Wavelet method determination of long period tidal waves and polar motion in superconducting gravity data X.-G.. Hu 1,2,*, L.T. Liu 1, Ducarme. B. 3, H.T. Hsu 1, H.-P. Sun 1 1Institute of Geodesy and Geophysics, Chinese Academy of Science, 340 Xu-Dong Road, Wuhan, China 2Graduate School of Chinese Academy of Sciences, Beijing, China 3Royal Observatory of Belgium, Brussels

SUMMARY Introduction Wavelet filtering Local atmospheric pressure effects on observation of the LP tidal waves Separation of eight long period tidal groups at periods 4~40 days Separation of LP wave Ssa Polar motion effect –Fitting cosine functions to the observed pole tide –Fitting theoretical pole tide in the time domain –Fitting theoretical pole tide in the frequency domain Discussion and Conclusions

Introduction 5060-day uninterrupted gravity records, extending from 15, 11, 1986 up to 22, 9, 2000, from SG GWR T003 at the Royal Observatory of Belgium have been until now the longest continuous tidal gravity observations. We select 4800-day stable records for the study of pole tide. The initial phases for EOP and gravity observations are referred to 12, 03, 1987, h UTC as initial epoch and 1, 05, 2000, h UTC as final epoch.

Wavelet filtering A discrete signal f N (t) can be expressed as a wavelet expansion by orthonormal basis functions derived from two closely related functions: scaling function  (t) and wavelet function  (t). where J, j and n are integer indices, N is the length of the discrete signal and suppose N=2 M, 0<J< M. The above expansion means that scaling and translating  (t) and  (t) lead to an orthogonal basis for the analysis of signal.

Wavelet filtering The scaling function  (t) and wavelet function  (t) used to construct the orthonormal basis are derived from h(k), g(k) are called scaling coefficients and wavelets coefficients respectively.

Wavelet filtering The 3-level wavelet analysis filter bank decomposes an input signal f(t) within Nyquist frequency band [0, F] into eight sub-band signals. Synthesis filter banks reconstruct the input signal only with w3,3, thus the output signal, which has the same length as the input signal, is located in sub-band [3F/2 3, 4F/2 3 ]

Wavelet filtering Table 1. Sub-bands for the analysis of 12 LP tidal groups p wavelet packet filter bank with a split factor of 3 or 2.

Wavelet filtering The moment condition of a Daubechies wavelet can be expressed in terms of the scaling coefficients as It indicates that a wavelet filter has L/2 vanishing moments, which means wavelet filter can suppresses parts of the signal which are polynomial up to degree L/2-1. instrument drift in the signal can be suppressed thoroughly, whether in long-term or in short-term, harmonic components are smoothly represented.

Wavelet filtering The wavelet bandpass filter is able to filter LP tidal groups into very narrow frequency bands with good frequency response and without phase shift and Gibbs-like phenomena. Moreover, it can thoroughly suppress instrumental drift in SG records.

Example of observation equations tidal group MStm, Mtm can be simply modeled as the combination of a set of cosine functions and atmospheric noise There are 27 tidal constituents in the MStm group and 24 constituents in the Mtm one. C is the unknown pressure regression coefficient

Local atmospheric pressure effects Pressure admittance for LP tidal bands We calculated frequency-domain admittances in these bands using the method proposed by Crossley et al. (1995).

Local atmospheric pressure effects

Separation of tidal groups at periods 4~40 days

Shallow water tides If the tidal forcing contains two frequencies ω 1 and ω 2, compound harmonic constituents at frequency ω 1 -ω 2 and ω 1 +ω 2 will be created. M 2 ( day period) and S 2 (2-day period) induce strong sea level oscillations at period of MS 4 (6.103 hours=1/(ωM 2 +ωS 2 )) and MSf (14.77 days=1/(ωS 2 – ωM 2 )). The amplitude of the compound tide MSf is about 2.53 cm in Ostend (Belgium) and its amplification factor is 17 with respect to the static tide. a shallow water tide MSm is induced by μ 2 and N 2 (31.81 days=1/(ωN 2 –ωμ 2 ) ), with a amplitude of about 1.75cm and amplification factor of 10.

Separation of LP wave Ssa

There are environmental noises of unknown origin at frequency about cpd and cpd. Observed Ssa is modulated with a period of about 4.5 year The fitting result gives δ=1.0257±0.0067, κ= ± and C= ± nms -2 hPa -1. The value of amplitude factor at the Ssa tide is considerably smaller than the theoretical value by DDW99.

Polar motion effect

In station Brussels, local atmospheric pressure correction did not reduce the variance of the residuals in the gravity signal at frequencies below 2 cycles/year.

Polar motion effect Theoretically for a rigid Earth, the gravity variations induced by polar motion can be predicted by An eight-level undecimated filter bank derived from Daubechies wavelet of order 127 is utilized to filter the gravity and local atmospheric pressure signal into a very narrow subband 0.5/2 8 ~ 0.5/2 7 cpd, i.e ~ cpd (period 256 to 512 day). Thus instrument drift and all long-period tides, except the annual term Sa, are removed. The theoretical pole tide is also filtered into this sub-band.

Fitting cosine functions to the observed pole tide The spectral analysis of 4800-day predicted pole tide suggests a 435 days Chandler period and an annual period at days. The SG records in the period band 256~512 day can be simply modeled as sum of three cosine functions (1=Chandler, 2=annual, 3=9 months) Similarly the theoretical pole tide is modeled as  = A 1 /B 1  =a 1 -b 1

Fitting cosine functions to the observed pole tide Our estimate yields the results δ=1.1681±0.0095, κ= o ± The negative value of κ implies a phase advance between the observed pole tide and the theoretically predicted pole tide. With atmospheric pressure correction δ=1.1784±0.0092, κ= o ± and C= nms -2 hPa -1.

Fitting cosine functions to the observed pole tide A tiny change of the selected Chandler frequency for harmonic analysis will cause a significant change of amplitude factor δ. – For example, at the period of 433 days δ=1.1578±0.0095, κ= o ±0.9064; –at the period of 438 days δ=1.1735±0.0092, κ= o ±

Fitting theoretical pole tide in the time domain Since the Earth response to the long-period tide is not strongly frequency dependent, we may accept that amplitude factor δ at the Chandler period is equal to that at annual period The parameters A 1, A 2, b 1, b 2 and amplitude factor δ are estimated by fitting  g(t) to  p(t). The optimum time lag  CH is determined by experimenting different values of  CH in adjustment of A 1, A 2, b 1, b 2 and δ

Fitting theoretical pole tide in the time domain At the optimum value of -3 days, this method yields δ=1.1638±

Fitting theoretical pole tide in the frequency domain Usually the gravimetric factor is defined in the frequency domain as: where f represents frequency, G(f) and P(f) are the Fourier transform of observed pole tide g(t) and theoretical pole tide p(t), We try to minimize |  G(f)| 2 with over a narrow frequency range f=1/454~1/418 cpd  G(f)= G(f)–δ(f)P(f)

Fitting theoretical pole tide in the frequency domain Using a Hanning window, we obtain δ=1.1631±0.0045, κ= o ± Note that this value of amplitude factor δ is very close to the value which we just directly obtained in the time domain, and the phase difference κ is near to -2 days.

Comparison of Results

CONCLUSIONS Wavelet filtering is used successfully to determine LP waves and polar motion amplitude factors and phase differences. Local pressure correction is reducing drastically the residues for periods up to 60 days. Shallow water tides loading effect is clearly seen at Msf frequency in Brussels. The determination of the pole tide amplitude factor depends strongly on the atmospheric correction model. If local pressure correction is applied the results are in good agreement with previous determinations.