U.S. Dept. of Commerce / NOAA / OAR / ERL / GFDL *Disclaimer  

 

1. CLIMATE DYNAMICS

GOALS

1.1 BACKGROUND FOR COUPLED CLIMATE MODELING AT GFDL

Two distinct coupled atmosphere-ocean climate models are actively being used in research on global warming and other aspects of climatic sensitivity and variability. This section summarizes the general characteristics of these models and the design of some key experiments using these models.

A low resolution model has provided the basis for much of GFDL's global warming, paleoclimate, and low-frequency variability research for over a decade and continues to yield new insights. This model consists of an R15 atmosphere (with a 7.5 longitude by 4.5 latitude grid) coupled to an ocean model of roughly 4 resolution, a simple free drift ice model, and a "bucket" land hydrology. While the resolution of this model is marginal, and, in some cases clearly insufficient, for the simulation of many atmospheric and oceanic phenomena, the low resolution and simplicity of this model and the flux adjustment strategy it employs are designed to allow integrations that would otherwise be computationally prohibitive. To cite some examples (described more fully below), this model has been integrated for 10,000 years with a stable climate, and an ensemble of nine CO2+aerosol global warming scenarios has been generated. This model is referred to as the R15 coupled model.

A medium resolution coupled model, designated R30, has been under development for the past several years, and has recently been integrated for 400 years with a reasonably stable climate. It has also been used for global warming simulations. It consists of atmosphere and ocean models with exactly twice the horizontal resolution as R15, with 50% higher vertical resolution in both atmosphere and ocean, and with similar ice and land surface parameterizations. The atmospheric component of this model has been extensively analyzed, most recently by the Observational Studies Group in a series of twelve 40-year experiments with prescribed observed sea surface temperatures (5.4), and has been shown to provide realistic responses to tropical sea surface temperature anomalies. It has been know for many years that the R30 model provides an atmospheric circulation superior to the R15 model. This is most evident in the simulation of the extratropical storm tracks, the strength of the tropical Hadley circulation and the intertropical convergence zone, and the strength of midlatitude westerlies in the Southern Hemisphere, all of which have important implications for the coupled model and the generation of global warming scenarios. The R30 coupled model is also run with flux adjustments, and the development of this model continues. Results are included in this chapter as a status report, and the reader is encouraged to keep their provisional nature in mind.

Both the R15 and R30 models have been used to simulate the evolution of the Earth's climate due to past changes in greenhouse gases and sulfate aerosols, as well as to project climate change into the 21st century. In these scenario integrations, the format follows that used in a previous R15 model study (1473) which, in turn, is based on Mitchell et al. (1995)1. An "equivalent" CO2 concentration is used to represent changes in all of the trace greenhouse gases, and changes in aerosol loading are modeled by changing the surface albedo. The "standard" scenario integrations begin in 1765 and proceed for 300 years, till 2065 with equivalent CO2 increasing at a rate of 1% per year after 1990. This projection is clearly uncertain. Some integrations are underway with a 0.5% increase in equivalent CO2 per year after 1990. Forcing functions for the 20th century are also uncertain, due to uncertainties in aerosol loading, indirect effects of aerosols on clouds, biomass burning, and possible solar insolation changes. Any climate projections must be considered with these uncertainties in mind.

1.2 COUPLED MODEL DEVELOPMENT

ACTIVITIES FY98

While a stable, long integration of the R30 coupled model has been achieved in the past year, research continues on modifying this model to reduce climate drift, minimize the shock of initialization, and generate more realistic tropical interseasonal variability.

A version of the R30 coupled model without flux adjustments has been run for several centuries to help identify any major problems in the model. As anticipated, the subtropical oceans become too warm due to the lack of boundary layer stratus clouds, and the oceanic salinities depart from those observed in several important regions, particularly in regions affected by substantial river runoff. The overturning circulation in the Atlantic is observed to weaken dramatically. Given the R15 model prediction of large changes in this thermohaline circulation in the 21st century, the stability of the overturning is a critical part of the coupled model requiring scrutiny. The conclusion is that the non-flux-adjusted model is not currently usable for global warming studies, even though it produces a realistic climate in many other respects and has little drift in global mean temperature. We continue to use flux adjustments at the air-sea interface, realizing on the one hand that both this procedure and overturning as an alternative are problematic, and on the other hand that the realism of the unperturbed climate is essential to studies of climate sensitivity.

The R30 coupled model with flux adjustments undergoes a shock following the multi-step initialization protocol within which the flux adjustments are computed. In particular, the strength of the thermohaline circulation in the Atlantic weakens by one-third and undergoes several large excursions (of 4-6 Sverdrups) that are unlikely to be characteristic of the final model climate (Fig. 1.1). Most of this shock appears to result from the fact that the coupled model has a somewhat different distribution of precipitation than that used when spinning up the ocean-only model to its climatic equilibrium as part of the initialization procedure. Modest changes in this pattern move rainfall from one drainage basin to another, so that rivers run off into different parts of the ocean, thereby altering oceanic salinities. Experiments are underway to minimize the modest changes in sea surface temperature that appear to be the cause of these shifts in precipitation. Alternative initialization strategies are also under consideration.

The exaggerated ENSO-like variability of the R30 model (1.3.2) has been found to be surprisingly sensitive to the horizontal diffusion in the ocean model. Experiments on this aspect of the R30 model are in progress.

The coupled model has been renovated in the past year to take advantage of upgrades to MOM (the Modular Ocean Model) (4.2.1) in a timely fashion. A version of the model coupled to MOM2 has been tested and found to produce a climate simulation similar to earlier versions of the model. Testing of a model incorporating the MOM3 code and its new isopycnal mixing schemes is now underway.

A new structure for atmospheric model development is under construction, jointly with the Experimental Prediction and Atmospheric Processes Groups, which will centralize laboratory activities aimed at improving the atmospheric component of the models used for climate studies, as well as for research in extended-range and seasonal forecasting (3.2).

PLANS FY99

The coupled model will make the transition to the new atmospheric modeling framework within the next year. Ongoing research on ice and land surface modeling will be incorporated into this new model in the next year as well. The goal on the one to two year time scale are to build new coupled models at approximately 3 atmospheric resolution and with oceanic resolutions ranging from 3 to 1, which can be stably integrated for a millennium with less dependence on flux adjustments at the air-sea interface. Simultaneously, a version of the atmospheric model at a resolution of roughly 1 will be built for the study of regional climate change using boundary conditions provided by the lower resolution coupled model, all in a software environment that encourages experimentation by a large number of users inside and outside of GFDL.

1.3 SENSITIVITY AND VARIABILITY STUDIES WITH THE
R30 COUPLED MODEL

ACTIVITIES FY98

1.3.1 Control Integration and Global Warming Experiments

A reasonably stable control run with the R30 coupled model has been achieved using an initialization and flux adjustment strategy similar to that used in earlier R15 experiments. This integration is currently over 400 years in duration. The drift in global mean sea surface temperatures over the first 400 years is less than 0.1 K.

Five extended integrations of this model with perturbed radiative forcing have been conducted. In the first of these integrations, the CO2 concentration, meant to represent all of the radiatively active trace gases, increases at a rate of 1% per year until the initial concentration is doubled, and is held constant thereafter. A second integration has the same structure, but the effective CO2 concentration does not level off until it quadruples.

In addition, integrations for three alternative scenarios are in progress using the standard framework (1.1), but with the integrations starting in 1865. The first two scenarios have CO2 increasing at 1% per year after 1990, but differ in initial condition with the second integration branching off from the control run 200 years later than does the first. The third scenario run branches off from the first at year 1990 and has CO2 increasing at 0.5% per year thereafter.

Shown in Fig. 1.2 are time series of global mean surface air temperature from observations, one of the 1% per year R30 scenario runs, and previously obtained results from the R15 coupled model. The larger variability in the R30 experiment is partially attributable to the model's exaggerated ENSO-like variability in the tropical Pacific (1.3.2). The R15 and R30 atmospheric models have similar equilibrium sensitivities to a doubling of CO2 when coupled to mixed layer oceans. For both, the global mean surface air temperature increase is close to 4 K, which is at the upper end of the 1.5 K-4.5 K range discussed by the IPCC (Intergovernmental Panel on Climate Change), so they are anticipated to have similar overall sensitivities when coupled to full ocean models as well. Fig. 1.2 is consistent with this expectation. The two models agree on the timing of the warming in these scenario integrations, although there is a hint that the warming in the R30 is delayed by roughly a decade compared to the R15. Despite their high sensitivity, both model integrations are also in relatively good agreement with the observed increase of surface air temperature over the last century. This fit must be interpreted with care, as it is dependent on the cooling effect of the poorly constrained sulfate forcing, and could also be modified by other radiatively important

perturbations to the system (i.e., solar constant changes, biomass burning) unaccounted for in this calculation.

Preliminary results have been obtained regarding changes in the Atlantic overturning circulation in the five transient integrations with varying CO2 and/or aerosol loading. The overturning eventually weakens in all of these integrations, but indications from the scenario integrations are that this weakening in less rapid in the R30 coupled model than in the R15. Given the large internal variability of the overturning in the R30 model, some of which could be related to the initialization process, additional experimentation will be needed before a more definitive analysis can be completed.

The spatial pattern of surface air temperature changes for years 2030-2050 minus 19761995 is shown in Fig. 1.3. Noteworthy features include the slower warming in the Southern Hemisphere compared to the Northern Hemisphere (particularly over the Southern Ocean), and the more rapid warming of the eastern tropical Pacific compared to the western tropical Pacific, which has the effect of decreasing the east-west temperature gradient. Given the large decadal variability in this model, the details in this pattern will undoubtedly differ from one realization to another.

1.3.2 Tropical Pacific Variability

An important test of coupled climate models is their simulation of tropical Pacific SST variability on interannual to decadal time scales. The R30 coupled model shows very pronounced ENSO-like (El Niño-Southern Oscillation) variability in the equatorial Pacific. Large magnitude events such as the super El Niños of 1982-1983 and 1997-1998 are common in the model. Indeed, they are too common, and have too long a duration. The model's tropical Pacific has a longer time scale (~8 years) than the observed ENSO (~4 years), and the oscillations are too regular and somewhat too large in amplitude on average, as can be seen in Fig. 1.4. However, preliminary analysis indicates that the dynamics underlying the simulated variability appears to be quite similar to the "delayed oscillator" mechanism to which the observed interannual variability is generally attributed (er).

The model's enhanced and relatively low-frequency ENSO-like Pacific variability results in enhanced decadal variability in the extratropical Pacific Ocean, through "atmospheric bridge" effects (1393), making it difficult to isolate extratropical sources of variability on these time scales. It also may create exaggerated decadal variations in global mean temperature, biasing studies of global warming detection. Therefore, it is important to isolate the reason for this larger-amplitude low-frequency ENSO-like variability in the model. One possibility is simply that greater resolution is needed in the ocean model, particularly within the narrow equatorial waveguide which is thought to play an important role in shaping ENSO dynamics. On the other hand, recently obtained results indicate that the tropical variability in the model appears to become more realistic when the horizontal diffusivity in the ocean is reduced, shifting to higher frequencies and with somewhat smaller amplitudes. Thus, while higher resolution may ultimately be needed to obtain satisfying ENSO simulations, it is possible that the R30 model's distorted variability may be at least partially ameliorated without an increase in resolution.

1.3.3 Southern Hemisphere Atmospheric Response to Global Warming

The various transient R30 coupled model integrations produce an interesting response in the surface winds in the Southern Hemisphere. The CO2-doubling experiment described above (effective CO2 is increased at 1% annually until doubled at about year 70, after which it is held fixed), provides a particularly clear example, illustrated in Fig. 1.5. The zonally averaged winds become more westerly, by roughly 1 m/s, in a latitude band centered at 60S, while becoming more easterly near 40S. This corresponds to a poleward shift and a modest strengthening of the westerlies.

This wind shift is presumed to be related to changes in the north-south temperature gradients that provide the energy source for the storms that maintain this wind distribution through their momentum fluxes. A robust feature of both R15 and R30 coupled model responses to transient greenhouse-gas increase is a significant difference between the hemispheres in the latitudinal pattern of the warming. In the Northern Hemisphere, the warming increases with latitude so that the magnitude of the north-south temperature gradient is always reduced. In contrast, there is initially very little warming of the Southern Oceans, from 50S to 70S, while the tropics do warm somewhat, so that the temperature contrast between low and high latitudes in the Southern Hemisphere is increased. The increase in the temperature gradient is more pronounced in the upper troposphere because the tropical warming increases with increasing height. As the integration proceeds, the polar amplification familiar from equilibrium integrations eventually sets in (as studied in detail in the R15 context in 1.4.5), reducing the temperature gradient near the surface.

Work is underway to better understand the connections between the changes in temperature gradient, both at the surface and in the upper troposphere, and the surface westerlies. This work involves analysis of eddy fluxes and eddy structures in the atmospheric model, experimentation with models with more idealized lower boundary conditions that produce similar responses, and experimentation with modified coupled models to see if these wind anomalies have a significant impact on the evolution of the ocean. The pattern of the wind anomalies in Fig. 1.5 is very similar to the dominant pattern of the month-to-month variability of these winds in atmosphere-only integrations with fixed ocean temperatures. Therefore, one focus of this research is the possibility that one phase of a natural "mode" of variability, akin to the classical "index cycle", is preferentially being excited by changes in oceanic temperatures.

PLANS FY99

Experimentation with incremental modifications to the model will focus on the characteristics of the model's ENSO-like variability and on the problem of initializing the model so as avoid exciting oscillations in the North Atlantic overturning that are not representative of the model's long term climate. Meanwhile, integration and analysis of the existing set of integrations will continue, with the immediate goal of clarifying which conclusions concerning climate sensitivity and variability obtained with the R15 model over the past decade are also supported by R30, and which may need to be modified. A focus on Southern Hemisphere circulation changes will continue. Besides being important in itself, the Southern Hemisphere response to climate change provides a relatively simple test bed for dynamical ideas that should prove useful in interpreting the more complex circulation of the Northern Hemisphere.

1.4 SENSITIVITY AND VARIABILITY STUDIES WITH THE
R15 COUPLED MODEL

ACTIVITIES FY98

1.4.1 A 10,000 Year Integration

The control integration of the R15 coupled model has now been extended to 10,000 years in order to study natural variability in the model on time scales up to 1000 years. A small drift in the global mean surface air temperature found in the first 1000 years of the integration continued until the year 2000, but the total drift in the global mean is only 0.7 K. After year 4000, there appears to be little or no drift in any part of the coupled system. From preliminary analysis it appears that most of the variability on time scales of 100 years and longer is found in the Southern Ocean. However, one very interesting event occurs near year 3100 in the North Hemisphere. Fig. 1.6 shows the time series of annual mean surface air temperature near the sinking site of the thermohaline circulation in the Northern Hemisphere. In this time period, an exceptional negative temperature anomaly occurs, reaching a magnitude of 4 K. Except for this event, anomalies in this region never reach significantly beyond 2 K. During this same time period, the Atlantic thermohaline circulation weakens from its normal value of 18Sverdrups to 14 Sverdrups. The cold anomaly is centered over the sinking region of this circulation and extends, with smaller amplitude, from Northern Canada, across Greenland and the North Atlantic, to Western Europe. This cold anomaly is also associated with changes in sea level pressure which likely act to enhance the cooling in the North Atlantic. This event appears to be a very high amplitude case of the model's dominant mode of interdecadal variability (1428).

In the paleo-record, there is a sharp cooling event in the Holocene, near 8,200 years before the present (Alley et al., 19972) with a somewhat longer time scale (100 years). This event has been thought to be associated with freshwater discharge from melting glaciers which slows down the thermohaline circulation, cooling the North Atlantic. This model integration suggests that this early Holocene event could be unforced variability of the coupled system that does not require large changes in freshwater input from melting glaciers. Natural events of this magnitude have the potential to make the detection and attribution of observed local climate change more difficult. The fact that only one such event occurs in this 10,000 year integration serves to underline the need for such long integrations. However, it should be noted as well that this large event has a negligible signature in the global mean temperature.

1.4.2 Two Stable Equilibrium Climates

An important result obtained in the past with the R15 coupled model is that it supports two distinct climate states with the same boundary conditions and forcing. One state has a realistic and active thermohaline circulation in the Atlantic Ocean. The other has a weak reversed circulation with extremely weak upwelling throughout the North Atlantic, although it maintains substantial sinking in the Southern Ocean. The temperature difference between these two climatic states at the surface in the Polar North Atlantic is as large as 8 K. In order to remove lingering doubts as to whether these two states remain stable on the very long diffusive time scales on which the deep ocean equilibrates, the model in its weak reversed circulation state has now been integrated for 7,000 years, to accompany the 10,000 year control run which is in the active circulation state. As shown in Fig. 1.7a, the results demonstrate unequivocally that this model does have two distinct equilibrium climates.

Partly to address questions raised by others concerning the stability of the weak, reversed circulation mode, particularly Schiller et al.3 using a version of the MPI (Max Planck Institute, Hamburg) coupled model, this simulation has been repeated with another version of the R15 coupled model using a larger vertical subgrid scale diffusion coefficient. A weak reversed thermohaline circulation is induced in the model's Atlantic Ocean by a massive discharge of freshwater, but the circulation eventually returns to its active, direct state (Fig.1.7b). The implication is that there is a critical value of the diffusivity below which two stable equilibria exist. The behavior of the more diffusive version of the coupled model resembles the response of the model of Schiller et al. Since the MPI model employs a first-order upstream differencing scheme which yields large, implicit diffusion, it appears likely that the difference in behavior of their model and the standard R15 model results from the fact that the former is more diffusive. Even the standard R15 model uses vertical diffusivities that are larger than those measured by direct release of artificial tracers into the Atlantic Ocean, so these results support the hypothesis that the Earth possesses two distinct climatic equilibria.

1.4.3 An Ensemble of R15 Scenario Integrations

A set of nine CO2-aerosol scenario integrations has been performed to facilitate studies of the emergence of the global warming signal from the noise of natural variability. The nine scenario experiments are conducted in three groups of three. They are initialized from various points of a long-running control model integration, with successive groups initialized at times separated by 500 model years. Within each group, experiments of 300, 200, and 150 years duration were performed, with radiative forcings starting from years 1765, 1865 and 1915, respectively. One motivation for this experimental design is to help determine whether or not it is important to start integrations of this type in the pre-industrial era, or if mid 19th or even early 20th century initial conditions are adequate. This issue becomes particularly relevant when working with more computationally intensive models. In the 1765 set of experiments, there is a smooth transition from equilibrium conditions, since the aerosol and greenhouse gas forcing for year 1765 are identical to the control model's. Scenario integrations begun in 1865 or 1915 experience initial discontinuities in radiative forcing.

The sensitivity of simulated 21st century global mean surface air temperatures to the choice of radiative forcing starting point (1765, 1865 or 1915) is detectable, but relatively small, and similar in magnitude to the inter-experiment variability that arises from variations in the choice of initial conditions for the coupled model. The range of responses in the global mean surface air temperature among all nine integrations is indicated in Fig. 1.8. Further analysis will be directed at the more slowly evolving parts of the system, such as the thermohaline circulation and sea level, in order to better characterize the systematic errors associated with starting a climate change scenario run in 1865 or 1915 rather than 1765.

1.4.4 A Scenario Integration Without Water Vapor Feedback

A modified version of the R15 coupled model has been constructed in which the water vapor feedback is disabled (the radiative transfer algorithms of the model use a water vapor distribution produced by a control integration, which is fixed when the model is perturbed). The models' response to a doubling of CO2 is reduced from 3.8C to 1.05C when modified in this way. This artificial modification results in a very stable climate with a response which is below the low-end of the plausible range of climate sensitivities suggested by IPCC and others. Previous work has analyzed the reduction in low-frequency variability in this fixed H2O model (fz), and ongoing work has focused more closely on a surprisingly sharp reduction in this model's ENSO-like variability.

A standard scenario integration (1.1) has been performed with this fixed H2O model. Fig 1.9 shows the resulting evolution of global mean surface temperature, along with the analogous result for the standard R15 model and observations. The modified model is unable to predict the observed warming during the 20th century with this scenario forcing. Consistency with observations would require much larger radiative forcing. In addition, the natural variability of global mean temperature in this model is reduced by nearly a factor of two on multidecadal time scales, so it is even more difficult for this model than it is for the unmodified model to produce, through natural variability, a temperature trend like that observed.

1.4.5 Comparison of the Fully Coupled and Mixed-Layer Models

A comparison of the equilibrium and transient climatic response to changes in greenhouse gases in fully-coupled and mixed layer ocean models allows one to isolate the role that ocean dynamics plays in shaping these responses. By utilizing flux adjustments, one attempts to produce control climates in the two models which have identical sea surface temperatures. Differences between the climate responses in the two models should then be related to changes in ocean heat transport.

The equilibrium response of the climate to a doubling of the CO2 concentration in the R15 model has been studied by integrating the model until a true equilibrium is reached. The control is of 10,000 years length. The doubled CO2 integration is 5,000 years in length. Equilibria are attained after about 4,000 years of integration. The globally averaged equilibrium surface air temperature (SAT) change due to CO2 doubling in the coupled model is found to be 4.5 K, and 3.7 K for the mixed layer model. While it is tempting to look for changes in ocean circulation responsible for this larger response, it appears instead to be at least partly a consequence of a small drift towards cooler temperatures in the coupled model control run. The cooler climate results in a larger snow-ice-albedo feedback and a larger sensitivity to increased CO2. As a result, it is uncertain whether the equilibrium response of the coupled model would be significantly larger than that of the atmosphere-mixed layer model if the control temperatures were more nearly identical.

The spatial pattern of equilibrium SAT response to doubled CO2 concentration (Fig.1.10b) in the coupled model is very similar over most regions to that in the mixed-layer model (Fig. 1.10a). The familiar polar amplification is seen in both models. An exception is found in the region southwest of Australia where the warming is at a minimum in the coupled model due to changes in vertical mixing in the ocean. Analysis of the time series indicates that this pattern is robust and does seem to be a feature of this model's equilibrium response. In general, however, the changes in the ocean model's heat transport are not large enough to change the SAT pattern substantially. Whether this will continue to be the case in higher resolution ocean models remains to be seen.

A project has also been initiated to compare integrations using the mixed layer and fully-coupled R15 models, following the standard scenario format (1.1). In addition, the mixed-layer ocean-atmosphere model has been integrated to equilibrium for five separate static settings of the total radiative forcing, corresponding to the years 1980, 2000, 2020, 2040 and 2060 of the scenario integrations. As discussed in the previous paragraph, the mixed layer model appears to provide a good estimate of the equilibrium response, at least in the context of this R15 model.

Figure 1.11 summarizes the resulting responses of global mean surface air temperature. Coupling to a full-ocean model delays the global mean warming by 20-30 years as compared to the 50 meter mixed layer results, with a somewhat longer delay near the end of the integration. The delay is primarily caused by the failure of the Southern Ocean and the extreme northern part of the Atlantic to warm appreciably in the coupled model. Other differences in the geographical pattern of the responses are under investigation.

Note that the equilibrium responses are close to those of the dynamic mixed layer model, indicating that even if the radiative forcing were somehow to be held fixed at today's value, the global mean SAT would rise by an additional 0.9C, which is larger than historical warming since the late 1800s. Differences in the geographical pattern of the responses are under investigation.

1.4.6 Detection of Global Warming Trends

A comparison has been completed of the variability of surface air temperature in 1,000-year coupled model integrations performed at the Hadley Centre (HadCM2), Max Planck Institute (HAML3) and at GFDL, where the R15 simulation was used for this purpose (hn). Integrations of this length are needed to study natural variability on the 100-year time scale that is most crucial when trying to attribute the observed 20th century warming to changes in radiative forcing or to natural variability. Models play an important role in this signal detection problem because the observational record for variability on this time scale is so limited.

The linear trends of surface air temperature in all three models are small over most areas of the globe. Although there are notable differences among the models, the simulated SAT variability is fairly realistic on annual to decadal time scales, both in terms of the geographical distribution and in the global mean, with the exception of the simulation of observed tropical Pacific variability. In the HadCM2 model, the tropical variability is overestimated, as in the GFDL R30 model (1.3.2), while in the GFDL R15 and HAM3L models it is underestimated. 1000-year integrations with models that provide better simulations of ENSO are clearly needed, but are not yet available.

None of the models generate a temperature trend as large as that observed (Fig.1.12). If the models' simulation of variability on long time scales is realistic, then the observed warming must be due to changes in the radiative forcing of the planet and not the result of internally generated variability.

A question of central importance is, therefore, whether all of these models could be grossly underestimating natural variability on the century time scale. It would be surprising if models that simulate ENSO better would affect this conclusion, since models that actually overestimate ENSO variability produce nothing resembling the observed trend. Mesoscale ocean eddies appear to be the most plausible remaining potential source of variability missing in these models. Such eddies could plausibly energize the variability in the oceans on the gyre scale and within convective regions, at higher levels than the rather diffusive, low resolution ocean models used here. Climate change studies resolving ocean eddies in global models will require greatly enhanced computational resources.

1.4.7 Decadal to Multi-Decadal Variability

Previous analyses have documented the existence of multi-decadal variability in the North Atlantic thermohaline circulation of the R15 coupled model (1428). In order to evaluate the role of the atmosphere in this variability, experiments have recently been conducted using only the oceanic component of this model subjected to different time series of surface fluxes. These fluxes were derived both from extended integrations of the coupled model and from integrations of the atmospheric component of the coupled model run with a prescribed seasonal cycle of ocean surface temperatures. A sequence of experiments has been performed in which the nature of the fluxes is systematically altered in order to explore the importance of air-sea coupling for the model's thermohaline variability. This study demonstrates that the thermohaline variability can be fully excited by surface flux forcing which is stochastic in time, implying that the thermohaline variability in this model is not due to feedback between the surface fluxes and the state of the ocean, and rules out explanations based on coupled air-sea modes. It also demonstrates that variations in surface heat flux are primarily responsible for exciting this thermohaline variability, rather than variations in freshwater flux.

Analyses have also been conducted to assess the higher frequency interannual to decadal variability simulated in the tropical North Atlantic by the R15 coupled model. Spectral analyses of both observed and simulated SST variations over this region show a weak peak on decadal time scales. The dominant spatial pattern of SST variability resembles that from observational analyses, as shown in Fig. 1.13. In the model, fluctuations in the intensity of the atmospheric subtropical high are associated with changes in the intensity of the trade winds, thereby causing changes in the air-sea flux of latent heat and providing the primary mechanism responsible for this model-simulated pattern of SST variability. Changes in oceanic heat advection also contribute to the SST pattern, but these are smaller in magnitude than the surface heat flux variations.

PLANS FY99

Work in the coming year with the R15 coupled model will focus on those areas in which such a fast, low-resolution model still plays a unique role in climate research. The 10,000 year integration will be analyzed to study natural variability on the 100-1,000 year time scale. Using the new nine-member ensemble of scenario integrations, which will likely be extended to more members, the emergence of the full variety of climatic signals from the noise of natural variability will be studied, and these integrations will be used as a test-bed for statistical methods that will be useful for future integrations with higher resolution models. Comparisons will be made between the multi-decadal thermohaline circulation variability in R15 and in the emerging long control integration with the R30 model. Finally, both the R15 and R30 models will continue to provide a testing ground for new techniques for initializing coupled models, and it will be upgraded as new ocean, atmosphere, land, and ice models are tested and become available.

1.5 HYDROLOGY AND CLIMATE

ACTIVITIES FY98

1.5.1 Summer Dryness in a R15 Scenario Integration

In a continuation of earlier work on the drying of continental interiors due to global warming (1473), this phenomenon has been re-examined in a R15 coupled model scenario run. Standard scenario forcing is used (1.1) except that three integrations are performed, one with equivalent CO2 forcing alone and one with aerosol forcing alone, in addition to the standard case of CO2 + aerosol forcing. The pattern of drying in the presence of CO2 + aerosols is similar to that found in previous integrations: summer drying and winter moistening in middle-to-high latitudes of North America and southern Europe, but drying through most of the annual cycle in southern North America, where the percentage reduction of soil moisture during summer is particularly large. A similar pattern is seen in other semi-arid regions in subtropical to middle latitudes, such as central Asia and the area surrounding the Mediterranean Sea. An analysis of the central North American region (Fig. 1.14) indicates that the reduction of summertime soil moisture exceeds one standard deviation of the summertime mean soil moisture by around year 2030. The corresponding emergence of the runoff signal is even later. The inclusion of sulfate aerosols delays the reduction of soil moisture by several decades. Using the same measure, the model's temperature response in this region has already emerged from the variability by 1980. Surface air temperature in the model is clearly a better early indicator of global warming than hydrologic quantities, such as precipitation, runoff and soil moisture. The high climatic sensitivity of this model, more than 4 K for the equilibrium response to doubling CO2, which is at the upper end of the range regarded as probable by the IPCC, should be kept in mind when considering these model responses.

1.5.2 Changes in Flood Frequency and Trends in River Discharge

The detectability of greenhouse-induced changes in extremes of river discharge has been examined using three R15 scenario runs and observational data from 45 large river basins, with periods of record typically 50-100 years. The corresponding model discharges were derived through a simple river-basin storage model, from global runoff fields generated in a 1000-year steady-climate experiment and from three standard scenario experiments (1.1). Preliminary analyses suggest that changes in flow characteristics of individual rivers could generally not be detected until well into the 21st century. For this reason, and to simplify subsequent analysis, the pooling of discharge measures from multiple river basins was explored. Fig. 1.15 shows the results of such an analysis. It can be seen that the fraction of basins experiencing a 30-year flood (flood whose magnitude is equaled or exceeded, on average, once every 30 years) in any year has fluctuated around the expected value of 1/30 during the past century, with no obvious trend. Additionally, the fluctuations are consistent with those expected from the steady-climate and greenhouse-warming experiments. Only at the start of the 21st century does the climate change signal indicate a significant rise in the incidence of flooding within this set of basins. In general, the analyses to date suggest that any globally-extensive changes in flood frequencies associated with greenhouse warming are too small to be detected at this time.

Previous studies of 20th-century river discharge revealed an upward trend in combined mean runoff of nine major world rivers (gb). The +8% trend over the century (significant at the 99% level) appears to be too large to be easily explained by natural variability, increased water-use efficiency of ecosystems resulting from heightened atmospheric carbon-dioxide levels, or direct water-balance effects of deforestation. It is also much larger than trends predicted by the R15 scenario runs. Because the Amazon River has the largest mean flow of the nine rivers, and because its individual flow trend has the highest level of statistical significance (95%), it was chosen for further study. Preliminary analyses of precipitation records suggest that precipitation trends were insufficient to explain the flow rise. Furthermore, the largest increases in runoff appear to have occurred during the dry season. These factors point toward decreased dry-season evaporative demand as the cause of the flow rise. Decreased evaporative demand could be driven by increase in cloud cover. An alternative explanation is a reduction in surface solar radiation caused by increased smoke aerosols, transported into the basin from nearby regions of intensive biomass burning.

1.5.3 Sensitivity of the Global Water Cycle to Stomatal Resistance of Vegetation

A series of numerical experiments has been conducted with the R-15 climate model with fixed ocean temperatures to assess the sensitivity of the global water cycle to stomatal resistance of continental vegetation. Historically, GFDL climate models have not included any representation of this resistance. Its direct physical effect is to impede evaporation and latent heat flux from the land surface to the atmosphere. Thus, the direct effect of incorporating stomatal resistance into a model should be to decrease mean evaporation and thus increase runoff from land to oceans. An equilibrium climate has been calculated for bulk stomatal conductance equal to 100 m/s, an estimated realistic value. Fig. 1.16 shows the global field of change in annual mean runoff resulting from inclusion of bulk stomatal resistance in the model. In the tropics and south Asia, the computed increase in runoff (5-15 cm over large areas) is qualitatively consistent with the expected direct model response. In contrast, in the middle and high latitudes, the dominant response is a decrease in runoff.

PLANS FY99

The summertime drying of continental interiors and changes in river runoff will be studied with the expanding set of scenario integrations being performed with the R15 and R30 coupled models. In addition, the counter-intuitive effect of stomatal resistance on runoff will be analyzed. Work will also begin on a new class of land surface models for inclusion in the developing modular atmospheric modeling framework.

1.6 PALEOCLIMATE MODELING

ACTIVITIES FY98

1.6.1 Tropical Cooling at the Last Glacial Maximum

The ongoing investigation of the sensitivity of tropical temperature to glacial forcing continued during the past year. Using specifications of glacial forcing established by the Paleoclimate Modeling Intercomparison Project (PMIP), changes in continental ice, orbital parameters, atmospheric CO2 and sea level were imposed in the R30 model coupled to a mixed layer ocean. These changes create a global mean radiative forcing of -4.20 Wm-2, with the vast majority of this forcing coming, in nearly equal portions, from the changes in continental ice and CO2. In response, the model's global mean surface air temperature decreases by 4.0C, with the largest cooling in the extratropical Northern Hemisphere. In the tropics, a more modest cooling of 2.0C (averaged from 30N to 30S) is simulated, but with considerable spatial variability resulting from the interhemispheric asymmetry in radiative forcing, contrast between oceanic and continental response, advective effects, and changes in soil moisture. Analysis of the tropical energy balance reveals that the decrease in top-of-the-atmosphere longwave emission associated with the tropical cooling is balanced primarily by a combination of increased reflection of shortwave radiation by clouds and an increased atmospheric heat transport to the extratropics.

Comparisons with paleodata indicate that the overall tropical cooling simulated by this model is comparable to paleoceanographic reconstructions based on alkenones and species abundances of planktonic microogranisms. The most comprehensive paleotemperature reconstruction of this kind comes from the CLIMAP (Climate: Long Range Investigation Mapping and Prediction) Project, in which statistical transfer functions were used to estimate glacial sea surface temperatures from the abundances of planktonic microfossils in deep sea sediment cores. The zonal mean glacial anomalies simulated by the model are quite similar to those reconstructed by CLIMAP (Fig. 1.17). Other paleotemperature proxies, such as noble gases in aquifers, pollen, snow line depression, and the isotopic composition of corals, indicate a larger glacial cooling than that simulated by the model. The differences in the magnitude of the reconstructed tropical cooling among the different proxies preclude a definitive evaluation of the realism of the tropical sensitivity of the model. Nonetheless, the comparisons with paleodata suggest that it is unlikely that the model exaggerates the actual climate sensitivity in the tropics.

1.6.2 Climate Variations During the Last Glacial Cycle

Preparations have begun for an R15 atmosphere-mixed layer ocean model experiment intended to explore the contributions of changes in orbital parameters, continental ice extent, and atmospheric CO2 content to climate variations during the past 120,000 years. This experiment is intended to investigate the internal workings of the climate system which communicated these three forcings to the paleoclimate record. Because an atmosphere-mixed layer ocean model will be used, mechanisms involving the atmosphere will be the focus. Pathways involving changes to the ocean's circulation will not be included

because of the large computational cost. To further reduce the computational requirements, the variations in climate forcing will be accelerated by a factor of approximately 30, reducing the integration to a manageable 4,000 years, a strategy that is consistent with the relatively short response time of the atmosphere-mixed layer ocean system. The experimental design will provide time series information that can be directly compared to a variety of paleoclimatic data sets. Most of the work during the past year has dealt with two issues: developing temporally consistent time series for each of the climate forcings (orbital parameters, ice extent, and atmospheric greenhouse gas concentrations) and adapting the existing version of the model to produce suitable output.

1.6.3 Paleohydrological Analysis of the Lake Victoria Basin

Recent field investigations4 at Lake Victoria have suggested that the lake was dry prior to 12,800 14C years ago. Through a simple water-balance analysis, it was subsequently inferred that precipitation in the drainage basin of Lake Victoria must have been smaller than its present value by at least a factor of 4 in order for the lake to remain dry5. This inference, together with the near-simultaneity of large lake-level changes at other locations on the globe, has been used to support the hypothesis of abrupt global climate change near the end of the most recent glaciation. In the present study, the water balance of Lake Victoria has been reconsidered, with attention to various feedbacks that would accompany regional drying. As precipitation decreases, not only does the runoff decrease, but the fraction of the precipitation that runs off decreases also, in response to general drying of the soil. This runoff feedback appears to be a major factor in the history of Lake Victoria. In contrast, the evaporation feedback, whereby lake-area reduction contributes to regional aridity and enhancement of evaporation from land, is apparently only a minor factor. Additionally, possible regional changes in surface radiation add significant uncertainty to the paleohydrologic analysis. Overall, the new analysis suggests a higher sensitivity of lake area to precipitation than previously recognized. A halving of precipitation in the Lake Victoria basin appears sufficient to dry out the lake. This revised estimate of lake sensitivity to precipitation makes it easier to explain the drying of Lake Victoria in terms of orbital forcing, without the need to invoke an abrupt shift in global atmospheric circulation.

PLANS FY99

The 120,000 year accelerated simulation of the last glacial cycle with the R15-mixed layer model will be generated, and the results will be compared to a variety of paleoclimatic time series. The experiment will be repeated forcing the model with only the changes in orbital parameters, ice sheets, or CO2, in order to study how each of these forcings contributes to the total response.

1.7 LARGE-SCALE ATMOSPHERIC DYNAMICS

ACTIVITIES FY98

1.7.1 Extratropical Forcing of Tropical Interhemispheric Asymmetry

Evidence from a variety of climate simulations has been suggestive of a relationship between the latitude of the intertropical convergence zone (ITCZ) and the interhemispheric temperature contrast. An experiment has been performed to address this issue more directly, using a climate model with idealized continental geometry. The model geography includes two continents, symmetric about the equator, which produce approximately the same land fraction as the actual continents. Poleward of 40 latitude, seasonally varying but zonally uniform sea surface temperatures are prescribed in the control run in approximate agreement with observations; elsewhere, a mixed layer ocean is present, with a heat flux adjustment to mimic the poleward transport of heat by the ocean. The resulting climate resembles a more zonally symmetric version of the actual climate, with the tropical ocean temperature maximum and ITCZ migrating from the southern tropics in boreal winter to the northern tropics in boreal summer.

To create an asymmetry in extratropical forcing, a second run of the hybrid model was made in which 4C positive (negative) surface temperature anomalies are imposed in the Northern (Southern) Hemisphere poleward of 40 latitude, in the region in which surface temperatures are prescribed. Any response of tropical climate to this perturbation must be transmitted through the model atmosphere. Dramatic changes in tropical circulation result, with a northward shift of ~6 in the latitude of the ITCZ and the generation of southerly anomalies in the tropical surface winds (Fig. 1.18). These features are most evident over oceanic regions, as monsoonal effects produce a more complicated pattern over the low latitude continents. The symmetric Hadley cell structure of the control experiment is replaced by a much more asymmetric structure in which the southern Hadley cell expands and intensifies at the expense of its northern counterpart.

This experiment clearly demonstrates that the ITCZ and the interhemispheric asymmetry of the Hadley circulation can be strongly influenced by extratropical forcing, a process that is likely to play an important role in the tropical response to ice age forcing and also in the transient response to global warming, in which there is substantial interhemispheric asymmetry in the extratropical climatic response.

1.7.2 Tropical Intraseasonal Oscillations

The sensitivity of tropical intraseasonal oscillations (TIOs) to cloud feedback has been examined using various R30 atmospheric models. The simulation of this oscillation continues to be a severe test of a model of the tropical atmosphere. Two separate modes of variability have been isolated in analyses of atmospheric models (1118), one at 20-30 days and one at 40-50 days. The observations have power at 40-50 days, but very little at 20-30 days, whereas in some models the 20-30 day peak is as strong, or even stronger than the lower frequency peak. The relative magnitude of these two peaks has been found to be sensitive to cloud feedbacks in the model.

When the distribution of cloud cover is fixed at observed zonal-annual mean values, as in previous work, the model's 40-50-day oscillation has an amplitude comparable to that of the 20-30-day oscillations, contrary to observations. With predicted clouds the 40-50-day oscillation is strengthened and the 20-30-day oscillation weakened, in much better agreement with observations. However, this modification changes the mean tropical climate, which in turn affects the oscillations. By prescribing the distribution of daily values of cloud cover to be that obtained from a one-year run of the predicted-cloud model, the correlation of clouds with other variables can be eliminated while leaving the climatic mean nearly unaffected. This experiment leads to a more complex picture of the effects of the cloud feedback. The direct cloud-oscillation feedback does in fact weaken the 20-30-day oscillation, but does not affect the 40-50-day oscillation. The latter is strengthened in the predicted cloud model through the changes in basic state in the tropics. Space-time regression analysis of radiative damping and amplification rates has been used to help rationalize these results.

One candidate mechanism for exciting the TIO is evaporation-wind feedback (EWF), in which changes in low level winds modify the evaporation field, which then alters the distribution of latent heat release that drives the wind field. This mechanism has recently been studied in both idealized and realistic atmospheric model settings, extending previous idealized studies (1452). The idealized model with globally uniform sea surface temperature has an easterly zonal-mean surface flow in the tropics, which is thought to be conducive to EWF. When the zonal-mean zonal flow in the parameterized surface heat fluxes is replaced with weak westerlies, the TIOs weaken drastically, consistent with the EWF mechanism. Similarly, when the fluctuations in the wind speed of the surface heat fluxes are eliminated, the TIOs weaken drastically, again consistent with the EWF mechanism.

The results are different in the standard R30 model. In this model the surface winds in the tropical western-Pacific region (90E-180E, 30N-30S) are biased towards being weak easterlies rather than weak westerlies as observed. When the surface winds in the flux formulation are replaced by weak westerlies, the TIO does not weaken dramatically, contrary to the EWF mechanism. Similarly, when the fluctuations of the surface wind in the flux formulation are eliminated from the tropical region (30N-30S), the TIOs again do not weaken drastically. While one can create idealized models in which EWF is dominant, the TIOs are generated in other ways in more realistic models (1490).

1.7.3 Atmospheric Test of a Method for Estimating Oceanic Eddy Diffusivities

G. Holloway6 has proposed a simple scaling that relates the rms eddy sea-surface height as measured by satellite altimetry to the effective tracer diffusivity of oceanic mesoscale eddies. This scaling, which could be of importance in the effort to parameterize fluxes due to oceanic eddies, has been tested using atmospheric data, for which the energetic eddies are well resolved. Using NCEP/NCAR reanalysis data, a diffusive estimate for the heat flux, with a diffusivity obtained from the scaled height variance, has been compared to the actual heat flux. At a given height in the lower troposphere, the proposed scaling appears to be successful to within a fairly uniform proportionality factor, as shown in the Fig.1.19. The results suggest that non-eddy resolving ocean GCMs require highly inhomogeneous diffusivities to match the inhomogeneity of the height variance in the altimeter data, and that the success of an eddy-resolving ocean model at reproducing the observed sea-surface height variability may also be a good measure of its ability to simulate the mesoscale eddy fluxes near the surface.

This atmospheric analysis also confirms that diffusive models can be very useful in emulating eddy heat fluxes in the atmosphere. Despite the success of atmospheric GCMs, replacing the complex dynamics of such models with simple diffusive parameterizations would still be of great value for some applications, such as slow transitions in and out of ice ages. The credibility of such models for climate studies is dependent on having a theory for the eddy diffusivities.

1.7.4 Linear Stochastic Models of the Midlatitude Storm Tracks

In recent years, a new approach to developing an understanding of the horizontal structure of midlatitude eddy statistics based on linear theory has emerged, in which one attempts to mimic the statistically steady state of the atmosphere with a stable linear operator forced stochastically by white noise (Farrell and Ioannou (1995))7. Some success has been achieved recently in simulating a GCM's storm tracks with such a model (gr). This linear model passes the important test of being able to qualitatively simulate the "midwinter suppression" "of the Pacific storm track, given the seasonal cycle of the three-dimensional mean flow. It is interesting that in this linear model, changes in the perturbations far upstream over Eurasia can influence the oceanic storm tracks. The extent of this influence is, however, sensitive to the manner in which one perturbs the atmosphere. As shown in Fig 1.20, if one forces the

temperature equation in the lower troposphere only, within the shaded region over central Asia, as in the upper panel, the Pacific storm track is excited, but the eddies refract rather quickly into the tropics and do not generate an Atlantic storm track of the correct intensity. In contrast, if one perturbs both the temperature and vorticity equations of the linear model throughout the troposphere, within the same central Asian region, the Atlantic storm tracks is simulated more realistically. The dependence of the result on the details of the stirring, given the same linear operator, complicates the interpretation of such models but raises important questions about the factors that control the interaction between the two oceanic storm tracks.

1.7.5 Zonally Asymmetric Wave-Mean Flow Interaction Theory

A new framework for examining wavemean flow interaction for zonally asymmetric flows has been developed (fj). A satisfying theoretical foundation has been in place for many years for the analysis of zonally symmetric wave-mean flow interactions in atmospheric models and observations. But it is necessary to generalize this theory to apply to zonally asymmetric flows in order to study the interaction of the oceanic storm tracks with the time-mean flow. The new framework exploits the analogy between the flux of mass between density surfaces in stratified flow, which has played an important role in recent attempts at parameterizing eddy fluxes in the ocean, and the flux of mass between potential vorticity surfaces in two-dimensional horizontal flow. Results include the derivation of a residual-mean circulation formulation for two-dimensional and quasi-geostrophic dynamics, and points of contact with previous results in zonally asymmetric wavemean-flow interaction theory. This approach helps explain in simple physical terms why there can be no simple "non-acceleration theorem" in the zonally asymmetric problem analogous to the well-known theorem in the zonally symmetric case.

1.7.6 The Surface Branch of the Zonally Averaged Mass Transport Circulation

The mean meridional mass transport within isentropic layers in the atmosphere is characterized by poleward flow in the upper troposphere and equatorward flow in the lower troposphere. The lower branch of this circulation is difficult to visualize, and has rarely been examined in detail, because much of the equatorward flow occurs in layers that are interrupted by the surface. The structure of this near-surface branch of the overturning mass transport circulation has now been analyzed theoretically, supported by a diagnosis of the circulation in the R30 atmospheric model (gm). This analysis helps explain how the structure of the return flow is related to the probability distribution of the surface potential temperature (Fig.1.21). In particular, an explanation is given for the fact that much of the equatorward mass transport occurs in isentropic layers that are colder than the mean surface potential temperature. This analysis also leads to an alternative formulation of the transformed Eulerian mean circulation in height-coordinates that corresponds more closely to the isentropic overturning.

1.8 PLANETARY FLUID DYNAMICS

ACTIVITIES FY98

[ Section omitted at author's request ]

  1. Mitchell, J. H., T. C. Johns, J. M. Gregory, and S. Tett, Climate response to increasing levels of greenhouse gases and sulfate aerosols, Nature, 376, 501-504, 1995.
  2. Alley, R. B., P. A. Mayewski, T, Sowers, M. Stuiver, K. C. Taylor, and P. U. Clark, Holocene climatic instability: A prominent, widespread event 8200 years ago, Geology, 25, 483-486, 1997.
  3. Schiller, A., U. Mikolajewicz, and R. Voss, The stability of the North Atlantic thermohaline circulation in a coupled ocean-atmosphere general circulation model, Clim. Dyn., 13, 325-347, 1997.
  4. Johnson, T.C., C.A. Scholz, M.R. Talbot, K. Kelts, R.D. Rickets, G. Ngobi, K. Beuning, I.Ssemmanda, J.W. McGill, Late Pleistocene desiccation of Lake Victoria and rapid evolution of Cichlid fishes, Science, 273, 1091-1093, 1996.
  5. Broecker, W.S., D. Peteet, I. Hajdas, J. Lin, E. Clark, Antiphasing between rainfall in Africa's Rift Valley and North America's Great Basin, Quat. Res., in press.
  6. Holloway, G., Estimation of oceanic eddy transport from satellite altimetry. Nature, 323, 243-244, 1986.
  7. Farrell, B., and P. Ioannou, Stochastic dynamics of the midlatitude jet. J. Atmos. Sci., 52, 1642-1656, 1995.


Back to Table of Contents

*Portions of this document contain material that has not yet been formally published and may not be quoted or referenced without explicit permission of the author(s).