Land seismic data quality can be severely affected by near-surface anomalies. The imprint of a complex near-surface can be removed by redatuming the data to a level below the surface, from where the subsurface structures are assumed to be relatively smooth. However, to derive a velocity-depth model that explains the propagation effects of the near-surface is a non-trivial task. Therefore, an alternative approach has been proposed, where the redatuming operators are obtained in a data-driven manner from the reflection event related to the datum. In the current implementation, the estimation of these redatuming operators is done in terms of traveltimes only, based on a high-frequency approximation. The accompanying amplitudes are usually derived from a local homogeneous medium, which is obviously a simplification of reality. Such parametrization has produced encouraging results in the past but cannot completely remove the near-surface complexities, leaving artefacts in the redatumed results. In this paper we propose a method that estimates the redatuming operators directly from the data, i.e., without using a velocity model, in a full waveform manner, such that detailed amplitude and phase variations are included. The method directly outputs the inverse propagation operators that are needed for true-amplitude redatuming. Based on 2D synthetic data it is demonstrated that the resulting redatuming quality is improved and artefacts are reduced.