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

 

Table of Contents 

4. EXPERIMENTAL PREDICTION

GOALS

4.1 FLEXIBLE/MODULAR FORECAST MODELING SYSTEM

ACTIVITIES FY97

4.1.1 Atmospheric Model Development

4.1.1.1 Global Atmospheric Grid Point Model

Development of the B-grid atmospheric general circulation model (AGCM) has progressed, addressing two distinct, but interrelated components: the dynamical core, and the physical parameterizations (4.1.1.3). The B-grid dynamical core retains most of the features of its E-grid predecessor (1351), while adding arbitrary horizontal and vertical resolution, tracer advection and diffusion, fourth-order vertical advection, improvements in horizontal diffusion, and output of model data in NetCDF (network Common Data Form) format. A large number of multiyear integrations of this model with both sigma and eta vertical coordinates have now been performed to test the Fortran90 implementation of the modular physical parameterizations (4.1.1.3).

4.1.1.2 Flexible Spectral Model

Development of a new Fortran90 spectral dynamical core has continued with emphasis on testing various configurations (horizontal resolution from T30 to T106 and vertical resolution from 20 to 100 levels, with sigma or hybrid vertical coordinates). Test have been run with simple moist physics, including large-scale condensation and convective adjustment, providing detailed documentation and improving performance. Several graduate students have used this code successfully in their research projects. Tests with realistic lower boundary conditions and a full physics package selected from the modular parameterizations (4.1.1.3) are underway. Extensions under active consideration include semi-implicit zonal advection, non-interpolating semi-Lagrangian vertical advection, and generalization of the vertical coordinate to allow a transition from sigma-coordinates near the surface to isentropic coordinates aloft.

4.1.1.3 Modular Physics Parameterizations

Work has continued on improving existing parameterizations while incorporating new modular parameterizations. The testing of changes continues to be performed in the B-grid AGCM (3.2.1.1). Modifications have included preliminary work on a prognostic cloud water and ice scheme (bw), arbitrary soil layer resolution, shallow convection scheme, improvements in the Mellor-Yamada turbulent closure scheme, an adjustable time step option for both convective schemes and large-scale condensation, and time-averaged input data for radiation and prognostic clouds. Additional work has been done on intermediate level routines that call the physical parameterization routines and output diagnostic quantities in NetCDF format.

A modular version of the full Arakawa-Schubert cumulus parameterization scheme is near completion. When completed, a version of the scheme with a prognostic cloud work function will be implemented. This will greatly reduce the cost of using the Arakawa-Schubert scheme.

4.1.1.4 Support Tools for Modular Models

A number of additional support tools for the flexible/modular modeling system have been completed. A general facility for creating a complete modular model from a variety of separately developed and managed components has been developed on both the Cray T90 and on the SGI workstations. This system allows a single copy of a module's source code to be used to build a variety of models on either computing platform. A time and calendar manager has been completed and is currently being incorporated in both atmospheric dynamical cores, as well as the modular atmospheric physics and the MOM 2 ocean model.

4.1.1.5 Coupled Model Development

A framework for coupling atmosphere, ocean, land surface, and ice models for both climate and seasonal/interannual prediction purposes has been developed. A set of modular tools for implementing this coupling framework is under development and has been tested in simple general circulation models.

PLANS FY98

Work will continue on the flexible/modular modeling systems. The tools for coupling models will be completed and used to construct one or more coupled general circulation models for climate and seasonal/interannual prediction. Additional physics modules for the atmosphere will be completed. These new modules, including gravity wave drag, a new radiation code, and improvements to the prognostic cloud water and ice scheme, will be tested using both the flexible spectral and B-grid dynamical core models. The land surface models will be removed from the atmospheric models and recreated as independent model components. Coupled and atmosphere-only models based on the flexible/modular cores will replace the current spectral model as the primary research tool for the experimental prediction group (although integrations as part of the coupled model ensemble prediction experiment (4.2.1) will continue with the old coupled model).

B-grid model development will focus on testing additional physics modules, optimizing performance, and reducing model bias. Longer integrations with higher resolution will be run to examine the sensitivity to physical parameterizations and the differences between the sigma and eta vertical coordinates.

The performance of the flexible spectral core model will be improved on GFDL's conventional vector and parallel computing platforms.

4.2 MODEL DEVELOPMENT FOR SEASONAL/INTERANNUAL PREDICTION

ACTIVITIES FY97

4.2.1 Development of Atmospheric Subgrid-Scale Parameterizations

The land surface parameterization being used in the NCEP/MRF (National Centers for Environmental Prediction/Medium Range Forecast) model has been obtained from NCEP. This parameterization includes some of the effects of vegetation type and soil type on the surface fluxes without the complexity and cost of other land surface models such as SiB (Simple Biosphere model). This scheme has been tested with off-line runs and with several one year runs in the full spectral GCM. The implementation of this scheme at GFDL involved a fruitful collaboration with Hua-Lu Pan at NCEP which has resulted in an improved parameterization at both GFDL and NCEP.

A prognostic cloud water parameterization, based upon the Del Genio formulation, has been developed and tested. The scheme incorporates bulk parameterizations of various cloud microphysics processes. Realistic simulations of climate, including the distributions of cloud water and of the earth radiation budget have been obtained, after adjusting a few parameters and incorporating the RAS (Relaxed Arakawa Schubert) parameterization of cumulus convection.

4.2.2 Ocean Model Development

4.2.2.1 Ocean Model Simulations

Assessing the error in the mean fields for the coupled model and reducing that error for the upper ocean remain key elements in the improvement of seasonal/interannual forecasts. Determining whether the source of bias is the surface forcing or the ocean model remains a challenge. Three main avenues of research have been applied to this problem: 1)evaluation of the sensitivity of the ocean model to surface forcing; 2) assessment of the impact of subgrid scale parameterizations (4.2.2.2); and 3) variation of the model grid resolution.

Because the surface fluxes drive the ocean circulation, information regarding their spatial and temporal variation is essential to understanding and modeling ocean variability over interannual time scales. Therefore, ocean model simulations are being run using surface fluxes from the reanalysis products of NCEP, NASA, and ECMWF (European Centre for Medium Range Weather Forecasts), and winds from Florida State University. Fluxes obtained from the atmospheric model, using observed SSTs, are also part of the study. The period covered by these experiments is 1979-1996.

4.2.2.2 Improved Physical Parameterizations

A major restructuring of the GFDL ocean model physical parameterizations has been proceeding for roughly two years. Crucial to this effort is the implementation of new tracer advection schemes, a reformulation of isoneutral tracer diffusion (em), a simplified implementation of the Gent-McWilliams parameterization (ev), and the implementation of the KPP (K-Profile Parameterization) vertical mixing scheme. The manner in which these schemes interact with each other is currently being investigated in a suite of idealized and realistic ocean model experiments.

Observations suggest that the role of tropical instability waves (TIWs) in the maintenance of the equatorial cold tongue in the east Pacific may be comparable to surface heat flux contributions. The sensitivity of the TIWs to subgrid scale mixing is being investigated. Water mass distribution in the equatorial thermocline has been shown to be sensitive to mixing schemes. Temperatures may vary by as much as 10 degrees between relatively warm, saline waters of South Atlantic origin, and colder, fresher water of the North Atlantic. The ability of the models to reproduce observed water mass distributions within the equatorial thermocline may be an important component in simulating tropical sea surface temperatures.

As a complement to the above, an idealized channel model has been constructed in which a portion of the ocean mesoscale eddy spectrum is explicitly represented. One purpose of this study is to understand the interaction of oceanic free convection with baroclinic eddies. The parameterization of this interaction turns out to be a crucial element in how the mesoscale eddy parameterizations are coupled to the parameterizations of oceanic mixed layers.

PLANS FY98

The prognostic cloud water scheme will be further developed and tested in coupled models using both the gridpoint and spectral atmosphere models. The development of land surface parameterizations appropriate for seasonal/interannual prediction will be accelerated.

The evaluation of the sensitivity of the ocean model to surface fluxes from the various reanalysis products will continue. The effects of subgrid-scale parameterizations on the upper ocean thermodynamic balances will be a focus of ocean model development. A comparison of model eddy flux convergence with drifter data and TAO (Tropical Atmosphere Ocean) mooring observations within the equatorial Pacific will be a powerful tool for diagnosing and improving model physics.

4.3 ATMOSPHERIC AND OCEANIC PREDICTION AND PREDICTABILITY

ACTIVITIES FY97

4.3.1 Coupled Model Ensemble Prediction Experiments

4.3.1.1 Experimental Design

A large set of atmosphere-only integrations and coupled model forecasts has been initiated in order to create a dataset which can be evaluated to better understand and improve the capabilities of seasonal/interannual forecast models. The first component of the experiment is an ocean data assimilation from 1979 through the present, made using a version of the GFDL MOM 2 ocean model. The SSTs from the ocean data assimilation are then used to force atmosphere-only integrations of the seasonal/interannual prediction spectral model. An ensemble of these atmosphere-only integrations is generated by slightly perturbing the initial atmospheric conditions for 1 January 1979. The ensemble of predictions is then made using ocean and atmosphere models that are identical to those used in the assimilation and the atmosphere-only integrations. The ocean initial conditions for the forecasts come from the ocean data assimilation (1457), while the atmospheric initial conditions come from the atmosphere-only integrations (1450). Coupled model predictions are made from initial conditions for 1 January, 1 April, 1 July, and 1 October and extend to lead times of up to a year or more. The initial ensemble size will be five members although this may be expanded if results warrant. A large set of auxiliary integrations of the coupled and atmosphere only models is also planned. These auxiliary experiments include: a set of coupled model forecasts with observed initial conditions; a large suite of one day integrations to be used in systematic error correction experiments; and an extended integration of the coupled model to study its internal variability. Currently, four ensemble members of both the atmosphere-only runs and coupled model forecasts are underway.

4.3.1.2 Frozen Spectral Atmospheric GCM

A stable atmospheric model configuration is an essential part of the coupled model ensemble prediction integrations involving both the atmosphere-only GCM (AGCM) and the coupled GCM (4.3.1.1). Implementation standards were established that included scientific, computational, operational, and documentation objectives. Primary amongst these objectives was the construction of a coupled model with state-of-the-art skill in seasonal/interannual forecasts of the tropical oceans without the use of explicit flux corrections. The major changes to the model physics from that used in the AMIP I ensemble (1300) were Arakawa Schubert convection (including a diurnal cycle with a complete radiation calculation every two hours) and the use of filtered orography to greatly reduce Gibbs error. These (and other) changes to the model physics were subjected to a series of coupled model test cases with the goal of determining an acceptable model configuration from a scientific perspective. In addition to the scientific considerations, model, archival and on-line post processing efficiencies were implemented to reduce the wall clock time per model month from nearly 16 hours to approximately 3.5 hours. In the spirit of assuring operational consistency between ensemble members and between the AGCM and the coupled GCM, a unified frozen model production run script was developed which uses frozen model executables. Finally, a description of the AGCM and the experimental implementation is currently in progress as a document accessible via the World Wide Web http://www.gfdl.gov/~jla/cmep.html).

4.3.2 Impact of Improved Seasonal Cycle on ENSO Forecasts

A suite of 13-year simulations was made with a coupled model for various treatments of low clouds. This included: 1) seasonally-varying low clouds specified from the ISCCP (International Satellite Cloud Climatology Project) over the oceans, but predicted over land; 2) predicted low clouds everywhere; 3) a hybrid configuration with ISSCP low clouds specified only over the eastern tropical Pacific; and 4) the absence of any low clouds over sea or both land and sea. Analysis of the results is continuing to clarify the mechanisms which control the annual cycle of SST and variations of SST on the ENSO time scale (1403).

The coupled model is very sensitive to low cloudiness over the eastern tropical Pacific. The integrations with ISCCP low clouds in that region are improved, generating colder SSTs and a larger amplitude annual cycle than those with predicted clouds, which fail to adequately simulate the marine stratus regime. In addition to the direct radiative response, stronger west-east (and north-south) temperature gradients, and hence stronger tradewinds and southerly meridional wind stresses, are set up in the former group of experiments. The stronger trades and associated equatorial upwelling push the cold tongue further west, inhibiting the eastward progression of warmer water from the west. When the low clouds are removed, the trades and west-to-east SST gradient weaken dramatically. The western tropical Pacific warms by ~2 to 3C, while 28C and at times, even 29C water extends across the eastern equatorial Pacific.

4.3.3 Sensitivity to Subgrid-Scale Parameterizations

4.3.3.1 Frequency of Radiation Calculations and Inclusion of Diurnal Variation

The frequency with which clouds and radiation are computed was increased from 12hours to 2hours. Diurnal variation was then added to both the AGCM and coupled models. The tropical zonal mean cloud-radiation response to the first change was more excessive cloud cover, and reductions of ~ 30W m-2 in OLR and ~ 20W m-2 in absorbed short wave radiation, which was surprisingly large, and unfavorable compared to observations. Similar results were obtained in coupled and uncoupled model integrations. In coupled runs, the SST warmed in the tropical eastern Pacific and convective activity shifted eastward. This change in the cloud-radiation response is presumably due to stronger interaction between clouds, radiation, and convection in the case with more frequent calculations. Also, in situations where the primary cloud predictor variable, relative humidity, oscillates about its threshold value, the shorter 2-hour time-averaging interval enhances cloud amount. Fortunately, when diurnal variation was switched on, this tropical sensitivity was greatly diminished, especially during non-transition seasons. Diurnal variation apparently weakens the radiative-convective interaction by modulating the intensity of the convection.

4.3.3.2 Modification of Cloud Prediction Scheme

Simulations by the coupled model have been observed to produce excessively warm SSTs over the higher middle latitude northern oceans, particularly over the North Pacific in summer. Insufficient cloud cover, with a maximum impact on solar insolation in summer, has long been suspected as a contributing cause. In an attempt to remedy this, the inhibiting effect of weak ascending or descending vertical motion on low cloud amount in the model's cloud prediction scheme was partially relaxed. In response, the summer bias of zonal mean absorbed short wave radiation with respect to ERBE (Earth Radiation Budget Experiment) observations decreased from greater than 40 W m-2 to ~15 W m-2 near latitude 50N. As a result, the surface insolation decreased over portions of the North Pacific by 25 to 40 W m-2, improving agreement with observations. In turn, the North Pacific summer warm SST bias decreased by more than 4C near latitude 50N (Fig. 4.1).

4.3.4 Tropical Intraseasonal Variability

A preliminary comparison of space-time power spectra for 850 (velocity potential at 850 hPa) from coupled and atmosphere-only GCMs indicates that the dominant modes for tropical intraseasonal oscillations (TIO) appear to have greater amplitude in the coupled GCM. The periods with largest amplitude are centered in the range of 40-50 days in the coupled model, versus about 30 days in the AGCM (zs). This can be seen in Fig. 4.2 which plots space-time power spectra for 850 for two individual years, 1985 and 1986. In this plot, the coupled results for an individual simulation are compared to a nine member ensemble of the AGCM (T42L18 resolution). In both these years (but especially in 1985), the spectral peak from the coupled prediction appears to be outside the range of individual spectral peaks from the AGCM ensemble members (cj). Based on these preliminary results, it is speculated that air-sea interaction as part of the evaporation-wind feedback mechanism plays a key role in the difference in amplitude and frequency of TIO in the coupled model.

4.3.5 Interannual Variability in Atmospheric Models

The eastward propagation of the ENSO signal in the spectral model and in the atmosphere has been investigated using a variety of techniques (ep), including the rotated complex principle component analysis and the Hayashi analysis. Insights from this analysis are aiding in an assessment of deficiencies of the coupled model's prediction of ENSO.

4.3.6 Simulation of Tropical Storm Frequency

A nine-member ensemble of 10-year integrations of a spectral AGCM forced by observed SSTs has been studied to evaluate the skill of a GCM in simulating interannual variability of tropical storm frequency (1455). An EOF (empirical orthogonal function) analysis of vertical wind shear, 850 mb vorticity and 200 mb vorticity has been performed in order to investigate the impact of the large scale circulation on simulated tropical storm frequency. The simulated large scale circulation has a significant impact on the simulated tropical storm frequency, intensity, and interannual track variability, in agreement with observations. The differences between simulated and observed tropical storm statistics over some ocean basins, particularly in the Southern Hemisphere, may be explained by differences between simulated and observed large scale circulation.

4.3.7 Reproducible Modes in Forced Atmospheric Model Ensembles

A method for assessing the potential predictability of the extratropical atmospheric seasonal variations in an ensemble of AGCM integrations has been developed which involves isolating reproducible forced modes and examining their contributions to the local ensemble mean. An EOF decomposition applied to the anomalies from the ensemble mean identifies some forced modes that are less affected by internal processes and thus appear to be highly reproducible. By developing a quantitative measure, the potential predictability index (PPI), which combines reproducibility (1300) with the local variance contribution, the local ensemble mean over selected areas in the extratropics was shown to result primarily from reproducible forced modes rather than internal chaotic fluctuations. Thus, the ensemble mean is potentially predictable over those areas, mainly over North America and part of the Asian monsoon regions (Fig. 4.3). Interestingly, the potential predictability over some preferred areas such as the Indian Monsoon region and central Africa occasionally results primarily from non-ENSO-related boundary forcing, although ENSO forcing generally dominates over most of the preferred areas. The PPI analysis has also shown that the preferred geographic areas have obvious seasonality. The boreal summer season possesses the largest potentially predictable area while the boreal winter appears to have the least.

4.3.8 Dynamics of Low-Order Coupled Systems

The Cane-Zebiak coupled model is being used to evaluate the validity of using forced atmospheric ensembles to evaluate the potential predictability of the coupled ocean atmosphere system. In this model, the ocean is highly constrained by the prescribed wind stress. An ensemble of forced integrations therefore tends to overestimate the predictability of the coupled system. Extending these results to more realistic models will aid in the interpretation of atmosphere-only integrations forced by observed SSTs.

4.3.9 North Atlantic Climate Predictability

The North Atlantic is one of the few places on the globe where the atmosphere is linked to the deep ocean through air-sea interaction. While the internal variability of the atmosphere by itself is only predictable over a period of one to two weeks, climate variations are potentially predictable for much longer periods (months or even years) because of coupling with the ocean. The current study provides a quantification of the predictability for simulated multidecadal climate variability over the North Atlantic (cm).

The model used for this investigation is the coarse resolution R15L9 GFDL coupled ocean-atmosphere climate model, which, nevertheless, captures fluctuations of the North Atlantic and high latitude oceanic circulation with variability concentrated in the 40-60 year range. Oceanic predictability is quantified through analysis of the time-dependent behavior of large scale empirical orthogonal function (EOF) patterns for various model fields. The results indicate that predictability in the North Atlantic depends on three main physical mechanisms.

The first mechanism involves the oceanic deep convection in the subpolar region which acts to integrate atmospheric fluctuations, thus providing a red noise oceanic response as elaborated by Hasselmann. The second involves the large-scale dynamics of the thermohaline circulation, which displays variations which have an oscillatory character on a multidecadal time scale. The third involves nonlocal effects on the North Atlantic arising from periodic anomalous fresh water transport advecting southward from the polar regions in the East Greenland Current. When the multidecadal oscillatory variations of the thermohaline circulation are active, the first and second EOF patterns for the North Atlantic dynamic topography have predictability time scales on the order of 10-20 years, whereas EOF-1 of SST has predictability time scales of 5-7 years. When the thermohaline variability displays weak multidecadal fluctuations, the Hasselmann mechanism is dominant and the predictability is reduced by at least a factor of two. When the third mechanism is in an extreme phase, the North Atlantic dynamic topography patterns realize a 10-20 year predictability time scale. Additional analysis of SST in a region of the Greenland Sea associated with the southward propagating fresh water anomalies indicates the potential for decadal scale predictability for this high latitude region as well. The model calculations also provide insight into regional variations of predictability, which might be useful information for the design of a monitoring system for the North Atlantic. Predictability appears to break down most rapidly in regions of active convection in the high latitude regions of the North Atlantic.

PLANS FY98

At least five ensemble members of the atmosphere-only and coupled model forecasts for the coupled model ensemble prediction experiment will be completed. Extensive analysis of the results will begin.

More detailed and comprehensive analyses of TIOs from the coupled model prediction ensembles will be pursued, with the goal of obtaining a better understanding of the relationship between TIOs and ENSO, and the role that TIOs play in coupled GCM predictions of ENSO.

The interdecadal sensitivity of ENSO forecast skill to realistic marine stratus clouds will be further examined using the frozen model. The sensitivity of the coupled model's annual cycle and its interannual response to various treatments of low cloud forcing will be further explored.

The sensitivity of the Asian Summer Monsoon circulation to changes in land surface forcing over the Tibetan Plateau will be studied in a coupled model with realistic cloud forcing given by a three-dimensional specification of ISCCP clouds.

The equatorial temporal variability of surface fluxes and other variables from long term coupled and uncoupled GCM integrations will be analyzed. Results from the uncoupled integrations (with specified SSTs) will be compared with NCEP reanalyses and COADS analyses.

4.4 DATA ASSIMILATION

ACTIVITIES FY97

4.4.1 Ocean Data Assimilation

As part of the coupled model ensemble prediction activity (4.3.1), the ocean data assimilation system was run from 1979 to 1996, saving daily SST analyses for the atmospheric runs and producing ocean initial conditions for the coupled model forecasts. TOPEX sea surface height data from JPL (Jet Propulsion Laboratory) was compared with the assimilation. One goal of this comparison was to evaluate how much of the difference between these initial conditions and the Topex sea surface height (SSH) might be due to salinity, which is not assimilated in the GFDL system. The climatological salinity-restoring boundary condition in the assimilation was then replaced with an observed water flux from the NCEP reanalysis. Sensitivity in the SSH field in the Pacific warm pool indicated the importance of good salinity measurements in that region.

4.4.2 Ensemble Data Assimilation

A fully nonlinear filtering system for data assimilation has been developed in low order models to create initial conditions for ensemble forecasts (1355). The filter produces a random sample of the probability distribution of the state of a dynamical system conditional on all previous observations of the system. This filtering system is the nonlinear version of the Kalman filter that has been used for some atmospheric data assimilation applications. The nonlinear filter can represent arbitrarily complex probability distributions and does not require the inversion of the operator that maps from the assimilating model's grid to simulated observations. It is a trivial task to apply this filtering system in low order models to produce a consistent set of ensemble initial conditions. These initial conditions can be used to evaluate the quality of ensemble initial conditions produced by other methods (ca). Attempts to extend the nonlinear filter to higher order forecast models have been successful, although a number of issues related to computational efficiency remain to be resolved.

PLANS FY98

The impact of salinity on the ocean data assimilation will be further clarified. The ocean data assimilation system will be used to study systematic errors in physical parameterizations, especially vertical mixing schemes.

The nonlinear filter for data assimilation will be extended to larger and more realistic forecast models. A number of possible heuristic simplifications can be tested to see if a practical operational ensemble data assimilation system can be obtained in this framework.

4.5 OCEAN-ATMOSPHERE INTERACTION

ACTIVITIES FY97

Despite the rapid progress during the 1980s in our ability to explain, simulate and predict the Southern Oscillation/El Nio phenomenon, the 1990s are bringing several surprises. The decade started with the unanticipated and prolonged persistence of warm surface waters over the eastern tropical Pacific. Next came El Nio of 1997 that some models failed to predict (even though those models succeeded in the prediction of earlier events). Analyses of time-series data from the TOGA-TAO array (bo) suggested that, between the 1980s and 1990s, there was a change in the properties of the equatorial thermocline because of a change in the processes that maintain it. Specifically, the thermocline in the eastern equatorial Pacific deepened because of a change in the oceanic exchanges between the tropics and extratropics. These exchanges, and their influence on interactions between the ocean and atmosphere, are being investigated because they critically affect a variety of phenomena ranging from decadal climate fluctuations to conditions during glacial and interglacial epochs.

The sharp, shallow tropical thermocline, the salient feature of the thermal structure of the tropical oceans, is maintained by a shallow meridional circulation driven by easterly winds that cause poleward Ekman flow in the surface layers, and equatorward geostrophic flow in the thermocline in response to the eastward pressure force established by the westward winds. Upwelling at the equator and subduction in the extratropics close the circulation. Studies with a realistic oceanic GCM show that one important subduction zone off the coast of Peru provides water parcels with a relatively direct window to the equatorial thermocline. Parcels that subduct in another important zone, off the coast of California, follow a circuitous route to the equator. First, they travel southwestward to approximately 10N and then either join an equatorward western boundary current or flow eastward in the North Equatorial Countercurrent while slowly moving equatorward. Presumably, these are the routes available to disturbances that bring unusually cold or warm waters from the surface layers of the extratropics to the equatorial thermocline. Initial results from experiments, with the MOM model forced with heat and momentum fluxes appropriate for the period 1970 onwards, indicate that, whereas interannual variations in the tropics are locally forced, decadal variations in low latitudes have an extratropical origin.

While the GCM studies described above are underway, consequences of variations in the exchanges between the tropical and extratropical oceans are being explored. Consider a disturbance in the form of unusually warm surface waters over the northern Pacific Ocean. Those waters converge onto the subduction zone and in due course find their way to the equatorial thermocline. This could be the origin of the warming of the eastern Pacific equatorial thermocline evident in the TOGA TAO time-series of the early 1990s. This change in the thermocline affects ocean-atmosphere interactions in such a manner as to result in conditions similar to El Nio: warm surface waters and relaxed easterly winds in the tropics, an equatorward shift of the jet stream, and relatively cold surface waters in the extratropics. In other words, the original warm disturbance in the extratropics in due course results in a cold disturbance in that region. These arguments imply decadal climate fluctuations that have been explored by means of a simple box-model of the ocean-atmosphere system (bo).

Consider next a disturbance in the form of unusually cold surface waters over the extratropics. This case is relevant to conditions during the last glacial maximum some 18,000 years ago and can shed light on the controversy concerning temperatures in the tropics during that phenomenon. Although paleoclimatologists at first believed that the tropics then were as warm as they are today, recent observations indicate that the tropics were significantly cooler, by as much as 5C. A simple coupled ocean-atmosphere model, of the type used to predict El Nio, has been used to explore how a cooling of the equatorial thermocline would affect ocean-atmosphere interactions and hence sea surface temperatures in low latitudes. The results indicate that the oceanic exchanges between the tropics and extratropics mentioned above can indeed result in a significant cooling of the tropics during glacial climates.

PLANS FY98

The development of coupled GCMs capable of realistic simulations of seasonal, interannual, and decadal climate fluctuations will continue. An attempt will be made to establish what must be known about an atmospheric model in order to anticipate how it will behave when coupled to a given ocean model. To this end, results from several different atmospheric models, all coupled to exactly the same ocean model, will be analyzed. The converse experiment will also be performed, coupling the same atmospheric model to different ocean models. In parallel, exploration of the tropical-extratropical exchanges that determine the properties of the equatorial thermocline will continue.

 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).