Click Here
JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 114, D06301, doi:10.1029/2008JD010600, 2009

Full Article

Geographic variability of nitrate deposition and preservation over the Greenland Ice Sheet
John F. Burkhart,1,2 Roger C. Bales,1,3 Joseph R. McConnell,4 Manuel A. Hutterli,5 and Markus M. Frey1,5
Received 12 June 2008; revised 12 October 2008; accepted 16 December 2008; published 18 March 2009.
[1] An analysis of 96 snow pit and ice cores distributed over the Greenland ice sheet is
used to determine the main drivers of variability in the preserved records of nitrate concentration. The data set provides samples from spatially distributed locations, allowing us to investigate the effect of snow accumulation rate, temperature, and sublimation on nitrate concentration. The mean ice sheet concentration in the dry snow zone (2000 ! mean annual sea level (masl)) is 132 ng g1, ranging between 47 and 265 ng g1 with a standard deviation of 37 ng g1. Nitrate flux varies between 1.1 and 14.7 mg cm2 a1 with a mean of mg cm2 a1. Large-scale spatial variability exists as a result of accumulation gradients, with concentration 5% greater in the northern plateau, yet flux over the northern plateau is 30% lower than the dry snow zone as a whole. While spatially, flux appears to be more dependent on accumulation, preservation of flux shows increasing dependence on concentration with increasing accumulation. The relationship between concentration and accumulation is nonlinear, showing less dependence in the low-accumulation regions versus high-accumulation regions. Accumulation alone is insufficient to account for the observed variability in nitrate flux in the low-accumulation regions, and evidence supports the need for additional components to a transfer function model for nitrate that includes photochemistry, temperature, and sublimation. Spatial variability across the ice sheet is nonuniform, and changes in nitrate concentration have occurred in some regions at a greater rate than others. While the data support that overall the ice sheet acts as an archive of paleoatmospheric concentration despite the effects of postdepositional processing, one needs to consider spatial variables to properly account for trends and variability in the records. This is tested by evaluating past spatial relationships and yields the result that the significant geographic shifts with respect to reactive N concentrations have occurred over the ice sheet in the past century.
Citation: Burkhart, J. F., R. C. Bales, J. R. McConnell, M. A. Hutterli, and M. M. Frey (2009), Geographic variability of nitrate deposition and preservation over the Greenland Ice Sheet, J. Geophys. Res., 114, D06301, doi:10.1029/2008JD010600.

1. Introduction

[2] Ice core records from the Greenland ice sheet provide some of the highest temporal resolution records of the paleoatmospheric environment. Ice cores can be used both to develop proxies, as in temperature, or to directly sample material presumably deposited in an earlier time. For some
1 School of Engineering, University of California, Merced, California, USA. 2 Department of Atmospheric and Climate Science, Norwegian Institute for Air Research, Kjeller, Norway. 3 Sierra Nevada Research Institute, University of California, Merced, California, USA. 4 Division of Hydrologic Sciences, Desert Research Institute, Reno, Nevada, USA. 5 British Antarctic Survey, Natural Environment Research Council, Cambridge, UK.
Copyright 2009 by the American Geophysical Union. 0148-0227/09/2008JD010600$09.00
compounds, such as lead and dust, these analysis can be straight forward as no postdepositional alteration of the record is anticipated in melt-free regions of the ice sheet. Some compounds, however, are expected to undergo postdepositional processes which may change the recorded (or preserved) concentration from what was originally deposited. Nitrate is one such compound; known to undergo recycling at varying degrees in the snow pack as a function of temperature, accumulation, and other factors [Grannas et al., 2007: Burkhart et al., 2004; Wolff et al., 2007; Wolff and Bales, 1996]. Yet, regardless of the recycling, temporal increases in nitrate deposition in Greenland ice cores are highly correlated with anthropogenic activities [Fischer et al., 1998a; Burkhart et al., 2006; Legrand and Mayewski, 1997; Mayewski et al., 1990]. Furthermore prior research has demonstrated preservation at many of the ice core sites is sufficient so as not to exclude the possibility of temporal analysis of the ice core records in the context of anthropogenic

D06301

1 of 15
BURKHART ET AL.: GREENLAND SPATIAL VARIABILITY OF NITRATE
Table 1. Ice Core and Snow Pit Records of Nitrate and Accumulation for the Period of Used for the Kriging and Geographic Variability Study
Point ID B16 B18 B21 P-6345 P-6642 P-6745 P-6839 P-6841 P-6938 P-6939 P-6941 P-6943 P-6945 P-7145 P-7245 P-7247 P-7249c P-7345 P-7551 P-7653a P-6345a P-6345b P-6644a P-6644b P-7035a P-7035b P-7035c P-CP P-D1 P-D2 P-D3 P-GITS-2 P-GITSc P-HUMBOLDT P-HUMB-E P-HUMB-N P-HUMB-S P-HUMB-W P-NASA-U1 P-NASA-U2 P-NASA-U3 P-SUM99 P-TUNU P-TUNU-E25c P-TUNU-E50 P-TUNU-N25 P-TUNU-N50 P-TUNU-S25 P-TUNU-W50 P-CNP1 P-CNP2 P-CNP3 P-JAV2 P-JAV3 P-KUL1 P-KUL2 P-KUL3 P-NSE P-S-DOME P-UAK1 P-UAK2 P-UAK3 P-UAK4 P-UAK5 P-D5 P-BASIN4 P-BASIN5 P-BASIN6 P-BASIN7 P-BASIN8 P-BASIN9 Period Latitude, Longitude (d.dd) 73.90, 76.60, 80.00, 63.80, 66.50, 67.50, 68.50, 68.00, 69.00, 69.60, 69.40, 69.20, 69.00, 71.50, 72.30, 71.90, 72.20, 73.00, 75.00, 76.00, 63.10, 63.10, 66.00, 66.00, 69.50, 69.50, 69.50, 69.90, 64.50, 71.80, 69.80, 77.20, 77.20, 78.50, 78.60, 78.70, 78.30, 78.50, 73.80, 73.80, 73.80, 72.80, 78.00, 78.00, 78.00, 78.30, 78.50, 78.30, 78.00, 73.20, 71.90, 70.50, 72.60, 70.50, 67.50, 66.80, 66.10, 66.50, 63.10, 65.50, 65.50, 65.50, 65.50, 65.40, 68.50, 62.30, 63.90, 67.00, 67.50, 69.80, 65.00, 37.60 36.40 41.10 45.00 42.50 45.00 39.50 41.00 38.00 39.00 41.00 43.00 45.00 45.00 45.00 47.50 49.40 45.00 51.00 53.00 44.80 44.80 44.50 44.50 34.50 34.50 34.50 47.00 43.50 46.20 44.00 61.10 61.10 56.80 55.70 57.20 56.80 58.00 49.50 49.50 49.50 38.00 34.00 32.90 31.90 33.90 33.80 33.80 36.10 32.10 32.40 33.50 47.10 46.10 39.00 40.10 41.00 42.50 46.40 44.50 43.50 42.60 46.10 46.50 42.90 46.30 46.40 41.80 40.40 36.40 44.90 Elevation (m) 2657 Concentration (ng g1) (s) 125 (2)b 108 (4)b 151 (7)b 134 (31) 101 (32) 147 (40) 129 (24) 115 (17) 129 (25) 149 (48) 178 (90) 145 (47) 132 (90) 115 (17) 126 (19) 138 (21) 181 (43) 137 (10) 164 (33) 141 (24) 101 (34) 110 (49) 105 (27) 142 (48) 107 (13) 129 (21) 135 (22) 118 (27) 76 (29) 112 (16) 115 (26) 248 (59) 265 (79) 111 (67) 213 (45) 195 (48) 123 (29) 125 (28) 106 (59) 117 (17) 131 (33) 110 (15) 127 (37) 207 (176) 142 (24) 159 (52) 175 (29) 199 (69) 157 (15) 111 (34) 74 (10) 125 (75) 126 (32) 117 (44) 74 (24) 100 (81) 47 (7) 66 (7) 99 (35) 114 (45) 48 (18) 51 (20) 103 (35) 99 (39) 156 (34) 152 (73) 145 (79) 116 (40) 110 (26) 124 (19) 137 (42) Accumulation (kg m2 a1) (s) 314 (70) 557 (155) 336 (50) 370 (68) 443 (73) 338 (56) 325 (39) 366 (66) 368 (77) 411 (84) 398 (69) 346 (57) 391 (76) 801 (101) 276 (60) 296 (63) 347 (46) 661 (95) 643 (94) 418 (79) 428 (55) 468 (93) 490 (98) 482 (114) 416 (158) 733 (107) 434 (97) 394 (64) 290 (55) 317 (61) 149 (57) 156 (48) 145 (24) 114 (52) 115 (51) 313 (131) 382 (154) 408 (259) 223 (25) 104 (29) 151 (21) 97 (23) 73 (13) 138 (30) 138 (31) 150 (26) 149 (36) 219 (37) 281 (32) 389 (91) 373 (71) 525 (156) 812 (168) 1150 (162) 442 (137) 537 83) 465 (61) 656 (142) 987 (387) 354 (129) 366 (109) 342 (69) 352 (154) 330 (81) 622 (116) 618 (83) 349 (48) 323 (57) Flux, (mg cm2 a1) (s) 49.9 (13) 61.8 (25) 49.1 (15) 47.7 (11) 50.7 (10) 45.2 (13) 44.1 (13) 59.5 (25) 53.9 (15) 42.2 (15) 46.2 (9) 43.7 (9) 53.7 (10) 147.7 (50) 36.8 (8) 44.6 (11) 44.2 (13) 63.8 (21) 79.1 (26) 47.6 (11) 56.0 (16) 50.5 (11) 62.8 (15) 64.9 (20) 49.3 (12) 39.6 (19) 30.1 (10) 27.4 (9) 61.5 (23) 93.0 (26) 14.2 (6) 18.4 (8) 17.5 (8) 15.5 (8) 16.8 (7) 31.3 (15) 41.4 (16) 38.7 (25) 15.7 (5) 33.1 (16) 75.8 (72) 26.6 (11) 35.6 (18) 43.6 (19) 49.5 (22) 42.8 (18) 15.2 (6) 19.9 (11) 28.7 (13) 41.5 (13) 40.1 (17) 42.1 (23) 80.1 (53) 55.0 (13) 25.8 (9) 46.2 (19) 36.6 (15) 29.9 (11) 44.4 (24) 39.2 (16) 44.4 (22) 54.3 (13) 59.1 (27) 49.8 (23) 71.8 (25) 64.3 (16) 38.8 (11) 39.1 (10) Temperaturea (C) (s) (0.19) (0.43) (0.64) (0.47) (1.02) (0.56) (0.25) (0.65) (0.26) (0.25) (0.28) (0.46) (0.43) (0.43) (0.29) (0.86) (0.55) (1.04) (0.89) (1.42) (1.42) (0.74) (0.74) (1.66) (1.66) (1.66) (1.09) (1.34) (0.39) (0.40) (0.98) (0.83) (1.23) (0.84) (1.55) (0.27) (0.27) (0.27) (0.22) (0.67) (0.65) (0.66) (0.83) (0.61) (0.83) (0.52) (0.61) (0.99) (0.54) (0.63) (0.70) (1.58) (0.70) (1.69) (1.02) (0.75) (0.74) (1.16) (2.43) (1.02) (1.06) (0.49) (1.68) (0.79) (1.16) (1.36) (0.28) (0.14)

2 of 15

D06301 Table 1. (continued)
Point ID NGT01-C NGT01 NGT02 NGT03 NGT04 NGT05-C NGT05 NGT06 NGT07 NGT08 NGT10 NGT11 NGT12-C NGT12 NGT13 NGT14 NGT15 NGT16 NGT17 NGT18-C NGT18 NGT19 NGT22 NGT23 NGT25 NGT26 NGT27 NGT28

Period 1990 1994

Latitude, Longitude (d.dd) 73.03, 73.03, 73.50, 73.94, 74.40, 74.85, 74.85, 75.25, 75.25, 75.25, 75.57, 75.65, 75.72, 75.72, 76.17, 76.62, 76.62, 76.62, 77.07, 77.52, 77.52, 78.00, 78.42, 78.83, 79.23, 79.62, 80.00, 80.36, 37.65 37.65 37.65 37.63 37.63 37.63 37.63 37.62 37.22 36.90 36.53 36.32 36.40 36.40 36.40 36.40 37.37 34.47 36.40 36.39 36.39 36.40 36.43 36.50 37.95 39.51 41.14 41.13

Elevation (m) 2185 2072

Concentration (ng g1) (s) 115 (40) 195 (99) 113 (22) 192 (150) 108 (36) 150 (26) 155 (34) 163 (37) 179 (56) 152 (10) 154 (30) 142 (32) 141 (40) 139 (11) 133 (20) 116 (52) 139 (42) 130 (37) 125 (16) 149 (33) 102 (45) 146 (35) 180 (43) 184 (45) 169 (45) 159 (28) 158 (28) 146 (18)
Accumulation (kg m2 a1) (s) 147 (55) 172 (33) 155 (26) 138 (34) 133 (8) 111 (19) 126 (32) 110 (18) 141 (7) 116 (44) 117 (26) 129 (12) 116 (22) 136 (42) 131 (27) 105 (11) 115 (18) 148 (6) 124 (13) 106 (29) 110 (30) 133 (40) 115 (37) 107 (30) 97 (25) 113 (41) 120 (34) 131 (28)
Flux, (mg cm2 a1) (s) 1.3 (6) 31.4 (10) 17.2 (1) 23.5 (12) 14.3 (4) 15.2 (5) 19.1 (3) 17.7 (3) 25.4 (9) 17.4 (6) 17.7 (4) 18.4 (5) 16.0 (5) 18.8 (5) 17.5 (5) 12.5 (7) 16.4 (7) 19.4 (6) 15.7 (4) 22.8 (11) 11.6 (8) 19.2 (8) 21.2 (9) 18.0 (4) 16.6 (6) 18.3 (9) 19.3 (7) 19.1 (5)
Temperaturea (C) (s) (0.16) (0.16) (0.13) (0.20) (0.14) (0.20) (0.20) (0.33) (0.27) (0.21) (0.25) (0.26) (0.24) (0.24) (0.28) (0.46) (0.34) (0.52) (0.42) (0.48) (0.48) (0.52) (0.58) (0.62) (0.52) (0.49) (0.63) (0.92)
Developed from Peterson and Vose [1997]. Three-year average records. c Excluded from evaluation.
alteration to the atmospheric burden of reactive nitrogen [Wolff et al., 2007; Burkhart et al., 2006]. [3] Current values from Greenland ice cores are nearly double those prior to 1940. It is unclear if this increase is uniform over the ice sheet or if the anthropogenic increases have resulted in a geographic shift in nitrate concentration. The spatial variability of nitrate concentration and nitrate flux over Greenland has been investigated in the northeast [Fischer et al., 1998b], a region that is subject to postdepositional snow-to-atmosphere chemical exchange, and spatial trends have been attributed to changes resulting from accumulation variability. Rothlisberger et al. [2002] evalu ated trends in nitrate over Greenland and Antarctica, showing high correlations between temperature and nitrate concentration, as well as an effect of volcanic sulfate in mobilizing nitrate in the preserved record. While their study evaluated a larger area of the Greenland ice sheet than did the Fischer et al. [1998b] analysis, their records were not all contemporaneous, nor did they cover as wide a range of accumulation regimes as is presented in the current analysis. This paper expands on the previous two investigations, and for the first time presents nitrate concentration and flux records from widely varied regions of the ice sheet covering several accumulation, temperature, and elevation regimes. [4] We have analyzed records from 96 contemporary ice core and snow pit locations geographically distributed over the ice sheet to evaluate the dependence of nitrate concentration and flux on environmental variables, i.e., accumulation and temperature, and to ascertain the magnitude of spatial variability of nitrate deposition over Greenland. Using the infor-

mation contained in time series data from each location, and the spatial relation of the mean values at each site, we address three questions. First, what factors drive variability in nitrate concentration and flux over the ice sheet, i.e., is the ice sheet a spatially consistent paleoatmospheric archive for nitrate? Second, has the geographic distribution of nitrate over the ice sheet shifted during the period of recent anthropogenic influence? Third, what is the spatial variability of nitrate deposition, and how is it related to accumulation variability?

2. Data and Methods

2.1. Ice Core and Snow Pit Records [5] The data set for the analysis was developed from three primary sources: (1) the Program for Arctic Climate Assessment (PARCA) cores collected from 1993 to 2003 [Bales et al., 2001a; Burkhart et al., 2006], (2) the North Greenland Traverse (NGT) cores and snow pits collected between 1993 and 1995 [Fischer, 2000, 2003] (and similar data sets therein), and (3) a Summit core, collected in 1999. [6] The PARCA cores were developed primarily to investigate the variability of Greenland Ice Sheet accumulation in the dry snow zone, generally defined as a region above 2000 masl [Bales et al., 2001a], and as such are well distributed to capture various accumulation regimes and spatial gradients (Table 1 and Figure 1). The dry snow zone is considered the region of the ice sheet which does not undergo annual melting, and hence percolation of soluble compounds. The contour should not be confused with an equilibrium line for the ice sheet.

3 of 15

Figure 1. Core and snow pit locations from PARCA and the NGT traverse. Groupings of cores used in the spatial variability analysis are indicated by circles. The gray circles (pre-1900 sites) represent cores used for Figures 3 and 4. One, two, and three thousand meter contours shown in dashed lines. [7] All PARCA cores were dated using a combination of seasonally varying markers (H2O2, d18O, dust, NO, NH+, and Ca+2) and specific time-stratigraphic markers (SO2, 4 beta radioactivity). The major ion and H2O2 records were developed using continuous flow analysis (CFA) [Anklin et al., 1998; Bales et al., 2001a, 2001b; Rothlisberger et al., 2000], while discrete samples were analyzed for d18O and dust. Subsections were measured for beta radioactivity and SO2 to capture specific volcanic peaks and beta horizons, 4 spikes in tritium resulting from nuclear bomb testing during the 1960s, as absolute age indicators [Mosley-Thompson et al., 2001]. This method enables multiparameter dating with accuracy of essentially zero years over the time periods considered in this paper. For the development of the annual concentration values, layer thickness for each year was determined from either the winter minimum of H2O2 or the spring peak in calcium. [8] The NGT data are available from the PANGAEA database (http://www.pangaea.de), including 28 snow pits and three deeper ice cores (B16, B18, and B21) collected along a south to north traverse (71N to 80N) through the northeastern portion of the Greenland ice sheet (Figure 1) [Fischer, 2003]. The snow pit samples and ice cores were analyzed at high resolution (8 samples/year) for major ion concentrations and d 18O [Fischer, 1997, 2000, 2003; Fischer et al., 1998a]. Annual concentrations were determined using the calcium maximum to define individual years. Dating of the cores is accurate to 1 year. [9] The Summit 99 core was collected 7 km northeast of the Greenland Environmental Observatory at Summit (GEOSummit). The core was analyzed in the field for NO, H2O2, Ca2+, NH+, and HCHO using CFA analysis. Further analysis was conducted in the laboratory after the field season using CFA with Trace Element analysis (CFA-TE) [Burkhart et al., 2006]. To convert the concentration records from depth to time and to determine layer thickness of individual years we used the winter Na+ maximum and H2O2 minimum. Specific time-stratigraphic markers (i.e., known volcanic events) were used as absolute depth-age markers. [10] Temperature and sublimation parameters for each core site are developed from gridded data sets. The temperature data is retrieved from the Global Historical Climatological Network (GHCN) data (http://www.ncdc.noaa.gov/

4 of 15

Figure 2. (a and b) Available time and period data and (b e) cumulative distribution of core locations for sites used in kriging analysis. oa/climate/research/ghcn/ghcngrid.html) and described further by Peterson and Vose [1997]. The sublimation data is from Box and Steffen [2001]. We calculated an ice core location value based on the four closest cells to the point, weighting the cells as a function of the distance to the cell center such that
Tsite a * TcellA b * TcellB c * TcellC d * TcellD
a b c d 1; where a di;cellA =D D P di;cellX
Here Tsite is the site temperature and TcellA is the temperature in one of four cells. The distance from the ice core site to the cell center (di,cellX) is used to determine the weighting factors, a, b, c, d. The same approach was used for the sublimation data set. 2.2. Kriging [11] Of the 96 points used in the analysis, record lengths for the period vary from 2 to 146 years, with an
average record length of 18 years (Figure 2a). For the period there are 99 records with at least 2 years of data (Figure 2a). Ice core locations were not significantly clustered and are well distributed with respect to elevation and latitude, centered at 2500 mean annual sea level (masl) and 72.5N. The meridional location of core and snow pits are well distributed in the southern region with semiregularly spaced ($km separation distance) locations distributed throughout (Figures 2c 2e). North of 70N the records are less distributed, with a large portion of the data from the NGT traverse. We used the UTM 24N coordinate system for our easting and northing directions. [12] The variogram modeling module for irregularly spaced data, gamv, from the Gstat software package [Pebesma and Wesseling, 1998], was used to model the variograms. For concentration we used a spherical model with sill of 25 (ng g1)2 and a nugget of 15 (ng g1)2. For flux our spherical model had a sill of 1.1 (mg cm2 a1)2 with a nugget of 0.9 (mg cm2 a1)2. We chose to preserve the spatial variability by using a relatively short lag increment (15 km) for 15 lags. The major range of the variogram model for concentration was 60 km and for flux 150 km. We evaluated anisotropy by examining our sample vario-

5 of 15

Figure 3. Kriged surfaces of nitrate (a) concentration and (b) flux over the Greenland Ice Sheet using 96 ice core and snow pits with at least 2 years of data from 1988 to 1994. Accumulation contours (kg m2 a1) from the Bales et al. [2001a] accumulation map are shown with white representing lower accumulation. grams in several search directions but found no improvement with directional variogram modeling. In most cases a search radius of 150 km allowed for at least 4 points being included in the estimate. In cases in which no data were available within that area, a prediction was not made. This was generally only necessary in the north central and eastern portion of the ice sheet. [13] Surface maps of nitrate concentration and annual flux were developed for the Greenland Ice Sheet with a mask employed below the 2000 m. contour using ordinary kriging. Note, we have employed the same mask as used in prior kriging studies over the ice sheet [Bales et al., 2001a] where the 2000 meter contour was chosen as a result of data availability and to assure samples exist only in the dry snow zone. For the flux data a log transform was required as a result of the high degree of skewness. Both second-order drift surfaces and log transforms of the data were accounted for in the geostatistical analysis. As a result of the log transformation for the flux data, we followed the backtransformation procedures standard to log normal kriging to generate the final prediction surfaces. [14] After evaluating the sample variogram for concentration and flux, three of the 99 records were removed from the data set as a result of their individual contributions to the sample variogram at short lag increments. One point was from the Gits location, another from the Tunu cluster, and a third shallow core located on the western portion of the ice sheet (P-7249). Both the Gits and Tunu cores were collected in addition to a main core, at each site. Reviewing the individual points, we found the value of the Tunu-E25 core was nearly twice in concentration of any surrounding core and also had a much greater standard deviation. At the P-Gits site, the core was also higher in concentration than the main core. Closer analysis of the data showed significant variability, which could be the result of melt layers causing nitrate to accumulate owing to percolation. For the P-7249 core, the concentration was within the range of the surrounding cores; however, the accumulation was twice that of the surrounding cores, indicating that it may have been located on an accumulation dome, or some other anomalous feature. Note that this particular point was also removed from the kriging analysis when the authors developed a new accumulation map for the Greenland Ice Sheet based on the PARCA data set [Bales et al., 2001a].

3. Results

3.1. Spatial Distribution of Nitrate Concentration and Flux [15] The average nitrate concentration from all points across the ice sheet for is 132 ng g1 with a range of ng g1 and a standard deviation of 37 ng g1 (Table 1). Standard deviation of the average ice sheet concentration is equivalent to the average standard deviation for all the individual time series, though for the time series the individual variability can be large, with one

6 of 15

Table 2. Regional Statistics From Ice Core Data
Accumulation Regime Low (<200 kg m a ) Moderate (kg m2 a1) High (>500 kg m2 a1) All area > 2000 masl
Area (km2) 602,500 377,500 127,500 1,107,500
Concentration (ng g1) (s) (18) (16) (14) (20)
Flux (mg cm2 a1) (s) 2.5 (0.75) 4.6 (0.60) 5.4 (0.75) 3.6 (1.4)
record having a standard deviation 84% of its mean (Tunu-E25). Nitrate flux for the data set varies between 1.1 and 14.7 mg cm2 a1, with a mean of mg cm2 a1. Note that concentration varies between sites by a factor of six, whereas the flux varies by a factor of ten. Water accumulation for all sites varies by a factor of 15, raging between 73 and 1150 kgH2O m2 a1, with a mean of kgH2O m2 a1. [16] Shown in Figure 3a is a kriged surface of concentration values for the dry snow zone. During our cross validation error analysis we found twenty points having errors greater than 20% (26 ng g1) of the mean ice sheet concentration. Despite the limited data points available for such a large region, the standard error for the concentration surface is only 28 ng g1, while root mean square prediction errors are 27 ng g1 indicating a good prediction of variability. The kriged flux surface ranges between 1.2 and 14.8 mg cm2 a1 with a mean of mg cm2 a1 (Figure 3b) which is well within the values of the point measurements. For flux the standard error was 1.3 mg cm2 a1 with root mean square prediction error of 1.1 mg cm2 a1. [17] Two features are prominent in the kriged surface of concentration: (1) a gradient of increasing concentration from the southeast to the northwest is notable, particularly over the northern plateau north of 70N, and (2) a strong west to east decrease between the 65 75N parallels (Figure 3a). In the southern region high variability and a general decrease in concentration are colocated with very large (>800 kgH2O m2 a1) accumulation rates, primarily around the vicinity of the eastern UAK cores (65.5N, 42.5W). Examining the ice sheet as a whole reveals few significant trends between longitude, elevation and concentration for the period. The trend with latitude, however, persists over the entire ice sheet with a slope of 18 ng g1 per degree latitude northward. Overall there is a general decrease in concentration from the low- to high-accumulation regions. However, in the moderate-accumulation region there are numerous areas where the contours of accumulation are perpendicular to the concentration gradients, indicating no correlation. [18] The flux map shows a stronger relationship to accumulation, with regions of high and low flux coincident with high and low accumulation, respectively. In the northern regions of the ice sheet flux is below 3 mg cm2 a1. There is a $50% increase in flux along the transition from the low- to moderate-accumulation region. In the intermediate regions of accumulation, patterns of accumulation and flux are in less agreement, particularly in the central region of the ice sheet ($68N, $42W). The largest fluxes are coincident with the highest accumulation in the southeastern portion of the ice sheet. Large-scale trends in flux appear in two directions; in the southern region decreasing from the southeast to the northwest, and on the northern plateau

decreasing in a northeasterly direction, though neither gradient is statistically significant. [19] In Table 2 we have grouped the records into regions of high (>500 kgH2O m2 a1), moderate (kgH2O m2 a1), and low accumulation (<200 kgH2O m2 a1). The division of the regions is based upon analysis of the histogram, which shows bimodal distribution of the data. The first two regions separate the two peaks in the distribution into the low and moderate regimes, while the high-accumulation band captures the tail of the distribution. For the dry snow zone the low-accumulation region accounts for $55% of the area. The moderate- and highaccumulation regions account for 34% and 11%, respectively. Nitrate concentrations in the low-accumulation area are $5% greater, while flux is $30% lower than the ice sheet mean. In the moderate region, concentration is representative of the mean of all areas, while flux is 30% greater. The concentrations in the high accumulation are 20% lower from the mean, yet flux is 50% greater. 3.2. Temporal Patterns of Distribution [20] To test whether anthropogenic components of variability have shifted the distribution of nitrate, we investigated the temporal history of geographic correlations using eleven multidecadal records (dark gray circles in Figure 1). Pearsons r was calculated for each year available for 11 sites against latitude, longitude, and elevation. We used a ten year boxcar window to reduce the noise in the data prior to the analysis. These results show changes in the distribution of concentration with latitude, longitude, and elevation across the ice sheet (Figure 4). Correlation coefficients (Pearsons r) for latitude were between 0.48 and 0.74 and are frequently significant at the 90th percentile until around 1960, indicating a slight gradient of increased concentrations northward on the ice sheet. A negative correlation between concentration and elevation ranges between 0.49 and 0.74, also significant until around 1960. The relationships for latitude and elevation were consistent over the period of 1850 1960. Trends in longitude, however, were progressively more negatively correlated over time, becoming significant around 1950. [21] Uncertainty in such an analysis is quite high, as the results of the linear regressions could be affected by a single core. We attempted to mitigate the small population of data for the longer record ice cores by repeating the analysis for the period of using an additional 19 ice cores available for that time period. The temporal patterns trends in the correlation coefficient between latitude, longitude, elevation and concentration during the past 35 years from 30 cores agreed with that of the 11 longer records (Figure 4). However, for the 96-core period data set, the results vary slightly in that the correlation coefficient for latitude is slightly significant (plotted as stars in Figure 4, bold where significant).

7 of 15

Figure 4. Pearsons r correlation coefficient plotted in time shows changes in linear relationship for nitrate concentration with longitude, latitude, and elevation for long-record (>150 years) ice cores shown as plus signs. Where the linear regression is significant at the 90th percentile, the values are shown with solid circles. The results from the addition of 19 ice cores to the analysis are shown with solid squares and solid triangles where significant. The correlation coefficients for the 96-core data set are plotted with a star (see text). 3.3. Spatial Variability of Nitrate [22] For seven geographic clusters of ice cores and three colocated cores, coefficients of variability were calculated on a year-by-year and core-by-core basis as the difference between an individual cores annual value and the mean of all other cores in the cluster. Temporal variability for each core is calculated as the difference between annual concentration value and the mean value for that core for the period of analysis. All time series were detrended using linear regression and normalized by dividing by the mean. In all cases, spatial variability was significant. For cluster 6838, which also has the largest mean distance (131 km) between core locations, the spatial variability actually exceeded the year to year variability. For the northern, low-accumulation Tunu and Humboldt clusters the spatial variability is 80% of the temporal, whereas in the high-accumulation regions in the south, spatial variability is in the range of 30 75% of the temporal variability despite the mean separation distance being twice as large ($50 km for Tunu and Humboldt versus $100 km for the southern clusters). Note, however, that even in the case of colocated cores (separation distance <1 km) the spatial variability is still in the range of 30 50% of the year-to-year variability. [23] Following McConnell et al. [2000a] and Fisher et al. [1985], we evaluated the effect of the spatial variability on the overall estimate of variability for each site. The spatial component of variability is developed assuming that the observed variability in an individual Var(x) is the sum of the synoptic-scale atmospheric variability, Var(C), and the pseudorandom spatial variability, Var(e), given the recorded signal x(t) = C(t) + ex(t). Note, that Var(x) denotes the variability of the record, and not the variance. For two colocated cores, the time series x(t) and y(t) will have Var(x) and Var(y). Signal and noise components of the time series were separated using the cross-correlation coefficient between the two sites, rxy, such that Var(C) = rxy Var(x) and Var(ex) = (1 rxy) Var(x). In cases of more than two cores per location, we used the average correlation coefficient from the central core to all other cores to estimate rxy. Two weaknesses in this method are as follows: (1) similar to the McConnell et al. [2000a] analysis, neither signal (x or y) is noise free, and (2) the time series for colocated and clustered cores is short ($10 years). [24] In most cases the spatial variability estimate exceeds the estimate of Var(C) (Table 3), except in the cases of colocated cores (<1 km separation). The average separation distance is large ($100 km) compared with most spatial variability studies, and for the two core groups that had field collection arrays designed, and hence lower separation distances, the accumulation rate is low.

Figure 5 shows there is clearly greater variability at the lowelevation sites and also provides evidence consistent with Burkhart et al. [2006] where it was found variability has increased in the postindustrial period. To test the null hypothesis that the distributions have equal mean and variance, we employ the two-tailed nonparametric Kruskal-Wallis (KW) test. The KW test statistic is an extension of the Mann-Whitney test, following approximately a chi-square distribution with k-1 degrees of freedom; P values are derived from this. For small samples the KW test statistic tables are preferred over the chi-square approximation [Conover, 1999]. For the two data sets we can reject the null hypothesis at the 95th percentile indicating that variability at the lower elevations is significantly greater than for the higher elevations (P > c2 = 0.96).
Figure 5. Mean 15-year windowed rate of change in nitrate concentration for two groups of ice cores from above (black) and below (gray) 2500 masl elevation. Shaded areas show the variance for each year. Greater variability at lower elevations is evident throughout the time series.

9 of 15

Figure 6. Average values from 96-core average data set of accumulation, concentration, and flux with latitude, longitude, elevation, temperature, and sublimation. Latitude is given as kilometers north of the 60 parallel. Longitude is given as kilometers from the 40 meridian (westward is negative; eastward is positive). Pearsons r values are shown when significant at the 0.95 level. Sublimation values are extracted in the same manner as temperature [Peterson and Vose, 1997] from the Box and Steffen [2001] data set. [27] In interpreting the plot, consider that a linear increasing trend, as can be seen in both the high- and low-elevation data sets after $1950, indicates an acceleration of the increases in nitrate concentration. This plot shows that nitrate concentrations increases deviated from the zeromean strongly during the 1970s. As has been seen in another analysis [Burkhart et al., 2006] there is also a decrease toward the end of the 20th century. It is interesting that the lower cores show a stronger decrease, while the higher-elevation cores seem only to level off. Note again, a leveling off in the growth rate indicates the concentrations are still increasing, while the lower-elevation cores show a decreasing concentration. While these differences could explain some of the shifts in regional patterns of deposition, the analysis would benefit significantly from a transport study, investigating differences in the specific sources for the ice cores. [28] The latitudinal gradient is also of interest owing to the possibility that it is a result of Arctic haze. Challenging the interpretation is the fact that the south to north increase in concentration also follows temperature, and to some degree the accumulation trends, two variables which significantly act on concentration (see below). This latitudinal gradient does, however, lose significance in time. As there are no accumulation trends in time, it is either driven by changes in temperature or as a result of increased pollution reaching the lower latitudes, driving an overall change in the spatial distribution of nitrate over the ice sheet. Two scenarios exist, and are possibly acting together. First, reactive N sources for the Greenland ice sheet are largely from North American and European continental emissions [Eckhardt et al., 2003; Burkhart et al., 2006], for the North American case, presumably these emissions would impact more greatly the southern region of the ice sheet, hence increasing these concentrations and decreasing the latitudinal gradient. Increases in Arctic temperatures would act to lower concentrations in the northern regions of the ice sheet (assuming temperature increases are nonuniform and

10 of 15

Table 4. Stepwise Multiple Linear Regression Results (r2) in Individual Accumulation Regimesa
Accumulation Regime Low Variable Latitude Longitude Elevation Temperature Accumulation Concentration Best subset RMSE
Moderate flx 0.07 0.47 0.27 0.81d 0.44 cnc 0.01 0.01 0.01 0.05 0.15 0.26b 0.94 acc 0.01 0.17 0.01 0.06 0.35c 0.85 flx 0.00 0.41 0.17 0.92d 0.3 cnc 0.22 0.23 0.17 0.02 0.36 0.55b 0.88
High acc 0.06 0.16 0.25 0.36 0.56c 0.77 flx 0.01 0.14 0.34 0.92d 0.33
cnc 0.04 0.08 0.01 0.08 0.06 0.26b 0.90
acc 0.06 0.16 0.25 0.36 0.68c 0.59
The components of the model used for regression are described in the footnotes that follow. Accumulation and T used as regressor variables (y = f(acc., temp.)). Elevation, T, lat, lon used as regressor variables (y = f(elv., temp., lat., lon.)). d Accumulation, concentration, and T used as regressor variables (y = f(acc., conc., temp.)).
stronger to the north). This too would decrease the north-tosouth gradient in concentration. Unfortunately, this data set does not extend far back enough in time to consider a period that is free of covarying anthropogenic driven changes temperature and reactive N. Understanding this gradient could provide insight into the history of Arctic haze, and whether it has existed prior to the 19th and 20th centuries. 4.2. Drivers of Spatial Patterns [29] We used stepwise multiple linear regression to evaluate the cause of geographic variability, and its relation to global transport. In order to properly assess the spatial patterns we must evaluate how snow concentrations depend on accumulation and temperature [Fischer et al., 1998b; Rothlisberger et al., 2002]. Within each accumulation regime, we evaluated the contribution to variability of each variable (latitude, longitude, elevation, accumulation, temperature, and sublimation) to determine whether the observed concentration and flux patterns may be explained simply by shifts in accumulation and/or temperature over the ice sheet rather than gradients in the atmosphere. We selected the records without bias to location using the simple criteria of mean annual accumulation to extract the records for each region. Not surprisingly, however, points for each regime were selected from common regions. The low-accumulation area is in the northeastern quadrant of the ice sheet. The high-accumulation locations are in the southeasternmost portion, and the moderate-accumulation areas fill in the remaining area. As can be seen in Figure 6, several of the predictor variables are correlated with one another. The approach of selecting samples from a common accumulation regime was beneficial in that the regression of accumulation with latitude, temperature, or elevation within each region was less significant, thereby reducing the colinearity of our predictor variables. [30] For all three regimes, accumulation and temperature were the best predictors of concentration (Table 4). Flux was a function of accumulation, concentration, and temperature, though the relation with temperature did not contribute greatly to the explained variance. Accumulation within a region was best explained by temperature, elevation, latitude, and longitude, the latter being most important in highaccumulation regions as it was essentially a measure of distance to the coast. [31] There are different relationships between accumulation, concentration, and temperature as predictors of flux in

[42] Despite the challenges and variability resulting from air-to-snow cycling, with sufficient accumulation (>200 kgH2O m2 a1), year to year variability is quantifiable. The data presented here supports the analysis of ice cores as a proxy for past atmospheres and demonstrates that regional changes in the tropospheric chemistry are being recorded across the ice sheet, thus should be considered when selecting ice core locations. [43] Acknowledgments. This research was funded in part by NASA grant NNG04GB26G and grants NSF 9813442 and NSF 0133204 from the National Science Foundations Office of Polar Programs and Geosciences Directorate, respectively. J. Burkhart was funded through a NASA Earth System Science Fellowship. Funding for some of the analysis was provided through the Norwegian Research Councils POLARCAT grant (175916). Three excellent anonymous reviews contributed valuably to the final manuscript. Research on the Greenland Ice Sheet is made possible with the generous permission of the Danish Polar Centre and Greenland Home Rule Ministry of Environment and Nature.

References

Anklin, M., R. C. Bales, E. Mosley-Thompson, and K. Steffen (1998), Annual accumulation at two sites in northwest Greenland during recent centuries, J. Geophys. Res., 103, 28,775 28,783, doi:10.1029/98JD02718. Bales, R. C., J. R. McConnell, E. Mosley-Thompson, and G. Lamorey (2001a), Accumulation map for the Greenland Ice Sheet: 1971 1990, Geophys. Res. Lett., 28, 2967 2970, doi:10.1029/2000GL012052. Bales, R. C., E. Mosley-Thompson, and J. R. McConnell (2001b), Variability of accumulation in northwest Greenland over the past 250 years, Geophys. Res. Lett., 28, 2679 2682, doi:10.1029/2000GL011634. Bergin, M. H., J. L. Jaffrezo, C. I. Davidson, J. E. Dibb, S. N. Pandis, R. Hillamo, W. Maenhaut, H. D. Kuhns, and T. Makela (1995), The contributions of snow, fog, and dry deposition to the summer flux of anions and cations at Summit, Greenland, J. Geophys. Res., 100, 16,275 16,288, doi:10.1029/95JD01267. Box, J. E., and K. Steffen (2001), Sublimation on the Greenland ice sheet from automated weather station observations, J. Geophys. Res., 106, 33,965 33,981, doi:10.1029/2001JD900219. Box, J. E., D. H. Bromwich, and L.-S. Bai (2004), Greenland ice sheet surface mass balance 1991 2000: Application of Polar MM5 mesoscale model and in situ data, J. Geophys. Res., 109, D16105, doi:10.1029/ 2003JD004451. Burkhart, J. F., M. A. Hutterli, and R. C. Bales (2002), Partitioning of formaldehyde between air and ice at 35C to 5C, Atmos. Environ., 36, 2157 2163, doi:10.1016/S1352-2310(02)00221-2. Burkhart, J. F., M. Hutterli, R. C. Bales, and J. R. McConnell (2004), Seasonal accumulation timing and preservation of nitrate in firn at Summit, Greenland, J. Geophys. Res., 109, D19302, doi:10.1029/2004JD004658. Burkhart, J. F., R. C. Bales, J. R. McConnell, and M. A. Hutterli (2006), Influence of North Atlantic Oscillation on anthropogenic transport recorded in northwest Greenland ice cores, J. Geophys. Res., 111, D22309, doi:10.1029/2005JD006771. Conover, W. J. (1999), Practical Nonparametric Statistics, 3rd ed., vol. viii, 584 pp., John Wiley, New York. Dibb, J. E., R. W. Talbot, J. W. Munger, D. J. Jacob, and S. M. Fan (1998), Air-snow exchange of HNO3 and NOy at Summit, Greenland, J. Geophys. Res., 103, 3475 3486, doi:10.1029/97JD03132. Eckhardt, S., A. Stohl, S. Beirle, N. Spichtinger, P. James, C. Forster, C. Junker, T. Wagner, U. Platt, and S. G. Jennings (2003), The North Atlantic Oscillation controls air pollution transport to the Arctic, Atmos. Chem. Phys., 3, 1769 1778. Fischer, H. (1997), Spacial variability in ice core time series of north east Greenland (in German), in Raumliche Variabilita in Eiskernzeitreihen t NordostgronlandsRekonstruktion klimatischer und luftchemischer Langzeittrends seit 1500 A.D., Univ. of Heidelberg, Heidelberg, Germany. Fischer, H. (2000), Nitrate and sulphate in ice core ngt14C93.2 from the North Greenland Traverse, doi:10.1594/PANGAEA.57092, PANGAEA, Network for Geol. and Environ. Data, Germany. Fischer, H. (2003), Chemistry on snow-pit ngt07S93 from the North Greenland Traverse, doi:10.1594/PANGAEA.133073, PANGAEA, Network for Geol. and Environ. Data, Germany. Fischer, H., D. Wagenbach, and J. Kipfstuhl (1998a), Sulfate and nitrate firn concentrations on the Greenland ice sheet: 2. Temporal anthropogenic deposition changes, J. Geophys. Res., 103, 21,935 21,942, doi:10.1029/ 98JD01886. Fischer, H., P. Wagenbach, and J. Kipfstuhl (1998b), Sulfate and nitrate firn concentrations on the Greenland ice sheet: 1. Large-scale geographical

R. C. Bales, J. F. Burkhart, and M. M. Frey, School of Engineering, University of California, P.O. Box 2039, Merced, CA 95344, USA. (jburkhart@ucmerced.edu) M. A. Hutterli, British Antarctic Survey, Natural Environment Research Council, High Cross, Madingley Road, Cambridge CB3 0ET, UK. J. R. McConnell, Division of Hydrologic Sciences, Desert Research Institute, 2215 Raggio Parkway, Reno, NV 89512, USA.

15 of 15