全球变化知识资源中心

Understanding the links between land-use change (LUC) and climate change is vital in developing effective land-based climate mitigation policies and adaptation measures. Although mitigation and adaptation are human-mediated processes, current global-scale modelling tools do not account for societal learning and other human responses to environmental change. We propose the agent functional type (AFT) method to advance the representation of these processes, by combining socio-economics (agent-based modelling) with natural sciences (dynamic global vegetation models). Initial AFT-based simulations show the emergence of realistic LUC patterns that reflect known LUC processes, demonstrating the potential of the method to enhance our understanding of the role of people in the Earth system.

In a world that faces continued population growth and changing consumption patterns while striving to achieve an equitable and acceptable level of human well-being, climate change and land-use change (LUC) are two of the foremost environmental challenges. They are also inseparably linked: land-use and land-cover change contribute to climate change by affecting ecosystem biogeochemical and biophysical processes^{1, 2}, and the climate shapes the way people use land by affecting food supply and pollution impacts on ecosystems^{3, 4, 5}. Nearly half of today's ice-free land surface has been converted from natural ecosystems into cropland and pastures⁶. Since around 1850, LUC resulted in an estimated release of more than 150 Pg C into the atmosphere — one third of the approximate total anthropogenic carbon emissions — and contributed 10–20% of CO₂ emissions during the late twentieth and early twenty-first centuries^{1, 7}. Most of the observed increase in atmospheric N₂O over the same period has been attributed to emissions from agricultural fertiliser use⁸. LUC-related climate forcing also occurs at the regional scale, either with a cooling or warming effect^{2, 9}, arising from changes to biogeophysical processes at the land surface that control the mixing of the near-surface air, and the surface radiation and energy balances¹⁰.

LUC will continue to contribute substantially to climate change in the future. A number of climate-change mitigation policies recognize the climate-regulating services of terrestrial ecosystems, which can be implemented through LUC^{11, 12, 13}. But despite the recognized need for a better understanding of the LUC–climate interplay, LUC is still poorly represented in the current generation of global circulation models (GCMs), which limits evaluations of the sensitivity of the climate system to LUC¹⁴. Moreover, the potentially adverse effects of climate change mitigation arising from indirect land-use change are largely ignored^{15, 16, 17}.

People will need to adapt land-use practices in response to climate change impacts, particularly in regions where climate change has been shown to be a threat to crop and pasture yields and water supply. Examples from history demonstrate that considerable economic and societal decline, even collapses of entire civilizations, can occur because of periods of unprecedented and persistent drought^{18, 19}. Conversely, examples of past successful responses to climate change exist through migration or the adoption of new models of sustenance^{18, 20}. Although changing supplies of natural resources combined with rapid rates of climate change certainly exert pressure on societies, it seems unlikely that a single driver (that is, climate change) is the sole cause of instability in socio-ecological systems, with their mix of vulnerable, but also stabilizing, facets.

Whether or not the adaptive capacity of today's societal actors is sufficient, globally, to withstand the impacts of projected climate change over the twenty-first century and beyond is a matter of debate²¹. Land-based mitigation or adaptation options at a certain locale may cause changes elsewhere with opposing effects^{16, 21}. Adaptive actions that don't seem promising in the short-term might become important for adaptation over a longer time horizon, whereas others may be increasingly ineffective when longer periods of time are considered. Thus a broad temporal and spatial perspective is needed when assessing adaptation and mitigation responses to climate change.

Adaptation and mitigation are processes. However, current attempts to represent adaptation and mitigation in climate change assessments have focused on top-down statistical indicators of the capacity to adapt²², or the capacity to mitigate²³, as proxies for these processes. It is axiomatic that statistical approaches are only valid within their calibration range and are therefore limited in their applicability under changing conditions beyond this range²⁴. Most importantly, current state-of-the-art modelling tools are unable to represent human agency, which underpins individual behaviour, decision making and adaptive learning and hence is important for understanding how societies will respond to challenges such as climate and other environmental changes.

Integrated assessment models (IAMs), often combined with computable general equilibrium (CGE) models, are the most commonly applied tools for creating projections of global LUC²⁵. These models combine representations of micro- and macro-economic theory with social and natural system constraints, and are widely used to project development pathways in climate change assessments^{14, 26, 27}. IAMs and CGEs have acknowledged strengths in providing comprehensive cross-sectoral analyses, and are an important component of a common scenario framework that bridges climate research communities²⁸. State-of-the-art IAMs analyse project changes in food exports or imports in response to market liberalisation, by considering environmental aspects²⁹. IAMs can provide estimates of the impacts of biofuel policies on LUC³⁰, or assess how changes in diet may affect agricultural greenhouse gas emissions³¹. But, such comprehensive cross-sectoral approaches come at the expense of simplifying the heterogeneity of human agency and human socio-cultural attributes. Assuming that the extant structural and functional relationships between people and their environment remain static, the capacity to explore adaptive learning across future scenarios is limited³². Models that are based on the principles of homogenous, utility-optimizing decision-making under equilibrium conditions¹⁴ tend to generate spatial patterns of land-use that conform to the underlying patterns of natural resources (see example in Box 1). They also generate outcomes that are very different when compared with models that have been developed and calibrated at regional scales³³.

Box 1: Simulating land-use change in a hypothetical region using AFTs.

The results from the example simulation shown in Fig. 3 are for a hypothetical region based on three farmer AFTs (high, medium and low intensity) and one conservationist AFT, which compete for capital resources that supply ecosystem services (simplified to 'food' and 'nature'). Conservationists only supply nature services. Farmers supply food and also provide nature services, but at a lower level than conservationists, and these increase from high- to low-intensity farmers. Therefore, the four AFTs are characterized by the relative level at which they supply each service, and by behavioural thresholds of resistance in response to stress and sensitivity to competition. Examples of behavioural parameterizations of AFTs are discussed in ref. 81.

The region is divided into 3,600 grid cells, each with a unique combination of capital attributes, which are limited here to natural and financial capital. Natural capital represents the provision of food for nature and is maximized in the bottom-right of the region, as indicated on the capital gradient map (Fig. 3a), whereas financial capital is maximized in the top-right. The resulting modelled land-use map, when a global demand for food and nature is applied uniformly across the whole area (Fig. 3c), reflects the gradients that are assumed in the distribution of capitals giving the optimal distribution of land-use based on resource (capital) availability. This type of outcome would be generated by utility-optimizing approaches or models that allocate land-use based on land suitability. By contrast, a quite different pattern emerges when the global food demand is partitioned spatially by dividing the demand equally between four sub-regions (Fig. 3d), but with the capital gradients across the whole area remaining unchanged. In this case, sub-region 1, is unable to meet its food demand due to low financial capital (Fig. 3b), and hence nearly the entire area is farmed, although only low-intensity AFTs can be sustained (Fig. 3d). As the high-productivity sub-regions 2 and 4 can meet the sub-regional demands, some grid cells are not needed for food or for nature supply and are abandoned (unmanaged). In sub-region 3, with higher financial than natural capital levels, food is relatively easily produced, so agents that produce nature have a slight advantage, since their unmet demand is greater. However, in this situation, there is no surplus land.

These example results demonstrate the basic functionality of the AFT concept, which could be applied to the global scale if parameterized with real data about location attributes and agent decision making. In the given example, society consumes (demands) a fixed amount of food and nature services. In principle, it would also be useful to model consumer trends with a similar agent-based approach that could draw on market-based agent-profiling (with the caveat that not all relevant information to achieve this would be easily accessible⁵⁰).

We propose a novel concept for developing LUC models that could overcome the difficulties outlined above. We focus on LUC because of its impact on climate change at the global and regional scales, the variety of land-based mitigation policies and the clear need for land-use adaptation to climate change. The concept is, however, sufficiently generic to be adopted in addressing other questions and interactions within broader socio-ecological systems. We argue for a new generation of global LUC models that are explicit about the role of human behaviour and decision making; models that can be linked to terrestrial ecosystem models to advance our understanding of the human–land system and its sustainable use in a changing world. Current ABM approaches that are applied at the local-scale are not practical for global-scale applications. Owing to rapidly increasing computing power, a model that mimics several billion individual actors might be technically feasible¹⁴ but properly parameterizing the attributes of billions of individuals is not possible in the absence of global socio-cultural data⁴⁹. This implies the need for a more limited set of generic agent types⁴⁹ that will also allow models to be applicable to a wide range of questions over long timescales.

The plant functional type (PFT) concept applied in dynamic global vegetation models (DGVMs) is used here as a template for developing typologies that operate at large spatial scales^{48, 49}. The basic principles that define PFTs are well grounded in fundamental ecology, plant physiology and biogeography (Fig. 1). Hence theory, rather than empiricism, is used as the starting point. In contrast to how large-scale typologies are created in, for example, marketing⁵⁰, the derivation of PFTs is much more transparent. Although concepts of theoretically grounded agents have been proposed before for the analysis of socio-ecological or economic systems, none of these concepts are explicitly for global-scale applications^{39, 49, 50, 51, 52}. The PFT approach is one of the few (perhaps the only) examples of the successful scaling-up of individuals (here plant species) to create global models. Thus, it is reasonable to learn as much as possible from this experience.

The realized niche is differentiated from the fundamental niche because it reflects interactions with environmental filters and other plants, modifying the relative abundance of a species within an area or within varying developmental stages of the ecosystem (for example, over time). Vegetation dynamics are represented through a limited number of plant functional types (PFTs) that group species with similar characteristics, growing in ecosystems of a similar type, even though these might be found in geographically very different locations (illustrated by the similar performance curves for species found in n environmental spaces, top centre). The biogeography and growth-components of a PFT are combined with process-based algorithms for plant and soil carbon, water, energy and nitrogen cycling (bottom left; see also Table 1). At a given location, a mix of PFTs interacts with the atmosphere and soil (and, more recently, humans). This mix can change in response to the ageing of the ecosystem, disturbances and environmental trends (bottom right). Typical outputs of DGVMs are carbon and water fluxes: net ecosystem exchange (NEE), gross primary production (GPP), net primary production (NPP), respiration (R_E) and evapotranspiration (E_T).

An assemblage of observable properties (traits) can be linked to plant biophysical and biogeochemical mechanisms that enable different species to cope with similar types of environment and/or competition, even when these are encountered in geographically very distant locations^{53, 54, 55}. DGVMs take advantage of this feature by coining functional units, PFTs, which can be thought of as representing groups of species with a similar expression of multiple traits in response to their environment^{53, 54, 55} (Fig. 1 and Table 1). Current DGVMs aim to represent the performance of plant species, and model the dynamics of plant–environment interactions, by combining climatic limits to growth with a strong footing in ecological theory and physiological mechanisms.

DGVMs typically define around 5–15 PFTs that embody the enormous variety of the Earth's plant species by collapsing diversity into the most general strategies to cope with variable sets of conditions. A universally agreed PFT scheme for global models does not exist^{53, 54, 55}, but by using a limited number of PFTs, DGVMs have been shown to adequately predict the formation and reformation of biomes in response to changing environments, and successfully reproduce patterns of terrestrial carbon and water fluxes^{53, 56}. Thus far, most DGVM applications have not explicitly accounted for human intervention in natural ecosystems, and their treatment of agricultural and forest management processes is immature. Different approaches are currently being explored^{57, 58} and further development of 'land-use enabled' DGVMs will facilitate the coupling of terrestrial ecosystem processes with the dynamics of human land-use systems. Eventually, such coupled models could be used to provide the scientific basis to assess trade-offs between immediate human requirements from ecosystems and the need to preserve the capacity of the terrestrial biota to supply these ecosystem services over the long term^URL: