The Role of Ocean-Atmosphere Interactions in Tropical Cooling During the Last Glacial Maximum

See allHide authors and affiliations

Science  27 Feb 1998:
Vol. 279, Issue 5355, pp. 1341-1344
DOI: 10.1126/science.279.5355.1341


A simulation with a coupled atmosphere-ocean general circulation model configured for the Last Glacial Maximum delivered a tropical climate that is much cooler than that produced by atmosphere-only models. The main reason is a decrease in tropical sea surface temperatures, up to 6°C in the western tropical Pacific, which occurs because of two processes. The trade winds induce equatorial upwelling and zonal advection of cold water that further intensify the trade winds, and an exchange of water occurs between the tropical and extratropical Pacific in which the poleward surface flow is balanced by equatorward flow of cold water in the thermocline. Simulated tropical temperature depressions are of the same magnitude as those that have been proposed from recent proxy data.

Although the CLIMAP project (Climate: Long-Range Investigation, Mapping, and Prediction) (1) estimated that, during the Last Glacial Maximum (LGM), tropical temperatures were only modestly lower than they are today, a growing body of evidence, including corals from Barbados (2), ground water from Brazil (3), inferred snow line depressions (4), and ice cores from Peru (5), indicates that the tropics were considerably cooler. There nonetheless continues to be a debate about the extent of this cooling over the low-latitude oceans (rather than land) because some oceanic data are consistent with the CLIMAP results (6).

Simulations of LGM conditions by means of atmospheric general circulation models reproduce tropical sea surface temperatures (SSTs) that are generally consistent with the CLIMAP results. However, these models neglect oceanic motion that can affect SST. In some of the simulations (4, 7, 8), the SST is specified; in others (9, 10), the ocean is allowed only vertical mixing. Such approximations neglect oceanic upwelling, horizontal advection, and the dynamical ocean-atmosphere interactions that profoundly affect SST and that are of central importance in the following three processes: (i) Tropical ocean-atmosphere interactions of the type that causes the interannual Southern Oscillation between complementary El Niño and La Niña states (11). These interactions involve a positive feedback between the SST and the winds that causes both to amplify or attenuate. Eolian deposits and analyses of planktonic foraminifera (12-14), which indicate that the low SST during the LGM was associated with an intensification of the trade winds and an increase in the slope of the thermocline, are consistent with these interactions. The type of ocean-atmosphere interactions that result in El Niño involves an adiabatic, east-west redistribution of warm surface waters in the equatorial Pacific that leaves the structure of the mean thermocline unchanged and has a time scale of a few years at most (15). (ii) Processes that can affect the thermocline itself, on decadal time scales. These processes include a shallow, wind-driven, tropical-extratropical, oceanic exchange in which poleward flow in the surface layers subducts in the subtropics, returns to low latitudes in the thermocline, and wells up at the equator (16). A decrease in SST of the extratropics in due course lowers the temperature of the water that subducts in the subtropics and subsequently alters the properties of the equatorial thermocline, thus influencing ocean-atmosphere interactions in low latitudes. (iii) The thermal structure of the deep ocean. This structure, on time scales of centuries and longer, depends on the thermohaline circulation that involves the sinking of cold, saline waters in certain high latitudes. The calculations to be described here simulate LGM conditions for a few decades at most so that processes (i) and (ii) are included but process (iii) is excluded. The thermal structure of the upper ocean, which depends on processes (i) and (ii), adjusts to LGM conditions, but that of the deep ocean remains close to the initially prescribed modern conditions.

We developed a fully coupled general circulation model of the atmosphere and ocean (17) in which the oceanic component permits motion in all three directions and the coupling scheme permits dynamical ocean-atmosphere interactions. The model domain is global and has, in the spectral atmospheric model, an equivalent spatial resolution of 2.25° in latitude by 3.75° in longitude. In the grid-point ocean model, the resolution is 2.00° in latitude by 3.62° in longitude. There are 14 and 15 vertical levels in the atmospheric and oceanic models, respectively. To simulate the LGM, we specified the following quantities: orbital parameters appropriate to 21,000 years before present to govern seasonal insolation, 200 parts per million atmospheric CO2, ice sheet topography (18), glacial albedo (1), and glacial sea level (19). The time integration proceeds asynchronously at 1-day intervals; at the end of each day, the daily mean values of SST and surface currents are passed as boundary conditions to the atmospheric model, and those of the wind stress, net heat flux, net freshwater flux, and net shortwave radiation are passed to the oceanic model.

Before proceeding to the LGM simulation, we performed a 40-year integration for the present-day climate, starting from Levitus's data (20) for the ocean, without applying any flux corrections to the model. The simulated climatology is reasonable except that the predicted amount of stratus cloud, as parameterized in (21), is too low in the eastern equatorial Pacific, so SST is higher than for the observed climatology in this region (Fig.1A). In the LGM simulation, which also starts from Levitus's data, the initial, rapid atmospheric adjustments (within weeks) to the glacial forcing (including lower atmospheric CO2 concentrations) are a decrease in temperature, substantial changes in the direction and speed of the midlatitude westerly jets, and intensification of the trade winds. Up to this stage, the simulation is in accord with earlier simulations (8, 9), but the adjustment in our model then continues in two more steps (Fig. 2): first, during the first year, a sharp drop in equatorial temperatures and a further intensification of the trade winds, all part of process (i), and, next, a gradual decrease in temperatures over a period of several years in response to process (ii), which allows extratropical surface conditions to affect the equatorial thermocline and hence the ocean-atmosphere interactions.

Figure 1

(A) Annual mean SST (in degrees Celsius) in the present-day simulation (averaged over the past 23 years; the contour interval is 1°C for temperatures above 26°C and 2°C for temperatures below 26°C). (B) The SST difference (LGM, averaged over the past 5 years, minus present day). The contour interval is 1°C with odd values contoured with short dashes and even values contoured with a solid line. The −2°C isotherm is contoured with long dashes for clarity. EQ, equator.

Figure 2

Time series of SST on the equator averaged between 180°E and 140°W for the LGM simulation. Open circles represent monthly mean values. The 5-year averaging interval over which annual mean results are obtained for the LGM is indicated.

When the simulated SSTs for today's climate are subtracted from those for the LGM (Fig. 1B), SST cooling on the order of 2° to 3°C can be seen over large parts of the globe (underneath the subtropical highs, for example). This cooling can be accounted for by the reduced atmospheric CO2 and the glacial boundary forcing, as in earlier models (9, 10). In our model, the extratropical cooling lowers temperatures in the equatorial thermocline over the next decade of simulation (because of the shallow, meridional, wind-driven circulation described above) and affects ocean-atmosphere interactions. Those interactions are primarily responsible for the tropical cooling of 4° to 6°C in the western Pacific, the strengthening of the trade winds (Fig.3A), the enhancement of the tilt of the thermocline (Fig. 3B), and the increase in the peak speeds of the South Equatorial Current and the Equatorial Undercurrent by a factor of nearly 2. These changes translate into stronger upwelling and zonal advection, processes that amount to an additional cooling (relative to today) of 18.6 W/m2 of the upper equatorial ocean. To compensate, there is an increase in shortwave radiation over the western Pacific because the region of strong convection shifts westward over the Indonesian archipelago. There the latent heat release produces what looks like a baroclinic response in the mean zonal circulation (22) that feeds back on the Pacific easterlies. In Fig.4, the anomalous LGM heating is the region of minimal cooling above the surface near 100°E; the enhanced surface easterlies are most likely a combined response to both the shift in convection and an increased flux of easterly momentum from midlatitude baroclinic waves.

Figure 3

(A) Simulated annual mean zonal wind U averaged from 10°S to 10°N as a function of longitude in present day (open circles) and the LGM (solid circles). (B) Simulated depth of the 18°C isotherm in the Pacific Ocean in present day (thin line) and the LGM (thick line). (The longitude scales in the two figures are not the same.)

Figure 4

Difference in annual mean temperature (shaded) as a function of longitude and height [950 to 100 mbar (mb)] averaged between 30°S and 30°N (LGM minus present day; quantities are temporally averaged as in Fig. 1). The bar on the right gives the scale (in degrees Celsius). Superposed in contours is the difference in the mean Walker circulation averaged between 30°S and 30°N (contour interval is 0.5 × 1011 kg/s).

In a zonal mean, atmospheric temperatures in the high northern latitudes are 6° to 8°C colder throughout the entire troposphere, whereas tropical tropospheric temperatures are 4° to 6°C lower (the colder values occur in the upper equatorial troposphere and arise from a decrease in latent heat release). Glacial forcing therefore amplifies the atmosphere's meridional baroclinic structure: The vertical shear in zonal winds in the North Atlantic storm track is 26% greater. This amplification would increase the frequency with which baroclinic eddies are spawned (23) and, as has already been demonstrated for the LGM (8), increases the eddy fluxes of heat and momentum. Evidence for this increase is in the strength of the Northern Hemisphere Hadley cell, which is, in the LGM simulation, 20% greater.

Mean annual atmospheric temperature depressions in the lowest model level (about 30 m above ground in a standard atmosphere) indicate more than 7°C cooling over the East African Highlands and Central America and 5°C cooling over much of the tropical Pacific Ocean. Continental interiors in the tropics are, in general, 5° to 6°C colder with a fair amount of spatial variability. The LGM atmosphere is 4.3°C colder near the surface (globally averaged). The glacial cooling is not as large, however, over the high-latitude Southern Hemisphere oceans, although temperature depressions in the central Antarctic reach 11°C. Differential cooling of the Northern and Southern Hemispheres in response to the LGM ice sheet distribution has been noted in previous studies (9, 24).

The globally averaged surface temperature of a planet depends primarily on two factors: the albedo and the atmospheric concentration of greenhouse gases. The various simulations of the LGM, including this one, all take into account the high albedo associated with glaciers and low atmospheric concentration of CO2. Part of the explanation for the lower temperatures in this simulation (relative to earlier simulations) is a decrease in the concentration of another greenhouse gas, water vapor. Simulated global mean LGM specific humidity is 10% less than in the present-day simulation. (Relative humidity, however, increases by 5% because the net decrease in saturation vapor pressure induced by colder temperatures is greater than the net decrease in specific humidity.) During the simulated LGM, the hydrological cycle is weaker than today; both evaporation and precipitation are reduced by about 10%.

Although the simulations described here are in short-term equilibrium (Fig. 2), further changes are likely on time scales of centuries when the thermohaline circulation alters the thermal structure of the deep ocean (25). Calculations for such extended periods are beyond the scope of this study.

  • * To whom correspondence should be addressed. E-mail: andrew.bush{at}


View Abstract

Stay Connected to Science

Navigate This Article