Two-dimensional prediction of the interface of geological formations: A comparative study

• Boreholes data from Singapore were used to predict geological interface locations. Prediction accuracies and uncertainties for different methods compared. MARS CRF generally have higher prediction than the TPSI method. shows spatial trend of interfaces more clearly TPSI. A zonation in a geologically complex area improves performance MARS. The location formations is an important piece information tunneling construction. As site investigation are usually limited, locating sections between boreholes can be large challenging estimate. suitable geostatistical method thus needed interfaces. In this paper, three commonly methods, namely multivariate adaptive spline regression (MARS), conditional random field (CRF) method, thin-plate interpolation (TPSI) evaluated two-dimensional cases using sites Singapore. accuracies, patterns predicted surfaces, obtained also proposed improve accuracy results indicate that show other two methods. produces undesirable oscillations surface may underestimate extreme values elevations. general, similar but For with very limited areas, larger errors However, former significantly improved if reasonable performed.