Post on 06-Oct-2020
CALIBRATION AND SENSITIVITY ANALYSIS OF A GROUND-WATER FLOW MODEL OF THE
COASTAL LOWLANDS AQUIFER SYSTEM IN PARTS OF LOUISIANA, MISSISSIPPI,
ALABAMA, AND FLORIDA
By Angel Martin, Jr., and C.D. Whiteman, Jr.
U.S. GEOLOGICAL SURVEY
Water-Resources Investigations Report 89-4189
Baton Rouge, Louisiana
1990
DEPARTMENT OF THE INTERIOR
MANUEL LUJAN, JR., Sejsretary
U.S. GEOLOGICAL SURVEYI
Dallas L. Peck, Director
For additional information write to:
District ChiefU.S. Geological SurveyP.O. Box 66492Baton Rouge, v LA 70896
Copies of this report can be purchased from:
U.S. Geological SurveyBooks and Open-File Reports SectionFederal Center, Bldg. 810Box 25425Denver, CO 80225
CONTENTSPage
Abstract................................................................. 1Introduction............................................................. 2Hydrogeology............................................................. 2Design of the ground-water flow model.................................... 4Model calibration........................................................ 9
Steady-state model.................................................. 11Transient model..................................................... 11Statistical optimization program.................................... 13Results of calibration.............................................. 14
Sensitivity analysis ..................................................... 24Method of study..................................................... 24Transmissivity and vertical leakance................................ 25Storage coefficient................................................. 34
Summary.................................................................. 53References............................................................... 54
ILLUSTRATIONS
Figure 1. Map showing location of the study area and major streams .... 32. Generalized hydrogeologic section showing zonation of the
aquifer system and direction of regional ground-water flow...................................................... 5
3. Map showing finite-difference grid with the active areaof the model.............................................. 7
4. North-south and east-west schematic sections through the coastal lowlands aquifer system showing representation of the permeable zones as model layers and the simulated boundary conditions....................................... 8
5. Maps showing the differences between model-simulatedand measured water-level altitudes for 1980............... 12
6. Graph showing results from the optimization program showing changes in parameter values and resulting changes in root-mean-square error of water-level residuals................................................. 15
7. Histogram showing the distribution of the residualsbetween model-simulated and measured water levels......... 18
8. Hydrographs showing comparison of model-simulated and measured water levels for selected wells in model layers 2 and 3............................................ 19
9. Hydrographs showing comparison of model-simulated and measured water levels for selected wells in model layers 4, 5, and 6........................................ 20
10-12. Maps showing areal distribution of:10. Transmissivity in the calibrated model for each
permeable zone....................................... 2111. Vertical leakance between the permeable zones in
the calibrated model................................. 2212. Storage coefficient in the calibrated model for
each permeable zone.................................. 23
III
ILLUSTRATIONS~Con1±nuedPage
Figure 13. Diagram showing root-mean-square errors of water-level residuals resulting from changes in transmissivity and vertical leakance of all mDdel layers................. 28
14. Diagram showing root-mean-square errors of water-levelresiduals resulting from changes in the transmissivityof model layer 2 and the vertical leakance betweenmodel layers 1 and 2...................................... 29
15. North-south water-level profiles showing the effect inmodel layers 2, 3, and 4 of changing the transmissivityof all model layers by factors; of 0.2 and 5............... 30
16. North-south water-level profiles; showing the effect in model layers 5 and 6 and in the average of all layers of changing the transmissivityf of all model layers by factors of 0.2 and 5...................................... 31
17. East-west water-level profiles showing the effect in model layers 2, 3, and 4 of changing the transmissivity of all model layers by factors of 0.2 and 5...................... 32
18. East-west water-level profiles showing the effect in model layers 5 and 6 and in the average of all layers of changing the transmissivity of all model layers by factors of 0.2 and 5...................................... 33
19. North-south water-level profiles: showing the effect inmodel layers 2, 3, and 4 of chianging the transmissivityof model layer 2 by factors of: 0.2 and 5.................. 35
20. North-south water-level profiles* showing the effect inmodel layers 5 and 6 and in the average of all layers ofchanging the transmissivity of model layer 2 by factorsof 0.2 and 5.............................................. 36
21. East-west water-level profiles showing the effect in model layers 2, 3, and 4 of changing the transmissivity of model layer 2 by factors of 0+2 and 5..................... 37
22. East-west water-level profiles showing the effect in model layers 5 and 6 and in the average of all layers of changing the transmissivity of model layer 2 by factors of 0.2 and 5................. f ............................ 38
23. North-south water-level profiles showing the effect inmodel layers 2, 3, and 4 of changing vertical leakance between all model layers by factors of 0.2 and 5.......... 39
24. North-south water-level profiles showing the effect in model layers 5 and 6 and in tfae average of all layers of changing vertical leakance between all model layers by factors of 0.2 and 5................................... 40
25. East-west water-level profiles ishowing the effect in model layers 2, 3, and 4 of changing vertical leakance between all model layers by factors of 0.2 and 5.................. 41
26. East-west water-level profiles showing the effect in model layers 5 and 6 and in the average of all layers of changing vertical leakance between all model layers by factors of 0.2 and 5...................................... 42
IV
ILLUSTRATIONS OcxitinuedPage
27. North-south water-level profiles showing the effect inmodel layers 2, 3, and 4 of changing vertical leakance between model layers 1 and 2 by factors of 0.2 and 5...... 43
28. North-south water-level profiles showing the effect inmodel layers 5 and 6 and in the average of all layers ofchanging vertical leakance between model layers 1 and 2by factors of 0.2 and 5................................... 44
29. East-west water-level profiles showing the effect in model layers 2, 3, and 4 of changing vertical leakance between model layers 1 and 2 by factors of 0.2 and 5.............. 45
30. East-west water-level profiles showing the effect in model layers 5 and 6 and in the average of all layers of changing vertical leakance between model layers 1 and 2 by factors of 0.2 and 5................................. 46
31. Diagram showing root-mean-square errors of water-levelresiduals resulting from changes in the storage coeffic ient of all model layers.................................. 47
32. North-south water-level profiles showing the effect in model layers 2, 3, and 4 of changing the storage coefficient of all model layers by factors of 0.2 and 5.................. 48
33. North-south water-level profiles showing the effect in model layers 5 and 6 and in the average of all layers of changing the storage coefficient of all model layers by factors of 0.2 and 5...................................... 49
34. East-west water-level profiles showing the effect in model layers 2, 3, and 4 of changing the storage coefficient of all model layers by factors of 0.2 and 5.................. 50
35. East-west water-level profiles showing the effect in model layers 5 and 6 and in the average of all layers of changing the storage coefficient of all model layers by factors of 0.2 and 5...................................... 51
TABLES
Table 1. Results of statistical analysis of water-level residualsof the steady-state calibrated model for 1980 conditions..... 11
2. Results of statistical analysis of water-level residualsof the transient calibrated model for the period 1957-82..... 16
3. Results of statistical analysis of water-level residuals showing the sensitivity of the steady-state calibrated model to changes in the transmissivity of all model layers... 26
4. Results of statistical analysis of water-level residualsshowing the sensitivity of the steady-state calibrated modelto changes in vertical leakance between all model layers..... 27
5. Results of statistical analysis of water-level residualsshowing the sensitivity of the transient calibrated model to changes in the storage coefficient of all model layers....... 52
CONVERSION FACTORS AND ABBREVIATIONS
For the convenience of readers who prefer System) units rather than the inch-pound units be converted by using the following factors:
to use metric (International used in this report, values may
Multiply inch-pound units By To obtain metric units
foot (ft) 0.3048
foot per mile (ft/mi) 0.1894
foot squared per day (ft2/d) 0.09290
inch (in.) 25.4
million cubic feet (Mft3 ) 28,320
million cubic feet per 0.3278
day (Mft3/d)
square mile (mi2 ) 2.590
mile (mi) 1.609
meter (m)
meter per kilometer (m/km)
meter squared per day (m2/d)
millimeter (mm)
cubic meter (m3 )
cubic meter per
second (m3/s)
square kilometer (km2 )
kilometer (km)
Sea level: In this report "sea level" refers to the National Geodetic Verti cal Datum of 1929 (NGVD of 1929) a geodetic datum derived from a general adjustment of the first-order level nets of both the Iftiited States and Canada, formerly called "Sea Level Datum of 1929."
VI
CALIBRATION AND SENSITIVITY ANALYSIS OF A GROUND-WATER FLOW MODEL
OF THE COASTAL LOWLANDS AQUIFER SYSTEM IN PARTS OF LOUISIANA,
MISSISSIPPI, ALABAMA, AND FLORIDA
By Angel Martin, Jr., and C.D. Whiteman, Jr.
ABSTRACT
The coastal lowlands aquifer system, consisting of aquifers in sediments of Miocene age and younger in southern Louisiana, Mississippi, and Alabama and in western Florida, is being studied as part of the U.S. Geological Survey's Gulf Coast Regional Aquifer-System Analysis program. This report describes the calibration and sensitivity analysis of a multilayer, finite-difference ground-water flow model developed to quantify flow in the aquifer system.
Initial calibration of the model by trial-and-error was followed by use of a parameter estimation program. Transmissivities of permeable zones within the aquifer system, vertical leakances between the zones, and the storage coefficient of the aquifer system were varied to obtain the best match between model-simulated and measured water levels for the period 1958-82. The mean error, root-mean-square error (RMSE), and standard deviation of the residuals between model-simulated and measured water levels were used to evaluate the progress of calibration, with greatest weight given to minimizing the RMSE.
Best calibration results were obtained in the model layer that repre sents the uppermost part of the aquifer system. Good results also were obtained in the subsurface part of the rest of the aquifer system where water- level gradients are relatively low and uniform. Calibration of the model is relatively poor in the outcrop areas of the lower part of the aquifer system and near some major pumping centers, where steep and irregular water-level gradients are difficult to simulate at the scale of the model.
Sensitivity analysis of the calibrated steady-state and transient models was performed by varying values of transmissivity, vertical leakance, and storage coefficient; the same parameters were varied during calibration. Changes in RMSE were used as the primary indicator of sensitivity. Near the calibrated values, the model is most sensitive to changes in transmissivity and almost as sensitive to changes in vertical leakance. By layer, the model is most sensitive to changes in the transmissivity of layer 2, which represents the upper part of the aquifer system, and in the vertical leakance between layers 1 and 2, which represents flow between a constant-head upper boundary and the top of the aquifer system. If transmissivity or vertical leakance is changed throughout the model, however, the effects are accentuated in the lower layers because much of the water flowing in these layers passes through and is affected by the overlying layers. The model is relatively insensitive to changes in the coefficient of storage because only a small part of the total flow is derived from storage.
INTRODUCTION
The coastal lowlands aquifer system is Geological Survey's Gulf Coast Regional program (Grubb, 1984). The GC RASA program tions that present a regional overview of the conditions in the principal aquifers of the objective of this study is to describe lowlands aquifer system. A digital flow model was the principal tool used to investigate
Aquif<sr
Gulf ground-water
flow
Calibration and sensitivity analysis wasdevelopment of a ground-water flow model of t^e coastal lowlands aquifer system. Calibration is the process by which model input parameters are adjusted so that model output matches observei conditions to a desired degree of accuracy. Sensitivity analysis involves changing the values of individual model inputs to observe the effects of the changes on model output. If a small change in input results in a large change in output, the model is said to be sensitive to that property. Conversely, if a large change in input produces only a small change in output, the model is insensitive to that
being studied as part of the U.S.-System Analysis (GC RASA)
includes a series of investiga- hydrogeologic and geochemical
Coastal Plain. A majorflow within the coastal
(McDonald and Harbaugh, 1988) in the aquifer system.
an integral part of the
property. Sensitivity analysis is useful in evaluating the confidence to beplaced in the accuracy of the values of aquifer properties adjusted during the calibration process.
This report describes the calibration ar|id sensitivity analysis of the ground-water flow model used to quantify flow in the coastal lowlands aquifer system. A steady-state version of the model for 1980 was used for calibration of transmissivities of permeable zones and vertical leakances between per meable zones within the aquifer system. A transient version of the model was used to calibrate the storage coefficients of! the sands, gravels, silts, and clays that make-up the aquifer system. Steady-state and transient calibra tions were based on miniinizing the mean errorr, RMSE, and standard deviation of the residuals between model-simulated and measured water levels. Initial calibration by trial-and-error was followed by use of an optimization program to check and improve calibration.
Sensitivity of the model was determined by comparing water-level residuals produced by the calibrated model with residuals produced by the model with one aquifer property changed. Propertie&s tested for sensitivity were aquifer transmissivity, vertical leakance, and storage coefficient. Results were evaluated on the basis of mean error, RMSE, and standard deviation.
Calibration and sensitivity results are comparisons of model-simulated and measured areal distribution of input parameter values
presented as statistical v/ater levels, plots showing the
and hydrographs.
HYDROGBOLOGY
The coastal lowlands aquifer system consists of sediments of Miocene age and younger which occur above the Jackson awl Vicksburg Groups of late Eocene and Oligocene age in parts of Louisiana, Misj-dssippi, Alabama, and Florida and adjoining offshore waters (fig. 1). Sand oaxirring near the top of the under lying Vicksburg Group in some areas was included in the aquifer system.
34'
33'
32'
31'
30
£
29'
28'
95° I
OKLA
HOM
A
94°
93°
92°
91C
90°
89°
88°
87°
.0
BO
UN
DA
RY
OF
THE
CO
AS
TA
L
LO
WL
AN
DS
ST
UD
Y A
RE
AG
ULF
20
I40 1
60
I80
100
MIL
ES
0 20
40
60
80
100
KIL
OM
ET
ER
S
Figure 1. Location
of t
he s
tudy are
a and ma
jor
stre
ams.
The aquifer system is characterized by off-lapping, coastward thickening wedges of fluvial, deltaic, and marine sediments of Miocene age and younger (Martin and Whiteman, 1989, p. 3). Deltaic px>cesses have been dominant during deposition of these sediments. Advanc:.ng deltaic fronts pushed the shoreline and its associated beach, dune, and lagoonal deposits seaward while blankets of fluvial sediments were deposited <xi the coastal plain inland, and extensive marine deposits formed offshore. Hie coastal lowlands aquifer system is underlain by clay, silt, and lime bods of the Jackson and Vicksburg Groups. The Jackson and Vicksburg Groups act as a lower confining unit below the coastal lowlands aquifer system. Flow across the confining unit has a negligible effect on flow in the coastal lowlands aquifer system (Williamson, 1987).
The coastal lowlands aquifer system cons lists primarily of alternating beds of sand and gravel, silt, and clay. The most extensive sand beds cannot be traced with certainty for more than 30 to 50 mi. Dip of individual sand beds is southerly, ranging from about 10 to 50 ft/mi in the outcrop area and shallow subsurface in the northern part of the study area. Dip increases to the south and with increasing depth to over 100 ft/mi at depths of more than 3,000 ft. Individual clay and silt confining beds are not areally extensive. Gravel is common in the northern part of the study area but becomes finer and less common southward. Grain size of the sand also decreases southward, grading to sandy clay or silt and finally to clay. Martin and Whiteman (1989) describe in detail the hydrogeologic setting and regional flow in the coastal lowlands aquifer system.
Regional ground-water flow in the coastal lowlands aquifer system is primarily from north to south. Figure 2 is a generalized schematic diagram showing predevelopment regional flow. Average annual rainfall ranges from about 48 in. on the northern and western parts of the study area to more than 65 in. on coastal areas of Louisiana, Mississippi, and Alabama. Water entering the aquifer system in upland terrace> areas that is not discharged locally to streams or by evapotranspiration moves downward to the regional flow system and then toward discharge areas eit lower altitudes in the coastal plain and along major stream valleys. In places, pumping of ground water hasaltered the natural predevelopment gradients much of the natural discharge area.
and has initiated recharge in
Saltwater occurs downdip in the marine and deltaic parts of the aquifer system. Freshwater moving downdip from recharge areas tends to push the saltwater ahead of it, but the downdip movement of saltwater is blocked where the sand beds in the aquifer system pinch-out or are displaced by faulting. Water can move out of the downdip part of the sand beds only by upward leakage through overlying sediments (Martin and Whitoman, 1989, p. 4).
DESIGN OF THE GROUND-WATER FLOW MODEL
The coastal lowlands aquifer system was divided into five permeablezones, A-E as shown in figure 2, in order tol use a digital ground-water flow model to investigate the lateral and vertical distribution of flow (Weiss and Williamson, 1985). The massive coastward-thLckening wedge of sediments was first divided into zones in intensively-pumpad areas (Baton Rouge, Louisiana,
OKLA.I ARK
GA.
'COCO
RECHARGE
GULF OF MEXICO MISSISSIPPI RIVER
A / DISCHARGE
LLUUJJ
-^^r-~"-H-Z-Z-~-~-/ '.'« * *.'. "' .*'.* ', '. .1.' .* 0'.,- o *' ' ,' .'o."-'-'^-*- c;Vj-_-~-_-_-_-~-i]//., .".' " ; . , ° .- ' .' , .'.- ';:
Not to scaleEXPLANATION
BASE OF THE FLOW SYSTEM STUDIED
GENERAL DIRECTION OF REGIONAL
GROUND-WATER FLOW
Figure 2. Generalized hydrogeologic section showing zonation of the aquifer system and direction of regional ground-water flow.
to the east and Houston, Texas, to the west) pumpage information. Emphasis was placed on top and progressively thicker downward for flow system where most of the freshwater flow extended along the strike of the beds by zone as the same percentage of the total aqui centers. The zones pinch-out in the updip pattern, where progressively older bands of The lower zones pinch-out in the downdip of individual sand beds and the rise through zone pinches out downdip along the line at contain water with a dissolved-solids (milligrams per liter).
concent; ration
The five permeable zones have been designated, from youngest to oldest,zone A (Holocene-upper Pleistocene deposits),
c*i the basis of water-level and r taking the zones thinnest at the
besrt resolution in the part of theoccurs. The zones were then
maintaining the thickness of eacher system as at the pumping
di]?ection simulating the outcropsediment are exposed (fig. 2).
direcrtion reflecting the pinching out:he section of saltwater. Each
which all sand beds in the zonegreater than 10, OCX) mg/L
zone B (lower Pleistocene-upperPliocene deposits), zone C (lower Pliocene-upjper Miocene deposits), zone D (middle Miocene deposits), and zone E (lower Miocene-upper Oligocene deposits) (Grubb, 1987, table 1). These permeable zones are defined as hydrogeologic units. The series designations are given as a general indication of relative age, and the zones may contain sediments younger or older than the series age designation.
A finite-difference grid consisting of 78 rows by 70 columns of uniform blocks 5 mi on a side (fig. 3) was constructed for use with the digital flow model. The grid covers an area of 390 mi by 350 mi, or 136,500 mi2 . The model grid covers an area considerably larger than the study area, which consists of 58,400 mi2 inland and 10,100 mi2 offshore. Six layers comprise the model (fig. 4). Layers 2-6 represent permeable zones A-E, respectively. Layer 1, the uppermost layer, represents a constant-head upper boundary.
The water level specified for each node of the constant-head upper boundary (layer 1) is the altitude of the wat;er table at the center of that block. Layer 1 acts as a source or sink for all water entering or leaving the simulated flow system through land surface oliher than that removed by pumpage (fig. 4). All of the lateral model boundaritis except the western boundary and a small part of the northern boundary are taiated as no-flow boundaries. Along the eastern and most of the northern sides, the no-flow boundaries for layers 2-6 are at the updip limit of the out<3?op-subcrop area for each layer. A relatively small amount of water flowing into or out of the northern edge of the model in zone A, in the Mississippi River valley, is accounted for by a specified-flux boundary in model layer 2 (fig. 3). Flows across this bound ary, derived from a model of the Mississippi i River alluvial aquifer (D.J. Ackerman, U.S. Geological Survey, written canmun., 1988), were adjusted at the start of each stress period. The southern boundary for each of these layers is at the line along which water in all sands represented by that layer exceeds a dissolved-solids concentration of 10,000 mg/L.
Analysis of water-level data indicated that flow occurs across the western boundary under natural conditions and ini response to pumpage from zone A(model layer 2) in the Lake Charles area in southwestern Louisiana. In modellayers 2-6, the western model boundary was farmed using general-head-boundarynodes (fig. 3). General-head-boundary nodes permit flow across the boundary
RO
WS
78
EX
PL
AN
AT
ION
+ A
CT
IVE
AR
EA
OF
TH
E M
OD
EL
G
EN
ER
AL-
HE
AD
BO
UN
DA
RY
SP
EC
IFIE
D-F
LU
X B
OU
ND
AR
YC
u-§3
8 O
BS
ER
VA
TIO
N W
ELL
A
ND
WE
LL N
UM
BER
80
100
KIL
OM
ET
ER
S
Figu
re 3
. Fi
nite
-dif
fere
nce
grid with
the
active a
rea of
the m
odel
.
00
ZONE B
iXX
XX
XX
XX
XX
XX
XX
X)P
XX
XX
Xi
ZON
E C
K-X
-XX
XX
XX
XX
XX
XX
XX
XX
XX
X *
XX
XX
Xi
ZON
E DB
y v
v w
ZO
NE
E
MO
DE
L A
1 LA
YE
R
AI M
OD
EL
B1
LAY
ER
CO
NS
TA
NT
HE
AD
XX
XX
X"X
XX
XX
XX
XX
XX
XX
XX
XX
XX
XX
XX
XX
XX
XX
XX
XX
XX
XX
XX
XX
^()
(XX
XX
XX
X
ZON
E D
xxxx
xxxx
xxxx
xxxx
xxxx
xxxx
xxxx
xxxx
xZ
ON
E E
CO
NS
TA
NT
-HE
AD
B
OU
ND
AR
Y
(SO
UR
CE
-SIN
K)
NO
-FLO
W B
OU
ND
AR
Y
EX
PLA
NA
TIO
N
SP
EC
IFIE
D-F
LU
X
BO
UN
DA
RY
GE
NE
RA
L-H
EA
D B
OU
ND
AR
Y
\AA
/ V
V
V
\/
7\A
!AA
Arf
RE
ST
RIC
TIO
N O
F
VE
RT
ICA
L
FLO
W B
ET
WE
EN
LA
YE
RS
DIR
EC
T
FLO
W B
ET
WE
EN
O
UT
CR
OP
A
ND
CO
NS
TA
NT
-HE
AD
L
AY
ER
Figu
re 4
. North-south
and
east
-wes
t schematic
sections t
hrou
gh t
he coa
stal
lowlands
aqui
fer
syst
em
showing
representation of
the permeable
zones
as m
odel
lay
ers
and
the
simu
late
d bo
unda
ry c
onditions.
based on the gradient between model-simulated water levels at the boundary and specified water levels outside of the boundary (McDonald and Harbaugh, 1988, p. 11-1 through 11-27). The specified water levels used in this study were derived from a model of the Texas coastal lowlands aquifer system (P.O. Ryder, U.S. Geological Survey, written commun., 1988) and adjusted at each stress period.
The lower boundary of the model in the northern part of the study area, where freshwater is present throughout the coastal lowlands aquifer system, is at the top of the thick clays of the Vicksburg and Jackson Groups. Water leaking vertically between the underlying Eocene sediments and the coastal lowlands aquifer system is accounted for by a specified-flux boundary in layer 6. The specified fluxes used in this study, adjusted at each stress period, were derived from a model of the underlying Mississippi Embayment aquifer system (J.K. Arthur, U.S. Geological Survey, written commun., 1988). The rate of leakage is small in comparison to the volume of flow in the coastal lowlands aquifer system. To the south, where the lower part of the aquifer system contains saltwater, a no-flow boundary is placed at the bottom of the lowest sand bed containing water with no more than 10,000 mg/L dissolved solids.
Regionally extensive confining units do not occur within the coastal lowlands aquifer system as defined for this study, but large water-level differences do occur vertically within the flow system. In order to simulate the vertical restriction to flow by interbedded clays within the aquifer system, clays in vertically adjacent zones were treated as being equivalent to a single clay bed between the zones (Bredehoeft and Finder, 1970, p. 884). Leakance between the zones was computed by dividing an average value of vertical hydraulic conductivity for the clays by the total thickness of clay between midpoints of the zones. Computing leakance in this way provides an area! variability in leakance that corresponds to area! variations in the restriction to vertical flow that occur within the aquifer system.
MODEL CALIBRATION
Model calibration consists of adjusting model input parameters, initial conditions, and boundary conditions so that the model simulates the aquifer system to a desired degree of accuracy. The calibration process involves matching water levels, water-level changes, hydraulic gradients, flow rates, volumetric budgets, or a combination of these. The model simulating the coastal lowlands aquifer system was calibrated to 1980 steady-state conditions by matching model-simulated water levels to measured water levels. Although not at true steady state throughout the area in 1980, the rate at which water levels were changing was not considered to be significant in relation to the scale of the aquifer system (Martin and Whiteman, 1989, p. 14). The model was calibrated for transient conditions by matching water levels for the period 1958-87. The mean error, RMSE, and standard deviation of the residuals between model-simulated and measured water levels were used as quantitative comparisons during calibration. The RMSE shows the variation of the residuals about measured water levels. Standard deviation shows the variation of the residuals about the mean of the residuals. The RMSE and standard deviation are defined by:
root-mean-square error =
N
standard deviation =
N
, and
where hs is the nodel-siinulated water level; h is the measured water level; h is the mean of the residuals; and N is the number of water-level pairs com pared. Due to the relatively coarse lateral and vertical discretization of the aquifer system for modeling, a precise water-level match was not expected.
Flow rates and volumetric water budgets cjould not be measured with enough accuracy for quantitative comparison with model results because inherent errors in measurement of surface-water flow in the study area may be greater than total flow in the ground-water system. Model-computed flow ratjes, however, provided qualitative checks or results. The model cannot be considered to be adequately calibrated even though model-simulated and meas ured water levels may closely match if siimilatied flows are not reasonable.
Parameters adjusted by trial-and-error and optimization methods during calibration were the transmissivities of the regional zones, the vertical leakances between zones, and the storage coefficients of the zones. Boundary conditions and pumpage were assumed to be correct and were not changed after initial refinements described in Martin and Whiteman (1989). Transmissivity of each zone was calculated as the product of total sand thickness within the zone and an average lateral hydraulic conductJ.vity of the sands and is pre sented in this report in units of foot squared per day. Vertical leakance between zones was calculated as the average vtartical hydraulic conductivity of the clays within and between the zones divided by the total thickness of clay between midpoints of the zones and is given in units of inverse day (day1 ). The storage coefficient of each zone was calculated by summing the products of the total thickness of sand times an average specific storage for sand and the total thickness of clay times an average specific storage for clay. Storage coefficient is dimensionless. |
Initial adjustments of transmissivity and vertical leakance were made over the full extent of each zone. After preliminary calibration, changes made to improve one part of the model would worsen calibration in other parts of the model. To further refine the calibration, each model layer was divided into areas based on hydrologic distinctions in the corresponding permeable zone, such as the outcrops and areas of intense pumpage.
10
Steady-State Model
Results of steady-state calibration are shown in terms of mean error, RMSE, and standard deviation of water-level residuals for each model layer and for the entire model in table 1. Values of RMSE and standard deviation are similar because the means of the residuals In all model layers are small. Model layer 2 (zone A), on average, is most accurately simulated in the model. Model layers 3-5 (zones B-D) show increasing values of RMSE, indicating that calibration of the model becomes progressively poorer downward. The RMSE of layer 6 (zone E) is somewhat lower than that of layer 5, but the mean error of layer 6 is higher than that of any other layer.
Table 1. Results of statistical analysis of water-level residuals of the steady-state calibrated model for 1980 conditions
Modellayer
23456
All
Numberof
observations
34973164278132996
Meanerror(feet)
-0.246.595.69.74
17.093.81
Root-mean-square error
(feet)
14.7533.2240.1156.1245.8439.74
Standarddeviation
(feet)
14.7532.5639.7056.1242.5439.56
The largest differences between measured and model-simulated water levels occur in the outcrop areas of the lower zones and near major pumping centers (fig. 5). Ground-water gradients are steep over much of the outcrop areas of the lower zones because of relatively large topographic relief. Gradients are also steep near major pumping centers. The finite-difference blocks used in the model are large in relation to the distances over which large water-level changes occur, making it difficult or impossible to accurately simulate water levels in these areas.
Transient Model
The transient simulations use nine stress periods to simulate the period 1898-1987. The first three stress periods, each 20 years in length, simulate the calibration period 1898-1957. Six periods, each 5 years in length, simulate the calibration period 1958-87. Water-level and pumpage data prior to 1958 were too sparse for quantitative use in calibration. Available water- level and pumpage data were used qualitatively for the first three stress periods to adjust pumpage to cause simulated water levels at the end of each period to match available measured levels as closely as possible. This allowed the fourth, stress period to begin with transient conditions similar to those present in the aquifer system.
11
Values of transmissivity and vertical leakance from the steady-state calibration were used as initial values in transient calibration and were not significantly changed as a result of the trial-and-error transient calibration process. Initial values of storage coefficients for the transient simulation were determined using values of specific storage of 1.0 X 10 " 6 for sand (Lohman, 1972, p. 8) and 4.0 X 10~6 for clay (Ireland and others, 1984, p. 148-149). Storage coefficients were adjusted uniformly throughout the model to achieve the best match between model-simulated and measured water levels.
Hydrographs comparing model-simulated and measured water levels for the period 1958-87 were used throughout transient calibration. In addition to wells with long-term hydrographs, many wells were measured once to a few times during the calibration period. Water-level measurements made near the ends of stress periods 3-9 (1957, 1962, 1967, 1972, 1977, 1982, and 1987) were com pared with model-simulated water levels for the ends of the stress periods. Mean error and RMSE of the water-level residuals calculated after each model run during the trial-and-error calibration process were used to guide changes in input parameters for the next model run. Water-level measurements were not available for 1987 for the entire model area at the time of calibration of the transient model, so statistics for stress period 9 (1983-87) were not used in the calibration process.
The model is relatively insensitive to changes in storage coefficient, as discussed later, so calibration was less useful in refining the values of storage coefficient than it was for transmissivity and vertical leakance. Conversely, a broad range of uncertainty in the values of storage coefficient is acceptable because of the insensitivity. Final values of storage coeffic ients used throughout the aquifer system were one-half of the initial values. Although sensitivity analysis showed that mean error and RMSE could be reduced slightly by lowering the storage-coefficient values to one-tenth of the values actually used, this was not done because the resulting values would be unrea sonably low.
Statistical Optimization Program
Following trial-and-error calibration, a statistical optimization program (Durbin, 1983) was applied to the model in an attempt to improve the calibration. This program uses a modified Gauss optimization technique (Wilde and Beightler, 1967, p. 299) based on minimizing an objective function propor tional to the RMSE. The program executes the model many times, changing the value of a single parameter for each run. Parameter changes may be made for the entire model, by layer or by areas within layers. After each run, a comparison of water-level changes versus an initial base run is made and the tested parameter is then returned to its former value. Testing of the param eters is continued until each parameter to be optimized has been tested once. The program solves simultaneously for new values of all the parameters to be optimized based on the previous tests. The new parameter values should improve the water-level match. The new values for all of the tested param eters are then used to make a model run that forms the base run for another round of parameter changes. This iterative process is continued until change
13
of the RMSE of the entire model from one iteration to the next is less than aspecified level (closure criterion) or until a is exceeded. The closure criterion allows thelittle improvement in model results occurred as a result of the latest itera tion. Specifying a maximum number of iteratiors prevents the program from running indefinitely if the closure criterion cannot be met.
A.total of 27 parameters were initially included transmissivity of each layer and of model layers, with several layers divided and-error calibration. Storage coefficients whole and not by layer or area. Because directly proportional to the number of were eliminated if no significant changes of the first few program iterations.
into WE see
execution parametcsrs
specified number of iterations optimization program to stop if
selected for optimization. These vertical leakance between each pair
subareas as in the trial- optimized for the model as atime of the program is
being optimized, parameters parameters occurred duringtiie
The optimization program made relatively small changes from trial-and- error calibration results. Figure 6 shows the results from an optimization test in which transmissivities and vertical leiakances were optimized by layer and the storage coefficient was optimized for the entire model from trial-and- error calibrated values. Transmissivity values for all layers except 4 and 5 and vertical leakance between all layers excepb 1 and 2 were slightly higher than for the trial-and-error calibration. Thei largest changes were for trans missivity of layer 2 (1.1 times the initial value) and vertical leakance between layers 4 and 5 (slightly less than 1.1 times the initial value). Transmissivity for layers 4 and 5 each decreased slightly and storage coeffi cient increased slightly. Improvement in the RMSE of the entire model was about 0.2 ft, achieved after 15 iterations. The optimization program did not significantly improve the trial-and-error calibration in terms of RMSE, and the tested model input parameters were not significantly changed.
Results of Calibration
The best optimization program results were used in the final version of the model. Values of mean error, FMSE, and standard deviation for each model layer and for the entire model for stress periods 3-8 are shown in table 2. Values of standard deviation generally show the same relation to RMSE as in the steady-state calibration.
As with the steady-state version of the model, layer 2 (zone A) showsthe best match and layer 5 (zone D) shows themeasured water levels. Significant difference MS in mean error and RMSE occur for the same model layer in different stress jjeriods. This may be primarily due to the varying number and locations of mecisured water levels available for comparison.
poorest match of simulated and
layerjs iNegative mean-error values for all layer)s in stress period 3 indicate that model-simulated water levels generally ate lower than measured water levels. One explanation for this might be that simulated pumpage in stress period 3 may have been greater than the actual pumpage. Positive mean-error values for stress periods 4-8 indicate that midel-simulated water levels generally are higher than measured water levels.
14
36.4
0
Ul
RO
OT
-ME
AN
-SQ
UA
RE
ER
RO
R
OF
W
AT
ER
-LE
VE
L
RE
SID
UA
LS
- X
- - S
TO
RA
GE
O
F
AL
L
LA
YE
RS
TR
AN
SM
ISS
IVIT
Y
Layer
2 <
>
Layer
3 D
Layer
4 A
Layer
5 O
Layer
6
VE
RT
ICA
L
LE
AK
AN
CE
B
ET
WE
EN
Laye
rs
1 an
d 2
Laye
rs
2 an
d 3
Laye
rs 3
and
4
Laye
rs 4
and
5
Laye
rs 5
an
d 6
ITE
RA
TIO
N
oc UJ
35.9
516
Figu
re 6
. Re
sult
s fr
om t
he o
ptim
izat
ion
program
show
ing
chan
ges
in p
aram
eter
val
ues
and
resulting
chan
ges
in r
oot-
mean
-squ
are
error of wat
er-l
evel
residuals.
Model layer
23456
All
23456
All
23456
All
23456
All
23456
All
the transient
Number of
observations
30155627020508
3449616718074
861
283121252222116994
348151301365150
1,315
34843716429555
calibrated model for the period 1957-82
Mean error (feet)
Stress period 3, 1
-0.82
Root-mean- igquare error
(feet)
957
18.27-17.22 33.17-18.00 32.42
-.23 44.57-6.00 38.05-4.81 27.86
Stress period 4, 1062
3.187.06-3.4916.9229.45
18.3530.5232.7046.5457.30
7.45 34.36
Stress period 5, 1967
3.876.94.39
12.4521.09
15.2527.8634.5248.0848.26
7.29 35.37
Stress period 6, :
-1.363.566.80
18.0118.168.68
.972
19.5731.3138.0653.8246.8640.00
Stress period 7, 1977
2.24 17.704.97 25.383.65 39.559.16 37.1527.304.74
54.4727.58
Standard deviation (feet)
18.2528.3626.9644.5737.5727.44
18.0829.6932.5143.3549.1633.54
14.7526.9834.5146.4443.4134.61
19.5231.1037.4550.7143.2039.04
17.5624.8939.3836.0047.1327.17
16
Table 2. Results of statistical analysis of water-level residuals of the transient calibrated model for the period 1957-82 Continued
Number Mean Root-mean- StandardModel of error square error deviationlayer observations (feet) (feet) (feet)
Stress period 8, 1982
2 349 -0.18 14.75 14.753 73 7.09 33.40 32.644 164 6.74 40.31 39.745 278 2.64 55.84 55.776 132 19.70 46.82 42.48
All 996 4.92 39.83 39.52
Overall model, stress periods 3-8, 1957-82
23456
All
1,973539
1,0171,179
5215,229
1.113.411.78
11.6120.395.77
17.4530.4936.5450.9048.9235.96
17.3029.4535.9349.1043.9934.50
A histogram of the differences between model-simulated and measured water levels for stress periods 3-8 (fig. 7) shows that layer 2 has the best fit of simulated to measured water-level altitudes with 1,615 of 1,973 simu lated water levels within 20 ft of the measured water levels. The poorest fit is for layer 5 with 460 of 1,179 simulated water levels differing by more than 40 ft from the measured water levels.
Hydrographs of measured and simulated water levels (figs. 8 and 9) show comparisons between measured and simulated water levels through time at dis crete points within the aquifer system. The simulated water levels used in these figures were interpolated to the locations of the measured water levels by distance-weighted averaging of water levels computed at the centers of the grid blocks encompassing the measured level. The model-simulated water levels, which represent the average water level in several grid blocks, do not closely match the measured water levels, but general water-level trends are reproduced. Model calibration, illustrated by figures 8 and 9 and table 2, is satisfactory given the limitations of the model and of the input and compari son data. Further calibration effort would not significantly improve water- level matches.
The areal distribution of calibrated values of transmissivity, vertical leakance, and storage coefficient are shown in figures 10-12. Calibrated transmissivity values (fig. 10) increase downdip in all zones as zone thickness and total sand thickness increase. Maximum transmissivity values generally occur some distance north of the downdip limit of the zones because the thickness of the zones decreases near their downdip limit. Transmissivity
17
900
800
700
600
oI I
en
LJtoCQ O
LJ CQ
500
400
300
200
100
1111 II
f
SIMULATED WATER 'LEVELS LOW ^
-
-
-
-
-
-
..Jl J
/
/
/
/
/
/
5 ^"" N '» '
> *" /
xJ X P>f ' xS / VJ /
/
J
f
s
j
f
/
/
/
s
s
f
I '
**'!..
> j .
NL
i i i i
SIMULATED WATERLEVELS HIGH
10DEL TOTAL NUMBER OFAVER OBSERVATIONS
Z?l 2 1,973
PHI 3 539
E2B 4 1,017
life 5 1,179
BE
N
S
-<ji
'<!
ll
*<*
1
^ 6 521
-
I
:;|
ill I n
> =: " : :;
> $ ^ :: < £
-120 -100 -80 -60 -40 -20 0 ;20 40 60
RESIDURLS, Ilk! FEET80 100 120
Figure 7. The distribution of the simulated and measured
18
residuals between model- water levels.
MODEL STRESS PERIOD
3 40
ft50: 100JKS*P 200
%- is« « 300
_
. . j:!
I
': ' '','
'.'. '.
1 I
:j ;
i i
r
'. '. '. '. X
WELL Cu-635
CALCASIEU PARISH. LOUISIANA
MODEL LAYER 2
Location shown on figure 3
MODEL-SIMULATED WATER LEVEL
MEASURED WATER LEVEL
140
MODEL-SIMULATED WATER LEVEL
WELL Or-42
ORLEANS PARISH. LOUISIANA
MODEL LAYER 2
Location shown on figure 3
MEASURED WATER LEVEL
- YEAR8 -
MODEL-SIMULATED WATER LEVEL
WELL EB-301
EAST BATON ROUGE PARISH, LOUISIANA
MODEL LAYER 3
Location shown on figure 3MEASURED WATER LEVEL
Figure 8. Comparison of model-simulated and measured water levels for selected wells in model layers 2 and 3.
19
MODEL STRESS PERIOD
I______I______I______I______I
MODEL-SIMULATED WATER LEVEL
WELL EB 90
EAST BATON ROUGE PARISH. LOUISIANA
MODEL LAYER 4
Location shown on figure 3
MEASURED WATER LEVEL
MEASURED WATER LEVEL
WELL EB 468
EAST BATON ROUGE PARISH. LOUISIANA
MODEL LAYER 5
Location shown on figure 3
MODEL SIMULATED WATER LliVEL
WELL M011
PERRY COUNTY MISSISSIPPI
MODEL LAYER 6
Location shown on ligure 3
MODEL SIMULATED WATER LEVELMEASURED WATER LEVEL
Figure 9. Comparison of model-simulated and measured water levels for selected wells in model layers 4, 5, and 6.
20
distributions generally foOJcw the same pattern as total sand thickness within each zone (Martin and Whiteman, 1989, figs. 14-18). Overall, transmissivities are highest in the upper zones and decrease downward. Transmissivities for large areas in zones A and B are greater than 30,000 ft2/d, whereas most values in zone E are less than 10,000 ft2/d.
Calibrated vertical leakance (fig. 11) varies widely within the model area. Values range from less than 10~6 day-1 to more than 10"4 day'1 . Ver tical leakance is highest in and near the outcrop areas of the permeable zones and generally decreases downdip as zone thickness and total clay thickness within the zones increase generally the thicker the clay, the lower the vertical leakance. Variations of this pattern occur where values of clay vertical hydraulic ccnductivity are significantly higher or lower than the regional average for the zone. Overall, vertical leakance decreases in the lower permeable zones. Most values between the constant-head boundary and zone A are greater than 10~ 5 day'1 , whereas most values between the lower zones are less than 10 ~ 5 day1 .
Calibrated storage coefficients range from 1.0 X 10" 5 to 5.0 X 10"3 (fig. 12). Values increase downdip as sand and clay thicknesses increase in each zone. Because the specific storage of sand is lower than that of clay, the storage coefficient of a zone varies inversely with the sand percentage of the zone.
SENSITIVITY ANALYSIS
The approach to sensitivity analysis and the presentation of results in the following section were patterned, in part, on a report describing sensi tivity analysis of the Southeastern Coastal Plain aquifer system (Pernik, 1987). Results of the sensitivity analyses from other flow-modeling studies covering parts of the study area were used as a guide in changing aquifer properties.
Method of Study
Sensitivity of the model to changes in transmissivity, vertical leak ance, and storage coefficient was evaluated using steady-state and transient versions of the calibrated model. Transmissivity and vertical leakance were varied from 0.01 to 100 times calibrated values in the steady-state model. Storage coefficient was varied from 0.1 to 100 times the calibrated value using the transient model. Analyses of water levels, flow rates, and volu metric water budgets during calibration indicate that transmissivity of layer 2 (zone A) and vertical leakance between layers 1 and 2 (the constant-head upper boundary and zone A) were more sensitive than equivalent properties of other layers.
In sensitivity analysis, transmissivity was varied uniformly for all layers through the range of values and independently in layer 2. Similarly, vertical leakance was varied uniformly throughout the model and independently between layers 1 and 2% Storage coefficient was varied uniformly in all layers through the range of values. The mean error and RMSE of the residuals
24
between model-simulated and measured water levels were used to quantify the sensitivity test results. As during calibration, the standard deviation of the water-level residuals was also calculated to show variation of water-level residuals about the mean.
Transmissivity and Vertical Leakance
Model sensitivity to changes in transmissivity and vertical leakance through the range of values for all model layers is shown in tables 3 and 4 and figure 13. Values of FMSE and standard deviation indicate that, in terms of water-level changes, the model is more sensitive to reductions of transmis sivity and vertical leakance than to increases in these parameters (tables 3 and 4). Figure 13 shows the effects of varying both transmissivity and vertical leakage throughout the model. The FMSE ranges from less than 40 ft for the calibrated model (parameter multipliers equal to 1.0) to 65 ft when both parameters are increased by two orders of magnitude (factor of 100) to 4,750 ft when both parameters are decreased by two orders of magnitude (factor of 0.01). Within an order of magnitude of the calibrated values, the model is more sensitive to changes in transmissivity than to changes in vertical leakance.
The effects of varying the transmissivity of layer 2 and the vertical leakance between layers 1 and 2 are shown in figure 14. The pattern of changes in RMSE closely resembles the pattern formed when the parameters of all layers are varied (fig. 13) except that the range of changes in FMSE is less. The FMSE increases from the calibrated value of slightly less than 40 ft to 47.8 ft when both parameters are multiplied by 100 and to more than 1,050 ft when both parameters are multiplied by 0.01.
NkDrth-south and east-west water-level profiles of individual model layers and the water-level average of all layers showing the effects of varying calibrated values of transmissivity of all layers by a factor of 5 are shown in figures 15 to 18. Higher transmissivity values generally produce lower water-level gradients, with lower water levels in the recharge areas and higher water levels in the discharge areas. Conversely, lower transmissivity values produce higher water-level gradients, with higher water levels in the recharge areas and lower water levels in the discharge areas. This effect is shown most clearly in the north-south profiles (figs. 15 and 16). Varying transmissivity in all layers accentuates water-level changes in the deeper layers because much of the water flowing into or out of the deeper layers passes through and is affected by overlying layers.
In contrast to the sensitivity of the model to decreases in transmis sivity shown by water-level changes, total flow circulating within the aquifer system is more affected by increases in transmissivity than by decreases. Increasing calibrated transmissivities of all layers by a factor of 5 increases the total flow circulating within the aquifer system under 1980 conditions by about 63 percent, from 380 to 621 Mft3/d. Lowering transmis sivities to one-fifth the calibrated values reduces total flow in the aquifer system by about 22 percent, from 380 to 297 Mft3/d.
25
Table 3. Results of statistical analysis of wajber-level residuals shewingthe sensitivity of the steady-state calibratJBd model to changes in thetransmissivity of all model layers
[Multiplier is the factor by which the calibratedvalues of transmissivity were varied]
Multiplier
0.01
.10
1
10
100
Modellayer
23456
All
23456
All
23456
All
23456
All
23456
All
Numberof
observations
34973164278132996
34973164278132996
34973164278132996
34973164278132996
34973164278132996
1Mean Root-mean-error square error(feet) (feet)
-186.69 926.75-201.77 470.60-113.34 332.32-93.17 397.60
-.31 159.60-124.92 619.70
-28.31 128.58-39.78 100.10-33.74 116.32-35.98 155.459.93 76.24
-27.07 127.54
-.24 14.756.59 33.425.69 40.11.74 56.12
17.09 45.843.81 39.74
24.21 36.0941.04 71.4752.13 84.5240.73 81.5936.91 67.9436.29 66.83
40.23 54.2043.09 100.7459.80 108.8443.59 97.8215.05 73.3841.22 84.20
Standarddeviation(feet)
909.05428.13313.37387.23160.21607.29
125.6192.49111.67151.5075.87124.70
14.7532.5639.7056.1242.5439.56
26.8058.9266.7470.8257.2556.15
36.3891.7091.2387.7272.0973.46
26
Table 4. Results of statistical analysis of water-level residuals showingthe sensitivity of the steady-statevertical leakance
[Multiplier
between all model
is the factor by
calibrated model to changeslayers
which the calibrated valuesvertical leakance were
Multi- Modelplier layer
23
0.01 456
All
234
.10 56
All
23
1 456
All
23
10 456
All
23
100 456
All
Number of
observations
34973164278132996
34973
164278132996
34973
164278132996
34973164278132996
34973164278132996
Mean error(feet)
-367.79-209.21-342.94-512.36-542.59-415.75
-77.37-16.74-58.35
-121.81-104.75-85.88
-.246.595.69.74
17.093.81
22.2426.8132.7740.2857.9434.07
24.7727.9935.7145.0462.6737.48
varied]
Root-mean- square error
(feet)
432.25258.14373.98544.37549.80465.08
101.4057.4185.99150.23127.36116.25
14.7533.4240.1156.1245.8439.74
32.2741.7456.4364.8473.9853.98
36.1344.8163.6570.6579.1159.10
in
of
Standard deviation(feet)
227.42152.29149.64184.2589.13
208.55
65.6355.2963.3688.0872.7278.39
14.7532.5639.7056.1242.5439.56
23.4232.2246.0750.9046.1841.89
26.3435.2052.8357.5048.4645.69
27
100
60 LINE OF MEAN Interval in is variable
0.01 0.1 1
MULTIPLIER FOR TRANSMISS VITIES OF ALL LAYERS
Figure 13. Root-mean-square errors of ting from changes in transmissivity model layers.
100
water-level residuals resul- and vertical leakance of all
28
100
cvj
ROOT-MEAN-SQUARE
.01 0.1 1 10 MULTIPLIER FOR TRANSMISSIVITY OF MODEL LAYER 2
100
Figure 14. Root-mean-square errors of water-level residuals re sulting from changes in the transmissivity of model layer 2 and the vertical leakance between model layers 1 and 2.
29
ALTITUDE ABOVE OR BELOW SEA LEVEL, IN FEET
en o o o
ro o o
GJ Oo
A -UO O1o o
8
gV)
a m~n 33O
Oc
m o O m
O o m
m0)
s
8
§
8
8
A'
AA
' A
450
40
0
35
0
h-
LU £
300
2 LTJ
250
LU <
20
0LU
(/
) O
150
LU
CO or
O LU O CO < LU
Q
100 50 50
-
100
150
Lay
er 5
- -
LAN
D S
UR
FAC
E
WA
TER
LE
VE
LS:
V
Incr
ease
d
Cal
ibra
ted
R
edu
ced
VE
RTI
CA
L S
CA
LE G
RE
ATL
Y
EX
AG
GE
RA
TED
Lay
er 6
Ave
rag
e o
f al
l la
yers
5010
015
020
050
100
150
200
5010
015
0 20
0
DIS
TA
NC
E
FR
OM
S
OU
TH
ED
GE
O
F M
OD
EL
, IN
M
ILE
S
Figure 16
. No
rth-
sout
h wa
ter-
leve
l pr
ofil
es s
howi
ng t
he eff
ect
in mod
el l
ayers
5 an
d 6
and
in t
he a
vera
ge o
f all
layers o
f changing t
he t
rans
miss
ivity of
all mod
el l
ayer
s by f
acto
rs o
f 0.
2 an
d 5.
B1
BB
1 B
to
UJ UJ uu > UJ UJ
CO UJ
CD UJ o CO -tu- <
400
35
0
300
250
200
150 50 50
100
150
LA
ND
SU
RF
AC
E
WA
TER
LE
VE
LS:
Incr
ease
d C
alib
rate
d R
educ
ed
VE
RT
ICA
L S
CA
LE
GR
EA
TLY
E
XA
GG
ER
AT
ED
Laye
r 2
i.i,
.11
.1,1
Lay
er 3
! /T
EN
N.
OK
.LA
.I
AR
K. f- T
\
^
/
i \
I"'"
' J
MIS
S.i
AL
A.\
w-<L
;=,,J
TE
X
B
Laye
r 4
100
200
30
04
00
100
200
300
400
10
0
20
0
30
04
00
DIS
TA
NC
E F
RO
M W
ES
T E
DG
E O
F M
OD
EL
, IN
MIL
ES
Figu
re 17. East-west
wate
r-le
vel
profiles s
howi
ng t
he eff
ect
in model l
ayers
2, 3, an
d of c
hanging
the
tran
smis
sivi
ty of
all model
layers by
factors
of 0.2
and
5.
B'
BB
1 B
(A)
CA)
ULJ
CO O
LU m DC O LU O
CD LU
O ID
40
0
350
30
0
250
20
0
150
100 50
I
50 100
-
150
LAN
D
WA
TER
LE
VE
LS:
......
......
......
in
crea
sed
Cal
ibra
ted
------
Red
uce
d
VE
RT
ICA
L S
CA
LE G
RE
ATL
Y
EX
AG
GE
RA
TE
D
GA
PS
IN
DIC
ATE
IN
AC
TIV
E N
OD
ES
Laye
r 5
1,1,1,1,1,1,1,1
Laye
r 6
100
200
300
400
100
200
30
04
00
100
20
0
30
0
40
0
DIS
TA
NC
E
FR
OM
WE
ST
ED
GE
O
F M
OD
EL
, IN
MIL
ES
Figure 1
8. E
ast-
west
water-level p
rofi
les
show
ing
the effect
in mo
del la
yers
5 a
nd 6
and
in
the a
vera
ge o
f al
l layers o
f ch
angi
ng t
he t
ransmissivity of
all
mod
el l
ayer
s by
fa
ctor
s of 0.2 a
nd 5
.
Increasing and decreasing only the transmissivity of layer 2 by a factor of 5 produces relatively small changes in the water levels of all layers (figs. 19-22). The effects of these changes ai?e relatively uniform throughout the model and are not accentuated in the deeper layers, as when the transmis- sivities of all layers are changed.
iNorth-south and east-west water-level profiles showing the results of
increasing and decreasing vertical leakance be'bween all layers by a factor of 5 from calibrated values are shown in figures !23-26. In general, these figures show that higher vertical-leakance values increase water levels throughout the aquifer system and lower values decrease water levels. Much of the effect of changing vertical leakances results from the effect these changes have en flow between the aquifer system and the constant-head upper boundary. As with transmissivity, changes in vertical leakances of all layers produce accentuated water-level changes in the deeper layers.
Total flow circulating in the aquifer system is more affected by increases in vertical leakance than by decreases. Increasing the calibrated values of vertical leakage by a factor of 5 increases total flow in the aquifer system by about 47 percent, from 380 to 558 Mft3/d. Lowering vertical leakances to one-fifth (0.2 times) the calibrated values reduces total flow in the aquifer system by about 22 percent, from 380 to 296 Mft3/d.
Varying vertical leakance between the constant-head upper boundary and layer 2 from 0.2 to 5 times the calibrated value produces relatively small and uniform changes In water levels throughout the aquifer system (figs. 27-30). These changes are not strongly accentuated in the lower layers.
Storage Ooefficisnt
Model sensitivity to changes in storage cx>efficient of all model layers is shown in table 5 and figures 31-35. The relative insensitivity of the model to changes in storage coefficient is besrt shown in figure 31. Large changes in storage-coefficient values result in relatively small changes in RMSE and standard deviation of individual layers and of all layers combined. Table 5 shows that mean error and RMSE could be lowered slightly by reducing the storage-coefficient values to one-tenth of the calibrated values; but, as noted in the section en calibration of the mcxtel, the resulting values would have been unreasonably low. The model is more sensitive to increases in the value of the storage coefficient above the calibrated value than to decreases.
Water-level profiles show the results of increasing and decreasing the calibrated value of storage coefficient by a factor of 5 (figs. 32-35). Water-level residuals resulting from changes in storage coefficient are greater in the lower model layers representing deeper zones. This is probably the result of the cumulative effect of water moving into or out of storage in the upper zones that impact the flow distribution in the lower zones.
Water levels rise throughout the aquifer system with increases in the storage coefficient and fall with decreases. More water enters the modeled aquifer system from storage with an increase [in storage coefficient, resulting in higher overall water levels. When storage; coefficients decrease fromcalibrated values, water levels must decline induce more recharge.
34
throughout the aquifer system to
A1
A1
CO
Oi
LU
LU UL Z _f
LU LU .j.
LU CO 0 _l
LU CD CC 0 LU O CD LU
Q b
4E>u
40
0
35
0
30
0
25
0
200
15
0
10
0
50 0
50
100
150
-
LAN
D S
UR
FAC
E
WA
TER
L
EV
EL
S:
......
......
......
......
.....
incr
ease
d
Calib
rate
d_ _
__
__
__
R
educe
d
VE
RT
ICA
L S
CA
LE G
RE
ATL
YE
XA
GG
ER
AT
ED
_
| If I/ r If // f j i
J
. .*/
-
Laye
r 2
i i
, i
i i
, i
0 50
10
0 15
0 20
0
OKLA.i ARK
Layer
3Layer
4
50100
150
200
50100
150
200
DIS
TA
NC
E F
RO
M S
OU
TH
ED
GE
OF
MO
DE
L,
IN M
ILE
S
Figure 1
9. North-south wa
ter-
leve
l pr
ofil
es s
howing t
he e
ffect
in mod
el l
ayer
s 2,
3»
and
4 of
changing
the
tran
smis
sivi
ty of model la
yer
2 by f
acto
rs o
f 0.2
and
5.
A1
A
GO a\
45
0
400
35
0
LJJ
30
0
LU >
LU
25
0
<
200
LJU
CO g LJJ
CD CC
O
LJJ
O
CO
150
100
50 50
100
150
Laye
r 5
V
LAN
D S
UR
FAC
E
WA
TER
LE
VE
LS
. ...
......
......
...
incr
ease
d
Cal
ibra
ted
-= - - - - -
Red
ucetL
VE
RT
ICA
L S
CA
LE
GR
EA
TLY
E
XA
GG
ER
AT
ED
Lay
er 6
Ave
rage
of
all
laye
rs
5010
015
0 20
050
100
150
200
5010
0 15
020
0
DIS
TA
NC
E F
RO
M S
OU
TH
ED
GE
OF
MO
DE
L,
IN M
ILE
S
Figure 2
0. N
orth
-sou
th wat
er-l
evel
profiles
show
ing
the
effect i
n mo
del
layers 5 and
6 and
in t
he a
verage of
all
layers o
f changing t
he t
ransmissivity of mod
el l
ayer
2 by
fact
ors of 0.2
and
5«
B1
BB
' B
CJ
45
0
400
350
LU LU
300
Lu
250
LU LU
CO 3 LU
CD CC
O LU O CD
< LU
Q
200
150
100 50 50
100
150
- -
LA
ND
SU
RFA
CE
WA
TER
LE
VE
LS
:
......
......
......
In
crea
sed
Cal
ibra
ted
______
Red
uced
VE
RTI
CA
L S
CA
LE
GR
EA
TLY
E
XA
GG
ER
ATE
D
Lay
er 2
Lay
er 3
OK
I.A
.!
AR
K
Lay
er 4
100
200
300
400
100
200
300
400
0 10
0 200
300
400
DIS
TA
NC
E
FR
OM
W
ES
T E
DG
E O
F
MO
DE
L,
IN M
ILE
S
Figure 2
1. E
ast-
west
wat
er-l
evel
profiles
show
ing
the
effect i
n mo
del
layers 2
, 3,
and
of c
hang
ing
the
transmissivity of mo
del
layer
2 by f
acto
rs o
f 0.2
and
5.
B'
BB
1 B
450
CO
00
uu
uu uu >
uu UJ
CO g uu CD
CC o uu > o 00 iJLI
Q
VE
RT
ICA
L S
CA
LE
GR
EA
TL
Y
EX
AG
GE
RA
TE
D
GA
PS
IN
DIC
AT
E
INA
CT
IVE
NO
DE
S
50
-
10
0
-
150
100
200
300
400
100
200
300
400
100
200
300
400
DIS
TA
NC
E F
RO
M W
ES
T E
DG
E O
F M
OD
EL
. IN
MIL
ES
Figure 22.
Eas
t-we
st w
ater-level p
rofiles
showing
the
effect i
n model la
yers
5 a
nd 6
and
in t
he a
verage of
all
laye
rs of
changing t
he t
rans
miss
ivit
y of
model layer 2
by
factors
of 0.2 a
nd 5
-
A1
A1
CO
VO
400
350
LU LU
30
0
LU
LU LU i LU
CO CC o LU O
CO < LU
O
250
200
150
100
50 50
-
100
-
150
LA
ND
SU
RF
AC
E
WA
TER
LE
VE
LS:
......
......
......
. in
crea
sed
Cal
ibra
ted
------
Red
uce
d
VE
RTI
CA
L S
CA
LE G
RE
ATL
Y
EX
AG
GE
RA
TED
Lay
er 2
Lay
er 3
TE
NN
._
_
OK
LA
.i
AR
K. f'~
"7"~
" \
Lay
er 4
5010
015
020
050
100
150
200
5010
015
020
0
DIS
TA
NC
E F
RO
M S
OU
TH
ED
GE
OF
MO
DE
L,
IN M
ILE
S
Figure 23.
Nor
th-s
outh
water-level p
rofiles
show
ing
the effect i
n model
laye
rs 2
, 3,
an
d 4 of
cha
ngin
g vertical leakance between
all model layers by
fact
ors of
0.2
and
5.
A1
AA
1 A
450
400
350
i- LU UJ
300
z Lu
250
LU <
200
LU O
150
_i LU
CO § 10
0LU 1
50
UJ
Q K b <50 10
0
150
Laye
r 5
LA
ND
SU
RFA
CE
WA
TER
LE
VE
LS:
Incr
ease
dC
alib
rate
d
Red
uced
VE
RT
ICA
L S
CA
LE G
RE
ATL
Y
EX
AG
GE
RA
TED
Laye
r 6
50
100
150
200
0 50
10
0 15
0 20
050
10
0 15
0 20
0
DIS
TA
NC
E F
RO
M S
OU
TH
ED
GE
OF
MO
DE
L,
IN M
ILE
S
Figure 2
4. North-south water-level pro
file
s sh
owin
g th
e ef
fect
in mo
del
laye
rs 5
and
6
and
in t
he a
vera
ge of
all la
yers
of ch
angi
ng ver
tica
l leakance bet
ween
all
mod
el l
ayer
s by f
actors o
f 0.2
and
5.
B1
BB
1 B
1 LU
LU LU Z _T LU LU LU g LU
CO CC O LU 0
CO LU
Q h- t3
<
tuu
400
350
300
250
200
150
100
50
0
50 100
ic;n
-
LAN
D S
UR
FA
CE
WA
TE
R L
EV
ELS
:
......
......
......
......
.....
incr
ease
d
Cal
ibra
ted
VE
RT
ICA
L S
CA
LE G
RE
ATL
YE
XA
GG
ER
AT
ED
- -
\ l\\\
\ J
i Y»
\ i \
,
1
/
'II
- u
_ -
Laye
r 2
, 1
,1,
,1
,1
,1
,1
,1
.
Laye
r 3
Laye
r 4
100
200
300
40
010
0 20
0 30
0 40
010
02
00
300
40
0
DIS
TA
NC
E F
RO
M W
ES
T
ED
GE
O
F
MO
DE
L,
IN M
ILE
S
Figu
re 2
5. East-west w
ater
-lev
el p
rofi
les
show
ing
the effect i
n model
layers 2
, 3»
and
of changing vertical l
eaka
nce be
twee
n all model
laye
rs b
y factors of 0.2
and
5-
B1
BB
1 B
to
LU Z LU >
LU LU
CO I
LU m
cc
O LU s m LU -a-
450
400
350
300
250
200
150
100 50
0 50 100
150
LA
ND
SU
RF
AC
E
WA
TER
LF
VE
LS
:
......
._...
......
. In
crea
sed
Cal
ibra
ted
-------
Red
uced
VE
RT
ICA
L
SC
AL
E
GR
EA
TLY
EX
AG
GE
RA
TE
D
GA
PS
IN
DIC
AT
EIN
AC
TIV
E N
OD
ES
Laye
r 5
Laye
r 6
100
200
300
400
100
200
300
400
100
200
300
400
DIS
TA
NC
E F
RO
M W
ES
T E
DG
E O
F M
OD
EL
, IN
MIL
ES
Figu
re 2
6. East-west water-level p
rofiles
show
ing
the
effe
ct i
n model
laye
rs 5
and
6 a
nd
in t
he a
vera
ge o
f al
l la
yers
of
changing ver
tica
l leakance bet
ween
all
model l
ayer
s by
factors of
0.2 a
nd 5
*
A1
A
CO
h- LU UJ
LL Z _T UJ
UJ < UJ
CO o 1
UJ m DC 0 UJ 0 CO UJ
O Z>
h- b <
<*ou
400
350
300
250
200
150
100 50
0
50 100
150
LAN
D S
UR
FAC
E
WA
TER
LE
VE
LS:
......
......
......
......
....
Incr
ease
d
Calib
rate
d
- -------
Reduce
d
VE
RT
ICA
L S
CA
LE G
RE
ATL
YE
XA
GG
ER
AT
ED
-
/ /i i/ ij II HI Ir If; 1 j IJF La
yer
2
i .
i ,
i .
i
0 50
10
0 15
0 20
0
Layer
3
i.i.i
Laye
r 4
50
100
150
20O
50100
150
200
DIS
TA
NC
E F
RO
M S
OU
TH
ED
GE
OF
MO
DE
L.
IN M
ILE
S
Figu
re 27.
Nor
th-s
outh
wat
er-l
evel
profiles
showing
the ef
fect
in model
layers 2
, 3,
and
4 of
cha
ngin
g ve
rtic
al l
eakance be
twee
n model
layers 1
and 2
by
fact
ors of
0.2
and
5.
A1
450
400
350
l- LU UJ
300
Lu
250
LU LU w
O HI
CO cc
O in O CO < LU
~O
^
ID
200
150
100 50 50 100
150
Laye
r 5
LA
ND
SU
RFA
CE
WA
TER
LE
VE
LS
:
......
......
......
. in
crea
sed
Cal
ibra
ted
------
Red
uce
d
V
VERTICAL SCALE GREATLY
EXAGGERATED
Laye
r 6
5010
015
0 20
050
100
150
200
50100
150
200
DIS
TA
NC
E F
RO
M S
OU
TH
ED
GE
OF
MO
DE
L,
IN M
ILE
S
Figure 28.
Nor
th-s
outh
wat
er-l
evel
pro
file
s showing
the
effect i
n model
laye
rs 5
and
6
and
in t
he a
vera
ge of
all
laye
rs o
f changing vertical
leakance between model layers
1 and
2 by
fac
tors
of 0.2
and
5«
B1
BB
1 B
Ul
UJ
UJ u. UJ
UJ
45
0
40
0
350
300
250
20
0
O
150
LU m
DC
O UJ O
m UJ
Q ID
100
50
50
-
100
- -
LAN
D S
UR
FAC
E
WA
TFR
LE
VE
LS
: ...
......
......
....
Incr
ease
d
-
Cal
ibra
ted
-------
Red
uced
VE
RTI
CA
L S
CA
LE G
RE
ATL
Y
EX
AG
GE
RA
TED
Lay
er 2
La
ye
r 3
La
ye
r 4
100
200
30
0
400
100
20
0
300
40
010
0 200
300
400
DIS
TA
NC
E
FR
OM
WE
ST
ED
GE
O
F M
OD
EL
, IN
M
ILE
S
Figure 2
9. East-west w
ater
-lev
el p
rofiles
showing
the
effect i
n model
layers 2, 3,
and
of c
hang
ing ve
rtic
al l
eakance
between model
layers 1 and
2 by
fac
tors
of 0.
2 an
d 5.
B'
BB
' B
450
400
350
h- LU
300
ULI
> LU LU
03 9 LU
CD tr
O LU
> O CO LU ~Q~
_D
250
200
150
100 50
0 50 100
150
- -
LA
ND
SU
RFA
CE
WA
TER
LE
VE
LS
:
......
......
......
. in
crea
sed
Cal
ibra
ted
_
__
__
_
Red
uce
d
VE
RTI
CA
L S
CA
LE G
RE
ATL
YE
XA
GG
ER
ATE
D
GA
PS
IN
DIC
ATE
INA
CT
IVE
NO
DE
S
Laye
r 5
100
200
300
400
100
200
300
400
100
200
300
400
DIS
TA
NC
E F
RO
M W
ES
T E
DG
E O
F M
OD
EL
, IN
MIL
ES
Figu
re 30.
Eas
t-we
st wat
er-l
evel
pro
file
s sh
owin
g the effect i
n model
layers 5
and
6 a
nd
in t
he a
verage of all
layers of
chan
ging
ver
tica
l le
akan
ce between model l
ayers
1 an
d 2
by f
acto
rs o
f 0.
2 an
d 5«
ROOT-MEAN-SQUARE ERROR, IN FEET
"0 H-
uo
rt- I
o<=>
CO rt O
CD
CD 3
O ^S"Mj O
o s H. pj CD rt D CD
H& & CO
CD [-S
CO CO' &
rt D*
oq
O
CD COH- Drt-81
A1
03
UJ LU >
LU UJ
CO 3 LU
CO cc
o UJ o CD
450
400
350
300
250
200
150
100 50
0
50 100
150
LAN
D S
UR
FAC
E
WA
TER
LE
VE
LS:
......
......
......
. in
crea
sed
Cal
ibra
ted
__
__
__
_
Red
uced
VE
RT
ICA
L S
CA
LE G
RE
ATL
Y
EX
AG
GE
RA
TE
D
Layer
2La
yer
3
;TENN._
-I"
\
. i
\
Laye
r 4
5010
015
020
050
100
150
200
50
100
150
200
DIS
TA
NC
E F
RO
M S
OU
TH
ED
GE
OF
MO
DE
L.
IN M
ILE
S
Figu
re 32. North-south water-level profiles
show
ing
the ef
fect
in mo
del
laye
rs 2
, 3, and
4 of cha
ngin
g th
e st
orag
e co
effi
cien
t of
all mod
el l
ayer
s by
fac
tors
of
0.2
and
5.
A'
A
III
450
400
350
300
LU
250
> LU _l <
200
HI
CO o _l
UJ m
rr
O LU
O m LU
Q
150
100 50 50 100
150
Laye
r 5
LA
ND
SU
RFA
CE
WA
TER
LE
VE
LS:
......
......
......
. In
crea
sed
Cal
ibra
ted
-------
Rod
ucod
1 '\
\/ \ V
VE
RT
ICA
L S
CA
LE
GR
EA
TLY
E
XA
GG
ER
ATE
D
Lay
er 6
! /T
EN
N
OK
I.A
i
AR
K
r ~
"T
Ave
rage
of
all
laye
rs
5010
015
020
050
100
150
200
5010
015
020
0
DIS
TA
NC
E F
RO
M S
OU
TH
ED
GE
OF
MO
DE
L,
IN M
ILE
S
Figu
re 3
3. N
orth
-sou
th wat
er-l
evel
profiles
showing
the effect i
n mo
del
layers 5
and
6 and
in t
he a
verage of
all
layers o
f ch
angi
ng t
he s
torage c
oefficient o
f al
l model
layers b
y factors of 0.2
and
5.
B1
BB
1 B
Ul
O
450
400
-
150
100
200
300
400
100
200
300
400
100
200
300
400
DIS
TA
NC
E F
RO
M W
ES
T E
DG
E O
F M
OD
EL
, IN
MIL
ES
Figure 34. East-west water-level p
rofi
les
showing
the ef
fect
in model
laye
rs 2
, 3,
and
of changing
the
storage
coefficient of
all
mod
el l
ayer
s by
factors of 0.2
and
5.
B1
BB
1 B
Ol
LU
UU LU
0) s LU
CD CC
O
UU O CD UU o
450
400
350
30
0
250
200
150
100 50 50 100
- -
LAN
D S
UR
FAC
E
WA
TER
LE
VE
LS.
......
......
......
. In
crea
sed
Cal
ibra
ted
______
Red
uce
d
VE
RT
ICA
L S
CA
LE
GR
EA
TLY
EX
AG
GE
RA
TE
D
GA
PS
IN
DIC
ATE
INA
CT
IVE
NO
DE
S
Laye
r 5
Lay
er 6
100
20
0
300
400
100
200
30
0
40
010
0 20
0 3
00
40
0
DIS
TA
NC
E F
RO
M W
ES
T E
DG
E O
F M
OD
EL
, IN
MIL
ES
Figu
re 35. East-west water-level p
rofi
les
show
ing
the
effect i
n model
layers 5
and
6 a
nd
in t
he a
vera
ge of
all
laye
rs o
f ch
angi
ng t
he s
torage c
oefficient o
f all model
layers b
y fa
ctor
s of
0.2 a
nd 5
.
Table 5. Results of statistical analysis of water-level residuals showingthe sensitivity of the transient calibrated. model to changes in thestorage coefficient of all model layers
[Multiplier is the factor by which the calibrated values of storage coefficient were varied]
Multi plier
0.10
1
10
100
Model layer
23456
All
23456
All
23456
All
23456
All
Number of
observations
1,973539
1,0171,179
5215,229
1,973539
1,0171,179
5215,229
1,973539
1,0171,179
5215,229
1,973539
1,0171,179
5215,229
Mean error (feet)
1.01
Root-mean- square error
(feet)
17.462.85 30.35.86 36.48
10.32 50.8118.61 48.105.02 35.80
1.11 17.453.411.78
11.6120.39
30.4936.5450.9048.92
5.77 35.96
2.66 17.367.918.5119.7626.8610.61
12.33
32.1537.8552.5853.4237.50
21.2614.55 36.2718.04 45.3627.8931.3219.07
57.9356.9042.26
Standard deviation (feet)
17.3129.3835.9849.2843.8434.53
17.3029.4535.9349.1043.9934.50
17.0730.3235.9648.5645.6434.58
18.8132.4940.0250.6846.7236.75
Increasing calibrated storage coefficients of all layers by a factor of 5 increases the amount of water derived from jstorage for the period 1898-1987 from 110,000 to 433,000 Mft3 . Lowering storage coefficients to one-fifth the calibrated values decreases the amount of wator derived from storage for the same time period from 110,000 to 23,000 Mft3 .| Although these results show a large difference in the amount of water derived from storage due to changes in storage-coefficient values, this amount of wafcer is only a small part of the total simulated flow in the aquifer system. With storage-coefficient values 5 times the calibrated values, the volume of water derived from storage is about 5 percent of the total simulated flow. With storage-coefficient values one- fifth of calibrated values, the volume of water derived from storage is less than 1 percent of the total simulated flow.
52
SUMMARY
A six-layer, finite-difference ground-water flow model was developed, calibrated, and tested for sensitivity in order to quantify regional flow in the coastal lowlands aquifer system of Louisiana, Mississippi, Alabama, and Florida. The model was initially calibrated by trial-and-error by varying values of transmissivity of the aquifers, vertical leakances between the aquifers, and the storage coefficients of the aquifer system in order to match model-simulated and measured water levels under steady-state and transient conditions. The mean error, RMSE, and standard deviation of the residuals between model-simulated and measured water levels were used to quantitatively evaluate the progress of calibration.
After trial-and-error calibration, a statistical optimization program was used to see if calibration could be improved. The program improves model calibration by minimizing an objective function which is proportional to the RMSE. The optimization program did not significantly change trial-and-error calibration results.
The model was calibrated best in layer 2, which represents the upper, youngest section of the coastal lowlands aquifer system. Calibration generally worsened in the lower model layers that represent older sections of the aquifer system. Relatively high residual errors in water levels in the lower model layers are largely due to topographic variations in the outcrop areas of the permeable zones represented by these layers. Large topographic variations within a grid block covering 25 mi2 produce steep, irregular water- level gradients which are difficult to reproduce in the model. Steep water- level gradients and relatively high water-level residuals may also occur near major pumping centers.
Sensitivity analyses of the calibrated steady-state and transient models show the effects of varying values of transmissivity, vertical leakance, and storage coefficients. In terms of mean error, RMSE, and standard deviation, results show that within an order of magnitude of calibrated values the model is most sensitive to changes in transmissivity. The model is more sensitive in terms of water-level change to reductions of transmissivity and vertical leakance than to increases in these parameters. The model is relatively insensitive to changes in storage coefficient over a wide range of values because only a small part of the total flow in the aquifer system is derived from storage.
When values of transmissivity or vertical leakance are changed one layer at a time, results obtained during calibration show that transmissivity of layer 2 and vertical leakance between layer 2 and the constant-head upper boundary are most sensitive. When transmissivity or vertical-leakance values are changed in all layers at the same time, the deeper model layers become more sensitive because flow between the constant-head upper boundary and the deeper layers is affected by the changes made in all intervening layers.
53
REFERENCES
Bredehoeft, J.D., and Finder, G.F., 1970, Digital multiaquifer groundwater systems: A quasi Water Resources Research Journal, v. 6, no
Durbin, T.J., 1983, Application of Gauss algorithm and Monte Carlo simulation to the identification of aquifer parameteirs: U.S. Geological Survey Open-File Report 81-688, 30 p. i
analysis of areal flow in three-dimensional model: 3, p. 883-888.
GulfGrubb, H.F., 1984, Planning report for theanalysis in the Gulf of Mexico coastal p Geological Survey Water-Resources
Coast regional aquifer-system ain, United States: U.S.
Report 84-4219, 30 p.Invest d .gations
1987, Overview of the Gulf Coast regional aquifer-system analysis, in Vecchioli, John, and Johnson, A.I., eds., Aquifers of the Atlantic and Gulf Coastal Plain: American Water Resoijirces Association, Monograph no. 9, p. 101-118.
Ireland, R.L., Poland, J.F., and Riley, F.S., Joaquin Valley, California, as of 1980: Professional Paper 437-1, 93 p.
Lohman, S. W., 1972, Ground-water hydraulics: Professional Paper 708, 70 p.
1984, Land subsidence in the San U.S. Geological Survey
U.S. Geological Survey
Martin, Angel, Jr., and Whiteman, C.D., Jr., 1989, Geohydrology and regional ground-water flow of the Coastal Lowlancls aquifer system in parts of Louisiana, Mississippi, Alabama, and Flojrida A preliminary analysis: U.S. Geological Survey Water-Resources Ifrwestigations Report 88-4100, 88 p.
McDonald, M.G., and Harbaugh, A.W., 1988, difference ground-water flow model:of Water-Resources Investigations, book
A modular three-dimensional finite- U.S Geological Survey Techniques
6, chap. Al, 576 p.
Pernik, Maribeth, 1987, Sensitivity analysis £>f a multilayer, finite- difference model of the Southeastern GoaWtal Plain regional aquifer system; Mississippi, Alabama, Georgia, atxl South Carolina: U.S. Geological Survey Water-Resources Investigations Report 87-4108, 53 p.
Weiss, J.S., and Williamson, A.K., 1985, Subdivision units into layers for simulation of 23, no. 6, December 1985, p. 767-774.
grot nd-waterof thick sedimentary flow: Ground Water,
Wilde, D.J., and Beightler, C.S., 1967, Foundations of optimization: Englewood Cliffs, New Jersey, Prentice-Hall, 480 p.
Williamson, A.K., 1987, Preliminary simulation of ground-water flow in the Gulf Coast aquifer systems, south-central United States, in Vecchioli, John, and Johnson, A.I., eds., Aquifers!of the Atlantic and Gulf Coastal Plain: American Water Resources Association, Monograph no. 9., p. 119- 137.
54