 |
 |
 |
 |
 |
 |
|
|
|
|
|
|
The El Niño-Southern Oscillation is the dominant phenomenon of global interannual climate variability, and seems to be the only one for which current prediction methods are more skillful than climatology or persistence. The Zebiak and Cane model is a physical coupled ocean-atmosphere model used to forecast sea surface temperatures in the ENSO-affected regions of the Pacific Ocean (Cane et al., 1986; Zebiak and Cane, 1987). While the predictive ability of this model has been demonstrated in the past (Latif et al., 1993; Graham, 1992), its prediction of the most recent 1997-1998 ENSO event was unsatisfactory. The dependence of its initialization schemes on wind fields derived solely from merchant ship observations proved to be a liability during that event (Chen et al., 1995). The deficiencies of the surface-based wind observations prevented the oceanic component of the model from attaining a realistic state during the year prior to the event.
TOPEX/Poseidon sea-level data of Cheney et al. (1994), with complete spatial but coarse time resolution, were tested in the model for their ability to improve predictions. The sea-level height anomaly product bins individual orbit observations into monthly boxes of 4° longitude by 1° latitude. This product is known to provide sea-level height anomaly values within 2 cm rms of monthly mean in situ measurements from equatorial tide gauge stations, and to capture propagating features, especially the equatorial Kelvin waves that are so critical to the ENSO mechanism. Indeed, the much better coverage of TOPEX/Poseidon (T/P) altimetry data of Cheney et al. (1994) yielded significantly improved forecasts.
The results show the TOPEX view to be comparable in quality to an in situ array of echo sounders, and adequate for prediction studies (Fig. 2). The major features of the differences (zonally elongated maxima in the NW and SW corners of the domain as well as the one near 10N and the eastern boundary) are present on the patterns of the comparison of sea-level fields of the NCEP ocean GCM with T/P altimetry. The NCEP fields assimilate both the T/P altimetry and subsurface temperature profiles.
Chen et al. (1999) showed that the wind fields obtained from the NSCAT scatterometer would have yielded good forecasts, had the satellite not failed in June 1997 after only nine months of operation. We have found, however, that other data can compensate for poor wind observations. Chen et al. (1998) achieved a high level of forecast skill, including an excellent forecast for the 97/98 El Niño event, by nudging sea-level height assimilated fields from 34 tropical Pacific IGOSS tide gauge stations into the initialization run of the model (Cane et al., 1996). The assimilation of observed sea-level height influences the representation of subsurface behavior and corrects a poor model initial state caused by simplified tropical ocean dynamics. This is similar to the assimilation of subsurface temperature used in the NCE ocean GCM with similar forecast results (Behringer et al., 1999; Ji et al., 1998).
In the long-term, we expect the problems in the wind forcing to be fixed by the Quickscat mission or alleviated by better ocean models and data assimilation. A shorter-term goal, developing better models for the error in the data assimilation procedures, will allow us to exploit the T/P altimetry data more completely. In addition, tuning the data assimilation parameters against the high-resolution T/P altimetry data will extend the benefits to the assimilation of in situ data in the pre-TOPEX era. The resulting improved hindcasts of the tropical Pacific sea level will be helpful for the development and improvement of reliable procedures for ENSO prediction.
The work on the project during 2000-2001 was concentrated on the use of satellite data for improving various stages of ENSO prediction technology: model initialization, process simulation, and data assimilation.
The utility of a formal data assimilation approach for initialization of El Niño predictions with the Zebiak-Cane model was studied with an approximation of the non-linear coupled model by a system of seasonally dependent linear models (Markov models). The low dimensional nature of such an approximation permits a sequence of ``perfect'' initial states, which define trajectory segments best fitting the observed data. Declaring these perfect initial conditions to be the ``true'' states of the model, we computed a priori parameters for data assimilation and test the ability of its solutions (optimal interpolation, Kalman filter, and optimal smoother) to produce an estimate of the ``truth'' superior to the less theoretically sound estimates. We found no discernible improvements and identified the violation of standard data assimilation ``textbook'' assumptions on temporal whiteness of observational errors and system noise as the reason for this failure (Canizares et al. 2001). We work on a variety of possible ways to overcome these difficulties (Canizares et al. 2000).
We developed a statistical method of model bias correction based on the regression of the leading EOFs of the model errors and its states. This amounts to adding an interactive statistical component to the model, and therefore it is not just a model output statistical (MOS) correction, but a statistical modification of the internal dynamics of the model. This approach has been applied to correcting the Zebiak-Cane coupled model biases in SST, wind stress, and sea level. The correction was constructed via multivariate regression in the space of leading EOFs of these three fields. The large biases in the model SST and wind fields were eliminated, and the model forecasts were improved, particularly for the 1997 El Niño event (Chen et al. 2000). The corrected model became a basis for the new LDEO ENSO forecast system (LDEO4). This model was then used to explore the impact of scatterometer wind data and altimeter sea level data on ENSO prediction in terms of the retrospective forecasts of the 1999-2000 La Niña conditions (Chen, 2001). The new QuickSCAT wind product and the updated TOPEX sea level product were evaluated. Either TOPEX or the QuickSCAT observations could have improved the model predictions of the 1999-2000 La Niña, with the former being more effective. Yet the best simulations and predictions were obtained when both of these data sets were used. Therefore, it is advisable to assimilate multiple data sets so that they can complement one another in providing the correct initial conditions for the model. The TOPEX and QuickSCAT data are now being routinely used in our operational experimental forecasting.
 Figure 1: Small-scale variability (SSV) in the wind and sea level height: (upper left) SSV in the sea level height estimated from the TOPEX altimetry; (upper right) SSV in the zonal wind estimated from the NSCAT data; (lower left) sea level height variability in an ocean model response to the stochastic forcing with characteristics matching the observed wind SSV; and (lower right) the same for the spatially uniform forcing with increased decorrelation scales. (Click for larger image) |
We attempted to use satellite data, specifically TOPEX/POSEIDON (T/P) altimetry and NSCAT winds, to tune assimilation of in situ (tide gauge) data in order to produce a better hindcast of the tropical Pacific Ocean states for the pre-satellite period (Kaplan et al. 2000, Kaplan et al. 2001). We use the reduced space optimal smoother with a linear wind-driven ocean model and the dominant large-scale patterns of model variability as the basis for the covariance update computation. Assimilation of in situ data or even entire TOPEX altimetry fields under the assumption of a spatially uniform wind stress error leaves the TOPEX altimetry data residuals with a variance pattern of striking similarity to those exhibited by simulation and assimilation products based on more sophisticated models. The areas of large residual variance (zonally elongated maxima in the northwest and southwest tropical Pacific Ocean as well as the one near 10N in the eastern Pacific) is characterized by high energy of small-scale sea level height variability, as inferred from the TOPEX altimetry data. Monte Carlo experiments suggest that this variability can be interpreted to a large extent as the ocean response to the small scale wind variability, as inferred from the NSCAT data (Fig. 1). Ocean models driven by wind products which misrepresent or lack the small scale wind variability bias the variance pattern of the ocean response. Quantification of this missing variability will hopefully allow us to improve error covariance models and achieve more realistic assimilation results.
CAFE Home *
Summary *
Team
Components:
ENSO,
Climate,
Marine,
Vegetation
Education *
Instruments *
Acquisitions *
Publications
|
|
|
