Dual?sensor (hydrophone and three?component geophone) data recorded
on the sea floor allow the elastic wavefield to be decomposed into
its upgoing and downgoing P? and S?wave components. Most decomposition
algorithms require accurate knowledge of the elastic properties of
the sea floor in the vicinity of the receivers and properly calibrated
sensors, in order for the data to be a faithful vector representation
of the ground motion. We present a multistep adaptive decomposition
scheme that provides the necessary information directly from the
data by imposing constraints on intermediate decomposition results.
The proposed scheme requires no a priori information and only a minimal
amount of user?defined input, thus allowing multicomponent data to
be decomposed in an automated data?driven fashion. The performance
of the technique is illustrated using seabed data acquired in the
North Sea with prototype single sensors (multicomponent geophones
individually sampled). Realistic sea floor properties and sensor
calibration operators are obtained, and elastic decomposition of
the calibrated data generally yields good results. Dominant water?layer
reverberations are successfully attenuated and primary reflections
are substantially enhanced in the computed upgoing P?wave potential
just below the sea floor. In contrast, the result for the upgoing
S?wave potential is somewhat less convincing; although the energy
of water?layer multiples is substantially reduced, notable amounts
of undesired multiple energy remain in this section after decomposition,
particularly at high offsets. These imperfections may point to inaccuracies
in the parametrization of the sea floor or remaining inaccuracies
in the vector fidelity of the horizontal geophone recordings. Nevertheless,
the results obtained with the extended data?driven decomposition
scheme are at least comparable to previously published results.