Development of a lithological model for Ta Thiet limestone deposit using geostatistics technique and machine learning

Abstract

The establishment of a 3D lithological model is a challenging task in Ta Thiet quarry limestone due to the sparseness of the exploration drilling pattern along with the imbalance of drill log data. This study proposes a hybrid Random Forest (RF) which is the combination of standard RF and geostatistical Indicator Kriging (IK). Using the spatial probability generated by IK, the RF model captures the geological context. The result shows that the hybrid RF achieved an overall accuracy of 93.0%, which outperformed the standard RF (80,0%) and IK (61,0%) and improved its ability in detecting minor rock types. The hybrid RF model also generates a high solution of lithological boundaries. This is potentially a reliable tool for selective mining and blending of cement raw mmaterials.