Late Paleocene and Early Eocene Paleoclimate Reconstructions: The Digital Leaf Physiognomy Approach

Transcription

1 Wesleyan University The Honors College Late Paleocene and Early Eocene Paleoclimate Reconstructions: The Digital Leaf Physiognomy Approach by Sharon Newman Class of 2008 A thesis submitted to the faculty of Wesleyan University in partial fulfillment of the requirements for the Degree of Bachelor of Arts with Departmental Honors in Earth and Environmental Science Middletown, Connecticut April, 2008

2 ACKNOWLEDGEMENTS I would like to thank those individuals who contributed to the thesis-writing process, and without whom I never would have been able to complete this work. First of all, I must acknowledge my thesis advisor, Dana Royer, who dedicated much time and effort in assisting me in all my thesis endeavors. Dana Royer has been a great help in both the research and writing process, providing advice, answering all of my questions, and editing my work. Only with his expertise and strong encouragement was I able to complete each essential phase of my thesis project. I must also give out a heartfelt thanks to my other two thesis readers, Suzanne O Connell and Scott Wing. Both have provided me with insightful comments on early drafts of my thesis. Their advice has helped me to improve upon my work. In addition, I would like to thank Wesleyan University and the Earth and Environmental Science Department. Without the support of both the department and the University, I would never have been able to complete this thesis. Finally, I would like to thank my family and friends. Their emotional support has aided me throughout the entire processes. I will forever be grateful to them. i

3 TABLE OF CONTENTS LIST OF FIGURES...iii LIST OF TABLES...iii ABSTRACT... iv 1. INTRODUCTION The Paleocene-Eocene Thermal Maximum Climate Proxies Phylogenetic Methods Physiognomic Methods Hypothesis METHODS AND MATERIALS Selection of Specimens and Site Locations Processing Leaf Images DiLP Calibration RESULTS DISCUSSION MAT Reconstructions MAP Reconstructions Precision of DiLP Improvements Over Past Research CONCLUSION APPENDICES WORKS CITED..54 ii

4 LIST OF FIGURES Figure 1. Locations of the five late Paleocene and early Eocene sites from the Bighorn Basin, Wyoming Figure 2. Representative examples of fossils used in this study Figure 3. Processing procedures for digital leaf physiognomy Figure 4. Determining major length and leaf area for damaged leaves Figure 5. Relationship between mean annual temperature and age of site Figure 6a. Relationship between mean annual precipitation and age of site Figure 6b. Relationship between mean annual precipitation and age of site Figure 7. Two Zizyphoides flabella specimens collected at Elk Creek LIST OF TABLES Table 1. Definitions and units for DiLP variables Table 2. Number of species and specimens included for DiLP analysis Table 3. MAT and MAP for present-day calibration sites Table 4. Multivariate and univariate regression model summaries Table 5. MAT reconstructions Table 6. MAP reconstructions Table 7. MAT and MAP estimates for fragmentary fossils iii

5 ABSTRACT The Paleocene-Eocene Thermal Maximum (PETM) was a time of distinct climatic change. Both precipitation and temperature patterns fluctuated during this time, with a pacing and magnitude similar to present-day climate variation. Extreme migration and extinction events also marked this interval. In the current study, I apply a new method that is based on the digital analysis of leaf size and shape (digital leaf physiognomy, DiLP) to fossil plant assemblages, in an effort to better constrain patterns of climate change. I reconstruct mean annual temperature (MAT) and mean annual precipitation (MAP) for five Paleocene and Eocene sites from the Bighorn Basin, Wyoming. Results reveal a trend of gradually increasing temperature throughout the late Paleocene, comparatively large increases in temperature during the PETM, and a return to pre-petm conditions during the early Eocene, which is consistent with paleoclimate reconstructions from previous research. MAP also decreased at the onset of the PETM, returning to late Paleocene conditions during the early Eocene. Most importantly, the DiLP approach, which relies on computer programs to calculate MAT and MAP values, provides a highly reliable and consistent methodology for reconstructing paleoclimates from fossil leaf assemblages. All physiognomic approaches (including leaf-area analysis, leaf-margin analysis, etc.) use variables related to both the size and shape of a leaf in order to calculate climate estimates. However, a comparison of climate reconstructions for the PETM, derived from various physiognomic methods, reveals DiLP methodologies to have a relatively high degree of precision. iv

6 1. INTRODUCTION 1.1 The Paleocene-Eocene Thermal Maximum As the Paleocene Epoch came to a close ~55 Ma, the Earth experienced distinct climatic changes that are recorded in the paleobiological and geologic records. During this interval, known as the Paleocene-Eocene Thermal Maximum (PETM), a global warming event occurred, which lasted only 100,000 to 200,000 years (Wing et al., 2005; Cariglino, 2007; Röhl et al., 2007). Both tropical sea surface temperatures and deep ocean temperatures increased by as much as 5 C; high latitude terrestrial sea surface temperature increased by up to 9 C (Thomas et al., 2002; Sluijs et al., 2006; Weijers et al., 2007). With the rapid onset of warming, coupled with the extinction and migration of many taxa (Gingerich, 2003; Zachos et al., 2003; Gingerich, 2006), the PETM resembles the magnitude and rate of presentday climate change. Through isotopic investigation as well as through detailed analysis of the fossil record, we are able to use Paleocene-Eocene warming as an ancient analog to modern climatic changes. Such work will enable us to predict future anthropogenic impacts on ecological, hydrologic, and atmospheric processes. The isotopic record indicates that during the PETM there was a negative carbon isotopic excursion (CIE); studies of benthic and planktonic foraminifera have shown that the carbon excursion was 3-4, indicating a massive introduction of 12 C into both the ocean and atmospheric reservoirs (Kennet and Stott, 1991; Thomas et al., 2002). Oxygen isotopic values from marine cores are also depleted during the PETM, indicating a warming, and the carbon and oxygen isotope records are largely

7 concordant. The initial depletion of isotopic values occurred quite rapidly, on the order of less than 20 kyr, while estimates for the near return to pre-petm isotopic conditions are on the order of 190 kyr (Zachos et al., 2005) Changes in precipitation patterns can also be inferred from the geologic record, although evidence from the PETM is often conflicting. Studies of clay and mineral deposits (Gibson et al., 2000) and the carbon isotopic composition of pedogenic carbonate in paleosols (Bowen et al., 2004) suggest a general increase in precipitation throughout the interval. However, later paleosol studies have revealed a decrease in precipitation at the end of the Paleocene, followed by a slow increase during the PETM (Kraus and Riggins, 2006). Kraus and Riggins (2006) have also found that precipitation levels returned to pre-petm conditions towards the end of the PETM interval (Kraus and Riggins, 2006). Leaf-area analysis (methodology described in detail below) has revealed a similar decrease in precipitation at the onset of the PETM (by as much as 40%), followed by an increased in precipitation during the PETM interval (Wing et al., 2005). Important extinction and diversification events mark the PETM. Most significantly, benthic foraminifera suffered a mass extinction on the order of 30 to 50 percent of all species (Jiang and Wise, 2006). Both infaunal and epifaunal benthic communities were affected by the global warming event; however, epifaunal organisms experienced the greatest decline in population (Kennet and Stott, 1991). Unlike their benthic counterparts, surface plankton underwent a temporary diversification during the PETM; this radiation event occurred quite rapidly, on the order of less than 10 kyr (Kelly et al., 1998). 2

8 At the Paleocene-Eocene boundary, mammalian species underwent a largescale diversification and migration event. For the first time, the APP taxa, which consist of Artiodactlya (even-toed ungulates), Perissodactyla (odd-toed ungulates), and primates migrated throughout the North American continent. Many PETM species, such as Ectocion parvus, experienced a temporary dwarfing effect (Gingerich, 2003). Such a decrease in size may have resulted from the increase in carbon dioxide during the PETM (Gingerich, 2003). In contrast to many animal groups, plants did not experience any significant extinctions or diversification events. However, in the Bighorn Basin, which contains the only known plant megafossils from the PETM, some taxa experienced a temporary northward migration. Taxa were found to migrate as far as 650 to 1500 km northward of their original late Paleocene distributions. Certain taxa were thought to have originated in locations as far south at the Gulf Coastal Plain and the Colorado Basin. During the PETM, the Bighorn Basin experienced negligible species turnover; the migrant species returned to their original southern habitats by the end of the PETM interval (Wing et al., 2005). Although the exact mechanisms for climatic change are unknown, a leading hypothesis is the release of seabed methane hydrates. The carbon in methane (CH 4 ) is strongly isotopically depleted (δ 13 C -60 ). Under certain pressure and temperature conditions, methane can form as a hydrate. However, increases in temperature and decreases in pressure can cause hydrates to destabilize, releasing large amounts of methane into the ocean and atmosphere. As temperatures rose during the PETM due, in part, to the radiative forcing of CH 4 (and its oxidative product, CO 2 ), more seabed 3

9 methane was released, forming a positive feedback loop (Dickens et al., 1995; Zachos et al., 2005). Other theories suggest that carbon dioxide (released from volcanic outgasing), thermogenic methane (organic matter exposed to high pressure and high temperature conditions (Archer, 2007)) and the burning of peat and coals may have contributed to the global warming event and the subsequent isotopic excursion (Kurtz et al., 2003; Svensen et al., 2004). 1.2 Climate Proxies Climate proxies are useful tools for reconstructing temperature and precipitation estimates from the geologic past. One broad class of climate proxies is based on floral composition or leaf physiognomy (size and shape). These approaches use the relationship between climate and plant taxonomy or physiognomy in order to provide qualitative and quantitative analyses of climate variability (Wolfe, 1995; Royer et al., 2005; Uhl et al., 2006). Plants are sensitive to their environment in part because they are sessile. Leaves are especially sensitive to environmental factors due to their large surface area to volume ratio. Several botanical methods discussed here are: the Nearest Living Relative (NLR), the Coexistence Analysis (CA) approach, leaf-margin analysis, leaf-area analysis, Climate-Leaf Analysis Multivariate Program (CLAMP), and Digital Leaf Physiognomy (DiLP) (Chaloner and Creber, 1990; Wolfe, 1995; Royer et al., 2005; Uhl et al., 2006; Cariglino, 2007). 4

10 1.2.1 Phylogenetic Methods The Nearest Living Relative Approach The Nearest Living Relative (NLR) approach compares fossil plant species with closely related extant species. Scientists can use the known climatic tolerances of modern flora in order to estimate tolerances of their ancient relatives. Through the evaluation of climatic constraints on both modern and fossilized floras, researchers can reconstruct paleoclimates (Chaloner and Creber, 1990; Wolfe, 1995). Several factors can impact the success of the NLR method, influencing the precision of the technique. In order to achieve maximum reliability, the modern and ancient flora must have a strong phylogenetic association. The modern taxa must also be distributed across its full climatic range (i.e., not a relictual distribution, such as Ginkgo or Metasequoia) (Wing and Greenwood, 1993; Cariglino, 2007). However, even when conditions are established for maximum success, there are many negative aspects involved in using these methods. The NLR approach does not account for evolutionary change, and it assumes that environmental tolerances remain identical for both modern and ancient relatives (Wolfe, 1995). Therefore, this approach becomes less accurate as one looks further back into geologic history because the evolutionary gap between modern and ancient flora increases. The scoring reliability is also poor, due to the fact that researchers will often provide different climatic interpretations for the same set of fossil data (Cariglino, 2007). 5

11 The Coexistence Analysis Approach The Coexistence Analysis (CA) approach was developed to improve the NLR approach. However, many of the fundamental flaws that are inherent in the original method remain. When using the CA approach, scientists establish the climatic tolerances of both the extant and ancient species, as well as various factors that may impede climatic adaptation (Utescher et al., 2000). In analyzing such factors prior to other investigations, researchers hope to find those living relatives which most closely resemble species from the fossil record and to establish a range of modern environmental tolerances that will help to identify those tolerances of the ancient flora. However, despite such improvements, the CA approach involves increased time demands, and, more importantly, fails to incorporate the impact of evolutionary trends in floral development. The CA approach also fails to provide accurate climate estimates for time intervals prior to the Paleogene period, due to a lack of nearest living relatives (Mosbrugger and Utescher, 1997; Utescher et al., 2000; Cariglino, 2007) Physiognomic Methods Leaf-margin analysis The leaf-margin analysis approach derives mean annual temperature (MAT) from the percentage of untoothed species in a given location. This method relies on the knowledge that untoothed species are more prevalent in warmer locations, while toothed species are more frequently found in colder locations (Wolfe, 1979; Wilf, 1997; Royer et al., 2005). Leaf margin analysis focuses only on the presence/absence of teeth, and has an accuracy of ± 2 to 3 C. A major advantage of leaf margin 6

12 analysis (and other physiognomic approaches) over CA and NLR is that they are not significantly impacted by phylogeny. However, despite the relative precision of leaf margin analysis, there are a number of potential drawbacks. First of all, current physiognomic methods are calibrated only for woody dicotyledonous species, and therefore cannot be used to reconstruct paleoclimates that pre-date the mid- Cretaceous Period. Also, leaf margin analysis relies on a binary character system (i.e., presence vs. absence of teeth), and does not fully describe the size and shape of the leaves. Lastly, leaf-margin analysis usually underestimates MAT for plants in physiologically-wet environments (Burnham et al., 2001; Kowalski and Dilcher, 2001). This is probably because leaf teeth are conduits for enhanced rates of transpiration (Royer and Wilf, 2006) and so are more adaptive in wet environments. The high prevalence of toothed specimens in riparian systems leads to a slight underestimation of MAT (Royer et al., 2005; Cariglino, 2007). Leaf-area analysis The leaf-area analysis approach uses the association between leaf size and moisture regimes in order to calculate MAP. Leaf-area analysis relies on the fact that precipitation patterns can greatly affect the morphology of leaf structures (Wilf et al., 1998; Wilf, 2000). For example, smaller leaves are often found in dry locations because such plants cannot afford the high water loss associated with larger leaf areas (Dolph and Dilcher, 1980; Wilf et al., 1998). While other methods have overestimated precipitation patterns, the leaf-area analysis approach consistently results in lower MAP estimates (Wilf et al., 1998). Leaf-area analysis reconstructions for the Eocene reveal reductions greater than fifty percent of previously estimated 7

13 MAP values. One drawback to the leaf-area analysis approach is that large leaves may become fragmented during the process of fossilization. If not correctly identified, such specimens may negatively impact the accuracy of the MAP estimates (Wilf et al., 1998). Climate-Leaf Analysis Multivariate Program The Climate- Leaf Analysis Multivariate Program (CLAMP), first proposed by Wolfe (1990), uses leaf characteristics to predict both temperature and precipitation. Unlike leaf-margin analysis and leaf-area analysis, CLAMP uses multiple characters to derive climate estimates. Although the method began with only 18 shape parameters, today twenty-nine leaf variables, relating to size and shape, are used to help create multivariate analyses. These variables are strongly correlated with environmental factors such as moisture and temperature. The results are then plotted on the ordinal plane system, revealing nonlinear relationships. Temperature and water stress are plotted on corresponding axes (Wolfe, 1995). In some cases, CLAMP and the NLR and CA approaches provide similar climate estimates. NLR estimates from the Miocene Latah formation in Washington resemble data obtained from CLAMP estimates (Wolfe, 1995). However, similar estimates of temperature in the Kissinger Lake assemblage reveal a difference between the two methods (NLR approach estimates ranged from C; CLAMP estimates were 15.2 C) (Wolfe, 1995). Despite CLAMP s innovative methodology, MAT estimates typically are no more reliable than those calculated from leaf-margin analysis (Jacobs and Deino, 1996; Wilf, 1997; Kowalski and Dilcher, 2001). Also, both CLAMP and leaf-margin analysis greatly underestimate MAT in riparian communities, due to the 8

14 overrepresentation of toothed species in wetland environments. Another significant source of error when using CLAMP involves the ambiguity of leaf parameter states. Researchers often disagree over the classification of leaf specimens, resulting in scoring inconsistencies (Wolfe, 1995; Wilf, 1997; Royer et al., 2005; Cariglino, 2007). CLAMP leaf states are also categorical in nature. Therefore, another factor which can affect method accuracy involves the complete rejection of continuous variables. Without fully exploring the continuous nature of leaf states, researchers limit their ability to study covariation between variables (Royer et al., 2005; Krieger et al., 2007). The eigenshape method, which bears some resemblance to CLAMP methodologies, was developed to increase scoring reliability as well as to include continuous variation patterns in leaf size and shape (Krieger et al., 2007). Eigenshape includes 20 character states. Unlike other methodologies, eigenshape does not combine size and shape into a single variable, but instead includes seven distinct leaf characters, one asymmetry character, as well as leaf area. Eigenshape also uses leaf asymmetry in order to produce reliable MAT calculations, which is an improvement over CLAMP methodologies. However, eigenshape does produce unreliable data when both lobed and unlobed leaf specimens are included in the data set (Krieger et al., 2007). In order to reconstruct paleoclimates, eigenshape also requires perfectly preserved fossil specimens, which are extremely rare. 9

15 Digital Leaf Physiognomy In order to maintain the advantages of a multivariate approach but to minimize some of the pitfalls of CLAMP, Huff et al. (2003) developed Digital Leaf Physiognomy (DiLP). This method requires the careful analysis of digital leaf images and the collection of continuous quantitative measurements (Table 1). Variables such as tooth count and tooth area (see Table 1 for definitions) are then used to create multiple linear regressions for both MAT and MAP estimates (Huff et al., 2003; Royer et al., 2005). DiLP is based on the assumption that if the proportional abundance of toothed species increases in colder climates, then leaves from cold climate floras should have both a greater number of teeth and a larger tooth area than plants from warmer climates (Royer et al., 2005). DiLP has also been associated with leaf economic traits such as nitrogen and leaf mass per area content. Thus, there is the potential to reconstruct both climatic and leaf economic information from leaf physiognomy. The DiLP approach represents a significant improvement over previously developed methodologies. Both leaf-margin analysis and CLAMP use categorical methods for defining leaf variables. (Leaf-margin analysis relies solely on a binary system: the presence or absence of leaf teeth; all CLAMP character states are categorical in nature.) In contrast, DiLP uses continuous variables, such as leaf area and tooth area (see Table 1 for variable definitions) (Huff et al., 2003). The use of such information should increase the accuracy and precision of climate estimates and allow for a better understanding of covariation between variables (Royer et al., 2005; Cariglino, 2007). DiLP methodologies also rely heavily on quantitative computer 10

16 analyses, removing elements of subjectivity that have proven problematic for some of the other physiognomic methods. By removing much of this ambiguity, DiLP ensures greater precision in climate estimations (Royer et al., 2005). Table 1. Definitions and units for DiLP variables. Physiognomic Variable Definition Margin Percentage Percent of untoothed species (%) Blade Area Inferred Leaf Area Perimeter (total Preserved) Internal Perimeter Perimeter Ratio Major Length Feret Diameter Feret Diameter Ratio Number of Primary Teeth Number of Secondary Teeth Number of Teeth (total) Number of Teeth: Perimeter Number of Teeth: Internal Perimeter Average Tooth Area Tooth Area: Perimeter Tooth Area: Internal Perimeter Tooth Area: Blade Area Area of reconstructed leaf blade (cm²) Area of pristine leaf blade (cm²) Perimeter of leaf blade (cm) Perimeter of leaf blade once teeth have been extracted (cm) Perimeter/ Internal Perimeter Longest unit of length across leaf blade (cm) Circle diameter than has an area identical to leaf area (cm) Feret Diameter/ Major Length (count) (count) Primary and Secondary Teeth (count) Number of teeth/ Perimeter (total preserved) (cm -1 ) Number of teeth/ Internal Perimeter (preserved) (cm -1 ) Tooth Area/ Number of Primary Teeth (cm²) Tooth Area/ Perimeter (total preserved) (cm) Tooth Area/ Internal Perimeter (preserved) (cm) Tooth Area/ Blade Area 11

17 DiLP calculations also may provide more accurate climate estimates of wetland environments compared to other physiognomic approaches. Both leaf-margin analysis and CLAMP significantly underestimate MAT for riparian and highly disturbed systems, as a result of the high proportion of toothed specimens found in such localities. MAT estimates from DiLP analyses of riparian environments have revealed error reductions of 3 and 5 ºC relative to leaf-margin analysis and CLAMP, respectively (Royer et al., 2005). Many fossil assemblages are recovered from wetland environments that are most vulnerable to the riparian effect. When analyzing fossil floras, the DiLP approach proves to be the most reliable of all modern physiognomic approaches (Royer et al., 2005; Cariglino, 2007). In addition, DiLP is successful at predicting MAT and MAP from fossil specimens, which are often poorly preserved and do not have fully intact margins. For slightly damaged specimens (25% damaged), DiLP is as accurate as leaf-margin analysis in reconstructing present-day climate. However, for leaves that are highly damaged (50% or more), DiLP produces more accurate results. Both leaf-margin analysis and CLAMP become less reliable as the amount of specimen damage increases as a result of false margin scoring. For example, many toothed species only have teeth near their apex, and in such cases if a fossil specimen does not have its apex preserved, it will be falsely scored as untoothed (Royer et al., 2005). The DiLP approach is also successful at compensating for rare species. Royer et al. (2005) have determined that DiLP MAT and MAP estimates remain accurate even when only a single specimen is available to represent an entire species. DiLP produces the most accurate estimations when at least 20 woody dicotyledonous 12

18 species are used. However, colder environments may require a higher number of species due to the high variability of physiognomic characteristics (Royer et al., 2005). Recently, DiLP has been used to reconstruct MAT for three Eocene floras collected from lake deposits in South America and North America. Cariglino (2007) suggests that the DiLP method provides more accurate paleoclimate reconstructions than either leaf-margin analysis or CLAMP. In order to show DiLP s limited sensitivity to the riparian effect, Cariglino compared DiLP and NLR climate estimates. The NLR approach, which is a phylogenic approach, does not underestimate climate in wetland environments. The correspondence of DiLP and NLR climate estimates reveal the increased reliability of DiLP over other physiognomic methodologies. Cariglino (2007) also demonstrated that the number of species analyzed, and not margin completeness, has the greatest effect on climate estimates. In general, limited research has been completed using the DiLP method on fossil assemblages and more research needs to be completed in order to determine the accuracy of the method in determining paleoclimate estimates. 1.3 Hypothesis Using the DiLP method, I intend to estimate both MAT and MAP for three sites from the late Paleocene (Skeleton Coast [Tiffanian 4a], Lur d Leaves [Tiffanian 5b], and Daiye Spa [Clarkforkian 3]), one site from the early Eocene (Elk Creek [Wasatchian 2]), and one site from the PETM (Hubble Bubble [Wasatchian 0]), all located in the Bighorn Basin in northern Wyoming (Figure 1; Table 2). I will then track climate changes throughout this interval and compare my results with climate 13

19 reconstructions for the previously mentioned locations using other climate proxies. I hypothesize that my results will closely match those found from leaf-margin analysis and leaf-area analysis. However, I also hypothesize that the DiLP calculations will provide more precise climate estimates. I will also attempt to further improve upon the DiLP approach by increasing the number of extant sites used to calibrate the multivariate regression models (Table 3). Most importantly, the previous calibration by Royer et al. (2005) was not suitable for developing an MAP proxy because the range in MAP was too narrow (MAP = cm for 16 sites, and one site with a MAP of 264 cm). My research will include 30 modern sites with a range in MAP from cm. 14

20 Figure 1. Locations of the five late Paleocene and early Eocene sites from the Bighorn Basin, Wyoming. 15

21 Table 2. Number of species and specimens included for DiLP analysis. The total number of species and specimens collected was provided by Currano et al. (2008). The number of species and specimens used for DiLP processing was lower due to the poor preservation of some fossil samples. Time Interval Paleocene Site and Zone Skeleton Coast (Tiffanian 4a) Lur d Leaves (Tiffanian 5b) Age (Ma) Total Number of Species Total Number of Specimens Number of Species Included in DiLP Processing Number of Specimens Included in DiLP Processing Daiye Spa (Clarkforkian 3) PETM Eocene Hubble Bubble (Wasatchian 0) Elk Creek (Wasatchian 2)

22 Table 3. MAT and MAP for present-day calibration sites. Site MAT (ºC) MAP (cm) 1 Hubbard Brook, NH Allegheny National Forest, PA Harvard Forest, MA Huyck Preserve, NY IES, NY Cockaponset, CT Hawk Mountain, PA York County, PA SERC, MD Duke Forest, NC Pee Dee, SC Big Hammock, GA Dilcher's Woods lowland, FL Dilcher's Woods upland, FL Archbold, FL Panther Refuge, FL Barro Colorado Island, Panama Ducke, Brazil* Parque El Rey, Argentina* Robinson Falls, Malaysia* Pasoh, Malaysia* Tangun Tuan, Malaysia* Kepong, Malaysia* Noah Creek, Australia* Chamela, Mexico* Alamos, Mexico* Empalme, Mexico* San Bartolo, Mexico* Guanica, Puerto Rico* Monte Guilarte, Puerto Rico* *Added since Royer et al. (2005). 17

23 2. METHODS AND MATERIALS 2.1 Selection of Specimens and Site Locations All leaf fossils and digital leaf fossil images were obtained from the paleobotany collections at the Smithsonian Institution. After careful examination, useable leaf images were selected and prepped. In total, 156 leaf specimens, encompassing 43 leaf species, were considered suitable for DiLP processing procedures (Table 2). Fossil specimens were excluded due to issues of major laminal damage which prevented proper quantitative analyses and tooth selection. Such damage included insect herbivory and poor preservation. The five sites were collected from alluvial deposits in Wyoming s Bighorn Basin (Figure 1; Table 2). Sites were dated using stratigraphic correlation, assuming continuous deposition. Wing et al. (2000) have created two age models for the Bighorn Basin, which are based upon previously published age estimates (Age Model 1) and polarity reversals (Age Model 2). The current study relies on Age Model 2, and assumes a PETM base of 55.8 Ma (Gradstein, 2004). The oldest site, Skeleton Coast (Tiffanian 4a, NMNH site number 42041), dates to approximately 58.9 Ma. In total, 840 leaf specimens, including seven leaf species, were collected from the site (Currano et al., 2008; Wilf et al., 2006). However, for DiLP purposes, only seven leaf specimens (four leaf species) were analyzed (Appendix A). Lur d Leaves (Tiffanian 5b, site collection number 42042), the second oldest site that was analyzed, dates to 57.5 Ma. A total of 1,362 leaf specimens and 16 leaf species were recovered at the site (Currano et al., 2008; Wilf et al., 2006); twenty-nine specimens from 10 species were suitable for DiLP analysis from this site (see Appendix B). The youngest 18

24 Paleocene site, Daiye Spa (Clarkforkian 3; site collection number 41643), dates to 55.9 Ma. A total of 857 leaf specimens were collected, which included 16 leaf species (Currano et al., 2008; Wilf et al., 2006). For DiLP, 22 specimens from seven species were deemed useable (Appendix C). Specimens collected from the PETM site, Hubble Bubble (Wasatchian 0, site collection number 42384), date to ~55.7 Ma. At this site, 995 specimens, composed of 29 species, were collected (Currano et al., 2008; Wilf et al., 2006). For this study, 85 specimens from 17 species were used (Appendix D). Finally, the early Eocene site, Elk Creek (Wasatchian 2, site collection number ), dates to 55.2 Ma. There were a total of 6 different leaf species collected at this site consisting of 1,008 specimens (Currano et al., 2008). Thirteen specimens from five species were used for DiLP processing (Appendix E). 2.2 Processing Leaf Images The various leaf images were processed using Adobe Photoshop (Adobe Systems, San Jose, California, USA) with minor adjustments to the Royer et al. (2005) and Huff et al. (2003) procedures (see below). Leaf specimens were first categorized as possessing either toothed or untoothed margins (Currano et al., 2008). Several of the species, such as the Macgnitiea gracilis, originally considered untoothed, were found to contain both toothed and untoothed specimens according to the criteria of Royer et al. (2005) (Figure 2). 19

25 Figure 2. Representative examples of fossils used in this study. (A) Cercidiphyllum genetrix, collected at Daiye Spa (toothed) (B) Hartley, collected at Hubble Bubble (untoothed) (C) Rhus, collected at Hubble Bubble (toothed) (D) Rhus, collected at Hubble Bubble (untoothed). Toothed leaves were first inspected for damaged sections in the leaf margin. The leaf image was then copied and pasted into a new layer in order to preserve the entire unprocessed leaf image for future investigation. Undamaged sections of leaf margin were selected using the polygonal lasso tool from the Adobe Photoshop toolbox (Figure 3a and 3b). The petiole was not included in the selected leaf section in order to prevent its strong influence on variables related to perimeter. 20

26 Figure 3. Processing procedures for digital leaf physiognomy (Cercidiphyllum genetrix from Skeleton Coast). (A) Original image of fossil, with scale. (B) Undamaged sections are extracted. The lasso tool is dragged from the base of the leaf (not including the petiole) through the midvein, to the point directly beneath the first section of margin damage. The lasso line is then directed perpendicular to the page (either vertical or horizontal, depending on the positioning of the midvein), and undamaged teeth are extracted. Subsequent undamaged sections are selected in this manner. (C) Teeth are extracted from the laminal body. 21

27 In order to calculate tooth area: blade area, I selected out the portion of blade area that is functionally related to the selection of undamaged margin. Following the methodology of Cariglino (2007), the lasso tool was dragged from the base of the leaf through the midvein, to the point directly beneath the first section of margin damage. At this point, the lasso line was directed perpendicular to the page, straight from the midvein (either vertical or horizontal, depending on the positioning of the midvein) in order to select out the undamaged sections (Figure 3b). This method differs slightly from the Royer et al. (2005) and Huff et al. (2003) procedure, in which the lasso was directed perpendicular to the midvein. However, both methods pose selection error problems, often resulting in selections that are outside of the leaf perimeter. This occurred most commonly with palmately lobed leaves, and in these cases I selected the blade area to the closest first-order vein, not the midvein. The lasso was then used to trace out the toothed and untoothed sections back toward the base of the leaf. For those leaves whose bases were damaged, the midvein point directly beneath the first section of pristine leaf matter was lassoed, and the entire undamaged margin was selected out. After selecting the first undamaged leaf section, the add to selection option in the lasso toolbox was chosen, so that additional undamaged images could be combined with the previous selection. Undamaged selections both above and below the midvein, from the base to the tip of the leaf, were identified and selected out. Once all of the pristine leaf sections had been isolated, the cut option was chosen from the edit scroll down menu in the menu bar. All undamaged sections were pasted into a new layer in the Adobe 22

28 Photoshop program and this entire layer was then copied and pasted into another distinct layer (Figure 3b and 3c). Primary and secondary teeth were identified on each leaf section, and were selected following Royer et al. (2005). Lobes were distinguished from teeth by using the lobe vs. tooth rule (Royer et al., 2005). If possible, teeth were selected by moving the lasso tool from sinus to sinus (sinus is defined as an incision between marginal projections, which can be angular or rounded after Ellis et al., 2008). However, this method often resulted in erroneous tooth selection, especially when the tooth in question was located at the base of the leaf. In such cases, sinus to sinus selection was impossible, due to a lack of a basal sinus. Therefore, a continuous lasso line was drawn from the previous tooth through the base tooth. If this second method also resulted in an erroneous selection, the solitary tooth rule was used (Royer et al., 2005). In this procedure, an imaginary line was drawn vertically through the tooth, and the lasso line was then drawn perpendicular to this line from the previous sinus. Although this third method often resulted in the selection of smaller tooth areas, it was used in order to maintain a repeatable methodology. After all teeth had been selected, they were extracted from the blade section. The two layers containing the 1) extracted teeth and cut perimeters and 2) undamaged selections with the teeth still attached were then merged together in order to prepare these images for further processing (Figure 3c). Although complete processing was not necessary for untoothed specimens, certain leaf variables including inferred leaf area and major axis length (see Table 1 for definitions) were calculated for both toothed and untoothed leaves. The lasso tool 23

29 was used to select out the entire leaf image. For those specimens that were missing leaf area calculations (Ellen Currano provided area estimates for a majority of the leaf files; Currano et al. 2008), the entire perimeter of the leaves were outlined in Adobe Photoshop, and whole leaf area estimates were made using the Image-J program. Often, leaf specimens were missing significant portions of the lamina, such as bases or tips. In such cases, it was necessary to make educated estimates of the original leaf shape. For those leaves whose tips were not preserved, major length was estimated by tracing a line across the midvein past the damage point. The outline of the tip was formed by two lines, beginning at the start of the site of damage. The point at which these two lines converged on the extended midvein was considered to be the end of the leaf tip (Figure 4). When an entire section of the leaf was completely damaged, the area above or below the midvein was calculated and then doubled in order to predict entire specimen area. Such calculations assume that each of the specimens were symmetrical about the midvein. However, for some specimens leaf area could not be determined. Prepped leaf files were then opened in Image-J ( to quantify all physiognomic variables (see Table 1 for list). Before such assessments could be made, the scale was set using the set scale feature in the analyze tool bar. Threshold levels were set to allow the program to differentiate between the leaf and the pure white background. Using the wand selection tool, the individual leaf sections were selected and leaf blade area, perimeter, and internal perimeter were then measured. The length of the cut leaf sections were measured using the line tool. The number of teeth were then 24

30 counted, and their areas were summed. The line tool was also used to determine the major length of the leaf. Figure 4. Determining major length and leaf area for damaged leaves (legume, collected at Daiye Spa). Oftentimes, reconstruction of the leaf tip was necessary to determine both major length and area of leaf lamina. First, a line was traced across the midvein, past the section of damage. The outline of the tip was formed by two lines, beginning at the start of the site of damage. The point at which these two lines converged on the extended midvein was considered to be the end of the leaf tip. Major length and area estimates could then be made using the Image-J program. 25

31 2.3 DiLP Calibration Statistical leaf-climate relationships were determined from 30 extant sites that had been previously processed for DiLP (Table 3) (Royer et al. 2005; Cariglino 2007; Royer, unpublished). Various models were run using SPSS 15.0 in order to find the best fit multiple linear regression model. Those regression models with the highest r² and the lowest standard error were considered to be the best fit models for the data. Additional constraints included: 1) models could not consist of more than four variables; a regression formula which consisted of more than four variables was thought to decrease the general applicability of the model; 2) significance (P) value of each variable in model must be less than 0.05; and 3) the variance inflator factor (VIF) must be less than 10; a VIF greater than 10 would indicate non-independence among variables. Once the best fit models were determined for both MAT and MAP, they were applied to the fossil floras. Site means of the physiognomic variables were calculated from species means and MAP and MAT were determined using the multiple linear regression models. 26

32 3. RESULTS The most reliable multivariate model for reconstructing MAT had an r 2 of 0.90 and a standard error of 2.4 C. By comparison, Royer et al. (2005) obtained slightly better r 2 values (0.93) and standard error (2.0 C) based on a more limited data set (17 sites; see Table 4 for comparison). Three variables that are included in the current MAT model are margin percentage, perimeter ratio, and tooth area: blade area (see Table 1 for definitions). The variance inflator factor (VIF) was below the preferred threshold (<10) for all variables. The significance (P) levels for margin percentage and perimeter ratio were well below 0.05, and tooth area: blade area was only slightly above the preferred cut-off value (P = 0.055) (Table 4). The best fit model for MAP had an r 2 value of 0.85 and a standard error of 26 cm. The physiognomic variables included in this model were leaf area, feret diameter ratio, and average tooth area (see Table 1 for definitions, see Table 4 for model summary). Both VIF and significance values for all variables were within the range of tolerance (<10 and <0.05, respectively). 27

33 Table 4. Multivariate and univariate regression model summaries for MAT and MAP. See Table 1 for variable definitions MAT results are calculated from the model of Royer et al. (2005), which is calibrated with 17 extant floras. Current MAT and MAP multivariate and univariate models obtained from the analysis of 30 extant floras. (SE= standard error, P = significance, and VIF = variance inflator factor.) r 2 SE Physiognomic Coefficient P Variable value Constant VIF Multivariate Model 2005 MAT model DiLP MAT model Margin Percent Perimeter Ratio Tooth Area: Blade Area Constant Margin < Percent Perimeter Ratio Tooth Area: Blade Area Constant < DiLP MAP model Feret < Diameter Ratio Average Tooth Area Leaf Area < Univariate Models MAT model MAP model Constant < Margin < Percentage Constant Leaf Area <

34 Climate reconstructions for the late Paleocene, early Eocene, and PETM sites were calculated using the best fit regression models. Site means for each of the relevant variables were entered into the multiple linear regression formulas to calculate climate estimates for each locality (See Table 4 for variable coefficients). MAT reconstructions reveal relatively high temperatures at the PETM site (Hubble Bubble = 19.8 ± 2.0 ºC), compared to those estimates from the Paleocene (Skeleton Coast = 10.4, Lur d Leaves = 12.0, and Daiye Spa = 12.5 ± 2.0 ºC) and Eocene (Elk Creek = 16.0 ± 2.0 ºC) sites (Figure 5; Table 5). MAP estimates show an initial decrease in precipitation during the late Paleocene, followed by a subsequent increase during the PETM (Figure 6a and 6b; Table 6). MAP values for Skeleton Coast, Lur d Leaves, and Daiye Spa (the three Paleocene sites), are 103, 76, and 68 ± 26 cm, respectively. The PETM site MAP was calculated at 120 ± 26 cm. Elk Creek, our Eocene site, yielded an anomalous MAP of -12 cm. However, after removing Zizyphoides flabella from the analysis (see below for further discussion), I obtained a new MAT of 15.6 ± 2.0 ºC and a MAP of 26 ± 26 cm. In order to compare the multivariate and univariate approaches, I calculated MAT and MAP for the five sites using leaf-margin analysis and leaf-area analysis (Table 5, Table 6). The leaf-margin analysis equation was obtained from Wing et al. (2000); the leaf-area analysis equation was obtained from Wilf et al. (2000). To calculate the MlnA (the mean of the natural log of the species leaf areas), I first took the natural log of each species mean leaf area, and then averaged the species means. 29

35 Figure 5. Relationship between mean annual temperature (MAT), in ºC, and age of site, given in Ma. Standard errors for MAT are plotted. 30

36 Figure 6a. Relationship between mean annual precipitation (MAP), in ºC, and age of site, given in Ma. Standard errors for MAP are plotted. 31

37 Figure 6b. Relationship between mean annual precipitation (MAP), in cm, and age of site, given in Ma, for the Paleocene- Eocene Thermal Maximum (PETM). Standard errors for MAP are plotted. MAP estimates for SW037 and SW0410 were calculated by Wing et al. (2005). Daiye Spa and Hubble Bubble MAP estimates were calculated in this thesis using DiLP. 32

38 Table 5. MAT reconstructions for the late Paleocene, early Eocene, and PETM sites. [1] DiLP MAT estimates were determined from the relationships developed in this thesis. [2] Leaf-margin analysis reconstructions were calculated for the five sites, using the formula: MAT = P, where P is the proportion of untoothed leaves (Wing et al. 2000). Error values were calculated according to Wing et al. (2000). [3] Leaf-margin analysis climate reconstructions compiled by Currano et al. (2008): Tiffanian MATs calculated by compiling flora from several sites; Daiye Spa and Elk Creek MATs calculated from corresponding coeval sites. Time Interval Site and Zone Age (Ma) MAT (ºC) Paleocene Skeleton Coast (Tiffanian 4a) Lur d Leaves (Tiffanian 5b) DiLP [1] Leaf- Margin [2] Leaf Margin Analysis [3] ± ± ± ± ± ± 2.9 PETM Eocene Daiye Spa (Clarkforkian 3) Hubble Bubble (Wasatchian 0) Elk Creek (Wasatchian 2) Elk Creek (Wasatchian 2) ± ± ± ± ± ± *16.0 ± **15.6 ± ± ± 2.7 *Ziziphoides flabella was included in the data set. **Ziziphoides flabella was excluded from the data set. 33

39 Table 6. MAP reconstructions for the late Paleocene, early Eocene, and PETM sites. [1] DiLP estimates were determined from the relationships developed in this thesis. [2] MAP estimates using the univariate relationship ln(map) = 0.548MlnA (Wilf et al. 2000). MAP estimates for sites SW0307 and SW0410 were obtained from Wing et al. (2005). Time Interval Site and Zone Age (Ma) DiLP [1] MAP (cm) Leaf-Area Analysis [2] Paleocene PETM Eocene Skeleton Coast (Tiffanian 4a) Lur d Leaves (Tiffanian 5b) Daiye Spa (Clarkforkian 3) SW0410 (from Wing et al., 2005) Hubble Bubble (Wasatchian 0) SW0307 (from Wing et al., 2005) Elk Creek* (Wasatchian 2) Elk Creek** (Wasatchian 2) ± / ± / ± / / ± / / ± / ± /-77 *Ziziphoides flabella was included in the data set. **Ziziphoides flabella was excluded from the data set. 34

40 4. DISCUSSION The MAT and MAP estimates generated in this study are consistent with reconstructions based on other physiognomic techniques (i.e., those derived from leaf-margin analysis and leaf-area analysis). However, a major advantage of the DiLP approach is that it is more precise. Also, DiLP climate estimates are probably more reliable than those derived from other physiognomic methods, in part, because DiLP is less affected by the riparian bias and by issues of fossil fragmentation (Royer et al., 2005; Cariglino, 2007). This study also improves on previous DiLP research by expanding the extant calibration. A total of 30 flora are included in the modern site database (compared to the 17 flora used by Royer et al and 20 by Cariglino 2007) (Table 3). By extending the modern site list, and thereby increasing the range of modern MAP, more reliable climate reconstructions can be made. Therefore, this thesis marks the first time that DiLP methodologies have been used to reconstruct MAP. 4.1 MAT Reconstructions MAT estimates derived from DiLP analyses closely parallel those estimates obtained from leaf-margin analysis (Currano et al., 2008). MAT reconstructions using both leaf-margin analysis and DiLP reveal a warming at the PETM site, and consistently lower MATs for the Paleocene and Eocene sites (Figure 5; Table 5), which is consistent with our current knowledge of climate trends throughout the interval (Kennet and Stott, 1991; Wilf, 2000; Zachos et al., 2005; Currano et al., 2008). MAT estimates for all five sites were compiled by Currano et al. (2008). 35

41 However, due to the low species abundances during the Tiffanian, Currano et al. (2008) established MAT estimates for Skeleton Coast and Lur d Leaves by combining floras from several different sites (Table 5). Also, for the other two non- PETM sites (Daiye Spa and Elk Creek), Currano et al. (2008) did not estimate MAT for these sites directly but instead for nearly coeval sites elsewhere in the Bighorn Basin. In order to make the most appropriate comparison between our DiLP estimations for each of the five sites and those reconstructed through leaf-margin analysis, it was necessary to calculate MAT for these sites directly (see Wing et al for MAT formula). The difference in MAT calculations between the two physiognomic methods is quite small, ranging from 0.1 ºC (at Hubble Bubble) to 1.7ºC (at Lur d Leaves). These differences lie within the standard errors of the technique, indicating that both DiLP and leaf-margin analysis yield corresponding MAT reconstructions. It is important to note that although our reconstructions generally match previously published reconstructions, all four of our non-petm sites are relatively species-poor (Table 2). Only 4 taxa were analyzed from Skeleton Coast, 10 from Lur d Leaves, 7 from Daiye Spa, and 5 from Elk Creek. The total number of leaf specimens were also correspondingly low. Wolfe (1993) and Wilf (1997) suggest that at least 20 taxa are necessary for reliable climate reconstructions from leaf physiognomy. Therefore, all climate reconstructions presented here from non-petm sites need to be verified with more species-rich sites. (Such comparisons may be difficult to make, due to the fact that many sites have less than 20 woody dicotyledonous species). However, the climate reconstructions for these taxa-poor 36

42 sites nonetheless allow us to compare across methods. The close parallel between leaf-margin analysis and DiLP reconstructions for the late Paleocene and early Eocene sites suggest that the DiLP methodologies are in line with the alternative methods. Due to higher plant richness in the PETM site (Table 2, Appendix D), climate reconstructions from this site can be more reliably placed within the broader context of climate change during this time. MAT reconstructions from both leaf-margin analysis and DiLP reveal a sharp peak in temperature during the PETM. DiLP reconstructions yielded MAT values of 19.8 ± 2.0 ºC, while leaf-margin analysis yielded an equivalent, but less precise, estimate of 19.9 ± 3.7 ºC. 4.2 MAP Reconstructions DiLP and leaf-area analysis MAP reconstructions also reveal similar precipitation patterns throughout the PETM interval (Table 6). Previous MAP reconstructions using leaf-area analysis have suggested a decrease in precipitation at the onset of the PETM, followed by a return to initial late Paleocene conditions during the early Eocene (Wing et al., 2005). Using DiLP, the three late Paleocene sites yield a MAP range of 68 to 103 ± 26 cm, which is considerably lower than the PETM MAP value, calculated to be 120 ± 26 cm (Table 6). PETM results are consistent with those of Wing et al. (2005), who reconstructed MAP for two PETM sites, one pre-dating and the other post-dating Hubble Bubble. (Stratigraphically, Hubble Bubble is 18 m above the base of the PETM; Wing s upper locality (SW0307) is at +37 m, and his lower locality (SW0410) is at +3 m.) Wing et al. (2005) calculated an MAP of /-56 cm for the lower flora and /

43 cm for the upper flora. DiLP reconstructions reveal a MAP estimate of 120 ± 26 cm, which, when combined with Wing et al. (2005) estimates, suggests an overall trend of increasing precipitation throughout the PETM (Figure 6b; Table 6). The initial decrease in precipitation, followed by a return to pre-petm conditions in the early Eocene is consistent with the study of paleosols in the Bighorn Basin, conducted by Kraus and Riggins (2006). These results contrast with Bowen et al. (2004), who suggest a continual increase in humidity throughout the PETM interval. Leaf-area analyses of fossil assemblages from southern Wyoming reveal a similar decrease in MAP during the late Paleocene and a subsequent return to pre- PETM values during the mid-eocene (Wilf, 2000). Throughout the Tiffanian, MAP reconstructions obtained from the sites Bison Basin (~58 Ma), Clarkforkian (~56 Ma), and Big Multi (~56 Ma) reveal a decrease from 150 cm to 130 cm; values then increase during the mid-wasatchian (~53 Ma) to 140 cm. Although DiLP and the leaf-area based estimates of Wilf (2000) reveal similar precipitation trends throughout the late Paleocene and early Eocene, a direct comparison between the data sets is difficult because the current study is based on sites from northern Wyoming, while the field area of Wilf (2000) is southern Wyoming. In order to estimate MAP with leaf-area analysis for the five sites investigated in this study, I applied the univariate model of Wilf et al. (1998) and Wilf (2000) to the current data set (Table 6). Once again, it is important to note that the Skeleton Coast, Lur d Leaves, Daiye Spa, and Elk Creek sites are relatively species-poor. Although it is possible to make generalized comparisons between the DiLP reconstructions and those obtained with leaf-area analysis, no conclusive patterns 38

44 throughout the Paleocene and Eocene could be established. However, the high species richness of the PETM site allowed for direct comparisons between DiLP and leaf-area analysis. Here, Hubble Bubble yielded MAP estimates that were wetter than leaf-area analysis reconstructions (120 vs. 103 cm, respectively; see Table 6). Although most of the DiLP reconstructions were consistent with previously published MAP trends, Elk Creek our early Eocene site yielded an anomalous estimate of -12 cm. Due to the limited species sampling at Elk Creek (only 5 taxa were included in our calculations), physiognomic aberrations of individual species have a strong impact on species means, leading to erroneous climate calculations. The MAP regression model, which includes the tooth area: blade area variable, is affected by the inclusion of large-toothed samples, especially in undersampled locations. The taxa Zizyphoides flabella, which has unusually large teeth (Figure 7), may have been the cause of our erroneous MAP estimates. To test this idea, I conducted a second MAP reconstruction, which excluded both Zizyphoides flabella specimens from our model. MAP values increased to 26 ± 26 cm. The revised Elk Creek reconstruction still greatly underestimated MAP, when compared with previously published estimates (Wing et al., 2005). However, the new value was much closer to leaf-area analysis MAP estimates than the original reconstruction. The underestimation of MAP at Elk Creek may be the direct result of the site s low species richness. In order to obtain a greater understanding of MAP trends throughout the early Eocene, it may be necessary to compare the results of the current study with DiLP reconstructions yielded from coeval sites with a greater species richness. 39

45 Figure 7. Two Zizyphoides flabella specimens collected at Elk Creek. 4.3 Precision of DiLP Relative to Leaf-Margin Analysis and Leaf-Area Analyses Despite the similar climate reconstructions between methods at all five Paleocene and Eocene sites, I consider the DiLP approach the most reliable (Huff et al., 2003; Royer et al., 2005; Cariglino, 2007; see also introduction to discussion section in the current paper). Standard errors for the MAT estimates were 40