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

 

4. EXPERIMENTAL PREDICTION

GOALS
  4.1 ATMOSPHERIC AND OCEANIC PREDICTION AND PREDICTABILITY

ACTIVITIES FY00

     4.1.1 Ocean Model Development for Seasonal/Interannual Prediction           The advent of more powerful computers will permit the next generation of ocean climate models to have a grid resolution necessary for studies of seasonal-to-interannual prediction. Consequently, over the past year the ocean model development efforts of the Climate Dynamics group have been closely coordinated with members of the Experimental Prediction group. The past year has seen prototype studies of ocean grid resolution and ocean parameterizations which are useful to both groups. Most notably, such work expands the domain of the experimental prediction group's ocean model into the Arctic.

          This development work represents part of a multi-year effort aimed at enhancing the realism of GFDL's ocean simulations (1535, 1536, 1676). In particular, this new ocean model will facilitate study of longer term phenomena, such as decadal fluctuations of ENSO and variations in the thermohaline circulation.

     4.1.2 Coupled Model Development for Seasonal/Interannual Prediction

          A new coupled model to be used for seasonal/interannual forecasting is under development. The configuration of the atmospheric component has not yet been determined. However, using Bombay release versions of the Flexible Modeling System (FMS) (1.3.3), an atmospheric model has been assembled with the following features: N45L18 B-grid dynamical core, 4th-order advection, Klein predicted clouds (3.2.4), diurnal variation of radiation, and RAS (Relaxed Arakawa Schubert) convection. The ocean component is MOM3 and is also used as the base model for a comparison project with the International Research Institute for Climate Prediction/Applied Research Centers (IRI/ARCS). This model is nearly global (no Arctic Ocean) with one degree resolution increasing to one-third degree meridional resolution within ten degrees north and south of the equator. The vertical resolution is 40 levels with a constant 10 m resolution to 200 m. The physical parameterizations include K-Profile Parameterization (KPP) vertical mixing, Smagorinsky viscosity (1720), Gent-McWilliams (GM) diffusion for horizontal mixing (1536), and an explicit free surface. The coupling between component models uses the FMS coupling framework (1.1.3). Modifications to pass fluxes to MOM and sea surface temperatures (SSTs) to the atmosphere, required because MOM3 is not a compliant FMS component model, have been completed. Preliminary coupled forecast experiments using this system are being run, although only rudimentary tuning has been done. Early indications are that the mechanics of the coupling are working, although considerable tuning will be required to produce acceptable simulations and predictions.

     4.1.3 Statistical Atmospheric Model Development

          A statistically-based atmospheric model has been developed using SST anomalies as predictors and ocean surface fluxes of heat, momentum, and freshwater as predictands. The model was initially developed using the National Centers for Environmental Prediction (NCEP) and European Centre for Medium Range Weather Forecasting (ECMWF) Reanalyses and the GFDL Atmospheric Model Intercomparison Project (AMIP) 2 model used for seasonal to interannual prediction. The model is capable of generating sustained oscillations when coupled with an ocean model without additional tuning. Considerable differences have been observed between atmospheric products (1624, na); for instance, the statistical version of the ECMWF Reanalysis model yielded sustained coupled oscillations, while the GFDL statistical model was highly damped when coupled to the ocean model. Subcriticality in the coupling between atmosphere and ocean resulted in diminished forecast skill, as measured by Niño3 SST correlation and RMS error.

     4.1.4 Sensitivity to Horizontal and Vertical Resolution in Atmospheric Models

          The impact of vertical and horizontal resolution in the B-grid atmospheric model (1.2.4) is under investigation in an attempt to determine the required resolution for improved coupled predictions. Initial results are focusing on the quality of simulations of basic atmospheric quantities. As an example, Fig. 4.1 shows the 1980-1981 mean winter and summer zonal jets from the NCEP reanalysis, and corresponding 2-year integrations from N45L18, N45L30, and N90L18 model simulations. Evaluation of many additional diagnostic quantities, aided by the capabilities of the GFDL web-based diagnostic facility (6.4) will determine the most cost-effective resolution.

     4.1.5 Impact of Land and Ocean Low-Level Tropical Clouds on ENSO Prediction

          The role of tropical land and ocean low-level clouds in the ability of a coupled atmospheric-ocean general circulation model (AOGCM) to predict ENSO has been studied (nl) through a set of ten 12-month model integrations. In the tropics, a strong bias in a GFDL Experimental Prediction AOGCM is evident in the western tropical Pacific, where overly strong convection diminishes the SST warm pool, reducing the east-west SST gradient, and effectively weakening the trade winds. This bias is exacerbated by the poor simulation of tropical marine stratocumulus clouds (MSc) in the eastern equatorial Pacific, which are essential to a proper SST annual cycle. The sensitivity of the AOGCM simulation of the tropical circulation to prescribing low-level clouds obtained from the International Satellite Cloud Climatology Project (ISCCP) was studied. The results showed that the combined prescription of global ocean and tropical land low-level clouds in the AOGCM resulted in the best simulation of the Walker circulation. The improved tropical circulation was accompanied by an improvement in the one-year prediction of ENSO by the AOGCM. It was determined that the combination of more realistic MSc in the eastern tropical Pacific and more realistic tropical land surface heating led to the improvements.

          The top panel in Fig. 4.2 depicts ENSO hindcast skill in the Niño3 region. The skill scores show that the addition of the low-level ISCCP clouds over the oceans alone (CMIO) has indeed resulted in an improvement in model skill compared to the fully predicted cloud

experiment (CM). The CMIOL experiment (prescribed clouds over both ocean and land) shows even more skill for the entire 12-month period. The experimental rms errors, shown in the middle panel of Fig. 4.2, indicate that only the CMIOL experiment separates itself from the other experiments as having reduced error. It follows that although the ocean low-level clouds are assisting the CMIO experiment to simulate the correct sign of the SST anomaly, these clouds alone are not sufficient to simulate the amplitude of the ENSO events. The lower panel of Fig. 4.2 shows that the CMIOL experiment has a more realistic simulation of precipitation than the other experiments, though it does err on the side of underpredicting precipitation.

          Multi-year simulation experiments with low (as well as multi-level) ISCCP cloud fractions and optical depths were performed with a newer version of the coupled model. However, relative to results obtained with the previous model (1719), the SSTs in the tropical Pacific cooled a few degrees and displayed much weaker interannual variability in the equatorial belt. Moreover, what variability remained was not ENSO-like in character, e.g., was not linked to a delayed oscillator mechanism. On the contrary, it appeared to be driven by westward propagating variations in surface heat flux associated with the ISCCP low clouds in the eastern tropical Pacific. This behavior is presumably due to one or more new features of the new coupled model: MOM3 instead of MOM2, RAS with a re-tuned threshold parameter instead of the full Arakawa-Schubert convective parameterization, variable high cloud emissivity, diurnally-varying radiation, and elimination of the 10% proportional surface evaporative flux correction.

     4.1.6 Atmospheric Model Predictability: Top of the Atmosphere Radiation Budget

          Bruce Wielicki of NASA/Langley is comparing several AGCM (Atmospheric GCM) simulations of the earth's tropical radiation budget to Earth Radiation Budget Experiment (ERBE) and to Clouds and the Earth's Radiant Energy System (CERES) satellite observations. GFDL has provided results that are relevant to evaluating the ability of GCMs to simulate interannual variability as well as secular trends. Two sets of AGCM runs with specified SSTs from 1997/1998 were analyzed: the single AMIP 2 run; and a 10-member ensemble of 28-month Monsoon Modeling Intercomparison Project (MMIP) integrations started from the AMIP 2 run.

          Both the phase and amplitude of the outgoing longwave radiation (OLR) and net absorbed shortwave anomaly fields (calculated with respect to the 1985-1989 monthly and 2-month means) verified extremely well compared to observations. Results from the 10 MMIP ensemble members indicated that the El Niño signal was very robust. In fact, the "GFDL-EP" "model (in contrast to the GFDL climate model) was among the top performers. These excellent interannual variability results may be attributed to the ability of the diagnostic cloud parameterization scheme to discriminate between anvil cirrus and non-anvil cirrus clouds in terms of cloud optical depth and emissivity. On the other hand, none of the models captured the apparent trend in global mean OLR suggested by a comparison of the CERES observations with earlier ERBE observations.

     4.1.7 Impact of MJO/WWB in the Far Western Equatorial Pacific on ENSO Forecasts

          The influence of the Madden-Julian Oscillation (MJO) on ENSO is an intriguing and controversial issue. Extraordinary MJO events coincided with the onset of the 1997/1998 warm event, which was unusual in its magnitude and rapidity of onset. In order to gain a quantitative understanding of the role of MJO and westerly windburst (WWB) events in ENSO prediction, a hybrid coupled model (HCM) forecast system was applied. The ocean model is MOM3 with 1 degree zonal and 1/3 degree meridional resolution. The statistical atmosphere (4.1.3) is based on ECMWF reanalysis surface windstress, evaporation, and radiation fluxes regressed to SST anomalies. The experimental design included ocean initial conditions from the GFDL ocean data assimilation system and control forecasts using the HCM for the 1980-1998 period. Another experiment was run using the observed wind time series within a region bounded by 120°E-165°E and 5°S-5°N. This case gave much better predictions of the large events in 1982-1983, 1988, and 1997, resulting in significantly better skill scores. The 1982 and 1997 El Niños were amplified by wind events from the preceding March. While the model predicted these warm events without the patch winds, the amplitude was significantly underestimated. The wind events were found to trigger ocean Kelvin waves which result in deepening of the thermocline in the Eastern Pacific with subsequent SST warming and amplification by coupled interactions.

     4.1.8 Interaction of MJO and Tropical Storms

          A number of observational studies have indicated the existence of a relationship between the frequency of tropical storm occurrence and the MJO. Tropical storm formation appears to be enhanced during the convectively active phase of the MJO. Data from a 20-year AMIP 2 run made with the experimental prediction AGCM have been analyzed and indicate that there is indeed a relationship. For 16 of the 19 years analyzed, over 50% of the tropical storms in the model formed during the convectively active phase of the MJO; for 7 of the 19 years, more than 70% formed during the convectively active phase of the MJO. This relationship was strongest in the west Pacific during the 1980s. During the 1990's the signal was stronger in the east Pacific and the Atlantic. During El Niño years the relationship seems to be weak or non-existent. The MJO also has an impact on the intensity of tropical storms in the model. During the 19 years analyzed, 75% of the tropical storms that reached hurricane intensity (maximum winds exceeding 75 mph) occurred during the convectively active phase of the MJO. In 7 of the 19 years, all of these "hurricanes" formed during the convectively active phase of the MJO.

     4.1.9 The Nature and Predictability of Tropical Intraseasonal Oscillations

          Results from both AMIP (1540) and coupled GCM integrations are being used to investigate tropical intraseasonal oscillations (TIO). As part of the Monsoon Modelling Intercomparison Project, a comparison of TIO/MJO behavior in the 1982-1983 and 1997-1998 ENSOs was performed using 10-member ensembles of AGCM integrations forced by observed SSTs. Longitude-time plots and power spectra indicate a difference in MJO behavior between the onset and mature phases of both ENSO events. Statistical tests indicate that some of the MJO interannual variability is significant, especially in the west-central equatorial Pacific Ocean. Considerable TIO activity is also seen in the Asian monsoon region and appears to be associated in part with northward propagation from the equatorial Indian Ocean. An association between equatorial MJO activity and Monsoon rainfall intensity is being sought.

          Results from Coupled Model Ensemble Prediction (CMEP) coupled GCM are being analyzed to better understand TIO as a coupled atmosphere-ocean phenomenon. Fig. 4.3 shows evidence of a robust TIO in the model using the first extended EOF of bandpassed (20-100 day) 200 hPa velocity potential. This EOF compares favorably to those from the NCEP/NCAR reanalysis, but it is stronger. By forcing the atmospheric AGCM with SSTs generated by the CGCM and comparing these results to those obtained using a weakly interacting slab mixed-layer ocean, a better understanding will be gained of how interactive ocean processes impact the behavior of TIO.

     4.1.10 Coupled Model Potential Predictability

          To increase understanding of the prediction and predictability (1539) of the coupled GCM and the coupled atmosphere-ocean system (ke), an ensemble of one-year forecasts was run with members differing only in small perturbations to the 850 mb temperatures in the atmospheric initial conditions. These results complement earlier results from the CMEP experiments in which somewhat larger initial condition perturbations were used. The spreads in the predicted SST anomalies over the east-central Pacific Ocean are as large as in the forecasts produced from initial conditions with much larger differences. For several of the years investigated, SST spreads of several degrees can appear in the tropical Pacific within a few months of the start of the forecast. This implies that predictions of the thermal structure of the tropical Pacific cannot be viewed as deterministic, even for lead times as short as one or two seasons, at least for certain ocean initial conditions.

     4.1.11 Indian Ocean Variability

           For the tropical Pacific and Atlantic Oceans, internal modes of variability that lead to climatic oscillations have been recognized, but in the Indian Ocean a similar interannual climate variability caused by air-sea interaction has not yet been found. However, recent observational analyses have suggested that the interannual variability of the Indian Ocean consists of a dipole mode in SST. These analyses further argue that the dipole mode is independent of ENSO and is the result of internal dynamics. The dipole mode has important implications for climate variability, as the shift of the convergence zone leads to floods in east Africa and drought in Indonesia. A 1/4 degree ocean model that covers the Indian Ocean basin was used to simulate the ocean during 1997/1998. Preliminary results indicate the importance of both the ocean dynamics and surface fluxes in the evolution of this mode. Fig. 4.4 shows the large subsurface temperature anomalies for November 1997 associated with the dipole mode. Investigations of the sensitivity to various forcing products and heat flux formulations are underway. These results are to be compared to the GFDL ocean data assimilation for verification and understanding.

PLANS FY01

          A key focus will be on developing the components of a coupled forecast system that will produce simulations and coupled predictions that are superior to those that were available from the models used for seasonal prediction experiments in the past five years. Initial efforts will focus on development of improved individual component models, particularly for the atmosphere. Model resolution, the choice of physical parameterizations, and the tuning constants available in existing parameterizations are all candidates for change. A coordinated diagnostic effort making use of new tools available at GFDL (6.4) along with the capabilities of the FMS (Section 1) are expected to facilitate model improvements. As component model improvements occur, they will be tested in a fully-coupled model with a final goal of producing an improved coupled model.

          MOM4's diagnostic capabilities and subgrid scale parameterization suite will be expanded and MOM4 will become the ocean model of choice for coupled model development.

          As improved coupled models are developed, they will be integrated for several decades and used to produce ensembles of forecasts from 1980-2000 to assess ENSO forecast skill. The primary focus will be on the tropical circulation and the use of physically based metrics for evaluating model performance. Among the metrics are: ENSO variability and forecast skill; MJO; monsoon circulation; tropical storm frequency and distribution; and precipitation patterns.

          The Hybrid-Coupled Model work will be extended to make additional comparisons between atmospheric models developed at GFDL and the IRI/ARCS community. An evaluation will be made of the relative merits of an anomaly coupled forecast system versus a fully coupled system in the context of ENSO prediction; i.e., are there significant and predictable air-sea interactions which are not captured by a diagnosed linear relationship to SST in any of the atmospheric models considered ?

          Further work using the Hybrid Coupled Model will focus on trying to gain an understanding of the relation between the ocean state and its response to MJO/WWB events. The impact of an event and its relation to the seasonal cycle will be investigated.

          The ability to perform integrations with ISCCP low and/or multi-layer cloud fractions and cloud optical depths will be added to the coupled FMS model, and reference integrations with ISCCP low clouds will be performed.

          An analysis will be performed of the sensitivity of the interannual variability (as well as the climatology) of the top of the atmosphere earth radiation budget in the FMS model to the diagnostic and prognostic cloud parameterizations.

          Many aspects of the onset and development of the Indian Ocean dipole mode remain to be investigated. Some of the questions to be studied are: Is there a preferred background state? Are there regime shifts that account for why there was no dipole in the 1980s but two in the 1990s? Is the dipole a coupled mode? How does it grow and decay? Is there a possible relationship with ENSO?

  4.2 DATA ASSIMILATION

ACTIVITIES FY00

     4.2.1 Ensemble Adjustment Filter           An established theory (1658) exists for generating an estimate of the complete probability distribution of the state of a model given a set of observations. This non-linear filtering theory unifies the data assimilation and ensemble generation problem that have been key foci of prediction and predictability research for numerical weather and ocean prediction applications. A novel Monte Carlo approximation to the fully non-linear filter has been developed and applied in perfect model experiments in a low order model, a non-divergent barotropic model in both perfect model and real data applications, and in a dry global primitive equation model. This ensemble adjustment method is able to produce assimilations with small ensemble mean errors while providing accurate measures of uncertainty in the assimilated variables. The filtering method uses information from an ensemble of model integrations to obtain an estimate of the covariance between model state variables and observations. Each available observation is then allowed to impact the prior distribution for each state variable independently. However, the way in which the required product of the observational error distribution and the prior state distribution is computed maintains much of the information about covariances of the prior state variables. The method is able to assimilate observations with a nonlinear relation to model state variables, and has also been demonstrated to be effective in using observations to adjust the value of unknown model parameters. The method is shown to have significant advantages over four dimensional variational assimilation in low order models (Fig. 4.5) and scales easily to much larger applications.

     4.2.2 Ocean Data Assimilation

          A nearly global, 40 level ocean model (MOM3) with 1 degree zonal resolution and 1/3-1 degree meridional resolution is being used for seasonal to interannual prediction at GFDL. Ocean initial conditions are generated by a statistical interpolation algorithm1 which assimilates ocean profile data to depths of approximately 400 meters. These ocean initial conditions and model configuration are being shared with several collaborators as part of an IRI/ARCS forecasting effort.

          Bi-weekly ocean initial conditions have been distributed to groups at IRI, NCEP, COLA (Center for Ocean, Land and Atmosphere studies), and NCAR for the period January 1980 through January 2000. The ocean model has been ported to parallel computational platforms at the respective centers. Comparisons of forecasts using identical ocean model and initial conditions but different atmospheric models are being made.

          Currently, monthly-averaged data from the ocean analysis are available on the web (data1.gfdl.gov) through interactive browsing software developed at PMEL by Steve Hankin. In addition to the ocean assimilation, a simulation which does not make use of subsurface data can be viewed in arbitrary latitude/longitude/depth/time sections and downloaded for further analysis. Other ocean assimilation products from Ming Ji and Dave Behringer at NCEP and Jim Carton at University of Maryland are also available.

           A four-dimensional ocean assimilation system using a version of MOM4 coupled to a statistical atmospheric model is currently being developed in collaboration with Eli Tzipermann and Eli Galanti at the Weizmann Institute in Israel. The tangent linear and adjoint versions of this model have been generated using a compiler developed by Ralf Geiring.

     4.2.3 Information Content of Surface Pressure Observations

          The ensemble adjustment filter (4.2.1) has been applied in a dry global primitive equation model to evaluate the information content of observations of surface pressure in a perfect model experiment. A control integration of the model is used to generate observations of surface pressure at a set of 600 randomly located points on the surface of the sphere every six hours. Using only these observations, the filter is able to reconstruct the structure of the free atmosphere (Fig. 4.6). Without the impact of the surface pressure observations, the ensemble

has an error doubling time of roughly 5 days and error saturates to climatological values which are several orders of magnitude larger than the RMS error of the assimilated ensemble mean after about 25 days. Applying the adjustment filter allows much more efficient use of data than most other known data assimilation methods and should improve our estimates of the predictability of the atmosphere.

     4.2.4 Targeted Observations

          The "targeted observation problem" tries to determine the location at which an additional observation should be taken some time in the future in order to improve significantly a forecast of some quantity at a time even further in the future. A number of heuristic methods using operational ensemble predictions have been used to attack this problem in recent years. The ensemble adjustment filter method (4.2.1) for ensemble assimilation and prediction provides a framework in which the targeted observation problem can be addressed. The filtering method uses information from an ensemble of model integrations to obtain an estimate of the covariance between model state variables and observations. Each available observation is then allowed to impact the prior distribution for each state variable independently. There is no reason that the impact of observations must be limited to those available at the present time. The impact of observations from previous times can also be computed in this filter using the information available from the time history of the ensemble of assimilations. Likewise, the impact of a hypothetical future observation on a forecast even further in the future can also be approximated. Initial results in a low order dynamical system have suggested that the ensemble adjustment filter targeted observation method is significantly superior to other methods currently available.

PLANS FY01

          The ensemble adjustment filter will be tested in high resolution primitive-equation models both with and without comprehensive physical parameterizations. Further study of the information content of surface pressure observations in more realistic models should lead to an understanding of operational implications. Initial application of the adjustment filter will be attempted in MOM. Strategies for targeted observations in both low order and realistic models will be refined.

          Additional ocean assimilation runs are being planned in collaboration with Mike Tippett at IRI using a reduced empirical orthogonal function (EOF) space estimate of model error. Forecasts using this technique will be compared to the current ocean analysis.

           Current plans are to assimilate data from the TAO array in the equatorial Pacific and wind data from the NCEP Reanalysis using the adjoint method. Working in collaboration with Geiring, his compiler is being updated to be Fortran 90 compliant in concert with the latest developments of MOM4 (4.1.1). Developing the ocean model and its adjoint compiler simultaneously is a key to this effort.

  4.3 OCEAN-ATMOSPHERE INTERACTIONS

ACTIVITIES FY00

          Although there has been considerable progress in our understanding of climate fluctuations on seasonal, interannual, and decadal time-scales, a wide gulf still seems to separate observational, theoretical, and modeling studies. Evidence of this divide includes the following.

          Theoretical studies of El Niño focus on the natural modes of oscillation and, therefore, take an eigenvalue approach. Measurements, on the other hand, pay essentially no attention to the theoretical results and describe the development of El Niño, of 1997 for example, in terms of westerly wind bursts that initiate an event, treating the evolution as an initial value problem. In a recent debate about the possible effect of global warming on El Niño, the various arguments being presented all concern statistical issues; no one is appealing to theoretical results to anticipate how global warming will affect El Niño. Numerous groups are developing coupled ocean-atmosphere GCMs and are simulating a bewildering variety of interannual oscillations. It is not at all clear how the available theoretical results can contribute to the development of realistic complex models. The remedy for these various problems is a stability analysis that establishes how the properties of the Southern Oscillation (its period, growth rate, etc.) depend on the background state described in terms of readily measurable quantities, such as the spatially averaged depth of the thermocline, the temperature difference across the thermocline, and the time-averaged intensity of the trade winds. Preliminary results from such an analysis have already been published (1710) and a more detailed analysis has been completed (ab). This past year, attention focused on application of these results to the interpretation of a realistic simulation of oceanic conditions over the past fifty years with a GCM, and to factors that can effect gradual (decadal) changes in the background state. The distinctiveness of each El Niño was another topic that was explored.

          Decadal changes in the depth of the thermocline and in the intensity of the time-averaged zonal winds along the equator can account for certain observed changes in the properties of El Niño from the 1960s to the 1980s. The period of the Southern Oscillation, for example, increased from approximately 3 years to 5 years, but does not explain why El Niño was prolonged and weak in 1992, but brief and intense in 1997. Westerly wind bursts near the date-line appear to have played a central role in the development of El Niño in 1997, but similar wind bursts were present on other occasions and failed to generate any El Niño. Up to now, studies of these wind bursts have concerned either their atmospheric aspects (whether they are associated with certain sea surface temperature patterns, for example) or the oceanic response they generate. Surprisingly, no one has investigated the response of the coupled ocean-atmosphere system to these winds. A study of the ocean-atmosphere interactions generated by westerly wind bursts, by means of a simple Cane-Zebiak type coupled model, indicates that the impact of the winds depends critically on the phase of the Southern Oscillation, if one is present. A swinging pendulum subjected to blows at random times is a good analogy. Westerly wind bursts can amplify El Niño enormously, if they occur when the Southern Oscillation is in the early stages of its El Niño phase, but have the opposite effect when they occur during La Niña. This result has the important implication that, although it should have been possible to anticipate El Niño of 1997 on the order of a year in advance, it was not possible to anticipate its enormous amplitude until the westerly wind bursts had actually occurred. Given this role of unpredictable westerly wind bursts, El Niño has very limited predictability. A long-term forecast can at best sound an alert that conditions are ripe for an intense El Niño should the appropriate wind bursts materialize. To make such a statement, the best quantity to monitor is the available potential energy (1694) of the tropical Pacific which is highly correlated with sea surface temperatures in the eastern tropical Pacific, but has the advantage of being a global rather than a local quantity.

          A realistic General Circulation Model of the ocean (MOM), forced with the winds observed over the tropical Pacific Ocean since 1950, provides data that permit a detailed study of both gradual, decadal, and rapid event-to-event changes in the properties of El Niño. Preliminary results indicate that the differences between one event and the next can indeed be attributed to the impact of random westerly wind bursts. Those winds contributed to the great intensity of El Niño in both 1982 and 1997. A further contributing factor was the change in the background state from the 1960s to the 1980s. The easterly winds weakened over that period, and the thermocline gradually deepened. This caused the structure of the Southern Oscillation to become closer to that of the delayed oscillator in which sea surface temperature variations in the eastern tropical Pacific are controlled mainly by vertical movements of the thermocline in response to winds near the dateline. (During the 1960s and 1970s, local winds in the eastern tropical Pacific had a stronger influence on sea surface temperatures.) Thus, even if the structure of westerly wind bursts remained unchanged over the decades, their impact increased in the 1980s and 1990s because of the preferred mode during that period.

          The extent to which oceanic exchanges between the tropics and subtropics contribute to the decadal variations in the background state is a topic of continuing interest. Experiments with a realistic oceanic GCM indicate that the disturbances that reach the equator come predominantly from the Southern Hemisphere.

PLANS FY01

          Studies planned for the coming year include the use of an intermediate coupled ocean-atmosphere model, composed of an oceanic GCM coupled to a statistical atmosphere in which the winds correspond to that component of the observed winds correlated with tropical sea surface temperature changes. The goal will be to explore the validity of the stability analysis which was obtained by means of a very simple coupled model with a great number of adjustable parameters in a more realistic setting. The strategy will be to focus on certain extreme situations where the differences between the possible modes are very large: the delayed oscillator mode that is favored when the thermocline is deep and the winds are intense, versus the local mode when the thermocline is shallow. (In the local mode sea surface temperatures depend not on vertical movements of the thermocline, but primarily on entrainment across the thermocline and on zonal advection.) The coupled model will also be used to explore how changes in the structure of the equatorial thermocline, induced by tropical-subtropical exchanges, affect ocean-atmosphere interactions.


1. Derber, J., and A. Rosati, A Global Oceanic Data Assimilation System, Journal of Physical Oceanography, 19(9), 1333-1347, 1989.

*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).