291 8.

2 Transient release and imbibition method (TRIM)

space aligned with the gravity direction z, the Richards equation can be written in the
suction head lz form as

iJ
- [K (h
iJz
) (ilh )] -+l
ilz
- iJ!:i (h) iJh
ilh iJ t
(4.15)

where !:i (h) and k(lz) µre the SWRC and HCF of the soil, respectively, and t i s time. In a
typical TRIM test, the initial and boundary conditions for the drying process are

h( z , t = 0) = O (8.la)
h( z = 0, t > 0) = hd (8. lb)
ilh(z = -/, I> 0) = O
(8. lc)
iJz

where hd is the applied increase in matric suction head, and l is the sample height i 1 plus the
thickness of the HAE ceramic stone / 2 • Conversely, the i nitial and boundary conditions for the
wetting process are

h(z , t = fa ) = h (z) (8.2a)

h( z = 0, f > fa ) = hw (8.2b)
iJ/J(z = -[ , I > td )
-------= 0 (8.2c)
iJ z

where td is the time when the drying loop is terminated, and hw is the applied one-step
decrease in matric suction head. Analytical solution of Equation (4.15) under the initial
and boundary conditions (8.1) or/and (8.2) are available under two strict assumptions: the
hydraulic conductivity is a constant, and the impedance of the HAE ceramic stone can be
ignored (e.g., Gardner, 1956; Miller and Elrick, 1958; Kunze and Kirham, 1961; Gupta et
al., 1974). These two assumptions are not valid for high impedance ceram ic stones and for
soil specimens in which suction and moisture content vary substantially. Therefore,
analytical approaches for solving this type of problem have been largely abandoned. With
the availability of computer power and development of nu merical solutions of partial
differential equations, Equation (4.15) can be solved numerically using finite-element and
finite-difference techniques. The numerical solution of Equation (4.15) under Equation
(8.1) conditions for the drying state leads to time series of outflow at the bottom of the
HAE ceramic stone:
c iJ h( z = 0, t ) _
q(O, 1) = Ks = qa(l , t) (8.3)
iJz

and the numerical solution of Equation (4.15) under Equation (8.2) conditions for the
wetting state can be expressed:

c il h( z = 0, 1) _
q(O, 1) = Ks = qw(l , 1) (8.4)
iJz

where KC, is the saturated hydraulic conductivity of the HAE ceramic stone.
 Hydro-mechanical properties

(+) -
----.
iJ'.;''(t,1)
fJ;;''(l,1)
!+--------- iE-------
1 Ks w measurement
!
Reset I
balance !
_ _ _\ -----

'
Water Water
outflow outflow
due to due to
small large
suction suction
increment increment

(-)
Time

Sample transient data obtained from typical TRIM test with complete drying and wetting loops (from Wayllace and
Lu, 2012).

Experi mental unsaturated flow data measured using the electronic balance for tran- sient
water outflow (q'P(/, t )) or inflow (q'P(/ , 1)) as a function of time (illustrated in rigure 8.4)
are used as objective functions i n an inverse numerical model to obtai n the unsaturated
hydraulic properties of the soil. This process is called parameter optimization and consists of
setting up a forward model with appropriate initial and boundary conditions and an ini tial
estimate of the optim i zed parameters. The anticipated or predicted system response (qd(L , t )
for drying or q.,,(L, t ) for wetting) is calculated using a numerical solution of the governing
flow equation (Equa tion (4. l 5)) with appropriate initial and boundary con- ditions (8.1) for the
dryi ng loop and (8.2) for the wetting loop. The predicted response is then compared to the real
system response, in this case the experimental data of water outflow as a function oftime UJ, x p(/
, I ) for drying or q'P(/ , t ) for wetting) . The system parameter s to be inversely estimated are
then iteratively adjusted and optimized until the differences between observed and expected
responses are within the degree of precision desired (van Gcnuchten, l 981; Kool el al., J 985;
van Dam et al ., 1994; Toride et al., 1995). To ensure parameter uniqueness, the inverse
model is typically run repeatedly using a range of initial parameter estimates to verify that the
solution converges to the same or similar results with some preset tolerances.
The results presented below were obtained by using the Hydrus-1 D (Simunek et al.,
2008) code that has an inverse modeling option. The code implements a Levenberg- Marquardt
non-linear optimization algorithm . The Levenberg-Marquardt algorithm com- bines the Gauss-
Newton algorithm and the method of gradient descent in order to min imize the deviations
between measured (q'P(/ , t ) for drying state, or q'P(/ , t ) for wetting state)
 293 8.2 Transient release and imbibition method (TRIM)

(a) (b) -Drying n"

-Drying 11'
1E+5 ---Welling nw
---wetl ng nw
1E+4

'"
c 1E+3
:.g
1E+2
" I

:< 1E+1 1 /ad I

1 =r=::::: :7.:.:::.:::.:. ·::7 ::·:.::::::."; .-
1E+O
!

0 Cl 25 l!,w Ii," Cl 5

Volumetric water content Volumelric waler content

llluwauon of hys1emls in (a) SWRCand (bl HCF.

and calculated system responses (qd(I, t ) or fJw(l , t )) (Levenberg, 1944; Simunek et al.,
2008).

Inverse modeling for drying conditions

Water flow into the system is modeled for a one-dimensional variably saturated med ium
with two materials: the ceramic disk and the soil. The constitutive relations between suction
head, water content, a nd hydraulic conductivity under drying conditions may be represented
using the van Genuchten ( 1980) and Mualem models ( l976), described by the following
equations:
I I

0 -0;1
{/ •l - (JJ
s r
[I+ ( a l h l )"'
1

] -
0
(8.5)

d J 1
1 _ (ad 11i1r"- [1 + (ad 11i1tT -r
(8.6)

K = Ks I I

{ 1 + (ad 11i 1rT-'""

where Fl s is the saturated volumetric water content, Fl , is the residual moisture content., n
and a are empirical fitting parameters with a being the inverse of the air-entry pressure
head and n the pore size distribution parameter, K, is the saturated hydraulic conductivity,
and the superscript d indicates drying state. The graphic definitions of these parameters are
illustrated in Figure 8.5.
As explained in the TRIM testing procedure (Section 8.3), two suction increments are
applied after the soil specimen is saturated. The first increment is set to slightly above the
air-entry value so that the sample is just in the unsaturated flow regime. Water outflow is
then monitored as a function of time (Figure 8.4). Once steady-state conditions are reached,
294 Hydro-mechanical properties

a second large suction increment is applied. The numerical model simulates this drying
process, taking into account both suction increments in the objective function. By using
two steps the model is better constrained in comparison with a single step (Toonnan et al .,
1992; van Dam et al., 1994; Dumer and Iden, 201 1) and the parameters obtained through
-inverse modeling are more accurate estimates.
Initial conditions in the numerical model are defined through pressure heads, with preset
values equal to zero. Boundary conditions are specified as no flow on top and a specified
pressure head on the bottom of the simulated flow cell. The value set for the latter changes
with time such that the applied suction increments are simulated in an identical manner to
the experi mental process.
The soil pa rameters that are estimated using inverse modeling are the residual moisture
content fi , parameters Old, 1l,and the saturated hydraulic conductivi ty K . Initial esti mates
of these parameters and a ra nge of minimum and maximum possi ble values are specified
by the user. Saturated hydraulic conductivity K1
is sensitive to any disturbance of the soil
specimen; thus, experimentally obtained values have a certain range of uncertainty. To
minimize the uncertainty in the esti mated K1
only a narrow range around the measured
saturated hydraulic conductivity value should be used in the inverse modeling procedure.
Hydrologic properties for the ceramic disk a re specified by the user. Once all soil properties
have been obtained, the SWRC and HCF for drying condi tions are fully defined.

Inverse modeling for wetting conditions

The numerical model setup for wetting conditions is sim ilar to that for drying, where both
the soil sample and ceramic disk are modeled in one-dimensional ftow, and hydrologic
constitutive relations may be defined using the Mualem ( 1976) and van Genuchten ( 1980)
models:

R - fl'" [ I ]'-;;\.
(8.7)
l:J: - ,'." = 1 + (DI '" I/JI)"•
2
1 - (aw l h l )"w _ , [ l + (aw lhl)""];;k -I }
{
K'" = K w -'----------.,--,---- (8.8)
s { l+ (aW l/Jl)""' ) j -p

where the superscript w i ndicates wetting. The definitions of these parameters are illus-
trated in Figure 8.5. Initial conditions for the wetting loop are specified using the pres-
sure head distribution at the time when the drying loop was terminated. Boundary con-
ditions are specified as no-flow on the top of the flow cell and a specified head on the
bottom.
The soil parameters estimated using an inverse model ing of wetting conditions are () ",
11" , and K .'i' . Hysteresis is expected in both the SWRC and HCF, thus, smaller values
1
aw ,
for (),, K,, and water entry pressure are commonly obtained (Mualem, 1976; Pham et al., 2005).
The saturated hydraulic conductivity K':' estimated using the inverse model may be
295 8.3 TRIM testing procedure

Tablea.1 Steps and testing times for <0mpleting a TRIM test

Step Description 1ime (hours)

Specimen preparation 2-6

2 System saruration 15-36
Data logging
4 Application of small suction increment 12-24
5 Application of large suction increment 48
6 Quantification of diffused air 0.5
7 Application of wetting conditions 7-24
8 Obtaining an objective function and performing inverse modeling 3
Total 88-138

verified with the experimental data considering steady flow through two materials in series
connection (e.g., Freeze and Cherry, 1979):

(8.9)

where Keq is the equivalent hydraulic conductivity of the system, K is the saturated
hydraulic conductivity of the ceramic disk, ls is the le11gth of the soil sample, and le is the
length of the ceramic stone. The residual moisture content for the wetting loop is set to the
fl, value obtained for the d rying loop. The estimated saturated volumetric water content for
wetting conditions can be verified experimentally by measuring the water content of the
sample after the tests are completed.

8.3 TRIM testing procedure

A typical testing program for measuring both drying and wetting SWRC and HCF involves
eight steps. The approxi mate testing time required to complete each step is provided i n
Table 8.1. Depending on the soil tested, constructing the principal dryi ng and a scanning
wetting SWRC and HCF with the TRIM method requires 4 to 9 days.

Step 1. Specimen preparation

The flow cell apparatus can accommodate both undisturbed and remolded samples as neither
suction probe nor moisture content sensor is needed. Undisturbed specimens can be placed
in their own mold inside the flow cell, while remolded sa.mples can be compacted directly
into the flow cell. Special care must be taken so that the ceramic disk is not damaged during
the compaction process. The experimenter(s) must then obtain accurate measurements of
296 Hydro-mechanical properties

the porosity np and of the saturated hydraulic conductivity. The former can be obtai ned by

1115
ll p = 1 - --- (8.10)
G,pwVr

where G, is the specific gravity, 111, is the mass of solids, V, is the total volume of the
specimen and Pw is the density of water. Saturated hydraulic conductivity is measured
independently and it may be accomplished by using either the constant head or the falling
head method after the specimen is fully saturated (in Step 2).

Step 2. Saturation of the system

Prior to testing, the experimenter(s) must saturate the entire system including the ceramic
disk, the thin water chamber underneath the disk, the bubble trap, the plumbing, and the soi l
specimen. Saturation of the disk is accomplished by partially and then fully submerging it
i n de-aired water for 1 2-24 hours while maintaining a vacuum of about 80 kPa in a glass
desiccator. A similar procedure is used to sah1rate the soil specimen; vacuum is applied
on the top of the sample while de-aired water is imbibed through the bottom. Change in
volume is m inimized by placing a coarse mesh and a spring between the top of the soi l
sample and the top cap of the flow cell. Dependi ng on the soil, sah1ration of the sample may
take 3 to 12 hours. Generally, a t the end of this step some excess water is observed at the
top of the soil sample. Saturation of the plumbing is accomplished by flowing water from
the large reservoir through the flow cell to the bubble trap and draining to the water jar that
sits on the balance. The bubble trap is filled with water using the two ports on its top cap.

Step 3. Data logging

Data are logged using a graphica 1 interface program written in LabVJEW. The experimenter
specifies the interval to record the mass and a pplied pressure time series. It is recommended
to log data every 10 seconds right after a ny changes in matric suction are applied to the
soil sample and every 10 minutes when conditions arc closer to steady state. An example
of typical data obtained during a TRIM test showing Steps 4 through 8 is provided in Figure
8.6.

Step 4. Application of small suction increment

After saturation of the specimen, any excess water on top of the sample is allowed to drain
by gravity; this process may be accelerated by applying a small pressure increment lower
than the air-entry pressure of the soil. Then, a small suction increment slightly above air
entry is applied to the sample to ensure suction is beyond air entry of the specimen. Water
outflow is monitored for steady-state conditions (Figure 8.4). lt takes approximately 12 to 24
hours to complete this step. This small increment in suction (close to the air-entry val ue) is
set to the magnitude beyond which water outflow is observed when matric suction is
increased in small i ncrements (0.1 kPa for sands, 0.5 kPa for silts). Typical values for sands
 297 8.3 TRIM testing procedure

05 0.3
(a) (b)
0 Experimental data -Fitied
04
C'
c
8 c" 02

:;; 03 0
u
1i)
2
"3:'
t)

"
2E Q2 ·c
0.1
!] :;

;g o Experimental data -Fitted

Q1

0 0
0 20 40 60 80 0 5 10 15
Time (hr) Time (hr)

Example of experimental and fitted objective functionsfor (a) drying conditions, (b) wetting conditions
(after Wayllace and Lu, 2012).

are I to 2 kPa and for si lts are 6 to 8 kPa. The flow rate observed is larger a t times closer to
zero due to the head gradient in the soil. As the total head distribution in the soil becomes
constant, the observed flow rate decreases to zero.

Step 5. Application of large suction increment

Next, the experimenter(s) must apply a la rge suction increment to the specimen while
measuring the water outflow as a function of time. The combination of data from Steps 4
and 5 is used as the objective function for the drying state qxp(I , t) (Figure 8.4). The li mit
of matric suction a pplied during this step is set by the air-entry pressure of the ceramic
disk. For example, if a three-bar ceramic stone is employed, it limits the largest suction
increment to about 300 kPa. If needed, the disk may be replaced with higher or lower air-
entry pressure. Water outflow due to the suction increment is moni tored and recorded until
steady-state or close to steady-state conditions are met. This step requires approxi mately
48 hours to be completed.

Step 6. Quantification of diffused air through ceramic disk

Any diffused air needs to be accounted for in order to accurately measure the water mass
changes on the electronic balance. As explained in the description of the apparatus, air
bubbles diffused through the ceramic disk are quantified by flushing them to the air bubble
trap. The flow cell may be turned on its side and tapped to help flush all bubbles out.
For 290 kPa of air pressure applied during 2 days, typically, l to 2 ml of diffosed air are
observed. When correcti ng the function of flow rate versus time for drying conditions, it is
298 Hydro-mechanical properties

assumed that air diffuses at a constant rate. The time required to complete this step is about
30 minutes.

Step 7. Application of wetting conditions

After all diffused air has been quantified, the experimenter must ensure that all plumbing
is saturated and that the balance is zeroed. Next, a large suction decrement is applied so
that the water is imbibed by the soil specimen. The mass of water inflow is monitored
and recorded in a similar way to the drying loop. If desired, a positive pressure head
can be applied to the bottom of the sample by decreasing the applied air pressure to
zero and adjusting the elevation of the flow cell relative to the elevation of the water jar
placed on the balance. Due to the hydraulic gradient created in this manner, water will
flow from the water jar on the balance to the flow cell even after the soil has reached
the wetti ng saturated water content B !: . A typical response for application of wetting
conditions is depicted in Figure 8.4. The rate of water imbibition is first i ncreasing then,
as the total head distribution becomes constant, steady-state conditions are reached. If a
gradient is created between the water reservoi r and the soil sample, the water mass in the
balance decreases l inearly with time after saturated wetting conditions are reached. The tests
presented in this chapter were performed with 0 kPa of applied air pressure and 8 cm of
elevation difference between the water jar and the ceramic disk (approximately -0.8 kPa of
matric suction). In the numerical model, the bottom boundary condition for the wetting path
is equal to the pressure head at the base of the ceramic disk (i.e., 8 cm). The transient
data considered for the objective function q 'P(/ , t ) must correspond only to the unsaturated
flow regime where moisture content of the sample increases with time. Depending on the
type of soil and specimen dimensions, this step takes 7 to 24 hours to complete.

Step 8. Obtaining objective function and performing inverse modeling

The objective function is defined as the water volume flowing in or out of the soil as a func-
tion of time. Two numerica l models that implement the Levenberg-Marquardt algorithm
are set up, one for drying and the other one for wetting conditions. To ensure uniqueness of
the results, the experimenter may check that the same parameters are obtained for differ-
ent initial estima ted values of the parameters that are identified through inverse modeling.
Since the numerical model simulates one-dimensional unsaturated flow, an equivalent three-
di mensional solution can be found by multiplying the water inflow or outflow given by the
model with the cross-sectional area of the soil sample. Figure 8.6 displays an exa mple of
change in volumetric water content of a soil for both drying and wetting conditions; the
hollow circles are experimental data (q'P(/ , t) for drying or q'P(t , 1) for wetting) and the
smooth lines are the responses predicted by the model. The time required to complete this
step is approximately 3 hours. Once all the hydrologic properties are obtained, the SWRC
and HCF for both drying and wetting conditions may be plotted.
 299 8.4 Validation of the TRIM method

50 (a) ------ Sand- TRIM (b)

-----·Sand-TRIM -Silly clay loam GP
:i: Sand - observod
x Sand- observed o Silty clay loam GP - - - Yolo light - TRIM

40 ---Yolo light· TRIM - observed a Yolo light - obseNed

a Yolo lighl - observed -Silty clay loam CP -TRIM

l 30 ,,. lfo .......-·...··-·-... 4 000 _ ........ H000 00000-00--0 w;;

o Silty day loam-observed

"j ;I --------0------------0
li 20 '
{2
"'\.
--,

500 1000

0.1 02
'
1500
i
f
1
03 04 Q5 0,6
lime (min) Volumolnc walot t;0<1lcn t

·:P;k:;:;1 Results obtained for verification of uniqueness using transient response data and inverse modeling for sand, silt, and
clay: (a) transient response, (b) obtained SWRC (from Wayllace and Lu, 2012).

8.4 Validation of the TRIM method

Three main approaches have been used to demonstrate the validity of the TRIM technique:
( 1) verification that the parameters obtained through inverse modeling are independent of
the initial values assigned to them, (2) repeatability of results during the experimental
portion, and (3) comparisons of the SWRC results for both drying and wetting with other
traditional methods (Wayllace and Lu, 20 12).

8.4.1 Uniqueness of results obtained by inverse modeling

The working principle for this technique is that when a large suction change is applied to
a soil subjected to fixed initial and boundary condi tions, the transient response is unique to
that soiL Thus, if information of water outow or inflow as a function of ti me is provided,
the hydrologic properties of the soil may be calculated with an inverse modeL The results
obtained must be independent of the initial estimates for the variables calculated this way,
For verification of uniqueness of results, a forward model for a given soil with known
properties was executed and the expected transient response obtained. These data were then
treated as "observed data" and a numerical inverse model was performed providing random
initial estimates for R" a , and 11; it was then verified that the results obtained with the inverse
model converge to the actual soil properties, The procedure described above was repeated
for 16 di fferent soils ranging from sand to clay. Typical transient response and SWRC for
sand, silt, and clay specimens are provided in Figure 8.7; the circles, squares, and crosses
represent the "observed data" (data obtai ned with the forwa rd model) while the solid lines

300 Hydro-mechanical properties