Statistical Analysis of the Atmsopheric Sulfate History Recorded in Greenland Ice Cores
STATISTICAL ANALYSIS OF THE ATMSOPHERIC SULFATE HISTORY RECORDED IN GREENLAND ICE CORES
DISSERTATION
Presented in Partial Fulfillment of the Requirements for
the Degree Doctor of Philosophy in the Graduate
School of The Ohio State University
By
Lijia Wei, M.S.
* * * * *
The Ohio State University 2008
Dissertation Committee:
Professor Ellen Mosley-Thompson, Advisor
Professor Lonnie G. Thompson Approved by
Assistant Professor Bryan G. Mark
Assistant Professor Catherine A. Calder ______Advisor Atmospheric Sciences Graduate Program
ABSTRACT
The Greenland Ice Sheet contains exceptionally valuable chemical and physical histories that allow reconstruction of paleoclimatic and paleoenvironmental conditions, particularly for the Northern Hemisphere. The chemical analyses of five multi-century long ice cores from the PARCA and Summit collections have yielded a high resolution volcanic aerosol history, which complements volcanic histories extracted from other
Greenland ice cores. A detailed ice-core volcanic index has been constructed and provides an improved estimate of the stratospheric sulfate burden which is an important input for models assessing the climatic impacts of volcanic eruptions. Additionally, these cores made it possible to confirm the timing of the arrival of the ash and aerosols from
Laki over Greenland. This time-stratigraphic horizon is an essential dating tool for high northern latitude ice cores, including those from Greenland.
The spatial characteristics of the sulfate aerosol deposition associated with specific eruptions provide information about the transport processes and the mechanisms dominating local deposition. Examination of the sulfate deposited from two eruptions, the
1783-84 A.D. Laki and the 1815 A.D. Tambora eruptions, reveals that precipitation
ii over the southeastern coastal regions in 1783 may have been suppressed by a regional cooling associated with Laki. This also suggests that Laki aerosols were likely deposited primarily by dry deposition. In contrast, the sulfate deposition from Tambora is more spatially homogeneous, suggesting primarily stratospheric transport and deposition primarily via wet processes. To quantify the impact of geographical factors on the deposition of volcanic sulfate over Greenland, a category explanatory variable analysis was conducted. The results indicate that the location of ice cores relative to north/south or east/west side of ice divide strongly affects EXS deposition, but the elevation of the core site is relatively unimportant.
Since 1850, the EXS flux extracted from Greenland ice cores has increased dramatically primarily as a result of anthropogenic sulfur emissions. To quantify this human impact as well as the effect of accumulation, a linear mixed model was applied.
The results indicate that for every Gg increase in the annual NH sulfur emissions, there is a 0.0013% increase in the annual non-volcanic excess sulfate flux. The impact of accumulation on sulfate deposition varies over Greenland, likely as a function of the dominant local depositional mechanisms. The linear slopes of accumulation versus sulfate were found to group naturally by the regional accumulation. The differences among the slopes likely reflect the regional strength of the role of dry deposition.
Additionally, local sources as well as the stochastic nature of depositional and post-depositional processes may also affect the sulfate flux deposition on the ice sheet.
Thus, it will be valuable to reconstruct the histories of other chemical constituents that contribute to the sulfate flux, such as those from marine biota and biomass burning. Also, close examination of the depositional processes, such as continuous observations of the
iii near surface and sulfate concentrations in fresh snow, may provide valuable information to improve our understanding of the relationship between the atmospheric sulfate background concentrations and the non-volcanic EXS flux deposited and preserved in the
Greenland Ice Sheet.
iv
Dedicated to my parents
Wei, Z. and Huang, B.
v ACKNOWLEDGMENTS
I wish to thank my advisor, Ellen Mosley-Thompson for the guidance, encouragement, and enthusiasm, which made this dissertation possible, and for her endless patience in correcting my linguistic errors. I am most grateful to Lonnie G.
Thompson, Bryan G. Mark, and Kate A. Calder for their scientific guidance and for serving as my dissertation committee members.
I am especially grateful to Mary Davis for her patience training me in the procedures for the dust analysis of ice core samples and her meticulous maintenance of the clean room that benefits the entire group. I would like to acknowledge Tracy
Mashiotta for chemical analyses, Ping-Nan Lin for δ18O analyses, Paolo Gabrielli for trace element analyses, and Victor Zagorodnov who was in charge of drilling many of the ice cores used in this study. I want to thank Peter F. Craigmile and Kate A. Calder who guided my development of the statistical model.
My colleagues and friends at the Byrd Polar Research Center and Geography
Department, Natalie Kehrwald, Lei Yang and Meng-Pai Hung, provided me much appreciated inspiration. My close friends at OSU, Yuxiong Ji, Fanyu Zhou, Xi Zhang, Yi
Zheng, and Yiqun Chen, made my four years in U.S. delightful.
Finally, I would like to express my appreciation to the various sources of financial
vi support I graciously received. The Ohio State University provided the University
Fellowship that supported me the first year of study while a U.S. National Science
Foundation award (NSF-OPP-0352527) supported three years of research and graduate education. My participation in the Crawford Point, Greenland drilling project was supported by the NSF Center for the Remote Sensing of Ice Sheets. The Dan David Prize
Scholarship provided support in my final year of research and the Rick Toracinta
Graduate Scholarship (OSU-BPRC) supported my presentation in 2008 at the Annual
Meeting of Association of American Geographers
vii VITA
June 21, 1979………Born – Xiamen, China
2001………………..B.S. Geology, Nanjing University
2001-2003…………Graduate Research Associate, Polar Research Institute of China
2003………………..M.S. Geochemistry, Nanjing University
2008………………..M.A.S. Statistics, The Ohio State University
2005 - present………Graduate Research Associate, The Ohio State University
PUBLICATIONS
Research Publication
1. Wei, L., E. Mosley-Thompson, P. Gabrielli, L. G. Thompson, and C. Barbante, Synchronous deposition of volcanic ash and sulfate aerosols over Greenland in 1783 from the Laki eruption (Iceland), Geophysical Research Letters, 35, doi:10.1029/2008GL035117, 2008.
2. Zhou, L., Y. Li, J. Cole-Dai, D. Tan, B. Sun, J. Ren, L. Wei, and H. Wang, A 780 year record of explosive volcanic eruptions from the DT263 ice core from East Antarctica, (Chinese) Science Bulletin, 51 (18), 2189-2197, 2006.
3. Wei, L. Y. Li, D. Tan, L. Zhou, M. Yan, K. Hu, J. Wen, B. Sun, and L. Liu, Research on Micro-particle Implicating Pollutants in Polar Regions, Advances in Earth Science, 25, 216-222, 2005.
4. Wei, L. Y. Li, D. Tan, L. Zhou, M. Yan, K. Hu, J. Wen, B. Sun, and L. Liu, Review of Research on Insoluble Micro-particle in the Polar Cores, Chinese Journal of Polar Research, 15, 274-282, 2003.
viii FIELDS OF STUDY
Major Field: Atmospheric Sciences
Minor Field: Statistics
ix TABLE OF CONTENTS Page Abstract………………………………………………………………………………..ii Dedication……………………………………………………………………………..v Acknowledgements…………………………………………………………………vi Vita………………………………………………………………………………viii List of Tables………………………………………………………………………..xii List of Figures……………………………………………………………………….xiv
Chapters:
1. Introduction…….………………………………..…………………………………1
2. Literature Review………………………………………..…………………………5
2.1 Greenland Ice Sheet overview………….……………………………………..5 2.2 Ice core paleoclimatology..………………….………………………...………8 2.3 General atmospheric circulation over Greenland……..………………………9 2.4 Annual snow accumulation…………….…………………….…………....…10 2.5 Surface temperatures…………………………….………………….………..15 2.6 Sulfate aerosols…………………………...………………….……..…….….17 2.7 Review of the various volcanic Indices……….…………………..…………21 2.7.1. Lamb’s Dust Veil Index (DVI)……………………………....……22 2.7.2. Mitchell Index………………..…..………...………………....……24 2.7.3. Volcanic Explosivity Index (VEI).……………………….……...…25 2.7.4. Sato Index…………………………..………………...... …...….26 2.7.5. Ice Core Volcanic Index (IVI) ……………….………..……..…….28
3. Data and methodology………………………………………..…….…………….30
3.1 Description of the ice cores used in this study.………..……………..……....32 3.1.1 PARCA ice cores…………………………………….…………..….34 3.1.2 Non-PARCA cores………………..…………………………………36 3.2 Laboratory analyses and data processing………..…………………………...38 3.2.1 Laboratory analyses……………………….………….……………..39 3.2.2 Ice core dating……………..………………………………….…….40 3.2.3 Calculation of fluxes and excess concentration…………….………42 3.2.4 Extraction of volcanic information……………………………..…...43 3.2.5 Estimating missing data…………………….……………………….44 3.3 Primary ice core data sets………………….….……………………………..55
x 3.4 Additional data sets…………………………………..……..……...………...56 3.4.1 Ice core-derived secondary data sets…………..……………………56 3.4.2 Non ice core-derived secondary data sets employed………………56
4. Volcanic aerosol history…………………………………..…………...………….59
4.1 An ice core derived volcanic aerosol history ………………….……….……60 4.2 Temperature anomalies associated with volcanic eruptions….………….…..68 4.3 Potential problems with ice core-derived volcanic indices……………...…...71 4.4 Timing of the arrival of Laki ash and sulfate aerosols to the GIS………..….72
5. Spatial variability of excess sulfate deposition over Greenland from the Laki and Tambora eruptions…...... 84
5.1 Review of two volcanic events…………………………….………..……….85 5.1.1 The 1783-84 A.D. Laki eruption (Iceland) …………………………85 5.2.2 The 1815 A.D. Tambora eruption (Indonesia) ……………….…..…86 5.2 Spatial variance of EXS…………………………………………..………….87 5.2.1 Raw data and synthesized variables………………….……….…….88 5.2.2 Spatial Interpolation…...……………………………..……………..91 5.2.3 EXS fluxes from Laki and Tambora to Greenland…..……….….…96 5.3 Analysis of variance for volcanic EXS………………….……………….…108
6. Analysis of non-volcanic excess sulfate time series………….……….…….……….112
6.1 Explanatory data analysis………………………………………………..…114 6.1.1 Assumptions regarding the distribution of model residuals………116 6.1.2 Temporal patterns and the inter-annual time dependency in the non-volcanic EXS time series……….…...……..……………...…120 6.1.3 Impact of regional accumulation rate and geographic factors on the non-volcanic EXS fluxes..……………………………….….……123 6.1.4 Impact of anthropogenic sulfur emissions on the non-volcanic EXS fluxes………………………………………………………..……124 6.2 Linear mixed model………………………………………………….……..126 6.2.1 Model development…………………………………….…….……127 6.2.2 Final model description……………………………………....……132 6.2.3 Checking model assumptions………………..……………..……...133 6.3 Inferences and climatic implication…..…………………………………….138
7. Conclusions and future work……………………………………………………….145
Bibliography………………...……….………………………...………...……………151
xi LIST OF TABLES
Table Page
2.1 Comparison of natural and anthropogenic sulfur emissions for 1990’s [data from Ramanathan et al., 2001]…………………………..…………………………...20
3.1 Information for the Greenland PARCA and Summit ice cores discussed in the text…………………………………………….………………………………...35
3.2 Information for earlier ice cores [Data from Clausen and Hammer, 1988]…….37
3.3 Missing sections in the Tunu, D3, and D2 ice cores…………………………….45
3.4 Correlation of EXS data in five Greenland ice cores from 1782 to 1850……….54
4.1 Annual volcanic EXS flux (kg/km2) for 5 Greenland ice cores…………………63
4.2 Common volcanic events from 1740 to 1985 in five Greenland ice cores……...66
4.3 Comparison of the different volcanic indices for selected eruptions……………67
4.4 Estimated background EXS and Laki eruption EXS in six Greenland ice cores..83
5.1 Laki EXS information from 19 Greenland ice cores……………………………93
5.2 Tambora EXS information from 17 Greenland ice cores……………………….94
5.3 Spherical model estimated parameters for the volcanic EXS flux, EXS density, NV-EXS flux, and accumulation in 1783 and 1816. The dash lines indicate the fitted spherical semivariogram models………………………………………….96
5.4 ANOVA table for the five-way main effect model…………………………….110
5.5 Contrast estimate and associated confidence intervals………………………...111
6.1 Non-volcanic EXS averages ( ) and variance (σ2) in five Greenland ice cores.122
xii 6.2 The square of the correlation coefficient (r2) between the annual NV-EXS and sulfur emission data…………………………………………………………..125
6.3 Comparison of the characteristics of five linear mixed models; AIC is the Akaike Information Criterion and BIC is the Bayesian information criterion, as discussed in p.129…………………………………..…………………………………….130
6.4 Parameter estimates for the fixed effects in the model………………………...139
6.5 Parameter estimates and confidence intervals (C.I.) for the fixed effect in the original scale…………...……………………………………………………....139
6.6 Parameter estimates for random effect in An…………………………………..142
xiii LIST OF FIGURES
Figure Page
2.1 Radar image of Greenland (courtesy of John P. DiMarzio and the ICESat Science Team) with the 1000, 2000, and 3000 m elevation contour lines (in red), added from Plate 1 in Bales et al. [2001a]………………………………………………7
2.2 Greenland accumulation map [modified from Bales et al., 2001a]……………..12
2.3 Locations of meteorological stations providing instrumental records and the locations of other places discussed in the text…………………………………..14
2.4 Global and regional sulfur emissions from 1850 to 2000 [data from Stern, 2005]…………………………………………………………………………….19
3.1 Map of ice core locations for this study…………………………………………33
3.2 Comparison of Ca profiles in Tunu1 and Tunu2 for overlapping sections……...46
3.3 Linear regression in estimating the trend component in the EXS flux in the D3 core……………………………………………………………………………..49
3.4 Residual plots for the D3 EXS data after removal of the trend: (A) residuals vs. time; (B) Normality plot; (C) ACF plot; (D) PACF plot. The dashed line represents the 95% confidence bound. The unit of lag is in year………………50
3.5 Comparison of EXS data in D3 using (A) the time series analysis, the bold line is the predicted value and the dash lines show the 95% prediction interval; (B) the multiple imputation (result of the first imputation)…………………………….51
4.1 Northern Hemisphere temperature anomalies, the IVI, and the VEI. Temperature anomalies are reconstructed from Briffa et al. [1992a, 1992b, 1995] and Esper et al. [2002], and modified by Jones and Mann [2004]. The VEI is constructed by Newhall and Self [1982]………………………………………………………...69
4.2 The annual average temperature for Ilulissat and Nuuk, Greenland, the IVI, and the VEI. The temperature data were compiled by Dr. Jason Box from Vinther et al. [2005] and Cappelen et al. [2006]. The VEI was constructed by Newhall and Self [1982]……………………………………………………………………...70
xiv
4.3 Seasonal variations in δ18O and insoluble dust used to date the entire core are shown along with the excess sulfate (EXS) for the time interval 1782-1819 from the (a) D2 and (b) D3 ice cores…………………………………………………75
4.4 δ18O in five Greenland ice cores for the 5-year interval containing the Laki eruption………………………………………………………………………….77
4.5 SEM photographs document the Laki tephra found in the D3 core (shaded boxes on the time axis). Photographs E-1 to E-3 and F-1 to F-3 are from filters E and F (in fine shading), respectively. These two filters with tephra contain snow deposited in the second half of 1783, coincident with the EXS peak (also shown). No volcanic fragments were found on the remaining 14 filters (see sampling discussion in section 3.2.1)……………………………………………………...80
4.6 Section of Core D2 (1782 to 1785 A.D.) containing the Laki eruption. (a) calculated concentrations of excess sulfate (EXS); (b) calculated concentrations of excess Tl (green), Bi (red), and Cd (blue); (c) normalized Lu/La ratio; (d) measured concentration of Cl-; (e) calculated ratio of Cl- to Na+; (f, g) seasonal variations in the insoluble dust concentrations and δ18O, respectively, used to construct the time scale. The shading highlights 1783 (darker) and 1784 (lighter)………………………………………………………………………….81
5.1 Empirical semivariograms for volcanic EXS flux, EXS density, NV-EXS flux, and accumulation for 1783 and 1816. The dash lines indicate the fitted spherical models.………………………………………………………………………….95
5.2 Spatial variability of the EXS fluxes from Laki and Tambora (units are kg/km2). (A) and (B) are the volcanic EXS fluxes in 1783 and 1816 shown on the same scale. (C) is the volcanic EXS flux in 1816 over a smaller range to highlight regional differences. The black points indicate the locations of the ice cores….97
2 5.3 Total SO4 (kg/km ) deposition from the Laki eruption to Greenland and surrounding areas averaged from the three simulations (modified from Oman et al., [2006]). The 2000 m elevation is shown for Greenland………………….....99
5.4 Spatial variability of the annual accumulation rate in 1783 and 1816 (units are mm w.e.). (A) is the 1783 An over a smaller range. (B) and (C) are An in 1783 and 1816 over the same range. The black points indicate the locations of the ice cores……………………………………………………………………………100
5.5 Accumulation histories for 5 Greenland ice cores (unit in mm w.e. a-1) from 1778 to 1788 and from 1810 to 1821. The long dash line indicates the mean of the period and the short dash line represents ±2σ…………………………………104
xv 5.6 Spatial variability of the NV-EXS fluxes in 1783 and 1816 (units are kg/km2). (A) is the NV-EXS flux in 1783 over a smaller range. (B) and (C) are the NV-EXS fluxes in 1783 and 1816 over the same range. The black points indicate the locations of the ice cores.……………………………………………………...105
5.7 Spatial variability of the volcanic EXS densities in 1783 and 1816 (units are 103kg/km2). (A) and (B) are the volcanic EXS densities in 1783 and 1816 on the same scale. (C) is the volcanic EXS density in 1816 over a smaller range. The black points indicate the locations of the ice cores……………………………107
6.1 Residual plots of the linear regression model using the original NV-EXS data. (A) Histogram of the residuals; (B) Normal Probability Plot of the residuals; (C) Residuals versus the fitted values; (D) Residuals versus the order of the data, entered as year 1782-1985 for Tunu, year 1782-1985 for Site A, year 1782-1985 for Site T, year 1782-1985 for D3, and year 1782-1985 for D2……………….118
6.2 Residual plots of linear regression model using Ln(NV-EXS) data. . (A) Histogram of the residuals; (B) Normal Probability Plot of the residuals; (C) Residuals versus the fitted values; (D) Residuals versus the order of the data, entered as year 1782-1985 for Tunu, year 1782-1985 for Site A, year 1782-1985 for Site T, year 1782-1985 for D3, and year 1782-1985 for D2……………….119
6.3 Annual fluxes of non-volcanic EXS (kg/km2) in five Greenland ice cores……121
6.4 ACF and Partial ACF plots of the residuals of the linear mixed model with a random An effect. (A) ACF for the within-group residuals; (B)-(F) PACF for individual ice core sites. The red lines indicate the 95% confidence bounds…131
6.5 Standardized residual plots for the linear mixed model. (A) Histogram; (B) Normal probability plot; (C) Residuals vs. fitted value plot…………………..135
6.6 Autocorrelation plot for the within-group residuals. The red lines indicate the 99% confidence bounds………………………………………………………..135
6.7 Normality plots of the residual for the linear mixed model. The blue lines indicate the 95% confidence interval for the data……………..………………136
6.8 Residual vs. fitted values plots for the linear mixed model for five ice core sites…………………………………………………………………………….138
6.9 Slopes for mixed effect in An…………………………………………………..143
xvi CHAPTER 1
INTRODUCTION
Ice cores from polar ice sheets, as well as from glaciers throughout the world, provide extraordinarily valuable records of Earth’s past paleoclimatic and paleoenviromental conditions. Greenland, as the second largest ice sheet, contains a unique record of the Northern Hemisphere’s climate history. The primary goal of this research project is to reconstruct an ice core-derived volcanic aerosol history using five multi-century cores from the NASA Program for Arctic Regional Climate Assessment
(PARCA) and Summit collections along with fifteen older records. These cores make it possible to examine the spatial variability of volcanic emissions, along with the preserved climate record, as they are widely distributed around Greenland. Their chemical analyses constitute a unique data set with the potential to quantify both the natural and anthropogenic contribution of sulfate aerosols to the Northern Hemisphere (NH) atmosphere.
Explosive volcanic eruptions release a variety of particles and gases into the atmosphere. Large explosive volcanic eruptions can inject gases and aerosols into the lower stratosphere and thereby significantly impact the climate on regional to global scales for up to 2-3 years. To assess the climatic impact of different volcanic eruptions, it is necessary to develop an accurate index that characterizes the explosive characteristics
1 of the eruptions and nature of the ejecta. Most of the previous efforts to reconstruct volcanic histories were based on geological criteria, such as the magnitude of the dust emissions [Lamb, 1970; Mitchell, 1970; Newhall and Self, 1982]. However, it is now known that most of their persistent, global scale climate effects result from the sulfur gases that convert to aerosols and modify Earth’s radiation balance. These gases also disrupt the stratospheric chemical equilibrium [Robock, 2000]. Recently, direct measurements of the stratospheric aerosol optical depth by satellite-borne sensors provide a more reliable index than previous indices [Sato et al., 1993], although its application is largely limited due to the relatively short length of observations (1979 to the present).
Overcoming these deficiencies, ice core-derived volcanic indices possess the advantages of high resolution and long time scales. The analyses of the volcanic signatures preserved in five well-dated ice cores, collected between 1985 and 1999 as part of PARCA and
Summit projects, are exploited to extract a reliable and accurate record for climatically influential volcanic events. Some well known volcanic horizons, such as those associated with the 1783-1784 A.D. Laki eruption, have been used for decades to constrain the time scales for Greenland ice cores. However, the timing of the arrival of the sulfate aerosols and volcanic fragments to the Greenland Ice Sheet (GIS) remains under discussion. Thus, another objective of this research is to resolve the timing of the deposition of the Laki sulfate horizon and provide an improved estimate of the associated stratospheric sulfate burden, which is an important input for models assessing climatic impacts of this volcanic eruption.
Furthermore, we incorporated data from fifteen other Greenland ice cores to examine the spatial variability of two volcanic events, 1783-1784 A.D. Laki and 1815
2 A.D. Tambora eruptions. These two events both produced global scale climate impacts, but are quite different in their location, their injection heights, and the sulfate transport time and trajectories. Thus their emissions should exhibit different depositional mechanisms and spatial patterns over the Greenland Ice Sheet (GIS). The analyses of these two events offer excellent insights that should prove valuable for interpretation of other past volcanic events, and for a more accurate assessment of their climatological effects.
Since the Industrial Revolution, the atmospheric sulfate aerosol concentration has increased as a result of human activities. The chemical analyses of these five PARCA and
Summit ice cores constitute a unique data set with the potential to quantify both the natural and anthropogenic contributions of sulfate aerosol to the Northern Hemisphere atmosphere. Moreover, these aerosol histories will be of particular interest for climate modelers who require better quantification of both the natural background and anthropogenic sulfate aerosol concentrations as well as estimates of the emissions contributed by specific volcanic eruptions. The PARCA and Summit cores provide an ideal opportunity to contribute to this body of knowledge.
Due to the extreme climate conditions and logistical challenges, only a limited number of ice cores exist. To extract as much information as possible from ice cores, many different analyses must be made from the limited mass available in these 100mm diameter cylinders. Unfortunately, when working with remnants of ice cores drilled and analyzed more than a decade ago for another project, as in the case of the PARCA cores, the complete cross-section at some depths in the core may have been depleted. Inevitably, situations of an inadequate supply of material and the resulting missing data occur in ice
3 core research. This is the case for the Tunu ice core from northeastern GIS, and the D2 and D3 cores from the western side of the GIS. Sections representing the late 1940s to late 1970s were consumed for beta radioactivity measurements. Earlier sections from the
1880s-1890s and late 1920s were also missing in D2. Fortunately, for the Tunu core, a nearby shallow core was available to replace the missing section. For the D2 and D3 cores, we explored two statistical methods to estimate the missing values based upon the data dependency. These approaches allow inferences to be made from the cores even though some data are missing. However, it should be noted that missing values lead to estimates with greater variability than would be expected if the original samples had been measured. This variability must be considered when using estimated parameters so that valid inferences may be drawn from “reconstructed” data.
4 CHAPTER 2
LITERATURE REVIEW
2.1 Greenland Ice Sheet overview
Greenland is the Earth’s largest island and 80% of its area is covered by permanent ice (Fig. 1). It is located between the Arctic and North Atlantic Oceans, and is bordered by North America to the west and Northern Europe to the east. Greenland extends ~2,670 km north to south and has a maximum width of 1300 km. The Greenland ice sheet covers an area of ~1.7 million square kilometers and is the Earth’s second largest ice sheet. Its average thickness is ~1600 m and reaches a maximum altitude of
~3,200 m in the central region [Thomas et al., 2001a; Nuttall, 2005]. The volume of ice on Greenland is about 2,350,000 km3 in water equivalent [Putnins, 1970], with almost
1/3 of its area lying below sea level. The marine sediment record suggests that formation of the GIS can be dated back to the end of Pliocene (~2.4 Ma B.P.) [Shackleton et al.,
1984; Funder, 1989]; however, the deepest ice core to bed rock in Greenland, the NGRIP ice core, extends back only ~123,000 yrs B.P. [NGRIP members, 2004].
Generally, the GIS gains mass by precipitation and loses mass through run-off of melt water, drifting snow, iceberg calving, basal melting, and surface water vapor loss by sublimation and/or evaporation [Paterson, 1998; Box and Steffen, 2001]. Estimation of the GIS mass balance is of great interest due to its potential contribution to global sea
5 level rise. During the Eemian, the contribution of the GIS to global sea level rise is estimated to have been ~4-5.5 m, a greater contribution than that from the West Antarctic
Ice Sheet [Cuffey and Marshall, 2000; Huybrechts, 2002]. The current ice sheet contains a water equivalent volume that could raise global sea level by about 6-7 m [Church and
Gregory, 2001; IPCC Fourth assessment report, 2007].
The Greenland Ice Sheet differs significantly from that on Antarctica. Unlike
Antarctica, which has a dominant influence on its own climate and on the surrounding ocean, Greenland’s climate is strongly affected by its proximity to nearby landmasses and the North Atlantic, with the Gulf Stream to the south and regions of North Atlantic deep water production to the east and west. Due to the height and location of the ice sheet, orographic influences are particularly pronounced. For example, shallow lows from the west are usually blocked and thus move northward along the west coast of Greenland. A low approaching the southern tip of Greenland may split into two parts under some circumstances, with one of the resulting lows moving off toward the northeast or east while the other moves northward along the west coast [Putnins, 1970].
6
Figure 2.1 Radar image of Greenland (courtesy of John P. DiMarzio and the ICESat
Science Team) with the 1000, 2000, and 3000 m elevation contour lines (in red), added from Plate 1 in Bales et al. [2001a].
7 2.2 Ice core paleoclimatology
The details of the Earth’s climate history extracted from ice cores reveal the range of past climate variability over multiple glacial and interglacial stages [GRIP Project
Members, 1993; Souchez, 1997; Spahni et al., 2005], and thereby allow 20th century changes to be assessed from a much longer-term perspective than that provided by meteorological and oceanographic observations. Since the 1960s when the first deep core was drilled to bedrock at Camp Century (now called GITS), numerous cores drilled in both Greenland and Antarctica have continually provided new details about the Earth’s climate history over many millennia. For example, two deep cores drilled in central
Greenland in the early 1990s, GRIP (3028.8 m) and GISP2 (3053.44 m), provided the most detailed reconstruction of the Northern Hemisphere climate history that extends back to ~110,000 years ago [GRIP Project Members, 1993; Grootes et al., 1993; Alley et al., 1993; 1995; Mayewski et al., 1996; Clausen et al., 1997]. Today, the deepest ice core in Greenland is the NGRIP core (3084.99 m), containing a climate history back to
~123,000 years ago [Flückiger et al., 2004; NGRIP members, 2004; Svensson et al.,
2005]. In Antarctica, a 906-m ice core drilled during 1977-78 at Dome C, extended back
~32,000 years [Lorius et al., 1979]. The deepest ice core from Antarctica was drilled to
3623 m at Vostok station and extended back through four glacial-interglacial cycles [Petit et al., 1997; 1999]. More recently, the long core (EDC99) drilled by EPICA (The
European Project for Ice Coring in Antarctica) at Concordia Station, Dome C, Antarctica, reached 3187m and extended back nearly 800,000 years [Jouzel et al., 2007; Loulergue et al., 2007, 2008; Lüthi et al., 2008]. The integration of the Greenland and Antarctic ice
8 core proxy records of stable oxygen isotopic ratio (δ18O), deuterium excess (δD), major ionic concentrations, and gas composition (CO2 and CH4) have provided a high resolution global picture of the Earth’s climate history [Bender, 1994; Anklin et al, 1997;
Chappellaz et al., 1997; Yiou et al., 1997; Blunier and Brook, 2001].
2.3 General atmospheric circulation over Greenland
The planetary scale atmospheric circulation around Greenland is characterized by semi-permanent low pressure systems. The primary systems include the Baffin trough, located over the eastern Canadian Arctic, and the Icelandic low pressure center, usually around 60ºN and between 30ºW and 50ºW [Barlow, 1997]. The ice sheet acts as a barrier in the prevailing paths of low pressure systems and strongly influences the atmospheric circulation [Putnins, 1970; Barry and Kiladis, 1982]. The large north-south temperature gradient and its orographic barrier effect enhance meridional flow and polarward heat exchange.
The major weather system producing precipitation over Greenland is the frontal cyclone [Chen et al., 1997]. The ice sheet affects cyclone dynamics by blocking, splitting, steering and cyclogenesis [Li, 2003]. Cyclogenesis is most favored along the southeast coast associated with the Icelandic Low. Statistical assessment of the mean cyclone tracks reveals a modest seasonal cycle in cyclone activities with a winter maximum [Serreze,
1995; Serreze et al., 1997]. This winter peak is partially due to the local development, with a significant contribution from the low-level baroclinicity associated with the land-sea boundary [Barlow et al., 1997]. As the frequent storm activity and cyclogenesis
9 in the vicinity of Greenland are coupled with the extensive surrounding open water, a considerable amount of water vapor associated with Greenland precipitation may be of local origin. However, isotopic assessment of the moisture origin indicates that the high-latitude moisture is not the dominant source [Johnsen et al., 1989]. A possible transport mechanism for the dominant moisture source proposed by Johnsen et al. [1989] is that cyclones originate from the polar front with a mean position of ~45º N and entrain low-level moisture originating from as far south as 30º N. As the systems decay, warm air is trapped above and loses contact with the ocean. As the cyclones containing this subtropical moisture continue to migrate poleward, water vapor becomes further depleted in the heavier isotopes through moist adiabatic and radiative cooling processes. The mechanism proposed by Johnsen et al. [1989] is different from, but is not necessarily in conflict with, the observation of frequent cyclogenesis in the vicinity of Greenland. In addition, much of the cyclogenesis can be traced to the redevelopment of decaying systems originating from further south [Barlow et al., 1997].
2.4 Annual snow accumulation
Changes in annual accumulation rate (An) are controlled by both local (drifting) and larger-scale processes. The large-scale atmospheric circulation exerts strong control over the motion of air masses that bring moisture to Greenland. Past temporal changes in accumulation, as well as variations in the spatial distribution, may provide additional insight to variations in past atmospheric circulation patterns. Likewise knowledge of the spatial distribution of the deposition of atmospheric sulfate from a known volcanic
10 eruption may also allow inferences to be drawn about past large-scale circulation patterns
[Thordarson and Self, 2003].
A comprehensive assessment of the net annual accumulation and mass balance history for Greenland was conducted by the PARCA project. A number of different approaches were employed including the use of historical and recent records [Bales et al.,
2001b], ice cores [Bales et al., 2001c; McConnell et al., 2001; Mosley-Thompson et al.,
2001], and satellite-borne radar and aircraft-borne laser altimeter data [Thomas et al.,
-1 2001b]. The average An is estimated to be ~300 mm water equivalent per year (w.e. a )
-1 with an average uncertainty (standard deviation) of roughly ±70 mm w.e. a . The An in northeast Greenland is poorly constrained due to lack of observations. Large spatial differences exist across the ice sheet (Fig. 2.2). Accumulation is higher in the south, southeast, and west and it is lower near the central area and in the northeast [Bales et al.,
2001a]. For interior region above 1800 m elevation, the average accumulation rate is also
~300 mm w.e. a-1.
11
Figure 2.2 Greenland accumulation map [modified from Bales et al., 2001a]
12 Over Greenland, accumulation rate and temperatures can increase significantly over a few years to decades [Alley et al., 1993, Steffensen et al., 2008]. Depending on the near-surface temperature, summer melting occurs over about 50% of the ice sheet surface
[Thomas et al., 2001a]. Overall, the high elevation parts of the ice sheet are in balance to within about 10 mm w.e. a-1 [Thomas et al., 2001b]. In contrast, most of the coastal regions thinned very rapidly during 1990s [Krabill et al., 2000].
Modeled annual precipitation distributions by Box et al. [2004] show that the precipitation maxima lie along the southeastern slope, near 72ºN along the western slope, and above Melville Bay (see Fig. 2.3 for location), consistent with the kriged accumulation map by Bales et al. [2001a] (Fig. 2.2) derived from ground-based measurements. Based on the model simulation, liquid precipitation constitutes up to 70% of the total annual precipitation.
13
Figure 2.3 Locations of meteorological stations providing instrumental records and other places discussed in the text
14 The spatial variability of accumulation reconstructed from ice cores indicates that
Greenland precipitation is strongly modulated by changes in North Atlantic atmospheric patterns [Mosley-Thompson et al., 2001]. Generally, under the negative phase of the
North Atlantic Oscillation (NAO), there is more large-scale atmospheric flow from the southwest, bringing more moisture to the ice sheet, especially to the southern region.
Conversely, under the positive phase of the NAO, stronger westerlies reduce southwesterly flow, resulting in an overall reduction of accumulation. The strongest linear relationship (negative) between snow accumulation and NAO is found to the west of central Greenland [Hurrell, 1995, Mosley-Thompson et al., 2005]. However, the spatial character of the precipitation response to NAO variability appears to have been influenced by the 20th century warming in the high Arctic [Mosley-Thompson et al.,
2005]. Using a Bayesian spatially varying coefficient regression model, Calder et al.
[2008] examined the spatial variation in the NAO signature preserved in the GIS. They confirmed that the accumulation in western Greenland (near the NASA-U core) is most strongly correlated with the NAO. However, they also identified a region southeast of the
NASA-U core, where the linear relationship between An and NAO is weaker than that for western Greenland, but is more predictable.
2.5 Surface temperatures
Monthly mean temperatures over the GIS usually show large inter-annual variability, increasing from south to north, especially during winter [Barry and Kiladis,
1981; Ohmura, 1987, Box et al., 2002]. The automatic weather stations (AWS) data from
15 the Greenland climate network confirm that the highest monthly mean temperatures usually occur in July; while the lowest monthly mean temperatures occur in February
[Steffen and Box, 2001]. The annual mean latitudinal temperature gradient over the western slope of the ice sheet is estimated to be -0.78 ºC per 1º latitude; while for the eastern slope, it is estimated to be -0.82 ºC per 1º latitude. Instrumental temperature records from 1873-2001 reveal a 1-3.5 ºC warming from 1885 to 1935 at five sites
(Ivigtut, Nuuk, Tasiilaq, Ilulissat, and Upernavik, see Fig 2.3 for locations) north of
Ivigtut, Greenland [Box, 2002]. In western Greenland, a 3-5 ºC warming was observed with a maximum in the 1930s, following by a 1-2 ºC decline until the resumption of a warming trend in the mid 1980s. Overall, the largest changes in average annual temperatures occur in winter and increase with latitude.
A longer temperature history for Greenland has been extracted from the measurements of the oxygen isotopic ratio (δ18O) on deep ice cores [Johnsen et al., 1995].
Basically, more negative δ18O values reflect cooler condensation conditions while less negative δ18O values reflect warmer air temperatures at the time of condensation. In
Greenland, the slope of the linear relationship between δ18O and surface air temperature ranges from -0.62 to -0.67‰ per ºC [Dansgaard et al., 1973; Johnsen et al., 1989].
However, Cuffey et al. [1995] caution that the slope values may also depend upon many factors other than local environmental temperature, such as the changes in sea-surface composition and temperature, atmospheric circulation, cloud temperature, and post-depositional isotopic exchange in the snowpack, that may vary over time.
16 Despite the large regional differences that exist among the Greenland δ18O records, the modest 20th century warming over Greenland does appear in most cores although to differing degrees [Mosley-Thompson et al., 2006]. Although this warming is contemporaneous with the 0.53 ºC increase in global mean temperature (GMT) over the past 150 years [Parker et al., 1995; Jones et al., 1999], a recent cooling (-1.29 ºC) was recorded from 1985 to 2001 at eight coastal stations in southern Greenland (Aasiaat,
Sisimiut, Kangerlussuaq, Tasiilaq, Nuuk, Paamiut, Narsarsuaq, and Qaqortoq, see Fig. 2.3 for location) [Hanna and Cappelen, 2003]. This strong cooling trend was significantly and negatively correlated with the 5-year unweighted running mean of the winter index of NAO. In other words, when the NAO is more strongly positive, temperatures over
Greenland tend to be colder.
2.6 Sulfate aerosols
Atmospheric aerosols play a very important role in the Earth’s energy balance
[Novakov et al., 2003; Hansen et al., 2004; 2005] and thereby have the potential to strongly force the climate system. In addition to their critical role in cloud condensation processes, aerosols exert a direct radiative forcing by scattering and absorbing shortwave radiation and by their role in heterogeneous chemical reactions [Andreae and Crutzen,
1997; Haywood and Boucher, 2000]. On regional scales, anthropogenic aerosols also affect surface and atmospheric heating, as well as precipitation processes [Ramanathan et al., 2001]. Therefore, it is essential to quantify the radiative contributions of the different
17 aerosols so that their various roles in climate forcing can be assessed [Charlson, 1992;
Haywood et al., 1999; Satheesh et al., 1999].
The atmospheric concentration of one particularly important radiatively active species, sulfuric acid, can produce significant climate perturbations. For example, sulfur aerosols emitted by large volcanic eruptions can have a strong radiative forcing and act as a surface for heterogeneous chemical reactions that destroy stratospheric ozone. Sulfate aerosols can also significantly affect the formation and maintenance of cirrus clouds
[Sassen et al., 1995], which effectively absorb long-wave radiation.
Atmospheric sulfur originates from both natural and anthropogenic processes.
Natural sulfur originates from sea salt, biogenic productivity, volcanic emissions, and terrestrially-derived windblown dust [Andreae et al., 1986; Legrand, 1995; Legrand et al.,
1997a]. Anthropogenic sulfur is emitted primarily from coal burning power plants and any other processes that burn fossil fuel [Georgii and Warneck, 1999; Smith et al., 2001].
Since the Industrial Revolution, the concentration of atmospheric sulfur has increased as a result of human activities, as shown in Figure 2.4 [Stern, 2005]. On a global scale, anthropogenic emissions of sulfate now exceed those from natural sources (see Table 2.1)
[Ramanathan et al., 2001].
18 W. Europe N. America 60000 E. Europe Asia Oceania Mid-East S. America Africa 40000 Northern Hemisphere Southern Hemisphere Gg S
20000
0 1860 1880 1900 1920 1940 1960 1980 2000
Year
Figure 2.4 Global and regional sulfur emissions from 1850 to 2000 [Data from Stern,
2005]
19 Flux Lifetime Column Burden Optical depth Source (Tg/a) (days) (mg/m2) scattering/ absorption
Natural:
Sulfate from biogenic gases 70 5 2 1.6 Sulfate from volcanic SO 2 20 10 1 0.8 (troposphere) Sulfate from Pinatubo (1991) 40 400 80 16
Anthropogenic:
Sulfate from SO2 140 5 3.8 3
Table 2.1 Comparison of natural and anthropogenic sulfur emissions for 1990’s [data from Ramanathan et al., 2001]
The current sulfate budget in the Northern Hemisphere is highly dominated by anthropogenic sulfur emissions. Today, the main sources of sulfate aerosols are SO2 emissions from fossil fuel burning (~72%), dimethyl sulfide emissions by marine phytoplankton (~19%), SO2 emissions from volcanic eruptions (~7%), and emissions from biomass burning (~2%) [Haywood and Boucher, 2000]. The history of the relative abundance of sulfur isotopes in the Greenland summit cores reveals the changing nature of the contributing sources [Nriagu et al., 1987, 1991]. Patris et al. [2002] report that the sulfur isotopic ratio [34S/32S] in the GRIP and EUROCORE ice cores indicate that the preindustrial background concentration of non-sea salt sulfate is dominated, on an annual average basis, by marine biogenic emissions (49%), followed by continental (44%, mainly sporadic volcanic activity and continental biota) and terrigenous contributions
20 (7%, mainly dust particles, such as CaSO4 or MgSO4), respectively. However, over the last 150 years, the sulfur isotopic ratio has become progressively smaller, consistent with the increase in human-derived sulfate emissions that are depleted in 34S. However, this isotopic change is not very evident in the Southern Hemisphere. Sulfur isotopic ratios from Dronning Maud Land, Antarctica, indicate no significant temporal change in sources over the last 1100 years [Jonsell et al., 2005].
2.7 Review of the various volcanic indices
Volcanism has long been regarded as an important mechanism affecting weather and climate variation. In 44 B.C., Roman authors, Seneca and Plutarch, observed the strange sunsets caused by the eruption of Mount Etna, and pointed out that it led to a cooling and the subsequent crop failures and famine in Rome and Egypt [Forsyth, 1988].
In 1816, people experienced “the year without summer”, which is said to have resulted from the eruption of Tambora in 1815. Modern studies of the climatic effects of volcanic eruptions started with the examination of the 1982 eruption of the El Chichón (Mexico)
[Oberbeck, et al., 1983; Hofmann and Solomon, 1989]. Recently, volcanic forcing has received more attention due to the availability of advanced technology such as satellite-borne sensor observations. By the 1991 eruption of Mt. Pinatubo (Indonesia), observations were well organized and more systematic [Hansen et al., 1992; McCormick and Veiga, 1992].
Examining the climate response to a known volcanic eruption helps better understanding of the important radiative and dynamical processes and provides realistic
21 and detailed data for modeling. Thus, efforts to reconstruct the atmospheric aerosol loading and transport associated with specific volcanic events, such as the eruption of
Laki (Iceland) in 1783-1784 A.D., have contributed to a better understanding of the natural fluctuations in the climate system [Thordarson and Self, 2003 and the references therein].
To investigate the climatic effect of a volcanic eruption, it would be desirable to have a robust index that is sensitive to and proportional to the impact of the volcanic eruption on climate. Ideally, a robust index should provide a consistent indication for the associated climate impact of eruptions. However, information derived from direct measurements is limited to those taken at the surface, or by aircraft and balloon.
Satellite-borne sensor measurements are limited to the last half century. Volcanic indicies have been developed by a number of scientists using a variety of approaches. The most widely discussed indices were developed by Lamb [1970], Mitchell [1970], Newhall and
Self [1982], Sato et al. [1993], Hammer [1977], and Robock and Free [1995]. At the same time, Pollack et al. [1976], Harshvardhan [1979], Hansen et al. [1992], and
Stenchikov et al. [1998] published theoretical studies of the radiative effects of volcanic eruptions.
2.7.1. Lamb’s Dust Veil Index (DVI)
The first extensive effort to construct a volcanic eruption index was by Lamb
[1970, 1977, 1983]. To provide a consistent numerical index for past volcanism, Lamb developed the dust veil index (DVI), taking into account the temperature perturbation and
22 the dust emissions associated with specific eruptions. Depending upon the available data, five parameters, RDmax, Emax, tmo, TDmax, and q were included into the formulae in the calculation of the DVI.
DVI = 0.97 RDmax Emax tmo (2-1)
DVI = 52.5 TDmax Emax tmo (2-2)
DVI = 4.4q Emax tmo (2-3)
In Lamb’s formulae, RDmax represents for the maximum percentage depletion of direct radiation. Emax is the related maximum geographical or spatial extent attained by the dust veil. For example, eruptions producing a globally dispersed dust veil would be assumed to have Emax = 1. And tmo is the total time in months between the eruption and last observation of the dust veil or its observed impact on monthly radiation or temperature. TDmax is the estimated reduction of the average temperature in degrees
Celsius over the mid latitude zone of the hemisphere affected in the year receiving most influence. The fifth parameter used in Lamb’s formulae is q, the estimated total volume of solid matter in cubic kilometers dispersed as dust in the atmosphere. In cases where ample information is available, Lamb’s three formulae provide alternative estimates of the same dust veil and an internal check for reliability. Using this method, Lamb [1970] reconstructed a detailed volcanic history for the past 500 years.
Lamb’s meticulous effort provided the pioneering work and a solid basis for later development of other volcanic indices. When Lamb attempted to relate great volcanic dust veils (he used only DVI >100) and climate, he found high correlation coefficients between DVI and numerous climatic parameters, such as the Northern Hemisphere
23 average temperature, Arctic sea ice along the coast of Iceland, and long-term atmospheric circulation over the North Atlantic, among others. For example, Lamb correlated DVI decadal values from 1790 to 1959 with the limited pressure data for the general circulation over the North Atlantic in January (represent by the overall range of pressure between regions of highest and lowest monthly mean pressure) and found a correlation coefficient of -0.41. Nevertheless, now it is recognized that the climate impacts caused by volcanic eruptions result primarily from the radiative effects of the emitted sulfur gases and aerosols that form rather than the dust. Also, Lamb’s DVI has been often criticized for its reliability as climatic information was used in DVIs derivation. Robock [1981] created a modified version of DVI without temperature information, but when both DVIs were used in climate model simulations, the results were not significantly different
[Robock, 2000].
2.7.2. Mitchell Index
J. Murray Mitchell also developed a volcanic index based on the total atmospheric loading by volcanic dust [Mitchell, 1970]. The reconstruction of his volcanic chronology came primarily from his research on the role of atmospheric pollution in global temperature fluctuations. Two globally extensive pollutants, carbon dioxide and particulates, were examined for their long-term changes and their impact on changes in the globally average temperature change over the past century. To separate the human contribution to atmospheric dust loading, Mitchell reconstructed the atmospheric dust loading associated with volcanic eruptions from 1855 to 1968. His index was based
24 mainly on volcanic eruption data from Lamb. He assumed 1% of the total ejected mass reached the stratosphere as fine dust with a uniform 14 months residence time for all eruptions. Mitchell neglected the remaining 99% of the ejecta, assuming that these large particles are usually removed rapidly from the atmosphere. Also, the severity of the eruptions south of 16°S was degraded because of their modest impact on the Northern
Hemisphere. Based on the estimated total ejected mass, Mitchell classified volcanic eruptions from 1855-1968 into four severity classes, i.e. class 1 (total ejected mass ~1010 metric tons), class 2 (~109 metric tons), class 3 (~108 metric tons), class 4 (~107 metric tons). Fractional classes were also used to represent compromises. For example, 1963
Agung eruption was rated to have a severity class of 1-1/2.
Unlike Lamb, Mitchell included eruptions with DVI < 100, and thus he reconstructed a more detailed volcanic index. However, similar to the DVI, Mitchell’s index focused on the climate effects due to the impact of dust, rather than the sulfur gases.
The assumption of a 14-month residence time for all eruptions, regardless of geographical location, is likely inaccurate as ejecta from eruptions with lower injection heights will have much shorter residence times and hence a more modest climatic impact.
2.7.3. Volcanic Explosivity Index (VEI)
Considering the incompleteness of Lamb’s DVI and Mitchell’s index, Newhall and Self [1982] designed a composite estimate of the magnitude of past explosive eruptions, the volcanic explosivity index (VEI), to extract a more complete history of volcanic eruptions. To extend the volcanic index further back in time, they considered
25 only the volcanological data (no atmospheric data). Depending upon one or more of
VEI’s criteria, such as, volume of ejecta, column height, eruption duration, tropospheric and stratospheric injection, eruptions were assigned a VEI on a scale of 0 to 8. For eruptions without further information, usually a default VEI of 2 was assigned. Using these criteria, VEI’s for over 8000 eruptions from 1500s to 1970s were estimated. In addition to this volcanic index, Newhall and Self [1982] also provided a comprehensive summary of the decadal frequency of each VEI category.
As the composite estimate for past volcanic eruptions, VEI was proposed as a semiquantitative compromise between poor data and the need in various disciplines for reconstructions of past volcanism. As a relatively complete volcanic index, VEI provides great assistance in assessing the completeness of other volcanic indices. However, the utility of VEI as an indicator of an eruption’s climatic influence is limited. For example, although the 1883 A.D. Krakatau eruption (Indonesia) was assigned a VEI of 6, it produced only a modest climate impact due to its significant output of ash particles but its limited quantity of injected gases.
2.7.4. Sato Index
More recently, the important climatic role of sulfur gases from volcanic eruptions has gained attention. Large explosive volcanic eruptions can eject sulfur gases into the stratosphere, where they are oxidized into sulfate aerosols that affect Earth’s radiation balance, primarily by reflecting and absorbing solar and terrestrial radiation. To assess the aerosol climate forcing, Sato et al. [1993] used stratospheric aerosol optical depths as the
26 principal parameter for their index. Data were divided into four periods depending on their sources. For example, aerosol optical depths for period 1 (1850-1882) were estimated based on the volume of ejecta from large known volcanoes. From 1883 to 1959, measurements of solar and stellar extinction became available and were used to estimate optical depths, although these measurements were mainly confined to mid-latitude regions. After 1960, the solar and stellar extinction measurements became more widespread and some in situ sampling of aerosol properties was conducted. Together, these data contributed to a more precise estimate of the stratospheric aerosol optical depths for period 3 (1960-1978). Since late in 1978, optical depths have been measured by the stratospheric aerosol monitor on the Nimbus 7 satellite. Therefore, during period from 1979 to 1990, estimates for aerosol optical depth was made from much more precise satellite measurements combined with other available data.
Sato’s method provides a new way to construct a more accurate index for past volcanism. His index is based upon aerosols, known to be the primary cause of climatic perturbations by volcanic eruptions. Therefore, compared with previous indices, Sato’s index is more reliable and precise. However, as the satellite-borne measurements are available only since the late 1970s, and the estimates for the early period (such as
1850-1882) were based only on the qualitative report of atmospheric optical phenomena,
Sato’s index is limited in length, and thus for its application to assess earlier volcanic impacts.
27 2.7.5. Ice core-derived volcanic Index (IVI)
The pioneering work on the ice core-derived volcanic index (IVI) was conducted by Hammer [1977], who demonstrated that volcanic eruptions may be detected by elevated electrical conductivities in the Greenland ice sheet. More recently, ion chromatography has been used to measure sulfate ion concentrations from which the sea salt contribution is removed to attain an estimate of excess sulfate (EXS). It is now widely recognized that the climatic effect exerted by large volcanic eruptions results primarily from the emitted sulfur aerosols, which produce a significant, albeit short-term, forcing on the climate system [Crowley, 2000; Mann et al., 2005].
Based on previous reconstructions of ice core-derived volcanic indices, Robock and Free [1995] and Robock [2000] combined ice core sulfate and ECM records to construct a global volcanic index. This new index correlated well with the existing non-ice-core volcanic indices. However, high-latitude volcanic eruptions were given too much weight due to the geographical proximity to the ice sheets. Also, the more recent increase in sulfate background concentrations and its inter-annual variability has obscured all but the largest sulfate peaks. Therefore, methods to extract volcanic signals from ice cores need to be selected with caution. As the sulfate background concentrations in snow have remained relatively constant in the Antarctic, Cole-Dai et al. [1997] suggested that when the concentration exceeds the background by 2 standard deviations
(2σ), the signal could be attributed to a volcanic eruption. In addition, tephra associated with the particular layer can be examined for chemical finger-prints to confirm the presence of an eruption and help to identify the volcanic source. A recent technique for
28 extracting volcanic signals was suggested by Gao et al. [2006]. They used a high-pass loess filter [Cleveland and Devlin, 1988] and assumed that signals exceeding the baseline plus 2 running median absolute deviations (MAD) were likely to be volcanic events. This approach allows relatively small volcanic peaks to be selected from the changing background and therefore it provides a practical approach for detecting volcanic events in sulfate time series from Greenland ice cores that are contaminated with anthropogenic sulfate.
29 CHAPTER 3
DATA AND METHODOLOGY
This research project is designed to reconstruct an ice core-derived volcanic aerosol history using the multi-century cores from the PARCA collection. As part of this project, the volcanic sulfate aerosols, trace elements, and tephra from two high resolution ice cores from the west side of the GIS, made it possible to confirm the timing of the ash particles and sulfate from the 1783-84 A.D. Laki eruption (now in press as Wei et al.,
2008). In addition, the spatial patterns of the sulfate deposition from the Laki eruption and the 1815 A.D. Tambora eruption were compared as they represent two distinct types of volcanic eruptions. The chemical analyses of these PARCA cores also constitute a unique data set with the potential to quantify both the natural and anthropogenic contributions of sulfate aerosol to the Northern Hemisphere atmosphere.
The following sections describe the ice cores used in this research project, the acquisition of the primary ice core data used, the construction of time scales, which is applicable to the generation of secondary ice core data sets. The discussion also includes the use of two different approaches to generate the data we substituted for missing data.
Other non-ice core related data sets used in the project are also described.
Three primary data sets were produced to fulfill the scientific objectives of this dissertation project. The first data set consists of the continuous (top to bottom) record of
30 the concentrations of the major anions and cations preserved in the five multi-century
PARCA and Summit ice cores (Fig. 3.1; Table 3.1). The second data set produced as part of this project is the detailed excess sulfate flux associated with two major volcanic eruptions—the 1783-84 A.D. Laki eruption (Iceland) and the 1815 A.D. Tambora eruption (Indonesia), as recorded in 19 Greenland ice cores (Tables 3.1 and 3.2). The third data set includes the chemical, particulate, and trace element concentrations for the ice core sections in two high accumulation regions associated with Laki eruption.
Two advanced data sets were produced from the primary data sets. The first data set consists of multiple high-resolution volcanic aerosol histories as preserved in the five multiple-century PARCA ice cores. The second data set consists of the non-volcanic EXS flux time series for these five ice cores, so as to examine both the natural and anthropogenic contributions to the atmospheric sulfate budget without the perturbations introduced by volcanic eruptions.
In addition to the ice core data sets, three other data sets were used in this research project. The 1850-2000 sulfur emission data set from Stern [2005] was used for comparison with the annual non-volcanic EXS flux in the ice cores. The NH temperature anomalies reconstructed from tree ring data [Briffa et al. 1992a, 1992b, 1995; Esper et al.
2002; Jones and Mann, 2004] were compared to our derived EXS history. In addition, the atmospheric temperature records from 1768 to 2005 from two Greenland coastal stations,
Ilulissat and Nuuk [Cappelen et al. 2006; Vinther et al. 2006], were used to examine the climate perturbations associated with large volcanic eruptions.
31 To explain the process of data set construction and the experimental approach, the rest of this chapter will discuss in the following topics: (1) information on the twenty-one ice cores used in this research project; (2) laboratory analyses and data processing, including ice core dating, flux and excess concentration calculations, volcanic signal extraction, and estimation of missing data; (3) construction of the primary ice core-derived data sets; (4) discussion of the additional data sets, including the secondary ice core-derived data sets and three other data sets used.
3.1 Description of the ice cores used in this study
Twenty-one ice cores (see Fig. 3.1, Tables 3.1 and 3.2) were used in this study, eight from the PARCA project, two Summit cores from earlier drilling projects in preparations for the Greenland Ice Sheet Drilling Program (GISP II), and eleven ice core records published by Clausen and Hammer [1988].
32
Figure 3.1 Map of ice core locations for this study
33 3.1.1 PARCA ice cores
Prior to 1995, less than a dozen multi-century long ice cores had been drilled and only a few of them were analyzed continuously and in high resolution for major anions and cations. Most efforts to reconstruct volcanic histories from ice cores were based on the analysis of selected sections, initially identified by continuous, detailed electrical conductivity measurements (ECM) as conductive, and thus as potentially volcanic. As part of PARCA, seven new multi-century long ice cores (see Table 3.1 and Fig. 3.1) were collected between 1995 and 1999, specifically for reconstruction of their annual net mass accumulation histories. These cores were analyzed at sub-annual resolution for insoluble dust and δ18O at The Ohio State University [Mosley-Thompson et al., 2001, 2003] and by
2+ continuous flow methods for H2O2 and Ca as well as other chemical species at the
University of Arizona but not for sulfate [Anklin et al., 1997; McConnell et al., 2000a,
2000b, 2001].
For the purposes of this research, three of the seven long cores (D2, D3, and
Tunu2) were analyzed continuously for major ions and used to construct the ice core derived volcanic history and the non-volcanic EXS data set. The remaining four cores were examined specifically for the 1783-84 A.D. Laki and the 1815 A.D. Tambora eruption. The shallow Tunu core (Tunu1, see Table 3.1) was used to replace a missing section in the long Tunu core (Tunu2).
34 Ave Core Elevation Year Bottom Bottom Lat., °N Long., °W Accum. Reference Name (m) Drilled Depth Age, A.D. (mm.w.e.) Site D2 71.75 46.16 2640 1999 132.40 1781 450 M-T2003
Site D3 69.80 44.00 2560 1999 150.73 1740 450 M-T2003
Raven 66.38 46.18 2053 1998 120.96 1717 326 M-T2003
Tunu1 78.02 33.99 2113 1996 20.00 1917 110 this work
Tunu2 78.02 33.99 2113 1996 69.00 1578 110 this work
GITS 77.14 61.09 1887 1996 120.50 1965 365 M-T2001
Humboldt 78.53 56.83 1995 1995 146.50 1153 141 M-T2001
NASA-U 73.84 49.50 2369 1995 151.24 1645 344 M-T2001
Site T 72.58 38.45 3200 1989 200.00 1212 222 M-T1993
Site A 70.38 35.49 3090 1985 101.00 1715 266 M-T1993
Note: M-T=Mosley-Thompson first author All cores were collected by PARCA (project#: NASA NAG-5032 and -6817) except Site T and A, which were collected at Summit by Dr. Ellen Mosley-Thompson with NSF funding (NSF project#: NSF-DPP-8520885).
Table 3.1 Information for the Greenland PARCA and Summit ice cores discussed in the text
35 3.1.2 Non-PARCA cores
Two earlier ice cores, Site T and Site A cores (see Table 3.1), were drilled in the central Greenland region in 1989 and 1985, respectively [Mosley-Thompson, et al., 1993].
The Site A core was drilled at a site 40 km east of the ice sheet crest as part of the site selection activities in preparation for GISP II. The Site T core was drilled at the summit of the GIS close to the GISP II drill site. These two cores were continuously analyzed by ion chromatography (IC) for anions and by Coulter Counter for insoluble dust concentrations and size distributions. However, due to the low accumulation rate in central Greenland the preserved δ18O signal is unreliable for dating because of diffusion during the firnification process. Thus, these cores were dated using seasonal variations in insoluble dust that is well-preserved.
To assess the spatial variability of the deposition of sulfate from the Laki and
Tambora eruptions, data from eleven Greenland ice cores were incorporated (Table 3.2)
[Clausen and Hammer, 1988]. These ice cores, drilled between 1974 and 1985, are widely distributed around the GIS (Fig. 3.1).
36 Core Name Lat., °N Long., °W Elevation (m) Year Drilled
Camp Century 77.18 61.11 1885 1977
North Central 74.62 39.60 2930 1977
Crête 71.12 37.32 3172 1974
Site B 70.63 35.82 3139 1984
Site D 70.65 37.48 3018 1984
Site E 71.76 39.62 3086 1985
Site G 71.15 35.85 3098 1985
Dye 3 65.18 43.83 2480 1979
4B 65.17 43.93 2491 1983
18C 65.03 44.39 2620 1984
Dye 2 66.48 46.33 2100 1977
Table 3.2 Information for earlier ice cores [Data from Clausen and Hammer, 1988]
37 These earlier ice cores were dated by counting the annual oscillations in δ18O
[Dansgaard et al., 1973], coupled with the use of reference horizons containing bomb-produced radioactivity. Detailed H+ concentrations were obtained by the ECM method [Hammer, 1980]. Chemical analyses and pH measurements were made on selected ice samples from annual layers containing Laki and Tambora eruptions [Clausen and Hammer, 1988; Langway and Goto-Azuma, 1988]. When combined with the continuous and highly-resolved chemical analyses for these PARCA cores, these ice core records provide an unprecedented opportunity to examine the spatial variation of the deposition and preservation of sulfate aerosols along with the climatic information associated with these two volcanic eruptions, which produced observable climate perturbations.
3.2 Laboratory analyses and data processing
Ice core analyses require state-of-art laboratories and equipment along with cautious construction of time scales before any further analysis can be performed. The first of the following sections describes the laboratory analyses of ice cores, including measurements of dust concentration, major ions, stable oxygen isotopes, trace elements and examination of volcanic fragments. The discussion of data processing includes construction of ice core time scales, calculation of annual sulfate fluxes and excess sulfate concentrations, volcanic signal extraction, and three approaches for estimating missing values in the ice core data.
38 3.2.1 Laboratory analyses
The analyses of δ18O, dust content, and major ions for the PARCA cores and
Summit cores were conducted by the Byrd Polar Research Center Ice Core Paleoclimate
Research Group at The Ohio State University (OSU). All the samples were cut using a pre-cleaned band saw and were then decontaminated and melted in a Class 100 clean room. Depending upon the physical property of the samples (for example, firn or ice), the decontamination procedure varies. For firn samples, ultra-clean gloves were used all through the sampling process to avoid contamination. Ice samples were cleaned by removing all outer surfaces using Milli-Q water before melting in the Class 100 clean room. Oxygen isotopic ratios were determined with a Finnigan-Mat mass spectrometer.
The dust constitution and size distributions were measured using a Coulter Multisizer.
Samples for major ion concentration analyses were prepared using standard ion chromatography procedures [Cole-Dai et al., 2000], and analyzed using a Dionex DX500 ion chromatograph with a GP40 gradient pump, a CD20 conductivity detector, and AS40 automated sampler.
Examination of volcanic fragments associated with certain events was conducted using scanning electron microscopes. Generally, ice cores from high An regions are selected in order to maximize the possibility of finding tephra. For example, the D3 ice
-1 core, which has an average An of 451 mm w.e. a , was specifically selected for the investigation of ash particles from the Laki eruption. Sections from one year before, during, and two years after the eruption (1782-1785) were examined for tephra. These 16 samples, 4 per year, were cut, melted, and filtered onto 0.2μm Nuclepore membranes in
39 the Class 100 clean room and were examined using Quanta and XL-30 Environmental scanning electron microscopes housed in OSU Campus Electron Optics Facility.
Certain trace elements are known to be associated with volcanic ejecta. To help address the timing of debris arrived from Laki, the D2 ice core was selected because of
-1 its good quality and high accumulation rate (An = 449 mm w.e. a ). Thirty-two samples in the section of 1782-1785 were extracted from the D2 core and prepared in the Class
100 clean room for trace element analyses. Low density polyethylene bottles were cleaned using established ultra-clean procedures [Gaspari et al., 2006]. Decontamination of the ice samples (10x3x2 cm3) was performed by means of a triple washing with ultra pure water by using stainless steel acid cleaned forceps. Procedural blanks were checked with ice samples prepared by freezing ultra pure water in 1 L LDPE bottles and were found to be negligible. The analyses of Cd, Bi, Tl, and Rare Earth Elements (REE) were conducted by Inductively Coupled Plasma Sector Field Mass Spectrometry (Element2,
Thermo) at the University of Venice [Gabrielli et al., 2006]. The concentrations of excess
Cd, Bi, and Tl were calculated by correcting for the terrestrial and marine contributions using the concentrations of Ba and Na, respectively.
3.2.2 Ice core dating
One of the most important but difficult aspects of ice core paleoclimate research is the construction of the age-depth relationship. Principal methods include the use of radioactive isotopic analysis, components with seasonal variations, reference horizons, theoretical models, and stratigraphic correlations. Radioactive isotopes that have been
40 used include 10Be, 14C, 36Cl, 39Ar, 81Kr, and 210Pb [Stauffer, 1989]. However, radiative decay is not a routinely used technique and is usually applied in regions with very low An, where other stratigraphic techniques are limited. On the other hand, certain ice constituents exhibit seasonal variations, such as δ18O and insoluble dust [Thompson et al.,
1975; Thompson, 1977; Hammer et al., 1978; Meese et al., 1995, 1997], are widely used to detect annual layers when Ān is high enough to ensure signal preservation [Johnsen et al., 1972]. Extremely accurate timescales can be obtained by cross-checking with well-known reference horizons, such as the radioactive fallout from nuclear bomb tests in the 1950s and 1960s, and major explosive volcanic events. In this way, ice core time scales can be constrained [Crozaz et al., 1966; Picciotto et al., 1971; Hammer et al.,
1977]. However, at greater depths, these dating techniques become problematic due to the compression of the ice. Thus, theoretically-based ice-flow models are used to calculate the age of the ice [Dansgaard and Johnsen, 1969; Reeh, 1989; Johnsen and Dansgaard,
1992]. For example, in the construction of time scales for the GRIP core, the last 14,500 years are based on counting layers downward from the surface. Beyond 14,500 years, a steady state ice flow model, based on the δ–dependent accumulation rate derived from all available annual layer thickness data, is used [Dansgaard et al., 1993]. In addition to these methods, certain stratigraphic features in ice cores, such as major changes in δ18O, can be correlated with δ18O changes in other proxy records to yield better chronological control [Dansgaard et al., 1982; Jouzel et al., 1987].
For this research project, all the other multi-century PARCA and summit cores, except the Tunu core, have been dated as accurately as possible using multiple seasonally
41 varying indicators (such as δ18O, dust, calcium, and nitrate, etc.) with an estimated dating uncertainty of ± 2 years. For ice cores from high An regions, such as the D2 and D3 cores, seasonal variations in δ18O and dust were well preserved with only modest δ18O smoothing and/or disturbance. In Greenland, δ18O is lowest (most negative) in late winter
(~February) and highest (most enriched or least negative) in mid-to late summer (July to
August) [Bolzan and Strobel, 1994; van der Veen and Bolzan, 1999]. Dust content usually peaks in spring and is lowest in winter [Hammer, 1977]. For regions with median to low
-1 An, such as the summit region (An ≈ 220 mm w.e. a ), dust was used as the primary dating parameter along with nitrate in the pre-1950 strata as the δ18O was smoothed by
-1 diffusion. For low An regions, such as Tunu site (An ≈ 110 mm w.e. a ), ice core dating was more complicated. A set of parameters, including dust, calcium, sodium and chloride, combined with the elevated βeta radioactivity in spring 1953 and winter 1961-1962 from thermo-nuclear bomb tests, and several known volcanic horizons, were used to constrain the time scales.
3.2.3 Calculation of fluxes and excess concentration
Generally, volcanic events are identified by elevated EXS deposited when gaseous compounds from explosive volcanic eruptions are oxidized to sulfuric acid
[Mosley-Thompson et al., 2003]. The EXS is routinely calculated by subtracting the sea-salt sulfate (SSS), estimated using the concentration of either chloride or sodium
2- measured in the same sample, from the total SO4 . The annual flux of EXS is calculated as ∑ cili for each core, where ci = the concentration of EXS in each sample and li i∈{1year}
42 is the length of that sample in w.e. For the D3, D2, and Tunu ice cores, sodium is used to estimate the SSS. SSS was also calculated using Cl- for comparison and no significant difference was found. For the Site A and Site T cores, chloride was used as no sodium data were available.
The concentrations of excess Cd, Bi, and Tl were calculated by correcting for the terrestrial and marine contributions using the concentrations of Ba and Na, respectively
[Nozaki, 1997; Wedepohl, 1995]. The crustal normalized Lu/La ratio, (Lu/La)N, which reflects changes in dust composition, is calculated by (Lusample/Lucrust) / (Lasample/Lacrust).
3.2.4 Extraction of volcanic information
There are multiple methods for extracting volcanic signals. Most involve selecting a specific standard deviation as the detection threshold [Robock and Free, 1995, 1996,
Cole-Dai et al., 2000]. For example, Cole-Dai et al. [2000] used the background concentration plus 2 times standard deviation (2Φ) of the annual EXS flux as the detection threshold for volcanic events. Naveau et al. [2003] developed an automatic extraction method based on a statistical multi-state space model, to provide a more objective estimation of the amplitude and subsequent time evolution of the signals, as well as a measure of confidence through the posterior probability for each event.
However, these methods all require the assumption of a normal distribution, but in reality, the ice core-derived sulfate time series typically follow a skewed distribution. Also, as the sulfate background concentrations have increased since 1850, it is likely that the 2σ
43 criteria will be affected so that less explosive low latitude volcanic eruptions, which may also be recorded in the Greenland ice sheet, will be eliminated in error.
To overcome these problems, we adopted Gao et al.’s method [2006], which applies a high-pass loess filter to the time series, and then examined the peaks that exceed twice the 31-year running median absolute deviation. This method provides robust estimates for a wide range of population distributions and does not require a normal distribution assumption. The step-wise signal extraction procedure is as following: (1) convert the sulfate concentration time series into fluxes (see previous section 3.2.3 for flux calculation); (2) remove the trend and the background variation with a high-pass loess filter, a locally weighted least squares quadratic estimate [Cleveland and Devlin,
1988] to remove signals longer than 31 years; (3) calculate the 31-year running median absolute deviation (MAD) of the residuals and select the potential volcanic peaks if they exceed the baseline plus 2 running MAD; (4) replace the selected peak values with the median of the original time series; (5) repeat step 2 using the time series with the peaks removed; (6) repeat step 3 to extract the volcanic peaks. Here the 31-year filter was selected because it falls at a spectral gap, which removes long-term variations while allowing for the full strength of volcanic peaks.
3.2.5 Estimating missing data
Because of the conflict between the limited ice core volume and the numerous analyses needed, situations of inadequate material or missing data inevitably occur in ice core research. In this research, missing sections were found in the Tunu2 core from the
44 northeastern GIS, and the D2 and D3 ice cores from the western GIS, as shown in Table
3.3. Sections between late 1940s to late 1970s were consumed for βeta radioactivity measurements and a few sections of the D2 core were missing for unknown reasons.
Core Name Tunu2 D3 D2 1881-1897, 1929-1930, Missing years 1942-1977 1951-1967 1947-1970, 1975-1977
Table 3.3 Missing sections in the Tunu, D3, and D2 ice cores
Due to the missing sections, dust and chemistry analyses for Tunu 2, and chemistry analyses for D2 and D3 are incomplete. Fortunately, a shallow core (Tunu1) was drilled nearby and was used in place of the missing section in Tunu2. For the D3 and
D2 cores, based upon the data dependency, different statistical methods were explored to estimate the missing values.
3.2.5.1 Estimate missing data with substitute ice core
For the missing sections of the Tunu2 core, part of Tunu1 chemistry profile is used to construct a continuous and integrated record for the Tunu site. During integration, overlapping sections (upper 7 meters and lower 15-20 meters) of dust and chemistry profiles were compared. Although these two cores are only 50 m apart from each other, the chemistry profiles contain observable differences (Figure 3.2). Thus, we compared
45 the main features in different species, such as Na+, Cl-, Ca2+, EXS, and dust to align these two cores. For example, Fig. 3.2 shows the calcium profile for the overlapping sections for Tunu1 and Tunu2 cores, where the peaks in 15.6 m, 17.4 m, 18.4 m, 18.6 m, 18.8 m, and 19.4 m agree well in both cores. In this way, we substituted the upper section of the
Tunu1 core (0-15.6 m) for the Tunu2 core (0-15.6 m). The length of the substituted sections was adjusted by fitting the 15.6 m section in the Tunu1 core proportionally into the 15.6 m gap in the Tunu2 core.
Figure 3.2 Comparison of Ca profiles in Tunu1 and Tunu2 for overlapping sections
3.2.5.2 Estimating missing data when substitute ice cores are not available
Since there are no nearby ice cores available for substitution into the D2 and D3 cores, we adopted several statistical methods to estimate the missing values. The two
46 imputation strategies selected are time series analysis and multiple imputation using parametric regression method. We also compared the results from these two methods, which may have implications for future research dealing with ice core missing data.
The time series method analyzes the inner dependency of the data and can be used to estimate missing values. The approach takes advantage of the fact that data points taken over time may have an internal structure, such as trend, seasonal variation, and autocorrelation. Here we followed the classical decomposition process and analyzed the possible internal structure in the EXS time series [Brockwell and Davis, 2002]. The data between 1782 and 1850 show a different pattern (slope) when compared to data from
1850 to 1985. Thus, we only used the latter data in the time series analysis. As similar approaches were applied to the D2 and D3 cores, the following discussion focuses only on the analysis of the D3 core.
Before conducting any analyses, the original data were plotted to examine the possible patterns in the time series. For example, the EXS flux has shown an upward trend evident since 1850 in the D3 core (see Fig. 3.3 and Fig. 3.5 A). No apparent seasonal pattern is found. A log transformation was performed on the original data in order to meet the normality assumptions. The examination of the sample autocorrelation function (ACF) on the transformed data also indicates the existence of a trend component, but no seasonal components. Thus, a simple model was fitted to the original data:
Xt = mt + Y (3-1) where mt is the trend component, i.e., a slowly changing function, and Yt is the random noise component.
47 A linear function (mt = 0.012Year -19.072) was fitted to the data using an
Ordinary Least Squares method to estimate the trend component (see Fig.3.3) that was then removed from the original data. The residuals were then examined for the internal relationship. Fig. 3.4 shows the normality plot, the ACF, and the partial autocorrelation function (PACF) plots for the residuals. The PACF is defined to be the conditional correlation of Xt and Xt–k given all the values from time t–k+1 to t–1 (i.e., values of Xt-k+1,
Xt-k+2, …,Xt-1). Like ACF, PACF also vary between-1 and +1 with values near ±1 indicating stronger correlation. Overall, the residuals are reasonably normally distributed with a small tail. Slightly high values in lag 1 and lag 2 fall outside of the 95% bounds in the ACF plot, but only for the first lag in the PACF plot. Hence, an AR(1) model was fitted to the residuals.
Yt = b Yt-1 + εt (3-2) where Yt is the observation in time t, b is the coefficient, and εt is the random noise.
The residuals from the AR(1) model were checked for normality and possible remaining pattern. The results show that the residuals are randomly distributed around zero and reasonably normally distributed. The ACF and PACF plots show that all the values fall within 95% confidence bounds (except lag 0 = 1 in the ACF plot), which suggests the validity and goodness of fit of the model.
48
Figure 3.3 Linear regression in estimating the trend component in the EXS flux in the D3 core
After fitting the time series model to the EXS flux data, it was then used to estimate the missing values. The results (solid line) are shown in Fig 3.5 A with the 95% prediction confidence interval (dashed line). Overall, this technique managed to estimate the average value of EXS flux, but failed to simulate the variability. Although the internal pattern is well explained by the time series model, the confidence interval is rather wide due to the substantial amount of missing data (17 years). Therefore, another approach, multiple imputation, was explored.
49
Figure 3.4 Residual plots for the D3 EXS data after removal of the trend: (A) residuals vs. time; (B) normality plot; (C) ACF plot; (D) PACF plot. The dashed line represents the
95% confidence bound. The unit of lag is in year.
50
Figure 3.5 Comparison of EXS data in D3 using (A) the time series method, the bold line is the predicted value and the dash lines show the 95% prediction interval; (B) the multiple imputation (result of the first imputation).
51 Multiple imputation provides a useful strategy for dealing with missing values by filling a set of plausible values that represent the uncertainty about the true value [Rubin,
1987]. Normally missing values lead to estimates that are more variable than those that would be obtained if the entire sample could be observed. As a result, variance estimates derived under the assumption of a complete data set are not valid in the presence of missing values. Therefore, the imputation strategy and/or the variance estimation approach must take imputation into account to make valid inferences. To properly reflect this uncertainty, multiple imputation attempts to estimate missing data using a set of random samples of the missing values. The process for making inferences under multiple imputation involves three steps: (1) generate m complete data sets by filling in the missing data m times; (2) analyze the m complete data sets using standard procedures; (3) combine the results from the m complete data sets for inference.
As the EXS data were arranged in a monotone missing pattern, we used the parametric regression method, which assumes multivariate normality to estimate the missing values. A monotone missing pattern exists under the following circumstance: when a variable Yj is missing for the individual i, all subsequent variables Yk, k>j, are missing for the individual i. In the regression method, a regression model is fitted for each variable with missing values, with the previous variables as covariate. That is, for the missing EXS data in D3, the model
YD3,t = β0 + β1YTunu,t + β2YSite A,t + β3YSite T, t + εt (3-3) is fitted with the existing observations. This process is repeated sequentially for variables
* * * * with missing values. For each imputation, values of parameters (β0 , β1 , β2 , β3 , and
52 2* σ D3), which are the simulated values for (β0, β1, β2, β3, and εt), respectively, are drawn from their posterior predictive distribution. The missing data values are then replaced by
* * * * * β0 + β1 yTunu + β2 ySite A + β3 ySite T + zi σ D3 (3-4) where yTunu, ySite A, and ySite T are the covariate values of the first three sites and zi is a simulated standard normal deviate.
In order to meet the normality assumption, a log-transformation was performed on all the EXS flux data before imputation. Five imputations were conducted using SAS. To demonstrate the variability simulated by the multiple imputation process, only the result from imputation 1 is shown (Fig. 3.5 B).
Although missing data in ice core paleoclimatology research are not often encountered, it is still inevitable due to the conflict between the limited ice core volume and the numerous analyses often required. To make inferences that can apply to the population, estimation of missing values becomes essential. Although the best way to estimate the missing values might be to know the mechanisms that drive the variability, limited scientific understanding may restrict this approach.
For time series analysis, the inter-correlation in the data, such as trend, seasonal variation, and autocorrelation were examined for further statistical inference. This approach is especially useful and effective when large dependencies exist in the data.
However, due to the short atmospheric life time of the sulfate aerosols (1-3 weeks in the troposphere), the dependency of EXS flux data in the ice cores is relatively modest, which limits the estimation of the missing value. The models fitted to the D2 and D3
EXS flux data are AR(1)’s, which rely largely on the 1 year dependency. This model is
53 exceptionally helpful in estimating a few missing data, such as 1-2 years, but not as useful for the D2 and D3 cores, which have 46 and 17 years of missing data, respectively.
Therefore, the time series approach is more reliable for estimating short term response, but is restricted in the estimation of the data variability.
In contrast, the multiple imputation technique takes into account the uncertainty involved with the presence of missing values. Inference was made based upon the multiple data sets produced by imputation. The imputation mechanism we employed is a regression method, which estimates missing values based upon the dependence between missing and existing EXS data in our cores, taking the posterior predictive distribution of the missing data into consideration. Although significant differences existed in all five cores due to different regional deposition mechanisms, high correlations were still observed for some ice core sites, as indicated by the bold values in Table 3.4 (p-values for all the correlations are <0.001 and unadjusted).
Tunu Site A Site T D3 D2 Tunu 1 0.25 0.41 0.52 0.29 Site A 1 0.35 0.19 0.39 Site T 1 0.44 0.43 D3 1 0.41 D2 1
Table 3.4 Correlation of EXS data in five Greenland ice cores from 1782 to 1850
54 As previously stated, the best way to estimate missing values is to understand the scientific mechanisms that drive the variability. For ice cores, the level of understanding is limited. These statistical methods provide a variety of practical and effective ways to estimate missing values. A plausible approach to test the accuracy of these methods is to use contemporary data from nearby ice cores. In the case of the D2 and D3 cores, the analysis of a new 150 m ice core from Crawford Point in the west central Greenland may have the potential to provide better estimates for the missing data.
3.3 Primary ice core data sets
The first primary data set consists of continuous chemistry, dust, and δ18O records for three PARCA ice cores (D2, D3, Tunu) and two Summit cores (Site A, Site T), which contain multiple century histories of 219, 260, 417, 271, and 259 years, respectively. For the chemistry and dust data, overall 2062 and 3330 chemistry samples for D2 and D3,
573 chemistry samples for Tunu 1, 1619 chemistry samples and 1762 dust samples for
Tunu2 were completed in this research project. The continuous, high-resolution sulfate records from these ice cores reveal the history of both major and minor volcanic events as well as the non-volcanic sulfate background concentrations over the past centuries. The second data set consists of the detailed EXS flux associated with 1783-84 A.D. Laki eruption from 19 ice core sites and those associated with 1815 A.D. Tambora eruption from 17 ice core sites. This data set was used to explore the spatial deposition pattern of the volcanic sulfate aerosols from these two events. The last data set consists of the
55 tephra and the trace element data from 32 ice core samples associated with the Laki eruption, which is essential confirming the timing of the arrival of Laki debris to the GIS.
3.4 Additional data sets
In addition to the primary data sets, two advanced ice core-derived data sets were generated for this research project. Three non ice core-derived data sets are employed in the investigation of climate perturbations resulting from volcanic eruptions and in the assessment of the impact of the NH sulfur emissions on the EXS fluxes over Greenland.
3.4.1 Ice core-derived secondary data sets
From the primary data sets, the first advanced data set is composed of the major and minor volcanic events extracted from the five multi-century Greenland cores, which contain 20, 19, 30, 24, and 32 events in D2, D3, Tunu, Site A, and Site T, respectively.
The second data set is the non-volcanic EXS flux time series, calculated by replacing the volcanic signals with the running median for all the five ice cores, which makes it possible to examine the natural and anthropogenic contributions to the atmosphere sulfate budget without the perturbations from volcanic eruptions.
3.4.2 Non ice core-derived data sets employed
To investigate the ice core-derived history of non-volcanic EXS, as they are affected by the anthropogenic sulfur emissions, the continuous time series of sulfur emissions developed by Stern [2005] were used (in Chapter 6). This data set was
56 produced using the observed data and the ASL database [ASL and Associates, 1997;
Lefohn et al., 1999]. Both global and individual country estimates of sulfur emissions for the period 1850-2000 were provided.
Stern’s emission estimates for the 1990s were based on observed data. Published estimates are available for ≈ 70 countries in Europe, the former Soviet Union, North
America, East and South Asia, Australia, and New Zealand. The primary data source for the period from 1850 to 1990 is the ASL database. For the remaining countries and for missing years, either an econometric emissions frontier model, an environmental Kuznets curve model, or a simple extrapolation, was applied to interpolate or extrapolate the estimates [Stern, 2005].
Based on this emission time series, Europe accounted for 88% of the total emissions except those from shipping in 1850. The locus of emissions then shifted first to North America, followed by Eastern Europe and the Soviet Union. In the late 1950s,
Asia emerged as a substantial emitter and became the largest source in the 1990s [Stern,
2005]. An initial peak in global emissions occurred in 1980 (73.2 Tg) and the maximum emission was attained in 1989 (74.1 Tg), primarily as a result of the rising emissions in
Asia. After 1989, sulfur emissions rapidly declined with punctuation due to the Kuwait oil fires. Emissions from Asia peaked in 1996 and have declined thereafter.
The NH temperature anomaly data used in Chapter 4 for investigation of the climate impact by volcanic eruptions were reconstructed based on tree ring records from the Briffa series [1992a, 1992b, 1995] and Esper et al. [2002], modified by Jones and
Mann [2004, and the reference therein]. For the Briffa series, 201 years (1740-1940) of
57 the data were used in this research project; while for Esper et al. data, 246 years
(1740-1985) were used (see section 4.2).
Temperature records from two Greenland coastal stations, Ilulissat and Nuuk, were extracted from Cappelen et al. [2006] and Vinther et al. [2005] and compiled by Dr.
Jason Box. This data set contains quarterly and annual temperature records from 1785 to
2005 with substantial missing data in the early years. Overall, the records from 1813 to
1985 at Ilulissat (continuous from 1840), and from 1785 to 1985 at Nuuk (continuous from 1866) were used to examine the climatic perturbations in Greenland contemporaneous with volcanic eruptions (see section 4.2).
58 CHAPTER 4
VOLCANIC AEROSOL HISTORY
Volcanic eruptions are an important natural driver of climatic variations. A large explosive volcanic eruption in the tropics can produce global scale climate impacts for up to 2-3 years; while the impacts of those from high latitude are usually limited to the hemispheric scale. Knowledge of the climatic response to a specific volcanic eruption helps provide a better understanding of the radiative and dynamic processes arising from volcanic forcing, but also contributes to improved climate models. As records of direct measurements of volcanic emissions are short, a longer, detailed volcanic index can help elucidate the regional and/or global scale climatic perturbations induced by volcanic eruptions. The ice core-derived volcanic index is a promising source of information on past volcanic aerosols as the ice contains physical evidence of atmospheric loading. The
EXS records in our three high resolution PARCA and two cores from central Greenland reveal the detailed history of volcanic eruptions in the high northern latitudes as well as for a number of large eruptions in the tropics over the past 400 years. This ice core-derived volcanic index provides an important supplement to the volcanic history previously derived from other ice cores, most containing longer histories but with lower temporal resolution. Additionally, this volcanic index provides an improved estimate of
59 the stratospheric sulfate burden and serves as an important input for models assessing climatic impacts of volcanic eruptions.
To attain a detailed volcanic history from ice cores, both thorough laboratory analyses and a robust volcanic signal extraction method are needed. Here we applied a high pass loess filter and examined EXS signals exceeding twice the 31-year running median absolute deviation [Gao et al., 2006]. We used this method as it allows the less explosive and/or lower latitude volcanic events to be isolated. Today, the sulfate budget in the Northern Hemisphere atmosphere is strongly dominated by anthropogenic sulfur emissions, clearly evident in the EXS histories from our five Greenland ice cores. The increasing background concentrations, that began in ~1850, obscure all but large volcanic signals in the ice cores. Gao et al.’s method [2006] effectively removes the trend and background variation and allows the smaller volcanic sulfate signals to be extracted.
Unfortunately by the mid 1950s, the mean EXS background concentrations are so high that it is likely some less explosive high latitude and/or tropical volcanic events are masked. For example, the 1991 eruption of Pinatubo did not leave a widespread easy to identify sulfate horizon over Greenland [Cole-Dai et al., 1999]
4.1 An ice core-derived volcanic aerosol history
The results from the extraction of the volcanic signals (see Chapter 3, section
3.2.4) are shown in Table 4.1 for the five ice cores. The largest signal in this ~250 year record is from the 1783-84 A.D. Laki eruption (Iceland), which produced more than ~200
Mt of H2SO4 aerosols in the atmosphere [Thordarson and Self, 2003]. For most of the
60 cores, the second largest signal is the 1816 EXS horizon from the 1815 A.D. eruption of
Tambora (Indonesia). In both Polar Regions, the EXS peaks associated with Tambora arrive in 2 sequential annual pulses, 1816 and 1817 A.D. The delay is a result of the height of injection into the tropical stratosphere and their long-distance transport poleward. These two known horizons have been used for decades to constrain the time scales for Greenland ice cores [Hammer et al., 1981; Clausen and Hammer, 1988; Laj et al., 1990].
Specific volcanic events were cross-checked among the five ice cores (Table 4.1) as potential dating errors might be expected. Most of the potential dating errors were identified in the ice cores from lower An regions. For example, the eruption Katmai
(Alaska, US) in 1912 produced modest EXS signals in all the five Greenland ice cores, with 3 dated as 1912 and 2 dated as 1913. As the 1913 signals occur in the Tunu and Site
A core, both from low An regions, it is very likely that this one year offset is due to dating errors although reworking of the near-surface snow can not be excluded. Similar situation arises for the 1809 A.D. Unknown eruption. EXS from the ‘Unknown’ eruption arrived over the GIS in the 1810 A.D. snow fall [Cole-Dai et al., 2000], producing EXS concentrations comparable to those deposited in 1816 A.D from Tambora
[Mosley-Thompson et al., 2003]. However, the timing of ‘Unknown’ in Site A is 1811, again suggesting a 1 year dating error. In the case of the 1831-1833 volcanic signals in all the five ice cores, it is more difficult to determine which cores likely contain the dating error. The low An cores (Tunu, Site A, and Site T) contain EXS signals in 1831-1832, but the high An cores (D2 and D3) show the elevated EXS in 1832-1833. Although the dating
61 for D2 and D3 cores is presumably more accurate, the contemporary volcanic records from VEI only show 3 candidate events during 1830-1831. These events only have VEI of 3, suggesting that they are unlikely to have produced a long-lived EXS signal extending to 1833.
It is possible that other non-volcanic anthropogenic and biogenic sulfate sources exist and obscure the volcanic signals. Ice cores may be differentially affected by glaciological noise as well as their distance and altitude relative to the sources. To avoid overinterpretation of our volcanic eruption index, we selected the common events with
EXS peaks appearing in at least 3 sites to construct an index of the most climatically influential volcanic eruptions from 1740 to 1985. Table 4.2 shows the eruption information for the possible candidate volcanic events remaining after applying this criterion. The IVI associated with the selected eruptions is calculated as described below for Laki. As Laki produced sulfate aerosols in 1783 and 1784, the total flux is calculated by summing up the volcanic EXS in these two years for each ice core. The IVI is then obtained by averaging over the 5 ice cores. The IVI was calculated this way to both take into account the spatial variability and the duration of the impact.
62 Tunu Site A Site T D3 D2 Year (1627)a (1715) (1731) (1740) (1782) 1632 5.57 NAb NA NA NA 1636 10.69 NA NA NA NA 1638 3.40 NA NA NA NA 1645 4.09 NA NA NA NA 1655 5.51 NA NA NA NA 1665 4.50 NA NA NA NA 1669 6.49 NA NA NA NA 1675 5.57 NA NA NA NA 1693 12.10 NA NA NA NA 1694 5.91 NA NA NA NA 1713 7.71 NA NA NA NA 1730 -c 9.03 NA NA NA 1731 - 22.95 - NA NA 1732 - - 11.15 NA NA 1736 - - 3.35 NA NA 1739 - - 6.44 NA NA 1740 - 8.66 16.97 37.78 NA 1741 - - 7.60 - NA 1743 16.07 - - - NA 1746 4.03 - - - NA 1750 9.73 - - - NA 1751 6.10 - - - NA 1755 - 12.31 3.71 17.89 NA 1756 - 8.45 - - NA 1761 - 9.15 - - NA 1762 - - 5.53 - NA 1766 - - 15.94 15.85 NA 1768 5.76 - - - NA 1779 4.91 - - - NA 1781 - - 15.09 NA 1783 81.67 171.80 116.18 300.91 236.50 1784 8.24 - - 11.07 3.73 1786 - - 3.81 9.39 4.60 1788 - - - 36.35 - 1789 - - - - 3.83 1794 5.97 - - - 4.52 1797 - 6.49 - - -
Continued
Table 4.1 Annual volcanic EXS flux (kg/km2) for 5 Greenland ice cores
63 Table 4.1 Continued
1798 - 3.81 - - - 1809 6.82*d - - - - 1810 12.38* - 25.26 40.37 37.21 1811 - 28.90 9.80 - - 1814 4.94 - - - - 1816 6.50 28.29 19.50 79.16 30.14 1817 - - 20.32 13.04 25.56 1822 - - 9.59 - - 1823 - - - 9.10 24.74 1831 11.17 28.81 14.78 - - 1832 - - 9.21 42.06* 30.16* 1833 - - - 9.57* 12.15* 1835 9.39 11.44 7.71 - - 1836 8.01 - 9.18 22.29* 21.28* 1837 - - - - 8.32* 1854 - - - - 8.04 1856 - 4.90 - - 11.70 1858 13.49 - - - - 1862 - - 5.86 - - 1863 - - - - 15.57 1864 - - - 15.28 - 1872 - - 5.37 - - 1877 - 10.82 - - - 1884 - 16.98 6.89 15.75 - 1885 - - 6.18 14.70 - 1887 16.34 - - - - 1902 - - - - 26.55 1905 - - 6.93 - - 1906 11.59* - - - - 1907 28.55* 13.30* - 13.85 18.02 1908 - - - 20.05 21.52 1911 - - 13.32 - - 1912 - - 14.17 31.41 19.67 1913 15.41* 8.25* - - 30.82 1917 - - - - 19.92 1920 13.56 - - - - 1924 14.26 - 13.23 -
Continued
64 Table 4.1 Continued
1925 11.99 - - - 17.61 1926 27.60 - - - 1928 - - 7.32 - 1931 - 10.10 - - 1933 9.23 - - - 1939 - - - - 16.50 1943 - - 7.77 - 1945 - - - 22.30 21.00 1952 - 22.23 6.59 - - 1956 26.17 - - - - 1964 26.99 11.29 - - - 1965 - 10.43 - - - 1970 - 16.39 - 17.54 - 1972 - - 15.65 - - 1974 - - 7.84 - - 1978 - - 8.42 - - 1982 - - 28.01 - - 1983 - - 20.53 - - 1984 19.90 - 14.48 - - anumber in the parentheses shows the bottom age of the core; bNA means the core does not extend back this far; c- means that the value is below the baseline established using Gao et al.’s approach [2006]. d*indicates possible dating errors due to lack of EXS contemporaneously among these cores
65 EXS Candidate volcano Eruption IVIb number Location VEIa years events year (kg/km2) of cores 1740 Shikotsu (Tarumai) Japan 1739 5 23.67 3 1755 Katla Iceland 1755 5 14.12 3 1783-84 Laki Iceland 1783-84 4 186.02 5 Pavlof Sister Alaska 1786 3+ 1786 5.93 3 Amukta Aleutian Island 1786 3+ 1810 Unknown 1809 30.78 5 1816-17 Tambora Indonesia 1815 7 44.50 5 Nyey Iceland 1830 3 1831-32 Shishaldin Aleutian Island 1830 3 31.58 5 St. Helen Washington, US 1831 3 1835-36 Cosiguina Nicaragua 1835 5 17.86 5 Krakatau Indonesia 1883 6 1884 20.17 3 Augustine Alaska, US 1883 4 1907 Ksudach Russia 1907 5 28.82 4 1912-13 Katmai Alaska, US 1912 6 23.95 5 Agung Indonesia 1963 4 1964 24.36 2c Sheveluch Russia 1964 4 aVEI values are from Newhall and Self [1982]. bIVI value is the calculated average EXS flux, as specified in the text. cThere are missing data in both D3 and D2 in this period.
Table 4.2 Common volcanic events from 1740 to 1985 in five Greenland ice cores.
66 Comparison of the VEI [Newhall and Self, 1982] and the IVI reveals that these two indices are not well correlated. Based on the VEI, the Tambora eruption is the most explosive eruption of this ~250-year period. However, its associated EXS flux as derived from Greenland ice cores is much less than that of Laki. This is not unexpected due to the long-distance transport of Tambora aerosols via the mid to upper stratosphere; while the main transport pathway for Laki EXS is probably through the upper troposphere/lower stratosphere as Laki (Iceland) is very close to Greenland.
Mitchell’s indexb Sato’s indexd IVIe Eruptions DVIa VEIc Severity Class Optical Depths Ave. EXS flux (kg/km2)
1783-84 Laki 2300 - 4 - 186.02
1815 Tambora 3000 - 7 - 44.50
1883 Krakatau 1000 1 6 0.081f 20.17f 1883 Augustine 70 3 4
1912 Katmai 500 2 6 0.032 23.95
1963 Agung 800 1-1/2 4 0.053 24.36 aDVI is from Lamb 1970; bFrom Mitchell 1970; cFrom Newhall and Self, 1982; dFrom Sato et al., 1993; eFrom this work, number of events recorded are the average over the five PARCA cores; fThese values are for the year 1883.
Table 4.3 Comparison of the different volcanic indices for selected eruptions
67
4.2 Temperature anomalies associated with volcanic eruptions
Numerous proxy records and meteorological data provide evidence that volcanic eruptions affect climate on scales ranging from local to global. For example, based upon the density of latewood in tree rings, a large negative temperature anomaly was observed in 1783 contemporaneous with the eruption of Laki [Briffa et al., 1988]. Generally, density variations in latewood are particularly valuable as they provide a useful indicator of climatic perturbations, such as temperature changes [Schweingruber et al., 1993]. To further evaluate our ice-core-derived volcanic index, we compared it to several temperature history data sets. Jones and Mann [2004] reconstructed the NH temperature anomalies using the Briffa tree ring series [Briffa et al., 1992a, 1992b, 1995] and the series by Esper et al. [2002]. Inspection of Fig. 4.1 reveals that most of our IVI events are associated with negative temperature anomalies. For example, the 1783 temperature anomaly, contemporaneous with the Laki eruption, reached -0.76 ºC. The amplitude of temperature anomalies is also related with the frequency of volcanic eruptions. For example, the large persistent cooling from 1809 to 1819 is associated with the closely spaced eruptions of Unknown in 1809 and Tambora in 1815.
68
Figure 4.1 Northern Hemisphere temperature anomalies, the IVI, and the VEI.
Temperature anomalies are reconstructed from Briffa et al. [1992a, 1992b, 1995] and
Esper et al. [2002], and modified by Jones and Mann [2004]. The VEI is constructed by
Newhall and Self [1982].
69
Figure 4.2 The annual average temperature for Ilulissat and Nuuk, Greenland, the IVI, and the VEI. The temperature data were compiled by Dr. Jason Box from Vinther et al.
[2005] and Cappelen et al. [2006]. The VEI was constructed by Newhall and Self [1982].
70 Large temperature perturbations following volcanic eruptions are also recorded in the meteorological data although long temperature histories for Greenland are scanty.
Two of the longest records are from Ilulissat and Nuuk in western Greenland (see Fig.2.3 for locations) [Vinther et al., 2005; Cappelen et al., 2006]. Although there are missing data before the 1860s, the two lowest annual temperatures observed in Nuuk occurred in
1811 (-4.96 ºC) and 1818 (-4.98 ºC), as shown in Fig. 4.2. The cooler temperatures at these two coastal towns were associated with the eruptions of Unknown and Tambora, although with a lag of several years.
4.3 Potential problems with ice core-derived volcanic indices
Compared with the earlier indices (DVI, Mitchell index, VEI, and Sato index), the
PARCA IVI yields a longer record and provides direct estimates of the associated sulfate aerosol burden. However, problems with the use of ice core-derived time series remain.
For example, other sources of acids, bases, and sulfate may produce noise and complicate the identification of a volcanic event. Low annual snow accumulation coupled with surface mixing processes due to blowing snow can decrease the reliability of the annually resolved ice core history. Furthermore, the transportation and deposition of sulfate from the stratosphere to the deposition site (the ice sheet surface) is not well understood.
Therefore, additional analyses and comparisons are essential to acquire a more accurate assessment of past volcanic eruptions.
For all the volcanic indices, the ability to capture the most climatologically significant volcanic eruptions and to quantify the relationship between the volcanic index
71 and magnitude of the climatic impact of the eruption are important. Eruptions in the mid to high southern latitudes will not be recorded by volcanic indices developed from
Greenland ice cores. Thus, combination of various indices may yield a composite global picture of past volcanism. Moreover, exploring other proxy-based indices, such as the footprint of volcanic eruptions in tree rings and other geological records, may prove fruitful to future research efforts to extract long and robust volcanic histories.
4.4 Timing of the arrival of Laki ash and sulfate aerosols to the GIS
The newly produced EXS histories from the D2 and D3 ice cores offer an excellent opportunity to finally confirm the year in which the sulfate from Laki arrived over Greenland. This is important as Laki provides the most commonly used time stratigraphic horizon for constructing Greenland ice core time scales over the last two centuries. Fiacco et al. [1994] reported that ash particles from Laki arrived over the GIS in 1783, while the associated sulfate pulse arrived in 1784 A.D. Their evidence consisted of a few glass shards found just below the layer with elevated sulfate concentrations in
-1 the GISP2 core from central Greenland (An = ~220 mm w.e. a ). Their dating of the core section containing this event was based solely on three Cl-/Na+ peaks interpreted as annual (1783-1785). Their inferred 1784 snowfall contained an almost negligible increase in dust and lacked the characteristic summer δ18O peak. Although their evidence for the
1784 annual layer was ambiguous, their conclusion has been used to assess the
1783-1784 atmospheric sulfate loading, with ~12.5% of the total H2SO4 aerosols
72 estimated to have been retained in the lower stratosphere for more than a year
[Thordarson and Self, 2003]. A recent model simulation of the sulfate aerosol distribution by Oman et al. [2006] suggested that the majority of Laki sulfate arrived in 1783, not in
1784. Oman et al. called for re-examination of the Fiacco et al. data and/or data from other ice cores to resolve the potentially spurious Cl-/Na+ peak (1784) that they suggested may reflect the emission of volcanic Cl-. Our two ice cores are from high accumulation regions and have been continuously analyzed and dated using multiple seasonally varying constituents [Mosley-Thompson et al., 2001], and thereby provide an optimal data set to explore the arrival of Laki emissions over Greenland. As surface and post-depositional processes are likely to distort preserved seasonal indicators (e.g., stable isotopic ratios and H2O2) in regions of low accumulation [Johnsen, 1977; Neftel, 1996;
Wolff, 1996], and/or remove all or part of a year’s accumulation, dating errors and uncertainty are often unavoidable. This is possibly the case of the Fiacco et al.’s missing
1784 δ18O peak in GISP2. Thus, we also incorporated three other ice cores, Raven,
NASA-U, and Site T, together with the D2 and D3 cores, to investigate of the behavior of
18 δ O with the changing of An level, which may provide a reasonable explanation for the supposedly missing 1784 δ18O peak in GISP2.
Although the Laki eruption lasted 8 months (June 8, 1783 to Feb. 7, 1784), ~96% of the SO2 was released during the first 5 months [Thordarson and Self, 2003], producing substantial societal and environmental impacts over Europe, North America, and the
Arctic. Located on the west-central side of Greenland, the D2 and D3 sites receive annual accumulation twice that in the Summit region (GRIP, GISP2 cores), which greatly
73 enhances the likelihood of complete preservation of the seasonally varying constituents.
The timing of the arrival of Laki emissions was independently established using multiple seasonally varying parameters as illustrated for sections of cores D2 and D3 from 1782 to
1820 A.D (Fig. 4.3). The dating was based primarily on seasonal variations in δ18O and
- dust content. NO3 , another seasonally varying constituent (not shown), was also used to assist with the dating. Generally, δ18O is lowest (most negative) in late winter (~February) and highest (most enriched or less negative) in mid to late summer (July to August)
[Bolzan and Strobel, 1994; van der Veen and Bolzan, 1999]. Dust content usually peaks in spring and is lowest in winter [Hammer, 1977]. Both parameters were well preserved in
D2 and D3 with little smoothing and disturbance. The top of this section is tightly constrained by the well dated 1816 EXS horizon from the 1815 eruption of Tambora
[Clausen and Hammer, 1988; Cole-Dai et al., 2000]. In both Polar Regions, the EXS peaks associated with Tambora arrived in 2 sequential annual pulses, 1816 and 1817 A.D., due to the height of injection into the tropical stratosphere and their long-distance transport poleward. Another large explosive tropical eruption, called the Unknown (1809) event, also provides an excellent time constraint. EXS from ‘Unknown’ arrived in both
Antarctica and the Arctic in the 1810 A.D. snow fall [Cole-Dai et al., 2000], producing
EXS concentrations comparable to those deposited in 1816 A.D from Tambora
[Mosley-Thompson et al., 2003]. By using these seasonal indicators and reference horizons, Laki sulfate is determined to have arrived over Greenland late in 1783.
74
Figure 4.3 Seasonal variations in δ18O and insoluble dust used to date the entire core are shown along with the excess sulfate (EXS) for the time interval 1782-1819 from the (a)
D2 and (b) D3 ice cores.
75 In the regions with low accumulation rate, such as northeastern Greenland, where
-1 An is usually 100-200 mm w.e. a or even less, determination of arrival time of Laki
18 sulfate and ash particles by δ O becomes problematic. As An decreases, the probability of distortion and occurrence of missing years increases. Smoothing in δ18O can be clearly seen. Fig. 4.4 shows the oxygen isotope for 5 sites, where ice cores were displayed in the order of decreasing An from left to right. As the An decreases, the annual signal smoothes
-1 and missing years become apparent. For regions with An greater than 400 mm w.e. a , the seasonal variation of δ18O is well defined with winter to summer differences around 10‰.
-1 18 For regions with An between 300 - 400 mm w.e. a , δ O becomes smoother, but annual cycles are still recognizable, However, in the Raven core, the seasonal signal has been smoothed in 1781-1782. In the Site T core, only three annual δ18O cycles are present in this five-year period. The An of Site T is the same as for the GISP2 core (~ 220 mm w.e. a-1). Thus, dating of the Raven and Site T cores was primarily based on the annual dust
-1 variations. For the Tunu ice core, where An is ~110 mm w.e. a (only ¼ that at D2), not
18 - only is δ O smoothed but seasonal variations in NO3 are indistinct, annual dust variations are ambiguous, and missing seasonal features in sea salt (Na+ and Cl-) are also observed. Because of the uncertainty involved, the timing of Laki EXS in the TUNU ice core could not be determined independently and was treated as reference horizon and set as 1783. As Laki is routinely used as a reference horizon in ice cores from low An region, resolving the year in which EXS arrives in Greenland snow is important to so that these ice core histories can be used in paleoclimatic investigations.
76
Figure 4.4 δ18O in five Greenland ice cores for the 5-year interval containing the Laki eruption.
In addition to the arrival of the EXS (Fig. 4.3), the timing of tephra deposition in
Greenland relative to the sulfate peak has implications for atmospheric transport processes. In D3, tephra fragments were found only on 2 filters associated with the layer coincident with the bulk (~97 to 98%) of the associated Laki sulfate peak, corresponding to the latter half year of 1783 A.D. (Fig. 4.5; Table 4.4). No fragments with distinctive volcanic features were observed on the remaining 14 filtered samples. Note that both D2 and D3 contain a prominent δ18O summer peak (Figs. 4.3, 4.4) inconsistent with the assumption of strong summer cooling invoked by Fiacco et al. [1994] to explain the missing δ18O summer peak in their core. In D2 and D3, there is a distinct two-sample peak in Cl- (Fig. 4.6d) that creates a spurious Cl-/Na+ signal late in 1783 (Fig. 4.6e). In general, volcanically derived Cl- is quickly scavenged by wet deposition before reaching
77 the stratosphere [Tabazadeh and Turco, 1993]. However, due to their close proximity it is quite likely that some of the Cl- emitted from Laki did arrive over the GIS creating the
Cl-/Na+ peak erroneously identified as annual by Fiacco et al., [1994, their Fig. 4]. In fact, as shown in Fig. 4.6, Cl-/Na+ is not a reliable indicator of seasonality and should be avoided as a dating tool for Greenland ice cores.
The Laki eruption also contributed additional trace elements to the atmosphere.
Hong et al. [1996] reported that excess Cd, Cu, and Zn were detected in the Laki fallout
-1 layer in the Greenland Summit core (An = ~230 mm w.e. a ) and that their deposition was constrained to a brief period. Their maximum Cd concentration was 12.6 ppt, coincident with the core section with a high electrical conductivity (an indicator of acidity). In our D2 core, the first peak of excess Cd (10.4 ppt) is coincident with the 1783 dust peak, and excess Cd reaches its maximum (80 ppt) near the end of 1783 (Fig. 4.6).
Other important volcanic tracers, such as excess Bi and Tl [Matsumoto and Hinkley,
2001], peaked at 9.3 ppt and 4.6 ppt, respectively, near the end of 1783 and quickly returned to their background values in 1784. Bi is derived mainly from volcanic emissions to the atmosphere and its average background concentration in central
Greenland snow is 0.15 ppt [Ferrari et al., 2000], consistent with the Bi background concentration in D2 (0.03 to 0.34 ppt in 1782, 1784-1785). The background concentration of Tl in D2 is 0.01 to 0.13 ppt (also in 1782, 1784-1785), consistent with the Tl background (0.03 to 1.3 ppt) from Devon Island (Arctic) ice samples [Krachler et al.,
2005]. Generally, Cd, Bi, and Tl have low mean element weight ash fractions (0.1-3%)
[Aiuppa et al., 2003], indicating that these trace metals were transported to the GIS
78 primarily by aerosols. In contrast, REE are refractory and usually associated with solid particles in precipitation [Freydier et al., 1998], consistent with the behavior of the crustal normalized Lu/La ratio that reflects a change in dust composition. A brief peak of
(Lu/La)N, calculated as (Lu/La)N = (Lusample/Lucrust) / (Lasample/Lacrust) [Wedepohl, 1995], is observed late in 1783, in close correspondence with other elevated trace elements (Fig.
4.6b). Thus, the contemporaneity of tephra and the rapid change of the (Lu/La)N ratio strongly suggest that the Laki ash arrived over the GIS late in 1783, along with the elevated concentrations of Cd, Bi, Tl, and sulfate aerosols. These ice core-derived observations agree well with the model simulation by Oman et al. [2006] and confirm the rapid transport of volcanic emissions to Greenland.
79
Figure 4.5 SEM photographs document the Laki tephra found in the D3 core (shaded boxes on the time axis). Photographs E-1 to E-3 and F-1 to F-3 are from filters E and F
(in fine shading), respectively. These two filters with tephra contain snow deposited in the second half of 1783, coincident with the EXS peak (also shown). No volcanic fragments were found on the remaining 14 filters (see sampling discussion in section 3.2.1).
80
Figure 4.6 Section of Core D2 (1782 to 1785 A.D.) containing the Laki eruption. (a) calculated concentrations of excess sulfate (EXS); (b) calculated concentrations of excess
Tl (green), Bi (red), and Cd (blue); (c) normalized Lu/La ratio; (d) measured concentration of Cl-; (e) calculated ratio of Cl- to Na+; (f, g) seasonal variations in the insoluble dust concentrations and δ18O, respectively, used to construct the time scale. The shading highlights 1783 (darker) and 1784 (lighter).
81 The decay time of volcanogenic sulfate signals in ice cores may provide information regarding the injection height of SO2 gas by the eruption and the residence time of sulfate aerosols in the upper troposphere/lower stratosphere. The EXS associated with Laki in both 1783 and 1784 was calculated along with the background concentration of EXS for six Greenland ice cores (Table 4.4). The percent of the total EXS flux (1783 and 1784) from Laki arriving over Greenland in 1784 ranges from near zero to 6%, consistent with the portion of SO2 estimated to have been produced by Laki during the last 3 months of the eruption (~4% from Nov. 1783 to Feb. 1784). Thordarson and Self
[2003] estimated that 200 Mt of sulfate aerosols were produced by Laki and that ~25 Mt
(12.5%) stayed aloft near the tropopause for >1 year. However, the relatively short atmospheric lifetime of the sulfate aerosols calculated from these 6 ice cores (Table 4.4) indicates that a much smaller portion of Laki aerosols remained in the lower stratosphere in 1784, suggesting both a relatively low injection height restricted to the lower stratosphere, and a short residence time (< 6 months) for most of the emissions from Laki.
(The results presented in this section have been accepted for publication [Wei et al.,
2008]).
82 a c Cores An Bkg. EXS Annual EXS flux Laki EXS flux EXS flux (mm w.e. a-1) fluxb (kg/km2) (kg/km2) (%) (kg/km2) (1781-1985) 1783 1784 1782 1783 1784 1783 1784 1784 D3 451 525 323 18.9 11.5 314.4 24.6 295.5 5.7 1.9 D2 449 596 442 12.7 10.3 249.1 16.2 236.4 3.5 1.5 NASA-U 344 208 312 11.0 11.0 167.0 20.6 156.0 9.6 5.8 Raven 316 304 304 9.5 7.6 154.7 14.7 125.2 5.2 4.0 Site A 228 244 161 4.8 7.3 176.0 3.1 171.2 -- -- Site T 221 261 260 7.5 7.7 123.0 9.2 115.5 1.6 1.4 a Locations are shown in Fig. 1 of Mosley-Thompson et al., 2001. b Background EXS flux is the average from 1778 to 1789, excluding 1783 and 1784. c Percent of EXS flux in 1784, relative to the total flux for 1783 and 1784.
Table 4.4 Estimated background EXS and Laki eruption EXS in six Greenland ice cores
83 CHAPTER 5
SPATIAL VARIABILITY OF EXCESS SULFATE DEPOSITION OVER
GREENLAND FROM THE LAKI AND TAMBORA ERUPTIONS
Explosive volcanic eruptions can inject a large quantity of sulfur dioxide into the
- troposphere and stratosphere. The sulfur dioxide reacts with OH and H2O to form sulfate
(H2SO4) on a timescale of weeks to months. Typically, tropospheric aerosols have atmospheric lifetimes of 1-3 weeks, while aerosols in the stratosphere are rapidly distributed around the globe and may remain there for up to 2-3 years [Robock, 2000].
Generally, large volcanic eruptions in lower latitudes, between roughly 0º - 20º in both hemispheres, are most likely to have a global climate impact as the associated sulfate aerosols are quickly distributed to the stratosphere of both hemispheres where they may remain for several years. Aerosols from high-latitude volcanic eruptions in the
NH are mostly confined north of 30ºN [Oman et al., 2006]. Thereby, the GIS contains the volcanic history of both large explosive eruptions in tropical areas as well as those in the higher latitudes of the NH. The concentrations of volcanically derived sulfate aerosols preserved in Greenland ice cores vary significantly over the ice sheet. These spatial differences result from factors such as the amount of volcanic ejecta, the injection height, the season of the eruption, the location of the volcano, the transport time, and the trajectory (pathway). On the ice sheet regional differences in sulfate deposition arise from
84 differing deposition and accumulation mechanisms (i.e., wet vs. dry) and the large range in annual mass accumulation rates.
The spatial characteristics of the sulfate aerosol deposition associated with specific eruptions may provide information about the transport processes and possible transport pathways. Comparing the sulfate deposited in different regions from the same eruptions may help to identify the primary deposition mechanisms. To explore this, we examined the spatial character of the EXS from two quite different eruptions, the 1783-84
A.D. Laki fissure eruption (Iceland: 64º N, 17º W) and the catastrophic eruption in 1815
A.D. of Tambora (Indonesia, 8 ºS, 118 ºE), in numerous spatially distributed ice cores.
5.1 Review of two volcanic events
The Laki and Tambora eruptions both produced strong EXS signals in the GIS although they differ in their geographic locations, quantity of sulfate aerosols injected to the atmosphere, aerosol transport time and pathways. The following sections provide a summary of these two eruptions as well as a brief discussion of their climate impacts.
5.1.1 The 1783-84 A.D. Laki eruption (Iceland)
The Laki eruption was a basaltic fissure type eruption that lasted from June 8,
1783 to February 7, 1784. It was estimated that ~15 km3 magma and a total of 122 megatons (Mt) of SO2 were released into the atmosphere, producing ~200 Mt H2SO4 aerosols [Thordarson and Self, 2003]. Although it was a multiple phase eruption, ~96% of the sulfur gases were released from the eruption episodes in the first five months. In
85 Europe and North America, the summer of 1783 was characterized by extreme and unusual weather, and was followed by a severe winter. A persistent dry haze appeared over Europe during the second half of 1783, known as “the great dry fog” [Stothers,
1996b]. It was reported that the winter mean temperature over Europe and eastern United
States decreased about -3ºC in 1783-1784 following the Laki eruptions and then exhibited a gradual recovery over the next 4 years [Thordarson and Self, 2003]. Almost a quarter of population in Iceland died from suffocation, famines, and injury from acid rain as a result of the Laki eruption [Finnsson, 1796]. In addition, increased death rates were reported across Europe [Grattan et al., 2005].
Using the NASA Goddard Institute for Space Studies modelE climate model coupled to a sulfur cycle chemistry model, Oman et al. [2006] simulated the atmospheric transformation and transport of the sulfate aerosols from Laki. They estimated that the top atmosphere net radiative forcing by the Laki eruption peaked at -27 W/m2 over the high latitudes and produced a global mean forcing of -4 W/m2 during the late summer in
1783 A.D.
5.1.2 The 1815 A.D. Tambora eruption (Indonesia)
The 1815 eruption of Tambora (Sumbawa Island, Indonesia) is the largest historically recorded eruption. The eruption injected ~60 Mt SO2 into the stratosphere, six times that injected by the 1991 Pinatubo eruption [Oppenheimer, 2003]. Following 3 years of rumbling, the eruption started on April 5, 1815, and intensified on April 10
[Stothers, 1984]. Although the eruptions finally ceased on July 15, “smoke” emissions
86 were still observed as late as August 23. The year 1816 was referred to as "Year Without a Summer" in Europe and North America [Stothers, 1984]. The Tambora eruption was also likely responsible for the phrase “Eighteen hundred and froze to death”. Famine was widespread because of crop failures. Over 71,000 people died during, or in the aftermath on Sumbawa and the neighboring islands in Indonesia.
Using Greenland ice core data, Clausen and Hammer [1988] estimated that ~220
Mt H2SO4 were injected to the atmosphere by the Tambora eruption. An estimated 60 Mt of sulfur was injected into the stratosphere, resulting in a global surface cooling of ~1 ºC in 1816. In both Polar Regions, the EXS peaks associated with Tambora arrive in 2 sequential annual pulses, 1816 and 1817 A.D. This is due to the height of injection into the tropical stratosphere and their long-distance transport poleward [Cole-Dai et al., 2000]. A detailed summary and discussion of the estimates of sulfur yield by the 1815 Tambora eruption can be found in Oppenheimer [2003] and references therein.
5.2 Spatial variation of EXS
Generally, the factors that may affect the deposited EXS associated with specific eruptions can be divided into three categories. The first category includes aspects associated with the nature of the eruption, such as the amount and type of ejecta, the injection height, and the location and season of eruptions. The second category includes the transportation process, such as primary pathways of the aerosols and transport time.
The last category consists of factors that affect the deposition of EXS at the different ice
87 core sites, including primary regional deposition mechanisms, the local annual mass accumulation rate, snow drifting and mixing, etc.
The availability of numerous GIS ice cores provides an excellent opportunity to assess both the spatial variability of EXS deposition from a single eruption and the differences in the depositional patterns of the EXS flux arising from different eruptions.
The data also offer the potential to learn more about the different depositional mechanisms. The EXS fluxes from two eruptions, Laki and Tambora, preserved in 20
GIS ice cores form the basis of the data set. As these ice cores are unevenly distributed around Greenland, kriging techniques [Matheron, 1963; Cressie, 1993a] were applied to estimate their spatial correlation and predict EXS fluxes for other locations that lack ice core data. Here the variables of interest include the volcanic EXS flux, the EXS density, the non-volcanic EXS flux (NV-EXS), and the annual net mass accumulation. The next section describes the raw data used to construct these variables, the kriging parameters, and the spatial patterns of the EXS flux deposited by Laki and Tambora across Greenland.
The final section presents a categorical explanatory variable analysis of the volcanically derived EXS fluxes.
5.2.1 Raw Data and Synthesized Variables
As described in Chapter 3, data characterizing the Laki and Tambora eruptions are available from 20 ice cores and thus allow an examination of the spatial variability of the associated EXS deposited over Greenland. For this comparison the data for the year of maximum impact and/or deposition were used, that is 1783 for Laki and 1816 for
88 Tambora. From the 20 ice cores, 19 of them contain the Laki horizon and 17 of them contain and the Tambora horizon. Unfortunately, different methods were used to identify and characterize the volcanic emissions so that the synthesized variables were not all constructed by identical approaches. The details of the data and variable calculations are discussed in the following section.
The four variables examined are the volcanic sulfate flux (volcanic EXS), the non-volcanic excess sulfate flux (NV-EXS), the excess sulfate density (EXS density) and the annual accumulation rate (An). For each of these 4 variables there are two values, one for 1783 and one for 1816, for each of the relevant ice cores. The volcanic EXS reflects the total mass of sulfate aerosols originating from Laki or Tambora and is calculated as the difference between the total flux and the calculated background flux for the specific year (1783 or 1816). The NV-EXS contains the non-volcanic excess sulfate aerosols deposited in the GIS. The volcanic EXS and NV-EXS are calculated differently depending upon the nature of the raw data available. For the PARCA and Summit ice cores the calculation of the total EXS is discussed in Section 3.2.3. Note that the NV-EXS flux was calculated by averaging the 5 years before the eruption and 5 years after [Wei et al., 2008]. This smaller time interval was used rather than that for the entire core to avoid more recent anthropogenic contributions to the atmospheric sulfate burden. For the earlier ice cores published by Hammer and Clausen [1988], the total deposition of sulfate flux was either calculated from the measured H+ values or from direct measurement of the sulfate concentrations. They calculated the volcanic sulfate flux as the sum of the values exceeding the background concentration calculated as the average for 1782 and 1784
89 (Laki) and 1815 and 1817 (Tambora). Hence their approach was quite different from ours
(see Section 3.2.3).
The third variable, EXS density, was calculated by dividing the volcanic sulfate flux by the annual accumulation rate, thereby standardizing the EXS flux with respect to the An. Thus, spatial variations in the EXS density should reflect depositional processes other than differences in accumulation rate and if the spatial variability of EXS density is found to be homogeneous, it would imply that most EXS is deposited by wet processes.
Alternatively, if the EXS density is spatially variable over Greenland, then regions with higher values would be expected to experience more dry deposition. By comparing the sulfate deposition associated with Laki and Tambora, the differences and similarities in the spatial distribution of EXS density should provide additional information about regional depositional mechanisms and/or possible transport pathways of the volcanic aerosols. Examination of the spatial variation of EXS density for a specific volcanic event should reveal the differences in more local depositional mechanisms. The final variable,
An, was obtained by adding all the sample lengths (in w.e.) for the year.
With regard to the differences among the ice cores, it is instructive to note that the
Raven (1998) and Dye 2 (1979) cores were drilled very close to each other but nearly 20 years apart. Their EXS and An values differ significantly as seen in Table 5.1. For example, the Laki EXS flux in Dye 2 is more than 1.5 times that in Raven although the accumulation for that year (1783) is only 75% that in Raven. These differences likely reflect advances in analytical techniques, improvements in temporal resolution, and the use of different calculation methods. As described by Clausen and Hammer [1988], the
90 Dye 2 (1979) core was analyzed using the electrical conductivity measurements (ECM)
2- that were then converted into anion (e.g., SO4 ) concentrations. However, sulfate in the
Raven (1998) core was measured directly using ion chromatography. Unfortunately, many of the older cores are either depleted or degraded so it is impossible to re-measure them using newer techniques.
5.2.2 Spatial Interpolation
In order to develop an ice-sheet-wide estimate of volcanic EXS flux, EXS density,
NV-EXS flux, and accumulation for interpolation, data from 20 ice core sites, 11 from
Clausen and Hammer [1988], and 9 from the PARCA and Summit collections were used.
The data from 19 cores containing Laki EXS (Table 5.1) and 17 cores containing
Tambora EXS (Table 5.2) were interpolated using kriging at the resolution of 24 km * 24 km [Bromwich et al., 2005]. As spatial prediction not only depends on the existing observation, but also on the embedded spatial dependency, the empirical semivariograms,
γ? z(h), were calculated based on the observed data. The semivariogram, denoted as γz(h), is a function describing the degree of spatial dependence of a spatially random field or stochastic process. It is defined as the expected squared increment of the values between
1 locations si and sj, i.e. γ (h) ≡ var[Z(s ) − Z(s )] [Cressie, 1993b; Calder and Cressie, z 2 i j in press]. To interpret the existing data, we consider a set of theoretical model to fit the calculated empirical semivariogram. Generally, three parameters, the nugget, the sill, and the range, are used to describe the semivariogram model. The nugget is defined as the height of the jump of the semivariogram at the discontinuity at the origin. The nugget
91 effect is likely to reflect measurement errors. It may also represent the microscale variation [Matheron, 1962]. The sill is the value of the variogram when the lag distance tends to infinity and represents the maximum possible amount of variability of the spatial process at distinct points. The range is the distance over which the difference of the semivariogram from the sill becomes negligible. From the estimates of these parameters, the spatial correlation for each of the four variables was defined.
Semivariograms were calculated for the ice core-derived data at lag increments of
72 km for 20 lags. Spherical models (5-1) were fitted to the semivariograms with different nuggets, sills, and ranges, as shown in Fig.5.1.
(5-1)
where a > 0, c0 > 0, cs > 0.
The parameters of the theoretical spherical models were estimated using the ordinary least squares estimation method. The results are shown in Table 5.3.
92 NV-EXS 1783 Laki EXS 1783 A 1783 Laki EXS Core name n (kg/km2) flux (kg/km2) (mm w.e. a-1) density (103 kg /km3) D3a 19 296 525 564 D2a 13 236 597 395 NASA-Ua 11 156 208 750 Ravena 10 125 304 411 Site Aa 5 171 244 701 Site Ta 8 116 261 445 Humboldta 10 52 149 347 Tunua 5 82 107 766 Camp Centuryb 14 280 195 1436 North Centralb 17 100 248 403 Crêteb 19 138 389 355 Site Bb 38 167 378 442 Site Db 60 291 471 618 Site Eb 25 62 275 225 Site Gb 20 145 282 514 Dye 3b 18 173 264 655 4Bb 13 168 220 764 18Cb 40 190 377 504 Dye 2b 36 231 233 991 aThese data were calculated as discussed in Wei et al., 2008; bThese data are from Clausen and Hammer, 1988.
Table 5.1 Laki EXS information from 19 Greenland ice cores
93 NV-EXS 1816 Tambora 1816 An 1816 Tambora EXS Core name (kg/km2) EXS flux (kg/km2) (mm w.e. a-1) density (103 kg /km3) D3a 15 80 586 136 D2a 13. 30 470 64 GITSa 13 20 259 75 Ravena 12 38 395 97 Site Aa 9 29 319 92 Site Ta 8 19 194 96 Tunua 5 7 70 100 Camp Centuryb 18 63 212 297 North Centralb 32 48 342 140 Crêteb 26 53 390 136 Site Bb 64 71 449 158 Site Db 71 129 587 220 Site Eb 25 13 300 43 Site Gb 53 94 584 161 Dye 3b 48 54 748 72 4Bb 89 98 871 113 18Cb 72 25 689 36 aThese data were calculated as discussed in Wei et al., 2008; bThese data are from Clausen and Hammer, 1988.
Table 5.2 Tambora EXS information from 17 Greenland ice cores
94
Figure 5.1 Empirical semivariograms for volcanic EXS flux, EXS density, NV-EXS flux, and accumulation for 1783 and 1816. The dash lines indicate the fitted spherical semivariogram models.
95 Nuggeta Sillb Rangec 1783 Laki EXS flux 260 4968 14 1783 Laki density 73 71965 19 1783 NV-EXS flux 74 193 18
1783 An 144 15167 13 1816 Tambora EXS flux 126 1100 22 1816 Tambora density 127 4004 19 1816 NV-EXS flux 165 679 17
1816 An 671 43375 19 aNugget: The units for EXS flux are (kg/km2)2. For EXS density, 3 2 -1 2 the units are (t/km ) , and for An, the units are (mm w.e. a ) ; bSill: The units are the same as those of nugget; cRange: The unit is in 72 km/grid distance;
Table 5.3 Spherical model estimated parameters for the volcanic EXS flux, density,
NV-EXS flux, and accumulation in 1783 and 1816
5.2.3 EXS fluxes from Laki and Tambora to Greenland
Because of its geographic approximity to Greenland, the Laki eruption produced a much larger concentration of EXS over Greenland, as seen by comparing the kriged maps for these two events on the same scale (Fig. 5.2 A and B). Based on the 19 ice cores containing the Laki event, the mean EXS flux is 168 kg/km2. In contrast, the average
EXS flux based on the 17 cores containing Tambora event is 51 kg/km2. Using their 16 common cores, on average, Laki deposited 119 kg/km2 more EXS over Greenland than
Tambora.
96 A. 1783 B. 1816 C. 1816 97
Figure 5.2 Spatial variability of the EXS fluxes from Laki and Tambora (units are kg/km2). (A) and (B) are the volcanic EXS fluxes in
1783 and 1816 shown on the same scale. (C) is the volcanic EXS flux in 1816 over a smaller range to highlight regional differences.
The black points indicate the locations of the ice cores.
97 The kriged map of the Laki EXS (Fig. 5.2 A) shows that the regions recording the largest fluxes lie in the southeast along the coast of the Denmark Strait (see Fig. 2.3 for location). Fluxes tend to decrease to the northwest, a pattern that is consistent with Laki’s location in Iceland across the Denmark Strait such that the highest EXS fluxes occur in the more proximate regions and decline with distance from the source as aerosols are removed from the air masses either by dry air fallout or by incorporation within precipitation (snow). Interestingly, the coastal area in northwestern Greenland also shows high fluxes of EXS, possibly reflecting the locally high accumulation rate. In northeastern
Greenland, where the annual precipitation rate is very low, the lowest EXS fluxes are also observed. This spatial pattern of EXS flux agrees well with the simulation result by
Oman et al. [2006] who used the GISS modelE coupled to a sulfur cycle chemistry model
(Fig 5.3, modified from Fig. 7 in Oman et al., 2006). Unfortunately, the dearth of ice cores from northeast Greenland results in a high kriging variance, and thus subsequent discussions are restricted to the central, southern, western regions plus the extreme northwestern coastal area.
The kriged map for the 1816 EXS (Fig. 5.2 C) also shows high flux in the south, but with a more homogenous pattern from east to west across central Greenland. Overall, central Greenland received little EXS from Tambora. In west central Greenland, there is a pronounced peak in EXS flux close to the D3 site, which is coincident with the region of maximum accumulation (see Fig. 2.2) [Bales et al., 2001c]. As discussed in Chapter 2, accumulation over Greenland is usually higher in the south, southeast, and west, and lower near the central area and in the northeast. The similarity between the EXS flux and the long term average accumulation pattern implies that the deposition of Tambora EXS
98 may largely depend on the regional accumulation rate. That is, Tambora EXS is very likely deposited primarily via wet deposition. However, accumulation over Greenland exhibits large inter-annual variability [Alley et al. 1993; Mosley-Thompson et al., 2001], thus it may range widely over the span of a single decade. As the An data are available for
1783 and 1816, it may be worthwhile to explore the effect of accumulation on the deposition of EXS.
2 Figure 5.3 Total SO4 (kg/km ) deposition from the Laki eruption to Greenland and surrounding areas averaged from the three simulations (modified from Oman et al.,
[2006]). The 2000 m elevation is shown for Greenland.
99 A. 1783 B. 1783 C. 1816 100 100
Figure 5.4 Spatial variability of the annual accumulation rate in 1783 and 1816 (units are mm w.e.). (A) is the 1783 An over a smaller range. (B) and (C) are An in 1783 and 1816 on the same scale. The black points indicate the location of the ice cores. The black points indicate the location of the ice cores.
100 To further explore the possible impact of differential An on EXS, the spatial distributions of An in 1783 and 1816 are examined. Fig. 5.4 B reveals a different spatial pattern for An in 1783 versus that in 1816 (Fig. 5.4 C). In 1783, An appears to be more homogenous with a maximum of ~450 mm w.e. a-1 in west central Greenland close to the
D3 site (Fig. 5.4 A). This is similar to the average accumulation pattern (Fig. 2.2) except that the southeastern coastal area does not show the pronounced high values. In fact, An in the southeast is 100-200 mm w.e. less than the long-term average (Fig. 2.2). When compared with the Laki EXS flux, the spatial pattern of An shows an inconsistency. In central Greenland, the maximum An is located on the west side of ice divide; while the maximum in the Laki EXS flux is observed on the east side. This reversed pattern suggests that wet depositional processes may have dominated in west central Greenland in 1783. However, precipitation over the southeastern coastal areas, where An is usually high, may have been suppressed due to its geographical proximity to Iceland, where a severe winter was reported for 1783 [Thordarson and Self, 2003].
In contrast, the 1816 kriged map shows very high An values in the south and southeast coast (~900 mm w.e. a-1) (Fig. 5.4 C). The long-term average accumulation pattern [Bales et al., 2001c] (Fig. 2.2) also shows high values in the south, southeast, and
-1 west (>800 mm w.e. a in southeast coastal area), and low An in the northeast. Thus, overall the An in 1816 appears rather similar to that of the long-term average, suggesting
Tambora had either a homogeneous or very modest impact on the accumulation for 1816.
A comparison of the Tambora EXS flux and An for 1816 shows a similar spatial pattern with high values in both the south and southeast with a maximum in the central
Greenland. This suggests a potentially high correlation between the Tambora EXS flux
101 and the 1816 accumulation, and likely reflects wet deposition process as the dominant process.
These spatial patterns for An suggest that volcanic eruptions likely affected
Greenland accumulation in both 1783 and 1816 time scales. As mentioned previously, accumulation over the GIS exhibits large inter-annual variability [Alley et al., 1993;
Mosley-Thompson et al., 2001]. Thus, the An history before, during, and after the eruption years was examined. Overall, large fluctuations were observed within this ten-year interval. When compared to the An in five years before and five years after, An in 1783 and 1816 shows no significant deviation around the 10-year mean (Fig. 5.5). The An in both years falls within the ±2σ at each site. For these two periods, the average An are similar, but σ values are more variable during 1778-1788.
To compare the impacts of Laki and Tambora on precipitation, we calculated the average difference of An recorded in ice cores from different locations of Greenland.
Based on 10 ice cores from the southeastern Greenland (Site A, B, D, E, G, T, Crête,
Dye3, 4B, and 18C), the An in 1816 is 197 mm w.e. higher than that in 1783. However, ice cores from the western Greenland (D2, D3, Raven, and Camp Century) only show an average difference of 10 mm w.e. Thus, the An records from ice cores do support the conclusion that the precipitation over the southeastern Greenland in 1783 was very likely suppressed as a result of the geographical proximity to Laki eruption.
As An exhibits large inter-annual variability in the decade surrounding the eruptions of Laki and Tambora, it might be enlightening to examine the average spatial pattern of the NV-EXS. As noted earlier, a linear relationship exists between the average annual accumulation rate and background EXS concentration [Mosley-Thompson et al.,
102 2003]. Thus, a similar spatial pattern between them should be expected unless there are other factors that significantly affect the deposition of NV-EXS. For example, in regions where dry deposition processes are dominant, NV-EXS values should be higher when compared to those in other regions with similar accumulation rates. Here we calculated the 10-year average of NV-EXS, excluding 1783 and 1816, for each eruption. A reasonably similar pattern of NV-EXS flux (Fig. 5.6) and An (Fig. 5.4) emerges with the highest values in central and southern Greenland for both 1783 and 1816 although the absolute values of NV-EXS are higher for 1816. The Pearson correlation coefficients between An and NV-EXS flux are 0.43 (significance level = 93.6%) for 1783 and 0.79
(significance level = 99.9%) for 1816 indicating that significant linear relationships exist for both eruptions. Note that these significance levels are not adjusted for serial correction. However, the higher correlation value for Tambora suggests that Laki may have produced a more heterogeneous perturbation of the sulfate budget over in the GIS.
As indicated earlier, NV-EXS is deposited primarily by wet deposition during non-volcanic years. During the time volcanic sulfate permeates the atmospheres, the presence or absence of wet deposition at a site will contribute significantly to the differential EXS concentrations recorded from site to site. To remove the impact of accumulation so that other possible influential factors can be detected, the EXS densities associated with Laki and Tambora were calculated by dividing the volcanic EXS flux by the accumulation (discussed in section 5.2.1). Assuming that other factors remain constant, a higher EXS density should reflect the stronger influence of the dry deposition process, and/or additional sources of sulfate, such as biogenic sulfate [Robock and Free,
1995].
103
Figure 5.5 Accumulation histories for 5 Greenland ice cores (unit in mm w.e. a-1) from
1778 to 1788 and from 1810 to 1821. The long dash line indicates the mean of the period and the short dash line represents ±2σ.
104 A. 1783 B. 1783 C. 1816 105
Figure 5.6 Spatial variability of the NV-EXS fluxes in 1783 and 1816 (units are kg/km2). (A) is the NV-EXS flux in 1783 over a smaller range. (B) and (C) are the NV-EXS fluxes in 1783 and 1816 on the same scale. The black points indicate the locations of the ice cores.
105 Different spatial patterns are observed in the kriged maps of the EXS density for
1783 and 1816 (Fig. 5.7). In 1783, higher EXS density is observed in the south and southeastern coastal areas (~900 103kg/km3) with a decreasing trend toward the northwest (Fig. 5.7 A). The central Greenland area close to the D3 site EXS density is
563 103kg /km3, with the decreasing trend to both the west and northwest. However, in the far north the highest values lies on both sides of the ice divide. On the west at Camp
Century the maximum is 1436 103kg/km3 while on the east side at Tunu the maximum value is 766 103kg/km3. The high value of EXS density in the southeast along the
Denmark Strait suggests that dry deposition dominates due to the geographical proximity to the source. During transportation further inland, the Laki sulfate aerosols are gradually depleted from the atmospheric through precipitation and/or direct fallout resulting in the fast decrease of the EXS density. In central Greenland, a more homogeneous spatial pattern of the EXS density was observed, but with a much higher values than that in 1816
(Fig. 5.7B).
In contrast, the spatial distribution of Tambora’s EXS density does not exhibit the northwestward trending decline (Fig. 5.7 C) but has a relatively homogeneous spatial pattern with highest values in the extreme south and northwest. This suggests that the
Tambora sulfate was deposited to the snow primarily through wet deposition, similar to the NV-EXS flux during non-volcanic years. The Camp Century core has the highest
EXS density for both 1783 and 1816, suggesting that there may be other sources of sulfate nearby. However, it is also possible that some of the Laki sulfate aerosols were transported there via the frequent northward tracking cyclones from southern Greenland
[Chen et al., 1997].
106 A. 1783 B. 1816 C. 1816 107 107
Figure 5.7 Spatial variability of the volcanic EXS densities in 1783 and 1816 (units are 103kg/km2). (A) and (B) are the volcanic EXS densities in 1783 and 1816 on the same scale. (C) is the volcanic EXS density in 1816 over a smaller range. The black points indicate the location of the ice cores.
107 5.3 Analysis of Variance for volcanic EXS
To determine the geographical factors that control the average deposition of EXS over the GIS, a five-way main effect ANOVA model was used assuming no interactions.
The data include the characteristics of the Laki and Tambora events and along with those for 7 other common volcanic events preserved in the five PARCA cores (see Table 4.1 and Table 4.2). The eruptions in 1786 and 1809 were excluded due to incomplete information of volcano locations. The following factors were considered: distance to the eruption source, geologic magnitude of eruption (VEI), and location of the ice cores
(such as north/south, east/west, and elevation). Here, we assumed that the regional factors and geologic magnitude of the eruption are independent. A log transformation was performed to meet the normality assumption of the model. Therefore, the model is as follows:
Yit = [ln(EXS)]it = β0 + β1NSi + β2WEj + β3LHi + β4Eli + β5Emi + β6VEIt + εit (5-2) where:
Yijklmt is the i-th log observation of the volcanic EXS flux from eruption with VEI level of t;
β0 is the overall mean level of log EXS;
β1,…, β6 are the regression parameters for the regional indicator variables and volcanic explosivity index variables, respectively.
NSi is the indicator variable of whether site i is located at south of 72°N:
NSi =
WEi is the indicator variable of whether site i is located at west of 40°W:
108 WEi =
LHi is the indicator variable of whether the eruption is from a tropical volcano:
LHi =
ELi is the indicator variable of whether the elevation of site i is low
ELi=
EMi is the indicator variable of whether elevation of site i is median
EMi=
VEIt is the volcanic explosivity index of level t, t=1, …, 8;
2 εit is the random error, εit N(0, σ )
The model assumptions, which include examination of independency, normality, equal variance, and any discernable pattern in the residuals, were checked before making further inferences. The results indicate that the residuals are randomly, independently, and normally distributed with constant variance. Thus, the model assumptions are satisfied.
The analysis of variance (ANOVA) table (Table 5.4) reveals that the p-values associated with the model are all smaller than 0.05 except for the elevation factor. This suggests that all the variables except elevation are significant at the 95% significance level. However, there is no evidence of a difference in EXS deposition at high versus low elevation sites.
109 Source DF Sum of Squares Mean Square F-value Pr > F Model 9 60.79 6.75 18.29 <.0001 WE 1 7.96 7.96 21.55 <.0001 NS 1 2.18 2.18 5.91 0.0183 LH 1 5.60 5.60 15.18 0.0003 VEI 4 45.02 11.26 30.48 <.0001 EL 1 0.01 0.01 0.04 0.8413 EM 1 0.01 0.01 0.03 0.8732 Error 56 20.68 0.37 Corrected Total 65 81.47
Table 5.4 ANOVA table for the five-way main effect model
To quantify the differential impact of the geographic factors and the distance to the eruption source on the deposition of EXS over the GIS, we compared the EXS fluxes deposited under the impact of these factors and corrected the bias induced by the previous log-transformation (details of correction is shown in section 6.3.3) [Cressie, 1993c]
(Table 5.5). None of the confidence intervals include 1, indicating that the differences are significant at the 95% level. The volcanic EXS flux on the west side is twice that on the east side. Ice cores in the south contain 1.81 times more EXS than those in the north.
These observations are consistent with the high accumulation rate in the western and southern parts of the GIS (see Fig. 2.2). On average, in Greenland, the An in the south is
~2-5 times than that in the north. In northern Greenland, the An on the west side of ice divide is 2-4 times higher than that in the east side. However, in southern Greenland, the
An is more similar on the west and east sides, with higher values along the coasts and
110 decreasing inland. The latitude of the eruptions also has a significant impact on the volcanic EXS deposited over the GIS. The sulfate burden from high latitude eruptions is
1.87 times greater than that from tropical eruptions. As expected, the short transport distance and time result in enhanced deposition of sulfate from high latitude eruptions. In contrast, sulfate from low latitude eruptions is transported to the GIS via stratosphere as evident by the delay of months to nearly a year in the arrival of sulfate to the GIS. The length of the delay is also influenced by the injection height and residence time of the ejecta. Sulfate aerosols remaining in upper troposphere/lower stratosphere may be quickly removed within a few months such that only a small portion remains in the mid to upper stratosphere for eventual deposition on the GIS.
Contrast Estimate 95% Confidence interval
EXSWest/EXSEast 2.01 (1.49, 2.72)
EXSSouth/EXSNorth 1.81 (1.31, 2.48)
EXSHigh/EXSLow 1.87 (1.38, 2.53)
Table 5.5 Contrast estimates and associated confidence intervals
111 CHAPTER 6
ANALYSIS OF NON-VOLCANIC EXCESS SULFATE TIME SERIES
Atmospheric aerosols affect the Earth’s radiation balance primarily by absorbing and scattering shortwave and longwave radiation, and hence produce both direct and indirect climate perturbations. Although quantification of their climate effects has improved, the IPCC’s Fourth Assessment assigned aerosols a medium to low level of scientific understanding [IPCC Fourth Assessment Report, 2007]. The climatic effects of aerosols differ considerably depending on their physical and chemical properties. As a major component of the atmospheric aerosol mass that originates from both natural and anthropogenic sources, sulfate is essentially a scattering aerosol across the entire solar spectrum with only a small degree of absorption in the near-infrared spectrum [Penner et al., 2001]. Climate models have estimated that sulfate aerosols have a total direct radiative forcing of -0.4±0.2 Wm-2 [Forster and Ramaswamy, 2007].
The increasing emissions of anthropogenic sulfate have perturbed the natural background concentrations, particularly in the latter half of the 20th century [Patris et al.,
2002]. Uncertainties in both the natural and anthropogenic sources and sinks in the sulfur cycle make it difficult to assess the climate forcing due to anthropogenic sulfur emissions.
However, understanding the role of sulfate aerosols is important for efforts to model future climate change scenarios such as those by the IPCC. Such assessments require
112 quantitative time series of the atmospheric sulfur burden so that the potential role of sulfate aerosols may be better constrained. As polar ice sheets provide the best archive of this information, we conducted an analysis of the annual non-volcanic sulfate flux preserved in three PARCA and two Summit ice cores.
Due to the variability of the regional atmospheric circulation that affects accumulation, wind patterns, and the distance to source regions, etc., the preserved ice core-derived EXS histories naturally contain spatial differences. Quantifying this spatial variability is important as it places constraints on the interpretation of the climate histories and as importantly, provides insight to the nature of changes in the dominant atmospheric circulation patterns in that region. In this research project, the primary interests include the quantification of the impact of anthropogenic sulfur emissions and the effect of accumulation on sulfate deposition at different ice core sites. The prior examination of volcanic EXS (Section 5.2.3) indicates that the primary depositional mechanism may vary with location. Thus, assessing the accumulation effect at different ice core sites may provide further insight. In addition to accumulation and emissions, time is also a significant factor in the deposition of EXS as the inter-annual correlations and effects from other time dependent factors are observed in our ice core time series
(section 6.3). To examine these factors and how they affect the deposition of the non-volcanic background EXS fluxes in the five ice core sites, a linear mixed model was employed. The contributions of volcanic eruptions were excluded as they are primarily non-periodic random events and therefore do not have a continuous and constant forcing on the atmospheric sulfate concentration.
113 The following chapter is divided into three sections and discusses the analyses of the non-volcanic EXS (NV-EXS) time series. The first section contains an explanatory data analysis of the NV-EXS time series from the five ice core sites. This provides a thorough examination of the temporal pattern and the factors considered later in the linear mixed model. Next a detailed description of the model is given, which includes the model development process and description of the final model. The last section discusses the inferences and climatic implications drawn from the linear mixed model.
6.1 Explanatory data analysis
Theoretically, understanding and accurately modeling the deposition of NV-EXS over the GIS requires full knowledge of the climate system processes delivering NV-EXS to the atmosphere over the GIS and regional deposition and preservation processes.
Limited by knowledge and observations, not all factors that control the NV-EXS deposition are known, but statistical models can be used to understand the relationships among the various factors and to evaluate the statistical uncertainty arising from the unknown factors. Also, a modeling approach can be used to quantify the impacts of the different factors and allows future inferences and predictions to be made. To properly select one model from a large set of potential models, it is important to examine the characteristics of the distribution of the residuals from the model. Generally, if the residuals follow a normal distribution, linear models (LM) can be applied. Otherwise, generalized linear models (GLM) should be considered based on the distribution type.
After the most appropriate model type is selected, a number of candidate models can be
114 examined to assess the various factors that affect the deposition of the NV-EXS over the
GIS. The factors examined in these models include the anthropogenic contribution, time, and regional accumulation rate. The data used in the model are the NV-EXS time series from five ice core sites in Greenland, which can be classified as longitudinal data. Thus, either linear mixed models (LMM) or generalized linear mixed models (GLMM) should be considered as they permit heterogeneity of variance in the data. Additionally, it is important to examine the correlation structure among the model residuals so that appropriate model assumptions can be made. For example, the inter-annual correlation among the NV-EXS data implies that the yearly sulfate deposition contains a contribution from previous years. In this case, a correlation structure in the residuals may need to be considered in the model.
This section provides a preliminary examination of the non-volcanic EXS time series. First, the assumption of the distribution of the model residuals is made so that the appropriate model type can be chosen. Depending upon the type of model selected, different assumptions will be made regarding the residuals. To validate the selected model, its assumptions will be checked before making any further inferences. The second section examines the four major factors that affect the concentration of NV-EXS in the
GIS. These are the anthropogenic sulfur contribution, the impact of regional deposition process, such as An, the geographical factors, and the time component, including the overall temporal trend and time dependency in the data. In the final section, these parameters are considered for model development.
115 6.1.1 Assumptions regarding the distribution of model residuals
Before we fit our statistical model, it is necessary to examine the possible distribution of the model residuals so that the proper model assumptions can be made.
Generally, linear models assume that the error terms are normally distributed. However, if the residuals are not normally distributed, either a transformation of the original data can be performed to obtain the normality, or a GLM, which allows different distributional data for the residuals, can be used. As we have not yet finalized the model, a linear regression model (6-1) was fitted to the original NV-EXS data to attain an approximate estimation for the distribution of the residuals.
(NV-EXS)i = β0 + β1(An)i+ β2Year i + β3Emissioni + εi (6-1)
The residuals from the linear regression model were examined for their distribution (Fig. 6.1). The histogram (Fig. 6.1 A) of the residuals shows a nice bell shape with slight right skewed tail. However, the normal probability plot does not contain a straight line (Fig. 6.1 B), suggesting that the residuals do not follow a normal distribution.
The plot of the residuals vs. the fitted values (Fig. 6.1 C) shows a megaphone shape pattern, indicating the unequal variance of the residuals. Therefore, a variance-stabilizing transformation should be applied. The plot of residuals according to observation order also shows a zigzag pattern, where the residuals seem to decrease with the increasing year at each site (Fig. 6.1 D).
To equalize the variances, a log transformation was performed to the NV-EXS data. A similar linear regression model (6-2) was then fitted to the transformed data to examine the possible distribution of the residuals.
116 Ln(NV-EXS)i = β0 + β1(An)i+ β2Yeari + β3Emissioni + εi (6-2)
As can be seen from Fig. 6.2, overall the variance was stabilized after transformation. The residuals seem to be randomly and equally distributed around zero.
Although for the larger fitted values, a smaller variance was generally observed (Fig. 6.2
C). The plot of residuals by observation order also does not contain a discernable pattern
(Fig. 6.2 D). The normal probability of the residuals forms a straight line except for a left skewed tail (Fig. 6.2 B), suggesting that the residuals are fairly well normally distributed.
The left skewed tail is coincident with the histogram and the outliers observed in Fig. 6.2
C and D.
As the residuals seem to better follow the normal distribution after the log transformation with a relatively constant variance, a linear model is selected to explore the different factors that control the deposition of NV-EXS in Greenland.
117
Figure 6.1 Residual plots of the linear regression model using the original NV-EXS data.
(A) Histogram of the residuals; (B) Normal Probability Plot of the residuals; (C)
Residuals versus the fitted values; (D) Residuals versus the order of the data, entered as year 1782-1985 for Tunu, year 1782-1985 for Site A, year 1782-1985 for Site T, year
1782-1985 for D3, and year 1782-1985 for D2.
118
Figure 6.2 Residual plots of the linear regression model using the Ln(NV-EXS) data. (A)
Histogram of the residuals; (B) Normal Probability Plot of the residuals; (C) Residuals versus the fitted values; (D) Residuals versus the order of the data, entered as year
1782-1985 for Tunu, year 1782-1985 for Site A, year 1782-1985 for Site T, year
1782-1985 for D3, and year 1782-1985 for D2.
119 6.1.2 Temporal patterns and the inter-annual time dependency in the non-volcanic
EXS time series
The average NV-EXS flux differs from site to site in our five ice cores. For each site, NV-EXS remained relatively constant from 1782 to 1850 (Fig. 6.3). After 1850, increasing NV-EXS concentrations were observed in all the five ice cores. A pronounced increase occurred in the late 1890s and peaked in the late 1910s. The NV-EXS fluxes decreased thereafter and reaching a minimum in the late 1930s to early 1940s after which they have increased steadily. Comparison of the averages for the pre-industrial and post-industrial periods reveals that the NV-EXS flux roughly doubled, in concert with anthropogenic sulfur emissions (Table 6.1). Regions of low An appear more sensitive to this human-produced contribution and show a greater increase in the NV-EXS. For instance, NV-EXS from the Tunu and Site A sites show 126% and 105% increases, respectively, while the rates of increase for D3 and D2 sites are more modest (96% and
79%, respectively). A sharp decreasing trend was observed in the mid 1970s in all cores except the Summit core. This decreasing trend appears to precede the overall reduction in anthropogenic sulfur emissions for NH in the late 1980s, but is contemporary with the decreasing emissions from North America and West Europe (Fig. 2.4), suggesting these regions are the primary sources of sulfur deposition over Greenland.
120
Figure 6.3 Annual fluxes of non-volcanic EXS (kg/km2) in five Greenland ice cores
121 Tunu Site A Site T D3 D2
-1 An (mm w.e. a ) 110 266 222 450 450 (kg/km2) 5.4 6.7 6.7 13.9 12.5 1782-1850 σ2 (kg/km2)2 4.1 6.1 2.6 13.2 9.5 (kg/km2) 12.1 13.7 13.3 27.2 22.3 1850-1985 σ2 (kg/km2)2 59.6 46.1 38.9 154.9 110.6 % increase in Average 126 105 100 96 79 % increase in Variance 1339 659 1414 1071 1067
Table 6.1 Non-volcanic EXS averages ( ) and variance (σ2) in five Greenland ice cores
As previously discussed, it is important to examine the inter-annual correlation in the NV-EXS data so that appropriate model assumptions can be made. To assess the possible time dependency in the data, it is essential to consider the physical and chemical properties of sulfur species in the atmosphere, such as their transformation processes and lifetimes. Initially, sulfur gases and dimethyl sulfide (DMS) are emitted to the atmosphere where they react (via gaseous phase) with the hydroxyl radical (OH-) and/or water vapor
(H2O) to form sulfate aerosols [Bekki, 1995; Koch et al., 1999]. The conversion rate of sulfur gases to aerosols is 0.06-0.6% day-1 in the upper troposphere and 0.005-0.01% day-1 in the stratosphere [Laj et al., 1990]. Therefore, it takes roughly a month for sulfur gases to transform into aerosols in the stratosphere, but only 2-15 days in the upper troposphere. Laj et al. [1990] noted that part of the volcanic sulfur was present as SO2 over the GIS at the time precipitation was forming, which suggests a fast transport time for SO2 and/or possibly a longer atmospheric lifetime. However, for the annual NV-EXS flux, the effects on the order of days to months are negligible. The atmospheric lifetime
122 for sulfate aerosols in the troposphere is only 1-3 weeks, but this increases with altitude.
In the stratosphere, the sulfate aerosol lifetime can be as long as 1 to 3 years. As the sulfur burden in the troposphere is much larger than that in stratosphere, it seems reasonable to assume that the current year’s sulfate fluxes do not affect those of the next year, i.e. the independency of the annual data. However, a modest correlation at lag 1 (1 year) was observed for the ice core time series (section 3.2.5.2), suggesting that the correlation structure in the residuals needs to be considered.
6.1.3 Impact of regional accumulation rate and geographic factors on the non-volcanic EXS fluxes
As implied from the linear relationship between An and the background EXS flux
[Mosley-Thompson et al., 2003], the regional accumulation rate can strongly affect the deposition of the NV-EXS. As shown in Table 6.1, the mean and variability of the
NV-EXS flux varies from site to site. Regions with higher An (D2 and D3) exhibit a larger variability than those with lower An (Tunu, Site A, and Site T). The magnitude of the increase of the mean NV-EXS also seems to depend upon the regional An. Regions with lower An exhibit a larger percentage increase in both the mean and variance, although it is greater for the variance than the mean. Overall, as a result of anthropogenic sulfur emissions, the variances at all the five sites increase by an order of magnitude after
1850 except Site A (Table 6.1).
The differences in the NV-EXS variability (σ2) from site to site over the GIS suggest the possibility of a geographic effect. This effect can be represented by a random
123 effect component in the model, which allows for residual correlation between observations at the same location. However, as the average An varies from site to site, the geographical effect on the NV-EXS is probably nested in the An effect. Thus, this will be taken into consideration during the later development of the model.
6.1.4 Impact of anthropogenic sulfur emissions on the non-volcanic EXS fluxes
As shown in Fig. 6.3, since 1850, increasing NV-EXS concentrations are observed in all the five ice cores and reflect the increasing sulfur emissions from human activities.
Each ice core contains a somewhat different concentration of anthropogenic sulfate, depending on its altitude, distance from the various sources, and the regional circulation patterns and different deposition mechanisms. To remove the local noise and examine the relationship between the NV-EXS time series from ice cores and the sulfur emission data
[Stern, 2005], three stacked data sets were produced and correlated with the emission data for the period from 1850 to 1985. The stacks were computed by summing up the normalized annual NV-EXS time series. Stack 1 contains the three ice core sites east of the ice divide, where accumulation rates are low to medium. Stack 2 contains two ice cores west of the ice divide, where accumulation rates are relatively high. Stack 3 contains all the five ice cores, considered to be representative to the entire GIS.
The stacked NV-EXS records were correlated with emissions from four major NH regions (Western Europe, North America, Eastern Europe, and Asia). These regions were selected as they play a significant role in global sulfur emissions. Based on the sulfur emission data from Stern [2005], Europe accounted for 88% of the global emissions in
124 1850. Later the locus of global emissions shifted first to North America, and then to
Eastern Europe and the Soviet Union. By the late 1950s, Asia emerged as a substantial emitter and by the 1990s became the single largest source region.
Cores in stack W. Europe N. America E. Europe Asia NH
Stack 1 Tunu, Site A, Site T 0.91 0.85 0.87 0.83 0.92
Stack 2 D2, D3 0.79 0.83 0.67 0.63 0.77
Stack 3 All 0.92 0.90 0.85 0.80 0.92
Table 6.2 The square of the correlation coefficient (r2) between the annual NV-EXS and regional sulfur emission data
The correlation summary is shown in Table 6.2. High positive correlations are observed for all the three stacks with the emission data from four major regions and the
NH, at a significance level of 99.9% (unadjusted for serial correlation). Overall, Stack 1 shows a remarkably high r2 (0.92) with NH emissions, while the score for Stack 2 is lower (0.77) (Table 6.2). This suggests that lower accumulation rate sites are more sensitive to changes in the sulfur budget of the atmosphere. Stack 1 is most strongly correlated with emissions from Western Europe (r2 = 0.91), suggesting that the east side of the GIS receives more sulfate input from Western Europe than from other source regions. On the other hand, Stack 2 is most strongly correlated with emissions from North
America (r2 = 0.82), rather than those from Europe and Asia, suggesting that the western
125 side of the GIS receives considerably more input of sulfate from North America. Stack 3 includes all the five ice cores, representing the impact on over much of the GIS.
Correlations between Stack 3 and EXS from Western Europe and North American are higher (r2 = 0.92 and 0.90, respectively) than those from Eastern Europe and Asia (r2 =
0.85 and 0.80, respectively). These results support the logical assumption that the atmospheric sulfur burden over the GIS is strongly dominated by the emissions from the most proximal sources, Western Europe and North American. However, emissions from
Asia and Eastern Europe also contribute to the NV-EXS level over Greenland.
6.2 Linear mixed model
From the previous explanatory data analysis, the possible factors that control the deposition of NV-EXS over the GIS include a time component, the impact from regional
An, and anthropogenic sulfur contributions. To quantify the effect of these factors on the concentration of NV-EXS over the GIS, a linear mixed model was selected. The log
NV-EXS data are used with the assumption of normally distributed residuals.
As previously discussed, it is quite likely that a geographical effect is nested in the An effect. If this is the case, a random effect component should be included in the An
(random slope of An) to simulate the regional variability. Otherwise, the An effect only contains a fixed component (fixed slope), that is, an homogeneous impact on the deposition of the NV-EXS. The geographical effect is then expressed as a random intercept for the model. Thus, in the following model development section, a series of models with different settings for the random intercept and/or random slope were
126 compared. The NH anthropogenic sulfur emissions and time component are both assumed to have a homogeneous impact on the NV-EXS over Greenland. To finalize the model, the time dependency in the residuals and between-site variance were examined to guide selection of appropriate model assumptions.
6.2.1 Model development
To understand the different effects contributed by the regional accumulation rate, time, and anthropogenic sulfur emissions with consideration of the geographical impact on the non-volcanic sulfate flux extracted from Greenland ice cores, a four-way linear model (6-3) was chosen:
yi = Ln(EXS)i = β0 + β1(An)i+ β2Year i + β3Emissioni + β4Sitei+ εi (6-3) where yi is the i-th log NV-EXS observation;
β0, β1, β2, β3, and β4 are the regression coefficients;
εi is the error term.
The time and sulfur emissions are treated as fixed effects in the model, i.e. these factors both have constant slope in the linear model since they tend to have identical effects on the NV-EXS over the GIS. The annual NH sulfur emission data [Stern, 2005] were used as they have an overall high correlation with the sulfate time series in all the five ice cores. A random effect approach was chosen to model the site effect in the sense that it attempts to describe the ice sheet wide effect beyond the particular ice core location. However, there is some judgment required in deciding whether to use a random
127 intercept and/or a random slope for the accumulation term at the different sites. A random intercept is appropriate under the condition that the accumulation effect is homogeneous over the GIS but the total EXS deposited differs among the regions due to geographical factors. Alternatively, the random slope for accumulation is a better choice when the geographic effect is nested in the An effect. In this case, the effect of An on the deposition of NV-EXS varies over the GIS. That is, for every unit change in An, the amount of deposited NV-EXS is distinct for different regions.
Generally, the error term is assumed to be independently normally distributed, i.e.
2 εijkmt ~ NID(0, σ ). However, considering the modest lag 1 autocorrelation in the time series, as discussed in section 3.2.6.2, we may need to include the correlation structure into the linear mixed model if the simple model fails to explain the internal relationship in the data. Also, as the variability in the EXS flux differs from site to site, models with and without constant variance in the residuals for all sites were compared.
With the constraints discussed above, the following linear mixed models were examined: (Model 1) a random intercept model; (Model 2) a random An slope model; and
(Model 3) a random intercept and a random An slope model. In addition to the selection from the previous three candidate models, we further compared (Model 4) model with temporal correlation structure, and (Model 5) model without constant variance in the residuals.
To guide the selection of either random intercept and/or random An slope models, the results of the comparison among the models are shown in Table 6.3. Among Models 1 to 3, random An slope model (Model 2) has the lowest Akaike Information Criterion (AIC)
128 [Akaike, 1974] and Bayesian information criterion (BIC) [Schwarz, 1978]. AIC and BIC are both indicators of goodness of fit for the model. They are relative measures of the precision and complexity of the model and describe the tradeoff between bias and variance in model construction. Usually, given a data set, the model with the lowest AIC and/or BIC would be considered as the best option. Although Model 3, where both the random intercept and the random An were considered, has an AIC similar to that of Model
2, the hypothesis test for selecting one of these two models has a p-value close to 1. This suggests it is not worth including random slope in the model. Therefore, Model 2 was selected for further examination of the correlation structure and constancy of variance in the residuals.
The residuals for Model 2 were examined for evidence of autocorrelation as shown in Fig. 6.4. The lag 1 autocorrelation coefficients are 0.33, 0.39, 0.39, 0.49, and
0.44 for Tunu, Site A, Site T, D3, and D2, respectively. These lag 1 autocorrelations are all higher than the 95% confidence interval, suggesting that Model 2 was not sufficient to explain the inter-annual correlation. Thus, we included the correlation structure
ARMA(1,1) into Model 4 as most of the lag 2 autocorrelations were within the 95% confidence bounds. The ARMA(p,q) model is the autoregressive moving average model with p autoregressive terms and q moving average terms [Box et al., 1994],
p q i.e.εt = γ t + ∑ϕiεt−i + ∑θ iγ t−i , where εt is the model residual in time t, γ t is the noise i=1 i=1
2 term in time t with zero mean and variance σ , ϕi and θ i are the parameters for the autoregressive and moving average components. Thus, the ARMA(1,1) model contains
129 the AR(1) and MA(1) models, i.e., εt = γt + ϕ1εt−1 + θ1γt−1. The AIC and BIC of the model indicate much lower values than those of Model 2 (Table 6.3), suggesting a better fit to
the data. It is estimated that parameters of ARMA(1,1) are ϕ1 =0.73 and θ1 = -0.37, suggesting a significant correlation in lag 1.
Model AIC BIC Log Likelihood Test Likelihood Ratio p-value Model 1 545.5 574.7 -266.8 Model 2 523.5 562.4 -253.7 1 vs. 2 26.0 <0.0001 Model 3 525.5 569.3 -253.7 2 vs. 3 0.0 1 Model 4 339.2 386.9 -159.1 2 vs. 4 189.3 <0.0001 Model 5 281.3 349.5 -126.7 4 vs. 5 64.9 <0.0001
Table 6.3 Comparison of the characteristics of five linear mixed models; AIC is the
Akaike Information Criterion and BIC is the Bayesian information criterion, as discussed on p.129.
130
Figure 6.4 ACF and Partial ACF plots of the residuals of the linear mixed model with a random An effect. (A) ACF for the within-group residuals; (B)-(F) PACF for individual ice core sites. The red lines indicate the 95% confidence bounds.
131 6.2.2 Final model description
The model development procedure described in the previous section led to the selection of the linear mixed model, with both time and sulfur emissions as fixed effects and An having both a random effect and fixed effect. The random effect is grouped by site as the accumulation effect varies across different ice core locations. Also, the variance for the errors is specified as constant only within groups (sites). Therefore, the final model is as follows:
yi = ln(EXS)i = β0 + β1(An)i + β2Year i + β3Emissioni + bm(An)i + εi (6-4) where yi is the i-th observation of the observed NV-EXS;
β0 is the intercept term;
β1, β2, and β3 are the fixed-effect coefficients for An, time, and sulfur emission
components, respectively. They are all identical for different sites;
Year i and Emissioni are the fixed-effect regressors, representing time, and the sulfur
emission effect, respectively;
(An)i is the mixed-effect regressor, which has both fixed-effect and random effect
components; bm is the random-effect coefficient for (An)i at site m, m=1,…,5. bm is assumed to be
2 normally distributed, i.e. bm ~ N(0, ψ ). Here bm is regarded as a random variable, not
as a parameter, and is similar in this respect to the errors εi;
132 εi is i-th error term of the NV-EXS. The errors for each site are assumed to be normally
distributed with the ARMA(1,1) correlation structure. The error terms are assumed to
have constant variance at each ice core site, but vary from site to site.
6.2.3 Checking model assumptions
In order to validate the model before we make any further inferences, it is necessary to check its goodness of fit and all the model assumptions. Generally, normal linear models assume the error variables are independently and normally distributed with constant variance. In contrast, general normal linear models contain a more relaxed assumption about the error terms. For example, correlation and non-constancy in the error terms are allowed in the models. As one of the general linear models, our linear mixed model assumes an ARMA(1,1) correlation structure in the residuals with constant within-group variance. A normal distribution for the error variables was also assumed.
Thus, we need to confirm that these assumptions are valid in our model before any further inferences are made.
Overall, the p-values associated with all the parameters are less than 0.005, suggesting a strong goodness of fit of the model (Table 6.4). Two types of residuals were calculated, standardized residuals (also known as Pearson residuals) and normalized residuals. Standardized residuals are calculated by dividing raw residuals (observation - fitted values) by the corresponding standard errors. As we assumed non-constant error terms, we intend to use standardized residuals instead of raw residuals in checking the normality and any other pattern in the residuals. Normalized residuals are calculated by
133 pre-multiplying standardized residuals with the inverse square-root factor of the estimated error correlation matrix. Besides the normality and potential patterns in the residuals, they remove the estimated correlation structure and allow us to check if there is any remaining correlation. If the original estimation sufficiently explains the correlation structure in the residuals, the normalized residuals should be randomly and equally distributed around zero.
Overall, the pooled standardized residuals seem to follow the normal distribution with a small left skewed tail (Fig. 6.5A and B). The hypothesis test for normality for all the data has a p-value of 0.139 (Fig. 6.6), suggesting that we can not reject the hypothesis that the residuals are normally distributed at the confidence level of 95%. The plot of residuals vs. fitted values shows that the residuals are randomly distributed around zero
(Fig. 6.5C). No discernable pattern was detected from the pooled standardized residuals.
The ACF of the normalized residuals was calculated and is shown in Fig 6.6. The correlation values (except lag 0) fall within the 99% confidence bounds, suggesting that our estimated ARMA(1,1) model sufficiently explains the correlation structure in the error terms.
134
Figure 6.5 Standardized residual plots for the linear mixed model. (A) Histogram; (B)
Normal probability plot; (C) Residuals vs. fitted values.
Figure 6.6 Autocorrelation plot for the within-group residuals. The red lines indicate the
99% confidence bounds.
135
Figure 6.7 Normality plots of the residual for the linear mixed model. The blue lines indicate the 95% confidence interval for the data.
136 In addition to the examination of the overall residuals, plots of the residuals for individual sites were constructed to examine the normality, randomness, and existence of any patterns neglected in the assessment using the pooled data. The residuals at each site appear to be normally distributed with a small left skewed tail (Fig. 6.7). The normality test for the residuals for the individual sites all have p-values greater than 0.18, indicating it is reasonable to assume the multinormality of the residuals.
In the plots of residuals vs. fitted values, the residuals appear to be randomly and evenly distributed around zero with several negative outliers (Fig. 6.8). No detectable pattern is found for the individual sites. The skewed tail observed in the normality plots and the outliers in the residuals vs. fitted values plot (Fig. 6.5 C and Fig. 6.8) probably reflect a low net An year, during which little precipitation was received or part of the annual snowfall may have been removed by snow drifting. The outliers may also reflect the incomplete removal of the volcanic contribution after 1900s due to the increasing sulfur background. However, the latter type of outliers should produce a right-skewed tail as they would likely be large values.
137
Figure 6.8 Plots of the residual vs. fitted values for the linear mixed model for five ice core sites
6.3 Inferences and climatic implication
From the model output (Table 6.4), the relationship between the non-volcanic
EXS flux with all the fixed effect components (An, time, and NH sulfur emissions), is as follows:
/\ E(yi ) = -4.04 + 0.0031(An)i+ 0.0030Yeari + 0.000013Emissioni (6-5)
All the p-values associated with the model parameters are less than 0.001, suggesting there are significant effects from accumulation, time, and NH sulfur emissions with a confidence level of 99.9%.
Since a log-transformation was performed on the original data before modeling, bias corrections are needed for the parameter estimates. Often, if Y is lognormal distributed, i.e. ln(Y) = X ~ N(μ,σ2), the transformed variable X is modeled. However, when the estimates of X are back transformed to the original scale, i.e.
138 and , biases are produced for the estimates of Y. To correct for bias, the following transformations are generally used [Cressie, 1993c].
(6-6)
(6-7)
By applying these bias corrections to the ln(EXS) data, the parameter estimates and C.I.’s on the original scale are calculated and shown in Table 6.5.
Estimate Std. Error t-stat P-value Intercept -4.04 1.16 -3.48 0.0005
An 0.0031 0.0011 2.93 0.0004 Year 0.0030 0.00062 4.81 0.0000 Emission 0.000013 0.0000019 7.14 0.0000
Table 6.4 Parameter estimates for the fixed effects in the model
Estimate Std. Error Approximate 95% C.I.
An 1.0031 1.0011 (1.0010, 1.0053) Year 1.0030 1.00062 (1.0017, 1.0042) Emission 1.000013 1.0000019 (1.000010, 1.000017)
Table 6.5 Parameter estimates and confidence intervals (C.I.) for the fixed effect in the original scale
139
On the regional scale, the EXS flux deposited over the GIS is significantly affected by local accumulation rate, time, and the atmospheric sulfate aerosol concentration, as implied by the model. Since 1850, the increasing sulfur emissions from human activities have greatly increased the atmospheric sulfate aerosol concentration as recorded in Greenland. Very quick responses were observed in the EXS deposited in the
GIS contemporaneous with the increasing emissions from North American and Western
Europe. This was also reflected by the high correlation coefficients between the EXS flux and sulfur emissions from these two regions (Table 6.2). Therefore, both the contemporaneity and the high correlation suggest that the primary sources of NV-EXS to the GIS are N. America and W. Europe. The model indicates that for every 1 Gg
(thousand metric tonnes) increase in the annual NH sulfur emissions, there is a 0.000013 increase in ln(EXS), or a 0.0013% increase in the annual non-volcanic sulfate flux (Table
6.5). For example, between 1850 and 1950, the total NH sulfur emission increased
~30290 Gg, which led to an average increase of 39.4% in the annual non-volcanic sulfate flux in the GIS relative to the 1850 sulfate background level, holding other components constant. Between 1950 and 2000, the total NH sulfur emissions increased ~12688 Gg, resulting in a 16.5 % increase in the sulfate flux in the GIS relative to 1950s level. A 95% confidence interval for this slope is (0.000010, .000017). Thus, with 95% confidence, we conclude that for every 1 Gg increase in the annual NH sulfur emission, there will be a
0.0010% to 0.0017% increase in the annual non-volcanic sulfate flux to the GIS.
The effect of accumulation contains both fixed and random components. The fixed effect represents the average effect of accumulation over the GIS on the deposition
140 of non-volcanic EXS flux. Thus, on average, every 1 mm (w.e.) increase in annual accumulation would result in a 0.0031 increase in ln(EXS), equivalent to a 0.31% increase in the annual non-volcanic sulfate flux (Table 6.5). A 95% confidence interval for this slope is (1.0010, 1.0053). Thus, with 95% confidence, we conclude that for every
1 mm (w.e.) increase in the annual accumulation, there will be 0.10% to 0.53% increase in the annual non-volcanic sulfate flux in the GIS.
The random component of the An suggests that there are different effects
(different slopes) at each core site. This is probably due to a difference in the mechanisms controlling local deposition. In response to contributions by different emission sources or different distances of the sources from the GIS, the sulfate aerosol density may vary considerably over the GIS. This is evident in Table 6.2. Therefore, a fixed amount of precipitation can bring different amounts of sulfate flux in the snowfall. Also, in the northern regions of the GIS, dry deposition appears more dominant than wet deposition, resulting in a greater amount of sulfate deposition in the region. The estimated slopes for the random effect An for different ice cores are shown in Table 6.6. It is evident that the slope of the random effect An becomes smaller as An increases. For example, the slope for the Tunu site (0.0041) is the only positive slope, suggesting that for a fixed amount of precipitation, a greater amount of EXS is deposited in the snow when compared to that for the overall GIS average. This implies that the dry deposition process likely provides a significant contribution to the total EXS deposited at the Tunu site.
141 Site Tunu Site A Site T D3 D2 Slope 0.0041 -0.0004 -0.0006 -0.0015 -0.0016
Table 6.6 Parameter estimates for random effect in An
The random effect and the fixed effect in An are then combined to produce a synthesized slope (Fig. 6.9). It is clear that the slopes group naturally by the regional An.
For example, the Tunu site (red) stands out as it has a steep slope compared with the rest ice core sites. Site A (blue) and Site T (green) share similar slopes as they have similar An levels. Likewise, D3 (purple) and D2 (black) are similar and have the lowest slopes as they are from the highest An region. If we assume that the non-volcanic EXS deposited in high An regions is primarily via wet deposition processes, then the difference between its slope and that of other regions must reflect the effect by dry deposition, if other factors are held constant. However, the presence of other local sulfate sources may add complications by increasing the local sulfate burden, therefore resulting in a higher
(steeper) slope of An vs. NV-EXS.
142
Figure 6.9 Slopes for the mixed effect in An
As suggested by the p-value associated with the Year component in the model, there is a significant time effect within the 99.9% confidence level. Holding other components constant, there is a 0.30% increase in ln(EXS) every year, or a 0.30% increase in the annual NV-EXS every year. A 95% confidence interval for this annual rate of increase is (1.0017, 1.0042). Therefore, with 95% confidence, we conclude that there will be a 0.17% to 0.42% increase in the annual non-volcanic sulfate flux to the GIS every year, holding sulfur emission levels and accumulation constant. It is quite likely that this time effect primarily results from the other factors with time dependency and the inter-annual correlation in the flux, considering the relatively constant sulfate background before 1850. However, the cause of this time effect is uncertain. For example, the changes in other natural sulfur sources, such as DMS, may also affect the NV-EXS
143 deposited over the GIS. However, a steady decrease of methanesulfonic acid has been observed in Greenland ice cores since 1945, which may reflect decreasing DMS emissions from marine biota at high northern latitudes [Legrand, 1997b]. Thus, DMS is unlikely to result in an increase of EXS with time. Other possible causes include fluctuations in other sulfate sources, the stochastic nature of deposition and post-deposition processes in the GIS. Thus, reconstruction of the histories of other components of the NV-EXS, such as contributions from marine and continental biota, and the examination of the deposition processes may provide valuable information in understanding the relationship between atmospheric sulfate background and the EXS flux over the GIS.
144 CHAPTER 7
CONCLUSIONS AND FUTURE WORK
Greenland is Earth’s second largest ice sheet and the largest in the Northern
Hemisphere and preserves an exceptionally valuable climate history for reconstructing paleoclimatic and paleoenvironmental conditions. The chemical analyses of five multi-century ice cores from the PARCA and Summit collections have yielded a high resolution volcanic aerosol history, which is an important complement to the volcanic histories extracted from other Greenland ice cores. This detailed ice-core volcanic index
(IVI) provides an improved estimate of the stratospheric sulfate burden and serves as an important input for models assessing the climatic impacts of volcanic eruptions. When compared with other volcanic indices, the IVI from our PARCA and Summit ice cores provides a more spatially representative record of the major volcanic events. When compared to climate histories extracted from other proxy indicators such as tree rings, the
IVI provides a reasonably robust assessment of the associated climatic impacts. However, as the IVI is strongly influenced by the eruption location and the climate system is strongly affected by explosive eruptions in low latitudes, a composite IVI derived from
EXS histories from both Greenland and Antarctic ice cores is desirable. Unfortunately, the number of Antarctic ice cores comparably analyzed by ion chromatography for sulfate concentrations with annual resolution is limited.
145 In these volcanic records reconstructed from Greenland ice cores, the largest excess sulfate (EXS) signal in the past 250 years is from the 1783-84 A.D. Laki eruption
(Iceland), which ejected more than ~200 Mt of H2SO4 aerosols into the atmosphere
[Thordarson and Self, 2003]. The second largest EXS event occurs in 1816 and arises from the 1815 A.D. Tambora eruption (Indonesia). These two known volcanic events have been used for decades to constrain the time scales for Greenland ice cores [Hammer et al., 1981; Clausen and Hammer, 1988; Laj et al., 1990]. However, whether the EXS from Laki arrived over Greenland in 1783 or 1784 has remained a point of discussion since it was first proposed in 1994. Two PARCA ice cores from western Greenland, analyzed with high temporal resolution, confirm that the sulfate aerosols from Laki arrived over Greenland late in 1783, concomitant with the tephra, elevated concentrations of Cd, Bi, and Tl, all indicators of volcanic emissions, and with a short-lived Rare Earth
Elements anomaly [Wei et al., 2008]. Thereafter sulfate deposition declined rapidly such that only very modest concentrations of sulfate arrived in the 1784 snowfall. Evidence from six Greenland ice cores suggests a relatively short (less than 1 year) atmospheric residence time and an injection height limited to the lower stratosphere. An improved estimate of the associated stratospheric sulfate burden has been calculated from these new analyses and provides an important input for models assessing the climatic impacts of the
1783-84 A.D. Laki eruption [Wei et al., 2008].
The volcanic history preserved in the GIS contains signatures from large explosive eruptions in the low latitudes as well as from eruptions of various magnitudes located in the higher latitudes of the NH. Examination of the volcanic signature associated with a single eruption in numerous ice cores across Greenland reveals that the
146 concentrations vary significantly over the GIS. Thus a closer examination of the spatial characteristics of the sulfate aerosol deposition associated with specific eruptions may provide additional information about the dominant depositional processes as well as about transport processes and pathways. For example, comparing the concentration of
EXS originating from the same eruption and deposited in different regions across the GIS may help to identify the primary deposition mechanisms. Using the two most pronounced volcanic events in our derived IVI, we examined the spatial character of the EXS from
Laki and Tambora. As expected, different spatial patterns were observed for these two eruptions, one located close by and one located in the tropics. The 1783 Laki EXS shows high flux to areas in the south and southeast, with a decreasing trend toward the northwest as distance from the eruption source increases. This spatial pattern is inconsistent with that of the ice core derived-net annual accumulation (An) in 1783, suggesting that precipitation over the southeastern coastal areas may have been suppressed due to the geographical proximity to Iceland where atmospheric temperatures were reportedly suppressed after the eruption. In contrast, the 1816 EXS also shows a higher flux in the southern region, and a more homogenous depositional pattern from east to west across central Greenland. The spatial patterns of the Tambora EXS and An are much more similar (than those for Laki) and this high correspondence strongly suggests that wet deposition is the key mechanism.
To better quantify the impact of various geographical factors including the distance of the eruption source on the deposition of EXS over the GIS, a category explanatory variable analysis was conducted on the volcanic EXS derived from multiple cores as described in Section 5.3. The results suggest that the location of ice cores relative
147 to north/south or east/west side of the ice divide strongly affects EXS deposition, while the elevation of the core site is relatively unimportant. The volcanic EXS on the western side is nearly twice that on the eastern side and ice cores in the south contain
~1.8 times more EXS than that sites in the north. These spatial patterns likely reflect the higher accumulation rates on the western and southern sides of the GIS. This analysis also reveals that the latitude of the eruption also strongly affects the EXS flux to the GIS, with the sulfate burden from higher latitude eruptions nearly double that from tropical eruptions.
The EXS flux extracted from Greenland ice cores shows an increasing trend since
1850, primarily as a result of anthropogenic sulfur emissions. High positive correlations are found between the NV-EXS fluxes and the NH anthropogenic sulfur emissions. For example, the correlation between the stacked EXS record from five ice cores and sulfur emissions from Western Europe and North American are high (r2= 0.92 and 0.90, respectively), but somewhat lower for Eastern Europe and Asia (r2= 0.85 and 0.80, respectively). Taking a more regional view, the ice cores from the eastern side of the
GIS (Tunu, Site A, and Site T) are most strongly correlated (r2= 0.91) with emissions in
Western Europe; while cores from the western side (D2 and D3) are most strongly correlated (r2= 0.83) with emissions from North America. These results imply that the atmospheric sulfur burden over the GIS is largely affected by the emissions from the surrounding continents - Western Europe and North America, although emissions from
Asia and Eastern Europe are also likely to arrive over Greenland as well. To quantify the impact of anthropogenic sulfur emissions on the NV-EXS, a linear mixed model was
148 applied. The model results indicate that for every 1 Gg increase in the annual NH sulfur emissions, there is 0.0013% increase in the annual NV-EXS.
On the regional scale, the NV-EXS flux deposited on the GIS is also expected to be affected by the local accumulation rate, as implied by the linear relationship suggested between An and background EXS flux [Mosley-Thompson et al., 2003]. On average, every 1 mm (w.e.) increase in annual accumulation would result in a 0.31% increase in the annual NV-EXS flux. However, the impact of An on the deposition of NV-EXS flux varies over the GIS, likely as a function of the differences in the dominance of local depositional mechanisms. From the results of the linear mixed model, it is clear that the linear slopes of accumulation versus NV-EXS group naturally by the regional An. For example, the Tunu site stands out from the rest of the cores as it has a very steep slope.
Sites A and T exhibit similar slopes as they have similar An levels. Likewise, cores D3 and D2 have the lowest slopes as they are from a higher An region. If we assumed that the
NV-EXS in high An regions is deposited primarily via wet deposition, then the differences among the slopes likely reflect the strength of the role of dry deposition in the other regions. However, the presence of sulfate from other local sources may complicate this interpretation, and therefore result in a higher slope of An versus NV-EXS.
A significant time effect on the deposition of NV-EXS is also detected. From the model, it is estimated that there is 0.30% increase in the annual NV-EXS every year, holding other factors constant. It is quite likely that this time effect is primarily due to unknown or unspecified factors associated with the time variable (such as factors with trends or periodic changes). In addition to the effects of accumulation, time related factors, and sulfur emissions, local sources as well as the stochastic nature of deposition
149 and post-deposition processes may also impact the sulfate flux deposition on the GIS.
Thus, it would be valuable to reconstruct ice core histories of other components of the
NV-EXS flux. For example, methanesulfonic acid and dimethyl sulfide are associated with marine biological activity, and levoglucosan is a biomarker for biomass burning. In addition, depositional processes should be examined more closely. This would provide important information needed to better quantify the relationship between the natural background concentrations of sulfate and the NV-EXS flux over GIS so that the ice core-derived proxy record can be better understood and more effectively utilized.
150 BIBLIOGRAPHY
Aiuppa, A., G. Dongarrà, M. Valenza, C. Federico, and G. Pecoraino, Degassing of trace volatile metals during the 2001 Eruption of Etna, in Volcanism and the Earth’s Atmosphere, Geophysical Monograph Series, vol. 139, edited by A. Robock and C. Oppenheimer, pp. 41-54, doi: 10.1029/139GM03, 2003.
Akaike, H., A new look at the statistical model identification, IEEE Transactions on Automatic Control, 19 (6), 716–723, 1974.
Alley, R. B., D. A. Meese, C. A. Shuman, A. J. Gow, K. C. Taylor, P. M. Grootes, J. W. C. White, M. Ram, E. D. Waddlington, P. A. Mayewski, and G. A. Zielinski, Abrupt increase in Greenland snow accumulation at the end of the Younger Dryas event, Nature, 362, 527-529, 1993.
Alley, R. B., A. J. Gow, S. J. Johnsen, J. Klpfstuhl, D. A. Meese, and T. Thorsteinsson, Comparison of deep ice cores, Nature, 373, 393-393, 1995.
Andreae, M. O., R. J. Charlson, F. Bruynseels, H. Storms, R. V. Grieken, and W. Maenhaut, Internal mixture of sea salt, silicates, and excess sulfate in marine aerosols, Science, 232, 1620-1623, 1986.
Andreae, M.O., and P. J. Crutzen, Atmospheric aerosols: biogeochemical sources and role in atmospheric chemistry, Science, 276, 1052-1058, 1997.
Anklin, M., J. Schwander, B. Stauffer, J. Tschumi, and A. Fuchs, CO2 record between 40 and 8 kyr B.P. from the Greenland Ice Core Project ice core, Journal of Geophysical Research, 102, 26,539-26,545, 1997.
ASL and Associates, Sulfur emissions by country and year, Report No: DE96014790, US Department of Energy, Washington, DC, 1997.
Bales, R. C., J. R. McConnell, E. Mosley-Thompson, and G. Lamorey, Accumulation map for the Greenland Ice sheet: 1971-1990, Geophysical Research Letters, 28, 2967-2970, 2001a.
151
Bales, R. C., J. R. McConnell, E. Mosley-Thompson, and B. Csatho, Accumulation over the Greenland ice sheet from historical and recent records, Journal of Geophysical Research, 106, 33,813-33,825, 2001b.
Bales, R. C., E. Mosley-Thompson, and J. R. McConnell, Variability of accumulation in northwest Greenland over the past 250 years, Geophysical Research Letters, 28, 2679-2682, 2001c.
Barlow L. K., J. C. Rogers, M. C. Serreze, and R. G. Barry, Aspects of climate variability in the North Atlantic sector: discussion and relation to the Greenland ice sheet project 2 high-resolution isotopic signal, Journal of Geophysical Research, 102 (C12), 26,333-26,344, 1997.
Barry, R. G., and G. N. Kiladis, Climatic characteristics of Greenalnd, in Climatic and physical characteristics of the Greenland Ice Sheet, edited by Radok, U., CIRES, Boulder, CO, USA, 7-33, 1982.
Bekki, S., Oxidation of volcanic SO2: a sink for stratospheric OH and H2O, Geophysical Research Letters, 22 (8), 913-916, 1995.
Bender, M., T. Sowers, M. Dickson, J. Orchardo, P. Grootes, P. A. Mayewski, and D. A. Meese, Climate correlations between Greenland and Antarctica during the past 100,000 years, Nature, 372, 663-666, 1994.
Blunier, T., and E. J. Brook, Timing of millennial-scale climate change in Antarctica and Greenland during the last glacial period, Science, 291, 109-112, 2001.
Bolzan, J. F., and M. Strobel, Accumulation-rate variations around Summit, Greenland, Journal of Glaciolology, 40 (134), 56-66, 1994.
Box, G., G. M. Jenkins, and G. C. Reinsel, Time series Analysis: Forecasting and Control, 3rd Edition, Prentice-Hall, 1994.
Box, J.E. and K. Steffen, Sublimation estimates for the Greenland ice sheet using automated weather station observations, Journal of Geophysical Research, 106 (D24), 33,965-33,982, 2001.
Box, J., Survey of Greenland instrumental temperature records: 1873-2001, International Journal of climatology, 22, 1829-1847, 2002.
Box, J.E., D. H. Bromwich, L-S Bai, Greenland ice sheet surface mass balance for 1991-2000: application of Polar MM5 mesoscale model and in-situ data, Journal of Geophysical Research, 109 (D16), doi:10.1029/2003JD004451, 2004.
152
Briffa, K. R., P. D. Jones, J. R. Pilcher, and M. K. Hughes, Reconstructing summer temperatures in northern Fennoscandinvia back to A.D. 1700 using tree-ring data from scots pine, Arctic and Alpine Research, 20, 385-394, 1988.
Briffa, K. R., P. D. Jones, T. S. Bartholin, D. Eckstein, F. H. Schweingruber, W. Karlén, P. Zetterberg, and M. Eronen, Fennoscandian summers from AD 500: Temperature changes on short and long timescales, Climate Dynamics, 7, 111-119, 1992a.
Briffa, K. R., P. D. Jones, and F. H. Schweingruber, Tree-ring density reconstructions of summer temperature patterns across western North America since 1600, Journal of Climatology, 5, 735-752, 1992b.
Briffa, K. R., P. D. Jones, F. H. Schweingruber, S. G. Shiyatov, and E. R. Cook, Unusual twentieth-century summer warmth in a 1,000-year temperature record from Siberia, Nature, 376, 156-159, 1995.
Brockwell, P. J. and R. A. Davis, Estimation and elimination of trend and seasonal components, in Introduction to time series and forecasting, Springer, Harrisonburg, VA, 2002.
Bromwich, D. H., L.-S. Bai, and G. G. Bjarnason, High resolution regional climate simulations over Iceland using Polar MM5, Monthly Weather Review, 133, 3527-3547, 2005. Calder, C.A. and Cressie, N. Kriging and Variogram Models, in: the International Encyclopedia of Human Geography. In press, 2008.
Calder, C. A., P. F. Craigmile, and E. Mosley-Thompson, Spatial variation in the influence of the North Atlantic Oscillation on precipitation across Greenland, Journal of Geophysical Research, 113, doi:10.1029/2007JD009227, 2008.
Cappelen, J, E. V. Laursen, P.V. Jørgensen, C. Kern-Hansen, DMI monthly climate data collection 1768-2005, Denmark, the Faroe Islands and Greenland, Danish Meteorological Institute, Technical Report 06-09, 53, 2006.
18 Chappellaz, J., E. Brook, T. Blunier, and B. Malaizé, CH4 and O of O2 records from Antarctic and Greenland ice: a clue for stratigraphic disturbance in the bottom part of the Greenland Ice Core Project and the Greenland Ice Sheet Project 2 ice cores, Journal of Geophysical Research, 102, 26,547-26,557, 1997.
Charlson, R. J., S. E. Schwartz, J. M. Hales, R. D. Cess, J. A. Coakley, J. E. Hansen, and D. J. Hofmann, Climate forcing by anthropogenic aerosols, Science, 255, 423-430, 1992.
153 Chen, Q.-S., D. H. Browmwich, and L. Bai, Precipitation over Greenland retrieved by a dynamic method and its relation to cyclonic activity, Journal of Climate, 10, 839-870, 1997.
Church, J. A. and J. M. Gregory, Changes in sea level, In Climate Change 2001: The Scientific Basis. Contribution of Working Group I to the Third Assessment Report of the Intergovernmental Panel on Climate Change, edited by Houghton, J. T. et al. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, pp. 641-693, 2001.
Clausen, H. B., and C. U. Hammer, The Laki and Tambora eruptions as revealed in Greenland ice cores from 11 locations, Annals of Glaciology, 10, 16-22, 1988.
Clausen, H. B., C. U. Hammer, C. S. Hvidberg, D. Dahl-Hensen, and J. P. Steffensen, A comparison of the volcanic records over the past 4000 years from the Greenland Ice Core Project and Dye 3 Greenland ice cores, Journal of Geophysical Research, 102, 26707-26723.1997.
Cleveland, W. S., and S. J. Devlin, Locally weighted regression: an approach to regression analysis by local fitting, Journal of the American Statistical Association, 83 (403), 596-610, 1988.
Cole-Dai, J., E. Molsey-Thompson, and L.G. Thompson, Annually resolved volcanic history from two Antarctic ice cores, Journal of Geophysical Research, 102, 16761-16771, 1997.
Cole-Dai, J., and E. Mosley-Thompson, The Pinatubo eruption in South Pole snow and its potential value to ice-core paleovolcanic records, Annals of Glaciology, 29, 1, 99-105, 1999.
Cole-Dai, J., E. Mosley-Thompson, S. P. Wight, and L. G. Thompson, A 4100-year record of explosive volcanism from an East Antarctica ice core, Journal of Geophysical Resesearch, 105 (D14), 24,431-24,441, 2000.
Cressie N., Effect of variogram parameters on kriging, in Statistics for spatial data, Wiley Interscience, pp127-134, 1993a.
Cressie N., Variogram, in Statistics for spatial data, Wiley Interscience, pp58-60, 1993b.
Cressie N., Lognormal and Trans-Gaussian, Kriging, in Statistics for spatial data, Wiley Interscience, pp135, 1993c.
Crowley, T.J., Causes of climate change over the past 1000 Years, Science, 289, 270-277, 2000.
154
Crozaz, G., C. C. Langway, and E. Picciotto, Aritifical radioactivity reference horizons in Greenland firn, Earth and Planetary Science Letters, 1, 42-48, 1966.
Cuffey, K. M., G. D. Clow, R. B. Alley, M. Stulver, E. D. Waddington, and R. W. Saltus, Large Arctic temperature change at the Wisconsin-Holocene glacial transition. Science, 270, 455-458, 1995.
Cuffey, M. K., and S. J. Marshall, Substantial contribution to sea-level rise during the last interglacial from the Greenland Ice Sheet, Nature, 404, 591-594, 2000.
Dansgaard, W. and S. J. Johnsen, A flow model and a time scale for the ice core from Camp Century, Journal of Glaciology, 8, 215-223, 1969.
Dansgaard W., S. J. Johnsen, H. B. Clausen, and N. Gunderstrup, Stable isotope glaciology, Meddelelser om Grønland, 197 (2), 1973.
Dansgaard, W., H. B. Clausen, N. Gunderstrup, C. U. Hammer, S. F. Johnsen, P. M. Kristinsdottir, and N. Reeh, A new Greenland deep ice core, Science, 218, 1273-1277, 1982.
Dansgaard, W., S. J. Johnsen, H. B. Clausen, D. Dahl-Jensen, N. S. Gundestrup, C. U. Hammer, C. S. Hvidberg, J. P. Steffensen, A. E. Sveinbjörnsdottir, J. Jouzel, and G. Bond, Evidence for general instability of past climate from a 250-kyr ice-core record, Nature, 364, 218-220, 1993.
Ferrari, C. P., S. Hong, K. Van de Velde, C. F. Boutron, S. N. Rudniev, M. Bolshov, W. Chisholm, and K. J. R. Rosman, Natural and anthropogenic bismuth in Central Greenland, Atmos. Environ., 34, 941-948, 2000.
Fiacco, R. J., T. Thordarson, M. S. Germani, S. Self, J. M. Palais, S. Whitlow, and P. M. Grootes, Atmospheric aerosol loading and transport due to the 1783-84 Laki eruption in Iceland, interpreted from ash particles and acidity in the GISP2 ice core, Quaternary Research, 42, 231-240, 1994.
Finnsson, H., Um mannfœkkun í Hallœrum (Decimation of the population in Iceland due to famines), in Rit þess konunglega islenzka lœrdόmslistafèlags, XIV, 30-226, 1796.
Flückiger, J., T. Blunier, B. Stauffer, J. Chapellaz, R. Spahni, K. Kawamura, J. Schwander, T. F. Stocker, and D. Dahl-Jensen, N2O and CH4 variations during the last glacial epoch: Insight into global processes, Global Biogeochemical Cycles, 18, doi:10.1029/2003GB002122, 2004.
155 Forster, P., and V. Ramaswamy, Changes in atmospheric constituents and in radiative forcing, In Working Group I Report “The Physical Science Basis”, the Forth Assessment Report of the Intergovernmental Panel on Climate Change, edited by Solomon, S. et al. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, pp. 289–348, 2001.
Forsyth, P. Y., In the wake of Etna, 44 B.C., Classical Antiquity, 7, 49-57, 1988.
Freydier R., B. Dupré, and J. P. Lacaux, Precipitation chemistry, in intertropical Africa, Atmospheric Environment, 32 (4), 749-765, 1998.
Funder, S., Quaternary geology of the ice-free areas and adjacent shelves of Greenland, in The Geology of North America, Geological Society of America, k-1, edited by Fulton R. J., 741-792, 1989.
Gabrielli, P., C. Barbante, C. Turetta, A. Marteel, C.F. Boutron, G. Cozzi, W. Cairns, C. Ferrari, and P. Cescon, Direct determination of Rare Earth Elements at the subpicogram per gram level in antarctic ice by ICP-SFMS using a desolvation system, Analytical Chemistry, 78, 1883-1889, 2006.
Gao, C., A. Robock, S. Self, J. B. Witter, J. P. Steffenson, H. B. Calusen, M.-L. Siggaard-Andersen, and S. Johnsen, The 1452 or 1453 A. D. Kuwae eruption signal derived from multiple ice core records: Greatest volcanic sulfate event of the past 700 years, Journal of Geophysical Research, 111, D12107, doi:10.1029/2005JD006710, 2006.
Gaspari, V., C. Barbante, G. Cozzi, P. Cescon, C.F. Boutron, P. Gabrielli, G. Capodaglio, C. Ferrari, J.R. Petit, and B. Delmonte, Atmospheric iron over the last deglaciation: Climatic implications, Geophysical Research Letters, 33, doi:10.1029/2005GL024352, 2006.
Georgii, H. W., and P. Warneck, Chemistry of the tropospheric aerosol and of clouds. In: Global aspects of atmospheric chemistry, edited by R. Zellner, Springer, Darmstadt, 111-180, 1999.
Grattan, J., R. Rabartin, S. Self , and T. Thordarson, Volcanic air pollution and mortality in France 1783-1784, Comptes Rendus Geosciences, 337, 641-651, 2005.
Greenland Ice-core Project (GRIP) Members, Climate instability during the last interglacial period recorded in the GRIP ice core, Nature, 364, 203-207, 1993.
Grootes, P. M., M. Stulvar, J. W. C. Whits, S. johnsen, and J. Jouzel, Comparison of oxygen isotope records from the GISP2 and GRIP Greenland ice cores, Nature, 366, 552-554, 1993.
156
Hammer, C. U., Past volcanism revealed by Greenland ice sheet impurities, Nature, 270, 482-486, 1977.
Hammer, C. U., H. B. Clausen, W. Dansgaard, N. Gunderstrup, S. J. Johnsen, and N. Reeh, Dating of Greenland ice cores by flow models, isotopes, volcanic debris and continental dust, Journal of Glaciology, 20, 3-26, 1978.
Hammer, C. U. Acidity of polar ice cores in relation to absolute dating, past volcanism and radio-echoes, Journal of Glaciology, 25 (93), 359-372, 1980.
Hammer, C. U., H. B. Clausen, and W. Dansgaard, Past volcanism and climate revealed by Greenland ice cores, Journal of Volcanology and Geothermal Research, 11, 3-10, 1981.
Hanna, E. and J. Cappelen, Recent cooling in coastal southern Greenland and relation with the North Atlantic Oscillation, Geophysical Research Letters, 30, 1132-1134, 2003.
Hansen, J., A. Lacis, R. Ruedy, and M. Sato, Potential climate impact of Mount Pinatubo eruption, Geophysical Research Letters, 19 (2), 215-218, 1992.
Hansen, J. E., M. Sato, A. Lacis, R. Ruedy, I. Tegen, and E. Matthews, Carbonaceous aerosols in the industrial era, Eos Trans American Geophysical Union, 85, 241-245, 2004.
Hansen, J., L. Nazarenko, R. Ruedy, M. Sato, J. Willis, A. D. Genio, D. Koch, A. Lacis, K. Lo, S. Menon, T. Novakov, J. Perlwitz, G. Russell, G. A. Schmidt, and N. Tausnev, Earth’s energy imbalance: confirmation and implications, Science 308, 1431-1435, 2005.
Harshvardhan, Perturbation of the zonal radiation balance by a stratospheric aerosol layer, Journal of Atmospheric Sciences, 36, 1274-1285, 1979.
Haywood, J., and O. Boucher, Estimates of the direct and indirect radiative forcing due to tropospheric aerosols: A review, Review of Geophysics, 38, 513-543, 2000.
Hofmann, D. J., and S. Solomon, Ozon destruction through heterogeneous chemistry following the eruption of El Chichon, Journal of Geophysical Research, 94, 5029-5041, 1989.
Hong, S., and J-P Candelone, C. F. Boutron, Deposition of atmospheric heavy metals to the Greenland ice sheet from 1783-1784 volcanic eruption of Laki, Iceland, Earth and Planetary Science Letters, 144, 605-610, 1996.
157
Hurrell, J. W., Decadal trends in the North Atlantic Oscillation: Regional temperatures and precipitation, Science, 269, 676-679, 1995.
Huybrechts, P., I. Janssens, C. Poncin, and T. Fichefet, The response of the Greenland Ice Sheet to climate changes in the 21st century by interactive coupling of an ANGCM with a thermomechanical ice sheet model, Annals of Glaciology, 35, 409-415, 2002.
IPCC Fourth assessment report, Climate change 2007: Synthesis report, Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, pp. 46–47, 2007.
Johnsen, S. J., W. Dansgaard, H. B. Clausen, and C. C. Langway, Oxygen isotope profiles through the Antarctic and Greenland ice sheets, Nature, 235, 429-434, 1972.
Johnsen, S. J., Stable isotope homogenization of polar firn and ice, in Isotopes and Impurities in Snow and Ice, pp. 210-219, IAHS-AISH Publication, Grenoble, 1977.
Johnsen, S. J., W. D. Dansgaard, and J. W. C. White, The origin of Arctic precipitation under present and glacial conditions, Tellus, Series B, 41 (4), 452-468, 1989.
Johnsen, S. J. and W. Dansgaard, On flow model dating of stable isotope records from Greenland ice cores, in The last deglaciation: absolute and radiocarbon chronologies, edited by E. Bard and W. Broecker, Springer-Verlag, Berlin, 13-24, 1992.
Johnsen, S. J. D. Dahl-Jensen, W. Dansgaard, and N. Gunderstrup, Greenland paleotemperatures derived from GRIP bore hole temperature and ice core isotope profiles, Tellus, 47B, 624-629, 1995.
Jones, P. D., M. New, D. E. parker, S. Martin, and I. G. Rigor, Surface air temperature and its changes over the past 150 years, Reviews of Geophysics, 37, 173-200, 1999.
Jones, P. D., and M. E. Mann, Climate over past millennia, Reviews of Geophysics, 42, doi:10.1029/2003RG000143, 2004.
Jonsell, U., M. E. Hansson, C. Morth, and P. Torssander, Sulfur isotopic signals in two shallow ice cores from Dronning Maud Land, Antarctica, Tellus 57B, 341-350, 2005.
Jouzel, J., C. Lorius, L. Merlivat, and J. R. Petit, Abrupt climatic changes: the Antarctic ice core record during the late Pleistocene, in Abrupt climatic change: evidence and implication, edited by W. H. Berger and L. D. Labeyrie, Dordrecht: D. Reidel, 235-245, 1987.
158 Jouzel, J., V. Masson-Delmotte, O. Cattani, G. Dreyfus, S. Falourd, G. Hoffmann, B. Minster, J. Nouet, J. M. Barnola, J. Chappellaz, H. Fischer, J. C. Gallet, S. Johnsen, M. Leuenberger, L. Loulergue, D. Luethi, H. Oerter, F. Parrenin, G. Raisbeck, D. Raynaud, A. Schilt, J. Schwander, E. Selmo, R. Souchez, R. Spahni, B. Stauffer, J. P. Steffensen, B. Stenni, T. F. Stocker, J. L. Tison, M. Werner, and E. W. Wolff, Orbital and millennial Antarctic climate variability over the past 800,000 years, Science, 317, 793-396, 2007.
Koch, D., D. Jacob, I. Tegen, D. Rind, and M. Chin, Tropospheric sulfur simulation and sulfate direct radiative forcing in the Goddard Institute for Space Studies general circulation model, Journal of Geophysical Research, 104 (D19), 23,799-23,822, 1999.
Krabill, W., W. Abdalati, E. Frederick, S. Manizade, C. Martin, J. Sonntag, R. Swift, R. Thomas, W. Wright, and J. Yungel, Greenland ice sheet: high-elevation balance and peripheral thining, Science, 289, 428-430, 2000.
Krachler, M., J. Zheng, D. Fisher, and W. Shotyk, Analytical procedures for improved trace element detection limits in polar ice from Arctic Canada using ICP-SMS, Analytica Chimica Acta, 530, 291-298, 2005.
Lamb, H. H., Volcanic dust in the atmosphere; with a chronology and assessment of its meteorological significance, Philosophical Transactions of the Royal Society London, Series A, 266, 425-533, 1970.
Lamb, H. H., Supplementary volcanic dust veil index assessments, Climate Monitoring, 6, 57-67, 1977.
Lamb, H. H., Update of the chronology of assessments of the volcanic dust veil index, Climate Monitoring, 12, 79-90, 1983.
Langway, C. C., and K. Goto-Azuma, Temporal variations in the deep ice core chemistry record from Dye 3, Greenland, Annals of Glaciology, 10, 209, 1988.
Lefohn, A. S., J. D. Husar, R. B. Husar, Estimating historical anthropogenic global sulfur emission patterns for the period 1850-1990, Atmospheric Environment 33, 3435-3444, 1999.
Legrand, M., Sulphur-derived species in polar ice: a review. In: Ice Core Studies of Global Biogeochemical Cycles, edited by R. J. Delmas. Springer, New York, 91-115, 1995.
Legrand, M, Ice-core records of atmospheric sulphur, Philosophical Transactions of the Royal Society, Ser. B, 352, 241-250, 1997a.
159
Legrand, M., C. Hammer, M. De Angelis, J. Savarino, R. Delmas, H. Clausen, and S. J. Johnsen, Sulfur-containing species (methanesulfonate and SO4) over the last climatic cycle in the Greenland Ice Core Project (central Greenland) ice core. Journal of Geophysical Research, 102, 26,664-26,679, 1997b.
Li, L., Greenland’s influence on cyclone activity, Ph.D thesis, The Ohio State University, 147, 2003.
Lorius, C., L. Merlivat, J. Jouzel, and M. Pourchet, A 30,000 yr isotope climatic record from Antarctic ice, Nature, 280, 644-648, 1979.
Loulergue, L., F. Parrenin, T. Blunier, J.-M. Barnola, R. Spahni, A. Schilt, G. Raisbeck, and J. Chappellaz, New constraints on the gas age-ice age difference along the EPICA ice cores, 0-50 kyr, Climate of the Past, 3, 527-540, 2007.
Loulergue, L., A. Schilt, R. Spahni, V. Masson-Delmotte, T. Blunier, B. Lemieux, J.-M. Barnola, D. Raynaud, T. F. Stocker, and J. Chappellaz, Orbital and millennial-scale features of atmospheric CH4 over the past 800,000 years, Nature, 453, doi:10.1038/nature06950, 383-386, 2008.
Lüthi, D., M. Le Floch, B. Bereiter, T. Bluiner, J.-M. Barnola, U. Siegenthaler, D. Raynaud, J. Jouzel, H. Fischer, K. Kawamura, and T. F. Stocker, High-resolution carbon dioxide concentration record 650,000-800,000 years before present, Nature, 453, doi:10.1038/nature06949, 379-382, 2008.
Mann, M. E., M. A. Cane, S. E. Zebiak, and A. Clement, Volcanic and solar forcing of the tropical Pacific over the past 1000 years, Journal of Climate, 18, 447-456, 2005.
Matheron, G., Principles of geostatistics, Economic Geology, 58, 1,246-1,266, 1963.
Matsumoto, A. and T. K. Hinkley, Trace metal suites in Antarctic pre-industrial ice are consistent with emissions from quiescent degassing of volcanoes worldwide, Earth and Planetary Science Letters, 186, 33-43, 2001.
Mayewski, P. A., P. C. Buckland, K. J. Edwards, L. D. Meeker, and S. O’ Brien, Climate change events as seen in the Greenland ice core (GISP2): implications for the Mesolithic of Scotland, The Early Prehistory of Scotland, Edinburgh University Press, Scotland, 74-83, 1996.
McConnell, J. R., R. J. Arthern, E. Mosley-Thompson, C. H. Davis, R. C. Bales, R. Thomas, J. F. Burkhart, and J. D. Kyne, Changes in Greenland ice sheet elevations attributed primarily to snow accumulation variability, Nature, 406, 877-879, 2000a.
160 McConnell, J. R., E. Mosley-Thompson, D. H. Bromwich, R. C. Bales, and J. D. Kyne, Interannual variations of snow accumulation on the Greenland Ice Sheet (1985-1996): New observations versus model predictions, Journal of Geophysical Research, 105, 4039-4046, 2000b.
McConnell, J.R., G. Lamorey, E. Hanna, E. Moslye-Thompson, R. C. Bales, D. Belle-Oudry, and J. D. Kyne, Annual net snow accumulation over southern Greenland from 1975 to 1998, Journal of Geophysical Research, 106, 33,827-33,838, 2001.
McCormick, M. P., and R. E. Veiga, SAGE II measurements of early Pinatubo aerosols, Geophysical Research Letters, 19 (2), 155-158, 1992.
Meese, D. A., A. J. Gow, P. Grootes, P. A. Mayewski, M. Ram, M. Stuiver, K. C. Taylor, E. D. Waddington, and G. A. Zielinski, Determination of GISP2 depth-age scale: techniques and results, CRREL Report No. 95, Hanover: USA Cold Regions Research and Engineering Laboratory, 1995.
Meese, D. A., A. J. Gow, R. B. Alley, G. A. Zielinski, P. M. Grootes, M. Ram, K. C. Taylor, P. A. Mayewski, and J. F. Bolzan, The GISP2 depth-age scale: methods and results, Journal of Geophysical Research, 102C, 26,411-26,423, 1997.
Mitchell, J. M., Jr., A preliminary evaluation of atmospheric pollution as a cause of the global temperature fluctuation of the past century, in Global Effects of Environmental Pollution, edited by S.F. Singer, pp. 139-155, Norwell, Mass., 1970.
Mosley-Thompson, E., L. G. Thompson, J. Dai, M. Davis, and P. N. Lin, Climate of the last 500 years: high resolution ice core record, Quaternary Science Review, 12 (6), 419-430, 1993.
Mosley-Thompson, E., J. R. McConnell, R. C. Bales, Z. Li, P. Lin, K. Steffen, L. G. Thompson, R. Edwards, and D. Bathke, Local to regional-scale variability of annual net accumulation on the Greenland ice sheet from PARCA cores, Journal of Geophysical Research, 106, 33,839-33,851, 2001.
Mosley-Thompson, E., T. A. Mashiotta, and L. G. Thompson, High resolution ice core records of late Holocene volcanism: current and future contributions from the Greenland PARCA cores, Volcanism and the Earth’s Atmosphere, 139, 153-164, 2003.
Mosley-Thompson, E., C. R. Readinger, P. Craigmile, L. G. Thompson, and C. A. Calder, Regional sensitivity of Greenland precipitation to NAO variability, Geophysical Research Letters, 32, doi: 10.1029/2005GL024776, 2005.
161 Mosley-Thompson, E., L.G. Thompson, and P. Lin, A multi-century ice core perspective on 20th cenury climate change with new contributions from high Arctic and Greenland (PARCA) cores, Annals of Glaciology, 43, 42-48, 2006.
Naveau, P., C. Ammann, H.-S. Oh, and W. Guo, An automatic statistical methodology to extract pulse-like forcing factors in climatic time series: Application in volcanic events, in Volcanism and the Earth’s Atmosphere, Geophysical Monograph Series, 139, edited by A. Robock and C. Oppenheimer, pp.177-186, AGU, Washington, D. C., 2003.
Neftel, A., The records of gases and reactive species in ice cores, and problems of interpretation, in Chemical Exchange Between the Atmosphere and Polar Snow, edited by E. W. Wolff and R. C. Bales, pp. 45-69, Springer-Verlag, New York, 1996.
Newhall, C. G. and S. Self, The volcanic explosivity index (VEI): an estimate of explosive magnitude for historical volcanism, Journal of Geophysical Research, 87, 1231-1238, 1982.
North Greenland Ice Core Project members, High-resolution record of Northern Hemisphere climate extending into the last interglacial period, Nature, 431, 147-151, 2004.
Novakov, T., V. Ramanathan, J. E. Hansen, T. W. Kirchstetter, M. Sato, J. E. Sinton, and J. A. Sathaye, Large historical changes of fossil-fuel black carbon aerosols, Geophysical Research Letter, 30, 1324-1327, 2003.
Nozaki, Y., A fresh look at element distribution in the North Pacific Ocean, EOS, Transaction American Geophysical Union, 78 (21), 221, 1997.
Nriagu, J. O., D. A. Holdway, and R. D. Coker, Biogenic sulfur and the acidity of rainfall in remote areas of Canada, Science, 237, 1189-1192, 1987.
Nriagu, J. O., R. D. Coker, and L. A. Barrie, Origin of sulphur in Canadian Arctic haze from isotope measurements, Nature, 349, 142-145, 1991.
Nuttall, M., (Eds), Encyclopedia of the Arctic, Volume 2, 778, Routledge, Abingdon, Oxon, United Kindom, 2005.
Oberbeck, V. R., E. F. Danielsen, K. G. Snetsinger, G. V. Ferry, W. Fong, and D. M. Hayes, Effect of the eruption of El Chichon stratospheric aerosol size and composition, Geophysical Research Letters, 10, 1021-1024, 1983.
Ohmura, A., New Temperature distribution maps for Greenland, Z. Gletscherkunde Glazialgeol, 23, 1-45, 1987.
162
Oman, L., A. Robock, G. L. Stenchikov, T. Thordarson, D. Koch, D. T. Shindell, and C. Gao, Modeling the deistribution of the volcanic aerosol cloud from the 1783-1784 Laki eruption, Journal of Geophysical Research, 111, doi: 10.1029/2005JD006899, 2006.
Oppenheimer, C., Climatic, environmental and human consequences of the largest known historic eruption, Tambora volcano (Indonesia) 1815, Progress in Physical Geography, 27 (2), 230-259, 2003.
Parker, D. E., C. K. Folland, and M. Jackson, Marine surface temperature: observed variations and data requirements, Climate Change, 31, 559-600, 1995.
Paterson, W. S. B., Mass balance, The physics of Glaciers, 3rd Ed., published by Butterworth-Heinernann, Woburn, MA, 26-52, 1998.
Patris, N., R. J. Delmas, M. Legrand, M. D. Angelis, F. A. Ferron, M. Stiévenard, and J. Jouzel, First sulfur isotope measurements in central Greenland ice cores along the preindustrial and industrial periods, Journal of Geophysical Research, 107, 4115-4125, 2002.
Penner, J. E. et al., Aerosols, their direct and indirect effects, In Climate Change 2001: The Scientific Basis. Contribution of Working Group I to the Third Assessment Report of the Intergovernmental Panel on Climate Change, edited by Houghton, J. T. et al. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, pp. 289–348, 2001.
Petit, J. R., I. Basile, A. Leruyuet, D. Raynaud, C. Lorius, J. Jouzel, M. Stievenard, V. Y. Lipenkov, N. I. Barkov, B. B. Kudryashov, M. Davis, E. Saltzman, and V. Kotlyakov, Four climate cycles in Vostok ice core, Nature, 387, 359-360, 1997.
Petit, J. R., J. Jouzel, D. Raynaud, N. I. Barkov, J.-M. Barnola, I. Basile, M. Bender, J. Chappellaz, M. Davis, G. Delaygue, M. Delmotte, V. M. Kotlyakov, M. Legrand, V. Y. Lipenkov, C. Lorius, L. Pépin, C. Ritz, E. Saltzman, and M. Stievenard, Climate and atmospheric history of the past 420,000 years from the Vostok ice core, Antarctica, Nature, 399, 429-436, 1999.
Picciotto, E., G. Crozaz, and W. De Breuck, Accumulation on the South Pole-Queen Maud Land Traverse 1964-1968, in Antarctic snow and ice studies II, edited by A. P. Crary, Washington, DC, Antarctic Research Series, American Geophysical Union, 257-316, 1971.
163 Pollack, J. B., O. B. Toon, C. Sagan, A. Summers, B. Baldwin, and W. Van Camp, Volcanic explosions and climatic change: A theoretical assessment, Journal Geophysical Research, 81, 1071-1083, 1976.
Putnins, P., The climate of Greenland, in World Survey of Climatology, 14, edited by Orvig, S., Elsevier Publishing Company, 3-112, Amsterdam, The Netherlands, 1970.
Ramanathan, V., P. J. Crutzen, J. T. Kiehl, and D. Rosenfeld, Aerosols, climate, and the hydrological cycle, Science, 294, 2119-2124, 2001.
Reeh, N, Dating by ice flow modeling: a useful tool or an exercise in applied mathematics? in the environmental record in glaciers and ice sheet, eited by H. Oeschger and C. C. Langway, Wiley, Chichester: 141-159, 1989.
Robock, A., A latitudinally dependent volcanic dust veil index, and its effect on climate simulations, Journal of Volcanology and Geothermal Research, 11, 67-80, 1981.
Robock, A. and M. P. Free, Ice cores as an index of global volcanism from 1850 to the present, Journal of Geophysical Research, 100, 11549-11567, 1995.
Robock, A., and M. P. Free, The volcanic record in ice cores for the past 3000 years, in Climatic variations and forcing mechanisms of the last 2000 years, edited by P. Jones, R. Bradley, and J. Jouzel, pp533-540, Springer, New York, 1996.
Robock, A., Volcanic eruptions and climate, Reviews of Geophysics, 38, 191-219, 2000.
Rubin, D. B., Multiple imputation for nonresponse in surveys, John Wiley & Sons, New York, 1987.
Sassen, K., D. O’C. Starr, G. G. Mace, M. R. Poellot, S. H. Melfi, W. L. Eberhard, J. D. Spinhirne, E. W. Eloranta, D. E. Hagen, and J. Hallett, The 5-6 December 1991 FIRE IFO II Jet Stream Cirrus Case Study: Possible Influences of Volcanic Aerosols, Journal of the Atmospheric Science, 52, 97-123, 1995.
Satheesh, S.K., V. Ramanathan, X. Li-Jones, J. M. Lobert, I. A. Podgorny, J. M. Prospero, B. N. Holben, and N. G. Loeb, A model for the natural and anthropogenic aerosols over the tropical Indian Ocean derived from Indian Ocean Experiment data, Journal of Geophysical Research, 104, 27421-27440, 1999.
Sato, M., J. E. Hansen, M. P. McCormick, J. B. Pollack, Stratospheric aerosol optical depths, 1850-1990, Journal of Geophysical Research, 98, 22987-22994, 1993.
Schwarz, G., Estimating the dimension of a model, Annals of Statistics, 6 (2), 461-464, 1978.
164
Schweingruber, F. H., K. R. Briffa, and P. Nogler, A tree-ring densitometric transect from Alaska to Labrador, International Journal of Biometeorology, 37, 151-169, 1993.
Self, S., and M. R. Rampino, The 1883 eruption of Krakatau, Nature, 294, 699-704, 1981.
Serreze, M. C., Climatological aspects of cyclone development and decay in the Arctic, Atmosphere-Ocean, 33, 1-23, 1995.
Serreze, M. C., F. Carse, R. G. Barry, and J. C. Rogers, Icelandic Low cyclone activity: climatological features, linkages with the NAO, and relationships with recent changes in the Northern Hemisphere circulation, Journal of Climate, 10, 453-464, 1997.
Shackleton, N. J., J. Backman, H. Zimmerman, D. V. Kent, M. A. Hall, D. G. Roberts, D. Schnitker, J. G. Baldauf, A. Despairies, R. Homrighausner, P. Huddleston, J. B. Keen, A. J. Kaltenbach, K. A. O. Krumsiek, A. C. Morton, J. W. Murrary, and J. Westberg-Smith, Oxygen isotope calibration of the onset of ice-rafting and history of glaciations in the North Atlantic region, Nature, 307, 620-623, 1984.
Smith, S.J., H., Pitcher, and T. M. L. Wigley, Global and regional anthropogenic sulfur dioxide emissions, Global and Planetary Change, 29, 99-119, 2001.
Souchez, R., The buildup of the ice sheet in central Greenland, Journal of Geophysical Research, 102 (C12), 26,317-26,323, 1997.
Stauffer, B. R., Dating of ice by radioactive isotopes, in The environmental record in glaciers and ice sheet, edited by H. Oeschger and C. C. Langway, Chichester: Wiley, 123-129, 1989.
Steffen, K. and J. Box, Surface climatology of the Greenland ice sheet: Greenland climate network 1995-1999, Journal of Geophysical Research, 106 (D24), 33,951-33,964, 2001.
Steffensen, J. P., K. K. Andersen, M. Bigler, H. B. Clausen, D. Dahl-Jensen, H. Fischer, K. Goto-Azuma, M. Hansson, S. J. Johnsen, J. Jouzel, V. Masson-Delmotte, T. Popp, S. O. Rasmussen, R. Rothlisberger, U. Ruth, B. Stauffer, M.-L. Siggaard-Andersen, A. E. Sveinbjörnsdόttir, A. Svensson, and J. W. C. White, High-resolution Greenland ice core data show abrupt climate change happens in few years, Science, doi:10.1126/science.115707, 2008.
165 Stenchikov, G. L., I. Kirchner, A. Robock, H.-F. Graf, J. C. Antuna, R. G. Grainger, A. Lambert, and L. Thomason, Radiative forcing from the 1991 Mount Pinatubo Volcanic eruption, Journal of Geophysical Research, 103, 13,837-13,857, 1998.
Stern, D.I., Global sulfur emissions from 1850 to 2000, Chemosphere, 58, 163-175, 2005.
Stothers, R. B., The great Tambora eruption of 1815 and its aftermath, Science, 224, 1191-1198, 1984.
Stothers, R. B., Major optical depth perturbations to the stratosphere from volcanic eruptions: Pyrheliometric period, 1881-1960, Journal of Geophysical Research, 101 (D2), 3,901-3,920, 1996a.
Stothers, R. B., The great dry fog of 1783, Climatic Change, 32, 79-89, 1996b.
Svensson, A., S. W. Nielsen, S. Kipfstuhl, S. J. Johnsen, J. P. Steffensen, M. Bigler, U. Ruth, and R. Röthlisberger, Visual stratigraphy of the North Greenland Ice Core Project (NorthGRIP) ice core during the last glacial period, Journal of Geophysical Research, 110, doi:10.1029/2004JD005134, 2005.
Tabazadeh, A., and R. P. Turco, Stratospheric chlorine injection by volcanic eruptions: HCl scavenging and implications for ozone, Science, 260, 1082-1086, 1993.
Thomas, R.H., and PARCA Investigators, Program for Arctic Regional Climate Assessment (PARCA): goals, key findings, and future directions, Journal of Geophysical Research, 106 (D24), 33,691-33,705, 2001a.
Thomas, R., B. Csatho, C. Davis, C. Kim, W. Krabill, S. Manizade, J. McConnell, and J. Sonntag, Mass balance of higher-elevation parts of the Greenland ice sheet, Journal of Geophysical Research, 106 (D24), 33707-33716, 2001b.
Thompson, L. G., W. L. Hamilton, and C. Bull, Climatological implications of microparticle concentrations in the ice core from “Byrd” station, western Antarctica, Journal of Glaciology, 14, 433-444, 1975.
Thompson, L. G., Variations in magnitude and concentrations, size distribution and elemental composition found in Camp Century, Greenland, and Byrd Station, Antarctica, deep ice cores, IAHS-AIS,. 118, 351–364, 1977.
Thordarson, T., and S. Self, Atmospheric and environmental effects of the 1783-1784 Laki eruption: A review and reassessment, Journal Geophysical Research, 108 (D1), 4,011-4,039. doi:10.1029/2001JD002042, 2003.
166 Van der Veen, C. J., and J. Bolzan, Interannual variability in net accumulation on the Greenland ice sheet: Observations and implications for mass balance measurements, Journal Geophysical Research, 104 (D2), 2,009-2,014, 1999.
Vinther, B. M., K. K. Andersen, P. D. Jones, K. R. Briffa, and J. Cappelen, Extending Greenland temperature records into the late eighteenth century, Journal Geophysical Research, 111 (D11), doi: 10.1029/2005JD006810, 2006.
Wedepohl, K. H., The composition of the continental crust, Geochimica et Cosmochimica Acta, 59 (7), 1217-1232, 1995.
Wei, L, E. Mosley-Thompson, P. Gabrielli, L. G. Thompson, C. Barbante, Synchronous deposition of volcanic ash and sulfate aerosols over Greenland in 1783 from the Laki eruption (Iceland), Geophysical Research Letters,35, doi:10.1029/2008GL035117, 2008.
Wolff, E., Location, movement and reaction of impurities in solid ice, in Chemical Exchange Between the Atmosphere and Polar Snow, edited by E. W. Wolff and R. C. Bales, pp. 541-560, Springer-Verlag, New York, 1966.
Yiou, P., K. Fuhrer, L. D. Meeker, J. Jouzel, S. Johnsen, and P. A. Mayewski, Paleoclimate variability inferred from the spectral analysis of Greenland and Antarctic ice-core data, Journal of Geophysical Research, 102, 26,441-26,454, 1997.
Zielinski, G. A., Use of paleo-records in determining variability within the volcanism-climate system, Quaternary Science Reviews, 19, 417-438, 2000.
167