A Baseline Statistical Model for
Tropical Pacific Wind Stress Anomalies

Andrew T. Wittenberg
Atmospheric and Oceanic Sciences Program, Princeton University

Matthew J. Harrison

14th Conference on Atmospheric and Oceanic Fluid Dynamics, paper #P4.2

Abstract: A simple statistical model of observed monthly-mean tropical Pacific wind stress anomalies is proposed as a baseline for evaluation and intercomparison. The model consists of (1) a linear regression onto the leading covariant patterns of stress anomalies and sea surface temperature anomalies (SSTAs), and (2) a red-noise model based on the leading patterns of the residual stress. Statistical models of this type are compared for 1961-79 and 1980-99 using the Florida State University (FSU) and NCEP/NCAR Reanalysis (NCEP) wind stress products.

The statistical model captures the stress signals most relevant to ENSO, and reveals large differences between the observational analyses. The NCEP stress response to SSTAs is weaker, more zonal, and extends farther east than that in FSU, especially prior to 1980. During 1961-79, there is hardly any propagation of stress anomalies despite a tendency for SSTAs to propagate westward; during 1980-99, both the stress anomalies and SSTAs propagate eastward. More than 75% of the stress anomaly variance is not linearly related to large-scale SSTAs, with FSU noisier than NCEP in this sense. The residual stress shows weak lag correlations with central Pacific SSTAs, such that westerly events are preceded 6-12 months earlier by cooling, and followed 6-10 months later by warming.

Just as illuminating as its successes are the model's failures. That the residual stress is autocorrelated in time, is not normally distributed, and has nonstationary variance suggests a discernable nonlinear relationship between the SSTAs and stress anomalies. In the equatorial west Pacific, the residual activity is stronger between November and March than during the rest of the year, and was particularly strong leading up to the 1997-98 warm event. Possible improvements and uses of the statistical model are discussed.

Poster in color (PDF, 1.6 MB)

Extended abstract in color (PDF, 0.8 MB)

Extended abstract in black-and-white (PDF, 0.8 MB)