Brendan T. Reilly for the Degree of Doctor of Philosophy in Ocean, Earth, and Atmospheric Sciences Presented on June 8, 2018

AN ABSTRACT OF THE DISSERTATION OF

Brendan T. Reilly for the degree of Doctor of Philosophy in Ocean, Earth, and Atmospheric Sciences presented on June 8, 2018.

Title: Deciphering Quaternary Geomagnetic, Glacial, and Depositional Histories Using Paleomagnetism in Tandem with Other Chronostratigraphic and Sedimentological Approaches

Abstract approved: ______Joseph S. Stoner

Stratigraphy and chronology are essential to sedimentological study of Earth system histories. And, stratigraphy and chronology are often challenging and interesting problems themselves. The Quaternary (2.588 Ma - present) experienced paleoenvironmental and paleo-geomagnetic variability well outside the range of the recent instrumental record, providing the opportunity to place recent observations in a more complete perspective. This dissertation presents three studies that combine paleomagnetism in concert with radiocarbon, stratigraphic correlation, and/or age-depth modeling to develop stratigraphy and assign chronology. This in turn, helps to better understand the evolution of these glacial, geomagnetic, and depositional systems. The first study investigates the glacial history of the Petermann Glacier, a major outlet glacier of the Greenland Ice Sheet, over the last ~7 ka. Petermann Glacier has been remarkably stable for as long as there have been historical observations apart from two anomalously large calving events of its floating ice tongue over the last decade. This is unique when compared with many other large marine terminating Greenland outlet glaciers. Yet, our geologic evidence clearly show the Petermann Ice Tongue was not present for much of the time recorded in the sediments of Petermann Fjord. While radiocarbon and paleomagnetic methods could not constrain the sediment’s chronology alone, due to large reservoir issues and uncertain regional paleomagnetic templates, using the two methods in tandem we determine the paleoenvironmental conditions that were required to maintain the stable ice tongue of the Late Holocene. Specifically, a stable ice tongue only formed

around 2-2.5 ka after sea ice conditions intensified, limiting Ekman transport of warm modified Atlantic Waters into the fjord, and surface air temperatures were within ~2o C of preindustrial conditions, slowing the subglacial run-off driven circulation of the fjord. The second study investigates the geomagnetic history of Western North America from ~35-15 ka. While the Holocene has been the focus of most studies about past directional changes of the geomagnetic field, the Late Pleistocene spans a greater range of field intensity variations that have a largely unknow relationship with field morphology. Yet, late Pleistocene sediments that could be used to investigate these questions, particularly from terrestrial archives of Western North America, are notoriously difficult to date. This makes comparison of millennial scale directional variations, like other studies have done for the Holocene, difficult. Using new data from Fish Lake, Utah and existing data from Bessette Creek, British Columbia, and Bear Lake on the Utah and Idaho Border, we construct a composite stacked record to define these variations and we account for radiocarbon and magnetic uncertainties in the stack’s chronology. We demonstrate that this PSV template can provide new insight to longstanding chronostratigraphic debates, such as the implications of various proposed chronologies of the sediments in the Wilson Creek Formation at Mono Lake, California on the outcrop’s chronostratigraphy and radiometric age estimates. The third study investigates the depositional history of the Bengal Fan over the last ~1.25 Ma. Regionally extensive hemipelagic deposits with good reversal magnetostratigraphy offer constraints on the evolution of the fan’s channel levee system through climate and sea-level transitions of the Pleistocene. Yet, it has been challenging to assign ages to the turbiditic sediments of the fan due to the absence of reliable chronostratigraphic markers. To address this issue, we model sediment accumulation rates at seven drill sites, incorporating all available age control points and integrating seismic observations to establish the stratigraphic relationships of paleo-channel-levee systems. The model results are stacked to create a composite regional signal for the Lower Bengal Fan, which, in the additional context of other regional archives, suggests growth of the spatial extent of the Bengal Fan channel-levee system along with increases in glacial-interglacial sea level amplitude.

Deciphering Quaternary Geomagnetic, Glacial, and Depositional Histories Using Paleomagnetism in Tandem with Other Chronostratigraphic and Sedimentological Approaches

by Brendan T. Reilly

A DISSERTATION

submitted to

Oregon State University

in partial fulfillment of the requirements for the degree of

Doctor of Philosophy

Presented June 8, 2018 Commencement June 2018

Doctor of Philosophy dissertation of Brendan T. Reilly presented on June 8, 2018

APPROVED:

______Major Professor, representing Ocean, Earth, and Atmospheric Sciences

______Dean of the College of Earth, Ocean, and Atmospheric Sciences

______Dean of the Graduate School

I understand that my dissertation will become part of the permanent collection of Oregon State University libraries. My signature below authorizes release of my dissertation to any reader upon request.

______Brendan T. Reilly, Author

ACKNOWLEDGEMENTS

I have benefited greatly from being surrounded by positive and supportive people through graduate school. Thanks for celebrating successes and helping me through failures.

My committee members Alan Mix, Anders Carlson, and Anthony Koppers taught three of my all-time favorite classes. Alan and Anders also provided the opportunity to participate in field studies of the Greenland Ice Sheet, which transformed the way I thought about past glacial activity. I am very grateful to their continued support and time. I also thank Francisca Belart for agreeing to serve as my GCR so close to my defense date.

I am grateful to Leslie and Mark Workman and the Oregon ARCS Foundation for their generous support and interest in my research.

Thanks to the ‘Quat Tea(m)’ and my other friends in the department and beyond for making it easy to love OSU and Corvallis. My adventure buddies in Greenland—Gaylen Sinclair, Anna Glueder, Jorie Clark, Chris Holm. The people I love to play music with—Scott Klasek, Andy Menky, and Suzette Savoie. My official and unofficial lab mates--Katherine Solada, Ben Frieberg, Leah Zeigler, Sarah Strano, Fenna Bergmann, Julie Velle, Torsten Haberzettl, and Doug Steen. My department collaborators who have challenged my perspectives—Mo and Paul Walczak, Andrea Balbas, Kelsey Winsor. The core lab—Maziet Cheseby and Cara Fritz—and all the people that helped me navigate CEOAS, particularly Lori Hartline. The professors that taught and mentored me. All the graduate students that have come and gone. What a community! I hope we will look for shells on warm sunny beaches, play sold out shows at Squirrel’s Tavern, and solve Earth’s mysteries for many years to come.

I learned the most while at sea and want to thank everyone involved with NBP1203, HLY1302, IODP Expedition 354, OD1507, OC1706B, and RR1718. These are experiences I will always treasure. Amy Leventer, Scott Ishman, Bruce Huber, Julia Wellner, Gene Domack, Mike McCormick, Kara Vadman, Natalie Elking, Lloyd Keigwin, Neal Driscoll, Ning Zhao, Chris

Maio, Chis Moser, Marti Jeglinski, Jenna Hill, Shannon Klotsko, Tom Cronin, Rachel Marcuson, Mark Abbott, Matt Finkenbinder, Aubrey Hillman, Laure Meynadier, Peter Selkin, Jairo Savain, Christian France-Lanord, Volkhard Spiess, Tilmann Schwenk, Adam Klauss, Mike Weber, Petra Dekens, Valier Galy, Trevor Williams, Jarrett Cruz, Lindsey Fox, JJ Bahk, Alan Baxter, Hendrik Lantzach, Yasmina Martos, Camilo Ponton, Anne Jennings, Martin Jakobsson, Liz Ceperly, Shaun Marcott, Larry Mayer, Laurence Dyke, Summer Praetorius, Jainghui Du, Heather Bervid, Coquille Rex, Masako Tominaga, Mitch Lyle, Anne Trehu, Rebecca Fowler, Ashley Long, Alexis Wright, Emily Schottenfels and many, many more—I thank you all for your mentorship, collaborations, and support. I loved learning new things with each and every one of you.

Most importantly, I have been very lucky to have incredible advisers.

Stefanie Brachfeld took me on as a graduate student when I was just starting in geology. Stefanie revealed a world of paleomagnetism, paleoceanography, and glaciomarine systems that captured my imagination. I wouldn’t be here without her.

Rob Hatfield was my second, unofficial adviser at OSU. We had a lot of adventures together and discussed a lot of fun ideas (some of which we’ve actually done!) I am lucky to have him as a mentor and a friend.

I will forever be grateful to Joe Stoner. Joe is my role model of a good scientist—he gets excited about his projects, he keeps an open mind on controversial subjects, he always supports his mentees, and he enables a balanced life for those around him. He has supported almost every idea I have had and provided more opportunities that I ever could have expected. Thanks for pushing me to think about my science more broadly and more rigorously. I am the scientist I am because of Joe.

CONTRIBUTION OF AUTHORS

Chapter 2: J. Stoner oversaw measurement and data collection for the CT, XRF, and magnetic analyses and directly supervised the project. A. Mix and M. Jakobsson oversaw the field campaign to collect the OD1507 samples and L. Mayer, K. Hogan, J. Stoner, and A. Jennings helped develop the project and select coring locations. A. Jennings, L. Dyke, M. Walczak, M. Cheseby, and J. Stoner described the cores and collected shipboard data. M. Walczak, A. Jennings, A. Mix, and S. Fallon contributed the radiocarbon materials and analysis. L. Dyke designed the maps. All authors provided input for the manuscript.

Chapter 3: J. Stoner and R. Hatfield oversaw measurement and data collection for the Fish Lake CT and magnetic analyses and supervised the project. M. Abbott, D. Larsen, S. Kuehn, A. Hillman, and M. Finkenbinder contributed the tephra and radiocarbon analysis and discussions on the stratigraphy of Fish Lake. J. Stoner, R. Hatfield, M. Abbott, D. Marchetti, and D. Larsen collected the cores from Fish Lake. C. Heil contributed and discussed the Bear Lake data. All authors provided input for the manuscript.

Chapter 4: F. Bergmann co-wrote the manuscript and provided expertise on the seismic context for the core data. F. Bergmann, T. Schwenk, V. Spiess, and M. Weber helped develop the idea and study design. J. Stoner supervised the project and provided feedback and ideas on the study design. M. Weber contributed to developing the chronostratigraphic age control and contributed text to an early version of this manuscript. C. France-Lanord, V. Spiess, and T. Schwenk oversaw the planning and implementation of IODP Expedition 354. All authors provided input for the manuscript.

TABLE OF CONTENTS Page

1. Introduction ...... 1

1.1 Forward ...... 1

1.1.1 A Brief Perspective on Magnetostratigraphy ...... 2

1.1.2 Paleosecular Variation Stratigraphy ...... 3

1.1.3 Considering the Timescales of Uncertainty in Paleomagnetic Data ...... 10

1.2 Project Objectives ...... 12

2. Past Collapse and Late Holocene Reestablishment of the Petermann Ice Tongue, Northwest Greenland ...... 15

2.1 Abstract ...... 16

2.2 Main Text ...... 16

2.3 Acknowledgements ...... 25

3. Regionally Consistent Western North America Paleomagnetic Directions from 15-35 ka: Assessing Chronology and Uncertainty with Paleosecular Variation (PSV) Stratigraphy ...... 26

3.1 Abstract ...... 27

3.2 Introduction ...... 27

3.2.1 Fish Lake, Utah ...... 29

3.3 Methods and Materials ...... 30

3.3.1 Sediment Cores and Computed Tomography (CT) Scans ...... 30

3.3.2 Sediment Magnetic Measurements ...... 32

3.3.4 Processing and Stacking Magnetic Data ...... 33

3.3.5 Additional Regional Records ...... 35

3.3.6 Age Control ...... 35

3.4 Results ...... 38

TABLE OF CONTENTS (Continued)

Page 3.4.1 Fish Lake, Utah Natural and Laboratory Remanent Magnetizations ...... 38

3.4.2 Stacking Paleomagnetic Directions for Fish Lake, Utah and Bear Lake, Utah/Idaho ...... 43

3.4.2.1 Stacking Fish Lake, Utah Cores ...... 43

3.4.2.2 Stacking Bear Lake, Utah/Idaho Cores ...... 44

3.4.3 Regional Comparison and Establishing an Integrated Regional Chronology45

3.5 Discussion ...... 50

3.5.1 Implications for Regional Chronologies and the Timing of Major Lithologic Transitions ...... 50

3.5.2 Assessing the Chronology of Mono Lake, California ...... 53

3.5.2.1 Scenario 1: The Laschamp Excursion Recorded at Mono Lake ...... 56

3.5.2.2 Scenario 2: The Excursion at Mono Lake Occurred between 30 and 34 cal ka BP ...... 58

3.6 Conclusion ...... 62

3.7 Acknowledgements ...... 63

4. Middle to Late Pleistocene Evolution of the Bengal Fan at 8o North: Integrating Core and Seismic Observations for IODP Expedition 354 Transect Age-Depth Modeling ...... 64

4.1 Abstract ...... 65

4.2 Introduction ...... 65

4.21 The Pleistocene Bengal Fan ...... 68

4.3 Materials and Methods ...... 73

4.3.1 System Specific Age-Depth Modeling ...... 73

4.3.2 Lithostratigraphic and Chronostratigraphic Inputs for Age-Depth Modeling ...... 75

TABLE OF CONTENTS (Continued)

Page 4.3.3 Stacking Model Results ...... 77

4.4 Results ...... 78

4.4.1 Lithostratigraphy ...... 78

4.4.2 Age-Depth Models ...... 81

4.5 Discussion ...... 86

4.5.1 Assessing the Ages of Regionally Extensive Brunhes-Aged Reflectors (BAR) ...... 86

4.5.2 Stacking Expedition 354 Records to Establish a Regional Signal ...... 87

4.5.3 Insights to the Middle to Late Pleistocene Evolution of the Bengal Fan ..... 90

4.6 Conclusion ...... 92

4.7 Acknowledgements ...... 92

5. Conclusions ...... 94

5.1 Chapter Summaries...... 94

5.1.1 Chapter 2: Using the Past to Better Understand Future Changes to one of Greenland’s Major Outlet Glaciers ...... 94

5.1.2 Chapter 3: Integrating Archives and Methods to Better Understand Uncertainty ...... 94

5.1.3 Chapter 4: Extracting a Regional Signal from a Complex Depositional System ...... 95

6. Appendices ...... 96

Appendix A. Supplementary Materials for Past Collapse and Late Holocene Reestablishment of the Petermann Ice Tongue, Northwest Greenland ...... 97

A.1 Materials and Methods ...... 98

A.1.1 Sediment Cores ...... 98

A.1.2 CT >2 mm Clast Index ...... 104

TABLE OF CONTENTS (Continued)

Page A.1.3 Sediment Magnetism ...... 105

A.1.4 Terrestrial Sediments ...... 107

A.1.5 Radiocarbon Dating and Age-Depth Estimation ...... 108

A.2 Supplementary Text...... 111

A.2.1 Core Recovery, Disturbance, and Fjord Stratigraphy ...... 111

A.2.2 Identifying Variations in and Signatures for Sediment Sources ...... 116

A.2.3 Conceptual Depositional Models for Petermann Fjord ...... 124

A.2.3.1 Well-Sorted Coarse Deposits ...... 125

A.2.3.2 Ice-Rafted Debris ...... 125

A.2.3.3 Suspension Settling ...... 128

A.2.4 Using IRD Gradients to Reconstruct Past Ice Tongue Extents ...... 128

A.2.5 Paleosecular Variation (PSV) and Sediment Core Chronology ...... 130

Appendix B. SedCT: MATLABTM Tools for Standardized and Quantitative Processing of Sediment Core Computed Tomography (CT) Collected Using a Medical CT Scanner 141

B.1 Abstract ...... 142

B.2 Introduction ...... 142

B.3 Description of SedCT and SedCTimage ...... 144

B.3.1 SedCT ...... 145

B.3.1.1 Importing CT Data ...... 145

B.3.1.2 Processing CT Data ...... 147

B.3.1.3 Stitching Together Cores Scanned in Multiple Intervals ...... 148

B.3.1.4 Creating Outputs ...... 149

B.3.2 SedCTimage ...... 149

B.4 Examples ...... 150

TABLE OF CONTENTS (Continued)

Page B.4.1 Extracting HU Values from Lacustrine Cores with Common Coring Imperfections ...... 151

B.4.2 Relationship Between HU and Density in Glaciomarine Cores ...... 155

B.5 Conclusions ...... 158

B.6 Acknowledgements ...... 159

Appendix C. Paleomagnetic Directions from IODP Expedition 354 Hole U1451A Cores 23H and 24H ...... 160

C.1 Abstract ...... 161

C.2 Introduction ...... 161

C.3 Methods ...... 164

C.4 Results ...... 166

C.5 Acknowledgements ...... 171

7. References ...... 172

LIST OF FIGURES

Figure Page Introduction Figure 1.1 IGRF 2015 Degree Power Versus Wavelength ...... 4 Figure 1.2 Magnetic Vectors ...... 5 Figure 1.3 Uncertainty in Declination Versus Latitude ...... 6 Figure 1.4 pfm9k.1a Northern Hemisphere Mid-Latitude PSV ...... 9 Figure 1.5 Timescales of Magnetic Lock-in Uncertainty ...... 11

Past Collapse and Late Holocene Reestablishment of the Petermann Ice Tongue, Northwest Greenland Figure 2.1 Petermann Fjord Cores and Near Surface Sediment Properties ...... 17 Figure 2.2 Ice Tongue Reconstruction ...... 20 Figure 2.3 Age-Depth Modeling of Fjord Sediments ...... 22 Figure 2.4 Paleoenvironmental Context for Ice Tongue Stability ...... 24

Regionally Consistent Western North America Paleomagnetic Directions from 15-35 ka: Assessing Chronology and Uncertainty with Paleosecular Variation (PSV) Stratigraphy Figure 3.1 Map of Radial Field Strength at Core Mantle Boundary ...... 29 Figure 3.2 Fish Lake, Utah Lithology and Composite Record ...... 31 Figure 3.3 Fish Lake, Utah Demagnetization Behavior ...... 40 Figure 3.4 Fish Lake, Utah Magnetic Results ...... 41 Figure 3.5 Deconvolving and Stacking Fish Lake, Utah Data ...... 42 Figure 3.6 Stacking Bear Lake, Utah/Idaho Data ...... 45 Figure 3.7 Development of the WNAM17 Stack and Age Scale ...... 48 Figure 3.8 Comparing the Ages of Lithologic Contacts ...... 51 Figure 3.9 PSV Tuning Exercise for Scenario 1 ...... 57 Figure 3.10 PSV Tuning Exercise for Scenario 2 ...... 59 Figure 3.11 Comparing Scenario 2 Excursion Age to Global Records ...... 61 Figure 3.12 Summary of Western North America PSV Records ...... 62

LIST OF FIGURES (Continued)

Page Middle to Late Pleistocene Evolution of the Bengal Fan at 8o North: Integrating Core and Seismic Observations for IODP Expedition 354 Transect Age-Depth Modeling Figure 4.1 Bay of Bengal Map with Core Locations ...... 67 Figure 4.2 Seismic Interpretation of 8o N Line, GeoB97-020/027 ...... 71 Figure 4.3 Stratigraphic Summary of Recovered Sediments ...... 72 Figure 4.4 Cross Plots of Physical Properties ...... 79 Figure 4.5 Comparison of Cluster Analysis to Described Lithologies ...... 80 Figure 4.6 U1452 Stratigraphic Comparison of L*, MS, and XRF Ca/Ti ...... 81 Figure 4.7 Age-Depth Modeling Results ...... 83 Figure 4.8 Sediment Accumulation Rates for Model Results ...... 85 Figure 4.9 Age Distributions for Brunhes-Aged Reflectors ...... 87 Figure 4.10 Bengal Fan 8o N Stacks and Comparison to Other Records ...... 89

Supplement to Past Collapse and Late Holocene Reestablishment of the Petermann Ice Tongue Figure 6.1 Fjord Map and Core Locations ...... 99 Figure 6.2 Outer Fjord Spliced Record ...... 100 Figure 6.3 Petermann Fjord Correlated Equivalent Depth ...... 101 Figure 6.4 Examples of Well-Sorted Coarse Deposits ...... 112 Figure 6.5 Upper 1 meter of Sediment Recovered ...... 113 Figure 6.6 Full Recovered Stratigraphy at Three Fjord Locations ...... 115 Figure 6.7 Examples of Regional Geology Observed During The Petermann 2015 Expedition ...... 116 Figure 6.8 Variations in Sediment Magnetism and Geochemistry ...... 118 Figure 6.9 Rock Magnetic Properties of 05UW Sediments ...... 119 Figure 6.10 Particle Size Specific Magnetic Susceptibility in Fjord Sediment Cores ...... 120 Figure 6.11 Particle Size Specific Magnetic Susceptibility in Terrestrial and Marine Sediments ...... 122 Figure 6.12 End-member Modeling of Particle Size Specific Magnetic Properties ...... 123 Figure 6.13 Conceptual Models for Depositional Processes in Petermann Fjord ...... 124

LIST OF FIGURES (Continued)

Page Figure 6.14 Examples of Ice Rafted Debris Observed During The Petermann 2015 Expedition ...... 127 Figure 6.15 Reconstructing Past Spatial Patterns in IRD Deposition ...... 129 Figure 6.16 Comparison of 210Pb and 14C in 38MC ...... 131 Figure 6.17 Constructing the Western Hemisphere Arctic PSV Template...... 132 Figure 6.18 Petermann Fjord Paleomagnetic Results ...... 134 Figure 6.19 Petermann Fjord PSV Stack ...... 135 Figure 6.20 Optimized ΔR Sensitivity Tests ...... 137 Figure 6.21 Constant ΔR versus variable ΔR PSV Optimize Age Model with No Magnetic Lock-in ...... 138 Figure 6.22 Constant ΔR versus variable ΔR PSV Optimize Age Model with 20 cm Magnetic Lock-in ...... 139

SedCT: MATLABTM Tools for Standardized and Quantitative Processing of Sediment Core Computed Tomography (CT) Collected Using a Medical CT Scanner Figure 6.23 Example Axial and Coronal CT Slices ...... 144 Figure 6.24 SedCT User Interface ...... 146 Figure 6.25 Pixels Sampled from Whole Round Core DICOM Files ...... 148 Figure 6.26 Example CT Values from Lake Sediment Core ...... 152 Figure 6.27 Comparison of CT Numbers Extracted Using Two Methods in Post Glacial Lake Sediments ...... 154 Figure 6.28 Comparison of CT Numbers Extracted Using Two Methods in Glacial Lake Sediments ...... 155 Figure 6.29 Comparison of GRA and CT Density Estimates ...... 158

Paleomagnetic Directions from IODP Expedition 354 Hole U1451A Cores 23H and 24H Figure 6.30 Map of the Expedition 354 Site Locations ...... 162 Figure 6.31 Core Photo Images for U1451A 23H and 24H ...... 165 Figure 6.32 Magnetic Results from U1451A 23H and 24H ...... 167 Figure 6.33 Zijderveld Plots of Samples from U1451A 23H and 24H ...... 168

LIST OF FIGURES (Continued)

Page Figure 6.34 Distributions of Paleomagnetic Directions ...... 171

LIST OF TABLES

Table Page Regionally Consistent Western North America Paleomagnetic Directions from 15-35 ka: Assessing Chronology and Uncertainty with Paleosecular Variation (PSV) Stratigraphy Table 3.1 Depth Table for Fish Lake, Utah 2014 Cores ...... 32 Table 3.2 Age Control Points ...... 37 Table 3.3 Major Tephra Glass Population Geochemical Data ...... 38 Table 3.4 Tie Points to Regional Depth Scale ...... 47

Middle to Late Pleistocene Evolution of the Bengal Fan at 8o North: Integrating Core and Seismic Observations for IODP Expedition 354 Transect Age-Depth Modeling Table 4.1 IODP Expedition 354 Sites ...... 67

Supplement to Past Collapse and Late Holocene Reestablishment of the Petermann Ice Tongue, Northwest Greenland Table 6.1 The Petermann 2015 Expedition Petermann Fjord Sediment Cores ...... 102 Table 6.2 Depth Table for 03TC-41GC-03PC Outer Fjord Splice ...... 103 Table 6.3 Depth Table for Conversion to Correlated Equivalent Depth ...... 103 Table 6.4 The Petermann 2015 Expedition Terrestrial Sediment Samples ...... 108 Table 6.5 Radiocarbon Results ...... 109 Table 6.6 Sediment Cores used in WHAP18 PSV Stack ...... 110 Table 6.7 Age-Depth Modeling Sensitivity Tests ...... 140

1. Introduction

1.1 Forward Secular variation of Earth’s magnetic field is measured as changes in the magnetic vector magnitude and direction (e.g. intensity, inclination, and declination) and reflects the dynamics of the geodynamo driven by thermochemical convection in Earth’s liquid iron- nickel outer core. While secular variation can be observed on annual to centennial timescales for which there is good data coverage from ship tracks, observatories, and satellites (e.g. Jackson et al., 2000; Finlay et al., 2010; Livermore et al., 2016), the velocities of outer core convection are about 4 orders of magnitude slower than the atmosphere— meaning these ~400 years of instrumental observation would be the equivalent of studying the convective pattern of the atmosphere for about 8 days (Bloxham et al., 1989). Luckily, geologic and archeologic materials can hold a memory of Earth’s magnetic field which allows for study of the paleomagnetic field on up to billion-year timescales. These recordings are often imperfect and data coverage is limited by where archives are available. Nevertheless, they have transformed our perception of the magnetic field from evidence for geomagnetic reversals (e.g. Matuyama, 1929; Vine and Matthews, 1963; Cande and Kent, 1995) to field intensity variations on a range of timescales (e.g. Valet and Meynadier, 1993; St-Onge et al., 2003; Tarduno, 2009; Ziegler et al., 2011; Shaar et al., 2016). On first order, Earth’s magnetic field can be approximated by a geocentric axial dipole (GAD). The GAD hypothesis, a fundamental idea in the study of paleomagnetism, states that over sufficient time (typically ~104 yrs) the time averaged field is dipolar and aligned with rotation axis of the earth (Merrill and McFadden, 2003; McElhinny, 2007). However, more detailed study of time averaged fields, using historical (Bloxham et al., 1989; Jackson et al., 2000), Holocene (archaeomagnetic, sediment, and volcanic) (Korte and Holme, 2010), and 0-5 Ma (volcanic) (Johnson and McFadden, 2007; Johnson et al., 2008) archives, suggest persistent or recurrent regions of high field intensity at middle to high latitudes at the core mantle boundary, often referred to as flux lobes, and lower intensity near the geographic poles. These structures likely result from the convective structure of the outer core and heterogeneities in the thermal and chemical structure of the lowermost mantle (Bloxham, 2002; Olson, 2016). At Earth’s surface, these features are attenuated and

2 the Holocene geologic record of field morphology has been argued to be described just as well by a simple wobbling non-axial dipole as with more complex spherical harmonic models (Nilsson et al., 2010). However, It may be these persistent or recurrent features that drive much of the change we observe in the geologic record (e.g. Nilsson et al., 2011; Stoner et al., 2013). Magnetostratigraphy of sedimentary archives, the common theme of this dissertation, leverages the reality that Earth is not a GAD and changes on a range of timescales, including greater than 10 ka reversal timescales, to multimillennial global intensity changes, to higher frequency field morphology changes.

1.1.1 A Brief Perspective on Magnetostratigraphy Since the development of a radiometrically dated reversal timescale from scattered locations around the world in the early 1960s (Cox et al., 1963; Mcdougall and Tarling, 1963), magnetostratigraphy has become an essential tool in Earth science, with some of the earliest applications assigning spreading rates at mid-ocean ridges with implications for plate tectonics (Vine and Wilson, 1965) and accumulation rates and chronology to geographically dispersed deep sea sediments with implications for the timing of paleoenvironmental changes (Opdyke et al., 1966). Since these early efforts, polarity magnetostratigraphy has been important for developing and accessing Quaternary timescales—notably providing an independent radiometrically dated framework for the timing and pacing of glacial-interglacial changes (Shackleton and Opdyke, 1973; Imbrie et al., 1984). Ultimately, Quaternary magnetic timescales were intercalibrated with orbitally tuned timescales based on Milankovitch Theory (Shackleton et al., 1990, 1995; Channell et al., 2008; Channell, Hodell, et al., 2016) which, in turn, has allowed for the assessment and discussion of orbitally and radiometrically derived timescales themselves (Tauxe et al., 1992; Channell et al., 2010; Mark et al., 2017a, 2017b; Channell and Hodell, 2017; Balbas et al., 2018). The common theme of magnetostratigraphic work, since its earliest application, is the potential to directly compare what might otherwise be considered disparate geologic archives. Following the initial development of the magnetic polarity timescale, researchers found that more subtle changes in the geomagnetic field, particularly morphology and

3 intensity changes that vary on shorter timescales than reversals, could also be used as regional and/or global stratigraphic tools. In the early 1970s, a number of studies began to document past directional changes, or paleosecular variation (PSV), recorded in higher- accumulation rate sediments and found cyclical changes that were hypothesized to reflect regular periodicities (e.g. Mackereth, 1971; Opdyke et al., 1972). It was thought that this periodicity could be used to assess and possibility calibrate the radiocarbon timescale (Stuiver, 1978); however, when a seminal study demonstrated that Holocene PSV signals were reproducible between lake basins and, when well-dated, did not have a clearly defined periodicity, strategies shifted to developing radiocarbon calibrated PSV dating curves (Thompson and Turner, 1979) and more direct comparison of geomagnetic signals. While the idea of millennial-scale cyclicity has been revisited with studies focusing on the last 8-14 ka (St-Onge et al., 2003; Nilsson et al., 2011; Lougheed et al., 2014), the direct comparison approach has led to a wealth of Quaternary PSV based magnetostratigraphic studies with creative applications, like synchronizing marine to terrestrial archives (Olafsdottir et al., 2013) and assessing radiocarbon reservoir age variation (Wündsch et al., 2016).

1.1.2 Paleosecular Variation Stratigraphy A number of studies in the last decade have highlighted the potential of developing PSV as a regional and even global stratigraphic tool. One such study illustrated that on Holocene timescales, reconstructions of the Earth’s field as a wobbling dipole with only a few Northern Hemisphere biased but high quality PSV records performed just as well, if not better, as a stratigraphic template than more complex spherical harmonic models (Nilsson et al., 2010). While all evidence suggests Earth’s magnetic field is not a simple dipole and variations are more complex than a wobbling dipole, descriptions of the field through spherical harmonic equations indicate the dipole terms are significantly more important than higher degree terms at the Earth’s surface for historical instrumental and Holocene geologic reconstructions (Jackson et al., 2000; Finlay et al., 2010; Korte et al., 2011; Nilsson et al., 2014; Constable et al., 2016). The wavelength (λ) associated with each spherical harmonic degree (l), with l = 1 being a dipolar field, can be approximated for low degrees by:

λ = 2πr/(l) (Equation 1.1)

and is depicted graphically for Earth’s surface (radius (r) = 6731 km) in Figure 1.1 by comparing the associated wavelength to the degree power used in the 2015 International Geomagnetic Reference Field (IGRF) model, illustrating the importance of long wavelength variations.

Figure 1.1. Comparison of spherical harmonic degree approximate half wavelength (i.e. 0.5λ, Equation 1.1) with associated degree power in the 2015 IGRF geomagnetic field model (Thébault et al., 2015). Greater degree power is typically associated with greater secular variation changes in these models derived from instrumental data. Distances of paleomagnetic comparisons discussed in Chapter 2, Chapter 3, and by Walczak et al. (2017) are included for comparison.

Recent studies by Stoner et al. (2013) and by Walczak et al. (2017) have demonstrated remarkable similarity between inclinations and intensity in Western North America/Northeast Pacific and declinations in the Northern North Atlantic, suggesting a common driver of the two signals which the authors link to oscillating regions of recurrent high magnetic flux under North America and Europe. We compare the length scale of this observation to paleomagnetic comparisons we discuss in Chapter 2 and Chapter 3 within the

5 framework of spherical harmonic degree wavelength and power. Smaller length scales equate to orders of magnitude lower power that could drive PSV differences between sites (Figure 1.1).

Figure 1.2. Magnetic vectors have three orthogonal components (X, Y, Z) and magnitude (B) and are generally described using inclination and declination. Left: Magnetic vectors for GAD predicted values at the latitude of Petermann Fjord (Brown), Fish Lake, Utah (Red), and the Expedition 354 Bengal Fan transect (Blue). Note the vectors are in the X-Z plane, as declination, driven by the Y vector, is always predicted to be zero by a GAD. Right: Equal angle plot for the same three magnetic vectors (filled circles). Open circles have the same inclination but have an angular distance of 5o from the closed circle, illustrating larger declination changes at higher inclinations for the same angular difference.

The convention for PSV studies is to describe the magnetic vector, with three orthogonal components (X, Y, and Z) and total magnitude (B), in terms of inclination (I) and declination (D) (Figure 1.2). Where:

푍 I = sin−1 (Equation 1.2) 퐵 and

푌 D = tan−1 (Equation 1.3) 푋

While this convention is suitable for direct comparison of proximal locations or distal sites in strategic locations (e.g. Walczak et al., 2017), inclination and declination data describe angles on a sphere relative to a pole and geometric effects may require use of the vector itself for certain correlations. For example, in Chapter 2, we investigate PSV near the geographic North Pole at ~80o N where, at GAD expected inclinations for this latitude, angular differences of 5o can translate to ~57o declination swings (Figure 1.2). This is opposed to our study site on the Bengal Fan at 8o N in Chapter 4, Appendix C, and France Lanord et al. (2016) where angular differences of 5o at GAD expected inclinations can translate to just a little greater than a 5o shift in declination. In Figure 1.3, this point is illustrated by plotting the associated declination uncertainty as a function of latitude, given a well-defined sedimentary

o paleomagnetic directions with α95 = 5 .

Figure 1.3. Declination 2σ uncertainty compared to site latitude for a well-defined paleomagnetic o direction with α95 = 5 . GAD is the Geocentric Axial Dipole predicted inclination.

As in Donadini et al (2009), 95% confidence intervals (2σ) for inclination and declination are calculated:

2σI = α95 (Equation 1.4) and

α95 2σD = (Equation 1.5) cos 퐼

At latitudes above ~70o, like the Petermann Fjord study area, uncertainty in declination can become quite large as small angular changes close to the geographic pole can translate to large declination differences. Again, this is the opposite of low latitude sites, like the Bengal Fan paleomagnetic directions discussed in Appendix C and France-Lanord et al. (2016), in which small angular changes equate to equally small declination changes. Another geometric effect considered in Chapters 2 and Chapter 3 is a correction for regional comparisons of paleomagnetic directions at distances from each other. In this correction, we assume a dipolar field, which is inaccurate but a good approximation to capture the regional field at the length scales discussed in these chapters (Figure 1.1). Using this assumption, we calculate virtual geomagnetic poles (VGP), or the associated geomagnetic pole of the paleomagnetic direction assuming a non-axial dipole field, and then recalculate paleomagnetic directions relative to that pole at a common location. This provides the same frame of reference for the PSV signal that could be distorted by latitudinal or longitudinal differences. And effectively allows for more direct comparison.

VGP latitude (θpole) and longitude (φpole) given a site co-latitude (θsite), longitude (φsite), inclination, and declination can be calculated by first solving for the paleomagnetic latitude (p):

2 p = tan−1 ( ) (Equation 1.6) tan 퐼 which can be used to find the associated VGP latitude:

−1 θpole = sin (sin 휃푠푖푡푒 cos 푝 + cos 휃푠푖푡푒 sin 푝 cos 퐷) (Equation 1.7)

and longitudinal difference (β) between the VGP and site location:

sin 푝 sin 퐷 β = sin−1 ( ) (Equation 1.8) cos 휃pole

which, if cos 푝 ≥ sin θ푠푖푡푒 sin θpole

φpole = φsite + β (Equation 1.9)

else, if cos 푝 < sin θ푠푖푡푒 sin 휃pole

o φpole = φsite + 180 – β (Equation 1.10)

To illustrate why this geographic correction is important and the stratigraphic opportunity provided by this approach, we use the pfm9k.1a spherical harmonic model for the last 9,000 years (Nilsson et al., 2014) to investigate the potential for mid-latitude Northern Hemisphere PSV stratigraphy. We calculate inclination and declination at 1551 evenly spaced locations (following the method of Deserno, 2004) between 20o and 70o N around the globe every 50 years between 5,000 and 50 years BP. These ranges were chosen as they have stronger data coverage to constrain the model than older times or other locations. To account for geometric effects on the paleomagnetic directions, each site was projected via their VGP paths to Seattle, WA (47.62o N, 122.35o W) and inclination and declination were calculated from this frame of reference (Figure 1.4). The pfm9k.1a output for Seattle show a prominent inclination feature between 3,000 and 2,000 years BP with a midpoint between lowest and highest inclination at 2,350 years BP. This feature is widely recognized in Western North American and Northeastern Pacific Data (Verosub et al., 1986; Hanna and Verosub, 1989; Hagstrum and Champion, 2002; Walczak et al., 2017). We randomly sample 120 of the projected timeseries and graphically correlate the midpoint of

9 what is interpreted to be the midpoint of the similarly high amplitude inclination of the equivalent feature. The result gives a distribution with a mean almost identical to the midpoint age of the Seattle feature and a standard deviation of 99 years. Similar tests should be performed with actual data, but we are limited in the number of very well dated and well-resolved PSV records to make this comparison. At this point, it is difficult to tell how well this analysis represents actual geomagnetic behavior due to limitations in Holocene paleomagnetic data. However, this test suggests that even if PSV is a mixture of global and regional patterns, the global component seems to be the most important and, after accounting for geometric effects, local signals only contribute to centennial scale uncertainty for this high amplitude feature in the pfm9k.1a model.

Figure 1.4. (a) 1551 evenly spaced inclination and declination predictions from the pfm9k.1a spherical harmonic model (Nilsson et al., 2014) between 20o and 70o North project to Seattle, WA via their VGP paths (colored lines) compared with predictions for Seattle, WA (black line). The mid-point of a large transition from low to high inclination occurs around 2350 years BP. (b) Distribution of correlated ages for what we consider equivalent high amplitude inclination features from 120 randomly sampled locations.

1.1.3 Considering the Timescales of Uncertainty in Paleomagnetic Data There are still uncertainties that are difficult to quantify in paleomagnetic data that complicate magnetostratigraphy studies. One of the most difficult uncertainties to constrain at this time is related to sediment magnetic acquisition processes. Understanding this uncertainty will likely be a major challenge in the next decades of research. While still poorly understood and perhaps widely variable, the sediment natural remanent magnetization (NRM) is thought to be acquired over a lock-in zone below the sediment water interface in most, if not all, depositional environments (Irving and Major, 1964; Løvlie, 1976; Verosub, 1977; Egli and Zhao, 2015). This magnetization is referred to as a post-deposition remanent magnetization (pDRM) and is unique to the depositional remanent magnetization (DRM) that is theoretically acquired around the time of deposition. The result of pDRM processes is the age of the sediment NRM is younger than the age of sediment deposition at the same horizon. The depth offset between the two can be referred to as the lock-in offset. Accordingly, when integrating magnetostratigraphy with radiometric, biostratigraphic, or climate correlation dating methods, one should not directly compare sediment and magnetization ages without considering the associated timescales of uncertainty between the two methods. Documentation of geologic estimates of lock-in offsets based on the direct comparisons of sediment and magnetic ages are still rare and biased towards deep sea sediments. Earlier efforts used the position of magnetic reversals relative to δ18O changes in deep sea sediments and found offsets of about 16-23 cm for sediments deposited at > 1 cm/ka (deMenocal et al., 1990; Channell and Guyodo, 2004). More recently, researchers have used independent estimates of magnetic field strength from the deposition of the cosmogenic nuclide 10Be and estimates of relative paleointensity (RPI) from normalizing the sediment NRM and have found offsets of about 15 cm around the last magnetic reversal (Suganuma et al., 2010) and about 2-14 cm in an almost 1.5 Ma timeseries (Simon et al., 2018). Similar estimates were also made for a deep sea environment with much higher sedimentation rates, on the order of 40-80 cm/ka, where age estimates from an independent radiocarbon age model and paleosecular variation correlation to a well- resolved Holocene regional template indicate offsets of about 15-25 cm (Stoner et al., 2013). Lock-in offsets have also been observed in independently dated varved lake

11 sediments through comparison of Late Holocene paleosecular variation to predictions from archaeomagnetic derived field models and sedimentary derived regional templates that, when only considering simple offset functions, are about 8-34 cm (Snowball et al., 2013; Mellström et al., 2015). However, some of the deeper estimates may be related to the presence of bacterial magnetosomes.

Figure 1.5. Offsets between sediment and magnetic ages as a function of sedimentation rate for lock- in offsets ranging from 2-25 cm. The associated timescales of uncertainty are defined for some of the paleomagnetic records discussed in this dissertation.

Without fully understanding the mechanisms driving these lock-in offsets or how lock-in offsets vary in different depositional environments, for magnetostratigraphic purposes we can assume that lock-in offsets are conservatively between 0 and 30 cm for most magnetic records. This range, while large can be used to characterize the timescales of

12 uncertainty associated with pDRM processes (Figure 1.5). These timescales can vary from multi-decadal to centennial for the very high accumulation rate (~150-250 cm/ka) North Iceland Shelf core, MD99-2269(Stoner et al., 2007, 2013) , used in the construction of the WHAP18 PSV reference template in Chapter 2, centennial to multi-centennial for high accumulation rate glaciomarine sediments in Petermann Fjord, multi-centennial to millennial for moderate accumulation Late Pleistocene lake sediments from Fish Lake, Utah, and multi-millennial or higher for low accumulation rate deep sea hemipelagic sediments on the Bengal Fan (France-Lanord et al., 2016). However, these range estimates can be used to assess how sensitive magnetostratigraphic integrated age-models are to lock-in offset uncertainty (as is done in Chapter 2 and Appendix A) or can be directly incorporated into age-depth modeling by perturbing the depth of age-control points in Monte Carlo simulations (as is done in Chapter 3).

1.2 Project Objectives This dissertation addresses three different geologic problems, study areas, and time intervals of the Quaternary. Each study uses a suite of sediment cores to better understand regional signals, whether they are related to the geomagnetic, glacial, or depositional histories. In each case, insight to the geologic problem is provided through a better understanding of time and stratigraphy—achieved through paleomagnetism in concert with other chronostratigraphic or sedimentological methods. Each project objective would be difficult to address with a single sedimentary archive, but the use of a network of cores helps address geologic uncertainty and provides a more complete perspective of the problem. In Chapter 2, we seek to understand the Holocene history of the Petermann Ice Tongue so that we can place modern changes in a more comprehensive perspective and explore the oceanographic and atmospheric forces the glacier is sensitive to. While the Petermann Ice Tongue has been remarkably stable in the historical record with the exception of large calving events in 2010 and 2012 (Falkner et al., 2011; Münchow et al., 2014), it was clear from our early analyses on sediment cores collected from Petermann Fjord that the ice tongue was not present for the entire duration of the Holocene, which has been well-supported with subsequent sedimentological and faunal work (Jennings et al.,

2018). This provides an opportunity to investigate the paleoenvironmental conditions both favorable and hostile to maintaining a stable ice tongue in Petermann Fjord, including investigating a number of model and/or observation driven hypotheses (Rignot and Steffen, 2008; Münchow et al., 2011, 2014, 2016; Shroyer et al., 2017; Cai et al., 2017). But, our observations suggest large and unconstrained offsets in radiocarbon ages, with the apparent radiocarbon ages significantly older than offsets due to expected marine reservoir ages when compared to independent constraints from 210Pb and paleomagnetic data. This prevents direct comparison to ice-core and lacustrine archives of atmospheric temperatures and driftwood based reconstructions of sea ice conditions—two potentially very important factors influencing the Petermann Ice Tongue. In Chapter 3, we seek to provide better regional age constraints for Late Pleistocene Western North American stratigraphy by developing a regional template for PSV from several imperfect (magnetic and geochronological) PSV records. This time and region has been notoriously difficult for establishing strong sedimentary chronologies due to a general lack of reliable materials for radiocarbon dating. Here we explore how these records can be used together to better address geologic uncertainty in both the radiocarbon dating and magnetization and provide a methodology to continue to improve this reference template as more, likely imperfect, records become available. This PSV reference template can be used to improve other lake sediment chronologies and, in future work, synchronize marine and terrestrial paleoclimate records. In Chapter 4, we build on the initial results of International Ocean Discovery Program Expedition 354 to investigate the Middle to Late Pleistocene evolution of the Bengal Fan. While on ship, magnetostratigraphy proved to be exceptionally important for providing a chronostratigraphic framework for the seven sites drilled in a transect across 8o N (France-Lanord et al., 2016). As the stratigraphy of each site was dominated by local sedimentary processes related to the position of the active channel-levee system, magnetostratigraphy proved to be the first common signal that could confidently be correlated between sites and provide a framework for interpreting other changes. In this chapter we use this framework to further explore the evolution of the fan, building on work that uses late Pleistocene sediments to demonstrate the potential for non-destructive methods in Bengal Fan stratigraphy (Weber et al., 2018) which can be applied to the Middle

Pleistocene within the context of magnetostratigraphy (Weber and Reilly, in review) and also work that reinterprets old seismic data using the new magnetostratigraphic framework (Bergmann et al., in prep.). With this new chronology, we can begin to assess if large changes in the depositional history of the fan over the Pleistocene were purely related to stochastic fan processes or if they could possibly be influenced by Pleistocene climate, tectonic, and/or sea level changes.

2. Past Collapse and Late Holocene Reestablishment of the Petermann Ice Tongue, Northwest Greenland

Brendan T. Reilly1, Joseph S. Stoner1, Alan C. Mix1, Maureen H. Walczak1, Anne Jennings2, Martin Jakobsson3, Laurence Dyke4, Kelly A. Hogan5, Larry A. Mayer6, Stewart Fallon7, Maziet Cheseby1

1 College of Earth, Ocean, and Atmospheric Sciences, Oregon State University, Corvallis, Oregon 97331, USA

2 Department of Geological Sciences and Institute or Arctic and Alpine Research, University of Colorado, Boulder, CO 80309, USA.

3 Department of Geological Sciences, Stockholm University, 106 91 Stockholm, Sweden.

4 Geological Survey of Denmark and Greenland, Department of Glaciology and Climate, Øster Voldgade 10, DK-1350, København K, Denmark.

5 British Antarctic Survey, Natural Environmental Research Council, High Cross, Madingley Road, Cambridge, CB3 0ET, UK

6 Center for Coastal and Ocean Mapping, University of New Hampshire, NH 03824, USA

7 Radiocarbon Dating Laboratory, Research School of Earth Sciences, The Australia National University, Canberra, ACT, Australia

In preparation for Science

2.1 Abstract Anomalous calving events in 2010 and 2012 of the Petermann Ice Tongue in Northwest Greenland have led many to speculate on its future stability. To place these observations in the context of a longer timeframe, we extend the historical record using the stratigraphy of Petermann Fjord sediments. Following the Early Holocene deglacial retreat of Petermann Glacier, the ice tongue only reached stable extents like those observed in the historical record during the last millennia. No ice tongue was present for nearly 5 ka during the Middle Holocene, when decadal mean regional surface air temperatures are estimated to be 0.8-2.9 oC warmer than preindustrial (1750 CE) and the Lincoln Sea had reduced seasonal sea-ice concentrations relative to the late Holocene.

2.2 Main Text The Greenland Ice Sheet (GIS) is losing mass at an accelerated rate (Rignot et al., 2011). Petermann Glacier, which drains 4% of the GIS by area (Rignot and Kanagaratnam, 2006), terminates as a ~45-50 km long floating ice tongue. Basal melting accounts for 80% of Petermann Ice Tongue’s negative mass balance, making it particularly sensitive to ice-ocean interactions (Rignot and Steffen, 2008; Münchow et al., 2014; Cai et al., 2017). Unlike other large marine terminating outlet glaciers, particularly Zachariæ Isstrøm (Mouginot et al., 2015) and Jakobshavn Isbræ (Joughin et al., 2014), Petermann Glacier has been relatively stable over the last few decades (Rignot and Kanagaratnam, 2006; Nick et al., 2012; Hogg et al., 2016). However, large calving events in 2010 and 2012 have reduced the ice tongue length from a historical range of ~70-90 km to extents shorter than previously observed, likely reflecting a departure from steady state mass balance (Münchow et al., 2014, 2016). The fate of the Petermann Ice Tongue is ultimately linked to its interactions with the warm modified Atlantic Water (AW) that fills the fjord at depth. This warmer water enters the fjord across a sill with maximum depth of 443 m (Jakobsson et al., 2018), having first circulated through the Arctic Ocean and Nares Strait (Heuzé et al., 2016). While some observations suggest that the deeper waters in Nares Strait are warming from inflowing AW (Münchow et al., 2011), recent modeling results highlight the mechanisms that could strengthen the fjord circulation and thereby increase the flux of warmer water interacting with the Petermann Ice Tongue. These include increased Ekman transport of AW into the

17 fjord resulting from decreased or more mobile sea ice in Nares Strait (Shroyer et al., 2017) and increases in surface air temperature driven subglacial run-off and increased entrainment of warm AW in buoyant meltwater (Cai et al., 2017).

Figure 2.1. (A) Petermann fjord sediment cores and historical ice tongue extents from 1959, before the large calving event in 2010, and before and after the large calving event in 2012. Landsat 8 OLI image from August 11, 2014. Data sources are listed in Figure 6.1. (B) Mean and standard deviation of the >2mm CT IRD Index in the upper 50 cm of the fjord correlated equivalent depth (ced). (C) Mean and standard deviation of a relative low (17-23 cm ced; filled) and relative high (28-36 cm ced; open) XRF Ti/Ca ratio. (D) Qualitative CT bioturbation index. (E) 2 mm thick CT slices of the uppermost recovered sediments, from top to bottom, 03TC, 10TC, 40TC, 06TC, 03UW, and 02UW. Dark blue shading represents the approximate range of historical ice tongue extents observed between 1879 and 2009. Light blue shading indicates the approximate range observed since the 2012 calving event.

In the early Holocene, near the end of the regional deglaciation from the last glacial maximum, Petermann Glacier retreated catastrophically by marine ice cliff instability from an advanced position into Petermann Fjord (Jakobsson et al., 2018). The Petermann 2015 Expedition (OD1507) recovered the first suite of sediment cores ever taken from the fjord, recording the history of the fjord’s glaciers since that retreat and providing a centennial- millennial timescale context for the historical record (Figure 2.1a, Figure 6.1, and Table 6.1). These sediment archives span a period of time in the middle Holocene when oceanographic conditions in Nares Strait were different (Jennings et al., 2011), regional surface air temperatures were warmer (Lecavalier et al., 2017; Lasher et al., 2017), and Lincoln Sea

18 seasonal sea-ice conditions were reduced (England et al., 2008; Funder et al., 2011) relative to late Holocene/preindustrial times. A transect of high quality cores were recovered from the fjord from slight bathymetric highs, including from beneath the modern ice tongue, at locations about 25, 54, 70, and 80 km from the modern grounding-line (Materials, methods, and a more detailed discussion of Petermann Fjord sediments, stratigraphy, and depositional processes are available in Section 6 Appendix A). Sedimentologic variations in a lithologic unit observed in the upper ~50 cm at all locations illustrates that the location of core sites relative to the Petermann grounding line and ice tongue has had a dominant influence on sediment stratigraphy. X-ray fluorescence (XRF) and computed tomography scans (CT) were used to track changes in bulk sediment geochemistry, >2 mm clasts, and sediment fabric across the core transect (Figures 2.1, 6.2-6.6; Appendix A). In the upper 50 cm, we observe gradients in the ratio of Ti/Ca, with Ti being enriched closer to the Petermann grounding line, and in the degree of bioturbation, with diffuse contacts and increased low-density burrow features in the outer fjord indicating enhanced bioturbation. The former observation tracks the relative contribution of carbonate and Paleozoic sedimentary rocks that comprise the local surficial geology (Dawes, Frisch, et al., 2000), relative to pink granite and other crystalline rocks, only observed locally in terrestrial glacial deposits that were excavated from deeper in the crust by the inland GIS (Figure 6.7-6.8). Our analysis of marine and terrestrial sediment samples indicates that the concentration of magnetic minerals, as tracked by particle size specific magnetic susceptibility (MS) measurements, of terrigenous sediments is a faithful tracer for the crystalline basement source (Figure 6.8-6.12) and offers the advantage of being easy to measure on small particle size fraction samples. We also observe a spatial pattern in the upper ~50 cm unit for >2 mm clast concentration, interpreted in this lithologic unit to reflect ice-rafted debris (IRD). In the sediment sampled from underneath the minimum historical ice tongue prior to 2010 (~70 km from the grounding line), there is little or no evidence for > 2 mm IRD deposition. Seaward of this, there are increased IRD concentrations. This is consistent with the idea of ice shelves acting as a ‘debris filter’ (e.g. Alley et al., 2005) and depositional models based on inferred sub-ice shelf sediments (e.g. Domack and Harris, 1998), where basal ice entrains

19 high quantities of sediment (Alley et al., 1997) which are removed by basal melting of the ice shelf near the grounding line where melt rates are the highest (Cai et al., 2017). While supraglacial and englacial expressions of sediment entrained during creation of medial and lateral moraines were observed in the Petermann Ice Tongue and its icebergs during the Petermann 2015 Expedition (Figure 6.14), these sediments would likely be smaller contributors to the flux of IRD compared to sediments entrained by basal ice (Andrews, 2000). Additionally, analysis of sediments from the left lateral ablation zone of Petermann Glacier indicates the coarse material of the lateral moraine has a low abundance of the high MS crystalline material (Figures 6.11), providing further support that MS is a reliable tracer of crystalline basement rocks entrained in basal ice. Using these fundamental sedimentologic observations, we reconstruct past extents of the Petermann Ice Tongue using the spatial distribution of >2 mm clasts on a fjord-wide correlated equivalent depth (ced) scale. The ced scale was created by correlating sediment lithology and XRF geochemistry across the fjord and our interpretation assumes that the spatial distribution of IRD in the upper most sediments reflects the historically observed 70- 90 km ice tongue extents (Figures 2.2e, 6.2-6.3, 6.15). We augment this reconstruction by measuring the MS of nine particle size fractions to trace Petermann crystalline basement source contributions to specific depositional processes (Figures 2.2c-d, 6.9-6.12). The lowermost recovered sediments in our core suite are glacial diamict, likely deposited near grounded ice after the abrupt retreat from the outer fjord sill (Jakobsson et al., 2018). Above this unit are IRD-poor laminated sediments, that we interpret to represent an extensive, but perhaps variable, ice tongue that was formed following retreat from the outer fjord sill. Based on near surface sediment gradients in bioturbation in Petermann Fjord, we interpret this sub ice-tongue unit as having been deposited while Petermann Glacier was at an advanced grounding-line position, likely the 540-610 m water depth basement-cored inner sill located about 25 km from the modern grounding line identified by geophysical surveys (Tinto et al., 2015). This inference is also supported by higher MS in the coarse silt fraction than in near surface IRD-poor sediments, suggesting coarser grain size for Petermann crystalline basement sourced material and, in turn, increased proximity to the Petermann sourced turbid melt-water layers (Figure 2.2, 6.11-6.13).

Figure 2.2. Ice tongue reconstruction on depth, documenting glacial retreat and ice tongue collapse, seasonally open marine conditions with no ice tongue in Petermann Fjord, and reestablishment and regrowth of the Petermann Ice Tongue. (A) XRF Ti/Ca in the outer fjord splice trace the relative abundance of Petermann Glacier sourced materials to bulk sediment composition (Figure 6.2, 6.7- 6.8). (B) Peterman Fjord stacked coarse material index, quantified using CT scans, for fjord cores 52 – 56 km from the modern grounding line (Figure 6.15). (C) MS of fine silt, calculated from particle size specific measurements of the outer fjord splice and a core 52 km from the modern grounding line, tracks the relative contribution of Petermann Glacier sourced material to fine sediments transported in the water column by turbid melt water plumes (Figure 6.10-6.12). (D) MS of coarse silt and sand, as in C, tracking the relative contribution of Petermann Glacier to IRD following the initial glacial retreat and ice tongue collapse. (E) Multi-decadal to centennial ice tongue extent estimates, relative to modern grounding line, from the spatial distribution of IRD in the fjord, with darker blue indicating the minimum and light blue indicating the maximum estimated ranges (Figure 6.15). (F and G) Illustrations of the Petermann Glacier when terminated with a stable ice tongue 70-90 km long as observed in the pre-2010 historical record (F) and where there was no stable ice tongue and seasonally open marine conditions in the fjord (G). Bathymetric profile is the gravity modeled east transect of Tinto et al. (2015) and ice tongue draft is after Münchow et al. (2014). Colored pins indicate coring locations used in this study’s transect. Small brown arrows indicate where deposition of high MS Petermann sourced coarse material would be deposited in each scenario.

This paleo-ice tongue collapsed around 402 cm ced and seasonally open marine conditions with no stable ice tongue persisted to around 160 cm ced, when a less extensive ice tongue reformed (Figure 2.2e). The newly established ice tongue only reached stable pre-2010 historical extents in the upper 55 cm ced. Following the paleo-ice tongue collapse, we observe a significant change in the composition of IRD, with IRD initially enriched in crystalline basement sourced sediments, indicating inclusion of basal ice in Petermann calved icebergs (Figure 2.2d). The relative contribution of crystalline basement sourced

21 sediments decays to negligible values in the coarse fraction when the new ice tongue is established, consistent with initiation of the ice tongue basal ice ‘debris filter.’ Meanwhile, the relative proportion of crystalline basement sourced sediments in the fine silt fraction increases, indicating a change in sediment flux or transport processes with the onset of the new ice tongue (Figure 2.2c). While the facies recovered below 400 cm ced were deposited during the retreat of Petermann Glacier from the outer fjord sill and at an advanced grounded ice position, the post paleo-ice tongue collapse sediments can be used to investigate the stability of Petermann Glacier and its ice tongue to paleoenvironmental conditions. Specifically, the conditions needed to prevent a stable ice tongue from forming (~160-402 cm ced) but also to maintain a ~70-90 km long ice tongue (i.e. consistent with pre-2010 extents; ~0-55 cm ced). Comparison of foraminifera and 210Pb age estimates in near surface sediments clearly indicate that a large radiocarbon age offset provides a challenge to age-depth modeling (Figure 6.16). The persistence of this age offset through the entire record is supported by comparison of well resolved paleomagnetic secular variation (PSV) directional changes with a PSV template based on high-resolution and well dated Western Hemisphere Arctic records (Stoner et al., 2013; Lund et al., 2016). A constant ΔR correction of 770 14C years, consistent with the 210Pb/radiocarbon comparison, is used in our discussion of chronology as it provides the best fit to the PSV data (Figure 2.3, 6.16-6.22). While there are uncertainties in our chronology that are difficult to quantify, sensitivity tests on the age-depth modeling support that our following discussion would not change significantly given other reasonable ΔR choices and time variations in ΔR (Table 6.7), as our regional comparisons are largely driven by a long-term decrease in northern hemisphere summer insolation over the Holocene (Figure 2.4).

Figure 2.3. Age-depth modeling of Petermann Fjord Sediments. (A) Goodness of fit (line is median, shading is 95% interval of results) between the Petermann Fjord PSV Stack and the WHAP18 Arctic PSV reference template for each of the 1,000 age-depth models generated for each ΔR choice, quantified as the mean of the cosine distance between the two timeseries where they overlap. (B) Comparison of the Petermann PSV stack inclination (red) on its median age to the WHAP18 template (black). (C) Median and 1σ uncertainty for the 1000 age-depth models using an optimized ΔR of 770 yrs. Calibrated probability distributions for ΔR of 0 yrs (gray) and 770 yrs (red) are also plotted. Additional details of the age depth modeling and sensitivity tests are presented in Figures 6.20-6.22 and Table 6.7.

We find that the Petermann Ice Tongue was not present for much of the Middle Holocene, with the collapse of the deglacial ice tongue estimated around 6.9 ka (1σ range: 6.82-6.98 ka; all ages are reported in calendar years before 1950 CE) and reestablishment of a small ice tongue around 2.18 ka (1σ range: 1.93-2.3 ka). During this time, surface air temperatures were 0.8-2.9 oC warmer than preindustrial times (defined as 1750 CE) based on the 95% interval decadal mean temperatures reconstructed from the Agassiz Ice Core between 6.9-2.18 ka (Lecavalier et al., 2017) and consistent with independently derived temperature estimates of maximum seasonal temperatures from Northwest Greenland paleolimnological observations (Lasher et al., 2017). At the same time, while shore-based studies suggest extensive land-fast sea ice in Northern Ellesmere Island and Greenland, the Clements Markham Inlet, Ellesmere Island, continued to have common driftwood deposition on its beaches until about 3.5 cal ka BP, suggesting extensive seasonally mobile sea-ice in the Lincoln Sea and likely Nares Strait (England et al., 2008; Funder et al., 2011). These observations indicate a stable Petermann Ice Tongue could not be supported when warmer air temperatures increased the flux of sub-glacial run-off and decreased sea-ice cover

23 increased Ekman transport of warm AW into Petermann Fjord, driving amplified melt rates through interaction with the AW. The reestablishment of a small Petermann Ice Tongue occurred during long-term regional cooling and following the onset of heavier sea-ice in the Lincoln Sea. A stable ice tongue with an extent similar to pre-2010 historical observations was not established until about 0.59 ka (1σ range: 0.39-0.92 ka), when surface air temperatures reached their coolest values of the Holocene. Recent atmospheric warming is amplified in the High Arctic, reversing a long insolation driven Holocene cooling trend, with modern regional temperatures warmer than those of the past 6.8 ka (Lecavalier et al., 2017). Using the Middle Holocene as an analog, our findings indicate that Arctic regional temperatures may have already past the threshold of Petermann Ice Tongue stability. While it is still uncertain if there would be a dynamic response of Petermann Glacier following ice tongue loss (e.g. Nick et al., 2013), the resulting modification of fjord circulation is estimated to increase melt rates by an order of magnitude (Cai et al., 2017) which will further increase the contribution of the GIS to future sea-level rise.

Figure 2.4. Petermann Ice Tongue history in the context of paleoenvironmental conditions. Shading indicated time intervals of deglacial retreat (gray), no ice tongue and seasonally open marine conditions in the fjord (red), reestablishment of the ice tongue (light blue), and an ice tongue with stable extents like the 1876-2010 historical record (dark blue) based on the median age of the M1 age model (Table 6.7). (A) 65o N summer insolation, illustrating the long-term Holocene reduction in northern hemisphere insolation by changes in Earth’s orbit (Laskar et al., 2004). (B) Regional surface air temperature (SAT) anomaly estimates reconstructed from the Agassiz Ice Core relative to 1750 CE (Lecavalier et al., 2017). Shading represents the 2σ confidence interval and line is the 10-year running mean. (C) Calibrated radiocarbon age distributions for driftwood deposited in the Clements Markham Inlet (CMI), Ellesmere Island, indicating seasonally reduced sea-ice conditions in the Lincoln Sea during the Middle Holocene (England et al., 2008). (D) Time intervals with evidence for human settlement in Northern Greenland of the Independence I, Independence II, and Thule cultures (Grønnow and Jensen, 2003). (E) Median age (circle) and 1σ uncertainty of the major transitions in Petermann Ice Tongue history using the M1 age model with ΔR of 770 yrs. (F) Like E, but using the M2 age model, which is a sensitivity test that explores the uncertainty in the depth the sediment remanent magnetization is acquired.

2.3 Acknowledgements We thank the Swedish Icebreaker Oden captain and crew and the Petermann 2015 Expedition scientific party; Keith Nichols, Paul Anker, Michael Brian, and Peter Washam for recovering cores beneath the ice tongue; Shaun Marcott for collecting sediment from the Petermann Glacier ablation zone; the Oregon State University Marine and Geology Repository for core archival and help sampling; Stefanie Brachfeld, Bernard Housen, and Robert Wheatcroft for generous use of their laboratories; Sam Albert for his help with XRF scanning; Jason Wiest for his help with CT scanning; and Robert Hatfield for his help with planning the particle size specific magnetic measurements. The Petermann 2015 Expedition (OD1507) and this work was funded by the National Science Foundation Office of Polar Programs (Awards 1418053 to AM and JS, 1417787 to LM, and 1417784 to AJ), the Swedish Polar Research Secretariat, and a Swedish Research Council (VR) grant to MJ. Additional support to BR came from the Oregon ARCS Foundation and a Geological Society of America graduate student research grant.

3. Regionally Consistent Western North America Paleomagnetic Directions from 15-35 ka: Assessing Chronology and Uncertainty with Paleosecular Variation (PSV) Stratigraphy

Brendan T. Reilly1, Joseph S. Stoner1, Robert G. Hatfield1, Mark B. Abbott2, David W. Marchetti3, Darren J. Larsen4, Matthew S. Finkenbinder5, Aubrey L. Hillman6, Stephen C. Kuehn7, Clifford W. Heil Jr.8

1 College of Earth, Ocean, and Atmospheric Sciences, Oregon State University, Corvallis, Oregon 97331, USA

2 Geology and Environmental Science, University of Pittsburgh, PA, 15260, USA

3 Geology Program, Western State Colorado University, Gunnison, CO, 81230, USA

4 Department of Geology, Occidental College, Los Angeles, CA, 90041 USA

5 Environmental Engineering and Earth Sciences, Wilkes University, Wilkes-Barre, PA 18766, USA

6 Department of Environmental Sciences, School of Geosciences, University of Louisiana at Lafayette, Lafayette, LA 70504, USA

7 Department of Physical Science, Concord University, Athens, WV, 24712, USA

8 Graduate School of Oceanography, University of Rhode Island, 215 South Ferry Road, Narragansett, RI 02882, USA

In preparation for Quaternary Science Reviews

3.1 Abstract Late Pleistocene Western North American lacustrine archives are notoriously difficult to date. When materials for radiocarbon dating are present, it can be difficult to assess if terrestrial materials are reworked or the degree to which lake reservoir effects may impact lacustrine materials. In some documented cases lake carbonate materials can be further complicated by the incorporation of modern carbon, yielding anomalous ages. Here we illustrate the potential of Paleomagnetic Secular Variation (PSV) stratigraphy, using past directional changes of Earth’s Magnetic Field, to assess these chronologies. Unlike relative paleointenisty (RPI) stratigraphy that utilizes variations in past intensity changes of Earth’s Magnetic Field to assign chronology, directional PSV stratigraphy is less impacted by the drastic changes in sediment lithology that are often found in Late Pleistocene lake sediments. We present new PSV data from Fish Lake, Utah, USA, which are used along with previously published regional records to build an independently dated Western North America (WNAM17) PSV stack from about 35 to 15 ka that quantifies dating and paleomagnetic uncertainties. We then discuss how this PSV template can be used to assess the timing of paleoenvironmental and paleomagnetic events in Western North America.

3.2 Introduction Well-dated and high resolution sedimentary records of geomagnetic paleosecular variation (PSV) have transformed our knowledge of geomagnetic field behavior on millennial timescales and provide a stratigraphic method independent of climate correlation for addressing a wide range of problems, including improvement of paleoceanographic and paleoclimate chronologies (Stoner et al., 2007; Darby et al., 2012), direct marine and terrestrial comparisons (Olafsdottir et al., 2013), and assessment of reservoir age changes through time (Wündsch et al., 2016). PSV describes the general spatial and temporal variability of the Earth’s magnetic field and typically is used in reference to directional variations during periods of stable polarity (Johnson and McFadden, 2007; Lund, 2007). While a number of stacks have been developed to define regional templates for PSV in the Holocene (Thompson and Turner, 1979; Snowball et al., 2007; Barletta et al., 2010; Zheng et al., 2014; Walczak et al., 2017), Late Pleistocene paleomagnetic stacks have mostly focused on regional or global relative paleointensity (RPI) changes (Stoner et al., 2002; Laj et al.,

2004; Nowaczyk et al., 2013). This is likely due in part to difficulty in establishing continuous, well-dated, and high-resolution records in Late Pleistocene sediments that resolve the amplitude and timing of the centennial to millennial directional changes characteristic of PSV. The Late Pleistocene experienced drastic climate variability often leading to large changes in paleoenvironmental and depositional processes in sedimentary archives. In many cases, this results in variations of magnetic remanence acquisition efficiency across lithologic transitions and complicates continuous reconstructions of RPI (e.g. Tauxe, 1993; Schwartz et al., 1996; Mazaud, 2006). Paleomagnetic directions are often less impacted by these lithologic variations and can be directly compared between archives without making scaling factor assumptions. This provides the opportunity to reconstruct past field morphology changes using approaches that integrate discontinuous PSV records and chronologies, such as efforts to extend the Scandinavian PSV reference curve through the deglacial interval (Lougheed et al., 2014). PSV has been central to a number of Western North America Late Pleistocene studies, and while many PSV records appear to have regionally consistent features (Negrini et al., 1984; Liddicoat, 1992; Benson, Lund, et al., 2003), the lack of strong independent chronologies has led researchers to use far field correlations to well-dated Western North Atlantic records (Benson et al., 1998, 2011; Lund et al., 2017) or rely on regional correlations to records with poorly constrained chronologies (Clague et al., 2003). Here we present new PSV data from Fish Lake, Utah (UT), which, when combined with previously published PSV data from Bear Lake, Utah/Idaho (Heil et al., 2009), and Bessette Creek, British Columbia (Turner et al., 1982), offers a step forward in building a well-defined and independently dated PSV reference curve that considers chronologic and magnetic uncertainties from ~35-15 ka (Figure 3.1). These sites are positioned on the southern margin of a region of high radial magnetic flux at the core-mantle boundary, or ‘flux lobe’, as recognized in the gufm1 historical geomagnetic field reconstruction (Jackson et al., 2000). This high flux region is likely persistent or recurrent on longer timescales (Bloxham, 2002). Evidence from independently dated Holocene records from Western North America suggests that the timing of high amplitude PSV features, particularly inclination, are largely consistent across the mid-latitude Western United States (Hanna and Verosub, 1988,

1989; Lund, 1996), Northeast Pacific (Walczak et al., 2017), and the Western Canadian Arctic (Barletta et al., 2008; St-Onge and Stoner, 2011). Global comparisons of these Holocene directional records suggest high amplitude features that occur on roughly millennial timescales are a reflection of large-scale geomagnetic field dynamics related to the field intensity over North America relative to other Northern Hemisphere flux lobes (Nilsson et al., 2010, 2011; Stoner et al., 2013; Walczak et al., 2017). These regular changes provide a stratigraphic context for both paleoenvironmental and paleomagnetic events recognized in Western North American sediments that can be integrated with the radiocarbon timescale to provide a different perspective than is possible with radiometric methods alone.

Figure 3.1. Radial Field Strength at the Core Mantle Boundary (CMB) from 1590-1990 time averaged field, based on the historical reconstruction GUFM1 (Jackson et al., 2000), with locations of mid- latitude western North American PSV sites discussed in the text, including Fish Lake, Utah, a British Columbia outcrop at Bessette Creek (Turner et al., 1982), Bear Lake on the Idaho/Utah border (Heil et al., 2009), and Mono Lake in California (Lund et al., 1988).

3.2.1 Fish Lake, Utah Fish Lake, Sevier County, Utah (38.54o N, 111.71o W; elevation 2,700 m), a common name for North American lakes including the important Holocene paleomagnetic site, Fish Lake, Oregon (Verosub et al., 1986), is the largest natural mountain lake in the state of Utah, USA. The lake basin is located in a northeast-southwest trending graben in the high plateau

30 of Utah, a transitional region between the Colorado Plateau and the Basin and Range (Bailey et al., 2007). Glacial geology and lake bathymetry suggest glaciers drained from the Fish Lake High-top (elevation 3,546) to the Fish Lake basin without overrunning the lake during the last and penultimate glaciations (Marchetti et al., 2011), providing the opportunity to recover a lacustrine record of Late Pleistocene regional glacial activity, paleoenvironment, and paleomagnetism. 3He cosmogenic exposure ages of moraine boulders suggest local last glacial maximum ice extent occurred around 21 ka with a possible readvance or standstill at ~15-18 ka (Marchetti et al., 2011). The lake dimensions are about 5 km by 1.7 km with a maximum water depth of about 37 m.

3.3 Methods and Materials

3.3.1 Sediment Cores and Computed Tomography (CT) Scans Twelve sediment cores, taken as individual ~1.8-2.0 m drives, were recovered using a 9 cm diameter UWITEC system from the frozen surface of Fish Lake, UT in February 2014. Three holes were cored, with two drives recovered from Hole A14, and five drives recovered from Holes B14 and C14 (Figure 3.2). A MS3 Bartington magnetic susceptibility meter with 100 mm diameter MS2C loop sensor was used to measure magnetic susceptibility following recovery to ensure overlap and inform coring decisions. Before being split, all 12 sediment cores were scanned on the Oregon State University (OSU) College of Veterinary Medicine Toshiba Aquillon 64 Slice medical computed tomography (CT) scanner at 120 kV. A “sharp” algorithm was used to produce 2 mm thick coronal slices, which have an effective resolution of 0.5 x 0.5 mm within the plane of each slice. SedCT MATLAB tools were used to extract downcore CT number profiles, which largely reflect changes in sediment density, and slice images (Reilly et al., 2017). A composite record splice was developed for the CT scans, based on the meters composite depth (mcd) scale developed by correlation of the CT scan number profiles and u-channel magnetic susceptibility (discussed below) and identification of tie points that correlate at the sub-cm scale in the CT number profiles (Table 3.1). Cores from Holes B14 and C14 were split at the OSU Marine and Geology Repository. Three principle lithologies were observed; a low density brown organic mud in

31 the upper 6.35 mcd, a massive gray silty clay from 6.35 mcd to about 9.34 mcd, and a gray mud with faint black laminations below ~9.34 mcd. These lithologic units correspond to major density changes, as observed in the CT scans, and are associated with large changes in the concentration of ferrimagnetic minerals, as reflected by measurements of magnetic susceptibility (Figure 3.2).

Figure 3.2. Simplified lithologic log, CT scan slices (3.5x horizontal exaggeration), CT numbers, and magnetic susceptibility for Fish Lake, Utah cores recovered during the 2014 field season on the meters composite depth scale (Table 3.1). The simplified lithologic changes (from top to bottom) of brown organic-rich mud, to massive gray silty clay, to gray silty clay with faint black laminations. Magnetic susceptibility data are offset by 10-4 SI, as some values measured in the upper 6.5 mcd are negative, and include u-channel (dark continuous line) and field whole round (light dashed line) measurements. The field whole round data are arbitrarily scaled by 40% to match the u-channel data. A composite record is made by splicing the CT scans together (Table 3.1) and stacking all u-channel magnetic susceptibility data on the mcd scale. Shading for the composite record indicates the hole used for each interval in the CT splice (Green = A14; Blue = B14; Red = C14). Note CT images have non-linear scaling to account for large density differences between lithologic units.

Table 3.1. Depth Table for 2014 Field Season Fish Lake, UT UWITEC Cores

Difference End U-Channel mcd Bottom Cap/Sediment Start in Offset Section # Top Section Section Hole Drive Start (m) Section (m)* (m)* in Splice Depth (m)* Depth (m)* 1 0.03 - 0.41 1 0 0.80 A14 2 0.04 - 2.33 - - - 1 0.04 0.00 0.46 2 0.75 1.25 2 0.02 0.15 2.41 4 0.62 1.70 B14 3 0.05 0.08 4.30 6 0.26 1.44 4 0.02 0.02 6.35 8 0.06 1.24 5 0.02 0.34 8.25 10 0.12 1.76 1 0.03 0.00 1.51 3 0.20 1.52 2 0.03 0.13 3.47 5 0.64 1.09 C14 3 0.01 0.00 5.47 7 0.27 0.94 4 0.01 0.06 7.45 9 0.15 0.92 5 0.02 0.18 9.44 11 0.55 1.79 * Measured from start of sediment in core liner

3.3.2 Sediment Magnetic Measurements U-channel samples, ~2 cm x 2 cm x up to 150 cm u-shaped plastic containers with a plastic lid, were sampled from the center of the working halves of Fish Lake, UT cores from Holes B14 and C14. As the UWITEC cores were commonly longer than 150 cm, u-channels were preferentially sampled away from the tops and bottoms of core sections to avoid potentially compressed or disturbed sediments and to ensure a complete sampling of the entire stratigraphy based on the field magnetic susceptibility correlations. The sediment natural remanent magnetization (NRM) and anhysteretic remanent magnetization (ARM) were measured at the OSU Paleo and Environmental Magnetism Lab on a 2G EnterprisesTM model 755-1.65UC superconducting rock magnetometer (SRM) with inline alternating field (AF) coils optimized for u-channel samples. Measurements were made every 1 cm with a 10 cm leader and trailer; however, the effective resolution is the integrated remanent magnetization within the 7.6 cm full width at half maximum (FWHM) response function of the magnetometer (see Oda and Xuan, 2014 for detailed discussion of the OSU system). In this study SRM data from the leader, trailer, and 5 cm at either end of each u-channel are not reported in the results but are necessary for the deconvolution experiments discussed below. The NRM of Core B14-D1 was measured before and after peak AF demagnetization every 5 mT from 10 to 70 mT and at 80 mT (15 measurements). Due to this sample’s weak magnetization and demagnetization behavior, the NRM of the other nine u-channels were measured every 2.5 mT from 10 to 40 mT, every 5 mT from 45 to 60 mT and every 10 mT

33 from 70 to 80 mT (20 measurements). Following demagnetization of the NRM, an ARM was applied using a 100 mT peak AF and 0.05 mT biasing field. The ARM was measured before and after peak AF demagnetization every 5 mT from 10 to 70 mT and at 80 mT (15 measurements). U-channel magnetic susceptibility was measured every 1 cm on a motion-controlled u-channel track built at OSU with a Bartington MS3 meter attached to an MS2C loop sensor with an internal diameter of 36 mm. Reported values are the mean of three repeat measurements. A magnetic susceptibility stack was created from u-channel data on their composite depth scale, after trimming the top and bottom 3 cm to account for edge-effects of the Bartington Loop sensor. The stack was created at 1 cm composite depth intervals using a weighted running mean with weighting based on a narrow Gaussian function filter with a FWHM of 1 cm.

3.3.4 Processing and Stacking Magnetic Data Flux jumps in the SRM data were monitored and corrected using UPmag MATLAB tools and Characteristic Remanent Magnetization (ChRM) directions, inclination and declination, were calculated using the PCA method (Kirschvink, 1980; Xuan and Channell, 2009). For intervals with very well defined ChRMs (B14 Drives 4 and 5 and C14 Drives 4 and 5), deconvolution experiments were performed with UDecon MATLAB Tools (Oda and Xuan, 2014; Xuan and Oda, 2015). After removing sections based on quality criteria (discussed in RESULTS) and rotating declinations based on agreement between overlapping drives, paleomagnetic directions were stacked every 1 cm by calculating a weighted running vector mean with weighting defined by a 5 cm FWHM Gaussian function filter after applying the error propagation technique described below. For the deconvolved Fish Lake, UT data, a 3 cm FWHM Gaussian function filter was used. Error related to measurement precision was propagated by scaling maximum angular deviation (MAD) values to 95% confidence intervals (after Khokhlov and Hulot, 2016) and selecting 1,000 random values from a normal distribution with standard deviations (σ) for inclination and declination calculated from the

95% confidence interval (after Donadini et al., 2009). At each stack depth (Ds), each set of

1,000 random values for each paleomagnetic measurement whose composite depth (Dcd)

34 were within ±5 σ (σ is the standard deviation of the Gaussian function and is equal to about 42% of the FWHM width) were binned and weighted using the Gaussian distribution function:

퐷 2 푔(퐷 ) = exp (− 푖 ), where 퐷 = 퐷 − 퐷 (Equation 3.1) 푖 2σ2 푖 푠 푐푑

The vector mean and Fisher statistics (Fisher, 1953) were calculated for the binned data at each stack depth, using the weighted vector magnitude (R) calculated by:

√ 푛 2 푛 2 푛 2 푅 = (∑푖=1 푔(퐷푖)푥푖) + (∑푖=1 푔(퐷푖)푦푖) + (∑푖=1 푔(퐷푖)푧푖) (Equation 3.2)

Where n is equal to the number of data in each bin and xi, yi, zi, are the east-west, north-south, and vertical components, respectively, of the ith binned value. The weighted mean component directions are calculated by:

∑푛 𝑔(퐷 )푥 푥̅ = 푖=1 푖 푖 (Equation 3.3) 푅

∑푛 𝑔(퐷 )푦 푦̅ = 푖=1 푖 푖 (Equation 3.4) 푅

∑푛 𝑔(퐷 )푧 푧̅ = 푖=1 푖 푖 (Equation 3.5) 푅

Which allows the mean inclination and declination to be calculated by the standard method. The Fisher statistic precision parameter (κ) and α95 are approximated by:

(푁−1) κ = , where 푁 = ∑푛 푔(퐷 ) (Equation 3.6) (푁−푅) 푖=1 푖

140 α95 = (Equation 3.7) √κC

Where C is equal to the number of cores that contribute to the bin and have a sum Gaussian weight for their measurements greater than one. We choose to normalize by the number of cores rather than the number of measurements used because the u-channel measurements are an integration of the magnetometer response function and not independent measurements (as discussed above).

3.3.5 Additional Regional Records For our regional comparison, we use paleomagnetic data from Bear Lake on the Utah and Idaho border from Heil et al. (2009) and the Bessette Creek Outcrop, British Columbia, originally studied by Oberg and Evans (1977) and later studied in detail by Turner et al. (1982). Sampling and paleomagnetic methods for these locations can be found in their original studies. While we use the complete Bessette Creek record, we focus only on the 9.75 to 25 meters interval in Bear Lake, due to the excellent agreement between inclination in the BL00-1D and 1E Holes in this interval documented by Heil et al. (2009) and more ambiguous relationship below that.

3.3.6 Age Control Terrestrial macrofossil material, including charcoal, wood, and a seed, were used for radiocarbon dating the Fish Lake, UT Sediment Cores (Table 3.2). Bulk sediment samples were disaggregated with 7% H2O2 and sieved using a 125 μm stainless steel screen. The remaining material on the sieve was examined and macrofossil material isolated using a fine detail paint brush. Macrofossils were pretreated using a standard acid-base-acid pretreatment (Abbott and Stafford, 1996) before being combusted and reduced to graphite for AMS radiocarbon measurements at the Keck Carbon Cycle Accelerator Mass Spectrometer Laboratory at the University of California, Irvine. All radiocarbon samples were calibrated using MatCal (Lougheed and Obrochta, 2016) and the IntCal13 calibration curve (Reimer et al., 2013). Five sediment samples were prepared for tephra analysis. Each sample was wet sieved, separated at a density of 2.5 g/cm3 using a solution of lithium heteropolytungstates, mounted in epoxy, polished to a 0.3 μm final grit, and carbon coated. Following Kuehn

(2016), Geochemical analyses for 11 major and minor elements (SiO2, TiO2, Al2O3, FeO, MnO,

CaO, Na2O, K2O, P2O5, and Cl) were performed at Concord University on an ARL-SEMQ electron microprobe automated by Probe for EPMA software using one large-area energy- dispersive spectrometer (EDS) for Si and Al and four wavelength-dispersive spectrometers (WDS) for all other elements. Analytical conditions were 14 kV accelerating voltage, 10 nA beam current, and 4 to 10 micron diameter spot. A time-dependent intensity correction was applied as needed for Na, and water-by-difference was incorporated into the X-ray matrix corrections. Mean atomic number modeled X-ray backgrounds were used for all WDS elements with a blank correction to enhance minor/trace level accuracy. Lipari ID3506 (rhyolite), BHVO-2g (basalt), and NKT-1g (nephelinite) reference glasses were included in each analytical session (Kuehn et al., 2011). Following analysis, representative backscatter electron images were collected for the tephra grains. Three of the five tephra were identified in Fish Lake, UT and matched to the Tabernacle Hill, Pavant Butte, and Pony Express tephra layers as identified in the Lake Bonneville Basin (Oviatt and Nash, 1989, 2014) based on major and minor trace metal chemistry and stratigraphic relationships (Table 3.2 and 3.3). Each of the samples contains a major population along with tephra grains of multiple other compositions, suggesting a relatively high background level of reworked tephra grains in the lake sediments. Samples B14-D5 91.5 cm (CU1344) is an unequivocal match to the approximately ~20 14C ka basaltic Pony Express tephra described by Oviatt and Nash (2014). Samples B14-D4 61 cm (CU1342) and B14-D4 91.5 cm (CU1343) contain compositionally indistinguishable basaltic glass compositions with compare closely to the ~15.6 14C ka Pavant Butte and ~14.4 14C ka Tabernacle Hill tephras described by Oviatt and Nash (1989), which we assign respectively based on stratigraphic order. Sample B14-D4 23 cm (CU1341) contains a rhyolitic major population of unidentified origin. Traces of glass of similar composition are present in all of the other samples, suggesting that this may represent reworked material. Sample C14-D3 116.5 (CU1345) contains a lower abundance of tephra grains than the other samples with a major population that could be correlative to either the Pavant Butte or Tabernacle Hill tephra. Previously published pollen extract radiocarbon dates from Bear Lake cores BL96-2, BL96-3, and BL00-1D (Colman et al., 2009, 2006) and various terrestrial radiocarbon samples from the Bessette Creek Outcrop (Westgate and Fulton, 1975) that cover the Late

Pleistocene were also used in this study. We follow the recommendation of Colman et al. (2009) for which radiocarbon samples to include and reject in Bear Lake and transfer depths to the BL00 depth scale using Table 3 in that publication.

Table 3.2. Age Control Points

Composite Depth in Section Depth BL00, 14C Depth Midpoint Bear UCIAMS age Core Interval (cm) (m) Lake (m) Material #/Lab # (BP) ± Ref.

Fish Lake, Utah B14-D1 89-90 1.355 - Charcoal 141387 2455 20 This Study C14-D1 31-33 1.830 - Wood 141379 2890 20 C14-D1 174-175 3.255 - Charcoal 141380 4585 50 C14-D2 55.5-56.5 4.030 - Wood 141381 5840 25 C14-D2 111.5-112.5 4.590 - Wood 141382 6630 80 C14-D2 136-137.5 4.838 - Charcoal 141383 7940 25 C14-D2 165-168 5.135 - Charcoal 141384 9150 30 B14-D3 151-152 5.815 - Wood 141388 11270 60 C14-D3 40-41 5.875 - Seed 141385 11325 40 C14-D3 60-61 6.075 9.21 Plant Material 152063 11760 250 B14-D4 61 6.960 12.08 Tephra Correlation - 14400 100 (Tabernacle Hill) B14-D4 91.5 7.265 12.84 Tephra Correlation - 15650 350 (Pavant Butte) B14-D5 91.5 9.165 17.54 Tephra Correlation - 20160 254 (Pony Express) C14-D5 125-126 10.695 20.00 Wood 152064 24300 2800 C14-D5 136-139 10.815 20.15 Wood 141286 28200 2100

Bear Lake, Utah/Idaho BL96-2-2D 7 10.05* 10.05 Pollen Extract WW-1774 12710 50 Colman et BL96-2-2D 8 10.08* 10.08 Pollen Extract WW-2602 12545 90 al., 2009 BL96-3-3E 15 16.72* 16.72 Pollen Extract WW-2606 22150 210 BL00-1D-6H2 109-110 18.00 18.00 Pollen Extract WW-6452 24280 110 BL00-1D-7H1 19-20 18.59 18.59 Pollen Extract WW-6453 23340 100

Bessette Creak, British Columbia Outcrop 19.82 18.25+ 16.26 Moss GSC-913 19100 240 Westgate Outcrop 6.90 6.90 20.21 Peat GSC-1945 25400 270 & Fulton, Outcrop 4.50 4.50 21.19 Wood GSC-1953 25300 320 1975 Outcrop 1.25 1.25 23.26 Peat GSC-1938 31100 480 Outcrop 0.00 0.00 24.06 Bark GSC-2031 31200 900 * Correlation to BL00 depth scale based on Table 3 of Colman et al. (2009). + Adjusted depth scale of Turner et al. (1982), after removing gravel layers

Table 3.3. Summary of geochemical data for major tephra glass populations, normalized to 100% 2+ 3+ totals. FeOt is total iron oxide reported as FeO (Fe ), some materials may contain significant Fe (Fe2O3); n = number of analyses; S.C. = similarity coefficient; Pavant Butte and Tabernacle Hill tephra glass compositions from Oviatt and Nash (1989); Pony Express glass composition from Oviatt and Nash (2014). Table with complete data, including outliers and reference glass are archived at the Oregon State University Paleo- and Environmental Magnetism Laboratory.

Sample SiO2 TiO2 Al2O3 FeOT MnO MgO CaO Na2O K2O P2O5 Cl n S.C.

CU1341 Mean 71.69 0.68 13.78 2.41 0.06 0.29 0.80 3.35 6.78 0.10 0.07 13 - Core B14- D4 23 cm StDev 0.73 0.04 0.34 0.13 0.01 0.03 0.09 0.38 0.25 0.02 0.01

CU1342 Mean 50.89 1.95 15.55 12.08 0.21 5.28 8.69 3.57 1.32 0.45 0.03 28 - Core B14- D4 61 cm StDev 0.52 0.07 0.39 0.61 0.02 0.46 0.41 0.22 0.11 0.04 0.01 Pavant Butte tephra (L-2) 51.40 1.74 15.82 12.19 0.19 5.00 8.55 3.33 1.34 0.34 0.02 0.97 Tabernacle Hill tephra (TH-3A) 50.56 1.82 15.69 11.69 0.18 5.34 9.62 3.32 1.28 0.49 0.02 0.96

CU1343 Mean 50.86 1.91 15.68 12.11 0.21 5.22 8.76 3.48 1.29 0.45 0.04 26 - Core B14- D4 91.5 cm StDev 0.36 0.06 0.29 0.49 0.03 0.44 0.42 0.18 0.10 0.04 0.01 Pavant Butte tephra (80-15) 51.19 1.66 15.83 11.82 0.18 5.32 8.93 3.37 1.29 0.34 0.02 0.98 Tabernacle Hill tephra (TH-3A) 50.56 1.82 15.69 11.69 0.18 5.34 9.62 3.32 1.28 0.49 0.02 0.97

CU1344 Mean 51.08 1.69 15.57 10.13 0.18 6.56 10.29 3.25 0.91 0.31 0.03 36 - Core B14- D5 91.5 cm StDev 0.21 0.05 0.15 0.20 0.02 0.15 0.13 0.17 0.04 0.02 0.02 Pony Express tephra (SK-2) 50.79 1.71 15.62 10.18 0.19 6.52 10.48 3.18 0.90 0.33 0.11 0.99

CU1345 Mean 50.73 1.97 15.76 11.66 0.20 5.35 9.09 3.52 1.25 0.45 0.03 13 - Core C14- D3 116.5 cm StDev 0.24 0.11 0.21 0.36 0.03 0.33 0.32 0.29 0.08 0.04 0.01 Pavant Butte tephra (PBN-SW) 51.26 1.78 15.71 11.48 0.16 5.37 9.22 3.31 1.27 0.35 0.02 0.98 Tabernacle Hill tephra (TH-3A) 50.56 1.82 15.69 11.69 0.18 5.34 9.62 3.32 1.28 0.49 0.02 0.98

3.4 Results

3.4.1 Fish Lake, Utah Natural and Laboratory Remanent Magnetizations NRM intensities are weak, typically between 10-4 and 10-3 A/m, in the low density brown organic mud above 6.35 mcd but increase by up to three orders of magnitude below 6.35 mcd (Figures 3.3 and 3.4). A viscous remanent magnetization (VRM) with anomalous

39 directions is present in all lithologies but can generally be removed with a 10 mT AF demagnetization. Following removal of the VRM, AF demagnetization shows systematic, although sometimes noisy, behavior with magnetic directions demagnetizing towards the origin on a Zijderveld plot (Figure 3.3). ARM intensities are stronger but follow the same general pattern as NRM intensities. There are large variations in ferrimagnetic mineral coercivity, tracked by the ratio of the ARM intensity remaining after 20 mT AF demagnetization relative to the primary ARM intensity (ARM20mT/ARM), with the lowest coercivity ferrimagnetic mineral assemblages stratigraphically above and below the high density massive gray clay lithologic unit (Figure 3.4). Based on the demagnetization behavior, ChRMs were calculated using a PCA over the 15-50 mT AF demagnetization range. MAD values are higher in the low NRM intensity unit of the upper 6.35 mcd, with the majority ranging from about 5 to 10o. Below, 6.35 mcd, MAD values indicate the ChRM is much better defined, with all values less than 5o. The average ChRM inclination value for all data after removing u-channel edges is 57.1o, similar to the 58o for the site that is predicted based on the assumption of a geocentric axial dipole. 95% of all inclination values are within 16.3o of the average value, consistent with observations of well-defined mid-latitude western North America Holocene paleosecular variation records from archeomagnetic, volcanic, and sedimentary archives (e.g. Verosub et al., 1986; Hagstrum and Champion, 2002; Hagstrum and Blinman, 2010). However, due to the weak NRM intensities in the upper unit, large changes in NRM intensities in the lower unit, and large changes in ferrimagnetic coercivity, we determine this site is not well-suited for a continuous relative paleointensity reconstruction through the entire recovered interval (e.g. Tauxe, 1993; Stoner and St-Onge, 2007), and focus our efforts on the directional PSV record. Future investigations to reconstruct relative paleointensity at Fish Lake, UT may need to consider the lithologic units separately or experiment with objective correction factors (e.g. Brachfeld and Banerjee, 2000; Mazaud, 2006).

Figure 3.3. AF demagnetization behavior of the 25 cm position in each u-channel taken from the Fish Lake, Utah cores. For each pair, (left) magnetization (J) of the NRM (black) and ARM (blue/red) during AF demagnetization normalized by the 20 mT AF demagnetization step and (right) Zijderveld plots (Zijderveld, 1967) of the vertical (black) and horizontal (blue/red) components of the NRM.

Figure 3.4. Magnetic Results for the Fish Lake, Utah cores from holes B14 (blue) and C14 (red). From top to bottom: NRM intensity measured before and after every 10 mT AF demagnetization step between 0 and 80 mT. ARM intensity at the same steps as the NRM intensity. ARM coercivity, tracked as the ratio of the ARM after 20 mT AF demagnetization to the ARM before demagnetization. ChRM Inclination and Declination calculated with a PCA for steps measured between 15 and 50 mT AF demagnetization (points) and mean stacked directions (black line) with α95 uncertainty (gray shading). Measurements that did not pass quality criteria are plotted in a lighter color. MAD values are plotted like the ChRM directions.

To further investigate the directional record from the well-defined ChRM interval below 6.35 mcd, we performed deconvolution experiments on the u-channel data, using UDECON MATLAB tools (Xuan and Oda, 2015) and a conservative smoothness parameter (ln(u)) of 2 in our final result (Figure 3.5). One of the key assumptions required in deconvolution of u-channel data is that the downcore magnetization varies as a smooth

42 function (Jackson et al., 2010). Ultimately, deconvolution models are a trade-off between the model-data misfit, some of which is likely due to measurement error, and smoothness (Oda and Shibuya, 1996; Jackson et al., 2010). Larger, more conservative, ln(u) values, similar to the ones employed here, result in a smoother deconvolution solution but are more robust to measurement noise by not allowing solutions that overfit the raw data. As a result, conceptually, our conservative approach to deconvolution may be more akin to simulating a magnetometer with a narrower response function than solving for discrete 1 cm magnetizations.

Figure 3.5. (Left) Results of deconvolution experiment for the Fish Lake, Utah cores for the 4th and 5th drive from holes B14 (blue) and C14 (red). Primary SRM data are plotted as gray dashed line. From top to bottom: Magnetic susceptibility measured with a 36 mm diameter Bartington loop compared with the NRM intensity after 20 mT AF demagnetization. A 30 cm bin moving correlation coefficient is used to quantify the general improvement in correlation of the detrended logarithmic NRM intensity compared to the detrended logarithmic magnetic susceptibility for each u-channel. ChRM Inclination, Declination and MAD values calculated with a PCA for steps measured between 20 and 50 mT. (Right) Comparison of PSV stacks with 1 σ uncertainty made with primary SRM data (black line/gray shading) and deconvolved data (blue line/light blue shading).

To test the results, we compare the deconvolved u-channel SRM data with the u- channel magnetic susceptibility data. The Bartington loop used to collect the magnetic susceptibility data has a narrower response function (~3.0 cm FWMH) than the SRM (~7.5

43 cm FWMH). Assuming the concentration of ferrimagnetic minerals is a primary control on both magnetic susceptibility and NRM intensity, we expect the correlation between the two parameters to improve following deconvolution of the NRM. To test this, we calculate the linear correlation coefficient in 30 cm bins between the detrended logarithmic magnetic susceptibility and detrended logarithmic NRM intensity after 20 mT AF demagnetization (Figure 3.5). We use the detrended logarithmic data to mitigate the impact of large susceptibility and intensity changes and long wavelength variations on the calculation. We find, in most instances, an improvement in correlation coefficients for the deconvolved SRM data. We also note that ChRM MAD values from calculating the PCA of the deconvolved data are not significantly different from those calculated before the deconvolution, suggesting the deconvolution is creating consistent results for each demagnetization step. These findings provide confidence in our application of the deconvolution method of Oda and Xuan (2014) and in using the deconvolution results in our discussion.

3.4.2 Stacking Paleomagnetic Directions for Fish Lake, Utah and Bear Lake, Utah/Idaho Paleomagnetic data from Fish Lake, UT and Bear Lake were stacked to increase signal to noise and assess uncertainty of the PSV estimates. We do not create a new stack for the Bessette Creek data, but use the data and uncertainty as reported by Turner et al. (1982), as discrete samples were directly sampled from the Bessette Creek outcrop and each stratigraphic horizon is already the mean of 3-4 specimens.

3.4.2.1 Stacking Fish Lake, Utah Cores Four intervals were not used in the creation of the Fish Lake, UT PSV stack, due to coring disturbances or anomalous magnetic results. High MAD values and negative inclinations in B14-D2 between 3.30 and 3.40 mcd were removed, which most likely result from remagnetization post recovery, although the exact origin is unknown. Significant gas expansion observed in the CT scans (Figure 3.2) at the base of C14-D2 below 4.71 mcd is associated with anomalously shallow inclination values with respect to overlapping sediments in B14-D3 and was also removed. We removed an interval in B14-D4 between 5.86 and 6.17 mcd with anomalously high ARM intensity and low ARM coercivity which also

44 has negative inclinations, suggesting the anomalous magnetic mineralogy is not suitable for paleomagnetic analysis. Finally, we remove the upper 19 cm in the B14-D4 u-channel, above 6.48 mcd, as we suspect the sediments near the top of the core were significantly twisted during recovery, leading to a large swing in declination with is not reproduced in C14-D3. As there was no azimuthal orientation of the cores during recovery, declinations are only relative changes within each core. Accordingly, we first rotated the declination in each core to a mean of zero. In all but one circumstance, we chose to use this simple correction as it provided a reasonable agreement between overlapping drives. The only exception is for C14-D5. There was a significant offset between B14-D5 and C14-D5 and we added an additional 25o eastward rotation to C14-D5 to account for it. Our estimates of uncertainty in the stacked Fish Lake, UT PSV record (Figure 3.4) suggest that while the low NRM intensity unit at the top of the core can only resolve longer wavelength PSV features, the higher NRM intensity unit below 6.35 m has well resolved PSV features. A second stack was created for B14-D4, B14-D5, C14-D4, and C14-D5 using the deconvolution results, with ChRMs calculated in the same manner as the primary data (Figure 3.5). The stack created with the deconvolution results does not create any new high amplitude PSV features when compared to the stack created with the primary measured data. The biggest difference is that well-defined PSV features in the deconvolved data stack are more sharply defined.

3.4.2.2 Stacking Bear Lake, Utah/Idaho Cores The Bear Lake PSV stack was created with all data between 9.75 and 25 m from Hole BL00-1D Cores 4H-9H and Hole BL00-IE Cores 4H-9H, after removing u-channel edges and rotating declinations of overlapping sediments to match (Figure 3.6). As there is more overlapping sediment between the Bear Lake Cores than the Fish Lake, UT cores, we believe this approach is justified over simply rotating each drive to a mean of zero. However, we found it difficult to rotate Bear Lake declinations with this strategy below 17 m without imparting a large ~8 m linear trend to the east (~7.5o/m). We interpret these linear declination trends to be twisting artefacts of coring. To address this issue, we subtract the linear trends found in BL00-1D Cores 7H and 8H and in BL00-1E Cores 7H-9H. The result improves the agreement in declination in overlapping sediments; however, we recognize

45 that uncertainty in this declination correction could only be fully quantified by more observations from additional overlapping cores. After detrending, the final rotations were 35, 55, -15, 60, 55, and -30 for BL00-1D Cores 4H-9H, respectively, and -50, -40, -30, 90, -10, and -125 for BL00-1E Cores 4H-9H, respectively, with positive values indicating eastward rotations.

Figure 3.6. Paleomagnetic data from Bear Lake Drill Holes BL00-1D and BL00-1E (Heil et al., 2009). All inclination data are plotted, but only corrected declination data from cores used in the stack as discussed in the text. While data were stacked from 9.7-25 m, only the 9.7-24 m data (yellow shading) were used in the WNAM17 PSV stack, as this was the interval that overlapped with the Fish Lake, UT and Bessette Creek, BC records. Horizontal black lines indicated the predicted values based on the geocentric axial dipole hypothesis for the Bear Lake latitude.

3.4.3 Regional Comparison and Establishing an Integrated Regional Chronology While we recognize broadly consistent PSV signals between the three sites for the Late Pleistocene, we also recognize that all three sites, like many Late Pleistocene records,

46 do not have the strong independent chronologies needed to make robust comparisons of these features on age. Accordingly, we adopt the approach of Stoner et al. (2007) by correlating the Fish Lake, UT (deconvolved), Bear Lake, and Bessette Creek records to a common depth scale through graphical correlation and transferring available age control points to a single age-depth model (Table 3.4; Figure 3.7). We choose the Bear Lake BL00 depth scale as the common depth scale, as the Bear Lake record spans the longest period of time. To account for any geometric effects due to variations in latitude and longitude between the three sites, directions were projected to Seattle, Washington (47.621 oN, 122.349 oW) via their virtual geomagnetic pole (VGP) paths. This site was chosen, as it is a mid-latitude location for Western North America and will be a central location for future work that will explore high resolution Northeast Pacific marine and Western North American terrestrial archives. While we find good agreement between the absolute projected inclinations in Fish Lake, UT and Bessette Creek, Bear Lake inclinations are on average about 8o steeper. This difference could be the result of coring artefacts or poorly understood differences in sediment magnetic remanence acquisition. As Fish Lake, UT and Bessette Creek agree and Bear Lake site inclinations are almost always greater than predicted values based on the geocentric axial dipole hypothesis (Figure 3.7a), we subtract 8o from Bear Lake inclinations for the purpose of this stack. We find good agreement between the three records declination values and do not apply any further correction. We consider these reasonable estimates for past absolute changes in declination, as the Bessette Creek data are derived from oriented outcrop samples and the furthest distance (~1500 km) between two sites is about half the distance associated with spherical harmonic degree 6-7 wavelengths, which is used as the distance for relative declination orientation in the pfm9k spherical harmonic model (Nilsson et al., 2014). Stacking the three records, using the same method as described above, reveals a well resolved regional signal with α95 values typically less than 10o, which we have named the WNAM17 (Western North AMerican 2017) PSV stack (Figure 3.7e-f).

Table 3.4. Tie Points to Regional Depth Scale

Fish Lake, UT Bessette Creek, BC Fish Lake, UT Bear Lake BL00 Bessette Creek Bear Lake BL00 Depth (mcd) Depth (m) Elevation (m) Depth (m) 0 0 14.45 17.15 6.4 9.7 8.2 19.04 6.56 10.8 6.93 20.19 6.77 11.18 4.07 21.28 7.01 12.32 2.11 22.65 7.42 13.15 7.77 14.09 7.87 14.29 8.01 14.46 8.25 15.15 8.42 15.42 8.85 16.98 9.16 17.53 9.67 18.66 10.85 20.25

The resulting age-depth relationship of the three integrated records contains a number of age reversals, notably the pollen extract radiocarbon ages between 16-19 m in Bear Lake, the peat radiocarbon ages from Bessette Creek, and a small mass (0.04 mg carbon) sample with large uncertainty from Fish Lake, UT. While this could result from mismatches in the PSV correlations, some age reversals are not surprising given the difficulty in dating Late Pleistocene terrestrial archives devoid of high quality macrofossils and, in the case of Bessette Creek, limitations in samples appropriate for radiocarbon dating before the advent of accelerated mass spectrometry (Westgate and Fulton, 1975). While post glacial Bear Lake pollen extract samples generally give younger ages by several hundred years than paired total organic carbon (TOC) and/or carbonate samples, suggesting less contamination from old carbon (Colman et al., 2006), Colman et al. (2009) found last glacial pollen extract ages were up to several thousand years older than stratigraphically lower samples and attributed this difference to higher relative proportions of refractory organic carbon in the extracts. As a result, the authors rejected many of these dates from their age model. The three pollen extracts used in their age model and this study from below the facies interpreted as the local glacial maximum were described as containing better preserved pollen. Similarly, the rest of our age control points, particularly the peat samples, are all samples that could have issues with being anomalously old and incorporating reworked carbon from the landscape. Few samples, if any, used in the creation of the WNAM17 age model are the high quality terrestrial macrofossils needed to closely

48 approximate the age of the sediment, such as deciduous leaf macrofossils (e.g. Howarth et al., 2013). Nevertheless, these terrestrial samples have little risk for modern carbon contamination, as has been demonstrated to impact some lake carbonate samples from Western North America by progressive leaching (Kent et al., 2002; Hajdas et al., 2004), or complications from old lake carbon reservoir issues. At a minimum, these ages can be considered robust maximum limiting ages.

Figure 3.7. PSV records, projected to Seattle, Washington via their VGP paths, with 1 sigma uncertainty and age control points for Bear Lake (red, (a)), Fish Lake (blue, (b)), and Bessette Creek (green, (c)). Correlated PSV records (d) were used to create the WNAM17 PSV Stack (e), which averages 2-3 records and typically has α95 uncertainty of less than 10o (f). Horizontal gray lines in (a- e) reflect predicted inclination and declination values based on the geocentric axial dipole hypothesis for the latitude of Seattle. (g) 14C age estimates, transferred to the common depth scale, were calibrated to calendar years and the median and 95% confidence interval age-depth model for the WNAM17 PSV stack was generated from 10,000 accepted model runs, as described in the text. Calibrated age PDFs from each site are indicated with colors matching (a-c) and the transferred maximum range for lock-in offset (30 cm) is indicated by a vertical line. The regional chronostratigraphic model factors three main sources of uncertainty: (1) the measurement precision and calibration of the radiocarbon measurements, (2) that only a subset of age estimates closely approximate the age of the sediment, and (3) that there is an unknown difference in the age of the sediment and the age of the magnetization.

Another source of age uncertainty for the WNAM17 PSV Stack is that the age of the physical sediment and age of the sediment magnetization are not necessarily equal due to sediment magnetization acquisition in a lock-in zone following deposition (e.g. Irving and Major, 1964; Løvlie, 1976; Verosub, 1977; Egli and Zhao, 2015). While sediment magnetization acquisition processes are not completely understood and may result from different processes in different depositional environments, studies commonly find decimeter offsets in the age of magnetizations in bioturbated and varved sediments where independent chronometers and/or superposition allow comparison (deMenocal et al., 1990; Channell and Guyodo, 2004; Suganuma et al., 2010; Stoner et al., 2013; Snowball et al., 2013; Mellström et al., 2015; Simon et al., 2018; Nilsson et al., 2018). Given these known uncertainties, we use an age-depth modeling approach inspired by Haslett and Parnell (2008) to calculate an ensemble of possible age depth combinations that capture the uncertainty structure of the WNAM17 age of magnetization. Assuming that all age control points are robust maximum limiting ages but only a subset approximate the actual age of the sediment, we start each iteration by randomly selecting eight of the sixteen (50%) age control points. The ages of the age control points are then randomly selected from the calibrated age probability distribution function (PDF) and the depth from a uniform distribution ranging from 0 cm (no offset in magnetization age) to 30 cm (large offset in magnetization age) and then transferred to the common BL00 Bear Lake depth scale. We believe this range is justified, as while some studies argue for little or no offset (Valet et al., 2014), many observations suggest offsets of up to 14-25 cm (Channell and Guyodo, 2004; Suganuma et al., 2010; Stoner et al., 2013; Simon et al., 2018) or in some cases greater offsets (Snowball et al., 2013). Additional synthetic age-depth pairs are then added at random between these age control points to allow for non-linear accumulation rates between dated horizons. While the spacing of synthetic age-depth pairs varies for each iteration, the result is approximately four real and/or synthetic age-depth pairs per meter. We reject the iteration if there are age reversals, as this violates the law of superposition, or if any of the sixteen dated horizons are older than its maximum limiting radiocarbon constraint, defined as the 99th percentile of the calibrated PDF. The model is run until 10,000 iterations are accepted. The result is an age-depth relationship for magnetization age of the

WNAM17 PSV Stack with large, but realistic uncertainty (median 95% confidence interval of ~1.5 ka), allowing for refinement with future work (Figure 3.7). The greatest strength of age-depth model is that be approaching chronostratigraphy from a regional perspective, assuming a common geomagnetic signal, we can address sources of geologic uncertainty that would be difficult to quantify using only one site.

3.5 Discussion

3.5.1 Implications for Regional Chronologies and the Timing of Major Lithologic Transitions Development of the independently dated WNAM17 PSV stack provides a template for improving or assessing regional chronologies. As a first exercise, we look at the impact on the timing of major lithologic changes observed at Fish Lake, UT and Bear Lake, UT/ID on their independent chronologies and WNAM17 ‘tuned’ chronologies. We use the original correlations defined between Fish Lake, UT and Bear Lake while building the stack and compare them to the ages of the same integrated horizons in the WNAM17 Stack (Table 3.4). This ensures each record has the same number of correlation points and each correlation point is at a PSV feature that was recognized at both sites. We generate PDFs from the ensemble of age-depth models discussed in Section 3.3 at each tie point, which then can be used to create a new ensemble of site age-depth models. Age-depth modeling is done as in Section 3.3, with a few modifications (Figure 3.8). First, we do not need to drop any age control points, as all dates are in stratigraphic order and are estimates for the sediment’s age of magnetization. Second, we do not need to use the maximum limiting age constraint for the same reason. Third, we do not perturb the depths of the age control points; rather, we run two simulations to illustrate the potential influence of constant offsets in magnetization age—with one assuming no offset and the other assuming a constant 15 cm offset. While it may be simpler for these two lakes to just take the WNAM17 age-distributions at horizons of interest, we create new age models in this way to illustrate how the WNAM17 stack could be used with other records where age- control points can only be defined by where there are strong PSV correlations. To illustrate the chronologic implications, we also use the published age-depth model of Colman et al.

(2009) for Bear Lake, and a new age-model for Fish Lake, UT, using only the Fish Lake, UT radiocarbon and tephra age control points, generated in the same fashion as the WNAM17 tuned chronologies.

Figure 3.8. Comparison of the ages of major lithologic contacts observed at Bear Lake, UT/ID and Fish Lake, UT. Age depth models for (a) Bear Lake, UT/ID and (b) Fish Lake, UT were generated using WNAM17 age PDFs (light red shading) at PSV correlation horizons (Table 3.4). Two age models were generated to illustrate age-depth relationships if sediment and magnetization ages are equal (0 cm lock-in; median age = red line) or if the age of the magnetization is offset by 15 cm (15 cm lock-in; median age = light blue line). For comparison, the previously published Bear Lake age model (Colman et al., 2009) and an age model generated only using age control points from Fish Lake, UT are plotted (black lines). (c) The resulting age PDFs for the two lithologic contacts for each set of age models.

We treat major lithologic transitions as events and generate PDFs of the event age from each of the age models described above (Figure 3.8c). For Bear Lake, the transitions are defined as the increase in magnetic susceptibility below the red siliciclastic unit (18 m) and the top of the red siliciclastic unit (10.4 m), interpreted as being deposited during the local glacial maxima (Colman et al., 2009; Heil et al., 2009; Kaufman et al., 2009). For Fish Lake, UT, the transitions are defined by the bottom (9.17 m) and top (6.41 m) of the high- density and high magnetic susceptibility unit (Figure 3.2). The focus here is to discuss the timing of the changes. A detailed study of the paleoenvironmental and glacial implications of these records will be discussed elsewhere. On each site’s independent chronology, the base of the lithologic transitions would be considered separate events, with the transition at Bear Lake preceding the transition at Fish Lake, UT (Figure 3.8c). Applying the WNAM17 tuned chronology pushes this event much younger at Bear Lake (~24-26 ka), with its new PDF overlapping with Fish Lake, UT independently of the depth in magnetization offset we use. Similarly, for the younger event, while the age of the event is poorly defined at Fish Lake, UT on its independent chronology, applying the WNAM17 tuned chronology provides a stronger constraint for its age (~14-16 ka). In both cases, a slightly better agreement can be generated by invoking magnetic lock-in offsets that are reasonable when compared to other published estimates, as discussed earlier. This occurs because in this exercise the constant magnetization offsets are applied in the depth domain, which results in a variable age offset depending on sediment accumulation rates. While there is no direct evidence that indicates that these lithologic changes need to be regionally consistent, we consider this anecdotal support for application of the WNAM17 template to problems such as the timing of geologic events. Within the context of regional PSV stratigraphy and uncertainties in radiocarbon dating and the magnetic acquisition processes, we cannot rule out the possibility that major glacial changes in the Late Pleistocene occurred in phase in these two basins.

3.5.2 Assessing the Chronology of Mono Lake, California In our second exercise, we apply the WNAM17 PSV stack as a reference to the magnetic record at Mono Lake, California. While, many are interested in the event age of the geomagnetic excursion found at this site, our goal in this discussion is not to give a definitive answer, but to explore the PSV stratigraphic context of the event and discuss the implications for proposed timings of that event. The chronology of sediment outcrops at Mono Lake have received considerable attention for resolving the age and uniqueness of the geomagnetic excursion recorded in its sediments (Kent et al., 2002; Benson, Liddicoat, et al., 2003; Zimmerman et al., 2006; Cassata et al., 2010; Vazquez and Lidzbarski, 2012; Lund et al., 2017) and the relationship between paleoenvironmental signals and abrupt climate and/or orbital climate signals (Benson et al., 1998; Benson, Lund, et al., 2003; Zimmerman, Hemming, et al., 2011; Zimmerman, Pearl, et al., 2011). Initial paleomagnetic work recognized a high amplitude declination feature around ash layer 15 in the Wilson Creek Formation (Denham and Cox, 1971), which was later defined with more detailed work as a significant and reproducible geomagnetic excursion recorded at Mono Lake (Liddicoat and Coe, 1979). Lund et al. (1988) compiled previously published and new PSV data for the Wilson Creek Formation, revealing a well-resolved and high amplitude record of Late Pleistocene PSV and placed the excursion event in the context of longer term geomagnetic change. As there is no community consensus on the chronology of the Wilson Creek Formation at Mono Lake, the WNAM17 PSV stack provides the opportunity to assess the Wilson Creek Formation chronology where they overlap in time. While the initial chronology for the complete record was constrained by radiocarbon dating of carbonate samples, including tufa and ostracods (Lund et al., 1988; Benson et al., 1990), progressive leaching of radiocarbon samples indicated significant contamination by young carbon, suggesting age estimates for these sediments were too young and radiocarbon dates may only represent a minimum limiting age (Kent et al., 2002; Hajdas et al., 2004), which seems to support older radiometric dates on ash layers surrounding the excursion (Kent et al., 2002; Zimmerman et al., 2006; Cassata et al., 2010; Cox et al., 2012; Vazquez and Lidzbarski, 2012). However, the radiocarbon ages are also complicated by unknown changes in potentially very large reservoir ages (on the order of

103 yrs; Broecker et al., 1988; Benson et al., 1990) and radiometric dating of ash layers have been argued as being only maximum limiting ages and not direct dates of the eruptions (e.g. Kent et al., 2002; Cassata et al., 2010; Negrini et al., 2014), requiring assessment by independent stratigraphic correlation. This independent stratigraphic correlation has previously been attempted using tephra and paleomagnetic correlation. For example, Benson et al. (2003) identified a tephra layer with similar chemical composition to the Wilson Creek Formation ash layer 15, which transects the excursion, in the Pyramid Lake Basin, Nevada with a 14C age of 28,620 ± 300 yrs (IntCal13 2σ: 31,693-33,474 cal yrs BP with no reservoir correction) based on total organic carbon, which may be impacted by reservoir ages (estimated at ~600 yrs in the late Holocene), but should not face the same young carbon contamination as the carbonate samples in Mono Lake. Excursional directions have also been documented just beneath this ash layer in sediment cores from Pyramid Lake (Lund et al., 2017) and outcrops within the basin (Liddicoat, 1992). Conversely, a paleomagnetic correlation, based on relative paleointensity, supports an older age for the excursion at Mono Lake, coeval with the age of the Laschamp Excursion documented elsewhere (Zimmerman et al., 2006) with an age of ~41 ka (Nowaczyk et al., 2012; Laj et al., 2014; Lascu et al., 2016). While this chronology is consistent with 238U-230Th ages of the most recent crystallization of allanite crystals at multiple ash layers (Vazquez and Lidzbarski, 2012), it is non-unique and difficult to reconcile with the leached carbonate radiocarbon samples (Cassata et al., 2010). Less visually appealing correlation scenarios than the Zimmerman et al. (2006) relative paleointensity correlation proposed by Cassata et al. (2010) could easily be justified as reflecting variations in magnetic remanence acquisition efficiency in different Wilson Creek Formation lithologic units, as the intervals with highest absolute values of normalized NRM intensity are consistently found in lithologic units with the lowest weight percent total inorganic carbon (e.g. Benson et al., 1998). Our WNAM17 PSV Stack offers an independent opportunity for stratigraphic correlation, using the well resolved PSV record of Lund et al. (1988) and transferred to the Wilson Creek Formation type section height by linear interpolation between ash layers. While PSV correlation to distal Western North Atlantic Sites has been conducted previously (Benson et al., 1998; Lund et al., 2017), we feel our approach is simpler and requires fewer

55 assumptions about the synchronicity of inclination and declination features over significantly larger length scales. For direct comparison with the WNAM17 PSV Stack, the Mono Lake directions are also projected to Seattle, Washington via their VGP path. For the purposes of this discussion we do not apply an offset to the magnetization, as we are comparing a magnetic age at Mono Lake to magnetic age in the WNAM17 template. We feel this is appropriate, but it is important to note that when comparing to the radiocarbon and ash layer dates our magnetic ages may indicate a slightly older age for the sediment, with age offsets depending on accumulation rates. Correlation of PSV features is straight forward for sediments from the top of the section to just below ash layer 7, with good agreement in inclination and declination without invoking large sedimentation rate changes. These PSV features are also in good agreement with the available radiocarbon samples, given their uncertainty, and the Vazquez and Lidzbarski (2012) 238U-230Th date of ash layer 7, giving us confidence in this approach and adding significantly stronger chronologic constraint to these sediments than previously available (Figure 3.9 and 3.10). Differences in 40Ar/39Ar, 238U-230Th, and 14C age estimates become more pronounced below this level. Moving down section, below ash layer 7, the Mono Lake PSV record has a number of high amplitude PSV features. The excursion occurs during a longer wavelength inclination steepening, overlain by a broad interval of shallower inclinations. The excursional feature also occurs during a longer wavelength period of eastern declinations, overlain by a broad interval of western declinations. This pattern is typical of what is described for other examples interpreted as the Mono Lake Excursion at other locations in Western North America (Negrini et al., 1984, 2014). However, we note that while the immediately older PSV pattern in the WNAM17 PSV Stack fits this description, there is no excursional behavior documented at any of our sites. Thus, there are at least two alternatives. As almost any stratigraphic correlation is non-unique, we present a case for each scenario with their respective magnetic, chronologic, and sedimentological implications. In each case, the PSV tie points from the top of the section to just below ash layer 7 are the same.

3.5.2.1 Scenario 1: The Laschamp Excursion Recorded at Mono Lake Reconciling the strong PSV based chronology above ash layer 7 and the 238U-230Th allanite dates below requires a decrease in sedimentation rates, relative to the overlying sediments, or a hiatus in sedimentation between ash layers 7 and 11 (Figure 3.9). We have difficulty finding a strong agreement in both inclination and declination consistent with a lower sedimentation rate scenario that still has good visual correlation with inclination and declination. Notably, the broad eastward declination and broad shallow inclination features centered around 4 meters elevation in the Wilson Creek Formation type section, would need to be stretched to encompass two large eastward declination and two shallow inclination features between about 18 and 23 meters in the WNAM17 PSV Stack. This could be the result of the lower sedimentation rates attenuating or distorting the PSV signal and reconciled by low and/or variable sedimentation rates through this time period (c.f. Balbas et al., 2018). It is also possible that there is an unrecognized hiatus. As we don’t have PSV based age control beyond the constraints of our stack, for this exercise we combine our tie points with age control points from the 238U-230Th allanite dates at ash layers 11, 15, 17, and 19 (Vazquez and Lidzbarski, 2012), as these age control points are in good agreement with the relative paleointensity age model of Zimmerman et al. (2006), supporting the interpretation that they are good estimates of eruption age. The results of 10,000 age model realizations suggests fairly low long-term sedimentation rates (on the order of 10 cm/ka) for the sediments deposited below about 4 m, although higher resolution variations in sedimentation rate cannot be fully assessed. Interestingly, in this scenario, the second largest inclination shallowing feature is given an age of around 30 ka which is within uncertainty of a recognized anomalous geomagnetic event and the age of the excursion at Mono Lake in Scenario 2, discussed below. It is our opinion that this should not be confused as evidence supporting the recording of the Laschamp at Mono Lake, as it is possible other short-lived (102 yr) field anomalies occurred around the Scenario 2 timing of this feature as well (e.g. Channell, Harrison, et al., 2016).

Figure 3.9. PSV Tuning exercise comparing the Mono Lake PSV record to the WNAM17 PSV Stack, following Scenario 1 in the main text (Section 3.5.2.1). (a) PSV record from Mono Lake (green line; Lund et al., 1988) transferred to the type section height by linear interpolation between ash layers. Vertical lines indicate the positions of Ash Layers 7 and 15. (b) Correlation of the Mono Lake PSV record to the WNAM17 PSV Stack (black line with 1σ uncertainty displayed in gray shading). Light blue lines are used to display tie points. In this scenario, PSV correlation is only used where there is good agreement between all chronometers to just below Ash Layer 7. (c) The resulting PSV record on age, using the age depth model in (d). (d) Probability distribution functions (PDF) of the ages of tie points to the WNAM17 PSV Stack (light green) and 238U-230Th radiometric constraints on Ash Layers 11-19 from Vasquez and Lidzbarski (2012) where used to develop 10,000 age models with median (black line) and 95% confidence interval (gray shading). This age model is compared to radiocarbon estimates from leached (Kent et al., 2002; Hajdas et al., 2004) and unleached (Lund et al., 1988; Benson et al., 1990) carbonate samples. Ash layers (pink) and the excursion, defined by the interval including the eastern most declination swing and shallowest inclination, (yellow) are included for reference. (e) Resulting sedimentation rates with 1σ uncertainty calculated from the linear sedimentation rates between tie points, compared with lithologic and paleolimnology proxies of Benson et al. (1998).

3.5.2.2 Scenario 2: The Excursion at Mono Lake Occurred between 30 and 34 cal ka BP In our second scenario, we continue tuning the PSV record below Ash Layer 7 to the extent of the WNAM17 template. Sedimentation rates are roughly comparable between ash layers 7 and 11 to the sediments overlying ash layer 7 and increase below ash layer 11 (Figure 3.10). In this scenario, 238U-230Th allanite dates would be anomalously old, meaning the final stages of allanite crystallization are not equivalent to the eruption ages, and Cox et al.’s (2012) (U-Th)/He ages of ash layer 15 would need to reflect multiple populations of ages, with only the youngest population (i.e. 34.9 ± 1.1 ka; 35.2 ± 1.1 ka) approaching the true eruption age (e.g. Negrini et al., 2014). The resulting excursion recorded at Mono Lake would be a short event (on the order of 102 yrs), meaning Bear Lake and Bessette Creek do not record the excursion because the signal is attenuated by pDRM processes related to differences in depositional processes and/or sedimentation rates. This is a common phenomenon, with geomagnetic excursions often recorded in some sedimentary archives and not others from the same region (e.g. Channell, 2017). To illustrate this point, we use a simple linear pDRM model with half lock-in depth of 15 cm (i.e. a 30 cm wide boxcar filter) on our PSV correlation depth scale (i.e. the depth scale of the BL00 cores at Bear Lake) and produce a signal similar to the WNAM17 PSV stack with no obvious excursion (Figure 3.10c). This exercise assumes that the higher resolution Mono Lake signal better approximates the true geomagnetic signal, for which a smoothed signal is recorded in other lower accumulation rate depositional systems.

Figure 3.10. PSV Tuning exercise comparing the Mono Lake PSV record to the WNAM17 PSV Stack, following Scenario 2 in the main text (Section 3.5.2.2). (a) PSV record from Mono Lake (green line; Lund et al., 1988) transferred to the type section height by linear interpolation between ash layers. Vertical lines indicate the positions of Ash Layers 7 and 15 (b) Correlation of the Mono Lake PSV record to the WNAM17 PSV Stack (black line with 1σ uncertainty displayed in gray shading). Light blue lines are used to display tie points. (c) The result of applying a simple 15 cm half lock-in linear pDRM model (i.e. 30 cm boxcar filter) to the correlated Mono Lake PSV record on the regional correlated depth scale of WNAM17, illustrating that the excursion at Mono Lake may not be recorded in all sedimentary archives from the region due to smoothing (e.g. Channell, 2017). (d) Probability distribution functions (PDF) of the ages of tie points to the WNAM17 PSV Stack (light green) were used to generate 10,000 age models with median (black line) and 95% confidence interval (gray shading). This age model is compared to radiocarbon estimates from leached (Kent et al., 2002; Hajdas et al., 2004) and unleached (Lund et al., 1988; Benson et al., 1990) carbonate samples. 238U- 230Th dating of allanite rims at Ash Layer 7, 11, and 15 (Vazquez and Lidzbarski, 2012), which are in good agreement with the relative paleointensity based chronology of Zimmerman et al. (2006), are also shown. Ash layers (pink) and the excursion, defined by the interval including the eastern most declination swing and shallowest inclination, (yellow) are included for reference. (e) Resulting sedimentation rates with 1σ uncertainty, calculated as the linear accumulation rates between correlation points, compared with lithologic and paleolimnology proxies of Benson et al. (1998).

Applying 10,000 age-depth model realizations using the age distributions of tie points to the WNAM17 PSV template places the excursion at Mono Lake, defined by the

60 shallowest inclination, between 29.8 and 31.7 cal ka BP with a median age of 30.6 cal ka BP on the IntCal13 timescale (range is 1σ), within uncertainty of anomalous directions dated regionally and globally, included basaltic lava groundmass 40Ar/39Ar estimates from New Zealand (31.8 ± 0.9 ka; Cassata et al., 2008) and the Canary Islands (32 ± 0.5 ka; Kissel et al., 2011) and radiocarbon estimates from total organic carbon near the correlative ash layer in the Pyramid Lake Basin after applying the 0.6 ka suggested reservoir correction (31.4 - 32.2 cal ka BP; Benson et al., 2003a) and from terrestrial macrofossils in anomalous magnetic directions in a drill core from the Santa Clara Valley (31.5 - 32.4 cal ka BP; Mankinen and Wentworth, 2004) (Figure 3.11). Interestingly, in comparison to the 10Be flux record from Greenland ice cores on the GICC05 timescale (Muscheler et al., 2014), these estimates are all more consistent with relatively high 10Be production period around 31 ka than the larger production episode around 34 ka. The Mono Lake PSV record is presented on this age model and compared to all the records discussed in this paper in Figure 3.12. In summary, both scenarios are possible, but have different stratigraphic, sedimentological, and magnetic implications when evaluated in the context of the WNAM17 template. In Scenario 1, the age of the excursion would be consistent with the global recognized Laschamp Excursion (~41 ka) and would be consistent with 238U-230Th allanite rim dates approximating true eruption ages. However, it is difficult to reconcile the lower sedimentation rates needed to fit this model without distorting the large amplitude PSV features older than 30 ka and invoking a hiatus or lower/variable accumulation rates complicating the signal. In Scenario 2, the age of the excursion would be consistent with independent age estimates for a shorter duration anomalous geomagnetic event recognized in a few locations between 30 and 32 ka. This scenario implies that sedimentation rates increase in the lithologic unit containing the excursion, approaching ~100 cm/ka or higher, allowing for better preservation of the signal compared to the lower accumulation rate (on the order of 101 cm/ka) sites used to build the WNAM17 template. However, this scenario also implies that 238U-230Th allanite dates are anomalously old for ash layers 11, 15, 17, and 19, but not ash layer 7.

Figure 3.11. Comparison of our WNAM17 PSV Stack Scenario 2 correlated age of the excursion at Mono Lake to independent radiocarbon estimates from other regional records (Benson, Liddicoat, et al., 2003; Mankinen and Wentworth, 2004), 40Ar/39Ar estimates from global volcanic records (Cassata et al., 2008; Kissel et al., 2011), and the 10Be flux in Greenland Ice Cores (Muscheler et al., 2014), smoothed using a 392 year FWHM Gaussian Filter. All radiocarbon estimates are calibrated to the IntCal13 timescale and 39Ar/40Ar are calibrated following the recommendation of Singer (2014). While these age estimates are all within uncertainty of each other, an age-depth model that gives an older age for the excursion recorded at Mono Lake, Scenario 1 displayed in Figure 3.9c gives an age for the excursion that is consistent with the high 10Be flux around 41 ka associated with the globally recognized Laschamp excursion.

Figure 3.12. Western North American PSV records projected to Seattle, WA discussed in the text and the WNAM17 PSV Stack placed on the WNAM17 median age timescale. Mono Lake data are presented on our Scenario 2 age-depth relationship. An age-depth model that gives an older age for the excursion recorded at Mono Lake, Scenario 1, is presented in Figure 3.9c.

3.6 Conclusion We present new PSV and geochronological data from Fish Lake, UT which in comparison to other regional PSV records provides stronger constraints on the regional PSV signal that can be used to address uncertainties in Late Pleistocene Western North American chronologies. In addition to radiocarbon uncertainties, we address uncertainties related to sediment magnetic acquisition processes to develop realistic magnetic age estimates for our regional PSV stack, WNAM17, spanning ~35-15 ka (Figure 3.12). We illustrate with two examples that an independently dated regional PSV template can be used to assess chronologies in difficult to date sediments. Comparison of the timing of local glacial maxima in the Bear Lake and Fish Lake, UT Basins using WNAM17 ‘tuned’ chronologies suggest major glacial advances and retreats could be in phase. Application of the stack to assess the controversial chronology of Late Pleistocene sediments at Mono Lake, California refines the

63 age model for sediments younger than 25 ka and offers new insight into magnetic, geochronological, and sedimentological implications for the age of the excursion recorded in the Wilson Creek Formation—specifically, that a 30-34 ka aged excursion would be recorded in high accumulation rate sediments and the WNAM17 and Lund et al. (1988) PSV signals could be reconciled by invoking reasonable smoothing associated with sediment magnetic acquisition processes, while a ~41 ka aged excursion would require either low and/or variable accumulation rates that would smooth and/or distort the Wilson Creek Formation PSV signal or an unrecognized hiatus to reconcile the WNAM17 and Lund et al. (1988) PSV signals. PSV stratigraphy is a powerful tool for assessing the chronologies of Late Pleistocene sediments in Western North America. Our regional chronostratigraphic model allows for evaluation of basin-specific age-depth models that have geologic uncertainties that are difficult to address otherwise. The WNAM17 PSV template offers a new independently dated and regionally developed tuning target. Moving forward, we can reduce uncertainties through identifying records with strong chronologies and well-defined PSV over discrete time intervals and furthering our understanding of sediment magnetic remanence acquisition processes.

3.7 Acknowledgements We are thankful to the 2014 Fish Lake, Utah coring team and working group, particularly Lesleigh Anderson, Andrea Brunelle, Vachel Carter, and Mitchel Power. The 2014 Fish Lake, Utah cores are stored at the OSU Marine and Geology Repository (www.osu- mgr.org; NSF-OCE1558679) and we thank Maziet Cheseby and staff for their help. We thank Jason Wiest and the OSU College of Veterinary Medicine for their help to CT scan these cores. We thank Chanda Bertrand for help with radiocarbon sample preparation and John Southon for radiocarbon measurements. Brendan Reilly is thankful for support from Leslie and Mark Workman and the ARCS Foundation Oregon Chapter. Funding from NSF- EAR1215888 to Joseph Stoner and NSF-EAR1215661 to Mark Abbott contributed to this product.

4. Middle to Late Pleistocene Evolution of the Bengal Fan at 8o North: Integrating Core and Seismic Observations for IODP Expedition 354 Transect Age-Depth Modeling

Brendan T. Reilly1, Fenna Bergmann2, Michael E. Weber3, Joseph S. Stoner1, Peter Selkin4, Tilmann Schwenk2, Volkhard Spiess2, Christian France-Lanord5

1 College of Earth, Ocean, and Atmospheric Sciences, Oregon State University, Corvallis, Oregon 97331, USA

2 Faculty of Geosciences, University of Bremen, Bremen, Germany

3 Steinmann-Institute, University of Bonn, Bonn, Germany

4 School of Interdisciplinary Arts and Sciences, University of Washington, Tacoma, Washington 98402, USA

5 Centre de Recherches Pétrographiques et Géochimiques, CNRS Université de Lorraine, Vandoeuvre les Nancy, France

In preparation for Earth and Planetary Science Letters

4.1 Abstract We investigate chronology and uncertainty for the Middle to Late Pleistocene Lower Bengal Fan using a novel age-depth modeling approach that factors litho-, magneto-, bio-, cyclo-, and seismic stratigraphic constraints, based on results from the International Ocean Discovery Program Expedition 354 Bengal Fan and analysis of the GeoB97-020/027 seismic line. The initial chronostratigraphic framework is established using regionally extensive hemipelagic sediment units and only age-depth models of fan deposits that respect the superposition of channel-levee systems are accepted. In doing so, we reconstruct signals of regional sediment accumulation rate and lithogenic sediment input that are consistent with more distal and more ambiguous Bay of Bengal and Bengal Fan records. This chronology allows us to discuss the Middle to Late Pleistocene Bengal Fan evolution within the context of sea level, climate, and tectonic controls. We hypothesize, based on the timing of accumulation rate changes, that growth and intensification of the Bengal Fan’s channel- levee system at 8o N was largely driven by increases in sea level amplitude during this time. However, it is also possible changes in Pleistocene climate occurring around the same time increased Himalayan erosion rates, resulting in greater sediment flux to the fan. Further work is needed to test these ideas.

4.2 Introduction The Bengal Fan is the final sink for a huge erosional system, containing the most complete record of Himalayan erosion since the collision of India and Asia in the early Cenozoic (Curray, 1994; France-Lanord et al., 2016). However, the climate and tectonic signals recorded in these sediments are convolved during sediment transport through rivers, floodplains, deltas, the continental shelf, and ultimately the fan’s channel-levee system. Understanding the complete history of the erosional system is important as the uplift of the Himalaya is thought to have a significant impact on global climate over the Cenozoic, including the development of the Asian monsoonal systems (Zhisheng et al., 2001) as well as carbon sequestration by increased silicate mineral weathering (Raymo and Ruddiman, 1992) and, more importantly, burial of organic carbon in its resulting deep-sea fan sediments (France-Lanord and Derry, 1997). Source to sink comparisons of organic carbon concentrations in river and fan sediments suggests near perfect burial efficiency—in large

66 part related to the very high accumulation rates of the Bengal Fan depositional system (Galy et al., 2007). Yet, complexities in fan depositional processes make it difficult to reconstruct time variations in integrated sediment flux and how these fluxes translate to sediment accumulation and associated organic carbon burial on the fan. To investigate the Pleistocene history of this depositional system, International Ocean Discovery Program (IODP) drilled a transect of seven sites along 8o North (~320 km; 85.85o – 88.74o East) and captured a range of lithologic units, including low accumulation rate hemipelagic deposits and high accumulation rate channel-levee influenced deposits which span at least the last ~1.25 Ma at all sites (Figures 4.1-4.3; Table 4.1; France-Lanord et al., 2016). A fundamental problem investigated by this study is that traditional age-depth modeling approaches for depositional systems like the Bengal Fan have trouble reconstructing accumulation rates that vary by orders of magnitude between interbedded lithologic units. This is especially true when age control at any one site is limited, as was the case for the seven sites in the Expedition 354 8o N transect. One cannot assign a single sedimentation rate prior (e.g. Blaauw and Christen, 2011) as an additional constraint to produce more realistic uncertainty nor should one allow for nearly every possible monotonic age-depth combination (e.g. Haslett and Parnell, 2008), as this ignores first order geologic observations that make a majority of these scenarios unreasonable. It would also be challenging to create age-depth models for each unit independently, as many of the fan units and thin calcareous clay units do not have reliable chronostratigraphic markers nor constraints on their start and end times. To explore accumulation rate variations for the last 1.25 Ma and work towards a complete 8o N Bengal Fan chronostratigraphic framework for future studies, we employ a system specific age-depth modeling approach that is inspired by well-established methods to address uncertainty (Blaauw and Christen, 2011) but also incorporates expert knowledge about the Bengal Fan. Sediments deposited by fan or hemipelagic depositional processes are objectively identified based on physical properties and each lithology is modeled using different strategies. The law of superposition provides additional constraints in fan deposits with few, if any site based constraints, by solving for all seven sites simultaneously and only accepting solutions that respect the stacking pattern defined by Bergmann et al. (in prep.) (Figure 4.2). Our results allow for reconstruction of the time variation in integrated sediment

67 flux at 8o North between 85.85o and 88.74o East and allow us to assess how representative a single site and/or the Expedition 354 transect is for reconstructing past dynamics of the Bengal Fan. Implications for fan evolution and associated sedimentary processes are discussed in greater detail by the companion paper to this work by Bergmann et al. (in prep.), while here we focus on the timing of depositional changes in relation to Pleistocene climate and sea level.

Figure 4.1. Locations of Bay of Bengal and Bengal Fan archives, including the IODP Expedition 354 transect (orange), IODP Expedition 353 U1444 (blue), SO93 cores (yellow), ODP Leg 116 sites (pink), and ODP Leg 121 Site 758 (white).

Table 4.1. IODP Expedition 354 Holes used in this study

Meters used in model/ Hole Latitude (˚ N) Longitude (˚ E) cluster analysis (total penetration) U1449A 8.01 88.11 120 (213.5) U1450A 8.01 87.67 190 (687.4) U1451A 8.01 88.74 110 (582.1) U1452B 8.01 87.18 195 (217.7) U1453A 8.01 86.79 180 (215.7) U1454B 8.01 85.85 170 (161.8) U1455C* 8.01 86.28 110 (949.0) * IODP Site U1455 is the reoccupation of DSDP Site 218, drilled during Leg 22.

4.21 The Pleistocene Bengal Fan Since the late Miocene, sediments have been transported by turbidity currents along channel-levee systems to 8o N (Schwenk and Spiess, 2009), that in at least recent times seem to be sourced to a single shelf canyon and extend the 2500 km length of the fan without bifurcation (Curray and Moore, 1974; Weber et al., 1997; Curray et al., 2003). On Pleistocene timescales, frequent channel avulsions have caused the migration of active deposition across the entire fan (Curray et al., 2003; Schwenk and Spiess, 2009). Bengal Fan channels are deeply incised into underlying deposits and show vertical aggregation and lateral migration (Schwenk et al., 2003, 2005). Fan depositional processes include the overspilling and flow-stripping of channelized turbidity currents, which build channel levees, deposit inter-channel sheet-like turbidites, and/or deposit massive sands (Piper and Normark, 1983; Schwenk et al., 2005). Proximity to the channel-levee system is the primary control on sediment accumulation rate. Accordingly, no single drill site should be expected to contain a complete record of fan deposition on timescales longer than the life of a stable channel-levee system. While it is likely that the highest fluxes of sediment were delivered to the fan during sea-level low stands (Curray et al., 2003), direct radiocarbon dating of late Pleistocene and Holocene levee turbidites (SO93 Cores 117- 120 KL, ~16.5o N) demonstrate continued sediment input through the deglacial sea-level rise and Holocene sea level high stand, with outer levee long-term accumulation rates around 70 cm/ka, but often exceeding 100 cm/ka while sea level rose, and long-term accumulation rates from a location very proximal to the active channel were around 120 cm/ka for the last 9.6 ka (Weber et al., 1997; Hein et al., 2017). Similar radiocarbon-based estimates preceding the deglacial sea level rise are not available at this time. Previous efforts have had difficulty assigning ages and unambiguous relationships of paleo-channel-levee systems, shelf-canyons, and depocenter spatial extents. However, seismic evidence from the upper-lower fan suggests that a series of ‘subfans’, resulting from spatial restriction of the active channel-levee systems, occurred during the Pleistocene (Curray et al., 2003; Schwenk and Spiess, 2009). During periods of local fan inactivity, when sediment routing moves the active fan depocenter elsewhere, calcareous clay sediments

69 drape inactive channel-levee systems and other fan deposits. These types of sediment are well documented in the upper stratigraphy of core SO93-47KL (11.18o N), where turbiditic activity ceased ~0.3 Ma (Weber et al., 2003) and U1452C-1H (8o N), where fan influenced sediments ceased deposition ~0.2 Ma (Weber et al., 2018). These observations provide constraints for the abandonment of local channel-levee systems on the eastern fan before the Late Pleistocene movement of the depocenter to the more western fan documented in seismic data (Subfan D1 of Schwenk and Spiess, 2009). Prior to sediments recovered during IODP Expedition 354 Bengal Fan, all older age constraints for the Pleistocene evolution of the upper-lower Bengal Fan were rooted in the stratigraphy of Deep Sea Drilling Program (DSDP) Site 218, which suffered partial coring and low recovery (59.4 m of sediment from a 773 m hole), limited biostratigraphic constraints, and ambiguity in core-seismic comparison beyond prominent continuous reflectors (Von der Borch et al., 1974). While this was sufficient to get a sense for long-term sediment accumulation rates at that site and to provide first order age estimates of Pleistocene depocenter changes within the context of seismic data (Curray et al., 2003; Schwenk and Spiess, 2009), its chronology could not constrain fan dynamics on 105-106 a timescales which are important for understanding fan evolution in the context of Pleistocene sea level and climate. There are clues that major changes in fan deposition occurred during the Pleistocene. Results from Ocean Drilling Program (ODP) Leg 116 Sites 717 and 719 on the distal Bengal Fan (~1o S) show a significant increase in sediment accumulation rates between the last occurrences of small Gephyrocapsa spp. dominance (~1.02 Ma) and Psuedoemiliania lacunosa (~0.44 Ma), with long term accumulation rates switching from values on the order of 100 cm/ka in the early and middle Pleistocene to on the order of 101 cm/ka after the switch (Gartner, 1990; ages updated according to Gradstein et al., 2012). This accumulation rate change was accompanied by sedimentological changes that could be interpreted as a change in the weathering regime of the Himalayan and flood plain source regions, but, within the context of more recent drilling, the lithologic changes are more likely a reflection of how sediments are transported through the fan (France-Lanord et al., 1993, 2016). Perhaps the best dated complete Pleistocene record from the region is ODP Leg 121 Site 758 on the Ninetyeast Ridge (5.38o N; recently redrilled as IODP Site U1443; Clemens et

70 al., 2016), whose benthic δ18O record (Chen et al., 1995) was used in the construction of the LR04 Benthic δ18O stack (Lisiecki and Raymo, 2005). While no fan deposits are found in the mostly pelagic stratigraphy of Site 758, researchers have argued that changes in the concentration of lithogenic sediments, tracked by magnetic susceptibility (MS), reflect a regionally integrated signal of climate, tectonic, and/or sea level controlled inputs to the Bay of Bengal and Bengal Fan (Farrell and Janecek, 1991; Klootwijk, Gee, Peirce, and Smith, 1992; Prell and Kutzbach, 1997; Zhisheng et al., 2001). Isotopic evidence from ~3-34 ka sediments of the Ninetyeast Ridge suggests that the lithogenic fraction may be a mix of Ganges and Brahmaputra sources, the primary sources to Bengal Fan SO93 Cores 117-120KL levee sediments (Lupker et al., 2013; Hein et al., 2017), and other river systems like the Irrawaddy (Ahmad et al., 2005). While the record suggests an increase in lithogenic sediments from the Middle to Late Pleistocene, it is difficult to understand the processes that drive this signal without context from the fan itself. IODP Expedition 354 reoccupied DSDP Site 218 (U1455) along with six other sites along 8o N (U1449-U1454; Figures 4.1-4.3; Table 4.1) with significantly better recovery and less disturbance than possible in 1972, thanks in part to the advent of the IODP half advanced piston corer (France-Lanord et al., 2016). Fortuitously, all seven sites recovered a regionally extensive calcareous clay unit with good reversal magnetostratigraphy—including the Matuyama-Brunhes Boundary (0.774 Ma), Jaramillo Subchron (1.071-0.990 Ma), and/or Cobb Mountain Subchron (1.208-1.187 Ma; ages according to Channell, Hodell, et al., 2016) and consistent physical properties (France-Lanord et al., 2016; Weber and Reilly, in review) that also appears as a prominent reflector in seismic imaging, providing firm isochron horizons and core-seismic correlation in the Middle Pleistocene (Bergmann et al., in prep.). In the upper part of this unit, some turbidites are intercalated with calcareous clay; however, there were no turbidites recovered at any of the Expedition 354 sites for at least 300 ka, between Marine Isotope Stage (MIS) 37 and 25, about 1.244 – 0.936 Ma (MIS ages according to Lisiecki and Raymo, 2005; Weber and Reilly, in review). The overlying channel-levee sandy turbidite deposits were delivered as a series of progressively larger fan units, separated by three regionally extensive high amplitude reflectors in seismic data, named the Brunhes Aged Reflectors (BAR) 1-3 (Bergmann et al., in prep.) (Figures 4.2 and 4.3). Where core recovery is good, these reflectors appear to be the

71 result of the impedance contrasts between thin calcareous clay sediments and coarse- grained fan sediments. Additionally, Late Pleistocene hemipelagic sediments were recovered in the uppermost stratigraphy at all sites except the westernmost Site U1454 located next to the most recently active channel. Using the Middle and Late Pleistocene hemipelagic units to establish an initial chronostratigraphic framework, we investigate age-depth relationships at all seven sites for the last ~1.25 Ma. This is an interesting period of time to study, as 1.25 Ma marks the start of increased 100 ka frequency amplitude in the benthic δ18O record driven by changes in the behavior of the major ice sheets (Clark et al., 2006), whose transition through the Middle and Late Pleistocene impacts the evolution of glacial-interglacial changes on sea level (Elderfield et al., 2012; Rohling et al., 2014) and monsoonal systems (Clemens et al., 1996; Sun et al., 2006) which have a largely unknown influence on the Bengal Fan sedimentary system.

Figure 4.2. Seismic interpretation of line GeoB97-020/027, as interpreted by Bergmann et al. (in prep), highlighting the sediments studied here. IODP Expedition 354 Sites U1449-U1455 are indicated with vertical pink lines. Channel-levee systems are numbered after Bergmann et al. (in prep) according to their stacking pattern (1=oldest; 40=youngest). Regionally extensive reflectors are highlighted, from oldest to youngest: The Middle Pleistocene Hemipelagic Layer (red shading; lower bound is green line, upper bound is purple bound); the Brunhes Aged Reflectors BAR1 (magenta line), BAR2 (blue line), BAR3 (green line); approximate reflector for the E. huxleyi datums (light blue line); and the base of the Late Pleistocene Hemipelagic Layer (orange line). Vertical exaggeration (VE) is about 400 times.

Figure 4.3. Stratigraphic summary of recovered sediments. Adapted from Weber and Reilly (in prep) and France-Lanord et al. (2016). Ages for important chronostratigraphic markers are noted for ash layers (Mark et al., 2017a), the upper and lower range of calcareous nanofossil datums (Gradstein et al., 2012), and the lower Jaramillo magnetic reversal (Channell, Hodell, et al., 2016). Where recovered, the OTT is always just beneath the Matuyama-Brunhes Magnetic Reversal. Light blue shading indicates the regionally extensive thick Middle and Late Pleistocene hemipelagic units. Seismic estimates of Bergmann et al (in prep.) for the depth ranges of Brunhes Aged Reflectors (BAR) (1 = Purple; 2 = Blue; 3 = Green), which are generally associated with thin hemipelagic units where recovery is good, are also indicated. Question marks indicate uncertainty in tracing the BAR reflectors, including multiple possibilities of where BAR 3 intersects with U1453 and U1449.

4.3 Materials and Methods Model design is discussed in Section 4.3.1 and inputs to the model in Section 4.3.2. For this study, we use the single hole at each IODP Expedition 354 site that best recovered middle to late Pleistocene aged sediments (Table 4.1), as most intervals were only recovered in one hole at each site. Hole recovery, lithology, and some important chronostratigraphic markers are summarized in Figure 4.3. A detailed discussion of all chronostratigraphic constraints is given in Weber & Reilly (in review), building on the results of France-Lanord et al. (2016), and briefly summarized in Section 4.3.2. Core depths in this study refer to the core depth below sea floor-B (CSF-B) scale, which applies a compression algorithm if recovery is greater than 100% to prevent overlap of recovered core with adjacent core intervals. This is necessary for this study, as uncertainty in drilling depth and core expansion led some stratigraphic markers to overlap on the primary CSF-A scale. When working with specific cores in each hole, CSF-B can be converted back to CSF-A using tables available through IODP. X-ray Fluorescence (XRF) data were collected from the surface of u-channel samples (2 x 2 x up to 150 cm plastic tubes) over discrete intervals from Site U1452 to supplement shipboard data using the ITRAX XRF core scanner at the Oregon State University Marine and Geology Repository. Data were collected every 1 mm with 5s count time and a Mo tube.

4.3.1 System Specific Age-Depth Modeling Our age modeling approach is inspired by well-established methods, primarily the BACON method of Blaauw & Christen (2011), and utilizes the ‘t-walk’ algorithm of Christen and Fox (2010), a Marcov Chain Monte Carlo (MCMC) sampler. The algorithm is implemented in MATLAB, using the code of Colin Fox with the corrections by Andreas Nilsson ((Nilsson et al., 2018); available: www.cimat.mx/~jac/twalk/). While the t-walk may be less efficient than other MCMC methods, it requires little tuning and can be applied to a diverse set of MCMC problems by non-experts. We refer the reader to Blaauw and Christen (2011) for a more detailed discussion of the t-walk and how it can be applied to sediment accumulation modeling. Here, we focus on the fundamental principles and highlight the differences between this method and BACON. The fundamental concept presented here is

74 that while we can define a prior distribution hemipelagic sediment accumulation due to good independent age constraints, fan deposits have unconstrained accumulation rates.

In our model, the Expedition 354 holes are broken into a series of sections (c1, c2,… ci) between depth intervals (d0, d1, d2,… di) with section width (Δci). Fan deposits are nominally divided into 10 m intervals, while calcareous clay deposits are divided into 1 m intervals. As each lithostratigraphic unit is not perfectly divisible by 10 m or 1 m, section widths are allowed to be smaller when needed. Sediment accumulation (A) in section i is modeled as:

퐴푖 = (1 − 푤푖)훼푖 + 푤푖훼푖+1 (Equation 4.1) where α is a randomly sampled sediment accumulation value and w is a memory parameter that determines the dependence of α on the preceding interval, where w = 0 means αi is independent and w = 1 means αi is equal to αi+1. In our model, w is always equal to zero at the base of the section, at lithologic transitions, and in fan deposits. Because we use unevenly spaced depth intervals, and w is dependent on interval spacing, for any interval wi

ΔCi = wp , where the memory has an exponential relationship with the lag (i.e. Δci in cm). In the calcareous clay sediments, wp is the memory value randomly selected according to the prior beta distribution (Beta(aw, bw)) (see Blaauw and Christen, 2011). We require that all values of w be between 0 and 1 and that all α values, for fan or hemipelagic deposits, are positive (i.e. monotonic accumulation). For hemipelagic sediments, we use a memory mean of 0.7 and strength of 30.

Sediment age at any depth (tD) is calculated as:

퐷 푡퐷 = ∑푑=1 ∆푐푑 ∗ 퐴푑 (Equation 4.2)

Given that the sediment water interface is at age zero.

We provide a prior in which hemipelagic sediments αh are described by a gamma distribution (Gamma(aαh, bαh)) (after Blaauw and Christen, 2011) while fan sediments are left unconstrained. Based on observations where age control is strong, we use accumulation mean of 3 cm/ka and 2 accumulation shape to define the distribution.

An ‘energy’ function is used as the input to the t-walk algorithm as U(α, w | y) = -log f(α, w | y1, y2,… yi) which represents the -log of the posterior distribution, where y are the magnetic reversal, tephra, and orbital tuning age constraints with uncertainty (σ) at depth

(D1, D2,… Di). In this case:

푈(훼, 푤 | 푦) = −(1 − 푎푤) log 푤푝 − (1 − 푏푤) log(1 − 푤푝) − 퐼 ∑푖=1(1 − 푎훼ℎ) log 훼푖ℎ + 푏훼ℎ훼푖ℎ − (Equation 4.3)

퐽 2 푦푗 − 푡퐷푗 ∑ log 휎푗 + ( ) /2 휎푗 푗=1

This ‘energy’ function differs with respect to BACON, as we have relatively simple probability distribution for our age constrains (i.e. we assume all normal distributions, rather than calibrated radiocarbon dates) and we set the sediment water interface to a constant value of 0 ka, thus the three lines of Equation 4.3, relate to fitting the memory, hemipelagic accumulation rates, and age constraint priors, respectively. To ensure that our solutions are geologically meaningful, we implement a few additional hard constraints, where if violated, the ‘energy’ function is set to -Inf and the iteration is rejected. First, all solutions must respect the maximum limiting date provided by the first occurrence of E. huxleyi, where only solutions that the first occurrence is younger than 290 ka are accepted. Second, all solutions must respect the channel-levee stacking pattern of Bergmann et al. (in prep.). This is achieved by solving for all seven sites simultaneously using the t-walk algorithm and rejecting any solution by setting the ‘energy’ function to -Inf where the base of a stratigraphically higher channel-levee system is modeled as being older than the top of a channel-levee systems that is stratigraphically lower.

4.3.2 Lithostratigraphic and Chronostratigraphic Inputs for Age-Depth Modeling Sediments recovered in the upper 200 m of each site were categorized as hemipelagic or fan (i.e. channel-levee, interlevee, and sand deposits) deposits by fitting a Guassian mixture model cluster analysis to three physical properties measured on ship: magnetic susceptibility (MS), natural gamma radiation (NGR), and sediment lightness (L*).

We removed measurements made within 3 cm from the edge of each section and normalized magnetic susceptibility and NGR by gamma ray attenuation (GRA) estimated wet bulk density using the method of Walczak et al. (2015) by smoothing all records to the same resolution using a Full Width at Half Maximum (FWHM) Gaussian filter of 6 cm and resampling every 5 cm. While L* was not normalized, we also filtered and resampled to allow for direct comparison. Intervals that were not recovered by coring were categorized as hemipelagic or fan deposits based on the closest overlying or underlying sediments recovered. If both agreed, all sediments in between were classified as being the same. If both disagreed, the depth of the contact was assigned at random. Bergmann et al. (in prep.) discuss the identification of channel-levee deposits and prominent reflectors in detail. We use their interpretation and number scheme of the channel-levee stacking pattern at 8o N (Figure 4.2). Their integration of cores and seismic data was conducted with constant velocities ranging from 1640-1680 m/s. Velocities were determined by the best fit of prominent reflectors with lithologic boundaries and physical properties. Velocities were also cross-checked with the shipboard measured p-wave velocities. While there are uncertainties in any time-depth conversion, broad agreement in the calculated depths of the high amplitude reflectors associated with calcareous clay and fan contacts and drilling depths lends confidence in their use in this study. We use age constraints from Weber and Reilly (in review) which establish an orbital resolution chronology in middle to late Pleistocene calcareous clay sediments through correlation of L* variations to the benthic δ18O timescale of Lisiecki and Raymo (2005) within the framework of biostratigraphic and magnetostratigraphic constraints. Along the Expedition 354 transect and other sites from the lower Bengal Fan, including cores with expanded calcareous clay lithologies SO93 Cores 22KL, 28KL, and 47KL, L* primarily reflects the relative proportion of biogenic calcareous to lithogenic sediments (Weber et al., 2003). At Site U1452, Late Pleistocene calcareous clay sediments were studied in detail and display L* variations on orbital timescales along with planktonic δ18O and other proxies that reflect marine productivity and sediment composition (Weber et al., 2018). In our model, we assign an uncertainty of 7 ka (1 σ) to account for uncertainties in correlation (human uncertainty), phase (geologic uncertainty) with the benthic δ18O reference template.

We include additional constraints from magnetic reversals, using the timescale of Channell et al. (2016), which is based on direct benthic δ18O and magnetic comparison in North Atlantic IODP Site U1308 and provides stronger constraints on the timing of the Cobb Mountain Subchron with respect to the benthic δ18O timescale than the Geologic Timescale of Gradstein et al. (2012) used in the Expedition 354 Proceedings (France-Lanord et al., 2016). These tuned reversal ages are assigned uncertainty of 5 ka. Two tephra layers recovered at most sites are interpreted as the Youngest Toba Tephra and Oldest Toba Tephra (France-Lanord et al., 2016; Weber et al., 2018; Weber and Reilly, in review). We assign the 40Ar/39Ar dates for these events with their analytical uncertainty, 73.7 ± 0.3 ka and 785.6 ± 0.7 ka, of Mark et al. (2017a) as additional age control points. Biostratigraphic datums are treated with caution, as the last occurrence of P. lacunosa (440 ka) was found in the same samples as E. huxleyi (first occurrence at 290 ka) in multiple holes (Holes U1450A and U1454B), suggesting reworking of very old materials in fan deposits, and many turbiditic and sandy samples were barren, making it challenging to establish first and last occurrence depths with confidence (all nannofossil ages after Gradstein et al., 2012). While these issues likely do not impact calcareous clay sediments, until further biostratigraphic work refines shipboard estimates, the biostratigraphic markers cannot be used as tie points for establishing a 105-106 a resolution chronology. In our model, however, we do include the first occurrence of E. huxleyi as a maximum limiting date, because even if the nannofossil is reworked from its actual first occurrence the interpretation doesn’t change (i.e. if a sample contains E. huxleyi, the sample must be less than 290 ka old). When additional constraints become available, they can be incorporated to improve the age-depth model results.

4.3.3 Stacking Model Results We build two stacks using the model results to represent the regional signals captures by the Expedition 354 8o N transect. Sediment accumulation rates were calculated for each of the age-depth combinations at each site over 1 ka intervals to capture 1 ka and greater changes in long term accumulation rates. These were used to calculate statistics for each site and for the transect as a whole. Similarly, a magnetic susceptibility stack was created by applying the median age depth relationship for each site to logarithmically transformed shipboard point magnetic susceptibility log(MS) records and resampling every 1

78 ka, after applying a smoothing Gaussian filter with FWHM of 2.5 ka. These smoothed and resampled log(MS) records were then stacked to extract a signal that is representative of the whole transect.

4.4 Results

4.4.1 Lithostratigraphy Results of the Gaussian mixture model are summarized in Figures 4.4 and 4.5. We choose a three cluster solution, as it does a good job of describing the data in a geologically meaningful way without overfitting the data. In comparison to the shipboard descriptions of primary lithology, cluster 1 was mainly described as calcareous clay, cluster 2 was mainly described as sands, and cluster 3 was described as a mixture of lithogenic clays, silts, and sands. Accordingly, we interpret cluster 1 as describing hemipelagic sediments and clusters 2 and 3 as describing fan sediments, such as massive sands deposited in sand lobes and turbidites (e.g. Schwenk et al., 2005). MS, which tracks the concentration of magnetizable material, is generally highest in sandy fan sediments (cluster 2) over muddy (clayey-sandy) turbidites (cluster 3), suggesting a particle size control on magnetic susceptibility values, consistent with previous observations for Bengal Fan sediments (Weber et al., 2003). Dilution of the relative proportion of lithogenic sediments by biogenic sediments is likely the primary control on the low magnetic susceptibility values in hemipelagic sediments (cluster 1). It is possible, but difficult to assess at this time, that there are either bedrock or weathering source controls on MS as well, as was suggested in the study of distal fan mud turbidities (Sager and Hall, 1990) and Bay of Bengal slope sediments (Phillips et al., 2014). L* values are similar for both fan sediment clusters, but exhibit a much greater range in hemipelagic sediments, reflecting orbital timescale controls on sediment composition in the hemipelagic sediments, as documented elsewhere (Weber et al., 2003, 2018). NGR, which tracks the amount of radioactive materials, is likely on first order related to the relative concentration of lithogenic materials and on second order related to the relative proportion of K, U, and Th bearing minerals in the lithogenic fraction. There is a strong negative correlation with L* in hemipelagic sediments, suggesting this reflects variations in biogenic and lithogenic

79 contributions. Conversely, there is a strong positive correlation in NGR and MS in sandy fan sediments, suggesting whatever is controlling the concentration of magnetic minerals is also controlling the concentration of K, U, and/or Th bearing minerals.

Figure 4.4. Comparison of physical properties and results of the cluster analysis. Natural gamma radiation (NGR) and magnetic susceptibility were normalized by gamma ray attenuation (GRA) estimates of bulk density following the method of Walczak et al. (2015).

Figure 4.5. Comparison of cluster analysis results with shipboard visual core descriptions. Note cluster 1 (a) was primarily described as calcareous clay and can be interpreted as hemipelagic sediments while clusters 2 (b) and 3 (c) were mostly described as various mixtures of lithogenic clays, silts, and sands and can be interpreted as fan sediments. Box and whisker plots (d-e) show the median (horizontal red line), range of the 2nd and 3rd quartile (box), and 99.3% range (whiskers) for the three physical properties used in the cluster analysis.

Figure 4.6. Shipboard measured physical properties, sediment lightness (L*) and magnetic susceptibility (MS), compared with the XRF ratio of Ca to Ti for hemipelagic sediments and their contacts with fan sediments at Site U1452. These contacts include the base of the Late Pleistocene hemipelagic layer (LPHL), the hemipelagic layer associated with Brunhes Aged Reflector 3 (BAR3), and the Middle Pleistocene hemipelagic layer (MPHL). Shipboard lithologic descriptions and images are included for reference. There is a stark contrast between high MS and low Ca/Ti in fan deposits and low MS and high Ca/Ti in hemipelagic sediments. L* highs in hemipelagic deposits can be used to identify sediments deposited during interglacial times (Weber et al., 2018). Blue shading indicates the hemipelagic sediments deposited with little influence from the fan, which is different than shipboard descriptions of calcareous clay sediments.

4.4.2 Age-Depth Models Following Blaauw and Christen (2011), we only sample every 100th realization of the t-walk. Goodness of fit relative to the prior distribution is tracked by the -log of the energy function (Equation 4.3). Since the initial conditions poorly fit the prior distributions, a set of saved iterations need to be removed, or burned-in, as the t-walk converges on solutions with good fit. We burn-in 15,000 of the saved iterations based on where the -log of the energy function seems to stabilize, leaving 70,284 iterations, or age-depth combinations that respect stratigraphic relationship for each site of the transect, to use in our analysis (Figure 4.7). As expected, based on our model inputs, sediments identified as hemipelagic by the cluster analysis have low accumulation rates, while fan sediments have higher accumulation rates.

Accumulation rates are widely variable on our model outputs, reflecting a combination in the actual accumulation rate variability and uncertainty of our model itself. We calculate the distributions of accumulations rates for hemipelagic and fan deposits, based on the designation from our cluster analysis results (Figure 4.8). Median accumulation rates for hemipelagic deposits range from 2.1-6.0 cm/ka for the seven sites and distributions range from about 100 – 101 cm/ka. The lowest median accumulation rates are at U1454 and U1449, the latter of which is the site with best chronologic control in the Middle Pleistocene and longest recovered sequence without intercalated turbidites, due to a thick sequence recovered in Core 18H that exhibits little turbiditic influence following the Jaramillo Subchron and Matuyama-Brunhes boundary relative to other sites. High median hemipelagic accumulation rates were modeled for U1450. This is likely in part a result of drilled intervals 33I and 35I (Figure 4.3), which limited recovery of Middle Pleistocene hemipelagic sediments and associated age control points—notably that U1450 was the only site that didn’t recover the Matuyama-Brunhes magnetic reversal or the Oldest Toba Tephra (France-Lanord et al., 2016). Other sites, like U1452, have distributions that include higher hemipelagic accumulation rates, were noted to have frequent fine grained turbidites intercalated with calcareous clay sediments near the top of the Middle Pleistocene hemipelagic layer.

Figure 4.7. Age-depth models with median age (black line) and 99% confidence intervals (gray shading) for all IODP Expedition 354 sites. Representative stratigraphy for fan (yellow) and hemipelagic (blue) deposits are included for reference. Levee systems of Bergmann et al. (in prep) highlighted in darker yellow and numbered as in Figure 4.2. A burn-in of 15,000 model runs was used to remove early iterations that have poor fits to the prior distribution, tracked by lower values of the - log of the energy function (Equation 4.3).

Fan deposit accumulation rate distributions are much wider, ranging from about 101-104 cm/ka. While we expect accumulation rates to be widely variable in these sediments, much of this spread is related to not using a fan accumulation rate prior in our model. Median accumulation rates range from 41.7 to 159.7 cm/ka, with the spatial pattern of median accumulation rate following the same pattern as seismic stratigraphic thickness between U1454 and U1449, with the thickest deposits found between the 85 East Basement Ridge and Ninetyeast Ridge (Schwenk and Spiess, 2009). The spatial pattern of median modeled sediment accumulation rates, calculated over 1 ka time intervals, indicate that earliest episodes of high, >101 cm/ka, long term accumulation rates occurred near the center of the transect following deposition of the Middle Pleistocene hemipelagic unit (Figure 4.8). These high accumulations are modeled earliest at U1450, which is consistent with the stratigraphically oldest channel-levee systems identified above the Middle Pleistocene hemipelagic layer in seismic imaging (Figure 4.2); however, as mentioned earlier, U1450 was also the only site to not recover the Matuyama- Brunhes Boundary and accordingly has worse age control than the other sites. High accumulation rates spread outwards, following the stacking pattern defined by Bergmann et al. (in prep), and ultimately switches to the highest accumulation rates at our westernmost site, U1454, between 200 – 300 ka. Site U1455 is the only site that experiences very high accumulation rates near the base of the Middle Pleistocene hemipelagic unit, likely related to incomplete recovery of the Middle Pleistocene hemipelagic unit leading to no constraints from the Cobb Mountain Subchron (France-Lanord et al., 2016).

Figure 4.8. Resulting sedimentation accumulation rates from the age-depth models presented in Figure 4.7. Top: Sediment accumulation rate distributions for fan deposits (orange) and hemipelagic deposits (blue) with median values in cm/ka indicated as horizontal lines for each site. Bottom: Median modeled accumulation rates calculated in 1 ka time increments plotted as an interpolated surface against latitude and time. The LR04 Benthic δ18O stack (Lisiecki and Raymo, 2005) with the magnetic polarity timescale (Channell, Hodell, et al., 2016) is plotted for reference, with greater δ18O values indicating times of increased ice volume, cooler deep ocean temperatures, and lower sea level. The blue shaded interval represents the time period where no turbidites were observed along the 8o N transect and age control is best (Weber and Reilly, in review). High accumulation rates at U1455 at the base of this unit is likely related to incomplete recovery of the Middle Pleistocene hemipelagic layer at this site (France-Lanord et al., 2016). The gray shaded interval indicates the interval where the depocenter is focused on the western fan, as recognized in seismic data (Schwenk and Spiess, 2009).

4.5 Discussion

4.5.1 Assessing the Ages of Regionally Extensive Brunhes-Aged Reflectors (BAR) Bergmann et al. (in prep) identify a series of three regionally extensive high amplitude reflectors above the thick Middle Pleistocene hemipelagic unit, BAR 1-3 (Figure 4.2 and 4.3). Often, where core recovery was good, these reflectors correlate with the locations of thin calcareous clay units, suggesting that the impedance contrast between hemipelagic and fan deposits create these reflectors and they represent time intervals of widespread low accumulation along the 8o N transect. L*, XRF geochemistry, and MS found in the unit that correlated with the youngest of the three reflectors, BAR-3, are consistent with a transition from glacial (Low MS, Low L*) to interglacial sediments (Low MS, High L*) with a sharp upper contact that could be erosional (Figure 4.6). It is more difficult to interpret the nature of sediments in BAR-1 and BAR-2, but this could be related core recovery or erosion of the upper hemipelagic surface. Although it is difficult to know the exact nature of these deposits, the presence of these regionally continuous reflectors suggest fan sediments deposited following deposition of the Middle Pleistocene hemipelagic unit were delivered in a series of pulses, with each pulse of fan activity delivering a larger volume of sediment (Bergmann et al., in prep.). We investigate the ages and potential timing of each BAR so that we can discuss possible fluxes of sediment to 8o N during each of the associated fan units. To do so, we calculate the age distributions from our age models of the predicted depth range (deepest estimate to shallowest estimate), according to Bergmann et al. (in prep.), of each BAR where they are observed to cross an IODP Expedition 354 Site (Figure 4.9). Age distributions are often wide, illustrating the large uncertainty of ages in fan sediments between the thick regionally extensive middle and late Pleistocene hemipelagic units; however, most age distributions are within uncertainty of each other. U1450 is an exception to this with BAR1 significantly older than estimates at other sites. This may be the result of limited recovery and fewer age control points in the Middle Pleistocene as discussed in Section 4.4.2.

Figure 4.9. Age distributions of Brunhes Age Reflectors (BAR) 1-3, calculated from all ages in the interval between the uppermost and lowermost depth estimates of Bergmann et al. (in prep) based on seismic data. The bottom panel includes the sum probability distribution of all observations. The LR04 benthic isotope stack with interglacial marine isotope stage (MIS) numbers is included for comparison (Lisiecki and Raymo, 2005).

The BAR3 age distribution only spans one major glacial-interglacial transition and was likely deposited around the transition from glacial MIS 8 to interglacial MIS 9, based on sedimentological observations from U1452 (Figure 4.6). BAR1 at U1452 and U1453 seems to be most consistent with being deposited during MIS 13. This is consistent with Bergmann et al (in prep.) tracing the layer to merge with the uppermost Middle Pleistocene hemipelagic unit at U1449, which experiences little turbiditic influence and can be dated to at least MIS 15, with additional hemipelagic sediment above that marker (Weber and Reilly, in review). Given that BAR1 and BAR3 are likely associated with interglacial sea level high stands, it is tempting to assign BAR2 to MIS 11. However, even though BAR2 maybe within uncertainty of MIS 11 in some instances of the age-depth modelling exercise presented here, there is little age control around this time and no direct support for this scenario at present.

4.5.2 Stacking Expedition 354 Records to Establish a Regional Signal As each Expedition 354 site has imperfect recovery, an incomplete record of fan deposition, and significant uncertainty in its age-depth model, we explore the regional signal

88 spanning the 320 km of the transect at 8o N. To do so, we build two data stacks, one for sediment accumulation rates and one for MS, using the age models presented here. These stacks are meant to simulate an integrated signal across 8o N from the most recently active channel to the west flank of the Ninetyeast Ridge. This signal would not include sediments deposited further to the west, which has been the main depocenter of the Bengal Fan since the establishment of the most recent subfan in the Late Pleistocene (Curray et al., 2003; Schwenk and Spiess, 2009; Bergmann et al., in prep.). While the Nicobar Fan may have been an important depocenter for parts of the Neogene, a prominent reflector that marks the end of high accumulation rates on the Nicobar fan was dated to the Early Pleistocene (McNeill et al., 2017), meaning the Nicobar Fan was likely not a significant sink for sediments during the time period discussed here. Our sediment accumulation rate stack median and 1 σ intervals are plotted in Figure 4.10. While median values give a sense for the transect wide changes in sediment accumulation, the variance could illustrate greater uncertainty in our age-models or increased variability between sites. Successions of active channel-levee systems with limited chronostratigraphic constraints would contribute to each of these interpretations. Our magnetic susceptibility stack shows similar changes, reflecting low values in the low accumulation hemipelagic deposits and higher values as lithogenic contribution increases and particles sizes become coarser. Similar to the accumulation rate stacks, increased variance can reflect increased chronologic uncertainty or increased variability between sites (Figure 4.10). We find that following the transect wide Middle Pleistocene Hemipelagic Layer, where no turbidites were recovered between MIS37 and MIS25 (1.244 – 0.936 Ma), sediment accumulation rate variance increased through about MIS17 (0.676 Ma) reflecting the onset of turbidite deposition at some sites along the transect. Accumulation rate median and variance begin to increase significantly following MIS17, reaching peak levels between 0.3 and 0.5 Ma before movement of the depocenter west of the 85 East Basement Ridge between 0.2 and 0.3 Ma.

Figure 4.10. Stacked Bengal Fan sediment accumulation rates and magnetic susceptibility at 8o N in the context of regional and global records. From top to bottom: The LR04 benthic δ18O stack which is a signal of deep sea temperature and global ice volume (Lisiecki and Raymo, 2005); Global sea level estimates from temperature corrected benthic δ18O (Elderfield et al., 2012; Rohling et al., 2014); Magnetic susceptibility from the Chinese Loess Plateau which is a signal of East Asian summer monsoon intensity (Sun et al., 2006); Heqing Lake Indian summer monsoon (ISM) Index which used a stack of proxies sensitive to temperature, precipitation and weathering in the lake’s watershed (Zhisheng et al., 2011); ODP Site 722 median grain size which tracks wind strength of the Indian summer monsoon in the Arabian Sea (Clemens et al., 1996; Clemens, 1998) Sedimentation rate stack in cm/ka median (line) and 1σ interval (red shading) of the seven Expedition 354 sites at 8o N on the Bengal Fan; Magnetic susceptibility stack mean (line) and 1σ interval (orange shading) of the seven Expedition 354 sites at 8o N on the Bengal Fan; Site 758 magnetic susceptibility (Farrell and Janecek, 1991) which has been interpreted as an integrated signal of sediment flux to the Bay of Bengal and/or Bengal Fan, with tephra layers removed and smoothed using a 40 ka FWHM Gaussian filter; and range for the possible age of Pleistocene sedimentation rate increases at Sites 717/719 to the southwest of the Expedition 354 transect (Gartner, 1990). Blue highlighted box indicates the time interval where no turbidites were recovered along the Expedition 354 transect and Middle Pleistocene chronology is best resolved (Weber and Reilly, in review). Gray shading indicates when the depocenter was focused on the more western fan, as recognized in seismic data (Schwenk and Spiess, 2009).

4.5.3 Insights to the Middle to Late Pleistocene Evolution of the Bengal Fan We compare our 8o N stacked records with what has been argued, but not demonstrated, to be a regionally integrated record of sediment input to the Bay of Bengal and Bengal Fan (Figure 4.10). We use the Site 758 (5.38o N) magnetic susceptibility record after removing magnetic susceptibility highs associated with tephra layers A through I (Dehn et al., 1991; Farrell and Janecek, 1991) with chronology assigned using the benthic δ18O record of Chen et al. (1995) aligned by Lisiecki and Raymo (2005). The magnetic susceptibility is smoothed using a 40 ka FWHM Gaussian filter and resampled every 1 ka to highlight the lower frequency signal. We find broad similarity with our reconstruction, including low magnetic susceptibility values coeval with the Expedition 354 Middle Pleistocene hemipelagic layer, an increase in magnetic susceptibly coeval with a general trend of increased variance and median accumulation rates across the Expedition 354 transect, and a decrease in magnetic susceptibility coeval with movement of the Bengal Fan depocenter to the west and more distal to the Ninetyeast Ridge. While it is difficult to make direct comparisons to the ODP Leg 116 Distal Bengal Fan drill sites, we note that increases in Pleistocene sediment accumulation rates at distal Bengal Fan Sites 717 and 719 (~1o S) must have occurred during the development of the Middle to Late Pleistocene channel-levee systems along our transect and/or subsequent shift of the depocenter to the more western fan at 8o N. Our observations suggest a major change in sediment routing on the fan during the Middle to Late Pleistocene that could be interpreted as the development of a more extensive channel-levee system after about 0.936 Ma. This is supported by multiple lines of evidence documenting regional changes following deposition of the regionally extensive Middle Pleistocene hemipelagic unit at 8o N and subsequent increases in magnetic susceptibility and accumulation rates. First, based on seismic observations of the BAR horizons along 8o N, fan sediments were delivered as a series of progressively larger pulses (Bergmann et al., in prep.). While there are still uncertainties in the timing of these pulses, seismic unit thickness illustrates each fan building episode was larger than the last, suggesting either closer proximity to the center of the active depocenter or a longer duration of channel-levee activity. Second, sites from the very distal fan see an increase in

91 accumulation during the time in which progressive episodes of fan building at 8o N become larger. And finally, the long-term pattern of sediment accumulation on the eastern lower fan at 8o N is remarkably similar to the long-term pattern in lithogenic concentration at the closest pelagic reference site, supporting earlier claims that this signal reflects an integrated record of sedimentation and sediment routing on the Bengal Fan, and likely, for this timeframe is influenced by processes occurring on the eastern lower Bengal Fan. It is important to note that these observations on their own do not necessarily make the Late Pleistocene unique, as channel levee systems were present at 8oN during times since the Late Miocene (Schwenk and Spiess, 2009; France-Lanord et al., 2016) and higher accumulation rates are observed at distal fan sites during Miocene and Pliocene times (Gartner, 1990). However, the Middle to Late Pleistocene is unique in that we have better age-control and can discuss this specific change within the context of regional and global records. In general, long term accumulation rates are controlled by changes in sediment supply, sediment transport, and accommodation space. It is always possible that the changes we observe at 8o N are entirely controlled by stochastic fan sediment transport processes. However, long term changes in sediment supply to the Bay of Bengal is likely influenced by long term trends in climatically or tectonically driven erosion and sea level variations influence continental shelf accommodation space and the connection between the major Himalayan river systems and the deep ocean. While evidence from the Chinese Loess Plateau indicate strengthening of interglacial summer monsoons and associated precipitation from the Middle to Late Pleistocene (Sun et al., 2006), this change is not in sync with where we observe our initial increases in accumulation rate, with the largest intensifications of the East Asian summer monsoon around 1.25 and 0.6 Ma. It also may not be representative of the Indian summer monsoonal system (Figure 4.10). For example, grain size changes in the Arabian Sea that reflect the Indian summer monsoon wind strength (Clemens et al., 1996; Clemens, 1998) and precipitation and weathering proxies from Heqing Lake in Southern China (Zhisheng et al., 2011) suggest that Indian summer monsoon may have weakened during the Late Pleistocene as Northern Hemisphere glaciation intensified.

We do observe, however, that the onset of the Middle to Late Pleistocene growth of the Bengal Fan channel-levee system, as detected at 8o N following 0.936 Ma, occurs during the first very low sea level stand of the Middle to Late Pleistocene, as recognized in the ice volume signal of temperature corrected benthic foraminifera δ18O (Elderfield et al., 2012; Figure 4.10) and sea-level controlled hydrologic cycle signals in Mediterranean Sea planktonic foraminifera δ18O (Rohling et al., 2014). Based on this observation, we hypothesize that the evolution of the Middle to Late Pleistocene Bengal Fan was driven by the increase in glacial-interglaciation sea level amplitude, which strengthened the connection between Himalayan Rivers and the Bengal Fan. However, changes in Pleistocene climate around the same time could have also strengthened the erosional regime in the Himalaya Mountains and increased sediment flux to the Bay of Bengal. Further work is required to investigate these hypotheses.

4.6 Conclusion We investigate the evolution of the Middle to Late Pleistocene Bengal Fan at seven sites along the IODP Expedition 354 8o N transect using a system specific age modeling approach that assesses uncertainty and incorporates expert knowledge of the fan. By quantifying sediment accumulation rate variations from the Middle to the Late Pleistocene, we find an increasing trend in accumulation rates following a 300 ka interval (~1.244 – 0.936 Ma) where no turbidites were observed across the 320 km transect. This observation, along with seismic observations, accumulation rates on the distal fan, and lithogenic sediment concentrations at a pelagic reference site, indicate development of the Bengal Fan channel- levee system during this time to a distribution network that covers a greater extent of the fan. The onset of this change is coincident the first major sea level low of the Middle to Late Pleistocene, leading us to hypothesize that increased sea level amplitude changes associated with amplification of polar glaciation was the major driver of Pleistocene Bengal Fan changes.

4.7 Acknowledgements We are very grateful to the captain, crew, IODP staff, and shipboard scientists on the JOIDES Resolution that made IODP Expedition 354 and subsequent research successful.

All samples and data were provided by IODP. This work was made possible by support from a United States Science Support Program post expedition award to BTR. We thank Leslie & Mark Workman and the Oregon ARCS Foundation for additional support to BTR. This study benefited greatly from a discussion with Andreas Nilsson, who recommended the use of the T-walk algorithm.

5. Conclusions

5.1 Chapter Summaries In this dissertation we explored three studies where networks of sedimentary archives can be used to learn about the history of Earth systems. Specific problems were addressed using paleomagnetism, which allowed for more direct comparison of seemingly disparate geologic archives, in concert with stratigraphic correlation, geochronological methods, and sedimentological methods.

5.1.1 Chapter 2: Using the Past to Better Understand Future Changes to one of Greenland’s Major Outlet Glaciers Through the study of the stratigraphy of Petermann Fjord, we learned that while the historical record suggests long-term stability of the Petermann Ice Tongue, the geologic record clearly shows that following regional deglaciation a stable ice tongue was not established until the late Holocene. A well-defined record of paleomagnetic secular variation allows for correlation to well-dated records from the Chukchi Sea and Northern North Atlantic, providing a means to objectively assess radiocarbon uncertainties and assign ages to glaciological transitions. These ages can then be compared to independently dated records of atmospheric temperature and sea ice conditions that are not complicated by marine reservoir issues and provide a perspective of the paleoenvironmental conditions necessary to maintain a stable ice tongue. Our findings indicate that recent human-driven environmental changes have already passed atmospheric temperature thresholds for the ice tongue’s stability.

5.1.2 Chapter 3: Integrating Archives and Methods to Better Understand Uncertainty By comparing the common paleomagnetic signals from a number of sedimentary basins around Western North America, we are able to assess uncertainties in late Pleistocene chronologies. We define a regional geomagnetic signal and provide a timescale for this signal that considers uncertainties related to the magnetics and radiocarbon. These

95 analyses provide stronger constraints on the ages of lithologic transitions and helps identify and/or confirm radiocarbon dates that are too old. We use this geomagnetic signal to provide a stratigraphic context for the age of a magnetic excursion recorded in the sediments of the Wilson Creek Formation at Mono Lake, CA. Our findings indicate that either the excursions is either in good agreement with another regionally and globally recognized excursion between 30 and 34 ka or the excursion is older and there is a significant drop in accumulation rates and/or a hiatus that complicates the paleomagnetic correlations.

5.1.3 Chapter 4: Extracting a Regional Signal from a Complex Depositional System Through correlation of seven drill sites, integration of seismic observations, and making assumptions about sediment accumulation rates in various lithologic units, we establish a chronology for complex and difficult to date sediments. While any one of these sites primarily reflects the local manifestation of a complex depositional system, considering all sites together and comparing to other regional records indicates that there were major changes to this depositional system during the Pleistocene. Our findings suggest growth of the spatial extent of the Bengal Fan’s channel-levee system over the middle to late Pleistocene that occurred along with intensification of glaciations and increased interglacial- glacial sea level change amplitudes. These observations will aid in the interpretation of older Bengal Fan sediments, recovered at fewer sites, and have implications for understanding the history of organic carbon burial in the deep sea.

6. Appendices

Appendix A. Supplementary Materials for Past Collapse and Late Holocene Reestablishment of the Petermann Ice Tongue, Northwest Greenland