., 2013;Ullah, 2013). Where these social processes go beyond historical precedents or available data, process-based modelling has the important final advantage of allowing fuller exploration of uncertainties (Gostoli & Silverman, 2020;Salganik et al., 2020).

| Impassable constraints
Whatever the reach of new forms of data and analysis, socialecological models cannot be parameterised to exactly represent reality. They can only ever be an approximate guide to system dynamics, and cannot be used to predict how systems will develop in the future because the systems in question-and especially crises in those systems-are inherently unpredictable (de Matos Fernandes & Keijzer, 2020;Oreskes et al., 1994). This limit not only highlights the importance of rigorous and transparent uncertainty analysis as a way of exploring the scope for unexpected developments (Gregr & Chan, 2015) but also highlights the need for a range of models and modelling approaches to be developed.
In fact, many theoretical and computational approaches may be equally valid or useful for simulating social-ecological crises.
Ecological and, especially, social theories remain diverse, difficult to precisely encode algorithmically, and legitimately open to differing representations (Watts, 2017). Making assumptions explicit and investigating associated uncertainties is essential in this context (Gregr & Chan, 2015). Modelling nevertheless remains, to a small but crucial extent, an imaginative, interpretative exercise that is hindered by methodological convergence (Feyerabend, 1993;Yusoff & Gabrys, 2011 Gregr & Chan, 2015; Oreskes et al., 1994), but perhaps their greatest value is to underscore the difficult questions that models raise and undermine the easy answers they sometimes appear to provide.

| CON CLUS IONS
The need for models that can simulate the emergence of crises related to climate and environmental change, biodiversity loss and associated human processes is growing. This need coincides with rapid advances in computational and data resources that allow for new forms of modelling to be developed. We argue that the simulation of social-ecological dynamics as emergent from 'high-level' processes or similar conceptual frameworks has particular promise. These processes should have broad thematic and geographical relevance, and the potential to be simulated in existing process-based models using efficient functional descriptions of social and ecological systems.

ACK N OWLED G EM ENTS
This research was supported by the Recruiting Initiative of the Helmholtz Association.

CO N FLI C T O F I NTE R E S T
The authors declare no conflicts of interest.

AUTH O R S ' CO NTR I B UTI O N S
C.B. conceptualisation, investigation, writing-original draft; M.R.
conceptualisation, writing-review and editing.