We develop a maximum likelihood based method of reconstructing band powers of the density and velocity power spectra at each wavenumber bins from the measured clustering features of galaxies in redshift space, including marginalization over uncertain
ties inherent in the Fingers-of-God (FoG) effect. The reconstruction can be done assuming that the density and velocity power spectra depend on the redshift-space power spectrum having different angular modulations of mu with mu^{2n} (n=0,1,2) and that the model FoG effect is given as a multiplicative function in the redshift-space spectrum. By using N-body simulations and the halo catalogs, we test our method by comparing the reconstructed power spectra with the simulations. For the spectrum of mu^0 or equivalently the density power spectrum P_dd(k), our method recovers the amplitudes to a few percent accuracies up to k=0.3 h/Mpc for both dark matter and halos. For the power spectrum of mu^2, which is equivalent to the density-velocity spectrum P_dv(k) in the linear regime, our method can recover the input power spectrum for dark matter up to k=0.2 h/Mpc and at both z=0 and 1, if using the adequate FoG model. However, for the halo spectrum, the reconstructed spectrum shows greater amplitudes than the simulation P_dv(k). We argue that the disagreement is ascribed to nonlinearity effect that arises from the cross-bispectra of density and velocity perturbations. Using the perturbation theory, we derive the nonlinear correction term, and find that the leading-order correction term is proportional to mu^2 and increases the mu^2-power spectrum amplitudes at larger k, at lower redshifts and for more massive halos. We find that adding the nonlinearity correction term to the simulation P_dv(k) can fairly well reproduce the reconstructed P_dv(k) for halos up to k~0.2 h/Mpc.
