TorresHernandez Asu 0010N 10702 PDF

Estimating the Soil–Water Characteristic Curve
Using Grain Size Analysis and Plasticity Index
by
Gustavo Torres Hernandez
A Thesis Presented in Partial Fulfillment

of the Requirements for the Degree
Master of Science
Approved April 2011 by the

Graduate Supervisory Committee:
Claudia Zapata, Chair

Sandra Houston
Matthew Witczak
ARIZONA STATE UNIVERSITY
May 2011
ABSTRACT
The infrastructure is built in Unsaturated Soils. However, the geotechnical
practitioners insist in designing the structures based on Saturated Soil Mechanics.
The design of structures based on unsaturated soil mechanics is desirable because
it reduces cost and it is by far a more sustainable approach.
The research community has identified the Soil–Water Characteristic
Curve as the most important soil property when dealing with unsaturated
conditions. This soil property is unpopular among practitioners because the
laboratory testing takes an appreciable amount of time. Several authors have
attempted predicting the Soil–Water Characteristic Curve; however, most of the
published predictions are based on a very limited soil database.
The National Resources Conservation Service has a vast database of
engineering soil properties with more than 36,000 soils, which includes water
content measurements at different levels of suctions. This database was used in
this study to validate two existing models that based the Soil–Water Characteristic
Curve prediction on statistical analysis. It was found that although the predictions
are acceptable for some ranges of suctions; they did not performed that well for
others. It was found that the first model validated was accurate for fine-grained
soils, while the second model was best for granular soils.
For these reasons, two models to estimate the Soil–Water Characteristic
Curve are proposed. The first model estimates the fitting parameters of the
Fredlund and Xing (1994) function separately and then, the predicted parameters
are fitted to the Fredlund and Xing function for an overall estimate of the degree
i
of saturation. Results show an overall improvement on the predicted values when
compared to existing models. The second model is based on the relationship
between the Soil–Water Characteristic Curve and the Pore-Size Distribution of
the soils. The process allows for the prediction of the entire Soil–Water
Characteristic Curve function and proved to be a better approximation than that
used in the first attempt. Both models constitute important tools in the
implementation of unsaturated soil mechanics into engineering practice due to the
link of the prediction with simple and well known engineering soil properties.
ii
DEDICATION
To my mother and my family
iii
ACKNOWLEDGMENTS
I would like to especially thank Dr. Claudia E. Zapata for the opportunity
to work on the 9-23A project, “A National Catalog of Subgrade Soil–Water
Characteristic Curve Default Inputs and Selected Soil Properties for Use with the
ME-PDG”, which was the beginning inspiration for this study. Thank you, Dr.
Claudia, for your teaching, patience and assistance during my time at ASU.
I would like to thank Dr. Matthew Witczak for the opportunity to work on
the 9-44A project, “Validating an Endurance Limit for HMA Pavements”, which
was an important means of support for me and my family during my last year at
ASU. I am immensely grateful to Dr. Witczak for his guidance with my studies,
for his help and support and for his invaluable advices about engineering.
I would like to thank Dr. Sandra Houston for being part of my committee.
It was a great honor for me. Finally, I have to express a special gratitude to my
parents, wife and daughters, for their permanent encouragement and for never
losing their faith in me.
This work was like building a bridge, where after thousands of difficulties
many cars can pass safely on it. Now, I am a little more prepared to teach my
daughters about life and to my students about engineering.
Thank you for doing something great: ‘To teach’.
iv
TABLE OF CONTENTS
Page
LIST OF TABLES ................................................................................................. xi
LIST OF FIGURES ............................................................................................. xiv
CHAPTER
1 INTRODUCTION .......................................................................................1
1.1 Overview ....................................................................................1
1.2 Historical Background ...............................................................3
1.3 Research Objectives ...................................................................5
1.4 Methodology ..............................................................................6
1.5 Organization of the Thesis .........................................................7
2 LITERATURE REVIEW ............................................................................9
2.1 Introduction ................................................................................9
2.2 Stress State Variables ...............................................................10
2.3 Moisture Flow in Unsaturated Soils ........................................13
2.4 Matric Suction and the Soil–Water Characteristic Curve ........16
2.5 Approaches to Obtain Unsaturated Soil Properties .................17
v
CHAPTER Page
2.5.1 Level 1 ......................................................................18
2.5.2 Level 2 ......................................................................21
2.5.3 Level 3 ......................................................................22
2.5.4 Level 4 ......................................................................22
2.6 Approaches to Predict the Soil–Water Characteristic Curve ...23
2.7 Final Remarks ..........................................................................26
3 DATABASE COLLECTION ....................................................................27
3.1 Source of the Database.............................................................27
3.2 Characteristics of the Database ................................................28
3.3 Data Collection ........................................................................30
3.3.1 Master File Properties and Characteristics ...............33
3.3.2 Preliminary Reduction of Soil Unit Data ..................34
3.3.3 Selecting the Proper Component to Represent the
Map Unit ...................................................................34
3.3.4 Properties Included in the SWCC Predicting
Analysis.....................................................................34
vi
CHAPTER Page
3.4 Working with the Database ......................................................46
3.4.1 Calculating Soil–Water Characteristic Soils
Parameters .................................................................46
3.4.2 Calculating Grain–Size Distribution Parameters ......53
3.4.3 Calculating Particle Size Values ...............................59
3.4.4 Other Calculations ....................................................62
3.5 Summary ..................................................................................66
4 VALIDATION OF AVAILABLE SWCC MODELS ...............................70
4.1 Introduction ..............................................................................70
4.2 Validating Zapata’s Model ......................................................71
4.2.1 Model for Plastic Soil ...............................................71
4.2.2 Model for Non–Plastic Soil ......................................77
4.2.3 Zapata’s Model Validation Analysis ........................78
4.3 Validating MEPDG Model (Witczak et al, 2006) ...................87
4.3.1 Predictive Equations for Fredlund and Xing SWCC
Parameters for Non-Plastic Soils ..............................89
vii
CHAPTER Page
Parameters for Plastic Soils ......................................92
4.3.3 MEPDG Model Validation Analysis ........................93
4.4 Summary ................................................................................100
5 A NEW SWCC MODEL BASED ON SWCC PARAMETERS ............107
5.1 Overview ................................................................................107
5.2 Database and Descriptive Statistics .......................................113
5.3 Correlations ............................................................................119
5.3.1 Correlations for Fine Grained Soils ........................119
5.3.2 Correlations for Granular Soils ...............................122
5.4 Physical Significance of SWCC Parameters..........................125
5.5 SWCC Prediction Models for Fine–Grained Soils ................130
5.5.1 Modeling SWCC Parameter a f ...............................132
5.5.2 Modeling SWCC Parameter b f ...............................137
5.5.3 Modeling SWCC Parameter c f ...............................141
5.5.4 Modeling SWCC Parameter h r ...............................146
viii
CHAPTER Page
5.6 New Prediction for Granular Soil ..........................................153
5.6.1 Modeling SWCC Parameter a f ...............................155
5.6.2 Modeling SWCC Parameter b f ...............................161
5.6.3 Modeling SWCC Parameter c f ...............................164
5.6.4 Modeling SWCC Parameter h r ...............................167
5.7 Measured versus Predicted Degree of Saturation ..................168
5.8 Procedure to Estimate the SWCC from the Proposed
Models...................................................................................172
5.9 Summary ................................................................................176
6 SWCC MODEL PREDICTED FROM GRAIN SIZE DISTRIBUTION183
6.1 Introduction ............................................................................183
6.2 Calculating Suctions from the Soil–Water Characteristic
Curve ......................................................................................188
6.3 Calculating Particle Size from the Grain–Size Distribution ..192
6.4 Ranges of wPI and Statistical Information ............................199
6.5 Calculating the Models for Each Range ................................200
ix
CHAPTER Page
6.6 Plotting the Models ................................................................204
6.7 Assessment of the Model for Fine–Grained Materials ..........210
6.8 Implementing the SWCC model based on GSD ....................214
6.9 Summary ................................................................................215
7 SIGNIFICANCE OF THE STUDY.........................................................218
7.1 Conclusions ............................................................................218
7.1.1 Conclusions Chapter 3 – Database Collection ........219
7.1.2 Conclusions Chapter 4 – Validating Models ..........220
7.1.3 Conclusions Chapter 5 – Approach 1 to Predict
the SWCC ...............................................................222
7.1.4 Conclusions Chapter 6 – Approach 2 to Predict
the SWCC ...............................................................227
7.2 Application.............................................................................229
7.3 Recommendations for Further Research ................................229
REFERENCES
x
LIST OF TABLES
Table Page
3-1 Initial Soil Properties Selected for the Master Database ...........................38
3-2 Description of the Soil Properties Initially Selected from the Main
Database .....................................................................................................40
3-3 Summary of Final Database Statistics .......................................................45
3-4 Process Developed in Preparing the Database ...........................................69
4-1 Comparative Analysis of Errors for SWCCs .............................................81
4-2 Error Analysis for Non–Plastic Soils .........................................................94
4-3 Error Analysis for Plastic Soils ..................................................................94
5-1 Database Available per Type of Soil .......................................................114
5-2 Descriptive Statistic Analysis on the Entire Database .............................116
5-3 Descriptive Statistic Analysis on the Entire Database .............................117
5-4 Summary of Descriptive Statistic Analysis on Fine–Grained Soils ........118
5-5 Summary of Descriptive Statistic Analysis on Granular Soils ................118
5-6 Correlation Matrix for Fine–Grained Soils ..............................................120
5-7 Correlation Matrix for Granular Soils......................................................123
xi
Table Page
5-8 Proposed Models for the SWCC Parameters for Fine–Grained Soils
Soils..........................................................................................................131
5-9 Summary of Trials Finding the Best Model (Parameter a f ) ....................135
5-10 Summary of Trials Finding the Best Model (Parameter b f ) ....................138
5-11 Summary of Trials Finding the Best Model (Parameter c f ) ....................142
5-12 Summary of Trials Finding the Best Model (Parameter h r ) ....................146
5-13 Proposed Models for the SWCC Parameters for Granular Soils .............155
5-14 Summary of Trials Use in Finding the Best Model for Parameter a f for
Granular Soils ..........................................................................................157
5-15 Summary of Trials Used in Finding the Best Model for Parameter b f ....161
5-16 Summary of Trials Used in Finding the Best Model for Parameter c f ....165
5-17 Soil Index Properties for Three Soils Taken from the Database .............180
6-1 Soil Properties for the Meegernot Soil.....................................................188
6-2 Calculating Suctions from the Degree of Saturation ...............................192
6-3 Calculated Particle Diameter from the Percent Passing Values ..............195
6-4 Relationship between Suction Values versus Particle Size .....................197

xii
Table Page
6-5 Database Divided by wPI Ranges and Statistics Associated with Each
Range of Values .......................................................................................200
6-6 Summary of Fitting Parameters Found for the Correlation .....................203
7-1 Soil Index Properties for Three Soils Taken from the Database .............225
xiii
LIST OF FIGURES
Figure Page
2-1 Soil–Water Characteristic Curve in Terms of Degree of Saturation .........17
3-1 Available Soil Survey Data ........................................................................36
3-2 Schematic Representation of Map Unit, Component and Soil Unit ..........37
3-3 Spreadsheet with the SWCC Parameter Calculations ................................48
3-4 Spreadsheet with the GSD Parameter Calculations ...................................55
3-5 Finding Particle Size Values ......................................................................60
4-1 Family of SWCC’s for Plastic Soils ..........................................................75
4-2 Relationship between Saturated Volumetric Water Content and wPI for
Plastic Soils ................................................................................................77
4-3 Family of SWCC’s for Non–Plastic Soils .................................................79
4-4 Combined Family of SWCC for Both Plastic and Non–Plastic Soils .......80
4-5 Measured vs. Predicted Volumetric Water Content based on Zapata’s
Model - Plastic Soils (Suction 1 kPa) ........................................................82
Model - Plastic Soils (Suction 10 kPa) ......................................................82
xiv
Figure Page
Model - Plastic Soils (Suction 100 kPa) ....................................................83
Model - Plastic Soils (Suction 1,000 kPa) .................................................83
Model - Plastic Soils (Suction 10,000 kPa) ...............................................84
Model for Non–Plastic Soils (Suction 1 kPa) ............................................85
Model for Non–Plastic Soils (Suction 10 kPa) ..........................................85
Model for Non–Plastic Soils (Suction 100 kPa) ........................................86
Model for Non–Plastic Soils (Suction 1,000 kPa) .....................................86
Model for Non–Plastic Soils (Suction 10,000 kPa) ...................................87
xv
Figure Page
4-15 Measured vs. Predicted Volumetric Water Content based on the MEPDG
Model for Plastic Soils (Suction 1 kPa) .....................................................95
Model for Plastic Soils (Suction 10 kPa) ...................................................95
Model for Plastic Soils (Suction 100 kPa) .................................................96
Model for Plastic Soils (Suction 1,000 kPa) ..............................................96
Model for Plastic Soils (Suction 10,000 kPa) ............................................97
Model for Non–Plastic Soils (Suction 1 kPa) ............................................98
Model for Non–Plastic Soils (Suction 10 kPa) ..........................................98
Model for Non–Plastic Soils (Suction 100 kPa) ........................................99
xvi
Figure Page
Model for Non–Plastic Soils (Suction 1,000 kPa) .....................................99
Model for Non–Plastic Soils (Suction 10,000 kPa) .................................100
4-25 Measured versus Predicted Volumetric Water Content Using
Zapata Model for Plastic Soils .................................................................103
4-26 Measured versus Volumetric Water Content Using
Zapata Model for Non-Plastic Soils.........................................................104
4-27 Measured versus Predicted Volumetric Water Content Using the MEPDG
Model for Plastic Soils .............................................................................105
4-28 Measured versus Predicted Volumetric Water Content Using the MEPDG
Model for Non-Plastic..............................................................................106
5-1 Graphical Representation of the Database Available for Each Type of
Soil ..........................................................................................................114
5-2 Changes in the SWCC Shape Due to Changes in the a f Parameter .........126
5-3 Changes in the SWCC Shape Due to Changes in the b f Parameter.........127
xvii
Figure Page
5-4 Changes in the SWCC Shape Due to Changes in the c f Parameter .........128
5-5 Changes in the SWCC Shape Due to Changes in the h r Parameter.........129
5-6 Model Predicting Parameter a f for Fine-Grained Soils ...........................136
5-7 Spreadsheet Used in Calculating Errors and R2 Values for the a f Model
for Fine–Grained Soils .............................................................................136
5-8 Measured Versus Predicted Analysis for SWCC Parameter a f ...............137
5-9 Model Predicting Parameter b f for Fine–Grained Soils ..........................140
5-10 Spreadsheet Used in Calculating Errors and R2 Values for the b f Model
5-11 Measured Versus Predicted Analysis for SWCC Parameter b f ...............141
5-12 Model Predicting Parameter c f for Fine–Grained Soils ...........................144
5-13 Spreadsheet Used in Calculating Errors and R2 Values for the c f Model
5-14 Measured Versus Predicted Analysis for SWCC Parameter c f ...............145
5-15 Model Predicting h r Parameter for Fine-Grained Soils ...........................148
xviii
Figure Page
5-16 Spreadsheet Used in Calculating Errors and R2 Values for the h r Model
5-17 Measured versus Predicted Analysis for SWCC Parameter h r ................150
5-18 Curve Gravimetric Water Content vs Void Ratio ....................................151
5-19 Histogram Showing Differences between Plastic Limit and Saturated
Volumetric Water Content .......................................................................153
5-20 Model Predicting Parameter a f for Granular Soils ..................................159
5-21 Spreadsheet Used in Calculating Errors and R2 Values for the a f Model
for Granular Soils .....................................................................................160
5-22 Measured versus Predicted Analysis for SWCC Parameter a f ................160
5-23 Model Predicting Parameter b f for Granular Soils ..................................162
5-24 Spreadsheet Used in Calculating Errors and R2 Values for the b f Model
5-25 Measured versus Predicted Analysis for SWCC Parameter b f ................164
5-26 Model Predicting Parameter c f for Granular Soils ..................................166
xix
Figure Page
5-27 Spreadsheet Used in Calculating Errors and R2 Values for the c f Model
5-28 Measured Versus Predicted Analysis for SWCC Parameter c f ...............167
5-29 Histogram of Parameter h r .......................................................................168
5-30 Measured vs Predicted Degree of Saturation for Fine–Grained Soils .....169
5-31 Spreadsheet Used in Calculating the Error Analysis for Measured versus
Predicted Degree of Saturation for Fine-Grained Soils ...........................170
5-32 Measured versus Predicted Degree of Saturation for Fine–Grained
Soils..........................................................................................................171
5-34 Calculating D 10 from the GSD curve.......................................................173
5-35 Approach to Estimate the SWCC Based on Statistical Correlation of
Fredlund & Xing Parameters with Soil Index Properties ........................175
5-36 Examples of SWCCs Using the Model Proposed ....................................181
5-37 Effect of a Low Initial a f Parameter Value in the SWCC........................182
6-1 Soil–Water Characteristic Curve .............................................................190
6-2 Spreadsheet for Calculating the SWCC Parameters ................................191
xx
Figure Page
6-3 Spreadsheet Used to Find the GSD Fitting Parameters ...........................194
6-4 Log Suction versus Log Particle Size for One Soil .................................198
6-5 Spreadsheet from Statistica® Used in Estimating the Best Models between
Particle Size and Suction Values .............................................................202
6-6 Plot of Log Suction versus Log Particle Size ..........................................205
6-7 Relationship between the Constant Values and the wPI..........................206
6-8 Suction as a Function of Particle Diameter and wPI ...............................208
6-9 Error Analysis ..........................................................................................209
6-10 Comparison of Measured versus Predicted Suction ................................209
6-11 Family of Curves for Granular Materials.................................................210
6-12 Family of Curves for Fine–Grained Materials .........................................211
6-13 Example Relationship Log Suction vs Log Particle Size ........................212
6-14 Grain–Size Distribution Example ............................................................213
6-15 Soil–Water Characteristic Curve Example ..............................................213
6-16 Approach to Estimate the SWCC Based on the GSD ..............................215
7-1 Examples of SWCC Using the Model Proposed .....................................226
xxi
CHAPTER 1
INTRODUCTION
1.1 Overview
The Soil-Water Characteristic Curve (SWCC) is the relationship between
matric suction and water content. This property is vital when solving engineering
problems or designing in unsaturated soils. For example, this function allows for
the determination of the hydraulic conductivity at different degrees of water
content or saturation, which is very important when estimating fluid flow
underneath covered areas such as foundations and pavement systems.
Most of the infrastructure is founded in unsaturated soils. Even though
constitutive relationships that utilize the concepts of unsaturated soils have been
proposed for the classic areas of interest to geotechnical engineers, the application
or implementation into engineering practice has been rather slow. One of the
reasons for the delay in the application of unsaturated soil mechanics in practice is
with no doubt the time required for the determination of the SWCC in the
laboratory, and also the specialized equipment and training needed.
In 2008, AASHTO approved the Mechanistic Empirical Pavement Design
Guide (MEPDG). This new pavement design guide incorporates the effects of
environmental conditions such as precipitation and temperature in the
determination of changes of unbound material properties during the life of the
pavement structure. This model makes use of unsaturated soil principles which in
1
turn requires the input of the SWCC. To aid in the implementation of the
MEPDG, an alternative way to determine the SWCC via laboratory testing, is a
method that estimates or derives the SWCC based on well-known soil index
properties.
Several attempts have been made to estimate the SWCC based on grain-
size distribution (GSD) and well-known index properties such as Plasticity Index.
Also, several approaches have been used to solve the problem including three
major approaches (Zapata, 1999):
1) Statistical estimation of water contents/degree of saturation at
selected matric suction values.
2) Correlation of soil properties with the fitting parameters of the
SWCC function by means of nonlinear regression analysis.
3) Estimation of the SWCC using a physics-based conceptual model.
A comprehensive comparison of the different approaches and models can
be found in Zapata, 1999; where yet another model was proposed based on the
second approach. This approach was also taken by Perera, 2003 and further
refined by Witczak et al., 2006.
In this study, the models published by Zapata, 1999 and Witczak et al.,
2006 (MEPDG model) were validated with a large database of matric suction and
other index soil properties collected as part of the National Cooperative Highway
Research Program (NCHRP) 9-23A project (Zapata, 2010). Furthermore, two

2
different procedures to estimate the SWCC are proposed. The first procedure is
based on the correlation of soil properties with the fitting parameters of the
SWCC analytical function proposed by Fredlund and Xing, 1994. The second
procedure is based on the estimation of the SWCC based on a physics-based
conceptual model which relates the grain-size distribution of the soil and index
properties with the pore-size distribution. The first procedure will greatly aid in
the implementation of the new MEPDG pavement design guide, while the second
procedure presents an alternative approach that is both conceptually sound and
easy to implement by engineering practitioners. Both procedures are based on the
database collected during the NCHRP 9-23A project, which consisted of soil
properties, including matric suction measurements, for more than 36,000 soils.
1.2 Historical Background
The NCHRP 9-23A project entitled "A National Catalog of Subgrade Soil-
Water Characteristic Curve (SWCC) Default Inputs and Selected Soil Properties
for Use with the ME-PDG" was carried out at Arizona State University in 2010
(Zapata, 2010). The objective of this project was the creation of a national
database of pedologic soil families that reflected the input soil properties for
subgrade materials needed in the implementation of the approved AASHTO ME-
PDG (Darter et al., 2006). The database focuses upon the Soil-Water
Characteristic Curve (SWCC) parameters, which are key parameters in the
implementation of Level 1 environmental analysis as well as measured soil index
properties needed in all hierarchical levels of the climatic/environmental engine of
3
the guide, the “Enhanced Integrated Climatic Model (EICM)”. These parameters
are primarily used to estimate the equilibrium moisture content in unbound
materials which directly affect the pavement performance due to changes in the
resilient modulus of the soil.
The NCHRP 9-23A project allowed for the creation of a database of more
than 31,000 soils throughout the continental US, Puerto Rico, Hawaii and Alaska,
which included soil index properties and moisture retention measurements for at
least two or three levels of suction. This extensive database, perhaps the largest
available, allowed for the determination of the fitting parameters to define the
SWCC Function. Other properties available included the grain-size distribution,
hydrometer analysis, liquid limit, plastic limit, and saturated hydraulic
conductivity.
As part of this work, the database collected allowed for the validation of
two models available for the determination of the fitting parameters for the
SWCC function. The first model, developed by Zapata in 1999 consisted in a
useful family of curves of SWCCs for both granular soils and fine-grained soils
based on GSD parameters such as the percent passing No. 200 sieve (P200), the
diameter corresponding to 60% passing (D60) and the PI. The equations
developed in this study were initially adopted in the NCHRP 1-37A project
entitled Design Guide for New and Rehabilitated Pavement Design and later
replaced by a set of equations initially developed by Perera, 2003 and then refined
as part of the NCHRP 1-40D project entitled Models Incorporated into the
4
Current Enhanced Integrated Climatic Model in 2006 (Witczak et al., 2006). The
refined equations gave rise to the second model that was validated as part of this
work.
This research work proposes two different procedures to estimate the
SWCC. The first procedure makes use of a statistical analysis to estimate the
SWCC fitting parameters needed in the Fredlund and Xing equation (Fredlund
and Xing, 1994), which could be easily incorporated into the EICM. The second
procedure makes use of a physics-based conceptual model and uses the entire
GSD to estimate the SWCC function, by relating the suction values with the
particle diameter.
1.3 Research Objectives
The main objectives of this research are to:
• Validate the SWCC prediction models previously proposed by
Zapata, 1999 (Zapata Model) and by Witczak et al., 2006
(MEPDG Model) by using the database collected as part of the
NCHRP 9-23A project from the National Resources Conservation
Service.
• Propose a new set of Soil–Water Characteristic Curve parameters
for the Fredlund and Xing equation based on correlations with soil
index properties.
5
• Propose a new approach to estimate the Soil–Water Characteristic
Curve based on a physics-based conceptual model whereas the
entire grain–size distribution and soil index properties are related
to the pore-size distribution of the soil.
1.4 Methodology
The methodology used to achieve the objectives of this research has been
divided into the following three main stages:
1. Creation of the database. This involves the acquisition of the data, the
conversion of the database into formats that makes it easier to manipulate, the
selection of variables of interest to this research work, the recognition and
elimination of inconsistent data, the statistical analysis of variables, and the
analysis of variability of the data. Under this task, the suction measurements
obtained will be used in the generation of the SWCC fitting parameters needed for
the possible correlations with several soil properties such as gradation and
consistency limits.
2. Validation of two existing models using the collected database. This task
allows for the evaluation of the models previously developed by Zapata,1999; and
the models developed by Perera, 2003, and enhanced as part of the project
NCHRP 1-40D in 2006 (MEPDG model). The validation study will serve as
benchmark to the effort pursue as part of this thesis work.
6
3. Generation of new models to predict the SWCC. As explained above,
two different approaches were followed. The first procedure consists in finding
relationships between the SWCC fitting parameters for the Fredlund and Xing
function and well-known index properties. This procedure is of interest because
the fitting parameters are needed as input values in the new pavement design
guide (MEPDG). The second procedure makes use of the entire grain-size
distribution curve and relates the particle diameter to their corresponding suction
value. Once again, some index properties such as passing #200 and Plasticity
index are used to calibrate the model found.
1.5 Organization of the Thesis
This research work is organized according to the outlined methodology as
follows:
Chapter 1 presents the Introduction of the thesis. This introduction
includes an overview of the importance of the work to be pursued, a brief
historical background, the objectives of the thesis and an outline of the
methodology followed in the development of the project.
Chapter 2 contains the Literature Review where the main geotechnical
concepts about unsaturated soil mechanics and the soil–water characteristic curve
are defined. Previous work on the prediction of the SWCC is summarized.
Chapter 3, Database Collection, presents the acquisition of the database.
This chapter includes the process followed to acquire the data, the conversion of
7
the database into formats easier to manipulate, the selection of appropriate
variables needed for this research work, the removal of data that presented
inconsistencies, the statistical analysis of variables, and the analysis of variability
of the data to determine the restrictions into the final results.
Chapter 4 presents the effort done as part of the Validation of two existing
SWCC predicting models by using the database collected and presented in
Chapter 3.
A new set of models to estimate the Soil-Water Characteristic Curve
fitting parameters of the model proposed by Fredlund and Xing (1994) is
presented in Chapter 5. These new set of equations are based on the soils index
properties and SWCC parameters obtained directly from testing results.
Chapter 6 proposes a different approach to estimate the SWCC, which is
based on the similarity between the SWCC and the GSD curves.
Finally, Chapter 7 includes the conclusions per chapter and
recommendations for future studies.
8
CHAPTER 2
LITERATURE REVIEW
2.1 Introduction
Considering the main objective of this work the prediction of the Soil–
Water Characteristic curve, it is convenient to review fundamental principles
related to the Unsaturated Soil Mechanics.
Four main concepts are reviewed in detail: the stress state variables and
the moisture flow in unsaturated soils; the matric suction and the soil–water
characteristic curve. Finally, the approaches that have been attempted to predict
the soil–water characteristic curve are presented.
In unsaturated soils, four phases in equilibrium are defining the system:
the soil particle, the contractile skin, and the phases of air and water. In each
phase, the measurable stresses (σ, u a and u w ) at equilibrium are formulated in
equilibrium under the context of continuum mechanics. Fredlund & Morgenstern,
1977, defined the stress state in an unsaturated soil by using two independent
stress tensors. The formulations are presented.
The second topic considered is the flow generated on the air and water
phases under the applied pressure gradients. That gives rise to the matric suction,
which has been considered the driving potential responsible of fluid flow,
Fredlund & Rahardjo, 1993. This concept is briefly analyzed due to the
9
importance in unsaturated soil mechanics. The clear knowledge about the causes
of the driving potential on the air and water phases is fundamental in
understanding the concepts of air and water flow in an unsaturated soil.
The concept of Matric Suction is explained under section 2-4. This
concept is fundamental to understand the Soil–Water Characteristic Curve
(SWCC) and its importance in unsaturated soil mechanics. The last reviewed
topic relates to the different approaches to obtain the SWCC Function. The
different approaches presented are based on the concepts outlined by Zapata, 1999
and Fredlund et al., 2003.
2.2 Stress State Variables
Terzaghi, 1943, introduced terms to understand the unsaturated soil
behavior. His works was focused on saturated soil for which he defined the
concept of “effective stress variable” as the most important variable or “effective”
variable to define the state of stress in such soil. The effective stress is defined as:
Effective Stress: σ ' = σ − u w .................................................................(2-1)
Where:
σ’ = Effective stress
σ = Total stress
u a = Pore water pressure
10
After Terzaghi, several researchers attempted to express the stress state of
unsaturated soils. In the 1950’s, Bishop, introduced the pore air pressure as an
independent and measurable variable in order to define the effective stress in
unsaturated soils (Bishop, 1959). Bishop proposed the following expression to
estimate the effective stress:
σ’ = (σ – u a ) + χ( u a – u w ) ....................................................................(2-2)
Where:
u a = Pore air pressure
χ = Parameter related to the degree of saturation
In the 1960’s, most of the research was focused in trying to define the
stresses driving the behavior of unsaturated soils or trying to vary Bishop’s
equation. In this decade, many experiments ere performed and theoretical
equations were presented by Donald, Blight, Aitchinson, Bishop, Coleman,
Jennings, Burland, Richards, Matyas and others (Fredlund, 1979). Most of the
models were based on measurable parameters such as total stress, pore–water
pressure and pore–air pressure.
In the 1970’s, Fredlund and Morgenstern, 1977, presented a new
theoretical stress analysis for unsaturated soils based on two independent stress
state variables: (σ – u a ) and ( u a – u w ), and considering the soil as a multiphase
element. Assuming soil particles incompressible and chemically inert, the analysis
11
inferred that any two of three possible normal stress variables could describe the
stress state in an unsaturated soil:
(σ – u a ) and (u a – u w ).............................................................................(2-3)
(σ – u w ) and (u a – u w ) ............................................................................(2-4)
(σ – u a ) and (σ – u w ) ..............................................................................(2-5)
Based on the stress equilibrium condition for an unsaturated soil, Fredlund
presented an equation of forces in equilibrium considering the phases: air, water,
and contractile skin. With these phases in equilibrium, he was able to establish
three stress state variables: u a (which can be eliminated assuming soil particles
and water are incompressible), (σ – u a ), and ( u a – u w ).
The stress state for an unsaturated soil can be expressed with the following
stress tensors:
(σ x − ua ) τ yw τ zx 

 τ xy (σ y − ua ) τ zy

 .........................................................(2-5)
 τ xz τ yz (σ z − ua )
and
(ua − u w ) 0 0 
 0 (ua − uw ) 0  .........................................................(2-5)
 
 0 0 (ua − uw )
Where:
12
τ ij = Shear stress in the i–plane and the j–direction
σ x – u a = Net normal stress in x–direction
σ y – u a = Net normal stress in y–direction
σ y – u a = Net normal stress in z–direction
2.3 Moisture Flow in Unsaturated Soils
The moisture flow can be analyzed in terms of energy or “head” when
water–air flows from a point of high energy to a point of low energy. This energy
gradient is known as “hydraulic head gradient”. The behavior of the moisture
flow is described under the principles of Bernoulli and Darcy. These principles
apply equally for both saturated and unsaturated soil.
Bernoulli’s law consider the total energy or head as the sum of three
heads: velocity head, pressure head and the position head. In geotechnical
practice, the velocity head is very low when comparing with pressure head and
position head (Fredlund & Rahardjo, 1993). The pressure head (p/γ w ) or total
suction is made of two components: matric suction and osmotic suction.
Therefore, the pressure head and position head, combined, define the hydraulic
head gradient in an unsaturated or saturated soil. Equation 2-8, expresses the
hydraulic head gradient, h, at any point in the soil mass.
h = (p/γ w ) + H………………………………………..………………. (2-8)
13
Where:
p = Total suction (matric suction + osmotic suction)
H = Position head (elevation)
On the other hand, Darcy’s law considers the flow of water–air
proportional to the hydraulic gradient. This law can be written as:
v = -k (dh/dx) …………………………...………………..…………. (2.9)
Where:
v = Velocity of water flow through an unsaturated soil
k = Coefficient of permeability or hydraulic conductivity
This hydraulic conductivity as a function of several factors: fluid
viscosity, pore–size distribution, grain–size distribution, voids ratio, roughness of
mineral particles and the degree of soil saturation. In unsaturated soil, the
hydraulic conductivity varies depending on the stress state of the soil (Fredlund,
2006) and particularly, on the matric suction which greatly affects the amount of
water into and out of the soil. It is important to recognize that the permeability of
the soil can be represented by two functions depending whether the process is
drying or wetting. Therefore, hysteresis in the soil–water characteristic curve
drives hysteresis in the permeability function and hence, a close connection
between the Soil–Water Characteristic Curve and the hydraulic conductivity
14
function (expressed as hydraulic conductivity versus soil suction) should be
expected.
One important concept related to the water in unsaturated soil was
developed by Lorenzo A. Richards in 1931 (Richards, 1931). The equation
described by Richards can be written as follows:
……………………………………………. (2.10)
Where:
K(θ) = Hydraulic conductivity as a function of volumetric water content
ψ = Pressure head
z = Elevation above a vertical datum
θ = Volumetric water content
t = Time
Richard’s concept, which describes the flow of water in unsaturated soils,
is based on the laws of hydrodynamics where the movement of water is due to
gravity and the pressure gradient acting on a soil element.
15
2.4 Matric Suction and the Soil–Water Characteristic Curve
Considering one of the three possible combinations of stress state
variables given by equations (2-3), (2-4), and (2-5), Fredlund (2006) defined the
net normal stress (σ – u a ) and the matric suction (u a – u w ) as the most applicable
combination in engineering practice. Net normal stress is the stress state caused
by external loads removed/applied to the soil, and matric suction is the stress state
caused by environmental conditions due to eliminate variations at the ground
surface or by the groundwater table fluctuations.
Matric suction is defined in the standard ASTM D 5298 – 03
Measurement of Soil Potential–Suction Using Filter Paper as:
“The negative pressure relative to ambient atmospheric
pressure on the soil water, to which a solution identical in
composition with the soil water must be subjected in order to
be in equilibrium through a porous permeable wall with the
soil–water”.
In other words, matric suction is the measure of negative pore–water
pressure due to changes in the relative humidity (water vapor pressure caused by
the difference in air and water pressures across the water surface).
Values of matric suction in the field (vadose zone) can range from high
pressures (1,000,000 kPa, Fredlund 2006) under dry conditions (zero water degree
of saturation on the ground surface, in some cases) to zero at the ground water
16
table. Soils close to the surface are frequently affected by environmental
conditions causing a negative effect on the soil. These soils are called expansive
or swelling, collapsible, and residual soils.
A plot between the matric suction and the volumetric water content or
degree of saturation is called Soil–Water Characteristic Curve (SWCC) (see
Figure 2-1). In engineering purposes, the degree of saturation (percentage of voids
filled with water) is commonly used.
100
90 Soil-Water Characteristic Curve
80
Degree of Saturation (%)
70
SWCC
60
Parameters:
50 af = 9.6230
40 bf = 1.1678
Points: cf = 1.6400
30
S = 20%, ψ = 33.3 kPa ψr = 500
20 S = 4%, ψ = 1,500 kPa
10
0
0.0001 0.01 1 100 10000 1000000
Matric Suction y, (kPa)
Figure 2-1. Soil–Water Characteristic Curve in Terms of Degree of
Saturation
2.5 Approaches to Obtain Unsaturated Soil Properties
In order to obtain the SWCC, several levels of sophistication have been
identified. For example, Fredlund et al., 2003, defined four hierarchical levels.
The following sub–sections will brief explain each level of analysis.
17
2.5.1 Level 1.
Level 1 implies direct measurement of suction values and then the use of
one of the universal models to fit the data to the whole range of suction. This level
properly applies for large projects with high risk and high consequences due to
failures. This level is usually followed by researchers that use equipment and
techniques with advance level of investigation in unsaturated soil mechanics.
Universal models are basically a series of empirical equations developed
by several researchers in order to best–fit the suction/water content values with a
soil–water characteristic model. The best fit parameters of these models can be
obtained by using a regression analysis that minimizes the least squared errors.
These models can be categorized as two–fitting–parameter and three–fitting–
parameter models. These parameters are related to the soil–water characteristic
curve in this manner: the first parameter depends on the air entry value of the soil;
the second parameter depends on the rate of water extraction of the soil after
exceeding the air entry value, and the third parameter, when it is used, is basically
a function of the residual water content at high values of suction.
Models with two parameters include those proposed by Garner (1958),
Brooks & Corey (1964), Brutsaert (1967), Laliberte (1969), Farrel & Larson
(1972), Campbell (1974), and McKee & Bumb (1987) (in Fredlund, 2006).
18
Some of the most commonly used models with three parameters were
proposed by Van Genuchten, 1980, Fredlund and Xing, 1994, Leong and
Rahardjo, 1997, and Sillers, 1997.
Van Genuchten, 1980, proposed the next equation with three fitting
parameters:
……………………..……………………… (2.11)
Where:
θ w = Volumetric water content
θ r = Residual volumetric water content
a = Soil parameter which is a function of the air entry value of the
soil
b = Soil parameter which is a function of the rate of water
extraction of the soil after exceeding the air entry value
c = Soil parameter which is a function of the residual water content
of the soil
Fredlund & Xing, 1994, proposed the following equation with three fitting
parameters:
19
………………………………… (2.12)
…………………………………… (2.13)
Where:
θ w = Volumetric water content
a = Soil parameter which is a function of the air entry value of the
soil (kPa)
n = Soil parameter which is a function of the rate of water
extraction of the soil after exceeding the air entry value
m = Soil parameter which is a function of the residual water
content of the soil
h r = Soil parameter which is basically function of the suction at which
residual water content occurs (kPa)
In summary, the implementation of unsaturated soil mechanics under level
1 requires testing to find directly the unsaturated soil property functions. Being
these tests highly expensive, this level should be considered mainly for projects of
great importance.
20
2.5.2 Level 2.
To find the SWCC under Level 2, there is no need for direct
measurements. In this case, conceptual model to infer the SWCC (unsaturated soil
property) from direct measurements of grain size distribution can be used. This
approach was postulated by Fredlund et al., 1997, where a model was presented to
estimate the SWCC from directly measured soil properties such as gradation, dry
density, void ratio, and specific gravity. In this case, Fredlund et al. used a
computational program (SoilVision®) to obtain the SWCC based on the least–
square errors curve fitting algorithm. Their model requires a conceptual model as
well as statistical computations of the SWCC parameters.
This approach has limitations. First, it requires a specific shape curve
(sigmoid). It also requires a minimum number of particle sizes, which have a
strong influence in the equation, and also requires three soil properties: Specific
gravity, void ratio, and dry density. The prediction of the SWCC following this
approach seems to be reasonable for non–plastic soils (Zapata, 1999).
In summary, at Level 2 analysis, unsaturated soil property functions can
be inferred from other function measured in the laboratory, such as the grain–size
distribution. This method has shown to be reliable for non–plastic soils. However,
it makes use of a physics–based conceptual model which makes it a good
candidate for reliable predictions provided a large database is used in the analysis.
21
2.5.3 Level 3.
At Level 3 analysis, a basic soil index property is correlated to estimate
the SWCC. This level is frequently used for the preliminary studies of most
projects. Statistical analyses are used at this level based on databases of previous
test results.
There are two ways to estimate the soil–water characteristic curve at this
level of analysis. First, the SWCC can be estimated from a database, by relating
the SWCC with the gradation, with the classification, or with other index soil
properties. Secondly, the SWCC can be estimated by relating a particular soil with
a similar soil for which a SWCC exists or has been measured. This level implies a
good criterion of estimation, and therefore the estimation at this level is less
accurate. It is appropriate for small projects or projects with low risk of failure.
2.5.4 Level 4.
At this level, correlations are used to estimate the SWCC. This level has
the lowest level of complexity, and could be applied to projects with low risk.
This level implies the use of the soil classification to estimate the SWCC, and
then to use this function to empirically estimate other unsaturated soil property
functions.
It is obvious that Level 1 has the highest level of accuracy while Level 4
involves correlations and hence, it has a great level of variability.
22
In the progress of unsaturated soil mechanics techniques, the soil–water
characteristic curve is a specialized test which involves laboratory equipment
which it is quite complex to operate. This situation has created the need to
estimate procedures, approaches, or use correlations to characterize unsaturated
soils. This methodology is fundamental in the implementation of unsaturated soil
mechanics into geotechnical engineering practice.
Zapata, 1999, presented a comprehensive review of approaches to predict
the soil–water characteristic curve. In her work, the predictions were organized in
three categories. The first category is based upon statistical estimation of water
contents at selected matric suction values. The second category encompasses
those models that, by regression analysis, correlate soil properties with the fitting
parameters of analytical equations that represent the SWCC. The models in the
third category estimate the SWCC using a physical conceptual model approach.
One particular approach followed under the last category converts the Grain–Size
Distribution into the Pore–Size Distribution which in turn can be developed into
the SWCC by means of a packing parameter to relate or calibrate the relationship
between the GSD and the SWCC.
2.6 Approaches to Predict the Soil–Water Characteristic Curve
Zapata, 1999, recognizes different approaches for the prediction of the
SWCC at Level 3 analysis:
23
Approach 1A
In this approach, the water content at a particular suction values is
estimated by using statistical correlation with grain–size distribution parameters
and volume mass soil properties.
Researchers such as Van Genuchten, 1980, Mckee & Bumb, 1987, Van
Genuchten & Mualem, 1980, Gardner, 1958, Williams et al., cited by Zapata,
2010, have predicted volumetric water content from equations calculated by
regressions which have fitting parameters that are function of soil properties such
as percentage of clay content, organic matter content, dry density, etc. These
models were found with a very limited database.
Approach 1B
Besides the grain–size distribution and volume mass soil properties,
several researchers have used one or more suction values to statistically estimate
the water content in the SWCC. They have found that by adding one or two
measurements of suction greatly improve the precision of the models. However,
this concept requires determining one point in the SWCC from one specific value
of matric suction, being this a limitation of this approach. As with the approach
1A, limited databases were used to develop these models.
24
Approach 1C
Similarly to the approaches mentioned above, this approach is base on
statistical estimation considering the grain size distribution and volume mass soil
properties. Particularly, this approach use models based on a small database from
a particular location. This approach has to be carefully considered understanding
clearly the assumptions, soil tested, the precision, and the soil properties
employed in this thesis.
Approach 2
In this approach, correlations are based on regressions analysis. The water
content can be computed by statistical correlations of soil properties with the
fitting parameters of the SWCC.
The uncertainty associated with predicting models depends on the
database used in the computations and the tests applied to validate the model. The
models proposed and based on Level 2 analysis have proofed reliable for granular
soils which operate in a low matric suction range of values. On the other hand,
predictions of the SWCC for fine–grained soils can be considered unreliable.
Some of features affecting the prediction of the SWCC on fine–grained soils
include the shape of soil particle, organics coating, the entrapped air, and some
adsorptive forces acting on the surface of soil particles. Nevertheless, this method
is desirable if a large database is available for proper calibration.
25
2.7 Final Remarks
The literature review shown in this chapter allows us visualize that the
estimation of the Soil–Water Characteristic Curve at any level of difficulty has
been approached by several researches and most of them use very limited
database and therefore, it is concluded that there is room for improvement of the
estimation.
26
CHAPTER 3
DATABASE COLLECTION
3.1 Source of the Database
The United States Department of Agriculture (USDA) has an organization
dedicated to conserving all natural resources particularly on private land. The
private land is more than 70% of the land in this country, and for this reason, it is
an objective of the Federal Government to ensure that the natural resources on
these lands are protected and conserved. This federal organization is called the
Natural Resources Conservation Service (NRCS) and has the objective of
collecting, storing, maintaining, and distributing the soil survey information for
private land owner in the United States (Soil Survey Staff, 1993).
Using the latest in science and technology, NRCS has been working with a
multidisciplinary group of professionals on projects and research to get productive
land in a healthy environment. These projects and research are developed in the
field directly, and are complemented with analysis from tests developed in the
laboratory (Manual, Soil Survey Division Staff, 1993).
One of the fundamental goals of the NRCS is soil conservation.
Understanding that soil is a foundation for agricultural and sustainable
development, it must be conserved between the highest standards of quality. In
this regard, the USDA is working consistently to have soils well characterized in
all private and public areas of the United States and its Trust Territories. The
27
characterization involves investigation, inventory, classification, mapping, and
interpretation of the quality of soils. This soil survey information is one of the
most important tools for the planning and management of the majority of urban
and rural projects where soil is involved. While the database was obviously
intended for agricultural purposes; the USDA entered into a joint agreement with
the then Bureau of Public Roads (BPR) to also measure key soil properties useful
into the field of highway/pavement engineering. The engineering properties from
this database will be used as the primary source of information in order to pursue
the main objectives of this research work (Soil Survey Staff, 1975 & 1994).
3.2 Characteristics of the Database
The information obtained from the NRCS is divided into three main
databases, which depend primarily on the scale used for mapping the different soil
units.
1. The Soil Survey Geographic (SSURGO) database, at the farm to
rural area scale ranging from 1:12,000 to 1:63,360
2. The State Soil Geographic (STATSGO) database, at the regional
scale of 1:250,000, and
3. The National Soil Geographic (NATSGO) database, at the national
scale of 1:5,000,000
28
The components of map units in each database are different and
correspond to different levels of detail. SSURGO database, for example, provides
the most detailed level of information. Its information is focused on local levels,
where the data is used for specific planning and management of farms, ranches,
and rural areas. STATSGO was designed for regional levels, river basins, states,
or regional governments with the purpose of planning, management and
monitoring natural resources, lands and aquifers. Its information cannot be used
for interpretation or planning at the county level. NATSGO has a lower level of
detail and is basically used for national and regional resource appraisal, planning,
and monitoring. Its information (maps and databases) was processed from states’
general soil maps, and its attributes were generated from generalization of
detailed maps.
It is important to note that each database contains the same amount if
detail in tabular form. However, the maps (spatial data) are a function of the area
scale. For example, a map unit at the SSURGO level represents a single soil
component; while at the NATSGO level, a map unit can contain up to 21
components. Each component represents a soil profile with information of up to
ten layers.
The source of information used in this study is based on the State Soil
Geographic (STATSGO) database. Due to the level of detail and scale, the
STATSGO database is specifically used by the NRCS in agricultural matters to
evaluate regional soils and water quality, soil erodibility, soil wind erosion, crop
29
growth, soil productivity, hydrology and ecology; and in general, to generate
environmental quality models. Although STATSGO was primarily oriented for
agricultural purposes, its information, has been used cautiously in studies of other
fields of science such as biology, chemical engineering, geology, geotechnical
engineering, environmental science, etc.
3.3 Data Collection
The US general soil map and data needed for this analysis was
downloaded from the website: http://soildatamart.nrcs.usda.gov/ (reference). It
contains two types of data: Spatial and Tabular files. The spatial files have
information necessary to process the graphical expressions of the different soil
units. They provide shapefiles, which allow the user to analyze spatial
information, edit data and create maps in a GIS® (Geographic Information
System) based format. The tabular files provide engineering and agricultural soil
properties in Microsoft® Access format. This format allows the user to handle an
immense volume of data. The tables are organized by group of attributes
according to the technological field; and therefore, it is possible to classify and
query the database. For the purpose of this study only the tabular information was
downloaded and processed. However, there are sufficient capabilities for mapping
the soil information, should further research requires visual representation of the
soil properties obtained.
30
The US general soil map downloaded from the USDA website was
obtained from a generalized analysis of detailed soil survey maps. In areas where
soil survey maps were not available, information about geology, topography,
vegetation, and climate, together with images was obtained from the Land
Remote Sensing Satellite (LANDSAT) that allowed for the definition of the most
probable attributes and areas for the STATSGO dataset (general soil map). Most
of the soil areas are defined cartographically by using the satellite images together
with the soil survey map. With regards to the characteristics, properties or
attributes of a particular soil unit area; these were obtained as estimates of
properties from sampling areas based on statistical extrapolation from areas where
its properties were well-known.
The database developed by the USDA-NRCS is being constantly updated
by organizing the layers according to new studies, validating soil properties from
new information received and including new soil properties according to
correlations provided by new research and studies. An example of the correlations
used by NRCS is mentioned in Feng et al., 2009 for the Saturated Hydraulic
Conductivity, Ksat. Ksat is expressed in mm/h and it is estimated by the equation
(3-1),
Ksat = 1930(SAT − θ 33 )
(3 − λ )
..................................................................(3-1)
Where:
SAT = Saturated Moisture at 0 kPa (%v)

31
θ 33 = Moisture at 33 kPa (%v)
1
λ= .....................................................................................................(3-2)
A
A=
[ln(1500) − ln(33)] ...........................................................................(3-3)
[ln(θ33 ) − ln(θ1500 )]
This equation and other correlations used in the database are also
presented by Saxton and Rawls, 2006.
The accuracy or errors encountered in the database are extremely
important for the objectives pursued in this research work. For example, a study
performed for the Western states (Feng et al., 2009) indicated a Root Mean
Squared Error (RMSE) for sand and clay content between 4% and 7%. This
validation was accomplished through comparisons with tests and studies which
were carried out directly in the field by the USDA-NRCS Soil Survey Laboratory.
In GIS format, general soil map units are linked to attributes in order to
indicate the location of each soil map unit and its soil properties. Although most
of the continental US areas are defined, some areas do not have available
information. The area where digital soil data is available is shown in Figure 3-1.
The tabular data contained in the database represent a mean range of properties
for the soil comprised in each soil map unit. The representative value is used in
this study to define the soil property for each type of soil. The tabular data contain
soil information that serves as an attribute of the soil map unit in GIS format.
32
The database main downloaded from the NCRS website contained
information for 1,227,117 soils throughout the continental US, Alaska, Hawaii
and Puerto Rico, with more than 150 geotechnical, chemical, and physical
properties for all the layers and soil unit maps considered at the SSURGO level.
The information is grouped in "map units". The “map units” are areas that
represent a group of soil profiles with generally the same or similar
characteristics. These map units contain information organized according to the
schematic diagram shown in Figure 3-2. Each map unit is identified with a code
called Mukey. Each Mukey or map unit is made of several "components", which
are soil profiles with slightly different soil properties. The percentage of area,
within the map unit, covered by each component is available. For the purpose of
this project, it was assumed that the component with the largest percentage of
coverage was representative of the entire map unit. Each profile is typically
comprised of 3 to 5 layers or "soil units", with some profiles containing
information of up to 11 layers. The depth covered by the typical profile averages
about 60 inches, with some profiles approaching 100 inches.
3.3.1 Master File Properties and Characteristics.
Soil properties that are known to impact the moisture retention properties of the
soil were extracted from the NRCS main database. From the main database, 52
soil properties were extracted in order to pursue the objectives of this research
work. These geotechnical properties that constitute the Master database are
summarized in Tables 3-1 and 3-2.

33
3.3.2 Preliminary Reduction of Soil Unit Data.
Each soil type found in the database had information from several boring logs. In
most of the cases, the information was similar or very similar and therefore, it
allowed for the initial reduction of the Master database. This process was
carefully performed by choosing the boring log with the most complete
information. In some cases, the information collected from two boring logs was
complimentary and therefore, the information was combined to produce a
complete description of the soil.
3.3.3 Selecting the Proper Component to Represent the Map Unit.
As previously noted, the Master database consisted of information for different
components within each map unit. It was necessary to further reduce the database
to reflect only one set of soil properties per map unit. For the purpose of this
project, it was assumed that the component with the largest percentage of
coverage was representative of the entire map unit. After this criterion was
applied, the number of soils was reduced to 36,462 items.
3.3.4 Properties Included in the SWCC Predicting Analysis.
Table 3-3 presents the final selection of soil properties included in the SWCC
predicting analysis and summarizes the percentage of data available, for each soil
engineering variable considered in the final database. The soil properties needed
to estimate the SWCC parameters include the volumetric water content at 10, 33,
and 1,500 kPa; and the saturated volumetric water content (i.e., satiated water
34
content or porosity). In addition, parameters such as grain-size distribution
values, consistency limits, saturated hydraulic conductivity, groundwater table
depth and bedrock information were included.
35
Figure 3-1. Available Soil Survey Data
(http://soildatamart.nrcs.usda.gov, June 2009)

36
Figure 3-2. Schematic Representation of Map Unit, Component and Soil Unit
37
Table 3-1. Initial Soil Properties Selected for the Master Database
Column Label Column Name

Component Name compname
AASHTO Classification aashtocl
AASHTO Group Index aashind_r
Unified Classification unifiedcl
Top Depth - Representative Value hzdept_r
Bottom Depth - Representative Value hzdepb_r
Thickness - Representative Value hzthk_r
#4 - Representative Value sieveno4_r
Total Clay - Representative Value claytotal_r
LL - Representative Value ll_r
PI - Representative Value pi_r
Db 0.1 bar H2O - Representative Value dbtenthbar_r
Db 0.33 bar H2O - Representative Value dbthirdbar_r
Db 15 bar H2O - Representative Value dbfifteenbar_r
Ksat - Representative Value ksat_r
0.1 bar H2O - Representative Value wtenthbar_r
0.33 bar H2O - Representative Value wthirdbar_r
38
Table 3-1. Initial Soil Properties Selected for the Master Database (Cont'd)
Column Label Column Name

15 bar H2O - Representative Value wfifteenbar_r
Satiated H2O - Representative Value wsatiated_r
LEP - Representative Value lep_r
CaCO3 - Representative Value caco3_r
Gypsum - Representative Value gypsum_r
CEC-7 - Representative Value cec7_r
Water Table Depth - Annual - Minimum wtdepannmin
Water Table Depth - April - June - Minimum wtdepaprjunmin
Bedrock Depth - Minimum brockdepmin
Corrosion Concrete corcon
Corrosion Steel corsteel
EC - Representative Value ec_r
Available Water Storage 0-150 cm aws0150wta
SAR - Representative Value sar_r
pH H2O - Representative Value ph1to1h2o_r
Kw kwfact
Kf kffact
AWC - Representative Value awc_r
Db oven dry - Representative Value dbovendry_r
Comp % - Representative Value comppct_r
Hydrologic Group hydgrp
MAAT - Representative Value airtempa_r
Elevation - Representative Value elev_r
ENG - Local Roads and Streets englrsdcd
Mapunit Key mukey
Component Key cokey
Chorizon Key chkey
Chorizon AASHTO Key chaashtokey
Chorizon Unified Key chunifiedkey
39
Table 3-2. Description of the Soil Properties Initially Selected from the Main
Database
Column Label Description

Mapunit Symbol The symbol used to uniquely identify the soil
mapunit in the soil survey.
Map Unit Name Correlated name of the mapunit (recommended name
or field name for surveys in progress).
Component Name Name assigned to a component based on its range of
properties.
AASHTO A rating based on a system that classifies soils
Classification according to those properties that affect roadway
construction and maintenance. Soils are classified
into seven basic groups plus eight subgroups, for a
total of fifteen for mineral soils. Another class for
organic soils is used. The groups are based on
determinations of particle-size distribution, liquid
limit, and plasticity index. The group classification,
including group index, is useful in determining the
relative quality of the soil material for use in
earthwork structures, particularly embankments,
subgrades, subbases, and bases. (AASHTO)
AASHTO Group Index The empirical group index formula devised for
- Representative Value approximately within-group evaluation of the "clayey
granular materials" and the "silty-clay materials".
Unified Unified Soil Classification System - A system for
classifying mineral and organo-mineral soils for
engineering purposes based on particle size
characteristics, liquid limit, and plasticity index.
Top Depth - The distance from the top of the soil to the upper
Representative Value boundary of the soil horizon.
Bottom Depth - The distance from the top of the soil to the base of
Representative Value the soil horizon.
Thickness - A measurement from the top to bottom of a soil
Representative Value horizon throughout its areal extent.
#4 - Representative Soil fraction passing a number 4 sieve (4.70mm
Value square opening) as a weight percentage of the less
than 3 inch (76.4mm) fraction.
40
Table 3-2. Description of the Soil Properties Initially Selected from the main
Database (Cont’d)

Total Clay - Mineral particles less than 0.002mm in equivalent
Representative Value diameter as a weight percentage of the less than
2.0mm fraction.
LL - Representative The water content of the soil at the change between
Value the liquid and plastic states.
PI - Representative The numerical difference between the liquid limit and
Value plastic limit.
Db 0.1 bar H2O - The oven dried weight of the less than 2 mm soil
Representative Value material per unit volume of soil at a water tension of
1/10 bar.
Db 0.33 bar H2O - The oven dry weight of the less than 2 mm soil
1/3 bar.
Db 15 bar H2O - The oven dry weight of the less than 2 mm soil
15 bars.
Dp Mass per unit of volume (not including pore space) of
the solid soil particle either mineral or organic. Also
known as specific gravity.
Ksat - Representative The amount of water that would move vertically
Value through a unit area of saturated soil in unit time under
unit hydraulic gradient.
0.1 bar H2O - The volumetric content of soil water retained at a
Representative Value tension of 1/10 bar (10 kPa), expressed as a
percentage of the whole soil.
0.33 bar H2O - The volumetric content of soil water retained at a
Representative Value tension of 1/3 bar (33 kPa), expressed as a percentage
of the whole soil.
15 bar H2O - The volumetric content of soil water retained at a
Representative Value tension of 15 bars (1500 kPa), expressed as a
percentage of the whole soil.
41
Database (Cont’d)

Satiated H2O - The estimated volumetric soil water content at or
Representative Value near zero bar tension, expressed as a percentage of
the whole soil.
LEP - Representative The linear expression of the volume difference of
Value natural soil fabric at 1/3 or 1/10 bar water content and
oven dryness. The volume change is reported as
percent change for the whole soil.
CaCO3 - The quantity of Carbonate (CO3) in the soil
Representative Value expressed as CaCO3 and as a weight percentage of
the less than 2 mm size fraction.
Gypsum - The percent by weight of hydrated calcium sulfate in
Representative Value the less than 20 mm fraction of soil.
CEC-7 - Representative The amount of readily exchangeable cations that can
Value be electrically adsorbed to negative charges in the
soil, soil constituent, or other material, at pH 7.0, as
estimated by the ammonium acetate method.
Water Table Depth - The shallowest depth to a wet soil layer (water table)
Annual - Minimum at any time during the year expressed as centimeters
from the soil surface, for components whose
composition in the map unit is equal to or exceeds
15%.
Water Table Depth - The shallowest depth to a wet soil layer (water table)
April - June - during the months of April through June expressed in
Minimum centimeters from the soil surface for components
whose composition in the map unit is equal to or
exceeds 15%.
Bedrock Depth - The distance from the soil surface to the top of a
Minimum bedrock layer, expressed as a shallowest depth of
components whose composition in the map unit is
equal to or exceeds 15%.
Corrosion Concrete Susceptibility of concrete to corrosion when in
contact with the soil.
Corrosion Steel Susceptibility of uncoated steel to corrosion when in
contact with the soil.
EC - Representative The electrical conductivity of an extract from
Value saturated soil paste.
42
Database (Cont’d)

Available Water Available water storage (AWS). The volume of water
Storage 0-150 cm that the soil, to a depth of 150 centimeters, can store
that is available to plants. It is reported as the
weighted average of all components in the map unit,
and is expressed as centimeters of water. AWS is
calculated from AWC (available water capacity)
which is commonly estimated as the difference
between the water contents at 1/10 or 1/3 bar (field
capacity) and 15 bars (permanent wilting point)
tension, and adjusted for salinity and fragments.
pH H2O - The negative logarithm to the base 10, of the
Representative Value hydrogen ion activity in the soil using the 1:1 soil-
water ratio method. A numerical expression of the
relative acidity or alkalinity of a soil sample. (SSM)
Kw An erodibility factor which quantifies the
susceptibility of soil particles to detachment and
movement by water. This factor is adjusted for the
effect of rock fragments.
Kf An erodibility factor which quantifies the
susceptibility of soil particles to detachment by
water.
AWC - Representative The amount of water that an increment of soil depth,
Value inclusive of fragments, can store that is available to
plants. AWC is expressed as a volume fraction, and is
commonly estimated as the difference between the
water contents at 1/10 or 1/3 bar (field capacity) and
15 bars (permanent wilting point) tension and
adjusted for salinity, and fragments.
Db oven dry - The oven dry weight of the less than 2 mm soil
Representative Value material per unit volume of soil exclusive of the
desiccation cracks, measured on a coated clod.
Comp % - The percentage of the component of the mapunit.
Representative Value
Hydrologic Group A group of soils having similar runoff potential under
similar storm and cover conditions.
43
Database (Cont’d)

MAAT - The arithmetic average of the daily maximum and
Representative Value minimum temperatures for a calendar year taken over
the standard "normal" period, 1961 to 1990.
Elevation - The vertical distance from mean sea level to a point
Representative Value on the earth's surface.
ENG - Local Roads The rating of the map unit as a site for local roads
and Streets and streets, expressed as the dominant rating class for
the map unit, based on composition percentage of
each map unit component.
Chorizon Key A non-connotative string of characters used to
uniquely identify a record in the Horizon table.
44
Soil-Property Unit n % Data Max Min Average Median Mode StDev
Top Depth of Layer cm 36,462 100 241 0 28 8 0 39
Bottom Depth of Layer cm 36,462 100 254 2 67 41 152 58
Thickness of the Layer cm 36,462 100 218 2 39 28 20 32
Passing Sieve # 4 % 36,462 100 100 19 85 93 100 18
Passing Sieve # 10 % 36,462 100 100 13 80 88 100 21
Passing Sieve # 40 % 36,458 100 100 6 69 73 95 22
Passing Sieve # 200 % 36,455 100 100 0 51 50 43 25
Passing Sieve 0.002 mm % 36,461 100 90 0 21 20 15 13
Saturated Hydraulic Conductivity µm/s 36,460 100 423 0 21 9 9 30
45
Volumetric Water Content at Suction 10 kPa % 4,357 12 37 0 15 15 14 5
Saturated Volumetric Water Content % 36,462 100 70 10 37 38 41 10
Liquid Limit % 32,494 89 125 0 32 30 25 12
Plasticity Index % 36,400 100 66 0 10 8 3 10
Elevation m 31,708 87 3,963 -5 973 823 305 753
Bedrock Depth - Minimum cm 10,218 28 202 0 54 41 77 36
Table 3-3. Summary of Final Database Statistics
Water Table Depth - Annual - Minimum cm 10,065 28 168 0 51 46 0 42

Water Table Depth - April - June - Minimum cm 9,363 26 168 0 51 46 0 42
3.4 Working with the Database
In addition to the initial set of properties extracted from the database, further
reduction of data was necessary in order to find the SWCC fitting parameters and
the GSD fitting parameters. The computed parameters were also incorporated
into the Final database generated from this study:
3.4.1 Calculating Soil-Water Characteristic Soil Parameters.
The Soil-Water Characteristic Curve (SWCC) is defined as the
relationship between soil water content or degree of saturation and soil matric
suction (Fredlund, 2006). Several researchers have proposed universal models to
define the SWCC, as previously discussed in Chapter 2. However, the model
implemented in the MEPDG is the model given by Fredlund & Xing, 1994 and
shown in the equations 2-12 and 2-13 in Chapter 2; and therefore, it was desirable
to find the fitting parameters of this model due to the practical application in
pavement design and analysis.
This model represents a sigmoid with four fitting parameters, named a f ,
b f , c f and h rf in the MEPDG. The best set of these fitting parameters can be found
by fitting the measured volumetric water content retained at tensions of 1/10 bars
(10 kPa), 1/3 bars (33 kPa) and 15 bars (1,500 kPa) which are obtained from the
primary database (explained in section 3-2 of this chapter).
The first step in order to find the best four SWCC fitting parameter was to
define the available points. Two points were obtained from the complete database:
46
the volumetric water content at 1/3 bar (33 kPa) and the volumetric water content
at 15 bars (1,500 kPa) of suction. With these data and the saturated volumetric
water content it was possible to calculate the Degree of Saturations at the same
suctions. Degree of Saturation is the ratio between the volumetric water content
and the saturated volumetric water content for a specific suction. The third point
is at zero suction when soil has 100% of saturation (Drying process was used for
these tests). The fourth point is assumed at 1,000,000 kPa when the lowest values
of saturation are reached, this assumption was defined by Dr. Fredlund (Fredlund
& Xing, 1994, and Sillers & Fredlund, 2001).
The second step was to calculate the best fitting parameters by a non-
linear least squared regression. This regression was developed by using the Tool
Solver of Excel® from Microsoft (Microsoft Corp., Redmond, WA). Figure 3-3
shows the spreadsheet used for this calculation.
47
Dry Sat vol
Suction Suction Density Vol. w/c w/c Sat
(%
(Bar) (kPa) (gm/cc) Note (%) (%) )
0.1 10 N/A 39.0
0.33 33.33 N/A 25.8 66.2
15 1500 N/A 14.2 36.4
SWCC Parameters
initia
Parameter final l
Objective
af 1.1972 10 Function
1.05642E-
bf 1.4156 1 09
cf 0.4969 2
hr 500.0 500
Vol. Water Constraint

xe Content: ye yp s
Suction
(psi) % Sat
0.0001 39.0 100.0 100.0 0.000
4.8309 25.8 66.2 66.2 0.000
217.3913 14.2 36.4 36.4 0.000
Figure 3-3. Spreadsheet with the SWCC Parameter Calculations
48
For this calculation, it is important to consider initial values for the Solver. These
initial values should be assumed to be very similar to the final parameters in order
to reach the lowest value of the objective function; which in this case is the
difference between the squared of degree of saturations measured and predicted.
In this project it was assumed:
For plastic soils: For non-plastic soils:
a f = 10 a f = 10
bf = 1 bf = 1
cf = 2 cf = 2
h rf = 500 h rf = 100
In order to simplify, especially because a database of more than 31,000
items was employed, these initial parameters were used. Developing more
detailed work would be ideal by using different initial parameters according to the
wPI value; however, that work would require too much time. Finishing this
calculation was necessary to re-process the Solver in many cases, especially with
the non-plastic soil where the calculations showed many errors or ‘not found’
values.
Once the calculation for the SWCC was defined, the third step was to
develop a VBA project which is usually called ‘Macro’ in Excel®. This program
allows the execution of one process several times. The code for this macro for
calculating the SWCC Parameters is as following:
49
Sub Macro1()
' Macro1 Macro
' Macro recorded 9/24/2010 by user
On Error GoTo Desc
Sheets(1).Select
Application.Goto Reference:="R1C1"
Range("A1:AR1").Select
Range(Selection, Selection.End(xlDown)).Select
Set RawData = Selection
Sheets(2).Select
For j = 2 To RawData.Rows.Count
If RawData(j, 27) <> "" Then
If RawData(j, 24) <> "" Or RawData(j, 25) <> "" Or RawData(j,
26) <> "" Then
Range("E3").FormulaR1C1 = RawData(j, 24)
50
Range("G3").FormulaR1C1 = RawData(j, 27)
Range("B8").FormulaR1C1 = 10
Range("D15").FormulaR1C1 = Range("C15").Text
SolverReset
SolverOk SetCell:="$D$9", MaxMinVal:=2, ValueOf:="0",
ByChange:= _
"$B$8:$B$11,$D$15:$D$17"
SolverAdd CellRef:="$E$15:$E$17", Relation:=2,
FormulaText:="0"
SolverAdd CellRef:="$B$8:$B$11", Relation:=3,
FormulaText:="0.0001"
51
SolverOptions MaxTime:=100, Iterations:=10000,
Precision:=0.000001, AssumeLinear _
:=False, StepThru:=False, Estimates:=1, Derivatives:=1,
SearchOption:=1, _
IntTolerance:=5, Scaling:=False, Convergence:=0.0001,
AssumeNonNeg:=True
SolverSolve True, 1
Range("B8").Select
Set Solution_ = Selection
RawData(j, "AO") = Solution_(1)
RawData(j, "AP") = Solution_(2)
RawData(j, "AQ") = Solution_(3)
RawData(j, "AR") = Solution_(4)
cHANGEsCALE
End If
End If
52
Next
Exit Sub
Desc:
MsgBox "Error"
End Sub
Finishing this step was necessary to re-process the data with errors in the
results. This part of the work showed errors for several reasons. Non-plastic soils
usually present problems because the rate of decrease of the degree of saturation
is high with small changes of suction; these cases cause problems when finding
the optimum in Solver. The Soils with high plasticity present problems as well,
because the sigmoidal shape is lost and the approximation to the objective
function takes a long time. As such, many times are shown as errors. It is
important to emphasize the fact that a regression with only two points is large and
not very precise. Many times the regression in the Solver cannot stop at an
appropriate point. In these cases, having good initial parameters is the best way to
reach the optimum. The complete database includes these parameters.
3.4.2 Calculating Grain-Size Distribution Parameters.
The Grain- Size Distribution Curve was represented by Wagner, 1994 as a
sigmoidal shape with a lognormal distribution, presenting a high similarity to the
53
soil-water characteristic curve given by Fredlund & Xing, 1994. The model
equation for the grain-size distribution given by Wagner, 1994 is:
   
7
 D 
  ln1 + r   
   D  
Pp (D ) =
1
gm 1 −    
...........................(3-4)
 g   D
gn
ln exp(1) +  a     ln1 + r
 
 D      Dm   
  
Where:
P p (D) = percent of Passing a particular grain-size, d
ga = fitting parameter associated to the initial break in the GSD,
gn = fitting parameter associated to the maximum slope of GSD,
gm = fitting parameter associated to the curvature of the GSD,
D= particle diameter in mm.
Dr = residual particle diameter in mm.
D m = minimum particle diameter in mm.
Similarly as the SWCC parameter were calculated, the GSD fitting
parameters were calculated finding the best set of GSD parameters. In this case
the points are given in the database and correspond to the Gradation of the soil,
Passing #4, #10, #40, and #200 US sieves. With these points, the regression was
54
developed using Solver of Excel®. A spread sheet similar to the spread sheet used
to calculate the SWCC parameters is showed in Figure 3-4.
Particle Size % Pass Grain-Size Distribution Parameters

# (mm) (%) Final Initial
4 4.750 95.0 ga 0.0001 1
10 2.000 91.5 gn 0.5837 0.5
40 0.425 82.5 gm 71.8474 0.5
200 0.075 57.5
Graph
Objective Function Particle Size (mm) Passing (%)
1.06438
0.0001 0.00
De % Passing % Passing Constraints 0.001 0.18
Diameter (mm) Measured Predicted 0.01 17.31
1,000 100.0 99.8 0.000 0.075 57.68
4.750 95.0 95.2 0.000 0.1 62.76
2.000 91.5 92.2 0.000 0.425 81.80
0.425 82.5 81.8 0.000 1 88.51
0.075 57.5 57.7 0.000 2 92.17
4.75 95.20
10 96.86
100 99.17
1000 99.78
Grain - Size Distribution Curve

100
80
Percent Passing (%)
60
40
20
0
0.001 0.01 0.1 1 10 100 1000
Particle Size, D (mm)
Figure 3-4. Spreadsheet with the GSD Parameter Calculations
55
For this calculation, is important to consider the initial values to use in the Solver
as well. For this point, each soil or item of the database was run three times in the
Solver. In this way, it was possible to define the best set of GSD parameters. The
idea was to change the a g parameter because this parameter defines the break
point of the curve. These three options allow for the choosing of the minimum
value for the objective function based on the least squared error:
Objective Function = (square measured value – squared estimated value)
Once the calculation was correctly defined, the next step was developed using
Macro in Visual Basic of Excel® in order to execute this process n times, n being
the number of items of the database. The code for this macro is:
Sub Macro2()
' Macro2 Macro
' Macro recorded 9/24/2010 by user
'On Error GoTo Desc
Sheets(1).Select
Application.Goto Reference:="R1C1"
Range("A1:AR1").Select
56
Set RawData = Selection
Sheets(2).Select
For j = 2 To RawData.Rows.Count
If RawData(j, 27) <> "" Then
If RawData(j, 24) <> "" Or RawData(j, 25) <> "" Or RawData(j, 26) <> ""
Then
Range("B10").FormulaR1C1 = 0.5
Range("B11").FormulaR1C1 = 0.5
57
SolverReset
SolverOk SetCell:="$D$10", MaxMinVal:=2, ValueOf:="0",
ByChange:= _
"$B$9:$B$11,$D$16:$D$20"
SolverAdd CellRef:="$E$16:$E$20", Relation:=2, FormulaText:="0"
SolverAdd CellRef:="$B$9:$B$11", Relation:=3,
FormulaText:="0.0001"
SolverOptions MaxTime:=100, Iterations:=30000, Precision:=0.000001,
AssumeLinear _
:=False, StepThru:=False, Estimates:=2, Derivatives:=2,
SearchOption:=1, _
IntTolerance:=5, Scaling:=False, Convergence:=0.00001,
AssumeNonNeg:=True
SolverSolve True, 1
Range("B9").Select
58
Set Solution_ = Selection
RawData(j, "AO") = Solution_(1)
RawData(j, "AP") = Solution_(2)
RawData(j, "AQ") = Solution_(3)
RawData(j, "AR") = Range("D10").Text
End If
End If
Next
Exit Sub
Desc:
MsgBox "Error"
End Sub
The data obtained after this process was added to the complete database.
These parameters are indispensable calculating the Particle-Size values.
3.4.3 Calculating Particle Size Values.
Particle Size (called Effective Size as well) is the grain size or grain
diameter of the soil through which a defined percentage of the total material is
59
passing. For example, D 60 is the diameter in the grain-size distribution curve
corresponding to the 60% finer. Figure 3-5 graphically shows the concept of
Particle Size. This particle size is a good geotechnical property to estimate the
hydraulic conductivity; many researches have shown similarities in the models
between Grain-Size Distribution, Soil-Water Characteristic Curve and the
Saturated Hydraulic Conductivity. The database developed for this work is an
excellent platform for finding more correlations between them.

100
90
80
Percent Passing (%)
70
60
50
40
30
20
10 D60
D10 D30 D90
0
0.001 0.01 0.1 1 10 100 1000
Figure 3-5. Finding Particle Size Values
Having the model equation for the grain-size distribution (see equation 3-
4), which was implemented by Fredlund (Fredlund at al., 1997) into SoilVision®
as a model to predict the SWCC, the next step was to calculate through this
equation the Particle Size for 10, 20, 30, 60, and 90% of Passing. This step
60
required calculating into the equation, the Dvalues from a defined percentage of
passing.
This work was developed using the Tool Goal Seek in Microsoft Excel®,
and using a Macro in Microsoft VBA® to repeat the process n times (n is the
number of items considered in the database). After this, which required a long
time to process, the database was considered complete.
Particle Size values are really important in the prediction of the SWCC
parameters for non-plastic soils. Many researches have studied the geotechnical
behavior for granular soils, and the effective size is an excellent measure for this
type of soil. Ayra & Paris, 1981, Gupta & Larson, 1979, Wagner & Ding, 1994,
Fredlund at al., 2000, are some researchers who have worked by using this
concept.
Two factors added using the particle size values are the Coefficient of
Uniformity, C u and the Coefficient of Curvature, C c . These expressions are
defined as:
D60
Cu = ................................................................................................(3-5)
D10
Cc =
( D30 )
2
......................................................................................(3-6)
(D10 )(D60 )
Where:
61
D 10 = grain diameter in mm corresponding to 10% passing, by weight
Coefficient of uniformity, C u is an expression to define the shape of the
grain-size distribution. When a granular soil is well graded, C u is higher than 15.
The sand of a beach poorly graded has a C u between 2 and 3. A granular soil with
a C u = 1, is a soil with particles of the same size. Coefficient of Curvature, C c is
an expression related to the shape of the particle size distribution, values of C c
between 1 and 3 are considered well graded soils. These geotechnical expressions,
which refer to the grain-size distribution are related to the SWCC, and are also
related to the saturated hydraulic conductivity.
3.4.4 Other Calculations.
The group index is an engineering concept developed by AASHTO that
categorizes the probable “service performance” of the soil, particularly when it is
used as a highway pavement subgrade. The group index can be calculated by the
empirical equation given in the standard AASHTO M 145-91 (Standard
Specification for Classification of Soils and Soil-Aggregate Mixtures for
Highway Construction Purposes):
GI=(P 200 -35)[0.2+0.005(LL-40)]+0.01(P 200 -15)(PI-10) .....................(3-7)
Where:
62
P 200 = Passing the No. 200 sieve
LL = Liquid Limit
PI = Plasticity Index
Note that the first term is related to the liquid limit and the latter to the
plasticity index. The final GI value is based on the following qualifications:
• If the GI calculated is negative, it is taken to be zero
• The GI calculated is rounded to the nearest whole number
• There is no upper limit
• The GI for the following soils must be taken as zero: A-1-a, A-1-b,
A-2-4, A-2-5, and A-3
• For soils A-2-6 and A-2-7, the GI must be calculated by the
equation:
GI=0.01(P 200 -15)(PI-10) ...........................................................(3-8)
Salient grain size distribution parameters such as the D 60 , Passing 200 and
Plasticity Index are required to estimate the weighted plasticity index (wPI). This
property is estimated as follows:
P200 × PI
wPI = ...................................................................................(3-10)
100
63
The wPI will depend on the type of soil being considered. For coarse soils
the wPI = 0, and for soils with more than 12% of fines, the wPI > 0.
The relationship between the Group Index (equation 3-7) and the wPI
(equation 3-10) is shown in Figure 3-6. For this analysis were used the entire
database with wPI greater than zero.
p p
70
60
wPI = 0.8486 GI + 2.2237
50
40
wPI
30
20
10
0
0 10 20 30 40 50 60 70
Group Index
Figure 3-6 Relationship between the Group Index and the Weighted
Plasticity Index
In order to have a complete database for this project (or for future
projects) these two properties were calculated: the California Bearing Ratio
(CBR) and the Resilient Modulus (MR).
64
The CBR is an empirical soil property that characterizes the strength of
materials in subgrades and unbound material. This characteristic allows for the
estimation of the resilient modulus by using the expression:
M R ( psi ) = 2,555 × CBR 0.64 .....................................................................(3-9)
This expression is used in the ME-PDG Methodology (Witczak et al,
2001). CBR values can also be estimated based on index soil properties like Grain
Size Distribution and Atterberg’s Limits. USCS and AASHTO classifications are
correlated to estimate typical CBR and MR values. However, one practical way is
to use the grain size distribution.For coarse soils (with wPI = 0), the CBR value is
referred to by the grain diameter at which 60% passes the grain size distribution
(D 60 ), in millimeters. The formula in this case is:
CBR = 28.09(D60 )
0.358
...........................................................................(3-11)
This expression has two limitations: for soils with D 60 less than 0.01 mm,
a CBR = 5 is used and for soils with D 60 greater than 30 mm, a CBR = 95 value is
used. For fine soils (with wPI>0), the expression that is used is:
75
CBR = .........................................................................(3-12)
1 + 0.728(wPI )
It should be realized that all of these conditions are approximations to the
real measured laboratory value for either CBR or Mr. Their use should be
confined to only level three applications of the design guide.
65
3.5 Summary
The database collection was a very important task for the development of
the work presented in this thesis. The vast amount of data points collected
contained a total of 36,394 different soils, with 4,518 items corresponding to non–
plastic materials and 31,876 plastic soils. The database was collected by the
National Conservation Resources Service for agricultural purposes and contains
chemical, physical and engineering soil properties which can be used in a number
of disciplines. The soils properties were obtained from studies developed during
many years through the continental US, Alaska, Hawaii, and Puerto Rico. The
database allowed for the estimation of parameters such as the wPI factor, Group
Index, the Soil–Water Characteristic Curve fitting parameters and the Grain–Size
Distribution fitting parameters.
Most of the properties were obtained directly from the laboratory or from
field testing while other properties were estimated from correlations or
estimations. Both sets of data or properties had some degree of uncertainty related
to them. The uncertainty of the data can be attributed to several factors: First, the
database was developed by collecting tests during a range of years (USDA–NRCS
was established in 1935); second, uncertainty associated with environmental
conditions and soil nature (samples were located all over the US territory); third,
the tests were performed by following protocols and standards which are being
constantly updated; and last, technological changes and advances in the field
66
allowed for new data interpretations during more than 70 years the data has been
collected.
In order to eliminate the variability encountered in the data, a moving
average technique was employed, whereas the data was organized or sorted
according to the geotechnical factor (predictor) that most affected the predicting
variable. This process is commonly used when the database presents high
variability in order to find the general trend of behavior (Graham, 1993).
It is important to emphasize that the vast database collected and presented
as part of this thesis work was drawn directly from laboratory testing. It is
perhaps the largest database of soil moisture retention curves available in the
world. These facts allowed for optimal models to estimate the Soil–Water
Characteristic Curve, as those presented in this work.
The process followed in this chapter had the objective of preparing a
geotechnical database with the biggest quantity of data available for modeling.
This database was used in correlations to create a new set of equations for the
Soil-Water Characteristic Curve Parameter Two approaches were utilized to
formulate these equations.
Table 3-4. Summarizes the process followed to prepare a complete
database for this work and for future work. The Primary database was
downloaded from the USDA–NRCS website which contains all the data used to
interface with maps that are allowed for working in all the US areas for
67
agricultural purposes. The Master database was condensed from the Primary
database and contains exclusively engineering properties; this database was
reduced to basically one with the largest amount of different soils. The Initial
database was extracted from the Primary database and contains only the soil
properties required for this work. This database presents the soil-properties and
the original data from the Primary database. The complete database contains the
initial database, as well as the soil–properties calculated such as the SWCC
parameter, GSD parameters and particle size or D-values, computed at level 1 by
using equations and correlations previously established in other research.
68
Table 3-4. Process Developed in Preparing the Database
Database # Items Obtained Contain Soil-properties

from
Main 1,227,117 USDA-NRCS Entire database Contains more

from US at than 150
SSURGO level chemical,
physical and
geotechnical
properties
Master 36,462 Main Database with 52 soil properties

all the to be used for
Engineering this and future
data projects
Final 36,462 Master Soil-properties 18 soil properties

required for selected from
this work NRCS plus 19
properties
estimated, for a
total of 37 soil
properties
69
CHAPTER 4
VALIDATION OF AVAILABLE SWCC PREDICTION MODELS
4.1 Introduction
This chapter intends to confirm the validity of two available SWCC
models that are based on index properties by using principles of statistics. In this
manner, it is possible to evaluate the bias of the published models towards a rather
limited database used in their development. A statistical analysis of errors will
enable the study to reach this objective. This chapter intends to check two
important projects previously developed at Arizona State University.
The first SWCC model to validate was proposed by Zapata in 1999. In her
dissertation titled: “Uncertainty in Soil-Water Characteristic Curve and Impacts
on Unsaturated Shear Strength Predictions.” She presented two sets of SWCC
fitting parameters (i.e. one set for plastic material and another set for non-plastic
soils) derived from a regression analysis from a set of 190 laboratory tests. The
SWCC model followed in this work corresponded to Fredlund & Xing, 1994.
The second SWCC model was developed in 2006 by an ASU research
team as part of the NCHRP 1-40D project, titled “Models Incorporated into the
Current Enhanced Integrated Climatic Model NCHRP 9-23 Project” which was
developed for the National Cooperative Highway Research Program (Witczak at
al, 2006). The model was a modification of the proposed equations by Perera,
2003 and had the main objective of validating the Enhanced Integrated Climatic
70
Model (EICM) to incorporate unsaturated soil properties and environmental
effects in the overall pavement design procedure. EICM is the module that ASU
research team included in the Mechanistic-Empirical Pavement Design Guide
(MEPDG) version 0.7. As part of this project, a new set of models was presented
for the af, bf, cf and hr SWCC fitting parameters under the primary SWCC model
published by Fredlund & Xing, 1994.
In short, the process developed in this chapter initially consisted in
preparing a complete database with the information required: the database
comprised geotechnical properties, such as wPI, % Grain Size Distribution
parameters (passing a particular sieve), Particles Size diameters (called Dvalues),
and the SWCC parameter predicted by the two models to be validated (Zapata’s
model and NCHRP 1-40D model). The database used in the validation process
was explained in detail in Chapter 3 (Database Collection) and consisted of
36,462 soils (36,462 soils are plastic and 4520 soils are nonplastic). The next step
was to calculate the volumetric water content for different suction values (1, 10,
100, 1,000 and 10,000 kPa) by using the two models. Finally, a statistical analysis
of the errors was conducted and the results presented in tables and plots.
4.2 Validating Zapata’s Model
4.2.1 Model for Plastic Soil.
The 190 data points used by Zapata, 1999, in developing her model were
classified into two types according to Plasticity Index; 70 soils with plasticity
71
index values greater than zero (plastic soils) and 120 soils with plasticity index
values equal to zero (non-plastic soils). Soil properties obtained in the laboratory
work allowed for an estimation of the best fitting SWCC parameters by using
SoilVision® (software created by SoilVision System Ltd). This software
estimated these parameters by fitting the primary SWCC model developed by
Fredlund & Xing, 1994.
For plastic soils (soils with PI > 0), Zapata used the percent of Passing #
200 sieve and the Atterberg Limits, and specifically the Plasticity Index to find
the models for the SWCC fitting parameters. Basically, the weighted plasticity
index or wPI factor was the main geotechnical concept used in her model for this
type of soil. wPI combines both properties the percent of Passing # 200 and the
Plasticity Index, and it is defined as follows:
P 200 × PI
wPI = ... ...............................................................................(4-1)
100
Where:
P200 = Passing # 200 US Standard Sieve, expressed in percentage
PI = Plasticity Index, expressed in percentage
The equations proposed by Zapata are:
0.00364( wPI ) 3.35 + 4( wPI ) + 11

af = ...................................................(4-2)
6.895
72
bf
= −2.313( wPI ) 0.14 + 5 ......................................................................(4-3)
cf
c f = 0.0514( wPI ) 0.465 + 0.5 ...................................................................(4-4)
hr
= 32.44e 0.0186 ( wPI ) ..............................................................................(4-5)
af
The family of curves for plastic soils obtained by Zapata is shown in
Figure 4-1.
Furthermore, a correlation to estimate the saturated volumetric water
content, s, as a function of wPI was proposed:
θ s = 0.0143(wPI )0.75 + 0.36 ...................................................................(4-6)
The validity of this equation was analyzed with the database available for
this project. In order to accomplish this, the equation was statistically evaluated
by calculating the mean algebraic and the mean absolute errors. Additionally, the
adjusted coefficient of determination, R2, and the Se/Sy parameter were computed
in order to assess the accuracy of the equation (Hines & Montgomery, 1990)
The following statistical expressions were used:
⎡ (θ m − θ p )100 ⎤
∑⎢ θm
⎥
Mean algebraic error: ea lg = ⎣ ⎦ ..................................(4-7)
n
73
⎡ (θ m − θ p )100 ⎤
∑⎢ θm
⎥
Mean absolute error: eabs = ⎣ ⎦ ....................................(4-8)
n
∑ (θ −θ p )
2
m
Sum of the squared error: S e = ...................................(4-9)
n− p
∑ (θ )
2
m −θ p
Mean of the squared error: S e = ..............................(4-10)
n− p
Adjusted Coefficient of Determination: R2 = 1 – (Se/Sy)2 ...................(4-11)
Where:
m = Measured volumetric water content
p = Predicted volumetric water content
m = Average measured volumetric water content
n = Number of data points
p = Number of parameters associated with the model
The statistical parameters found for the s equation when using the
database employed in this project are:
Database with n = 36,394
ealg = 98.87
74
1.2
1.0 wPI = % Passing #200* PI
0.8
wPI = 50
0.6 40
30
75
20
10
wPI=0.1 3 5
Degree of Saturation
0.4 15
D60 =1 mm D60 =0.1 mm

0.2
Figure 4-1. Family of SWCC’s for Plastic Soils

0.0
1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E+5 1E+6
Matric Suction (kPa)
eabs = 98.87
Se/Sy = 0.99
R2 = 0.0224
These results suggest a very weak, if at all, correlation. In order to find out
if there is any correlation between wPI and s, the database was regressed. 36,394
points were used in the regression.
The following correlation was found:
θ s = 7.92(wPI )0.27 + 25 ........................................................................(4-12)
The plot obtained for this correlation is showed in Figure 4-2.
In order to eliminate some scatter, a moving average technique was
employed in the analysis. This technique required the wPI values sorted from the
smaller to the largest. Then every 300 points were averaged.
The relationship obtained confirmed the fact that there exists a correlation
between volumetric water content and wPI. It also confirms the Zapata’s
correlation is only valid for the database used in her analysis, which is very
limited.
76
50
Saturated Volumetric Water Content 45
40
Represent average of 300 soils

35
θs = 7.92 (wPI)0.27 + 25
30 n = 36,394
R2 = 0.9978
Se/Sy = 0.05
25
20
0 10 20 30 40 50
wPI parameter
Figure 4-2. Relationship between Saturated Volumetric Water Content and wPI
for Plastic Soils
4.2.2 Model for Non–Plastic Soil.
For non-plastic soils, Zapata used the Diameter D60 as the main soil
property or predictor to correlate with the SWCC parameters. The equations
presented in her dissertation are:
0.8627( D60 ) −0.751

af = .......................................................................(4-13)
6.895
b f = 7.5 ............................................................................................... (4-14)
77
c f = 0.1772 ln( D60 ) + 0.7734 ...............................................................(4-15)
hr 1
= ................................................................................(4-16)
a f D60 + 9.7e − 4
The family of curves for non-plastic soils obtained by using Zapata
correlations is shown in Figure 4-3. The Combined families of curves for plastic
and non-plastic soils were shown in Figure 4-4.
4.2.3 Zapata’s Model Validation Analysis
In order to validate the model, the Fredlund & Xing, 1994, model was
fitted to the matric suction data for each soil in the database. A comparative
analysis was developed between the SWCCs fitted to the measured data and the
SWCCs obtained with the models developed by Zapata, 1999. In order to estimate
the errors, the measured volumetric water content, m, was compared with the
estimated volumetric water content, p, by using Zapata’s model. The analysis
was performed by comparing these values at different suction values. In this
manner, it was possible to assess the behavior of the model in a wide range of
values.
The results of the error were assessed for both plastic and non-plastic
soils. The results are shown in Table 4-1.
78
1.2
Predicted SWCC based on D60
1.0
0.8
0.6
79
Degree of Saturation
D60=0.1 mm
0.4
0.2 D60=1 mm
Figure 4-3. Family of SWCC’s for Non-Plastic Soils

0.0
0.1 1 10 100 1000 10000 100000 1000000
1.0
0.9
0.8
0.7
wPI =
0.6 wPI=
50
0.5 0 .1
0.4
80
0.3 D 60 = D 60 =
Degree of Saturation, S
1 mm 0 .1 mm
0.2
0.1
0.0
0.1 1 10 100 1000 10000 100000 1000000
Figure 4-4. Combined Family of SWCC for both Plastic and Non-plastic Soils
Table 4-1. Comparative Analysis of Errors for SWCCs
Analysis for Non Plastic Soils
Parameter ψ = 1 kpa ψ = 10 kpa ψ = 100 kpa ψ = 1,000 kpa ψ = 10,000 kpa
n 4,518 4,518 4,518 4,518 4,518
ealg 26.88 67.84 62.18 61.00 59.05
eabs 38.22 73.80 68.62 68.70 69.23
Se/Sy 0.93 1.35 0.57 0.46 0.42
2
R 0.1271 -0.8119 0.6800 0.7872 0.8225
Analysis for Plastic Soils
n 30,672 30,672 30,672 30,672 30,672
ealg -8.05 -30.89 -38.59 -24.71 -20.67
eabs 8.08 31.76 40.24 35.04 40.57
Se/Sy 0.42 1.16 1.01 0.63 0.62
2
R 0.8262 -0.3503 -0.0248 0.6002 0.6110
Figures 4-5 through 4-9 show the plots measured versus predicted
volumetric water content for plastic soils. These plots correspond to comparisons
made at suctions of 1 kPa, 10 kPa, 100 kPa, 1,000 kPa, and 10,000 kPa,
respectively.
81
Figure 4-5. Measured vs. Predicted Volumetric Water Content based on Zapata’s
Model - Plastic Soils (Suction 1 kPa)
82
83
Figures 4-10 through 4-14 show the plots of measured versus predicted
water content for non-plastic soils. These plots were developed for suctions of 1
kPa, 10 kPa, 100 kPa, 1,000 kPa, and 10,000 kPa.
84
Model for Non–Plastic Soils (Suction 1 kPa)
85
Model for Non–Plastic Soils (Suction 1,000 kPa)
86
4.3 Validating the MEPDG Model (Witczak et al, 2006)
The model developed under the NCHRP 1-40D project made used of a
database of 217 data points; 154 data points corresponded to non-plastic soils and
63 corresponded to plastic soils. These data were obtained by combining the soil
used by Zapata in 1999 and a database obtained under the NCHRP 9-23 project
titled Environmental Effects in Pavement Mix and Structural Design Systems,
Houston, Mirza, & Zapata (2006).
The development of the SWCC prediction equations presented in the
NCHRP 1-40D project was performed in a similar manner as those developed by
87
Zapata (1999). However, a greater number of soils were included in an attempt to
find correlations.
From the total database, 52 plastic soils were corrected by volume change.
This correction was necessary because the change in volume due to the suction
applied create errors in the SWCC, especially in the residual condition of the
SWCC (high suction levels) where the function is very sensitive to changes in the
density undergone by plastic soils. The pressure plate used to obtain the SWCC
for these soils allows volume change measurements in the determination of the
SWCC. The density is calculated at each point of the test and, therefore, the
volume change correction is possible.
The procedure used to achieve the volume change correction is not
included in this work, but it was clearly explained in both projects: Zapata, 1999
and NCHRP 9-23 project (Witczak et al., 2006). To estimate the corrected
volumetric water content due to changes in density was given by the following
expression:
Gs w
θ w-corr = ....................................................................................(4-17)
1+e
Where:
w-corr = Corrected volumetric water content
Gs = Specific gravity of solids
88
w = Gravimetric water content
e = Void ratio
In addition, the NCHRP 1-40D predictive equations did not consider the
effect of hysteresis. This project assumes that the difference between the wetting
curve and the drying curve would be very insignificant. Furthermore, solute
suction is not considered in this project and therefore, only matric suction was
measured.
The analysis to predict the set of SWCC parameters was made separately
for plastic soils and non-plastic soils, and the final equations presented in that
project are given in the following section.
Parameters for Non-Plastic Soils.
For non-plastic soils, the following equations were proposed by Witczak
et al., 2006 (MEPDG model), to find the af parameter:
a f = 1.14a - 0.5 ...................................................................................(4-18)
a = −2.79 − 14.1log( D20 ) − 1.9 ×10−6 P200

4.34
+ 7 log( D30 ) + 0.055D100 .....(4-19)
Where:
⎡ 40 ⎤
⎢ + log( D60 ) ⎥
⎢⎣ m1 ⎥⎦
D100 = 10 .............................................................................(4-20)
89
30
m1 = ..................................................................(4-21)
[log( D90 ) − log( D60 )]
af = SWCC fitting parameter
D20 = Grain diameter corresponding to 20% of passing by weight, in mm
P200 = Percent passing U.S. standard sieve #200
To find the bf parameter:
b f = 0.936b - 3.8 .................................................................................(4-22)
Where:
bf = SWCC fitting parameter
⎪⎧ ⎡ ⎛ D ⎞⎤ ⎫
1.19 ⎪ 0.1
b = ⎨5.39 − 0.29 ln ⎢ P200 ⎜ 90 ⎟ ⎥ + 3D00.57 + 0.021P200 ⎬ m1 .................(4-23)
⎩⎪ ⎣ ⎝ D10 ⎠ ⎦ ⎭⎪
⎡ −30 ⎤
⎢ + log( D30 ) ⎥
m
D0 = 10 ⎣ 2 ⎦
..............................................................................(4-24)
90
20
m2 = ..................................................................(4-25)
[log( D30 ) − log( D10 )]
To find the cf parameter:
c f = 0.26e0.758c + 1.4 D10 .......................................................................(4-26)
Where:
cf = SWCC fitting parameter
⎛ 1 ⎞
c = log ( m1.15
2 ) - ⎜⎜1 − ⎟⎟ ......................................................................(4-27)
⎝ bf ⎠
The SWCC fitting parameter hrf was defined as a constant:
hrf = 100 ..............................................................................................(4-28)
These equations have the following constraints:
If af < 1, then af = 2.25 P2000.5 + 5
0.3 < bf < 4
91
Parameters for Plastic Soils.
For plastic soils, the following equations were proposed by Witczak et al.,
2006 (MEPDG model):
a f = 32.835 {ln( wPI )} + 32.438 ..........................................................(4-29)
b f = 1.421( wPI )
−0.3185
.........................................................................(4-30)
c f = −0.2154 {ln( wPI )} + 0.7145 ........................................................(4-31)
hrf = 500 .............................................................................................(4-32)
Where:
wPI = weighted Plasticity Index as defined before
The constraints required for these equations are:
If af < 5, then af = 5
If cf < 0.01, then cf = 0.03
For the special case where wPI is less than 2 for plastic soils, a weighted
average is used for the af parameter. For af parameter the following model was
proposed:
af avg = a fn +
wPI
2
( )
a fp − a fn ...........................................................(4-33)
92
Where:
af avg = af average
afn = af value using the model for non-plastic soils
afp = af value using the model for plastic soils
For the parameter bf, cf, and hrf equations 4-30 to 4-32 apply.
4.3.3 MEPDG Model Validation Analysis.
Similar to the analysis used in the validation of Zapata’s model, the
validation of the MEPDG models was performed by a comparative analysis
between the measured volumetric water content and the predicted. The measured
volumetric water content were found by fitting the Fredlund & Xing, 1994, model
to measured data. Several suction values that cover a wide range of suctions were
chosen for the comparison with the predicted volumetric water content. The
statistical equations 4-12 through 4-16, presented previously, were used for the
error analysis.
The results of the error analysis were independently evaluated for non-
plastic and plastic soils. The results are shown in Tables 4-2 and 4-3, respectively.
93
Table 4-2. Error Analysis for Non-Plastic Soils
n 4,487 4,487 4,487 4,487 4,487
ealg 4.38 4.86 -30.54 -22.07 -11.02
eabs 15.92 30.87 62.08 66.17 70.56
Se/Sy 0.63 1.21 1.38 1.24 1.21
2
R 0.5971 -0.4666 -0.9105 -0.5284 -0.4528
Table 4-3. Error Analysis for Plastic Soils
n 30,561 30,561 30,561 30,561 30,561
ealg -5.08 -23.57 -45.27 -47.27 -46.83
eabs 7.62 28.19 58.80 74.54 81.67
Se/Sy 0.41 1.10 1.60 1.56 1.43
2
R 0.8329 -0.2008 -1.5690 -1.4296 -1.0584
Figures 4-15 through 4-19 show the plots of measured versus predicted
volumetric water content for plastic soils. These plots represent comparisons at
suctions of 1 kPa, 10 kPa, 100 kPa, 1,000 kPa, and 10,000 kPa.
94
Figure 4-15. Measured vs. Predicted Volumetric Water Content based on MEPDG
Model for Plastic Soils (Suction 1 kPa)
95
Model for Plastic Soils (Suction 1,000 kPa)
96
Model for Plastic Soils (Suction 10,000 kPa)
Figures 4-20 through 4-24 show the plots Measured versus Predicted for
non-plastic soils. These plots were developed for suctions of 1, 10, 100, 1,000,
and 10,000 kPa. In this way, it is possible to evaluate the range of suctions.
97
98
99
4.4 Summary
Table 4-1 shows the errors found for the validation of Zapata’s models for
plastic and non-plastic soils. The validation was performed at different suction
levels: 1, 10,100, 1,000 and 10,000 kPa. For non-plastic soils, the R2 values
ranged between 68% and 82%. Relatively good predicted water contents were
found for suction values higher than 100 kPa. For plastic soils, the highest R2
(82%) was found at suction values lower than 1 kPa and relatively acceptable R2
(60%) was found for suction values higher than 1,000 kPa.
100
The figures 4-5 through 4-14 show the graphs Measured versus Predicted
of volumetric water content values for granular and fine-grained soils separately.
These figures include all the predicted volumetric water contents obtained at 1,
10, 100, 1,000 and 10,000 kPa of suction. The error analyses for these figures are
summarized in Table 4-1.
Figure 4-25 and 4-26 show the measured versus predicted volumetric
water content values obtained by using the model proposed by Zapata, 1999, for
plastic and non-plastic soils, respectively. These figures include all the predicted
water contents estimated at suctions of 1, 10, 100, 1,000 and 10,000 kPa. For
plastic soils, the model developed by Zapata, 1999, although it presented an
overall R2 of 0.70, it was found to be biased towards overprediction for most of
the data points. For non-plastic soils, the Zapata’s model presents a different
behavior, in which most of the data points were underpredicted and yielded a low
overall R2 value of 0.40.
In general, the models proposed by Zapata, 1999, present acceptable errors
considering that it was developed 10 years ago with few data points, when
compared to the vast database used in this project.
Tables 4-2 and 4-3 show the error analysis performed for the MEPDG
models for non-plastic and plastic soils, respectively. For non-plastic soils, an R2
value of 60%, which was considered to be acceptable, was found only for suctions
101
values lower than 1 kPa. Similarly, for plastic soils, the highest R2 value (83%)
was found for suction values lower than 1 kPa.
The figures 4-15 through 4-24 show the graphs Measured versus Predicted
of volumetric water content values for granular and fine-grained soils separately.
These figures include all the predicted volumetric water contents obtained at 1,
10, 100, 1,000 and 10,000 kPa of suction. The error analyses for these figures are
summarized in Table 4-1.
Figures 4-27 and 4-28 show the measured versus predicted volumetric
water content values obtained by using the MEPDG model for plastic and non-
plastic soils, respectively. These figures include all the predicted water contents
obtained at 1, 10, 100, 1,000 and 10,000 kPa of suction. It was observed that for
plastic soils, the volumetric water content was consistently overestimated and
yielded an R2 of 0.49. However, for non-plastic soils, the MEPDG model
presented an acceptable prediction of volumetric water content with an R2 value
equal to 0.91.
In general the MEPDG models presented acceptable estimations
considering the amount of data analyzed. The MEPDG model can be considered
to be a better model for non-plastic soils, while the model proposed by Zapata,
1999 can be considered to perform better for fine-grained materials.
102
Measured vs Predicted θw for Plastic Soils - Zapata, 1999's Model
70
60
50
40
103
30
20 n = 30,672
Se/Sy = 0.55
Zapata Model for Plastic Soils

R2 = 0.70
Predicted Volumetric Water Content

10
0
0 10 20 30 40 50 60 70
Measured Volumetric Water Content
Figure 4-25 Measured versus Predicted Volumetric Water Content Using

Measured vs Predicted θw for Non-Plastic Soils - Zapata, 1999's Model
45
40
n = 4,518
35 Se/Sy = 0.78
R2 = 0.40
30
25
104
20
15
10
Zapata Model for Non-Plastic Soils

5
0
0 5 10 15 20 25 30 35 40 45
Figure 4-26 Measured versus Volumetric Water Content Using

Measured vs Predicted θw for Plastic Soils - Witczak, 2006's Model
50
40
30
105
20
n = 30,561
10
MEPDG Model for Plastic Soils

Se / Sy = 0.71
R2 = 0.49

0
0 10 20 30 40 50

Measured vs Predicted θw for Non-Plastic Soils - Witczak, 2006's Model
50
40
30
106
20
n = 4,487
Se / Sy = 0.30
10 R2 = 0.91
MEPDG Model for Non-Plastic Soils

0
0 10 20 30 40 50

CHAPTER 5
A NEW SWCC MODEL BASED ON SWCC PARAMETERS
5.1 Overview
The purpose of Chapter 5 is to propose an improved set of models for the
SWCC parameters based on the equation given by Fredlund and Xing in 1994.
This equation was shown in terms of volumetric water content in Chapter 2
(equation 2-12 and 2-13), but for other engineering purposes it can also be
expressed in terms of degree of saturation as follows:
 
 
  ψ   
 ln1 +  
θw   hr  1 
S (%) = = 1−   ............................(5-1)
θs  
 1,000,000     bf
 
cf

 ln1 +    ψ 
  hr   ln e +  a 

 
   f   
   
Where:
S(%) = Degree of Saturation, in Percentage
ψ = Matric Suction in kPa
a f , b f , c f , h r = SWCC Fitting Parameters
θ w = Volumetric Water Content
θ s = Saturated Volumetric Water Content
107
The parameters a f , b f , c f and ψ r were estimated, based on non-linear
regression analysis, from laboratory measured values of suction and volumetric
water content as indicated in Chapter 3 - Database Collection. The effect of the
hysteresis is not considered in this analysis. Hysteresis is important in cases when
the soil has air trapped or when the structure of the soil (pores and connectivity
between pores) permits a different behavior of the soil under drying or wetting
conditions. The data collected in this study was obtained from pressure plates,
which is usually tested under drying conditions. The database used consisted of
36,394 soils obtained from the National Resources Conservation Service (NRCS)
to work on a new model. 31,876 corresponding to plastic soils, 4,518 of non–
plastic soils and 68 soils did not have enough information to be classified or
defined.
The analysis was developed separately for plastic soils (fine grained soils)
with wPI greater than zero and non–plastic soils (granular soils) with wPI equal to
zero. The concept of wPI (previously explained) is a geotechnical expression
where the Plasticity Index and the Gradation are directly involved in the analysis.
The Weighted Plasticity Index usually called wPI is expressed as follows:
P200 × PI
wPI = ....................................................................................(5-2)
100
Where:
P 200 = Material Passing # 200 US Standard Sieve, in Percentage
108
PI = Plasticity Index, in Percentage
Once the database was divided according to wPI, the next step was to select soil
properties most related to the moisture retention characteristic for each group of
soils. This analysis was based on published work compile from several authors.
The properties assessed by Zapata, 1999, Witczak et al. 2006, and some other
studies by Fredlund, served as basis for the preliminary election of the properties.
For plastic soils, the properties considered into the analysis were: Group
Index, the gradation available (percent passing #4, #10, #40, and #200), the total
percent of clay (% of soil finer than 0.002 mm), Liquid limit, Plasticity Index and
wPI. For non–plastic soils, the properties collected were the Group Index, the
gradation (percent passing #4, #10, #40, and #200), the particle sizes (D 10 , D 20 ,
D 30 , D 60 , D 90 ), and the shape parameters C u and C c . For both sets, volumetric
water content values at 0.1, 0.33 and 15 bars of suction were available.
Group Index, GI, is an engineering parameter associated with AASHTO
classification and used extensively for the analysis of pavement subgrades. The
Group Index expression combines two important soil properties: gradation and
consistency. The Group Index is expressed as:
GI = (P200 − 35)[0.2 + 0.005(LL − 40 )] + 0.01(P200 − 15)(PI − 10 ) ...........(5-3)
Where:
P 200 = Percent Material Passing # 200 US Standard Sieve
109
LL = Liquid Limit
PI = Plasticity Index = LL – PL
On the other hand, the Weighted Plasticity Index or wPI, is a geotechnical
property defined by Zapata, 1999 and shown in the Equation 5-2, which presents
a narrow similarity with the Group Index. Both factors, wPI and GI, are functions
of gradation (P 200 ) and consistency limits (LL and PL). Despite of this similarity
and the close correlation between them, both indexes were considered in the
statistical analysis of fine–grained soils. It is important to recognize that the
Group Index is a factor that is easily recognized by the pavement design
community and therefore, it is a good candidate for the application into the
SWCC equation included into the MEPDG. The assessment of the predicted
values versus the measured values was performed based on an “Error Analysis”,
(Zapata & Houston, 2008). The new models were analyzed through the following
statistical concepts:
Absolute Mean Error, e abs . This concept indicates how the predicted
values are dispersed about the best fitting curve.
100 ∑
(y m − yp )
ym
eabs = ...............................................................(5-4)
n
110
Where:
y m = Measured value
y p = Predicted value
n = Number of data points
Algebraic Mean Error, e alg , indicates how well the curve fit is centered on
the data. A low value of e alg indicates a prediction well centered and with a very
little bias. The sign of this factor describes the direction of the bias.
100 ∑
(y m − yp )
ym
ea lg = ...............................................................(5-5)
n
Where:
y m = measured value
y p = predicted value
n = number of data points
Standard Error Divided by the Standard Deviation, S e /S y . This ratio is an
expression describing how spread out the data is,
∑( y m − y p )
2
Se = ....................................................................(5-6)
n− p
111
∑( y m − y m )
2
Sy = ...................................................................(5-7)
n −1
Where:
p = number of parameters associated with the proposed functions
y m = average of measured values
The Adjusted Coefficient of Determination, R2 (adjusted). This coefficient
defines how well the regressed predicted function approaches the measured data
points,
2
S 
R ( Adjusted ) = 1 −  e  ......................................................................(5-8)
2
S 
 y
As an overview, the procedure followed to find the new SWCC model is:
1. The database was classified according to the wPI property. Those soils
with wPI>o were considered plastic soils, while the soils with wPI=0 were
considered non–plastic.
2. For each type of soil, the measured SWCC parameters were treated as
dependent variables and correlated with all the soil properties affecting the SWCC
(independent variables). Arithmetic functions were considered as well as
transformed functions, including squared values, log arithmetic values, natural
log, arithmetic values and powered functions.
112
3. Each SWCC parameter was subjected to a statistical non–linear
regression analysis against all possible combination of parameters. The adjusted
R–square value, the algebraic mean error, the absolute mean error, the standard
errors and the standard deviation were computed for each analysis.
4. Based on statistical, geotechnical and applicability considerations, the
best model was chosen for each SWCC parameter.
5. Each Measured fitting parameter was compared to the Predicted value.
6. The predicted degree of saturation was obtained by fitting the predicted
SWCC parameter to the Fredlund & Xing function.
7. Final model for plastic and non-plastic soil were proposed.
5.2 Database and Descriptive Statistics
From a total of 36,462 data, 36,394 soils or items were available for this
project. Table 5-1 and Figure 5-1show the number of data points available for
each type of soil. Most of the soils were found to be fine grained soils (classified
from A-4 to A-7-6). In total, the database available consisted of 31,876 plastic
soils (wPI > 0) and 4,518 non–plastic soils (wPI = 0).
113
Table 5-1. Database Available for Each Type of Soil
Type of Soil Data Available

A-1-a 445
A-1-b 1335
A-2-4 4256
A-2-5 28
A-2-6 878
A-2-7 283
A-4 12611
A-5 174
A-6 6237
A-7-5 799
A-7-6 4830
31876
14000
12000
10000
Frequency
8000
6000
4000
2000
0
A-1-a
A-1-b
A-2-4
A-2-5
A-2-6
A-2-7
A-7-5
A-7-6
A-4
A-5
A-6
Figure 5-1. Graphical Representation of the Database Available for Each Type of
Soil
114
A descriptive statistical analysis was performed on the available database for this
project. The descriptive statistical analysis allowed for the preliminary assessment
of the central tendency and variability of the database. This analysis was
developed initially for the entire database, and included each soil property
collected. Table 5-2 summarizes the data found for each parameter and includes
the average, maximum, minimum value, as well as the median, mode, and
standard deviation. The same analysis for selected parameters was developed
separately for plastic and non–plastic soils and the results are shown on Tables 5-
4 and 5-5. Only 12% of the items have available data for the volumetric water
content at 10 kPa of suction, this created a problem because only two measure
points were available to estimate the soil–water characteristic curve. However, in
addition to these two points, the extremes of the SWCC function could be
defined. For 100% of saturation or at very low suction the saturated volumetric
water content was available. Also, a suction of 1,000,000 kPa can be assumed at
zero degree of saturation (Fredlund & Xing, 1994). In this way, the regression
analysis considered four points and in some instances, five points. A descriptive
statistical analysis is presented in table 5.3 for selected properties. It can be seen
that the property values cover a wide range of values.
115
Statistical Group Passing Passing Passing Passing Passing Liquid Plasticity wPI
Parameter Index #4 #10 #40 #200 0.002 mm Limit Index
Mean 5.7 84.6 80.0 69.4 51.4 21.4 32.0 10.3 6.8
Median 1.0 92.5 87.5 72.5 50.0 19.5 30.0 7.5 3.7
Mode 0.0 100.0 100.0 95.0 42.5 15.0 25.0 2.5 0.0
Standard Deviation 9.4 18.3 20.8 22.2 24.7 13.2 12.2 9.7 8.3
Sample Variance 88.3 335.5 433.7 493.5 611.0 175.4 149.9 95.0 68.6
116
Kurtosis 5.6 0.5 -0.1 -0.7 -1.0 0.7 1.7 2.0 4.2
Skewness 2.3 -1.2 -1.0 -0.6 0.0 1.0 1.1 1.4 2.0
Range 68.0 81.0 87.5 94.0 100.0 89.7 125.0 66.0 58.6
Minimum 0.0 19.0 12.5 6.0 0.0 0.0 0.0 0.0 0.0
Maximum 68.0 100.0 100.0 100.0 100.0 89.7 125.0 66.0 58.6
Count 36,394 36,462 36,462 36,458 36,455 36,461 32,494 36,400 36,394
% data available 99.8 100.0 100.0 100.0 100.0 100.0 89.1 99.8 99.8
Table 5-2. Descriptive Statistical Analysis on the Entire Database
Soil Property Unit n % Data Max Min Average Median Mode StDev
Top Depth of Layer cm 36,462 100 241 0 28 8 0 39
Bottom Depth of Layer cm 36,462 100 254 2 67 41 152 58
Thickness of the Layer cm 36,462 100 218 2 39 28 20 32
Passing Sieve # 4 % 36,462 100 100 19 85 93 100 18
Passing Sieve # 10 % 36,462 100 100 13 80 88 100 21
Passing Sieve # 40 % 36,458 100 100 6 69 73 95 22
Passing Sieve # 200 % 36,455 100 100 0 51 50 43 25
Passing Sieve 0.002 mm % 36,461 100 90 0 21 20 15 13
Saturated Hydraulic Conductivity µm/s 36,460 100 423 0 21 9 9 30
117
Soil Properties

Saturated Volumetric Water Content % 36,462 100 70 10 37 38 41 10
Liquid Limit % 32,494 89 125 0 32 30 25 12
Plasticity Index % 36,400 100 66 0 10 8 3 10
Elevation m 31,708 87 3,963 -5 973 823 305 753
Bedrock Depth - Minimum cm 10,218 28 202 0 54 41 77 36
Water Table Depth - Annual - Minimum cm 10,065 28 168 0 51 46 0 42
Water Table Depth - April - June - Minimum cm 9,363 26 168 0 51 46 0 42
Table 5-3. Descriptive Statistical Analysis on the Entire Database for Selected
Table 5-4. Summary of Descriptive Statistical Analysis on Fine–Grained Soils
Mean 6.5 85.0 80.6 71.5 55.9 23.6 32.5 11.8 7.7
Median 2.0 92.5 87.5 75.0 55.0 21.5 30.0 9.0 4.5
Mode 0.0 100.0 100.0 95.0 42.5 15.0 25.0 2.5 3.0
Sample Variance 95.9 318.6 410.8 453.7 511.2 159.7 141.2 91.2 70.8
Kurtosis 4.6 0.6 0.0 -0.6 -1.0 0.9 1.8 2.0 3.7
Skewness 2.1 -1.3 -1.1 -0.6 0.0 1.0 1.3 1.4 1.9
Range 68.0 81.0 87.5 94.0 100.0 89.7 125.0 66.0 58.6
Minimum 0.0 19.0 12.5 6.0 0.0 0.0 0.0 0.0 0.0
Maximum 68.0 100.0 100.0 100.0 100.0 89.7 125.0 66.0 58.6
Count 31,876 31,944 31,944 31,940 31,937 31,943 31,854 31,882 31,876
% data available 99.8 100.0 100.0 100.0 100.0 100.0 99.7 99.8 99.8
Table 5-5. Summary of Descriptive Statistical Analysis on Granular Soils
Mean 0.3 81.9 76.0 54.5 19.7 5.8 9.2 0.0 0.0
Median 0.0 92.5 87.5 55.0 17.5 5.5 7.0 0.0 0.0
Mode 0.0 100.0 100.0 60.0 20.0 2.5 7.0 0.0 0.0
Sample Variance 0.2 446.2 578.0 522.3 166.4 10.0 51.4 0.0 0.0
Kurtosis 29.2 -0.2 -0.8 -1.0 3.9 2.9 1.9 4,518.0 -
Skewness 2.6 -1.0 -0.8 0.0 1.6 1.2 0.9 67.2 -
Range 9.0 80.0 87.5 92.5 95.0 32.0 55.0 1.0 0.0
Minimum 0.0 20.0 12.5 7.5 0.0 0.5 0.0 0.0 0.0
Maximum 9.0 100.0 100.0 100.0 95.0 32.5 55.0 1.0 0.0
Count 4,518 4,518 4,518 4,518 4,518 4,518 640 4,518 4,518
% data available 100.0 100.0 100.0 100.0 100.0 100.0 14.2 100.0 100.0
118
5.3 Correlations
5.3.1 Correlations for Fine Grained Soils.
There are several computer programs to statistically analyze the database.
In this work, Microsoft® Excel, Statistica®5.5 and Minitab®15 were employed.
There are advantages and disadvantages associated with each program, but all of
them were used to complement, calculate, find better model predictors and check
the results. These programs were employed to calculate the descriptive statistical
parameters such as the average (arithmetic mean), the median, the mode, the
standard deviation, the variance, the kurtosis, and the skewness. These programs
were also used to define the possible statistical correlations between the SWCC
parameters and the variables or soil properties considered in the analysis, to find
the best predictors and finally to define the best models.
119
Table 5-6. Correlation Matrix for Fine–Grained Soils
Variables af bf cf hr
Group Index 0.061 0.045 -0.031 0.208
Ln(GI+1) 0.098 0.013 -0.041 0.164
LnGI+1^2 0.081 0.032 -0.035 0.196
Ln(GI+e) 0.093 0.020 -0.039 0.178
LnGI+e^2 0.082 0.031 -0.036 0.195
sieveno4 0.024 0.004 -0.016 0.059
LnP4 0.023 0.003 -0.016 0.054
sieve10 0.028 0.005 -0.016 0.065
LnP10 0.028 0.003 -0.016 0.059
sieve40 0.057 0.012 -0.015 0.087
LnP40 0.053 0.008 -0.015 0.076
sieve200 0.088 0.024 -0.010 0.111
LnP200 0.086 0.017 -0.010 0.094
LnP200^2 0.088 0.020 -0.010 0.100
clay.002 0.063 0.023 -0.051 0.201
Lnclay 0.071 -0.006 -0.058 0.158
Lnclay^2 0.071 0.003 -0.058 0.175
ll_r 0.056 0.032 -0.039 0.197
pi_r 0.074 0.030 -0.042 0.202
clayPI 0.047 0.053 -0.034 0.222
wPI 0.072 0.041 -0.034 0.207
wPI^-2 -0.020 0.003 0.003 -0.019
wPI^-1 -0.068 0.008 0.023 -0.065
wPI^-0.5 -0.093 0.005 0.035 -0.105
120
Table 5-6. Correlation Matrix for Fine–Grained Soils (Cont’d)
wPI^0.5 0.092 0.024 -0.040 0.187
wPI^2 0.032 0.059 -0.023 0.203
LogwPI 0.101 0.007 -0.041 0.150
LogwPI^2 0.090 0.027 -0.039 0.193
P200LogwPI 0.094 0.022 -0.034 0.168
LnwPI 0.101 0.007 -0.041 0.150
LnwPI^2 0.090 0.027 -0.039 0.193
P200/PI -0.043 0.028 0.058 -0.084
PI/P200 0.033 0.005 -0.038 0.116
P200^2PI 0.063 0.046 -0.029 0.203
P200PI^2 0.039 0.057 -0.025 0.209
ksat_r -0.082 -0.006 0.008 -0.065
D10 -0.019 -0.003 -0.005 -0.007
D20 -0.018 -0.002 -0.001 -0.013
D30 -0.018 0.000 0.004 -0.018
D60 -0.006 -0.002 0.008 -0.028
D90 -0.023 -0.003 0.012 -0.044
D100 -0.003 0.002 -0.003 0.001
Cu 0.009 -0.005 0.003 -0.021
Cc -0.007 -0.002 0.002 -0.010
af 1.000 0.094 0.011 -0.046
bf 0.094 1.000 0.934 -0.009
cf 0.011 0.934 1.000 -0.006
hr -0.046 -0.009 -0.006 1.000
The values for the correlations obtained in Table 5-6 are considered low.
Low correlation is caused usually by the high variability of the data. Some of the
reasons of this variability were presented in Chapter 3.
121
Based on the results presented in Table 5-6, the parameter a f (related to
the air entry value) presented the highest correlation (although poor) with the wPI
and Group Index. The parameter b f presented the best correlation when related
with the parameter a f , while the parameter c f presented a very good correlation
with b f . Finally, the parameter h r showed some correlation with several variables,
but in all cases yielded poor results.
This type of situation is analyzed by several researchers in different ways.
For this particular work, the SWCC parameters for plastic soils were organized
according to the wPI. The data points were grouped by taking 300 consecutive
data points. A moving average and the median were calculated for each sub–
group. These two statistical properties defined proper values to represent every
sub–group of data. This procedure was used for each variable, except for the
Group Index, in which case the sub–groups were formed by the group index
number.
5.3.2 Correlations for Granular Soils.
In order to find the correlation matrix for granular or non-plastic material,
the same considerations applied to the analysis done for fine–grained soil were
use; however, the variables considered in the analysis were different. For these
soils, the grain-size distribution parameter such as particle sizes, percent passing,
the coefficient of uniformity, Cu, and the coefficient of curvature, Cc, were
considered. Table 5-7 shows the correlation matrix for non-plastic material.
122
The parameter a f shows the best correlation when related with the percent
passing #200 (P 200 ). Parameter b f shows a significant correlation with parameters
a f and c f . Parameter c f presents some correlation with P 200 , Particle Size D 10 and
D 90 , and showed to be inversely proportional to b f parameter.
In summary, the results from the correlation matrix for granular material
showed that the SWCC fitting parameters are not independent for each other.
Table 5-7. Correlation Matrix for Granular Soils
Group Index 0.135 0.081 0.032 -0.021
Ln(GI+1) 0.170 0.097 0.041 -0.027
LnGI+1^2 0.124 0.075 0.029 -0.019
Ln(GI+e) 0.163 0.094 0.039 -0.026
LnGI+e^2 0.148 0.087 0.035 -0.023
sieveno4 -0.230 -0.056 -0.170 0.021
Log P4 -0.224 -0.068 -0.153 0.016
LnP4 -0.224 -0.068 -0.153 0.016
sieve10 -0.227 -0.057 -0.181 0.021
Log P10 -0.223 -0.074 -0.162 0.016
LnP10 -0.223 -0.074 -0.162 0.016
sieve40 -0.219 -0.086 -0.166 0.022
Log P40 -0.231 -0.103 -0.152 0.018
LnP40 -0.231 -0.103 -0.152 0.018
123
Table 5-7. Matrix of Correlation for Granular Soils (Cont’d)
sieve200 -0.199 -0.249 0.047 -0.014
P200^ 0.5 0.318 0.281 0.035 -0.010
P200^2 -0.087 -0.188 0.088 -0.025
P200^3 -0.019 -0.141 0.102 -0.030
P200^4 0.016 -0.110 0.104 -0.032
Log P200 -0.307 -0.288 -0.016 0.003
Log P200^2 -0.277 -0.279 0.005 -0.004
LnP200 -0.297 -0.285 -0.008 0.000
LnP200^2 -0.271 -0.277 0.009 -0.005
ll_r -0.144 -0.059 -0.092 0.005
pi_r 0.039 0.038 -0.006 -0.004
ksat_r 0.177 0.209 -0.040 -0.018
D10 0.207 0.221 0.023 -0.011
Log D10 0.232 0.229 -0.002 -0.034
Ln D10 0.232 0.229 -0.002 -0.034
D20 0.202 0.177 0.048 -0.011
Log D20 0.192 0.229 -0.059 -0.014
Ln D20 0.192 0.229 -0.059 -0.014
D30 0.192 0.132 0.054 -0.008
Log D30 0.166 0.179 -0.025 -0.003
Ln D30 0.166 0.179 -0.025 -0.003
D60 0.181 0.060 0.103 -0.005
Log D60 0.186 0.093 0.108 0.004
Ln D60 0.186 0.093 0.108 0.004
D90 0.210 0.066 0.129 -0.011
D90/D10 -0.025 -0.137 0.182 -0.011
P200 * D90 -0.016 -0.139 0.197 -0.019
Log D90 0.220 0.048 0.175 -0.028
Ln D90 0.220 0.048 0.175 -0.028
D100 -0.043 -0.028 -0.016 0.014
Cu -0.043 -0.104 0.102 0.002
Log Cu -0.123 -0.170 0.061 0.035
Ln Cu -0.123 -0.170 0.061 0.035
Cc -0.142 -0.094 -0.051 0.036
Log Cc -0.097 -0.017 -0.102 0.029
af 1.000 0.704 -0.028 -0.121
bf 0.704 1.000 -0.464 -0.108
cf -0.028 -0.464 1.000 0.067
hr -0.121 -0.108 0.067 1.000
124
5.4 Physical Significance of SWCC Parameters
It is important to understand the effect of each parameter into the shape of
the SWCC given by Fredlund and Xing, 1994 (see equation 5-1). A sensibility
analysis of each fitting parameter on the shape of the SWCC function I depicted
in Figures 5-2 through 5-5.
Figure 5-2 shows four SWCCs, where the parameters b f , c f and h r are
fixed while the parameter a f is varying. This parameter a f is associated with the
initial break of the SWCC, commonly known as the air–entry value. At this point,
the air starts entering the soil filling up the larger pores. As the a f parameter
increases, the matric suction increase. Fine grained soils have higher air–entry
values than granular material, and therefore, fine grained soils require more
pressure than granular soils to remove the same amount of water. It should also be
notice that the suction at the inflection point correspond to the parameter a f . This
is an important observation because historically a f has been related to the air entry
value and not with the inflection point of the SWCC.
In Figure 5-3 the parameter b f varies while the other parameters remain
fixed. This graph clearly shows that b f parameter is intimately related with the
slope of the SWCC. The higher the bf value, the steeper the SWCC becomes.
In Figure 5-4 the parameters a f , b f and h r are fixed while the parameter c f
changes. It can be observed that the parameter c f is related to the parameter b f ;
because the slope of the SWCC is steeper as C f increases. Furthermore, the
125
100
90 SWCC
af = 500 Parameters
80
af varies
bf = 1
70 cf = 2
af = 100 hr = 500
60
126
50
af = 10 af = 1.0
40
af = 10
30
af = 1.0 af = 100
20
af = 500
10
0
0.001 0.01 0.1 1 10 100 1000 10000 100000 1000000
Matric Suction, kPa
Degree of Saturation, %
Figure 5-2. Changes in the SWCC Shape Due to Changes in the a f Parameter
100
90 SWCC
Parameters
80 af =10
bf varies
70 cf = 2
hr = 500
60
50 bf = 0.5
127
bf = 0.5 bf = 1.0
40 bf = 2
bf = 5
30
bf = 1
20
10 bf = 2
bf = 5
0
0.001 0.01 0.1 1 10 100 1000 10000 100000 1000000
Matric Suction, kPa
Figure 5-3. Changes in the SWCC Shape Due to Changes in the b f Parameter
100
SWCC
90 Parameters
80 af =10
bf =1
cf varies
70 hr = 500
60
50 cf = 0.5
128
cf = 0.5
40 cf = 1
cf = 1
30 cf = 2
20 cf = 2
cf = 500
10 cf = 500
0
0.001 0.01 0.1 1 10 100 1000 10000 100000 1000000
Matric Suction, kPa
Figure 5-4. Changes in the SWCC Shape Due to Changes in the c f Parameter
100
90 SWCC
hr = 500 Parameters
80 af =10
hr = 10 bf =1
70 cf = 2
hr varies
60
hr = 1
50 hr = 0.1
129
hr = 0.1
40
hr = 1.0
30
hr = 10
20
hr = 500
10
0
0.001 0.01 0.1 1 10 100 1000 10000 100000 1000000
Matric Suction, kPa
Figure 5-5. Changes in the SWCC Shape Due to Changes in the h r Parameter
residual degree of saturation decreases as c f parameter increases. Finally, Figure
5-5 shows the variation of the SWCC Shape when hr parameter varies while a f ,
b f , and c f remains constant. It can be seen that the sensibility of the SWCC due to
changes in the hr parameter is relatively low when compared to the changes
observed due to the variation of the other parameters.
5.5 SWCC Prediction Models for Fine–Grained Soils
Table 5-8 presents a summary of the models proposed for the SWCC
parameters for fine–grained soils. A detailed process of how the models were
obtained is presented in the next four sections.Briefly, the process used to find
models is as follows. This process was used, in general to find the models for the
four SWCC parameters, a f , b f , c f and h r :
Step 1. Choosing the Best Predictors: Based on the results obtained with
the correlation matrix, the best predictions were chosen to be correlated with the
SWCC fitting parameter. The database was sorted by the best predictor from the
lowest value to the highest value. The data points were grouped by taking 300
consecutive data points. Either the average or median was chosen to represent the
values of the predictor as well as the value of the fitting parameter.
Step 2. Regression Analysis: The best predictors of each parameter were
then use to regress several models. The statistical package MiniTab® 15 was used
for the analysis. In some instances, the analysis was complemented with features
from Statistica® 5.5 and Microsoft Excel®.
130
Table 5-8. Proposed Models for the SWCC Parameters for Fine–Grained Soils
 
 
 
S (%) = C(ψ ) × 
1
cf 
Modified    ψ   
bf
 
 ln e +     
Soil-Water Characteristic  
   a   
 
Curve Equation
(Fredlund and Xing, 1994)  ψ 
ln 
1 + h  
C (ψ ) = 1 −  r 
  1,000 ,000 
ln 1 + 
 

  hr 
 2.7 
 0.69− 
SWCC Parameter af, kPa 1+ e 4 − 0.14 GI 
a f = 10 
 0.78 
 6.75 −0.19 GI 
SWCC Parameter bf
b f = 10  1+ e 
SWCC Parameter cf c f = 0.03 + 0.62 × e

(− 0.82 (log a f − 0.57 )2 )
660
SWCC Parameter hr, kPa hr = 494 +
1 + e (4 − 0.19 GI )
Where:
S = Degree of Saturation, %
ψ = Matric Suction, kPa
GI = Group Index
Non–linear regression analyses were performed for each combination of
variables by using the least squared error criterion, in which the sum of the
differences between the squares of measured values and estimated values are
131
minimized. The regression results yielded the best model. A plot of measured
versus predicted values is presented.
Step 3. Error Analysis: The error analysis was performed by using
equations 5-4 through 5-8 to assess the accuracy of the model proposed. The data
was analyzed and summary tables showing the errors found in each case, were
created.
Step 4. Final Assessment of the SWCC parameters Models: A final
assessment of the validity of the four models proposed (one for each parameter)
was performed. In order to accomplish that, the predicted parameters were applied
to the Fredlund and Xing equation and the predicted degree of saturation was
obtained. This value was compared with the measured degree of saturation.
The following sections detail the steps follow to obtain the models.
5.5.1 Modeling SWCC parameter a f .
Based on the correlation matrix shown in Table 5-6, the Group Index (GI)
and the wPI parameters were chosen as the best predictor of the a f parameter,
despite the low correlation observed. Following the moving average procedure
described above, the data was grouped and the median and mean values were used
in the regression analysis.
Several trial models were analyzed. The models were carefully chosen
based on previous published corrections and the trends observed in the correction
132
matrix. The mean and the median values were used; but in general, the best
correlations were found to be related to the mean values.
Table 5-9 summarizes the best correlations found out of many trials. In
general, the Group Index showed better performance as predictor than the wPI
value. Also, the logarithmic of a f parameter was found to correlate better than the
arithmetic value.
Based on the results shown in Table 5-9, the model proposed for this
parameter is Model 3.
log(a f ) = 0.69 −
2.7
.................................................................(5-5)
1 + e (4−0.14GI )
Where:
GI = Group Index, expressed in Chapter 3, equation 3-7 as follows:
GI=(P 200 -35)[0.2+0.005(LL-40)]+0.01(P 200 -15)(PI-10)
Where:
P 200 = Material Passing # 200 US Standard Sieve
LL = Liquid Limit
PI = Plasticity Index
Although this equation has not the best correlation, this model was
selected for two simple reasons. First, the third order polynomial equations given
133
by models 1 and 2 present maximum and minimum that implies the correlation
will yield the same a f parameter value for different group indexes; and second, the
expressions are rather complicated for the little gained of accuracy.
The plot of the model selected is shown in Figure 5-6. The statistical
analysis yielded the following results:
Number of Data Points, n = 31,835
Absolute Mean Error, e abs = 30.01
Algebraic Mean Error, e alg = -12.19
Standard Error divided by the Standard Deviation, S e /S y = 0.21
Adjusted Coefficient of Determination, R2 (adjusted) = 0.9552
The spreadsheet used to calculate the statistical errors is shown in Figure
5-7. This spreadsheet was used on a check for the R–squared calculation, by using
a similar expression.
To evaluate de goodness of the model proposed (in other words the model
accuracy), the Figure 5-8 present graphically the relationship between the
measured values versus the predicted values.
134
Table 5-9. Summary of Trials Finding the Best Model (Parameter a f )
# Model Equation R2 Type of Data

3 2
1 log af = 0.00004GI - 0.004GI + 0.041GI + 0.469 0.9633 Average
3 2
2 log af = -0.1964Ln (GI+1) + 0.7766Ln (GI+1) - 0.7248Ln(GI+1) + 0.5549 0.9579 Average
3 log(af) = 0.69-2.7 / (1+EXP(4-0.14GI) 0.9552 Average
4 log(af) = 0.6338-2.6978 / (1+EXP(3.9958-0.1362GI) 0.9549 Average

2
5 log af = -0.0003GI - 0.0433GI + 0.8487 0.9392 Average
2
6 log af = -0.5067Ln (GI+1) + 1.5514Ln(GI+1) - 0.2216 0.9094 Average
3 2
7 log af = 0.00004GI - 0.0035GI + 0.0298GI + 0.5147 0.802 Average
2
8 log af = 0.0003GI - 0.0708GI + 1.0549 0.7628 Average
9 log af = -0.0729Ln3 (GI+1) + 0.0823Ln2 (GI+1) + 0.2902Ln(GI+1) + 0.3281 0.7375 Average
10 log af = -0.4317Ln2 (GI+1) + 1.2851Ln(GI+1) - 0.0623 0.7311 Average
11 log af = 0.00009wPI3 - 0.0074wPI2 + 0.1008wPI + 0.2317 0.6129 Average
12 af = 0.00007GI3 - 0.0045GI2 - 0.0457GI + 4.7922 0.5982 Median
13 log af = 0.00008GI3 - 0.0082GI2 + 0.1591GI + 0.0602 0.5964 Median
14 af = 0.0002GI3 - 0.022GI2 + 0.5614GI + 5.5527 0.5815 Average
15 af = 0.00008wPI3 - 0.0054wPI2 - 0.0081wPI + 4.3602 0.5793 Median
16 af = -9.5517log3 (wPI) + 13.898log2 (wPI) + 3.9952log(wPI) + 1.6786 0.5747 Median
17 af = -0.5543Ln3 (GI+1) + 2.0125Ln2 (GI+1) + 0.6174Ln(GI+1) + 4.1266 0.5684 Average
18 af = 0.0004wPI3 - 0.039wPI2 + 0.9793wPI + 3.1511 0.5615 Average
19 af = 0.0016wPI2 - 0.1746wPI + 5.1796 0.5528 Median
20 af = 0.0016log2 (wPI) - 0.1746log(wPI) + 5.1796 0.5528 Median
135
Group Index vs Log (af) for Fine-Grained Soils
1
0.5 log(af) = 0.69 - 2.7 / (1 + EXP(4 - 0.14GI))

R2 = 0.9552
n = 31,835
0
Se/Sy = 0.21
Log (af) Measured
-0.5
-1
-1.5
-2
-2.5
-3
0 10 20 30 40 50 60 70
Group Index (Measured)
Figure 5-6. Model Predicting Parameter a f for Fine-Grained Soils
2 2
ealg eabs (Sm - Sp) (Sm(avg) - Sp)
Model: ealg = -12.19
log(af) = 0.69 - 2.7 / (1 + exp (4 - 0.14GI)) eabs = 30.01
S= 1.91 42.65
ngroup = 51 51
Checking Calculation p= 1 1
R2 adj = 1-((SSE/(n-p))/(SST/(n-1))) Se = 0.20
Sm(Avrg) -0.4885
2
SSE = S (Sm - Sp) 1.91 Sy = 0.92
2
SST = S (Sm(avg) - Sm) 42.65 Se/Sy = 0.21
R2 adj = 0.9552 R2 = 0.9552
Figure 5-7. Spreadsheet Used in Calculating Errors and R2 Values for the a f
Model for Fine–Grained Soils
136
SWCC Parameter af (Measured vs Predicted)
1
log(af) = 0.69 - 2.7 / (1 + exp (4 - 0.14GI))
0.5 R2 = 0.9552
n = 31,835
log (af), Predicted
0 Se/Sy = 0.21
-0.5
-1
-1.5
-2
-2.5
-2.5 -2 -1.5 -1 -0.5 0 0.5 1
log (af), Measured
Figure 5-8. Measured versus Predicted Analysis for SWCC Parameter a f
5.5.2 Modeling the SWCC parameter b f .
Using the same procedure used for the analysis of parameter a f , the model
obtained for the SWCC parameter b f was based on correlations of this parameter
with the soil properties that were consider the best predictors. Despite the low
correlation observed (see Table 5-5) when the data treated independently,
parameters such as wPI and GI were chosen as the best predictors of the bf
parameter. Table 5-10 summarizes the results of non–linear regressions performed
for several trials. The average values used corresponded to the mean of the values
per Group Index number, in cases where the variable independent was different of
GI the average correspond to the mean of 300 consecutive values, arranged in
increasing order of wPI.
137
Table 5-10. Summary of Trials Performing in Finding the Best Model for
the Parameter b f for Fine–Grained Soils
# Equation R2 Type of Data
log(bf) = -0.00002wPI + 0.0019wPI2 - 0.0337wPI + 0.1353

3
1 0.8652 Avg 300 data points
2 log (bf) = 1/(1+EXP(5-(0.125*GI))) 0.8612 Average
3 log (bf) = 0.78/(1+EXP(6.75-(0.19*GI))) 0.8509 Average
4 log(bf) = -0.00003wPI3 + 0.0025wPI2 - 0.0465wPI + 0.1842 0.7890 Average
bf = -0.00002GI + 0.0021GI2 - 0.0392GI + 0.1366

3
5 0.7677 Average
6 bf = -0.0007wPI3 + 0.0669wPI2 - 1.2122wPI + 7.463 0.6981 Average
7 bf = -0.0006wPI3 + 0.0556wPI2 - 0.8916wPI + 5.9533 0.6358 Avg 300 data points
8 log(bf) = -0.00004wPI3 + 0.0035wPI2 - 0.0663wPI + 0.2656 0.4701 Average
9 bf = -0.001wPI3 + 0.0851wPI2 - 1.471wPI + 8.2345 0.4008 Average
10 log(bf) = -0.00003E-05wPI3 + 0.0033wPI2 - 0.0712wPI + 0.2753 0.3319 Median
11 bf = -0.0007wPI3 + 0.0636wPI2 - 1.3822wPI + 6.5846 0.2285 Median
12 bf = 0.00001af3 - 0.0031af2 + 0.235af + 3.0982 0.0273 Total
The model proposed for this parameter is the Model 3 on Table 5-10:
log(b f ) =
0.78
(6.75−0.19 GI ) .........................................................................(5-6)
1+ e
Although this model did not yield the best correlation, it was selected for
two reasons: The third–order polynomial equation, given by trial #1, presents
maximum and minimum values which will yield the same bf parameter for
different wPI values. On the other hand, the second expression (Trial #2) yields
very high values for intermediate values of GI.
138
The plot of the model selected is shown in Figure 5-9. The statistical
analysis of errors yielded the following values:
Algebraic Mean Error, e alg = 87.04
5-10. This spreadsheet allows for the check of the R–squared calculation by using
a similar expression than that used by Statistica®.
Figure 5-11 presents the relationship between the measured b f parameters
and the predicted values. This figure allows evaluating the goodness of the fit or
accuracy of the model proposed for the b f parameter.
139
Group Index vs Log (bf) for Fine-Grained Soils
1.25
1 log (bf) = 0.78/(1+EXP(6.75-(0.19*GI)))

R2 = 0.8509
n = 31,833
Log (bf) (Measured)
0.75 Se/Sy = 0.39
0.5
0.25
-0.25
0 10 20 30 40 50 60 70
Group Index (Measured)
Figure 5-9. Model Predicting Parameter b f for Fine–Grained Soils
ealg eabs (Sm - Sp)2 (Sm(avg) - Sp)2

Model: ealg = 21.31
log (bf) = 0.78/(1+EXP(6.75-(0.19*GI))) eabs = 87.04
S= 1.09 7.32
ngroup = 58 58
Sm(Avrg) 0.3060
2
SSE = S (Sm - Sp) 1.09 Sy = 0.36
2
R2 adj = 0.8509 R2 = 0.8509
Figure 5-10. Spreadsheet Used in Calculating Errors and R2 Values for the b f
140
SWCC Parameter bf (Measured vs Predicted)
1.4
log (bf) = 0.78/(1+EXP(6.75-(0.19*F3)))
1.2 R2 = 0.8509
n = 31,833
1 Se/Sy = 0.39
log (bf), Predicted
0.8
0.6
0.4
0.2
-0.2
-0.2 0 0.2 0.4 0.6 0.8 1 1.2
log (bf), Measured
Figure 5-11. Measured versus Predicted Analysis for SWCC Parameter b f
5.5.3 Modeling SWCC parameter c f .
The trials shown in Table 5-11 reflect some of the attempt to find a good
correlation. As shown by the correlation matrix (Table 5-6), the c f parameter
seem to be highly correlate with parameter b f . However, when the data was
analyzed by the average of 300 consecutive data points, a strong correlation
between the a f and the c f parameter were found; as well as an important
correlation with the GI parameter.
141
Table 5-11. Summary of Trials Performed in Finding the Best Model for
the Parameter c f for Fine-Grained Soils.

GI
1 cf=0.08+(0.59*0.94 ) 0.9825 Average per GI
2 cf=0.03+0.62*(exp(1)^(-0.82*((logaf-0.57)^2))) 0.9215 Avg of 300 data
3 cf=0.65*(exp(1)^(-0.65*((logaf-0.58)^2)))) 0.9038 Average

3 2
4 cf = -0.0302log (af) - 0.0992log (af) + 0.1907log(af) + 0.5281 0.8266 Avg of 300 data
3 2
5 cf = 0.0007af - 0.0208af + 0.1651af + 0.2745 0.6777 Avg of 300 data
Based on the results from different trials, the model proposed for the
parameter c f is the Model 2 presented in the Table 5-11:
c f = 0.03 + 0.62 × e
(−0.82 (log a f −0.57 )2 )
..........................................................(5-7)
Equation 5-7 was selected due to the convenience of being able to
mathematically link the c f parameter with the a f parameter. On the other hand,
equation 5-7 yields an acceptable adjusted R2 value of 0.9215. The plot of the
model selected is shown in Figure 5-12. The statistical analysis of errors yielded
the following:
142
The spreadsheet used to calculate the statistical errors is shown on Figure
5-13. This spreadsheet allows for checking the R–squared calculation using a
similar statistical expression as that used by Statistica®.
To evaluate de goodness of the fit or accuracy of the model proposed, the
measured values were plotted against the predicted values as shown in Figure 5-
14. The fine line shows a linear regression between the measured and predicted
points, which indicates a relative unbiased prediction.
143
Log (af Parameter) vs Parameter cf for Fine-Grained Soils
0.8
0.7
0.6 cf=0.03+0.62*(EXP(-0.82*((logaf-0.57)^2)))
Parameter cf, Measured
R2 = 0.9215
0.5 n = 31,520
Se/Sy = 0.28
0.4
0.3
0.2
0.1
0
-4 -3 -2 -1 0 1 2 3
log af - Measured
Figure 5-12. Model Predicting c f Parameter for Fine–Grained Soils
ealg eabs (Sm - Sp)2 (Sm(avg) - Sp)2

Model: ealg = -1.41
cf=0.03+0.62*(EXP(-0.82*((logaf-0.57)^2)) eabs = 10.25
S= 0.23 2.99
ngroup = 107 107
2
R adj = 1-((SSE/(n-p))/(SST/(n-1))) Se = 0.05
Sm(Avrg) 0.5323
2
SSE = S (Sm - Sp) 0.23 Sy = 0.17
2
R2 adj = 0.9215 R2 = 0.9215
Figure 5-13. Spreadsheet Used in Calculating Errors and R2 Values for the c f
144
SWCC Parameter cf (Measured vs Predicted)
0.75
cf=0.03+0.62*(EXP(-0.82*((logaf-0.57)^2)))
0.6 R2 = 0.9215
n = 31,520
Se/Sy = 0.28
0.45
145
0.3
0.15
,f Predicted Parameter c
0
0 0.15 0.3 0.45 0.6 0.75
Parameter cf, Measured
Figure 5-14. Measured versus Predicted Analysis for SWCC Parameter c f

5.5.4 Modeling SWCC parameter h r .
Table 5-12 summarizes the regression trials attempted to correlate the hr
parameter with the GI value. Several trials were attempted using the wPI with no
success. The average and media values as well as grouping 300 consecutive data
points were attempted.
Table 5-12. Summary of Trials finding the best model (Parameter h r )
1 hr = 494 + 660 / (1 + exp(1)^(4 - 0.19GI)) 0.9041 Avg of 300 data
2 hr = -0.0082GI3 + 0.5996GI2 + 4.342GI + 496.67 0.8775 Avg of 300 data
3 hr = -0.0012GI3 + 0.0082GI2 - 0.2679GI + 499.65 0.8392 Avg of 300 data
4 hr = 9.0009GI + 570.31 0.4563 Average
5 hr = 0.0169GI3 - 2.0349GI2 + 65.747GI + 238.02 0.4314 Average
6 hr = 0.1449GI2 - 6.5654GI + 546.35 0.4188 Median
The model proposed for the hr parameter is Model 1 shown in Table 5-12:
660
hr = 494 + ..........................................................................(5-8)
1 + e (4−0.19GI )
This equation was selected because it has the best coefficient of
determination and behaves asymptotically on the extremes. The plot of the model
selected is shown in Figure 5-15. The statistical analysis of errors yields the
following results:
146
5-16. This spreadsheet allows for checking the R–squared calculation using a
similar expression as that used by Statistica®.
To evaluate de goodness of the fit or accuracy of the model proposed for
the hr parameter for fine–grained soils the measured versus the predicted values
were plotted as shown in Figure 5-17. It can be seen that the model is unbiased,
which is reflected in the low e alg found.
147
Group Index vs SWCC Parameter hr for Fine-Grained Soils
1300
1200
1100
1000
900
148
800
700 hr = 494 + 660 / (1 + EXP(4 - 0.19GI)

R2 = 0.9041
600 n = 31,839
Se/Sy = 0.31
500
400
0 10 20 30 40 50 60 70
Group Index - Measured
Figure 5-15. Model Predicting h r Parameter for Fine-Grained Soils

Parameter hr, kPa, M
2 2
ealg eabs (Sm - Sp) (Sm(avg) - Sp)
Model: ealg = -1.39
hr = 494 + 660 / (1 + EXP(4 - 0.19GI) eabs = 4.25
S= 272,094.83 2,837,347.14
ngroup = 107 107
2
R adj = 1-((SSE/(n-p))/(SST/(n-1))) Se = 50.66
Sm(Avrg) 578.2088
2
SSE = S (Sm - Sp) 272,094.83 Sy = 163.61
2 2
SST = S (Sm(avg) - Sm) 2,837,347.14 Se/Sy = 1 - (Se/Sy) 0.31
2 2
R adj = 0.9041 R = 0.9041
Figure 5-16. Spreadsheet Used in Calculating Errors and R2 Values for the h r
149
SWCC Parameter hr (Measured vs Predicted)
1400
1200
1000
150
800
hr = 494 + 660 / (1 + EXP(4 - 0.19GI)
600 R2 = 0.9041
n = 31,839
Se/Sy = 0.31
400
400 600 800 1000 1200 1400
Parameter hr, Measured
Parameter hr, k
Figure 5-17. Measured versus Predicted Analysis for SWCC Parameter h r

A final checking was performed on the database to validate the data information
associated with the plastic limit and saturated volumetric water content.
Figure 5-18 shows the Shrinkage Curve. In this curve are indicated the
limits of consistency when the soil is drying process. The point when immediately
the soil begins the desaturation (or when the curve starts) is considered the plastic
limit. This point is associated with the Air Entry Value (AEV) (Fredlund, et al.,
2011)
Figure 5-18 Curve Gravimetric Water Content vs Void Ratio
151
A statistical analysis of the differences between the plastic limit and the
saturated volumetric water content can be observed in the histogram shown in
Figure 5-19. For plastic soils the values for the saturated volumetric water content
and the plastic limit should be really close.
The statistical analysis was performed on a range of data-points with the
95% of confidence considering a normal distribution like can be observed in
Figure 5-19. The data points out of this range can be considered suspicious and
should not be included in any regression analysis.
A difference between the plastic limit and the saturated volumetric water
content greater than 10 is affecting the validity of the model and makes the fit and
data suspicious. At this point, further attempt to find a better model for plastic
material should be consider the following:
Eliminate soil data that look suspicious. The comparison of plastic soil and
saturated volumetric water content is a starting step that can help in the
elimination process.
The initial SWCC fitting parameters for the Fredlund and Xing model
should be analyzed carefully. a f parameter for plastic materials should be higher
than 10, because a regression analysis with low values of a f create in SWCC a
bimodal shape. A low value of a f force the curve to have an initial decreasing and
152
the same time to approach to the given points of water content at 33 and 1500
kPa. This effect can be observed in the soil 1 in Figures 5-36 and 5-37
700
13.8%
600
Difference = 100(θsat-PL)/θsat
Mean = 13.8 %
500 n = 30,290 data points
At 95% of Confidence the range
Frequency
400 is between -88% and 100%
300
200
100
0
-88 -78 -67 -57 -46 -35 -25 -14 -4 7 17 28 39 49 60 70 81 92
Difference in Percentage
Figure 5-19 Histogram Showing Differences between Plastic Limit and
Saturated Volumetric Water Content
5.6 SWCC Prediction models for Granular Soil
Table 5-13 presents a summary of the SWCC parameters models proposed
for granular or non-plastic soils. The complete analysis is presented in the
following four sections. These parameters apply to the Soil–Water Characteristic
Curve (equation 5-1) defined by Fredlund & Xing, 1994.
The equation for a f is a function of Particle Size D 10 . The effective
particle size D 10 has been related to the coefficient of permeability in the past
153
(Hazen, 1911) and therefore, it seems logical that it correlates well with moisture
retention characteristic. It can be seen from the correlation matrix (Table 5-7) that
the correlations among the parameters yielded the highest correlation values. The
SWCC parameter b f was found to correlate with a f and parameter c f is inferred
from b f parameter. The dependency between b f and c f with a f is found to be
convenient because it eliminates the possibility of not getting a continuous SWCC
function once the fitting parameters are put together in the Fredlund & Xing
equation. Finally, the parameter h r yielded a constant value of 100.
The process followed to estimate the fitting parameter for granular soils
was the same used for fine–grained soils. Refer to section 5.5 for details.
154
Table 5-13. Proposed Models for the SWCC Parameters for Granular Soils
 
 
 
S (%) = C(ψ ) × CF(ψ ) × 
1
cf 
Modified     ψ  b f   
Soil-Water Characteristic  ln e +     
       
a
Curve Equation
(Fredlund and Xing, 1994)  ψ 
1 + h 
ln  
 r 
C (ψ ) = 1 −
  1,000 ,000 
ln 1 + 
 

  hr 
a f = −967 .21 D10 + 218 .37 D10 − 2.7

2
SWCC Parameter af, kPa
Constraint: if D10 < 0.020, then af = 1.28
SWCC Parameter bf b f = 10
(−0.0075a f
3
)
+ 0.1133a f 2 −0.3577a f + 0.3061
c f = 0.0058a f − 0.0933a f + 0.4069a f + 0.3481

3 3
SWCC Parameter cf
SWCC Parameter hr, kPa hr = 100
Where:
S = Degree of Saturation, %
ψ = Matric Suction, kPa
D 10 = Grain Diameter at 10% Passing by Weight
5.6.1 Modeling the SWCC Parameter a f for Granular Soils.
The model obtained for the SWCC parameter a f is given as a function of granular
soils properties found to be the best predictors according to results presented in
Table 5-7 and explained in section 5.3.2. Table 5-14 summarizes some of the
trials used to estimate the best model. The variables used in the regressions were
155
the effective particle size D 10 , percentage passing sieve #200 and the grain–size
distribution shape parameters Cu (Coefficient of uniformity) and Cc (Coefficient
of curvature). These geotechnical properties were considered in previous studies
such as the dissertation written by Zapata, 1999, and Perera, 2003.
The database used for the non–linear regression analysis consisted of 4,485 data
points. After the database was sorted from minimum to maximum value of the
SWCC parameter, the average of 50 consecutive points was obtained. Groups of
50 data points were desired due to the fact that the database corresponding to
granular materials was smaller than the database obtained for fine–grained soils.
From the results presented in Table 5-14, it can be seen that the expression
makes use of all the soils properties found to be good predictors of parameters af.
Expression given by trial 2 does not consider the Cu and Cc properties and has a
light low R2. Equation 3 uses only the particle size D 10 as independent variable,
and yet keeps a R2 value of 0.72. The expression shown for trials 4 through 8 are
rather complex and do not improve the correlations.
156
1 af=1.574+0.1296P200+0.0442D90+0.3962D60+1.2088D30- 0.7420 Avg of 50 data
24.6642D20+145.7467D10+1.5322logD90+2.5409logD10-12.2653/cc+95.1692/cu
2 af = 1.1319 + 0.1032P200 + 0.0357D90 + 0.4277D60 + 0.035D30 - 20.5359D20 + 136.7032D10 + 1.8858*logD90 0.7377 Avg of 50 data
+ 2.1319logD10
3 af = -967.21D102 + 218.37D10 - 2.7006 0.7167 Avg of 50 data
4 af = 1.13 + 0.103 P200 + 0.0357 D90 + 1.89 log(D90) + 0.43 D60 + 0.0 D30 - 20.5 D20 + 137 D10 + 2.13 log(D10) 0.7120 Avg of 50 data
157
5 af = 0.28 + 0.149 P200 + 0.0028 D90 + 2.05 log(D90) + 1.67 D60 - 9.2 D30 - 10.1 D20 + 142 D10 + 2.29 log(D10) 0.7070 Avg of 50 data
- 0.000020 Cu + 0.00133 Cc
6 af = - 1.51 + 0.060 P200 + 0.0485 D90 + 1.30 log(D90) - 1.29 D60 + 10.5 D30 - 29.8 D20 + 130 D10 + 1.59 0.7050 Avg of 50 data
log(D10) + 10.4 log(D60) - 11.9 log(D30) + 5.7 log(D20)
Parameter a f for Granular Soils
7 af = - 3.42 + 0.123 P200 + 0.0074 D90 + 1.51 log(D90) - 0.26 D60 + 5.5 D30 - 26.5 D20 + 137 D10 + 1.82 0.7010 Avg of 50 data
log(D10) - 0.000028 Cu + 0.00249 Cc + 14.5 log(D60) - 20.6 log(D30) + 10.9 log(D20)
8 af = 11.6 - 0.010 P200 + 1.03 log(D90) + 3.57 log(D10) + 13.6 log(D60) - 29.9 log(D30) + 21.3 log(D20) 0.6530 Avg of 50 data
Table 5-14. Summary of Trials Use in Finding the Best Model for
The model proposed for the parameter a f is the Model 3 presented in
Table 5-14:
a f = −967.21D10 + 218.37 D10 − 2.7 ......................................................(5-9)

2
Constraint: if D 10 < 0.020, then a f = 1.28
Figure 5-20 shows the plot of the selected model. The error analysis
yielded the following results:
5-21.
The goodness of the model proposed 5-9 is evaluated and illustrated in
Figure 5-22 where the measured values of a f are plotted versus the predicted
values.
158
D10, mm versus SWCC Parameter af for Granular Soils
12
10
159
4 af = -967.21D102 + 218.37D10 - 2.7
R² = 0.7167
n = 4,485
2 Se/Sy = 0.53
1.28
,f kPa (Measured)
0
0.00 0.02 0.04 0.06 0.08 0.10 0.12
Parameter a
D10, mm (Measured)
Figure 5-20. Model Predicting Parameter af for Granular Soils

ealg eabs (Sm - Sp)2 (Sm(avg) - Sm)2
Model:
2
af = -967.21D10 + 218.37D10 - 2.7006 ealg = -139.48
eabs = 170.62
Σ= 310.50 1,096.16
ngroup = 90 90
p= 1 1
R2 adj = 1-((SSE/(n-p-1))/(SST/(n-1))) Se = 1.87
Sm(Avrg) 4.7697
2
SSE = S (Sm - Sp) 310.50 Sy = 3.51
2
SST = S (Sm(avg) - Sm) 1,096.16 Se/Sy = 0.53
R2 adj = 0.7167 R2 = 0.7167
Figure 5-21. Spreadsheet Used in Calculating Errors and R2 Values for the a f
Model for Granular Soils
SWCC Parameter af for Granular Soils - (Measured vs Predicted)

10
af = -967.21D102 + 218.37D10 - 2.7
9
R2 = 0.7167
Parameter af, kPa, Predicted Values
8 n = 4,498
Se/Sy = 0.53
7
0
0 1 2 3 4 5 6 7 8 9 10
Parameter af, kPa, Measured Values
Figure 5-22. Measured versus Predicted Analysis for SWCC Parameter a f
160
5.6.2 Modeling the SWCC parameter b f .
The first step taken in order to analyze the b f parameter for granular
materials was to choose the best soil properties predictors from the matrix of
correlation shown in Table 5-7 and explained in detail under section 5.3.2. Table
5-15 shows some of the trials used in calculating the R–squared. During this
process the database was worked in two ways; initially, all the data was used, then
the average/median values of grouped parameters were analyzed.
Table 5-15. Summary of Trials Used in Finding the Best Model for Parameter b f
1 log(bf) = -0.0075af3 + 0.1133af2 - 0.3577af + 0.3061 0.9668 Avg of 50 data points
3 log(bf) = 0.1384log3(af) + 0.548log2(af) + 0.1755log(af) - 0.0216 0.7067 Avg of 50 data points
4 bf = -0.075af3 + 1.0141af2 - 2.653af + 2.619 0.6418 Avg of 50 data points
5 bf = 0.7458log3(af) + 3.4994log2(af) + 2.1039log(af) + 0.7301 0.4655 All data points
The best model corresponded to that obtained by trial # 1, which presented
the highest R2. The model proposed for the b f parameter is:
log b f = −0.0075a f + 0.1133a f − 0.3577a f + 0.3061 ...................................(5-10)

3 2
The graph corresponding to this model is given in Figure 5-23.Even
though third order polynomials present inflection points that might not reflect the
measure data, in this case it is a good representation of the data obtained.
161
SWCC parameter af (kPa) vs log(bf) for Granular Soils
1.5
1.25
0.75
0.5
162
- )f
0.25
Measured
0 log(bf) = -0.0075af3 + 0.1133af2 - 0.3577af + 0.3061
R² = 0.9668
log (b
-0.25 n = 4,485
Se/Sy = 0.18
-0.5
0 2 4 6 8 10 12
Figure 5-23. Model Predicting Parameter b f for Granular Soils

Parameter af - Measured
Equations 5-4 through 5-8 were used in the error analysis, which gave the
following results:
5-24.

Model:
3 2
log(bf) = -0.0075af + 0.1133af - 0.3577af + 0.3 ealg = 25.73
eabs = 89.80
S= 0.41 12.26
ngroup = 90 90
R2 adj = 1-((SSE/(n-p-1))/(SST/(n-1))) Se = 0.07
Sm(Avrg) 0.3779
2
SSE = S (Sm - Sp) 0.41 Sy = 0.37
2
R2 adj = 0.9668 R2 = 0.9668
Figure 5-24. Spreadsheet Used in Calculating Errors and R2 Values for b f Model
in Granular Soils
163
Figure 5-25 presents the relationship between the measured values versus
the predicted values in order to evaluate de goodness of the model proposed.
SWCC Parameter bf for Granular Soils (Measured vs Predicted)

1
0.75
Parameter bf, Predicted
0.5
0.25
log(bf) = -0.0075af3 + 0.1133af2 - 0.3577af + 0.3061

0 R² = 0.9668
n = 4,497
Se/Sy = 0.18
-0.25
-0.25 0 0.25 0.5 0.75 1
Parameter bf, Measured
Figure 5-25. Measured versus Predicted Analysis for SWCC b f Parameter
5.6.3 Modeling the SWCC parameter c f .
The correlation matrix yielded a f parameter as the best predictor for granular or
non-plastic soils for c f parameter. The results of several correlation trials are
shown in Table 5-16.
The model proposed for this parameter is the Model 1 from Table 5-16:
c f = 0.0058a f − 0.0933a f − 0.4069a f + 0.3481 ...............................(5-11)

3 2
164
Table 5-16. Summary of Trials Used in Finding the Best Model for Parameter c f
2
# Equation R Type of Data
1 cf= 0.0058af3 -0.0933af3 + 0.4069af + 0.3481 0.8735 Avg of 50 data points
2 cf = 2.8649 e -1.227 bf (bf < 3) 0.7978 Avg of 50 data points
2 cf = 1.9 bf -0.58 (bf >= 3) 0.6395 Avg of 50 data points
3 cf = -0.00008bf3 + 0.0031bf2 - 0.0675bf + 0.9272 0.5054 Avg of 50 data points
4 cf = 0.0003bf3 - 0.0086bf2 + 0.0348bf + 0.7391 0.2224 All data points
The model selected is shown in Figure 5-26. The statistical analysis of
errors yielded the following results:
5-27. The goodness of fit or accuracy of the model can be visualized in figure 5-
28, where the measured c f parameter is presented against the predicted value. The
results yielded an unbiased model.
165
1.2
1
SWCC Parameter cf (Measured)
0.8
0.6
0.4
cf= 0.0058af3 -0.0933af3 + 0.4069af + 0.3481
R² = 0.8735
0.2
n = 4,450
Se/Sy = 0.36
0
0 1 2 3 4 5 6 7 8 9 10
SWCC Parameter af (Measured)
Figure 5-26. Model Predicting Parameter c f for Granular Soils

Model: ealg = 1.55
3 2
cf = 0.0058af - 0.0933af + 0.4069af + 0 eabs = 4.74
S= 0.17 1.38
ngroup = 89 89
p= 1 1
Sm(Avrg) 0.6994
2
SSE = S (Sm - Sp) 0.17 Sy = 0.13
2
R2 adj = 0.8735 R2 = 0.8735
Figure 5-27. Spreadsheet Used in Calculating Errors and R2 Value for c f model
for Granular Soils
166
SWCC Parameter cf for Granular Soils (Measured vs Predicted)
1
cf= 0.0058af3 -0.0933af3 + 0.4069af + 0.3481
R² = 0.8720
0.9
n = 4,450
Parameter cf - Predicted
Se/Sy = 0.36
0.8
0.7
0.6
0.5
0.4
0.3
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Parameter cf - Measured
Figure 5-28. Measured Versus Predicted for SWCC Parameter c f
5.6.4 Defining the SWCC parameter h r .
The statistics for the measured hr parameter are as follows:
Average = 100.17
Mode = 100.10
Median = 100.4
Standard Deviation = 1.68
Given these results, it is appropriate to use a value equal to 100 to represent this
parameter. It is important to notice that in order to fit the SWCC equation to the
measured data points, the optimization process requires to assign initial values to
af, bf, cf and hr parameters. For the hr parameter for granular or non-plastic
167
materials, an initial value of 100 was assigned. This implies that the shape of the
SWCC is basically independent of this parameter. Therefore a constant is
proposed:
h r = 100 ................................................................................................(5-12)
The histogram of measured h r parameters illustrates the reasonableness of
this selection.
2500
Mean = 100.17 Histogram
Std Error = 0.03 Parameter hr
2000
Median = 100.4 for Granular
Mode = 100.1 Soils
Frecuency
1500 Std Dev = 1.68

n = 4,478
1000
500
0
90 95 100 105 110
SWCC Parameter hr, kPa
Figure 5-29. Histogram of Parameter h r
5.7 Measured versus Predicted Degree of Saturation
To assess the validity of the model proposed independently for each
parameter, a comparison between measured versus predicted degree of saturation
is presented. The measure values used in the analysis correspond to the saturation
168
at 33 kPa and 1,500 kPa. These values were calculated based on the volumetric
water content directly obtained from testing. On the other hand, the predicted
degree of saturation at 33 kPa and 1,500 kPa of suctions were estimated from the
Fredlund & Xing equation (eq. 5-1) by fitting the predicted equations proposed.
The first analysis was developed for fine–grained soils. Figure 5-30 shows
the graph for Measured versus Predicted degree of saturation, while Figure 5-31
presents the spreadsheet used for the error analysis. As it can be seen the
combined equations yielded an R2 of 0.56.
Measured vs Predicted - Fine Grained Soils

1.2
1
Degree of Saturation, Predicted
0.8
y = 0.5218x + 0.3481
0.6 n = 31,869
R2 = 0.5615
Se/Sy = 0.66
0.4 log(af) = 0.69 - 2.7 / (1 + EXP(4 - 0.14GI))
log (bf) = 0.78/(1+EXP(6.75-(0.19*GI)))

0.2
cf=0.03+0.62*(EXP(-0.82*((logaf-0.57)^2)))
0 hr = 494 + 660 / (1 + EXP(4 - 0.19GI)

0 0.2 0.4 0.6 0.8 1 1.2
Degree of Saturation, Measured
Figure 5-30. Measured vs Predicted Degree of Saturation for Fine–Grained Soils
169
SS err SS tot ealg eabs (Se) (Sy)
2 2
(Sm - Sp) (Sm - Sm(avg)) 100*(Sm - Sp)/Sm - (Sm - Sp) (Sm(avg) - Sm)2
2
Σ = 6.13 13.98 ealg = -25.48

eabs = 25.52
Sm(avg) = 0.5176 Σ= 6.13 13.98
ngroup = 102 ngroup = 102 102
p= 1 p= 1 1
Se = 0.25 Se = 0.25
Sm(avg) = 0.5176
Sy = 0.37 Sy = 0.37
Se/Sy = 0.66 Se/Sy = 0.66
R2 (adj) = 0.5615 Predicted R2 (adj) = 0.5615
Figure 5-31. Spreadsheet Used in Calculating the Error Analysis for Measured
versus Predicted Degree of Saturation for Fine-Grained Soils
The analysis for granular soils followed the same for fined–grained soils.
The predicted values of degree of saturation were calculated at suctions of 33 and
1,500 kPa using the Fredlund & Xing equation and the predicted equation found
for the fitting parameters for granular or non-plastic materials. These predicted
values were compared to the measured values of degree of saturation available in
the database. Figure 5-32 shows the plot of Measured versus Predicted degree of
saturation values for granular soils, for suctions of 33 kPa and 1,500 kPa together.
Figure 5-33 shows the spreadsheet used for the calculations of errors.
170
Measured vs Predicted - Granular Soils y = 0.9128x + 0.017
0.50
af = -967.21D102 + 218.37D10 - 2.7
0.45
log(bf) = -0.0075af3 + 0.1133af2 - 0.3577af + 0.3061
0.40
cf= 0.0058af3 -0.0933af2 + 0.4069af + 0.3481
Degree of Saturation, Predicted
0.35
hr = 100
0.30 n = 4,498
ealg = -3.54
0.25 eabs = 16.43
R2 = 0.8930
0.20 Se/Sy = 0.33
0.15
0.10
0.05
0.00
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50
Degree of Saturation, Measured
Figure 5-32. Measured versus Predicted Degree of Saturation for Granular Soils
SS err SS tot ealg eabs (Se) (Sy)

2 2
(Sm - Sp) (Sm - Sm(avg)) 100*(Sm - Sp)/Sm - (Sm - Sp) (Sm(avg) - Sm)2
2
Σ = 0.28 2.58 ealg = -3.54

eabs = 16.43
Sm(avg) = 0.2137 S= 0.28 2.58
u = 180 ngroup = 180 180
e= 1 p= 1 1
Se = 0.04 Se = 0.04
Sm(avg) = 0.2137
Sy = 0.12 Sy = 0.12
Se/Sy = 0.33 Se/Sy = 0.33
R2 (adj) = 0.8930 Predicted: R2 (adj) = 0.8930
Figure 5-33. Error Analysis for Degree of Saturation for Granular Soils
171
5.8 Procedure to Estimate the SWCC from the Proposed Models
The following is the procedure to obtain the SWCC for Fine-Grained
(plastic) and Granular (non-plastic) soils, based on the work presented in this
Chapter.
Step 1. Obtain the Grain–Size Distribution and the Atterberg’s limits
(liquid limit, LL and plasticity limit, PL)
Step 2. Calculate the Plasticity Index PI = LL – PL
Step 3. Calculate the Group Index, GI, by using equation 5-3
Step 4. Calculate Weighted Plasticity Index, wPI by using equation 5-2
Step 5. Plot the Grain–Size Distribution Curves
Step 6. Obtain the Particle Size, D 10 , (see Figure 5-34)
Step 7. Define the model to be used. For soil with wPI > 0, the soil is
categorized as plastic. If the wPI = 0, the soil is categorized as
non–plastic or granular. If the soil is plastic, continue with step 8.
If the soil is granular or non–plastic soils, continue with step 17
Step 8. Calculate the SWCC parameter a f according to equation 5-5. The
parameter a f is a function of the Group Index GI
172
100
Sieve #200
90
Sieve #4
80
Percent Passing (%)
70
60
50
40
30
20
10
0
0.001 0.01 0.1 D10 1 10 100 1000 10000
Figure 5-34. Calculating D 10 from the GSD curve
Step 9. Calculate the SWCC parameter b f according to equation 5-6. The
parameter b f is a function of the Group Index, GI
Step 10. Calculate the SWCC parameter c f according to equation 5-7. The
parameter c f is a function of the SWCC parameter a f , calculated in step 8
Step 11. Calculate the SWCC parameter h r according to equation 5-8. The
parameter h r is a function of Group Index, GI
Step 12. Calculate the degree of saturation, S(%), according to equation 5-
1, which correspond to the SWCC function defined by Fredlund & Xing, 1994
Step 13. Plot the soil–water characteristic curve
173
Step 14. If the particle size D 10 , calculated in step 6, is less than 0.020, use
a f = 1.28. Otherwise, go to step 15
Step 15. Calculate the SWCC parameter a f according to equation 5-9. In
this equation the parameter a f is a function of particle size D 10
Step 16. Calculate the SWCC parameter b f according to equation 5-10.
The parameter b f is a function of SWCC parameter a f calculated in steps 17/18
Step 17. Calculate the SWCC parameter c f according to equation 5-11.
The parameter c f is a function of SWCC parameter a f obtained in steps 17/18
Step 18. Use h r = 100
Step 19. Calculate the degree of saturation, S%, using the Fredlund &
Xing equation 5-11 and the fitting parameters estimated in steps 15 to 18
Step 20. Plot the soil–water characteristic curve
The procedure outlined before is summarized in Figure 5-35.
174
Input:
Atterberg Limits and
Gradation (Step 1)
Steps 2 to 4: Calculate
Plasticity Index, Group Index,
and Weighted Plasticity Index
Steps 5 to 6: Plot Grain-Size

Distribution Curve, and find D10
wPI > 0 wPI = 0

Step 7: NON-PLASTIC
PLASTIC SOILS
wPI SOIL
Steps 8 to 11:
Calculate af (eq. 5-5), D10 < 0.020 Step 14: D10 ≥ 0.020
bf (eq.5-6), cf (eq. 5-7), D10
hr (eq.5-8)
Steps 15 to 18:
Step 12:
Calculate af (eq. 5-9),
Calculate Degree of Saturation, eq. 5-1 af = 1.28
bf (eq. 5-10), cf (eq. 5-11).
(Fredlund & Xing, 1994)
Use hr=100
Step 13:
Plot SWCC Step 19:
Calculate Degree of
Saturation, (eq. 5-1)
(Fredlund & Xing, 1994)
Step 20:
Plot SWCC
Figure 5-35. Approach to Estimate the SWCC Based on Statistical Correlation of
Fredlund & Xing Parameters with Soil Index Properties
175
5.9 Summary
Tables 5-8 and 5-13 present the proposed models for the SWCC
parameters for the Fredlund and Xing equation, for plastic and non-plastic soils,
respectively. The models proposed for plastic soils were estimated in function of
the Group Index, which is in turn a function of passing sieve #200, liquid limit
and plasticity index. On the other hand, the models proposed for non-plastic soils
were estimated as function of the particle diameter D 10 .
The procedure followed to develop this Chapter 5 had the following order:
• The database was classified according to the wPI property.
• The measured SWCC parameters were correlated with all the soil
properties affecting the SWCC.
• Each SWCC parameter was subjected to a statistical non–linear
regression analysis against all possible combination of parameters,
followed of a statistical error analysis.
• Based on statistical, geotechnical and applicability considerations,
the best model was chosen for each SWCC parameter.
• Each Measured fitting parameter was compared to the Predicted
value.
176
• The predicted degree of saturation was obtained by fitting the
predicted SWCC parameter to the Fredlund & Xing function.
• Final model for plastic and non-plastic soil were proposed.
The models proposed in Chapter 5 to estimate the SWCC parameters,
present the following advantages:
• The models proposed can be implemented in the Enhanced
Integrated Climatic Model (EICM), which is incorporated in the
Mechanistic–Empirical Pavement Design Guide (MEPDG). The
format required allows including the SWCC fitting parameters.
• The database is vast. With more than 31,000 data points for plastic
soils and 4,500 data points for non-plastic soils, it contains the
most important Soil Index Properties obtained directly from
laboratory testing or in the field. This database can be considered
the largest in the world containing unsaturated soil properties.
• The models proposed are very simple to be implemented. For
plastic soils, the Atterberg’s Limits and the Passing US sieve #200
are needed as input parameters; while for non-plastic materials,
only the particle diameter D 10 is needed. These are parameters
commonly used by practicing engineers and therefore, this model
becomes an excellent candidate for practical applications.
177
The approach and the models proposed in this chapter have the following
limitations:
• The models proposed for the SWCC fitting parameters were
estimated independently of each other. That makes it difficult to
control the shape of the Soil–Water Characteristic Curve.
• Due to the tremendous amount of data points in the database, a
moving average estimate of the parameters during the statistical
analysis was necessary. The fine-grained material database was
sorted according to the wPI and the data were averaged in groups
of 300 consecutive data points. Groups of 50 consecutive data
points were used for non-plastic soils, based on the D 10 parameter.
While working with the moving average allows finding a clear
tendency of the data and hence, better correlations; the variability
gets somehow masked within each range. A deeper study of
variability should be performed.
When comparing the model for plastic soils found by Zapata, 1999
(Figures 4-25 and 4-26) with the model proposed in this work (Figures 5-30 and
5-32), it can be seen that the R2 improved from 0.70 to 0.81; while the R2
improved from 0.40 to 0.89 for non-plastic materials. Even though the R2
improved marginally for plastic soils, it can be observed that the Zapata's model is
biased towards overprediction for most of the dataset. For non–plastic soils the
178
results obtained with the Zapata’s model are underpredicting most of the data
points.
On the other hand, when comparing the model proposed in this work
(Figure 5-30) with the plastic model used in the MEPDG model, it can be seen
that the R2 improved greatly from 0.49 to 0.81. For non-plastic or granular soils,
the R2 for both models are somewhat similar. Even though the non–plastic model
from the MEPDG model is almost similar than the model proposed in this work,
the later model is much simpler and easier to implement because it only depends
on one gradation parameter, the particle size at 10% Passing or D 10 .
It can be concluded that the new models proposed in this Chapter 5 will
enhanced the prediction of the SWCC and therefore, it is recommended to
consider applying them in practical applications.
As an example of the predicted SWCC based on the models proposed in
Chapter 5, the soil index properties shown in Table 5-17 were used to calculate
the SWCC fitting parameters. The three soil–water characteristic curves obtained
are shown in Figure 5-36.
179
Table 5-17. Soil Index Properties for Three Soils Taken from the Database
Classification Group Predicted SWCC Parameters

Soil USCS AASHTO P200 LL PI D10 wPI Index af bf cf hr
1 CH A-7-6 85 65 40 - 34.00 37 0.0709 2.1570 0.0849 1,102.1
2 CL A-6 65 40 25.5 - 16.58 14 1.8436 1.0489 0.6047 691.5
3 SP-SM A-1-a 5 0 0 0.07 0.00 1 8.0760 7.1855 0.6041 100.0
In Figure 5-36, it can be observed that the SWCC for fine-grained material
loses its sigmoidal shape. This this result might represent a “dual porosity” for
soils that are highly plastic (Zhang & Chen, 2005). The first air–entry value
(i.e.0.05 kPa) might be associated with a macro–porosity while the second air–
entry value (i.e. about 1,000 kPa) might be associated with a micro–porosity of
the soil. This is reasonable given the fact that the measurements of suctions were
obtained from natural clods and not from slurries.
Another important aspect that should be considered is the initial fitting

parameters when making analysis regressions. In this work was used 10 kPa as
the initial a f parameter for plastic and non–plastic materials. This initial value for
the parameter a f is low and creates a change in the sigmoidal shape for the SWCC
when the curve is forced also to approximate the water content at 33kPa and 1500
kPa. Figure 5-37 illustrate the effect on the shape of the SWCC when the fitting is
changed by a low value of the a f parameter.
180
100
90
80
70
60
50 Soil 1
Soil 2
181
40
Soil 3
30
20
10
0
0.001 0.01 0.1 1 10 100 1000 10000 100000 1000000
Matric Suction, kPa
Figure 5-36. Examples of SWCCs Using the Model Proposed

Soil-Water Characteristic Curve
100
90 Measured SWCC - Soil 1
80
70 Predicted SWCC - Soil 1
60
50
40
30
20
10
0
0.0001 0.01 1 100 10000 1000000
Figure 5-37 Effect of a Low Initial a f Parameter Value in the SWCC
182
CHAPTER 6
SWCC MODEL PREDICTED FROM GRAIN–SIZE DISTRIBUTION
6.1 Introduction
In Chapter 5, it was proven that the Soil–Water Characteristic Curve
(SWCC) can be predicted by independently predicting of fitting parameters of the
SWCC function. Even though the prediction using this procedure yielded
acceptable results, there is a concern about discontinuities in the function due
primarily to the fact that most of the fitting parameters are independent from each
other, particularly the ones defined for fine-grained materials. In order to avoid
the uncertainty associated with the statistical process, a new model for the Soil–
Water Characteristic Curve based on the entire Grain–Size Distribution (GSD)
function is presented in this chapter. This new model is founded on two equations,
the SWCC function given by Fredlund & Xing (1994) and the GSD function that
can be also represented by a model as that given by Fredlund et al (2002). The
Soil–Water Characteristic Curve is primary associated with two concepts: The
Pore–Size Distribution and the Capillary Pressure. The soil is considered as a
group of spherical pores interconnected and therefore, the pore–size distribution is
directly related to the grain–size distribution. The Capillary Pressure is the tension
into the soil–water that allows the water to flow upward from the static
groundwater table. This pressure is commonly called Matric Suction (u a – u w ),
where u a is the pore–air pressure and u w is the pore–water pressure. In this work,
183
the retention characteristic of the soil is assumed to be primarily related to matric
suction, and therefore, the osmotic suction is considered to have a negligible
effect.
The objective of this chapter is to present a method to correlate the Soil–
Water Characteristic Curve with the Grain–Size Distribution. In this process,
some assumptions are considered:
• The solids of the soil have spherical shape (soil–texture is not
considered).
• Stress history is not considered (test were performed at zero
overburden pressure).
• The hysteresis effect is not considered.
• It is assumed that each suction value is associated with one pore–
size or grain–size.
The last assumption is perhaps the most important. The Soil–Water
Characteristic Curve equation proposed by Fredlund & Xing, 1994, is
directly linked to the pore–size distribution; the function is a sigmoid,
which is the same shape followed by the Grain–Size Distribution. The
GSD model proposed by Fredlund et al., 1997 is based on models
developed by Wagner & Ding, 1994, which are based in a modified
lognormal distribution. When the distribution is plotted, in cumulative
184
way, it presents also a sigmoidal shape. This principle allows relating both
equations.
The soil–suction and its corresponding water content should be fit
properly to the SWCC equation. In order to fit the suction–moisture pair of values
to obtain the SWCC, the least squared error were minimized. The objective
function to minimize, in terms of degree of saturation is as follows:
ObjectiveFunction = Σ[S meas − S pred ] ....................................................(6-1)

2
Where:
S meas = Measured Degree of Saturation
S pred = Predicted Degree of Saturation
For the case of the GSD equation, the best function is obtained in the same
way. That is the least squared error minimized by comparing the percent passing
measured versus the predicted values:
ObjectiveFunction = Σ[% Passmeas − % Pass pred ] ..................................(6-2)

2
Where:
%Pass meas = Measured percent of passing
%Pass pred = Predicted percent of passing
185
Since both functions have a range between 0% and 100%, that allows a
one–to–one comparison by normalization
The following is an overview of the procedure follow to estimate the
SWCC function based on the grain–size distribution:
1. The fitting parameters for the SWCC function were obtained by
minimizing the squared errors between measures versus predicted degree of
saturation values. The Fredlund & Xing, 1994 equation was used in this process.
2. The fitting parameters for the grain–size distribution function were
obtained by minimizing the squared errors between measured versus predicted %
Passing values.
3. For values of % Passing that corresponded to the same value of degree
of saturation, the corresponding particle size and suction values were obtained and
compared.
4. Correlations between the particle size and suctions values were
developed based on other simple index properties such as the wPI factor.
The Weighted Plasticity Index usually abbreviated wPI is:
P200 × PI
wPI = ....................................................................................(6-3)
100
Where:
P 200 = Material Passing # 200 US Standard Sieve, expressed in %
186
PI = Plasticity Index, expressed in %
This process was followed for all the 33,210 soils available in database.
The soil properties collected in database are presented in Table 6-1 for Meegernot
soil.
187
Table 6-1 Soil Properties for the Meegernot Soil
Soil Property Unit Item

Component Name - Meegernot
AASHTO Classification - A-1-a
AASHTO Group Index - 0
Unified Soil Classification System - GP
Top Depth of Layer cm 147
Bottom Depth of Layer cm 168
Thickness of the Layer cm 21
Passing # 4 % 20
Passing # 10 % 15
Passing # 40 % 12.5
Passing # 200 % 5
Passing Sieve 0.002 mm % 7.5
Liquid Limit % 22.5
Plasticity Index % 2.5
Weighted Plasticity Index % 0.125
Specific Gravity g/cm3 n/a
Saturated Hydraulic Conductivity (Ksat) mm/s 91.7432
Volumetric Content of Soil Water Retained at
a Tension of 1/10 bar % 5.1
a Tension of 1/3 bar % 3.7
a Tension of 15 bars % 1.3
Elevation m 2469
6.2 Calculating Suctions from the Soil–Water Characteristic Curve
This step consisted in finding, for every soil in the database, the suctions
values corresponding to several degrees of saturation (5%, 10%, 15%, and so on
100%). In order to achieve this, the measured data was fitted to the Fredlund and
188
Xing equation, and the SWCC parameters a f , b f , c f , and hr were obtained. That
allowed for the development of the entire SWCC function
The process to calculate the suction values required using either the Solver
or Goal Seek utilities available in Excel®. These utilities were necessary because
the suction is the independent variable into the Fredlund & Xing’s equation (6-4),
and in order to mathematically solve for matric suction as a function of the degree
of saturation is really quite complicated:
 
 
  ψ   
 ln1 +  
θw   hr  1 
S (%) = = 1−   ............................(6-4)
θs   1,000,000     bf
 
cf

 ln1 +    ψ 
  hr   ln e +  a 

 
   f   
   
Where:
S(%) = Degree of Saturation, in percentage
ψ = Matric Suction, in kPa
a f , b f , c f , h r = SWCC Fitting Parameters, a f and h r in kPa
θ w = Volumetric Water Content
θ s = Saturated Volumetric Water Content
Equation (6-4) represents a sigmoidal model as shown in Figure 6-1:
189
100
90
SWCC Parameters:
80 af = 9.6230
bf = 1.1678
70 cf = 1.6400
ψr = 500
60
50
40 Points:
S = 20%, ψ = 33.3 kPa
30
S = 4%, ψ = 1,500 kPa
20
10
0
0.0001 0.01 1 100 10000 1000000
Matric Suction ψ, (kPa)
Figure 6-1. Soil–Water Characteristic Curve
The procedure followed to find the SWCC parameters was explained in
detail in Chapter 3 under section 3.3.1. The spreadsheet shown in Figure 6-2 was
used to estimate first the SWCC parameters and then to calculate the data shown
in Table 6-2 by using Solver or Goal Seek Functions in Excel®.
190
Suction Vol. w/c Sat vol w/c Degree Saturation Graph
(Bar) (kPa) (%) (%) (%) Suction, kPa Saturation, %
0.1 10 13
0.33 33.3 3.7 28.5 0.0001 100.00
15 1500.0 1.3 10.0 0.001 100.00
0.01 100.00
SWCC Parameters initial SWCC 0.016 100.00 Graph
af 9.0869 10 Objective Function 0.027 100.00 Suction, kPa Saturation, %
bf 5.6921 1 2.67671E-11 0.045 100.00
cf 0.6236 2 0.074 100.00 0.00 100.00
hr 500.0 500 0.122 100.00 7.04 95.00
0.201 99.99 8.14 90.00
ψ θ Smeasured Spredicted Constraints 0.331 99.99 8.96 85.00
Suction (kPa) Vol. Water Content 0.546 99.99 9.68 80.00
0.0001 13.0 100.0 100.0 0.000 0.900 99.98 10.38 75.00
33.3333 3.7 28.5 28.5 0.000 1.484 99.96 11.09 70.00
1,500.0000 1.3 10.0 10.0 0.000 2.447 99.92 11.86 65.00
4.034 99.67 12.74 60.00
6.651 96.24 13.77 55.00
Soil-Water Characteristic Curve 10.966 70.88 15.07 50.00
18.080 42.13 16.79 45.00
100 29.810 30.12 19.26 40.00
191
49.148 24.09 23.14 35.00
81.031 20.34 30.04 30.00
80 133.5973 17.68 44.70 25.00
220.2647 15.62 85.66 20.00
363.155 13.91 261.55 15.00
60 598.7414 12.41 1,500.08 10.00
987.1577 11.05 18,158.64 5.00
1627.548 9.80
40 2683.373 8.65
4424.134 7.58
7294.164 6.60
20
12026.04 5.69
19827.59 4.86
0 32690.17 4.08
0.0001 0.01 1 100 10000 1000000 53896.98 3.36
Figure 6-2 Spreadsheet for Calculating the SWCC Parameters

88861.1 2.69
Matric Suction (kPa) 146507.2 2.07
241549.5 1.48
Degree of Saturation (%
398247.8 0.93
656599.7 0.41
1000000 0.00
Table 6-2 Calculating Suction Values from the Degree of Saturation
Degree of Saturation, % Suction, kPa
100 0.00
95 7.05
90 8.15
85 8.97
80 9.69
75 10.39
70 11.10
65 11.87
60 12.74
55 13.77
50 15.07
45 16.79
40 19.26
35 23.14
30 30.04
25 44.70
20 85.66
15 261.55
10 1,500.08
5 18,158.64
6.3 Calculating Particle Diameter from the Grain–Size Distribution
In order to calculate the particle diameter at each percent passing from the
grain–size distribution, a model presented by Fredlund et al., 2002. In their article:
“Use of the Grain–Size Distribution for Estimation of the Soil–Water
Characteristic Curve”. This model is shown in equation 6-5:
192
   
7

  ln1 + D 
r
 
   D  
Pp (d ) =
1
gm 1 −    
............................(6-5)
  ga   D
gn
ln exp(1) +      ln1 + r
 
 D      Dm   
  
Where:
P p (D) = percent passing a particular grain–size D
g a = fitting parameter related to the initial break point in the grain–size
curve
g n = fitting parameter related to the steepest slope of grain–size curve
g m = fitting parameter related to the curvature (fine section) of grain–
size curve
D = particle diameter (mm)
D r = residual particle diameter (mm), related to the fines
D m = minimum allowable particle diameter (mm)
The last part of the equation corresponds to a correction factor which
adjusts the extremes of the model properly.
In order to define the GSD function for each soil of the database, it was
necessary to estimate the fitting parameters. This process was explained in detail
in Chapter 3 section 3.3.2.
193
Particle Size Passing
# (mm) (%)
4 4.750 20.0
10 2.000 15.0
40 0.425 12.5
200 0.075 5.0
GSD parameters initial Graph Table

ga 62.8595 1 Objective Function Particle Size (mm) Passing (%) Particle Size (mm) Passing (%)
gn 1.7000 0.5 11.46316221
gm 1.0716 0.5 0.0001 3.49 36,923.075 100.00
0.001 4.32 202.717 95.00
Dmeasured Pmeas Ppred Constraints 0.01 5.54 128.664 90.00
Diameter (mm) Sieve # % Passing % Passing 0.075 7.34 96.358 85.00
1,000 100.0 99.6 0.000 0.1 7.69 77.011 80.00
4.750 4 20.0 20.3 0.000 0.425 10.10 63.597 75.00
2.000 10 15.0 15.0 0.000 1 12.35 53.473 70.00
0.425 40 12.5 10.1 0.000 2 15.01 45.377 65.00
0.075 200 5.0 7.3 0.000 4.75 20.32 38.626 60.00
194
10 28.39 32.815 55.00
100 85.73 27.682 50.00
Grain - Size Distribution Curve 1000 99.64 23.051 45.00
100 18.799 40.00
14.844 35.00
80 11.139 30.00
7.685 25.00
4.562 20.00
60 1.995 15.00
0.405 10.00
40 0.004 5.00
20
Figure 6-3 Spreadsheet Used to Find the GSD Fitting Parameters

0
0.001 0.01 0.1 1 10 100 1000 10000
Percent Passing (%)

The spreadsheet shown in Figure 6-3 was used to estimate the GSD
parameters g a , g n ,g m and consequently to calculate the values shown in Table 6-3
by using the Goal Seek function in Microsoft Excel®. Two parameters needed to
be assumed in the process: the residual particle diameter (Dr) and the minimum
allowable particle diameter (Dm). The values assumed were 0.001 mm and
0.00001 mm, respectively.
Table 6-3 Calculated Particle Diameter from the Percent Passing Values
Passing (%) Particle Size, D (mm)
100 36,923.136
95 202.718
90 128.664
85 96.358
80 77.011
75 63.597
70 53.473
65 45.376
60 38.626
55 32.816
50 27.682
45 23.051
40 18.799
35 14.843
30 11.139
25 7.685
20 4.562
15 1.995
10 0.405
5 0.004
Once the suction (kPa) was obtained from the degree of saturation (%)
based on the SWCC, and the particle size (mm) obtained from the percentage
195
passing (%) base on the GSD; the following step consisted in relating the suction
values with the particle diameter values. Figure 6-4 shows the relationship
between these values for one specific soil. The values are summarized in Table 6-
4.
It is important to note that the relationship between particle diameter and
suction values was possible because the SWCC was obtained in terms of degree
of saturation. In this case, the degree of saturation, which ranges between 0% and
100% can be normalized or scale to the same range of variation of the % passing
in the grain–size distribution curve. The same process used to obtain the
relationship shown in Figure 6-4 for one soil was used to obtain the same
relationship for every soil available in the database. This analysis was possible
due to the creation of a program Macro in Excel®. Note that there is a clear
connection between the particle diameter and the suction value due to the
relationship between the grain–size distribution and the pore–size distribution
(PSD) of the soil.
196
Table 6-4 Relationship between Suction Values versus Particle Size
Particle Size, D (mm) Suction, ψ (kPa)

36,923.14 0.00
202.72 7.05
128.66 8.15
96.36 8.97
77.01 9.69
63.60 10.39
53.47 11.10
45.38 11.87
38.63 12.74
32.82 13.77
27.68 15.07
23.05 16.79
18.80 19.26
14.84 23.14
11.14 30.04
7.68 44.70
4.56 85.66
1.99 261.55
0.41 1,500.08
0.00 18,158.64
197
5
4.5
3.5
)ψ
2.5
198
2
1.5
y = 0.0917x3 - 0.0216x2 - 1.2206x + 2.7158
R² = 0.9986
1
Log Suction (
0.5
Figure 6-4 Log Suction versus Log Particle Size for One Soil
-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5
Log Particle Diameter (D)
6.4 Ranges of wPI and Statistical Information
The database used in developing this model consisted of more than 660,000 soils
as shown in Table 6-5. Given the large amount of data available it was deemed
necessary to subdivide it and grouped the soils based on a soil property
representative of the moisture retention characteristic of the soil. The weighted
plasticity index (wPI) was proven to be significantly related to the SWCC
(Zapata, 1999) and therefore it was selected as the basis for the grouping process.
Furthermore, several authors have attempted to relate the SWCC with the GSD
(Arya and Paris, 1981; Fredlund et al., 1997) without success for all ranges of soil
encountered in the field. This might be due to the variability associated with the
porous materials, and therefore, a process that group soils with relatively the same
characteristics would eliminate some of this variability.
Table 6-5 presents the data divided by wPI ranges along with the number
of soil in each range and statistic. Note that at higher values of wPI, the variability
inferred from higher values of standard deviation and variance, is high.
199
Table 6-5 Database Divided by wPI Ranges and Statistics Associated with Each
Range of Values
items in file Range Count Average Median Mode St dev Var

86,100 wPI = 0 4305 0.00 0.00 0.000 0.000 0.000
79,260 0 < wPI ≤ 1 3,963 0.67 0.69 0.750 0.221 0.049
85,680 1 < wPI ≤ 2 4,284 1.48 1.49 1.063 0.297 0.088
111,160 2 < wPI ≤ 4 5,558 2.95 3.00 3.000 0.575 0.331
126,480 4 < wPI ≤ 8 6,324 5.73 5.63 4.500 1.146 1.313
64,640 8 < wPI ≤ 12 3,232 9.88 9.75 10.500 1.170 1.370
38,520 12 < wPI ≤ 16 1,926 13.98 13.95 13.500 1.132 1.282
27,000 16 < wPI ≤ 20 1,350 17.97 18.00 18.000 1.131 1.279
34,320 20 < wPI ≤ 30 1,716 24.18 24.00 2.744 7.530 22.500
9,160 30 < wPI ≤ 40 458 33.97 33.75 2.631 6.921 34.000
1,880 40 < wPI ≤ 60 94 45.54 43.73 4.933 24.336 42.750
664,200 33,210
6.5 Calculating the Models for Each Range
Several plots like the one shown in Figure 6-4 were processed to get an
idea of the best models to fit the particle diameter versus suction data. These plots
showed that a third order polynomial model had the highest correlation as implied
from the high R2 values obtained (0.96 and higher). This process was performed
by using Excel® and Minitab® 15 software packages. Having defined the type of
model to use for the non–linear regression analysis and the ranges based on wPI
values, the following step was to find the constants that yielded the highest R2
values. Statistica® 5.5 was used for the determination of the best model. Figure 6-
5 is an example of a spreadsheet used in the estimation for only one range of wPI
values. The same process was repeated for all the selected ranges.
200
The general form of the model used to relate particle diameter and suction
values is as follows:
logψ = k1 + k 2 log D + k 3 log D 2 + k 4 log D 3 ........................................... 6-6
Where:
ψ = Suction in kPa
D = Particle Diameter in mm
k 1 , k 2 , k 3 , k 4 = Regression Constants
201
Figure 6-5 Spreadsheet from Statistica® Used in Estimating the Best Models
between Particle Size and Suction Values
A summary of the regression constants found is presented in Table 6-6 for
each range of wPI values selected.
202
Table 6-6 Summary of Fitting Parameters Found for the Correlation
between Log of Particle Size Versus log of Suction for each Range of wPI Values
wPI Ranges
2
From To n k1 avg k2 avg k3 avg k4 avg R
0 0 4305 1.05310 -0.79466 0.00209 0.00725 0.5752
0 1 3963 1.29284 -0.84537 -0.01175 0.00658 0.6904
1 2 4284 1.08684 -0.72871 -0.01610 0.00461 0.6206
2 4 5558 1.28120 -0.64835 -0.01631 0.00112 0.6429
4 8 6324 1.09260 -0.77777 -0.01294 0.00620 0.6131
8 12 3232 1.20260 -0.84218 -0.02023 0.00813 0.5882
12 16 1926 1.18353 -0.86519 -0.01715 0.00848 0.6064
16 20 1350 0.97223 -0.96512 -0.01304 0.01141 0.5934
20 30 1716 0.60121 -1.06883 -0.01141 0.01260 0.5637
30 40 458 0.52383 -1.16329 -0.03659 0.01030 0.5532
40 60 94 0.88131 -1.37328 -0.08086 0.01249 0.5029
203
6.6 Plotting the Models
The relationship between suction and particle diameter was plotted for
each range of wPI values as shown in Figure 6-6. At higher ranges of suction
(small particle sizes) the curves essentially merged. However, as the suction
decreases (large particle sizes) a dependency of the relationship on wPI can be
noted. This is particularly true for particle size greater than 1 mm. In order to
assess the dependency of this relationship on wPI, the k i avg constants obtained at a
wPI corresponding to the mid-point of each range were plotted against wPI
values. These relationships can be seen in Figure 6-7. It can be seen that
particularly constants k 1 and k 2 present high correlation with wPI. Constants k 3
and k 4 are also correlated to lower degree.
204
6
Model: log (ψ) = k1avg+ k2avg(log D) + k3avg(log D)2 + k4avg(log D)3 Ranges
Suction, ψ (kPa)
Particle Size, D (mm) wPI = 0
4
wPI 0-1
wPI 1-2
2 wPI 2-4
wPI 4-8
ψ
wPI 8-12
0
wPI 12-16
205
wPI=0 wPI 16-20
L
-2 wPI 20-30
16<wPI<20 wPI 30-40
20<wPI<30 wPI 40-60

-4
30<wPI<40
40<wPI<60
Figure 6-6 Plot of Log Suction versus Log Particle Size

-6
-5 -4 -3 -2 -1 0 1 2 3 4 5 6 7
LOG Particle Size, D
1.5
y = 5E-05x3 - 0.003x2 + 0.03x + 1.1355
R2 = 0.8735
1
k1
0.5
y = 0.0002x + 0.0056
Constant, K
R2 = 0.5905 k4
0
y = -0.0011x - 0.0044 k3
2
R = 0.7177
-0.5
-1
y = -0.0126x - 0.7285
R2 = 0.9238 k2
-1.5
0 10 20 30 40 50 60
wPI
Figure 6-7 Relationship between the Constant Values and the wPI
Based on these observations, the following equations are proposed for
constants k 1 , k 2 , k 3 and k 4 :
k1 = 0.00005 (wPI)3 - 0.003 (wPI)2 + 0.03 (wPI) + 1.1355 .................... 6-7
k2 = - 0.0126 (wPI) - 0.7285 ................................................................... 6-8
k3 = -0.0011(wPI) - 0.0044 ..................................................................... 6-9
k4 = 0.0002(wPI) + 0.0056 .................................................................... 6-10
206
Equation 6-7 yielded an R2 of 0.87 while equation 6-8 yielded an R2 of
0.92. Furthermore, equation 6-9 and equation 6-10 yielded R2 values of 0.72 and
0.59, respectively.
Substituting equation 6-7 to 6-10 in to the equation 6-6, we obtain:
log ψ = 0.00005 (wPI)3 - 0.003 (wPI)2 + 0.03 wPI + 1.1355 - (0.0126 wPI
+ 0.7285) log D - (0.0011wPI + 0.0044) log D2 + (0.0002wPI +
0.0056) log D3 ........................................................................... 6-11
This equation is valid for plastic and granular materials.
For non-plastic granular materials, the equation gets reduced to:
log ψ = 0.0056 log D3 + 0.0002wPI log D3- 0.0044 log D2
- 0.0011wPI log D2- 0.7285 log D - 0.0126 wPI log D
+ 0.00005 (wPI)3 - 0.003 (wPI)2 + 0.03 wPI + 1.1355 ............ 6-12
Figure 6-8 presents a family of curves representatively the relationship
given by equation 6-11
207
5
wPI = 60
4
wPI = 50
wPI=0
2 wPI=5
wPI=10
1 wPI=15
wPI=20
208
wPI=25
0
wPI = 0 wPI=30
lo
wPI = 5 wPI=40
-1
wPI = 10 wPI=50
wPI = 15 wPI=60
-2
wPI = 40
wPI = 30
-3
wPI = 25
wPI = 20
-4
Figure 6-8 Suction as a Function of Particle Diameter and wPI

-6 -4 -2 0 2 4 6 8
log Particle Size (mm)
SS err SS tot SS reg ealg eabs (Se) (Sy)
2 2 2
(Sm - Sp) (Sm - Sm(avg)) (Sp - Sm(avg)) 100*(Sm - Sp)/Sm - (Sm - Sp) (Sm(avg) - Sm)2
2
Σ = 3,358.71 18,047.62 9,095.74 ealg = -27.10

eabs = 143.04
Sm(avg) = 2.3567 S= 3,358.71 18,047.62
υ = 6,642 ngroup = 6,642 6,642
e= 1 p= 1 1
Se = 0.71 Se = 0.71
Sm(avg) = 2.3567
Sy = 1.17 Sy = 1.65
Se/Sy = 0.61 Se/Sy = 0.43
R2 = 0.8139 R2 = 0.8139
Figure 6-9 Error Analysis
5
Log Predicted Suction
1 n = 664,200
Se/Sy = 0.39
0 R2 = 0.846
-1
-1 0 1 2 3 4 5 6 7
Log Measured Suction
Figure 6-10 Comparison of Measured versus Predicted Suction
209
6.7 Assessment of the Model for Fine–Grained Materials
The families of curves for granular and fine–grained soils by using the
approach indicated in this Chapter 6 are shown in Figure 6-11 and 6-12.
100
90
80 Soil 1
Degree of Saturation, S(%)
70
Soil 2
60
50 Soil 3
40
30
20
10
0
0.0001 0.001 0.01 0.1 1 10 100 1000 10000 1000001000000
Matric Suction, ψ (kPa)
Figure 6-11 Family of Curves for Granular Materials
It is clear that the model doesn’t perform well for suction values greater
than about 10,000 kPa when the material is fine–grained, mostly. The reason for
this is the fact that the particle diameter on the lower part is not accurately
determined. Equation 6-5 that defines the grain–size distribution makes use of two
parameters: D r and D m which correspond to the residual particle diameter and the
minimum allowable particle diameter respectively.
210
100
90 wPI = 24.0
Degree of Saturation, S(%) 80 wPI = 8.4 wPI = 10.5
70
wPI = 5.0
60
wPI = 1.0
50
40
30
20
10
0
0.0001 0.001 0.01 0.1 1 10 100 1000 10000 1000001000000
Figure 6-12 Family of Curves for Fine–Grained Materials
The assumed values for these two parameters were 0.001 mm and 0.00001
mm, respectively. These parameters appear to be underestimated for the fine–
grained materials and they should be probably not constant values but rather a
function of the soil plasticity.
In order to check if indeed these parameters were affecting the prediction
of the SWCC, an example that included 5 different soils is presented below. First,
the GSD curve was fitted to the measured values and a value of 10-14 mm for D m
was used this time. A new relationship between particle diameters and suctions
was obtained as shown in Figure 6-13. The equation that best represents this
relationship is:
211
 6.774 
logψ = 5.797 −  −0.634 log D 
....................................................... 6-13
 0.83 + e 
5
New Model:
4
Log Particle Size
0
-14 -12 -10 -8 -6 -4 -2 0 2
Log Suction
Figure 6-13 Example Relationship Log Suction vs Log Particle Size
Just as an example, Equation 6-13 used in the example before. These soils
were used to estimate the SWCC for the five soils with wPI values ranging from 1
to 24. The grain–size distribution is presented in Figure 6-14. Based on the GSD,
the SWCC presented in Figure 6-14 were obtained. Note how the prediction
improves dramatically.
It is then recommended to perform the analysis for all the soils in the
database to obtain a more reliable model.
212
100
90
80
wPI = 24.0
70
Percent of Passing
60 wPI = 10.5
50 wPI = 8.4
40
30 wPI = 5.0
20 wPI = 1.0
10
0 #200 #40 #4
#10
0.0001 0.001 0.01 0.1 1 10 100 1000
Figure 6-14. Grain–Size Distribution Example
100
wPI = 24.0
90
wPI = 10.5
80
Degree of Saturation, S(%)
wPI = 1.0
70
wPI = 5.0
60
50 wPI = 8.4
40
30
20
10
0
0.0001 0.001 0.01 0.1 1 10 100 1000 10000 1000001000000
Figure 6-15. Soil–Water Characteristic Curve Example
213
6.8 Implementing the SWCC model based on GSD
The Procedure to estimate the SWCC based on the entire GSD is
summarized as follows:
Step 1: Obtain the grain–size distribution of the soil and Atterberg limits.
Step 2: Calculate the Weighted Plasticity Index, wPI.
Step 3: Estimate the particle diameter, D values for different values of %
passing: 5%, 10%, 15%, 20%, 25%, 30%… 95%, 100%. These values can be
obtained by fitting the gradation data to the sigmoid function presented by
Fredlund et al., 2002 (equation 6-5) and then using the Goal Seek function to
solve for D. alternatively, the particle diameter values can be found by simply
reading them off the GSC graph.
Step 4: Estimate the suction values by using equation 6-11 for the same
degree of saturation equivalent to the % passing for which the particle diameter D
was calculated in step 3.
Step 5: Plot the suction values versus degree of saturation pair of values
found in step 4 and fit the SWCC Fredlund & Xing Function by using a non-
linear regression package such as Solver in Excel®.
Figure 6-16 illustrates the procedure described to estimate the SWCC from
the Grain–Size Distribution
214
Step 1, Input data:
Atterberg Limits and Gradation
Step 2: Estimate wPI
Step 3: Estimate the Particle Diameter at different

percent of Passing (10%, 20%, 30%, etc.)
Step 4: Using the model proposed (eq. 6-11),

estimate Suctions for different % of Degrees of
Saturation equivalent to the same % passing
used in step 3
Step 5: Plot Suction versus Degree of

Saturation par of values and fit to the
SWCC Function
Figure 6-16. Approach to Estimate the SWCC based on the Grain–Size
Distribution.
6.9 Summary
An approach to estimate the SWCC function based on the grain–size
distribution of the soil has been proposed in Chapter 6. This approach relied on
33,210 soil–water characteristic curves that corresponded to 664,200 data suction
points measured by the Natural Resources Conservation Service. The procedure
215
described in Figure 6-1 explains the process to estimate the SWCC from the
Atterberg limits and the plasticity index. The model proposed in Equation 6-11 is
used to estimate the suction values for the same degree of saturation equivalent to
the % passing for which the particle diameter D is defined. This model expresses
the suction as a function of the wPI and the particle diameter, D.
The advantages of the proposed model include:
• The model makes use of the complete grain–size distribution data.
• The model produce a continuous SWCC function as opposed to the
method proposed in Chapter 5.
• The model is based on the physical concept that relates the grain–
size distribution to the pore-size distribution of the soil.
• The model is based on simple and routinely measured soil index
properties such as gradation and Atterberg limits.
• Its implementation is straight forward.
A disadvantage of the model lies on the fact that the database provided
only 2 to 3 measured data points and therefore, in order to fit a complete SWCC
function, the author had to rely on the extremes of the function. That is, two extra
data points were included in the fitting process: 100% saturation was assigned to a
very low suction value and 0% saturation was assumed to occur at 1,000,000 kPa.
216
(Figures 4-25 and 4-26) with the model proposed in this work (Figure 6-10), it
can be seen that the R2 improved from 0.70 to 0.81; while the R2 for non-plastic
soils improve from 0.40 to 0.81. On the other hand, when comparing the model
proposed in this Chapter 6 (Figure 6-10) with the MEPDG model (Figure 4-27
and 4-28), it can be seen that the R2 improved greatly from 0.49 to 0.81 for plastic
soils. For non–plastic soils, the R2 values were somewhat similar.
The new models proposed in this Chapter 6 based on the entire GSD did
not perform well for soils with high plasticity. New equations for k 1 , k 2 , k 3 and
k 4 parameters should be found. The procedure used in Chapter 6 looks promising
and can be used to repeat the analysis once all the problems have been
recognized.
217
CHAPTER 7
SIGNIFICANCE OF THE STUDY
7.1 Conclusions
This thesis work proposes a new set of models for the prediction of the
SWCC fitting parameters based on the equation given by Fredlund and Xing in
1994. These models were estimated by following two different approaches. The
first approach, explained in detail in Chapter 5 was based on a statistical
regression analysis from values of matric suction and water content data points
found in a large database maintained by the National Resources Conservation
Service (NRCS). The second approach, presented in Chapter 6, proposes a model
based on the Grain–Size Distribution, which is based on the relationship between
the GSD and the Pore–Size Distribution (PSD) of the soil. Both methods have
practical application. The first method can be directly implemented in the
Enhanced Integrated Climatic Model (EICM), which is the model that
incorporates environmental effects in the estimation of the resilient modulus of
unbound materials. The EICM is an important component of the new
Mechanistic–Empirical Pavement Design Guide (MEPDG).
A summary of the conclusions from each chapter is presented below.
218
7.1.1 Conclusions Chapter 3 – Database Collection.
The database collection was a very important task for the development of
the work presented in this thesis. The vast amount of data points collected
contained a total of 36,394 different soils, with 4,518 items corresponding to non–
plastic materials and 31,876 plastic soils. The database was collected by the
National Conservation Resources Service for agricultural purposes and contains
chemical, physical and engineering soil properties which can be used in a number
of disciplines. The soils properties were obtained from studies developed during
many years through the continental US, Alaska, Hawaii, and Puerto Rico. The
database allowed for the estimation of parameters such as the wPI factor, Group
Index, the Soil–Water Characteristic Curve fitting parameters and the Grain–Size
Distribution fitting parameters.
Most of the properties were obtained directly from the laboratory or from
field testing while other properties were estimated from correlations or
estimations. Both sets of data or properties had some degree of uncertainty related
to them. The uncertainty of the data can be attributed to several factors: First, the
database was developed by collecting tests during a range of years (USDA–NRCS
was established in 1935); second, uncertainty associated with environmental
conditions and soil nature (samples were located all over the US territory); third,
the tests were performed by following protocols and standards which are being
constantly updated; and last, technological changes and advances in the field
219
allowed for new data interpretations during more than 70 years the data has been
collected.
In order to eliminate the variability encountered in the data, a moving
average technique was employed, whereas the data was organized or sorted
according to the geotechnical factor (predictor) that most affected the predicting
variable. This process is commonly used when the database presents high
variability in order to find the general trend of behavior (Graham, 1993).
It is important to emphasize that the vast database collected and presented
as part of this thesis work was drawn directly from laboratory testing. It is
perhaps the largest database of soil moisture retention curves available in the
world. These facts allowed for optimal models to estimate the Soil–Water
Characteristic Curve, as those presented in this work.
7.1.2 Conclusions Chapter 4 – Validating Existing Models.
Based on the database collected and described in detail in Chapter 3, the
validation of two existing models to estimate the SWCC was possible. The
analysis is presented in Chapter 4. The models corresponded to those proposed
by Zapata, 1999, and the MEPDG model (Witczak et al., 2006). These validations
were statistically calculated separately for fine–grained plastic soils and for
granular non–plastic materials.
Table 4-1 shows the errors found for the validation of Zapata’s models for
plastic and non-plastic soils. The validation was performed at different suction
220
levels: 1, 10,100, 1,000 and 10,000 kPa. For non-plastic soils, the R2 values
ranged between 68% and 82%. Relatively good predicted water contents were
found for suction values higher than 100 kPa. For plastic soils, the highest R2
(82%) was found at suction values lower than 1 kPa and relatively acceptable R2
(60%) was found for suction values higher than 1,000 kPa.
Figure 4-25 and 4-26 show the measured versus predicted volumetric
water content values obtained by using the model proposed by Zapata, 1999, for
plastic and non-plastic soils, respectively. These figures include all the predicted
water contents estimated at suctions of 1, 10, 100, 1,000 and 10,000 kPa. For
plastic soils, the model developed by Zapata, 1999, although it presented an
overall R2 of 0.70, it was found to be biased towards overprediction for most of
the data points. For non-plastic soils, the Zapata’s model presents a different
behavior, in which most of the data points were underpredicted and yielded a low
overall R2 value of 0.40.
In general, the models proposed by Zapata, 1999, present acceptable errors
considering that it was developed 10 years ago with few data points, when
compared to the vast database used in this project.
Tables 4-2 and 4-3 show the error analysis performed for the MEPDG
models for non-plastic and plastic soils, respectively. For non-plastic soils, an R2
value of 60%, which was considered to be acceptable, was found only for suctions
221
values lower than 1 kPa. Similarly, for plastic soils, the highest R2 value (83%)
was found for suction values lower than 1 kPa.
Figures 4-27 and 4-28 show the measured versus predicted volumetric
water content values obtained by using the MEPDG model for plastic and non-
plastic soils, respectively. These figures include all the predicted water contents
obtained at 1, 10, 100, 1,000 and 10,000 kPa of suction. It was observed that for
plastic soils, the volumetric water content was consistently overestimated and
yielded an R2 of 0.49. However, for non-plastic soils, the MEPDG model
presented an acceptable prediction of volumetric water content with an R2 value
equal to 0.91.
In general, the MEPDG models presented acceptable estimations
considering the amount of data analyzed. The MEPDG model can be considered
to be a better model for non-plastic soils, while the model proposed by Zapata,
1999 can be considered to perform better for fine-grained materials.
7.1.3 Conclusions Chapter 5 – Approach 1 to Predict the SWCC.
Tables 5-8 and 5-13 present the proposed models for the SWCC
parameters for the Fredlund and Xing equation, for plastic and non-plastic soils,
respectively. The models proposed for plastic soils were estimated in function of
the Group Index, which is in turn a function of passing sieve #200, liquid limit
and plasticity index. On the other hand, the models proposed for non-plastic soils
were estimated as function of the particle diameter D 10 .
222
The models proposed in Chapter 5 to estimate the SWCC parameters,
present the following advantages:
• The models proposed can be implemented in the Enhanced
Integrated Climatic Model (EICM), which is incorporated in the
Mechanistic–Empirical Pavement Design Guide (MEPDG). The
format required allows including the SWCC fitting parameters.
• The database is vast. With more than 31,000 data points for plastic
soils and 4,500 data points for non-plastic soils, it contains the
most important Soil Index Properties obtained directly from
laboratory testing or in the field. This database can be considered
the largest in the world containing unsaturated soil properties.
• The models proposed are very simple to be implemented. For
plastic soils, the Atterberg’s Limits and the Passing US sieve #200
are needed as input parameters; while for non-plastic materials,
only the particle diameter D 10 is needed. These are parameters
commonly used by practicing engineers and therefore, this model
becomes an excellent candidate for practical applications.
The approach and the models proposed in this chapter have the following
limitations:
223
• The models proposed for the SWCC fitting parameters were
estimated independently of each other. That makes it difficult to
control the shape of the Soil–Water Characteristic Curve.
• Due to the tremendous amount of data points in the database, a
moving average estimate of the parameters during the statistical
analysis was necessary. The fine-grained material database was
sorted according to the wPI and the data were averaged in groups
of 300 consecutive data points. Groups of 50 consecutive data
points were used for non-plastic soils, based on the D 10 parameter.
While working with the moving average allows finding a clear
tendency of the data and hence, better correlations; the variability
gets somehow masked within each range. A deeper study of
variability should be performed.
(Figures 4-25 and 4-26) with the model proposed in this work (Figures 5-30 and
5-32), it can be seen that the R2 improved from 0.70 to 0.81; while the R2
improved from 0.40 to 0.89 for non-plastic materials. Even though the R2
improved marginally for plastic soils, it can be observed that the Zapata's model is
biased towards overprediction for most of the dataset. For non–plastic soils the
results obtained with the Zapata’s model are underpredicting most of the data
points.
224
On the other hand, when comparing the model proposed in this work
(Figure 5-30) with the plastic model used in the MEPDG model (Figure 4-27), it
can be seen that the R2 improved greatly from 0.49 to 0.81. For non-plastic or
granular soils, the R2 for both models are somewhat similar. Even though the
non–plastic model from MEPDG models is almost similar than the model
proposed in this work, the later model is much simpler and easier to implement
because it only depends on one gradation parameter, the particle size at 10%
Passing or D 10 .
It can be concluded that the new models proposed in this Chapter 5 will
enhanced the prediction of the SWCC and therefore, it is recommended to
consider applying them in practical applications.
As an example of the predicted SWCC based on the models proposed in
Chapter 5, the soil index properties shown in Table 7-1 were used to calculate the
SWCC fitting parameters. The three soil–water characteristic curves obtained are
shown in Figure 7-1.
Table 7-1. Soil Index Properties for Three Soils Taken from the Database
Classification Group Predicted SWCC Parameters

Soil USCS AASHTO P200 LL PI D10 wPI Index af bf cf hr
1 CH A-7-6 85 65 40 - 34.00 37 0.0709 2.1570 0.0849 1,102.1
2 CL A-6 65 40 25.5 - 16.58 14 1.8436 1.0489 0.6047 691.5
3 SP-SM A-1-a 5 0 0 0.07 0.00 1 8.0760 7.1855 0.6041 100.0
225
100
90
80
70
60
50 Soil 1
Soil 2
226
40
Soil 3
30
20
10
0
0.001 0.01 0.1 1 10 100 1000 10000 100000 1000000
Matric Suction, kPa
Figure 7-1. Examples of SWCCs Using the Model Proposed

In Figure 7-1, it can be observed that the SWCC for fine-grained material
loses its sigmoidal shape. This result might represent a “dual porosity” for soils
that are highly plastic (Zhang & Chen, 2005). The first air–entry value (i.e.0.05
kPa) might be associated with a macro–porosity while the second air–entry value
(i.e. about 1,000 kPa) might be associated with a micro–porosity of the soil. This
is reasonable given the fact that the measurements of suctions were obtained from
natural clods and not from slurries.
7.1.4 Conclusions Chapter 6 – Approach 2 to Predict the SWCC.
An approach to estimate the SWCC function based on the grain–size
distribution of the soil has been proposed in Chapter 6. This approach relied on
33,210 soil–water characteristic curves that corresponded to 664,200 data suction
points measured by the Natural Resources Conservation Service. The procedure
described in Figure 6-1 explains the process to estimate the SWCC from the
Atterberg limits and the plasticity index. The model proposed in Equation 6-11 is
used to estimate the suction values for the same degree of saturation equivalent to
the % passing for which the particle diameter D is defined. This model expresses
the suction as a function of the wPI and the particle diameter, D.
The advantages of the proposed model include:
• The model makes use of the complete grain–size distribution data.
• The model produce a continuous SWCC function as opposed to the
method proposed in Chapter 5.
227
• The model is based on the physical concept that relates the grain–
size distribution to the pore-size distribution of the soil.
• The model is based on simple and routinely measured soil index
properties such as gradation and Atterberg limits.
• Its implementation is straight forward.
A disadvantage of the model lies on the fact that the database provided
only 2 to 3 measured data points and therefore, in order to fit a complete SWCC
function, the author had to rely on the extremes of the function. That is, two extra
data points were included in the fitting process: 100% saturation was assigned to a
very low suction value and 0% saturation was assumed to occur at 1,000,000 kPa.
(Figures 4-25 and 4-26) with the model proposed in this work (Figure 6-10), it
can be seen that the R2 improved from 0.70 to 0.81; while the R2 for non-plastic
soils improve from 0.40 to 0.81. On the other hand, when comparing the model
proposed in this Chapter 6 (Figure 6-10) with the MEPDG model (Figure 4-27
and 4-28), it can be seen that the R2 improved greatly from 0.49 to 0.81 for plastic
soils. For non-plastic soils, the R2 values were somewhat similar.
The new models proposed in this Chapter 6 based on the entire GSD did
not perform well for soils with high plasticity. New equations for k 1 , k 2 , k 3 and
k 4 parameters should be found. The procedure used in Chapter 6 looks promising
228
and can be used to repeat the analysis once all the problems have been
recognized.
Similarly to Chapter 5, it can be concluded that the new models proposed
in this Chapter 6 will enhanced the prediction of the SWCC and therefore, it is
also recommended to consider applying them in practical applications.
7.2 Application
One of the most important applications for the models proposed in this
thesis work is in pavement design; where the unsaturated soil mechanics plays an
important role in the performance of the pavement structure, mainly on the
resistance and deformation of the soil. These characteristics of the soil are mainly
due to variations in matric suction, which could take place due to changes on
external conditions especially due to presence of water, changes in temperature,
depth of the ground water table, external loads, etc. This relationship between the
matric suction and the amount of water into the soil has been considered in the
Enhanced Integrated Climatic Model (EICM) as part of the Mechanistic–
Empirical Pavement Design Guide (MEPDG).
7.3 Recommendations for Further Research
This work was focused on soil properties which are affecting the SWCC
such as the volumetric water content for suctions of 10, 33 and 1,500 kPa, the
percentage of passing sieve # 200, and the Atterberg’s limits. The database
229
initially considered in this thesis has more physical and engineering properties
which could be used to establish more estimations or correlations.
Many studies relate the soil–water characteristic curve with other soil–
properties to obtain unsaturated soil property functions. The SWCC is related with
the grain–size distribution as applied in Chapter 6. Furthermore, the SWCC is
related with shear strength parameters to predict shear strength functions; and it is
also related to other hydraulic properties to obtain the water seepage constitutive
function. The database contains enough information to attempt correlations that
include the SWCC, the saturated hydraulic conductivity and other properties in
order to estimate the unsaturated hydraulic conductivity function.
In summary, unsaturated soil mechanics allows relating constitutive
relationships (water seepage, air flow, heat flow, shear strength and volume–mass
change), and compaction properties with the soil–water characteristic curve and
this database could be used to obtain other estimations.
230
RERERENCES
ASTM Standard D 5298–10, 2010, “Standard Test Method for Measurement of

Soil Potential (Suction) Using Filter Paper,” ASTM International, West
Conshohocken, PA, 2010, DOI: 10.1520/D5298–10, www.astm.org.
Arya, L.M., & Paris J.F. (1981) “A Physicoempirical Model to Predict the Soil
Moisture Characteristic from the Particle–Size Distribution and Bulk
Density data, Soil Science Soc. American Journal, Vol. 45, pp. 1023–1030
Bishop, A. W., 1959, “The principle of effective stress”, Norwegian Geotechnical

Institute, Vol. 39, pp. 859–863.
Darter, M.I., Mallela, J., Titus-Glover, L., Rao, C., Larson, G., Gotlig, A., Von
Quintus, H., Khazanovich, L., Witczak, M., El-Basyouny, M., El-Badawy,
S., Zborowski, A. and Zapata, C. (2006). Research Results Digest 308:
Changes to the Mechanistic-Empirical Pavement Design Guide Software
through Version 0.900, National Cooperative Highway Research Program,
Transportation Research Board of the National Academies, September
2006. 22 pages.
Feng, G., Sharratt, B., Vaughan, J., & Lamb, B. (2009) “A Multiscale Database of
Soil Properties for Regional Environmental Quality Modeling in the
Western United States”, Journal of Soil and Water Conservation, Vol. 64,
No. 6, December 2009, pp 363–373.
Fredlund, D.G. (2006). “Unsaturated Soil Mechanics in Engineering Practice”,

Journal of Geotechnical and Geoenvironmental Engineering, Vol. 132,
No. 3, pp. 286–321.
Fredlund, D.G., & Morgenstern, N.R. (1976) “Constitutive relations for volume
change in unsaturated soils”, Canadian Geotechnical Journal, V. 13, No. 3,
pp. 261–276.
Fredlund, D.G., & Morgenstern, N.R. (1977) “Stress State Variables for
Unsaturated Soils”, Journal of Geotechnical and Geoenvironmental
Engineering, Vol. 130, No. GT5, May 1977, pp. 447–466.
Fredlund, D.G. (l979), “Second Canadian Geotechnical Colloquium: Appropiate

Concepts and Technology for Unsaturated Soils”, Canadian Geotechnical
Journal, Vol. 169, No. 1, pp. 121–139.
Fredlund, D.G., & Rahardjo, H. (1993). Soil Mechanics for Unsaturated Soils.
John Wiley & Sons, Inc., New York, NY.
231
Fredlund, D.G., & Xing, A. (l994), “Equations for the Soil–Water Characteristic
Curve”, Canadian Geotechnical Journal, Vol. 31, No. 3, pp. 521–532.
Fredlund, D.G. (1995). “The Prediction of Unsaturated Soil Functions Using the
Soil–water Characteristic Curve. In Proceedings of the Bengt Broms
Symposium in Geotechnical Engineering”, 13–16 Dec., 1995, Singapore,
World Scientific, River Edge, NJ, pp. 113–133.
Fredlund, D.G. (1981). “Seepage in Saturated Soil”, 13–16 Dec., 1995,

Singapore, World Scientific, River Edge, NJ, pp. 113–133.
Fredlund, M.D., Fredlund, D.G., & Houston, S.L., Houston, W. (2003).

“Assessment of Unsaturated Soil Properties for Seepage Modeling through
Tailings and Mine Wates”, Proceedings of the Tenth International
Conference on “Tailings and Mine Wastes”, Vail, Colorado, U.S., pp. 149
– 157, Oct. 12-15
Fredlund, M.D., Fredlund, D.G., & Wilson, G.W. (1997). “Prediction of the Soil–
Water Characteristic Curve from Grain–Size Distribution and Volume–
Mass Properties”, Third Brazilian Symposium on Unsaturated Soils, Rio
de Janeiro, Brazil, 1977
Fredlund, M.D., Fredlund, D.G., & Wilson, G.W. (2000). “An Equation to
Represent Grain–Size Distribution”, Canadian Geotechnical Journal, Vol.
37, No. 4, pp. 817–827
Fredlund, D.G., Stone, J., Stianson, J., Sedgwick, A., (2011) “Obtaining
Unsaturated Soil Properties for High Volume Change Materials”, Golder
Associated Ltd. & Total E.P. Canada
Fredlund, M.D., Wilson, G.W., & Fredlund, D.G. (2002). “Use Of The Grain–
Size Distribution For Estimation Of The Soil–Water Characteristic
Curve”, Canadian Geotechnical Journal, Vol. 39, No. 5, pp. 1103–1117
Gitirana Jr. G., & Fredlund, D.G. (2004). “Soil–Water Characteristic Curve
Equation with Independent Properties”, Journal of Geotechnical and
Geoenvironmental Engineering, Vol. 130, No. 2, pp. 209–212
Graham, R.C. (1993) “Data Analysis for the Chemical Science: A guide to
statistical techniques”, VCH Publishers Inc., New York, NY
Gupta, S.C., & Larson, W.E. (1979). “Estimation Soil–water Retention

Characteristics from Particle–size Distribution, Organic Matter Percent,
232
and Bulk Density”, Water Resources Research Journal, Vol. 15, No. 6, pp.
1633–1635
Hazen, A. (l911). “Discussion of Dams on Sand Foundation”, A.C. Koenig

Transactions, ASCE, Vol. 73, pp. 199–203.
Hines, W.W., & Montgomery, D.C. (1990). Probability and Statistic in

Engineering and Management Science, John Wiley & Sons, Inc., Canada,
732 p
Houston, W.N., Mirza, M.W. and Zapata, C.E. (2006). “Environmental Effects in
Pavement Mix and Structural Design Systems. Calibration and Validation
of the ICM Version 2.6. NCHRP 9-23 project”. Preliminary Draft. Final
Report - Part 2. Submitted to National Cooperative Highway Research
Program. (Under panel revision).
Leong, E.C., & Rahardjo, H. (1997). “Review of Soil–Water Characteristic Curve

Equations”, Journal of Geotechnical and Geoenvironmental Engineering,
Vol. 123, No. 12, pp. 1106 – 1117
Perera, Y. Y. (2003). Moisture Equilibria Beneath Paved Areas. Ph. D.

Dissertation, Arizona State University, Tempe, United States.
Richards, L. A. (1931). Capillary conduction of liquids through porous mediums.

In Black, C. A. (Ed.). Methods of Soil Analysis, Part 1: Physics (pp. 318–
333).
Saxton, K.A., & Rawls, W.J. (2006). “Soil Water Characteristic Estimates by
Texture and Organic Matter for Hydrologic Solutions”, Soil Science
Society of America Journal, Vol. 70, pp. 1569–1578
Sillers, W.S., & Fredlund, D.G. (2001). “Statistical Assessment of Soil–water

Characteristic Curve Models for Geotechnical Engineering”, Canadian
Geotechnical Journal, Vol. 38, No. 6, pp. 1297–1313
Soil Survey Division Staff, (1993) Soil Survey Manual—United States

Department of Agriculture handbook no. 18 (rev. ed.): [Washington,
D.C.], U.S. Department of Agriculture, 437 p.
Soil Survey Staff, (1994) Keys to Soil Taxonomy (6th ed.): [Washington, D.C.],
U.S. Department of Agriculture, Soil Conservation Service, 306 p.
Soil Survey Staff, (1993) National Soil Survey Handbook, title 430 VI (issue 3):
[Washington, D.C.], U.S. Department of Agriculture, Soil Conservation
Service [variously paged].
233
Soil Survey Staff, (1975) Soil taxonomy—A Basic System of Soil Classification
for Making and Interpreting Soil Surveys, agriculture handbook no. 436:
[Washington, D.C.], U.S. Department of Agriculture, Soil Conservation
Service, 754 p.
Taylor, J.K. (1987). Quality Assurance of Chemical Measure, Lewis Publishers,

1987, 328 p. (QD51.T38 1987)
Terzaghi, K. (1943). “Theoretical Soil Mechanics”, John Wiley & Sons, Inc.,
New York, NY
U.S. Department of Agriculture, National Resources Conservation Service, Soil

Data Mart, Order U.S. Soil Map (STATSGO) Data. June 2008. Retrieved
from: http://soildatamart.nrcs.usda.gov/USDGSM.aspx
Van Genutchen, M.Th. (1980). “A Closed–Form Equation for Predicting the

Hydraulic Conductivity of Unsaturated Soils”. Soil Science Society of
American Journal, Vol. 44, No. 5, pp. 892-898.
Wagner, L.E., Ding, D. (1994). “Representing Aggregate Size Distributions as

Modified Lognormal Distribution”, Soil and Water Division of American
Society of Agricultural Engineers, Vol. 37, No. 3, pp. 815-821
Witczak, M.W., Zapata, C.E. and Houston, W.N. (2006). Models Incorporated
into the Current Enhanced Integrated Climatic Model: NCHRP 9-23
Project Findings and Additional Changes after Version 0.7. Final Report.
Project NCHRP 1-40D. Inter Team Technical Report. Arizona State
University. Tempe.
Zapata, C.E. (1999). Uncertainty in Soil–Water Characteristic Curve and Impacts

on Unsaturated Shear Strength Predictions. Ph. D. Dissertation, Arizona
State University, Tempe, United States.
Zapata, C.E. (2010). Research Results Digest 347: A National Catalog of

Subgrade Soil–Water Characteristic Curves and Selected Soil Properties
for Use with the MEPDG. National Cooperative Highway Research
Program, Transportation Research Board, of the National Academies.
ISSN 0077-5614. ISBN: 978-0-309-09929-5. Library of Congress Control
Number 2008924251. pp. 62.
Zapata, C.E., Andrei, D., Witczak, M.W., & Houston, W.N. (2007)
“Incorporation of Environmental Effects in Pavement Design”,
International Journal of Road Material and Pavement Design, Vol. 8, No.
4, 2007, pp. 667–693
234
Zapata, C.E., Houston, W.N. (2008). “NCHRP Report 602, Project 9-23:
Calibration and Validation of the Enhanced Integrated Climatic Model for
Pavement Design”, National Cooperative Highway Research Program.
Transportation Research Board, Washington D. C.
Zapata, C.E., Houston, W.N., Houston, S.L., & Walsh, K.D. (1999). “Soil–Water
Characteristic Curve Variability”, Department of Civil and Environmental
Engineering, Arizona State University.
Zhang, L., & Chen, Q. (2005). “Predicting Bimodal Soil–Water Characteristic

Curves”, Journal of Geotechnical and Geoenvironmental Engineering,
Vol. 131, No. 5, pp. 666-670
235

TorresHernandez Asu 0010N 10702 PDF

Uploaded by

Copyright:

Available Formats

TorresHernandez Asu 0010N 10702 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

TorresHernandez Asu 0010N 10702 PDF

Uploaded by

Copyright:

Available Formats

Estimating the Soil–Water Characteristic Curve

Using Grain Size Analysis and Plasticity Index

Gustavo Torres Hernandez

A Thesis Presented in Partial Fulfillment

Approved April 2011 by the

Claudia Zapata, Chair

ARIZONA STATE UNIVERSITY

The infrastructure is built in Unsaturated Soils. However, the geotechnical

practitioners insist in designing the structures based on Saturated Soil Mechanics.

The design of structures based on unsaturated soil mechanics is desirable because

it reduces cost and it is by far a more sustainable approach.

The research community has identified the Soil–Water Characteristic

conditions. This soil property is unpopular among practitioners because the

laboratory testing takes an appreciable amount of time. Several authors have

attempted predicting the Soil–Water Characteristic Curve; however, most of the

published predictions are based on a very limited soil database.

The National Resources Conservation Service has a vast database of

content measurements at different levels of suctions. This database was used in

For these reasons, two models to estimate the Soil–Water Characteristic

compared to existing models. The second model is based on the relationship

between the Soil–Water Characteristic Curve and the Pore-Size Distribution of

Characteristic Curve function and proved to be a better approximation than that

implementation of unsaturated soil mechanics into engineering practice due to the

To my mother and my family

to work on the 9-23A project, “A National Catalog of Subgrade Soil–Water

losing their faith in me.

daughters about life and to my students about engineering.

Thank you for doing something great: ‘To teach’.

LIST OF TABLES ................................................................................................. xi

LIST OF FIGURES ............................................................................................. xiv

1.1 Overview ....................................................................................1

1.2 Historical Background ...............................................................3

1.3 Research Objectives ...................................................................5

1.4 Methodology ..............................................................................6

1.5 Organization of the Thesis .........................................................7

2 LITERATURE REVIEW ............................................................................9

2.1 Introduction ................................................................................9

2.2 Stress State Variables ...............................................................10

2.3 Moisture Flow in Unsaturated Soils ........................................13

2.4 Matric Suction and the Soil–Water Characteristic Curve ........16

2.5 Approaches to Obtain Unsaturated Soil Properties .................17

2.5.1 Level 1 ......................................................................18

2.5.2 Level 2 ......................................................................21

2.5.3 Level 3 ......................................................................22

2.5.4 Level 4 ......................................................................22

2.6 Approaches to Predict the Soil–Water Characteristic Curve ...23

2.7 Final Remarks ..........................................................................26

3 DATABASE COLLECTION ....................................................................27

3.1 Source of the Database.............................................................27

3.2 Characteristics of the Database ................................................28

3.3 Data Collection ........................................................................30

3.3.1 Master File Properties and Characteristics ...............33

3.3.2 Preliminary Reduction of Soil Unit Data ..................34

3.3.3 Selecting the Proper Component to Represent the

Map Unit ...................................................................34

3.3.4 Properties Included in the SWCC Predicting

3.4 Working with the Database ......................................................46

3.4.1 Calculating Soil–Water Characteristic Soils

3.4.2 Calculating Grain–Size Distribution Parameters ......53