In ocean-bottom seismic (OBS) data processing, wavefield separation results are sometimes affected by high levels of noise on the vertical component Z, while the pressure component P is in general of good quality. Nonetheless, Z is needed to achieve complete pre-stack wavefield separation and also to drive processes such as mirror imaging and up-down deconvolution. To address the problem of noise on Z affecting wavefield separation results, we propose a new method which first estimates a multiple model from the downgoing wavefield in a least-squares fashion. Next, this multiple model is used as a measure of seismic signal coherency to calculate an enhanced upgoing wavefield with minimal noise degradation. We show the benefits of this new method on a real data example from the Caspian Sea