Download as pdf or txt
Download as pdf or txt
You are on page 1of 19

PREDICTING GROUND-WATER RESPONSE

TO PRECIPITATION
By D w i g h t A. Sangrey, 1 M . ASCE, Kingsley O . Harrop-Williams, 2
a n d Jeffrey A. Klaiber, 3 Associate M e m b e r s , ASCE
Downloaded from ascelibrary.org by New York University on 05/17/15. Copyright ASCE. For personal use only; all rights reserved.

ABSTRACT: High ground-water levels are a principal cause of slope instability


and of significance in other geotechnical engineering problems. A methodology
has been developed, and is described in this paper, for predicting the fluctua-
tion of ground-water levels as a function of precipitation. The approximate
method uses measured characteristics of the site or region and a short period
of calibration from a well record. The linkage of precipitation to ground-water
fluctuation can be used in either deterministic or probabilistic methods. The
new methodology has been tested by application in several different geological
and climatological areas. Based on these field studies, the sensitivity of the
methodology to calibration period, aquifer response time, evapotranspiration
and other characteristics has been evaluated. The overall conclusion of the work
has been a demonstration that the fluctuation of ground-water levels, including
extreme values, can be predicted with sufficient accuracy for many design and
analysis problems.

INTRODUCTION

The vast majority of landslides are related directly or indirectly to pre-


cipitation. Runoff of rainfall over a slope can contribute to sliding. Ero-
sion, especially by streams cutting at the toe, can reduce the stability of
slopes and produce failure. A general effect of precipitation is infiltration
into the ground contributing to two factors which influence stability: an
increase of unit weight as saturation increases, and seepage forces. The
general relationship between precipitation and landslides has been the
subject of a few specific field studies. Among the most comprehensive
are the results of Campbell (2), Fig. 1, illustrating several important con-
clusions from previous work. Perhaps the most important example in
Fig. 1 is the complex, rather than simple, relationship between rainfall
and landslides. Slope failures do not occur in direct proportion to pre-
cipitation quantity. In some years a given quantity produces many land-
slides while in other years the same quantity produces few landslides.
The second principal observation from Fig. 1 is the significance of time
in the process. All types of landslides lag somewhat behind the precip-
itation and certain types have greater time effects than others.
Qualitative studies, such as the one shown in Fig. 1, provide a val-
uable perspective on the important relationship between precipitation
and landsliding but cannot be extrapolated from one area to another.
Furthermore, many important problems, such as risk assessment, re-
gional modeling, and design of control or remedial measures require a
'Prof, and Head, Dept. of Civ. Engrg., Carnegie-Mellon Univ., Pittsburgh, Pa.
15213.
2
3
Asst. Prof., Dept. of Civ. Engrg., Carnegie-Mellon Univ., Pittsburgh, Pa. 15213.
Engr., D'Appolonia Consulting Engrs., Pittsburgh, Pa. 15146.
Note.—Discussion open until December 1, 1984. To extend the closing date
one month, a written request must be filed with the ASCE Manager of Technical
and Professional Publications. The manuscript for this paper was submitted for
review and possible publication on October 13, 1982. This paper is part of the
Journal of Geotechnical Engineering, Vol. 110, No. 7, July, 1984. ©ASCE, ISSN
0733-9410/84/0007-0957/$01.00. Paper No. 18992.
957

J. Geotech. Engrg. 1984.110:957-975.


TIME IN YEARS
I
10
I
15
I I
20
I I ~T~I—r~i—r 1 T
40
CUMULATIVE / Z5 30 35
'•>s PRECIPITATION /
LONG-v'TERM \ MEAN /
/
Downloaded from ascelibrary.org by New York University on 05/17/15. Copyright ASCE. For personal use only; all rights reserved.

—DRY- -WET-

SKIN
SLIDES' n
ii
I I r.
1 f
» I I i\
I I

!.MVi h'!'!(
SLIDES IN
SURFICIAL
DEPOSITS *'S"W
A./
BEDROCK
SLIDES w \ />

FIG. 1.—Different Relationships between Landsliding and Precipitation which


Produce Changes in Ground-Water Level (After Ref. 2)

quantitative description. The fundamental relationship underlying all of


these problems is the time-dependent relationship between precipitation
and change in the position of the phreatic surface or ground-water table.
A quantitative tie between precipitation and ground-water levels is es-
pecially valuable because some precipitation records exist for most lo-
cations. If a link can be established between these records and a specific
phreatic surface position, then a hindcasting projection of the ground-
water level for the entire period of rainfall record is possible. Such a time
record would be very valuable in application to the problems previously
noted.
This paper presents a description of a methodology developed to link
precipitation and ground-water response. The principle of the meth-
odology was proposed by Johnson (6) and extended by Sangrey, et al.
(7). The latter study also applied the methodology to a number of sites
in an attempt to gain perspective on the reliability of the methodology
in a variety of natural environments. Partial support for work by San-
grey, et al. (7) was provided from funds related to the severe landsliding
which occurs in the coastal areas of southern California. Consequently,
many of the field studies were done there.
The methodology described in this paper has direct application to
landsliding. Many different facets of landslide prediction, prevention,
and engineering control can be improved through application of this
methodology. These are not described in the paper, although a brief
illustration of probabilistic applications is given. The methodology also
has potential application to several other important problems such as
958

J. Geotech. Engrg. 1984.110:957-975.


ground-water contamination, ground-water control, water supply, min-
ing, and agriculture. This paper does not address any of these applications.

THEORETICAL BASIS FOR METHODOLOGY

The objective of this methodology is to predict the height of the phreatic


surface in an aquifer from the precipitation experienced in the area. Thus,
Downloaded from ascelibrary.org by New York University on 05/17/15. Copyright ASCE. For personal use only; all rights reserved.

it deals with meteorological elements such as precipitation and temper-


ature and the geological characteristic of the aquifer. It is convenient to
divide the theoretical model into two parts. The first part, called the
Hydrometeorological Model links the amount of precipitation, and other
factors such as temperature, to the effective ground-water recharge. The
second part, called the Geohydrological Model, links the recharge to the
height of ground-water in the aquifer.
This theoretical approach is similar, in principle, to some methods used
in ground-water hydrology (3-5,8,9) where a theoretical model is used
as the basis for a correlation of precipitation records and ground-water
levels. These previous contributions from ground-water hydrology, plus
others, have been formulated with various degrees of mathematical
complexity. The significance of many parameters and assumptions used
in these methods has also been examined.
The unique contribution of this paper is to propose a methodology
which is developed particularly for geotechnical engineering applica-
tions, such as landsliding. The various assumptions used and the par-
ticular elements selected for inclusion in the methodology have been
chosen after comparative analysis of those available from ground-water
hydrology. The result is a method with appropriate balance between
complexity and accuracy, considering the quality of available input data
and the needs for practical design and analysis.

HYDROMETEOROLOGICAL MODEL

The hydrometeorological model is the interface between hydrology and


meteorology. It determines the recharge to the ground-water and is based
on meteorological factors such as temperature and precipitation. Basi-
cally, the recharge R, can be obtained by subtracting the runoff, QR,
and evapotranspiration E, from the precipitation P. Other factors that
may affect the recharge are storage Sf, and synthetic elements, Sy . Stor-
age is primarily encountered in temperate zones, e.g., the accumulation
of snow in winter and its subsequent melting in the spring. Water in
the soil above the phreatic surface is also part of St. Synthetic factors
refer to unnatural irrigation and pumping. The recharge to the ground-
water can then be expressed as
R = P-QR-E±St±Sy (1)
In the present study, for simplification, the contributions due to S( and
Sy will be neglected. This simplification is valid for the case studies pre-
sented herein but may lead to significant errors in other areas. If S, or
Sy, or both, are known and cannot be neglected, the generality of the
model allows for their incorporation,
Since both QR and E are functions of P, it follows that R is a function

959

J. Geotech. Engrg. 1984.110:957-975.


of P also. Goldschmidt (3) suggested that both QR and £ are linear func-
tions of P as follows:
QR = c0 + ClP (2)
E = d0 + diP (3)
Thus, from Eq. 1, R is a linear function of P:
Downloaded from ascelibrary.org by New York University on 05/17/15. Copyright ASCE. For personal use only; all rights reserved.

R = (-co - do) + (1 - cx - d1)P (4)


The coefficients c0, C\ and d0, dx are obtained by linear regressions of
recorded values of P, QR and E. The validity of the assumptions that QR
and E are linear functions of P is limited to the steady state. Other fac-
tors such as the condition of soil moisture, anticedent precipitation, soil
cover, and others can also affect linearity. In general, the longer the base
time period for QR, E and P, the more valid the linear model will be.
However, there will be a corresponding increase in the possible error
for any individual precipitation event.
An alternative approach in the determination of the recharge was pro-
posed by Thornthwaite (8). He evaluated the monthly potential eva-
potranspiration PE m , in terms of the mean monthly temperature, Tm.
This can be expressed as
/ior \h
PEm = (16)1* ( - j ^ j (5)
1,514
in which 7 = 2(0.2r„,) is called the annual heat index, the summation
being taken over the year. The coefficient b2 is a correcting factor for
unequal month and day lengths (tabulated by month and latitude), and
&! = (6.75 X lfT 7 )! 3 - (7.71 x lCT5)/2 + (1.792 x 10~2)I + 0.49239.
Assuming that most of the runoff is eventually recharge to the ground-
water, Johnson (6) suggested that the monthly recharge can be taken as
the difference between the monthly precipitation and the monthly eva-
potranspiration. Thornthwaite's monthly recharge can then be obtained
as
Rm = Pm- PE„, (6)
in which P,„ = the monthly precipitation.
Johnson further suggested a combination of the Goldschmidt and
Thornthwaite methods because Thornthwaite's equation is in terms of
potential evapotranspiration rather than evapotranspiration. He sug-
gested that the monthly evapotranspiration can be written in the terms
of annual evapotranspiration, Ey, as: E,„ = (PE,„/PEy)Ey in which PEy
= ~2PEm; a n d Ey = d0 + diPy.
The coefficients d0 and dx are from Goldschmidt's regression equation
of the annual evapotranspiration, Ey, to the annual precipitation Py.
Combining these factors gives the Goldschmidt-Thornthwaite recharge
formula:
Rm = 1 m ~ Em or Rm = do,,, + d\mi m (7)
in which d0m = -d0{PEJPEy); dlm = 1 - d 1 (PE m /PE y )(P y /P m ); P,„ = mean
monthly precipitation; and Py = mean annual precipitation.
An illustration of the relative distribution of monthly precipitation to
960

J. Geotech. Engrg. 1984.110:957-975.


5-
4- PRECIPITATION-
3- POTENTIAL
EVAPOTRANSPIRATION
2
I -
0 •
I m
Downloaded from ascelibrary.org by New York University on 05/17/15. Copyright ASCE. For personal use only; all rights reserved.

-I - EFFECTIVE •RECHARGE
-2 RECHARGE /POTENTIAL^
^NEGATIVE j
-3-
< "4-
-5-

FIG. 2.—Relative Distribution of Monthly Precipitation to Recharge and Evapo-


transpiration for Semiarid Climate

recharge and evapotranspiration for a semi-arid climate is given in Fig.


2. The Goldschmidt-Thornthwaite theory has been used to define eva-
potranspiration. If another evapotranspiration theory had been used, in-
cluding the Goldschmidt theory, the Thornthwaite theory or any other
model for evapotranspiration, the resulting distribution would be dif-
ferent from that shown in Fig. 2 (see Fig. 7).
As noted in Fig. 2, an important characteristic of precipitation and re-
charge records is the deficit when potential evapotranspiration is greater
than precipitation. This condition may occur for most months in an arid
area, but even in temperate climates there will be some periods with a
potential deficit. During a period of deficit recharge there is no contri-
bution of new precipitation to the ground water. In theory, there is po-
tential for the deficit to be manifest in a withdrawal of ground water
(e.g., through deep-rooted vegetation). This is a complex issue and for
the present study has been left out by assuming that there is no mech-
anism for withdrawing water to satisfy the deficit in evapotranspiration.
This hypothesis is found to be reasonable in the studies described in
this paper: however, there may be situations where this assumption is
not reasonable.

GEOHYDROLOGIC MODEL

The geohydrologic model relates the elevation of the ground-water ta-


ble to the recharge. The basic assumptions in the development of this
model are that the aquifer is homogeneous, the recharge is uniform time-
varying, and that there is a constant-head outflow. A schematic illus-
tration for reference in derivation is Fig. 3.
The amount of flow for a unit element of aquifer section can be ex-
pressed as (Darcy's Law):
Q= -kHVH (8)
in which k = the permeability of the aquifer; and H = the height of the
phreatic surface above a reference elevation. For continuity, the increase
in storage (S) in the aquifer due to the recharge, R, is found to be [Fig.
3(a)]
961

J. Geotech. Engrg. 1984.110:957-975.


Downloaded from ascelibrary.org by New York University on 05/17/15. Copyright ASCE. For personal use only; all rights reserved.

Q+VQ

FIG. 3.—Schematic Depiction of Development of Geohydrologic Model: (a) De-


velopment of Boussinesq Equation; (A) Boundary Conditions for 3 Different Aqui-
fer Configurations

BH
S— = R-VQ (9)
at
Combining Eqs. 8 and 9 gives the Boussinesq equation
r) H
S — = R + kV(HVH) (10)

Substituting an average value, ha, for the first appearance of H, this


equation can be linearized as follows:

S— = R + khaV2H (11)
dt
For a sloping aquifer, as shown in Fig. 3(a), the flow may be considered
to be one dimensional. Thus Eq. 11 becomes
962

J. Geotech. Engrg. 1984.110:957-975.


dH , d2H
S — = R + kha — - . (12)
2
dt dx
The equations for other aquifer configurations are shown in Fig. 3(ib).
The decay solution of Eq. 12 for the boundary conditions of a sloping
aquifer [H = 0 at x = L and dH/dx = 0 at x = 0 (Fig. 3)] is
Downloaded from ascelibrary.org by New York University on 05/17/15. Copyright ASCE. For personal use only; all rights reserved.

RL216 A
-{2n
ma IT „=o
t TTX
+ 1)2 {In + 1)~3 cos {In + 1) (13)
toJ 2L
in which t0 = ^SL2lttikha •
In this derivation, it is assumed that the recharge over some time in-
terval (t; to tt + At) is uniform and equal to K,, and the amount of re-
charge varies for each time interval. This is shown in Fig. 4, in which
(a) shows discretized recharge hyetograph with solution H{x, t); (b) shows
isolation recharge "pulse" with simpler solution Hj{x,t - ti); (c) shows
resolution of pulse into two abruptly ending steady states with decay
solutions Hd{x,t - £,) and Hd[x,t — {tt — At)]; {d) shows that superposed
difference of decay solutions gives pulse solution Ht{x,t - f,) = Hd{x,t
- t^i - Hd{x,t - tt + Af); and (e) shows that superpositions of pulse
solutions produce effect of entire hyetograph H{x,t) = 1,iHi{x,t - tt).
The contribution of the total recharge to the ground water can then
be approximated by the superposition of the recharge of each time in-
terval. In this way the contribution of one pulse of recharge between
times separated by time length At is the difference of the decay function
before and after At, i.e.
Hi{x,t - U) = Hd{x,t- ti) - Hd{x,t - ^ + At) (14)
Generalizing, the height of the phreatic surface can be obtained as

«(*'*) = 2 H<(*>* -*<)•• <15)


If the effective recharge, R * (t), is defined as the recharge to the ground-
water at any time, t, then R*{t) can be written as

R*{t)= R{t')C{t')dt'. (16)

in which R{t) = the recharge due to the precipitation at time t; and C{t)
= the relative contribution of each precipitation that occurred in the past.
C{t) is found from Eq. 13 to be (6)

L2 32 °° nxx
1 1 (2n + 1) -(2n
C(0 = -T77T—T,
2 2 " 3 1 1 (- )"(2» + I)" cos exp
tl{L - x ) TT ^ o 2L

+ D2l7 (17)
oJ
As an approximation to Eq. 17, Jacob (4,5) proposed a linear function
963

J. Geotech. Engrg. 1984.110:957-975.


Recharge (R)
(a.)

ULT : k»
Downloaded from ascelibrary.org by New York University on 05/17/15. Copyright ASCE. For personal use only; all rights reserved.

-At time (t)

Hi
(b.)
m
tf-At ti

(c.)

-*-t

_e» t
tfAt

(d.)
\""V Hd(x,t-t|)
s
Hd(x,t-trAt)->sx x^

Hp{x,M|)-
=i ». f

FIG. 4.—Superposition of individual Recharge Pulses Giving Final Height to Phreatle


Surface
964

J. Geotech. Engrg. 1984.110:957-975.


C(,,=
^ir D
In a later study, Venetis (9) proposed an exponential decay weighting
function which results in a better fit. Venetis' theory allowed for the
representation of the current effective recharge, R*, in terms of the re-
charge of the current precipitation, R„, and the latest effective recharge,
Downloaded from ascelibrary.org by New York University on 05/17/15. Copyright ASCE. For personal use only; all rights reserved.

K*_i. This expression is a function of t0 and can be written as (6):


R* = [1 - exp H o 1 ) ] * . + exp (-t^R^. (18)

Precipitation Records
0) ( P)
•o
o
2 Mean Monthly Temperature
"5
o (Tm)
O Storage ± (S t )
Synthetic Precipitation
Evapotranspiration Model
(PE m ) Equa. 5
a)
E
o Monthly Recharge (R m ) Equa. 6 or 7
-& ( linear interpolation for daily recharge '
>-.
I
Initialize Model with Assumed
Mean Recharge (R*) and a Lead
Groundwater Records (H n Time Period Equa. 18

Assume Value of ( t „ ) Calculate ( R*) Equa. 18

a)
o

a
o
'5>
_o
o
Assume Calibration Time and Predict Groundwater ( H p )
Equa. 19
o
(1)

Report Model Characteristics ( t 0 ) ,


(Calibration time), (a 0 ),(a,)

Application Any Precipitation


Record

Predicted Groundwater Record

FIG. 5.—Flow Chart Indicating Steps Taken to Calibrate Model

965

J. Geotech. Engrg. 1984.110:957-975.


This new effective recharge is the weighted average of the previous
one and the recharge of the present precipitation, with the relative weights
being determined by the drainage time, t0, of the aquifer. Eq. 18 is an
iterative equation which can be initialized by assuming a value for R*,
say the mean recharge at some point well before the first ground-water
observation, and cycling through enough of the iterations that the error
in the original assumption is reduced.
Downloaded from ascelibrary.org by New York University on 05/17/15. Copyright ASCE. For personal use only; all rights reserved.

To simplify the solution, the height of the phreatic surface (or depth
to ground-water in the aquifer) is assumed to be linearly related to the
effective recharge. Thus, at any time, t, one obtains
H(t) = a0 + aiR*(t) (19)
The coefficients a0 and a\ are obtained by a linear regression of known
ground-water records, H, and the calculated effective recharge, R*, dur-
ing that period of record. In order to do this, one must choose a period
(calibration period) that gives the most efficient values of a0 and d\.
Fig. 5 represents a flow chart showing the steps to be taken in cali-
brating the model. The main parameters of the model are the drainage
time, t0, to be used in the recharge relationship (Eq. 18) and the cali-
bration time that determines the coefficients of Eq. 19. The drainage time,
t0, can be evaluated by arbitrarily selecting a period for which both ground-
water and precipitation records are available, assuming different values
of t0 and selecting the value of t0 that gives the maximum correlation
between the actual ground-water record and the predicted effective re-
charge (or phreatic surface elevation). In like manner, the calibration pe-
riod can be obtained by selecting sections of record of various lengths.
A prediction is made using each section of rainfall record, and compar-
ing the goodness of fit between the actual and predicted ground-water
levels. The calibration time is then selected as the period where there is
no further increase in the goodness of fit as the length of period increased.
In principle, the criterion for establishing goodness of fit might be dif-
ferent for different situations; e.g., in most cases a measure of goodness
of fit would involve matching the actual and predicted values for small
time intervals throughout the entire record. In contrast, for a situation
where the objective of a prediction was the peak elevation of the phreatic
surface in every year, a matching of actual and predicted peaks could
be the basis for goodness of fit. In such a case, the remainder of the
year's record would be ignored when evaluating fit.

APPLICATION AND EVALUATION

The theoretical methodology previously described has been applied to


a variety of actual situations to evaluate the technique and to provide
insight into the parameters affecting the model. All of the observations
and conclusions presented in the remainder of this paper are based on
results from similar evaluation studies. The basic data required for each
of these studies were: (1) A precipitation record; and (2) a long-term
record of ground-water level observations. Other data, such as runoff,
temperature, pumping, or irrigation records, were also collected and used
in several specific studies.
The experimental plan involved use of particular sections of a ground-
966

J. Geotech. Engrg. 1984.110:957-975.


Lo Honda - Slope I - well 5 P
Calibration period ( Jan. 1979 - June 1980)
Downloaded from ascelibrary.org by New York University on 05/17/15. Copyright ASCE. For personal use only; all rights reserved.

'/77 '/78 '/79


Day of Prediction Period ( t 0 = 155 Doys)

FIG. 8.—Comparison of Predicted and Actual Ground-Water Level

water record for calibration against the available precipitation record.


Having calibrated a particular well, other sections of the precipitation
record could be used to predict ground-water levels. These predictions,
in turn, were compared with the actual record for that period. In es-
sence, this was no different from calibrating a new well using the data
as they were collected and then predicting the ground-water levels as
additional precipitation records became available. Finally, these predic-
tions could be compared with newly collected records of ground-water
level measurements. The final product of such studies is a comparison
of predicted and actual ground water, as shown, in Fig. 6.
To describe the goodness of fit of predictions to actual ground-water
records, the relationship used was
\Hg - Hp\
K=l (20)

in which Ha = actual depth to the phreatic surface for a prediction pe-


riod; and Hp = predicted depth to the phreatic surface for a prediction
period.
It is also useful to describe some relationships in terms of statistical
correlation using
Nlzy - 2z~2y
Py Z =
(21)
{[N2z2 - (Ez)2][N2y2 - (2y)2]}1/2
in which y and z = variables. The results which follow are presented in
terms of these indicators.
The numerical calculations and data processing involved in this meth-
odology for predicting ground-water response to precipitation require
use of a digital computer. Programs for use with a DEC system (7) or
an IBM system (6) are available. The more recent and complete version
by Klaiber is the TEC program (7) which was used for all results reported
herein.

RESULTS OF FIELD VERIFICATION

Each of the elements of the methodology for predicting ground-water


967

J. Geotech. Engrg. 1984.110:957-975.


TABLE 1.—Study Areas and Characteristics
Mean annual
Mean annual temperature,
precipitation, in degrees
Area Location in inches Fahrenheit Geological setting
(D (2) (3) (4) (5)
Downloaded from ascelibrary.org by New York University on 05/17/15. Copyright ASCE. For personal use only; all rights reserved.

La Honda Northern 30 56 Foothills in sedimentary


California shales and sandstones
Camp Pendleton Southern 13 59-63 Foothills in weathered
California granite and
metasediments
16 sites Upstate New 36 47 Various glacial and la-
York custrine land forms
Northern Western 34 51 Colluvial soils on sedi-
Appalachian Pennsylvania mentary bedrock
Big Rock Mesa Southern 17 59 Marine terraces
California

response to precipitation has been evaluated through field studies. Five


different general locations have been used, each having a somewhat dif-
ferent geological and climatological environment. These five areas and
their characteristics are summarized in Table 1.
Based on all of these studies, the sensitivity of the overall prediction

FIG. 7.—Difference in Effective Recharge Predicted by Different Evapotranspira-


tfon Theories for Site in Southern California: (a) Thornthwaite Theory; (b) Qold-
schmidt-Thornthwalte Theory

968

J. Geotech. Engrg. 1984.110:957-975.


to variation within the particular elements of the methodology was eval-
uated. Some factors are of more interest than others, e.g., while the
accuracy of temperature and precipitation records is very significant, the
uncertainty associated with these data is no different than for other ap-
plications where extensive research has been done. On the other hand,
a factor such as t0, the characteristic drainage time for an aquifer, is
Downloaded from ascelibrary.org by New York University on 05/17/15. Copyright ASCE. For personal use only; all rights reserved.

unique to this methodology and is of great interest in a sensitivity anal-


ysis. The factors of greatest interest are those which follow.
Evapotranspiration Model.—The hydrometeorological model is used
to define the effective recharge to the ground water, R, by subtracting
runoff and evapotranspiration from precipitation. Other factors such as
storage or synthetic elements can also be included, but the principal
variable in most cases is the estimate of evapotranspiration. Numerous
theories have evolved for predicting evapotranspiration, as described in
this paper. These different theories would predict different amounts of

Camp Pendleton - 13 R
Calibration Period ( 1 / 7 6 - 5 / 7 9 )
TEC Model

'/7I '/72 '/73 '/74 1/75 >/76 1/77


Day of Prediction Period ( t 0 = 8 0 5 Days)

Camp Pendleton well 13R


Calibration Period (Z76-5/79)

DAY VS GROUNDWATER LEVELS (predicted vs actual) h ff

PREDICTED
ACTUAL

%a '69 '70 V7I </„ 773 774 '/75 ''76 ''77 '78 '79
Day of Prediction Period ( % = 1205 Days)

FIG. 8.—Difference In Predicted Depths to Phreatic Surface Using: (a) Thornth-


waite Evapotranspiration Theory; (b) Goldschmidt-Thornthwaite Evapotranspira-
tion Theory
969

J. Geotech. Engrg. 1984.110:957-975.


evapotranspiration for any given situation and, therefore, the effective
recharge would differ. These differences are shown in Figs. 7(a)-7(b) for
a site in southern California.
The significance of having different estimates of effective recharge is
that this characteristic is the basis for all of the calculations and corre-
lations within the hydrogeologic model presented previously. Conse-
quently, the empirical coefficients linking the prediction of ground-water
Downloaded from ascelibrary.org by New York University on 05/17/15. Copyright ASCE. For personal use only; all rights reserved.

levels to precipitation will be entirely different if a different evapotran-


spiration model is used. Ultimately, the quality of prediction will depend
on the model used. Fig. 8 shows how different the predictions may be;
the Thornthwaite evapotranspiration theory [Fig. 8(a)] and the Gold-
schmidt-Thornthwaite theory [Fig. 8(b)] give distinctly different predic-
tions. Note that both of these predictions are quite reasonable in terms
of overall correlation with the actual record (phh = 0.92 and phh - 0.84).
The methodology described in this report has been applied in many
different geographical areas having distinctly different climatic condi-
tions. Working in the temperate continental areas of New York and
Pennsylvania, Johnson (6) concluded that the hybrid Goldschmidt-
Thornthwaite theory worked best. However, in the more arid and dis-
tinctly seasonal rainfall areas of coastal California, the more severe eva-
potranspiration theory of Thornthwaite defined an effective recharge
which provided better predictions than any other evapotranspiration
model. It is probably premature to draw any general conclusion that the
Thornthwaite theory works better in drier climates and the Gold-
schmidt-Thornthwaite theory has advantages in temperate climates. Both
models will work and should be considered when applied in a new area.
Additional experience with this methodology may justify a stronger con-
clusion in the future.
Evaluating Drainage Time, t0.—A major component of the geohy-
drological element is the determination of the slope's drainage time, f0 •
The drainage time is a function of a given geometry as well as perme-
ability. Since these are generally unknown or costly to determine, John-
son (6) proposed a process of iteration and evaluation. By assuming a
range of drainage times, t0, and evaluating the accuracy of subsequent
predictions of ground water, a drainage time can be selected which max-
imizes the correlation between actual and predicted ground-water levels
for a set prediction period. Fig. 9 shows this process, in which the drain-
age times were determined for five wells. The distinct peaks in the plots
of phh versus t0 defined the best value for the drainage time, t0.
A conclusion indicated by the data in Fig. 9 is that wells in the same
general area tend to have similar values of t0 . The potential to apply this
methodology to regional planning and evaluation depends on such con-
sistency throughout a region.
Defining Required Calibration Period.—The principle of the geohy-
drologic model is to link the effective recharge record to actual ground-
water levels through a calibration process. In general, the longer the
calibration period, the more closely the predicted value approaches the
actual record. However, there must be a practical limit where the ad-
ditional accuracy obtained from longer calibration time is relatively
insignificant.
The present study defined a goodness of fit factor which measured
970

J. Geotech. Engrg. 1984.110:957-975.


LA HONDA SLOPE I

VARIATION of / h h with t0
Downloaded from ascelibrary.org by New York University on 05/17/15. Copyright ASCE. For personal use only; all rights reserved.

<\

160 240 320 400 480


Droinoge Time (days)

FIG. 9.—Method of Selecting Optimum Drainage Time, t0, and Its Similarity for
Wells in Same General Area

the incremental improvement of the predicted ground-water levels com-


pared to the actual ground-water levels (Eq. 20). A typical illustration of
the increase in goodness of fit as the calibration period lengthened is
shown for Camp Pendleton Well 13R in Fig. 10, in which t0 = 905 days;
and goodness of fit parameter: K = 1 - 2(1H A — Hp/HA). The best good-
ness of fit (approximately 0.9) was achieved using a calibration period
of 36 months, but a calibration period of only 12 months was nearly as
good and represents a satisfactory practical limit in this case. This pro-
cedure is recommended to define a calibration period consistent with
the need for accuracy and the requirement for expediency in collecting

K VS CALIBRATION PERIOD

Calibration Period (months)

FIG. 10.—Increase in Goodness of Fit of Predicted and Actual Depth to Phreatic


Surface with Calibration Period
971

J. Geotech. Engrg. 1984.110:957-975.


1.0

0.8

P 0.6 A .(mean = 0.60)


-'hh
Downloaded from ascelibrary.org by New York University on 05/17/15. Copyright ASCE. For personal use only; all rights reserved.

0.4

0.2

0
0 1976 1978 1980

YEAR

FIG. 11.—Insensitivlty of Model to Time when Calibration Period was Chosen (Note:
Data Plotted at Midpoint of Calibration Period)

data to make predictions. Based on experience to date with this meth-


odology, the overall fit may not be of as much interest as predicting the
peak fluctuations. One trend noticed throughout the course of the study
was that calibration periods consisting of large ground-water fluctua-
tions generally resulted in good peak predictions of ground water. In
fact, the shorter the calibration period incorporating large fluctuation,
the better the peak predictions and the worse the overall goodness of
fit. These results tend to suggest that for a particular length of calibration
period, the correlation or goodness of fit factor may vary depending upon
the amount of precipitation during the calibration process.
To test whether there is a significant difference in the accuracy of pre-
dictions at a given site depending upon when the model is calibrated,
studies were done in which a constant length of calibration period was
taken at different times within the rainfall record. A typical example,
from the La Honda area of northern California, is shown in Fig. 11.
Three different calibration periods of 29 months each were used to cal-
ibrate the model. Different empirical coefficients were obtained but, as
shown in Fig. 11, the differences between these coefficients and the dif-
ferences in prediction accuracy were negligible.
The conclusions from these studies are that a rational procedure for
defining an acceptable calibration period has been developed. The meth-
odology is relatively insensitive to when the calibration period is taken
within the precipitation record so long as a representative period of rain-
fall is included. For seasonal precipitation areas, the calibration period
should include at least one wet and one dry season.

PROBABILISTIC APPLICATION

For geotechnical engineers, one important application of this model is


to provide a method for assessing the risk associated with landsliding
which is caused, at least in part, by high levels of ground water. It is
an underlying assumption that most landslides occur during periods when
the phreatic surface is high. If long records of ground-water levels exist,
these can be used to predict extreme states of the phreatic surface ele-
972

J. Geotech. Engrg. 1984.110:957-975.


vation. Unfortunately, in most cases these records do not exist. Long
periods of precipitation data are available for most areas, however, and
these can be used to create ground-water data using the methodology
presented in this paper.
Using the calculated record of ground-water levels, extreme states of
the record can be calculated using familiar statistical methods (1). For
landsliding problems the maximum ground-water rise expected in a par-
Downloaded from ascelibrary.org by New York University on 05/17/15. Copyright ASCE. For personal use only; all rights reserved.

ticular interval (e.g., a 20-yr return period) is a key characteristic for long-
term stability predictions. As an illustration, if H mta is defined as the
minimum depth to ground water during that period, its mean value Hmin
and coefficient of variation, VHmb, can be evaluated from the record.
Without loss of generality, Hmin can be assumed to be modeled by a
Weibul probability distribution (Type III, smallest extreme). It follows
that the probability of the depth to ground water being less than some
given value, h is represented mathematically as (1):
PiHrnin S ft) = 1 - *-<""»>" (22)
2
in which u = H m i n /r(l + l/«); a is obtained from V Hm!n = |T(1 + 2 / a ) /
T 2 (l + l/a)] - 1; and T( ) = the gamma function.
Therefore, the recurrence interval of this particular ground-water depth,
h, is

T = (23)
P(H m l n </ l ) '
and the risk associated with it in i years can be evaluated as

Rk=l-(l-^j (24)
Assuming that the slip surface location and soil properties of a slope
are known, the slope stability problem reduces to defining the ground-
water levels which may occur during the existence of the slope. Thus,
if H is the depth to ground water at which sliding would start, then the
expected life of the slope is given by Eq. 23. Remedial measures would
be necessary if T proves to be too short a time. In this way, an individual
and specific slope can be evaluated to define the level of risk associated
with particular levels of ground water, and the cost of remedial mea-
sures which might be applied to increase its stability.
As an example, slope 1 at La Honda (shown in Fig. 12) was consid-
ered. Neglecting cohesion because the slip surface was taken as an old
surface of sliding, Wieczorek (10) conducted an infinite slope analysis
and found that the factor of safety of the slope can be expressed as
/ wVw\ tan cb
F= 1 (25)
V 7 / tan (3
in which ()> = the angle of internal friction; p = the angle of slope; 7 =
the unit weight of the soil; yw = the unit weight of water; and m = the
ratio of the height to the phreatic surface above the slip surface to the
height of soil above the slip surface. He also found that for average con-
ditions of the slope (<>
| = 17°; 7 = 117 pcf; and (J = 9.5°), failure was
expected to occur whenever m was greater than 0.85. From Fig. 12 this
973

J. Geotech. Engrg. 1984.110:957-975.


Downloaded from ascelibrary.org by New York University on 05/17/15. Copyright ASCE. For personal use only; all rights reserved.

FIG. 12.—Geometry of Slope 1 at La Honda

indicates that if the depth to ground water should be less than 7.5 ft
below the ground surface, failure will occur.
Using a past rainfall record for the site, the corresponding ground-
water record was constructed, from which the mean minimum annual
depth to ground water was found to be Hmin = 11.33 ft with a coefficient
of variation of VHmin = 0.47. Thus, from Eq. 22, the probability of the
depth to ground water being less than 7.5 ft in an annual period is 0.26.
This indicates that the slope is expected to fail with a 3.85 yr (Eq. 23)
recurrence interval.

CONCLUSIONS

The relationship between precipitation and geotechnical engineering


problems, such as slope stability, can be established through the pre-
dictability of ground-water response to precipitation. A methodology has
been developed and demonstrated for modeling the response of ground
water to the intensity, duration, and time history of precipitation. The
method is approximate and, therefore, the validity of applications de-
pends on the quality of data used.
Based on field studies to implement this methodology, these conclu-
sions have been demonstrated:

1. Different evapotranspiration models can be used in the method.


Certain evapotranspiration models work better under given conditions
than others.
2. The aquifer drainage time, t0, can be evaluated by iteration using
ground-water records. Values of t0 for a similar region are similar thus
providing a basis for extending this methodology to regional studies.
3. The required calibration period can be defined according to the re-
liability desired.
4. This methodology can be applied in risk-oriented probabilistic studies.
5. The predictability of ground-water fluctuations with precipitation
can exceed 90%. Extreme events also can be modeled with similar reli-
ability if sufficient data are available.
974

J. Geotech. Engrg. 1984.110:957-975.


ACKNOWLEDGMENT

Support for part of the work reported in this p a p e r w a s provided by


the U.S. Geological Survey u n d e r Grant 9-6102-05813. The writers ap-
preciate the data and valuable comments from G. F. Wieczorek of the
U.S. Geological Survey. C. O'Leary a n d F. Bukes assisted in preparation
of the manuscript.
Downloaded from ascelibrary.org by New York University on 05/17/15. Copyright ASCE. For personal use only; all rights reserved.

APPENDIX.—REFERENCES

1. Benjamin, J. R., and Cornell, A. G., Probability, Statistics, and Decision for Civil
Engineers, McGraw-Hill, New York, N.Y., 1970.
2. Campbell, R. H., "Soil Slips, Debris Flows, and Rainstorms in the Santa Monica
Mountains and Vicinity, Southern California," U.S. Geological Survey
Professional Paper, No. 851, 1975.
3. Goldschmidt, "Hydrometeorological Methods of Quantitative Estimation of
Annual Underground Water Replenishment," Internat. Assoc, of Sci. Hydrol-
ogy, No. 52, 1960, pp. 272-278.
4. Jacob, C. E., "Correlation of Groundwater Levels and Precipitation on Long
Island, N.Y., Part 1," Trans. AGU, 1943, pp. 564-573.
5. Jacob, C. E., "Correlation of Groundwater Levels and Precipitation on Long
Island, N.Y., Part II: Correlation of Data," Trans. AGU, 1944, pp. 928-939.
6. Johnson, K. H., "A Predictive Method for Groundwater Levels," thesis pre-
sented to Cornell University, at Ithaca, N.Y., in 1977, in partial fulfillment
of the requirements for the degree of Master of Science.
7. Sangrey, D. A., Harrop-Williams, K. O., and Klaiber, J. A., "Groundwater
Response to Precipitation with Application to Slope Stability," Research Report
R-82-131, Dept. of Civ. Engrg., Carnegie-Mellon Univ., Jan., 1982.
8. Thornthwaite, C. W., "An Approach Toward a Rational Classification of Cli-
mate," Geographical Rev., No. 38, 1948, pp. 55-94.
9. Venetis, C , "A Study of the Recession of Unconfined Aquifers," Bull. In-
ternat. Assoc, of Sci. Hydrology, No, 14(4), Dec, 1969, pp. 119-125.
10. Wieczorek, G. F., "Ground Water Level and Precipitation Data for Slopes
Near La Honda, California," Open File Report 81-367, U.S. Geological Survey,
1981, 25 pp.

975

J. Geotech. Engrg. 1984.110:957-975.

You might also like