Estimation of autocorrelation distances for in-situ geotechnical properties using limited data

A simultaneous estimation of the vertical and horizontal autocorrelation distances (ACDs) for geotechnical properties under the condition of limited data (fewer than one hundred data points) is not well studied. This paper evaluates the performance of the maximum likelihood estimator (MLE) in estimating autocorrelation distances using some simulated data. The effect of the autocorrelation structure on the accuracy of ACD estimation is also investigated. It is found that MLE may produce highly biased estimations if limited data are available. Hence, a clustered borehole layout scheme and a resampling method are proposed to improve the estimation accuracy. The methods are illustrated using some simulated data, standard penetration test (SPT) data and cone penetration test (CPT) data. It is found that the clustered borehole layout scheme and resampling method yield more accurate estimations of ACDs in analyses using both the simulated data and real data.