Introduction

Since the beginning of the industrial revolution, and maybe even earlier, the Earth has entered the so-called “Anthropocene” [1], an era in which human activities have started to affect profoundly the characteristics of the climate system. For example, the atmosphere’s composition has been modified by massive emissions of greenhouse gases (GHGs) of anthropogenic origin, such as carbon dioxide (CO2) and methane (CH4), which have caused global warming and consequent changes in a range of climate features (e.g. [2]). As another example, emissions of a wide variety of gaseous and particulate pollutants are affecting air quality, and similarly to the atmosphere, the ocean’s chemical composition is being altered by water pollutants along with liquid and solid waste, such as plastics (e.g. [3]). The surface of the Earth is extensively modified by agriculture, rapid urbanization and deforestation, which can modify regional climates. Finally, the land and ocean biosphere is modified by activities such as soil overuse, excessive fishing and hunting, forest management etc. [4].

Not only human activities are affecting the Earth system, but human societies are in turn strongly influenced by environmental stresses, and respond to them, possibly generating feedback mechanisms. An illustrative example is given by mitigation policies aimed at curbing GHG emissions (e.g. [4, 5]), which can be considered as a response to global warming, in this case providing a negative feedback mechanism: greater warming would (presumably) lead to the implementation of more effective mitigation policies, which would in turn reduce the warming itself. As a second example, climate and environmental stresses in specific vulnerable regions may induce massive migrations (e.g. [5]), and this would lead to changes in land use, GHG and pollution emissions, with consequent regional effects on climate. Clearly, in order to fully understand the evolution of twenty-first century climate under the influence of human activities, the human dimension has to be considered as an integral and interactive component of the climate system.

Despite this realization, although today’s climate system models (CSMs) have reached a high level of complexity, with the inclusion of interactive atmosphere, ocean, cryosphere, biosphere and chemosphere components, simulations of twenty-first century climate still consider the human component as an “external forcing”. In other words, the forcing due to human activities, e.g. the increase in GHG concentrations due to the use of fossil fuels or land use management and change, is prescribed as input to the climate models. In addition, the impacts of climate change to different socioeconomic sectors are calculated off line using impact models driven by the output of climate simulations. This approach obviously cannot account for possible feedbacks between the physical climate system and human responses. An exception is represented by the category of so-call “Integrated Assessment Models (or IAMS)” [6], in which some aspects of climate response to impacts are described interactively. In these models, however, the climate component is extremely simplified, and is represented using bulk variables, such as global temperature, and therefore their use for informing policy decision is relatively limited.

It is thus evident that a major need in climate modeling towards improving our understanding of the possible climate evolutions throughout the twenty-first century is the inclusion of an interactive human component in CSMs. This is indeed a formidable task which calls for an interdisciplinary approach going well beyond the state-of-the-art of today’s climate modeling, and will likely require a decadal modeling perspective. A few international programs have started to address this scientific challenge through the concept of “Digital Twins” of the Earth System [7], but they are still in their infancy.

Based on these considerations, the aim of this chapter is to present some considerations on this new frontier facing the climate modeling community. The chapter starts with a brief summary of the structure of today's CSMs and their application to twenty-first century climate change simulations. This information will serve as background for introducing the concept of the inclusion of interactive humans in climate models towards the development of “Populated Climate System Models” or “Pop-CSMs”.

The Basic Structure of Today’s CSMs

During the last 4 decades, CSMs have evolved from what were essentially atmospheric models to extremely complex systems including different components that fully interact with each other: atmosphere, oceans, cryosphere, land and marine biosphere, chemosphere. These models are three dimensional numerical representations of the basic equations that regulate the behaviour of their components and the physical processes at their interfaces. The models are numerically integrated in time to provide the evolution of the climate system using some of the most powerful supercomputers today available. As input, they require information such as topography, land-use distribution, atmosphere and ocean background composition, incoming solar radiation. The model resolution is determined essentially by the availability of computational resources, and the spatial resolution of most global models used for the latest generation simulations of historical and twenty-first century climate varies in the range of 50–100 km [8]. Physical processes that occur at scales smaller than the model resolution, for example cumulus convection, cloud microphysics, boundary layer turbulence and some radiative transfer processes, are typically “parameterized” in terms of resolved variables using modules based on the process physics understanding and on calibration against field observations.

Many impact applications, for example in hydrology or energy production, require climate information at regional to local scales that are not captured sufficiently well at the spatial and temporal resolutions of global CSMs. For this reason a number of “downscaling” techniques have been developed which use as input the coarse scale meteorological fields from global CSM simulations to produce fine scale climate information. These vary from the use of high resolution limited area regional climate models (RCMs, [9], to variable resolution global models [10] and a wide range of empirical-statistical downscaling techniques [11]. Current generation RCMs can reach spatial resolution of a few km (the so-called “convection-permitting” resolution), at which some processes parameterized in CSMs, such as cumulus convection, can be explicitly represented [12]. The various downscaling approaches have different advantages and limitations and their use depends on specific applications.

The performance of global CSMs in simulating the behaviour of the atmosphere has considerably improved over the years, with the increase in model comprehensiveness and resolution, to the point that present day models can reproduce reasonably well the basic features of the global atmosphere and ocean circulation, both in its climatological mean and basic modes of variability (e.g. [13]). The performance of RCMs and other downscaling techniques has similarly improved [14], so that the effect of local forcing, such as due to complex topography, coastline and land-use features, can also be well described. Despite these improvements, some key deficiencies are still there, particularly in the description of clouds, convection and precipitation, which are among the most difficult processes to simulate and are the main contributors to the different behaviours of the models [13]. More information on three dimensional climate models can be found in [15].

The Process of Producing Twenty-First Century Climate Change Projections and the Assessment of Related Uncertainties

The problem of climate “prediction” is very different from that of weather forecast. In the latter, the aim is to predict how the system will evolve given knowledge of its initial conditions. This is a deterministic, initial condition problem (also referred to as “prediction of the first kind”, [16], and due to the chaotic nature of the atmosphere there is a predictability limit of ~10–15 days, depending on specific weather patterns. Climate prediction, (or “prediction of the second kind” [16] is a boundary value problem, whose aim is to investigate how the climate system responds in a statistical sense and over long periods of time to changing external (boundary) forcings, e.g. solar radiation or GHG concentrations. In other words, climate change simulations can be considered not as predictions but as sensitivity experiments to changing forcings [17], and therefore the term “projection” (or “scenario”) is most often used instead of “prediction”.

An aspect that makes climate change simulation even more difficult is the unpredictability of the forcings themselves, since for example it is virtually impossible to predict socio-economic developments leading to given trajectories in GHG emissions. What can be done is to generate plausible hypotheses, or scenarios, of socio-economic development, and thus GHG emissions, input these scenarios into the climate models and assess how the system responds over long periods of time, typically order of a century (end of the twenty-first century). In other words, the question posed in climate projection is not to predict the actual future climate, but to characterize the full distribution of possible future climates under different forcing scenarios and their probability to occur. Thus climate prediction (or better “projection”) is not deterministic, but has a probabilistic nature.

The process of completing climate change projections for the twenty-first century and related impacts essentially follows a number of sequential steps: (1) develop a range of possible socio-economic scenarios –> (2) derive a range of GHG and aerosol emission scenarios (and in some cases land-use change scenarios) –> (3) derive GHG and aerosol concentrations –> (4) Input this information in climate models to simulate the global climate response (although some models have interactive aerosols or interactive carbon cycle) –> (5) downscale the global climate information to regional and local scales –> (6) use this information for assessments of impacts in support of the development of policy response options. For each step, typically an ensemble of models is used to estimate the full range of possibilities, since there is no single “perfect” model available and different processes in the models are represented in a number of ways. Climate models used for twenty-first century projections have been developed by a multitude of research groups worldwide, and this leads to possibly hundreds of projections available for users.

Each step of the projection procedure just described is affected by its own sources of uncertainty, which compound sequentially in a cascade process leading to an overall uncertainty range in possible future climate outcomes [17]. This uncertainty needs then to be fully characterized in order to provide robust information to relevant stakeholders. In this context, it is important to conceptually separate the full uncertainty range into a portion related to the intrinsic variability of the climate system and the external forcings, and one related to incomplete knowledge of processes and deficiencies in models and observations. The former needs to be fully characterized in a quantitative way, because, most often, extreme outcomes, although low probability ones, are most relevant for impacts. Within this context, different realizations with the same model using different initial conditions of the slow components of the climate system (e.g. the oceans) is necessary to sample the internal variability of the climate system. The latter portion of the uncertainty range is for example related to the existing wide spread in model responses to the same forcings, which is due to different and imperfect physics and dynamics representations in the models. This needs to be reduced through the improvement of models, observations and physics understanding.

The provision of climate change information, including assessment of uncertainties, is thus based on large multi-model ensembles carried out by several tens of laboratories worldwide under the auspices of large international programs such as the Climate Model Intercomparison Project (CMIP, [8] or the Coordinated Regional Downscaling Experiment (CORDEX), [18]. Therefore, the finite amount of available computing resources needs to be shared among three model needs: increase in resolution, increase in ensemble size (to better characterize uncertainties) and increase in model complexity. A continuous discussion across the modeling community has been ongoing on what is the priority among these three directions for a most effective improvement in climate projections. In the next section it will be argued that the latter one is clearly an important, albeit extremely challenging, direction which should have a high priority.

The Need and Challenge of Including an Interactive Human Component in Climate Models

It is by now recognized as unequivocal that human activities are modifying the climate system both globally and regionally [2] and that in turn, the resulting changes in climate characteristics are affecting a range of socioeconomic [4]. Under most plausible GHG emission scenarios, these interactions are due to strengthen in the next decades [2]. Therefore, building a digital twin of the climate system cannot neglect the representation of the mutual interactions across the natural and human systems.

We have already seen the example of mitigation as a negative feedback mechanism across these two systems and migration as a mechanism by which they interact at regional scales. Additional examples may be useful. Most current CSMs include dynamical vegetation modules mostly considering vegetation as “natural”, i.e. unmanaged. Yet, it can be argued that the fraction of total continental land that is truly unmanaged by humans is relatively small and mostly limited to remote areas. The assumption underlying present modeling of vegetation in climate models is thus, to say the least, of limited value. The implementation of pollution control measures is another example of a possible negative feedback across the natural and human systems. As we enter more ubiquitously into the Anthropocene, the two systems are destined to become increasingly intertwined, and therefore it is paramount that their interactions are represented in the next generation CSMs.

Today there is a range of population dynamics and socioeconomic models [5], so that the basic modeling frameworks for carrying out their coupling with CSMs is already available. Once the population response to environmental stresses is described, its feedback on the climate system occurs primarily through emissions of GHG, aerosols and other pollutants along with modifications of the earth's surface. This interaction should be represented in a distributed way on a common spatial grid across all model components. The key bottleneck in this approach is to disentangle the human response (or lack of) to environmental stresses from the response to other socio-economic factors. For example, population migration can have a multitude of causes, often deeply interconnected, which depend on specific socio-economic conditions. While conceptual models can certainly be constructed to address this issue, large field campaigns are necessary to provide sufficient data to build response models.

Given the dependence of this coupling exercise on specific environmental and socioeconomic settings, it may be useful to first carry out pilot studies over limited areas using RCMs as basic modeling systems [19]. One region that may be especially suitable as a pilot case is for example the Sahel, since agriculture and population dynamics are strongly dependent on climate variability [20] and in turn the region's climate features, e.g. the monsoon dynamics, are significantly affected by human forcings such as biomass burning, urbanization and deforestation [21, 22]. This is also a region were both accurate climate and population dynamics data may be scarce, so that innovative data production approaches are necessary. The extension of results to different climate and socioeconomic settings is not trivial, but at least from the methodological viewpoint such a pilot study may represent an extremely useful testbed, which would pave the way for other similar activities.

It can be argued that, of the three directions identified to improve the actionable value of climate change information, i.e. increased model resolution, increased ensemble size of model projections and increased model complexity, the first two are substantially, albeit certainly not exclusively, of technological nature. The third one, in particular concerning the natural-human system coupling, is an outstanding scientific challenge which will require truly interdisciplinary efforts and innovative modeling and field campaign approaches. Although the climate and socioeconomic modeling communities have increased their interactions within the realm of the climate change debate, they are still far from speaking a common scientific and methodological language. Training of a new generation of scientists lying at the interface between these two modeling communities is thus necessary.

As mentioned above, such coupling efforts will entail the production and analysis of very large datasets, and within this context mobile and internet technologies, along with machine learning techniques will play a central role, tying in with other research communities. In addition, the occurrence of natural geophysical disasters, such as earthquakes, tsunamis and volcanic eruptions, is another important element interacting with the climate system with strong socio-economic implications, and there are suggestions that global warming, through the induced changes in glacier mass, may actually affect the statistics of earthquake occurrence (e.g. [23]). Inclusion of this geophysical component in twenty-first century projections is therefore another promising and highly innovative area of further research.

In conclusion, the development of what we have called Pop-CSMs represents one of the main frontiers in climate, and more generally, Earth system modeling. It is also one of the research pillars of the development of Digital Twins of the Earth System, a major upcoming enterprise in climate system modeling which cannot prescind from modeling in an interactive way the role of humans in the climate system. These efforts will offer very challenging and exciting opportunities for innovative research, especially for a new breed of truly interdisciplinary scientists, and it will constitute a qualitative step forward towards a better understanding of the Anthropocene.