Quasi-periodic, global oscillations in sea level pressure on intraseasonal timescales

Abstract

The sea level pressure (SLP) variability in 30–60 day intraseasonal timescales is investigated using 25 years of reanalysis data addressing two issues. The first concerns the non-zero zonal mean component of SLP near the equator and its meridional connections, and the second concerns the fast eastward propagation (EP) speed of SLP compared to that of zonal wind. It is shown that the entire globe resonates with high amplitude wave activity during some periods which may last for few to several months, followed by lull periods of varying duration. SLP variations in the tropical belt are highly coherent from 25°S to 25°N, uncorrelated with variations in mid latitudes and again significantly correlated but with opposite phase around 60°S and 65°N. Near the equator (8°S–8°N), the zonal mean contributes significantly to the total variance in SLP, and after its removal, SLP shows a dominant zonal wavenumber one structure having a periodicity of 40 days and EP speeds comparable to that of zonal winds in the Indian Ocean. SLP from many of the atmospheric and coupled general circulation models show similar behaviour in the meridional direction although their propagation characteristics in the tropical belt differ widely.

Appendix I

The purpose here is to explore if the pressure signal, after removing the zonal average, has a standing wave mode or not at zonal wavenumber one. In the case of a standing wave, the amplitudes (hence the power) of positive and negative wave numbers are equal. Hayashi (1979) proposed a method to separate travelling and standing modes in a variable using Fourier spectral analysis, and we basically followed this procedure. However, the standard wavenumber–frequency spectrum, based on the entire time series, will not reveal if the eastward and westward travelling waves were simultaneously present or not. It is desirable to have information on temporal variation of wave activity for the desired wavenumbers in the specified frequency domain. Wavelet transform can provide such information. But for the global data at wavenumber one, edge effects (cone of influence) can affect the coefficients obtained when we use wavelets in spatial domain. Therefore, here we carried out a combined FFT-wavelet analysis to extract the information on wave activity. First FFT is performed for each day in the zonal direction to calculate the coefficients at different wavenumbers. The resulting FFT coefficients are functions of wavenumber and time. Then for each wavenumber, the time series of these coefficients are used to calculate wavelet coefficients for different time scales τ. We used complex Morlet wavelet as the mother wavelet function ψ (e.g., Torrence and Compo 1998). The wavelet coefficient W _k corresponding to wavenumber k at time t _o and time period τ is given by,

$$ W_k (t_0 ,\tau ) = \frac{1}{{\sqrt \tau }}\int\limits_{ - \infty }^\infty {f_k (t)\psi ^* \left( {\frac{{t - t_0 }}{\tau }} \right){\text{d}}t} $$

(A1)

where f _k is the FFT coefficient for wavenumber k at time t ₀ and τ is the timescale. To obtain the energy in the timescales of interest (τ₁ to τ₂), the spectral power is calculated from the following expression.

$$ \overline {W_k ^2 } (t_0 ) = \sum\limits_{\tau = \tau _1 }^{\tau = \tau _2 } {\left| {W_k (t_0 ,\tau )} \right|^2 } $$

(A2)

Using this method, the wavenumber–frequency spectra similar to that shown in Fig. 3 of Wheeler and Kiladis (1999) can be reproduced. In this study, we took the sum of all the coefficients corresponding to timescales from 30 to 60 days for k = ±1 to obtain power in eastward and westward moving waves, respectively.