Probability of False Correlations of Layer Properties With Geophysical Multidimensional Attributes

Multivariate linear regression is widely used for predicting geological properties. When the amount of reference hard data is limited the risk of false correlations between predicted parameters and attributes becomes significant. These false correlations do not correspond to any real relationships and result in principally wrong predictions. We performed stochastic simulation using random sampling of both reference data and attributes which allowed to quantitatively evaluate probablity of high rank false correlations and to substantiate the recommendations of parameter selection for applications of multivariate regressional analysis for geophysical data interpretation. The most important parameters include the number of attributes, the numebr of informative resgressors, determination coefficient.