Comparing suites of models, not just with observations but also with one another, can illuminate areas of uncertainty within individual models as well as more broadly, across multiple simulations.
Earth System Models (ESMs) numerically model the atmosphere, oceans, land and sea ice, and have biogeochemical components (for example, dynamic global vegetation models) to study the carbon cycle. Earth System Models of Intermediate Complexity (EMICs) use simpler physics or coarser resolution [Flato et al., 2013, Flato, 2011] to focus on selected processes.
One major contributor to global sea-level rise, the transfer of land ice mass to oceans via melting, remains difficult to model. ESMs with ice-sheet coupling capability do exist, but were not used in the most recent CMIP series, known as CMIP 5.
Since the ocean absorbs heat and carbon dioxide from the atmosphere, the ocean component of ESMs is essential to form a complete climate picture. The ESMs used in this latest series, however, include ocean components with too coarse a resolution (on the order of 100 kilometers) to resolve “mesoscale” eddies—the whirling ocean equivalent of atmospheric storms. These features, occurring on 50 to 200 kilometer length scales, carry heat and nutrients. Such eddies, for example, are responsible for much of the meridional (north-south) heat transport across the Antarctic circumpolar current. The features are modeled with conceptual approximations, or “parameterized,” in current ESMs, which is less accurate than resolving them directly with mathematical representations. Nonetheless, the ocean model components of the ESMs do a reasonable job of reproducing observed sea surface temperature (SST), although with a cold bias, and with smaller errors in the tropical regions. These models’ “mean dynamic topography”—the height of the ocean surface above a horizontal surface at rest—correlate at the 95 percent level with observations. Another feature of ocean models is the amount of heat they gain from their numerically represented atmospheres. Older ESMs (labeled CMIP 3) that failed to include volcanic eruptions absorbed too much heat. But the ocean heat uptake of current CMIP 5 models, which include volcanic injections into the atmosphere, is more in line with observations—or as Flato et al (2013) state, “There is high confidence that many CMIP5 models reproduce the observed increase in ocean heat content since 1960.”
Adding volcanic effects to the modeling, however, could introduce a negative bias in thermosteric sea level rise using the standard atmospheric-ocean general circulation models [Gregory et al., 2013]. Because these were used in the CMIP 5 series, Church et al. [Church et al., 2013] recommended adding 0.1 millimeter per year to the models’ mean rate of sea-level rise due to thermal expansion between 1971-2010. That brought the modeled rate close to the observed rate, though with a statistically non-significant excess for 1993-2010 [Flato et al., 2013].
Ice sheet modeling is critical to any projection of future sea level rise, but remains in its early stages. Still, recent attempts to model surface mass balance of the Greenland and Antarctic ice sheets during the latter half of the 20th century agree well with observations—an important step in validating projections of future changes in mass balance and their sea level equivalent [Shepherd et al., 2012]. Such projections are typically derived from regional climate models or downscaled atmosphere-ocean general circulation models [Flato et al., 2013].
The Ice Sheet System Model (ISSM), an impressive tool developed in recent years, integrates data from a variety of sources to simulate the higher-order physics of ice flow, and to develop mass-balance projections for the Greenland and Antarctic ice sheets [Larour et al., 2012]. This high-resolution model, which includes three-dimensional capability, represents an improvement over previous “hybrid” models (these combine Shallow Ice Approximation and Shallow Shelf Approximation simulations with simplified ice-flow mechanics, but do not always capture ice sheets and ice shelves realistically). The ISSM was shown to model ice flow surface velocities in good agreement with observations for Greenland [Larour et al., 2012], and has been used to improve Antarctic ice-shelf velocity modeling [Larour et al., 2014].
For the Greenland ice sheet, a variety of models show no trend of statistical significance from the 1960s to the 1980s, then a significant trend toward an increasing contribution to sea-level rise beginning in the 1990s [Flato et al., 2013]. By 2011, Rignot et al. were using the Regional Atmospheric Climate Model (RACMO2) to show that changes in surface mass balance could account for roughly 60 percent of loss in ice mass since 1992 [Rignot et al., 2011, Flato et al., 2013].
Antarctic simulations come with greater uncertainty, inspiring among the authors of the latest IPCC report only medium confidence in modeling results for the late 20th century, including a RACMO2 estimate of a negative contribution to global sea level rise: minus 5.5 millimeters per year, plus or minus 0.3, between 1979 and 2000 [Lenaerts et al., 2012].
A recently published evaluation of Antarctic climate modeling by 20 experts [Bracegirdle et al., 2015] included a number of recommended improvements:
- Elimination of large biases in positioning of the southern hemisphere mid-latitude tropospheric jet, which drives westerly winds, in CMIP 5 models, to better capture the effects of stratospheric ozone depletion and recovery.
- Better modeling of clouds over the southern ocean.
- Evaluation of energy fluxes over Antarctica in climate models, with the aim of developing a proper atmospheric energy budget.
- Using reconstructions of paleoclimate to create more effective simulations of the relative warming rate over the southern continent, known as “polar amplification.”
- Reliably simulating connections between the tropical Pacific and Antarctica, especially the climatic effects of an expected eastward shift in southerly directed Rossby waves—a shift projected by most models.
The evaluation also recommended improved modeling of Southern Ocean circulation.
Modeling land hydrology
Modelers in recent years have worked to simulate the wide variety of land-surface conditions influencing evapotranspiration and runoff, both key factors in broader climate modeling. And while their efforts have met with some success, challenges remain in accurately predicting annual variations in river discharges, matching soil-moisture trend simulations to observations, and modeling the effects of temperature change on tropical and boreal ecosystems, as well as other land-surface changes [Flato et al., 2013].
A variety of specific land-surface models, for example, were used in one study as inputs for a broader model meant to simulate discharge from 30 of the world’s largest rivers [Materia et al., 2010]. The authors reported strong simulations for two of the better developed land surface schemes used as input, though they said simpler schemes were hampered by multiple biases; the broader model, called a hydrology discharge routing model, also showed significant limitations in estimating mean annual river discharge.
The CMIP 5 models, meanwhile, had difficulty matching up with observations of soil moisture and its effects on precipitation in the tropics and in the Sahel [Flato et al., 2013, Williams et al., 2012,Taylor et al., 2012].
The latest IPCC report commends recent efforts to include agriculture, urbanization and deforestation in Earth System Models, and the effects of such land-use changes on carbon dioxide emissions [Flato et al., 2013]. But again, increasing model complexity led to increasing variation in results among the models [de Noblet-Ducoudre et al., 2012].