Human impacts on the biosphere are a matter of urgent and growing concern, with ecologists increasingly being asked to project biodiversity futures. The Intergovernmental Platform on Biodiversity and Ecosystem Services (IPBES) is likely to comprehensively assess such projections, yet despite being widely used and potentially critical tools for analysing socio-environmental futures, integrated assessment models (IAMs) have received little attention from ecological modellers. We aim to raise awareness and understanding of IAMs among ecologists by describing the structure and composition of IAMs, assessing their utility for biodiversity projections and identifying limitations that hamper greater interaction between scientists using IAMs and those using ecological models. We also hope to inspire more accessible and applicable models by suggesting development needs for IAMs.
We conduct a systematic review of four state-of-the-art IAMs, which describes and contrasts key model features and analyses six aspects of IAMs that are of fundamental interest to ecologists.
IAMs could be valuable for modelling biodiversity futures; however, current IAMs were not developed for this application and challenges remain for ecologists looking to use their outputs. Separating and understanding the differences resulting from IAM formulation and those resulting from specific scenario assumptions is currently problematic, and current IAMs may be unable to accurately represent environmental conditions for both Earth-system projections and for building robust models of biodiversity because key ecological processes are absent. We suggest that model intercomparisons would identify differences in model dynamics, and detailed studies of how dynamical interactions between components influence behaviour would address why such differences arise. Bio-economic fisheries models and agriculture pollination models provide starting points for integrating key ecological feedbacks within IAMs. Ultimately, making IAMs more accessible within the multidisciplinary study of global change, drawing on user-centred research, would enable more resolved, reliable and accurate assessment of how Earth's socio-ecological system is approaching planetary boundaries.
The major driver of terrestrial biodiversity loss currently, and in the near future, is land-use and land-cover change (Bawa & Dayanandan, 1997; Tilman et al., 2001; Mace et al., 2005). However, most studies exploring biodiversity futures have focused on climate change in isolation (Huntley et al., 1995; Erasmus et al., 2002; Peterson et al., 2002; Thomas et al., 2004; Thuiller et al., 2005; Lawler et al., 2009) with few considering the separate or combined effects of other drivers (Van Vuuren et al., 2006b; Jetz et al., 2007). This represents a large knowledge gap at the science–policy interface (Pereira et al., 2010; Schweiger et al., 2012; Spangenberg et al., 2012). This gap is particularly important for the newly formed Intergovernmental Platform on Biodiversity and Ecosystem Services (IPBES), which aims to aggregate understanding of the relationships between biodiversity and people, and to support policy formulation and implementation (Perrings et al., 2011; Turnhout et al., 2012).
Integrated assessment models (IAMs) could be important for understanding and predicting the effects of human development on the biosphere. Such models combine representations of human energy use, industrial development, agriculture, land-use/land-cover changes and scenarios of the future development of human societies in order to make projections about the future of anthropogenic and natural ecosystems (see Box 1 for a glossary; Fig. 1).
Box 1. Definitions for important terms used in integrating assessment modelling
A scenario can have various meanings, such as: ‘hypothetical sequences of events constructed for the purpose of focusing attention on causal processes and decision-points’ (Kahn & Wiener, 1967), or ‘a description of potential future conditions produced to inform decision-making under uncertainty’ (Parson et al., 2007)
Here scenarios represent qualitatively different descriptions of hypothetical futures to be used as inputs to integrated assessment models (IAMs), developed to inform decision-making under uncertainty. Elsewhere, outputs from IAMs are used as scenario inputs for other models, for example emissions scenarios for climate studies (Nakicenovic et al., 2000; Meehl et al., 2007)
When explored using an IAM, a scenario is represented by assumptions regarding the value of parameters and the selection and structuring of elements in the modelled system, as well as the inputs of exogenous driving information to that system (e.g. socio-economic development trend or a carbon mitigation strategy)
The process for understanding uncertainty through the exploration of a range of possible futures (scenarios). The process for which IAMs are commonly used
A quantitative forecast within the assumptions of an individual scenario. Publically available projections from IAMs include spatially explicit emissions of greenhouse gases and land-use changes
Representative concentration pathway (RCP)
The temporal trajectory of atmospheric greenhouse gas (GHG) concentrations. They are referred to as representative owing to the fact that they are one of a set of scenarios that could give rise to similar radiative forcing and emissions characteristics (Moss et al., 2008)
RCPs are the output from IAMs and feed into climate models. The concept behind RCPs is that a range of different socio-economic development and mitigation options may give rise to similar pathways of atmospheric concentrations. Therefore, identifying representative pathways enables the earth system community to study the climate effects of these whilst the IAM community can explore the scenarios that give rise to such pathways
General equilibrium models
Are sector (energy or food) or economy-wide models accounting for all linkages between goods in that domain? At the heart of these models are equations that describe the assumed behaviour of consumers, producers and regulators of goods. Consumers create demand for goods as a function of their price, whilst producers supply goods to the market also as a function of price. General equilibrium models make the assumption that effects of the changing price of a good will on its own lead to a price at which supply and demand are equal. Once this price is calculated, the use of resources and distribution of income resulting from the transaction is calculated (see for example Borges, 1986). As opposed to partial equilibrium models (see below), general equilibrium models represent all interactions between goods in the sector covered and calculates prices such that all supply and demand balance is met for all goods in the sector. So, for example, competition for labour may take place between producers of different goods. General equilibrium models have been criticized for the fact that they cannot reflect phenomena that arise from imbalances of supply and demand – for example unemployment, where supply of producers outweighs demand from consumers (Böhringer & Löschel, 2006). Due to the greater complexity of the interacting commodities, general equilibrium models generally represent fewer commodities than partial equilibrium models, thus providing a coarser representation of regional and global economic activities
Partial equilibrium models
Partial equilibrium models, similar to general equilibrium models (described above) represent one or more sectors of an economy and at their core are equations that describe the assumed behaviour of, for example, consumers, producers and regulators of goods. Consumers demand different amounts of goods depending their price, whilst the quantity producers supply to the market is also determined by the price of the good. Partial equilibrium models focus on a specific good and calculate the price at which supply and demand of that good are equal, similar to general equilibrium models making the assumption that price alone will lead to this equilibrium. In contrast to general equilibrium models, partial equilibrium models assume that other factors that affect the supply and demand of this good remain constant (Francois & Hall, 1997). So, for example, a policy change implemented by a regulator is assumed to only affect the price of that good. Furthermore since the cost of that good has decreased, consumers will have more income available to spend on other goods; however, this ‘spill-over’ demand is neglected in partial equilibrium models (Piermartini & Teh, 2005). The advantage of partial equilibrium models is that they are computationally tractable and can therefore resolve a greater number of goods within the sector(s) of focus
Current IAMs are largely used for modelling climate change and investigating options for climate mitigation. Their key predictions in this context are of anthropogenic greenhouse gas emissions. However, they also provide projections for other variables, including, for example: land cover and land use (including deforestation rates, agricultural crop cultivation areas, types and inputs); and technology application within the energy or industry sectors.
Most ecologists engaged in forecasting ecological futures are familiar with climate change scenarios and projections, such as those contributing to the Intergovernmental Panel on Climate Change (IPCC) (Thomas et al., 2004; Thuiller et al., 2005; Lawler et al., 2009) and the emissions scenarios on which these are based. However, fewer are familiar with the IAMs that underpin these emission scenarios (Box 1) (Nakicenovic et al., 2000; Meehl et al., 2007). This bias by ecologists, preferentially using climate model outputs instead of land-cover predictions, is perhaps surprising given the potential use of IAMs in ecological forecasting. This might be explained by the greater accessibility of climate models and their outputs compared with IAMs (Houghton et al., 1996, 2001; Solomon et al., 2007; Kang et al., 2011; Min et al., 2011; Overpeck et al., 2011; Pall et al., 2011).
Rather than using IAMs as black boxes, however, there are significant benefits to be gained for practitioners analysing the outputs of IAMs in detail and exploring both the scenario assumptions and the model mechanisms giving rise to those projections. For example, it is important to note that the subsystems and their representations relevant for simulating the climate change dynamics and interactions with anthropogenic activities – the focus of current IAMs – may not be the same as those relevant for providing ecologically relevant outputs.
This paper is motivated by the fact that many IAMs have limited, but often highly technical, documentation that is difficult for non-specialists to grasp, and this might limit the uptake of IAMs and their projections by ecologists. Here, in an attempt to make IAMs more accessible to ecologists, we use published documentation to describe (1) components of and philosophy behind a typical IAM, (2) similarities and differences among four state-of-the-art IAMs, (3) the ecological components that these models include, (4) the projections made by these models, (5) why IAMs may be valuable for ecologists, and (6) how to select an IAM for specific purposes. Then we examine two developments that integrated assessment modellers could implement to make IAMs more accessible and applicable to a wider community, including ecological modellers. Specifically we propose: (1) promoting greater understanding, and (2) incorporating key ecological concepts within IAMs.
Six Things Ecologists Should Know about Integrated Assessment Models
Earth is a complex system which includes a huge number of interacting natural and anthropogenic subsystems. IAMs are a formal quantitative representation of a subset of the Earth system (Parson & Fisher-Vanden, 1997), in which scientific and technical knowledge about the behaviour of, and interactions among, the socio-economic, human energy use and environmental subsystems have been synthesized.
Inputs to IAMs are based on scenarios, qualitatively different descriptions of hypothetical futures (Fig 1 & Box 1; Settele et al., 2012). They are not intended to describe future conditions with complete precision but rather highlight essential characteristics of potential futures (Parson et al., 2007). In general, scenarios are developed by a process of gathering expert opinion using techniques such as workshops, panels and surveys (Börjeson et al., 2006). The constraining limits to scenarios are determined by the people generating them, but are influenced to a large degree by the type of scenario to be generated; for example, scenarios aiming to establish ‘what could happen in the future’ may allow a wider scope than that those aiming to describe ‘what will happen’. Internal consistency is a key characteristic of well-designed scenarios (Spangenberg et al., 2012) and is most often evaluated by peer review, although there are other formal tools to aid with this evaluation (Börjeson et al., 2006).
A scenario is represented in an IAM by the selection and structure of elements in the modelled system, and the exogenous driver information input to that system (e.g. socio-economic development trends). Used in conjunction with scenarios, IAMs offer projections (Box 1) on the plausible changes in those subsystems over the coming decades.
Integrated assessment modelling has been applied across a range of research areas, including water resource management (Jakeman & Letcher, 2003) and global earth system change (where IAMs were used to study the cross-disciplinary issue of acid rain). Subsequently, the use of IAMs was focused on attempting to predict the dynamics of and consequences for climate change arising from different anthropogenic future climate scenarios. Emission scenario outputs from IAMs are essential to climate modellers for assessing future climate change projections (Moss et al., 2010), and as such they have become cornerstones of the global environmental policy debate. However, it is important for the ecological modelling community to recognize this focus on projecting climate change because those subsystems necessary for simulating climate dynamics and interactions with plausible human futures may not be the same as those subsystems necessary for providing ecologically relevant outputs.
Components of a typical IAM
A typical IAM includes both natural and human subsystems (Fig. 1; Weyant et al., 1996; Parson & Fisher-Vanden, 1997). Natural subsystems comprise both biological and physical components. Terrestrial vegetation models, also combining biological and physical components, are used to represent both agricultural and natural autotrophs. Modelled natural vegetation then enters a simplified representation of the terrestrial carbon cycle. Hydrological models provide inputs for vegetation models and models of nutrient cycles. Physical models of the ocean–atmosphere subsystem capture changes in atmospheric constituents and the resultant radiative feedback, and calculate changes in sea level, whilst biogeochemical ocean models incorporate marine sources and sinks of carbon, though they do not incorporate biophysical autotroph models analogous to the terrestrial vegetation models. Many other ecological processes are not represented in current IAMs (as indicated in Fig. 1) and some important omissions are discussed further below.
In the human subsystem, energy supply and demand are calculated such that they meet the requirements of the simulated human population, and may take into account variables such as primary fuel supply, commodity prices and the characteristics and development of technology (Messner & Schrattenholzer, 2000; De Vries et al., 2001; Brenkert et al., 2003). Industrial production is determined according to scenario-specific economic and demographic trajectories (Messner & Schrattenholzer, 2000; Brenkert et al., 2003). Food production and consumption are affected by land availability, food commodity prices, human population size, consumption rates and crop vegetation biophysics (Brenkert et al., 2003; Vanmeijl et al., 2006), whereas capture fisheries and aquaculture are not included in the models.
State-of-the-art IAMs produce spatially explicit estimates of human land use and greenhouse gas emissions. For example, outputs such as land-use changes are available at grid cell resolution from 0.5° in IMAGE (integrated model to assess the global environment) (Van Vuuren et al., 2006a). However, some components of the human system, such as energy supply and demand, are modelled at coarser resolution, being resolved based on regional groupings of countries (IIASA, 2006; MNP, 2006). IAMs typically operate with a time-step duration on the order of years, due to the high computational requirements for such complex and large-scale models. However, some subsystems, such as vegetation, may be modelled at a higher temporal resolution.
Having described the general structure of current state-of-the-art IAMs above, the next section will draw comparisons between four of the principal models.
The principal IAMs, their similarities and differences
Whilst there are many IAMs (Weyant et al., 1996; Moss et al., 2008), we focus on the four chosen by the Integrated Assessment Modelling Consortium to constitute the representative concentration pathways (RCPs) (Box 1) to be used in the fifth assessment report of the IPCC (hereafter AR5) due to be published in 2014 (Moss et al., 2010). Each RCP is designed to calculate a different trajectory for the evolution of greenhouse gas concentrations in the atmosphere (Moss et al., 2010; van Vuuren et al., 2011). Four IAMs were chosen that met the RCP selection criteria and are likely to become widely used once reported in AR5. The four models are: AIM (Asia-Pacific integrated model), IMAGE, MESSAGE (model of energy supply strategy alternatives and their general environment) and GCAM (global climate assessment model) (summarized in Box 2).
Box 2. Summaries of each of the four integrated assessment models reviewed in this paper, namely those used to construct the representative concentration pathways for the Fifth Assessment Report of the Intergovernmental Panel on Climate Change
AIM (Asia-Pacific integrated model)
AIM comprises a set of inter-related models developed at National Institute for Environmental Studies (NIES) and Kyoto University and used for assessing policy options on sustainable development particularly in the Asia-Pacific region (Kainuma et al., 2002a). The AIM framework employed for the representative concentration pathways (RCPs) simulations consisted of AIM/Impact [Policy], a dynamic optimization model (Hijioka et al., 2006, 2008), which calculated global emission projections that were subsequently used to constrain AIM/CGE [Global], a general equilibrium model, which calculates regional activity and emissions across a range of economic sectors (Masui et al., 2011)
AIM/Impact [Policy] has at its core a global but single-region model of energy and economy, which aggregates global economic activities into a single product. The model calculates an economic trajectory that maximizes the total satisfaction derived globally from consumption of that product under specified constraints, for example under global warming control targets (Hijioka et al., 2006; Masui et al., 2011). The economic impacts of projected emissions are calculated using a simple climate model and an impact assessment model (Hijioka et al., 2006)
The AIM/CGE [Global] is a general equilibrium model that resolves the human world in 24 divisions but represents the Asia-Pacific region in greater detail. Within each region the supply and demand of 21 types of economic goods and services are simulated. The model is calibrated against global trade analysis project (GTAP) and International Energy Agency (IEA) data for economic activity and energy in the year 2001 (Masui et al., 2011)
IMAGE (integrated model to assess the global environment)
The IMAGE model originated with the RIVM (National Institute for Public Health and the Environment) and is now developed by PBL (Netherlands Environmental Protection Agency). The first version of the IMAGE model was a single region and aimed to explore the causes and effects of climatic change. The most recent model version is IMAGE 2.4, which defines 24 world regions (plus Antarctica and Greenland) (PBL, 2011)
The core of the IMAGE model comprises: the IMAGE energy regional (TIMER) model, which simulates regional supply and demand of 12 energy carriers (De Vries et al., 2001); agriculture, as modelled by the GTAP model (Hertel, 1997); a terrestrial vegetation model (TVM) (Vanmeijl et al., 2006); a land-cover model (Alcamo et al., 1998); and it has been linked with the model for the assessment of greenhouse-gas induced climate change (MAGICC) (Hulme et al., 1995)
Economic, demographic, energy supply demand and the agricultural economy are all represented at the regional level, whilst land use and cover is defined for a 0.5° × 0.5° grid of the world (PBL, 2011)
MESSAGE (model of energy supply strategy alternatives and their general environment)
MESSAGE is used to describe a collection of models and modules developed by International Institute for Applied Systems Analysis (IIASA) and forming part of the integrated assessment scenario analysis framework
The MESSAGE framework is global in scope resolving 11 world regions: North America, western Europe, Pacific Organization for Economic Cooperation and Development, central and eastern Europe, newly independent states of the former Soviet Union, centrally planned Asia and China, South Asia, other Pacific states, Middle East and North Africa, Latin America and Caribbean and sub-Saharan Africa (IIASA, 2006). The model configures the evolution of the integrated system from base year to the end of the time horizon in intervals of 10 years
The MESSAGE model is itself a systems engineering optimization model, providing a framework for representing an energy system from resource extraction or imports through conversion and distribution to the provision of end-use services. MESSAGE is coupled with: the MACRO model, a top-down macroeconomic equilibrium model (Messner & Schrattenholzer, 2000); DIMA (dynamic integrated model of forestry and alternative land use) (Rokityanskiy et al., 2007), a spatially explicit representation of the interactions and feedbacks between ecosystems and human land-use; and the agricultural-ecological zones (AEZ) module that represents crops and natural vegetation (Fischer et al., 2002b)
GCAM (global climate assessment model)
GCAM is descended from MiniCAM (mini climate assessment model), and is available as a community tool; efforts are being made to expand the user community and development of the model outside of the Pacific Northwest National Laboratory (PNNL) group
The GCAM framework represents the world in 14 geopolitical regions and combines economics-based energy and agricultural models with a coupled climate assessment model
GCAM is a partial-equilibrium model comprising of three modules (Brenkert et al., 2003): energy supply and demand using an updated version of the Edmonds–Reilly–Barnes model (ERB) (Edmonds et al., 1986; Brenkert et al., 2003); the agriculture and land use model (AgLU) (Brenkert et al., 2003; Sands & Leimbach, 2003); and, atmospheric composition and global-mean climate changes using the model for the assessment of greenhouse-gas induced climate change (MAGICC) (Hulme et al., 1995). Model solutions for these modules are calculated every 15 years within the GCAM model, although recent developments have seen the model using a shorter time-step of 5 years or 1 year
In order to be an RCP candidate, each IAM projection must among, other things: (1) be peer reviewed and published; (2) describe future changes in all of the required sources of greenhouse gas emissions, including land use/land cover, through to 2100; and (3) incorporate the global carbon cycle and atmospheric chemistry. Each RCP is modelled by a different IAM and represents a qualitatively different scenario of human development (Table 1), comprising a different set of socio-economic assumptions and contains an independent set of components and internal structure.
Table 1. The four integrated assessment model–representative concentration pathway (RCP) scenarios, summarizing the model used to interpret the pathway and the socio-economic development trajectories assumed for each
GDP (trillion $)
GHG, greenhouse gas; CCS, carbon capture and storage.
*Dollar value year is not specified.
aAbsolute gross domestic product (GDP) unknown for this RCP but GDP growth rates roughly follow reference scenario of World Energy Outlook to 2030 (IEA, 2004), then converge to prior SRES B2 trajectory.
bIn 1990 dollars.
cIn 2000 dollars.
Asia-Pacific integrated model representative concentration pathway 6.0 (AIM RCP6.0)
Based on the AIM interpretation of the special report on emission scenarios (SRES) B2 scenario, with a GHG concentration stabilization target of 860 p.p.m.
Integrated model to assess the global environment representative concentration pathway 2.6 (IMAGE RCP2.6)
Follows baseline B2 scenario but with a 400 p.p.m. CO2-eq. target and makes optimistic assumptions for land use, energy efficiency and bio-energy (both the available potential and the combination of bio-energy and CCS)
Productivity change over time based on Bruinsma (2003) to 2030 and then converges to 0.25% year−1 thereafter
Smith & Wigley (2006), Clarke et al. (2007), Wise et al. (2009)
We compared the four IAM implementations based on their publicly available documentation, and below we summarize the similarities and differences among the energy and industry, agriculture and food, vegetation and ocean–atmosphere subsystems. (Appendices S2–S5 in the Supporting Information contain compact comparisons of each of these subsystems across the IAMs.)
Energy and industry subsystems
The structure of, and high-level inputs to, each IAM energy subsystem model are similar, including a description of technologies used in energy and industrial production, technological development pathways and assumptions for population trends and economic development (Appendix S2). However, there are differences in model implementation: AIM employs an aggregate, global dynamic optimization model coupled to a general equilibrium model (Box 1) to sequentially calculate the state of the energy and industry subsystems (Kainuma et al. 2002b; Fujino et al. 2006; Masui et al., 2011), whilst the GCAM model makes use of the Edmonds–Reilly–Barns (ERB) partial equilibrium model (Edmonds et al., 1986) (Box 1).
In return for greater internal consistency, general equilibrium models sacrifice some computational tractability compared with partial equilibrium models, which calculate the market price of one or more commodities for which supply and demand are in perfect balance but assume that supply and demand for other commodities are held constant. General equilibrium models calculate the market price of all commodities taking into account the interactions between them, for example competition for land production of crops or forestry products. They therefore provide a relatively coarser representation of regional and global economic activities.
MESSAGE uses an optimization model with perfect foresight. As the name implies, the optimization model calculates the optimal developments within the system of study for a set of given constraints, which in the case of MESSAGE could be, for example, maximum reduction of greenhouse gas emissions or the lowest economic cost reduction. Finally, IMAGE makes use of a deterministic rule-based model, which attempts to re-create the decision-making processes for actors in the energy and industry systems. For example, decisions by electricity generating organizations are simulated by modelled investment in generation technologies, where competition for technology investment is based on economic performance in terms of capital and operational costs.
Agriculture and food subsystems
There is greater similarity across the IAM representations of agriculture and food subsystems (Appendix S3), with AIM, IMAGE and MESSAGE making use of general equilibrium models to calculate the price of agricultural commodities at which supply and demand reach equilibrium across all representative agricultural and food products simultaneously.
The general equilibria for agriculture and food subsystems are likely to be consistent with those calculated for the energy and industry subsystems, since models for both subsystems appear to be driven by the same assumptions about the socio-economic scenario (e.g. Riahi et al., 2011; Van Vuuren et al., 2011). There also is some direct linkage between these subsystems in the IMAGE and MESSAGE models, in which the demand for bio-energy crops is calculated within the energy model in each and input to the agriculture and food subsystem models (Fischer et al., 2005; Vanmeijl et al., 2006). In contrast, the GCAM model makes use of a partial equilibrium model, which as a result of the simpler assumption of constant prices of complementary and substitute products can resolve 14 representative agricultural products, a greater number than any of the general equilibrium models.
The treatment of agriculture and food is inadequately described in the public documentation of the AIM model at present and thus the details of the formulation are unclear beyond the fact that a general equilibrium approach is adopted. In all models, with the exception of AIM for which this feature is unclear, demand for agricultural products is dependent on population, price and calorific requirement. Supply of agricultural products is determined by the availability and productivity of land within the region of interest; as the price of agricultural commodities rises, more land is made available to supply this produce. Productivity information is derived from externally supplied data (e.g. Bruinsma, 2003) in GCAM and calculated endogenously through vegetation models in IMAGE and MESSAGE.
IMAGE, MESSAGE and the natural vegetation component of AIM model vegetation using different biophysical models of productivity (Appendix S4) that take into account both environmental conditions and characteristic physiological properties of natural and agricultural vegetation types (Leemans & Solomon, 1993; Leemans & van Den Born, 1994; Alexandrov et al., 2002; Fischer et al., 2002a, b; Ito, 2010). Productivity in GCAM comes from externally supplied yield data (e.g. Bruinsma, 2003), so there is no feedback of future changes in the physical environment on crop yields. Land is allocated to uses (e.g. unmanaged, crops, pasture, forest) according to a set of rules, which vary across the models, but generally on the basis of yield information or to maximize economic performance.
Three of the four IAMs reviewed (IMAGE, MESSAGE and GCAM) link with the model for the assessment of greenhouse-gas induced climate change (MAGICC) (Hulme et al., 1995) to represent the ocean–atmosphere subsystem and natural carbon cycling (Appendix S5). Using greenhouse gas emissions profiles and global deforestation as inputs, the MAGICC model then calculates internally the flux of atmospheric CO2 to the oceans and to terrestrial vegetation and soils; the resulting atmospheric concentrations of greenhouse gases; the radiative forcing of these atmospheric concentrations; global mean temperature changes and concomitant sea level rise. In the IMAGE and MESSAGE models, the calculated changes in global mean temperature feed back to crop productivity through the biophysical vegetation models. The ocean–atmosphere subsystem model for AIM is limited in its documentation and therefore it is unclear how environmental changes calculated in this subsystem affect the rest of the model.
This subsection has described the relative complexity of four state-of-the-art IAMs and has demonstrated their structural similarity but variable subsystem implementation. Of particular ecological importance is the common model assumption in the agriculture and food subsystem that as the demand for agricultural produce, and thus the price for these goods, rises, the area of land converted to agricultural production will increase without an apparent limit on availability of agricultural land (other than the physical limit of land area). Another emergent message is that amongst the subsystems included in the models, the natural world, and ecological systems in particular, are under-represented relative to other components.
Ecological components in IAMs
Simulations of terrestrial vegetation are the only explicit ecological component included in IAMs (except for human activities).
When considering the ecology of natural vegetation, IMAGE groups natural plant species into plant functional types (PFTs) (Leemans & Van Den Born, 1994), sets of plant species exhibiting similar responses to environmental conditions and having similar effects on the dominant ecosystem processes, e.g. tropical evergreen trees or warm grasses and shrubs (Noble & Gitay, 1996). Climatic thresholds for each PFT (e.g. mean temperature of the coldest month) are used in conjunction with a dominance class, which defines the PFTs that will dominate at a particular location, to determine the vegetation composition. MESSAGE groups natural vegetation zonally, such as tropical rain forest, grasslands or subhumid woodlands/scrub. The distribution of zones is not modelled explicitly, but taken from the International Geosphere–Biosphere Programme land-cover dataset (Obersteiner et al., 2006). Productivity for biomes is modelled as a function of environmental information and parameterized against a dataset of net primary production (NPP) (Alexandrov et al., 2002). AIM uses a Holdridge classification scheme to classify land areas according to precipitation, temperature and evapotranspiration (Holdridge, 1947; Matsuoka et al., 2001). It is unclear from AIM publications if this approach is employed in the current modelling framework. Masui et al. (2011) also describe the VISIT model used in AIM, which attempts to capture the ecophysiological processes that drive carbon and nitrogen cycles in natural ecosystems (Ito, 2010). GCAM represents a land-use type for forests managed for wood products within its agriculture and land-use model, AgLU, but yields are derived from exogenous sources.
Both land use, such as pasture or cropland, and land cover given by crop types, are represented in the agricultural vegetation subsystems of IAMs as follows: common varieties of agricultural crops (12 varieties) are modelled in IMAGE by defining information on climatic requirements, such as growing period length or temperature of coldest month, specific to the crop variety (Leemans & Van Den Born, 1994). Productivity for suitable crop varieties is calculated using specific functions for the rates of photosynthesis and respiration, and may be adjusted according to location-specific environmental information (Leemans & Van Den Born, 1994). MESSAGE calculates crop suitability in a similar manner to IMAGE, but includes a more comprehensive array of crop, fodder and pasture types (154 in all; Fischer et al., 2002b). Crop yields are calculated following Kassam (1977) and Fischer & Antoine (1994), using maximum photosynthetic rate, temperature and photosynthetic pathway, and are subsequently constrained according to agro-climatic conditions. Crop productivity in the AIM framework depends on climatic and environmental parameters such as photosynthetically active radiation, potential evapotranspiration, temperature, soil moisture and defined crop characteristics (Matsuoka et al., 2001). Again, it is unclear if this approach is still used in the current version of AIM. GCAM, as described above, does not model the biophysics of crops, but instead makes use of exogenous yield information.
Absent ecological processes
As well as describing the ecological components included in IAMs, it is important to highlight the ecological processes that are absent from IAMs. We do not intend to criticize integrated assessment modellers, who did not develop these models with the goal of ecological realism, but to identify areas where ecologists could help to improve the realism of IAMs in order to incorporate feedbacks from ecological processes to other subsystems, and to improve the applicability of these models for predicting ecological futures.
Most notably, all of the IAMs reviewed here focus solely on the land surface, and do not incorporate any ecological aspects of aquatic realms, either marine or freshwater (Fig. 1). The MAGICC model of the ocean–atmosphere subsystem, to which the IMAGE, MESSAGE and GCAM models have been linked, simulates a variable flux of carbon from the atmosphere to an inorganic ocean reservoir (Wigley, 1993). The absence of aquatic ecological processes is especially surprising given the importance of aquatic ecosystems for food production, with fish from marine and freshwater sources representing 15.7% of animal protein consumed globally by humans (FAO, 2010), representing on average 8% of aquatic primary productivity and as much as 24–35% on continental shelves (Pauly & Christensen, 1995). There have already been enormous declines in many wild fish stocks globally (Myers & Worm, 2003; Worm et al., 2009), with marine biodiversity being disproportionately concentrated in areas of high human impact (Tittensor et al., 2010), and these declines are likely to continue into the future (Worm et al., 2009; Garcia & Rosenberg, 2010). Aquaculture has recently grown substantially, and is likely to continue to grow in the future, supplanting capture fisheries as the most important source of fish and other aquatic food (Bostock et al., 2010).
The development of marine capture fisheries and aquaculture is closely linked with terrestrial agriculture, such that, for example, the supply of fish has an impact on the aggregate demand for protein from terrestrial sources (Brashares et al., 2004). From both the perspective of identifying the causes of climate change and of increasing applicability for studying ecological futures, there is a strong need to include marine ecology and fisheries in IAMs.
Humans rely on the biosphere for a range of services, including the provision of food and fresh water, nutrient cycling and soil formation (Reid et al., 2005; Díaz et al., 2006; Carpenter et al., 2009). Other than the effects of the abiotic environment on plant productivity, none of the models incorporate any feedback from the biosphere on human activities or populations (indicated in Fig. 1). This may lead to unrealistic projections within the defined scenarios of socio-economic development. For example, actions that reduce the number or composition of species in natural systems may compromise ecosystem functioning, as the ability of ecosystems to provide services may depend on both these aspects (Loreau et al., 2001; Tilman et al., 2001; Hooper et al., 2005; Isbell et al., 2011). Such feedbacks, while not fully understood, could be incorporated into future IAMs to more explicitly assess the impacts of human actions on the biosphere and resultant consequences for the human system.
Details of IAM scenarios and projections
Projections generated by IAMs depend on the specific scenario assumptions the models have been applied to. Several scenario analyses have used IAMs to generate projections (Nakicenovic et al., 2000; UNEP, 2007; Bakkes et al., 2008; IAASTD, 2009). Here we focus on the RCP projections for illustrative purposes since they are the most recent publically available IAM outputs including spatially explicit projections of land-use change.
The RCP scenarios vary in specific assumptions (Table 1): first, the greatest human population growth is assumed by the MESSAGE-RCP8.5 scenario and the lowest by the GCAM-RCP4.5 scenario; second, the GCAM-RCP4.5 scenario assumes the highest global gross domestic product (GDP) trajectory and the MESSAGE-RCP8.5 scenario the lowest. Note that GDP values are not reported for the IMAGE-RCP2.6 scenario.
A range of outputs are available for each RCP from version 2.0.5 of the database (IIASA, 2009), including atmospheric concentrations of the greenhouse gases CO2, CH4 and N2O, spatially explicit emissions of greenhouse gases commonly broken down by major source activity and spatially explicit land use and land-use transitions.
Because each RCP is the product of a combination of a different model implementing a different socio-economic trajectory, direct comparisons and attributions of observed differences between them are near impossible. However, in the following two subsections we perform global and regional comparisons of the spatially explicit projections of land-use and land-cover changes made by each IAM and attempt to summarize broad-scale differences. For each RCP projection the area of each land-use type was calculated from the proportion of each grid square belonging to each land-use type and the area of the cell (for details of the methodology see Appendix S1).
Global land-use comparison
At the global scale in all RCP projections the area of primary vegetation will decrease over time and the area of secondary vegetation will increase (Fig. 2a, b). The greatest decrease in area of primary vegetation by 2100 is projected by MESSAGE-RCP8.5, conforming to expectations as this scenario assumes the highest trajectory of population growth and gives rise to the highest atmospheric concentrations of greenhouse gas. The lowest decline in primary vegetation is projected by AIM-RCP6.0, which is somewhat surprising given the higher population growth assumed by this scenario relative to IMAGE-RCP2.6 and GCAM-RCP4.5. However, AIM-RCP6.0 assumes limited expansion of livestock globally, reducing demand for agricultural land, whilst both the GCAM-RCP4.5 and IMAGE-RCP2.6 scenarios assume strong expansion of bio-energy, and IMAGE-RCP2.6 further assumes increased demand for meat amongst the population, adding further demand for agricultural land.
For crop, pasture and urban land (Fig. 2c–e), the direction of change varies across the IAM outputs. For example, cropland extent in the GCAM-RCP4.5 projections decreases over time, while the other IAMs show increasing cropland area. For pasture land the AIM-RCP6.0 projects a reduction in area, whilst the GCAM-RCP4.5 projects a decline by 15% until 2070, after which it increases slowly. IMAGE-RCP2.6 projects a small decline (2% by 2100), whilst MESSAGE-RCP8.0 projects a 10% increase in area. AIM-RCP6.0 and MESSAGE-RCP8.5 project a steep increase in urban land area, whilst urban area remains constant in IMAGE and GCAM, as the extent of urban area is fixed by the models.
The GCAM-RCP4.5 scenario assumes strong expansion of bio-energy, particularly commercial electricity generation from biomass (Clarke et al., 2007), and some crop and pasture lands are converted to bio-energy crops. Necessarily, this projection produces a strong intensification of global agriculture to support a larger global population on reduced crop and pasture land area. All other projections suggest that global cropland areas will expand throughout this century. It is difficult to assign this outcome to a particular difference in IAMs because variation in model formulation is confounded with different scenario assumptions across the RCPs. This lack of comparability is of fundamental importance and will be covered in more detail in the discussion below.
When comparing the four projections by continent (Fig. 3), the consensus across all scenarios and across all continents is that area of primary vegetation will decline, showing the continued global extent of and demands from direct effects of human on terrestrial ecosystems. In absolute terms the most substantial declines in primary land area are projected to occur in Asia, where all models predict a near linear decline.
The most striking divergences are seen in the extents of pasture and cropland. For example, according to the GCAM-RCP4.5 projection, the area of cropland decreases in all regions, whilst pasture area also shows small declines. In contrast, AIM-RCP6.0 projects substantial decreases in pasture land area but increases in cropland area, thus indicating a shift in agricultural practice away from extensive livestock farming.
Value of IAMs for ecologists
Despite their limited representation of ecological processes, IAMs could be of fundamental importance for understanding future human development and its effect on the trajectory of the biosphere. They capture representations of energy, industrial development, agriculture and land-use/land-cover changes within a single modelling framework. Apart from projections of greenhouse gas emissions, IAM projections of changes in land use/land cover are currently publicly available (IIASA, 2009).
To project biodiversity changes in the future ecologists frequently employ projections of global temperature changes (Thomas et al., 2004; Thuiller et al., 2005; Lawler et al., 2009), which are underpinned by emission projections generated by IAMs. Given the importance of changes in land use and land cover as a driver of ecological change (Bawa and Dayanandan, 1997; Tilman et al., 2001; Rindfuss et al., 2004; Mace et al., 2005), spatially explicit projections of land-use/cover change (Figs 2 & 3) are likely to be highly sought after for integration in ecological research (see Appendix 1 for further details on extracting these data). Furthermore, projections of industrial activity and energy generation could inform estimates of resource extraction and pollution from by-products, whilst proximity to forestry and roads could help determine the extent of hunting. Once extracted from IAMs and published, this suite of information is likely to be valuable for ecologists studying the future of the biosphere.
Selecting an IAM to use
Depending on the question being considered, the best approach would be to use projections from a range of IAMs for a scenario that closely matches the question to be addressed, since variation in the mathematical and computational approaches, as well as the interpretation of the scenario in each model, provides some measure of variability and uncertainty in the future state of the system. In reality, outputs may not be readily available from all IAMs for the particular scenario of interest, and there are also subtle variations in model formulation that might make particular models less applicable to a particular questions. For example, an IAM using a static urban land mask would not be useful if urban land area were a variable of interest.
Choosing which model to use is unfortunately inseparable from the choice of scenario at present, since there is currently only one RCP scenario projection for each model. Therefore, it is likely that a user's choice of model or models will be determined by the scenario(s) which is/are most appropriate.
Nevertheless, transparency is an important consideration for users aiming to understand IAM projections. We have collated an indication of the availability of documentation for each IAM broken down by subsystem (Tables S1–S4). This takes into account elements of the transparent and comprehensive ecological modelling (TRACE) documentation structure proposed by Schmolke et al. (2010), principally the description of the conceptual model and detailed description of the actual model.
Future Development of Integrsted Assessment Models: Two Suggestions to Increase Accessibility and Expand Applicability by Ecologists
IAMs have the potential to play an important role in assessing the future state of biodiversity, and in particular for the IPBES. Although the IPBES is in the process of deciding on work programmes, its remit is to provide decision-makers with information on relationships between biodiversity, ecosystem services and people, so that informed decisions about the adoption and implementation of adequate local, national and international policies can be made (IPBES, 2013). The IPBES therefore has a need to project biodiversity and ecosystem services into the future, and in order to achieve this it will require robust, credible and understandable projections of the major drivers of biodiversity change into the future (Pereira et al., 2010, 2013; Vohland et al., 2011).
IAMs in particular have the potential to be extremely valuable for modelling large-scale biosphere change by (1) providing futures of human development to drive models of ecosystems and biodiversity and (2) investigating the impacts of key policy decisions on ecosystem change, such as large-scale trade-offs between land sharing and land sparing, networks of protected areas or fisheries regulation.
In order to fulfil these roles IAMs need to develop in two ways: the first relates to enabling ecologists to understand the underlying mechanisms operating in the models, and the second to improving the relevance of these models for ecological projections by incorporating more ecology.
Differences in IAM projections
Within a particular scenario, we can distinguish two principal mechanisms giving rise to differences between IAM projections: the first is the uncertainty arising from the interpretation of the scenario storyline; the second arises from differences in model formulation. Ecologists employing IAM projections will need to understand the importance of these difference sources of uncertainty. For example, does variation arise because of differences in how ecological responses to drivers of change are modelled, or because of differences in the methodologies generating the projections of those drivers?
Model intercomparison – adopting a standardized experimental protocol that modelling groups can follow to generate comparable projections that can be used to systematically document differences in model outputs using quantitative methods (Bennett et al., 2013) – helps to identify the causes of differences (Lambert & Boer, 2001) and allows researchers from within and outside the community to scrutinize models (Randall et al., 2007). Model intercomparisons have been widely adopted by the Earth system modellers as a means to isolate and better understand model differences (Cess et al., 1989; Gates et al., 1998; Meehl et al., 2000).
If a set of IAMs were to make projections using identical detailed constrained experimental protocols – an intercomparison exercise – then the outputs would enable users to observe the differences arising from variability in formulation between models and at least begin to investigate the mechanisms giving rise to those differences. Furthermore, such results would demonstrate the degree of consensus or disparity between projections for the same scenario, so informing the degree of confidence in a particular outcome for a specific scenario.
To date no formal IAM intercomparison has been published. The collaborative exercise that gave rise to the RCPs considered here established a ‘benchmark’ simulation with a radiative forcing target to be achieved by 2150 for all IAM groups to follow, but this still allowed flexibility in the specific scenario assumptions made by each IAM group (Weyant et al., 2006).
The potentially large role that scenarios and projections may play in IPBES should be seen as an opportunity for integrated assessment modellers, as IPBES may provide a community and infrastructure to facilitate intercomparison exercises and to interpret the ecological implications of differences between IAM outputs. This would provide an interface between IAM modellers and ecologists, helping to ensure the latter do not duplicate IAM-like models that already exist.
Understanding the model components that are responsible for driving dynamics
Model intercomparisons only achieve part of what will be required to understand why IAM projections differ: such exercises are good at revealing how they differ but can be limited in revealing precisely why.
Understanding how existing processes influence model behaviour requires detailed studies of how the dynamical interactions between components influence the behaviour of models. Ecologists would be interested in understanding how ecological processes currently influence the dynamics of IAMs in general: do the existing processes exert any strong influences on the overall dynamics of IAMs or do they act more passively, being driven by the dynamics of other processes? Furthermore, ecologists would like to understand which processes are primarily responsible for driving the dynamics of variables of fundamental interest for predicting biodiversity futures, such as land-use change. Given that understanding, are the strengths of the existing dynamical couplings realistic? Once sufficient understanding has been achieved, ecologists might want to propose modifications either to increase the accuracy of the representation of ecological processes for IAMs or to obtain predictions of greater relevance to their studies.
Understanding these interactions will enable ecologists to assess the validity of the representations of existing ecological processes and propose modifications.
In order to better understand what is driving the dynamics of IAMs and how existing ecological processes affect that behaviour, it will be necessary to abstract from the full models to consider how the various model subsystems are coupled together and to determine whether differences in these structures could give rise to differences in model predictions. If we compare the IMAGE and GCAM models as an example, three differences can be identified in the coupling of the agricultural subsystem with other model subsystems. We present these differences below and suggest what the impact of these differences on model behaviour might be.
Temporal changes in crop yields in the GCAM are supplied exogenously, so the model would not show environmentally driven changes in yield. Depending on the exact model formulation and the exogenous yield assumptions, this could lead to mismatched yields and environmental conditions in the GCAM, for example the internally calculated surface temperature may increase substantially but this change would not feed back on agricultural yields. By contrast, yields are dynamically influenced by the internally calculated environmental conditions in IMAGE, hence agricultural yields and land-use types should track environmental conditions directly. If we assume that yields increase with increasing temperatures, then the area allocated to farming would be larger in GCAM as a result of this absent feedback compared with IMAGE, all other things being equal.
IMAGE models the intensification of agriculture by substituting capital and labour for land, whilst the GCAM does not dynamically model intensification of agriculture. Therefore, in conditions of high commodity prices and limited land availability, IMAGE should respond by increasing yields and sparing land through intensification, whilst GCAM might instead allocate more land to agriculture regardless of what is left. This difference might give rise to a lagged effect of increased commodity prices on land-use conversion in the IMAGE model compared to GCAM, all other things being equal.
Finally, the two models use difference equilibrium assumptions for calculating agricultural commodity prices; IMAGE uses a general equilibrium model and GCAM a partial equilibrium model formulation. This difference means changes in the energy/industrial markets affecting the price of these goods have an immediate effect on the income available for households to spend on food in the IMAGE model. This might not be the case for GCAM. Thus IMAGE should display immediate effects of changes in other components of the economy on demand for different food types, which has implications for the land area required to supply these different foods. On the other hand GCAM may demonstrate lagged land-use and land-area responses to energy/industrial developments, depending on the exact formulation of the model.
Model transparency and accessibility
For ecologists to interpret IAM projections and the uncertainties associated with them they need to have access to detailed descriptions of the models that gave rise to the projections. This paper attempts to give ecologists an insight into IAMs. However, we suggest that despite the substantial effort invested in IAMs, they are not easily interpretable by those outside the IAM community. There are some differences that appear perplexing and are not adequately explained. Firstly, urban land cover is projected to remain constant at 2005 levels in the IMAGE and GCAM models, an assumption of constancy that is not stated explicitly in the model documentation but is important for human effects, such as the conversion of agricultural to urban land around the extremities of settlements. Secondly, there is no discussion of how transport infrastructure, such as road networks, a strong driver of agricultural development (Rudel et al., 2009), is represented in each model nor of the implications for modelled agricultural development.
For IAMs to be viewed as more than black box models and to be incorporated within ecology requires that they be communicated in such a way that ecologists, and others, find useful. The criteria for inclusion of IAMs in the RCP development process for the IPCC include technical soundness and replicability, which are implicitly determined by publication in a scientific peer-reviewed journal (Moss et al., 2008). Here we have identified some gaps in the description of IAMs used for the RCPs (see for example Tables S2–S5), which could affect the assessment of technical soundness and prohibit replication of the results by others.
There are many proposed structures for documentation of models For example, the TRACE framework has been developed to facilitate transparent modelling for decision support, and although specifically designed for ecological models it is applicable to other models (Schmolke et al., 2010). TRACE proposes a structure that covers model description, model testing and analysis and model application.
Replicability of computational studies like IAM projections requires more than documentation: computational findings supported by descriptions alone can be considered not to be reproducible (Peng, 2011) without source code and input data. Source code is publicly available for the GCAM and IMAGE models only. In addition to source code, input data to drive the models need to be made available for all IAMs.
To maximize engagement with a broad scientific audience, we argue that models in all fields should ideally be described to users in all the following forms: words, mathematically using equations, algorithmically using pseudo-code and finally the computer code manifestation of the model. However, we propose a minimal set of requirements for IAMs and models more broadly would be a complete description of the model using words, equations or pseudo-code, combined with publication of the computer code manifestation of the model – the code being critical to replication of specific model simulations.
IAMs already include some ecological processes, and we have already identified that the outputs from IAMs could be a valuable tool for ecologists to use as inputs to explore biodiversity futures. However, we argue that this may be insufficient to accurately represent future environmental conditions, both to drive Earth system model projections and to represent a solid basis on which to build models of biodiversity and ecosystem services.
A representation of marine ecological components is of great importance for a realistic representation of human futures. Yet current state-of-the-art IAMs do not contain any representation of marine ecology. The interaction between capture fisheries and aquaculture, and terrestrial agriculture are fundamentally important – large-scale collapse of fisheries would results in significantly elevated pressure on agricultural land to supply the unmet demand for protein.
The ecology of marine organisms is also critical to the global carbon cycle (Falkowski et al., 1998; Falkowski, 2000), and projected changes from the oceanic uptake of CO2 are likely to affect many of these organisms (Orr et al., 2005; Beaufort et al., 2011), with consequences for both the marine food web and the role of the ocean as a carbon sink. This scale of abstraction would require a greater level of ecological processes to be incorporated than is currently represented in any of the IAMs.
On land, heterotrophs can be fundamental to the provision of ecosystem services such as pest and disease regulation and pollination (Kremen et al., 2004; Ricketts et al., 2004; Sekercioğlu et al., 2004; Nichols et al., 2008; Boyles et al., 2011; Estes et al., 2011). The loss or deterioration of these services through human impacts could significantly alter the economics of agriculture and affect human development. For example, a review of global crop data showed that 35% of the production volume is dependent on pollinators (Klein et al., 2007). Current IAMs neither represent heterotrophic organisms (with the exception of farmed livestock) nor their effects on the modelled subsystems.
We propose that incorporating further ecological processes into IAMs could improve the accuracy of projections of future environmental conditions used to drive Earth system models. Furthermore, we suggest that the inclusion of more ecological processes would provide more robust and consistent models on which to develop biodiversity/ecosystem services models for projecting biodiversity futures.
The outputs from IAMs can already be used by ecological modellers to investigate fundamentally important questions linked with human society, such as: the effect of human development and future land-use changes on the provision of ecosystem services such as pollination (Burkle et al., 2013; Tylianakis, 2013); interactions between demand for agricultural land and habitat conservation globally (Soares-filho et al., 2006); and the future emergence and transmission of zoonotic diseases following conversion of pristine habitats to agricultural production (Keesing et al., 2010).
We suggest that the most effective approach for starting to incorporate the necessary ecology into IAMs is for ecologists to use the current outputs from IAMs to study questions such as those we propose above. The findings and development challenges of these studies should then feed into the development of a next generation of IAMs, in which fundamental ecological feedbacks have been incorporated.
Here we provide some examples of current research that could contribute to expanding the scope of natural systems and level of ecological detail that is included in IAMs:
Bioeconomic fisheries models such as Ecoseed (Beattie et al., 2002), a component of the Ecopath with EcoSim model (Christensen & Walters, 2004), Atlantis (Fulton et al., 2011) and ISIS-Fish (Pelletier et al., 2009) represent the dynamics of fish populations, economically driven activities of fishing fleets and the management actions resulting from policies and regulations. These provide a potential starting point from which to develop simplified versions for application over greater spatial extents and inclusion in IAMs to capture fisheries and their economic interactions with agricultural production.
Models exist that link the effects of landscape structure on pollinators and the services they provide (Lonsdorf et al., 2009, 2011). Such models provide a useful starting point for incorporating at least some of the feedbacks to agriculture from human-driven landscape changes into IAMs.
More broadly, models such as InVEST (Kareiva et al., 2011) enable calculation of human impacts on the provision of ecosystem services, including hydrological flow regimes and sediment and nutrient retention, areas that would improve IAMs. IAMs also overlook the implications of land-use changes and agriculture for freshwater ecosystems and wetlands in particular, which in addition to important ecological systems, also represent significant components of the global carbon cycle – inland waters mineralize and bury an amount of carbon equivalent to 20% of that sequestered in terrestrial biomass and soils globally (Battin et al., 2009).
Finally, the community of stakeholders involved in the process of developing the next generation of IAMs could be brought together to agree a new set of objectives and generate a more inclusive, transparent and flexible modelling paradigm. Participatory modelling approaches are well suited for this, providing a common language for interaction and improving communication between formerly disconnected groups of stakeholders (Gaddis et al., 2010; Voinov & Bousquet, 2010). Establishing guidelines for model development, standards for communication between system components, integration of data across disciplines and techniques for evaluating model components and systems will enable interaction between parties and push forward the scientific development of integrated models (Laniak et al., 2013). In the spirit of participatory model development, ecologists are beginning to develop complex general ecosystem models, integrated models of the ecological world, in a way that aims to make them a resource for the community – the intention is that they are available to and accessible by as many ecologists as possible (Purves et al., 2013).
However, unlike ecosystem models, IAMs have their roots in a separate community from ecologists. Therefore accessibility, which we define in both the sense of naive model users – here ecologists – being able to employ IAMs, and integrated assessment modellers accessing or engaging with those users, imposes high demands on IAM modellers. The documentation and description of models will not necessarily be sufficient. Ecologists need to have a dialogue that opens this field of models to discussion, and points to areas of useful future interaction, through which integrated assessment modellers and ecologists can form a new collaborative community. User-centred research (McGuire et al., 2008), focusing on what resources ecologists use when investigating biodiversity futures, why and in what way, might be a pragmatic approach to shed light on the path to achieve wide accessibility and to build a collaborative community.
Conclusions: Integrated Assessment Models for Ecological Futures
IAMs are well developed, particularly with respect to the construction of emissions scenarios for use in climate change assessments (Nakicenovic et al., 1998, 2000; Morita et al., 2001). However, as interest in projecting the future state of the Earth is broadening (Reid et al., 2005), as new types of model are being conceived (Purves et al., 2013) and as ecologists are now being asked to project the future state of the natural biosphere (Pereira et al., 2010), there are huge opportunities for IAMs to play important roles in these endeavours. Working with future projections is likely to be central to the activities of the IPBES (Pereira et al., 2013), and therefore projections of human development are needed that are fit for the purpose of studying ecological futures.
IAMs combine the ecology of terrestrial vegetation with subsystem models of human energy use and food and agriculture, but lack important ecological components such as marine ecosystems and terrestrial heterotrophs. Despite this, IAMs are still useful to ecologists in providing futures of human development in the terrestrial realm, for exploring aspects of ecosystem and biodiversity futures, and the impacts of policy options on ecosystem change. As a result, IAMs have the potential to be a core element within the work of the IPBES assessing future states of biodiversity. Involvement within the IPBES should be seen as a great opportunity for integrated assessment modellers – IPBES may provide a community and infrastructure to facilitate interaction between the IAM community and ecologists. This interaction may be helpful when addressing two key areas of development that we conclude are needed for IAMs to take advantage of this opportunity.
First, we need a greater understanding of the structure and dynamics of IAMs. A first step towards this understanding will be to identify the differences between the projections made by different IAMs under the same scenario assumptions – in other words, intercomparison of IAMs. We also need to understand the processes that are driving the dynamics of IAMs, how these affect predictions of futures, and how the existing ecological processes drive the behaviour of the current models. A first approach would be to consider simplified abstractions of the full IAMs, and understand how the couplings of subsystem components may give rise to different model behaviours. To facilitate this process, greater model transparency will be essential and as we think about a new community of ecologists working with IAMs and their model developers, the demands for accessibility will move beyond traditional scientific documentation of the model. User-centred research that identifies what projections ecologists use to explore biodiversity futures, and how they use them, will be needed to shape the engagement needed to build an effective participatory modelling community.
Second, current ecological processes within IAMs may be insufficient for accurately representing environmental conditions both to drive Earth system model projections and to form a robust basis on which to build models of biodiversity and ecosystem services. Major ecological processes missing from IAMs include representation of marine ecology and any feedback from terrestrial ecosystems on human development. The implications of these deficiencies may be profound, for example investigations of climate mitigation (Rogelj et al., 2013) could be missing important feedbacks. Two avenues to begin addressing these deficiencies are simplification and extension of bio-economic fisheries models and models that capture the feedback from pollination services to agricultural productivity.
Ultimately, both the ecological community and the IAM community stand to benefit from greater interaction and collaboration. However, the way ecologists and integrated assessment modellers interact around IAMs needs to change for this to occur: IAMs should become a community resource available for development and be used for projecting critical ecological and environmental futures. Achieving this will be challenging and the demands for enabling accessibility for such a complex resource will be high. However, we believe that making IAMs more accessible within the multidisciplinary study of global change will enable greater resolution, reliability and fidelity for assessing how Earth's socio-ecological system is approaching planetary boundaries.
We gratefully acknowledge useful comments from three anonymous referees that improved the manuscript.
The research group of which the authors are members has the development of a general ecosystem model (http://www.madingley-model.org) as one of its primary objectives. Interest in IAMs was born of research into available projections of human impacts on ecosystems globally. Author contributions: M.H., T.N., D.P.T. and J.P.W.S. conceived the ideas. M.H. collected data, analysed data and led the writing. T.N., D.T, J.P.W.S., G.M. and M.S. reviewed and significantly enhanced the manuscript.