Although a wide range of Earth System models are appearing on the research landscape, three approaches can be highlighted: (i) models based on the extension of global climate models (i.e. General Circulation Models (GCMs)) extended to include more physical and biological processes as interacting components in the model; (ii) Earth System models of Intermediate Complexity (EMICs), which are constructed from the outset to include all critical biophysical processes in the appropriate balance; and (iii) integrated models that fully include the human dimensions and the biophysical dynamics of the Earth System in an interactive way, at any level of complexity.

Following the weather

Global climate models, which had their origins in numerical weather prediction models, are now being extended in step-wise fashion to evolve from atmosphere-dominated models towards more holistic Earth System models. An early major advance was the inclusion of ocean and sea-ice dynamics. The first aspects of atmospheric chemistry were then incorporated in the form of sulphate aerosols, which act to cool the Earth’s surface, in opposition to greenhouse gases. Current challenges in the further evolution of GCMs are to include an interactive carbon cycle, which will capture the potentially large feedbacks that would act to reinforce warming, and to add more complete atmospheric chemistry.

Earth System models of GCM origins are becoming increasingly adept at simulating the behaviour of the climate system, particularly with regard to aggregated global indicators like mean temperature. They are also improving significantly at simulating other aspects of the climate, including atmospheric dynamics, the overall behaviour of the hydrological cycle, and some aspects of storms and severe weather.

Components
Figure 1

Because of their origins, however, Earth System models based on GCMs have a more complete and detailed representation of the atmosphere compared to the other components of the Earth System, a bias that might be called “the tyranny of the atmosphere” (Figure 1). Although this atmospheric bias certainly makes sense in terms of numerical weather prediction, which operates on timescales of days and weeks, it is less appropriate for the longer time scales where “slow” processes in other parts of the Earth System become critically important for the functioning of the system as a whole. For example, up to now relatively less effort and resources have been committed to ice-sheet dynamics, the dynamics of terrestrial vegetation and the carbon cycle.

Intermediate complexity

The development of EMICs over the past decade or so has given Earth System science a complementary modelling tool to GCMs to explore the behaviour of the planetary environment. EMICs are based on a different philosophy, which entails (i) an equivalent level of detail in all important components of the Earth System; (ii) an emphasis on the interactions between the components as much as the processes within components; (iii) a trade-off between spatial resolution and the capability to simulate Earth System functioning over very long timescales. Thus, GCMs are the preferred models for simulating changes in climate over a century (or less) and at higher spatial resolution, whereas EMICs are more skilful at exploring long-term dynamics (e.g. at the scale of ice age dynamics).

In another way, EMICs appear to have an advantage over GCM-based Earth System models, and that is in their ability to simulate abrupt changes and threshold effects, features known from palaeo records to be important for Earth System functioning. For example, EMICs have been successful in simulating the rapid transition about 6000 years ago of the Sahel/Sahara savanna ecosystem into a desert, the abrupt warm events that punctuated the last ice age climate in the north Atlantic region, and the sudden cold spell that occurred when the Earth was warming rapidly at the end of the last ice age. In addition, EMICs have been able to simulate the potential shutdown of the north Atlantic thermohaline circulation – the “Gulf Stream” – as the climate warms; this is a threshold effect, with a small increase in atmospheric carbon dioxide concentration tipping the oceanic current into a new state.

Much of the success of EMICs in simulating some of the more complex dynamics of the Earth System is probably due to their emphasis on the interactions between the components of the system, and so their ability to capture complex system behaviour. However, neither EMICs nor GCMs have yet to tackle the most daunting challenge facing Earth System modelling – fully including the human enterprise as an interactive part of the system.

Just add humans

Although to many, the inclusion of human processes in Earth System models is a very new challenge, the origins of integrated human-biophysical models at the global scale go back at least to the early 1970s with the World2 and World3 models used in the Club of Rome’s “Limits to Growth” analysis. Integrated human-biophysical models have evolved considerably since the early days of those globally aggregated models, and a wide array of approaches is now being developed.

Much of the variety in contemporary approaches to integrated Earth System modelling lies in the human, or socio-economic, part of the Earth System model. A common approach is to link a global economic model with a GCM, both being quantitative in nature. More exploratory approaches to modelling the human dimensions include agent-based modelling (where humans are modelled as “agents” with rules governing behaviour in a given situation), network theory and modelling, game theory and simulation, and scenario analysis. This last approach is achieving significant popularity, as it allows narrative storylines to be developed about the future with more quantitative biophysical models providing internally consistent projections within the common analytical framework. The IPCC family of scenarios and the Millennium Ecosystem Assessment suite of scenarios are good examples of this approach.

Complexity
Figure 2

Whatever approach is used, modelling human dynamics in the context of the Earth System presents significant challenges that are yet to be solved:

Integration and balance. Large questions remain regarding the degree to which biophysical and human processes can and should be integrated, about the level of complexity in the two parts of the model and in the model as a whole, and the balance of emphasis between the two parts of the model. A recent survey of integrated Earth System models (Costanza et al. 2006 – Figure 2) highlights the characteristics of the current genre of integrated Earth System models from the perspective of integration level and balance. As seen in the figure, most models begin with a strong emphasis on either the biophysical or the human part of the Earth System and this carries through in the first generation, at least, of the integrated model.

Spatially explicit social data. Most social data is collected and reported by jurisdictions rather than in a spatially explicit way. This creates difficulties in parameterising and testing Earth System models, which are now all spatially explicit although resolution varies. Land-cover patterns are an exception to this, and methods are being developed to represent the drivers of land-use change in a spatially explicit way as well.

Temporal range. Most integrated ESMs are designed to run for the next 100  years and perhaps to be tested against observations for the past 100 years at most. This is an awkward timescale for social processes. The last 100 years is not nearly long enough to capture the rich array of human-environment relationships in the past, while projecting social dynamics for 100 years into the future greatly increases the probability of occurrence of “contingent events” – chance events that defy prediction but can change the course of history.

It is likely that a single approach to building integrated Earth System models will not be the optimum strategy. Rather comparing and synthesizing results from a range of different approaches would be a more robust plan. Alternatively, perhaps a number of individual modelling approaches can be integrated into a singly hybrid modelling framework. The next few years will see interesting times for Earth System modeling.