Geostatistical analysis of cone penetration test (CPT) sounding using the modified Bartlett test

Abstract

More in situ tests are typically carried out over the same volume of soil in comparison to laboratory tests on undisturbed borehole samples. Hence, geostatistical analysis of in situ test records should in principle provide a more accurate and representative overview of spatial variation. A natural probabilistic model for correlated spatial data is the random field. Although the random field provides a concise description of spatial variation, it poses considerable practical difficulties for statistical inference because of the underlying autocorrelation structure. This note presents an extended discussion of the modified Bartlett random field estimation procedure, which is capable of rejecting the null hypothesis of weak stationarity for spatially correlated data. In comparison with simple visual inspection and the standard run test, the modified Bartlett test is shown to provide three advantages: (i) it is a more consistent measure that is unaffected by the vagaries of subjective interpretation; (ii) it is sufficiently discriminative to decide if a section is stationary, even when visual clues are ambiguous; and (iii) it is capable of accommodating realistic constraints (e.g., short record length). The possibility of identifying secondary soil boundaries that may not be readily apparent from visual inspection of cone soundings, its robustness to alternate transformations of the cone data, and the sensitivity of the proposed procedure to different levels of significance are discussed.Key words: geostatistics, random field, stationarity, modified Bartlett test, level of significance, run test.