The Triassic-Jurassic deep sandstone reservoirs in onshore Denmark are known geothermal targets that can be exploited for sustainable and green energy for the next several decades. The economic development of such resources requires accurate characterization of the sandstone reservoir properties, namely, volume of clay, porosity, and permeability. The classic approach to achieving such objectives has been to integrate well-log and prestack seismic data with geologic information to obtain facies and reservoir property predictions in a Bayesian framework. Using this prestack inversion approach, we can obtain superior spatial and temporal variations within the target formation. We then examined whether unsupervised facies classification in the target units can provide additional information. We evaluated several machine learning techniques and found that generative topographic mapping further subdivided intervals mapped by the Bayesian framework into additional subunits.