Nonlinear aspects of sea surface temperature (SST) in Monterey Bay are examined,
based on an 85-year record of daily observations from Pacific Grove, California.
Oceanic processes that affect the waters of Monterey Bay are described, processes
that could contribute to nonlinearity in the record. Exploratory data analysis
reveals that the record at Pacific Grove is non-Gaussian and, most likely, nonstationary.
A more recent test for stationarity based on a power law approximation to the slope of
the power spectrum indicates that the record is stationary for frequencies up to ~8
cycles per year (~45 days), but nonstationary at higher frequencies. To examine the
record at Pacific Grove for nonlinear behavior, third-order statistics, including the
skewness, statistical measures of asymmetry, the bicorrelation, and the bispectrum,
were employed. The bicorrelation revealed maxima located approximately 365 days apart,
reflecting a nonlinear contribution to the annual cycle. Based on a 365-day moving window,
the running skewness is positive almost 80% of the time, reflecting the overall impact of
warming influences. The asymmetry is positive approximately 75% of the time, consistent
with the asymmetric shape of the mean annual cycle. Based on the skewness and asymmetry,
nonlinearities in the record, when they occur, appear to be event-driven with time scales
possibly as short as several days, to several years. In many cases, these events are related
to warm water intrusions into the bay, and El Nino warming episodes.
The power spectrum indicates that the annual cycle is a dominant source of variability in
the record and that there is a relatively strong semiannual component as well. To determine
whether or not the annual and semiannual cycles are harmonically related, the bispectrum and
bicoherence were calculated. The bispectrum is non-zero, providing a strong indication of
nonlinearity in the record. The bicoherence indicates that the annual cycle is a major source
of nonlinearity and further implies that the annual and semiannual cycles are harmonically related.
Based on the wavelet power spectrum (WPS), the appearance of the semiannual cycle is transitory;
however, pathways between the annual and semiannual cycles appear at certain times when nonlinear
interaction between them could occur. Comparisons between the WPS and the running skewness suggest
that there is a tendency for periods when pathways exist, to coincide with increased positive skewness,
and, often, with El Nino warming episodes. The Hilbert-Huang transform, a relatively new tool for
nonstationary and nonlinear spectral analysis, was used to further examine the origin of the semiannual
cycle. The time-dependent Hilbert spectrum reveals large and erratic variations in frequency associated
with semiannual cycle but far greater stability associated with the annual cycle. As a result, the
time-integrated Hilbert spectrum does not indicate the presence of a semiannual cycle. The method of
surrogates from the field of nonlinear dynamics was also employed to test the Hopkins record for
nonlinearity. Differences between the data and the surrogates were found that were statistically significant,
implying the existence of nonlinearity in the record. Using the method of surrogates together with a
one-year moving window, El Niño warming episodes appear to be a likely source of nonlinearity, consistent
with the other analyses that were performed. Finally, the influence of stochastic variability due to
serial correlation in the data was examined by comparing standardized statistics for the observations
and for simulations based on an autoregressive model whose properties were obtained from the observations.
The magnitude of the variability for the simulations was found to be far less than that associated with
the original data, and thus stochastic variability does not appear to be a factor that significantly
affects the interpretation of our results.