The potential of discrimination and grouping of digital levels of image classification methods based on machine-learning algorithms allow obtaining good results in the classification of land coverages. The machine-learning algorithm Random Forest is considered a robust algorithm for classification and regression, presenting good performance for data of high dimensionality, as is the case of the satellite imagery stored in the Colombian Data Cube (CDCol). This paper aims to present the implementation of Radom Forest on the CDCol infrastructure for land cover classification, on the Orinoquía Natural Region in Colombia. We used Landsat 8 OLI imagery data for 2016 at surface reflectance level and seven thematic land cover classes for the supervised classification. The overall thematic accuracy assessment was 86% and Kappa index of agreement was 79%. The results suggest that this method produces an accurate sub-pixel characterization of the land cover classes that is acceptable for practical applications.