Oil-source correlation in the slope of the Qikou Depression in the Bohai Bay Basin with discriminant analysis

The commonly-used tools for oil-source correlation, such as mass chromatograms of biomarkers and bivariate cross-plots of geochemical parameters, cannot deal with multiple geochemical parameters and plenty of samples simultaneously, leading to uncertainties in the results and even failures sometimes. In this paper, the discriminant analysis (DA) is selected from supervised machine learning algorithms, as it is superior to the commonly-used tools as well as other multivariate statistical methods, with the accumulation of geochemical data of source rocks and oils. In the slope of the Qikou Depression, the main source rocks in the third (Es3) and first members (Es1) of the Paleogene Shahejie Formation were deposited in similar depositional environments. The source rocks cannot be distinguished with the commonly-used tools. We firstly extended geochemical parameters and then used stepwise DA to select informative parameters and to develop a discriminant model for oil-source correlation. The 22 selected parameters are supported by geochemical characteristics of the source rocks in the study area. The DA of these parameters for oil-source correlation achieved a high correct rate of original validation (96.8%) and leave-one-out cross-validation (89.4%), indicating a sufficient discriminatory power. The oil-source correlation results with high posterior probabilities, showing strong similarity between the sources and oils, coincide with geological conditions and illustrate that there is still much exploration potential in the study area, especially for the Es3 petroleum system. All these illustrate that DA is one of the most useful tools for oil-source correlation with the accumulation of geochemical data of source rocks and oils.