Saltwater Intrusion and Hydraulic Conductivity Estimation in East Baton Rouge Parish, Louisiana

1
Saltwater Intrusion and Hydraulic Conductivity
Estimation in East Baton Rouge Parish, Louisiana

• Frank T-C. Tsai and Xiaobao Li
• Department of Civil and Environmental Engineering
• Louisiana State University, Baton Rouge
• (20th Salt Water Intrusion Meeting, June 24, 2008)

2
Outline

• Background and motivation
• Groundwater model development
• Hydraulic conductivity estimation
• Generalized parameterization (GP)
• Bayesian model averaging (BMA)
• Groundwater model calibration and prediction
  results
• Intrusion simulation/prediction for 90 years.
• Closing remarks

3
Project benefits from

• USGS Louisiana Water Science Center at Baton
  Rouge
• Louisiana Capital Area Ground Water Conservation
  Commission (CAGWCC)
• Louisiana Department of Transportation and
  Development (DOTD), Water Resources Programs
• Louisiana Geological Survey

4
Southern Hills aquifer system
5
Source of saltwater salt domes
Salt dome
R. Brent Bray and Jeffrey S. Hanor(1990)
6
Saltwater intrusion in Southern Hills aquifer
system
7
Saltwater intrusion in 1500-foot sand in Baton
Rouge
Freshwater
Saltwater
water.usgs.gov/wid/html/la.html
8
Groundwater model development

• Data
• Groundwater production rates (CAGWCC, DOTD)
• Groundwater head data (USGS)
• Electrical resistivity log data (USGS, LGS)
• USGS/DOTD reports and CAGWCC Newsletters
• Model components
• Modeling area and grid size (200 m by 200 m)
• Boundary conditions
• Initial condition (01/1990)
• Aquifer parameters (aquifer thickness, specific
  storage, hydraulic conductivity, and BR fault
  permeability)
• Sinks and sources (pumping wells and connector
  well)
• Solver (MODFLOW-2000, Harbaugh et al., 2000)

9

• Box (1976)
• All models are wrong, some are useful.

• All models are wrong, some are useful with
  careful calibration.

10
Challenges in traditional hydraulic conductivity
estimation (inverse problem)

• Traditional parameterization methods are not
  flexible.
• Non-kriging interpolation methods are not
  applicable to spatially correlated K fields
• Non-uniqueness of parameterization is not
  considered in the inverse problem.
• Model structure uncertainty.

11
Bayesian model averaging (BMA) with generalized
parameterization (GP)

• GP integrates zonation and interpolation methods
  under a geostatistical framework (Tsai 2006 WRR).
• BMA takes into account multiple GP methods (Tsai
  and Li, 2008 WRR Tsai and Li 2008 GW).
• BMA integrates multiple GP methods and multiple
  simulation models.

12
GP
Data weighting coefficient
Conditional estimator
(plnK)
Conditional covariance
BMA
Law of total probability
Law of total expectation
Model weight
Law of total variance
Within-model variance
Between-model variance
13
K estimation and head predictions through GP and
BMA
Maximum likelihood estimation to estimate b
values in GP
Predictions MODFLOW (2D) MT3D (2D)
K estimation in MODFLOW (2D)
Trans. parameters estimation in MT3D (2D)
SEAWAT (3D)
14
Numerical Verification
15
Identification Results
BMA
GP
Interpolation
16
Seven (7) parameterization methods
17
Three (3) model structures
(2) Impermeable-fault model (HC0/day)

• Low-permeable-fault
• model (HC0.000155/day)

(3) No-fault model (no HC)
Model output
observation
Groundwater head at 1/1/2005
18
Results model weights and method weights
Low-permeable-fault model
Impermeable-fault model
No-fault model
19
Results model outputs vs. observations
There are significant differences between these
methods
Time step (day)
Head(m)
Seven methods under low-permeable-fault model
head output at observation well EB-168
20
Hydraulic conductivity distributions using
Multi-GP and BMA
BMA
NN-VT
Within-GP
Between-GP
BMA
OK-VT
ID-VT
21
(No Transcript)
22
Within-method head variance
Between-method head variance
Between-model head variance
Total head variance
23
Saltwater intrusion simulation (1990-2080) MT3D
(Zheng and Wang, 1999)
24
Saltwater intrusion prediction (1990-2080)
25
Saltwater intrusion prediction at 2080
26
Closing remarks

• Parameter estimation and model predictions need
  to consider non-uniqueness of parameterization
  and model structure uncertainty
• Generalized parameterization (GP) provides a
  flexible way for parameterizing spatially
  correlated parameters.
• Bayesian model averaging (BMA) provides a
  statistically rigorous approach to integrate
  multiple models. It can be used for parameter
  estimation, model prediction, and model
  application.
• Using GP and BMA obtains better hydraulic
  conductivity estimation and avoids
  over-confidence in the results using one method
  and one model.

27
Acknowledgements

• This research was supported in part by
• Department of the Interior, USGSNIWR
  (05HQGR0142)
• Louisiana Water Resources Research Institute
  (06HQGR0088)
• Louisiana Board of Regents (LEQSF(2005-08)-RD-A-1
  2)