Detecting Tropical Cyclone Signals in Tree Rings of Longleaf Pine (Pinus palustris Mill.), Valdosta, Georgia, U.S.A.

Transcription

1 Detecting Tropical Cyclone Signals in Tree Rings of Longleaf Pine (Pinus palustris Mill.), Valdosta, Georgia, U.S.A. A Thesis Presented for the Master of Science Degree The University of Tennessee, Knoxville Savannah Anne Collins December 2014

2 Copyright 2014 by Savannah Anne Collins All rights reserved. ii

3 DEDICATION I was blessed with the greatest parents I could have ever had. One week before I began my Master s program, we lost my stepfather, Dan. He was a resounding rock of love and encouragement through my undergraduate career, and continues to be an inspiration through my graduate work. I miss him every day. I could not have gotten this far in my education and life without the love and support of my parents. Therefore, this thesis is dedicated with humility and gratitude to them: Mark, Retha, Sharon, Dave, and Dan. iii

4 ACKNOWLEDGEMENTS First and foremost, my greatest gratitude goes to my advisor, Dr. Henri Grissino- Mayer, not only for the knowledge, data, and prolific editing he provided during this process, but for being my most prevalent professional advocate and enthusiast, and for offering me counsel and guidance whenever I was falling apart and needed reassurance. I would not be in graduate school at all without the confidence you had in me. To my committee members Dr. Sally Horn, who always had fantastic advice and light- hearted words of encouragement, and Dr. Kelsey Ellis, for her delightful enthusiasm, terrific ideas, and introducing me to coffee. To Dr. Liem Tran, for his support in the department and instilling in me that the knowledge I walk away with is the most important impression to gain from this experience. To Dr. Nicholas Nagle and Gengen He, for finally getting statistics through to me and making it enjoyable, and introducing me to R, both of which were essential to this thesis. To my incredible graduate department: Anne Meltzer, Sarah (& Martin) Lewis- Gonzales, Anna Alsobrook, Vi Tran, Matthew Kerr, Lauren Stachowiak, Sarah Jones Wayman, Maegan Rochner, Neil Conner, Matthew Cook, Jose Izquierdo, Ruth Bowling, Mathew Boehm, Tyler Sonnichsen, Joanne Ballard, Julie McKnight, Rachel Craig, Joseph Roberts, Helen Rosko, Adam Alsamadisi and many others I was fortunate enough to work with, for their endless and warmhearted support of myself and each other through this arduous process. The faculty and students in this department are phenomenal, and it was a blessing to work with them. To my loving and wonderful family immediate and extended for supporting and assisting me the last several years. To my little sisters, Rachel, Amanda, and Cate, for letting me iv

5 vent to them on a regular basis and for always giving me sound guidance. To my Opa, Dr. Bob Collins, for habitually sending me every article on weather and climate change he has ever come across, some of which made it into this thesis. To my father, Dr. Mark Collins, for his support and camaraderie during the most eccentric two years of my life. To my mother, Retha Alexander Shreve, for her limitless and enthusiastic support, even from several states away. To Dustin Walker and the management and staff at Lakeside Tavern, for amiably accommodating my ever- changing schedule; to Stephanie Wheatley, for providing me with a wonderful path to teaching; and to the Fox and DeLaura families, who made my education a priority along with their children s educations, and for graciously cooperating with my hectic schedule. To Tara and Philip Rhodes, Melissa Rogers, Sarah Wegman, Whitney Kaul, Jamie Wright, Earl Ray Rader, Jordan Stephens, Dave Gotthold, Matthew Davenport, Rebecca Cherry, Crystal Brand, and my therapist Steven Matheny for the years of friendship and encouragement. And finally, to my Key people, Jeffrey and Amanda, for allowing me to practice presentations on you, taking care of me when I was ill, forcing me to spend the holidays with you when I would otherwise be alone, but most of all, for anchoring my sanity on a daily basis; it is safe to say I would not have succeeded during the last two years without your friendship and never- ending support. An enormous thank you to all of you. v

6 ABSTRACT The study of past hurricanes to help interpret the patterns of current and future tropical storms is vital to our economy, society, and infrastructure. Understanding how hurricanes are influenced by a warm climate is critical, and hurricane reconstructions from former periods of the Holocene (the last ~11,500 yr) will be beneficial. Paleotempestology is the study of past tropical cyclones and uses historical, biological, and geological proxies to reconstruct tropical cyclone activity to create a record of historical hurricane patterns. A tropical cyclone (TC) is a chaotic weather event that is influenced by several elements, including warm ocean waters from which TCs directly draw energy. In this study, linear regression was iteratively performed on tree- ring data developed at Lake Louise in Valdosta, Georgia, with monthly climate values (precipitation, temperature, PDSI) and monthly oscillation indices (ENSO, AMO, NAO). Residuals were drawn from the tree growth responses and then compared to data obtained from the National Hurricane Center s North Atlantic Hurricane Database (HURDAT), which comprised all storms that entered two buffer zones (100- km and 150- km) during the period 1894 to Low values in residuals of tree- ring growth (narrow rings) were determined to be less than 0.3 (very low growth) and 0.4 (extremely low growth). Twelve residuals displayed very low growth while seven residuals displayed extremely low growth. For the storms that occurred within the 100- km buffer, 6 of the 12 very narrow rings and 4 out of 7 extremely narrow rings occurred the year directly after a TC event. For the storms that occurred within the 150- km buffer, 8 of the 12 very narrow rings and 6 out of 7 extremely narrow rings occurred the year directly after a TC event. A superposed epoch analysis was also used and found a vi

7 statistically significant relationship between hurricane events and extremely low growth years. The latter analysis further corroborated the negative response of tree- ring growth to hurricane disturbances. Additional research stemming from this study should include a strengthening of the disturbance signal by accounting for events that could also have affected tree growth, such as fire, insects, and human activity. vii

8 TABLE OF CONTENTS 1. INTRODUCTION Why is it important to study trends in tropical cyclones? Hypotheses Research Questions LITERATURE REVIEW Climate Change and Tropical Cyclones Climate Oscillations that Affect the Southeastern US El- Niño Southern Oscillation Atlantic Multidecadal Oscillation North Atlantic Oscillation Pacific Decadal Oscillation Tropical Cyclone History from Sediment Cores The Effects of Tropical Cyclones on Stand Dynamics Detecting Climate Oscillations and Tropical Cyclones with Dendrochronology Analyses of Ring- width Growth with Linear Regression Analyses of Oxygen- 18 Isotopes DETECTING TROPICAL CYCLONE SIGNALS IN TREE RINGS OF LONGLEAF PINE (PINUS PALUSTRIS MILL.), VALDOSTA, GEORGIA, U.S.A Introduction Study site Methods Tree- Ring Data Hurricane Data Climate Data Statistical Analyses Superposed Epoch Analysis viii

9 3.4 Results Discussion Climate Response Benefits of Multiple Buffer Zones Tropical Cyclone Events Superposed Epoch Analysis CONCLUSIONS After the effects of climate have been removed, do tree- ring data reveal a disturbance signal that can be attributed to known tropical cyclone events? Does superposed epoch analysis reflect an association between years of narrow tree rings and tropical cyclone events? Can the results from this research contribute to reconstructions of tropical cyclones that precede the satellite period? Suggestions for Future Research and Concluding Remarks REFERENCES APPENDICES VITA ix

10 LIST OF TABLES Table 1. TCs that approached within 100 km of Valdosta Table 2. TCs that approached within 150 km of Valdosta Table 3. Correlation coefficients from linear regression Table 4. Residual years and TC years x

11 LIST OF FIGURES Figure 1a. Location of Lake Louise Figure 1b. Aerial photo of Lake Louise Figure 2a. Example of stump being cored Figure 2b. Samples being taken for crossdating Figure 3. Chart of samples from tree- ring chronology Figure 4. Actual precipitation versus predicted precipitation Figure 5a. TC tracks that approached within 100 km of Lake Louise Figure 5b. TC tracks that approached within 150 km of Lake Louise Figure 6a. Tree- ring growth after removal of residuals from precipitation Figure 6b. Tree- ring growth after removal of residuals from temperature Figure 6c. Tree- ring growth after removal of residuals from ENSO Figure 6d. Tree- ring growth after removal of residuals from AMO Figure 6e. Tree- ring growth after removal of residuals from NAO Figure 7. Plotted residuals with very low and extremely low growth Figure 8a. SEA results 100- km buffer Figure 8b. SEA results 150- km buffer xi

12 CHAPTER ONE INTRODUCTION 1.1 Why is it important to study trends in tropical cyclones? Tropical cyclones cause destruction and loss of life every year in regions across the world. These storms produce high winds, tornadoes, and storm surges that cause sudden and significant flooding along coastlines. In the United States, 19 million people live within one kilometer of the coast, and the percentage of the U.S. population living near or on a coastline is rapidly increasing (Lam et al. 2009; Lin et al. 2010). Approximately 11.6 million people live at elevations below three meters, including in Florida and Louisiana, which have the largest population living below three meters of elevation, and are the two states with the highest number of landfalling Atlantic hurricanes (Elsner et al. 2000b; Fogarty et al. 2009; Lam et al. 2009). From 1925 to 1995, the average annual cost of damages from tropical cyclones in the U.S. was $4.8 billion (Pielke and Landsea 1998); however, in 2004 and 2005, the US saw over $150 billion in hurricane damages, and the average annual cost rose to $10 billion (Pielke et al. 2008). Insured property losses associated with hurricanes and storms have risen substantially in the last few years (Arpe and Leroy 2009). With the numbers of communities affected by hurricanes rapidly rising, research on future tropical cyclone trends and how they will be influenced by a warming climate is pertinent to human life as well as our economy and infrastructure. Examining tropical cyclone (TC) patterns associated with previous climate episodes is fundamental because such studies allow comparisons with present- day TC trends (Doyle and 1

13 Gorham 1996; Reams and Van Deusen 1996; Rodgers et al. 2006). An estimated 300,000 to 500,000 people have been killed by hurricanes in the North Atlantic basin alone in the last 500 years, and the United States experiences billions of dollars in annual hurricane damages (Rappaport and Fernandez- Partagas 1997; McCloskey and Keller 2009). If TCs experience an increase in frequency and/or intensity, the change could be detrimental to human life and economy (Lam et al. 2009; Corral 2010; Retchless et al. 2014). The past provides a template for comparing current and future trends. Research that focuses on reconstructing hurricane events will improve our understanding of TCs and how they will operate in a dynamic future climate. Climate has been vigorously investigated in recent years in an attempt to separate anthropogenic from natural mechanisms potentially responsible for climate variations (Webster et al. 2005; Emanuel et al. 2008; Screen and Simmonds 2010). Paleoclimatic research suggests that if our environment experiences a major variation in climate, the shift will not be gradual, but abrupt. Abrupt climate change is defined as a rapid warming (+4 C) that is detectable within a human lifetime (National Research Council 2002). Because climate change has likely been exacerbated by industrialization in the last century, thresholds in our current system are being approached rapidly, which could result in abrupt and significant variations in climate (Alley et al. 2003). Our knowledge of past abrupt climate change is limited, so understanding how thresholds are reached is pertinent to predicting future change that could occur in our lifetime (National Research Council 2002). Recently, analyses that focus on effects of climate shifts on TCs have become more prominent and methodical. The timescale of TC activity can be extended by investigating 2

14 historical records, such as newspapers, ship logs, and diaries that provide details of past storms (Dodds et al. 2009; Chenoweth and Mock 2013). However, TC events are also discovered through the biological and geological record (Nott 2004). Paleotempestology investigates past TC activity using historical records, biological proxies such as tree rings, and geological proxies such as overwash and beach deposits in sediment cores (Liu and Fearn 2000; Nott 2004; Mora et al. 2006; Emanuel et al. 2008). Using these proxies helps develop timescales of past trends in TC activity, and these trends assist researchers in determining the association between climate shifts and the frequency of catastrophic weather events. TC activity is susceptible to warming trends because an increase in sea surface temperatures (SSTs) may contribute to increased TC frequency and/or intensity. Researchers are evaluating if weather pattern variations could signal ongoing climate change in our environment. However, to assume that climate change causes a single weather event is inaccurate because many mechanisms determine the development, intensity, track, and lifetime of an individual storm. It is problematic to claim that a single hurricane is caused by a fluctuating climate. Lakoff (2012) produced an assessment shortly after the landfall of Hurricane Sandy in October 2012 and used Sandy as an example of distinguishing between a direct event and an episode that is caused systemically. He postulated that direct causation is a force applied that creates an immediate change. However, the results of systemic causation are more difficult to detect, and more challenging to predict as they are common in atmospheric systems. Numerous studies have analyzed the potential effect of a warming trend on TCs and the results are varied (Trenberth 2005; Webster et al. 2005; Elsner and Jagger 2010). The temporal 3

15 expanse of TC data provides a challenge for discerning a distinguishable pattern of increasing/decreasing frequency or intensity (Knutson et al. 2010). Scientists are collecting data to extend the paleoclimate record further back so we can compare present- day climate to a similar past climate that experienced frequent or intense TC patterns. The precise effect of a warming Earth on TC activity has not been fully established. The processes altered by climate variability are extensive and often interrelated, and many can influence the characteristics of a TC (Landsea 1993; Emanuel et al. 2008; Elsner and Jagger 2010). Several other climatic properties influence TC activity. One of the most influential is the El Niño- Southern Oscillation (ENSO) (Gershunov and Barnett 1998; Blake and Gray 2004; Xie et al. 2005). For example, when ENSO is in a warm phase, upper- level wind shear increases and amplified wind shear hinders TC formation. Additionally, the North Atlantic Oscillation (NAO) and Atlantic Multidecadal Oscillation (AMO) influence TCs. Distinct phases of the NAO based on the pressure gradient between the Icelandic Low and Azores High influence the track of northern Atlantic TCs through the placement of the Azores High, while the AMO influences SST anomalies (Trenberth 2005; Mora et al. 2007; Iizuka and Matsuura 2009). The Pacific Decadal Oscillation (PDO), while occurring over the Pacific Ocean, is associated with weather changes affected by SST variations and can also affect Atlantic TCs (Mantua and Hare 2002; MacDonald and Case 2005; Nateghi et al. 2010). Developing a better understanding of climate oscillations in the past could advance our understanding of past TC trends. 4

16 1.2 Hypotheses An increase in hurricane frequency and intensity has been suggested over the last 20 years; however, recent trends are not broad enough to draw any conclusion on the influence of a warming climate on TCs (Arpe and Leroy 2009). Applying proxies to determine occurrences of TCs further into the past can demonstrate how hurricane trends are affected by climate shifts. A growing body of research on reconstructing past climate and TCs exists; however, further studies are needed to determine appropriate methods of detecting past TCs and comparing them to paleoclimates. My research compares years of narrow tree rings to years in which TC events occurred in south central Georgia. I hypothesize that, by removing effects of climate on tree growth, temporal patterns of TC events can be discerned by isolating the narrowest tree rings in the growth record of longleaf pine (Pinus palustris Mill.). I postulate that the TC signal will be stronger once the climate signal is removed, and this statistical process can be replicated at different sites that are prone to North Atlantic TCs. The ultimate goal in this research is to develop improved methods for isolating TC events in the proxy record. 1.3 Research Questions In this study, I address: 1) After the effects of climate have been removed, do tree- ring data reveal a disturbance signal that can be attributed to known tropical cyclone events? 5

17 2) Can superposed epoch analysis reveal an association between years of narrow tree rings and tropical cyclone events? 3) Can the results from this research contribute to reconstructions of tropical cyclones that precede the satellite period? 6

18 CHAPTER TWO LITERATURE REVIEW 2.1 Climate Change and Tropical Cyclones In recent years, scientists are debating whether the earth is currently experiencing climate change and, if so, whether anthropogenic forcing is a cause. However, a segment of climate change that has not been closely examined is abrupt climate change (Alley et al. 2003). The likelihood of our environment experiencing an abrupt shift in climate is significant. Rapid climate variations are suspected causes of several past civilization failures and economic crises such as the fall of the Mayan and Akkadian civilizations and the American Dust Bowl of the 1930s. Sudden and drastic variations in climate can greatly disturb weather patterns. Paleoclimatic records have shown sudden deviations in TC frequency, flood regimes, and extreme droughts through the Holocene as a result of abrupt climate change (National Research Council 2002; Alley et al. 2003). Past changes in long- term hurricane trends have been suggested as a driver of agricultural loss and ultimately migration in some regions (McCloskey and Keller 2009). If a climate shift occurred over a few years rather than decades, studies suggest that we could see intensification of Atlantic hurricanes in the first half of the 21 st century (Walsh 2004). The poles are particularly sensitive to climate variations. Arctic amplification (AA) explains the increased sensitivity of polar regions to climate change, and can be detected by intensified warming of the poles. The amount of sea- ice in polar regions that has been lost since the 1980s as a result of melting is comparable to 40% of the contiguous United States (Francis 7

19 and Vavrus 2012). Calving of glaciers has occurred at an accelerated rate, with 87% of Antarctic glaciers on the retreat since 1950 and 100% of the glaciers experiencing calving. This increased climatic sensitivity at the poles can influence large- scale processes that disturb weather trends in the mid- latitudes, such as the track and longevity of TCs. The strength of the poleward pressure gradient directly influences the speed of upper level winds, such as the jet stream. When the pressure gradient is weakened, zonal winds are slower, and the outcome is elongated (deeper) peaks of ridges northward where blocking patterns can occur for large weather systems. AA is an example of systemic causation related to TCs because blocking patterns can change TC tracks and sometimes cause the coastline to become more susceptible to storms (Screen and Simmonds 2010). The probability of these prolonged conditions and persistent weather patterns caused by AA has been increasing, causing cold waves, heat waves, flooding, drought, and TCs (Francis and Vavrus 2012). An additional study indicated that upper- level blocking patterns caused by melting sea- ice and stalled deep- ocean convection could instigate a spontaneous or abrupt climate change (Drijfhout et al. 2013). Further evidence of the relationship between trends in climate change and trends in TC events is the influence of fluctuating sea surface temperature (SST). The temperature of the ocean during the development and lifetime of a hurricane is the most significant factor in tropical cyclogenesis (Elsner and Jagger 2010). TCs derive their energy from warm ocean water: when the water evaporates within the storm, it rises and cools adiabatically, releasing latent heat that energizes and intensifies the storm, surface wind, and evaporation (Arpe and Leroy 8

20 2009). Hence the subsequent warming of the oceans is important when modeling the frequency of future TCs. Research has indicated that TC frequency is positively associated with increasing SSTs, and anthropogenic climate change can increase the destructive potential of a TC (Emanuel 1987; Emanuel 2005). Models determining the impact of warmer climate and SSTs on TCs suggested an increase in intensity since the 1970s, while also predicting a doubling of category 4 and 5 hurricanes by the end of the 21 st century, particularly in the Atlantic (Bender et al. 2010). Other studies have found SSTs are positively correlated with TC intensity, and stated that SSTs are directly linked to the potential intensity of a TC (Elsner et al. 2000a). Conversely, some studies suggest the increase of upper- level wind shear another significant factor in cyclogenesis could contradict the influence of a warming ocean in the northern Atlantic (Arpe and Leroy 2009). Another theory concerning the effect of climate change on TC trends is that continental aerosols will work to decrease the strength of hurricanes by invigorating premature convection, making dissipation more likely prior to landfall (Khain et al. 2010). However, Webster et al. (2005) found that the North Atlantic has already shown an increase in frequency and duration of TC days that occur positively with rising SST trends, leading to speculation that the increase is a product of the warming trend. Mann and Emanuel (2006) contended that aerosol forcing facilitated cooling in the late 20 th century and storm activity was suppressed. However, they stated that hurricanes were already reflecting the increase in Atlantic temperatures because correlations showed SSTs resolved over half of Atlantic TC trends (r = 0.61), suggesting that 9

21 large- scale anthropogenically- forced warming will lead to intensification of storms at a rapid rate (Mann and Emanuel 2006). 2.2 Climate Patterns that Affect the Southeastern US and Atlantic TCs El Niño- Southern Oscillation The El- Niño Southern Oscillation (ENSO) is a large- scale atmosphere- ocean interaction, influenced by a band of either warm or cool SST anomalies in the Pacific Ocean, that ultimately impacts atmospheric circulation throughout the world (Ropelewski and Halpert 1986; Alexander et al. 2002). The El Niño (warm) and La Niña (cool) phases of ENSO are associated with high surface pressure or low surface pressure in the western Pacific, respectively. Changes in SSTs are monitored within the El Niño 3.4 region, which lies between the 120 th and 170 th west meridians. ENSO has a periodicity of 2 4 years and is strongest during the winter season, though the oscillation has displayed temporal and spatial unpredictability (Trenberth and Stepaniak 2001). ENSO indices can be reconstructed through proxies that contain records of past climate, such as ice cores, corals, and tree rings (D Arrigo and Jacoby 1991; Wilson et al. 2010; Li et al. 2013). Though a strong El Niño phase greatly influences western U.S. climate, ENSO shifts can also cause variability in the southeastern U.S., specifically by increasing winter precipitation during El Niño phases (Mojzisek and Mock 2009). Hence, El Niño (warm) phases are associated with higher precipitation while La Niña (cool) cause drier conditions over the southeast region (Seager et al. 2009). The cause of this precipitation variability also influences the development of 10

22 Atlantic TCs and the volume of storms observed within an Atlantic hurricane season. During a La Niña (cool) phase, storm tracks shift to the northern U.S., which creates quieter wind patterns across the southern U.S. and North Atlantic. The change in storm trajectory and subsequently calmer winds generate weakened upper- level wind shear and create better conditions for tropical cyclogenesis (Bove et al. 1998; Gershonov and Barnett 1998; Cleaveland et al. 2003). Periods of increased hurricane frequency have been documented through anomalously cool years exhibiting La Niña- like conditions over the Pacific Ocean (Dodds et al. 2009) Atlantic Multidecadal Oscillation The Atlantic Multidecadal Oscillation (AMO) is an index of SST anomalies in the northern Atlantic Ocean basin. Values of the AMO can reflect phases of warm (positive) and cool (negative) SSTs and are associated with the force of the thermohaline circulation, which controls heat transport through the oceans (McCabe et al. 2004). With a periodicity of years, AMO exhibits one of the lowest frequencies of oscillations that have been discovered (Enfield et al. 2001). Tree- ring reconstructions have been able to reveal more than five centuries of SST variability related to AMO, and show a consistent trend in values for several observed cycles. Despite its low frequency, the AMO has a strong influence on climates in the northern hemisphere. Ocean temperatures regulate a good percentage of seasonality, so SST fluctuations can produce noticeable changes in climate. A positive AMO can create a higher probability of drought in the United States, especially when combined with a negative phase of the Pacific Decadal Oscillation, and has been associated with major droughts in the last century. 11

23 AMO activity has exhibited regularity before and after the start of the Industrial era, suggesting that recent 20 th century warming has not affected this low- frequency oscillation. However, because of its long- term variability, AMO can perhaps to mask the effects of anthropogenically- induced climate change (Dima and Lohmann 2007). AMO indices have not displayed a correlation with SSTs in other ocean basins, except the high- latitude northern Pacific, as both are influenced by the tropospheric polar vortex (Gray et al. 2004). As of the mid- 1990s, AMO has been in a steady positive phase (Enfield et al. 2001; Gray et al. 2004) North Atlantic Oscillation The North Atlantic Oscillation (NAO) is expressed as the sea- level pressure gradient between the Icelandic subpolar low- pressure system and the Azores (Bermuda) subtropical high pressure system. The NAO is associated with the shift in surface westerlies over the North Atlantic Ocean (Lamb and Peppler 1987). These shifts in ocean- wind circulations can cause temperature and precipitation anomalies in western Europe and the eastern United States. The NAO is considered a low- frequency anomaly because it fluctuates at a periodicity of years, though a precise temporal variability has not been recognized. Past trends in the NAO can be detected through proxy records such as ice cores and tree rings, and indices can be reconstructed based on proxies that captured extreme shifts in NAO (Hurrell 1995; Cook et al. 1998). NAO can be especially extreme in the winter, as this is when the two pressure systems are at their strongest and induce a greater pressure difference. The strong pressure gradient 12

24 creates a warming over Europe and over the eastern U.S. during winter seasons when in a positive phase (Linderholm et al. 2003). Hurricane activity tends to be less frequent during positive phases of NAO, whereas activity picks up when NAO (or sea- level pressure) is weaker. The tracks of tropical cyclones are also influenced by phases of NAO: the eastern coast of the U.S. is struck more often during a positive NAO phase, as the subtropical Azores High is located more to the north and east, allowing hurricanes to sustain a more northern track. Previous studies have found a significant correlation between past periods of Gulf (East) Coast TC strikes with weaker (stronger) NAO values (Elsner et al. 2000b; Scott et al. 2003) Pacific Decadal Oscillation The Pacific Decadal Oscillation (PDO) is similar to ENSO but regulates SST anomalies over a larger portion of the Pacific basin. However, PDO periodicity has a lower frequency (30 50 years) and it tends to impact the mid- latitudes rather than equatorial regions (Mantua and Hare 2002; MacDonald and Case 2005). Though PDO appears over the Pacific basin, its effects can be felt as far as the southeastern US, with warm PDO phases generating wetter conditions, and cool phases associated with warm, dry conditions. When the PDO aligns with an ENSO phase, the outcome is enhanced properties of the latter oscillation, producing a stronger, more stable ENSO event (Gershunov and Barnett 1998). When both are in a positive (warm) phase, the Aleutian low is strong and creates moist, southward storm tracks across the southern U.S. Alternatively, when both oscillations are in a negative (cool) phase, storm tracks shift to the 13

25 northwestern U.S. while the southern region experiences a dry, less stormy period. A combination of ENSO and PDO phases can also create particular conditions for North Atlantic hurricane activity. When PDO and ENSO are in negative (cool) phases and SLP is low over the southwestern US, these conditions support enhanced TC activity in the month of August (Blake and Gray 2004). August is considered a pivotal month during the Atlantic hurricane season as 70% of hurricanes that form during the active season do so during the months of August and September (Retchless et al. 2014). 2.3 Tropical Cyclone History from Sediment Cores Since the satellite- documented TC record only reaches back to the 19 th century, other proxies must be used to detect TC occurrences. Liu and Fearn (1993) performed one of the leading studies to detect hurricane events using overwash deposits found in sediment cores at Lake Shelby, Alabama. They identified nutrient- rich peat layers overlaid by sand at the nearby Western Lake that were caused by storm surge from landfalling storms. Sandy layers were formed during a hurricane by tidal overwash and dune erosion, and a thicker layer would indicate a strong TC. Extratropical and winter storms can also cause sand erosion resulting in sandy layers, but their associated storm surges would not be strong enough to leave large deposits in Lake Shelby. Liu and Fearn (1993) calibrated hurricane strikes by the radiocarbon method and found that substantial TC events had a recurrence interval of ca. 600 years. The study noted that an abrupt shift in climate occurred around 3.2 ka. By the magnitude of the perturbations, they suggested a large- scale shift in climate resulted in an increase in hurricane 14

26 events. In a later study, Liu and Fearn (2000) discovered through lake- sediment cores that 12 catastrophic storms made landfall in the past 3400 years. They developed a year timescale of TC activity from cores at Western Lake, Alabama, and distinguished periods of hyperactive and low TC frequency. The sedimentary proxy analyses strengthened past TC reconstructions from the late Holocene. Local environmental effects can create bias in single- lake studies of sediment cores. To prevent this bias, Noren et al. (2002) took core samples from 13 separate lakes in the northeastern U.S. and established a millennial- scale storm chronology. They were able to detect storms with exceptional rainfall and determine patterns of storm frequency, specifically four peaks of storminess throughout the past 13,000 years. The study also documented that storm variability correlated with the long- term periodicity of the Arctic Oscillation (AO). Noren et al. (2002) ultimately established storm chronologies from several different sites and calibrated them to the chronology of AO, which also explains a significant amount of climate variability in North America and Europe during the Holocene. Scileppi and Donnelly (2007) evaluated hurricane activity through sediment cores in salt marshes from Long Island, New York, and found that four intense hurricanes had struck the area in the last 350 years based on deposits left from high storm surges. The sediment type, color, grain size, macrofossils, and layers of sediment cores were documented, while the loss- on- ignition (LOI) determined the amount of organic material in the core. The results revealed storm activity over the last ~3500 years, with increased activity occurring through the B.P. and pre B.P. periods and decreased storm frequency between 900 and 250 B.P. Four 15

27 major hurricanes occurred in 1693, 1788, 1821, and Hurricane frequency increased during the last half of the Little Ice Age, although SSTs at that time were 1 C cooler than present- day SSTs. Scileppi and Donnelly (2007) ultimately reconstructed a timeline of activity that correlated with TC patterns down the eastern seaboard and into the Caribbean. Liu et al. (2008) used pollen samples within sediment cores to identify the relationship between hurricane strikes and fire history in Little Lake near Gulf Shores, Alabama. The study found that halophytic plants (Chenopodiaceae) and heliophytic shrubs (Myrica) populations, as shown by pollen percentages in the cores, expanded after a hurricane strike in response to saltwater intrusion from storm surge. The relationship was recognized through data that displayed a decrease in pine populations (Pinus spp.) caused by intense fires that followed hurricane events. Not only did the study determine hurricane strikes through the strong presence of pollen in lake sediment cores, but it also linked the TC pattern to historical fire patterns, solidifying the importance of sedimentary records for climate reconstruction and exposing complex processes that follow TC events (Liu et al. 2008). Mann et al. (2009) compared a year TC record from sediment cores against a statistical model of TC trends based on proxy- derived climatic oscillations (SSTs, ENSO, NAO). Sediment- based overwash reconstructions were taken from 10 sites that covered five regions in the North Atlantic basin. The study found that the two independent records were strongly consistent. Both models described a peak in TCs during the Medieval Period (A.D ) followed by a decrease in activity around A.D Other climatic properties, such as additional oscillations, could create a stronger correlation between the statistical and 16

28 sedimentary models. However, the comparison validated the use of geological and biological proxy techniques to detect past TC trends. The study called for the development of further multi- proxy reconstructions of TC activity (Mann et al. 2009). 2.4 The Effects of Tropical Cyclones on Stand Dynamics Characterizing the range of forest responses to a disturbance can help predict tree mortality and stand dynamics, especially in the face of potentially intensified storms (Busby et al. 2009). Pillow (1931) found that compression wood could be detected from samples of hurricane- damaged trees. This type of reaction wood can result from particularly strong windstorms but is not limited to weather events. The abnormalities of compression wood in comparison to normal, healthy wood are the striations of cell walls, relatively wide annual growth rings, and wood that is not hard, flinty, and or has a lifeless appearance. Compression wood is found in the upper portion of the tree more often than the lower portion and possibly occurs in this manner because the upper part of the tree is more susceptible to abrupt or sustained wind shifts. Similar disturbance- based reaction wood is found in response to debris flows (Fantucci 1999), ice storms (Bragg et al. 2003), and insect attacks (Hartig 1896; Balch et al. 1964). Therefore there is difficulty determining if the reaction wood developed because of a specific disruption. Nevertheless, the presence of compression wood in a tree or tree- ring year known to have experienced a TC can strengthen the interpretation of other attributes (decreased growth) found within the tree that can be explained by a hurricane event. 17

29 Coastlines are an ideal geographic area to gather evidence for TC activity. In particular, barrier islands can provide TC information because of their placement along coastlines and their role as a natural barricade to incoming storms (Sallenger 2000). Windstorms are found to disturb the growth response of stands and alter mortality rates of forests in the years following a particularly strong windstorm. Trees that reside along the coastline are not only evaluated for windstorm damage, but also for damage caused by overwash and storm surge (Sallenger 2000). For example, Busby et al. (2009) used dendroecological methods to understand stand dynamics in response to landfalling hurricanes on Naushon Island, Massachusetts. They attempted to associate TC events with extreme growth responses in sampled trees over the last 150 years. Early 19 th century logging events were also taken into account when determining the responses. However, only one out of seven known storms was found to trigger a major change in tree growth. They hypothesized that storms can sometimes disturb only a portion of the stand or more susceptible tree species. Moreover, the winds of a hurricane, tornado, or downburst can be greater in defined areas of a storm and winds will never be evenly distributed throughout its lifespan. For example, the eye wall of a hurricane contains its strongest winds. Failure to take into account the characteristics of individual storms and different tree species, assuming that a weak storm will not significantly affect growth, could lead to underestimation of the frequency of disturbance (Wakimoto and Black 1994; Busby et al. 2009). Elsner et al. (2008) studied the differing effects of TC winds that impacted forest stands near Lake Shelby, Alabama. The research found that areas on the left side of a hurricane (facing 18

30 the direction of motion) were directly hit if they were within a radial distance equal to the maximum wind of the storm; a mean radius of 47 km was generated from hurricanes from 1893 to Subsequently, affected sites on the right side of the TC were considered direct hits if the radial distance was twice the maximum wind of the storm. One cannot assume wind damage caused by TCs will influence the stand dynamics of an entire forest; researchers must consider the angle and direction taken by a landfalling hurricane (Busby et al. 2009). In 1989, Hurricane Hugo made landfall as a Category 4 storm along the Luquillo Mountains of Puerto Rico (Turner et al. 1997). Following this event, several Cyrilla racemiflora L. trees, a lower montane subtropical rain forest species, were surveyed for up to four years (Drew 1998). Immediately after Hugo s landfall, four trees experienced severe branch detachment and defoliation. The following year, the same trees experienced an immediate stress response that included a decrease in branch elongation and smaller annual rings caused by low xylem development (Drew 1998). However, Drew (1998) also reported heavy flowering among the trees following the TC event and attributed this to nutrients being used for foliation, then becoming limited for branch growth and new xylem formation. Everham and Brokaw (1996) observed a generally quick recovery time for hurricane- impacted trees in comparison to other disturbances, such as earthquakes, landslides, fire, and anthropogenic practices. They proposed that windstorms cause less internal injury and the damage is less direct than the other types of disturbances. Research has also suggested that stronger hurricanes can result in different microclimates within a stand, creating differing levels of recovery (Ackerman et al. 1991; Everham and Brokaw 1996). 19

31 Batista and Platt (2003) examined a mixed- hardwood forest in northern Florida to investigate the effects of the 1985 Hurricane Kate event. Four classifications were designated to describe the tree responses to the disturbance: Resilient, Usurper, Resistant, and Susceptible. These classes were based on recruitment, growth, and mortality of the 10 most populous tree species before and after Kate made landfall. The results showed high resistance to hurricane disturbances by the most abundant old- growth species, whereas other classifications (such as Resilient and Usurper) depended upon periodic large- scale disturbances. Their findings categorized the various reactions of different tree species to a disturbance, and also described how hurricanes were a significant part of species diversity within the stand. Turner et al. (1997) also found that forest composition and structure were directly impacted by high- frequency TC trends in the Luquillo Mountains of Puerto Rico following the Hurricane Hugo event. 2.5 Detecting Climate Oscillations and Tropical Cyclones with Dendrochronology Analyses of Ring- width Growth with Linear Regression A climate oscillation is a fluctuation of climate on a seasonal, decadal, or millennial timescale and generally affects a multi- continental region (Cook et al. 1998). Common fluctuations such as the AMO, NAO, PDO, and ENSO indirectly structure the track, pattern, and frequency of TCs (Nott 2004; Mora et al. 2006; Mann et al. 2009; Grissino- Mayer et al. 2010). Past oscillations can be detected through biological and geological proxies and linked to SSTs, wind patterns, precipitation, and other meteorological patterns, while typically exhibiting a distinct periodicity (Mora et al. 2007). 20

32 A tree- ring chronology can reveal climate trends, atmosphere- ocean oscillations, and disturbances such as fires, insect infestations, and storm events (Rodgers et al. 2006). Johnson and Young (1992) provided evidence of landfalling hurricanes in tree- ring width data from several barrier island locations along the Virginia shoreline. They compared ninety- four loblolly pine (Pinus taeda) samples to storm data from 1930 to 1989 and detected five narrow rings in years following known hurricanes (1953, 1958, 1967, 1972, and 1989). They noted considerable variability among the tree- ring widths at their sites, and stressed the significance of precipitation causing the variations. Their results also showed that tree- ring width increased as latitude decreased, most likely owing to the expansion of the growing season as the climate becomes subtropical and the winters become mild. Drew (1998) supported this finding, as his Puerto Rican study sites were located in a more tropical region than the sites studied by Johnson and Young. Reams and Van Deusen (1996) showed that hurricane strikes corresponded to the most influential years diagnosed within a regression model between tree- ring width and annual moisture the year of a hurricane strike and the following year displayed the narrowest ring widths. Baldcypress (Taxodium distichum (L.) Rich.) data collected from southern Louisiana strongly correlated to PDSI and growing season precipitation, and these variables were used to model tree growth to better understand relationships between drought years and years of known hurricane strikes. The chronology showed reduced growth similar to drought that suggested a TC event. Doyle and Gorham (1996) also discovered narrow or missing rings following a TC event after removing age and climate trends from slash pine (Pinus elliottii) 21

33 along the northern Gulf Coast. They concluded hurricanes could affect tree growth responses as far as seven years after a hurricane made landfall, as seen with Hurricane Camille (1969). Years marking a disturbance or some traumatic development were termed pointer years. The degree of reduced growth following a hurricane year correlated with the strength and magnitude of a storm (Doyle and Gorham 1996). Rodgers et al. (2006) used linear regression to model tree- ring index values as a function of climate. In their study, climate was removed from a slash pine chronology collected in coastal Alabama. The residuals were used to isolate years of growth release (increase) and suppression (decrease), then compared to the hurricane database (HURDAT) to link sizeable growth changes to known hurricane strikes. After the authors determined that relatively no climatic noise remained, they found that hurricane years coincided with release years, especially during a period of high TC activity. These high- release values were the result of one major storm and not a cluster of repetitive landfalls. In contrast, they found suppression of growth occurred during periods of low TC activity, suggesting that tree growth was affected by both individual storms and periods of hurricane activity Analyses of Oxygen- 18 Isotopes A critical yet complex process to detect a TC signal in tree rings is determining oxygen- 18 isotope ( 18 O) levels in the alpha- cellulose of the tree ring. Previous studies have shown that TC precipitation is depleted of 18 O by as much as 10 compared to an average thunderstorm, enabling researchers to determine whether the source water of a tree was from rainfall 22

34 associated with TC activity (Miller et al. 2006; Mora et al. 2007; Lewis et al. 2011). The 18 O depletion amounts are dependent on storm size, the soil type in which the tree is growing, and pre- existing moisture conditions in the soil. However, this is only a signal of TC occurrence and not an indicator of the strength or magnitude of a storm. Miller et al. (2006) evaluated the 18 O values of longleaf pine (Pinus palustris Mill.) collected in Valdosta, Georgia, and compared the values with TC events that struck within 400- km of the sampled site. The buffer of 400- km was chosen because the most 18 O- depleted portions of a TC are the outermost rain bands, which can stretch hundreds of kilometers from the eyewall. A 1- year autoregression or AR(1) model was applied to remove autocorrelation. Low residual values of 0.5 to 1.0 indicated years of low 18 O values. The study found that all but three low residuals were associated with a TC event between 1940 and 1990 when compared to the HURDAT record. Years prior to 1940 indicated that 22 TCs affected the site, and historical records identified 21 of these potential storms. They also found differences between earlywood (EW) and latewood (LW) 18 O compositions, suggesting seasonal influences on source water. Proxy data can aid in identifying both TCs and climate oscillations (Emanuel et al. 2008). When isolating hurricane events in the oxygen isotope time series, climate oscillations are discernible because they dictate precipitation and temperature patterns that can also influence tree growth and the chemical composition of tree rings (Mora et al. 2007). For example, results from spectral analysis of the EW and LW oxygen isotope series can be compared to the AMO to reveal patterns of past SSTs. By determining whether the AMO was positive or negative, 23

35 researchers can establish whether SSTs were warmer or cooler, respectively, which further enhances our knowledge of past TC activity (Mora et al. 2007). Lewis et al. (2011) compared 18 O values of a longleaf pine chronology collected in Big Thicket National Preserve, Texas, to TC events that occurred within a 250- km radius of the site. Thirty trees at two sites were collected, and four trees from each site were selected for analysis over a 25- year period ( ). After applying an AR(1) model to the dataset, negative residuals less than 0.1 were determined as disturbances. Isotopic values were found to vary between individual series, but all four trees in the first site identified three known TCs (1986, 1998, and 2002). Four additional TC events were identified by the first site, but were not associated with known storms and were considered false positives. All four trees at the second site indicated one known TC, and four events were falsely indicated the same false years detected in the first site. A total of six false positives occurred, indicating a disturbance other than hurricanes was affecting isotopic values at the sites. Lewis et al. (2011) suggested that although 18 O values can determine storm frequency, climate variability could complicate the identification of individual TC events. 24

36 CHAPTER THREE Detecting Tropical Cyclone Signals in Tree Rings of Longleaf Pine (Pinus palustris Mill.), Valdosta, Georgia, U.S.A. This chapter is intended for submission to the journal Tree- Ring Research. The research topic was originally developed by me and my advisor, Dr. Henri Grissino- Mayer. The use of we throughout the text refers to me and Dr. Grissino- Mayer, who assisted with site selection, data collection, project development, and text editing. My contributions to this chapter include statistical data analysis, interpretation and graphic displays of results, and writing of the manuscript. Abstract The study of past hurricanes to help interpret the patterns of current and future tropical storms is vital to our economy, society, and infrastructure. Paleotempestology uses historical, biological, and geological proxies to reconstruct tropical cyclone (TC) activity to create a record of historical hurricane patterns. In this study, linear regression was iteratively performed on tree- ring data developed at Lake Louise in Valdosta, Georgia, with monthly climate values (precipitation, temperature, PDSI) and monthly oscillation indices (ENSO, AMO, NAO). Residuals were drawn from the tree growth responses and then compared to data obtained from the National Hurricane Center s North Atlantic Hurricane Database (HURDAT). Two buffer zones (100- km and 150- km) were created and we used HURDAT data that included storms that entered either buffer zone during the period 1894 to Low values in residuals of tree- ring growth (narrow rings) were determined to be less than 0.3 (very low growth) and 0.4 (extremely low growth). Twelve residuals displayed very low growth while seven residuals displayed extremely low growth. For the storms that occurred within the 100- km buffer, 6 of the 12 very narrow rings and 4 of the 7 extremely narrow rings occurred the year directly after a TC 25