PerspectiveClimate Change

What Are Climate Models Missing?

See allHide authors and affiliations

Science  31 May 2013:
Vol. 340, Issue 6136, pp. 1053-1054
DOI: 10.1126/science.1237554

Fifty years ago, Joseph Smagorinsky published a landmark paper (1) describing numerical experiments using the primitive equations (a set of fluid equations that describe global atmospheric flows). In so doing, he introduced what later became known as a General Circulation Model (GCM). GCMs have come to provide a compelling framework for coupling the atmospheric circulation to a great variety of processes. Although early GCMs could only consider a small subset of these processes, it was widely appreciated that a more comprehensive treatment was necessary to adequately represent the drivers of the circulation. But how comprehensive this treatment must be was unclear and, as Smagorinsky realized (2), could only be determined through numerical experimentation. These types of experiments have since shown that an adequate description of basic processes like cloud formation, moist convection, and mixing is what climate models miss most.

From GCMs to Earth System Models

Smagorinsky's GCM was designed around the premise that studies of the general circulation required a model capable of resolving the heat transport from the equator to the poles. Its formulation was the next logical step in a program of hierarchical model development best known for its pioneering contributions to numerical weather prediction (3). The work paved the way for fundamental studies of the atmospheric general circulation, and hence Earth's climate.

Over the past half century, many of these studies have focused on the types of numerical experiments anticipated by Smagorinsky. Beginning with basic processes like moist convection and cloud formation, which have long been appreciated as central to the energetics of the troposphere, a long succession of processes and couplings have been added to primitive-equation descriptions of the atmospheric general circulation. In so doing, GCMs have gradually morphed into Global Climate Models, and with the more recent incorporation of models of the biosphere and the associated cycles of important chemical nutrients, Earth System Models (4, 5).

Wide variation.

The response patterns of clouds and precipitation to warming vary dramatically depending on the climate model, even in the simplest model configuration. Shown are changes in the radiative effects of clouds and in precipitation accompanying a uniform warming (4°C) predicted by four models from Phase 5 of the Coupled Model Intercomparison Project (CMIP5) for a water planet with prescribed surface temperatures.

Key Uncertainties

The increase in complexity has greatly expanded the scope of questions to which GCMs can be applied (5). Yet, it has had relatively little impact on key uncertainties that emerged in early studies with less comprehensive models (6). These uncertainties include the equilibrium climate sensitivity (that is, the global warming associated with a doubling of atmospheric carbon dioxide), arctic amplification of temperature changes, and regional precipitation responses. Rather than reducing biases stemming from an inadequate representation of basic processes, additional complexity has multiplied the ways in which these biases introduce uncertainties in climate simulations (7, 8).

For instance, a poor understanding of what controls the distribution of tropical precipitation over land, and hence vegetation dynamics, limits attempts to understand the carbon cycle (9). Similarly, uncertainties in arctic amplification of warming hinder predictions of permafrost melting and resultant changes in soil biogeochemistry.

Although the drive to complexity has not reduced key uncertainties, it has addressed Smagorinsky's question (2) as to what level of process detail is necessary to understand the general circulation. There is now ample evidence that an inadequate representation of clouds and moist convection, or more generally the coupling between atmospheric water and circulation, is the main limitation in current representations of the climate system.

That this limitation constitutes a major roadblock to progress in climate science can be illustrated by simple numerical experiments. In idealized simulations of a waterworld that neglect complex interactions among land surface, cryosphere, biosphere, and aerosol and chemical processes (see the figure), the key uncertainties associated with the response of clouds and precipitation to global warming are as large as they are in comprehensive Earth System Models (10).

Differences among the simulations in the figure are especially evident in the tropics, where the sign of cloud changes and the spatial structure of the precipitation response differ fundamentally between models. This diversity of responses arises because, at low latitudes, the coupling between water and circulation is disproportionately dependent on the representation of unresolved processes, such as moist convection and cloud formation (11, 12). The mid-latitudes show more robust responses because much of the energy transport is carried by baroclinic eddies; these, too, are fundamentally coupled to water, but they are much better described and resolved by modern GCMs, as foreseen by Smagorinsky (1).

The uncertain interplay between water and circulation that underlies differences in the response of the climate system to warming (see the figure) can be expressed in terms of more specific questions. For instance, how do marine boundary-layer clouds depend on their environment? Or how do atmospheric circulations couple to moist convection through surface and radiative fluxes? The first question ends up being key to explaining the intermodel spread in climate sensitivity (13, 14), the second to the pattern of the regional response to warming. Differences in regional responses also influence ocean circulations, and hence how oceans take up heat, as well as patterns of precipitation, and hence how the land biosphere takes up carbon.

Back to Basics

A deeper understanding and better representation of the coupling between water and circulation, rather than a more expansive representation of the Earth System, is thus necessary to reduce the uncertainty in estimates of the climate sensitivity and to guide adaptation to climate change at the regional level. This knowledge should help focus efforts and lead to progress in reducing the imprecision of climate models in the next 50 years. Here, Numerical Weather Prediction (NWP) provides a good example. By focusing on key limitations in the model initialization, spatial resolution, and the representation of key parameterized processes, NWP has improved forecast skill substantially over the past 30 years (15).

It is time to draw lessons from the era of experimentation that Smagorinsky launched half a century ago, and focus climate modeling efforts on advancing understanding and improving the numerical representations of how clouds, moist convection, and heating couple to the general circulation.

References and Notes

  1. In (1), Smagorinsky concluded: "In pursuing the objective to generalize theoretical models we must ask ourselves whether greater detail in formulating the contributing processes is warranted by truncation errors, by sensitivity of the results to detail, by the resulting increase in computational complexity and time, and by ignorance of the way these processes really work. Very often this cannot be determined in advance, but must wait for computational experiments to be performed.
  2. Acknowledgments: The simulations presented in the figure are taken from Phase 5 of the Coupled Model Intercomparison Project. We acknowledge the World Climate Research Programme's Working Group on Coupled Modelling, which is responsible for CMIP5, and thank the climate modeling groups for making available their model output. Additional support was provided through funding from the European Union Seventh Framework Programme (FP7/2007–2013) under grant agreement number 244067.

Navigate This Article