University of Montana
ScholarWorks at University of Montana
Theses, Dissertations, Professional Papers
Graduate School
2013
FORECASTING GROUNDWATER
RESPONSES TO DAM REMOVAL
Antony Ray Berthelote
The University of Montana
Follow this and additional works at: http://scholarworks.umt.edu/etd
Recommended Citation
Berthelote, Antony Ray, "FORECASTING GROUNDWATER RESPONSES TO DAM REMOVAL" (2013). Theses, Dissertations,
Professional Papers. Paper 1402.
This Dissertation is brought to you for free and open access by the Graduate School at ScholarWorks at University of Montana. It has been accepted for
inclusion in Theses, Dissertations, Professional Papers by an authorized administrator of ScholarWorks at University of Montana. For more
information, please contact [email protected].
FORECASTING GROUNDWATER RESPONSES TO DAM REMOVAL
By
Antony Ray Berthelote
Masters of Science, University of Alaska Fairbanks, Fairbanks, Alaska, 2005
Bachelors of Science, University of Montana, Missoula, Montana, 2003
Dissertation
presented in partial fulfillment of the requirements
for the degree of
Doctor of Philosophy
in Geosciences
The University of Montana
Missoula, MT
May 2013
Approved by:
Sandy Ross, Associate Dean of The Graduate School
Graduate School
Dr. William W. Woessner, Committee Chair
Department of Geosciences
Dr. Marco Maneta
Department of Geosciences
Dr. Johnnie Moore
Department of Geosciences
Dr. Steven D. Sheriff
Department of Geosciences
Dr. David Shively
Department of Geography
COPYRIGHT NOTICE TEMPLATE
© COPYRIGHT
by
Antony Ray Berthelote
2013
All Rights Reserved
ii
Table of Contents
Acknowledgements
.................................................................................................................
viii
Preface
.................................................................................................................
ix
Chapter/Paper No. 1
.................................................................................................................
1
Contents
Responses of Groundwater Systems to Dam Removal
List of Figures
.................................................................................................................
iv
List of Appendices
.................................................................................................................
iv
Abstract
.................................................................................................................
1
1.0
Introduction
.................................................................................................................
1
1.1
The Role Of The Dam ..........................................................................................................
2
1.2
Hydrologic Changes Associated With Dam Emplacement ..................................................
2
1.3
Hydrologic Changes Associated With Dam Removals ........................................................
4
1.4
Challenges Of Forecasting Groundwater Impacts From Dam Removal ..............................
5
2.0
Research Purpose ...............................................................................................................
5
3.0
Conceptual Groundwater Response Model .....................................................................
5
4.0
A Generic Box Model to Assess Groundwater Level Responses ....................................
7
5.0
Case Study - Milltown Reservoir ......................................................................................
10
6.0
Results
.................................................................................................................
11
7.0
Discussion
.................................................................................................................
13
8.0
Conclusions
.................................................................................................................
13
9.0
References
.................................................................................................................
15
9.0
Appendices
.................................................................................................................
21
iii
List of Figures
Figure 1 Generalized Illustration Of The Typical Dam And Dam Removal Impact Research...........
3
Figure 2 Conceptual Models Of Groundwater Responses To Dam Removal ....................................
7
Figure 3 Set Up For Generalized Box Model Used To Determine Flow To Conductivity Ratios (Q/K)
8
Figure 4 Groundwater Impacts Associated With Various Flow To Hydraulic Conductivity Ratios (Q/K)
9
Figure 5 Study Area Map .................................................................................................................
10
Figure 6 Observed Groundwater Data For Study Period ....................................................................
12
Figure 7 Groundwater Impacts From Q/K Model For Milltown ........................................................
13
List of Appendices
Appendix 1A
Box Model Input Files for Q/K ratios (100, 1000, 3000, 10000) ............
21
Chapter/Paper No. 2
.................................................................................................................
22
Contents
Proactive Mitigation of Domestic and Municipal Groundwater
Supplies During Dam Removal Actions, Milltown Reservoir,
Western Montana
List of Figures
.................................................................................................................
v
List of Tables
.................................................................................................................
vi
List of Appendices
.................................................................................................................
vi
Abstract
.................................................................................................................
22
1.0
Introduction
.................................................................................................................
22
2.0
Milltown Dam Site ..............................................................................................................
23
3.0
Conceptual Model of Impacts to Groundwater Levels from
Dam and Reservoir Removal .............................................................................................
25
4.0
Methods
.................................................................................................................
26
4.1
Baseline Conditions And Groundwater Level Monitoring ...................................................
27
4.1.1
Well Monitoring Network ....................................................................................................
27
4.1.2
Surface Water Monitoring Network .....................................................................................
27
4.1.3
Water Level Data Analyses And Establishment Of Baseline Conditions ............................
27
4.2
Data Sets Needed To Formulate A Predictive Groundwater Model.....................................
27
iv
4.2.1
Water Budget
.................................................................................................................
27
4.3
Formulation Of The Numerical Groundwater Modeling......................................................
30
4.4
Development Of Alternative Conceptual Models /Uncertainty Analyses ............................
31
4.5
Mitigation Process ................................................................................................................
31
4.5.1
Model Forecasting And Mitigation Requirements ...............................................................
31
5.0
Results
.................................................................................................................
32
5.1
Groundwater Base Case .......................................................................................................
33
5.2
Observed Groundwater Changes During Stage Drawdowns and Complete Dam Removal
33
5.3
Water Budget and Surface Water Groundwater Exchange ..................................................
35
5.4
Numerical Modeling Results ................................................................................................
35
5.4.1
The Forecasting Of Groundwater .........................................................................................
35
5.5
Mitigation Process ................................................................................................................
37
6.0
Discussion and Conclusion ................................................................................................
39
6.1
Challenges of Predicting Groundwater Impacts ...................................................................
39
6.2
Evaluation of Remediation Actions......................................................................................
39
6.3
Recommend Process for Mitigation Planning ......................................................................
40
7.0
Summary
.................................................................................................................
42
8.0
References
.................................................................................................................
43
9.0
Appendices
.................................................................................................................
47
List of Figures
Figure 1
Study Area Map ................................................................................................................
24
Figure 2
Conceptual Models Of Groundwater Responses To Milltown Dam Removal .................
25
Figure 3
A Timeline Summarizing The Sequencing Of Events Described In This Work ..............
26
Figure 4
Generalized Conceptual Model Illustrating The Components Of The Milltown Water Balance
27
Figure 5
Spatial Distribution Of Data Used To Estimate Bedrock Depths .....................................
29
Figure 6
Locations Of Direct Channel Measurements Used To Evaluate
Surface Water Groundwater Exchange .............................................................................
30
Figure 7
Outline Of Research Objectives And Work Flow ............................................................
32
Figure 8
Graphical Representation Of The Determination Of Base Case Conditions ....................
33
Figure 9
Water Table Maps Of Observed Groundwater Levels ......................................................
34
v
Figure 10 Map Of Observed Impacts Compared To Model Forecasts .............................................
37
Figure 11 Spatial Comparisons Of Mitigation Recommendations And Actions ..............................
38
Figure 12 Comparison Of The Actual And Predicted Mitigation Actions .......................................
38
Figure 13
41
Risk Based Decision Process For Groundwater Mitigation Resulting From Dam Removals
List Of Tables
Table 1
Table Of Techniques, Methods, And Equipment Used To Generate Data Sets
Describing The Hydrological System ...............................................................................
29
Table 2
General Mitigation Decision Path .....................................................................................
32
Table 3
Pre Model Estimated Water Balance Summary ....................................................................
35
Table 4
Comparison Of Pre Model Estimated And Simulated Water Balances ............................
36
List of Appendices
Appendix 2A
Appendix 2B
Appendix 2C
Appendix 2D
Data For Figure 3 (Timeline) ...................................................................
Compiled Water Level And Climate Data For Study Period ...................
Final Calibrated MODFLOW Model Input Files ....................................
Well Survey Data .....................................................................................
47
48
49
50
Chapter/Paper No. 3
.................................................................................................................
51
Contents
Role of Drawdown Data in ANN Forecasting of Water Table Responses to Dam and
Reservoir Removals
List of Figures
.................................................................................................................
vii
List of Tables
.................................................................................................................
vii
List of Appendices
.................................................................................................................
vii
Abstract
.................................................................................................................
51
1.0
Introduction
.................................................................................................................
51
2.0
Study Site And Background ..............................................................................................
53
3.0
Methods
.................................................................................................................
55
3.1
Observed Water Level Changes ...........................................................................................
55
3.2
Three Dimensional MODFLOW Model ..............................................................................
55
3.3
Application Of ANN To The Milltown Study......................................................................
56
vi
3.3.1
Data Set
.................................................................................................................
56
3.3.2
Ann Input Selection Method ................................................................................................
57
3.4
Statistical Methods ...............................................................................................................
57
4.0
Results
.................................................................................................................
58
4.1
Observed Data
.................................................................................................................
58
4.2
MODFLOW Modeling Results ............................................................................................
59
4.3
Ann Modeling Results ..........................................................................................................
59
5.0
Discussion
.................................................................................................................
61
6.0
Conclusion
.................................................................................................................
63
7.0
References
.................................................................................................................
64
8.0
Appendices
.................................................................................................................
66
List of Figures
Figure 1
Study Area Map ................................................................................................................
54
Figure 2
Typical Representations Of Ann Architecture And Processing ........................................
56
Figure 3
Spatial Representations Of The Observed Groundwater Changes ....................................
58
Figure 4
Observational Data Results For The Entire Study Period Compared To Ann And Modflow
Forecasts
Figure 5
.................................................................................................................
59
Plot Of Model Performace Vs. Distance To Surface Water .............................................
61
List of Tables
Table 1
List Of Inputs Used In ANN Models ................................................................................
57
Table 2
Comparative Statistical Results For ANN And MODFLOW ...........................................
59
List of Appendices
Appendix 3A
Appendix 3B
ANN Code ...............................................................................................
ANN Data (Inputs and Results) ...............................................................
vii
66
67
ACKNOWLEDGEMENTS
This project was only possible with the help and support of many people and organizations. I would like to
thank the Environmental Protection Agency, Montana Department of Environmental Quality, Missoula
County Water Quality District, and The University of Montana for the opportunity to work on, and funding
support for, this research. My gratitude goes out to the American Indian College Fund, The National
Center for Earth-Surface Dynamics, the American Geological Institute, Hopa Mountain, Alfred P. Sloan
Foundation, Inland Northwest Research Alliance, and others for additional funding support. Many
students, staff, and faculty assisted in the work throughout the years and I am thankful for all their work,
particularly the work and friendship of Anthony Farinacci and Aaron Deskins. I want to thank my
committee members for their time, patience, and thoughtful review of my dissertation and especially Dr.
William Woessner for his continued support, motivation, mentoring, and overall guidance throughout the
process which actually began in his office in 1999 when I decided to return to school as an
undergraduate. I want to thank all those at Salish Kootenai College who have supported me during the
last four years while I finished this dissertation work. Lastly, I want to thank my family, without whose
support, I could not have completed. Both of my children have spent their entire lives watching me work
through school and I look forward to spending more time with them in the coming years. I am particularly
grateful for the continued support from my wife Shawnna and mother Walene who seemed to always be
at my side throughout this journey.
viii
PREFACE
This dissertation contains three papers describing groundwater system responses to dam removals and presents
current and new methodologies that managers can use to proactively forecast and mitigate those impacts. The three
papers bring together literature, data gathering, data analysis, testing, and modeling techniques that apply
groundwater science to forecasting the response of groundwater systems to dam removal actions. Conceptual and
numerical models developed as part of this work provide scientists, environmental consultants, regulators and
managers with tools to assess the consequences of the removal of stream reservoirs on the adjacent and underlying
groundwater system
The first chapter/paper, Responses of Groundwater Systems to Dam Removal is a review paper on the connection
between groundwater systems and artificial impoundments. The synthesis of materials was compiled into a general
conceptual model of the effects of dam and reservoir emplacement and removal on associated groundwater systems.
Additionally, a method is proposed and tested to forecast the magnitude of impacts to groundwater levels using a
generalized lumped parameter approach that allows the ratio of aquifer discharges to hydraulic conductivity to vary
depending on the hydrogeological setting.
The second chapter/paper, Proactive Mitigation of Domestic and Municipal Groundwater Supplies During Dam
Removal Actions, Milltown Reservoir, Western Montana is an applied case study and outlines the process for
mitigating water supplies from engineering actions associated with the removal of the Milltown Dam in western
Montana between 2006 and 2009. This paper is a summary of four technical reports completed and submitted by
these authors to the EPA (http://www.epa.gov/region8/superfund/mt/milltown/techdocs.html) outlining the
groundwater mitigation processes in detail. The paper summarizes data collection and modeling approaches
undertaken to provide practical forecasts of groundwater level changes. A range of forecasts are compared to
completed mitigation actions and model performance is evaluated. A risk management framework is proposed and
tested.
The third chapter/paper, The Role of Drawdown Data in ANN Forecasting of Water Table Responses to Dam and
Reservoir Removals examines the applicability of using Artificial Neural Networks (ANNs) to forecast groundwater
levels changes resulting from a dam removal. Two specific ANN models were developed and analyzed to
specifically examine the need for training data inclusive of a temporary or partial drawdown. Results for the
Milltown Dam removal are compared to observed water levels and results of standard numerical techniques
(presented in the second paper). ANN modeling shows promise as a tool to forecast likely groundwater responses to
dam removals as it requires less detailed hydrogeological data sets and is executed more efficiently than standard
numerical models.
ix
Chapter/Paper 1
Berthelote, Antony, Doctor of Philosophy, May 2013
Geosciences
Responses of Groundwater Systems to Dam Removal
Chairperson: Dr. William W. Woessner
ABSTRACT:
Dams are constructed to generate hydropower, provide water for crops and human consumption,
manage floods, create navigable waterways, provide recreational opportunities, create new or enhance
existing wildlife habitats, and capture contaminated sediments. Research on physical impacts of
building or removing dams on local hydrology typically focuses on modifications to the river system,
however, impacts to groundwater systems also occur. Dam emplacement or removal actions modify
adjacent groundwater system boundary conditions and often result in a rise or fall in the underlying
and adjacent water table. Unfortunately, few dam removal efforts have documented changes to
associated groundwater making the formulation of impact magnitude and time forecasts challenging.
This research develops descriptive and generic semi-quantitative conceptual models of the response of
groundwater systems to dam and reservoir emplacement and removal. Numerical generic models are
constructed using a set of dimensionless parameters and the ratio of aquifer discharges to values of
hydraulic conductivity. The generic conceptual model forecasted changes in water table positions after
a dam removal are then compared to observed groundwater responses during a 8.5 m high dam
removal in western Montana. The simulated water table declines compared favorably with the
observed declines of 0.5 to 3.0 m. Future dam emplacement and removal actions need to recognize the
likely response of the local water table and, if necessary, develop pre and post dam groundwater
impact mitigation measures.
KEY WORDS
Milltown Reservoir; Dam Removal Mitigation And Management; Groundwater Surface
Water Interactions; Groundwater Level Forecasting;
1.0
INTRODUCTION
Construction of new dams, hydropower systems, and reservoirs provide immediate benefits to many of the
two billion people lacking access to electricity and seven billion expected to face water scarcity by 2050
(ICOLD, 2007; Pegg, 2004; World Commission on Dams, 2000). Pegg (2004) reported that 1500 large
dams were under construction globally while, in more developed countries like the United States, dam
removals were outpacing new construction. Decisions to remove dams are based on identified adverse
ecological and social impacts, safety conditions associated with aging dams, and appreciation for societal
values linked to healthy rivers and fisheries (American Indian Law Center, 1999; Collier et al., 1996;
Collins et al., 2007; Graf, 2002, 2003, 2005; Johnson and Graber, 2002; Pejchar and Warner, 2001;
Pennsylvania Organization for Watersheds and Rivers et al., 2004; Pohl, 2002; Whitelaw and MacMullan,
2002; World Commission on Dams, 2000). Recent interests in dam removals are reflected in a steady
stream of dam removal articles appearing in the popular press (Babbitt, 2002; Francisco, 2004; Hart and
Poff, 2002; Landers, 2004; Martin, 2004; McCool, 2004; O'Conner et al., 2008; Tweit, 2006). There were
over 23 news articles in the month of June 2008 alone, related to dam removals (e.g., Aun, 2008; Bouma,
2008; Caduto, 2008; Dean, 2008; Egan, 2008). Dam removal rates have been steadily increasing (Bowman
et al., 2002; O'Conner et al., 2008) with 60 U.S. dams removed in 2010 alone (McClain, 2012). The last 13
years account for over 450 of the 888 large U.S. dams removed in the past century (McClain, 2012).
The impacts and changes to associated groundwater systems as a consequence of the emplacement or
removal of dams and reservoirs have historically been overlooked (e.g., Doyle et al., 2003a; Doyle et al.,
2003b; Evans et al., 2000a; Graf, 2003; Hart et al., 2002). In addition to the anticipated water table
changes, new water table positions are also secondarily associated with extent and function of wetlands, the
degree of disconnection or reconnection of groundwater with aquatic ecosystems, and impacts to surface
water and groundwater quality (Aseltyne et al., 2006; Constantz, 2003; Constantz and Essaid, 2007). It is
1
the physical response of the adjacent alluvium dominated groundwater system to dam and reservoir
construction and more specifically dam removals that is the focus of this work.
1.1
The Role Of The Dam
Starting with the first large-scale dam constructed in 5000 B.C., with the exception of the Great Wall of
China, dams are the largest structures ever built (PBS, 2008). The U.S. has approximately 2.5 million
small dams (less than 1.8 m high), 80,000 large dams (over 1.8 m high), and 8036 major dams (greater then
15 m high) (American Rivers et al., 1999; Bowman et al., 2002; USACE, 2008b). The U.S. accounts for
nearly 20% of the 45,000+ major dams in the world today. One in four of the major dams in the U.S. are
built in river locations with unconsolidated and semiconsolidated sand and gravel aquifers. Such valley
settings host major aquifers (U.S. Dept. of the Interior, 2008; USACE, 2008b). Dams are operated to
generate hydropower, provide water for crops and human consumption, manage floods, create navigable
waterways, provide recreational opportunities, create new or enhance existing wildlife habitats, and capture
contaminated sediments. Such dams also often represent an important aspect of a community’s history
(American Rivers et al., 1999; Bowman et al., 2002; World Commission on Dams, 2000). Their
construction commonly determined the locations of towns, industries, and trade routes. Communities
benefitted from hydroelectric power, recreation opportunities, flood protection, and access to reservoir
water or elevated groundwater levels for municipal and agricultural water supplies (American Rivers et al.,
1999; Evans et al., 2000a; Pyle, 1995; World Commission on Dams, 2000).
1.2
Hydrologic Changes Associated With Dam Emplacement
While some studies have looked at larger-scale historical channel changes in rivers from dam
emplacements (Beyer, 2005; Gregory et al., 2002; Nilsson et al., 2005; Renwick et al., 2005; Wootton et
al., 1996), the impacts of dams on rivers have been most typically viewed as a surface water phenomenon
involving geomorphology, surface-water hydraulics, sediment transport, fisheries, benthic and riparian
ecology, as well as a plethora of aesthetic issues (e.g., Graf, 2005) (Figure 1). Changes in river conditions
at dam sites include pool development, increased water depth, changes in river temperature, possible pool
water density stratification, loss of light penetration due to increased water depths and turbidity, retention
of nitrates and phosphates, growth of plankton and algae, and changes in aquatic ecosystems from lentic to
lotic species (Baxter, 1977; Petts, 1984; Poff and Hart, 2002). In addition, it has been long documented that
sediment accumulation in reservoirs will result in continual declines in water storage capacity (Dendy,
1968; Rãdoane and Rãdoane, 2005).
Downstream of a dam, a river typically reestablishes its sediment load by eroding bed and bank materials,
causing incision and channel widening, and preferential transport of fine grained material (Evans et al.,
2000c; Faulkner and McIntyre, 1996; Graf, 2005; Grams and Schmidt, 2005; Petts and Gurnell, 2005; Poff
et al., 1997; Rãdoane and Rãdoane, 2005; Renwick et al., 2005). The inevitable result is channel
embedding, which can have an adverse impact on benthic ecosystems (Petts, 1984; Petts and Gurnell,
2005). The most pervasive long-term downstream effect, however, is aggradation that results from flow
regulation (Grams and Schmidt, 2005; Marston et al., 2005). The dam serves to attenuate the flood peaks
that govern sediment transport in an unregulated river. The resulting reduction in sediment deposition
downstream affects channel morphology, substrate, and flood stages (Chin et al., 2002; Collier et al., 1996).
Downstream impacts to the river system can include alteration of the thermal structure of the river due to
release of water from below the thermocline of the reservoir (Muth et al., 2000), riparian plant communities
(Bayley, 1995; Doyle et al., 2005; Magilligan and Nislow, 2005; Muth et al., 2000; Petts and Gurnell,
2005; Shafroth et al., 2002), and the restriction of anadromodous fish migration (Baxter, 1977; Bayley,
1995; Nislow et al., 2002; Wootton et al., 1996).
2
Figure 1
Generalized illustration of the typical dam and dam removal impact research, separated by river reach.
Illustrations modified from Hart et al. (2002).
As suggested by Constantz (2003), all these previously studied vantage points possess merit, but neglect
any associated physical and/or ecological responses of adjacent shallow groundwater systems. It has been
long documented that increases in surface water elevation tend to increase the height of the water table in
the areas immediately behind impoundments (Leopold and Maddock, 1954). A few researchers have
attempted to describe and quantify likely ground water responses to reservoir pool elevation changes.
Engineered pools and beaver constructed ponds are hydraulically similar to lakes and reservoirs. Thus,
beaver dam studies are a natural (small to large-scale dam) analog for understanding how groundwater
responds to reservoir pool manipulations. Where beaver dams span the entire valley, the main hydrologic
feature will be an upstream pond that elevates groundwater levels adjacent to the pond (Butler and
Malanson, 2005; Chen and Chen, 2003; Mertes, 1997; Naiman et al., 1988; Westbrook et al., 2006; Woo
and Waddington, 1990). However, where valleys are unconfined, yet rivers are narrow enough to be
dammed by beaver, the hydrologic effects may extend far beyond the edge of the pond (Lowry and
Beschta, 1994). Guo (1997) outlined several historical and new analytical solutions for transient
groundwater flow between a reservoir and a semi-infinite unconfined aquifer. He suggested that following
release of water from bank storage into a reservoir, the rate of hydraulic-head change in the aquifer should
decrease with distance and time. Sawyer et al. (2009) applied similar techniques to examine the effects of
frequent river stage fluctuations caused by dam operation on the hyporheic zone. All of these solutions are
limited to special conditions (e.g. saturated, homogeneous, and semi-infinite) seldom found in natural
3
settings. Modeling of groundwater where a river enters a reservoir has demonstrated connections between
the river delta, floodplain terrace, reservoir stage, and groundwater levels (Rains et al., 2004).
Lewandowski, et al. (2009) completed an experimental study to identify drivers of water level fluctuations
and hydrological exchanges between groundwater and surface water in a saturated oxbow system.
Aseltyne et al. (2006) modeling showed increases in the depth that surface-water penetrates the river bed
sediments following a reservoir-stage rise as anticipated. Finally, Heilweil et al. (2005) monitored rising
groundwater levels underneath and adjacent to a newly constructed reservoir atop consolidated materials.
Groundwater level changes in response to dam and reservoir construction in these settings is expected, yet,
as cited above, rarely documented. Principally, the degree of groundwater level change will be a direct
result of: 1) the magnitude of reservoir stage rise, 2) the rate of exchange between the reservoir, river
sections, and the groundwater (losing or gaining conditions), 3) the regional groundwater conditions
(constrained or unconstrained valley sediments), and 4) boundary conditions. Generally, as dams and
reservoirs are constructed, the surrounding adjacent groundwater systems are likely to expand the zone of
saturation raising local water tables.
1.3
Hydrologic Changes Associated With Dam Removals
The emerging science of observing and forecasting impacts of dam removal has been reviewed elsewhere
(Doyle et al., 2003b; Evans et al., 2000b; Graf, 2003; Hart et al., 2002). Dam removals involve transient
effects that introduce new concerns for watershed management and river restoration as summarized in
Figure 1. (Collins et al., 2007; Hewitt et al., 2001; Pennsylvania Organization for Watersheds and Rivers et
al., 2004). Major concerns typically include modifications to channel morphology (Cantelli et al., 2007;
Williams and Wolman, 1984), stream and floodplain exchange processes (Graf, 2006; Kondolf, 1998), the
fate of reservoir sediments (Cui et al., 2006; Doyle et al., 2002, 2003a; Evans et al., 2002; Evans et al.,
2000c; Lorang and Aggett, 2005; Pizzuto, 2002; Stanley and Doyle, 2002), and the potential generation of
downstream flood hazards (Roberts, 2006). Other concerns include the ecological consequence and risks to
human health of released or exposed contaminated reservoir sediments (DesGranges et al., 1998; James,
2005; Shuman, 1995; World Commission on Dams, 2000). Recognized impacts to fluvial systems upstream
of a reservoir include changes in the flood regime (Batalla et al., 2004; Leopold and Maddock, 1954; Poff
et al., 1997; Power et al., 1996; Rowntree and Dollar, 1999), riparian ecosystems (Doyle et al., 2005; Petts,
1984; Shafroth et al., 2002), sediment budgets (e.g., Faulkner and McIntyre, 1996) or some combination of
these three factors (Beyer, 2005; Bushaw-Newton et al., 2002; Graf, 2006; Lytle and Poff, 2004;
Magilligan and Nislow, 2005; Poff et al., 2006). Dam removal projects also involve numerous legal issues
(Bowman, 2002; Lindloff and Wildman, 2006; Nadeau and Rains, 2007).
As stated previously, dam removal science has also been historically limited to a plethora of
geomorphologic studies that focus on surface processes with no connection to subsurface groundwater
levels or floodplain aquifer impacts (e.g., Collier et al., 1996; Doyle et al.; Evans et al., 2007; Evans et al.,
2000c; Graf, 2003; Hart et al., 2002). The same basic hydrogeological principles that pertain to increasing
groundwater levels following reservoir construction (Heilweil et al., 2005) would suggest reservoir levels
would decline to pre-dammed levels following a dam removal. With the exception of a few studies on
beaver dams that have documented reductions in groundwater levels following removals (Butler and
Malanson, 2005; Chen and Chen, 2003; Mertes, 1997; Naiman et al., 1988; Westbrook et al., 2006; Woo
and Waddington, 1990), only a small number of studies have addressed how dams alter the rates and
locations of near surface hyporheic exchanges, with little emphasis on any connection to the local water
table position (Alexander and Caissie, 2003; Dahm et al., 1998; Hayashi and Rosenberry, 2002; Hendricks
and White, 1991; Palmer, 1993; Pusch et al., 1998; Stanford, 1998; Stanford and Ward, 1993; Valett et al.,
1990). The United States Army Corps of Engineers’ Dam Removal Research Office, which is responsible
for all U.S. dam removal oversight; American Rivers, a nonprofit organization dedicated to the protection
and restoration of North America's rivers and possibly the leading authority on dam removals; University of
California’s Clearing House for Dam Removal Information; and Oregon States’ Dam Removal Listserv all
report no information covering reservoir-groundwater level linkages other than those by these authors
(American Rivers, 2008; CDRI, 2012; Oregon State, 2008; USACE, 2008a).
A central Vermont village experienced a water shortage following a dam removal (Pyle, 1995) which
resulted in the Federal Energy Regulatory Commission (FERC) allowing the shallow groundwater supplied
4
village to buy control of a second dam scheduled for decommissioning to prevent its removal (Graf, 2002).
More recently, unforeseen “dry” wells occurred following the 39 m Condit Dam Removal (Oct 2011) on
the White Salmon River in Washington (Learn, 2011). The homeowners are asking the PacifiCorp Power
Company who owned and removed the dam to assist in mitigating the loss of local groundwater supplies.
The associated Condit Dam Environmental Impact Statement (EIS) stated “Significant unavoidable adverse
impacts were not identified with respect to groundwater” (Sandison, 2010). Alternatively, the EIS for the
upcoming 2.75 m Finesville Dam removal on the Lower Musconetcong River in New Jersey states that a
“Potential drop in water table may result in lower water levels in some wells” (USDA, 2010). The EIS for
the Gold Ray Dam Removal on the Rogue River in Oregon stated “Wells upstream of the dam could be
affected by lower water levels” (NMFS, 2010). During the 6 m Wadsworth and 4.4 m Sterling Lake Dam
removals in the Mantua Creek Watershed in New Jersey, several residents noted that the water table
decreased following declines in reservoir levels, and they expressed concern about the need for well
mitigation following the complete dam removal (Wyrick et al., 2009).
The proposed removal of the Rodman Dam on the Ocklawaha River in Florida was accompanied by a
suggested research plan (the first mention of this type of research) to determine the impacts of complete
drawdown from dam removal on groundwater levels, including effects on water levels in residential wells,
discharge from springs and water levels in nearby lakes and wetlands (Shuman, 1995). When the initial
water study was completed, it incorporated a numerical groundwater model that was used to determine
sustainable groundwater pumping yields following the dam removal, but did not specifically address
drawdown induced groundwater level impacts and/or impacts to hyporheic exchange and their effects on
ecological system responses (Hall, 2005).
Current EIS statements and public reactions are highlighting potential negative impacts to groundwater
systems from dam removal actions; however, rigorous proactive research and planning for groundwater
impacts are conspicuously absent. This may necessitate costly mitigation and litigation (Bowman et al.,
2002; Bowman, 2002). Legal issues of liability over well production losses are complex and often involve
conflicting regulations from multiple agencies (local, tribal, state, and Federal). Attempts to plan for and
mitigate groundwater, channel and riparian system impacts from dam removal actions can be accomplished
if anticipated (Hart et al., 2002; Hart and Poff, 2002).
1.4
Challenges Of Forecasting Groundwater Impacts From Dam Removal
Hydrogeological science provides a clear framework within which to develop cause and effect models of
groundwater level response to dam and reservoir construction and removal actions. However, impacts at
both the local hyporheic (e.g., Alexander and Caissie, 2003; Dahm et al., 1998; Hayashi and Rosenberry,
2002; Hendricks and White, 1991; Palmer, 1993; Pusch et al., 1998; Stanford, 1998; Stanford and Ward,
1993; Valett et al., 1990), and valley wide (regional groundwater) scale have rarely been measured (e.g.,
Heilweil et al., 2005) or modeled. Constantz and Essaid (2007) attempted to highlight how changing
reservoir and river management could impact downstream water supplies in California. They used a
MODFLOW model to predict generic responses of downstream pumping water levels following a dam
removal in a groundwater basin with altered groundwater recharge resulting from the conversion from
perennial stream leakage to ephemeral stream leakage.
2.0
RESEARCH PURPOSE
This research was designed to develop and test a conceptual model that will provide a framework within
which changes to groundwater levels in response to dam removals can be formulated and evaluated. It is
tested using generic settings and develops a relationship that can be used to develop initial assessments of
water level impacts. A case study is also presented where observed groundwater level changes are
compared to the proposed conceptual model groundwater responses.
3.0
CONCEPTUAL GROUNDWATER RESPONSE MODEL
Developing a conceptual model for a wide range of stream and hydrogeological settings is desirable;
however, such an approach involves a large number of variables. We developed a conceptual model that
was constrained to represent natural, dammed and restored river reaches located in upland confined, semiconfined and un-confined or broad valley settings for both high and low water table systems (Figure 2).
The high water table scenario includes a floodplain/valley water table that is generally higher than the river
5
channel stage (overall gaining stream), and the low water table scenario features a water table that is
generally lower than the river stage (losing stream). In general, a confined valley system will have a narrow
floodplain that contains sediments that are coarse grained, with a steep upland topography, and a high
longitudinal riverbed gradient. Dams placed in such settings often fill the river valley. Whereas, the wider
un-confined floodplain setting generally has a much lower relief, floodplain sediments include larger
quantities of finer material, and the longitudinal riverbed gradient is more gradual. Dams constructed in
these settings are often wider than higher and partially fill the floodplain. The relationships between
riverbed/reservoir sediment budgets, surface water distributions, groundwater levels, and surface water
exchanges are illustrated for each stage of the river evolution.
Dams impound rivers. The reduction in longitudinal riverbed gradients and increase in cross sectional
wetted area at the reservoir-river transition slows water velocities and begins aggrading sediments (if they
are present in the system). Though the aggradation of these new transition zones with fine sediments may
potentially reduce riverbed leakage to the adjacent groundwater, an increase in wetted surface area, and rise
in river/reservoir stage will generally act to locally raise the associated groundwater levels. The reservoir
head causes seepage from the reservoir and beneath the dam that discharges to the river reach downstream
of the dam. The potential below dam reduction in the channel sediment budget degrades and coarsen the
river channel, in some settings, and may lead to an increase in channel bed leakage extending far
downstream (a factor that may also cause some increase in below dam groundwater levels). Channel
incising may also induce additional groundwater discharge and effectively lower associated groundwater
levels in some settings.
Immediately following a dam removal, a portion of the accumulated reservoir sediment will migrate
downstream and aggrade or embed the channel causing a decrease in groundwater surface water exchange
and a change in the channel conditions that can increase the potential for downstream flooding. This may
also temporarily reduce the downstream groundwater levels (or increase them if groundwater is discharging
to the channel). Changes in the local hyporheic exchange process, sites, timing and magnitudes is also
likely to occur.
6
Figure 2
An illustration of a basic grid of conceptual models for unconsolidated sand and gravel aquifers in; A) a
confined floodplain, B) a semi-confined floodplain, and C) an un-confined floodplain during the following
conditions; 1) natural or restored, 2) dammed, and 3) immediately following a dam removal for both high
and low water table systems.
4.0
A GENERIC BOX MODEL TO ASSESS GROUNDWATER LEVEL RESPONSES
It is apparent from the conceptual model presented above that a number of groundwater responses can
occur from dam and reservoir construction and removal. In an attempt to simplify the wide range of factors
influencing groundwater level responses a generic box model was developed. This model assumes steady
state, isotropic and homogeneous and rectangular boundary conditions. Using the relationships expressed
in the groundwater flow governing equations, a number of hydrogeologic scenarios can be assessed.
Proceeding with the understanding that for any given conditions, the primary factors driving the relative
position of groundwater levels beneath a river channel atop unconsolidated materials are found in the Darcy
flow equation:
Q=KiA
where
Q = Volumetric Flow Rate (L3/t)
K =Hydraulic Conductivity (L/t)
i =Hydraulic Gradient (L/L)
7
A =Cross Sectional Area (L2)
Under steady state conditions, for the same values of i and A variations in the value of K will yield
specific groundwater flows, Q. It is also evident that corresponding groundwater head values, h (where i=
h/l and l is constant) will also vary under different ratios of Q/K. Thus a set of generic models can be
developed to evaluate how the groundwater head distribution will vary within a number of hydrogeologic
settings expressed solely by ratios of Q/K (Figure 3). A steady state three layer homogeneous unitless
groundwater model was designed using the MODFLOW graphical user interface program available from
Environmental Simulations Group (Groundwater Vistas 5).
The dimensionless hydrogeologic setting model included three vertical layers consisting of two base layers
each 50 units thick with a 250 unit thick surface layer. Model dimensions were 5000 units wide by 25000
units long with an overall depth of 350 units. The model consisted of 187,500 individual 100 unit2 cells.
The MODFLOW river package was used to place a river down the central transect at a slope of 0.004., A 5
unit river stage, 2 unit riverbed thickness, and 10 unit riverbed hydraulic conductivity were designated for
each river cell. In the center of the model at cell location 100 (upstream) from the left boundary, a dam was
represented with a 1 cell thick wall (K = 10 units) in layer 1 that extended 10 cells to either side of the river
(2000 units total width). The reservoir was represented by river cells behind the dam extending upstream 50
cells (5000 units). These reservoir river cells maintained the same architecture as the central river cells
with the reservoir stage (depth of reservoir pool) varying from 8 units at the upstream end to 28 units
immediately behind the dam. The horizontal to vertical hydraulic conductivity ratios were set to 10:1 for all
model cells. The downstream face of the model was represented with a general head boundary (where the
water leaves the model) and the upstream face was set with a constant flux boundary by using a series of
wells producing a cumulative volumetric flow rate of 11,920,000 (length 3/time)(groundwater entering the
model). Remaining boundaries were no flow. Proportional hydraulic conductivities values were set to
represent Q/K ratios of 100, 1,000, 3,000, and 10,000. A compilation of model input files is located in
Appendix 1A
Using this framework if a meter scale was applied, it would represent a set of conditions where the total
aquifer thickness was 350 m, the reservoir has a head of 28 meters at the dam and covers an area of 10 km 2.
The rectangular block of sediments (aquifer) cover a valley that is 5 km wide and 25 km long. The inflow
to the model (Q) is in m3/d and K is in m/d. Q/K ratio would be in units of square meters. Alternatively, if
a decimeter scale were applied the total aquifer thickness would be reduced to 35 m, the reservoir head
would be 2.8 m covering 0.1 km2, and the block of sediments would represent an area that is 0.5 km wide
by 2.5 km long. Q/K ratios would be square decimeters.
Figure 3
Illustration of a simplified homogeneous unitless (any consistent length and time unit could be used) three
layer box model of a valley river system containing a dam and associated reservoir. This model was
constructed to generally approximate the Milltown Reservoir valley in Western Montana (Berthelote et al.,
2007). It was built in Groundwater Vistas using MODFLOW with both the reservoir in place and removed
in order to conceptually examine groundwater impacts associated with various aquifer conditions (Q/K
ratios).
8
The relative difference in head between the river stage/reservoir stage and the groundwater for
hydrogeologic settings where the flow to hydraulic conductivity ratios (Q/K) are 1,000, 3,000, and 10,000
are presented in Figure 4. Under undammed conditions or post dam removal settings ( Figure 4A),
modeling suggests that when the Q/K ratio is large the water table is at or near the stream stage creating a
gaining stream reach in which groundwater discharges to the stream (Figure 2A conditions). In contrast,
when Q/K ratios are small the water table is lower than the stream channel stage and the river channel leaks
water into the groundwater system (losing channel Figure 2C). When a dammed setting is represented by
the model ( Figure 4B), larger ratio values appear to extend gaining river reaches farther upstream with
overall higher water table positions throughout the system (leaking of reservoir water into the
groundwater). Only the lowest ratio tested suggests the water table elevation proximal to the reservoir
would be lower than the river channel elevation (losing stream) below the dam.
Figure 4
Illustration of the modeled groundwater head profile coincident with the river channel (Figure 3).
Groundwater positions relative to the river channel elevation in each cell are presented for Q/K ratios of
1,000, 3,000 and 10,000 In A, the undammed scenario, as you increase the Q/K ratio by either increasing
the flow or decreasing hydraulic conductivity the groundwater level increases. Conversely it decreases
with a reduction in the ratio. In B, the dammed scenario, the area proximal to and downstream of the
reservoir is highly influenced by the recharge from the reservoir.
The box model results (Figure 4) illustrate steady state conditions with and without a dam and reservoir in
place. The figure can be used to illustrate the groundwater transition from free flowing stream conditions
(A) to a dammed scenario (B), or the response of a dammed system (B) to a dam removal (A). Though
these modeling results represent stabilized steady state settings it is realized that in some cases the dam
removal process is not instantaneous. Large scale dam removals are often done in stages. As a result, in
some settings, the river and groundwater systems may take months to years to fully revert to steady state
conditions following a dam removal. The results of the generic modeling will be evaluated by comparing
them to the observed groundwater response to the 2006 to 2009 removal of the 8.5 m high Milltown dam
and reservoir , western Montana (Berthelote 2013).
9
Figure 5
Illustration of the 9 km study area reach of the Clark Fork River Valley in Western Montana extending from
Hellgate Canyon to Turah Bridge. The Blackfoot river joins the Clark Fork River at the Milltown Dam and
associated reservoir in the center of the graphic. Locations of project monitoring wells and 500+ domestic
wells are also indicated for reference. USGS river data were obtained online for the station below the dam
indicated on the figure. Wells A, B, and C are highlighted as comparison wells and used to represent water
level changes that occurred during dam removal.
5.0
CASE STUDY - MILLTOWN RESERVOIR
The 8.5 m high Milltown Dam construction was completed in 1907 at the confluence of the Clark Fork and
Blackfoot rivers. Over the next 100 years, the Milltown reservoir filled with mining and smelter wastes
from the Butte and Anaconda area located 140 km upstream. The Milltown Reservoir was designated a
CERCLA (EPA Superfund) site in 1983 as water seeping from the reservoir sediments recharged the
adjacent coarse-grained aquifer and contaminated local wells with dissolved arsenic, iron and manganese
(ARCO 1992; Harding Lawson Associates (HLA) 1987; Moore and Woessner 2002; Udaloy 1988;
Woessner et al. 1984).
The river systems and reservoir are located in a semi-confined mountain valley setting in which the valley
floor sediments (6 to 60 m thick) are dominated by fluvial sand, gravel, cobbles and boulders deposited by
the ancestral Clark Fork and Blackfoot Rivers. Additional coarse grained sediments were deposited from
receding Glacial Lake Missoula floods. These sediments are bounded by steep mountain boundaries
composed of argillite, quartzite and limestone metasediments of the Precambrian Belt Series (Gestring,
1994). Residents located adjacent to and proximal to the reservoir utilize the prolific unconfined valley
aquifer (with water table depths of 2 to 35 m below land surface) for all domestic and municipal water
supplies (Berthelote and Woessner, 2009).
Initial pre-dam removal field data revealed the rivers and groundwater systems formed a complex
hydrogeological system. The highly conductive aquifer (range of hydraulic conductivity from 90 to
>27,000 m/d) was recharged from four sources: 1) a perched and leaking (losing) Blackfoot River arm of
the reservoir; 2)d most all of the losing Clark Fork River channel from just below the dam to Hellgate
Canyon (vertical riverbed hydraulic conductivities range from 0.4 to 12.8 m/d and leakage rates from
68,000 to 284,000 m3/d); 3) lateral underflow in the Clark Fork and Blackfoot river valley and; 4) limited
recharge from the mountain boundaries. Aquifer discharge occurred in a short gaining reach directly below
the dam (300m), as underflow through the valley aquifer located in Hellgate Canyon to the west (~283000
m3/d ), and locally in gaining stream sections of the Clark Fork River above the reservoir. Generally,
groundwater flowed towards the reservoir area from the upper Clark Fork River valley and converges just
above the reservoir with the groundwater entering at the mouth of the Blackfoot River canyon, then flows
northwest down valley (Figure 5). Valley widths are approximately 1 km +/- 0.5 km.
Mean annual flows for the Clark Fork River below the dam (USGS station # 12340500) ranged from 70
m3/d to 86 m3/d during the study period. The aquifer flow rates are measured in tens of meters per day due
to the coarse grained nature of the sediments, porosities of ~20% and prevailing groundwater gradients
10
(0.0013 near the upper Blackfoot River arm to about 0.066 near well B) (Berthelote et al. 2007; Berthelote
et al. 2010; Gestring 1994; Moore and Woessner 2002; Tallman 2005; Woessner et al. 1984). Field Data
report an average riverbed gradient of 0.004.
In 2004 the U. S. EPA, the State of Montana and other stakeholders decided to remove the 8.5 m high
Milltown Dam and 1.9 of the 5.0 mcm (million cubic meters) of contaminated reservoir sediments. The
goals of the removal efforts were to restore groundwater quality, provide fish passage, and return the two
rivers to a natural, free-flowing state (River Design Group et al., 2008; Westwater Consultants et al., 2005).
The Milltown Dam and the associated reservoir were removed during the period of 2006 to 2009
(Envirocon 2006; Westwater Consultants, River Design Group, and Geum Environmental Consulting 2005;
Envirocon 2006). This estimated $100+ million remediation/restoration project required three drawdowns
starting in 2006 (3.5 m in March 2007, 3.5 m in March 2008, and 1.2 m in April 2009) that correlated with
engineering tasks prior to reaching the final free flowing state in 2009. Before the dam removal process
formally began, project engineers initiated a 3.5 m temporary drawdown in November of 2005 to examine
the submerged portion of the dam. It was immediately observed that groundwater levels in some wells
adjacent to the reservoir declined and a few shallow domestic wells became inoperable. These conditions
resulted in the initiation of an expanded water level monitoring network and the construction and
calibration of an industry standard three dimensional numerical groundwater model (Berthelote et al., 2007;
Berthelote et al., 2010). The monitoring network consisted of 78 wells located at 56 locations. Ground
water levels at 22 wells were recorded at intervals no greater than 60 minutes using Solinst ® continuous
water level recorders (recording pressure transducers corrected with readings from a separate Solinst ®
barlogger). The remaining wells were measured monthly using an electric water level tape.
The State of Montana and U.S. EPA implemented a well replacement and mitigation program that
attempted to proactively mitigate water supply issues prior to likely well failures. A comparison of the
observational field data and the previously described conceptual and box models follows.
6.0
RESULTS
The Milltown Reservoir site would be classified as a semi-confined valley high water table system (Figure
2 as a B2 type scenario) from the middle of the reservoir upstream. The remaining valley below the
Blackfoot river arm is represented by a lower water table E2 type scenario. This difference is caused by the
widening and deepening of the valley sediments at the river confluence and is concurrently dependent on
the presence of high aquifer hydraulic conductivities.
The observed groundwater level data presented in Figure 6 suggests portions of the groundwater system
declined 2 to 3 meters as a consequence of the reservoir and dam remediation activities. This conclusion is
based on comparing the observed March 31, 2006 water table levels with the observed post dam out March
water levels. March water levels are consistently the lowest annual groundwater levels in this groundwater
system. The computed spatial distribution of the observed changes during the low water table period was
considered the best representation of the total impact to the groundwater levels from the removal activities
as observed in 2010. The water level declines were not limited to the reservoir area but extended at least 6
km downstream and 2 to 3 km upstream. The magnitude of change was dependent on the proximity to the
reservoir and groundwater flow directions.
11
Figure 6
Map showing the decline in meters of the March low water table position from March 2006 to March 2010.
Wells A, B, and C illustrate transient observation data at locations downstream at, and upstream of the
reservoir, respectively. The actual full pool wetted reservoir area was interpolated as no well data were
available in this area (pre-dam removal wells were lost to construction activities and no 2010 head
differences were available). However, the magnitude of this interpolated data (directly under the reservoir
bed) is consistent with anticipated declines underneath the removed 8.5 m deep reservoir.
Milltown Site was box modeled in units of feet with the dimensions, boundaries and gradients shown in
Figure 3. The dam was 2 8 ft high, the reservoir about 2000 ft by 5000 ft, the aquifer thickness 350 ft and
Q/K ratios ranged from 500 ft 2 to 2000 ft2. Generally, Q/K ratios average approximately 900 ft2 below the
dam (wells A and B) and 1300 ft2 above the dam (well C). Hhydraulic conductivity values in the Milltown
Aquifer site cover four orders of magnitude. A model was constructed using “average” sites conditions
with a Q/K ratio of 1,100 ft2. Then, in an attempt to better represent the variation in Q/A ratios known to
occur, a hybrid model was created by combining the results of two additional constant Q/K value models
(Figure 7). For the hybrid model, groundwater level changes above the dam area were derived from
completing the modeling with a constant Q/K ratio of 1,300 ft2. Results from a second set of model runs
using a constant Q/K ratio of 900 ft2 was used to represent conditions below the dam location Both the
use of average Q/K ratios and the combined hybrid model yield similar results and are inline with
observations (Figure 7B).
12
B
-1
-1
-2
-3
Q/K = 1000
-4
Q/K = 3000
-5
River/Pool Elevation
-2
Well C
Groundwater Decline (m)
Groundwater Decline
Box Model Results for Milltown Dam Removal
0
River/Pool Elevation
Well B
Box Model Results for generic dam removals
0
Well A
A
-3
-4
Hybrid Q/K (Milltown)
Observed Declines (2006 to 2010)
-5
Q/K = 10000
Average Q/K (Milltown)
-6
Distance Upstream (Cells)
Figure 7
250
200
150
100
50
0
250
200
150
100
50
0
-6
Distance Upstream (Cells)
A. Re-plotted Figures 4B. Illustration of the expected groundwater level declines from a dam removal
(differences between groundwater levels with the dam in place and without the dam, zero represents the
relative groundwater levels with the dam and reservoir in place). Values are illustrated as a profile
coincident with the river profile.
B. Illustration of the box model results for the Milltown Dam
removal representing the difference between dammed groundwater conditions (along the river profile, zero
equals dammed groundwater elevations) and post dam removal groundwater conditions (groundwater
declines) for a 1100 Q/K ratio (average conditions) and hybrid model that combines results of two
additional constant ratio models.
7.0
DISCUSSION
Study site heterogeneities often make it difficult to develop generic approaches for assessing responses of
groundwater systems to changes in stream or reservoir water levels. Hydrogeological conditions at the
Milltown Dam site were generally represented by the generic model developed for this research The
simplified box model (Figure 4) generally produced the expected outcomes when a site based average Q/K
ratio was uniformly applied. The simplified Q/K block model forecasted groundwater level declines that
compared favorably with observed changes, within about 1m.
The hybrid approach presented here simply combined results from two models that used different constant
Q/K ratios. These modeling results also closely approximated observations. Where conditions at Milltown
may be somewhat unique as both flows and hydraulic conductivities are high, modeling suggests that in
aquifers impacted by a dam and reservoir where the area is dominated with gaining stream reaches, impacts
may not be as large as observed at Milltown. In contrast, at sites where dams are removed, Q/K ratios are
small, and streams are losing groundwater changes may be larger than observed. In addition to the
uncertainties related to site conditions represented in any model used to forecast possible groundwater level
responses from a dam removal, isolating the response of the groundwater system that is solely attributed
to the dam removal action is not as straightforward as might be thought. As the model forecast produces
predicted changes under steady state conditions, in reality during the period of dam removal variations in
river flows and natural or induced groundwater recharge and discharge may produce additional
groundwater changes. These conditions may transiently reduce or enhance the groundwater response to a
dam removal operation. For example, at the Milltown site, a series of drought years prior to removal and
both normal water budget and drought conditions during the three year period of dam removal influenced
final observed water levels. Groundwater systems associated with reservoirs that are principally recharged
by the reservoir and river systems are most likely to have the largest changes in groundwater levels during
dam removals. It is recommended that environmental and hydrogeologic monitoring prior to, throughout,
and following future dam removals will provide needed data sets that can be used to develop predictive
groundwater level change models for future dam removals. When extensive pre-removal hydrogeologic
data sets are not available, the development of general box models such as the ones developed here can
provide managers with the general magnitudes and distributions of likely groundwater responses. Future
groundwater level mitigation activities associated with dam removals will undoubtedly benefit from
effective forecasts of groundwater level changes.
8.0
CONCLUSION
Project managers need to be keenly aware of the connection between dam removal activities and the likely
magnitude and distribution of groundwater level changes. This research derived a general conceptual
13
model of the effects of dam and reservoir emplacement and removal on associated groundwater systems.
Relationships between valley width, sediment scour, aggradation, and surface water groundwater
interactions are highlighted for each physical domain in a high and low water table setting (Figure 2). The
conceptual model tested used generic settings that showed that groundwater Q/K relationships are inversely
proportional to the likely magnitude of groundwater level impacts from dam removals. A case study is
presented and compared to the proposed conceptual model of groundwater responses. For the Milltown
Dam removal in western Montana, hydrographs and spatial extrapolation of observed groundwater level
declines indicate significant overall changes in the water table position since reservoir remediation.
Observed declines in groundwater levels following the dam removal were 2 m, 3 m, and 0.5 m at wells
located below the reservoir, at the dam site, and above the reservoir, respectively. The observational data
indicated that the magnitude in change was dependent on the proximity to the dam, and in this case,
extended over 6 km down stream of the dam location and 2 km upstream. Application of generic box
modeling that used both average Q/K site conditions and a hybrid approach produced similar magnitudes
and patterns of groundwater level declines. Forecasting likely groundwater changes prior to dam removal
activities will provide managers with information needed to initiate groundwater level mitigation planning.
14
9.0
REFERENCES
Alexander, M.D., and Caissie, D., 2003, Variability and Comparison of Hyporheic Water Temperatures and
Seepage Fluxes in a Small Atlantic Salmon Stream1: Ground Water, v. 41, p. 72-82.
American Indian Law Center, I., 1999, Tribal Model Research Code With Materials for Tribal Regulation
for Research and Checklist for Indian Health Boards, p. 29.
American Rivers, 2008, Groundwater Linkages to Dam Removal Research, in Wildman, L., ed., p.
Personal Communication.
American Rivers, Friends of the Earth, and Trout Unlimited, 1999, Dam Removal Success Stories:
Restoring Rivers Through Selective Removal of Dams That Don't Make Sense. Final Report, in
Maclin, E., and Sicchio, M., eds., p. 114.
Aseltyne, T., Rowe, H., and Fryar, A., 2006, Stable isotopic fingerprint of a hyporheic–hypolentic
boundary in a reservoir: Hydrogeology Journal, v. 14, p. 1688-1695.
Aun, F.J., 2008, Dam removal makes it cooler for trout, The Star-Ledger.
Babbitt, B., 2002, What Goes up, May Come down: BioScience, v. 52, p. 656-658.
Batalla, R.J., Gómez, C.M., and Kondolf, G.M., 2004, Reservoir-induced hydrological changes in the Ebro
River basin (NE Spain): Journal of Hydrology, v. 290, p. 117-136.
Baxter, R.M., 1977, Environmental Effects of Dams and impoundments: Annual Review of Ecology &
Systematics, v. 8, p. 255-283.
Bayley, P.B., 1995, Understanding large river-floodplain ecosystems: Bioscience, v. 45, p. 153-158.
Berthelote, A.R., Farinacci, A., Storb, M., and Woessner, W.W., 2007, Determining the impact of Milltown
Reservoir Drawdowns on the Adjacent Unconfined Aquifer: Remediation Stage 1 Observed and
Stage 2 Forecasted, University of Montana, p. 174.
Berthelote, A.R., and Woessner, W.W., 2009, Mitigating Groundwater Impacts from the Removal of a
Large Dam in an Unconfined Aquifer: Milltown Reservoir Remediation, University of Montana,
p. 178.
Berthelote, A.R., Woessner, W.W., and Thompson, J.H., 2010, Assessing Groundwater Level Behavior and
Numerical Model Forecasts of Water Table Changes Related to Removal of a Large Dam:
Milltown Reservoir Remediation, University of Montana, p. 113.
Beyer, P.J., 2005, Introduction to the special issue: Dams and geomorphology: Geomorphology, v. 71, p. 12.
Bouma, K., 2008, Cahaba River dam removal restores river life, The Birmingham News Birmingham,
Alabama.
Bowman, M., Higgs, S., Maclin, E., McClain, S., Sicchio, M., Souers, A., Johnson, S., and Graber, B.,
2002, Exploring Dam Removal: A Decision-Making Guide, American Rivers Trout Unlimited, p.
80.
Bowman, M.B., 2002, Legal Perspectives on Dam Removal: BioScience, v. 52, p. 739-747.
Bushaw-Newton, K.L., Hart, D.D., Pizzuto, J.E., Thomson, J.R., Egan, J., Ashley, J.T., Johnson, T.E.,
Horwitz, R.J., Keeley, M., Lawrence, J., Charles, D., Gatenby, C., Kreeger, D.A., Nightengale, T.,
Thomas, R.L., and Velinsky, D.J., 2002, An Integrative Approach Towards Understanding
Ecological Responses To Dam Removal: The Manatawny Creek Study: Journal of the American
Water Resources Association, v. 38, p. 1581-1599.
Butler, D.R., and Malanson, G.P., 2005, The geomorphic influences of beaver dams and failures of beaver
dams: Geomorphology, v. 71, p. 48-60.
Caduto, M.J., 2008, Un-damming our rivers, Rutland Herald: Rutland, Vermont.
Cantelli, A., Wong, M., Parker, G., and Paola, C., 2007, Numerical model linking bed and bank evolution
of incisional channel created by dam removal: Water Resources Research, v. 43, p. 1-16.
CDRI, 2012, Clearinghouse for Dam Removal Information, University of California Riverside.
Chen, X., and Chen, X., 2003, Stream water infiltration, bank storage, and storage zone changes due to
stream-stage fluctuations: Journal of Hydrology, v. 280, p. 246-264.
Chin, A., Harris, D.L., Trice, T.H., and Given, J.L., 2002, Adjustment of Stream Channel Capacity
Following Dam Closure, Yegua Creek, Texas: Journal of the American Water Resources
Association, v. 38, p. 1521-1531.
Collier, M., Webb, R.H., and Schmidt, J.C., 1996, Dams and Rivers: Primer on the Downstream Effects of
Dams, U.S. Geological Survey Circular 1126.
15
Collins, M., Lucey, K., Lambert, B., Kachmar, J., Turek, J., Hutchins, E., Purinton, T., and Neils, D., 2007,
Stream Barrier Removal Monitoring Guide, Gulf of Main Council on the Marine Environment,
Volume 2008, www.gulfofmaine.org/streambarrierremoval.
Constantz, J., 2003, Dams and downstream ground water: Hydrological Processes, v. 17, p. 3533-3535.
Constantz, J., and Essaid, H., 2007, Influence of groundwater pumping on streamflow restoration following
upstream dam removal: Hydrological Processes, v. 21, p. 2823-2834.
Cui, Y., Parker, G., Braudrick, C., Deietrich, W.E., and Cluer, B., 2006, Dam Removal Express
Assessment Models (DREAM). Part 1: Model developement and validation: Journal of Hydraulic
Research, v. 44, p. 291-307.
Dahm, C.N., Grimm, N.B., Marmonier, P., Valett, H.M., and Vervier, P., 1998, Nutrient dynamics at the
interface between surface waters and groundwaters: Freshwater Biology, v. 40, p. 427-451.
Dean, C., 2008, Follow the Silt, The New York Times: New York, New York.
Dendy, F.E., 1968, Sedimentation In The Nation's Reservoirs: Journal of Soil and Water Conservation, v.
23, p. 135-137.
DesGranges, J.L., Rodrigue, J., Tardif, B., and Laperle, M., 1998, Mercury accumulation and
biomagnification in ospreys (Pandion haliaetus) in the James Bay and Hudson Bay regions of
Quebec: Archives of Environmental Contamination & Toxicology, v. 35, p. 330.
Doyle, M.W., Stanley, E.H., and Harbor, J.M., 2002, Geomorphic Analogies For Assessing Probable
Channel Response To Dam Removal: Journal of the American Water Resources Association, v.
38, p. 1567-1579.
Doyle, M.W., Stanley, E.H., and Harbor, J.M., 2003a, Channel adjustments following two dam removals in
Wisconsin: Water Resources Research, v. 39, p. 1011.
Doyle, M.W., Stanley, E.H., Harbor, J.M., and Grant, G.S., 2003b, Dam Removal in the United States:
Emerging Needs for Science and Policy: EOS, Transactions, American Geophysical Union, v. 84,
p. 29-36.
Doyle, M.W., Stanley, E.H., Orr, C.H., Selle, A.R., Sethi, S.A., and Harbor, J.M., 2005, Stream ecosystem
response to small dam removal: Lessons from the Heartland: Geomorphology, v. 71, p. 227-244.
Egan, D.A., 2008, Dam removal might bring walleye boom, The Plain Dealer: Cleveland, Ohio.
Evans, J.E., Huxley, J.M., and Vincent, R.K., 2007, Upstream Channel Changes Following Dam
Construction and Removal Using a GIS/Remote Sensing Approach1: Journal of the American
Water Resources Association, v. 43, p. 683-697.
Evans, J.E., Levine, N.S., Roberts, S.J., Gottgens, J.F., and Newman, D.M., 2002, Assessment Using GIS
And Sediment Routing Of The Proposed Removal Of Ballville Dam, Sandusky River, Ohio:
Journal of the American Water Resources Association, v. 38, p. 1549-1565.
Evans, J.E., Mackey, S.D., Gottgens, J.F., and Gill, W.M., 2000a, From Reservoir to Wetland: The Rise
and Fall of an Ohio Dam, in Schneiderman, J.S., ed., The Earth Around Us: Maintaining a Livable
Planet: Boulder, Colorado, Westview Press, p. 256-267.
Evans, J.E., Mackey, S.D., Gottgens, J.F., and Gill, W.M., 2000b, Lessons from a dam failure: The Ohio
Journal of Science, v. 100, p. 121-131.
Evans, J.K., Gottgens, J.F., Gill, W.M., and Mackey, S.D., 2000c, Sediment Yields Controlled by
Intrabasinal Storage and Sediment Conveyance over the Interval 1842-1994: Chagrin River,
Northeast Ohio, U.S.A: Journal of Soil and Water Conservation, v. 55, p. 264.
Faulkner, D., and McIntyre, S., 1996, Persisting Sediment yields and Sediment Delivery Changes: Water
Resources Bulletin, v. 32, p. 817-829.
Francisco, E., 2004, Tales Of The Undammed: Science News, v. 165, p. 235-237.
Gestring, S.L., 1994, The interaction of the Clark Fork River and the Hellgate Valley Aquifer near,
Milltown, Montana. M.S. Thesis, Geology Department, University of Montana.
Graf, W.L., 2002, Dam Removal: Science and Decision Making, The H. John Heinz III Center for Science,
Economics and the Environment, p. 236.
Graf, W.L., 2003, Dam Removal Research: Status and Prospects, in Graf, W.L., ed., The Heinz Center's
Dam Removal Research Workshop: Warrenton, Virginia, The H. John Heinz III Center for
Science, Economics and the Environment, p. 1-151.
Graf, W.L., 2005, Geomorphology and American dams: The scientific, social, and economic context:
Geomorphology, v. 71, p. 3-26.
Graf, W.L., 2006, Downstream hydrologic and geomorphic effects of large dams on American rivers:
Geomorphology, v. 79, p. 336-360.
16
Grams, P.E., and Schmidt, J.C., 2005, Equilibrium or indeterminate? Where sediment budgets fail:
Sediment mass balance and adjustment of channel form, Green River downstream from Flaming
Gorge Dam, Utah and Colorado: Geomorphology, v. 71, p. 156-181.
Gregory, S., Li, H., and Li, J., 2002, The Conceptual Basis for Ecological Responses to Dam Removal:
BioScience, v. 52, p. 713-723.
Hall, G., 2005, Ocklawaha River Water Allocation Study, in Greeneville Hall, P.D., ed.: Palatka, Florida,
St. Johns River Water Management District.
Hart, D.D., Johnson, T.E., Bushaw-Newton, K.L., Horwitz, R.J., Bednarek, A.T., Charles, D.F., Kreeger,
D.A., and Velinsky, D.J., 2002, Dam Removal: Challenges and Opportunities for Ecological
Research and River Restoration: BioScience, v. 52, p. 669-681.
Hart, D.D., and Poff, N.L., 2002, [Introduction]: A Special Section on Dam Removal and River
Restoration: BioScience, v. 52, p. 653-655.
Hayashi, M., and Rosenberry, D.O., 2002, Effects of Ground Water Exchange on the Hydrology and
Ecology of Surface Water: Ground Water, v. 40, p. 309-316.
Heilweil, V.M., Susong, D.D., Gardner, P.M., and Watt, D.E., 2005, Pre- and Post-Reservoir GroundWater Conditions and Assessment of Artificial Recharge at Sand Hollow, Washington County,
Utah, 1995-2005: Salt Lake City, Utah, U.S. Geological Survey.
Hendricks, S.P., and White, D.S., 1991, Physicochemical Patterns within a Hyporheic Zone of a Northern
Michigan River, with Comments on Surface Water Patterns: Canadian Journal of Fisheries and
Aquatic Sciences, v. 48, p. 1645-1654.
Hewitt, L., Graber, B., and Lindloff, S., 2001, Restoring the Flow: Improving Selective Small Dam
Removal Understanding and Practice in Great Lakes States, Small Dam Removal Workshop &
Working Meeting: Kohler, Wisconsin, p. 1-25.
ICOLD, 2007, Dams and the world's water: An educational book that explains how dams help to manage
the world's water: Paris France, International Committee on Large Dams, 66 p.
James, L.A., 2005, Sediment from hydraulic mining detained by Englebright and small dams in the Yuba
basin: Geomorphology, v. 71, p. 202-226.
Johnson, S.E., and Graber, B.E., 2002, Enlisting the Social Sciences in Decisions about Dam Removal:
BioScience, v. 52, p. 731-738.
Kondolf, G.M., 1998, Lessons learned from river restoration projects in California: Aquatic Conservation:
Marine and Freshwater Ecosystems, v. 8, p. 39-52.
Landers, J., 2004, River Renaissance: Civil Engineering (08857024), v. 74, p. 52-59.
Learn, S., 2011, Well owners want PacifiCorp to pay more for damage done by removal of Condit Dam,
The Oregonian.
Leopold, L.B., and Maddock, T.M., 1954, The flood control controversy, Big dams, little dams, and land
management.
Lindloff, S.D., and Wildman, L.A.S., 2006, Permitting Dam Removal: The State of (Several) States, The
Association of State Dam Safety Officials Annual National Conference: Boston, Massachusetts,
American Rivers.
Lorang, M.S., and Aggett, G., 2005, Potential sedimentation impacts related to dam removal: Icicle Creek,
Washington, U.S.A: Geomorphology, v. 71, p. 182-201.
Lowry, M.M., and Beschta, R.L., 1994, Effects of a beaver pond on groundwater elevation and
temperatures in a recovering stream ecosystem in Marston, R.A., and Hasfurther, V.R., eds.,
Annual Summer Symposium of the American Water Resources Association: Effects of HumanInduced Changes on Hydrologic Systems: Bethesda, Maryland, American Water Resources
Association, p. 503-516.
Lytle, D.A., and Poff, N.L., 2004, Adaptation to natural flow regimes: Trends in Ecology & Evolution, v.
19, p. 94-100.
Magilligan, F.J., and Nislow, K.H., 2005, Changes in hydrologic regime by dams: Geomorphology, v. 71,
p. 61-78.
Marston, R.A., Mills, J.D., Wrazien, D.R., Bassett, B., and Splinter, D.K., 2005, Effects of Jackson Lake
Dam on the Snake River and its floodplain, Grand Teton National Park, Wyoming, USA:
Geomorphology, v. 71, p. 79-98.
Martin, G., 2004, Battle of Battle Creek: Which way to save salmon?; Environmentalists are split over plan
that would remove 5 of 8 small PG&E hydropower dams -- some think it's not enough.(NEWS):
San Francisco Chronicle, p. A1.
17
McClain, S., 2012, 60 dams removed to restore rivers in 2010, Volume 2012, American Rivers.
McCool, D., 2004, As dams fall, a chance for redemption, High Country News, Volume 36.
Mertes, L.A.K., 1997, Documentation and Significance of the Perirheic Zone on Inundated Floodplains:
Water Resources Research, v. 33, p. 1749-1762.
Muth, R.T., Crist, L.W., LaGory, K.E., Hayse, J.W., Bestgen, K.R., Ryan, T.P., Lyons, J.K., and Valdez,
R.A., 2000, Flow and Temperature Recommendations for Endangered Fishes in the Green River
Downstream of Flaming Gorge Dam. In: Upper Colorado River Endangered Fish Recovery
Program Final Report, Volume 48: Lakewood, Colorado, p. 1-344.
Nadeau, T.-L., and Rains, M.C., 2007, Hydrological Connectivity of the Headwaters to Downstream
Waters: Introduction to the Featured Collection: Journal of the American Water Resources
Association, v. 43, p. 1-4.
Naiman, R.J., Johnston, C.A., and Kelley, J.C., 1988, Alteration of North American Streams by Beaver:
BioScience, v. 38, p. 753-762.
Nilsson, C., Reidy, C.A., Dynesius, M., and Revenga, C., 2005, Fragmentation and Flow Regulation of the
World's Large River Systems Science, v. 308, p. 405-408.
Nislow, K.H., Magilligan, F.J., Fassnacht, H., Bechtel, D., and Ruesink, A., 2002, Effects Of Dam
Impoundment On The Flood Regime Of Natural Floodplain Communities In The Upper
Connecticut River: Journal of the American Water Resources Association, v. 38, p. 1533-1548.
NMFS, N.M.F.S., 2010, Environmental Assessment For The Arra Rogue River Restoration - Gold Ray
Dam Project: Jackson County, National Marine Fisheries Service, p. 110.
O'Conner, J., Major, J., and Grant, G., 2008, Down with the Dams: Unchaining U.S. Rivers: Geotimes.
Oregon State, 2008, Groundwater in Dam Removal Research, in Dam Removal Listserv, ed., Oregon State,
p. Personal Communication.
Palmer, M.A., 1993, Experimentation in the Hyporheic Zone: Challenges and Prospectus: Journal of the
North American Benthological Society, v. 12, p. 84-93.
PBS, 2008, Building Big: Dam Basics, in WGBH Educational Foundation, ed., Volume 2008, Public
Broadcasting System.
Pegg, J.R., 2004, Dams: Blessing and Curse for World's Largest Rivers in Environment News Service, ed.,
Volume June, International Daily Newswire.
Pejchar, L., and Warner, K., 2001, A River Might Run Through It Again: Criteria for Consideration of
Dam Removal and Interim Lessons from California: Environmental Management, v. 28, p. 561575.
Pennsylvania Organization for Watersheds and Rivers, American Rivers, Alliance for Aquatic Resources
Monitoring, and Alliance for the Chesapeake Bay, 2004, Citizen's Guide to Dam Removal
Monitoring and Restoration: Harrisburg, PA p. 1-36.
Petts, G.E., 1984, Impounded Rivers. Perspectives for Ecological Management: Chichester, John Wiley and
Sons, 326 p.
Petts, G.E., and Gurnell, A.M., 2005, Dams and geomorphology: Research progress and future directions:
Geomorphology, v. 71, p. 27-47.
Pizzuto, J., 2002, Effects of Dam Removal on River Form and Process: BioScience, v. 52, p. 683-691.
Poff, N.L., Allan, J.D., Bain, M.B., Karr, J.R., Prestegaard, K.L., Richter, B.D., Sparks, R.E., and
Stromberg, J.C., 1997, The Natural Flow Regime: BioScience, v. 47, p. 769-784.
Poff, N.L., Bledsoe, B.P., and Cuhaciyan, C.O., 2006, Hydrologic variation with land use across the
contiguous United States: Geomorphic and ecological consequences for stream ecosystems:
Geomorphology, v. 79, p. 264-285.
Poff, N.L., and Hart, D.D., 2002, How Dams Vary and Why It Matters for the Emerging Science of Dam
Removal: BioScience, v. 52, p. 659-668.
Pohl, M.M., 2002, Bringing Down Our Dams: Trends In American Dam Removal Rationales: Journal of
the American Water Resources Association, v. 38, p. 1511-1519.
Power, M.E., Dietrich, W.E., and Finlay, J.C., 1996, Dams and Downstream Aquatic Biodiversity:
Potential Food Web Consequences of Hydrologic and Geomorphic Change: Environmental
Management, v. 20, p. 887-895.
Pusch, M., Fiebig, D., Brettar, I., Eisenmann, H., Ellis, B.K., Kaplan, L.A., Lock, M.A., Naegeli, M.W.,
and Traunspurger, W., 1998, The role of micro-organisms in the ecological connectivity of
running waters: Freshwater Biology, v. 40, p. 453-495.
18
Pyle, M.T., 1995, Beyond Fish Ladders: Dam Removal as a Strategy for Restoring America's Rivers:
Stanford Environmental Law Journal, v. 14, p. 97.
Rãdoane, M., and Rãdoane, N., 2005, Dams, sediment sources and reservoir silting in Romania:
Geomorphology, v. 71, p. 112-125.
Rains, M.C., Mount, J.F., and Larsen, E.W., 2004, Simulated changes in shallow groundwater and
vegetation distributions under different reservoir operations scenarios: Ecological Applications, v.
14, p. 192-207.
Renwick, W.H., Smith, S.V., Bartley, J.D., and Buddemeier, R.W., 2005, The role of impoundments in the
sediment budget of the conterminous United States: Geomorphology, v. 71, p. 99-111.
River Design Group, I., Westwater Consultants, I., and Geum Environmental Consulting, I., 2008,
Restoration Plan for the Clark Fork River and Blackfoot River near Milltown Dam, p. 72.
Roberts, S.J., Gottgens, J.F., Spongberg, A.L., Evans, J.E., and Levine, N.S., 2006, Integrated approach to
assessing potential dam removals: An example from the Ottawa River: Journal of Environmental
Management, Northwestern Ohio.
Rowntree, K.M., and Dollar, E.S.J., 1999, Vegetation controls on channel stability in the Bell River,
Eastern Cape, South Africa: Earth Surface Processes and Landforms, v. 24, p. 127-134.
Sandison, D., 2010, Condit Dam Removal Final Second Supplemental Environmental Impact Statement:
Yakima. WA, Department of Ecology, p. 85.
Shafroth, P.B., Friedman, J.M., Auble, G.T., Scott, M.L., and Braatne, J.H., 2002, Potential Responses of
Riparian Vegetation to Dam Removal: BioScience, v. 52, p. 703-712.
Shuman, J.R., 1995, Environmental Considerations For Assessing Dam Removal Alternatives For River
Restoration: Regulated Rivers: Research & Management, v. 11, p. 249-261.
Stanford, J.A., 1998, Rivers in the landscape: introduction to the special issue on riparian and groundwater
ecology: Freshwater Biology, v. 40, p. 402-406.
Stanford, J.A., and Ward, J.V., 1993, An ecosystem perspective of alluvial rivers: connectivity and the
hyporheic corridor: Journal of the North American Benthological Society, v. 12, p. 48-60.
Stanley, E.H., and Doyle, M.W., 2002, A Geomorphic Perspective on Nutrient Retention Following Dam
Removal: BioScience, v. 52, p. 693-701.
Tweit, S., 2006, Tearing Down The Elwha River Dam, Popular Mechanics.
U.S. Dept. of the Interior, 2008, National Atlas of the United States, Volume 2008,
http://nationalatlas.gov/atlasftp.html#dams00x.
USACE, 2008a, Groundwater Linkages to Dam Removals or Emplacements, in Conyngham, J., ed., United
States Army Corps of Engineers Environmental Laboratory's PI in Dam Removal Research, p.
Personal Communication.
USACE, 2008b, National Inventory of Dams NID, in United States Army Topographic Engineering Center,
ed., http://crunch.tec.army.mil/nidpublic/webpages/nid.cfm.
USDA, 2010, Environmental Assessment Lower Musconetcong River Restoration Project Finesville Dam
Vicinity: Somerset, New Jersey, USDA Natural Resources Conservation Service.
Valett, H.M., Fisher, S.G., and Stanley, E.H., 1990, Physical and Chemical Characteristics of the
Hyporheic Zone of a Sonoran Desert Stream: Journal of the North American Benthological
Society, v. 9, p. 201-215.
Westbrook, C.J., Cooper, D.J., and Baker, B.W., 2006, Beaver dams and overbank floods influence
groundwater–surface water interactions of a Rocky Mountain riparian area: Water Resources
Research, v. 42, p. 12.
Westwater Consultants, I., River Design Group, I., and Geum Environmental Consulting, I., 2005,
Restoration Plan for the Clark Fork and Blackfoot River near Milltown Dam – October 2005.
Whitelaw, E., and MacMullan, E., 2002, A Framework for Estimating the Costs and Benefits of Dam
Removal: BioScience, v. 52, p. 724-730.
Williams, G.P., and Wolman, M.G., 1984, The downstream effects of dams on alluvial rivers, United States
Geological Survey.
Woo, M.-K., and Waddington, J.M., 1990, Effects of Beaver Dams on Subarctic Wetland Hydrology:
Arctic, v. 43, p. 223-230.
Wootton, J.T., Parker, M.S., and Power, M.E., 1996, Effects of Disturbance on River Food Webs: Science,
v. 273, p. 1558-1561.
World Commission on Dams, 2000, Dams and Development: A New Framework For Decision-Making:
London and Sterling, VA, Earthscan Publications Ltd, 404 p.
19
Wyrick, J.R., Rischman, B.A., Burke, C.A., McGee, C., and Williams, C., 2009, Using hydraulic modeling
to address social impacts of small dam removals in southern New Jersey: Journal of
Environmental Management, v. 90, Supplement 3, p. S270-S278.
20
Appendix 1A
Box model Input Files for different Q/K ratios (100, 1000, 3000, 10000) for
dammed and natural systems (Files available in digital format only)
21
Chapter/Paper 2
Berthelote, Antony, Doctor of Philosophy, May 2013
Geosciences
Proactive Mitigation of Domestic and Municipal Groundwater Supplies During Dam Removal
Actions, Milltown Reservoir, Western Montana
Chairperson: Dr. William W. Woessner
ABSTRACT:
It has recently been recognized that reservoir recharged groundwater supplies are increasingly being
threatened by dam removals. Managing groundwater system responses to dam and reservoir removals
and the resulting economic consequences require time effective mitigation strategies informed by
groundwater level forecasting. In western Montana project managers of the recent 8.5 m high Milltown
Dam and the contaminated reservoir sediment removal implemented a well replacement mitigation
strategy that attempted to proactively mitigate groundwater supply impacts. At the initiation of dam
removal activities, an extensive groundwater and river system monitoring network was established to
observe water level changes. A suite of multi- layer three dimensional finite difference groundwater
models calibrated to both historical data and observed groundwater responses to staged reservoir
drawdowns provided groundwater level forecasts of post-dam out conditions. These data were inputs
to a decision tree that identified wells needing replacement, lowering of pumps, further data collection,
or were considered not at risk. Uncertainty was evaluated using sensitivity analyses and resulting
alternative conceptual models. Observed results showed water table changes were up to 3 m near the
reservoir and impacts extended 6 km downstream and about 2 km upstream of the reservoir. Model
forecasts of groundwater level changes were greater than post dam removal observed levels, as
expected, because it is likely that natural variations in annual recharge and river leakage that occurred
during the dam and sediment removal period did not match low water table position forecasts. Risk
tree analyses showed up to 115 wells were at risk. Proactive mitigation included the construction of 80
new wells and lowering of 20 pumps. Most future dam removal projects would benefit from
groundwater impact analyses and proactive groundwater supply mitigation as needed. This process
should include both observations of changes and pre-dam removal forecasting of groundwater
responses. Forecasts should then be assessed and actions taken based on a logical project based
decision tree such as the one developed here.
KEY WORDS
Groundwater Level Forecasting; 3D Numerical MODFLOW Modeling; Milltown
Reservoir; Dam Removal Mitigation And Management; Groundwater Surface Water
Interactions
1.0
INTRODUCTION
There is increasing interest in removing dams in the United States to remedy adverse ecological impacts,
eliminate risks associated with the deteriorating conditions of aging dams, and address societal pressures to
restore rivers to more natural settings (Babbitt, 2002; Farinacci, 2009; Hart et al., 2002; Landers, 2004;
O'Conner et al., 2008). Concerns have been raised that emptying reservoirs as dams are removed will have
local to regional impacts on domestic, municipal, and agricultural groundwater availability and use. Until
recently, dam removal projects were commonly planned without consideration of the likely corresponding
changes in groundwater conditions (Berthelote, 2013Chapter 1). Previous evaluations of water table
impacts resulting from dam removals have been limited to comparative studies of small scale beaver dam
removals, a model that may not adequately represent groundwater responses for large scale dam removal
actions (Butler and Malanson, 2005; Chen and Chen, 2003; Mertes, 1997; Naiman et al., 1988; Westbrook
et al., 2006; Woo and Waddington, 1990). It is encouraging that recent pre-dam removal environmental
assessments for the 2.75 m high Finesville Dam (Musconetcong River in NJ) (USDA, 2010) and 11.2 m
Gold Ray Dam (Rogue River in OR) (NMFS, 2010) report the possibility of well failures. However, “after
the fact” well mitigation is more the norm. As an example, the 39 m high Condit Dam (White Salmon
22
River in WA) Environmental Impact Statement (EIS) predicted “... no significant unavoidable adverse
impacts…” from groundwater level changes (Sandison, 2010). However, well mitigation was being
considered to replace multiple failed wells following the dam removal (Sandison, 2010). Though current
EIS statements and public reactions are highlighting potential negative impacts to groundwater systems
from dam removal actions, rigorous proactive research and planning are conspicuously absent, a condition
that may result in costly mitigation and litigation in some settings (Bowman et al., 2002; Bowman, 2002).
This research was designed to assess a process used to forecast groundwater level changes prior to and
during a dam removal, and evaluate if mitigation strategies executed prior to the final dam out scenario
were effective in reducing actual and perceived impacts to associated groundwater supplies. The following
hypothesis was tested:
Hypothesis: Model forecasts of the water table position resulting from planned sequential drawdowns and
final dam removal will identify at least 70% of the water supply wells requiring groundwater level response
mitigation.
We evaluated this hypothesis during the staged 8.5 m high Milltown Dam removal project that occurred
between 2006 and 2009. The dam created Milltown Reservoir located in the semi-confined alluvial valley
of the Clark Fork River in western Montana. Just prior to the initiation of the first phase of planned
reservoir drawdowns, a groundwater monitoring network was initiated. Numerical groundwater modeling
was developed to forecast the response of the water table to reservoir drawdowns and final dam out
conditions. As staged removal plans were executed, a decision tree populated by modeling forecasts was
created that assessed the risk of impairment of valley domestic and municipal water supplies. The
processes used to derive and assess observations and model results within a risked based decision tree are
the subject of this work.
2.0
MILLTOWN DAM SITE
The 8.5 m high Milltown Dam construction was completed in 1907 at the confluence of the Clark Fork and
Blackfoot Rivers (Figure 1). Mean annual flows for the Clark Fork River below the dam (USGS station #
12340500) ranged from 70 m3/day to 86 m3/day during the 2006 to 2010 study period. Over the last
century, the Milltown reservoir filled with mining and smelter wastes from the Butte and Anaconda area
located 180 km upstream. The Milltown Reservoir was designated a CERCLA (U.S. EPA Superfund) site
in 1983 as water seeping from the reservoir sediments recharged the adjacent coarse-grained aquifer and
contaminated local wells with dissolved arsenic, iron, and manganese ((ARCO) Atlantic Richfield
Company, 1992; Harding Lawson Associates (HLA), 1987; Moore and Woessner, 2002; Udaloy, 1988;
Woessner et al., 1984).
23
Figure 1
The 15 km study area reach of the Clark Fork River valley in Western Montana extending from Hellgate
Canyon to Turah Bridge. The Blackfoot River joins the Clark Fork River at the Milltown Dam forming the
Milltown Reservoir. Locations of project monitoring wells and 500+ domestic wells are also indicated for
reference. USGS river data were obtained for the stations illustrated. The blue arrows identify the general
direction of groundwater flow.
The river systems and reservoir are located in a semi-confined mountain valley setting in which the valley
floor sediments (predominantly 6 to 60 m thick) are dominated by fluvial sand, gravel, cobbles, and
boulders deposited by the ancestral Clark Fork and Blackfoot Rivers. Additional coarse grained sediments
were deposited during the Glacial Lake Missoula floods. These sediments are bounded by steep mountain
boundaries composed of argillite, quartzite, and limestone metasediments of the Precambrian Belt Series
(Gestring, 1994). Residents located adjacent to and proximal to the reservoir utilize the prolific unconfined
valley aquifer (with water table depths of 2 to 35 m below land surface) for all domestic and municipal
water supplies (Berthelote and Woessner, 2009).
In 2004 the U.S. EPA, the State of Montana, and stakeholders decided to remove the 8.5 m high Milltown
Dam and 1.9 of the 5.0 mcm (million cubic meters) of contaminated reservoir sediments to restore
groundwater quality, provide fish passage, and return the two rivers to a natural free-flowing state (River
Design Group et al., 2008; Westwater Consultants et al., 2005). The Milltown Dam and the associated
reservoir were removed during the period of 2006 to 2009. Remediation and restoration plans were
designed to be completed in stages over a number of years (Envirocon, 2006; River Design Group et al.,
2008). This estimated $100+ million remediation/restoration project required three planned drawdowns
starting in 2006 (3.6 m in June 2006, 3.6 m in March 2008, and 1.3 m in April 2009) that correlated with
engineering tasks prior to reaching the final free flowing state in 2009.
Before the dam removal process formally began, project engineers initiated a 3.5 m temporary drawdown
in November of 2005 to examine the submerged portion of the dam. It was observed that groundwater
levels in some wells adjacent to the reservoir declined and a few shallow domestic wells became
inoperable. These observations prompted the development of an assessment tool that could be used to
identify wells that would likely be impaired and preemptively replace or remediate them prior to the
initiation of the next phase of drawdowns. As the second and third (final dam removal) drawdowns
occurred, the assessment tool was revised to incorporate new observations and to evaluate if additional
remediation was needed.
24
3.0
CONCEPTUAL MODEL OF IMPACTS TO GROUNDWATER LEVELS FROM DAM
AND RESERVOIR REMOVAL
Relying on the literature and basic hydrogeological theory, Berthelote (2013) developed generic conceptual
models of how groundwater systems associated with dam removals would respond under various
hydrogeological and geomorphic settings. At the Milltown site, groundwater conditions were generally
higher than river levels in the river section above the dam and lower than river levels at and below the
reservoir area. Berthelote’s (2013) conceptual models fit to Milltown conditions would combine semiconfined high water table (above the dam) and low water table (below the dam) conceptualizations (Figure
2). This model illustrates the progression of hydrogeologic changes expected in Milltown following the
breach of the dam. Above the dam, in the high water table setting, slight alterations in the surface water
groundwater exchanges will ensue in response to removal of the reservoir and river
reconstruction/restoration activities (new gradients, river configurations, bank stabilization, etc). Proximal
to and below the dam it is anticipated that vertical channel bed gradients (influent conditions) will all
convert to or remain downward, and valley wide groundwater levels will decline with the magnitude of
changes being inversely proportional to the distance down gradient from the dam,.
Figure 2 Hybridization of Berthelote’s (2013) conceptual models fit to Milltown conditions which combines his semiconfined high water table (above the dam) and low water table (below the dam) conceptualizations. This
model illustrates the progression of hydrogeologic changes expected in Milltown following the breach of
the dam. Generalized hydrographs for wells above and below the dam are presented to demonstrate likely
groundwater responses to dam removals immediately following a breach and after the hydrogeologic
system is fully restored.
Prior to implementing any monitoring, modeling, or mitigation strategies, project managers understood that
they would need to employ adaptive mitigation management strategy that relied on historical data and
interpretations, newly collected groundwater level and flux data, and professional knowledge. It was
understood from the initiation of the project that potential alterations to the planned engineering activities
(timing of each staged drawdown, magnitude of the individual drawdowns, timing of the diversion into a
planned bypass channel), issues with monitoring data (sparse or erratic historical data and maintaining
consistent access to monitoring or private wells), multiple modeling calibration issues (irregular or transient
data sets that would be updated as new data became available), and other transient hydrological concerns
would likely arise throughout the project. The key goal of this work was to provide project managers
(regulators) with the adequate tools and knowledge that they could use to plan and implement water supply
remediation. The timeline in Figure 3 illustrates the accessible transient knowledge base. The subsequent
methods section describes the key methodological components of this timeline. Though the project
extended over 5+ years, the first wells were mitigated in April of 2006, only one month after the
monitoring and modeling process commenced. The time constraints for all mitigation actions (majority
25
completed during 2007 and early 2008) dictated that the first year of data collection and analyses would
formulate the key information used to make the majority of mitigation decisions.
Figure 3 A timeline summarizing the sequencing of events described in this work (vertical columns), changes in site
conditions, groundwater level data collection, and the timing and number of wells mitigated in response to modeling
results. Data for this figure is located in Appendix 2A
4.0
METHODS
To forecast the spatial distribution and magnitude of groundwater level change, a strategy was developed to
collect appropriate data sets and to perform required detailed analyses. The strategy had five primary parts;
1) locate and correlate all historical groundwater data; 2) establish a well monitoring program to evaluate
groundwater level changes throughout the life of the project including the cataloging of well construction
and pumping parameters; 3) construct a three dimensional groundwater model that was capable of
adequately representing the historical and current groundwater levels and fluxes; 4) use the model to
geospatially and temporally forecast likely groundwater level declines resulting from each staged
drawdown and final dam removal; and 5) develop a risk based decision tree to assist project managers with
mitigation activities.
This methods section will first present a detailed description of the water level monitoring network and the
establishment of background pre reservoir drawdown groundwater level conditions. Next, a brief
description of the construction, calibration, and application of an industry standard three dimensional
numerical groundwater model is presented. Thirdly, the development of a risk based decision tree and its
application to well mitigation is described.
26
4.1
Baseline Conditions and Groundwater Level Monitoring
4.1.1
Well Monitoring Network
The monitoring network consisted of 78 wells located at 56 locations (Figure 1). Groundwater levels at 22
wells were recorded at intervals no greater than 60 minutes using Solinst ® continuous water level recorders
(recording pressure transducers corrected with readings from a separate Solinst ® barlogger). The remaining
wells were measured monthly using an electric water level tape. Appendix 2B contains a compilation of
historical water level data and data derived from this work (1981 to 2010).
4.1.2
Surface Water Monitoring Network
A network of surface water stage gauges was established. Surface water elevations were obtained from
USGS gauging locations on the Blackfoot River at Bonner (#12340000), Clark Fork River at Turah
(#12334550) and Clark Fork River above Missoula (#12340500). Monthly river stage data were
supplemented with project installed staff gauges and a continuous water level recorder operated at the
Milltown Dam (prior to stage 2 drawdown). Initially, the staff gauge spacing was set up at approximately
equal distant intervals (no greater than every 800 meters) along the river channels depending on river
accessibility (Figure 1). Both river stage gauges and wells without established measuring point elevations
were surveyed using a real-time kinematic survey-grade Trimble 5800 GPS surveyor using standard
techniques (Trimble, 2008).
4.1.3
Water Level Data Analyses and Establishment of Baseline Conditions
Water level data were analyzed by constructing and evaluating well hydrographs, flow nets, and regional
water table maps, and reviewing river discharge and climatic data. Spatial and temporal water level data
were used to calibrate numerical models and assess the net change in groundwater levels from simulated
and observed changes during dam removal.
A second phase of the analyses identified sets of water level data that could be used to compare
groundwater conditions prior to dam removal with forecast and observed water levels following dam
removal. A database was compiled with groundwater data from 1982 to 2006. It was strategized that an
effort should be made to forecast the maximum likely reduction in the water table position so remediation
decisions would minimize the need for additional future work. As no pre-reservoir construction
groundwater data were available, this involved establishing a pre-dam baseline data set to be combined
with model forecasts of changes. Baseline development required a sufficient spatial and temporal
groundwater level data set that represents the area likely to be impacted, and information of the natural
variation of seasonal water levels. Previous groundwater studies found the lowest groundwater levels
occurred in late winter, a period when river flows and river stages (Gestring 1994). Also, previous work
had shown annual groundwater levels were lower than annual monthly averages during periods of less than
normal stream flow (Berthelote 2013). The most spatially complete data set was derived as part of this
study (2006-2007). The lowest groundwater level conditions were associated with March 2006. In
addition, low stream flows conditions had occurred in the previous year (64% probability of exceedence).
Based on data availability, March 2006 water table elevation data were used to represent the pre dam
removal base case from which modeling results would be combined to provide forecasts of post dam
removal groundwater conditions
4.2
Data Sets Needed to Formulate a Predictive Groundwater Model
Interpretation of the groundwater level data and stream stage data sets as well as review of previous
investigations related to the Milltown CERCLA site supported the reported aquifer conditions that allowed
for the transmission of large volumes of groundwater. A water budget was prepared for the study site.
4.2.1
Water Budget
An annual groundwater balance for the study area was formulated as follows:
In = Out +/- Change in Storage
GWinCFR + GWinBFR + GWinDC + GWinMC +BFRleak + CFRleak + Resleak + GWinBR =
GWoutCFR + GWConsP + GSWout+/- GWS
27
where:
GWinCFR is lateral groundwater underflow into the model area from the Clark Fork River Valley at Turah
Bridge; GWinBFR is lateral groundwater underflow into the model area from the Blackfoot River valley;
GWinDC is lateral groundwater underflow into the model area from Deer Creek; GWinMC is lateral
groundwater underflow into the model area from Marshall Creek; BFRleak is seepage (recharge) from the
Blackfoot River channel into the valley aqufier,; CFRleak is seepage (recharge) from the Clark Fork River
channel into the valley aquifer; Resleak is seepage (recharge) from Milltown Reservoir into the underlying
aquifer; GWinBR is the seepage into the model domain from a bedrock groundwater system; GWoutCFR is
lateral groundwater underflow from the Clark Fork River Valley at Hellgate Canyon out of the model
domain; GWConsP is consumed groundwater pumped from wells; GSWout is groundwater seepage into
the Clark Fork River within the model area; +/- GWS is the net change in groundwater storage (volume of
water annually removed or added to the aquifer) (Figure 4).
Figure 4 Generalized conceptual model illustrating the components of the water balance.
The groundwater flow system was strongly influenced by Clark Fork River and Blackfoot River leakage at
the reservoir site and downstream. A number of the water budget components were estimated using Darcy
Law calculations based on aquifer thickness estimates, local gradients, and estimated hydraulic
conductivities. Initial aquifer property information was compiled from existing studies ((ARCO) Atlantic
Richfield Company, 1995; Newman, 1996; Woessner et al., 1984), study collected data (Table 1)
(Berthelote et al., 2007), and previous modeling studies (Brick, 2003; Gestring, 1994). Hydraulic
conductivity data sets were extrapolated by analyzing well hydrograph data using stage peak lag time
methods (Pinder et al., 1969), flow net analyses (Fetter, 2001), and historical pumping tests (Harding
Lawson Associates (HLA), 1987; Walton, 1987; Woessner and Popoff, 1982).
28
Table 1 Table of techniques, methods, and equipment used to generate data sets describing the hydrological system in
the study area.
Though many wells derive water from the aquifer, most wells do not extend through the full saturated
thickness of the aquifer and into the underlying bedrock. The geometry of the lateral and bottom boundaries
of the unconsolidated valley aquifer were revised by compiling and analyzing historical borehole data,
construction site borings, well logs, topographic projections of mountain slopes into the subsurface as well
as some previous surface geophysical estimates of bedrock depths (Figure 5) (Evans, 1998; Gradient
Geophysics, 1991; Nyquest, 2001; Sheriff and others, 2007; Woessner et al., 1984). To further refine
aquifer saturated thickness estimates, additional gravity data were collected and interpreted using a Scintrex
CG3 Microgal Gravity Meter.
Figure 5 Spatial distribution of geophysical data, and site boring and well log data used to estimate bedrock depths.
In order to establish spatial and temporal surface water and groundwater exchange rates and locations
before and after dam removal, several direct in-channel measurement techniques were applied (Figure 6).
This work established the locations, directions, and rates of river leakage into the aquifer and groundwater
discharge to the river channel. The assessment also compared river stage elevations to nearby shallow well
water levels and interpreted water table maps to evaluate if a section of river channel could be classified as
gaining, losing, flow-through, or parallel flow (Woessner, 2000).
29
Figure 6 Locations of direct channel measurements used to evaluate surface water groundwater exchange rates and
locations. Using river stage, VHG (vertical hydraulic gradient) and flux calculations, and analyses of water table maps,
channel reaches were assigned as either losing or gaining. Point data were extrapolated by assuming they were
representative of conditions one half the distances between adjacent data collection sites. Quantitative assessment of the
exchange rates required installation of instruments to characterize vertical hydraulic gradients, river bed hydraulic
conductivities, and river bed flux rates using falling head tests with single or clusters of steel piezometers (Baxter,
1977; Bouwer, 1989; Farinacci, 2009; Scalon et al., 2002). Vertical temperature arrays were also installed in the river
bed to estimate flow directions and fluxes (Constantz et al., 2003; Farinacci, 2009; Hsieh et al., 2000; Johnson et al.,
2005).
4.3
Formulation of the Numerical Groundwater Modeling
More than two decades of hydrologic investigation of the Milltown Reservoir Superfund site resulted in the
construction of two earlier two dimensional numerical groundwater models of portions of the study area
(Figure 1) (Brick, 2003; Gestring, 1994). For this research, a third set of numerical models (Y1, Y2 and Y3
Figure 3) were developed using Ground Water Vistas graphical user interface to the USGS MODFLOW
code (ESI, 2004; Harbaugh, 2005; Harbaugh et al., 2000). These three dimensional groundwater models
contained up to 7 layers and were discretized into 44,837 active 46 by 46 m cells. Initial forecasts of water
level responses were based on original staged drawdowns and dam removal engineering plans. The
response of the groundwater system to stage 1 drawdown was included in the year one (Y1) model
forecasts and each of the subsequent drawdowns were incorporated into revised models (Y2 and Y3) which
produced new sets of forecasts (Berthelote et al., 2007; Berthelote and Woessner, 2008, 2009; Berthelote et
al., 2010).
Boundary condition, hydraulic conductivity and storage coefficient distributions, river reservoir stages, and
river bed hydraulic conductivities were principal model inputs. Hydraulic conductivity data were assigned
to zones based on previous modeling efforts and field data, knowledge of the sediment distribution,
professional judgment, and borehole stratigraphy-hydraulic conductivity relationships (Berthelote et al.,
2007; Berthelote and Woessner, 2008, 2009; Berthelote et al., 2010). Over 780,000 input parameters were
required to populate the model. The groundwater flow system was simulated in both steady state and
transient conditions. Model calibration was achieved by comparing head responses, river-groundwater
spatial and temporal exchange rates, and computed lateral down-valley groundwater flows into and out of
the study area. Both trial and error and parameter optimization methods (Pest pilot points and
regularization) were applied (Anderson and Woessner, 1992a; Berthelote et al., 2007; Berthelote and
Woessner, 2008, 2009; Berthelote et al., 2010; Doherty, 2000). During the three years of analyses, the base
model (Y1) was first calibrated to a base case (minimum annual groundwater level) steady state conditions
(March 31, 2006), transient conditions (March 31, 2006 to April 21, 2007), and history matched with
October 8, 1992 steady state data and a transient data set, 10/8/92 to 7/7/93 (e.g. Anderson and Woessner).
30
The final Y3 model set utilized MODFLOW 2000 (Berthelote et al., 2010; Harbaugh et al., 2000) to
forecast and test calibrations using both steady state and transient conditions (1992 to 1993 and 2006 to
2010). Appendix 2C contains the final set of input files. Complete documentation of the model
development, calibration, forecasts, and uncertainty analyses are presented in a number of project reports
(Berthelote et al., 2007; Berthelote and Woessner, 2008, 2009; Berthelote et al., 2010). A summary table of
groundwater and surface water model inputs is found in Berthelote, 2010, Appendix B. The forecast March
water levels for individual stages of reservoir drawdown were compared to the March 2006 water level
reference in order to compute water level change impacts from planned reservoir drawdowns.
It is clearly recognized that predicting future groundwater responses from planned reservoir stage changes
have many challenges. In addition to having data on the planned reservoir stage declines, timing, and
levels, assumptions were needed to represent the stream stage and flows, changes in aquifer recharge, and
river leakage. These conditions were generally estimated from construction designs, by assuming the river
bed properties did not change and the stream stages mirrored the 2006-2007 hydrograph for steady state
March water level forecasts. Stream stage data within and immediately proximal to the reservoir were
modified during the modeling process to reflect the changes in reservoir stage, physical reservoir channel
changes, and stream bed elevations.
4.4
Development of Alternative Conceptual Models /Uncertainty Analyses
Parameters assigned to active model nodes, and hydrological changes in stream stages and lateral valley
inflows and outflows are required to calibrate transient and steady state models. However, these models do
not produce unique solutions. Though results from a single calibrated model may become the principle tool
used to assess impacts from dam removals, a methodology to assess uncertainty in model forecasts is
needed. Two approaches to defining prediction uncertainty are often exercised: 1) development of
alternative conceptual models (variations in parameters or physical changes in boundaries and/or source of
sink terms that bracket likely ranges of predicted impacts); or 2) geostatistical model averaging or analyses
(random variation of key parameters constrained by assigned probability density functions) (Singh et al.,
2010; Ye et al., 2010). Our approach was to conduct a standard sensitivity analyses on the extensively
calibrated model used to make post dam removal groundwater level forecasts (Anderson and Woessner,
1992a; Hill, 1998; Hill et al., 1998; Hill and Tiedeman, 2007), and then develop alternative
parameterization that would likely bound (create a range of key sensitive values) predictions.
The sensitivity analyses uniformly varied each zoned group of parameters (formational hydraulic
conductivity, river bed hydraulic conductivity, specific yield, and storativity,) by +/- 10 and 20% while
holding all other parameters at the calibrated model value (Anderson and Woessner 1992). Changes in the
Root Mean Squared Error (RMSE) of the difference between observed and simulated heads were used to
assess the sensitivity of each parameter to the assigned changes. Parameters were considered sensitive if
any zone variation of 10 or 20% resulted in a RMSE change greater than or equal to 0.1m. Once the
sensitive parameters were identified a combination of the most sensitive parameters was used to create a set
of alternative models to bracket the calibrated model forecasts. The two most sensitive parameters
(horizontal hydraulic conductivity and river bed conductance) were chosen to develop the first two
alternative conceptual models. One alternative model used a combination of uniformly raising the hydraulic
conductivity values by 20% and uniformly lowering the river bed conductance by 20% to produce the
likely lowest forecast water table. The second alternative model reversed the magnitude change of these
parameters to produce the likely highest forecast water table. These models were used to evaluate how
reasonable combinations of sensitive parameters changed model forecasts of reservoir drawdown impacts.
A second set of alternative conceptual models tested the impact of uniformly changing the elevation of the
bottom boundary condition +/- 5 m (an uncertainty value identified in the geophysical data analyses).
Decision tree analyses using the original forecast model results and the alternative model results provided
managers with a range of the number of wells likely to require some form of remediation.
4.5
Mitigation Process
4.5.1
Model Forecasting and Mitigation Requirements
Forecast post drawdown groundwater level minimums (March) were used to develop a decision tree/risk
analyses methodology (Figure 7). Threats of reducing or losing productivity from individual wells were
31
ranked. The wells closest to the reservoir or having a shallow depth were investigated first. The data base
we developed using standard well drillers logs (online Montana Bureau of Mines and Geology GWIC)
contained well depths but no information on the depth of the pump intake. As not all wells had well logs
and in most cases pump setting depths were absent, a large number of wells were field visited and
additional data collected. These data were then assessed using a decision tree develop in consultaton with
the regulators, projecct managers and stakeholders (Table 2).
Figure 7 A) Outline of research objectives for each year; B) work flow required to identify the level of risk of
impacting groundwater supplies and the steps used to select wells for mitigation.
Decisions fell into five categories (Table 2). No action (OK) was recommended when a forecast water level
was at least one meter above the elevation of the top of the pump, and the bottom of the well (or well
intake) was greater than three meters below the predicted water level. Second, the well was replaced
(NEW) because the forecast water level was less than three metes above the well botom. Third, lower the
pump (LOWER) because more than three meters of water were in the well bore but the pump set results in
less than one meter of water over the pump. Fourth, Check the pump set (CHECK) as the forecast water
level was greater than three meters but the pump set elvelation is unknown. Fifth, Pull the pump (PULL) to
determine pump set or well depth because they were unknown and can’t not be determined by sounding the
well. Table 3 outlines the decision paths that were used to make specific recommendations.
Table 2 General mitigation decision path for seven hypothetical well construction scenarios. Well scenario numbers
and elevations (in meters) are shown strictly as examples and do not relate to well numbers at the study site. The
predicted water elevation (water table) (B) is representative of the model forecast results used for each analyses
(calibrated or alternative conceptual model prediction).
A
Well
Scenario
1
2
3
4
5
6
7
Static Water
Elevation
920
920
920
920
920
920
920
B
Predicted
Water
Elevation
914
914
914
914
914
914
914
C
D
Pump
Elevation
Bottom of
Well Elevation
Decision
Action
917
912
917
?
?
912
917
915
908
908
908
913
?
?
If B < D + 2
If B > D + 3 & C < B - 1
If B > D + 3 & C > B - 1
If B > D + 3 & C is unknown
If B < D + 3 & C is unknown
If B > C + 1 & D is unknown
If B < C + 1 & D is unknown
New
OK
Lower
OK
Check
OK
Pull
5.0
RESULTS
The dam removal process at Milltown was not a single operation taking only a few days or weeks to
complete. Because the reservoir was filled with contaminated sediments, the dam and reservoir removal
plan was staged. The process included removal of 1.9 of the 5 mcm (million cubic meters) of sediment
which took three years. The sediments removal action was designed to lower the reservoir stage to allow
32
dewatering of portions of the reservoir sediments so they could be “dry” excavated. In addition, annual
drawdowns were planned to minimize river metal, arsenic and turbidity conditions. Initially the stage 1
drawdown was initiated on June 1st, 2006 after the spring river discharge peak. Due to surface water quality
concerns, it was suspended from July 7th, 2006 to September 18th, 2006 and then resumed. Final Stage 1
drawdown was completed on November 12th 2006. The maximum stage 1 drawdown from full pool was
~3.6 m (Figure 3). To avoid possible future summer water quality issues Stage 2 drawdown (additional 3.6
m) began on March 28, 2008 and the final Stage 3 (additional 1.3 or 8.5 m total) drawdown was completed
on March 27, 2009 with the removal of the spillway coffer dam.
5.1
Groundwater Base Case
Establishment of a base case from which to reference observed and forecast impacts was complicated at
this site by the availability of only short, intermittent, and spatially discontinuous historical groundwater
level data. To overcome this limitation, all water level data at each historical monitoring well were plotted
by month of collection on a 1 year time scale. (1982 to 2006) (Figure 8). An envelope surrounding the
points was used to indicate temporal variability (gray shading). It was decided that initial data collected as
part of this work, water levels for March 2006, would be used to generally represent the pre dam removal
groundwater conditions from which the groundwater impacts would be measured. The March 2006 data
were consistently near the historical lower boundary of the groundwater level position shown in the
historical measurements. A plot of project collected 2006 to 2007 water table data shows a response to the
June through September Stage 1 reservoir decline.. Groundwater level impacts resulting from reservoir
drawdowns and removal were generated by using numerical models each of the first three project years
(Y1, Y2 and Y3) to forecast future March groundwater conditions and then subtracting these elevations
from the March 2006 base case values
Figure 8 Hydrograph of Well 01 (closest well to the dam) showing the relative variation of 1982 to 2006 historical
groundwater level elevations (squares and gray shaded area) and 2006 to 2007 groundwater levels with respect to a
March 2006 groundwater level ( used as the reference point from which to measure change) (see Appendix 2B for
complete data sets).
5.2
Observed Groundwater Changes during Stage Drawdowns and Complete Dam removal
Water level data collected between 2/24/1982 to 5/9/2010 included 1,805 groundwater monitoring days
with over 87,245 measurements at 229 locations and 1,276 surface water monitoring days with over 5,028
measurements at 25 locations (Appendix 2B). March water table maps for each year (1993, 2006, 2007,
2008, 2009, & 2010), and a representative hydrograph from a well near the dam (well 05) show that
following the dam and reservoir removal the general flow directions and seasonal water level responses
(2010) remained similar to pre drawdown conditions (1993 and 2006) (Figure 9). The difference between
the base case observed water levels and the observed 2010 water levels are illustrated in Figure 10A. In
addition, observed March groundwater levels for each subsequent year were differenced from the base case
March 31 2006 groundwater conditions to quantify observed changes. Though these data show changes in
groundwater levels, deciphering the influence of the magnitude of changes resulting solely from the
drawdowns and dam removal is complicated and can be highly dependent on previous streamflow
conditions and recharge regimes (Berthelote, 2013 Chapter 3). During the dam removal period both above
and below average stream flow conditions occurred (Figure 9). These conditions likely resulted in variable
33
rates and durations of surface water groundwater exchanges that directly influenced overall observed
groundwater level elevations. When dam removal groundwater impacts are based on base case and
observed post-dam removal measurements, portions of the groundwater system declined 2 to 3 m as a
consequence of the reservoir and dam remediation activities. Mapped water level declines were not limited
to the immediate reservoir area but rather extended at least 6 km downstream and 2 to 3 km upstream
(Figure 9) .
Figure 9 Observed groundwater levels. Valley water table maps for March low water levels from 1993, and 2006 to
2010. Water level contours indicate that groundwater flow directions remained uniform in areas above and below the
dam site, however some variaitons were noted proximal to and north of the reservoir area (central portion of the
map).The bottom graph illustrates the transient groundwater hydrograph for a well immediately below the dam site
(well 05). River discharge is presented for the above Missoula USGS guage for comparison. The red dots highlight the
March 31 water level for each of the project years.
34
5.3
Water Budget and Surface Water Groundwater Exchange
Initial pre-dam removal field data revealed the rivers and groundwater systems formed a complex
hydrogeological system. Groundwater recharge is primarily from four sources: 1) a perched and leaking
(losing) Blackfoot River and Clark Fork River arms of the reservoir and most of the Clark Fork River from
just below the dam to Hellgate Canyon (vertical riverbed hydraulic conductivities range from 0.4 to 12.8
m/d and leakage rates from 68,000 to 284,000 m3/d); 2) lateral down-floodplain underflow at the Turah
Bridge Boundary in (11,900 m3/d to 45,300 m3/d), 3) underflow from the Blackfoot River floodplain into
the Clark Fork valley (1,600 m3/d to 19,500 m3/d) and, 4) limited recharge from the mountain boundaries (
less 4400 m3/d) (Berthelote et al., 2007; Farinacci, 2009; Tallman, 2005; Woessner and Popoff, 1982).
Aquifer discharge occurred in a short gaining reach directly below the dam (300 m), as underflow through
the valley aquifer located in Hellgate Canyon to the west (59,400 m3/day to 566,300 m3/d), and locally in
gaining stream sections of the Clark Fork River above the reservoir. Recharge from precipitation and
discharge from evapotranspiration were considered to be negligible as they were estimated to be a small
percentage of the total water balance (Woessner et al., 1984). Groundwater inflow from the bedrock
boundary was also estimated to be small and not a significant component of the water balance(Woessner et
al., 1984). The groundwater was assumed to be at steady state for the year period.
Generally, groundwater flowed towards the reservoir area from the upper Clark Fork River valley. This
component of the flow system converged just north of the reservoir with groundwater moving from the
Blackfoot River Canyon south towards the reservoir area (Figure 9). These systems combined with the
northward moving groundwater from the reservoir and the northwesterly groundwater flow continued west
eventually discharging as underflow through Hellgate Canyon
The aquifer is highly conductive (range of hydraulic conductivity from 90 to >27,000 m/d with a mean of
3600 m/d) with groundwater velocities calculated in tens to hundreds of meters per day. Hydraulic
gradients range from 0.0013 (upper Blackfoot River arm) to 0.066 (near the dam), and estimated porosities
are 0.20 (Berthelote et al., 2007; Berthelote et al., 2010; Gestring, 1994; Moore and Woessner, 2002;
Tallman, 2005; Woessner et al., 1984). Surface water and groundwater exchange locations and volumes
ranged from 2.3 to 43 m3/(day m2) as computed using hydraulic properties, gradients, and temperature
modeling (Table 3).
Table 3 Water Balance Summary: Groundwater inflow outflow spatial descriptions, magnitudes, ranges, potential
errors, and sources
Water Balance
Parameter
Inflow source
Minimum m3/day
Maximum m3/day
Possible
Error
Range of
Values
GWinCFR
GW Underflow Clark Fork
River
1.8E+04
3.4E+04
33%
1.2e4 - 4.5e4
GWinBFR
GW Underflow Blackfoot
River
2.4E+03
1.5E+04
33%
1.6e3 - 2.0e4
Marshall Creek underflow
4.0E+00
GWinMC
GWinDC
Previously Determined Values m3/day
Water Balance
Parameter
3.1E+03
60%
1.6e0 - 5.1e3
3.1e3(Gestring, 1994)
8.8e3 (Brick, 2003)
Deer Creek Underflow
7.9E+02
8.8E+03
60%
3.1e2 – 1.4e4
3.96E+00
3.96E+00
50%
2.8e4 - 9.9e4
Leakage Blackfoot River (I90 Bridge to BFR4)
3.96E+00
3.96E+00
50%
1.4e4 - 4.2e4
1.9e4-5.1e4 (Gestring, 1994)
Leakage Clark Fork river
Below dam to well HGD
3.96E+00
3.96E+00
54%
3.1e4 - 5.1e5
6.8e4-2.0e5 (Gestring, 1994)
GWoutCFR
Possible
Error
Range of
Values
3.96E+00
3.96E+00
Leakage Clark Fork river
Above dam CFRA8
3.96E+00
Leakage Reservoir
3.96E+00
3.96E+00
Total Inflow
1.6E+05
1.1E+06
3.96E+00
60%
60%
4
6.8e - 2.7e
GW under Hellgate Canyon
9.3E+04
4.2E+05
33%
6.0e4 - 5.7e5
1.2e5 (Brick, 2003)
1.2e5 (Brick, 2003)
GSWout
GW discharge:Clark Fork
River CFRA2
5.4E+04
5.4E+04
60%
2.1e4 - 8.5e4
GW discharge: Clark Fork
River CFRA5
5.1E+04
5.1E+04
60%
2.0e4 - 8.2e4
GW discharge: Clark Fork
River (CFRB1 to CFRB4)
5.7E+04
5.7E+04
60%
2.3e4-9.1e4
Pumping Wells
1.4E+03
2.2E+03
5%
1.4e3 – 2.3e3
Total Outflow
2.5E+05
5.7E+05
Difference
-9.3E+04
5.7E+05
1.1e1-2.2e5 (Gestring, 1994)
5
6.5e3 - 2.6e4
GWConsP
50%
Previously Determined Values m3/day
9.3e4-1.9e5 (Tallman A.A., 2005)
5.7e4 (Gestring, 1994)
Resleak
Maximum m3/day
1.1e5-2.2e5 (Gestring, 1994)
BFRleak
CFRleak
Minimum m3/day
7.6e3 (Brick, 2003)
6.8e2 (Popoff M.A., 1985)
Leakage Blackfoot River
(BFR4 to BFR6)
Leakage Clark Fork river
CFRA3 to CFRA4
Outflow Source
2.4e2 - 1.4e5
2.6e5 - 6.0e5
5.4e4-7.4e5 (Popoff M.A., 1985)
9.1 e4 (Moore and Woessner, 2002)
5.4
1.9e5 - 6.2e5
Numerical Modeling Results
5.4.1
The forecasting of groundwater
Forecasts of future groundwater levels were generated with extensively calibrated models. Simulated
groundwater levels were calibrated so they were consistently less than 2 m different from observed values
(up to 77 network wells depending on the model run). The resulting model water budgets were within the
targeted baseline pre drawdown ranges of uncertainties presented in Table 3 (Table 4). Fitted hydraulic
conductivity and river leakage parameters were within measured ranges. The simulated water balance
35
results were consistently stable and differences between inflow and outflow components were less than
0.02 percent.
The simulated Y1 base case steady state water budget compared favorably with the pre-model estimated
steady state budget (Table 4). Seepage from the Clark Fork River into the underlying groundwater and the
flow of valley groundwater into gaining portions of the river were slightly less than original budget
estimates. The calibrated steady state heads, boundaries, and fluxes were used as initial conditions during
transient model calibration. The comparison of simulated heads to observed heads (March 31, 2006 to April
21, 2007), and the pre-modeling estimated water balance revealed the model reasonably produced observed
conditions under this more demanding evaluation. For the transient simulation 60 observed heads were
used as calibration targets. They were distributed as follows: 23 in layer 1; 7 in layer 2; 8 in Layer 3; 7 in
layer 4; 8 in layer 5; 3 in layer 6, and 4 in layer 7. Results comparing observed and simulated heads over
time show relatively good fits of simulated water level positions with observed levels at most sites.
Table 4 Comparison of the pre-model estimated (includes error estimate) and simulated steady and transient state
water balances. Transient water balance was presented for the last stress period of April 21, 2007.
GWinCFR
Estimated range (m3/day)
Y1 Modeled value
(SS 3/31/06)
(m3/day)
Y1 Modeled value
(T 4/21/07)
(m3/day)
1.2e4 - 4.5e4
3.7E+04
3.4E+04
GWinBFR
1.6e - 2.0e
4
1.6E+03
5.1E+03
GWinMC
1.6e0 - 5.1e3
2.5E+03
2.5E+03
GWinDC
3.1e2 – 1.4e4
8.8E+03
8.8E+03
BFRleak
1.4e4 - 9.9e4
5.7E+04
4.2E+04
CFRleak
3.1e4 - 5.1e5
9.1E+04
1.2E+05
5.4E+04
2.8E+04
Resleak
Total Inflow
3
2
5
5
6
2.4e - 1.4e
1.6e - 1.2e
2.5E+05
2.4E+05
Y1 Modeled value
(T 4/21/07)
(m3/day)
Outflow
Estimated range (m3/day)
Y1 Modeled value
(SS 3/31/06)
(m3/day)
GWoutCFR
6.0e4-5.7e5
2.3E+05
2.0E+05
GSWout
4
2.0e -9.1e
5
2.0E+04
2.4E+04
Total Outflow
5
5
2.5E+05
4.8E+03
2.6e -6.0e
These calibrated models (Y1, Y2, and Y3) were used principally to forecast the March groundwater levels
after each stage of reservoir drawdown was completed. Model Y1 and the two hydraulic conductivity and
river leakage alternative model forecasts (Y1 Hi and Y1 Low) were used to forecast the response of the
system to the stage 2 drawdown. The following year Model Y2 was recalibrated to the initial stage 1 and
stage 2 groundwater level response and was used to forecast impacts from stage 3 drawdowns. Finally,
model Y3 was revised using all the observed stage-drawdown responses and to forecast the final dam out
impacts (Figure 10). To illustrate the type and analyses of model results, water level data for well 01 is
presented (Figure 10) .The forecast responses at well 01 are indicative of the simulation results at the other
77 wells used for calibration. Transient calibrations of each model had difficulty closely matching
groundwater levels at the peak hydrograph periods (Figure 10 C). However, the models (Y1, Y2, and Y3)
were well calibrated with all root mean square errors less than 0.51. The alternative conceptual models that
employed varying bedrock elevations (Y2 B+15 (ft) and Y2 B-15 (ft)) revealed that changes in this
boundary did not extend the range of impacts forecast by the other conceptual models during Y2 modeling
(Figure 10B). As a result they were not used as viable alternative conceptual models in Y3 modeling.
Examining the results of the calibrated models and alternative models at well 01 suggest that the initial Y1
modeling adequately represented likely groundwater level changes resulting from stage 2 drawdowns. Y2
and Y3 modeling were also reasonable representations of hydrogeologic conditions and useful in
forecasting the groundwater response to stage 3 and final dam out actions.
36
Figure 10. A) A spatial representation of the difference between the observed 3/31/06 base case and the observed 2010
March groundwater levels. B) Composite hydrograph of well 01 showing a graphical representation of the range of
modeling results for the calibrated (green data) and four alternative conceptual models (red, blue, purple, & orange) for
each year. The simulation in which the hydraulic conductivity was uniformly lowered by 20% and the river leakage
uniformly raised by 20% always resulted in the modeled water table being at a higher elevation (blue points). The
simulation in which the hydraulic conductivity was uniformly raised by 20% and the river leakage uniformly lowered
by 20% always resulted in the modeled water table being at a lower elevation (red points). Observed groundwater level
data are represented by the brown line. The brown horizontal line represents the March 31st 2006 base case preremoval low water table position. C) A transient illustration of the Y3 modeling results bounded by the alternative
conceptual modeling results compared to the observed groundwater levels for well 01.
5.5
Mitigation Process
The goal of the agencies responsible for groundwater impact mitigation was to provide dependable supplies
of water to every resident without significant service gaps. Year 1 groundwater modeling results were used
by project managers prior to the winter of 2007 to identify groundwater supply wells that were likely to
require mitigation (Figure 11A). By the completion of our evaluation (Spring 2008), all 515 surveyed wells
were evaluated for possible impacts with 286+ well sites physically visited. Sixty wells were identified and
remained in the “check” category as needing additional information. Wells identified as likely to be
impacted were mitigated based on the Table 3 decision tree.
37
Figure 11 Spatial comparisons of mitigation recommendations and actions at the end of the project in 2010 (Y3). A)
Distribution of well mitigations for the 515 wells separated into action recommended (red dots), wells predicted to be
unimpacted (OK) (green dots), and the wells needing to be checked (black dots), B) Illustration of the wells that were
either mitigated or recommended for mitigation. Individual wells were either mitigated based on the recommendations
calculated by the decision tree (Table 3) (green dots), not mitigated based on the recommendations (red dots), or
mitigated when no model based action was recommended (black dots). It was observed that many mitigation well
locations were coincident with regions where large water level impacts were predicted. However, shallow wells
located farther away from the reservoir area were also affected. The fact that area wells were developed over a century
with varying depths and construction philosophies resulted in a non uniform distribution of mitigation actions.
Y1 forecasts resulted in identifying between 62 and 137 wells needing some sort of mitigation which
included between 42 and 84 wells needing replacement (New), 13 to 36 pumps that would require lowering
(Lower) and up to 17 pumps that would need to be pulled (Pull) to determine total well depth. Hundreds of
wells were initially identified as needing more information (Check). Based on Y1 results, the field
campaign data were used to update the mitigation recommendations. The final results illustrated in Figures
11 and 12 indicate the final numbers generated by the decision tree at the end of the project. As the
mitigation analyses proceeded we performed additional modeling refinements; however, project managers
continued to base all mitigation decisions on the Y1 modeling scenarios primarily due to the fact that
subsequent models were produced after the majority of the wells had been mitigated (Figure 3). Figure
11B illustrates the variance between the actual mitigation actions completed and the recommendations
computed by the decision tree. All subsequent modeling efforts confirmed intital actions and only
minimally refined the recommended actions (Figure 12).
Mitigated Wells
Y1 (Stg 2) Calibrated Forecast
Y1 (Stg 2) Alternate Hi
Y1 (Stg 2) Alternate Low
Y2 (Stg 2) Calibrated Forecast
Y2 (Stg 2) Alternate Hi
Y2 (Stg 2) Alternate Low
Y2 (Stg 2) Alternate B+15
Y2 (Stg 2) Alternate B-15
Y2 (Dam Out) Calibrated Forecast
Y3 (Stg 2) Calibrated Forecast
Y3 (Dam Out) Calibrated Forecast
400
350
# of Wells
300
250
200
OK Action New Lower Pull Check
357
98
81
17
0
60
377
78
57
16
5
60
318 137
84
36
17
60
393
62
42
13
7
60
385
75
56
9
10
55
351 115
80
20
15
49
404
54
44
6
4
57
382
76
56
10
10
57
368
84
60
12
12
63
358
97
65
21
11
60
369
86
62
19
5
60
375
80
57
19
4
60
450
Mitigated Wells
Y1 (Stg 2) Calibrated Forecast
Y1 (Stg 2) Alternate Hi
Y1 (Stg 2) Alternate Low
Y2 (Stg 2) Calibrated Forecast
Y2 (Stg 2) Alternate Hi
Y2 (Stg 2) Alternate Low
Y2 (Stg 2) Alternate B+15
Y2 (Stg 2) Alternate B-15
Y2 (Dam Out) Calibrated Forecast
Y3 (Stg 2) Calibrated Forecast
Y3 (Dam Out) Calibrated Forecast
400
350
300
# of Wells
450
250
200
150
100
150
50
100
0
50
OK
Action
Mitigated Wells
Y1 (Stg 2) Alternate Hi
New
Lower
Pull
Check
0
Y1 (Stg 2) Calibrated Model
Y1 (Stg 2) Alternate Low
OK
Action
New
Lower
Pull
Check
Figure 12 Comparison of the predicted mitigation for 515 surveyed wells. There is only a small percent variation in
predicted mitigation from the use of multiple conceptual models. All but two of the models represented Stage 2
conditions. The final stage 3, 1.3 meter drawdown was forecast to have minimal impacts on planned and executed
mitigation efforts.
Using the results of this work, the project manager would inform the well owner of the perceived risk and
offer to mitigate the well. Many well owners with well depths recommended for mitigation that were close
to the decision tree threshold decided to opt out of immediate mitigation to wait and see what would
happen. This was presumably due to the decision makers’ public statement that they would mitigate any
38
wells affected by dam removal for up to one year following the final stage 3 removal. Many homeowners
whose wells were close to the decision tree threshold, but not recommended for remediation, insisted on a
replacement. Often these well owners received new wells. Appendix 2D contains a compilation of the well
survey data by location.
6.0
DISCUSSION AND CONCLUSIONS
Groundwater supply mitigation implemented in response to the dam and reservoir removal cost close to
$1,000,000. This work showed a well calibrated numerical groundwater model as an appropriate tool with
which to forecast groundwater response to the removal action. However, though challenges remain, the
applied methodology appears to present a reasonable approach when attempting to identify groundwater
impacts..
6.1
Challenges of Predicting Groundwater Impacts
Throughout the mitigation process some wells became inoperable and were mitigated before
recommendations were completed. All of these wells were initially identified by the decision tree as
needing more information (Check or Pull), and at the time of production loss had not been visited. Several
wells were mitigated without the direct need (identified as OK) in an attempt to appease social concerns
where small (<1m) differences in the initial forecast would have altered the recommended action from OK
to Action. A few of these individual well owners found that the deeper waters had a different chemistry that
required the installation and maintenance of treatment systems to remove increased concentrations of
manganese, iron, and other minerals.
When the year 1 and 2 observational data was included in the Y3 modeling efforts, it suggested that the
earlier modeling had over estimated water table changes . This is attributed to an underestimation of the
hydrogeologic properties assigned to represent the bed of a temporary bypass channel constructed to assist
in removal of some portions of the reservoir sediments. Further model calibration revealed channel leakage
into the groundwater from the by pass channel needed to be doubled to meet calibration targets. The
increased leakage acted to reduce forecast groundwater level declines and bring them more inline with
observations. Attempts were made to verify these calibrated rates by instrumenting the new channel,
however high river flows resulted in the loss of field instrumentation, and access was limited by
construction activities.
Though the selection of the base case data set (March 2006) was dictated by available historical data and
was within a low stream discharge year (64% probability of exceedence), it is not unlikely that future
groundwater levels will be less than those predicted by the model (36% probability of lower river discharge
years or drought). Efforts should be made to forecast future minimal flow groundwater levels to make this
impact analyses more complete. This raises the question of what happens in the future if water level
decline in wells so that water supplies are interrupted? Certainly mitigation decisions and actions that
formed this work should be well documented in anticipation of further questions that may arise regarding
quantifying impacts from removal actions. In this setting, the river exchange with the groundwater controls
the seasonal variations and overall position of the water table, thus the maximum and minimum annual
water table elevations. During low flow drought periods water table positions are lower than average and
wetter than normal periods have the opposite affect. It is recognized that this 2006 to 2010 climate signal is
over printed on the observed water table response to dam removal. How the river stage and flows will
behave in the forecast period is of course unknown, thus the predicted response of the groundwater system
does not mirror 2010 observations. This makes testing the model forecast difficult as pre established
variations in groundwater conditions were assumed to produce the forecast. In the Milltown case, only if
the antecedent flow and groundwater conditions matched the March 2006 water table elevations could the
accuracy of the forecast be evaluated.
6.2
Evaluation of Remediation Actions
In most cases, the proactive mitigation activities were completed prior to a well becoming inoperable, so
any direct measure between the recommended number of wells needing action and the number actually
needing mitigation is difficult. This is because old wells were immediately sealed when the new ones
became operational. Thus, a direct comparison of the mitigation vs. recommendation numbers (as presented
in Figures 11 and 12) does not provide an evaluation of the effectiveness of the mitigation process. The
39
new well can, however, provide a proxy for the resultant groundwater elevation due to the close proximity
of the two wells. If, however, actual observational data (reactive) rather than forecast water table data
(proactive) were used to determine wells requiring remediation, fewer wells would have been mitigated
because 2008 to 2010 March water levels were higher than the base case levels.
The decision tree developed for this project seemed to successfully identify 100% of the wells requiring
mitigation action when appropriate input data was available (pump set depth, total well depth, and an
accurate forecast of groundwater levels resulting from dam removal actions) as would be expected. A
general comparison of spatial annual baseline water levels and subsequent model forecasts of post
drawdown and dam removal seasonal low water table conditions did effectively identified 80% of the water
supply wells requiring groundwater level response mitigation in the first year (Y1 stage 2 forecast).
Subsequent models all identified 77% or more of the wells originally identified as possibly being impacted
with up to 99% of the wells requiring mitigation identified by the Final Y3 modeling effort. This was not
unexpected as the Y3 model was calibrated to well responses from all of the staged drawdowns.
The goal of the project managers was to provide consistent supplies of water to every resident without
noticeable gaps. In a few circumstances, wells became inoperable after they were recommended for
mitigation due to lack of availability of well drillers (there were only 3 local drillers who were often times
all working on other properties). The goal was met with temporary mitigation measures by piping water in
from wells on adjacent properties to the households until the new well was completed. Water supply losses
to individual residences were not reported to exceed 36 hours and in general never exceeded 6 hours. This
was the true test of success for this project and was reflected in the positive public perception of the entire
mitigation process.
6.3
Recommend Process for Mitigation Planning
As a dam removal is planned, a groundwater level impact analyses, and if needed, a mitigation process
should be implemented. These actions require a number of steps:
1. Establish reference groundwater levels: A base case set of groundwater level conditions has to be
established. For the Milltown study, historical and project gathered data on regional surface water
groundwater interactions and flow systems were collected. This process required community
support, planning, manpower, time, and hydrogeologic expertise. The extensive data sets
provided an invaluable resource to use for numerical model creation and calibration. In retrospect,
transient and continuous data both prior to and following the dam removal were necessary.
Having pre dam removal groundwater level and river stage monitoring data provides a baseline to
compare with post removal data which may avert potential litigation issues.
2. Forecast post dam out groundwater levels: Water table changes during and after dam removals
need to be predicted. For the Milltown study a standard extensively calibrated numerical
groundwater model was used. Our model was calibrated with historical, steady state and transient
data sets . Models reproduced observed historical and initial monitored water levels with minimal
residuals and the pre modeling groundwater budget suggests models were representative of
hydrogeologic conditions. Forecasting future water levels is highly dependent on this calibration
process and must also include observational data throughout a drawdown period to adequately
evaluate system responses to engineering activities. Forecasts also need to be reported with some
degree of uncertainty framed by using alternative conceptual modeling (model averaging) or
geostatistical approaches. Reasonable forecasts of groundwater levels are a critical component of
the decision tree input data and must be acquired through modeling or similar techniques. It has
been proposed that simplified Q/K modeling (Berthelote, 2013Chapter 1) or Artificial Numerical
Modeling (Berthelote, 2013Chapter 3) may be useful alternative tools for forecasting future
groundwater levels, both of which require a more limited input data set.
3. Determine impacts: Simply predicting future groundwater levels resulting from a reservoir
drawdown or dam out scenario does not allow decision makers to assess the risk of the water
levels dropping below a well bottom or pump intake elevation. Well information (total depth or
well intake and pump set depth) must be obtained in a data collection phase. For the Milltown
study, well drillers logs and hundreds of well site visits were used to compile these data. Wells
with the greatest potential threat (closest to the reservoir or shallowest well depths) were
investigated first.
40
4.
Develop a Mitigation Strategy: Threats of reduction or loss of productivity from individual wells
were ranked. Once the individual well data were gathered and the predicted impact quantified
(model results), then a recommended action was derived to assist managers in proactively
mitigating wells with the greatest risk of failure from dam removal activities. A more descriptive
explanation of this strategy is presented next.
This research resulted in the development of a risk based framework for proactive well mitigation
necessitated by alterations to groundwater levels from a dam removal. We believe these approaches will be
applicable to similar dam removal projects where well production losses are possible. The specific process
used throughout this research (Figure 7B) can be re configured and modified to fit different sets of
hydrological conditions (Figure 13). It is important to remember that while no approach will guarantee 100
percent protection all of the time, effective risk management reduces risks and increases the feasibility and
effectiveness of remedial control or preventative options. Redundancies should be built into the system
wherever feasible. These actions will mitigate repercussions when, and if, failures occur in the system and
also help demonstrate that the mitigation managers have acted with due diligence.
Figure 13
Risk based decision process for groundwater mitigation resulting from dam removals.
Milltown managers used the decision tree presented in Table 2 to prioritize risk into manageable actions.
In its most basic form, a risk assessment can simply be a ranking of hazards against designated benchmarks
for the protection of consumers. Many standard risk matrices exist in different contexts. A mitigation plan
reconfirms objectives (outlined in Figure 7) that were chosen for assessment as management targets or
goals against which management actions will be evaluated. Making decisions that benefit stakeholders
while maintaining watershed objectives can be challenging. The decision-making process is
multidisciplinary in nature and must integrate variables such as scientific, socioeconomic, and political
knowledge. All mitigation actions considered the risk framework, public perception, expert opinion, and
cost benefit analyses
In the Milltown mitigation process, regular public forums were conducted where forecasted impacts to
groundwater levels and mitigation actions were presented to and discussed with the public. These public
meeting generally started with discontent, suspicion, and misunderstanding but ended with cooperation,
mutual understanding, and collaborations. Effective risk communication ensures all participants adequately
understand the risk management process and how decisions are made. Educational activities executed
during the Milltown dam removal included resource materials, seminars, workshops and public meetings.
41
Educational goals were to encourage awareness, understanding, and more informed decision-making.
Public participation is the process by which all interest groups (stakeholders and the general public) in a
community are provided the opportunity to make their views known.
7.0 SUMMARY
It is recognized that dam and reservoir removals will have some level of impact on adjacent aquifer
systems. Managing groundwater system responses to dam and reservoir removals and the resulting
economic and sociological consequences requires development of time effective mitigation strategies
informed by groundwater level forecasting. In western Montana, project managers of the recent 8.5 m high
Milltown Dam and the contaminated reservoir sediment removal implemented a well replacement
mitigation strategy that attempted to proactively mitigate groundwater supply impacts from project
activities. At the initiation of dam removal activities an extensive groundwater and river system monitoring
network was established to observe groundwater level conditions and changes. A suite of multi-layer, three
dimensional, finite difference groundwater models calibrated to both historical data and observed
groundwater responses to staged reservoir drawdowns provided groundwater level forecasts of post dam
out conditions. These data were inputs to a decision tree that identified wells needing replacement,
lowering of pumps, further data collection, or were considered not at risk. Uncertainty was evaluated using
sensitivity analyses and resulting alternative conceptual models. Observed results showed water table
changes were up to 3 m near the reservoir and impacts extended 6 km downstream and about 2 km
upstream of the reservoir. Model forecasts of groundwater level changes were greater than post dam
removal observed levels, as expected, because natural variations in annual river driven groundwater
recharge that occurred during the dam and sediment removal period were not used as model input. Risk
tree analyses showed up to 115 wells were at risk. Proactive mitigation included the construction of 80 new
wells and lowering of 20 pumps. Future dam removal projects should include groundwater impact analyses
and proactive groundwater supply mitigation as needed. This process should include both observations of
changes and pre-dam removal forecasting of groundwater responses. Forecasts should then be assessed and
actions taken based on a logical project based decision tree such as the one developed here.
(Anderson and Woessner, 1992b; Berthelote et al., 2007; Berthelote and Woessner, 2009; Brick, 2003; Caldwell and Bowers, 2003; Cordell and Henderson, 1968; Croft, 2006;
Doherty, 2000; ESI, 2004; Evans, 1998; Fetter, 2001; Gestring, 1994; Gradient Geophysics, 1991; GWIC, 2008; Harbaugh, 2005; Harbaugh et al., 2000; Hennes, 2002; Hill,
1990, 1992, 1998; Hill et al., 1998; Hill and Tiedeman, 2007; Hsieh et al., 2000; Jahns, 1966; Janiszewski, 2007; Johnson et al., 2005; Land and Water Consulting, 2004,
2005; Landon et al., 2002; Lorang et al., 2005; Newman, 1996, 2005; Nyquest, 2001; Pinder et al., 1969; Stanford et al., 2005; Tallman, 2005; Trimble, 2008; USGS, 2011;
Walton, 1987; Woessner et al., 1984; Woessner and Popoff, 1982)
42
8.0
REFERENCES
(ARCO) Atlantic Richfield Company, 1992, Summary of Four-Step Process, Addressing Wetlands Issues
in Upper Clark Fork River Superfund Sites, Letter from Ms. Sandra Stash, ARCO, Anaconda, MT
to Mr. Donald Pizzini and Mr. Robert Fox, USEPA: Helena, MT.
(ARCO) Atlantic Richfield Company, 1995, Milltown Reservoir Sediments Operable Unit, Final Draft
Remedial Investigation Report, in Corp, T.E., ed.
Anderson, M.P., and Woessner, W.W., 1992a, Applied groundwater modeling, simulation of flow and
advective transport, Academic Press, 382 p.
Anderson, M.P., and Woessner, W.W., 1992b, The role of the postaudit in model validation: Advances in
Water Resources, v. 15, p. 167-173.
Babbitt, B., 2002, What Goes up, May Come down: BioScience, v. 52, p. 656-658.
Baxter, R.M., 1977, Environmental Effects of Dams and impoundments: Annual Review of Ecology &
Systematics, v. 8, p. 255-283.
Berthelote, A.R., 2013, Forecasting Groundwater Responses to Dam Removal [Dissertation thesis]:
Missoula, University of Montana.
Berthelote, A.R., Farinacci, A., Storb, M., and Woessner, W.W., 2007, Determining the impact of Milltown
Reservoir Drawdowns on the Adjacent Unconfined Aquifer: Remediation Stage 1 Observed and
Stage 2 Forecasted, University of Montana, p. 174.
Berthelote, A.R., and Woessner, W.W., 2008, The Use of Observed and Model Forecasted Groundwater
Levels to Identify Likely Impacts to Local Water Supply Wells in the Vicinity of a Dam Removal
Project, Milltown Montana, University of Montana, p. 31.
Berthelote, A.R., and Woessner, W.W., 2009, Mitigating Groundwater Impacts from the Removal of a
Large Dam in an Unconfined Aquifer: Milltown Reservoir Remediation, University of Montana,
p. 178.
Berthelote, A.R., Woessner, W.W., and Thompson, J.H., 2010, Assessing Groundwater Level Behavior and
Numerical Model Forecasts of Water Table Changes Related to Removal of a Large Dam:
Milltown Reservoir Remediation, University of Montana, p. 113.
Bouwer, H., 1989, The Bouwer and Rice slug test- an update: Groundwater, v. 27, p. 304-309.
Bowman, M., Higgs, S., Maclin, E., McClain, S., Sicchio, M., Souers, A., Johnson, S., and Graber, B.,
2002, Exploring Dam Removal: A Decision-Making Guide, American Rivers Trout Unlimited, p.
80.
Bowman, M.B., 2002, Legal Perspectives on Dam Removal: BioScience, v. 52, p. 739-747.
Brick, C., 2003, Preliminary groundwater modeling to estimate effects of dam and sediment removal on the
alluvial aquifer in Milltown, Montana. Clark Fork Coalition. Report., p. 24.
Butler, D.R., and Malanson, G.P., 2005, The geomorphic influences of beaver dams and failures of beaver
dams: Geomorphology, v. 71, p. 48-60.
Caldwell, R.R., and Bowers, C.L., 2003, Surface-water/Ground-water interaction of the Spokane River and
the Spokane Valley/Rathdrum Prarie Aquifer, Idaho and Washington, Water-Resources
Investigations Report, p. 03-4239.
Chen, X., and Chen, X., 2003, Stream water infiltration, bank storage, and storage zone changes due to
stream-stage fluctuations: Journal of Hydrology, v. 280, p. 246-264.
Constantz, J., Cox, M.H., and Su, W.G., 2003, Comparison of heat and bromide as groundwater tracers
near streams: Ground Water, v. 41, p. 647-650.
Cordell, L., and Henderson, R.G., 1968, Iterative three-dimensional solution of gravity anomaly data using
a digital computer: Geophysics, v. 33, p. 596-601.
Croft, B., 2006, Using Gravity to Determine Depth to Bedrock in the Rattlesnake Valley, Missoula,
Montana. Unpublished Senior Thesis, University of Montana.
Doherty, J., 2000, PEST, model-independent parameter estimation: Townsville, Watermark Numerical
Computing.
Envirocon, 2006, Remedial design work plan, Final, For stages 1,2, and 3 drawdowns and related
construction activities Milltown Reservoir Sediments Site.
ESI, 2004, Guide to Using Groundwater Vistas, Version 4 Volume 2006, Environmental Simulations Inc.
Evans, C.A., 1998, A constrained gravity model of the Central Missoula Valley and shape of the Ninemile
Fault [Masters thesis]: Missoula, MT, University of Montana.
43
Farinacci, A.J., 2009, Surface water and groundwater exchanges in fine and coarse grained river bed
systems and responses to initial stages of dam removal, Milltown, Montana [Unpublished Masters
Thesis thesis]: Missoula, University of Montana.
Fetter, C.W., 2001, Applied hydrogeology, 4th ed.: New York, NY, Macmillan Publishing Company, 598
p.
Gestring, S.L., 1994, The interaction of the Clark Fork River and the Hellgate Valley Aquifer near,
Milltown, Montana. M.S. Thesis, Geology Department, University of Montana.
Gradient Geophysics, 1991, Missoula-Milltown area seismic refraction determined bedrock depth data,
processed for Land and Water Consulting, Inc., unpublished.
GWIC, 2008, Ground-Water Information Center, Montana Bureau of Mines and Geology, Montana Tech
of the University of Montana, http://mbmggwic.mtech.edu/sqlserver/v11/help/go/signin.asp.
Harbaugh, A.W., 2005, MODFLOW-2005, The U.S. Geological Survey Modular Ground-Water Model—
the Ground-Water Flow Process, 253 p.
Harbaugh, A.W., Banta, R.B., Hill, M.C., and McDonald, M.G., 2000, Modflow-2000, The U.S.
Geological Survey Modular Ground-water Model—User guide to Modularization Concepts and
the Ground-water Flow Processes. U.S. Geological Survey Open File Report 00-92. .
Harding Lawson Associates (HLA), 1987, Milltown Reservoir Data Report Supplemental Investigations
conducted under the Feasibility Study, Report prepared for Montana Department of Health and
Environmental Sciences Solid and Hazardous Waste Bureau: Helena, Montana, p. 132.
Hart, D.D., Johnson, T.E., Bushaw-Newton, K.L., Horwitz, R.J., Bednarek, A.T., Charles, D.F., Kreeger,
D.A., and Velinsky, D.J., 2002, Dam Removal: Challenges and Opportunities for Ecological
Research and River Restoration: BioScience, v. 52, p. 669-681.
Hennes, A., 2002, Structural Contours at an Extensional Fault Intersection; Gravity Modeling of the
Northeastern Missoula Valley, Montana. Unpublished Senior Thesis, University of Montana.
Hill, M.C., 1990, PreConditioned Conjugate-Gradient 2 (PCG2), A computer crogram for solving groundwater flow equations, in Survey, U.S.G., ed.: Denver, Colorado, p. 43.
Hill, M.C., 1992, A computer program (MODFLOWP) for estimating parameters of a transient, threedimensional, ground-water flow model using non-linear regression, U.S. Geological Survey OpenFile Report 91-484, p. 358.
Hill, M.C., 1998, Methods and guidelines for effective model calibration, USGS Water Resources
Investigations Report 98-4005, p. 15.
Hill, M.C., Cooley, R.L., and Pollock, D.W., 1998, A controlled experiment in ground water flow model
calibration, v. v36, p. p520(16).
Hill, M.C., and Tiedeman, C.R., 2007, Effective groundwater model calibration: Hoboken, NJ, John Wiley
& Sons, Inc., 455 p.
Hsieh, P.A., Wingle, W., and Healy, R.W., 2000, VS2DHI- A graphical software package for simulating
fluid flow and solute or energy transport in variably saturated porous media, USGS Water
Resources Investigations Report 9., p. 16.
Jahns, H.O., 1966, A rapid method for obtaining a two-dimensional reservoir description from well
response data: Society of Petroleum Engineers Journal, v. 6, p. 315-327.
Janiszewski, F.D., 2007, Seismic reflection and gravity constraints on the bedrock configuration in the
greater East Missoula Area [Masters thesis]: Missoula, MT, University of Montana.
Johnson, A.N., Boer, B.R., Woessner, W.W., Standford, G.C., Poole, G.C., Thomas, S.A., and O'Daniel,
S.J., 2005, Evaluation of an inexpensive small-diameter temperature logger for documenting
ground water-river interactions: Ground Water Monitoring and Remediation, v. 25, p. 68-74.
Land and Water Consulting, I., 2004, Aquifer/Groundwater Analysis, Aquifer Test of Plywood Processing
Well Impacts from Drawdown of the Blackfoot River, Prepared for the Stimson Lumber Company
Bonner, Montana.
Land and Water Consulting, I., 2005, Well #1 Pumping Test Results And Water Availability Report,
Prepared for Canyon River Development, LLC: Missoula, Montana, p. 11.
Landers, J., 2004, River Renaissance: Civil Engineering (08857024), v. 74, p. 52-59.
Landon, M.K., Rus, D.L., and Harvey, F.E., 2002, Comparison of instream methods for measuring
hydraulic conductivity in sandy streambeds: Ground Water v. 39, p. 870-885.
Lorang, M.S., Whited, D.C., Hauer, F.R., Kimball, J.S., and Stanford, J.A., 2005, Using Airborne
Multispectral Imagery To Evaluate Geomorphic Work Across Floodplains Of Gravel-Bed Rivers:
Ecological Applications, v. 15, p. 1209-1222.
44
Mertes, L.A.K., 1997, Documentation and Significance of the Perirheic Zone on Inundated Floodplains:
Water Resources Research, v. 33, p. 1749-1762.
Moore, J.N., and Woessner, W.W., 2002, Arsenic contamination in the water supply of Milltown, Montana,
In: A.H. Welch and K.G. Stollenwerk (editors), Arsenic in Ground Water: Geochemistry and
Occurrence: Boston, Kluwer Academic Publishers.
Naiman, R.J., Johnston, C.A., and Kelley, J.C., 1988, Alteration of North American Streams by Beaver:
BioScience, v. 38, p. 753-762.
Newman, H., 1996, Groundwater Source and Availability of Two Wells Serving Canyon Pines
Homeowners Association, Piltsville, Montana, Prepared for Canyon Pines Homeowners
Association: Missoula, Montana, p. 6.
Newman, H., 2005, Technical Report for Beneficial Water Use Permit for Turah Meadows SUbdivision,
Turah, Montana, Aquifer Test Results Seepage Loss Estimates and Augmentation Plan for Turah
Meadows Subdivision, Turah, Montana: Missoula, Montana, p. 24.
NMFS, N.M.F.S., 2010, Environmental Assessment For The Arra Rogue River Restoration - Gold Ray
Dam Project: Jackson County, National Marine Fisheries Service, p. 110.
Nyquest, D.L., 2001, A depth-to-bedrock model of the Hellgate Canyon and Bandmann Flats area, East
Missoula, Montana using constrained inversion of gravity data. M.S. thesis, Geology Department,
University of Montana, p. 186.
O'Conner, J., Major, J., and Grant, G., 2008, Down with the Dams: Unchaining U.S. Rivers: Geotimes.
Pinder, G.F., Bredehoeft, J.D., and Cooper, H.H., 1969, Determination of aquifer diffusivity from an
aquifer response to fluctuations in river stage: Water Resources Research, v. 5, p. 850-855.
River Design Group, I., Westwater Consultants, I., and Geum Environmental Consulting, I., 2008,
Restoration Plan for the Clark Fork River and Blackfoot River near Milltown Dam, p. 72.
Sandison, D., 2010, Condit Dam Removal Final Second Supplemental Environmental Impact Statement:
Yakima. WA, Department of Ecology, p. 85.
Scalon, B.R., Healy, R.W., and Cook, P.G., 2002, Choosing appropriate techniques for quantifying
groundwater recharge: Hydrogeology Journal, v. 10, p. 18-39.
Sheriff, S.D., and others, 2007, Unpublished gravity data for the Missoula-Milltown region, University of
Montana.
Singh, A., Mishra, S., and Ruskauff, G., 2010, Model Averaging Techniques for Quantifying Conceptual
Model Uncertainty: Ground Water, v. 48, p. 701-715.
Stanford, J.A., Lorang, M.S., and Hauer, F.R., 2005, The shifting habitat mosaic of river ecosystems,
International Association Of Theoretical And Applied Limnology, Volume 29: Stuttgart, p. 123136.
Tallman, A.A., 2005, Sources of Water Captured by Municipal Supply Wells in a Highly Conductive
Aquifer, Western Montana, Masters Thesis, Department of Geology, University of Montana, p.
291.
Trimble, 2008, Research and White Papers for GPS Survey Systems, GPS and GNSS Surveying Systems
White Papers, Trimble.
Udaloy, A.G., 1988, Arsenic mobilization in response to the draining and filling of the reservoir at
Milltown, Montana. M.S. Thesis, Department of Geology, University of Montana.
USDA, 2010, Environmental Assessment Lower Musconetcong River Restoration Project Finesville Dam
Vicinity: Somerset, New Jersey, USDA Natural Resources Conservation Service.
USGS, 2011, National Water Information System Real-Time Data for Montana Streamflow, United States
Geological Service, http://waterdata.usgs.gov/mt/nwis/current?type=flow.
Walton, W.C., 1987, Groundwater pumping tests, design and analysis, Lewis publishers, 201 p.
Westbrook, C.J., Cooper, D.J., and Baker, B.W., 2006, Beaver dams and overbank floods influence
groundwater–surface water interactions of a Rocky Mountain riparian area: Water Resources
Research, v. 42, p. 12.
Westwater Consultants, I., River Design Group, I., and Geum Environmental Consulting, I., 2005,
Restoration Plan for the Clark Fork and Blackfoot River near Milltown Dam – October 2005.
Woessner, W.W., 2000, Stream and Fluvial Plain Ground Water Interactions: Rescaling Hydrogeologic
Thought: Ground Water, v. 38, p. 423.
Woessner, W.W., Moore, J.N., Johns, C., Popoff, M.A., Sartor, L.C., and Sullivan, M., 1984, Arsenic
Source and Water Supply Remediation Action Study, Milltown, Montana. Final Report for Solid
45
Waste Bureau, Montana Dept. Health and Environmental Sciences, in Department of Geology,
U.o.M., ed.: Helena, Montana, p. 448.
Woessner, W.W., and Popoff, W.A., 1982, Hydrogeologic Survey of Milltown, Montana and Vicinity.
Report for the Water Quality Bureau, Department of Health and Environmental Sciences: Helena,
MT.
Woo, M.-K., and Waddington, J.M., 1990, Effects of Beaver Dams on Subarctic Wetland Hydrology:
Arctic, v. 43, p. 223-230.
Ye, M., Pohlmann, K.F., Chapman, J.B., Pohll, G.M., and Reeves, D.M., 2010, A Model-Averaging
Method for Assessing Groundwater Conceptual Model Uncertainty: Ground Water, v. 48, p. 716728.
46
Appendix 2A
Data For Figure 3 (Timeline) (Excel file available in digital format only)
47
Appendix 2B
Compiled Water Level And Climate Data For Study Period (Excel file available
in digital format only)
48
Appendix 2C
Final Calibrated MODFLOW Model Input Files (Files available in digital
format only)
49
Appendix 2D
Well Survey Data (Excel file available in digital format only)
50
Chapter/Paper 3
Berthelote, Antony, Doctor of Philosophy, May 2013
Geosciences
The Role of Drawdown Data in ANN Forecasting of Water Table Responses to Dam and Reservoir
Removals
Chairperson: Dr. William W. Woessner
ABSTRACT:
Planning for and mitigating a groundwater system response caused by dam and reservoir removals
requires the development of methods to forecast post dam removal groundwater levels. A standard
approach to generate the required data involves the development of extensive field based
hydrogeological data sets and the application of sophisticated mechanistic models that solve for the
three-dimensional distribution of fluxes and heads. Such models require large amounts of costly data
and typically require long run times. An alternative is to use statistical models that capture the
relationship between surface processes and the response of the groundwater system. This research
assesses if Artificial Neural Network methods (ANN) can be used to forecast groundwater level
changes likely to occur from a dam removal action. A groundwater level response data set obtained
during the removal of the Milltown Reservoir in western Montana was used to assess ANN model
performance. To further evaluate the ANN modeling forecasts, results were also compared with
forecasts made with a three dimensional MODFLOW deterministic model. ANN modeling was
conducted using MATLAB software and associated tool boxes. ANN forecasts of groundwater levels
utilized daily river discharge, temperature, and sets of field measures of pre reservoir drawdown
reservoir stage ( ANN model AM1). However, ANN forecasts were improved by including
groundwater level data collected during a partial reservoir drawdown (ANN model AM2). The ANN
model trained without the reservoir pool drawdown signal (AM1) residuals for the two subsequent
staged reservoir drawdown forecasts were 1.1 m and 5.8 m. Model AM2 produced a residual of 0.7m
for both forecasts. Average Root Mean Squared Error (RMSE) for AM1 and AM2 forecasts over a
two year forecast period were 2.1 m and 0.7 m. In comparison, the RMSE for the calibrated
deterministic model were 1.3 and 1.4. It was concluded that AM2 ANN modeling produced post dam
out groundwater level forecasts that were similar to both the deterministic model results and field
observations. It is suggested that ANN groundwater level forecasting inclusive of training data
containing a preliminary or temporary reservoir pool drawdown will provide managers with a
reasonable representation of post dam out groundwater conditions.
KEY WORDS
Groundwater Level Forecasting; Artificial Neural Networks (Anns); 3D Numerical
MODFLOW Modeling; Milltown Reservoir; Dam Removal Mitigation And
Management; Groundwater Surface Water Interactions
1.0
INTRODUCTION
There is increasing interest in removing dams in the United States to remedy adverse ecological impacts,
eliminate risks associated with the deteriorating conditions of aging dams, and address societal pressures to
restore rivers to more natural settings (Babbitt 2002; Farinacci 2009; Hart et al. 2002; Landers 2004;
O'Conner, Major, and Grant 2008). Dams and associated reservoirs provide a variety of economic,
environmental, and societal benefits, including recreation, flood control, water supplies, hydroelectric
power, waste management, river navigation, and wildlife habitat. In the United States, there are
approximately 2.5 million small dams less than 1.8 m high, 80,000 large dams over 1.8 m high, and 8000
major dams greater than 15.2 m high (American Rivers, Friends of the Earth, and Trout Unlimited 1999;
Bowman et al. 2002; USACE 2008). To date, dam removal studies have focused principally on
geomorphologic changes (Doyle et al. 2003; Evans et al. 2000; Graf 2003; Hart et al. 2002; Collier, Webb,
and Schmidt 1996). Dam removal projects are commonly planned without consideration of the likely
changes in groundwater conditions that will occur in adjacent aquifers.
51
Industry, agriculture, and population centers are often established in close proximity to dam sites when a
reservoir is created. One in four of the major dams in the United States are constructed in settings with
underlying and/or adjacent unconsolidated and semi-consolidated sand and gravel aquifers. Groundwater
from these systems is often used for primary water supplies (U.S. Dept. of the Interior, 2008; USACE,
2008). A few investigators have reported that after reservoir creation groundwater levels in the adjacent
landscape would locally increase (e.g. Leopold and Maddock, 1954). Heilwell (2005) documented the
presence of additional groundwater in a fractured bedrock system after the construction of a reservoir. It
can therefore logically be presumed that based on hydrogeological principals and the literature that, in most
settings, groundwater levels will typically rise underneath and adjacent to newly constructed reservoirs.
Conversely, it is reasonable to expect that groundwater levels will decline to their previous levels following
a dam removal. In response, water level declines have the potential to impact existing groundwater use and
groundwater supported ecological systems that have developed over the life of the reservoir.
Interestingly, the body of published literature addressing both observed and predicted groundwater changes
during and after dam removals is extremely limited. Recent pre-removal environmental assessments have
begun to mention possible well failures from groundwater head loss following reservoir pool removals (e.g.
the 2.75 m high Finesville Dam (Musconetcong River in NJ) (USDA, 2010) and 11.2 m high Gold Ray
Dam (Rogue River in OR) (NMFS, 2010)). However, “after the fact” well mitigation is more the norm. The
39 m high Condit Dam (White Salmon River in WA) Environmental Impact Statement predicted “... no
significant unavoidable adverse impacts…” to groundwater resources. However, in October 2011 well
mitigation was being discussed to replace multiple failed wells following the dam removal (Sandison,
2010). Wyrick (2009) outlined the social concern for well mitigation from the 6 m high Wadsworth and 4.4
m high Sterling Lake in the Mantua Creek Watershed.
The lack of literature on groundwater impacts resulting from dam and reservoir removals may, in part, be
because few large dams have been removed up to this time. With the exception of water table changes
resulting from the removal of small scale beaver dams which are only a meter or two in height and of
limited areal extent, no comprehensive pre- and post-dam groundwater response studies have been reported
in the literature (Butler and Malanson 2005; Chen and Chen 2003; Mertes 1997; Naiman, Johnston, and
Kelley 1988; Westbrook, Cooper, and Baker 2006; Woo and Waddington 1990). To avoid future conflicts
with water users adjacent to reservoirs planned for removal, resource managers will need appropriate
methods to forecast the consequences of dam removals on groundwater systems.
Standard groundwater modeling approaches include analytical and numerical methods. An ideal model
would produce a reasonable prediction (based on the post removal groundwater level data) at a degree of
uncertainty that is appropriate for the related management action (e.g. identifying the number of domestic
well replacements or remediation, and/or the economics of increased pumping costs). Analytical models are
typically designed to represent relatively simple space and time causes and effects; thus, in most dam
removal settings, they would be of limited value (Guo, 1997). Deterministic numerical models such as
MODFLOW (Anderson and Woessner, 1992; Harbaugh, 2005) and FEFLOW (Diersch and Kolditz, 1998)
allow for the representation of complex hydrogeologic settings and conditions, and can simulate steady
state and transient condition in two or three dimensions. Such models require extensive field derived
physical and hydrogeological input data. When extensive data sets are available and model calibration can
be completed, deterministic modeling approaches are likely to provide adequate tools for post-dam removal
groundwater level forecasting (Berthelote, 2013). Alternatives to analytical and deterministic modeling
approaches include geostatistical (Wiese and Nutzmann 2011; Diodato and Ceccarelli 2006) and non-linear
time-series analysis methods. Our interest here is to assess if Artificial Neural Networks (ANN) modeling
forecasts of groundwater responses to a dam removal provide a reasonable alternative approach to standard
numerical deterministic methods. Citations of the success of ANN modeling in other disciplines report
their use as a more practical and cost effective alternative to predicting outcomes than complex
deterministic modeling approaches. Applications of ANN modeling of groundwater levels driven by
environmental and climatic stresses have recently been examined in a large scale multi-level confined
aquifer system, a lake recharged aquifer system, to represent dynamic head boundaries in an arid
environment, and the impacts of pumping and possible changes in climatic conditions on water table
elevations in a semiconfined glacial aquifer (Coppola et al., 2005; Dogan et al., 2008; Huo et al., 2011;
Nourani et al., 2008).
52
Given that ANN has been used successfully to forecast groundwater levels in a variety of steady state
systems, we investigate its potential to forecast groundwater levels following a large system change, the
response of a groundwater system to the removal of a dam and reservoir. Unlike the previous research
using ANN forecasts, once a reservoir is removed, the entire system is altered resulting in potentially
different surface-subsurface responses. The focus of this research is in three areas: first, assess the
capability of ANN modeling to adequately forecast groundwater responses to large system changes (dam
removal) as an alternative to applying industry three dimensional deterministic numerical modeling
approaches; secondly, assess the value of training data (reservoir pool drawdown signal) that approximate
the system change; and thirdly, propose ANN modeling strategies for forecasting responses of groundwater
systems to dam removal actions. Hydrological data sets and deterministic modeling of the groundwater
response to the removal of the Milltown Dam and reservoir complex in western Montana provided the
observational data and comparative forecasts by which ANN modeling was evaluated. Specifically we
hypothesized that an ANN model trained with pre dam out groundwater level data alone will poorly predict
post dam out groundwater levels. In addition, we hypothesized that groundwater level data used for
training the ANN model that also captures a partial or temporary pre dam out reservoir drawdown data set
will provide predictions that more closely match observations and deterministic model forecasts.
2.0
STUDY SITE AND BACKGROUND
The groundwater response data were generated during the three year Milltown Dam removal project
located 8 km east of the city of Missoula, Montana (Figure 1). The study area extends from Hellgate
Canyon at the eastern edge of the Missoula City limits upstream 13 km to Turah Bridge in western
Montana (Figure 1). The Milltown Dam was located at the confluence of the Blackfoot and Clark Fork
rivers at the center of the study area. The reservoir extended approximately 1.5 km upstream and had a full
pool width of approximately 0.75 km (Berthelote et al., 2007). The Milltown area has a semi-arid climate
with a mean annual temperature is 13.7 oC and the mean annual precipitation of 35.1 cm
(http://www.ncdc.noaa.gov). The average discharge of the Clark Fork River 4.5 km downstream of the
reservoir (USGS gauge number 12340500) is 83.2 m3/s (USGS, 2011).
53
Figure 1: Study area location map. Blue arrows represent groundwater flow directions. Well identifiers are the sites
corresponding with the well responses forecast by ANN modeling. Well 03 data was incomplete and therefore removed
from subsequent analyses and comparisons.
The Milltown Dam construction was completed in 1907 at the confluence of the Clark Fork and Blackfoot
rivers. Over the next 100 years, the Milltown reservoir filled with mining and smelter wastes from the Butte
and Anaconda area located 140 km upstream. The Milltown Reservoir was designated a CERCLA (U.S.
EPA Superfund) site in 1983 as water seeping from the reservoir sediments recharged the adjacent coarsegrained aquifer and contaminated local wells with dissolved arsenic, iron and manganese (ARCO 1992;
Harding Lawson Associates (HLA) 1987; Moore and Woessner 2002; Udaloy 1988; Woessner et al. 1984).
Residents located adjacent to and proximal to the reservoir utilize the bedrock bounded 6 to 60 m thick
coarse sand, gravel, and boulder unconfined valley aquifer for all domestic and community water supplies.
Aquifer recharge is principally from four sources: 1) river channel leakage (losing channels of the
Blackfoot River and the Clark Fork River (below the dam to Hellgate Canyon); 2) leakage from the
Milltown Reservoir pool; 3) lateral valley underflow at the Turah Bridge and Blackfoot Canyon
boundaries; 4) limited inflows at the bedrock boundaries. Groundwater discharges principally as underflow
through the valley aquifer located in Hellgate Canyon to the west, and locally in gaining stream sections of
the Clark Fork River (above the reservoir and, at one time, just below the dam). The groundwater flows
towards the reservoir area from the upper Clark Fork River valley and converges just above the reservoir
with the groundwater entering at the mouth of the Blackfoot River canyon, then flows northwest down
valley (Figure 1). Aquifer flow rates are measured in 10's of m/d, the coarse grained nature of the
sediments allows rapid flow. Hydraulic conductivities range from 90 to >27,000 m/d and reflect the high
energy deposition environment and coarse grained nature of the sediments. With measured horizontal
hydraulic gradients 0.0013 (upper Blackfoot River arm) to 0.066 (near well 05) groundwater velocities are
measured in 10’s of meters per day (porosity estimated at 0.20). Vertical riverbed hydraulic conductivities
range from 0.43 to 12.8 m/d and groundwater river channel exchange rates range 0 to 4 m3/(day·m2)
54
(Berthelote et al. 2007; Berthelote, Woessner, and Thompson 2010; Gestring 1994; Moore and Woessner
2002; Tallman 2005; Woessner et al. 1984).
In 2004 the U.S. EPA, the State of Montana, and stakeholders decided to remove the Milltown Dam and
1.9 of the 5 mcm (million cubic meters) of contaminated reservoir sediments. The 8.5 m high Milltown
Dam and the associated reservoir were removed during the period of 2006 to 2009. Remediation and
restoration plans were designed to be completed in stages (Envirocon, 2006a, b; Westwater Consultants et
al., 2005). This estimated $100+ million remediation/restoration project required three drawdowns starting
in 2006 (3.5 m in March 2007, 3.5 m in March 2008, and 1.2 m in April 2009) that correlated with
engineering tasks prior to reaching the final free flowing state in 2009. Before the dam removal process
formally began, project engineers initiated a 3.5 m temporary drawdown in November of 2005 to examine
the submerged portion of the dam. It was observed that groundwater levels in some wells adjacent to the
reservoir declined and a few shallow wells became inoperable. These conditions resulted in the initiation of
an expanded water level monitoring network, and the construction and calibration of an industry standard
three dimensional deterministic numerical groundwater model (Berthelote et al. 2007; Berthelote,
Woessner, and Thompson 2010). A MODFLOW model was setup in 2006 and transiently calibrated
throughout the initial drawdown. This model was used to forecast likely groundwater responses to
reservoir stage drawdowns and complete dam and reservoir removal. The MODFLOW model forecasts
combined with a decision tree resulted in remediation actions (Berthelote, 2013). The State of Montana
and U.S. EPA implemented a well replacement and well mitigation program that attempted to limit water
supply impacts before further reservoir stage declines were implemented. The numerical groundwater
model was updated annually with new observation data and used to re-evaluated potential groundwater
level changes resulting from planned drawdowns. It is the field observational data and numerical modeling
results from this previous effort that will be compared to the ANN modeling output described in this work.
3.0
METHODS
We evaluated the performance of the ANN by examining to what degree it could forecast the observed
groundwater levels prior to and after the dam removal. We used variations in the composition of the
datasets used to train the ANN. We were specifically interested in evaluating if the inclusion of preliminary
reservoir drawdown information in the ANN training process improved the forecasts of post dam out
groundwater level predictions. The ANN predictions were also benchmarked against the results of the
deterministic modeling performed with MODFLOW (considered an industry standard mechanistic
groundwater model).
3.1
Observed Water Level Changes
Groundwater levels were monitored using recording transducers and electric hand operated water level
monitoring tapes (Berthelote et al. 2007; Farinacci 2009; Tallman 2005). Data derived from a 74 well
network was combined with historical non-continuous hydrologic data dating back to 1981 and compiled
into a single database containing 226 individual wells with over 2000 measurement days (Berthelote et al.,
2010). Farinnaci (2009) recognized that approximately 80% of the observed response of the water table
was controlled by the river stage/discharge conditions. Therefore, the observation database included the
reservoir pond and tailrace elevations (North West Energy, 2007) as well as the available USGS river stage
data (USGS, 2011). The timing and magnitude of the unconfined groundwater system response to reservoir
stage and dam removal activities was determined by analyzing pre- and post-dam removal groundwater
level trends at wells located throughout the study site (measurement error of 0.02 m) (Berthelote et al.,
2007).
3.2
Three Dimensional MODFLOW Model
More than two decades of hydrologic investigation of the Milltown Reservoir Superfund site resulted in the
construction of two earlier two dimensional numerical groundwater models; however, only portions of the
study area were modeled (Brick, 2003; Gestring, 1994). Managers and regulatory agencies responded to
well failures induced by the 2005 temporary drawdown by funding an integrated three year extensive field
data collection campaign and development of a three dimensional groundwater model to forecast the
magnitude, timing, and location of likely physical changes in the groundwater system (Berthelote et al.
2007; Berthelote and Woessner 2008, 2009; Berthelote, Woessner, and Thompson 2010). This model was
used to inform mitigation planning. The model was updated and adapted as drawdown and dam removal
55
activities progressed, operation scheduling changed, and new observational data became available. As a
consequence of the dynamic nature of this large de-construction effort, forecasts were adjusted and revised
as new information became available. Complete documentation of the model development, calibration,
forecasts and uncertainty analyses are presented in the referenced documents (Berthelote et al. 2007;
Berthelote and Woessner 2008, 2009; Berthelote, Woessner, and Thompson 2010). A summary table of
groundwater and surface water model inputs can be located in Berthelote (2010) Appendix B. The final six
layer three dimensional numerical MODFLOW model (Berthelote et al., 2010) consisted of 53,192 active
46 m by 46 m cells. Using standard techniques, it was parameterized and calibrated to steady state
conditions (March 31, 2006), transient conditions (March 31, 2006 to May,9, 2010), and history matched
with October 8, 1992, steady state data and a second transient data set, 1992-1993 (e.g. Anderson and
Woessner). Calibration used automated Least Mean Squares analyses and root mean square error (RMSE)
analysis to evaluate model fit. It should be noted that calibration of this mechanistic model to observational
data required data inclusive of system responses from initial and subsequent drawdowns. Well level
responses, river fluxes, changing reservoir configurations, and pool level changes were among the data sets
needed to complete calibration of the changing system.
3.3
Application of ANN to the Milltown
Study
We used MatLab software and the associated
neural network toolbox
(http://www.mathworks.com) for our ANN
modeling (Figure 2). We trained two networks with
identical architecture but with different training
datasets. Each model was executed using data
representative of a period where both inputs and
the groundwater response (measured change in
water level at a monitoring well) were known
A two layer feed-forward network with LevenbergMarquardt back propagation learning algorithm
(trainlm) was utilized. Two non-linear transfer
functions (Tan-Sigmoid Transfer function and
Purelin Transfer function) were applied to layers
Figure 2 Typical representation of ANN architecture and
processing
one and two respectively. Our approach was to
develop the model using the most parsimonious
(lowest number of neurons) network architecture needed to achieve satisfactory results. This was done to
decrease the chances of data over-fitting. We started with an ANN structure with one node in the hidden
layer and increased the number of nodes by 1 in each successive runs until the ANN was considered
successfully trained or the number of nodes required for a successful run exceeded sixteen. We considered
the network trained if the RMSE of the predictions was better than 0.75 m in less than 200 iterations.
Acceptable convergence was generally achieved in fewer than 100 iterations. Performance goals were
determined based on behavioral observations for each well and averaged 0.23 m for all 12 wells. A copy of
the ANN Code used for this research is located in Appendix 3A.
3.3.1
Data Set
For the Milltown site, observed spatially distributed variations in groundwater levels at 12 wells were used
to assess ANN predictions of drawdown impacts for each of two ANN model architectures. A total of 731
days of input training data were used. Input training data sets included two years of daily measurements
beginning a month prior to the initial reservoir drawdown (March 2, 2006) to just before the second
drawdown (March 01, 2008). This enabled the training/validation model to capture groundwater level
behavior impacted from drawdown activities, required for the testing/forecast model to mimic the impact
during each staged drawdown and future dam out river conditions. Training and validation used 80% and
20% of the available input data, respectively, interlaced for the time period examined.
The observed groundwater level data described above were initially subsidized with reservoir pool level
data (North West Energy, 2007) and potential ANN modeling inputs derived from nineteen continuous
56
climatologic- and hydrologic-data sets acquired from two public internet sources
(http://wrcc.dri.edu/wraws/ and USGS). Climate data was acquired from the Missoula FTS Montana
RAWS data station located at Latitude 46° 51' 00", Longitude 114° 03' 00", 15.4 kilometers from the site
(at an elevation of 976 m). Hydrologic data were acquired from the USGS real time stream data database
for station number 12345000. Independent pre-processing used linear cross correlation to remove input
parameters that had minimal (<0.75 correlation coefficient) or no signal contribution to the groundwater
levels observed for the training period. The final three selected inputs were reservoir pool level, river
discharge, and air temperature. These input data were normalized to a mean of 0 and a standard deviation
of 1 prior to the ANN analysis to avoid undue influence of data types with relatively large values
(magnitude) and variability during the training process.
3.3.2
ANN Input Selection Method
Two ANN models were constructed to evaluate how the nature of the training data, specifically how the
absence or inclusion of groundwater responses to a reservoir pool drawdown event, impacted forecasts of
groundwater level responses. The two models used identical inputs for training, validation, and forecasting
with one exception. AM1training used reservoir pool level data that did not include a reservoir drawdown
signal AM2 training alternatively used pool level data inclusive of a drawdown signal during the training
and validation period. For both models a forecast was conducted using the same 3 input types used for
training and validation where a pool level step function representative of the actual total staged reservoir
drawdown for the forecast period was substituted for the pool level inputs for both models. The other two
inputs were simply duplicated for the forecast period (assuming that the river discharge and temperatures
represented a steady state system and that no future data would be known). The two forecast responses of
groundwater levels to the dam out conditions allowed for a controlled experiment evaluation of how the use
of drawdown data during the training and validation process impacted predicted groundwater responses t.
Table 1 illustrates the variations in the data sets used for each model:
Table 1 We trained two networks with identical architecture but with training datasets varied by one input. This table
identifies the input data used for the two models AM1 and AM2 (Appendix 3B). Inputs were river discharge (Q), daily
temperature (T), and reservoir pool level represented by a full pool (P 1) or reservoir pool levels inclusive of a staged
drawdown (P2). Head or water table elevation in each well is represented as h. AM1 and AM2 varied only by input 3
during training and validation but both used a step function to represent the staged drawdown of the reservoir pool (P)
for model forecasting along with the original Q and T data used for training.
For this modeling evaluation, since the ANN models were also being evaluated to determine how well they
forecast groundwater level responses to reservoir drawdown and dam removal actions at Milltown, the
forecasts were compared with observed groundwater responses. In addition, ANN modeling results were
also compared and contrasted with three dimensional groundwater modeling forecasts to assess prediction
similarities or differences.
3.4
Statistical Methods
Three techniques were used to evaluate performance of the two ANN models relative to observed and
deterministic model groundwater levels: 1) a subjective visual hydrograph comparison of residuals of ANN
57
model transient forecasts and groundwater level observations; 2) a standard root mean square error (RMSE)
analyses of the residuals used to evaluate fit for transient paired data:
where subscripts m and s represent the measured and simulated outputs, respectively, and P is total number
of events considered; and 3) a T-Test evaluation of the model residuals to evaluate the statistical relevance
of the reservoir pool level input.
4.0
RESULTS
4.1
Observed Data
Natural historical groundwater level fluctuations proximal to the dam seasonally vary 1 to 2 m. Below the
dam site, water table elevations typically vary annually up to 4 m, whereas, above the reservoir site annual
variations in groundwater levels are usually less than a meter. Well hydrographs mimic stream stage
changes with the highest groundwater levels occurring during snowmelt driven high river stages in late
spring and low water levels corresponding with late winter low flow conditions. Farinnaci (2009) observed
groundwater responses to changes in river stage occurred with little or no time lag
Figure 3 Spatial representations of the observed groundwater changes (background shaded map) from March 31 2006
to
March 31 2010. Graphs illustrate the observed data (blue lines) in Wells 12, 05, and 10 below, at, and above the
Milltown Dam, respectively. Red boxes and connected dashed highlight the occurrence of low groundwater levels in
March each year and the annual variation in this level from year to year.
Groundwater levels were also impacted by reservoir drawdowns during the reservoir drawdown and
removal process (Stage 1 drawdown 3.5 m; Stage 2 3.5 m; and Stage 3 1.2 m) (Figure 3). The impact of the
first drawdown in June 2006 coincided with the natural declining limb of the hydrograph making reservoir
drawdown only impacts to groundwater less obvious (possibly an increased slope in the hydrographs). The
remaining drawdowns were scheduled to coincide with the historical March low river flows. Once again
separating reservoir drawdown groundwater responses from well hydrographs is partially masked by
antecedent groundwater recharge conditions that also influenced March groundwater levels. When the
entire groundwater level records are reviewed and the changes in March 31 water levels used as reference
points, reservoir removal appears to account for about a 3 m reduction in groundwater levels proximal to
the dam.
58
4.2
MODFLOW Modeling Results
The most parsimonious model became the 244 cell by 169 cell 6 layer MODFLOW model with 53,192
active cells. The model required input parameters (conductivity, recharge, initial head, etc.) for every stress
period and layer at each cell location. Additional parameters for each boundary condition cell were
required (head boundaries, river cells, drains, wells, etc.) resulting in more than 783,000 input parameters.
Detailed information on model construction, calibration and uncertainty analyses are reported in the work
of Berthelote et al. (2007, 2008, 2009, 2010). The groundwater levels (modeled verses observed residuals
were generally less than 2 m for each of the network wells used in calibration) generated by deterministic
modeling were considered to be an adequate representation of groundwater conditions both before dam
removal and after dam removal. Model calibration suggest that MODFLOW results appear to generally
under predict the highest and lowest portions of the groundwater level trends at some locations.
Predicting future conditions has many challenges as both hydrogeological and stream elevations, and river
bed leakage properties and river stages (river discharges and durations) are unknown but required forecast
input parameters and must be approximated. In addition, the construction of the calibrated numerical
model is not a complete representation of the complex hydrological system which is an unfortunate artifact
of all mechanistic models that try to mimic complex systems. However, the aggregate of computed post
dam removal forecast ranges provided managers with sufficient information to assess groundwater level
impacts and develop a proactive remediation program (Berthelote and Woessner 2009; Berthelote,
Woessner, and Thompson 2010). Each model forecast provided groundwater levels for single day in the
future that represented the groundwater response to a set of new conditions (reservoir level drawdowns).
As stated previously, new groundwater level data acquired during the three year dam removal phase was
continually used to update, refine and recalibrate the working model. Both the simulated heads of the final
deterministic calibrated model and the two forecast head distributions from the initial deterministic
modeling were used for comparison to the ANN results presented below.
4.3
ANN Modeling Results
Compared to daily observed or monthly deterministically simulated results, ANN solutions tended to
produce higher frequency signals around a central trend line. However, ANN results captured observed
temporal and spatial spring groundwater level peaks and winter declines (Figure 4). Mechanistic or other
spatial modeling suffer from compounding errors due to parameter uncertainties, interpolation,
extrapolation, or other similar techniques. Since ANN models do not utilize this approach they are not
affected by these types of compounding errors
The well hydrograph shown in Figure 4 illustrate comparisons of the observed and forecasted results for
well 01 that is closest to the Milltown Reservoir. This well location captured the largest magnitude of water
level changes.
59
1.2 m Drawdown
1001
3.5 m Drawdown
Groundwater Elevation (m)
1006
Stage 2 Forecast
Stage 3 Forecast
AM1
-1.1
-5.8
Residuals
AM2
0.7
-0.7
MOD
-1.3
-1.4
996
991
986
981
Mar-10
Dec-09
Sep-09
Jun-09
Mar-09
Dec-08
Mar-08
Jun-08
Sep-08
976
Observed
MOD Forecast
AM1 Forecast
AM2 Forecast
AM2
AM1
A
1001
1.2 m Drawdown
Groundwater Elevation (m)
1006
3.5 m Drawdown
MOD Calibrated
Stage 2 Forecast
Stage 3 Forecast
AM1
-1.1
-5.8
Residuals
AM2
0.7
-0.7
MOD
-1.3
-1.4
996
991
986
981
Mar-10
Dec-09
Sep-09
Jun-09
Mar-09
Dec-08
Sep-08
Jun-08
Mar-08
976
Observed
MOD Forecast
AM1 Forecast
AM2 Forecast
AM2
AM1
MOD Calibrated
B
Figure 4 Changes in measured and modeled groundwater elevations at Well 01. Stage 2 drawdown (March 2008) and
final stage 3 drawdown (March 2010) forecast are shown in blue dots, red diamonds, and green triangles for AM1,
AM2, and MODFLOW models, respectively. A Observed groundwater data are represented by the black line. Raw
ANN groundwater level forecast results are represented by the blue line (AM1), and the red line (AM2. MODFLOW
simulated results are presented in the monthly green squares B Post processed ANN model results using a 30 day
running average to remove high frequency noise plotted with observations, monthly MODFLOW simulated heads and
model forecasts. Appendix 3C contains the compilation of ANN results for each well.
Table 2 Statistical residual and RMSE results representing the differences between simulated water levels and
observed water levels at each of the 12 wells generated from AM1, AM2 and MODFLOW. Wells located below the
dam are shaded in gray.
Steady State Stage 2
Steady State Stage 2
Steady State Stage 2
Steady State Stage 3
Steady State Stage 3
Steady State Stage 3
Transient 2008 to 2010
Transient 2008 to 2010
Transient 2008 to 2010
Transient 2008 to 2010
Transient 2008 to 2010
Well
Residual Initial MODFLOW Model Forecast
Residual AM1 Forecast
Residual AM2 Forecast
Residual Initial MODFLOW Model Forecast
Residual AM1 Forecast
Residual AM2 Forecast
RMSE Final Calibrated MODFLOW Model Simulation
RMSE AM1
RMSE AM2
stdev AM1
stdev AM2
Data for t-test from absolute value of residuals
t-test mean AM1
t-test mean AM2
t-test variance AM1
t-test variance AM2
t-test error Pcalc AM1 vs AM2
Mean
0.2
0.8
0.9
0.7
2.2
0.4
0.5
2.1
0.7
2.2
0.7
1
0.8
0.4
0.8
2.3
5.8
0.7
0.5
4.1
1.0
4.1
1.1
2
0.0
0.6
0.9
1.3
3.0
1.2
0.4
2.9
0.7
2.6
0.8
4
0.7
1.5
1.2
1.8
1.7
0.7
0.6
2.2
0.7
2.1
0.8
5
0.5
0.2
1.1
0.9
1.2
0.6
0.6
2.2
1.1
3.6
1.0
6
3.0
0.3
0.5
1.6
3.3
0.6
0.7
1.2
0.3
1.2
0.3
7
2.5
0.9
0.4
2.4
0.1
0.0
0.4
0.6
0.4
0.6
0.4
8
0.6
0.5
0.9
1.8
1.6
0.8
0.5
1.4
0.8
1.4
0.9
9
1.1
0.8
0.7
0.2
0.3
0.2
0.6
1.2
0.6
1.2
0.6
10
2.6
0.2
0.3
1.5
0.2
0.3
0.3
0.6
0.2
0.6
0.2
11
0.2
2.4
1.2
0.5
3.5
0.1
0.5
1.5
0.8
1.4
0.8
12
0.0
0.3
1.3
0.4
4.3
0.1
0.4
5.2
1.0
5.2
1.0
13
0.1
1.7
0.8
0.4
1.3
0.2
0.4
2.5
1.1
2.6
1.1
2.2
4.3
2.8
2.1
3.3
1.2
0.6
1.4
1.3
0.5
1.7
4.7
2.6
0.8
1.2
0.9
0.7
1.1
0.4
0.3
0.9
0.7
0.2
0.7
1.1
1.3
24.6
52.9
30.0
16.7
46.6
4.6
1.2
7.2
4.8
1.7
5.4
104.0 20.5
1.6
2.7
0.5
1.4
3.2
0.3
0.5
1.8
1.0
0.1
2.0
2.8
2.6
3.E-23 4.E-73 6.E-59 1.E-52 2.E-49 4.E-67 4.E-22 2.E-26 5.E-32 4.E-30 8.E-81 3.E-58 7.E-41
Steady state AM2 results for stage 3 (Dam removed) forecast maintain a mean residuals of 0.4 m, which is
comparable to the 0.7 m residual of the initial MODFLOW model forecast and significantly better than the
AM1 residuals of 2.2 m. Transient results for the period 2008 to 2010 maintained similar performances
with RMSE values for AM1, AM2, and MODFLOW of 2.1 m, 0.7 m, and 0.5 m. The Transient
MODFLOW values however do not reflect a true forecast and are representative of the final calibrated
model results with all data for the period included in the model. Similar residual comparisons are
60
documented for the 12 individual study wells in Table 2 demonstrating the effectiveness of ANN modeling
in capturing general groundwater level trends during the Milltown staged reservoir drawdowns and
subsequent complete dam removal. The maximum RMSE’s for the 12 wells tested were 5.2 m and 1.1 m
for AM1 and AM2, respectively and maximum residuals for steady state forecasts were 5.8 m and 1.2 m
respectively. Individual well RMSE values for AM2 that met or exceeded 1m (wells 01, 05, 12, & 13)
were the closest wells to the reservoir area or down river from the dam. All the other wells were less
impacted by drawdown events as they were upstream of the main reservoir or closer to the valley
boundaries less proximal to the river. A plot of the distance from surface water (reservoir or river) vs
RMSE or variance (from the t-test) demonstrated that ANN modeling residuals for individual wells
increased proportionally with the proximity to surface water (Figure 5). This is a function of signal
dampening produced by the subsurface.
3.50
AM2 Mean
3.00
AM2 Variance
Meters
2.50
2
R = 0.8
2.00
1.50
1.00
R2 = 0.8
0.50
0.00
100
150
200
250
300
350
400
450
500
550
600
Distance from Surface Water (m)
Figure 5 T-test results for AM2 residual variance and mean plotted against minimum distance to surface water
(reservoir pool or river channel). Higher variability is observed closer to water bodies.
Statistical residual comparisons of RMSE, standard deviation, and mean values for ANN models all
demonstrate that AM2 which included reservoir and groundwater drawdown training information produces
forecast groundwater levels that are more representative of observations than did the AM1. The t-test
results comparing the absolute values of the residuals for these two models with a significance level set to
0.001 demonstrates this fact with an average calculated p value of 3E-23 with individual well p values
ranging from 8E-81 to 4E-22. The fact that the p value is far less than the significance level and is
approaching zero concludes that the observed effects were unlikely to be the result of chance alone and that
the null hypothesis is false (Goodman, 2008). The AM2 results are therefore statistically significant.
5.0
DISCUSSION
The literature suggested that ANN modeling could be used to forecast hydrologic time series (Dogan,
Demirpence, and Cobaner 2008; ASCE and The Task Committee on Application of Artificial Neural
Networks in Hydrology 2000), including groundwater level forecasting (Dogan, Demirpence, and Cobaner
2008; Nourani, Mogaddam, and Nadiri 2008). However, its application to forecast the response of
groundwater levels to dam and reservoir removals is a new application. Resource managers and regulatory
agencies overseeing dam removal projects need to consider how actions will affect adjacent groundwater
systems. This responsibility can be addressed using two basic methods, monitoring or modeling. It can be
argued that establishing both spatially and temporally pre-dam and reservoir removal baseline
hydrogeologic data (groundwater levels and river and pool stages) is a critical step in assessing the impacts
of removal actions on the adjacent groundwater resources. Observing groundwater levels and stage
changes during and after dam removal processes will likely result in mitigation following the removal
process (reactive approach). A more proactive approach would be to integrate pre dam removal
hydrogeologic information and forecast water level changes resulting from dam removal plans. This
approach allows for the development of mitigation plans and allows execution of plans prior to observing
impacts.
Forecasting likely groundwater level changes requires some degree of groundwater modeling. Physicallybased deterministic numerical groundwater models such as MODFLOW are widely used to identify
impacts of natural and human-induced changes in the subsurface environment. Such models enable us to
conduct a series of numerical experiments to analyze subsurface flow and transport phenomenon under
61
varying physical, biological and chemical processes. However, because of the practical difficulties of
representing all the natural subsurface complexity, model results include a degree of uncertainty. The
implications of these uncertainties are particularly significant when the models are used in practical
applications for prediction or extrapolation purposes under varying environmental conditions (Demissie
2008 ; Mohammadi, Eslami, and Qaderi 2008). This modeling methodology includes costly data
requirements which can take years of data discovery and interpretation. Finally, forecasts require a
representation of future conditions including groundwater recharge rates and timing, river stage changes,
and groundwater discharge rates and timing, and changes in stresses such as pumping. Accuracy of
forecasted groundwater level responses is dependent on how well future modeling scenarios match reality.
Additionally, if the entire system is altered as in the case of a dam removal, model forecasts can be poor if
they do not integrate some observational changes of initial system responses to changing conditions
(observations measured during a drawdown event including; changes in fluxes from altered river or
reservoir configurations, groundwater level responses etc).
Our data support the hypothesis that using ANN to forecast groundwater levels following a dam removal in
the Milltown setting is feasible when the training datasets include pool levels during a drawdown event.
ANN results were compared to temporal and spatial observational data and contrasted with MODFLOW
results. We concluded that ANN solutions are similar to those obtained using a standard numerical
groundwater model in this case.
It was determined that transient data must contain an adequate representation of hydrogeologic conditions
such as an annual cycle of water level change and, as stated previously, information on how the
groundwater system responds to a change in reservoir stage (a response signal). In our case, this response
signal data set was the observed groundwater level changes in response to the first stage of the reservoir
drawdown. Such a data set is also needed to build an appropriate mechanistic model including a dam
removal event. . Ideally, to maximize the information from observing a drawdown the timing of a response
signal data set should coincide with a hydrological period in which few additional factors are influencing
groundwater level change. Unfortunately, at the Milltown site, the observed groundwater level data set
representing the response to the reservoir level decline coincided with noisy background signals making
training and calibration not ideal. Analyses of data correlations minimize the number of inputs for the ANN
modeling. It also may be useful to precondition data inputs to minimize high frequency signals (low pass
filter), a process that would result in smoothing ANN modeling solutions. Such data processing would
likely avoid anomalous responses to high frequency signal inputs (e.g. large precipitation events and/or ice
dam induced recharge events). It is also recognized that further data reduction techniques, such as
dimensional analysis or principal component analysis may allow for maximization of information and
minimize any redundancies like interaction effects between inputs.
It is not surprising that groundwater level forecasts are predominantly dependent on river discharge,
climatic data that serves as proxies for snow melt and groundwater recharge timing (maximum and average
temperature, solar radiation, and total heating degree days), and reservoir pool levels. ANN input
parameters will likely change when attempting to predict groundwater level responses in other
hydrogeological settings dominated by alternative sources of recharge and discharge. For example, a
precipitation dominated groundwater recharge system would presumably utilize any number of
precipitation inputs and be less dependent on temperature data. The advantage of the ANN methodologies
presented here is that input selection can be semi-automated, allowing for the evaluation of a wide variety
of data sets.
It should be cautioned that if one or more available inputs are independent or weakly related to the
dependent variable, the ANN may have difficulty reaching a viable solution. The explanatory value of input
variables is highly dependent on the type of system under consideration. For example, if a dam is removed
in a low permeable (bedrock) or confined system where the aquifer recharge does not originate from the
reservoir leakage, then: 1) the impacts to the local groundwater system would presumably be negligible or
absent, and therefore 2) any attempt to use the ANN methodologies presented in this work would lack the
correlations required to reach an acceptable solution. More importantly, due to the poor extrapolation
power of ANN, any assessment of impacts under conditions beyond these represented in the calibration
data set may be potentially subject to large errors. Standard numerical groundwater models (MODFLOW)
62
are better suited to predict complex systems where physical impacts in the system or where strong transient
alterations of the inputs beyond calibration (training) conditions are expected. ANN solutions are limited to
point location, and independent interpolation techniques are needed to reconstruct the spatial distribution of
the water table. In contrast, depending on the space and time discretization, deterministic modeling
provides solutions at every cell in its domain although it is important to keep in mind that these models are
vulnerable to errors in the generation of the spatially distributed parameter fields needed to run them. Both
modeling approaches require information about future conditions to allow for accurate forecasting. ANN
modeling would need future climatic, river and pool stage data sets. Realizing that there is a large degree
of speculation and uncertainty in these data, a reasonable approach would be to run a number of likely
scenarios and create a forecast tempered with probability information. In our case, based solely on observed
data it is not clear how the future March low water table positions will be modified by natural variations in
the magnitude, duration and distribution of steam stage and discharge, and changes in the river bed
sediment character (leakage properties). For example, it is likely that a series of drought years may limit
stream/aquifer recharge and result in lower water table positions than measured in 2010, one year after the
dam was removed.
Despite the limitations of ANNs, they are a convenient method that permits real-time continuous
improvement of the forecasts as new data become available for further training and permits predictions and
offer useful information even for poorly understood systems (Hertz 1991; Zurada 2006.; Sung 1998). This
may provide decision makers with a convenient tool for managing water resources without the need for
extensive data collection. Second, the increasing number of dam removals, particularly larger dams, will
necessitate a rapid and cost-effective consideration of the impacts on local groundwater systems that
ANN’s can provide
6.0
CONCLUSIONS
Two potential ANN scenarios were developed and used to evaluate the importance of training data that
contained groundwater level responses to reservoir drawdowns in providing post dam removal estimates of
groundwater levels. AM1 forecast impacts to groundwater without a drawdown response training data set
and AM2 utilized such information. AM2 forecasts more closely matched forecast generated from a
MODFLOW model and observed groundwater levels than the AM1 models. The maximum RMSE’s for
the 12 wells tested were: AM1 (5.2 m) and AM2 (1.1 m), with respective averages of (2.1 m and 0.7 m).
The MODFLOW model required 3 years of extensive field data collection and iterative expert model
calibrations to produce forecasts. In contrast the semi automated ANN solutions used data from 2 internet
data sources (USGS and RAWS), reservoir pool levels acquired from the dam operation records, and at
least one set of groundwater level data proximal to the dam for calibration matching. Though many
environmental data sets are available for inputs into ANN models, they should be limited to those that have
some correlation with aquifer recharge. Input conditioning must also include data normalization. . We
found that it would be practical to implement ANN modeling to forecast the response of individual and
groups of wells to dam and reservoir removal actions. This approach would be useful to water managers
and project leaders as the consequences of remediation and restoration are considered. The reduced data
and manpower requirements for ANN modeling make it a practical methodology in hydrogeologic settings
where a reservoir contributes to aquifer recharge for forecasting how groundwater levels are likely to
change when a dam is removed.
63
7.0
REFERENCES:
Anderson, M.P., and Woessner, W.W., 1992, Applied groundwater modeling, simulation of flow and
advective transport, Academic Press, 382 p.
Berthelote, A.R., 2013, Forecasting Groundwater Responses to Dam Removal [Dissertation thesis]:
Missoula, University of Montana.
Berthelote, A.R., Farinacci, A., Storb, M., and Woessner, W.W., 2007, Determining the impact of Milltown
Reservoir Drawdowns on the Adjacent Unconfined Aquifer: Remediation Stage 1 Observed and
Stage 2 Forecasted, University of Montana, p. 174.
Berthelote, A.R., Woessner, W.W., and Thompson, J.H., 2010, Assessing Groundwater Level Behavior and
Numerical Model Forecasts of Water Table Changes Related to Removal of a Large Dam:
Milltown Reservoir Remediation, University of Montana, p. 113.
Brick, C., 2003, Preliminary groundwater modeling to estimate effects of dam and sediment removal on the
alluvial aquifer in Milltown, Montana. Clark Fork Coalition. Report., p. 24.
Coppola, E.A., McLane, C.F., Poulton, M.M., Szidarovszky, F., and Magelky, R.D., 2005, Predicting
Conductance Due to Upconing Using Neural Networks: Ground Water, v. 43, p. 827-836.
Diersch, H.J.G., and Kolditz, O., 1998, Coupled groundwater flow and transport: 2. Thermohaline and 3D
convection systems: Advances in Water Resources, v. 21, p. 401-425.
Dogan, A., Demirpence, H., and Cobaner, M., 2008, Prediction of groundwater levels from lake levels and
climate data using ANN approach: Water Sa, v. 34, p. 199-208.
Envirocon, 2006a, Milltown Reservior Sediments Operable Unit Remedial Action Monitoring Plan, p. 263.
Envirocon, 2006b, Remedial design work plan, Final, For stages 1,2, and 3 drawdowns and related
construction activities Milltown Reservoir Sediments Site.
Farinacci, A.J., 2009, Surface water and groundwater exchanges in fine and coarse grained river bed
systems and responses to initial stages of dam removal, Milltown, Montana [Unpublished Masters
Thesis thesis]: Missoula, University of Montana.
Gestring, S.L., 1994, The interaction of the Clark Fork River and the Hellgate Valley Aquifer near,
Milltown, Montana. M.S. Thesis, Geology Department, University of Montana.
Goodman, S., 2008, A Dirty Dozen: Twelve P-Value Misconceptions: Seminars in Hematology, v.
doi:10.1053/j.seminhematol.2008.04.003, p. 135-140.
Guo, W., 1997, Transient groundwater flow between reservoirs and water-table aquifers: Journal of
Hydrology, v. 195, p. 370-384.
Harbaugh, A.W., 2005, MODFLOW-2005, The U.S. Geological Survey Modular Ground-Water Model—
the Ground-Water Flow Process, 253 p.
Heilweil, V.M., Susong, D.D., Gardner, P.M., and Watt, D.E., 2005, Pre- and Post-Reservoir GroundWater Conditions and Assessment of Artificial Recharge at Sand Hollow, Washington County,
Utah, 1995-2005: Salt Lake City, Utah, U.S. Geological Survey.
Huo, Z.L., Feng, S.Y., Kang, S.Z., Mao, X.M., and Wang, F.X., 2011, Numerically modelling groundwater
in an arid area with ANN-generated dynamic boundary conditions: Hydrological Processes, v. 25,
p. 705-713.
Leopold, L.B., and Maddock, T.M., 1954, The flood control controversy, Big dams, little dams, and land
management.
NMFS, N.M.F.S., 2010, Environmental Assessment For The Arra Rogue River Restoration - Gold Ray
Dam Project: Jackson County, National Marine Fisheries Service, p. 110.
North West Energy, 2007, Milltown Dam Daily Readings, in Montana, U.o., ed.: Missoula, Montana,
Northwest Energy.
Nourani, V., Mogaddam, A.A., and Nadiri, A.O., 2008, An ANN-based model for spatiotemporal
groundwater level forecasting: Hydrological Processes, v. 22, p. 5054-5066.
Sandison, D., 2010, Condit Dam Removal Final Second Supplemental Environmental Impact Statement:
Yakima. WA, Department of Ecology, p. 85.
U.S. Dept. of the Interior, 2008, National Atlas of the United States, Volume 2008,
http://nationalatlas.gov/atlasftp.html#dams00x.
USACE, 2008, Groundwater Linkages to Dam Removals or Emplacements, in Conyngham, J., ed., United
States Army Corps of Engineers Environmental Laboratory's PI in Dam Removal Research, p.
Personal Communication.
USDA, 2010, Environmental Assessment Lower Musconetcong River Restoration Project Finesville Dam
Vicinity: Somerset, New Jersey, USDA Natural Resources Conservation Service.
64
USGS, 2011, National Water Information System Real-Time Data for Montana Streamflow, United States
Geological Service, http://waterdata.usgs.gov/mt/nwis/current?type=flow.
Westwater Consultants, I., River Design Group, I., and Geum Environmental Consulting, I., 2005,
Restoration Plan for the Clark Fork and Blackfoot River near Milltown Dam – October 2005.
Wyrick, J.R., Rischman, B.A., Burke, C.A., McGee, C., and Williams, C., 2009, Using hydraulic modeling
to address social impacts of small dam removals in southern New Jersey: Journal of
Environmental Management, v. 90, Supplement 3, p. S270-S278.
65
Appendix 3A
ANN Code (Files available in digital format only)
66
Appendix 3B
ANN Input Files ANN Data (Inputs and Results) (Files available in digital
format only)
67