Case Study/

Effect of Sea-Level Rise on Salt Water Intrusion

near a Coastal Well Field in Southeastern Florida
by Christian D. Langevin1 and Michael Zygnerski2

Abstract
A variable-density groundwater ow and dispersive solute transport model was developed for the shallow
coastal aquifer system near a municipal supply well eld in southeastern Florida. The model was calibrated for
a 105-year period (1900 to 2005). An analysis with the model suggests that well-eld withdrawals were the
dominant cause of salt water intrusion near the well eld, and that historical sea-level rise, which is similar to
lower-bound projections of future sea-level rise, exacerbated the extent of salt water intrusion. Average 2005
hydrologic conditions were used for 100-year sensitivity simulations aimed at quantifying the effect of projected
rises in sea level on fresh coastal groundwater resources near the well eld. Use of average 2005 hydrologic
conditions and a constant sea level result in total dissolved solids (TDS) concentration of the well eld exceeding
drinking water standards after 70 years. When sea-level rise is included in the simulations, drinking water standards
are exceeded 10 to 21 years earlier, depending on the specied rate of sea-level rise.

Introduction
There is little dispute that global mean sea level has
been rising, and there is recent evidence to suggest that
the rate of rise is accelerating. Recent satellite altimetry
data collected from 1993 to 2003 show an increased
rate of 3.1 0.7 mm/year (Cazenave and Nerem 2004).
This rate is almost twice the rate observed during the
20th century (1.7 0.5 mm/year; Bates et al. 2008), but
owing to the relatively short period of time, it is possible
that part of the increased rate could be due to natural
variability. Predictions of future rates of sea-level rise
continue to improve as the science evolves, as new data
are collected, and as associated uncertainties are more
fully addressed. In the Third Assessment Report (TAR),
the Intergovernmental Panel on Climate Change (IPCC)
1 Corresponding author: U.S. Geological Survey, 411 National
Center, Reston, VA 20192; 703-648-4169; fax: 703-648-6693;
langevin@usgs.gov
2 Broward County Environmental Protection and Growth
Management Department, 115 South Andrews Avenue, Fort
Lauderdale, FL 33301.
Received January 2012, accepted September 2012.
Published 2012. This article is a U.S. Government work and is
in the public domain in the USA.
doi: 10.1111/j.1745-6584.2012.01008.x
reported possible increases for the 21st century that range
from 0.24 to 0.88 m, with a median value of about 0.48 m
(Church et al. 2001). As part of the Fourth Assessment
Report (AR4) by the IPCC, Meehl et al. (2007) provide
an estimated range of 0.18 to 0.59 m for the expected rise
in sea level by the end of this century. Bates et al. (2008)
provide insight into the apparent differences between the
2001 and 2007 studies: the upper values of the ranges
(reported in Meehl et al. (2007)) are not to be considered
upper bounds for sea-level rise. Meehl et al. (2007) noted
that dynamic ice ow processes are poorly understood.
For this reason, they did not include Greenland and
Antarctic ice sheet losses in their projections. By including
the effect of land ice, Pfeffer et al. (2008) suggest that
a 2.0 m rise in sea level by the end of the century is
possible if variables are quickly accelerated but that a
0.8 m rise is more plausible. Improving these projections
has been the subject of recent IPCC investigation on ice
sheet instabilities (IPCC 2010).
Several studies have attempted to quantify and char
acterize, in a generic way, the effect of sea-level rise
on salt water intrusion into a coastal aquifer. Using a
steady-state analysis with an analytical solution, Werner
and Simmons (2009) identied the major hydrogeologic

NGWA.org Vol. 51, No. 5GroundwaterSeptember-October 2013 (pages 781803) 781

controls on the impact of sea-level rise on salt water intru
sion in unconned coastal aquifers. They differentiated
between ux-controlled and head-controlled systems and
showed that sea-level rise is more problematic for head-
controlled systems because inland water levels do not rise
with rising sea level. Chang et al. (2011) also found that
for ux-controlled conned aquifers, sea-level rise may
not have an impact on fresh water volumes. Werner et al.
(2012) extended the analysis of Werner and Simmons
(2009) to include unconned and conned aquifers and
proposed quantitative vulnerability indicators that can be
calculated based on boundary condition type and hydro-
geologic parameter values. Webb and Howard (2010) and
Watson et al. (2010) investigated the migration aspect and
response time of salt water movement. Webb and Howard
(2010) focused solely on the head-controlled system as
the consequences are more severe for that case. Their
simulation results indicated that in certain situations, sev
eral centuries may be required for the salt water interface
to reach equilibrium with sea-level change. Watson et al.
(2010) found markedly different response times depending
on the type of indicator. For example, the representative
response time for the vertical center-of-mass was much
shorter than the response time for the toe position, indicat
ing that care should be given to select indicators relevant
to the study purpose. These studies generalize the effect
of sea-level rise on salt water intrusion for hypothetical
and simplied conditions.
Several efforts have addressed the effect of sea-
level rise on a specic coastal setting. Masterson and
Garabedian (2007) predicted the response for the Lower
Cape Cod aquifer system and found that sea-level rise
increased groundwater discharge into streams causing
a reduction in the total volume of fresh water. Using
Werner and Simmons (2009) terminology, the Lower
Cape Cod aquifer would be classied as a head-controlled
system. In contrast, Rozell and Wong (2010) found that
Shelter Island, New York, would act as a ux-controlled
system and that the effects of sea-level rise on the fresh
water volume would be relatively minor. Interestingly,
they found that an increase in sea level might actually
increase the fresh water lens volume. They attributed this
counterintuitive response to the presence of a marine clay
layer that truncates the base of the fresh water lens; thus,
the volume of fresh water in the aquifer is less if the
marine clay layer were absent. As the prescribed sea level
rose in the model, there were no overlying head controls
and so fresh water accumulated in the unsaturated zone.
Vulnerability of low-lying coastal areas to sea-level rise
has been addressed by Lebbe et al. (2008) for the Belgian
coastal plain, by Oude Essink (1999) and van der Meij
and Minnema (1999) for the Netherlands, by Feseker
(2007) for northwestern Germany, and by Giambastiani
et al. (2007) for an unconned coastal aquifer near
Ravenna, Italy. Fujinawa et al. (2009) evaluated the
effect of climate change (including sea-level rise) for the
eastern Mediterranean coastal region of Turkey. Loaiciga
et al. (2012) concluded for the seaside area sub-basin in
Monterey County, California, that groundwater extraction
782 C. Langevin and M. Zygnerski Groundwater 51, no. 5: 781803
would have a larger effect on sea water intrusion than
sea-level rise. For northern Miami-Dade and southern
Broward Counties, a sensitivity analysis by Guha and
Panday (2012) suggests that water levels and chloride
concentrations could increase by as much as 15 and 640%,
respectively, for coastal parts of the Biscayne aquifer. All
of these studies used a mathematical modeling approach to
predict the impact of sea-level rise on salt water intrusion.
This paper adds to our understanding of the impact
of sea-level rise on salt water intrusion by quantifying
historical changes in fresh water resources and quantifying
process sensitivity for a low-lying coastal aquifer in
southeastern Florida subjected to municipal groundwater
withdrawals. The shallow coastal aquifers of southern
Florida, which include the Biscayne aquifer, offer a unique
opportunity to evaluate the effect of sea-level rise; the
limestone aquifer is highly permeable, and thus, effects
on fresh water resources may be seen more quickly than
for less permeable clastic aquifers. Southeastern Florida
also generally ts into the head-controlled category of
Werner and Simmons (2009) because of an extensive
canal network that overlies the entire region; these canals
have been shown to be in direct hydraulic connection with
the underlying permeable aquifers and act as a strong
head control. Combined with a thin unsaturated zone,
high propensity for damaging oods and high rates of
evapotranspiration, there is little volume available in the
thin unsaturated zone for future rises in the water table.
Southern Florida is also representative of many coastal
areas because of its large population. The combined 2009
population of the three counties comprising mainland
southeastern Florida (West Palm Beach, Broward, Miami-
Dade Counties) is about 5.5 million. Collectively, these
conditions suggest that southeastern Florida may be more
highly susceptible to accelerated salt water intrusion
caused by sea-level rise than other coastal areas.
This investigation uses a numerical groundwater ow
and dispersive solute transport model to evaluate the
relative importance of sea-level rise compared to the
other dominant hydrologic processes for a municipal well
eld in southeastern Florida. The model represents the
hydrologic changes that occurred as the area transformed
from a natural coastal environment into an agricultural
setting and then into an urban corridor (Renken et al.
2005). The model was then used to predict the impact
of future rises in sea level on salt water intrusion
near the well eld. Bredehoeft (2003) summarizes the
general premise that model predictions tend to be more
accurate when the calibration period contains events
and conditions, and encompasses time scales that are
comparable to those expected in the future. The model
presented here was calibrated for a 105-year period using
measured heads and salinity concentrations at monitoring
wells. During the calibration period a salt water intrusion
event was observed near the well eld followed by
a subsequent freshening of the aquifer. Also during
this period, sea level rose by about 25 cm, which is
similar to the lower-bound estimate of the IPCC (Church
et al. 2001). The model was calibrated using highly
NGWA.org
Figure 1. Map of study area showing physiographic features, surface water control structures, municipal groundwater wells,
and monitoring wells. Lines in Florida map delineate county areas.

parameterized inversion techniques to help ensure that
the model was a reasonable representation of the physical
system. Challenges encountered with the calibration effort
are described here for others working on sea-level rise
groundwater simulations.
Description of Study Area
This study focuses on the Pompano Beach well
eld in northeastern Broward County, Florida (Figure 1).
The study area is dened as the active model domain
boundary shown in Figure 1. The climate of the area
and southeastern Florida in general is characterized by
distinct wet (May through October) and dry seasons. The
extreme seasonal rainfall variability combined with the
desire to reclaim large parts of the former Everglades for
urban and agricultural uses necessitated the construction
of an extensive water management system throughout
most of southeastern Florida. This water management
system consists of a series of levees, canals, pumps, and
gates, which are used to control the elevation of the water
table. A structure is a spillway, culvert, or weir located
within a canal that can be used to control the water surface
elevation. Primary structures are controlled and operated
by the South Florida Water Management District and by
the U.S. Army Corps of Engineers. Secondary and smaller
drainage features are operated by the county and local
drainage districts. During the wet season and hurricane
events, excess water is released in to the Atlantic Ocean
as a mechanism for providing ood protection. During the
dry season, the canal system is used to provide aquifer
recharge in coastal areas to prevent salt water intrusion
into municipal well elds. The water management system
is also used by the agricultural community during dry
periods as a source of irrigation water. East of the
easternmost control structure, canals are tidally inuenced
and can have salinities close to that of sea water. Tidal
nger canals, which were dredged to provide waterfront
property with ocean access, can be seen in Figure 2 in the
area east of the Pompano Beach well eld.
Prior to the extensive development that occurred
during the 20th century, northern Broward County was
characterized by Everglades fresh water wetlands that
extended from inland areas to the western side of the
Atlantic Coastal Ridge (Parker et al. 1955). The Hillsboro
River and Cypress Creek (presently the Hillsboro Canal
and the Cypress Creek Canal) owed eastward through
low areas in the Atlantic Coastal Ridge called the
Peat Transverse Glades (Parker et al. 1955). With land
elevations exceeding 7 m, it is unlikely that the Atlantic

NGWA.org C. Langevin and M. Zygnerski Groundwater 51, no. 5: 781803 783


Figure 2. Map of the area of interest showing the salt water
intrusion lines mapped by Dunn (2001) for the top of the
production zone. The lines are contours of the 250 mg/L
chloride concentration for different years.
Coastal Ridge in this area would ever have been inundated
by the fresh water wetlands to the west. During the
20th century, the landscape of northern Broward County
changed considerably. What were once the fresh water
wetlands of the Everglades were transformed rst into
agricultural areas and then into the expansive urban
corridor of today (Renken et al. 2005).
Hydrostratigraphy and Aquifer Properties
This study focuses on the highly permeable, shallow
surcial aquifer system, which is the primary source of
potable water in Broward County (Klein and Hull 1978;
Causaras 1985). The underlying Floridan aquifer system,
which is hydraulically separated from the surcial aquifer
system by an extensive conning unit, is not discussed
in this paper or represented in the model. A study on the
effect of long-term (100,000 year) sea-level changes on
the Floridan aquifer system is reported by Hughes et al.
(2009). The surcial aquifer system, which increases in
thickness from west to east, is dened on the top by the
water table and at the bottom by the top of the Hawthorn
conning unit (Fish 1988). The base of the surcial
aquifer system slopes downward from an elevation of
about 40 to 55 m in the western part of the study area
to more than 110 m in the eastern part. In the Pompano
Beach well eld, Fish (1988) dened the base of the
surcial aquifer system at an elevation of about 114 m.
In this paper, all elevations, including those referring to
784 C. Langevin and M. Zygnerski Groundwater 51, no. 5: 781803
sea level, are referenced to the National Geodetic Vertical
Datum (NGVD) of 1929.
The surcial aquifer system in southern Florida
contains the highly transmissive Biscayne aquifer and
the gray limestone aquifer. According to Fish (1988)
and Reese and Cunningham (2000), the gray limestone
aquifer is not present within the study area. The Biscayne
aquifer, however, is present within the study area and is
the primary water producing part of the surcial aquifer
system. Fish (1988) denes the Biscayne aquifer as
that part of the surcial aquifer system in southeast
Florida comprised (from land surface downward) of
the Pamlico Sand, Miami Oolite, Anastasia Formation,
Key Largo Limestone, and Fort Thompson Formation
all of Pleistocene age, and contiguous highly perme
able beds of the Tamiami Formation of Pliocene age
where at least 10 feet of the section is very highly per
meable (a horizontal hydraulic conductivity of about
1,000 ft/d or more).
With this denition, Fish (1988) mapped the base of
the Biscayne aquifer in the western part of the study area
at an elevation of about 37 m. At the Pompano Beach
well eld, Fish (1988) mapped the base of the Biscayne
aquifer at an elevation of about 98 m, which is slightly
higher than the elevation of 122 m suggested by Tarver
(1964, 8) for the Pompano Beach well-eld area.
Implicit in the Fish (1988) denition is that the top
of the Biscayne aquifer coincides with the water table.
Restrepo et al. (1992) and Dunn (2001), however, note
that a blanket of less permeable sand (of the Pamlico Sand
and Anastasia Formation) is present in most areas. They
dene the top of the Biscayne aquifer as being the rst
occurrence of highly permeable limestone. For the present
study, a similar distinction is made and the overlying less
permeable sands are not included as part of the Biscayne
aquifer. Accordingly, this paper discusses three parts of
the surcial aquifer system: the upper part, the Biscayne
aquifer, and the lower part. Ranges of aquifer properties as
summarized from the literature are presented in Table 1.
Fish (1988) constructed a transmissivity map using
values from selected aquifer tests representative of the
surcial aquifer system. The transmissivity estimates used
to construct that map were used here with estimates of
Biscayne aquifer thickness to calculate hydraulic conduc
tivity. These hydraulic conductivities are thought to be
representative of the average hydraulic conductivity over
the entire Biscayne aquifer thickness. Hydraulic conduc
tivities of individual zones within the Biscayne aquifer are
probably much different than these average values. Never
theless, these average values are used as starting hydraulic
conductivities for the numerical model, which were then
adjusted as part of the calibration process.
Salt Water Intrusion near the Pompano Beach Well Field
Construction of the Pompano Beach well eld began
in 1926 with the completion of the rst well in 1927 (Dunn
2001). The well eld was located on the Atlantic Coastal
Ridge because the underlying surcial aquifer system near
NGWA.org
 Table 1
Summary of Aquifer Properties for the Surcial Aquifer System

Property Value References Comment

Kh (upper part of surcial 15 m/d Fish (1988)

aquifer system)
Kh (Biscayne aquifer) 8020,000 m/d Fish (1988, Table 4) Values calculated from multiple
pumping test results and aquifer
thickness at different locations
Kh (lower part of surcial 0.120 m/d Fish (1988)
aquifer system)
Kv:Kh (Biscayne aquifer) 1:7 to 1:49 Camp and McKee, Inc. (1980)
Sy (Biscyane aquifer) 0.0040.30 Fish (1988)
Sy (Biscayne aquifer) 0.0930.25 Camp and McKee, Inc. (1980)
Sy (Biscayne aquifer) 0.200.25 Merritt (1996a) Analysis based on rainfall-event-based
water level uctuations in
Miami-Dade County
L , T (Biscayne aquifer) 110, 0.11 m Langevin (2001, 2003) Calibration of variable-density
groundwater model in Miami-Dade
County
L , T (Biscayne aquifer) 76, 0.03 m Merritt (1995) Calibration of solute transport model in
Miami-Dade County
L , T (Biscayne aquifer) 2.02.5 m Renken et al. (2008) From a tracer test in Miami-Dade
County
n (Biscayne aquifer and 0.370.48 Fish (1988) Analyses performed on core-scale
lower part of surcial samples
aquifer system)
n (surcial aquifer system) 0.20 Merritt (1996b)
n (Biscayne aquifer) 0.20 Langevin (2001)
n (Biscayne aquifer) 0.4 Renken et al. (2008) Based on one-dimensional simulations
of a tracer test in Miami-Dade County

Notes: A range is reported for some properties because more than one value is reported in the literature. Kh is horizontal hydraulic conductivity; Kv is vertical
hydraulic conductivity; Sy is specic yield; L is longitudinal dispersivity; T is transverse dispersivity; n is porosity.

the ridge tends to have better groundwater quality than
areas to the west (Tarver 1964). Five additional production
wells were drilled during the 1950s. By 1972, the well
eld consisted of a total of 16 production wells (Figure 2).
These wells were completed in a production zone of the
Biscayne aquifer that extends from about 22 to 43 m
below sea level. Production well 1 was abandoned in
the mid-1980s (Dunn 2001) because of elevated chloride
concentrations.
Using measured chloride concentrations at monitor
ing wells and an estimate of the vertical chloride con
centration gradient, Dunn (2001) mapped the temporal
evolution of the position of the 250 mg/L isochlor near
the well eld. Contours of the 250 mg/L isochlor at the
top of the production zone (about 22 m below sea level)
are shown in Figure 2 for selected years between 1972 and
1999. The isochlor advanced to its furthest inland position
in 1984 and then moved seaward to its last mapped posi
tion in 1999. Identifying the contributing factors, such as
drought and groundwater withdrawals, to the advance and
subsequent retreat of saline groundwater is not straight
forward as there are likely many factors contributing to
salt water movement.
Relevant data for the Pompano Beach well-eld
area are shown in Figure 3 to summarize the hydrologic
conditions that led to the advance and subsequent retreat
of saline groundwater in the surcial aquifer system.
Rainfall variations have been suggested by Dunn (2001)
as one of the primary drivers for the salt water intrusion
event that began in the mid-1970s. For the 1970 to 1981
period, 11 out of the 12 years had rainfall values less than
the long-term mean, and this period corresponds to a time
of salt water intrusion.
Groundwater withdrawals from the Pompano Beach
well eld are probably a dominant cause of the salt
water intrusion event. From 1950 to 1980, withdrawals
at the Pompano Beach well eld continued to increase.
By 1980, groundwater withdrawals reached 1 105
m3 /d (Figure 3). Based on a simple Theis analysis
of predicted drawdown, Tarver (1964) warned that
withdrawals exceeding about 7.6 104 m3 /d could cause
salt water intrusion and suggested that an expansion
of the well eld to the north and west would reduce
the potential for salt water intrusion by distributing the
withdrawal effects. The withdrawal threshold calculated
by Tarver (1964) was rst exceeded in 1971. In 1984,
the City of Pompano Beach constructed the Palm Aire
well eld about 5 km west of the Pompano Beach well
eld (Figure 1). The late 1980s to the present shows
a redistribution of groundwater withdrawals from the

NGWA.org C. Langevin and M. Zygnerski Groundwater 51, no. 5: 781803 785

Figure 3. Plots of rainfall, groundwater withdrawals, water levels, and TDS concentration for selected monitoring wells.
Pompano Beach well eld to the Palm Aire well eld
(Figure 3). Reluctance by water managers to construct
new well elds in the western part of the county was
due to the occurrence of poor quality groundwater (Howie
1987).
Water levels of the Atlantic Ocean, Cypress Creek
and Hillsboro Canals, and the G-853 monitoring well pro
vide insight into the salt water intrusion event (Figure 3).
Both the Hillsboro and Cypress Creek Canals maintain
relatively constant stages from about 1970 onward. The
Atlantic Ocean, however, shows an increase of about
25 cm from 1900 to 2005. By itself, the rise in sea level
does not explain the salt water intrusion event, but it may
have been a contributing factor. The most striking feature
of the water levels in Figure 3 is the sharp decline in the
G-853 monitoring well, which is located near the center
of the well eld. Water levels in this well remained near
or below sea level for the 1970 to 1990 period. A water
table map constructed by Sherwood et al. (1973) for May
1971 showed water levels 1-m below sea level for much
of the Pompano Beach well-eld area.
An interesting characteristic of the salt water intrusion
event was that salt water intruded more rapidly in the
786 C. Langevin and M. Zygnerski Groundwater 51, no. 5: 781803
production zone than in the layer beneath the production
zone. The G-2055A monitoring well was open to the
production zone and salinity concentrations started to rise
in 1974. The G-2055 monitoring well, located next to
G-2055A, but open in a deeper zone, did not begin to
show elevated salinity concentrations until about 1983.
Data from these two wells indicate the presence of an
isolated salt water wedge in the middle part of the
aquifer.
The City of Pompano Beach owns and maintains the
municipal golf course adjacent to the Pompano Beach
well eld (Figure 2). The golf course is irrigated using
treated waste water. Irrigation rates were intentionally
increased above what is needed to maintain the golf
course in order to provide articial recharge and prevent
salt water intrusion. Irrigation with treated waste water
began in August 1989. The average irrigation rate from
1989 to 2005 is about 4300 m3 /d. Averaged over the
area of the golf course, this rate is about 120 cm/year,
which is similar to the average annual rainfall rate of
about 150 cm/year. The importance of excess golf course
irrigation on minimizing the potential for salt water
intrusion was evaluated with a sensitivity analysis.
NGWA.org
Model Development and Calibration
A variable-density groundwater ow and solute trans
port model was developed for the northern part of
Broward County to evaluate the causes of salt water intru
sion near the Pompano Beach well eld and to determine
if historical sea-level rise was a factor. The model was
then used to predict the effect of alternative rates of
sea-level rise on salt water intrusion. To ensure that the
inversion process had the exibility to extract the most
information from the observation data set, the model was
calibrated for a 105-year simulation period (1900 to 2005)
using a highly parameterized approach. Simulation of this
a long time period is computationally intensive, and so it
is worthwhile to comment on the rationale for choosing
the calibration period length, which was established early
in the study. First, sea level has risen by about 25 cm
over this calibration period; therefore, sensitivity analyses
can be used with the calibrated model to test the effect
of that 25-cm rise on salt water intrusion in the area. It
may not be possible to resolve the importance of sea-
level rise with shorter simulations. Second, it is difcult
to assign initial conditions to salt water intrusion mod
els. Models with long simulation periods tend to be less
sensitive to errors in initial concentrations than models
with short simulation periods. Lastly, there was no way
to quantify how long it would take for saline groundwater
to respond to hydrologic variability. Because the hydrol
ogy changed drastically over the 105-year period, a long
calibration period seemed necessary in order to ensure that
it contained the hydrologic forcings responsible for caus
ing salt water movement. Parameter estimation with ow
and transport observations has not been applied to three-
dimensional sea water intrusion problems (Carrera et al.
2010); however, Dausman et al. (2010) applied automated
inversion techniques for a related problem of buoyancy-
driven plume migration. This study, therefore, is among
the rst to apply sophisticated calibration strategies to a
three-dimensional salt water intrusion model.
Several numerical models of groundwater ow have
been developed for Broward County. Restrepo et al.
(1992) designed a groundwater model to address problems
associated with water supply; however, the model did
not include a variable-density component. Two models
designed to evaluate salt water intrusion in southern
Broward County, south of the present study, are described
by Andersen et al. (1988) and Merritt (1996b). Other
variable-density models developed for nearby areas to
evaluate groundwater ows or salt water intrusion are
described by Langevin (2001, 2003), Dausman and
Langevin (2005), and Guha and Panday (2012).
Simulation Codes
SEAWAT is a coupled version of MODFLOW
and MT3DMS designed to simulate variable-density
groundwater ow and solute transport (Guo and Langevin
2001; Langevin et al. 2003; Langevin and Guo 2006). The
program has been used to address a variety of issues,
such as submarine groundwater discharge (Langevin
2001, 2003) and salt water intrusion (e.g., Shoemaker
NGWA.org C. Langevin and M. Zygnerski Groundwater 51, no. 5: 781803
and Edwards 2003; Rao et al. 2004; Shoemaker 2004;
Masterson 2004; Dausman and Langevin 2005; Hughes
et al. 2010), for example. The simulations reported here
were performed using SEAWAT Version 4 (Langevin et al.
2008), which is based on MODFLOW-2000 (Harbaugh
et al. 2000) and MT3DMS Version 5 (Zheng and Wang
1999; Zheng 2006).
For the present application, the solute concentration
(C) simulated by the model is the total dissolved solids
(TDS) concentration of sea water salts. Fresh water is
assumed to have a TDS concentration of zero; sea water
is assumed to have a TDS concentration of 35 g/L. Fluid
density () is calculated by SEAWAT using a linear rela
tion subject to the constraints that fresh water has a uid
density of 1000 kg/m3 and sea water has a density value
of 1025 kg/m3 . The resulting equation of state used for
all of the simulations reported here is: = f + 0.714 C.
In some instances, chloride concentration measurements
were available. These concentrations were converted to
TDS concentrations using a simple linear relation between
sea water, which has a chloride concentration of about
19,800 mg/L, and fresh water, which is assumed to have
a chloride concentration of zero. A chloride concentration
of 250 mg/L is commonly used as a maximum concentra
tion for potable water. In terms of TDS, this equates to a
concentration of 0.44 g/L.
Many of the preliminary simulations used the implicit
nite-difference solution method in MT3DMS and SEA
WAT to solve the solute transport equation. Later tests
revealed, however, that this solution scheme was caus
ing an excessive level of numerical dispersion, resulting
in a high level of parameter surrogacy, and difculties
were encountered in trying to reproduce observed salin
ity variations in monitoring wells. Parameter surrogacy
occurs when the inversion process adjusts parameter val
ues in order to compensate for errors in the model, such
as numerical dispersion. The simulations reported here
used the explicit third order, Total variation diminishing
(TVD) scheme in MT3DMS and SEAWAT as an alter
native to the standard implicit nite-difference scheme.
TVD is mass conservative and can minimize numerical
dispersion, but because it is an explicit scheme, it is sub
ject to time step constraints and can be computationally
demanding. TVD simulations better represented the pre
sumed level of hydrodynamic dispersion as evidenced by
an improved ability to represent observed salinity varia
tions compared with nite-difference transport solutions.
Related work by Langevin and Hughes (2009) showed that
calibration of a highly parameterized salt water intrusion
model can result in parameter surrogacy, such as hetero
geneity artifacts in the presence of numerical dispersion.
These artifacts can be reduced by using high levels of grid
resolution or TVD schemes that minimize numerical dis
persion and also by using uniform concentration weighting
schemes for calibration instead of assigning weights that
are proportional to the concentration value.
Preliminary simulations of the salt water intru
sion event had difculties reproducing the relatively
quick response of salinity concentrations in groundwater
787
monitoring wells. Numerous attempts to capture the
response with alternative parameterization approaches and
parameter values repeatedly failed until the conceptual
model for transport was revised. The surcial aquifer sys
tem in southern Florida is highly heterogeneous in both the
vertical and horizontal directions. Recent work in south
eastern Florida (Cunningham et al. 2006, 2009; Renken
et al. 2008) has identied the presence of preferential ow
pathways that likely play a key role in transport even
though they comprise only a fraction of the aquifer total
thickness. To accommodate these important groundwater
ow pathways, the dual-domain capabilities in MT3DMS
(and SEAWAT) were used (Zheng and Wang 1999). With
the dual-domain approach, the aquifer is conceptualized
as having a fast moving mobile domain and an immo
bile domain. All advective transport occurs within the fast
domain, and solute exchange between the two domains
occurs based on an exchange coefcient and the con
centration difference. Lu and Luo (2010) demonstrate
the effect of the dual-domain conceptual model on salt
water intrusion simulations. The dual-domain approach
was used for all the simulations reported here.
The salt water intrusion model was calibrated
using the PEST software suite (Doherty 2009a, 2009b).
PEST uses the Gauss-Marquardt-Levenberg algorithm to
estimate parameters by minimizing weighted residuals
between observations and simulated equivalents. To avoid
problems with numerical instabilities and to allow for the
estimation of many more parameters than there are obser
vations, PEST contains several options for regularizing
the problem into one that is tractable. For example, PEST
contains subspace regularization methods (singular value
decomposition [SVD]) as well as Tikhonov methods. For
the present application, the SVD-assist technique (Doherty
2009a, 2009b), which is a combination of both subspace
and Tikhonov methods, was the approach used for model
calibration. Parameter estimation methods based on per
turbation sensitivities can benet greatly from paralleliza
tion (Carrera et al. 2010). To facilitate tractability of the
parameter estimation process, a cluster computer with 232
computer cores was used.
Spatial and Temporal Discretization
The model grid consists of 115 rows and 160
columns (Figure 4). Each model cell is 150 by 150 m.
In the Universal Transverse Mercator (UTM) Zone
17 coordinate system and the horizontal 1983 North
American Datum (NAD 83), the southwest corner of the
model grid is located at x = 570,000 and y = 2,898,350.
There is no rotation of the model grid from the UTM
coordinate system. The model is bounded on the west by
Water Conservation Area 2A, to the north by the Hillsboro
Canal, to the south by the Cypress Creek Canal, and to
the east by the Intracoastal Waterway and the Atlantic
Ocean. Although the extent of the model grid includes
the barrier island system, groundwater ow within the
shallow isolated lens of the barrier island is only roughly
approximated owing to an insufcient grid resolution
relative to the island width.
788 C. Langevin and M. Zygnerski Groundwater 51, no. 5: 781803
Figure 4. Model grid, inland and coastal pilot points, and
layer 1 boundary conditions for stress period 783 (December
2005).
Nine model layers were used to discretize the surcial
aquifer system. Model layers 1 and 2 correspond to the
unconsolidated sediments of low to moderate permeability
that overly the Biscayne. Layers 3 through 8 correspond
to the highly transmissive Biscayne aquifer, and layer 9
represents the lower part of the surcial aquifer system,
which tends to be less permeable than the Biscayne
aquifer. Land surface elevation was estimated using 10
m horizontal resolution, U.S. Geological Survey (USGS)
digital elevation models. The bottom of model layer 1 was
set uniformly at an elevation of 5.0 m. This elevation
was set lower than the lowest anticipated water table
elevation so that wetting and drying problems common to
MODFLOW-based codes could be avoided. The bottom
of model layer 2 was specied using elevation data from
the bottom of layer 2 of an existing Broward County ow
model (Restrepo et al., 1992). Layer 2 of that model also
corresponded to the lower permeability sands overlying
the Biscayne aquifer. Spatial interpolation using estimates
of the bottom of the Biscayne aquifer (Fish 1988) was
used to assign elevations for the bottom of model layer
8. The thickness of the Biscayne aquifer (bottom of
layer 8 subtracted from bottom of layer 2) was then
divided equally among model layers 3 through 8. Spatial
interpolation using estimates of the bottom of the surcial
aquifer system (Fish 1988) was used to assign elevations
for the bottom of model layer 9.
SEAWAT follows the MODFLOW and MT3DMS
convention of stress periods, ow time steps, and
transport time steps (Langevin et al. 2003). Hydrologic
stresses remain constant for each stress period, with the
exception of specied heads. Specied heads are linearly
interpolated within a stress period from starting and
ending head values assigned for each stress period. Time is
further discretized in SEAWAT using transport time steps.
NGWA.org
For each transport time step, SEAWAT rst solves the ow
equation and then solves the transport equation. Although
SEAWAT contains options for iteratively solving the ow
and transport equations until the solution meets a specied
convergence criterion, this option was not used for the
present study.
The 105-year simulation period, beginning January 1,
1900, and ending December 31, 2005, was divided into
783 stress periods. The rst three stress periods represent
the 40 years from January 1, 1900 to December 31, 1940.
The rst stress period represents the time period prior
to the construction of major canals. The second stress
period starts January 1, 1907, which is the approximate
construction date of the Hillsboro Canal, and extends
through December 31, 1929. The third stress period starts
January 1, 1930, which is the approximate construction
date of the Pompano Canal, and ends December 31, 1940.
One ow time step, which can be used in SEAWAT
to control the frequency of writing output, was used
per stress period. Lengths of transport time steps were
calculated during the simulation using a specied Courant
number of 0.75.
Representation of Hydrologic Stresses
Hydrologic stresses were included in the model as
boundary conditions or as internal sources and sinks.
In most instances, representation of hydrologic stresses
required specication of a ux or head-dependent con
dition and the specication of a solute concentration or
ux. Accordingly, each hydrologic stress is discussed
both in terms of its effect on groundwater ow and
solute transport. The hydrologic features and the MOD
FLOW/SEAWAT package used for their representation in
the model are summarized in Table 2.
A simplied linear equation was used to estimate
the Atlantic Ocean stage relative to NGVD 1929 for the
1900 to 1940 period (C. Zervas, National Oceanic and
Atmospheric Administration [NOAA], written communi
cation, 2007):
Stage = 2.39 mm/year (year 2000) + 22.6 mm.
For the remainder of the simulation period, data from
three NOAA tide stations were combined. From January
1941 to June 1981, tide data from the NOAA Miami
Beach tide station (station identication number 8723170)
were used. From August 1981 to August 1992, data
from the Haulover Pier tide station (station identication
number 8723080) were used. From February 1994 to
December 2005, data from the Virginia Key station
(station identication number 8723214) were used. The
resulting Atlantic Ocean stage record, as used in the
model, is shown in Figure 5.
Parameterization, Regularization, and Initial Parameter
Values
Application of formal parameter estimation tech
niques requires parameterization of aquifer properties and
initial parameter values from which calibration takes
NGWA.org C. Langevin and M. Zygnerski Groundwater 51, no. 5: 781803
place. The initial parameter values should be assigned
based on existing system information. For highly parame
terized models, some form of regularization must also be
applied for the problem to be tractable. For the present
study, preferred value regularization was applied to all
parameters using the initial parameter value.
An irregular distribution of 97 pilot points (Doherty
2003), with a higher density of points near the Pompano
Beach well eld, was used to parameterize Kh (Figure 4).
These 97 pilot points were used for each model layer.
Ordinary kriging was used with an isotropic exponential
variogram and a range of approximately 7.5 km to
interpolate between pilot points. Statistics on the initial
Kh parameter elds and the minimum and maximum
restricted values during calibration are provided in
Table 3. Initial Kh pilot-point values for the Biscayne
aquifer (model layers 3 to 8) were assigned using a
spatially variable Kh map prepared using the aquifer tests
reported in Fish (1988). At each pilot point, 80% of the
Biscayne aquifer transmissivity was apportioned evenly
among layers 3 through 5 (the production zone) and
used as the initial value for calibration. The remaining
20% of the Biscayne aquifer transmissivity was evenly
apportioned among model layers 6 through 8. In coastal
areas, Kh pilot-point values were allowed to vary for
all layers. For inland pilot points, however, a single Kh
multiplier was used to scale the initial Kh values in model
layers 3 through 8 by a single value.
A similar pilot-point methodology was used for the
Kh:Kv ratio and n. An initial parameter value of 100 was
assigned to Kh:Kv for all model layers. For inland pilot
points, the Kh:Kv ratio in model layers 3 through 8 were
adjusted by a single factor. The Kh:Kv ratio was restricted
to a range between 1 and 10,000. An initial parameter
eld of 0.25 for the mobile domain porosity (n) was
assigned for model layers 1, 2, and 9; based on preliminary
simulations of salt water movement, an n value of 0.10
was assigned to the Biscayne aquifer (layers 3 through
8). For calibration, n was restricted to a range between
0.05 and 0.40. Use of a lower n value than found in the
literature for layers 3 through 8 was required to match the
salt water intrusion event and subsequent ushing. The
model could not match the timing of these events with
higher n values, providing further support for the concept
of preferential ow zones in the Biscayne aquifer. Similar
to Kh and Kh:Kv, n values for model layers 3 to 8 were
adjusted by a single factor at inland pilot-point locations.
During parameter estimation, the conductance for
each canal reach was updated using the spatially variable
Kh eld because these dredged canals typically have
good hydraulic connection with the adjacent aquifer.
This approach provided the inversion process with a
mechanism for adjusting aquifer-canal interaction. Other
parameters estimated as part of the calibration process
are included in Table 4. These parameters do not vary
spatially or temporally. In the absence of literature values,
results from preliminary sensitivity simulations were used
to determine initial values for these parameters. For
example, a relatively low value was required for the
789
 Table 2
Hydrologic Feature and the Package Used to Represent the Stress

MODFLOW/SEAWAT
Hydrologic Feature Package Comment

Atlantic Ocean and CHD Cells with center elevations above the Atlantic Ocean sea oor are
Intracoastal Waterway represented as time-varying specied heads with the TDS
concentration for inow specied as 35 g/L. Intracoastal waterway
cells are also included as time-varying specied heads, but with a
TDS concentration of inow specied as 27 g/L (BCDPEP 2001)
and only in model layer 1.
Predevelopment fresh GHB Everglades fresh water wetlands were represented in western parts of
water wetlands the model in stress period 1 with a TDS concentration of zero. The
stage was set to 4 m, and the hydraulic conductance was calculated
using the cell area, half the cell thickness of model layer 1, and the
estimated vertical hydraulic conductivity of model layer 1.
Primary water GHB Primary water management canals (Hillsboro, L36, Pompano, and
management canals Cypress Creek) were represented in model layers 1 and 2 using
historical stage measurements. Canals were activated in the model
based on construction date. Hydraulic conductance was calculated
for each canal cell using an estimate of the aquifer-canal contact
area, the estimated horizontal hydraulic conductivity of the aquifer at
that cell, and a ow length of 50 m. A TDS concentration of 8 g/L
was specied for the tidal part of the Hillsboro Canal and 18 g/L
was assigned for the tidal parts of the Pompano and Cypress Creek
Canals. These concentrations were calculated using water-quality
data reported in BCDPEP (2001).
Secondary water RIV Secondary and tertiary water management canals were represented in
management canals model layer 1 using the RIV Package. Canal activation date, stage,
and hydraulic conductance were assigned using the same procedure
described for the primary water management canals.
Tidal canals GHB and RIV Tidal nger canals were assigned Atlantic Ocean stages and TDS
concentrations of the adjoining water body (Intracoastal Waterway,
Hillsboro Canal, or Cypress Creek Canal). Tidal canal activation
date and hydraulic conductance were assigned using the same
procedure described for the primary water management canals.
Recharge RCH A spatially uniform recharge rate was assigned to model layer 1 based
on measured rainfall totals. No attempt was made to subtract runoff,
interception, and unsaturated zone evapotranspiration quantities. This
approach was used by Merritt (1996a) and Langevin (2001, 2003) for
similarly constructed groundwater models of Miami-Dade County.
Evapotranspiration EVT The evapotranspiration surface was calculated by subtracting a value of
1.0 m from land surface to approximate microtopographic effects of
small depressions. The extinction depth was set to 7.0 m; this
relatively large depth was explained by Merritt (1996a) as
approximating other processes not represented by the model. For the
rst three stress periods, a maximum evapotranspiration rate of
151 cm/year was assigned (Merritt 1996a). For the remaining stress
periods, the maximum evapotranspiration rate varied by month
according to the rates estimated by Merritt (1996a).
Well-eld withdrawals WEL Withdrawals at public supply wells were specied in the model based
on estimated pumping records for each well. For public supply wells
with open-hole intervals that spanned multiple model layers, the
withdrawal rate was apportioned based on the estimated horizontal
hydraulic conductivity at that cell.
Golf course irrigation WEL Excess golf course irrigation (articial recharge) was modeled by
specifying a ux to layer 1 model cells within the Pompano Beach
municipal golf course. Measured irrigation totals not available for
19931994 and 20022005 were estimated from other years. The
percentage of the irrigation water that recharges the aquifer was
calculated as part of the calibration process. The TDS concentration
of the irrigation water was calculated using an average chloride
concentration of 400 mg/L.
BCDPEP, Broward County Department of Planning and Environmental Protection.

790 C. Langevin and M. Zygnerski Groundwater 51, no. 5: 781803 NGWA.org


Figure 5. The Atlantic Ocean stage record (relative to the
National Geodetic Vertical Datum of 1929) as used in the
model.
dual-domain mass transfer rate, which indicates that the
system is advection dominated with little mass transfer
between the mobile and immobile domains. An advection
dominated system with slow exchange between the mobile
and immobile domains is supported by the relatively fast
rates of observed salt water intrusion and subsequent
aquifer ushing, and by the lack of a pronounced tail on
the observed TDS concentration plots (Figure 3).
Initial conditions can be complicated to estimate for
transient salt water intrusion models because they are
rarely known with any certainty, and they can have a
large effect on model predictions. A common procedure
is to perform a steady-state simulation and then use the
resulting salinity eld as input for a subsequent transient
analysis. This approach was used early in the model
development process; however, steady-state conditions
were difcult to estimate, concerns over changes in sea
level raised questions about the defensibility of this
approach, and long runtimes were required to achieve
steady-state conditions. As an alternative, a variant of
the pilot-point methodology, as described in Doherty
(2009c), was used to parameterize the initial salinity eld.
Initial heads were not parameterized in this manner as
they equilibrated quickly relative to the length of the
simulation. Initial interface elevations were assigned to
the pilot points shown in Figure 4 using the salinity
eld from a preliminary steady-state simulation. Ordinary
kriging was then used to spatially interpolate the two-
dimensional interface surface to the model grid. This
interface surface was then intersected with the three-
dimensional model grid. Model cells with centroids above
the surface were assigned an initial TDS concentration
of zero; model cells with centroids below the interface
surface were assigned an initial TDS concentration of
35 g/L. To represent a diffuse interface, an interface width
parameter was introduced whereby TDS concentration
decreased upward and increased downward according to
a sigmoidal function. The interface width parameter was
assigned an initial value of 10 m and was limited to a
range between 1 and 50 m. Dausman et al. (2010) used
a similar approach to parameterize a salinity eld. The

Table 3
Statistical Description of the Spatially Variable Horizontal Hydraulic Conductivity (Kh) Fields Prior
to Calibration

Parameter Mean Standard Deviation Initial Kh Pilot Initial Kh Pilot Calibration Calibration
Group (log[Kh]) (log[Kh]) Point (Min) Point (Max) Minimum Limit Maximum Limit

Kh1 1.176 0.0 15 15 0.001 100

Kh2 1.176 0.0 15 15 0.001 100
Kh3 2.327 0.303 43 572 0.001 10,000
Kh4 2.327 0.303 43 572 0.001 10,000
Kh5 2.327 0.303 43 572 0.001 10,000
Kh6 1.726 0.303 11 143 0.001 10,000
Kh7 1.726 0.303 11 143 0.001 10,000
Kh8 1.726 0.303 11 143 0.001 10,000
Kh9 0.845 0.0 7 7 0.001 100

Note: Kh values are in m/d.

Table 4
Spatially Uniform Model Parameters Estimated as Part of the Calibration Process

Parameter Initial Value Calibration Minimum Limit Calibration Maximum Limit

Specic storage 1 105 /m 1 107 /m 1 103 /m

Specic yield 0.20 0.10 0.40
Evapotranspiration Extinction depth 7m 0.1 m 10 m
Multiplier for golf course Irrigation 0.20 0.1 0.9
to Aquifer recharge
Dual-domain mass Transfer rate 1.02 107 /d 1 1010 /d 1.0 /d
Immobile domain porosity 0.30 0.2 0.5

NGWA.org C. Langevin and M. Zygnerski Groundwater 51, no. 5: 781803 791

 Table 5
Observation Groups Used for Model Calibration
Observation
Group Description
HEADS Water levels in monitoring wells outside
the area of interest
HEADS_POMP Water levels in monitoring wells within
the area of interest
CONCS Salinity concentrations in monitoring
wells outside the area of interest
CONCS_POMP Salinity concentrations in monitoring
within the area of interest (this group
also contains a time series of salinity
concentration difference at the G2055
and G2055A nested monitoring wells)
C_WELL Measure of the total salt mass withdrawn
at each public supply well over the
entire simulation period (concentrations
less than 0.4419 g/L not included in
calculation)
Note: The area of interest is shown in Figure 2.
advantage of this approach is that the parameter estimation
process is given the freedom to adjust the initial salinity
eld, if necessary, in order to better match observed salt
water intrusion patterns, and thus lengthy steady-state runs
can be avoided.
Observations and Weights
Water levels and TDS concentrations in groundwater
monitoring wells and public supply wells comprised the
observation dataset used to calibrate the model. Temporal
and spatial interpolation of model results was used to
derive simulated values that corresponded in time and
space to the observations. The observation data set was
divided into ve observation groups (Table 5). Weights
were assigned uniformly to observations within a group.
Weights assigned to each group were manually adjusted
to achieve the intended contribution of the observation
group to the composite measurement objective function.
A wide variety of weighting schemes and weight values
were tested as part of the calibration process. For example,
concentration weights are typically related to the inverse
of the concentration value to accommodate the assumed
level of measurement error (Hill and Tiedeman 2007;
Sanz and Voss 2006). While this approach tended to
improve the match for low concentrations, simulated
TDS breakthrough curves did not adequately characterize
the salt water intrusion event. Ultimately, the weights
presented in Table 5 were used.
The contribution of each observation group to the
composite objection function was assigned on the basis
of modeling objectives, an assessment of measurement
error, and experience gained from preliminary calibra
tion runs. The C_WELL observation group was intention
ally assigned the highest contribution to the measurement
792 C. Langevin and M. Zygnerski Groundwater 51, no. 5: 781803
Weight Assigned Number of Objective Function
to Individual Observations with Value Prior to
Observations a Nonzero Weight Calibration
0.0 0 0
3.45 1735 6347
2.55 1659 2064
7.43 2138 9209
0.04 130 11,085
objection function. Historical water-quality records and
discussions with well-eld personnel indicated that with
the exception of the public supply well at the south
end of the well eld, TDS concentrations of withdrawn
groundwater never exceeded the potable limit (a chloride
concentration of 250 mg/L, which equates to a TDS con
centration of about 0.4419 g/L). TDS concentrations of
withdrawn groundwater simulated by the uncalibrated
model (using the initial parameter values), however,
exceeded potable limits at certain times indicating that
salt water had intruded into the Pompano Beach well
eld. Accordingly, the C_WELL observation group was
assigned a relatively large weight to improve the capa
bility of the model to represent fresh water conditions
at municipal wells. TDS concentrations in monitoring
wells near the Pompano Beach well eld (Figure 2;
CONCS_POMP) were weighted the next highest. The
CONCS_POMP group also contains a time series of con
centration differences at monitoring wells G2055 and
G2055A. These derived observations were added to
help the inversion process reproduce the isolated salt
water wedge in the middle of the aquifer. Water lev
els near the Pompano Beach well eld and then TDS
concentration differences at other monitoring wells were
weighted the next highest. Outside of the Pompano Beach
well-eld area, heads were assigned a weight of zero for
two reasons. First, there was generally good agreement
between simulated and observed heads with the uncali
brated model. This was not by chance, as many different
conceptualizations, parameter sets, and boundary imple
mentations were tested. The mean error and mean abso
lute error for the HEADS group were 0.03 and 0.41 m,
respectively. Second, because groundwater levels in the
Biscayne aquifer are highly dependent on exchanges with
NGWA.org
the surface water system, errors in assignment of canal
boundary levels have a large effect on simulated heads.
Data exist for assigning some canal boundary levels, but
they were derived or interpolated when missing. Con
sequently, when nonzero weights were assigned to the
HEADS group, the parameter estimation process adjusted
hydraulic conductivity as the sole option for improving
the head match (canal levels were not parameterized).
Consequently, the resulting parameter elds did not seem
reasonable as they tended to compensate for the structural
error caused by errors in assigned canal stage. Because
of this weighting approach and the focus on the Pompano
Beach well eld, the domain outside the area of interest
was not formally calibrated using PEST. This issue did not
seem to affect the area of interest because the secondary
canal network is restricted to only the westernmost part of
the area of interest, and because historical stage measure
ments for the Pompano and Cypress Creek Canals were
generally of good quality.
Model Calibration
PEST was used with the SVD-assist methodology
for model calibration (Doherty 2009a, 2009b) to estimate
a large number of parameter values, many of which
were highly correlated. A preferred value regularization
constraint was set for all of the estimated parameters
based on literature values and results of preliminary
calibration attempts with fewer parameters. The strength
of the regularization constraints was controlled through
PEST using a tuning variable. This variable was adjusted
until a good t was obtained with the measurements
and the estimated parameter values and distributions
were reasonable. The parameter estimation process made
substantial progress in improving the t between measured
values and simulated equivalents as shown in Table 6. The
inversion process was manually terminated on the 10th
optimization iteration as progress toward reducing the
measurement objective function had slowed considerably.
For some previous calibration runs, the inversion process
was allowed to continue for more than 40 optimization
iterations, and while the matches between observed
and simulated values were extraordinary, the resulting
parameter elds contained a high level of heterogeneity
that was not considered reasonable.
Overtting of the model to observations can reduce
the accuracy of predictions (Doherty and Welter 2010).
Even if the model was provided with the best possible
set of parameter values, there would still be disagreement
between observed values and simulated equivalents. This
is because the model observations contain measurement
error and because of structural errors in the model
caused by numerical errors, simplications of physical
processes, spatial and temporal averaging, inaccurate
boundary values, and other model inadequacies. Thus,
if the calibration process is allowed to overt the
observations, parameter values may become polluted by
measurement and structural errors. This overtting may
reduce the predictive capability of the model if the
prediction is dependent on the affected parameters. To
minimize the potential for this problem, the estimated
parameter elds and values were carefully evaluated to
ensure that the level of calibration achieved with the
estimation process was consistent with the quality of the
observations and model errors.
Selection of the appropriate level of calibration
was based on residual statistics, time-series plots of
observed versus simulated values, plots of spatially
varying parameter elds, and estimated parameter values.
To facilitate the discussion, the uncalibrated model is
referred to as Opt.0. The calibrated model for the rst
optimization iteration is referred to as Opt.1, and so forth.
Opt.6 was selected as the model used for sensitivity and
scenario analyses and is referred to later as the base case
calibrated model.
Time-series plots of water level and TDS concen
tration (Figure 6) highlight the progression of the cali
bration procedure for several of the key wells near the
Pompano Beach well eld. Simulated water levels at the
G-853 monitoring well near the center of the well eld
are in good agreement with observed water levels. This

Table 6

Table of Residual Statistics for Selected Observation Groups

Optimization Iteration
Group Statistics 0 1 2 3 4 5 6 7 8

ME 0.260 0.241 0.193 0.138 0.104 0.077 0.046 0.021 0.030

HEADS_POMP MAE 0.433 0.409 0.364 0.314 0.279 0.253 0.228 0.211 0.207
RMS 0.307 0.275 0.219 0.164 0.130 0.108 0.089 0.077 0.074
ME 13.113 12.350 10.891 8.641 6.483 4.950 2.207 0.820 0.471
CONCS MAE 16.193 15.313 13.541 11.011 8.685 7.133 4.446 3.032 2.479
RMS 478.334 426.133 329.704 216.841 136.608 93.741 43.902 28.892 22.575
ME 4.059 3.990 3.791 3.151 2.575 2.051 1.844 1.322 1.190
CONCS_POMP MAE 9.687 9.105 8.178 6.736 5.631 4.956 4.320 3.951 3.465
RMS 195.063 173.486 140.950 98.411 70.578 53.031 39.869 33.100 26.653

Notes: Residual statistics were calculated for those with nonzero weights. The number of values used to calculate these statics are listed in Table 5.
ME = mean error; MAE = mean absolute error; and RMS = root-mean-square error.

NGWA.org C. Langevin and M. Zygnerski Groundwater 51, no. 5: 781803 793

Figure 6. Plots of observed and simulated water levels and TDS concentrations for selected monitoring wells. The t between
observed and simulated values improves with increasing optimization number. Solid lines are for the Opt.6 base case model,
which was used for the sensitivity analyses. Faint dashed lines are for other optimization iterations.

is true for all of the optimization iterations, including
the uncalibrated model. Simulated TDS concentrations are
highly affected by the calibration process and it is clear
that calibration has improved the t between observed
and simulated values, but there are some obvious de
ciencies. For monitoring wells G-2054, G-2055A, and
G-2063, for example, maximum simulated concentrations
do not match with maximum observed concentrations. In
the mid-1980s, simulated concentrations are as much as
10 g/L less than observed concentrations. Another model
deciency is the inability to accurately represent the
isolated salt water wedge detected at the G-2055A
and G-2055 wells. In addition to the observed TDS
concentration values used for calibration, a separate obser
vation set of temporal concentration differences at these
two wells was also used for calibration. Although there
are many explanations for this model deciency, the lead
ing explanation is numerical dispersion caused by a lack
of vertical model resolution. Thus, while broad salt water
transition zone characteristics over the width of the aquifer
may be adequately represented, concentration differences
between layers may be underestimated.

794 C. Langevin and M. Zygnerski Groundwater 51, no. 5: 781803 NGWA.org


(a)
(b)
Figure 7. Horizontal hydraulic conductivity (Kh, as a base
10 logarithm) of the Biscayne aquifer estimated from (a) Fish
(1988) aquifer tests and from (b) the Opt6. model calibration.
In general, estimated parameter values were within
10% of their specied initial values, but there were some
exceptions. For example, the multiplier used to convert
the irrigation ux to a net recharge ux was increased
from 20% in Opt.0 to 70% in Opt.6. Heterogeneity was
also introduced in Kh, Kh:Kv, and n. Plots of the Opt.0
and Opt.6 Kh elds (log transformed) for the Biscayne
aquifer are shown in Figure 7. The vertically averaged
Kh elds were calculated by summing the transmissivity
values for model layers 3 through 8 and dividing by the
Biscayne aquifer thickness. The Opt.0 and Opt.6 Kh elds
share similar characteristics because the Fish (1988) data
were used as initial parameter values and as preferred
value regularization information. Thus, in the absence
of informative observation data, estimated Kh values
remained at their initial values. Near the Pompano Beach
well eld, heterogeneity in the Kh eld was introduced as
part of the calibration process to improve representation
of the spatial and temporal pattern of the salt water
intrusion event. Most importantly, a band of lower Kh was
identied near the well eld. There are no other sources of
data to suggest if this lower Kh band is real or not, but the
ability of the calibrated model to match G-853 water levels
NGWA.org C. Langevin and M. Zygnerski Groundwater 51, no. 5: 781803
(Figure 6) within the highly stressed well eld provides
some assurance that the estimated Kh eld is reasonable.
Plots of Kh:Kv and n (not shown) show similar degrees
of heterogeneity as shown for Kh.
Prior to calibration, simulated TDS concentrations of
groundwater withdrawn at the Pompano Beach well eld
were higher than observed values. Historical data indicate
chloride concentrations of the pumped groundwater never
exceeded drinking water standards for chloride (approxi
mately equal to a TDS concentration of 0.44 g/L) except
in one withdrawal well. All the simulations show an
increase and subsequent decrease in the TDS concentra
tion of pumped groundwater. As the optimization number
increases, the TDS concentrations of pumped groundwa
ter decrease to more realistic values near or below the
potable limit, which is consistent with historical obser
vations. In particular, there appears to be a large TDS
decrease and improvement in the simulated pumped con
centration between Opt.5 and Opt.6. This improvement
between Opt.5 and Opt.6 is caused primarily by a low
ering of the Kh in the area to the east of the well eld
(Figure 7).
The model does a good job representing many of
the important characteristics of the ow system at the
well-eld scale. Simulated heads are in good agreement
with measured heads at G-853, for example, and the
simulated water table map (Figure 8) is consistent with
previously published water table maps (e.g., Tarver 1964).
Most importantly for the present investigation, the model
qualitatively represents characteristics of the salt water
intrusion event and subsequent ushing. Figure 8 shows
simulated TDS concentrations in model layer 3 for six of
the years evaluated by Dunn (2001). Figure 8 also shows
the 0.44 g/L TDS contour mapped by Dunn (2001) for
the top of the Biscayne aquifer. Thus, the inland extent of
the colored salt water zones in Figure 8 can be compared
directly with the Dunn (2001) contours also shown in
the gure. The model shows a gradual salinization of the
aquifer during the late 1970s and 1980s, when drawdowns
are the largest, and a subsequent freshening during the
1990s, after water levels had risen. The model does not,
however, represent some details of the events, such as
the exact spatial patterns of the intrusion or the precise
timing of the retreat. At the coastline, the model simulates
a zone with TDS concentrations less than 0.44 g/L. This
zone forms at the top of the Biscayne aquifer due to
fresh groundwater recharge from above. There are no
groundwater salinity data to conrm whether or not the
Biscayne aquifer is fresh in this area, and so these results
should be evaluated with caution. Deeper model layers
show elevated TDS concentrations for this area.
Although a slight cone of depression can be seen in
the water table for 1999, water levels clearly increased
from 1984 to 1999 (Figure 8). The higher water table
elevations had a positive impact by ushing out some
of the salt water at the top of the Biscayne aquifer. The
ushing can be attributed to a reduction in groundwater
withdrawals, an increase in rainfall relative to the drought
period, and articial recharge at the golf course. Effects of
795
Figure 8. Comparison between simulated TDS concentrations in model layer 3 and the mapped 0.44 g/L TDS contour of Dunn
(2001). Contours of the simulated water table elevation are also shown to indicate the effect of groundwater withdrawals on
groundwater ow patterns.

the 25-cm rise in sea level over the 105-year simulation
period cannot be directly quantied from the calibrated
model, nor can the relative importance of the other
hydrologic stresses. For this reason, a sensitivity analysis
was performed to isolate the relative importance of these
factors on salt water intrusion.
Effect of Historical Sea-Level Rise
A qualitative sensitivity analysis was used to compare
the importance of historical sea-level rise to several
other key hydrologic factors: well-eld withdrawals,
annual recharge variations, and articial recharge at
the golf course. The evaluation was performed by
making a targeted adjustment to the input for the base
case calibrated model (Opt.6) and then rerunning the
simulation. This approach is consistent with the approach
outlined by Loaiciga et al. (2012) for isolating the effect
of different stresses on salt water intrusion. In the case
of historical sea-level rise, a constant sea level at the
estimated 1900 level was used for the entire 105-year
simulation for the tidal canals, the Intracoastal Waterway,

796 C. Langevin and M. Zygnerski Groundwater 51, no. 5: 781803 NGWA.org

and the Atlantic Ocean. The importance of well-eld
withdrawals was evaluated by running a simulation
without groundwater withdrawals. The next simulation
was used to evaluate the importance of droughts and
annual variations in recharge. Dunn (2001) hypothesized
that the 1971 to 1982 period of below average rainfall was
partially responsible for the observed salt water intrusion
event at the well eld. To evaluate this hypothesis, a
simulation was performed using a constant recharge rate
calculated from the average annual rainfall total. Lastly,
the utility of the articial recharge system was evaluated
by performing a simulation without articial recharge.
As shown in Figure 9, historical sea-level rise does
not have a large effect on the position of the 1 g/L
TDS contour in model layer 4, compared to the effect
of pumping, but the effect is discernible. In 1955, for
example, the 1 g/L contour for the simulation without
sea-level rise is about 100 m seaward of the 1 g/L contour
for the calibrated model. The largest effect from historical
sea-level rise can be seen in 1995 near the southeastern
part of the well eld. In this area, the 1 g/L contour is as
much as 1 km farther inland for the calibrated model than
for the simulation without sea-level rise. This difference in
the contour position is the result of the larger fresh water
ux toward the coast for the simulation without sea-level
rise. Thus, as sea level rises, the hydraulic gradient is
reduced, the fresh groundwater ux decreases, and salt
water intrusion occurs. This response is consistent with a
head-controlled system.
An analysis of simulated well-eld TDS concentra
tion (not shown) also indicates the effect of sea-level rise.
Figure 9. Results from the qualitative sensitivity analysis showing simulated results for the base case and for simulations of
constant recharge, no sea-level rise, no groundwater withdrawals, and no articial recharge. Contours are of the 1 g/L TDS
concentration in model layer 4.
NGWA.org C. Langevin and M. Zygnerski Groundwater 51, no. 5: 781803
The calibrated model shows an increase in TDS beginning
in about 1971 and with TDS values exceeding the potable
limit in 1984. Without a rise in sea level, the increase in
TDS occurs within about a year of the calibrated model,
but the well-eld TDS concentration never exceeds the
potable limit. Most importantly, well-eld TDS concen
trations are consistently about 0.2 g/L less from about
1980 to 2002 for the case without sea-level rise. Although
this is a relatively small difference, it equates to a differ
ence in chloride concentration of about 100 mg/L, which
is important considering the drinking water standard is
250 mg/L.
To further evaluate the head-controlled nature of
the system in response to sea-level rise, the simulated
water table from the calibrated model was compared to
the simulated water table from the sensitivity simulation
without sea-level rise. Over most of the model domain,
the elevation of the water table compares to within about
0.02 m or less. Along a narrow band near the coast,
however, the water table for the calibrated model is higher
than the water table for the simulation without sea-level
rise. Specically, to the east of a line that connects
structure G56 with G57 (the easternmost structures that
separate the fresh water canals from the tidally inuenced
canals; Figure 1), there are no fresh water canals to
act as a strong head control and the relatively high
land surface elevations along the Atlantic Coastal Ridge
allow the water table to rise without much restriction by
evapotranspiration.
The sensitivity analysis clearly indicates that well-
eld withdrawals have the largest effect on the position
797
of the 1 g/L TDS contour (Figure 9). By eliminating
pumping altogether, the 1 g/L TDS contour does not
change appreciably and the slight changes shown in
Figure 9 can be attributed to construction of the tidal
nger canals, historical sea-level rise, and recharge vari
ations. To further evaluate the effect of well-eld with
drawals, a range of different withdrawal increases and
decreases were simulated. By eliminating withdrawals
altogether, there is no historical rise in TDS concentra
tions at the well eld. Doubling withdrawals, however,
show substantial increases in well-eld TDS concentra
tions. Slight decreases in well-eld withdrawal rates also
have a large effect on well-eld TDS concentrations. Had
the actual well-eld withdrawals been 25% to 50% less,
model results suggest that salt water intrusion may not
have been a concern.
Sensitivity results indicate that rainfall variations and
articial recharge can affect salt water intrusion, but the
effects are much less important than effects of well-eld
withdrawals. Results from the simulation with a constant
recharge rate are similar to the base case calibrated model,
suggesting that the 1971 to 1982 period of less-than
average rainfall was not a predominant cause of salt water
intrusion near the well eld (Figure 9). A likely expla
nation is that surface water was brought into the area
during that time to maintain water levels of the primary
canals (Hillsboro and Cypress Creek). These canals do not
show a decrease in stage during that period (Figure 3), and
would have provided recharge to the aquifer to compen
sate for the drought conditions. Model results suggest that
articial recharge at the Pompano Municipal Golf Course
has a benecial impact on salt water intrusion. In 1995,
for example (Figure 9), elimination of articial recharge
in the sensitivity simulation results in the 1 g/L TDS con
tour being located as much as 1.5 km landward of the
position in the base case calibrated model. Contour posi
tions in 2005 also suggest that articial recharge helps to
prevent salt water from intruding near the well eld.
Sensitivity to Projected Rates of Sea-Level Rise
A sensitivity analysis was performed with the Opt.
6 model using four different rates of projected sea-level
rise and using the average annual hydrologic conditions
(well-eld withdrawals, canal stages, rainfall and articial
recharge, and evapotranspiration rates) from the last year
of the calibration period (2005). Results from these 100
year simulations cannot be used to predict future rates
of salt water intrusion in response to sea-level rise,
because the simulations do not include anthropogenic
changes, alternative rainfall patterns from climate change,
or well-eld management strategies. The results can be
used, however, to investigate the sensitivity of salt water
movement to different rates of projected sea-level rise.
For the rst simulation, sea level was held constant at
the average annual 2005 level. For the remaining three
simulations, sea level linearly increased over the 100
year simulation at rates of 24, 48, and 88 cm/century as
estimated in the IPCC TAR (Church et al. 2001). Sea-level
798 C. Langevin and M. Zygnerski Groundwater 51, no. 5: 781803
Figure 10. Simulated well-eld TDS concentration for the
sensitivity analysis of projected rates of sea-level rise. The
concentrations were calculated as a volumetric average for
groundwater extracted from municipal wells at the Pompano
Beach well eld.
rise was represented in the model by linearly increasing
the stage of the Atlantic Ocean and tidal canals. Intra-
annual variations in sea level were not represented in these
simulations.
Figure 10 shows a plot of well-eld TDS concen
tration relative to time for the four simulations. The
well-eld TDS concentration was calculated as a volu
metric average using the withdrawal rates and simulated
TDS concentrations at individual extraction wells. Use
of average 2005 hydrologic conditions and a constant
sea level result in TDS concentrations of the well-eld
exceeding drinking water standards after 70 years. This
nding suggests that the 2005 withdrawal rates may not
be sustainable with the 2005 hydrologic conditions. When
sea-level rise is included in the simulations, drinking
water standards are exceeded 10 to 21 years earlier (after
60 years for a rise of 24 cm/century; 55 years for a rise of
48 cm/century; and 49 years for a rise of 88 cm/century).
Apparent rates of lateral salt water intrusion in model
layer 4 were calculated from these sensitivity simulations
using the 1 g/L TDS contour. They are referred to here
as apparent because there is an upward component of
groundwater ow near the well eld, and thus, intrusion
is not limited to horizontal movement. Apparent lateral
intrusion rates are 15, 17, 18, and 21 m/year for the 0, 24,
48, and 88 cm/century sea-level rise rates, respectively.
Webb and Howard (2010) reported lateral salt water
intrusion rates (referred to as interface velocity in their
work) for different ratios of hydraulic conductivity to
recharge and for different rates of sea-level rise. Their
largest reported intrusion rate was 4 m/year, which is
about four to seven times less than the rates reported here,
but similar considering the substantial differences between
their simplied two-dimensional system and the Pompano
Beach well-eld area.
Discussion
The Pompano Beach well-eld area and nearby
coastal areas in southeastern Florida represent an end-
member in the spectrum of impacts of sea-level rise on
NGWA.org
salt water intrusion. The following are a number of general
observations about southeastern Florida that help explain
why the shallow coastal aquifer system is particularly vul
nerable to salt water intrusion caused by sea-level rise.
1. As shown by Werner and Simmons (2009), systems
that are head controlled are more susceptible to salt
water intrusion caused by sea-level rise than those
that are ux controlled. For conned aquifers that
are ux controlled, sea-level rise may not have any
effect on salt water intrusion (Chang et al. 2011). The
widespread canal system in southern Florida places an
extensive head control on water levels in the shallow
surcial aquifer. The head control is particularly strong
in southeastern Florida due to the direct hydraulic
connection between canals and the highly permeable
Biscayne aquifer. Also, land surface is relatively at
with little relief, and the unsaturated zone is thin (typ
ically less than a meter or two). Flooding from high
water tables can be a problem in many neighborhoods.
Evapotranspiration rates are also relatively high and
can be similar to rainfall rates during the summer
months. These combined conditions effectively elim
inate the possibility for groundwater levels to rise as
sea level rises. Consequently, the seaward hydraulic
gradient and associated fresh groundwater ow toward
the coast is expected to decrease.
2. Southeastern Florida has many tidally inuenced
canals that extend inland into the permeable coastal
aquifer. In some cases, tidal canals extend inland as
far as municipal well elds. These canals provide
ocean access for a thriving boating community. Near
the Pompano Beach well eld, the tidal portion of
the Cypress Creek Canal extends inland as far as the
Pompano Beach well eld. Tidal canals are also present
between the well eld and the Intracoastal Waterway.
These tidal canals, which have elevated salinities, have
stages at or near the stage of the Atlantic Ocean. A
rising sea level makes it difcult to maintain a seaward
hydraulic gradient that is strong enough to prevent salt
water intrusion.
3. The highly permeable shallow aquifer system also
contributes to the susceptibility of southeastern Florida
to sea-level rise. The high hydraulic conductivities
serve to reduce the seaward hydraulic gradient, cause
rapid water level declines after aquifer recharge events,
and allow salt water to intrude the aquifer at relatively
fast rates.
4. Southeastern Florida is heavily populated with a large
water demand for potable as well as for environmental
purposes; nearly all of the potable water is derived
from the shallow aquifer system, although there have
been recent efforts to explore alternative water sources.
Most groundwater is withdrawn near the coast at well
elds located along the Atlantic Coastal Ridge. The
Atlantic Coastal Ridge is the preferred location for well
elds because additional treatment is often required for
poorer quality groundwater withdrawn farther inland.
Construction of the Palm Aire well eld, located west
NGWA.org C. Langevin and M. Zygnerski Groundwater 51, no. 5: 781803
of the Pompano Beach well eld, was an effective
mechanism for reducing groundwater withdrawals near
the coast; shifting withdrawals inland raised the water
table near the coast and reduced the threat of salt water
intrusion.
The modeling analysis described in this investiga
tion can be used to quantify effects of sea-level rise
for other areas. Therefore, it is important to summarize
important and transferable lessons such as those related
to dual-domain transport, grid resolution, computational
issues, and modeling approach. Numerous challenges
were encountered in the development and calibration of
the county-scale (300 km2 ) dispersive salt water intrusion
model. Representing solute transport with a dual-domain
approach is consistent with geological knowledge of per
meable ow zones in the Biscayne aquifer and seemed to
provide a more accurate representation of salt water intru
sion than the traditional advection-dispersion approach.
Without the dual-domain approach, there was no way to
calibrate the model (with a reasonable parameter set) so
that it could simultaneously represent the salt water intru
sion event and subsequent retreat of saline groundwater.
Numerical dispersion and use of appropriate grid
resolution and transport schemes were among the most
difcult challenges. Sanford and Pope (2010) encountered
the same problem for a large 2000 km2 salt water intrusion
model of the Eastern Shore of Virginia and questioned
whether concentrations at an individual well can be
accurately simulated by a numerical model of that scale.
They suggest that in some instances, 10-cm thick model
layers may be required to accurately characterize the
transition zone between fresh and saline groundwater.
Owing to computational limitations, Sanford and Pope
(2010) were not able to use the TVD scheme in
MT3DMS/SEAWAT for their problem, which would have
helped to reduce numerical dispersion as it did for the
present application. For computation reasons, Sanford
and Pope (2010) used the relatively fast, implicit nite-
difference scheme for solute transport. This made the
problem tractable with present computing technology.
Additional numerical resolution would have been useful
for the present study. The model was unable to represent
some of the observed salinities in monitoring wells
(e.g., G-2055); maximum concentrations tended to be
underestimated, for example. There are many possible
reasons why the model had difculty in simulating the
details of the salt water intrusion event (e.g., errors in
the conceptual model or problems with the data); it
would have been useful to eliminate numerical dispersion
as a possibility. As shown by Langevin and Hughes
(2009), calibration of a salt water intrusion model with
a high level of numerical dispersion can have deleterious
effects on the predictive capability of the model. If, for
example, some of the hydraulic conductivity heterogeneity
was introduced by the calibration process in order to
compensate for the effects of numerical dispersion, then
model predictions would be in error if the predictions were
sensitive to that heterogeneity. As mentioned by Sanford
799
and Pope (2010), the capability to add resolution where
necessary is desirable in this situation. A nite element
or nite volume model would have been one option.
Another option for future studies would be to implement
a local-grid renement approach (Mehl and Hill 2002)
in SEAWAT as a way to increase horizontal and vertical
model resolution in areas where transport is important
and predictions depend on accurate representation of large
hydraulic gradients. For these types of larger-scale studies
with potential difculties in simulating dispersive solute
transport, one might also consider an entirely different
modeling approach based on an existing sharp interface
formulation, which is available in the Salt Water Intrusion
(SWI) Package for MODFLOW (Bakker 2003; Bakker
and Schaars 2005). The sharp interface approach was
designed for regional salt water intrusion modeling, and
while it cannot simulate solute concentrations, it is by
design, free of any type of dispersion including numerical
dispersion.
Calibration of the salt water intrusion model within
a highly parameterized context was found to be useful
for this study. The Pompano Beach well eld consists
of 16 groundwater withdrawal wells. Extraction rates for
the wells are highly variable between wells and through
out time. The system also has many other spatially and
temporally variable stresses (variations in fresh and tidal
canal levels, recharge, articial recharge, rainfall, evapo
transpiration) that confound interpretation of hydrologic
records. Karst aquifers, such as the Biscayne aquifer, are
highly discontinuous and heterogeneous, both in their spa
tial and temporal functioning, and it is commonly difcult
to separate the signal caused by a natural hydrologic event
from one caused by an anthropogenic event. It was dif
cult to infer from the data subsurface areas that may
be more or less permeable, and there were few reliable
point measurements of hydraulic conductivity. With the
highly parameterized calibration approach, features of the
system were quantied by monitoring where and how
parameters changed. In many instances, unrealistic param
eter distributions were used to identify deciencies in
the numerical model, such as an erroneous prescribed
canal level or an error in a monitoring well location.
Although not considered in this assessment, a next step in
this type of analysis is to quantify prediction uncertainty.
Robust uncertainty measures can be calculated within the
highly parameterized context provided one can assign
measures of parameter uncertainty. Future efforts to quan
tify sea-level rise impacts would benet from considering
uncertainty quantication.
Uncertainties in the predicted salt water intrusion
patterns that result from sea-level rise were not quantied,
but experience with the model suggests that there is a high
level of uncertainty. As suggested by Konikow (2011)
and experienced in this effort, solute transport models
are particularly difcult to develop, and one should not
expect the same level of reliability as one might expect
for a groundwater ow model. Uncertainties caused by
structural model errors can be reduced by evaluating
differences in model simulations instead of focusing on
800 C. Langevin and M. Zygnerski Groundwater 51, no. 5: 781803
a specic numeric value produced from a single forward
model run (Doherty and Welter 2010). For the sensitivity
analysis of the projected sea-level rise, this means it is
preferable to state that well-eld concentrations exceed
the potable limit 10 to 21 years sooner than for the case
without sea-level rise.
This study focused on evaluating the sea-level rise
projections reported by the IPCC (Church et al. 2001).
More recent projections by the IPCC (Meehl et al. 2007)
seem to project sea level rising at a slower rate, but
the revised estimates do not include some feedback
mechanisms that are anticipated to occur, such as rapid
ice sheet melting. Recent studies (Pfeffer et al. 2008)
have shown that sea level may rise by 0.8 to 2.0 m
by 2100. Heimlich et al. (2010) summarize some of the
recent sea-level rise projections and their possible effects
on southeastern Florida. These larger rates of sea-level
rise were not tested with the model. It is reasonable to
assume that increased rates of sea-level rise much larger
than those considered here would have serious impacts
on the fresh coastal groundwater supplies of southeastern
Florida. For these larger rates of sea-level rise, it is unclear
if adaptation measures and changes to the infrastructure
could meet potable water demands while simultaneously
providing ood protection.
Conclusions
Conditions near the Pompano Beach well eld in
northern Broward County, Florida, provide a unique
opportunity to examine the effects of historical and
projected sea-level rise on the fresh groundwater resources
of a low-lying highly permeable coastal aquifer. Results of
a numerical modeling analysis suggest that groundwater
withdrawals were the dominant cause of a multi-decade
salt water intrusion event, and that historical sea-level rise
(about 25 cm for the simulation period) contributed to the
extent of the intrusion by about 1 km. The historical rate
of sea-level rise was similar to the lower-bound estimate
(24 cm) of the IPCC (Church et al. 2001) projection for
the next century. A sensitivity analysis of four projected
rates of sea-level rise (24, 48, and 88 cm/century)
comparatively illustrates the relative severity of the
situation in south Florida. Even if sea level does not rise
in the future, model simulations suggest that corrective
actions would likely be required to protect the aquifer
from salinization. Corrective actions would be required as
much as 21 years sooner depending on the future rate of
sea-level rise. The ndings from this study are consistent
with general observations about the vulnerability of
southeastern Florida to salt water intrusion caused by sea-
level rise. Southeastern Florida is particularly vulnerable
because of (1) the overlying canal system, which acts as
a strong head control, (2) the presence of tidal canals
that extend inland, (3) the highly permeable shallow
aquifer system, which includes the Biscayne aquifer, and
(4) the large groundwater withdrawals from the coastal
aquifer.
NGWA.org
Acknowledgments
Special acknowledgments to Darrel Dunn and Katie
Lelis of the Broward County Environmental Protection
Department/Water Resources Division, Francis Henderson
and Dave Markward of the Broward County Ofce
of Environmental Services/Water Management Division,
and Randy Brown, Maria Loucraft, and Alan Clark
of Pompano Beach Utilities Department for providing
extensive information and data on the study site. A special
thanks to Winnie Said and Krista Guerrero-Reger of the
South Florida Water Management District as well as Guy
Bartolotta of Broward County Environmental Operation
Division for providing model data sets and historical
well-eld pumping records, and to John Doherty for his
continued support with the parameter estimation program
(PEST). Technical assistance was also provided by Alyssa
Dausman, Joann Dixon, and Roy Sonenshein. The authors
are grateful to John Masterson, Jeremy White, and three
anonymous reviewers for providing thoughtful comments
that substantially improved the manuscript.
References
Andersen, P.F., J.W. Mercer, and H.O. White. 1988. Numerical
modeling of salt-water intrusion at Hallandale, Florida.
Ground Water 26, no. 5: 619630. DOI: 10.1111/j.1745
6584.1988.tb00796.x
Bakker, M. 2003. A Dupuit formulation for modeling seawater
intrusion in regional aquifer systems. Water Resources
Research 39, no. 5: 1131.
Bakker, M., and F. Schaars. 2005. The Sea Water Intrusion
(SWI) Package Manual, version 1.2. http://bakkerhydro.org/
swi/index.html (accessed January 2011).
Bates, B.C., Z.W. Kundzewicz, S. Wu, and J.P. Palutikof, ed.
2008. Climate Change and Water. Technical Paper of the
Intergovernmental Panel on Climate Change, 210. Geneva:
IPCC Secretariat.
Bredehoeft, J.D. 2003. From models to performance assessment:
The conceptualization problem. Ground Water 41, no. 5:
571577.
Broward County Department of Planning and Environmental
Protection (BCDPEP). 2001. Broward County, Florida
Historical Water Quality Atlas: 1972-1997, Technical
Report Series TR: 01-03, 418.
Camp, Dresser, and McKee, Inc. 1980. Prospect well eld
impact analysis. Consultant report to the City of Fort
Lauderdale, 96 p.
Carrera, J., J.J. Hidalgo, L.J. Slooten, and E. Vazquez-Sune.
2010. Computational and conceptual issues in the calibra
tion of seawater intrusion models. Hydrogeology Journal
18, no. 1: 131145.
Causaras, C.R. 1985. Geology of the surcial aquifer system.
U.S. Geological Survey Water-Resources Investigations
Report 84-4068. Broward County, Florida, 167, 2 sheets.
Cazenave, A., and R.S. Nerem. 2004. Present-day sea level
change: observations and causes. Reviews of Geophysics
42, no. 3, RG3001. DOI: 10.1029/2003RG000139
Chang, S.W., T.P. Clement, M.J. Simpson, and K.K. Lee. 2011.
Does sea-level rise have an impact on saltwater intrusion?
Advances in Water Resources 34, no. 10: 12831291.
Church, J.A., J.M. Gregory, P. Huybrechts, M. Kuhn, K. Lam-
beck, M.T. Nhuan, D. Qin, and P.L. Woodworth. 2001.
Changes in sea level. In Climate Change 2001: The Sci
entic Basis, ed. J.T. Houghton, Y. Ding, D.J. Griggs,
M. Noguer, P.J. Van Der Linden, X. Dai, K. Maskell, and
NGWA.org C. Langevin and M. Zygnerski Groundwater 51, no. 5: 781803
C.A. Johnson, 639694. New York: Cambridge University
Press.
Cunningham, K.J., M.C. Sukop, H. Huang, P.F. Alvarez, H.A.
Curran, R.A. Renken, and J.F. Dixon. 2009. Prominence
of ichnologically inuenced macroporosity in the karst
Biscayne aquifer: stratiform super-K zones. Geological
Society of America Bulletin 121, no. 1/2: 164180.
Cunningham, K.J., R.A. Renken, M.A. Wacker, M.R. Zygner
ski, E. Robinson, A.M. Shapiro, and G.L. Wingard. 2006.
Application of carbonate cyclostratigraphy and borehole
geophysics to delineate porosity and preferential ow in the
karst limestone of the Biscayne aquifer, SE Florida. In Per
spectives on Karst Geomorphology, Hydrology, and Geo
chemistryA Tribute Volume to Derek C. Ford and William
B. White, 191208, ed. R.S. Harmon, and C. Wicks.
Geological Society of America Special Paper 404.
Dausman, A.M., J. Doherty, C.D. Langevin, and J. Dixon. 2010.
Hypothesis testing of buoyant plume migration using a
highly parameterized variable-density groundwater model
at a site in Florida, USA. Hydrogeology Journal 18, no. 1:
147160. DOI: 10.1007/s10040-009-0511-6
Dausman, A.M., and C.D. Langevin. 2005. Movement of the
saltwater interface in the surcial aquifer system in response
to hydrologic stresses and water-management practices,
Broward County, Florida. U.S. Geological Survey Scientic
Investigations Report 2004-5256. Reston, Virginia: USGS.
Doherty, J. 2009a. PEST, Model Independent Parameter Estima
tion. Addendum to user manual: 5th ed. Brisbane, Australia:
Watermark Numerical Computing. http://www.sspa.com/
PEST/.
Doherty, J. 2009b. PEST, Model Independent Parameter Estima
tion. User manual: 5th ed. Brisbane, Australia: Watermark
Numerical Computing. http://www.sspa.com/PEST/.
Doherty, J.E. 2009c. PEST Groundwater Data Utilities. Bris
bane, Australia: Watermark Numerical Computing.
Doherty, J. 2003. Groundwater model calibration using pilot
points and regularisation. Ground Water 41, no. 2:
170177.
Doherty, J., and D. Welter. 2010. A short explanation of
structural noise. Water Resources Research 46: W05525.
DOI: 10.1029/2009WR008377
Dunn, D. 2001. Three-dimensional analysis of saltwater Intru
sion: City of Pompano Beach Area, Broward County,
Florida. Technical Report Series TR-01-06.
Feseker, T. 2007. Numerical studies on saltwater intrusion in
a coastal aquifer in northwestern Germany. Hydrogeology
Journal 15, no. 2: 267279.
Fish, J.E. 1988. Hydrogeology, aquifer characteristics, and
ground water ow of the surcial aquifer system, Broward
County, Florida. U.S. Geological Survey Water-Resources
Investigations Report 874034.
Fujinawa, K., T. Iba, Y. Fujihara, and T. Watanabe. 2009.
Modeling interaction of uid and salt in an aquifer/lagoon
system. Ground Water 47, no. 1: 3548. DOI: 10.1111/
j.1745-6584.2008.00482.x
Giambastiani, B.M.S., M. Antonellini, G.H.P. Oude Essink, and
R.J. Stuurman. 2007. Saltwater intrusion in the unconned
coastal aquifer of Ravenna (Italy): A numerical model.
Journal of Hydrology 340, no. 12: 91104.
Guha, H., and S. Panday. 2012. Impact of sea level rise on
groundwater salinity in a coastal community of South
Florida. Journal of the American Water Resources Associ
ation (JAWRA) 48, no. 3: 510529. DOI: 10.1111/j.1752
1688.2011.00630.x
Guo, W., and C.D. Langevin. 2002. Users guide to SEAWAT: A
computer program for the simulation of three-dimensional
variable-density ground-water ow. U.S. Geological Survey
Open-File Report 01-434.
Harbaugh, A.W., E.R. Banta, M.C. Hill, and M.G. McDonald.
2000. MODFLOW2000, the U.S. Geological Survey
801
 modular ground water modelUser guide to modular
ization concepts and the ground water ow process. U.S.
Geological Survey Open-File Report 0092.
Heimlich, B.N., F. Bloetscher, D.E. Meeroff, and J. Murley.
2010. South Floridas resilient water resources: Adapta
tion to sea level rise and other impacts of climate change.
Florida Atlantic University, Center for Urban and Envi
ronmental Solutions and Department of Civil Engineering,
Environmental, and Geomatics Engineering. http://www.
ces.fau.edu/les/projects/climate_change/
SE_Florida_Resilient_Water_Resources.pdf (accessed Feb
ruary 2010).
Hill, M.C., and C.R. Tiedeman. 2007. Effective Calibration of
Groundwater Models, with Analysis of Data, Sensitivities,
Predictions, and Uncertainty. New York: John Wiley and
Sons.
Howie, B. 1987. Chemical characteristics of water in the surcial
aquifer system, Broward County, Florida. U.S. Geological
Survey Water-Resources Investigations Report 864330.
Reston, Virginia: USGS.
Hughes, J.D., C.D. Langevin, and L. Brakeeld-Goswami. 2010.
Effect of hypersaline cooling canals on aquifer saliniza
tion. Hydrogeology Journal 18, no. 1: 147160. DOI:
10.1007/s10040-009-0511-6
Hughes, J.D., H.L. Vacher, and W.E. Sanford. 2009. Temporal
response of hydraulic head, temperature, and chloride con
centrations to sea-level changes, Floridan aquifer system,
USA. Hydrogeology Journal 17, no. 4: 793815. DOI:
10.1007/s10040-008-0412-0
Intergovernmental Panel on Climate Change (IPCC). 2010.
Workshop Report of the Intergovernmental Panel on Climate
Change Workshop on Sea Level Rise and Ice Sheet Instabil
ities, ed. T.F., Stocker, D. Qin, G.-K. Plattner, M. Tignor,
S. Allen, and P.M. Midgley, 227. Bern, Switzerland: IPCC
Working Group I Technical Support Unit, University of
Bern.
Klein, H. and J.E. Hull. 1978. Biscayne aquifer, southeast
Florida. U.S. Geological Survey Water-Resources Investi
gations 79112. Reston, Virginia: USGS.
Konikow, L.F. 2011. The secret to successful solute-transport
modeling. Ground Water 49, no. 2: 144159, DOI:
10.1111/j.1745-6584.2010.00764.x
Langevin, C.D. 2003. Simulation of submarine ground water
discharge to a Marine Estuary: Biscayne Bay, Florida.
Ground Water 41, no. 6: 758771. DOI: 10.1111/j.1745
6584.2003.tb02417.x
Langevin, C.D. 2001. Simulation of ground-water discharge to
Biscayne Bay, southeastern Florida. U.S. Geological Survey
Water-Resources Investigations Report 00-4251. Reston,
Virginia: USGS.
Langevin, C.D., and J.D. Hughes. 2009. Effect of numerical
dispersion as a source of structural noise in the calibration
of a highly parameterized saltwater intrusion model. The
PEST Conference 2009, November 2-4, 2009, Potomac,
Maryland.
Langevin, C.D., D. Thorne, A.M. Dausman, M.C. Sukop, and
W. Guo. 2008. SEAWAT Version 4: A Computer Program
for Simulation of Multi-Species Solute and Heat Transport.
U.S. Geological Survey Techniques and Methods Book 6,
Chapter A22. Reston, Virginia: U.S. Geological Survey.
Langevin, C.D., and W. Guo. 2006. MODFLOW/MT3DMS
based simulation of variable density ground water ow and
transport. Ground Water 44, no. 3: 339351.
Langevin, C.D., W.B. Shoemaker, and W. Guo. 2003. MOD
FLOW2000, the U.S. Geological Survey modular ground
water modelDocumentation of the SEAWAT2000
version with the variable density ow process (VDF) and
the integrated MT3Dms transport process (IMT). U.S. Geo
logical Survey Open-File Report 03426. Reston, Virginia:
USGS.
802 C. Langevin and M. Zygnerski Groundwater 51, no. 5: 781803 NGWA.org
Lebbe, L., N. Van Meir, and P. Viaene. 2008 Potential impli
cations of sea-level rise for Belgium. Journal of Coastal
Research 24, no. 2: 358366.
Loaiciga, H.A., T.J. Pingel, and E.S. Garcia. 2012. Sea water
intrusion by sea-level rise: Scenarios for the 21st century.
Ground Water 50, no. 1: 3747.
Lu, C., and J. Luo. 2010. Dynamics of freshwater-seawater
mixing zone development in dual-domain formations. Water
Resources Research 46: W11601. DOI: 10.1029/2010WR
009344
Masterson, J.P. 2004. Simulated interaction between freshwater
and saltwater and effects of ground-water pumping and
sea-level change, Lower Cape Cod Aquifer System, 78.
Massachusetts: USGS SIR 2004-5014.
Masterson, J.P., and S.P. Garabedian. 2007. Effects of sea-level
rise on ground water ow in a coastal aquifer system.
Ground Water 45, no. 2: 209217. DOI: 10.1111/j.1745
6584.2006.00279.x
Mehl, S., and M.C. Hill. 2002. Development and evaluation
of a local grid renement method for block-centered
nite-difference groundwater models using shared nodes.
Advances in Water Resources 25, no. 5: 497511.
Meehl, G.A., T.F. Stocker, W.D. Collins, P. Friedlingstein,
A.T. Gaye, J.M. Gregory, A. Kitoh, R. Knutti, J.M. Mur
phy, A. Noda, S.C.B. Raper, I.G. Watterson, A.J. Weaver,
and Z.-C. Zhao. 2007. Global climate projections. In Cli
mate Change 2007: The Physical Science Basis. Contribu
tion of Working Group I to the Fourth Assessment Report
of the Intergovernmental Panel on Climate Change, ed.
S. Solomon, D. Qin, M. Manning, Z. Chen, M. Marquis,
K.B. Averyt, M. Tignor, and H.L. Miller. Cambridge, UK
and New York: Cambridge University Press.
Merritt, M.L. 1996a. Simulation of the water table altitude
in the Biscayne aquifer, southern Dade County, Florida,
water years 194589. U.S. Geological Survey Water-
Supply Paper 2458.
Merritt, M.L. 1996b. Assessment of saltwater intrusion in
southern coastal Broward County, Florida. U.S. Geological
Survey Water-Resources Investigations Report 96-4221,
133. Reston, Virginia: USGS.
Merritt, M.L. 1995. Numerical simulation of a plume of brack
ish water in the Biscayne aquifer originating from a
owing artesian well, Dade County, Florida. U.S. Geo
logical Survey Open-File Report 95-402. Reston, Virginia:
USGS.
Oude Essink, G.H.P. 1999. Impact of sea level rise in the
Netherlands. In Seawater Intrusion in Coastal Aquifers
Concepts, Methods, and Practices, ed. J. Bear, A.H.
D. Cheng, S. Sorek, D. Ouazar, and I. Herrera, 507530.
Dordrecht, The Netherlands: Kluwer Academic Publishers.
Parker, G.G., G.E. Ferguson, S.K. Love, and others 1955. Water
Resources of Southeastern Florida. U.S. Geological Survey
Water-Supply Paper 1255.
Pfeffer W.T., J.T. Harper, and S. ONeel. 2008. Kinematic
constraints on glacier contributions to 21st-century sea-level
rise. Science 321, no. 5894: 13401343.
Rao, S.V.N., V. Sreenivasulu, S.M. Bhallamudi, B.S. Than
daveswara, and K.P. Sudheer. 2004. Planning groundwa
ter development in coastal aquifers. Hydrological Sciences
Journal 49, no. 1: 155170.
Renken, R.A., K.J. Cunningham, A.M. Shapiro, R.W. Harvey,
M.R. Zygnerski, D.W. Metge, and M.A. Wacker. 2008.
Pathogen and chemical transport in the karst limestone
of the Biscayne aquifer: 1. Revised conceptualization of
groundwater ow. Water Resources Research 44: W08431.
DOI: 10.1029/2007WR006058
Reese, R.S., and K.J. Cunningham. 2000. Hydrogeology of
the Gray Limestone aquifer in southern Florida. U.S.
Geological Survey Water-Resources Investigations Report
99-4213. Reston, Virginia: USGS.
Renken, R.A., J. Dixon, J. Koehmstedt, S. Ishman, A.C. Lietz,
R.L. Marella, P. Telis, J. Rogers, and S. Memberg. 2005.
Impact of anthropogenic development on coastal ground
water hydrology in southeastern Florida, 19002000.
USGS Circular 1275. Reston, Virginia: USGS.
Restrepo, J.I., C. Bevier, and D. Butler. 1992. A three
dimensional nite difference ground water ow model of the
surcial aquifer system, Broward County, Florida. South
Florida Water Management District Technical Publication
DRE-314.
Rozell, D.J., and T.F. Wong. 2010. Effects of climate change on
groundwater resources at Shelter Island, New York State,
USA. Hydrogeology Journal 18, no. 7: 16571665.
Sanford, W.E., and J.P. Pope. 2010. Current challenges using
models to forecast seawater intrusion: lessons from the
Eastern Shore of Virginia, USA. Hydrogeology Journal 18:
7393.
Sanz, E., and C.I. Voss. 2006. Inverse modeling for seawa
ter intrusion in coastal aquifers: Insights about parameter
sensitivities, variances, correlations and estimation proce
dures derived from the Henry problem: Advances in Water
Resources 29, no. 3: 439457.
Sherwood, C.B., H.J. McCoy, and C.F. Galliher. 1973. Water
resources of Broward County, Florida. Florida Department
of Natural Resources, Bureau of Geology. Report of
Investigation 15.
Shoemaker, W.B. 2004. Important observations and parameters
for a salt-water intrusion model. Ground Water 42, no. 5:
829840.
Shoemaker, W.B., and K.M. Edwards. 2003. Potential for
Saltwater Intrusion into the Lower Tamiami Aquifer near
Bonita Springs, Southwestern Florida. U.S. Geological
Survey Water-Resources Investigations Report 03-4262, 74.
Reston, Virginia: USGS.
NGWA.org C. Langevin and M. Zygnerski Groundwater 51, no. 5: 781803
Tarver, G.R. 1964. Hydrogeology of the Biscayne aquifer in the
Pompano Beach Area, Broward County, Florida. Florida
Geological Survey, Report of Investigations No. 36, 47 p.
van der Meij, J.L., and B. Minnema. 1999. Modelling of the
effect of a sea-level rise and land subsidence on the
evolution of the groundwater density in the subsoil of the
northern part of the Netherlands. Journal of Hydrology 226,
no. 3: 152166.
Watson, T.A., A.D. Werner, and C.T. Simmons. 2010. Tran
sience of seawater intrusion in response to sea level rise.
Water Resources Research 46: W12533. DOI: 10.1029/
2010WR009564
Webb, M.D., and K.W. Howard. 2010. Modeling the tran
sient response of Saline intrusion to rising sea-levels.
Ground Water 49, no. 4: 560569. DOI: 10.1111/j.1745
6584.2010.00758.x
Werner, A.D., and C.T. Simmons. 2009. Impact of sea-level rise
on sea water intrusion in coastal aquifers. Ground Water 47,
no. 2: 197204. DOI: 10.1111/j.1745-6584.2008.00535.x
Werner, A.D., J.D. Ward, L.K. Morgan, C.T. Simmons, N.I.
Robinson, and M.D. Teubner. 2012. Vulnerability indicators
of sea water intrusion. Ground Water 50, no. 1: 4858.
Zheng, C. 2006. MT3DMS v5.2 Supplemental Users Guide.
Technical Report to the U.S. Army Engineer Research and
Development Center, Department of Geological Sciences,
University of Alabama, 24.
Zheng, C., and P.P. Wang. 1999. MT3DMS, A modular three-
dimensional multi-species transport model for simulation
of advection, dispersion and chemical reactions of contam
inants in ground-water systems; documentation and users
guide. Contract Report SERDP-99-1. Vicksburg, Missis
sippi: U.S. Army Engineer Research and Development
Center.
803