NON-INVASIVE FLOW PATH CHARACTERIZATION in a MINING-IMPACTED WETLAND by James C Bethune

NON-INVASIVE FLOW PATH CHARACTERIZATION IN A MINING-IMPACTED WETLAND

by James C Bethune A thesis submitted to the Faculty and the Board of Trustees of the Colorado School of Mines in partial fulfillment of the requirements for the degree of Master of Science (Hydrology).

Golden, Colorado

Date ______

Signed: ______James C Bethune

Signed: ______Dr. Kamini Singha Thesis Advisor

Golden, Colorado

Date ______

Signed: ______Dr. David Benson Professor and Program Director Hydrological Science and Engineering

Signed: ______Dr. Paul Santi Professor and Department Head Department of Geology and Geological Engineering

ii" " ABSTRACT

Time-lapse electrical resistivity (ER) is used in this study to capture the annual pulse of acid mine drainage (AMD) contamination, the so-called ‘first-flush’ driven by spring snowmelt, through the subsurface of a wetland downgradient of the abandoned Pennsylva- nia Mine workings in Central Colorado. Data were collected from mid-July to late October of 2013, with an additional dataset collected in June of 2014. ER provides a distributed measurement of changes in subsurface electrical properties at high spatial resolution. In- version of the data shows the development through time of multiple resistive anomalies in the subsurface, which corroborating data suggest are driven by changes in total dissolved solids (TDS) localized in preferential flow pathways. Because of the non-uniqueness inherent to deterministic inversion, the exact geometry and magnitude of the anomalies is unknown, but sensitivity analyses on synthetic data taken to mimic the site suggest that the anomalies would need to be at least several meters in diameter to be adequately resolved by the inversions. Preferential flow path existence would have a critical impact on the extent of attenuation mechanisms at the site, and their further characterization could be used to parameterize reactive transport models in developing quantitative predictions of remediation strategies.

iii TABLE OF CONTENTS

ABSTRACT ...... iii

LIST OF FIGURES AND TABLES ...... vi

ACKNOWLEDGMENTS ...... viii

CHAPTER 1 GENERAL INTRODUCTION ...... 1

CHAPTER 2 NON-INVASIVE FLOW PATH CHARACTERIZATION IN A MINING-IMPACTED WETLAND ...... 6

2.1 Abstract...... 6

2.2 Introduction ...... 7

2.3 FieldSiteDescription...... 9

2.4 Material and Methods ...... 12

2.5 Inversion ...... 14

2.6 Evaluating Error ...... 15

2.7 Results ...... 17

2.7.1 Supportingdata...... 20

2.8 SensitivityAnalysis...... 23

2.9 Discussion and Conclusion ...... 24

2.10Acknowledgments...... 26

CHAPTER 3 FUTURE WORK ...... 27

3.1 Long-term monitoring ...... 27

3.2 Reactive Transport ...... 28

iv 3.3 CharacterizationofPennsylvaniaMineLeakage ...... 30

APPENDIX A - EXTENDED METHODS ...... 32

A.1 ResistivityDataAnalysis...... 32

A.2 FiniteElementMeshDesignandGmsh ...... 33

APPENDIX B - MISCELLANEOUS DATA ...... 35

REFERENCESCITED ...... 43

v LIST OF FIGURES AND TABLES

Figure 1.1 Geological map of the Pennsylvania Mine area, including the hypothesized Montezuma shear zone. Modiﬁed from Bird, 2003...... 3

Figure 2.1 Map of study region with Peru Creek, resistivity array, and borehole sample locations...... 10

Figure 2.2 Resistivity inversion of data collected on July 12th, 2013. Electrodes (E1-E72), model ﬁtting parameter results, borehole logs, and the general characterofvegetationareshown...... 18

Figure 2.3 Resolution of inversion of data collected on July 12th, 2013. Note, because of smoothing issues, only data for 1 m x 1 m pixels are shown. . . 18

Figure 2.4 Time-lapse percent changes in resistivity, relative to background inversionof12July2013data...... 19

Figure 2.5 Time-lapse absolute change in resistivity, relative to background inversionof12July2013data...... 19

Figure 2.6 Flow diagram of the sensitivity modeling process. ’Summarized region’ denotes the area over which the total resistivity anomaly is calculated. . . 23

Figure2.7 Sensitivitymodelingresults...... 25

Figure B.1 All wetland inversions with ﬁtting results. All changes are relative to the background inversion of data from 12 July 2013...... 36

Figure B.2 Resolutions of all inversions through the wetland...... 37

Figure B.3 Temperature (A) and conductivity (B) measurements taken from boreholesinthewetlandarea...... 38

Figure B.4 Average temperatures measured in the boreholes at shallow <1.5 m bgs., and deep depths...... 39

FigureB.5 Sensitivitymodelingresults...... 39

vi Figure B.6 Additional measurement locations, including pressure transducers and stilling well. HOBO W1 denotes the location of the transducer installed in Peru Creek. HOBO A1 denotes the location of the air pressure transducer from October to November. HOBO A2 denotes the air pressuretransducerleftatthesiteoverwinter...... 40

Figure B.7 Water temperature (A) and pressure (B) measurements of HOBO W1. Pressure has been corrected for air pressure and converted to cm water. . 41

Figure B.8 Air temperature (A) and pressure (B) measurements of HOBO A1. . . . . 42

TableB.1 DischargemeasurementsfromPeruCreek...... 36

vii ACKNOWLEDGMENTS

It took the support of many people and institutions to make this project possible. My advisor, Dr. Kamini Singha, and committee members Dr. Rob Runkel and Dr. Alexis Navarre-Sitchler were all instrumental in their contributions. Kamini’s constant source of knowledge and direction throughout the project was deeply appreciated, as was Rob’s valuable assistance at the field site and with the manuscript. The inspiration to work in an AMD impacted site arose from a chance conversation with Dr. Katie Walton-Day, following a pre- sentation she gave at the Colorado School of Mines. Conversations with Je↵ Graves, Mark Rudolph, and Dr. Stan Church would all later provide valuable insights for the project. Many volunteers tirelessly supported this project in the field, often by trudging through the mucky wetland, carrying heavy batteries, and nearly always with inclement weather quickly approaching. In particular, fellow HSE students Ben Bader, Skuyler Herzog, Em- manual Padilla, and Mike Sanders, were all kind enough to dedicate multiple days to the project. My time at CSM was supported by a teaching assistantship provided by the Geology Department. The experience was beyond rewarding, and has inspired me to continue to incorporate teaching into my life in some capacity. It also proved to be an excellent field work volunteer recruiting position. Finally, Jackie Randell provided key assistance throughout the project, in the field, with the text, and in the lab. Without Jackie, this project would be in a very di↵erent place, and Idon’tthinkIcanthankherenough.

viii CHAPTER 1 GENERAL INTRODUCTION

Weathering of sulﬁde deposits throughout the Montezuma Mining District in Central Colorado presents a major environmental water quality issue for the Snake River and its tributaries. Sulﬁde oxidation produces acid and releases high concentrations of metals, resulting in ecologically toxic discharge known as acid rock drainage (ARD). Because of its abundance, pyrite (FeS2) is the primary mineral responsible for ARD production. There are multiple pyrite oxidation reaction pathways, but in the acidic conditions observed at mine sites, pyrite is oxidized by ferric iron (Fe3+)inthefollowingmicrobiallymediatedreaction (Hallberg, 2010):

3+ 2+ + FeS +14Fe +8H O 15Fe +2HSO +14H (1.1) 2 2 ! 4 Mining operations greatly accelerate the sulﬁde weathering process through augmentation of available reactive mineral surface area (Alpers et al., 2007). To di↵erentiate it from naturally occurring ARD, discharge from mined lands is called acid mine drainage (AMD). Current mining practices seek to minimize impact on water resources, but the Montezuma District contains many historic and abandoned mines that pre-date recent impact concerns and regulations. Equation 1.1 typically proceeds until all available pyrite is consumed, as as aresultthee↵ects of AMD can persist for decades or even centuries after mining operations have ceased (Younger, 1997). Because of its persistent and pervasive nature, AMD has been described as the greatest water quality issue facing the western US today (Da Rosa et al., 1997). Analyses of water and sediment samples taken from throughout the Snake River and its tributaries found that concentrations of zinc consistently exceed acute and chronic toxicity thresholds for trout (Fey et al., 2001). Indeed, the Snake River currently needs to be re- stocked with trout each spring because they cannot survive the winter in the mining-impaired

1 habitat (Fey et al., 2001). There is some debate as to the existence of a shear zone, locally known as the Montezuma shear zone, cutting through the site (Figure 1.1) that may be regionally enhancing the rate of pyrite weathering (Wood et al., 2005). Some have argued that a linear zone of ductile and brittle features across the front range represent a major strain feature of the crust. Others have documented features in the area that would be inconsistent with a large crustal strain feature, and instead suggest that deformation associated with the area is related to Laramide deformation (Caine et al., 2010). In any event, the bedrock of the region contains a large density of fractures that serve as fundamental hydrogeological conduits (Caine & Tomusiak, 2003). The Snake River becomes signiﬁcantly more impacted with metals after its conﬂuence with Peru Creek, its largest tributary. Although some of the dissolved metals loads are the result of naturally occurring ARD (Verplanck et al., 2009), the U.S. Geological Survey came to the following conclusion after extensive sediment and water chemistry sampling (Fey et al., 2001):

Primary targets for remediation should target identiﬁed mining sources draining into those reaches of Peru Creek. Other sources of metals in the watershed are minor by comparison.

In particular, the Pennsylvania Mine, which is near Peru Creek about 4 miles upstream of its confluence with the Snake River, was identified as a major contributor of metals to the watershed (Fey et al., 2001). The Pennsylvania Mine was historically the largest in the area, yielding a total of over 105 kg of gold, 26,000 kg of silver, 2,800 kg of lead, 27,000 kg of copper, 336,000 kg of zinc (Bird, 2003; Lovering, 1935). After the mine was closed in 1953 (Bird, 2003), it changed hands several times, eventually falling into management by the US Forest Service as a part of a broader e↵ort to restore the watershed (County, 2005). Initial mass balance calculations showed that the major surface inflows of the upper reaches of Peru Creek could not account for the extent of local metal loading (Fey et al., 2001), a fact which was attributed to additional loading from the Pennsylvania Mine, but

2 pC

Qal

Peru Creek

Pennsylvania Mine Penn. Mill Cinnamon Gulch Qal

EXPLANATIONYg Delaware Qal Sur!cialpC deposits pC Precambrian DIVIDE X Mine Yg Porphyritic quartz Montezuma monzonite of the Shear zone Montezuma Stock

CONTINENTAL

Figure 1.1: Geological map of the Pennsylvania Mine area, including the hypothesized Montezuma shear zone. Modiﬁed from Bird, 2003. which was not investigated further at that time. Additional synoptic sampling performed along the Pennsylvania Mine reach of Peru Creek in September 2009 identiﬁed a di↵use source of contamination emanating from a wetland between the mine and Peru Creek (Runkel et al., 2013). There are three potential contaminant transport pathways through the wetland that could be contributing to the metals loading in Peru Creek. First, the wetland could be generating contamination in several large deposits of mining waste rock. Waste rock often contains disseminated pyrite, and has been documented to discharge severely contaminated water at other sites (Smith, 1995). Second, the wetland could have a direct hydrogeological connection with the mine, as indicated by the recovery of tracers injected directly in multiple wells in the wetland (Mark Rudolph, Colorado Geological Survey, personal communication

3 of unpublished data). The mine workings are extensive and remain poorly mapped due to hazardous structural collapses (Lovering, 1935; Rudolph, 2010). As a result, flow through the mine workings remains poorly understood. Third, chemical analyses of groundwater downgradient of the mine outflow suggest that the mine outflow is infiltrating into groundwater, opening the possibility that water from the mine outflow is also reaching the wetland area (Rudolph, 2010). It is unclear how flow through the wetland is a↵ecting downstream transport of AMD contaminants. Wetlands have the capacity to precipitate metal sulfides (Sheoran & Sheoran, 2006) and adsorb positive metal ions to negatively charged clay particles or organic material (Johnson & Hallberg, 2005). The longer flow paths and slower velocities of subsurface flow allow for greater contact time with biogeochemically active attenuating features, therefore flow through the subsurface can be particularly important to promoting removal processes (Gandy et al., 2007; Mulholland & DeAngelis, 2000). However, the wetland was previously found to be ine↵ective in remediating redirected mine discharge (Emerick et al.,1988). The presence of preferential flow paths through the wetland could limit the extent of attenuation mechanisms, while complicating interpretations of downstream breakthrough curves. Preferential flow paths result in earlier breakthrough time, lower residence time, and more pronounced tailing (Brusseau, 1994). The existence of mining waste piles in the wetland increases the likelihood that preferential flow exists in the wetland because deposition of mining waste often results in vertical grading, with larger grains tumbling down and finer grains settling over the surface (Smith, 1995). Slug tests on boreholes in the wetland reveal substantial variability in hydraulic conductivity, which also suggests preferential flow (Emerick et al.,1988). Ongoing remediation e↵orts begun in the summer of 2012 include entering the mine to identify potential sources of contamination, evaluating the potential for a bulkhead installation to stymie the outflow of water from the mine, and moving the mine tailings farther from Peru Creek. A recent characterization of water flowing through the mine found that

4 a substantial volume of relatively clean water drains the crosscut of the lowest mine level upstream of the main mine workings, while a smaller volume of highly impacted water drains the inner mine works (Personal Communication Mark Rudolph, 2013). As of the time of this writing, the plan is to install two separate bulkheads in the lower cross-cut. However, success of the bulkhead installation is contingent on its ability to plug the mine, saturate the mine workings, and limit further sulfide oxidation. If the wetland is in direct hydrogeological connection with the mine workings, it would indicate that the mine workings may be leaking internally, and may not hold the water required to maintain fully saturated conditions. The goal of this research is to contribute to the understanding of subsurface flow within the wetland, and to explore those results in light of recent remediation activities and with regard to AMD transport processes more generally. The results from this research have been compiled into the following manuscript for submission to the Journal of Contaminant Hydrology. After the paper, the reader will find a closing statement in which future directions for this research are explored, followed by a number of appendices containing extended documentation of methods and data, and discussion of several topics in the main body of the paper.

5 CHAPTER 2 NON-INVASIVE FLOW PATH CHARACTERIZATION IN A MINING-IMPACTED WETLAND

A paper to be submitted to the Journal of Contaminant Hydrology James Bethune1, Jackie Randell2, Robert L. Runkel3, Kamini Singha4

2.1 Abstract

1Primary Author and Researcher, Graduate Student, Colorado School of Mines 2Field Technician, Colorado School of Mines 3Scientist, U.S. Geological Survey 4Associate Professor, Colorado School of Mines

6 2.2 Introduction

Weathering of sulfide deposits presents a major environmental water quality issue by creating acidic conditions and mobilizing heavy metals (reviews include Da Rosa et al., 1997; Nordstrom, 2011a). Although acid rock drainage naturally forms as a byproduct of sulfide oxidation, mining operations can increase the weathering rate by up to three orders of magnitude through the augmentation of reactive mineral surface area (Alpers et al.,2007). In the western U.S., acid mine drainage (AMD) impacts between 8,000 and 16,000 km of streams on Forest Service land alone (US Forest Service, 1993). The e↵ects of AMD can persist for decades or even centuries after mining operations have ceased through continued oxidation and dissolution of acid-releasing minerals (Younger, 1997). E↵ective remediation of AMD requires detailed knowledge of contaminant transport through the subsurface, where longer retention times may allow for extended contact with attenuating agents (Zhu et al., 2002). Heterogeneity and preferential flow path development in AMD settings has been shown to decrease the eciency of contaminant attenuation (Malmström et al., 2008), likely because preferential flow paths reduce the residence time of solutes in the subsurface and contact with attenuating agents (Brusseau, 1994). Deposition of mining waste piles typically results in graded bedding, through which most discharge is concentrated into a small of the total rock volume (Morin & Hutt, 1994; Smith, 1995). Un- fortunately, the subsurface is rarely mapped to a sucient extent to identify and characterize flow paths, especially at historical mine sites, where e↵orts generally contend with a lack of site data and highly disturbed aquifer material (e.g., Nordstrom, 2011b; Oram et al.,2010). Many AMD remediation projects expend considerable e↵ort constraining flow and transport parameters through tracer injections (Benner et al., 2002), hydrograph separation (Smith, 1995), flow balance calculations (Gélinas et al., 1994), and aquifer permeability tests, or are otherwise forced to make simplifying assumptions regarding subsurface homogeneity. The high conductivity of AMD has been demonstrated to be a useful tracer for mapping mining contamination (Gray, 1996), and makes it an excellent target for electrical geophysical

7 methods (Merkel, 1972). Electrical resistivity (ER) is a geophysical technique that measures the electrical conductance of the subsurface by both establishing and measuring a potential gradient between one or more pairs of electrode (Binley & Kemna, 2005). The procedure is repeated for many di↵erent electrode locations and current configurations to develop a spatially distributed dataset of subsurface conductance (See Loke et al., 2013, for a recent review). ER has been previously used to image both the extent and concentrations of subsurface mining contamination (e.g., Oldenburg & Li, 1994; Rucker et al., 2009). However, these studies were limited in that they assumed a consistent relationship between resistivity and TDS (Day-Lewis et al., 2005), which, due to heterogeneities in the aquifer and in the resolution of the inversion, can be dicult to define (Singha & Gorelick, 2006). Time-lapse ER circumvents the reliance on petrophysical relationships by attributing changes in measured resistivity to changes in pore fluid conductivity. Many time-lapse ER studies inject a conductive tracer to facilitate flow path imaging (e.g., Ward et al.,2010); however the ‘first-flush’ behavior demonstrated by many mine sites creates an ideal natural electrical signal to capture and define contaminant transport using time-lapse ER. This sea- sonal pulse of AMD can be used in place of a tracer, eliminating assumptions about contamination source location that are implicit in injected tracer studies. The largest contaminant loads are typically, though not always, coincident with large storms following prolonged dry conditions (Miller & Miller, 2007; Nordstrom, 2009). The goals of this paper are twofold: first, to demonstrate the use of time-lapse ER to map AMD flow paths with application to characterizing contaminant transport. Second, to demonstrate the sensitivity of ER to di↵erent flow path geometries. Inverting ER measurements using a standard L2-norm necessarily involves smoothing (Day-Lewis et al.,2005), which may lead to some smaller features being lost. An understanding of the capabilities of ER to resolve small features is crucial for actionable analysis in AMD settings. ER has been previously used to characterize the extent of AMD contamination, but the novel approach outlined in this paper is the first time that time-lapse ER has been used to image natural

8 conductivity changes in an AMD setting.

2.3 Field Site Description

This research was conducted in a wetland between the historic Pennsylvania Mine and Peru Creek, a headwater stream to the Colorado River in Central Colorado (Figure 2.1). The Peru Creek basin is bracketed on the north and east by the Continental Divide, and drains west into the Snake River. Because 80% of precipitation falls as snow, the hydrograph is dominated by spring snowmelt pulse (Crouch et al.,2013). The local geology includes part of a heavily mined Oligocene quartz monzonite porphyry of the Montezuma district (Figure 1.1). The Montezuma stock intruded the precambrian schist and gneiss, causing extensive fracturing and faulting and widely disseminating pyrite (Fey et al., 2001). The mineral assemblage includes abundant sulﬁdes, in particular pyrite

(FeS2), sphalerite ([Zn,Fe]S), and galena (PbS) (Lovering, 1935). The Snake River contains ecologically toxic concentrations of zinc, cadmium, and copper as a result of natural and anthropogenically-induced pyrite weathering (Wood et al., 2005). Secondary porosity associated with the Colorado Mineral Belt has been suggested to enhance the rate of pyrite weathering in both mining impacted and unimpacted areas, though the precise nature and cause of that porosity has been debated (Caine & Tomusiak, 2003; Wood et al.,2005). A study of the nearby Handcart Gulch, an unmined drainage near the edge of the Mon- tezuma district, found deposits of ferricrete (iron oxide) coating the streambed (Verplanck et al., 2009), indicating that background metals concentrations are high even in unmined drainages in the area, likely due to natural weathering of sulﬁde minerals. Regionally, sulfate concentrations, which are a common proxy of mining contamination, are highest in areas with inactive mines and with extensive hydrothermal alteration (Fey et al., 2001). Many abandoned mines are scattered throughout the region, but water and sediment chemistry analyses of the Snake River reveal that one of the largest contributor of metals is the Pennsylvania Mine reach of Peru Creek (Todd et al.,2005).

9 The Pennsylvania Mine was one of the largest mines in the region during its operation from 1885 to 1953 (Bird, 2003). The extensive underground mine workings were historically accessed via 6 adits, two of which remain at least partially open today (Lovering, 1935; Wood et al., 2005). A surface ﬂow exits the lower adit and discharges into Peru Creek approximately 100 m upgradient of the wetland (Figure 2.1).

GW3 > Electrode #72

> GW5 Creek

Peru Electrode #1 > >> GWC1 MW02-04

E Electrode > Sample well

Resistivity survey Cinnamon Gulch

Pennsylvania Mine Peru Creek Surface inflow Elevation contour (10 m) 0 0.05 0.1 0.2 Kilometers Wetland area ±

Figure 2.1: Map of study region with Peru Creek, resistivity array, and borehole sample locations.

Because of its high impact on local water sources, the surface water chemistry of Peru Creek has been studied extensively (Fey et al., 2001; McKnight & Bencala, 1990; Runkel et al., 2013; Sullivan & Drever, 2001). There is a large seasonality to both surface flow and contaminant concentration and loading in Peru Creek (Sullivan & Drever, 2001). The peak flow occurs in late May to early June and is typically 5-10 times as great as low flow during early spring (Sullivan & Drever, 2001; Todd et al., 2005). Metals concentrations in the mine

10 outflow are highest during the high spring flows, consistent with the first-flush behavior identified at other mine sites (Sullivan & Drever, 2001). In September of 2009, synoptic sampling along the Pennsylvania Mine reach of Peru Creek identified a di↵use contaminant source emanating from the wetland area (Runkel et al., 2013). Peru Creek discharge was found to increase from 55 L/s to 100 L/s over the wetland reach (Runkel et al.,2013). Sampled inflows show higher metal concentrations than Peru Creek, indicating that water discharging from the wetland is mining-impacted (Runkel et al., 2013). The constant pH and increasing sulfate concentrations over the wetland reach also indicate that the unsampled contributing water is mining impacted (Runkel et al., 2013), since unimpacted water would be expected to dilute the acidity and sulfate concentration. Concentrations of specific metals were variable over the wetland reach: in-stream concentrations of Cu, Zn, and Cd increased, while concentrations of Al, Fe, Pb decreased (Runkel et al.,2013). The mine outflow carries higher metals loads to Peru Creek (Runkel et al.,2013),but groundwater discharging to Peru Creek from the wetland is still fundamental in addressing the Pennsylvania Mine impact as a whole. The wetland has large deposits of potentially AMD generating waste rock. Using an average precipitation total of 91 cm/year, and estimating the total waste rock area as 4,600 m2, the Colorado Geological Society estimates that an annual average of 0.5 m3/hr of flow could be passing through waste rock and into groundwater each year (Wood et al., 2005). Water budget calculations from Cinnamon Gulch (Figure 2.1) show that the vast majority (>95%) of discharge to Peru Creek is from groundwater inflow (Wood et al., 2005). Furthermore, a tracer injected directly into the mine was recovered in boreholes in the wetland about 100 days after injection, indicating there is a hydrological connection with the mine (Mark Rudolph, Colorado Geological Survey, personal communication of unpublished data). Chemical analyses of groundwater sampled downgradient of the mine outflow suggest that the mine outflow is infiltrating into groundwater, implying that the mine outflow is infiltrating into the wetland area (Rudolph, 2010). TDS levels are highest in the deeper wells, indicating that the wetland connection with the mine

11 workings or other tailings piles is through the deeper fractured granite bedrock (Rudolph, 2010). Mixing and end-member analyses of metal concentrations indicate that the wetland could be receiving drainage from the Pennsylvania Mine as well as neighboring mines in Cinnamon Gulch (Runkel et al.,2013). Flow through the wetland was previously studied in evaluation of the site’s capacity to naturally attenuate redirected mine e✏uent (Emerick et al., 1988). Boreholes drilled in the wetland show three stratigraphic units: from 0-6 meters depth there are interbedded layers of clay, silt, and peat; from 6-12 meters depth there is a sand and gravel layer; below 12 meters depth there is a layer of fractured granite bedrock (Rudolph, 2010). Interpolation of the borehole logs suggests that the upper layer of clay and silt in the wetland is bowl-shaped, roughly 5 m thick in the middle and tapering out toward the edges (Emerick et al.,1988). The hydraulic conductivity of this uppermost layer was found to be highly variable, with recovery rates from bailing tests of the boreholes spanning orders of magnitude (Emerick et al., 1988). The highly variably recovery rates were attributed to anomalous 2-3 inch thick layers of ﬁne grained clayey sand encountered in multiple boreholes (Emerick et al.,1988). The upper 10 cm of the wetland soil is characterized as 41% organic, with the texture of loam or clayey loam (Emerick et al., 1988). The upper 3-4 cm of the soil is oxidized, with meter-scale surface patches of metal oxide deposits. Vegetation is dominated by water sedge and patches of bog birch (Emerick et al., 1988). The ground surface has localized 10-15 m patches of standing water up to 5 cm deep. Tailings were dumped haphazardly throughout the eastern half of the wetland, but the extent of these deposits is poorly mapped (Rudolph &Mackenzie,2009).

2.4 Material and Methods

An array of 72 electrodes with 5 meter spacing was installed east to west, through the wetland area and across the mine outlet, running roughly parallel to the creek (Figure 2.1). Data were collected on a 645 dipole-dipole quadripole sequence, which was collected in 3 replicates each ﬁeld session to better estimate measurement error. Each stored quadripole

12 measurement represents the average of a stack of 3-6 separate measurements, resulting in atotalof9-18measurementscollectedperquadripoleperfieldsession.Thedipole-dipole geometry allows for up to 10 measurements to be collected simultaneously with a 10-channel Syscal Pro resistivity meter (IRIS Instruments, Orleans, France), resulting in a total collection time of approximately 15 minutes per sequence. Initial resistivity data were collected on July 12th, 2013. Subsequent time-steps were collected at approximately 2 week intervals, until the road was inaccessible in late October, 2013. An additional dataset was collected in June of 2014. Electrodes were constructed from 75 cm X 1.27 cm schedule 40 PVC, wrapped with 8 cm of conductive foil tape approximately 10 cm from one end. Each electrode was installed to 20 cm below ground surface (bgs) and connected to the resistivity meter using 18 gauge tinned copper wire and prebuilt cables. Electrodes were left in place throughout the field campaign, including over the winter season. Contact resistance was checked at each electrode before each survey, and was typically less than 1 kohm-m in the wetland, indicating excellent electrical connection with the ground. Elevations of each electrode were recorded using a Trimble XT6000 GPS unit, and post-corrected with GPS Pathfinder Oce 2, resulting in sub-decimeter accuracy in the horizontal direction, and 10-20 cm accuracy in the vertical direction. Ancillary data that were collected to facilitate interpretation of the ER measurements include: pore fluid conductivity, temperature, and water levels in 6 pre-existing wells (identified as MW02, MW03, MW04, CGW1, GW3, and GW5 on Figure 2.1) using a Solinst water level probe. In each borehole, temperature and conductivity data were collected at water level, as well as the top, middle, and bottom of the screened interval. Borehole measurements were made synchronous to, or immediately following, ER measurements. The water level probe was rinsed with water from bailers installed in each borehole prior to collecting measurements. MW02 was screened into the deeper granite bedrock (from 14-16.8 m bgs), MW03 was screened into the sandy gravel layer (5.5-8.5 m bgs), GW5 (1.5-3 m bgs), MW04

13 (1.5-3 m bgs), and CGW1 (0.3-1.4 m bgs) were screened into the wetland clay and peat, and GW3 is screened into Peru Creek alluvium (1.5-4.6 m bgs). Peru Creek flow was gaged near the center of the resistivity array using velocimeters on Oct 11th, 2013; thereafter, flow was monitored continuously with two pressure transducers until Nov 4th, 2013. Pressure transducers were left in four monitoring wells (MW02, MW03, MW03, CGW1) over winter and spring to monitor water level, temperature, and pore fluid conductivity.

2.5 Inversion

ER measurements were inverted using the R2 research code (v2.7, Generalized 2-D In- version of Resistivity Data, (described in Binley & Kemna, 2005)). Inversions are inherently non-unique and ill-posed because model unknowns typically greatly outnumber measurements (LaBrecque et al., 1996), and hence require additional model constraints. To satisfy that requirement, R2 uses regularized optimization, which seeks to minimize both data misﬁt and model roughness (Tikhonov & Arsenin, 1977). The objective function, (), takes the form:

(m)=(W (m)[d f(m)])2 + ↵(W [m m ])2 (2.1) d m ref where

m model vector d measured resistance data Wd data weighting matrix f(m)forwardsolutionoperatingonthemodel Wm model weighting matrix that typically evalu- ates model roughness ↵ weight that controls the relative importance of the two terms on the right side of the equation mref starting model guess

14 Conceptually, the term on the far right of Equation 2.1 measures model roughness, while the next term to the left measures model misfit. Inversions require an initial guess or model starting value, mref . In time-lapse ER, mref refers to the inversion of the initial dataset. A finite element mesh was designed with 1 m x 1 m elements to 20 m depth, below which element size gradually increased, resulting in a total of 10,152 elements. Element nodes needed to be placed at each electrode location in the model, resulting in slightly non-uniform element sizes. Because of the regularization term in Equation 2.1, the resulting tomograms represent smoothed depictions of the subsurface bulk electrical resistivity. However, the degree of smoothing varies over the model space, resulting in an inversion having less resolving power in some regions than others. Resolution matrices inform on the degree of smoothing associated with a given pixel (Day-Lewis et al., 2005). The resolution matrix, R, is quantified as:

mˆ Rm (2.2) ⇡ true in whichm ˆ is the model estimate, and mtrue is the true resistivity value of the measured domain. The diagonal of R quantiﬁes the degree to which the value of a given pixel in the inversion is informed by the data corresponding to that pixel, as opposed to the smoothing inﬂuence of the regularization term. To limit interpretation of results dominated by smoothing, resistivity results corresponding to values less than Log[-2.5] in the diagonal of their resolution matrix were clipped from the analysis.

2.6 Evaluating Error

Appropriately deﬁning error is vital to achieving proper inversion ﬁt. Error estimates that are too low result in a noisy model with inversion artifacts, while error estimates that are too high result in an overly smooth model with low resolving power (LaBrecque et al.,

1996). The data weighting term in Equation 2.1, Wd, is typically of the form diag(1/✏1,...,

1/✏n), in which ✏i is the percent standard deviation associated with a stack of quadripole resistance measurements.

15 Measurement error was reported as the percent standard deviation of each stack; however, inspection of the data revealed that the measured resistances between replicate stacks were much more variable than the individual stack errors. Accordingly, the total measurement error for each quadripole was calculated as the global percent standard deviation from the three replicate stacks. Final measurement error was then either the total measurement error, or the reported precision of the Syscal Pro unit (0.2%), whichever was greater. R2 also allows for measurement error to be calculated based on a generalized error model, but this method was deemed less appropriate after inspection of the data measurement errors revealed that error values were highly variable between quadripoles, particularly when comparing the errors of the flatter wetland area, where we typically had excellent contact, and errors of the steeper upland area, where contact resistance was generally higher and where electrodes replaced before each field session. Measurement errors generally decreased over the field campaign, likely because soil settling around the electrodes promoted better electrical contact with the ground. Some quadripoles covering the far eastern part of the survey measured unreasonably large changes in resistance between time-steps, possibly due to electrode placement issues or local construction activities. To make sure that these suspect quadripoles did not negatively a↵ect the analysis, quadripoles measuring resistance values with a coecient of variation greater than 1 over the survey duration were filtered out of the analysis. Model error was assessed by comparing the apparent resistivities resulting from a forward solution on an homogeneous model with a flat surface boundary (Binley & Kemna, 2005). The average model error was 0.5%. Total error for each quadripole was taken to be the sum of measurement error and model error.

16 2.7 Results

The resistivity data (Figure 2.2) conform to lithology interpretation made by boreholes (Emerick et al., 1988; Rudolph, 2010). There is a bowl-shaped, low resistivity (<50 ohm-m) unit in the wetland with a maximum thickness of about 5 m that tapers out toward the edges of the wetland area. The resistivity of this unit is typical of clay (<100 ohm-m), which corresponds well with the interbedded clays, silts, and peat logged in boreholes MW02-04 and CGW1 (Telford & Sheri↵, 1990). There appears to be a more resistive unit extending from electrode 48 at the surface down and to the west that corresponds to the layer of sand and gravel (80-120 ohm-m) seen in MW02 and GW05 (Telford & Sheri↵,1990).Belowthe sand layer, resistivity increases to about 700 ohm-m, typical of the granite bedrock observed in the bottom of MW02, though this depth is near the resolution limit of this study. The bedrock appears to outcrop at electrode 48. There is a small surface flow here at electrode 48, likely because of the contact between sand and granite. Work planned for the summer of 2014 will confirm the existence of the granite at this location. Resolution on the east side of the profile was impacted by local construction activities and the added complication of reinstalling electrodes with each survey (Figure 2.3). Conse- quently, the east side of the profile has lower resolution, especially near the mine outflow. There are scattered high resistivity anomalies at various depths, possibly due to natural landslide or rockfall deposits or construction activities related to the emplacement of the road. The high resistivities under electrodes 68-72 likely correspond to dry granite bedrock. Time-lapse resistivity data collected over four months show the development of two resistive anomalies at approximately 5 m bgs (Figure 2.4), and the development of a more extensive resistive feature in the near-surface (<3 m) of the wetland. Note that because resistivity increases with depth at this site, any changes at depth have to be larger in magnitude to produce the same percent change; as a result, the surficial resistivity anomalies, though they represent a larger percent change, are actually of a lower magnitude than the changes at depth (Figure 2.5).

17 West East Mine out!ow E72

Large surface !ow Road Wetland Area

E48 E24 GW5 E1 MW02 CGW1

Borehole Log Key Clay and peat Sand and gravel Granite bedrock

Figure 2.2: Resistivity inversion of data collected on July 12th, 2013. Electrodes (E1-E72), model ﬁtting parameter results, borehole logs, and the general character of vegetation are shown.

Figure 2.3: Resolution of inversion of data collected on July 12th, 2013. Note, because of smoothing issues, only data for 1 m x 1 m pixels are shown.

18 Figure 2.4: Time-lapse percent changes in resistivity, relative to background inversion of 12 July 2013 data.

Figure 2.5: Time-lapse absolute change in resistivity, relative to background inversion of 12 July 2013 data.

19 2.7.1 Supporting data

Assuming that mineralogy and surface conduction remained constant over the study period, three parameters could explain the development of the resistive anomalies at depth: saturation, temperature, and pore fluid conductivity. Field observations suggested and water levels in the boreholes confirmed that the wetland stayed saturated throughout the field campaign, as the static water level was typically within 0.5 m of the ground surface. As a result, no saturation correction was necessary for the resistivity data and saturation di↵erences cannot explain the development of the resistive anomalies. Localized enhanced communication with surface water could produce imagable temperature anomalies in the subsurface, in e↵ect acting as a temperature tracer (Musgrave & Binley, 2011). To explore the possibility of the anomalies being temperature based, the resistivity response to small changes in temperature was modeled linearly (Schon, 2004):

⇢(T ) ⇢(T )= 0 (2.3) 1+(T T ) 0 where

⇢ resistivity (ohm-m) T temperature ( C) T0 initial temperature ( C) 1 constant, equal to 0.025 ( C )

Over the course of the ﬁeld campaign, temperatures in the top 2 m bgs generally increased by about 2.5 CbytheendofSeptember,beforedecreasingbyabout2.5 Cbytheend of October According to Equation 2.3, a temperature decrease of 2.5 Cwouldproduce aroughly6.8%increaseinresistivity,whichwouldnotcompletelyexplaintheresistivity anomalies. However, the development of a localized layer of ice, which is much more resistive

20 than water, at the site in October indicates that some areas had more pronounced cooling than others. A temperature decrease of 8 Cwouldproducearesistivityincreaseof25%, which is entirely within the range of the observed data. It therefore likely that the resistivity changes observed at the surface are primarily temperature driven.

However, water temperatures in the boreholes remained relatively constant (+/- 1 C) at depths greater than 1.5 m bgs, and any changes were typically increases from July to October. Even if some localized temperature decrease occurred away from the boreholes, the average starting temperatures at depth was approximately 3.5 C, indicating that a highly improbably phase change would need to occur to produce the resistive anomalies at depth. Therefore, it is unlikely that the resistivity anomalies could be completely explained by temperature changes. This leaves conductivity change as the only possible explanation for the development of resistivity anomalies. Conductivity decreases could have occurred if preferential flow paths exist in the wetland that allow for the flushing of contamination. Since the contamination is typically produced during the dry season, any additional flow through the system after the spring snowmelt pulse is likely to have lower TDS concentrations and produce a more resistive signal. There are multiple lines of evidence in the borehole data to suggest that such preferential

flow paths exist in the wetlands. There was a small (1 C) but consistent positive temperature anomaly observed in two of the boreholes (MW02, MW03) at approximately 4 m depth, which indicates localized hydrological connection with either sulfide oxidation, or upgradient surface waters. Some AMD piles have been known to reach internal temperatures of 65 , driven by the exothermic nature of Equation 1.1 (Lefebvre et al., 2001). Note that sulfide oxidation would not be favorable at such depths below the water table, so the anomalously warm water would need to have transported from upgradient. TDS initial values were more variable and changes were more localized than they were for temperature. There was little consistency between boreholes screened into the same aquifer; for example, CGW1 had TDS

21 concentrations nearly an order of magnitude higher than MW02, even though they are 100 m apart and both screened into the surface aquifer. Furthermore, the trends in TDS over the study period were not consistent between boreholes: MW04 had the largest relative variability in TDS, as it nearly doubled in TDS from 180 to just over 300 µS from the initial July 12th to the ﬁnal Oct 28th measurement. TDS in GW05 gradually decreased from about 900 µS to just over 700 µS until October 1st, before rebound back to over 800 µS. There was little TDS change in GCW1 or MW02. Petrophysical relationships allow for examination of the feasibility that TDS is controlling the trends in resistivity. Archie’s law relates pore ﬂuid conductivity to bulk conductivity (Archie, 1942; Yuval & Oldenburg, 1996): = (2.4) w a✓m where

w pore ﬂuid conductivity µS/cm bulk conductivity µS/cm a constant ✓ porosity (-) m cementation factor

To calculate the pore fluid conductivity change necessary to produce a resistivity change of 50 ohm-m, Equation 2.4 was used with a = 1.2 and m = 2, both values consistent with clay-free unconsolidated rock (Yuval & Oldenburg, 1996). The results of the calculation depend on initial resistivity, but near the center of the anomalies, where starting resistivity is approximately 130 ohm-m, an decrease in pore fluid conductivity of about 200 µS/cm would be required to produce the observed signal. This value is within the range of observed conductivities at the site (up to 1600 µS/cm), therefore TDS flushing could explain the development of resistive anomalies at depth.

22 2.8 Sensitivity Analysis

As a result of the non-uniqueness of the inverse problem, a large number of potential solutions exist that ﬁt the data suciently well. The sensitivity of the inversion to changes in the model resistivity was investigated with synthetic data to develop an understanding of how regularization impacts the inversion’s ability to resolve features of di↵erent sizes and contrasts, and to constrain the range of possible features that would be expected to produce the tomograms in Figure 2.4. Synthetic data parameters were chosen to mimic the resistive conditions of this study, such that the results inform on real-world, non-uniform settings and geometries. The below process (schematically depicted in Figure 2.6) was followed:

Sensitivity Model Flow Diagram July 12th Resistivity

Calculate anomaly Add anomalies magnitude to background

Modeled Oct 28th Resistivity anomaly Summarized region R2 forward

R2 inverse V I

Equipotential Data errors lines from July 12th

Captured anomaly + 1 % noise

Figure 2.6: Flow diagram of the sensitivity modeling process. ’Summarized region’ denotes the area over which the total resistivity anomaly is calculated.

Total resistivity change was calculated in a region encompassing the two anomalies • that developed between the background and ﬁnal datasets, respectively collected on July 12th and Oct 28th.

23 The recovered resistivity change was then split into two anomalies and added onto the • background resistivity model in locations that replicated the observed anomalies. The anomalies were emplaced at y = -5 m with circular geometries that had variable radii. Total mass remained constant and equal to the original anomaly mass, such that an areally larger anomaly would have lower contrast against the background.

A forward solution was computed for the synthetic anomalies using the same quadripole • sequence that was used to collect ﬁeld data.

The resulting forward model data were given 1% random noise to replicate the noise of • ﬁeld data. The same error parameters that were used to evaluate the ﬁeld data were assigned to the forward model data.

The inverse solution was then computed in R2 (Figure 2.7). •

In short, forward and inverse solution was computed for the observed anomalies from this study using di↵erent geometries to test the inversion’s ability to resolve anomalies of di↵erent sizes and contrasts. As was expected, results show that the model is less sensitive to smaller plume structures (Figure 2.7). The inversion was unable to detect anomalies with radii of 0.5 to 1 m at relatively shallow depths of 5 m, even though the contrasts of these anomalies are much larger. The inversions of the actual data had much more weight given to the data, and would be expected to outperform these synthetic demonstrations.

2.9 Discussion and Conclusion

Time-lapse ER techniques allowed for non-invasive location and characterization of flow paths in an AMD impacted wetland. Borehole measurements support the interpretation that preferential flow is occurring. Because of smoothing inherent in inversions, the exact geometry of the flow paths is unknown, but the sensitivity tests reveal that they must be at least several meters in diameter. Petrophysical relationships suggest that the localized

24 Figure 2.7: Sensitivity modeling results. changes in TDS within the flow paths are on the order of hundreds of µg/L, which is within the temporal variability of the borehole data. The results of this study indicate that contamination may be discharging from the wetland to Peru Creek that would have been missed by spatially localized traditional borehole sampling. Synoptic samples along Peru Creek would also fail to account for the pulse-like character of the contaminant transport. If the pore fluid changes are driven by leakage from the mine workings, there may be substantially more flow leaking from the mine workings than previously thought, which would have substantial implications for the hydrogeology of the mine workings and proposed remediation e↵orts. Furthermore, this study shows that the wetland metals contributions would be expected to change over time as the pulse of contaminant travels through the wetland. The results of this study could be a significant contribution to the development of a reactive transport model of the site. Reactive transport models are an important tool for developing quantitative predictions of remediation activities (e.g., see Benner et al.,2002; Walton-Day et al., 2012); however, collecting sucient data in the field to parameterize these

25 models in AMD impacted regions can be dicult, especially considering that heterogeneous distributions of either potentially contaminant-retarding agents or hydraulic conductivity can dramatically complicate interpretation of contaminant breakthrough curves. In particular, the presence of preferential ﬂow paths has the same e↵ect on resulting breakthrough curves as geochemical attenuating processes, which if not properly distinguished could lead to overestimates of the total amount of attenuation taking place (Malmstr¨om et al.,2008). The results of this study demonstrate the need to develop much more control on subsurface processes at legacy AMD sites.

2.10 Acknowledgments

Dr. Rob Runkel’s time was supported through the USGS Toxic Substances Hydrology Program. Site access was permitted by the U.S. Forest Service, in particular through Paul Semmer and Brian LLoyd. Dr. Alexis Navarre-Sitchler provided helpful guidance throughout the project. Many people put in long hours in the ﬁeld to make this work possible. In particular, CSM graduate students Ben Bader, Skuyler Herzog, Emmanual Padilla, and Michael Sanders were of tremendous assistance. The authors would also like to thank Dr David Benson, Dr. Katie Walton-Day, Dr. Stan Church, Mark Rudolph, and Je↵ Graves for their contributions.

26 CHAPTER 3 FUTURE WORK

This study was successful in its goal of using the natural first-flush of mining contamination to characterize subsurface flow paths. As with any research project, important questions remain. The AMD setting has many attributes that make it ideal for studying hydrological processes, such as strong annual signals and a rich history of data. There is also a significant human impact and the opportunity to intersect pure science research with water resource and engineering needs. Three possible extensions of this work are outlined below, including long-term monitoring, reactive transport imaging, and mine workings characterization.

3.1 Long-term monitoring

The most direct extension of this project would be to continue to observe the identified anomalies, seeking both to develop a better constrained quantitative model of subsurface changes, and to observe the e↵ects of interannual flow variability on transport through the site. It would have been logistically challenging to sample the subsurface over an entire survey of this extent, especially with the complexity of multiple possible contaminant sources. Better control data should be easier to obtain with a more targeted study, possible now given these new measurements of subsurface hydrology. The stream-quality consequences of flow variability in AMD generating systems can be tremendous. Comparison of two stream tracer studies conducted during the extreme drought 2002 water year (Todd et al., 2005), and the relatively wet years of 2004-2005 indicate that lower flows are more likely to have higher contributions from di↵use groundwater sources (Rudolph, 2010). Furthermore, there is evidence that climate change is increasing the sulfide weathering rate and in turn increasing in-stream zinc concentrations in the Snake River watershed (Crouch et al., 2013; Nordstrom, 2009). During this study, there was large storm in September, in which more than 12 cm of rain more than doubled the average monthly

27 precipitation. If the anomaly behavior is substantially di↵erent between the results seen in this study versus a more normal or drier water year, such observations could provide insight into the e↵ects climatic change will have on long-term stream quality.

3.2 Reactive Transport

The application of ER to imaging reactive processes is a promising new field that had previously been largely confined to laboratory settings (e.g., Regberg et al.,2011),butisnow starting to emerge as a viable field technique (Flores Orozco et al., 2011). With its strong annual signals and high total solute concentrations, AMD o↵ers an ideal setting to pursue the development of ER to characterize reactive transport. Contaminants were assumed to be conservative in this study because the focus was on the deeper subsurface, where low concentrations of nutrients typically depress reaction kinetics. However, AMD transport in the near-surface is complicated by attenuation mechanisms, including reactions of sorption, complexation, precipitation-dissolution, oxidation-reduction, and biological uptake (Gandy et al., 2007). In particular, wetlands have been identified as settings that host many AMD attenuating reactions due to their steep redox and dissolved organic matter (DOM) concentration gradients (Johnson & Hallberg, 2005). Wetlands with clay deposits can also host sorption reactions as positively charged metals adsorb to negatively charged clay particles or organic material (Sheoran & Sheoran, 2006). The resistive anomalies in the shallow subsurface in this study were interpreted to be primarily driven by temperature changes from localized surface-groundwater communication, but where there is enhanced communication with surface water there is also likely to be mixing of organic-rich surface water to provide nutrients for enhanced sulfate and metal oxide reduction. Therefore, there is a likely TDS component to the observed very near-surface signal (<2m)beyondtheresolutioncapacity of this study. ER is uniquely capable of informing on reactive transport of AMD. Subsurface residence time is a crucial parameter in predicting the extent of AMD attenuation because the kinetics of typical attenuating reactions are slow (Gandy et al., 2007). Furthermore, solubilities and

28 bioavailabilities of many mine-related solutes are sensitive to redox conditions, and many studies have demonstrated that redox conditions become more reducing along subsurface flow paths due to microbial respiration and mixing behavior (e.g., Findlay et al., 1993). Traditional stream-groundwater field studies have been unable to distinguish between mobile and less-mobile domains of subsurface solutes (Ward et al., 2010), but ER is sensitive to all subsurface domains and could be used to more accurately quantify residence time. Several modifications to the existing survey of this study would need to take place to e↵ectively capture reactive transport. The electrode spacing would need to be tightened because these reactive processes are smaller scale and are confined to the shallow near-surface. More temperature and chemistry control data should also be developed. In particular, pH control data would be crucial, because pH has been described as a master variable with regards to metals transport in streams, as it a↵ects metals speciation and precipitation/dissolution through solubility limits (McKnight & Bencala, 1990). Much of the near-surface of the wetland would lack the first-flush pulse of contaminants that acted as a natural tracer for this study, but there were several surface flows in the wetland that could be targeted. If first-flush dynamics prove dicult to capture in a stream setting near the Pennsylvania Mine because of site access issues or snow cover, natural contaminant pulses from storm events could be used instead. Small storms have been documented to increase the flow of streams near the continental divide by a factor of 2 to 3 (Campbell & Clow, 1995), and there has been well documented hysteresis storm event cycles, with TDS concentrations in the rising limb being larger than in the corresponding falling limb of the stream hydrograph. This e↵ect was attributed to flushing solutes that accumulated in the hyporheic zone, where the longer flow paths allowed for more interactions with the bu↵ering capabilities of soils (Campbell & Clow, 1995; Nagorski et al., 2003). However, the opposite e↵ect has also been observed in which the rising hydrograph has lower solute concentrations (Harvey et al.,2012). The hypothesized mechanism was storm events pushing solutes deeper into the hyporheic zone, thereby decreasing the solute concentration through dilute and by slowing the reentry

29 of solutes into the stream (Harvey et al., 2012). Again, because it provides a distributed measurement of mobile and less-mobile domains, ER provides an ideal measurement for distinguishing between these theories. Reactive transport in an AMD wetland would have a significant biogeochemical component because of the importance of microbes in catalyzing many of the AMD attenuating reactions. In particular, feedback mechanisms can be important components of transport. For example, water velocity and turbulence can a↵ect microbial community structure, which can alter hydrodynamic conditions through the precipitation of metal oxides (Bottacin-Busolin et al., 2009). Developing quantitative descriptions of these processes in AMD settings is crucial to predicting AMD transport and designing remediation schemes, and would be a significant contribution to the emerging field of biogeophysics.

3.3 Characterization of Pennsylvania Mine Leakage

At the time of this writing, a plan is underway to emplace at least one and possibly two bulkheads inside the mine. For the bulkhead to be e↵ective it needs to keep the mine workings saturated, such that no further sulfide oxidation takes place. Success of the plan is contingent on the inner mine workings being suciently watertight so as to prevent leakage around the bulkheads. All of the evidence to date, including tracer tests, groundwater chemistry analyses, and this research, suggest that contamination is leaking from the mine workings into the wetland. However, these results do not inform on the nature or location of these leaks. It would be a substantial contribution to characterize the flows exiting the mine. Un- fortunately, identifying and characterizing fracture flow remains extremely dicult for most geophysical techniques because of resolution issues inherent to settings in which the vast majority of the flow is contributed by a small fraction of the medium. In this case, the seeps and fractures through which flow discharges from the mine workings would likely be on the order of centimeters, while the adjacent mine workings are on a scale from meter to tens of meters. However, this could perhaps be overcome if, instead of imaging the frac-

30 tures directly themselves, the material immediately down gradient of the mine workings was imaged with resistivity. Any significant flows exiting the mine are likely to a↵ect localized water content levels and would therefore be represented as conductive anomalies that develop through time. Installation of the bulkhead should produce large head gradients within the mine, which could drive more flow through the fractures and make the anomalies easier to detect, but, of course, any information obtained after the bulkhead installation will be of little predictive value for the current construction e↵orts.

31 APPENDIX A - EXTENDED METHODS

A.1 Resistivity Data Analysis

Due to the construction activities, several events took place that influenced the quality of results on the eastern side of the profile. The road adjacent to the mine outflow was regraded. Electrodes 1-9 had to be manually replaced, and while an e↵ort was made to maintain electrode spacing, the regraded terrain was significantly altered such that the data from those electrodes became suspect. Further complicating the issue, the road debris was dumped over the embankment under electrode 9, substantially altering the material with which it was in contact. In the first week of October, the mine outflow was redirected through a culvert buried about 0.25 m under the road, running parallel with the survey at this location. As a result, the mine outflow began to pool around electrodes 9-11. Significant ferricrete deposits were visible at the bottom of the embankment by the end of the 2013 field season. Because of the construction, some of the measurements from the east side of the array had higher error than was reflected in the data stacks. To filter out the quadripoles that had been a↵ected by the construction activities, the coecient of variation was calculated for each quadripole over the extent of the survey. Quadripoles that had coecients of variation greater than 0.25 in the construction zone were removed from the analysis. The high resistivities of that area make such large resistive variations unlikely. As a result, most of the quadripoles involving electrodes 6-9 were removed from the analysis, resulting in a large loss of resolution near the surface in that region. Individual errors for each quadripole were used to weight the data in the inversions of this analysis, however, several studies report success fitting a general error model to the data instead (Binley & Kemna, 2005; Musgrave & Binley, 2011). The general error model approach has the advantage of smoothing out chance occurrences of high or low error on

32 relatively sparse datasets, but it is only appropriate when measurement errors are expected to be similar across the resistivity survey. The east and west sides of the survey of this study di↵ered enough in their error that individual data error was maintained. Chance deviations from the measurement error were minimized through the collection of many measures per time-step. Measurement error is often quantified from the variance between reciprocal quadripole measurements, but that not was not done in this study due to the large amount of time necessary to collect reciprocal measurements. The benefit of using the Syscal Pro unit is that 10 measurements can be collected simultaneously with the dipole-dipole configuration, but the geometry of reciprocal measurements allows for only one measurement at a time. Again, the collection of a large number of stacks is thought to make up for the lack of reciprocal measurements. R2 o↵ers the ability to do regular inversion, inversion relative to a background dataset, and di↵erence inversion, in which the di↵erence between the dataset and the background dataset are inverted. Di↵erence inversion was chosen for this analysis because inversion errors from the electrode configuration and numerical errors in the model tend to be constant through time, and thereby cancel in time-lapse analyis, allowing a fit with lower error and fewer artifacts (LaBrecque & Yang, 2001). Di↵erence inversion is also faster than standard inversions since the new model is generally similar to the background model. R2 is capable of performing singularity removal in the forward model, but only if the model surface is represented by the straight line y=0, because singularity removal is only available for analytical solutions. Because of the significant topographical variability at the site, it was decided to forgo singularity removal and instead model the site topography as recommended by the model creator (Personal communication, Andrew Binley).

A.2 Finite Element Mesh Design and Gmsh

Inversion requires the development of a ﬁnite element mesh (FEM) over which the model space can be discretized and the forward solution calculated. Several guiding principles were

33 adopted for this project:

Nodes must be placed at all electrodes. •

At least three nodes must be present between electrodes. •

Element spacing must be tighter in the shallow subsurface and in the areas of interest. •

As the study area had regions of rather abrupt topography, a triangular element FEM was tested first, because, in principle, triangular elements can handle more complex geometries than quadrilateral elements. The program Gmsh was used to generate the first FEMs (Geuzaine & Remacle, 2009). Gmsh loads a user-created input script outlining the model geometry and desired element size and outputs a FEM. A simple matlab script is then used to transform the coordinates of the FEM to an appropriate format for R2. This method requires a little more upfront time investment than the FEM imbedded into R2, but it has several significant advantages, including flexible element design and stronger mesh visualization tools. Drawbacks of using a triangular mesh are that R2 seems to have a more stringent total element count for triangular meshes (around 15,000 as compared to 30,000 for quadrilateral meshes), the output resolution matrices are less clean, and the computation is increased. Final RMS error for the inversions also increased, even with finer element spacing and the same error parameters. The disadvantages of the triangular mesh were perceived to outweigh the advantages, and so a quadrilateral mesh was used instead for this study.

34 APPENDIX B - MISCELLANEOUS DATA

More data was collected than ultimately made it into the main body of the paper. Those data are presented here for completeness, including:

Resistivity time-steps that did not contribute signiﬁcantly to the trends and so were • omitted from the ﬁnal paper. All collected resistivity data are presented in Figure B.1 and Figure B.2.

Data used to parse the resistivity signal, such as the borehole temperature and con- • ductivity measurements (shown in Figure B.3, temperature averaged by depth in Fig- ure B.4).

Sensitivity modeling results, displayed in absolute resistivity changes (Figure B.5). • At depths greater than 5 m, the inverted anomalies of the sensitivity analysis show substantial smoothing beyond the boarders of the synthetic injected anomalies. This may be because of the substantial amount of weight given to the smoothing term in the inversion, as indicated by the large alpha values (shown in Figure 2.7).

Flow conditions in Peru Creek and meteorological conditions of the surrounding area • (locations show on Figure B.6). Data were collected using pressure transducers as a part of an abandoned e↵ort to design a flow model of the site. The temperature and pressure data recorded in Peru Creek are displayed in Figure B.7, while the correcting air pressure and temperature are in Figure B.8. The stream was also gaged, and the resulting discharge measurements are recorded in Table B.1. Note that as of 18 July 2013, transducers are still in the field, and the resulting data will be incorporated into the final draft of this paper.

35 Figure B.1: All wetland inversions with ﬁtting results. All changes are relative to the background inversion of data from 12 July 2013.

Table B.1: Discharge measurements from Peru Creek.

Date Gage Height Discharge 10/28/2013 0.22 m 0.14 m3/s 11/13/2013 0.21 m 0.13 m3/s

36 Figure B.2: Resolutions of all inversions through the wetland.

37 8 A GWC1

7 MW04 MW02 MW03 6

) GW05 C (

e r u t

a 5 r e p m e T 4

2 07/01/13 08/01/13 09/01/13 10/01/13 11/01/13

1400 B

1200

1000 ) S ( 800 y t i v c u

d 600 n o C

400

200

0 07/01/13 08/01/13 09/01/13 10/01/13 11/01/13

Figure B.3: Temperature (A) and conductivity (B) measurements taken from boreholes in the wetland area.

38 8 < 1.5 m bgs 7.5 > 1.5 m bgs 7

6.5 ) C (

e 6 r u t a r 5.5 e p m e

T 5

4.5

3.5

3 07/01/13 08/01/13 09/01/13 10/01/13 11/01/13

Figure B.4: Average temperatures measured in the boreholes at shallow <1.5 m bgs., and deep depths.

Figure B.5: Sensitivity modeling results.

39 > GW3

HOBO W1 D HOBO A1 > GW5

> >> GWC1 MW02-04

D Stream gage HOBO A2 Pressure transducer

Electrode > Sample well Resistivity survey Pennsylvania Mine

Peru Creek Surface inflow Elevation contour (10 m) Wetland area 0 0.025 0.05 0.1 Kilometers ±

Figure B.6: Additional measurement locations, including pressure transducers and stilling well. HOBO W1 denotes the location of the transducer installed in Peru Creek. HOBO A1 denotes the location of the air pressure transducer from October to November. HOBO A2 denotes the air pressure transducer left at the site over winter.

40 10 A 9

7 ) C °

( 6

e r u t

a 5 r e p

m 4 e T

0 9/28/2013 0:00 10/8/2013 0:00 10/18/2013 0:00 10/28/2013 0:00 11/7/2013 0:00

22 B 21

19 ) 0 2 18 H

m c ( 17 e r u s s

e 16 r P 15

12 9/28/2013 0:00 10/8/2013 0:00 10/18/2013 0:00 10/28/2013 0:00 11/7/2013 0:00

Figure B.7: Water temperature (A) and pressure (B) measurements of HOBO W1. Pressure has been corrected for air pressure and converted to cm water.

41 20 A

10 ) C ° (

e 5 r u t a r e p 0 m e T

-5

-10

-15 9/28/2013 0:00 10/8/2013 0:00 10/18/2013 0:00 10/28/2013 0:00 11/7/2013 0:00

69.5 B

68.5 ) a P k (

r 68 u s s e r P 67.5

66.5 9/28/2013 0:00 10/8/2013 0:00 10/18/2013 0:00 10/28/2013 0:00 11/7/2013 0:00

Figure B.8: Air temperature (A) and pressure (B) measurements of HOBO A1.

42 REFERENCES CITED

Alpers, Charles N., Nordstrom, D. Kirk, Verosub, K. L., & Helm-Clark, C. 2007. Paleomeg- netic Determination of Pre-Mining Metal Flux Rates at the Iron Mountain Superfund Site, Northern California. Eos, Trans. AGU, 88(23), Suppl.

Archie, G. E. 1942. The electrical resistivity log as an aid in determining some reservoir characteristics. Transactions of the American Institute of Mining, Metallurgical, and Petroleum Engineers, 146,54–62.

Benner, S.G, Blowes, D.W, Ptacek, C.J, & Mayer, K.U. 2002. Rates of sulfate reduction and metal sulﬁde precipitation in a permeable reactive barrier. Applied Geochemistry, 17(3), 301–320.

Binley, Andrew M., & Kemna, Andreas. 2005. DC resistivity and induced polarization methods. Chap. DC Resisti, pages 129–156 of: Rubin, Y., & Hubbard, S.S (eds), Hydro- geophysics. N.Y: Springer.

Bird, David A. 2003. Characterization of anthropogenic and natural sources of acid rock drainage at the Cinnamon Gulch abandoned mine land inventory site, Summit County, Colorado. Environmental Geology, 44(8), 919–932.

Bottacin-Busolin, Andrea, Singer, Gabriel, Zaramella, Mattia, Battin, Tom J, & Marion, Andrea. 2009. E↵ects of streambed morphology and bioﬁlm growth on the transient storage of solutes. Environmental science & technology, 43(19), 7337–42.

Brusseau, ML. 1994. TRANSPORT OF REACTIVE CONTAMINANTS MEDIA IN HET- EROGENEOUS POROUS MEDIA. Reviews of Geophysics,285–313.

Caine, Jonathan Saul, & Tomusiak, Stephanie R.a. 2003. Brittle structures and their role in controlling porosity and permeability in a complex Precambrian crystalline-rock aquifer system in the Colorado Rocky Mountain Front Range. Geological Society of America Bulletin, 115(11), 1410.

Caine, Jonathan Saul, Ridley, John, & Wessel, Zachary R. 2010. Nature and varied behavior of structural inheritance in the Proterozoic basement of the eastern Colorado Mineral Belt over 1.7 billion years of earth history. Geological Society of America Field Guide, 18, 119–140.

Campbell, DH, & Clow, DW. 1995. Processes controlling the chemistry of two snowmelt- dominated streams in the Rocky Mountains. Water Resources . . . , 31(11), 2811–2821.

43 County, Summit. 2005. Peru Creek Brownﬁeld Report - Part 1. Tech. rept.

Crouch, Caitlin M., McKnight, Diane M., & Todd, Andrew S. 2013. Quantifying sources of increasing zinc from acid rock drainage in an alpine catchment under a changing hydrologic regime. Hydrological Processes, 27(5), 721–733.

Da Rosa, Carlos D, Lyon, James S, Hocker, Philip M, & Udall, Stewart L. 1997. Golden dreams, poisoned streams: how reckless mining pollutes America’s waters, and how we can stop it. Washington, DC: Mineral Policy Center.

Day-Lewis, Frederick D., Singha, Kamini, & Binley, Andrew M. 2005. Applying petrophysical models to radar travel time and electrical resistivity tomograms: Resolution-dependent limitations. Journal of Geophysical Research, 110(B8), B08206.

Emerick, JC, Huskie, WW, & Cooper, DJ. 1988. Treatment of discharge from a high elevation metal mine in the Colorado Rockies using an existing wetland. Proceedings from a conference at the annual meeting of the American Society for Surface Mining and Recla- mation,345–350.

Fey, D L, Church, S E, Unruh, D M, & Bove, D J. 2001. U.S. Geological Survey Open-File Report 02-0330 Water and Sediment Study of the Snake River Watershed, Colorado. Tech. rept.

Findlay, Stuart, Strayer, David, Goumbala, Cheikh, & Gould, Kim. 1993. Metabolism of streamwater dissolved organic carbon in the shallow hyporheic zone. Limnology and Oceanography, 38(7), 1493–1499.

Flores Orozco, Adri´an, Williams, Kenneth H., Long, Philip E., Hubbard, Susan S., & Kemna, Andreas. 2011. Using complex resistivity imaging to infer biogeochemical processes associated with bioremediation of an uranium-contaminated aquifer. Journal of Geophysical Research, 116(G3), G03001.

Gandy, C J, Smith, J W N, & Jarvis, a P. 2007. Attenuation of mining-derived pollutants in the hyporheic zone: a review. The Science of the Total Environment, 373(2-3), 435–46.

G´elinas, P, Lefebvre, R, Choquette, M, & Isabel, D. 1994. Monitoring and Modeling of Acid Mine Drainage from Wate Rock Dumps. Tech. rept. June. Report GREGI 12.

Geuzaine, Christophe, & Remacle, JF. 2009. Gmsh: a three-dimensional ﬁnite element mesh generator with built-in pre- and post-processing facilities. International Journal for Numerical Methods in Engineering, 0,1–24.

Gray, N. F. 1996. Field assessment of acid mine drainage contamination in surface and ground water. Environmental Geology, 27(4), 358–361.

44 Hallberg, K B. 2010. New perspectives in acid mine drainage microbiology. Hydrometallurgy, 104(3-4), 448–453.

Harvey, J. W., Drummond, J. D., Martin, R. L., McPhillips, L. E., Packman, a. I., Jerolmack, D. J., Stonedahl, S. H., Aubeneau, a. F., Sawyer, a. H., Larsen, L. G., & Tobias, C. R. 2012. Hydrogeomorphology of the hyporheic zone: Stream solute and ﬁne particle interactions with a dynamic streambed. Journal of Geophysical Research, 117(Oct.), 1–20.

Johnson, D Barrie, & Hallberg, Kevin B. 2005. Acid mine drainage remediation options: a review. The Science of the total environment, 338(1-2), 3–14.

LaBrecque, Douglas J, & Yang, Xianjin. 2001. Di↵erence Inversion of ERT Data: a Fast Inversion Method for 3-D In Situ Monitoring. Journal of Environmental & Engineering Geophysics, 6(2), 83–89.

LaBrecque, Douglas J., Miletto, Michela, Daily, William, Ramirez, Aberlardo, & Owen, Earle. 1996. The e↵ects of noise on Occams inversion of resistivity tomography data. Geophysics, 61(2), 538–548.

Lefebvre, R, Hockley, D, Smolensky, J, & G´elinas, P. 2001. Multiphase transfer processes in waste rock piles producing acid mine drainage 1: Conceptual model and system characterization. Journal of Contaminant Hydrology, 52(1-4), 137–64.

Loke, M.H., Chambers, J.E., Rucker, D.F., Kuras, O., & Wilkinson, P.B. 2013. Recent developments in the direct-current geoelectrical imaging method. Journal of Applied Geo- physics, 95(Mar.), 135–156.

Lovering, T.S. 1935. Geology and ore deposits of the Montezuma Quadrangle, Colorado. Tech. rept. Professional Paper 178. United States Geological Survey.

Malmström, Maria E., Berglund, Sten, & Jarsjö, Jerker. 2008. Combined e↵ects of spatially variable flow and mineralogy on the attenuation of acid mine drainage in groundwater. Applied Geochemistry, 23(6), 1419–1436.

McKnight, Diane M., & Bencala, Kenneth E. 1990. The Chemistry of Iron, Aluminum, and Dissolved Organic Material in Three Acidic, Metal-Enriched, Mountain Streams, as Controlled by Watershed and in-Stream Processes. Water Resources Research, 26(12), 3087.

Merkel, R. H. 1972. The use of resistivity techniques to delineate acid mine drainage in ground water. Ground water, 10(5).

Miller, Jerry R, & Miller, Suzanne Orbock. 2007. The water column-concentration and load. Pages 103–126 of: Contaminated Rivers. Springer Netherlands.

45 Morin, KA, & Hutt, NM. 1994. An empirical technique for predicting the chemistry of water seeping from mine-rock piles. In: Proceedings of the Third International Conference on the Abatement of Acidic Drainage.

Mulholland, PJ, & DeAngelis, DL. 2000. Surface-Subsurface Exchange and Nutrient Spi- ralling. In: Streams and Ground Waters.

Musgrave, Heather, & Binley, Andrew. 2011. Revealing the temporal dynamics of subsurface temperature in a wetland using time-lapse geophysics. Journal of Hydrology, 396(3-4), 258–266.

Nagorski, Sonia a., Moore, Johnnie N., McKinnon, Temple E., & Smith, David B. 2003. Geo- chemical response to variable streamﬂow conditions in contaminated and uncontaminated streams. Water Resources Research, 39(2), n/a–n/a.

Nordstrom, D. K. 2011a. Mine Waters: Acidic to Circmneutral. Elements, 7(6), 393–398.

Nordstrom, D. Kirk. 2009. Acid rock drainage and climate change. Journal of Geochemical Exploration, 100(2-3), 97–104.

Nordstrom, D. Kirk. 2011b. Hydrogeochemical processes governing the origin, transport and fate of major and trace elements from mine wastes and mineralized rock to surface waters. Applied Geochemistry, 26(11), 1777–1791.

Oldenburg, Douglas W., & Li, Yaoguo. 1994. Inversion of induced polarization data. Geo- physics, 59(9), 1327–1341.

Oldenburg, Douglas W., & Li, Yaoguo. 1999. Estimating depth of investigation in dc resistivity and IP surveys. Geophysics, 64(2), 403–416.

Oram, Libbie L, Strawn, Daniel G, Morra, Matthew J, & M¨oller, Gregory. 2010. Selenium biogeochemical cycling and ﬂuxes in the hyporheic zone of a mining-impacted stream. Environmental Science & Technology, 44(11), 4176–83.

Regberg, Aaron, Singha, Kamini, Tien, Ming, Picardal, Flynn, Zheng, Quanxing, Schieber, Jurgen, Roden, Eric, & Brantley, Susan L. 2011. Electrical conductivity as an indicator of iron reduction rates in abiotic and biotic systems. Water Resources Research, 47(4), 1–14.

Rucker, Dale F., Glaser, Danney R., Osborne, Tom, & Maehl, William C. 2009. Electrical Resistivity Characterization of a Reclaimed Gold Mine to Delineate Acid Rock Drainage Pathways. Mine Water and the Environment, 28(2), 146–157.

46 Rudolph, Mark. 2010. Dye injection and sampling and analysis plan Cinnamon Gulch/Penn- sylvania Mine site 2009 ﬂuorescent dye tracer study. Tech. rept. COLORADO DEPART- MENT OF PUBLIC HEALTH AND ENVIRONMENT.

Rudolph, Mark, & Mackenzie, Jean. 2009. Dye injection and sampling analysis plan Cinna- mon Gulch/Pennsylvania Mine Site 2009 Fluorescent Dye Tracer Study. Tech. rept.

Runkel, Robert L, Walton-day, Katherine, Kimball, Briant A, Verplanck, Philip L, & Nimick, David A. 2013. Estimating instream constituent loads using replicate synoptic sampling, Peru Creek, Colorado. Journal of Hydrology, 489,26–41.

Schon, JH. 2004. Physical Properties of Rocks: Fundamentals and Principles of Petrophysics. Elsevier Science & Technology Books.

Sheoran, A.S., & Sheoran, V. 2006. Heavy metal removal mechanism of acid mine drainage in wetlands: A critical review. Minerals Engineering, 19(2), 105–116.

Singha, Kamini, & Gorelick, Steven M. 2006. E↵ects of spatially variable resolution on ﬁeld-scale estimates of tracer concentration from electrical inversions using Archies law. Geophysics, 71(3), G83–G91.

Smith, LA. 1995. Hydrogeology of waste rock dumps. Tech. rept. July. British Columbia Ministry of Energy, Mines and Petroleum Resources and CANMET.

Sullivan, Annett B, & Drever, James I. 2001. Spatiotemporal variability in stream chemistry in a high-elevation catchment a↵ected by mine drainage. Journal of Hydrology, 252(1-4), 237–250.

Telford, William Murray, & Sheri↵,RobertE.1990.Applied geophysics. Vol. 1. Cambridge university press.

Tikhonov, AN, & Arsenin, VY. 1977. Solutions of ill-posed problems. Winston, Washington, DC.

Todd, AS, McKnight, DM, & Duren, SM. 2005. Water quality characteristics for the Snake River, North Fork of the Snake River, Peru Creek, and Deer Creek in Summit county, Colorado: 2001 to 2002. Tech. rept. University of Colorado, Institute of Arctic and Alpine Research.

US Forest Service. 1993. Acid Drainage from Mines on the National Forests: a Management Challenge. Tech. rept.

47 Verplanck, Philip L., Nordstrom, D. Kirk, Bove, Dana J., Plumlee, Geo↵rey S., & Runkel, Robert L. 2009. Naturally acidic surface and ground waters draining porphyry-related mineralized areas of the Southern Rocky Mountains, Colorado and New Mexico. Applied Geochemistry, 24(2), 255–267.

Walton-Day, Katherine, Runkel, Robert L., & Kimball, Briant a. 2012. Using Spatially Detailed Water-Quality Data and Solute-Transport Modeling to Support Total Maximum Daily Load Development. JAWRA Journal of the American Water Resources Association, 48(5), 949–969.

Ward, Adam S., Goose↵, Michael N., & Singha, Kamini. 2010. Characterizing hyporheic transport processes Interpretation of electrical geophysical data in coupled streamhy- porheic zone systems during solute tracer studies. Advances in Water Resources, 33(11), 1320–1330.

Ward, Adam S, Goose↵, Michael N, & Singha, Kamini. 2012. How does subsurface characterization a↵ect simulations of hyporheic exchange? Ground water, 51(1), 14–28.

Wood, Robert H II, Bird, David A, & Sares, Matthew A. 2005. Mine site history and watershed characterization of the Cinnamon Gulch Area, Dillon Ranger District, White River National Forest, Summit County, Colorado. Tech. rept. Colorado Geological Survey, Department of Natural Resources.

Younger, P L. 1997. The longevity of minewater pollution: a basis for decision-making. The Science of the total environment, 194-195(Feb.), 457–66.

Yuval, Douglas, & Oldenburg, W. 1996. DC resistivity and IP methods in acid mine drainage problems: results from the Copper Cli↵ mine tailings impoundments. Journal of Applied Geophysics, 34(3), 187–198.

Zhu, C, Anderson, GM, & Burden, DS. 2002. Natural Attenuation Reactions at a Uranium Mill Tailings Site, Western U.S.A. Groundwater, 40(1), 5–13.