To what extent is climate change adaptation a novel challenge for agricultural modellers?

Modelling is key to adapting agriculture to climate change (CC), facilitating evaluation of the impacts and efficacy of adaptation measures, and the design of optimal strategies. Although there are many challenges to modelling agricultural CC adaptation, it is unclear whether these are novel or, whether adaptation merely adds new motivations to old challenges. Here, qualitative analysis of modellers' views revealed three categories of challenge: Content, Use, and Capacity. Triangulation of findings with reviews of agricultural modelling and Climate Change Risk Assessment was then used to highlight challenges specific to modelling adaptation. These were refined through literature review, focussing attention on how the progressive nature of CC affects the role and impact of modelling. Specific challenges identified were: Scope of adaptations modelled, Information on future adaptation, Collaboration to tackle novel challenges, Optimisation under progressive change with thresholds, and Responsibility given the sensitivity of future outcomes to initial choices under progressive change.

Introduction

Agriculture must feed a growing world population and deliver essential ecosystem services, while providing economic, social, and cultural value (Chaudhary et al., 2018; Howden et al., 2007; Thornton, 2010). Ensuring that the sector adapts effectively to the multi-faceted impacts of climate change (CC) (Iglesias and Garrote, 2015; Olesen, 2017) is therefore vital. Proactive adaptation undertaken today is likely to be less costly and more effective in reducing the societal impacts of CC than delayed or reactive responses (Stern, 2007). However, there is uncertainty around essential knowledge regarding CC impacts at local level (Diogo et al., 2017) and the effectiveness of adaptation strategies under different future scenarios (Mandryk et al., 2017; Schaap et al., 2013). Such strategies interact with a range of wider societal concerns, including the need to achieve sustainable development goals (Chaudhary et al., 2018), mitigate greenhouse gas (GHG) emissions (Del Prado et al., 2013), safeguard ecosystem services (Balbi et al., 2015; Hamidov et al., 2018), ensure food security (Godfray et al., 2010) and avoid damaging land use change (Foley et al., 2005).

Modelling is a key tool for characterising the likely environmental, economic, and social impact of CC on agricultural systems but, to reflect reality, models must incorporate adaptive responses to these impacts (Reidsma et al., 2010; Reilly and Schimmelpfennig, 2000). Models need to incorporate adaptation to test the effectiveness of adaptive responses and reveal synergies and trade-offs between adaptation to CC and other objectives (Del Prado et al., 2013; Kipling et al., 2016a; Lobell et al., 2008). In relation to any specific modelled system, CC adaptations can be autonomous (responses occurring without external intervention) or, non-autonomous (planned actions taken pre-emptively or due to experience of CC impacts) (FAO, 2007; Reilly and Schimmelpfennig, 2000). For example, in a regional scale model, autonomous adaptation might include predicted responses of farmers to environmental change (such as altering sowing dates) while a policy decision to fund irrigation systems might be a non-autonomous adaptation investigated by altering model inputs. In addition, modelling strategies investigated as potential CC adaptations might include responses to non-climatic systemic pressures with adaptive or maladaptive consequences (Grüneis et al., 2016; Mitter et al., 2018). In the context of this study, CC adaptation is defined as including: non-autonomous adaptations, any strategy explored by modellers as a potential CC adaptation, and autonomous human adaptations. Autonomous biophysical responses of the system, and actions not recognised as CC adaptations within a specific modelling exercise, are considered to be part of the context within which CC adaptation occurs.

The literature on agricultural modelling of CC impacts and adaptive responses is vast and growing (Challinor et al., 2014; Özkan et al., 2016; Rötter et al., 2018; Ruiz-Ramos et al., 2018; Wheeler and Reynolds, 2013; Zhang et al., 2017), with diversity in scope and focus. This complexity makes it hard to unpick the nature of the modelling challenges. The question arises as to whether efforts to model CC impacts and to improve agricultural modelling in general, are sufficient to support adaptive actions by stakeholders and policymakers or, whether there are agricultural modelling challenges specific to CC adaptation and thus requiring focussed attention from researchers and modellers. The aim of the current study was to search for and (if found) define challenges specific to CC adaptation modelling in agriculture. Research was based on the gathering and analysis of agricultural modellers’ views of challenges to modelling agricultural CC adaptation.

Materials and methods

The study proceeded in three stages: i) modelling challenges were identified by modellers within workshops and analysed to identify challenge themes and categories, ii) findings were triangulated by comparing the identified themes with modelling challenges described in existing reviews. This process was used to validate the workshop data and to identify themes likely to include elements specific to modelling CC adaptation, iii) the subset of challenges considered to have CC adaptation specific aspects was considered in the light of a review of the wider literature on CC adaptation, to highlight those novel elements.

Identifying challenge themes and categories

Two workshops were held to understand and explore modellers' views on the challenges to modelling CC adaptation, bringing together researchers from across the Modelling European Agriculture with Climate Change for Food Security (MACSUR) knowledge hub (http://macsur.eu). The workshops engaged 22 modellers from 21 institutes across 11 European countries, with participants representing a purposive sample of agricultural modellers with a specific interest in modelling CC adaptation (Yin, 1989). Within this sample, 16 agricultural modelling groups were represented (Appendix A) from across crop, grassland, livestock farm-scale and economic modelling disciplines. Workshops gathered participants’ views through two structured discussions in which attendees were asked to map adaptation strategies for agriculture and related modelling challenges. Participants were asked what the challenges to modelling climate change adaptation were. They recorded their ideas on sticky notes (one challenge per note) to reduce bias towards the views of vocal participants which can arise in group discussions (Kitzinger, 1995). Notes were collected and reviewed by the group to remove duplicates, clarify unclear contributions, and give participants a chance to add further ideas after considering the question during the session and in the light of other responses.

Data (responses recorded by participants on sticky notes) were analysed following a grounded-theory approach using thematic coding (Ritchie et al., 2014). Grounded theory seeks to draw information from data, rather than fitting them to a pre-conceived categorisation. Themes in the data are identified by thematic coding, for example, identifying that several contributions relate to data quality. Themes are then compared and contrasted to identify underlying characteristics linking them into broader categories relevant to the research question. In this way, categories are grounded in (derived from) the original data, ensuring relevance and openness to emerging issues (Charmaz, 2014). Qualitative approaches have been widely used to investigate the views, perspectives and characteristics of agricultural stakeholders (Mitter et al., 2018; Morris et al., 2017) but, to a lesser extent to explore agricultural research processes themselves. Exceptions include Reed et al. (2014) who used a grounded-theory approach to identify key principles of knowledge exchange in environmental management, and Kipling and Özkan et al. (2016) who analysed questionnaire data to reveal discourses underlying the perspectives of agricultural modellers on the challenges to communication with stakeholders. These examples demonstrate the practical value of grounded theory in revealing underlying patterns in complex topics.

After the identification of themes through coding of the workshop data, these themes were compared and contrasted to reveal underlying categories with relevance to the research topic (Ritchie et al., 2014). To ensure that the identified categories were robust and properly grounded in the data, results were checked by co-authors not involved in the analysis, following Bitsch (2005). In addition, intermediate findings were presented and discussed at an internal MACSUR project meeting.

An important aspect of grounded theory methodology is to ensure data saturation (Morris et al., 2017) where no new themes or issues arise from the data. To check this, specific modelling challenges were identified in the text of a global review focussed on crop modelling of the impacts of and adaptation to CC (Rötter et al., 2018). These challenges were coded to ascertain whether any new themes were present, or whether workshop themes were sufficient to accommodate the challenges described (indicating saturation). The article also defined challenges specific to modelling tropical plant production systems, providing a test of whether themes arising from the contributions of European modellers involved in the present study have relevance beyond the region.

Triangulation with previous reviews and the identification of CC adaptation specific challenges

A recent review of challenges for Climate Change Risk Assessment (CCRA) for adaptation policy (Adger et al., 2018) offered a comparison between the data from the current study, and challenges identified within a discipline focussed specifically on CC adaptation, but which encompasses change in all sectors (not only agriculture) and which may, but does not necessarily, draw on modelling. This comparison could therefore, reveal or expand on challenges related to CC adaptation itself that modellers may not have considered. Themes were identified in the review using thematic coding, following the method applied to the data from the workshops (2.1). The themes defined in this process were then compared with the themes derived from the current study to identify similarities and differences. A second set of comparisons were made between workshop data and two recent reviews of modelling challenges in the context of CC; for grassland modelling, and for animal health and disease modelling (Kipling et al., 2016b; Özkan et al., 2016). These two articles were chosen as they applied a similar approach to that used here in order to derive the challenges they presented, allowing a straight comparison with the themes identified in the current study. The disciplines of grassland and health and disease modelling lie within the broader agricultural modelling community focussed on in this study, but the reviews reflected on CC in general, only briefly treating CC adaptation. They could therefore be used to reveal which of the challenges from the current study were also wider challenges for modellers, and therefore not specific to CC adaptation modelling. In the context of a wider review of literature on CC adaptation, triangulation of these different comparisons was used to draw out specific CC adaptation challenges for agricultural modelling.

Results and discussion

Challenges to modelling adaptation

Grounded theory analysis of challenges to modelling adaptation expressed by modellers, identified 18 themes (see Appendix A for full description of each), and three underlying categories: Content, Use, and Capacity (Fig. 1).

Fig. 1

Themes and underlying categories derived from workshop data. White boxes = themes; grey boxes = underlying categories.

Content of models

Many comments made by participants related to the fundamental question of how (and how well) models characterise systems. For some, the effects of external processes on systems was important “Not possible to model landscape adaptation strategies such as creating synergies between districts for producing feeds where it is more feasible: How to assess the impacts at farm-scale?” including top-down political influences “Changes in policies can make previous changes in farm strategic planning (investments) useless”. Other comments highlighted challenges of modelling different types of change over time “Solutions can be applied in many different ways (gradually, in one step, in a series of steps – ‘timeframe of choices’) so that the dynamism of adaptation represents another level of complexity” and in particular, sudden change in biophysical systems “The length and severity of extreme events may limit available management choices, and this is hard to model (e.g. a model may usually apply irrigation in a drought, but previous droughts, or a long drought may mean that irrigation water is not available)” and changes in the adaptive choices available “Disruptive technology: One of the areas where the struggle is predicting the arrival of disruptive technology partly because it is behavioural”. Issues were also raised relating to when choices are made “When it becomes preferable to change the system (production) rather than to adapt”. Underlying these challenges, were those relating to modelling interactions within systems in general, e.g. “The application of fertilisers and its effects is highly complex, for example interactions in the soil and in relation to climate change”, with adaptation adding further complexity “Incorporate adaptation strategies adequately into models in a way that allows you to study feedbacks and side effects without prescribing too many of them as inputs, and that reflects the technical characteristics of the measure”.

Participants reflected on unevenness in the coverage of different systems by modelling, which may be limited in regions currently facing the most negative CC impacts “Difficulties in modelling Mediterranean grassland systems dominated by annual self-reseeding species: the majority of models were developed for temperate grasslands” or in relation to previously marginal production approaches that might be important adaptation strategies “Nitrogen cycle: To create a zero-sum long term N balance - what happens in a semi-arid soil? What happens under agro forestry? What happens in ley plus arable?”. The category is bound together by a sense from the data of a research community that is itself being asked to adapt. This mirrors the progressive nature of CC, how this is changing modelling priorities, and how it disrupts a research community of previously discrete, specialised modelling groups, forcing them to broaden their focus and the application of their models (see also Section 3.1.3).

Use of models

Many participants raised issues that related to how modelling could and should be interpreted and applied. Some considered the need to highlight different outcomes “Adaptation to protect ecosystem services – social context – motive of farmers (values, policy context, market failure, etc.). Demonstration of importance of these services” while others viewed modelling as part of a wider process “Demonstrate use of modelling in participatory projects” with the capacity to alter the focus of stakeholders and also of research “Modelling imagined situations to produce simulations can draw attention to a problem and stimulate the data production required to improve such estimates”. While these comments view the role of models as stimulating understanding and interest, others focussed on how to fit findings to users' needs “Policy makers are asking ‘How do we do X?’ while scientists are answering ‘What happens if?’ questions – this can create communication problems” and “Understanding of the requirements of key players (policy, farmers)”. Some specific interests believed to be important for stakeholders were highlighted, along with the challenge of tailoring outputs to specific conditions “Cooling, ventilation: Adaptation designs are very farm-specific, e.g. requirement for a very detailed design and approach”. Other comments considered how stakeholder engagement and model relevance were related “Actors (e.g. farmers) have to be involved in the research pathway from the beginning in order to co-design research questions and co-develop win-win adaptation strategies“. A final element was the challenge of communicating findings, which may be related to the complexity of the results “Distinguishing between descriptive forecasting and prescriptive (normative) information and results” or the skills of modellers “Talking is important - modellers can put too little effort into communication skills”. Some participants suggested ways to overcome communication barriers when sharing results, e.g. “Incorporation into media products like animated films“. The Use of models category therefore has both pragmatic (What do stakeholders want? How to communicate?), and normative (What should be modelled and explored?) elements.

Capacity of models

In contrast to comments about what models characterise and how (Content), a distinct set of contributions were related to information and support for modelling. Many participants highlighted historical and resource-related reasons for model limitations; “Some model limitations come from the development of models over time. E.g., [MODEL NAME] was developed when it was only technically possible to send management information to the biophysical model in (what now seems) a limited way. Management experts then moved on to other projects, and [MODEL NAME] became more biophysical”. Model evolution was seen to create problems in the capacity of successive generations of researchers to use models effectively “Most models contain vast amounts of implicit knowledge […] Continuity of human capital is too short - this rapidly degrades the future utility of models despite huge latent potential”. Some participants referred to how model capacity can be shaped by the interests of funders, which may not align with priorities identified by researchers “With disease, endemic diseases are more important than incursions, but less attractive to funders (e.g. liver fluke)”. Collaboration was seen as a way to tackle issues of capacity by drawing on wider expertise: “Linking groups 'interdisciplinarily' to ensure models are fit for purpose for the end user”

A distinct capacity-related element in responses referred to the data on which models rely. A particular focus was issues relating to data on the impacts of future CC conditions on modelled systems “Lack of data for forage crops response to fertilizer under varied extreme event conditions”, future climatic conditions “Focussed climate scenarios needed (e.g. northern Europe is likely to face wetter conditions and heat stress is not an issue)” and the likely adaptive behaviour of stakeholders “Data on risk perception of farmers: are they likely to use the strategy, why? Past experiences?”. There was awareness that data issues often related to a lack of interaction between researchers in different regions and disciplines “There are examples of systems in extreme climates: We in north western Europe have little sense about them or data that may exist on them” and to variation in available data quality “Heat stress modelling work requires wider data availability to capture differences in impacts between regions (EU database). Some variation between countries can reflect differences in data quality and availability, rather than real differences in conditions”. The need for data about projected futures was raised, and particularly limitations in approaches to gaining such information “Subjective expert knowledge on 'probability' of events and shocks”. Comments on data sometimes focussed on the need for better data sharing systems “Inventory of modelling and experimental work to allow better access to available information” and barriers to this “Often one of the limits is parties holding onto data and models to protect their turf and/or obtain cash and rights”. Again, the underlying thread in this category was how tackling CC adaptation created a need to overcome the constraints of fragmented research structures.

Evaluation of analysis

Comparing the challenges identified here with those defined for crop modelling by Rötter et al. (2018) (Appendix C) no new challenge themes were discovered, indicating data saturation in relation to the themes and categories derived from our workshop data. Challenges from Rötter et al. (2018) aligned with a subset of six of the 18 themes identified from workshops. The article also discussed the use of model ensembles to tackle issues related to uncertainty, with this topic treated as an aspect of progress rather than a future challenge. If included as a challenge to modelling, this topic would be accommodated within the ‘uncertainty modelling’ theme identified in the workshop (Appendix C). As the article included specific challenges for modelling tropical plant production, the fact that no new themes were revealed also provides an indication that the themes presented here are also relevant to adaptation modelling in non-European contexts, although specific challenges within themes are likely to vary. Further investigations of challenges to modelling other non-European farming systems would be important to confirm this wider applicability of the categorisation presented.

Triangulation with previous reviews

Consolidating challenges to modelling CC adaptation into three categories (model Content, Use and Capacity) provided a useful conceptual overview. However, many issues raised related to broader modelling challenges. In particular, modelling CC impacts on agriculture is a complex challenge in itself that has been recently reviewed by a number of authors (Kipling et al., 2016a, 2016b; Özkan et al., 2016; Rötter et al., 2018). Comparisons with previous work therefore sought to further elaborate and differentiate CC adaptation-specific issues from wider modelling challenges.

CCRA review comparison

Comparison of the challenge themes derived from workshop data with the CCRA review (Adger et al., 2018) identified workshop challenges also recognised as issues for the adaptation (but not agriculture) specific field of CCRA for adaptation policy. The aim was to highlight challenges with potentially adaptation-specific aspects for further consideration (for full details of themes from the CCRA review and of the comparison, see Appendix D). In relation to the category of Capacity in the current study, several themes were found in both modelling and CCRA studies, specifically: Collaboration, Data availability, Data quality, Novel scenarios and Uncertainty (Fig. 2). Challenges identified by modellers relating to resource availability (Resources for modelling), and Data accessibility (i.e. due to communication and ownership of data rather than due to whether they exist) were not raised in relation to CCRA for adaptation policy. This difference may reflect the very specific data requirements of models, and the fact that agricultural modelling must come together across specific disciplines to incorporate CC adaptation, while CCRA is already a united community explicitly focused on this aim and working in direct support of policy. In general, these two challenge themes are clearly not a specific issue for modelling CC adaptation in agriculture, but broader challenges to model development and application.

Fig. 2

Comparison of challenges identified by participants with themes drawn from Adger et al. (2018) and challenges from Kipling et al. (2016b) and Ozkan et al. (2016). Within the three categories of Use, Content and Capacity defined from the analysis of data, white and grey boxes indicate challenges identified by participants in the current study: i) listed as wider challenges for modelling in reviews (white); ii) in which some element was considered unique to CC adaptation in reviews (light grey); and iii) considered specific to adaptation in reviews (dark grey). Black boxes = challenges only raised in CCRA review. Asterisks = challenges identified by participants, and also in CCRA review. Dashed ovals delineate groupings of challenges contributing to one of the specific CC adaptation modelling challenges depicted in Fig. 3 (denoted by letters A-E).

Fig. 3

Challenges for modelling CC impacts and management change and how they interact in specific challenges for modelling CC adaptation. Letters A-E reference the groupings in Fig. 2. White arrows indicate the key interaction between required model capacity and content and their use.

In relation to the category of model Content, the CCRA review highlighted challenges relating to the Interdependence of systems, and how adaptations and their impacts cascade outwards. This theme overlaps with the modelling challenges of Interactions, Scale Interactions, External limitations modelling and Dynamic change modelling. In relation to the latter, the specific issue of accounting for Time lags in adaptation was highlighted. There is no CCRA theme that relates to Discrete events, except for a reference within the Collaboration theme to the benefits of linking to disaster risk management researchers. In relation to Management modelling, the CCRA review focused on the specific issue of Cognitive bias, and how it affects decision-making.

The biggest differences between the current study and the CCRA review were found in the category of Use of models. Participants in the current study expressed awareness of practical challenges relating to how to improve model relevance (User focus), the Engagement approaches that modellers need to use, the Role of modelling (when and how to engage) and the challenge of effective Communication with stakeholders. However, they did not consider how such issues might interact with differing stakeholder limitations and agendas – which was highlighted in the CCRA review within several different themes (Fig. 2, A – black boxes). These differences in relation to Use, may reflect the different characteristics of modelling versus risk assessment: In a CCRA, data form the core content, with the scope of the assessment and understanding of interactions (biophysical, political and economic, across scales and sectors etc.) explicit in the subsequent interpretation of those data (an issue relating to the capacity to carry out this interpretation). In contrast, in modelling, data are required to develop and use the model (relating to the category ‘Capacity’) while scope and understanding of how systems work form the Content of the model. As a result, model outputs may be shared and used by decision makers without these underlying issues (contained within the model) being considered. Interpretation of results becomes a much more contested space within CCRA, with uncertainty and limitations of knowledge interacting with the sometimes conflicting subjective agendas of stakeholders (Adger et al., 2018). Comparison of the CCRA review and the present study therefore highlights the importance for agricultural CC adaptation modelling of taking better account of ethical issues relating to the presentation of findings, what they include and exclude, and with whom they are shared – i.e., issues relating to the diverse motives, perspectives and values of different societal groups.

Comparison with modelling reviews

Comparing challenges to grassland modelling and animal health and disease modelling under CC with the workshop data (Appendix D) revealed two themes highlighted only in relation to CC adaptation (not as wider challenges to modelling). The first was the development of Novel scenarios of future adaption (Fig. 2). Novel scenarios relate to the challenges of model Scope, with the difference being between elements of future change provided as model inputs (scenarios) and elements that are endogenous to models (scope). The second theme only arising in relation to CC adaptation was Management modelling. However, it is apparent that incorporating all relevant aspects of decision making, including Cognitive bias, into models is a general challenge for modellers seeking to represent any form of decision making. Studies of approaches to incorporating management into agricultural models (Moore et al., 2014; Robert et al., 2016) suggest that technical solutions exist in modelling to characterise adaptation, including constraints and changes that take place over time but, that actually representing proactive management is difficult. The fact that Management modelling was not framed as a general challenge in the reviews used in the comparison exercise, may relate to their focus on biophysical challenges, reflecting the implicit nature of decision-making assumptions in biophysical modelling (i.e. as rational responses to biophysical cues with perfect knowledge). This view is reinforced by the consideration in both reviews of Interactions between management and biophysical and economic systems, suggesting a greater focus on biophysical impacts and triggers of choices, than on characterisation of the decision-making process itself. Again, although Interactions modelling was considered as both a general and an adaptation-specific challenge in the reviews, improving how the Interactions of biophysical, economic and management systems are characterised is a general modelling challenge, with only the Scope of the systems modelled increasing to facilitate the modelling of novel adaptations. Data accessibility was another theme identified as a challenge for both general and CC adaptation modelling in the grassland and livestock health and disease reviews (the need to collate available data on future adaptations). However, overcoming issues of Data accessibility (ownership, sharing) is clearly a general challenge for modellers, and no specific elements of it were detailed in the reviews, or the literature.

Taking account of the discussions above, ten of the original challenges were either, not mentioned in the modelling reviews, only mentioned as general challenges to modelling under CC, or were determined to be general after further consideration (Fig. 2, white boxes). A CC adaptation specific element was suggested for three of these (Fig. 2, A – white boxes) as a result of their relation to the CCRA challenges associated with subjective aspects of dealing with stakeholders. Seven more of the original challenges were given both adaptation-specific and wider relevance (Fig. 2, light grey boxes). Determining the precise nature of the adaptation-specific elements within these challenges, required further consideration in the context of understanding from the wider literature, which is the focus of the following section.

Identifying challenges specific to modelling adaptation

The challenges identified in the previous section as having CC adaptation-specific elements, as well as associated themes from the CCRA review (Fig. 2., boxes within dashed ovals) are explored below in the light of key characteristics of CC adaptation, in order to focus on the underlying specific issues they present. Climate change differs from most other issues in that it overlays pre-existing socio-economic (Iglesias and Garrote, 2015) and environmental challenges, and represents a progressive and sustained change over time. As CC affects the biophysical systems on which we rely in multiple ways, it produces cascades of interacting impacts and feedbacks within and between sectors, making studies of CC issues particularly complex (Terzi et al., 2019). So, while other types of change affecting farming may also be progressive (e.g. increasing demand for meat and dairy products, advances in technology) CC is unique as a sustained, progressive change in the biophysical systems that farmers rely on, rather than just in the socio-economic context in which farming takes place. Path dependency in relation to processes of economic and political change over time, including in agricultural systems (Kay, 2003) (see Martin and Sunley (2006) for a critical review) means that our iterative responses to progressive CC may lead us down particular pathways, each with different implications for different societal groups, regions and biophysical systems. For example, investment to install and improve irrigation systems may make increasing crop water supply more cost effective for a farmer than changing towards more water efficient systems as CC advances, with implications for other water users and the environment. In Sardinia, Dono et al. (2016) found that intensive dairy farming reliant on irrigation systems is likely to be less vulnerable to CC than traditional, low input sheep production reliant on natural water supplies. Therefore, pathways of adaptive response to progressive CC need to be explored in order to facilitate informed and reflective decision making that take such issues into account. In this light, the Scope of models to explore the future consequences of CC adaptation strategies is revealed as a CC adaptation-specific element of the workshop theme of Scope (Fig. 3, B).

The issue of path dependency is also relevant to the ‘Use of models’ challenge category. In the literature on CC adaptation, even the need for intervention to ensure agricultural adaptation to CC is contested, with some suggesting that market forces will automatically adjust systems to change, while others argue that progressive CC will require well-planned responses beyond the autonomous, incremental change already undertaken by agricultural stakeholders (Anwar et al., 2013; Reilly and Schimmelpfennig, 2000). Relying on autonomous responses or intervening to completely manage CC adaptation, are two extremes in a continuum of approaches. Which adaptive pathway (different types of planned response or, reliance on autonomous change) appears most favourable depends on chosen system boundaries (e.g. biophysical processes, economic processes, social processes) and the nature of CC change (Reilly and Schimmelpfennig, 2000) but, also on desired outcomes and on whose desires are considered. Although profit or production maximising objectives may be assumed in ‘hard systems’ (van Paassen et al., 2007) research approaches, this assumption has been described as representing an ‘implicit sociology’ (Jansen, 2009) of unexplored motives and opinions. If particular motives and objectives for change have already been assumed in a model, this represents a move towards more instrumental engagement with stakeholders (to improve research outcomes or increase the implementation of recommendations) and away from normative engagement (involvement of stakeholders and incorporation of their views and needs as a right) (Reed et al., 2009). Using Freeman's (1984) classification of affected and affecting stakeholders, this focus shifts attention from those who may be affected by change, towards those that can affect change. In this context, and given that the quantification of information (e.g., in models) is understood to fundamentally alter how things are perceived and valued (Espeland and Stevens, 2008) it is important that the aims modellers focus on, what models include, who they are for, and how they are communicated, are critically reflected on by modellers in general. Within the current study, the more normative aspects of the ‘Use of models’ challenge category reflected awareness among modellers of the potential for models to affect the direction of choices (including adaptive responses) and of how, in some cases, modellers are facing the challenge of assuming new roles, e.g. recognising a “Paradigm shift in the research praxis: from observer to co-researchers/knowledge brokers”. Much previous work considers these issues, with recent reviews focussing on best practice in stakeholder involvement, model development, use, and evaluation (Fulton et al., 2015; Hamilton et al., 2019; Jakeman et al., 2006; Voinov et al., 2016) including the development of specific engagement processes drawing on understanding of ‘soft systems’ approaches (Martin, 2015). However, with pathway dependency in the context of progressive CC, the potential impacts of model findings beyond the implementation of a given modelled choice, add an extra dimension to issues of model use. This additional element can be seen as a CC adaptation-specific challenge to model use.

As discussed above, CC adaptation modellers (including biophysical modellers as well as bio-economic modellers) need to consider how social conflicts, power relations and sectoral interests may influence their work and its use (Lang et al., 2012; Newell and Taylor, 2018; Reed et al., 2009) in the context of progressive CC and escalating adaptive responses. Such considerations enable modellers to recognise the implications of their focus (on which stakeholders, which objectives, which adaptations and which impacts) and to identify ways to ensure that the wider context of non-modelled strategies and impacts is conveyed to stakeholders. This may be carried out by the modellers themselves where they have the required expertise and sufficient resources but, may also be achieved through Collaboration with social scientists, to try to avoid unintended consequences arising from the use of model outputs, and to achieve best practice (Fig. 3, E). Taken together, these considerations represent the specific CC adaptation aspects of the challenges of Communication, Engagement approaches, Role of modelling and User focus grouped as ‘Responsibilities of modelling under progressive (climate) change’ in Fig. 3, A).

Related to the progressive nature of CC, and adaptive responses to it, a second key characteristic of CC adaptation was revealed explicitly in workshop data. Modellers expressed the need to better understand and incorporate likely stakeholder choices under progressive CC in which their expectations and experiences of CC evolve over time, distinct from likely responses to other types of change (such as one-off shocks or opportunities to increase efficiency). One participant, for example, highlighted the importance of understanding “Reasons or other triggers for farmer decisions on the number of cattle they have and the type of grassland management they apply, and the point when they begin to care about climate change and take action”. Addressing this issue, which contributes to the adaptation-specific elements ‘Optimisation’ and ‘Information’ (Fig. 3) of challenge themes in groups C and D in Fig. 1, requires the development of CC adaptation scenarios which are relevant to likely future conditions, and which provide data about the context of decision making and (depending on the type of model) define at least some aspects of decision making itself. Constructing adaptation scenarios is complex, not least because of the issue of path dependency in iterative adaptive responses to progressive CC, discussed above. In addition, choices are likely to be affected by dynamic changes in stakeholder understanding as conditions change (Anwar et al., 2013). Data for scenarios may come from social science models or, be gathered from stakeholders or experts, and will therefore incorporate Uncertainty. In addition, data needed for scenarios includes information on the likely efficacy and impacts of adaptation strategies themselves, which can also be considered to be CC adaptation-specific. Given that participants in the current study highlighted the limitations to the data on adaptation efficacy, including relating to reliance on expert views, Uncertainty about the likely effectiveness of CC adaptation strategies can also be considered an adaptation-specific challenge within the cluster of challenges relating to Information available for model development, testing and use (Fig. 3, D). However, uncertainty relating to models themselves is common to modelling in general, while issues around the quality of data from climate models are important for both adaptation and CC impact modelling (Cammarano et al., 2017). Scenario development therefore brings together the CC adaptation-specific elements of Data availability and quality, Uncertainty and Novel scenarios, as ‘Information on adaptive responses to progressive change’ (Fig. 3, D).

Under progressive CC, the period over which stakeholders seek to optimise systemic outputs is important, as long-term and short-term goals may not align. How this trade-off is viewed is likely to alter with the considered time periods or the assumed pace and certainty of CC (Reilly and Schimmelpfennig, 2000). This is a specific challenge for CC adaptation modelling with the goal of Optimisation (Fig. 3, C), and represents the CC adaptation-specific aspect of Dynamic change modelling. Recent work has started to consider the application of approaches from other disciplines to agricultural settings, in order to build understanding of how changes in the efficacy of CC adaptations over time, and uncertainty in conditions and outcomes, can be incorporated into assessments of adaptation strategies (Dittrich et al., 2017).

Barriers to inter-disciplinary research collaboration have been well documented (Siedlok and Hibbert, 2014) and the need for coordination across disciplines and institutes to tackle CC challenges has been recognised (Soussana et al., 2012). Key to challenges A and C (Fig. 3), is Collaboration with social scientists with expertise in managing stakeholder engagement (Nguyen et al., 2014; Reed et al., 2014) and particularly those with expertise in normative and critical engagement approaches. However, inter-disciplinary research communities require time, resources, appropriate structures and the application of specific skillsets to flourish (Kipling et al., 2016c; Tomassini and Luthi, 2007). Initiatives such as MACSUR and the Agricultural Model Intercomparison and Improvement Project (AgMIP) (Rosenzweig et al., 2013) have driven progress in agricultural model development and use (Ewert et al., 2015; Sándor et al., 2017) and supported the application of inter-disciplinary expertise to region-specific CC issues (Dono et al., 2016; Özkan Gülzari et al., 2017; Schönhart et al., 2016).

The need to characterise a wider range of sometimes transformative adaptations in agricultural models, makes it essential to include, smaller and geographically marginal research groups in inter-disciplinary networks, to capture the diversity of expertise in the research community. These groups are vital to fully leveraging existing expertise, along with ‘core’ research groups that may find it easier to engage (Saetnan and Kipling, 2016). Although differences in context may prevent data on management responses to CC conditions in one location being used as a reliable predictor of change in another (Reilly and Schimmelpfennig, 2000), linking local research expertise across regions offers the opportunity to explore novel solutions, cross-pollinating ideas between scientific communities within and between disciplines. The need for integrated modelling approaches to investigate CC impacts and adaptation has been widely recognised (Reidsma et al., 2015a,b; Rötter et al., 2018), and the closer involvement of stakeholders in modelling processes is vital to the generation of model outputs with real-world relevance (Bellocchi et al., 2015; Hamilton et al., 2019). The distinct aspect of Collaboration for CC adaptation (Fig. 3, E) is therefore the urgency of the need to work together (resulting from the progressive nature of CC) (Hallegatte, 2009), focussing efforts on the specific challenges to agricultural modelling identified above (Fig. 3, A-D).

Illustrative reviews of the five CC adaptation-specific modelling challenges identified (Fig. 3) are provided in Appendix E, giving richer descriptions of how they are tackled by specific modelling communities.

Conclusions

This study sought to answer the question, to what extent is CC adaptation a novel challenge for agricultural modellers? The findings indicate that there are a number of CC adaptation specific aspects to the challenges of adaptation modelling identified by modellers. Within the three challenge categories of Use, Content and Capacity derived from the data, the theme of creating Novel (adaptation) scenarios was found to be entirely specific to CC adaptation modelling. Seven challenge themes, such as Resources for modelling and Scale interactions, represented essential pre-requisites for CC adaptation modelling but, were not considered specific to it. Ten other themes were considered general modelling challenges but, with CC adaptation-specific aspects. Most fundamentally, the importance of understanding and managing the influence of model focus, limitations, use and presentation on adaptive responses and their consequences was highlighted for both, bio-economic modellers and biophysical modellers. CC adaptation modelling draws agricultural modellers into social and political contexts in which their approaches and findings affect who wins and who loses, what is valued and what sacrificed, in the adaptation of agriculture to progressive CC. In modelling CC adaptation in agriculture, there is a need for the agricultural modelling community to focus on the aspects of model content and capacity relating to scope, optimisation, and information, on collaboration across disciplines and institutes, and on the responsibilities of modelling evolving responses to progressive CC.

Type	Name	Focus
Biophysical	Eco-DREAMS-S	animal
	DSSAT Platform	field
	PaSim	field
	FarmAC	whole farm
	Holos Nor	whole farm
	Melodie	whole farm
Economic	PAFAMO	whole farm
	Scotfarm	whole farm
Biophysical & economic	Dairy Wise	whole farm
	FarmDesign	whole farm
	FSSIM	whole farm
	MODAM	whole farm
	DiSTerFarm	whole farm/regional
	SFARMOD	whole farm/regional
Biophysical & economic (coupled)	MAgPIE	regional
	PASMA	regional
	FAMOS	whole farm

Table B.1

Types of data availability challenge

Data type	Management	Economic	Biophysical
Current systems	Data on risk perception of farmers: are they likely to use the strategy, why? Past experiences?	Information on economic costs of disease and treatment	Limited knowledge on the interactions between grassland productivity and associated ecosystem services
Different systems	Few long term datasets on Mediterranean grasslands Lack of data for low input grassland systems
Different systems as predictors of extremes	Lack of data for forage crops response to fertilizer under varied extreme event conditions there are examples of systems in extreme climates: We in NW Europe have little sense about them or data that may exist on them
Different scales	Scarcity of data is an important problem – we need to know what the ‘business as usual’ state of a farm is Usually very few data on management practices and productivity at territorial scale
Future predictions	Cost and availability of new technologies (e.g. breeds, soil management options). Also, the change in management required (e.g. new feed regime for new breeds)		Focussed climate scenarios needed (e.g. Northern Europe is likely to face wetter conditions and heat stress is not an issue) Limited knowledge on the interactions between grassland productivity and associated ecosystem services

Table C.1

Text relating to challenges to modelling agricultural impacts of and adaptations to climate change identified by Rötter et al. (2018) and its relation to themes identified in workshop data. TS = challenges identified by Rötter et al. (2018) as specific to tropical plant production systems.

Text extracts from Rötter et al. (2018) relating to modelling challenges	Themes
Economic models need to be combined with Crop Simulation Models in whole-farm assessments to better evaluate management practices	Collaboration; Interactions
Management and land use does not only respond to climate change, but also to changing socio-economic conditions, such as liberalization of markets or changes in dietary habits. Crop Simulation Modelling thus needs to be integrated into a larger modelling framework.	Collaboration; Interactions; Scale interactions
Statistical Models are constrained in many cases by the availability of adequate, representative yield data	Data availability; Data quality
Substantial mismatches between Crop Simulation Models and Statistical Models may indicate knowledge gaps regarding the mechanisms/processes that cause under-/overestimation of yield, et cetera	Data availability; Interactions
Despite some efforts, the effect of tillage on carbon storage has so far only been modelled with limited success, mainly due to insufficient field data to develop mechanistic descriptions in the models	Data availability; Interactions
At larger scales, Crop Simulation Modelling is severely hampered by lack of data for parameterization and calibration and management systems are often unknown. Large uncertainties persist — especially related to variability in managerial practices and spatial response patterns.	Data availability
Likewise, important tropical crops have been much less investigated in experiments regarding their exposure to agro-climatic extremes than those for temperate systems. Even with some progress in data availability, there is a need for both — more experiments and modelling — (as propagated by TROPAGS, see Fig. 3) to understand the underlying mechanisms. TS	Data availability
A key constraint to realistically upscaling the productivity of such systems (and how it is affected by climate change) to region level is, for instance, that fields of smallholder systems are not clearly defined, and a wide range of crop types can be found within a field. TS	Data availability
Besides improving crop models, fast track methods are needed to characterise and inventory smallholder fields as a basis for upscaling. Thereby the typical simplistic focus of modelling climate change impacts on sole crops (usually maize) in smallholder systems of Africa can be overcome. TS	Data availability
We need much better understanding of how climate effects scale with changes in low input systems TS	Data availability
Still, by far the majority of Crop Simulation Models deal with single season, single crop runs	Dynamic change
There is also an increasing interest in the role that agricultural management has on environmental impact, such as carbon sequestration or GHG emissions. However, carbon stocks need years to build up, thus long-term simulation over multiple years that also reflect the current deviation from the equilibrium state are necessary to capture that	Dynamic change
While ensemble approaches helped to make model predictions more robust and quantify the uncertainties, the next logical step was to improve responses to heat and the fundamental temperature functions in individual models, to eventually reduce the uncertainty by proposing improved functions and parameterization	Interactions modelling
… and also lack information on critical interactions of factors such as weather, soil and management practices	Interactions modelling
So far, crop models are not capable of capturing the multi-species interactions within one ‘field’ and the associated services delivered TS	Interactions modelling
Moreover, many systems are integrated crop-livestock systems, which makes the common use of the model output variable ‘yield produced per unit area’ difficult TS	Interactions modelling
One advantage of this method (use of statistical models) is that it inherently covers also indirect yield limiting factors, which are linked to climate variables, like pest and diseases. Process-based crop models so far largely ignore their effects, and thus fail to estimate farmer yields accurately in regions and years where biotic stresses are significant	Scope
While these (multi-model ensemble approaches) can be the basis for systematically exploring critical parameters and assumptions, they do not compensate for exploring missing mechanisms	Scope
Global model runs suggest strong effects of climate change on the crop production systems in the Global South, especially in Africa. However, such runs were done mainly for water limited and/or nutrient-limited yield, hence, with yields not limited by biotic stresses. That makes the results of little use to understand the actual effect of climate change on these systems, as many tropical plant production systems are heavily restricted by combinations of severe abiotic and biotic stresses TS	Scope
Many tropical systems are arguably more complex including agroforestry/intercropping. Unfortunately, crop models have been rarely tested/applied in such systems. TS	Scope

Table D.1

comparison of agricultural adaptation modelling challenge themes, themes drawn from Adger et al. (2018) (See S2.1 for description of themes), CC adaptation challenges previously identified for grassland and for animal health modellers, and broader climate-related challenges identified for these modelling disciplines. The Adger et al. (2018) theme ‘Developing new CCRA methods’ is not included in the table as it would associate with all themes within the ‘Capacity’ category in the current study.

Category (current study)	Theme (current study)	Themes from Adger et al. (2018)	Grassland modelling (adaptation specific)	Grassland modelling (general climate change)	Animal health & disease modelling (adaptation specific)	Animal health & disease modelling (general climate change)
Capacity	Collaboration	Inter-disciplinary and trans-disciplinary approaches		Links to other disciplines to explore impacts of changes in grasslands on the nutritional value of the sward for animals, & on the economics of systems; Fit-for-purpose models (use of model platforms & modular approaches for model integration)	Links to other disciplines to understand health interactions with other aspects of production; improved regional economic modelling of CC & socio-economic impacts of health changes & adaptive responses	Terminology and measurements: differences in international and interdisciplinary collaboration; Fit-for-purpose models (use of model platforms and modular approaches for model integration)
	Data availability	Data on adaptations (availability and quality); Socio-economic drivers (data availability); Biophysical knowledge (data availability)	Collation of data on adaptation strategy, their efficacy and impacts	Data for models (including availability, accessibility, quality); fitting model & data scale	Data on costs and efficacy of both health issues & interventions	Implicit in other challenges, e.g. need for data on different systems, on pathogen, pest & host ecology, at different scales & in relation to model scope
	Data quality					Need for data collection protocols, agreed standards, approaches and terminology
	Data accessibility	DATA ACCESSIBILITY NOT MENTIONED			Collating adaptation options related to different health conditions	Issues relating to data ownership and sharing in a competitive context
	Novel scenarios	Scenarios	Context specific adaptation scenarios based on stakeholder needs		Creating adaptation scenarios
	Resources for modellers	RESOURCE LIMITS NOT MENTIONED
	Uncertainty	Uncertainty	Implicit across challenges in terms of need to improve data and modelling capacity	Implicit across challenges in terms of need to improve data & modelling capacity & adequacy for stakeholder requirements – including using ensemble modelling approaches	Ability to model a range of climate change scenarios and the robustness of adaptive solutions across these; uncertainty in uptake likelihood for adaptation strategies	Implicit across challenges in terms of need to improve data and modelling capacity and adequacy for stakeholder needs/expectations, including accuracy
Content	Discrete events modelling	Mentioned only in relation to collaboration with disaster risk management research		Modelling the impact of extreme events
	Dynamic change modelling	Interdependence; Time lags in adaptation	Dynamics of uptake and implementation, threshold changes and carry-over effects		Incorporating implementation of adaptations over time	Pathogen and vector spread
	External limitations modelling	Interdependence; Scope		Implicit in need to extend model scope and to link models of different types		Implicit in need to extend model scope and to link models of different types
	Interactions modelling	Interdependence	Management driven by/a driver of biophysical change; Interactions between management changes and other systemic processes, & between different management changes	Modelling livestock & pasture interactions; Modelling plant responses to environmental change	Capturing farm and policy level strategies and their impacts; Improved modelling of environmental impacts on health, and of the biophysical processes via which adaptations cause change	Impacts of climate on health; Impacts of health on GHG emissions; Impacts of health on production; Interactions between health conditions, pathogens and interventions
	Management modelling	Cognitive bias	Inclusion of realistic decision making
	Scale interactions	Interdependence		Model & data scales – model linking, scaling data etc.		Modelling across spatial & temporal scales
	Scope	Scope; Diversity and number of factors	Novel adaptations (e.g. novel breeds) and systems (e.g. silvopasture)	Modelling different regions & production systems (to assess CC impacts); Challenges on incorporating into models (nutrient balances, GHGs, ecosystem services, soil variables & processes, pests & pathogens, overwintering, multi-species swards, nutritional variables)	Capturing farm and policy level strategies and their impacts	Variation in capacity between systems and nations; Nutrition and health; Pathogen, vector and host ecology; Genetics of health; Land use change and health
Use	Communication	Communication; Ethics; Expectations; Experience of decision makers; Interests		Fit-for-purpose modelling – engaging stakeholders to improve model relevance and understanding		Fit-for-purpose modelling – engaging stakeholders to improve model relevance and understanding
	Engagement approach			Fit-for-purpose models: Information on models and their capabilities made easily available for stakeholders, including limitations		Stakeholder involvement – to gain local info of disease patterns, & build trust & relevance through engaging in model development; Fit-for-purpose models: Info on models & their capabilities made easily available for stakeholders, inc. limitations
	Role of modellers
	User focus		Included in the need for context specific scenarios relevant to stakeholders	Making models fit-for-purpose in relation to stakeholder needs	Improved evaluation of model assumptions and performance, particularly for empirical models	Fit-for-purpose modelling meeting stakeholder needs; Validation of empirical relationships under CC