scope:

Preface

This Geotechnical Special Publication consists of a series of papers presenting a wide range of methods of probabilistic site characterization and demonstrating them using soil data obtained at the National Geotechnical Experimentation Sites (NGES), in particular the sites at Texas A&M University, Treasure Island Naval Station (California), the University of Massachusetts at Amherst, the University of Houston and Northwestern University. An earlier volume in this series, Geotechnical Special Publication No. 93, provides relevant background material on the layouts of borings and cone penetrometer test (CPT) soundings, the geological setting, and the types of field and laboratory tests performed at each site.

Among the topics covered in this publication are, in sequence, statistical estimation procedures based on homogeneous random field theory, with as principal parameters the mean, the coefficient of variation and the scale of fluctuation (O'Neill and Yoon; Wu; Akkaya and Vanmarcke), two different fractal representations of spatial variation in soil deposits (Kulatilake and Urn; Fenton and Vanmarcke), a neural network model (Juang and Jiang), and a new approach to correlating CPT data with soil type and engineering properties based on the concept of fuzzy subsets (Zhang and Tumay). There arc results for both horizontal and vertical variation of measured soil properties - most often CPT data -- under various assumptions about trends-in-the-mean and layering by soil type, or about the validity of combining, for statistical analysis purposes, data from similar non-contiguous soil deposits (e.g.. data on variation-with-depth of CPT tip resistance from different clay layers at one site or from all layers at all five sites).

Useful perspectives on the different sources of uncertainty of soil properties is offered, in this volume, by Kulatilake & Urn and Akkaya & Vanmarcke. Equations for the variogram and several equivalent representations of second-order statistics of homogeneous (or stationary) one-dimensional random variation, can be found in the papers by O'Neill & Yoon and Kulatilake & Um. The paper by Fenton and Vanmarcke focuses on alternative spectral representations. The results for the various second-order descriptors (correlation functions, variograms, spectral density functions) and their parameters (like the scale of fluctuation) differ greatly depending on whether (and how) trends-with-depth or layering-by- soil-typc arc modeled, as informed cither by the site-specific measurements or by information about site geology. The papers by O'Neill and Yoon (using data from the University of Houston NGES) and Wu (integrating new data from the Northwestern University NGES) report stable estimates for coefficients of variation and scales of fluctuation based on analyses of either the raw data originating from nominally homogeneous domains or normalized data (obtained from measured values by first subtracting trends or layer-specific means and perhaps dividing residuals by a local standard deviation).

When trends and other site-specific information are ignored in the data processing, evidence of fractal behavior, implying fluctuations across a range of spatial scales, appears in the second-order statistics (Fenton and Vanmarcke, Kulatilake and Um); the sampling interval and the size ot the domain sampled or analyzed also atlcct the parameter estimates. Transformation of the raw data prior to statistical analysis, for example by first calculating the logarithms of the measured CPT tip resistance values (Fenton and Vanmarcke). further complicates interpretation of the results. The paper by Akkay and Vanmarcke reports estimated coefficients of variation and scales of fluctuation for many different soil properties at the Texas A&M NGES sand and clay sites. Kulatilake and Urn concentrate on statistical analyses of cone tip resistance data from the Texas A&M NGES clay site, while Fenton and Vanmarcke aggregate all the CPT data from the five NGES, disregarding trends and other site-specific information. Only Zhang and Tumay's paper deals explicitly with cross- correlation between measured values, focusing on how CPT data inform about soil type and engineering properties.

Knowledge about the inherent variability of soil properties is of critical importance in reliability analysis of gcotcchnical facilities, risk assessment for decision support or regulatory control, and planning and optimization of site-specific exploration and testing. The dual aim of this publication is to present an overview of traditional and novel statistical methods of soil profile modeling and provide a set baseline statistics, representing the well- documented NGES. that can serve as a priori information, in a Bayesian sense, about pattern of spatial variation of soil properties in probabilistic site characterization worldwide.

The papers in this volume, having been accepted for publication by the editors based on a process of peer review in accordance with the standards of ASCE and the Geo-Institute, are eligible for discussion in the Journal of Gcotcchnical and Geo-Environmental Engineering and for ASCE awards. Early versions of most of the papers were presented, discussed and criticized at a workshop held in conjunction with the Geo-Institute Conference in Seattle. WA (1998) and sponsored by the G-I Committee on Risk Assessment and Management.

The Federal Highway Administration (FHWA) and the National Science Foundation (NSF) provided funding for the NGES program during the 1990's. Support for the project on probabilistic site characterization at the NGES, conducted by the G-I Committee on Risk Assessment and Management, and for related research by Dr. Akkaya and co-editor Dr. Fenton during extended visits to Princeton University, came from FHWA through the Geo- Institute of ASCE. Special thanks are due to Albert F. DiMillio of FHWA for his leadership in the NGES program and to Carol Bowers of the Geo-Institute for critical project management support.