Sea states are usually the combination of several time-evolving wave systems whereas the classical spectral estimation methods assume stationarity. A method that adapts to the dynamical evolution of the spectral components is proposed to improve both omnidirectional and directional sea wave spectral estimations. In this method, periodograms are computed for each sea state as in the conventional methods, and rather than only smoothing individual periodograms, the overall time-history of periodograms are simultaneously smoothed in frequency and time dimensions. Since a simple two dimensional averaging would not be appropriate because the temporal evolution of the wave systems reflects typical non-stationary behaviors, we use either kriging or adaptive 2D kernel density estimators that allow the taking care of the spectral component frequency–time evolutions. The method is successfully validated on sequences of spectra typical of sea-state conditions in West Africa. The comparison with the simple 2D averaging method and individual periodogram smoothing method shows that the proposed method gives higher effective numbers of degrees of freedom, better estimates of the spectral shape and reliable spectral moments. The method also provides a tool for sea wave spectra interpolation and may thus be used to fill in missing values and improve wave systems tracking for storm identification purposes.