Studying and Projecting Climate Change with Earth System Models

A model organizes what we think we know about something in order to predict how it might behave in the present, future, or past as well as how it might respond to external influence. Models are especially useful when direct, controlled experiments are difficult or impossible. A model can be a simple concept, for instance: the heart is a circulating pump. A possible prediction from this model is that blood in the body regularly moves through the heart. Or a model can be a series of mathematical equations that are solved by computers. This second type of model can be built upon the first model by applying the physics of pumps to calculate how much blood cycles through the heart in a minute.

Humans are now conducting a long-term, uncontrolled climate experiment by emitting greenhouse gases into the Earth's atmosphere, converting forest into farmland, and otherwise changing the natural environment. Climate models solve mathematical equations that describe the physics of the atmosphere, ocean, and the land surface (left side of Figure 1) to help us understand how Earth's climate is changing. Even if humans had no influence on climate, climate models would help us understand and predict natural variations in the climate, like the El Niño Southern Oscillation (ENSO).

Climate, however, is not shaped by the physics of radiation and fluids alone. Biological and chemical processes impact climate. Both over the course of the year and over the course of Earth's history, growing plants and respiring microbes change atmospheric CO₂ levels and the albedo (reflectivity) of the Earth's surface. And biological processes are affected by climate. The circulation of the ocean controls the nutrients available to phytoplankton, while temperature and precipitation affect the metabolism of biological organisms on land. Studying how biological processes and climate are related requires a new type of climate model: the Earth system model (ESM).

ESMs include physical processes like those in other climate models but they can also simulate the interaction between the physical climate, the biosphere, and the chemical constituents of the atmosphere and ocean (right side of Figure 1). ESMs include processes, impacts, and complete feedback cycles; for example, they can simulate droughts as well as the resulting change in plant cover due to the drought, which may lead to more or less drought. They can even include the impact of human decision-making. ESMs never perfectly simulate the processes they include, but they are useful tools for extrapolating what we know about the present Earth system to the past and the future. There are several types of ESM, but we focus here on models of full complexity that simulate the atmosphere and ocean in three dimensions.

ESMs simulate physical processes much like other climate models and weather prediction models. ESMs are composed of model components that simulate individual parts of the climate system (such as the atmosphere, ocean, land, and sea ice) and the exchange of energy and mass between these parts.

The atmosphere is where most weather occurs and therefore where humans mostly experience the climate. The atmospheric model component of an ESM simulates the movement of mass and energy over large distances, as well as the nanometer-scale interactions between cloud droplets and water vapor. In order to capture this broad range of scales, full-complexity atmospheric models divide the atmosphere into thousands of grid-boxes and solve the fundamental equations of motion and energy conservation within each grid-box, which might be 50-200 km in size. There are many important processes (such as clouds, precipitation, and radiation) that are smaller than a grid-box and are simulated using "parameterizations." Parameterizations, which treat small-scale processes as a function of average large-scale properties, are the source of many of the uncertainties in climate projections.

The Earth's surface plays a crucial role in earth system modeling, not only because humans live at the land surface, but also because 2/3 of the sunlight absorbed by the Earth is absorbed at the land and sea surface. Almost all of this energy is eventually transferred to the atmosphere. How this transfer occurs is critical for climate. For instance, if solar energy is absorbed by dry soil, it simply heats the soil. If absorbed by wet soil, some energy may evaporate the water, depriving the land and living things of moisture, providing moisture to the atmosphere, and limiting heating of the soil. Therefore, the land model component simulates how water moves from its sources (where it rains or snows) to sinks such as the ocean or aquifers, a process simultaneously involving physical, biological, and anthropogenic aspects. The model can consider the effect of topography on drainage or how water use by plants affects soil moisture. Since land is fixed in space, most of the processes simulated on land occur within just one grid-box: only water and any energy or nutrients that water carries moves between grid-boxes.

Like the land, the ocean exchanges sensible and latent heat (the energy associated with the evaporation of water) with the atmosphere. However, the ocean's great depth and water's high heat capacity give the ocean an energy storage capacity about a thousand times greater than the atmosphere. In contrast to the land, the ocean can transport energy from the warm tropics to colder high latitudes. Thus, as with the atmosphere, ocean models simulate large-scale movement of mass and energy. In addition, small-scale processes, such as the sinking of cold, salty water near the poles, are parameterized within the ocean model.

The cryosphere, the part of the earth system that is ice, covers a small proportion of the Earth's surface but plays a large role in the climate system. Sea ice has an albedo about ten times that of ocean water. In addition, sea ice insulates the atmosphere from the ocean and affects the exchange of energy and mass between them. ESMs therefore simulate sea ice formation and loss. Snow on land is also simulated within the land model, including modifications to energy and mass fluxes, as well as albedo. Most ESMs do not directly simulate the growth and decay of ice sheets on land, but ice sheet model components are being developed to address the potential for ice sheet collapse in the future.

ESMs are chiefly distinguished from climate models by their ability to simulate the carbon cycle. If the sum of all CO₂ emitted into the atmosphere between 1966 and 2008 is compared with the observed level of atmospheric CO₂, approximately one of out of every two CO₂ molecules appears to be missing (Figure 2). This extra CO₂ has not vanished entirely. It has been incorporated into land and ocean reservoirs, often in carbon fixed by organisms during photosynthesis. Whether all of it will stay there and what proportion of future emissions will remain in the atmosphere are open questions, which have motivated the development of land model components that can predict the spatial distribution of vegetation, how its growth varies through the year, and the exchange of carbon between it and the soil. Similar model components exist to simulate the marine biosphere and chemistry.

ESMs are also distinguished by the sophistication of their atmospheric and oceanic chemistry. In the ocean, biological productivity is limited by the availability of nutrients, ranging from fundamental cellular model components such as N to trace nutrients such as Fe, which is a key ingredient in many enzymes. The distribution of these nutrients is controlled by transport and biology but also by reactions with dissolved organic matter and inorganic constituents. Nutrient cycling involving atmospheric transport is also important on land and can be affected by atmospheric chemical reactions. In addition, climate, as well as water and soil chemistry, can determine whether microbes decompose organic matter into methane or carbon dioxide. And once in the atmosphere, methane, a more efficient greenhouse gas, oxidizes to carbon dioxide, a less efficient greenhouse gas.

There is a Russian proverb, often quoted by both Vladimir Lenin and Ronald Reagan, that goes: "Trust, but verify." This proverb is not appropriate only for politicians, but also for scientists, who must trust that the natural world can be understood and predicted but must constantly verify their understanding to justify the credibility of future predictions. All models are wrong: they simplify complicated systems. ESMs are verified by comparing the model's behavior with the observed behavior of the climate system. Where and when the model is wrong can help identify what the model is simulating incorrectly or ignoring. What a model does well identifies what problems the model can be used to study.

Comparing an ESM with observations is not always straightforward. Observations are expensive to make and often limited in time and space. Even if they are available, they may have large uncertainties, or may be only indirectly related to what the model simulates. Fifty years ago, the vast majority of climate observations were of weather at the land surface in populated areas. Since that time, remote sensing of the Earth by instruments on artificial satellites and aircraft have expanded the type, spatial coverage, and frequency of climate observations. These observations now are crucial to testing models. Translating the amount of radiation that the satellite observes to the appropriate model variable, however, often requires another model, adding uncertainty to the observations. Thus, both in situ (directly measured) and remote sensing data are necessary for comparison with models.

Models are compared with observations at many different scales of time and space, and with as many types of observation as possible. Most model parameterizations are based on in situ experiments. Once included in an ESM, parameterizations are tuned to simulate current climate mean conditions as well as particular, well-observed events and their statistical distribution on longer timescales (e.g., El Niño-La Niña events or temperature changes since 1870). Different model variables such as temperature, the amount of radiation observed at the top-of-atmosphere, ocean heat content, and greenhouse gas concentrations are all open to evaluation. To be considered credible, the individual model parameterizations and the full ESM must perform well on all these tests.

Comparing the land carbon cycle model component of an ESM with observations could involve three types of test. First, a process, such as the response of organic matter decomposition by soil microbes to changes in temperature and soil moisture, can be measured at an individual site. Second, satellite data can be used to study larger spatial scales than can be measured directly. For example, leaf area predicted in the model can be compared with satellite-observed values to see how well the model simulates the annual cycle of growth. Third, the model can be compared to longer-term compilations of data to see how it performs on decadal time scales and over large spatial scales. Since satellite data is limited to the last few decades, these compilations often mix satellite data with in situ measurements and economic data (such as reports of lumber harvested from a nation's forests). The model's prediction of how much deforestation as well as reforestation has changed the carbon taken up by land over the past few decades can be compared to independent estimates based on atmospheric concentrations of carbon dioxide, carbon isotopes, and land-cover-change datasets.

Another test of an ESM is the simulation of past climates (paleoclimates). Projecting future climate change applies models tested against current conditions to new conditions and different concentrations of greenhouse gases and aerosols. One way to feel more secure about these projections is to simulate paleoclimates. Paleoclimate data are collected from rock, ice, marine sediment, and lake sediment cores, as well as from other data sources such as tree rings and corals. This information usually has lower temporal and spatial resolution and larger uncertainties than current climate observations and often includes the use of proxies, which means that a variable of interest (such as temperature) is not directly measured; a variable (the proxy) that correlates well with the variable of interest is measured Instead.

Glacial-interglacial fluctuations in proxy temperature, atmospheric CO₂ and CH₄ levels_, and dust concentration measured in ice cores closely parallel one another (Figure 3). Combining measurements like these can provide an independent estimate of the climate sensitivity: the change in temperature due to a change in the balance of incoming and outgoing radiation resulting from, for example, greenhouse gases, changes in incoming solar radiation, or aerosols from volcanic eruptions. Paleoclimate data also has identified extreme events in the past, providing a baseline for testing a model's ability to simulate the speed of climate change as well as its magnitude.

As we discussed above, one goal of ESMs is to find out where CO₂ from fossil fuel emissions would go in the Earth system. We focus on the land sink as an example of progress in this area. Additional carbon storage on land can be viewed as a competition between four processes: (1) vegetation growing more efficiently in more CO₂-rich air, thus fixing carbon; (2) climate change driving changes in precipitation, which drives changes in vegetation growth; (3) warming climate increasing microbial respiration of fixed carbon in soils; and (4) carbon fixation slowing as vegetation deplete soil nutrients like nitrogen. Even when they assume the same emissions from burning fossil fuels, different ESMs make projections of future atmospheric CO₂ levels that vary by more than 300 ppmv at 2100. To put this in context, so far humans have only changed atmospheric CO₂ levels by 100 ppmv.

ESM predictions differ for two reasons. First, models do not agree on the details of how climate, especially the distribution of precipitation, will change. Second, land carbon models are more or less sensitive to the four processes mentioned above. Further comparisons of these models with observational data will be necessary to reduce these uncertainties and suggest what improvements are needed.

The cyclical process of model development and comparison with observations may seem circular. It is more progressive than it appears. The changes in land carbon model sensitivity that resulted from adding a nitrogen cycle to simulate depletion of soil nutrients or including a strong response of microbial respiration to warming have motivated closer attention and more precise measurement of these processes. In these and other ways, ESMs have helped determine the answer to a critical question for us all: how the future of the Earth's climate will be shaped by the responses of its biosphere.