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

 

3. EXPERIMENTAL PREDICTION

GOALS

3.1 FLEXIBLE/MODULAR MODELING SYSTEM

ACTIVITIES FY98

3.1.1 Atmospheric Model Development

3.1.1.1 Global Atmospheric Grid Point Model

A highly flexible and modular version of the B-grid dynamical core (1351) with more user-friendly interfaces has been developed using Fortran-90 enhancements. This code has been optimized on the GFDL Cray T90 with a reduction of about 20 percent in CPU time. The code has been incorporated into the full Atmospheric General Circulation Model (AGCM) and modular physics (3.1.1.3) and has been used in developing the framework for the new flexible coupled models (3.1.2).

3.1.1.2 Flexible Spectral Model

The flexible spectral dynamical core has undergone continued development to improve efficiency on the Cray T90 and to make the spectral model interfaces consistent with those of the B-grid model (3.1.1.1). Tests with various physical parameterizations originally developed in the B-grid context have been completed. Extensions to incorporate a semi-Lagrangian advection scheme are being considered; initial tests with such advection schemes in simple models have been completed.

3.1.1.3 Modular Physics Parameterizations

A modular version of the SiB (Simple Biosphere Model) land surface parameterization has been created and tested in off line runs. Development of a modular version of the full Arakawa-Schubert cumulus parameterization scheme has been completed.

3.1.1.4 Spectral Model Parallelization

The design of GFDL codes is evolving apace with computer architectures and compiler technologies. In particular, codes are now in transition from an era dominated by vector supercomputers to one where massively parallel processing architectures may dominate. The new spectral core is likely to be of vital importance to GFDL in coming years, and will be a key component of a coupled ocean-atmosphere model now under development. Substantial effort has been devoted to its parallelization for distributed memory architectures.

The relevant feature of spectral models is that each model field has a spectral and a grid representation, and operations are performed on each representation at each timestep. Linear terms are treated in the more succinct spectral representation, while physical parameterizations are applied on the grid representation. The problem arises with non-linear terms, such as in advection, which, if explicitly expanded, result in a sum of terms quadratic in the order of the expansion. The transform method is used to convert spectral fields to grid space for the computation of non-linear terms and then back. Since this method involves frequent transformations between grid fields and spectral fields, efficient numerical transform methods have been developed, including parallel methods.

In keeping with the modular approach, a data decomposition and communication tool called "mpp_mod" has been developed for the spectral core. This module has a flexible interface for specifying decompositions of the global grid among processors. Conceptually, the module distinguishes the "computational domain" (the set of grid points for which computations are to be done on any processor in a distributed environment) and the "data domain" (the set of points whose values need to be available in order to carry out the computation). If points in the data domain have been altered, the data domain must be updated, by acquiring data as required from other processors, before the computation can go forward. The "mpp_mod" module maintains the processor map as a linked list, and contains routines for managing these communications with a straightforward interface. Internally, communication is carried out on the GFDL Cray T3E using either the MPI standard (http://www.mcs.anl.gov/mpi) or the Cray-native SHMEM library, which is faster, but proprietary. The domains described here are conceptually general, and can be used to specify computational domains with halo regions for distributed grid-point models as well.

The spectral core is now being parallelized using the approach above. Scaling studies have been carried out at various resolutions (R30, T42, T63, T106, T213) using a one-dimensional decomposition where spectral fields are distributed along the Fourier wavenumber and grid fields along latitude, requiring a data transposition between the Legendre and Fourier transforms, which are carried out on-processor. The code is now under modification to be run in a shared-memory parallel mode as well.

3.1.2 Coupled Model Development

In order to support coupled modeling in a flexible framework, it is desirable to eliminate, as much as possible, the impacts of decisions made in one component model, say the ocean, on another component model, for instance the atmosphere. A coding framework for allowing the coupling of component models with arbitrary grids is under development. One result of this effort that is new to GFDL is that models of the land surface can be developed on their own grid and mostly independently of atmospheric models. The framework also minimizes the impact of design choices in atmosphere, ocean and ice models on each other.

To support this coupled model coding framework, a tool for conservative interpolation between arbitrary grids has been developed. In its present form, this tool accepts information describing the grids of various component models and creates an interpolation mapping that can be used to transfer information between the different model grids. The module has been developed so as to isolate the impacts of parallel computer architectures in a small portion of the interpolation code. Initial versions of atmosphere-ocean-ice-land models coupled with this tool have been completed and successfully integrated. The interpolation tool is relatively efficient on vector architectures and efforts to produce efficient MPP implementations are underway.

3.1.3 Support Tools for Modular Models

Work has continued on a number of software tools to support the development of flexible models of components of the climate system (atmosphere, ocean, land, ice). The time and calendar manager completed last year has been incorporated successfully into the MOM ocean model and into the grid point and spectral model cores. A flexible interface using the facilities of Fortran-90 to read and write NetCDF format files has been completed and incorporated into the atmospheric dynamical cores and physics packages. An improved version of the modular model compilation tool has been developed and is in use to create complete modular models from separately developed and managed components on both the Cray T90 and 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 and does not require the creation of any additional files (such as Makefiles). An HTML-based system for documenting flexible modeling components is under development. In addition to documenting individual component modules, this system interacts with output of the compilation tool to produce a set of graphical analyses of the components of a particular model configuration.

PLANS FY99

Complete atmosphere-land-ice-ocean coupled GCMs using both B-grid and spectral dynamical cores and the coupling tool will be developed and integrated.

Atmosphere and coupled integrations with flexible GCM dynamical cores will be performed as the flexible modeling system becomes the operational tool for experimental prediction at GFDL. Climate versions of these models will be developed in parallel and evaluated by GFDL climate groups.

New physical parameterizations, for clouds in particular, will be completed and tested in the flexible modeling system.

Work will continue on producing scalable versions of the spectral dynamical core that can be run efficiently on a variety of architectures. The B-grid dynamical core will be converted to a scalable form.

3.2 MODEL DEVELOPMENT FOR SEASONAL/INTERANNUAL PREDICTION

ACTIVITIES FY98

3.2.1 Sensitivity to Subgrid-Scale Parameterizations

Removal of an upper troposphere-lower stratosphere cap on convection in the RAS cumulus convection scheme yielded a more realistic coupled model simulation of the ITCZ and a reduced SST cold bias in the western tropical Pacific (cl) (Fig. 3.1). Conversely, the model's tropical tropopause rose to 50 hPa, while the 100 and 200 hPa levels warmed by approximately 10K and 6K, respectively. These unrealistic responses had adverse consequences for the atmospheric general circulation. In an attempt to have the best of both worlds at the surface and near the tropopause, the high cloud component of the atmospheric model's cloud prediction scheme was re-tuned. Two modifications produced rather dramatic results. First and foremost, the quadratic, temperature-dependent parameterization of optical depths of cold, cirrus clouds (based upon a scheme of Harshvardhan) was re-activated for the non-anvil class of high clouds. This modification reduced the emissivity of high clouds in convectively inactive regions, including those in the tropics. Second, the bases of high clouds were permitted to reside one sigma layer beneath their tops, if warranted by the local vertical profile of relative humidity. This modification altered the vertical profile of long wave radiative heating/cooling generated by the model's high clouds. Hence, the spatial pattern of outgoing longwave radiation (OLR) improved and the warm bias at 100 hPa and 200 hPa decreased at least to the levels experienced in the previous version of RAS, while the ITCZ and SST improvements were retained.

3.2.2 Ocean Model Simulations

The verification of the ocean model is complicated by the strong dependence of the model solution on surface forcing. This is a dilemma for tropical ocean modeling in particular due to the strong interaction between planetary wave dynamics in the thermocline and surface mixed layer processes. In order to study the ocean model response to surface forcing, subgrid scale parameterization, and variation of model grid resolution, several experiments were run. The model domain covered the Pacific basin and the simulation period was 1979-1997. Some of the more notable sensitivity tests were:

Evaluation of these experiments was performed in the context of comparisons to observed data, i.e., TAO (Tropical Atmosphere Ocean) moorings, TOPEX altimetry, and COARE data. Some of the current phenomenological studies using these simulation runs are:

3.2.3 Hybrid Coupled Model

Traditional ocean model development has involved using a prescribed boundary condition for momentum and a constraint on sea surface temperature (SST) and salinity (SSS) values through linear damping terms. Such constraints on the ocean do not allow for a complete evaluation of the ocean model in the context of a fully coupled GCM. A statistical atmosphere has been implemented to help further ocean model development for seasonal to interannual tropical forecasts. A singular vector decomposition of the observed wind vector and SST covariance matrix is performed in order to extract patterns of covariance in these fields. The issue of the SST and SSS constraints are addressed through the use of linear damping terms using empirically derived heat flux/SST relationships.

Surface heat fluxes are not well observed and are quite sensitive to closure assumptions for the near-surface atmospheric boundary layer, as well as assumptions about the impact of clouds on the surface radiation budget. As a result, drift is inevitable due to imbalances in the ocean/atmosphere heat flux requirements. By running the component models uncoupled, but tied to observed temperatures near the ocean surface, the corresponding heat fluxes can be compared and will, in the absence of perfection, differ. Such imbalances are the basis for "flux adjustments" used in some climate simulations. One can imagine that the climate system is somehow tied to a climatological state at the surface (i.e., a mean annual cycle) and a damping restoring force exists which leads the system back to the climatological state. This is, in fact, verifiable in the tropics owing to the influence of the Clausius-Clapeyron relation, which means the saturation vapor pressure of air immediately above the ocean surface increases with rising SST and leads to an increase of turbulent latent heat transport to the atmosphere, thus reducing SST. While cloud and wind feedbacks can mitigate this process, such a constraint seems to be justified in the tropics. This is not the case in higher latitudes where, in general, circulations are more complicated owing to the variability associated with synoptic disturbances. Local regressions of surface heat flux anomalies (with the mean annual cycle removed) to interannual SST anomalies are performed using the NCEP reanalysis and the experimental prediction group atmospheric model. These regressions are used in the hybrid coupled model runs in addition to the wind SST relationship. Regression values for SSS are set to a constant value since a clear relationship between SSS and evaporation/precipitation anomalies is not observed.

The hybrid coupled model allows for a free coupled system, not constrained too strongly to climatology, which can mimic to some extent the behavior of a fully coupled GCM (with a flux adjustment). This allows for a more complete assessment of the impacts of ocean sub-grid parameterizations and resolution on tropical variability.

3.2.4 Correction of Systematic Errors in Coupled GCM Forecasts

A method called the prognostic tendency (PT) correction is used to reduce systematic errors in coupled GCM forecasts with realistic initial conditions. The idea is simple: assess the systematic prognostic tendency error (STE) of the coupled model and subtract it from the discrete prognostic equations. The STE can be estimated by calculating a climatologically-averaged tendency between the forecast value at a very short lead time and the observed initial value and discarding the part associated with the mean seasonal cycle. The STE may be defined as a function of season or as a climatological annually-averaged constant. The PT correction is currently only applied to the three-dimensional ocean temperatures, for which the STE is computed using a very large ensemble of very short forecasts with the coupled GCM. Large values of the STE are found in the subsurface as well as at the surface, in the high latitudes and in the tropical regions. In the tropical Pacific, the dominant pattern for the STE from the surface through the subsurface is characterized by a warming tendency error in the east and cooling error in the west, while in the high latitudes the STE is confined to the surface with a cooling tendency error in the winter hemisphere and warming in the summer hemisphere. The three-dimensional STE structure assessed from the very short forecasts for the oceanic temperature is roughly consistent with the drift behavior of the uncorrected coupled model. The PT correction was incorporated into the coupled GCM system, and two sets of 12-month forecasts with January initial conditions were produced. One uses the annual cycle correction which subtracts the STE defined as a function of season, and the other uses the annual mean correction with the STE defined as a constant. These were compared to a set of forecasts without any correction. The results are summarized in Fig. 3.2, and show that both corrections can greatly reduce the drift of the coupled model and maintain a more realistic mean seasonal cycle in the oceanic temperature field, especially the annual cycle in the eastern Pacific. The impact of the PT correction on the ENSO-related interannual variability and forecasting skill was also examined. The annual mean correction tends to be more helpful in producing a higher skill of ENSO prediction with a longer lead time. The feedback mechanisms responsible for the improvement of mean annual cycle due to corrections and possible impact of corrections on the extratropical seasonal forecasts were also investigated.

PLANS FY99

Ocean simulation runs will be used to explore further the phenomena mentioned in 3.2.2 through the analysis of heat and momentum budgets. The sensitivity studies will be used

in an attempt to configure coupled GCMs so that mean bias is reduced and forecast skill is improved.

Investigation of ocean model development and prediction issues using the hybrid coupled model will continue. External stochastic forcing will be included for the coupled model simulations. Stochastic forcing (1339), for the 1997 event in particular, has been discussed as an important factor in warm event development.

The inter-decadal sensitivity of ENSO forecast skill to realistic marine stratus clouds will be further examined, using the ensemble of model simulations and forecasts (3.3.1). New coupled model ENSO forecasts will be made with the modified RAS cumulus convection scheme and re-tuned cloud prediction scheme. The analysis of the sensitivity of the coupled model's annual cycle and its inter-annual response to various treatments of low cloud forcing (1403) will continue, using additional analysis techniques such as singular value decomposition (SVD) analysis.

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

3.3 ATMOSPHERIC AND OCEANIC PREDICTION AND PREDICTABILITY

3.3.1 Coupled Model Ensemble Prediction Experiments (CMEP)

A large set of experiments consisting of ocean data assimilation, atmosphere-only integrations, and coupled model forecasts using a frozen version of the atmosphere (version V197) and ocean (MOM II) models has been completed. The experiments were motivated by a number of goals designed to improve seasonal-interannual prediction, including:

The core of the experiments is an ensemble of simulations and predictions with the atmosphere and coupled models. Each member of an ensemble is composed of an atmosphere-only integration (each starting from slightly perturbed initial conditions) for the duration of the ocean data assimilation period (currently 1979 to 1997), plus a set of coupled model forecasts started every six months (1 January and 1 July) with initial ocean conditions from the assimilation and initial atmosphere conditions from the atmosphere-only integration. An initial ensemble of six members has been completed. One coupled run from the first ensemble member has been extended to several decades to examine the internal variability of this coupled model.

A number of auxiliary experiments have also been completed. These include a set of coupled predictions with observed atmospheric initial conditions from the NCEP reanalysis, a large suite of single timestep coupled runs to derive the systematic error tendency (3.2.4), a limited ensemble of coupled predictions with systematic error correction (3.2.4), an extended coupled integration with systematic error correction, and a number of atmosphere-only and coupled runs from different ocean assimilations.

Output from the coupled model ensemble prediction experiments is available in a standard NetCDF format and has been made available to members of the GFDL University Consortium, as well as to researchers within GFDL for analysis.

3.3.2 Interannual and Interdecadal Predictability of Tropical Storms

Tropical storms simulated by a nine-member ensemble of GCM integrations forced by observed SSTs have been tracked by an objective procedure for the period 1980-1988 (1455). Statistics on tropical storm frequency, intensity and first location have been produced. Statistical tools (Chi-Square or Kolmogorov-Smirnov tests) indicate that there is significant potential predictability of the interannual variability of tropical storm frequency, intensity and first location over most of the basins. This implies that SSTs play a fundamental role in model tropical storm interannual variability. An EOF analysis of local SSTs over each ocean basin and a combined EOF analysis of vertical wind shear, 850 mb vorticity and 200 mb vorticity have been performed. Over some basins like the western North Atlantic, the impact of SSTs on simulated tropical storm statistics is an indirect effect through the large scale circulation, as in observations.

It has been observed that the number of Atlantic tropical storms was higher in the 1950s than in the 1970s. To test the ability of the GCM to simulate such decadal change, a 10-member ensemble of atmospheric GCM integrations forced by observed climatological SSTs from the 1950s has been performed. The results are summarized in Fig. 3.3, which shows a significantly (99% significance) higher number of tropical storms compared to similar integrations using climatological SSTs from the 1970s. Further examination indicates that it is the

local cooling of tropical North Atlantic SSTs that is responsible for the decrease of simulated tropical storm activity in the 1970s.

3.3.3 Tropical Intraseasonal variability

Space-time power spectra for the 850 mb velocity potential from a five-case ensemble of integrations with the current (v197) atmospheric GCM (AGCM) configuration show a dramatic improvement relative to earlier AMIP I ensemble results (cj), with the main difference being attributed to changes in the convection scheme (i.e., Relaxed Arakawa-Schubert in v197 versus moist convective adjustment in the AMIP I). These AGCM ensemble mean space-time spectra, along with v197 coupled GCM ensemble mean spectra, are plotted in Fig 3.4. It is evident that the dominant modes for tropical intraseasonal oscillations (TIO) in the v197 AGCM ensembles have more than twice the amplitude of the dominant TIO modes in AMIP I and the period is shifted to the 40-60 day range as compared to approximately 30 days. In addition, tropical intraseasonal oscillations (TIO) from the (five case) ensemble of v197 coupled GCM predictions appear to be of somewhat stronger amplitude and somewhat slower speed (approximately 60 days versus approximately 50 days) when compared to the v197 AGCM ensemble results. This is presumably an impact of the interactive ocean. Studies are currently underway to investigate the potential predictability of TIO by looking at the significance of interannual TIO fluctuations.

3.3.4 Relationship Between Tropical Convection and SST

This study examines the relative roles of the large-scale circulation and SST on the relationship between tropical convection and SST. The SST, outgoing longwave radiation (OLR), and wind divergence fields from the atmosphere-only, coupled model predictions, and coupled model long run of the CMEP experiment (3.3.1) have been compared to each other and to observations in order to understand better the complex relations between these fields. The comparisons to observed fields showed significant bias in the convection scheme, a problem now under investigation (3.2.1).

3.3.5 Coupled Model Equatorial Response to Different Specifications of Low Clouds

In a previous multi-year coupled model sensitivity experiment, the seasonal cycle of SSTs in the eastern equatorial Pacific was found to improve substantially when the model-predicted low clouds over the ocean were replaced with a specification from the ISCCP (International Satellite Cloud Climatology Project) database. This result was attributed to the fact that the ISCCP clouds quasi-realistically simulated the marine stratus regime. Now, an additional suite of multi-year sensitivity experiments has been performed to help elucidate cloud feedbacks upon the equatorial dynamics in the context of the annual mean SST and its seasonal cycle in the equatorial Pacific, as well as SST variability on the ENSO time scale. All members of the suite of experiments may be viewed as employing specified low clouds over the oceans based upon ISCCP multiplied by a scale factor SF. The distinguishing characteristic of each experiment amongst the suite is the value of SF: 1.0 (full ISCCP), 0.8, 0.5, 0 (zero low clouds), and a hybrid specification corresponding to full ISCCP (SF = 1.0) in the tropical Pacific, east of longitude 120W, and no low clouds (SF = 0) elsewhere. In all experiments, low clouds over land as well as high and middle clouds over both land and sea are predicted by the atmospheric model's empirical cloud prediction scheme.

A modified view of the feedback of clouds upon coupled model equatorial dynamics has emerged from the above suite of sensitivity experiments. The presence of marine stratus leads to an intensification of the trade winds and upwelling. In turn, the strength and westward extent of the eastern equatorial Pacific affects the SSTs in the western equatorial Pacific. However, the efficiency of this circulation is modulated by the sign of the SST and heat content biases in the western equatorial Pacific cold tongue on a somewhat delayed time scale. More specifically, a cold bias (warm bias) in the western equatorial Pacific reinforces (opposes) the positive feedback of the marine stratus upon the equatorial dynamics. Meanwhile, the temperature bias in the western tropical Pacific itself can change sign due to a relatively modest change in low cloudiness and short wave radiation at the ocean surface. Overall, the SF = 0.8 simulation is perhaps the most realistic, as it retains most of the strength of the seasonal cycle of the simulation with the impact of ISCCP low clouds, while exhibiting a reduced cold bias in the western tropical Pacific and somewhat stronger interannual variability on the ENSO time scale. The results also suggest that the SST simulation across the tropical Pacific is very sensitive to cloud amount and cloud optical properties. The 0.8 ISCCP vs. 1.0 ISCCP differential low cloud amount field is quite small in the western tropical Pacific, and may be difficult for cloud parameterization schemes to predict.

3.3.6 Predictable Component Analysis

A conceptual framework has been developed for a unified treatment of issues arising in a variety of predictability studies. The predictive power (PP), a predictability measure based on information-theoretical principles, lies at the center of this framework. The PP is invariant under arbitrary linear coordinate transformations and applies to multivariate predictions without making assumptions about the probability distribution of prediction errors. For univariate Gaussian predictions, the PP reduces to the conventional predictability measure that is based on the ratio of the rms prediction error over the rms error of a "prediction" drawn randomly from the climatology.

Climatic variability on intraseasonal to interdecadal time scales follows an approximately Gaussian distribution. For multivariate Gaussian predictions, the predictability measure PP makes it possible to discriminate a system's predictable components from its unpredictable components. Predictable components can be extracted by predictable component analysis, a procedure derived from discriminate analysis: seeking components with large PP leads to an eigenvalue problem, whose solution yields uncorrelated components that are ordered by PP from largest to smallest.

The application of the PP and the predictable component analysis in different types of predictability studies has been described. Studies are being considered that use either ensemble integrations of numerical models or autoregressive models fit to observed or simulated data.

A number of studies using the results from the CMEP experiments will be performed to address the goals listed in 3.3.1. In particular, forecast skill and potential skill will be examined for both tropical and extratropical fields. The impacts of ocean and atmosphere initial conditions on skill and potential skill will also be studied.

Experiments will continue, using the CMEP experiments as a baseline, to evaluate possible improvements to coupled models for seasonal/interannual prediction. Of particular interest will be improvements to atmospheric and ocean model physical parameterizations, especially cloud and convective parameterizations. An effort to understand methods for reducing initial drift of coupled model forecasts will begin.

Efforts to explore the relationship between SST and convection in model runs and observations will continue, with a focus on seasonal and interannual variability of the ITCZ.

Predictable component analysis will be applied to atmosphere-only and coupled model integrations from the CMEP experiments.

3.4 DATA ASSIMILATION

ACTIVITIES FY98

3.4.1 Nonlinear Filter for Ensemble Data Assimilation

Knowledge of the probability distribution of initial conditions is central to almost all practical studies of predictability and to improvements in stochastic prediction of the atmosphere. Traditionally, data assimilation for atmospheric predictability or prediction experiments has attempted to find a single "best" estimate of the initial state. Additional information about the initial condition probability distribution is then obtained primarily through heuristic techniques that attempt to generate representative perturbations around the "best" "estimate (1509). However, a classical theory for generating an estimate of the complete probability distribution of an initial state given a set of observations exists. This non-linear filtering theory can be applied to unify the data assimilation and ensemble generation problem and to produce superior estimates of the probability distribution of the initial state of the atmosphere (or ocean) on regional or global scales. A Monte Carlo implementation of the fully non-linear filter has been developed and applied to several low order models. The method is able to produce assimilations with small ensemble mean errors while also providing random samples of the initial condition probability distribution. The Monte Carlo method can be applied in models that traditionally require the application of initialization techniques without any explicit initialization. Fig. 3.5 demonstrates the application of the Monte Carlo filter in the simple 3variable Lorenz-63 dynamical system. Initial application to larger models is promising, but a number of challenges remain before the method can be extended to large realistic forecast models. The method has been applied successfully in a spectral barotropic model at T42 resolution. An effort is underway to apply the data assimilation algorithm in the context of the flexible Bgrid model (3.1.1.1) with simple physics.

3.4.2 Ocean Data Assimilation

Perhaps the main utility of Ocean Data Assimilation (ODA) is to help correct for the mean bias of both the ocean model and atmospheric forcing (1457). However, continued improvements of the type discussed earlier (3.2.3) are needed. Naturally, efforts at improving the ODA are intimately related to ongoing model development efforts.

In order to produce ocean initial conditions that are more consistent with the coupled model, the ODA system was run from 1979-1997 forced by daily winds from the atmosphere-only runs. This is a first attempt to address the issue of initialization shock. While the forecast runs indicate a sensitivity to the ocean initial conditions using different wind forcing, the forecast results were comparable to those ocean initial conditions from the ODA forced with NCEP reanalysis winds.

PLANS FY99

The Monte Carlo Filter will be applied to the B-grid model to determine if the method is capable of scaling to realistic prediction models.

Aspects of the ODA scheme that improve initialization of the coupled model with regard to a reduction of shock and improved forecast skill will be investigated. In particular, the sensitivity to forcing, the first guess and observational statistics, and data coverage will be examined.

3.5 OCEAN-ATMOSPHERE INTERACTIONS

ACTIVITIES FY98

Although much progress has been made in our understanding of El Niño during the 1980's -- no one anticipated the event of 1982, but by 1987 there were coupled ocean-atmosphere models capable of skillful predictions -- the 1990's brought surprises. First there was the unexpected persistence of unusually warm surface waters over the eastern tropical Pacific after El Niño of 1990, then there was El Niño of 1997 which the models predicted with mixed success. Has there been a change in the properties of El Niño between the 1980s and 1990s? Analyses of long time-series indicate that El Niño is subject to decadal modulations (dw), energetic up to the early 1930s, then practically disappearing for a few decades before reappearing in the late 1950s. The possible causes of these changes are being investigated via analysis of data from the TOGA-TAO array, analyses of gridded data sets provided by a GCM that assimilates the available measurements, and through modeling studies.

A change in the mean state of the tropics is one possible explanation for the decadal modulation of El Niño. (Simple coupled ocean-atmosphere models in which the mean depth of the thermocline is specified show that the amplitude of the simulated Southern Oscillation is very sensitive to changes in that depth.) The processes that maintain the thermocline include a shallow, wind-driven meridional circulation that involves the subduction of surface waters in the subtropics off the western coasts of the Americas, Africa, and Australia. Studies with a realistic ocean GCM indicate striking differences between the three oceans. Whereas water parcels can reach the equator from either hemisphere in the Pacific, this is possible only from the Southern Hemisphere in the much smaller Atlantic and Indian Oceans. The implications of this result, which suggests that low frequency variability in low latitudes can have very different origins in the three oceans, have been explored for the case of the Pacific by means of a simple coupled model of self-sustaining decadal oscillations (bo).

The continual exchange of surface waters between the tropics and extratropics means that a change in high latitude conditions will, in due course, affect the tropics. The manner in which the unusually low surface temperatures of the polar regions during the Last Glacial Maximum (LGM) affected low latitudes has been a matter of much debate. At first, it was believed that, during the LGM, the tropics were only slightly colder than they are today. However, recent data indicate that the tropics were significantly colder. Calculations with coupled GCMs show that, because of the shallow meridional circulation that links the tropics and extratropics, a cooling of the polar regions caused the tropical thermocline to shoal. The resultant decrease in sea surface temperatures were then amplified by local interactions between the ocean and atmosphere, interactions of the type that characterizes El Niño.

The importance of the mean state to interannual fluctuations emerges from a study of the energetics of the Southern Oscillation. Data from a realistic simulation over a prolonged period indicate that the surface winds do positive work on the ocean and create available potential energy during half the cycle, and destroy that available potential energy during the other half. If the work done is represented by the terms where bars indicate a time average, and primes indicate the departure from a time-average, then the first term is found to be dominant by a large margin. This result suggests that the background state that supports El Niño as part of an interannual oscillation is characterized by certain time-averaged winds with which are associated a certain zonal temperature gradient, and a zonal slope of the thermocline. Current research is focussed on determining whether there were changes in these fields between the 1980s and 1990s.

PLANS FY99

Studies planned for the coming year include the effect of warm and cold extratropical surface anomalies on the tropical thermocline, and the dependence of the spectrum of interannual, coupled ocean-atmosphere modes of oscillation on the mean state (time-averaged climate) of the tropics. A principal question will be whether the time-averaged zonal component of the wind has to exceed a certain value before interannual oscillations are possible. These studies are part of a long-term effort to develop realistic coupled ocean-atmosphere models.


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).