INVERSION OF A DOUBLE BSR IN THE EAST NANKAI ACC...
Line HR09was acquired by SFJ2000（Seismogenic zone experiment by French-Japanese cooperate study）in the eastern Nankai area in summer of 2000（Fig.1）.The seismic signals were received with the 360channel hydrophone array with 12.5m group interval，sampled at 1ms.The record length was 6s.Shot interval was 25m.Source to near trace offset was 275m.Source was a 151cubic inch array made of five air guns of various chamber size.The processing was focused on obtaining better image of sub-bottom shallow structure similar to SONG et al.［3］，and then investigating the seismic characteristics of BSR distribution.Processing consisted of geometry definition，common mid-point（CMP）sort，velocity analysis，normal moveout（NMO）correction，stack and migration.The section of migration profile is shown in Fig.2.The first BSR and the second BSR are both strong amplitude reflectors with negative polarities，crossing the bedding reflectors.
Fig.1 Bathymetry and location of seismicreflection profile HR09in the area of eastern Nankai Trough.
Fig.2 Section of Migration profile of HR09
At CMP 2140，the first BSR and the second BSR are clear，where the slope of the first BSR is about 4°，both the seafloor and the second BSR subparallel to it and can be approximated a one-dimensional structure.Four CMP gathers around CMP 2140are combined to form a supergather with the group interval of 12.5m to avoid the spatial alias problem.Then it is transformed to the data in intercept time-slowness domain（Fig.3）.Five reflectors are picked in this domain（Fig.4）.They are the sea floor reflector，the first BSR，the second BSR and two reflectors below the second BSR.Travel-time information is input to the very fast simulated annealing algorithm，and interval velocities are obtained as the result of the minimum of the observed travel-time and the calculated travel-time while vertical travel-times vary around the picked ones in the range of 32ms in the procedure.Ten random models are used as initial models for the inversion.The final velocity structure based on the inversion results is shown in Fig.5，and the observed and the calculated travel-times are shown in figure 4.The smoothed version of this velocity structure is obtained and used as the initial model of the full waveform inversion.
The seismic wavelet is carefully estimated from the seabed reflector for the full waveform inversion.The array directivity correction is done in the inversion.Because the source array is vertical to the profile，only receiver array directivity correction has been conducted.The S-wave velocities and the densities are derived from the P-wave velocities by using the relationship formula by Hamilton［18］and Gardner's equation.The S-wave velocities and densities are kept fixed in the inversion procedure，while S-wave velocities and densities are assumed to change little across BSR.
Fig.3 Transformed seismic data of CMP2140-2137in intercept time-slowness domain (https://www.daowen.com)
Fig.4 The observed and calculatedtravelling time of five reflectors in intercept time-slowness domain
Fig.5 Comparison of velocity models 1：the inverted velocity model by Simulated annealing algorithm；2：velocity trend model based on curve 1；3：the inverted 10m-interval velocity model by full waveform inversion using curve 2as the starting model；4：the inverted l0m-interval velocity model by full waveform inversion using curve 1as the starting model.
Fig.6 Comparison of velocity models1：the inverted 5m-interval velocity model by full waveform inversion using curve 3in Fig.5as the starting model；2：Velocity trend model based on curve 1in Fig.6；3：the inverted 5m-interval velocity model by full waveform inversion using curve 4：in Fig.5：as the starting model.
The initial trend model is discretized to a layered model of which each layer is 10m thick and the inverted model is shown in Fig.5after the first run of inversion.Then the inverted model is discretized to a 5m-thick-layer model and another run of inversion is carried out.The final inverted model is shown in Fig.6.So multi-grid parameter technique is used in our full waveform inversion.
We also compare the different inverted results from different initial models for considering the sensitivity of our inversion.The 10m-thick-layer model from initial model is also shown in Fig.5and its variation is similar to the model from the smooth initial model.The 5m-thick-layer model by using curve 4in Fig.5as the initial model is also shown in Fig.6and it is very close the other one.The comparison indicates that inversion result is reliable.
There are two zones which have higher velocity than the trend velocity，and there are low velocity zones below these high velocity zones（Fig.6）.The first BSR corresponds to the first negative velocity contrast and the second BSR corresponds to the second negative velocity contrast.The high velocity zones above the first BSR and the second BSR are interpreted as gas hydrate bearing sediment layers.The low velocity zone beneath the second BSR can be related to gas bearing sediments，while the low velocity zone beneath the first zone can be related to non-gas-hydrate-bearing/water-saturated sediments or free gas-bearing sediments with little concentration.
