A common application in seismic imaging of machine learning algorithms (Artificial Neural Networks) is to produce petrophysical models at seismic scale combining well-log information and seismic data. Here we use these resulting models as prior inputs in full-waveform inversion (FWI). We compute instantaneous seismic attributes to a stacked P-wave reflected seismic section in the Tenerife field located in Colombia and train Artificial Neural Networks (ANN's) to estimate P-wave velocity Vp, density ρ and volume of clay Vclay at seismic scale. The logs are provided by a well near the survey line, allowing images of different rock properties to be used in the inversion of velocities. We extend the use of the volume of clay by constraining it with the well lithology, consequently categorizing the Vclay by two classes: sands and shales. This process allows us to build an initial estimate of the earth property model, which is iteratively refined to produce a synthetic seismogram (by means of forward modeling) to match the observed seismic data. We use the 1-D Kennett method as forward modeling tool to create synthetic data using the images of Vp, ρ and the thickness of layers (sands or shales) obtained with the ANN's. A nonlinear least-squares inversion algorithm minimizes the residual (or misfit) between observed and synthetic full-waveform data improves the P-wave velocity resolution. The interpreter can thus delineate the intercalation of sands (saturated with hydrocarbons) and shales that are undetected using current techniques. This approach is similar to the one given in Parra et al. (2017). Presentation Date: Wednesday, October 17, 2018 Start Time: 9:20:00 AM Location: Poster Station 1 Presentation Type: Poster