Modeling at the Global Scale

By Melanie Lenart | The University of Arizona | September 12, 2008

Climate models have made great strides since the turn of the millennium.

On one hand, it’s good news that people can trust the newer models more than previous versions. On the other hand, it means there’s even more reason to believe projections that the Earth’s average temperature will rise by 4 to 8 degrees Fahrenheit if carbon dioxide levels double compared to 18th century values.

Advances in the past decade have given researchers more confidence in coupled atmosphere-ocean Global Climate Models (coupled GCMs) typically used to make projections for temperature and precipitation.

• Coupled GCMs have improved with computers and increasing knowledge about climate patterns
• “Ensemble” runs averaging results from multiple models consistently yield better results than individual models
• Other models are improving our understanding of processes beyond temperature and precipitation

Coupled GCMs have improved

Several factors have combined to enhance today’s models compared to those of previous decades.¹

For one, an increase in computer capacity allows modern models to run at higher resolutions (Figure 1).

Many GCMs run at resolutions as fine as about 1 degree latitude by 1 degree longitude. This means grid squares average roughly 4,000 square miles—about eight times the size of metropolitan Phoenix—at the equator, and increasingly smaller toward the poles. Some “spectral” models have similar resolutions, but with grids that minimize size discrepancies between the equator and the poles.

The improvements have allowed researchers to produce coupled GCMs that do not rely on “flux corrections” to resolve problems. Earlier models required corrections of heat, momentum, and other fluxes to keep results from drifting over virtual time to unrealistic climate states. Most of the current batch of models can run on their own without these artificial adjustments.

More computer power leads to improved understanding of climate processes by speeding up computations, thus allowing for more testing of how modeling equations relate to actual climate data.

The higher resolution and improved understanding, in turn, has allowed for a more sophisticated modeling of climate factors and related parameters. That was the conclusion of a pair of researchers who systematically compared 57 models developed over the decades based on their performance on up to 14 different factors.²

Along with the results for temperature, precipitation, and sea surface temperature, Thomas Reichler and Junsu Kim considered how the models reproduced factors such as winds, specific humidity, the fractions of snow and sea ice, and sea salinity. The time frames tested depended on the available data.

Based on comparisons to climate patterns of the 20th century, coupled GCMs have consistently improved in their ability to capture average seasonal conditions. While there’s no guarantee that future climate will feature the same patterns as those of the historical past, such comparisons remain the best means of assessing a model’s skill.

Ensembles: Better than the sum of its parts

Researchers have found that the average of all available models—sometimes called the ensemble mean—almost always weigh in with more accuracy than any one model (Figure 2). Although the reason for this remains somewhat unclear, the finding recurs consistently through different model generations.

These findings from two assessments of GCM skill, both published in 2008, have given climate scientists renewed confidence in the 2007 projections by the Intergovernmental Panel on Climate Change (IPCC).

The IPCC generally used ensemble means when developing its projections that average annual temperature for the globe will rise by 4 to 8 degrees Fahrenheit if carbon dioxide rates double over its levels during pre-industrial times, the period before people began burning fossil fuels such as gas, coal, and oil. This greenhouse gas is expected to reach double its pre-industrial value sometime between the middle and end of the century; exactly when this occurs depends on several factors—most notably, whether society takes action to reduce its greenhouse gas emissions.

Incidentally, the 4- to 8-degree temperature range has made little advance in precision over the years, despite increasing confidence among scientists in its accuracy. In large part, the wide range relates to uncertainties over the future of clouds and other factors that could dampen or exacerbate global warming.¹

Other types of models

The confidence applies to the results of coupled GCMs, and comes from extensive comparisons with actual climate conditions captured in historic records. Many other kinds of models exist in addition to the coupled GCMs, but few of them have been tested rigorously against historic conditions. Scientists thus have less confidence in their projections than they do in projections from coupled GCMs.

Given their complexity, coupled GCMs are rarely linked directly to models assessing factors beyond the atmosphere’s interaction with oceans, sea ice, and land surfaces over time and space. Scientists employ less complex models when considering the potential impact of climate changes on vegetation dynamics, the carbon cycle, and ocean chemistry.

“Simple models” are based on a combination of coupled GCMs, expert judgments, and equations drawn from observations.

Earth System Models of Intermediate Complexity, known as EMICs, include many of the same processes simulated in GCMs, but they operate at a coarser resolution than the typical modern GCM.

EMICs also rely more heavily on specified equations—or parameterizations—than GCMs do. While GCMs must resort to parameterizations for some factors, such as thunderstorms and other processes occurring at scales smaller than a grid cell, they derive their results largely based on dynamic processes.

Because the resolution of EMICs falls well below that of modern coupled GCMs, results from the intermediate models apply only at the large scales of the globe or a continent.³ To get results down to the scale of a region smaller than a continent, scientists often apply downscaling techniques to coupled GCMs.