Timescales for dust variability in the Greenland Ice Core Project (GRIP) ice core in the last 100,000 years

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 104, NO. D24, PAGES 31,043-31,052, DECEMBER 27, 1999 Timescales for dust variability in the Greenland Ice Core Project (GRIP) ice core in the last 100,000 years Katrin Fuhrer Physics Institute, University of Bern, Bern Eric W. Wolff British Antarctic Survey, Natural Environment Research Council, Cambridge, England S igfus J. Johnsen Geophysical Institute, University of Copenhagen, Copenhagen Abstract. The calcium (representing dust) concentration record of the last 100,000 years from the Greenland Ice Core Project (GRIP) ice core shows a huge dynamic range (factor > 100). The relationship between dust concentrations and temperature (represented by the oxygen isotope ratio) is not a simple one, as has often been assumed. A rapid alternation (factor of 5-10) between low concentrations during the Dansgaard-Oeschger interstadial periods and high levels in colder periods is superimposed on a long-term trend encompassing a further factor of Within climate periods, there is only a very weak relationship between Ca concentration and temperature. Previous authors [Biscaye et al., 1997] have suggested that the most likely source for the increased dust is eastern Asia. For the first time, we consider each possible cause of both rapid and slow increases from source to deposition. We suggesthat, to account for the size and rapidity of the fast changes, significantly higher wind speed in the source area is required, although changes in atmospheric residence time could also play a role. For the slower long-term variability, changes in transport speed or, possibly, route are probably also involved. Changes in the size of the source area could give some change on longer time periods. The probable importance of changes in source area wind speed, almost simultaneous with Greenland temperature changes, confin'ms that climatic parameters in high and low latitudes were strongly coupled through the atmosphere during glacial climatic changes. This adds to evidence that the atmosphericirculation system underwent almost instantaneous large-scale changes during the last glacial period. 1. Introduction Atmospheric dust concentrationshowed dramatic increases worldwide in the last glacial period. For example, large increases are seen in Antarctic ice cores [Petit et al., 1981 ]; ice cores from South America [Thompson et al., 1995]; ocean sediment cores from the North Pacific, Indian, and North Atlantic oceans [Rea, 1994]; and various ice cores from Greenland [Hammer et al., 1985; Hansson, 1994; Mayewski et al., 1994]. The importance of mineral aerosol in climate forcing in present conditions has been recognized and assessed [Sokolik and Toon, 1996; Tegen et al., 1996]. The climatic impact of ice age aerosols has been estimated in the past, and dust was believed to have contributed to the cooling [Harvey, 1988], but the conclusions were challenged [Anderson and Charlson, 1990]. A more recent assessment suggested that high dust concentrations would have Now at Ionworks, Houston, Texas. Copyright 1999 by the American Geophysical Union. Paper number 1999JD /99/1999JD ,043 caused regional warming downwind of sources [Overpeck et al., 1996]. However, an improved appraisal of dust forcing for past climate states will require that climate models incorporate the increased levels and regional distribution correctly, which in turn requires a better knowledge of the causes of increased atmospheric concentrations recorded in the paleorecords. Similarly, iron fertilization of surface waters [Martin andfitzwater, 1988], which has been suggested to have a role in increased CO2 uptake in the last glacial period, cannot be assessed correctly until the causes of increase dust deposition to the ocean are understood. The evidence for large increases in dust concentration in the polar regions is particularly strong, and it has proved difficult to satisfactorily explain it. Atmospheric general circulation models (GCMs) have, until now, been unable to reproduce the dust increase [Joussaume, 1993; Andersen and Genthon, 1996]. Only one recent modeling exercis that used biosphere, dust source, and transport models together has managed to produce an order of magnitude increase in dust flux to Greenland in the Last Glacial Maximum (LGM) [MahowaM et al., 1999]. Large increases the concentrations of dust in Greenland have been discussed before, and most scientists have chosen to ascribe them to a single cause. Early studie suggested that the increased dust concentration in glacial periods was due to increased areas of loess and newly exposed continental shelves [Cragin et al., 1977;

2 31,044 FUHRER ET AL.' DUST IN GRIP ICE CORE Hammer et al., 1985]. The possible influence of stronger winds and atmosphericirculation was acknowledged [Delmas and 2. GRIP Ca Record Legrand, 1989] but not explored in more detail. The influence of Ca was measured continuously on the Greenland Ice Core atmospheric lifetime of aerosol (due to changes in precipitation) Project (GRIP) ice core (Summit, Greenland) from 1300 m to the was also considered using simple models [Hansson, 1994; Yung et al., 1996]. More recently, the concept of a polar circulation index (PCI) was introduced [Mayewski et al., 1994]. The PCI is a silty ice at 3021 m. The method used a continuous flow melting device and an absorption technique down to 2280 m, giving an estimate depth resolution of 35 mm [Fuhrer et al., 1993]. Below statistical parameter that emerges from empirical orthogonal 2280 m (an age of- 40 kyr), detection was by a fluorescence function (EOF) decomposition of Greenland ice core chemical data as the dominant EOF. Because it explains a high degree of variance of chemical species that have different sources, it must describe a large-scale feature of Northern Hemisphere climate and is considered to represen the size and intensity of the polar technique, with estimated resolution of 10 mm. At these resolutions each data point represents <1-3 years. Records from the GRIP core for many parameters agree well with those from the nearby Greenland Ice Sheet Project (GISP)2 core for the top 90% of the core (representing 110 kyr), but the two circulation cell. Thus, in cold periods, when dust concentrations records diverge strongly below this. In this paper, we refer only to are high, it is considered that a more vigorous polar circulation extends to lower latitudes entraining additional dust as well as other sources [Mayewski et al., 1994]. The PCI is a useful concept, but at the mechanistic level it includes a number of source areas and processes. the last 100 kyr, where it is clear that disturbed stratigraphy has not affected the core [Alley et al., 1995b]. The Ca record from GRIP (Figure 1) generally agrees well with the less detailed record from the GISP2 core [Mayewski et al., 1997]. The general features of the record are also seen in the In summary, a large number of possible causes for increased published GISP2 insoluble dust record [Ram and Koenig, 1997]. dust in Greenland in the last glacial period have been considered, It has been shown that Ca is well correlated with insoluble dust but there has been no systematiconsideration of the importance content in discrete samples from the GRIP core [Steffensen, 1997], of each. Few studies have considered the detailed record with time, instead discussing the difference between the "dusty" glacial and the cleaner interglacial. Most of these studies have been unable to look at the detailed relationship between species within glacial periods because of low data resolution or incomplete data although the constant of proportionality may differ between climatic periods. Where dust mass and Ca have been compared, [Steffensen, 1997] it appears as if the dust:ca ratio is, if anything, higher at higher concentrations of each, suggesting that the relative changes with time in the dust concentration are actually greater sets; this is information which might allow different mechanisms even than those of Ca reported here. We therefore treat the Ca to be distinguished. In this paper, we dissect the timescales of dust changeseen in Greenland over the last 100,000 years, relate dust changes particularly to the oxygen isotope temperature proxy, and consider the full range of factors influencing the concentration. record as a proxy for the amount of terrestrial dust reaching the ice sheet, while realizing that noncalcareous dust may also be present. There are insufficient published quantitative dust data to carry out a similar analysis using dust mass itself. Although records from Age / kyr B.P. Figure 1. Ca and oxygen isotope record from the Greenland Ice Core Project (GRIP) ice core for the last 100 kyr. The depths have been converted to a linear timescale on the basis of annualayer counting back to years B.P. and ice flow modeling beyond that [Dansgaard et al., 1993]. Data are shown here with a 100 year resampling so that variability can be compared along the core. Hol is Holocene, and YD is Younger Dryas. Interstadial (IS) numbers marked on the oxygen isotope profile refer to the mild periods within the glacial period and have been defined previously [Dansgaard et al., 1993]. IS 1 is the Allerod/Bolling warm stage, while IS 21 is generally identified as marine isotope stage 5a. To avoid adding extra terminology, we refer in the text to the cold periods by their surrounding IS numbers (thus cold 5/6 is the cold period between IS 5 and IS 6).

3 FUHRER ET AL.: DUST IN GRIP ICE CORE 31,045 other Greenland sites do not have the resolution needed to see all the interstadial events in detail, it is clear that the main features of the GRIP Ca record are seen throughout Greenland [Cragin et al., 1977; Hammer et al., 1985; Hansson, 1994]. A number of featurestand out in the Ca record. (1) There is a huge dynamic range (from typical Holocene concentrations of 0.2 pm to maxima of >20 pmjust before 20 kyr B.P.). (2) There is a very close correspondence with major shifts in oxygen isotope ratio (assumed to represent temperature change). In particular, the mild Dansgaard-Oeschger (DO) events of the last glacial period (previously numbered as interstadials (IS) [Dansgaard et al., 1993]) are recorded equally clearly as periods of low Ca concentrations. (3) Changes in Ca concentrations are extremely rapid [Fuhrer et al., 1993] Rapidity of Ca Changes The third point is clearly illustrated in Figure 2. Particularly during warmings, the Ca concentration often reduces by an order of magnitude in about a decade, with 50% reductions within 2-3 years. As an example, the warming into IS 3 (Figure 2a) is seen in the oxygen isotope record as occurring over-1 m (40 years); it is not clear that the isotopichange has occurred until the Ca reduction that accompanies it is complete. The Ca change, in contrast, occurs in two discrete steps (shown by vertical lines in Figure 2a), lasting <4 (2 years) and 6 cm (3 years), respectively. After the first step, Ca is already at a concentration lower than any seen in the previous 600 years, so thathe outside observer would see an unprecedented and irreversible change occurring within 2 years. Any mechanisms for this type of change must be capable of producing such rapid shifts. Coolings are rather slower. Taking as an example the cooling at the end of IS 5 (Figure 2b), the Ca increase that accompanies it takes a total of 4 m (-150 years). However, even here there are two distinct steps (marked by vertical lines), each lasting only years. In the first of these, Ca again reaches levels higher than any seen for the previous 400 years Ca Relationship At first sight, Ca and 8 O show a good logarithmic correlation Figure 2. Ca (solid line) and b 80 (dashed) across four transitions. (Figure 3a), as already noted by previous authors [DeAngelis et al., (a) Start of IS 3, (b) end of IS 5, (c) end of IS 8, and (d) start of IS 1997; Yiou et al., 1997]; the best fit line through all the data would 20. At the warmings, Ca concentrations reduce to interstadial suggest a 50% increase Ca concentration for a 1%o decrease in values in two or three jumps (marked by verticalines), each 8 O. This has been taken to imply that dust increases are linked occurring 2-3 years. The Ca increases at the two coolings with changes in temperature, either through changes in ice sheet (Figures 2b and 2c) are more gradual but still include some rapid size or through changes in circulation patterns. jumps that are not reversed until the next interstadial. However, this simple picture masks the fact that different sensitivities apply within a climate period (Figures 3b-3d and 4). The Ca-8 relationship is rather weak within each climate period, but there is a very strong jump in Ca from mild to cold periods. and Ditlevsen, 1997] it was stated that there was a strong This same feature was found in the relationship between snow relationship between calcium and temperature over timescales accumulation rates and temperature Summit [Kapsner et al., >200 years. However, this disguises the true nature of the 1995]. For instance, both the Holocene and the cold period 2/3 relationship, which is seen to be not with temperature but with (2/3 denotes between IS 2 and IS 3), the Ca-8 sensitivity is just shifts in the entire climate regime, which in fact, occurs over very 10%/%0 (Figure 3b), but for a given 8 value, Ca is 28 times higher short time periods but at millennial or greater intervals. in 2/3 than in the Holocene. The variability in b O within a Although Ca switches rapidly between a "high" and "low" climate period is comparable to the typical cold-mild shift in b O; concentration state, the average concentration is not the same for for Ca the variability within a climate period is small (even on a all warm periods nor is it the same for all cold periods. Through logarithmic scale) compared to the shift seen between cold and the last 100 kyr, there is a pattern of Ca concentration in successive mild periods. For example, in Figure 3c, there is considerable cold periods, with generally low Ca values before-80 kyr, and a overlap in values of 15 O between the interstadial (IS 8) and the maximum-20 kyr (LGM)(Figure 5). The warm periodshow a adjacent cold period but no overlap in Ca values. As a result, Ca is similar pattern, although it is more subdued until---50 kyr B.P. actually a much more distinct indicator of switches in the climate In summary, the dynamic range of 100 in the Ca concentration state. In an earlier time series analysis of the same data [Marsh can be seen to be composed of (1) a variability (factor of-1.5-2), lo 8 2 o o o Depth / m

4 31,046 FUHRER ET AL.' DUST IN GRIP ICE CORE 10 residence time)' and (6) deposition in Greenland. In sections 4-6 the importance of these factors in the rapid and the long-term changes will be assessed Relationship Between Ice Concentration and Atmospheric Concentration x x ß ß Chemical deposition to ice sheets is made up of wet and dry deposition components. The relativefficiency of wet processes will depend on factors such as cloud type and height and the chemical and physical characteristics of aerosol particles. The relativ efficiency of dry processes will depend on particle sizes and on meteorological and snow surface properties. The relative importance of wet and dry deposition depends on the snow accumulation rate. If dry deposition is important, then a reduction in snow accumulation rate leads to an increase in snow concentration in the absence of any changes in atmospheric concentrations or other factors. Since snow accumulation rate also changes rapidly during climate shifts [Kapsner et al., 1995], a part of the change in Ca concentration could result from the dilution effect, reflecting the changing precipitation rate rather than changing Ca concentration in the atmosphere. However, in present-day circumstances, Ca deposition at Summit is dominated by wet deposition; for summer 1993 it was estimated that 76% of Ca was deposited in either snow or fog deposition and only 18% as dry deposition [Bergin et al., 1995]; even with a threefold reduction in accumulation rate in the Last Glacial Maximum [Kapsner et al., 1995], wet deposition would still be the largest contributor. Any quantitativ estimate would be highly uncertain given that the present-day data are limited in time and that scavenging ratios could have altered in earlier periods for other reasons, but the overall conclusion that wet deposition is more important over most of the record is robust. In these circumstances, snow concentration rather than deposition flux is the better measure of atmospheri concentration trends ( 180/ %0 Figure 3. Scatterplots of Ca concentration versus 6]80. Ca data have been binned to the highest resolution available for the oxygen isotope data. This resolution was 27.5 cm from 1300 to 1500 m, cm between 1500 and 1900 m, cm from 1900 to 2694 m, and 3.44 cm below that. The equivalentime resolution varies from 2 to 18 years. (a) The entire data set from 1300 to 3021 m. (b) Data from the Holocene (squares) and cold 2/3 (crosses) only. (c) Data from IS 8 (squares) and the adjacent cold 8/9 (crosses). (d) Data from IS 20 (squares) and cold 19/20 (crosses) only. r k... x-----x 20/21 - x... 12/13 10 "'."'... x-... x 11/12 > - :..L-' ' :::.-:-.-.--'-' t::: x... x > " ' " :"- -- x... x 213 ' 1 i x--- _.... YD... _ _ _-N,,.---x... IS ' IS12... IS1 "-_-'.- '.' -.. :..." -... Hol "-" 4' ' "'"' --- slightly related to 5 80, within climate periods; (2) large and rapid jumps (factor of even for apparently similar temperatures) between mild and adjacent cold periods; and (3) a long-term trend 0.1 (factor ) through the complete climate cycle between successive cold or successive mild stages. To explain the changes, we will discuss qualitatively the factors ( 180/ %o that control the delivery of Ca to Greenland. These are illustrated schematically in Figure 6, and are (1) size and location of sources, Figure. 4. Best fit lines from regressions of log Ca versus ] O which can expand or decrease in size and can be located favorably ratio for different climatic periods. In each case, a meter or two at or unfavorably for transport to Greenland; (2) mobilization of clay- transitions has been discarded before the regressions were carried size particles from the source; (3) uplift of particles to higher levels out. The lines are centered on the average ] O for the period and of the atmosphere; (4) strength and location of transport pathways extend +2 rs on either side. The solid line is the regression for the for long-range transport; (5) deposition en route (affecting entire data set.

5 FUHRER ET AL.: DUST IN GRIP ICE CORE 31, Age / kyr Figure 5. Average Ca (solid lines) and estimated Ca at b 80=-38 %o(dashed lines, based on regressions of log Ca versus b 80) for (top) each cold period and (bottom) each IS. For each of the 22 IS events (plus the Holocene) and for each intervening cold period, the average Ca and the estimate for O=-38 %o(from the regression) are plotted. Some meters of ice at each transition were excluded from either the warm or cold periods for this analysis. [Alley et al., 1995a]. On the basis of measured fluxes in individually dated years, Alley et al. [1995a] concluded also that wet deposition has dominated in the Younger Dryas and that changes in snow concentration between the Younger Dryas and the Holocene overestimate changes in atmosphericoncentration only by -20% (with a large uncertainty). We therefore conclude that atmospheric dust concentrations over Greenland changed by similar factors and at similar speeds to concentrations seen in ice cores. The effect of this assumption is included as one of the factors accounting for temporal changes in our analysis in section 4.4. For the LGM, mineralogical and isotopic analysis of dust in the GISP2 core suggests east Asian deserts as the main source [Drab, 1998] and tends to rule out North America and the Sahara. Higher kaolinite/chlorite ratios in interstadials than in cold periods [Biscaye et al., 1997; Maggi, 1997] suggesthat the source shifted southwar during interstadials. Although not entirely settled, these findings imply that we should look to changes in delivery of Asian dust to Greenland for most of the variability in the GRIP Ca record. One further source region we cannot rule out is the continental shelf areas of north Asia and Beringia that became uncovered when sea level was lower. It has been suggested that these areas were covered in a soil that may itself have derived from east Asia (therefore isotopically indistinguishable from Asian dust) and that could have been an active source for dust [DeAngelis et al., 1997]. On the other hand, at least the land bridge part of Beringia seems to have been well vegetated at the LGM [Elias et al., 1997]. A recent modeling exercise that included a biosphere model to generate dust source areas agreed that Asia was a significant source for Greenland in the LGM (50%) but identified northern Siberia and the Canadian Arctic as important also [Mahowald et al., 1999]. For the rest of this paper, we will write as if Asia was the main source for increase dust, while bearing in mind that many of the mechanisms would be similar for other sources. 4. Stadial-Interstadial Changes We consider first the rapid fivefold to tenfold changes occurring between cold periods and intersradials. Often the variability of the ao signal makes it hard to decide the phasing between Ca and 8 ao at each transition (Figure 2). In some warmings, Ca drops to low concentrations a few years before 8 ao appears to change, but the opposite feature also exists. This implies that b 80 and Ca changes are not causally related, although the same phenomenon provokes a change over decades or less to the influx of warmer, wetter, less dusty air. Coolings are often slower (several decades), 3. Source Area for Dust in the Holocene and Glacial Periods Asia [Biscaye et al., 1997], North America [DeAngelis et al., 1997], and the Sahara [Mosher et al., 1993] have all been suggested as potential dust sources to Greenland in the Holocene. Our data suggest that North America is not a dominant influence in causing higher concentrations during the glacial period: Ca is obviously not affected by the large-scale changes in surface area with the retreat of the Laurentide ice sheet during the early Holocene; instead, it reaches low Holocene concentrations immediately after the transition from the Younger Dryas. We also note that there was no increase in Ca concentration in Summit snow during the dust bowl years of the 1930s [Mayewski et al., 1993], suggesting that there is no strong connection between southwestern North America and Greenland. Ten-day back trajectories calculated from Summit for the last 44 years suggest that the vast majority of 700 and 500 hpa trajectories arrive from the west, which implies that the Sahara has minimal influence [Kahl et al., 1997]. As many as 67% of the 500 hpa trajectories in winter originate as far back (to the west) as Asia or Europe [Kahl et al., 1997]. The trajectories from the west travel over North America dominantly north of 50øN, again suggesting that likely dust sources in North America are not well connected to Greenland. port westerlies over pole ' ' '" deposition en route uplift to transport levels f by mobilization wind source source area deposition to ice ice core Figure 6. Schematic diagram of factors affecting dust concentrationseen in ice cores (see text).

6 31,048 FUHRER ET AL.: DUST IN GRIP ICE CORE Table 1. Possible Factors Contributing to Increased Calcium in the GRIP Ice Core Over Different Timescales Timescale Long-Term StadiaMnterstadial Within Climate Maximum Pattern State Change Example LGM/early cold Cold stage/ Early pre-boreal/ Coldest LGM/ stage adjacent mild mid-holocene warmest stage Holocene Factor Source area <2 a < 1.5 a 1 a <2 a Source wind speed Uplift to transporting level Transporting wind speed NE NE 1 a NE a a 1 a 2 a Transport route NE 1 1 b NE (northern branch jet) Losses en route (residence 1 a 2 a 1.4 a time) 3 a Accumulation rate (dilution 1 b in core) Observed 5-10 u 5-10 b 1.5 u 100 u NE is not evaluated. Confidence of our estimates is moderate. Confidence of our estimates is high. For unmarked estimates, confidence is low. particularly in the temperature record. The sparser data that exist for sea salt [DeAngelis et al., 1997] show a smaller (though still rapid) change in these components at the rapid transitions (typically a factor of 2 compared to 5-10 for Ca) Source Area The transitions are far too rapid to be explained by changes in the exposure of continental shelves (caused by sea level changes). This is especially true during coolings, when unfeasibly rapid buildup of ice volume would be required to uncover any significant area of shelf in years to decades, as the calcium data require (Figure 2). Changes in aridity of source areas could potentially be fast. These affect dust flux mainly through changes in the threshold velocity for dust mobilization. Existing arid area sources already have low threshold velocities, and a further reduction in precipitation may even act to reduce dust storm frequency [Rea, 1994]. A conversion of semiarid into arid areas would cause a decrease in the threshold velocity for bare soils and a possible sharp change through loss of vegetation. However, the present size of semiarid areas relative to arid areas in Asia [Pye, 1987] seems insufficient to allow changes of several hundred percent in dust uplift through changes in area alone. We also note (Figure 2) that during warmings, factor of 2 decreases in Ca occur in 1-2 years. It seems unlikely that the appearance of vegetation (reducing threshold velocities) over huge areas could be so instantaneous. In Table 1 we have tried to give very approximate quantitative limits to the changes in Ca (dust) from various causes. We consider a 50% increase or decrease in source area to be the maximum realistic value for the years to decades timescale of these changes Dust Mobilization and Uplift More efficient mobilization of dust would require stronger wind speeds at the Asian source area. The amount of dust mobilized is dependent on U 3 or even U 4 (where U is wind speed) above the threshold speed [Gillette and Passi, 1988], so that small changes in wind speed could have a very large effect. Dust storms are rather episodic, so that the strongest wind speeds have an excessive impact; it is probably small changes in maximum wind speeds that would be important. Past wind speeds at the source might be available from paleoclimatic data near the source and from GCM model runs. Grain sizes from Chinese 1oessequences [Porter and An, 1995], interpreted as indicators of winter monsoon strength in the Asian source region, show high-frequency (millennial) structure. Changes in diameter of the order of 30% are seen for faster changes that may be connected to millennial-scale climate change seen elsewhere, and increases of 100% are seen between the Holocene and the LGM [Porter andan, 1995]. These changes imply substantial changes in wind speed. There is no accepted method for converting these diameters into wind speeds, but we note that a 20% increase in wind speed would be sufficient to double the dust mobilization, while a doubling of speeds would increase mobilization by an order of magnitude. Although surface winds do tend to be increased during the LGM in most models,

7 FUHRER ET AL.: DUST IN GRIP ICE CORE 31,049 they cannot generally capture local changes very precisely and have been used mainly to study LGM-Holocene changes, not stadial-interstadial changes. In Table 1 we assume a 20-30% increase in wind speed at the source to calculate a twofold to threefold change in dust mobilization across rapid transitions. To achieve long-range transportoward Greenland, dust must be raised up to higher levels in the atmosphere. This is most easily achieved by cyclonic activity. In the present day, vertical motion of dusty air is often associated with deep depressions moving across China from the west, mainly in spring [Pye and Zhou, 1989]. Although the tracks and frequency of these depressions would probably have changed as other aspects of circulation changed during rapid transitions, we do not feel we have sufficient information to estimate the sign or magnitude of this effect. particle size that can be transported long range is dependent on the transporting wind speed (square root dependence). In practice, we can expect a more complex dependence, especially along such a long transport route on which the speed certainly changes with longitude. Nonetheless, we can use the particle sizes measured in ice cores to get a qualitative indication of changes. From the limited number of particle sizes available for dust in Greenland ice, it is found that sizes in cold periods are somewhat larger than those in the adjacent warm periods [Steffensen, 1995, 1997]. This could be taken to imply a 20-50% increase transporting wind speed in cold periods compared to warm, which is qualitatively consistent with the model findings discussed above [Fawcett et al., 1997], and is reflected in the value used in Table 1. In the above discussion, transport in the jet is assumed, and the geographically more direct route over the pole is not considered Transport and Residence Time Long-range transport of dust is achieved at several kilometers altitude in at least the lower levels of the westerly jet stream [Pye and Zhou, 1989; Duce, 1995]. Some atmospheric GCM (AGCM) reconstructions of the LGM suggesthat the westerly jet split into two branches around the North American ice sheet [Kutzbach et al., 1993], although this split may not be present [Sloan et al., 1995] when newer, lower reconstructions of the Laurentide ice In the present-day atmosphere, analysis of 1 O-day back trajectories [Kahl et al., 1997] shows that few trajectories at 500 hpa (3-7%, depending on season) follow a polar route from Asia, while as many as 58% in winter follow a zonal route from East Asia. At 700 hpa (close to the altitude of the Asian sources and the Greenland ice sheet), the zonal route also dominates (though trajectories are shorter). More detailed model calculations would be needed to assess whether the polar route could be more important during the glacial period. sheet [Peltlet, 1994] are used. The weaker northern branch of the jet would be a particularly efficient route for delivering cold, dry 4.4. Greenland Conditions air containing Asian dust to Greenland, and the rapid splitting and closing of the northern jet has been suggested [Stuiver et al., 1995; DeAngelis et al., 1997] as a contributor to the rapid changes. However, assuming the split is controlled mainly by the altitude of the Laurentide ice sheet, it is hard to see why any switch for this branch would react so rapidly and sensitively and throughouthe 100 kyr, despite large variations over this timescale in the size of the Laurentide. Without further evidence we consider the rapid splitting and joining of the jet stream to be unlikely. Nonetheless, the strength and position of the jet stream can be expected to alter. Changes in accumulation rate for the most recent transitions have been ascribed to changing storm tracks [Kapsner et al., 1995; Fawcett et al., 1997] and show that some significant change in circulation must have occurred. GCM experimentshow that switching off Nordic Sea heat flux (to simulate reduced North Atlantic Deep Water (NADW) formation, while leaving the ice sheets unchanged) both moves and strengthens the storm tracks (and therefore, presumably, the associated jet stream) over North America [Fawcett et al., 1997]. This is to be expected since large polar temperature changes lead to greater latitudinal temperature gradients. In fact, the efficiency of long-range transport should be a balance of two factors: the travel time between the source and Greenland and the residence time of aerosol. Transport is enhanced if transit times are shorter (as they should be with a faster jet) and if residence times are longer. Precipitation rates in Greenland were certainly lower in cold periods and changed very fast [Kapsner et al., 1995]. Assuming that the lower accumulation rate applied to the dust-transporting air masses en route to Greenland, this would lead to increased residence times and therefore increased dust concentrations even in the absence of other changes [Hansson, 1994]. With a very simple physical model [Hansson, 1994] and reasonable present-day transit and residence Finally, as discussed in section 2.3, the lower snow accumulation rate could also have affected the concentrations seen at Summit through an increased relative importance of dry deposition. However, studies of fluxes in individual annualayers [Alley et al., 1995a; DeAngelis et al., 1997] suggesthat wet deposition was dominant even in the LGM and that the direct effect of changing accumulation rates was relatively small (up to 30%) Summary of Rapid Changes We have not considered possible more specialized causes, such as changing seasonal snow cover at the source or change in relative proportion of (dustier) spring snowfall at Summit. In fact, the model calculations that simulated a switch off of NADW [Fawcett et al., 1997] found that the reduction in precipitation was mainly in winter and spring. Assuming that dust input was still at its maximum in spring as in present-day snow, this would work in the wrong direction for explaining the dust changeseen in Greenland, since the seasonality would bias the ice core average toward the less dusty summer period, thus reducing the average dust concentration. Our conclusion is that the factors likely to be significant for the rapid dust changes (factor 5-10) were a combination of increased residence times (possibly affecting also sea salt, although the presumed Atlantic source for these suggests no direct linkage) and increased dust mobilization at the Asian source (a factor of 3 would require wind speeds increased by 30-40%). These two factors would clearly be linked (and therefore synchronized) by changes in circulation patterns throughout the high northern latitudes; for example, GCM simulations ofthe LGM with reduced North Atlantic sea ice cover show more northerly storm tracks in the Atlantic and changed pressure distributions over central Asia [Kutzbach et al., 1993]. Table 1 summarizes our estimates. times, it seems possible that a factor of 2 decrease in precipitation en route (and hence a reduction in wet deposition) could lead to a factor of 2 increase in concentration reaching Greenland. Faster transporting winds would also give an increase in concentration Greenland. All other factors being equal, the 5. Long-Term Pattern The shape of the longer-term changes in dust concentration (Figure 5) seems to be reflected in other records, such as Antarctic

8 31,050 FUHRER ET AL.: DUST IN GRIP ICE CORE dust [DeAngelis et al., 1997] and Pacific and Indian Ocean sediment cores [Rea, 1994]. The shape is similar to that of changes in global ice volume. Increased exposure of continental shelves is unlikely to be a factor, given the mineralogical and isotopic evidence for an Asian source [Biscaye et al., 1997] and that the sea-salt components [DeAngelis et al., 1997], unaffected by changes in dust sources, show a long-term trend of similar shape and magnitude, particularly for the sequence of interstadials. This also argues againstrends in aridity as the major factor. Evidence from lake levels has previously been used to argue against greater aridity in the dust-producing area of central Asia in the LGM [Phillips et al., 1993]. However, the pattern of aridity is certainly geographically variable with greater aridity in some areas at or just after the LGM [Pachur et al., 1995; Sun andding, 1998], and it is quite possible that there were shifts in desert areas into areas with a different atmospheric circulation regime. The long timescales here do not rule out changes in aridity at a mechanistic level, so in Table 1 we put an upper limit on source area changes based on an estimate of the semiarid area potentially available. If the newly arid areas were particularly sensitive (low threshold velocity or strong winds), then synergism between factors could lead to a greater than linear response to area changes. The global ice volume is dominated by changes in volume of the Laurentide ice sheet. The height of the Laurentide appears to control the strength of the proposed northern branch of the jet stream [Kutzbach et al., 1993]. The size and height ofthe ice sheet is also important in controlling the latitudinal temperature gradient, which should determine the overall strength and location of the transporting winds at -500 hpa. These factors would affect the concentrations of dust in Greenland through either or both of (1) provision of a more direct route to Greenland through the northern jet stream if it existed and (2) shorter transit times (and therefore less removal en route) through faster wind speeds in the main southern branch of the jet. These changes would affect the concentrations in both warm and cold periods in similar ways through the 100 kyr period, which is what we observe in the data (Figure 5). The strengthened northern branch cannot easily explain increase dust flux to the Pacific: although the location of the jet over the Asian source could also be affected, evidence from northwestern Pacific sediment cores suggested that the latitude of dust flux maximum remained fixed over the last 30 kyr [Rea, 1994]. Because more recent models do not have a split jet, we do not assign an increase to this cause, although our argument against rapid splitting does not apply in this case. However, a faster jet stream is to be expected. For the strengthened transporting wind, again using a simple model [Hansson, 1994], a factor of 2 increase in transit time (as predicted by GCMs between the LGM and the Holocene [Kutzbach et al., 1993]) could only cause about a factor of 2 decrease in dust concentrations in Greenland and far smaller over the Pacific. The limited GRIP dust particle-size data [Steffensen, 1997] show little evidence for even this magnitude of change in transporting wind; the mode particle radius in the Holocene and in IS 3 is very similar, while that in the LGM is larger than that in the Younger Dryas, but only large enough to support a 40% increase in wind speed, not a doubling. The range given in Table 1 reflects this uncertainty. In Greenland, there is little further change in precipitation rate between the Younger Dryas (considered a mild stadial) and the LGM (a cold stadial) [Kapsner et al., 1995], and therefore no direct accumulation rate effect in Greenland. In the absence of other evidence about precipitation rates elsewhere, we have assumed that the residence time for aerosol does not change at these longer time periods. A further enhancement of wind speeds at the sources is implied by the loess grain size record [Porter andan, 1995]. In summary (Table 1), it appears that the long-term changes are due to a combination of further wind speed changes at the source and changes in transporting wind speed. We also cannot rule out changes in source area as a contributor at this timescale. 6. Changes Within Climate Periods The Ca-b180 sensitivities within individual climate periods are low and very similar for cold and warm periods, suggesting that the same mechanism at play. Close inspection of the Holocene data suggests that the relationship between Ca and b180 disappears completely over shortimescales and that the correlation is actually controlled by century-scale trends in both parameters (e.g., the warming during the pre-boreal). Among the factors involved could be changing proportions of winter snow or changes in accumulation rate. The change in accumulation rate in the pre- Boreal [Kapsner et al., 1995] is not large enough to fully explain the correlation through an increased proportion of dry deposition. However, if it reflects lower precipitation rates en route, it could sufficiently alter the residence time of material travelling to Summit to give the necessary change in Ca concentration. We do not have the information to extend this very tentativ explanation to earlier periods at present. 7. Implications It is clear that several mechanisms are needed to account for the changes in dust seen. Our argumentsuggest that wind speed at the source, along with factors associated with long-range transport, are likely to be dominant causes of changing dust concentration; changes in source area can only contribute significantly to a 100- fold increase if there are nonlinear synergisms with other factors, such as wind speed changes. Change in deposition efficiency in Greenland is unlikely to be a major factor. Detailed work is needed to confirm our tentative conclusions: the Asian origin of the dust needs to be confirmed; the role of changing hydrology (residence time) might be estimated by looking at the phasing of dust and accumulation rate changes within the GRIP or GISP2 core, and it may be possible to correlate the Chinese 1oess records with the GRIP and GISP2 dust profiles in a more convincing way. Further insights into the causes of the dust changes will require improved modeling experiments. In a recent study, sensitivity tests were carried out to determine which factors were capable of giving the observed LGM dust increases; a particular conclusion was that only large increases in dust production in eastern Asia could account fbr the paleodata, although none of the modeled factors gave a satisfactory explanation for the cause of such an increase [Reader et al., 1999]. This suggests that only with finer grids and with better coupling of different model components will a satisfhctory simulation be achieved. Our conclusions underline the need for dust modelers to review the parameterizations of dust uptake and to ensure that the transport and wet deposition in the models lead to realistic transit and residence times. Such studies will lead to a better understanding of why models do not reproduce the observed increases. Model sensitivity studies [Reader et al., 1999] could also set more realistic limits on some of the values we have estimated in Table 1. If we have correctly identified the relevant factors for the rapid changes, then we find that substantial changes in surface wind speed in Asia must have occurred almost simultaneously with temperature changes in Greenland. Atthe very least, some important aspect of conditions at the source has to be

9 FUHRER ET AL.: DUST IN GRIP ICE CORE 31,051 very tightly coupled to Greenland climate. This is strong evidence that the atmosphericirculation was capable of undergoing instant large-scale reorganization, affecting both Greenland and east Asia. Model simulations are needed to show which factors are really capable of triggering an almost instant reorganization of atmosphericirculation, affecting both the North Atlantic and east Asia. Finally, if the relative importance of different mechanisms can be quantified, it will allow models to give realistic estimates of the global distribution of dust changes. This is needed to quantify the effects of dust on radiation and on iron input to the oceans. Acknowledgments. This work is a contribution to the Greenland Ice Core Project (GRIP), a European Science Foundation program with eight nations (Belgium, Denmark, France, Germany, Iceland, Italy, Switzerland, and United Kingdom), and the EU, collaborating to drill through the central part of the Greenland ice sheet. We thank Richard Alley, Natalie Mahowald, David Peel, Bernhard Stauffer, and many other colleagues for helpful discussions based on earlier versions of this manuscript. References Alley, R. B., R. C. Finkel, K. Nishiizumi, S. Anandakrishnan, C. A. Shuman, G. R. Mershon, G. A. Zielinski, and P. A. Mayewski, Changes in continental and sea-salt atmospheric loadings in central Greenland during the most recent deglaciation: Model-based estimates, J. Glaciol., 41, , 1995a. Alley, R. B., A. J. Gow, S. J. Johnsen, J. Kipfstuhl, D. A. Meese, and T. Thorsteinsson, Comparison of deep ice cores, Nature, 373, , 1995b. Andersen, K. K., and C. Genthon, Modeling the present and last glacial maximum transportation of dust to the Arctic with an extended source scheme, in Environmental Science and Technology Librat3, Vol. 11, The Impact of African Dust Across the Mediterranean, edited by S. Guerzoni and R. Chester, pp , Kluwer Acad., Norwell, Mass., Anderson, T. L., and R. J. Chaffson, Ice-age dust and sea salt,nature, 345, 393, Bergin, M. H., J.-L. Jaffrezo, C. I. Davidson, J. E. Dibb, S. N. Pandis, R. Hillarno, W. Maenhaut, H. D. Kuhns, and T. Makela, 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, Biscaye, P. E., F. E. Grousset, M. Revel, S. VanderGaast, G. A. Zielinski, A. Vaars, and G. Kukla, Asian provenance of glacial dust (stage 2) in the Greenland Ice Sheet Project 2 ice core, Summit, Greenland, J. Geophys. Res., 102, 26,765-26,781, Cragin, J. H., M. M. Herron, C. C. Langway Jr., and G. Klouda, Interhemispheric comparison of changes in the composition of atmospheric precipitation during the late Cenozoic era, in Polar Oceans, edited by M. J. Dunbar, pp , Arctic Inst. of N. Am., Calgary, Alberta, Canada, Dansgaard, W., et al., Evidence for general instability of past climate from a 250-kyr ice-core record, Nature, 364, , DeAngelis, M., J.P. Steffensen, M. Legrand, H. Clausen, and C. Hammer, Primary aerosol (sea salt and soil dust) deposited in Greenland ice during the last climatic cycle: Comparison with east Antarctic records, J. Geophys. Res., 102, 26,681-26,698, Delmas, R. J., and M. Legrand, Long-term changes in the concentrations W. B. Lyons, and M. Prentice, Major features and forcing of highof major chemical compounds (soluble and insoluble) along deep ice latitude Northern Hemisphere atmosphericirculation using a 110,000- cores, in Dahlem Konferenzen: The Environmental Record in Glaciers year-long glaciochemical series,j. Geophys. Res., 102, 26,345-26,366, and Ice Sheets, edited by H. Oeschger and C. C. Langway Jr., pp , John Wiley, New York, Mosher, B. W., P. Winkler, and J.-L. Jaffrezo, Seasonal aerosol chemistry Drab, E., Saisonalit, source et alteration de l'a rosol mineral archiv dans les glaces r centes de Summit (Groenland), Ph.D. thesis, 218 pp, Universit Paris VII, Paris, Duce, R. A., Sources, distributions, and fluxes of mineral aerosols and their relationship to climate, in Dahlem Workshop on Aerosol Forcing of Climate, edited by R. J. Chaffson and J. Heintzenberg, pp , John Wiley, New York, Elias, S. A., S. K. Short, and H. H. Birks, Late Wisconsin environments of the Bering Land Bridge, Palaeogeogr. Palaeoclimatol. Palaeoecol., 136, , Fawcett, P. J., A.M. Agustsdottir, R. B. Alley, and C. A. Shuman, The Younger Dryas termination and North Atlantic Deep Water formation: Insights from climate model simulations and Greenland ice cores, Paleoceanography, 12, 23-38, Fuhrer, K., A. Neftel, M. Anklin, and V. Maggi, Continuous measurements of hydrogen peroxide, formaldehyde, calcium and ammonium concentrations along the new GRIP core from Summit, central Greenland, Atmos. Environ., Part A, 27, , Gillette, D. A., and R. Passi, Modeling dust emission caused by wind erosion, J. Geophys. Res., 93, 14,233-14,242, Hammer, C. U., H. B. Clausen, W. Dansgaard, A. Neftel, P. Kristinsdottir, and E. Johnson, Continuous impurity analysis along the Dye 3 deep core, in Greenland Ice Core: Geophysics, Geochemistry and the Environment, Geophys. Monogr. Ser., Vol. 33, edited by C. C. Langway Jr., H. Oeschger, and W. Dansgaard, pp , AGU, Washington, D.C., Hansson, M. E., The Renland ice core: A Northern Hemisphere record of aerosol composition over 120,000 years, Tellus, Ser. B, 46, , Harvey, L. D. D., Climatic impact of ice-age aerosols, Nature, 334, , Joussaume, S., Paleoclimatic tracers: An investigation using an atmospheric general circulation model under ice age conditions, 1, Desert dust, J. Geophys. Res., 98, , Kahl, J. D. W., D. A. Martinez, H. Kuhns, C. I. Davidson, J. L. Jaffrezo, and J. M. Harris, Air mass trajectories to Summit, Greenland: A 44-year climatology and some episodic events, J. Geophys. Res., 102, 26,861-26,875, Kapsner, W. R., R. B. Alley, C. A. Shuman, S. Anandakrishnan, and P.M. Grootes, Dominant influence of atmospheric circulation on snow accumulation in Greenland over the past 18,000 years ature, 373, 52-54, Kutzbach, J. E., P. J. Guetter, P. J. Behling, and R. Selin, Simulated climati changes: Results ofthe COHMAP climate-model experiments, in Global Climates Since the Last Glacial Maximum, edited by H. E. J. Wright et al., pp , Univ. of Minn. Press, Minneapolis, Maggi, V., Mineralogy of atmospheric microparticles deposited along the Greenland Ice Core Project ice core, J. Geophys. Res., 102, 26,725-26,734, Mahowald, N., K. Kohfeld, M. Hansson, Y. Balkanski, S.P. Harrison, I.C. Prentice, M. Schulz, and H. Rodhe, Dust sources and deposition during the last glacial maximum and current climate: A comparison of model results with paleodata from ice cores and marine sediments, J. Geophys. Res., 104, 15,895-15,916, Marsh, N. D., and P. D. Ditlevsen, Observation of atmospheric and climate dynamics from a high-resolution ice core record of a passive tracer over the last glaciation, J. Geophys. Res., 102, 11,219-11,224, Martin, J. H., and S. E. Fitzwater, Iron deficiency limits phytoplankton growth in the north-east Pacific subarctic, Nature, 331, , Mayewski, P. A., L. D. Meeker, M. C. Morrison, M. S. Twickler, S. I. Whitlow, K. K. Ferland, D. A. Meese, M. R. Legrand, and J.P. Steffensen, Greenland ice core "signal" characteristics: An expanded view of climate change, J. Geophys. Res., 98, 12,839-12,847, Mayewski, P. A., et al., Changes in atmosphericirculation and ocean ice cover over the North Atlantic during the last 41,000 years, Science, 263, , Mayewski, P. A., L. D. Meeker, M. S. Twickler, S. Whitlow, Q. Z. Yang, at Dye 3, Greenland, Atmos. Environ., Part A, 27, , Overpeck, J., D. Rind, A. Lacis, and R. Healy, Possible role of dustinduced warming in abrupt climate change during the last glacial period, Nature, 384, , Pachur, H.-J., B. Wunnemann, and H. Zhang, Lake evolution in the Tengger Desert, Northwestern China, during the last 40,000 years, Quat. Res., 44, , Peltier, W. R., Ice age paleotopography, Science, 265, , 1994.

10 31,052 FUHRER ET AL.: DUST IN GRIP ICE CORE Petit, J. R., M. Briat, and A. Royer, Ice age aerosol content from east Steffensen, J.P., The size distribution of microparticles from selected Antarctic ice core samples and past wind strength, Nature, 293, 391- segments of the Greenland Ice Core Project ice core representing 394, different climatic periods, J. Geophys. Res., 102, 26,755-26,763, Phillips, F. M., M. G. Zreda, T.-L. Ku, S. Luo, Q. Huang, D. Elmore, P. W. Stuiver, M., P.M. Grootes, and T. F. Braziunas, The GISP26 80 climate Kubik, and P. Sharma, 23øTh/234U and 36C1 dating of evaporite deposits record of the past 16,500 years and the role of the sun, ocean, and from the western Qaidam Basin, China: Implications for glacial-period volcanoes, Quat. Res., 44, , dust export from central Asia, Geol. Soc. Am. Bull., 105, , Sun, J., and Z. Ding, Deposits and soils of the past 130,000 years at the desert-loess transition in northern China, Quat. Res., 50, , Porter, S.C., and Z. S. An, Correlation between climate events in the North Tegen, I., A. A. Lacis, and I. Fung, The influence of climate forcing of Atlantic and China during the last glaciation, Nature, 375, , mineral aerosols from disturbed soils, Nature, 380, , Thompson, L. G., E. Mosley-Thompson, M. E. Davis, P.-N. Lin, K. A. Pye, K., Aeolian Dust and Dust Deposits, 334 pp., Academic, San Diego, Henderson, J. Cole-Dai, J. F. Bolzan, and K.-B. Liu, Late glacial stage Calif., and Holocene tropical ice core records from Huascaran, Peru,Science, Pye, K., and L.-P. Zhou, Late Pleistocene and Holocene aeolian dust 269, 46-50, deposition in north China and the northwest Pacific Ocean, Yiou, P., K. Fuhrer, L. D. Meeker, J. Jouzel, S. Johnsen, and P. A. Palaeogeogr. Palaeoclimatol. Palaeoecol., 73, 11-23, Mayewski, Paleoclimatic variability inferred from the spectral analysis Ram, M., and G. Koenig, Continuous dust concentration profile of pre- of Greenland and Antarctic ice-core data, J. Geophys. Res., 102, Holocene ice from the Greenland Ice Sheet Project 2 ice core: Dust 26,441-26,454, stadials, interstadials, and the Eemian, J. Geophys. Res., 102, 26,641- Yung, Y. L., T. Lee, C.-H. Wang, and Y.-T. Shieh, Dust: A diagnostic of 26,648, the hydrologic cycle during the last glacial maximum, Science, 271, Rea, D. K., The paleoclimatic record provided by eolian deposition in the , deep sea: The geologic history of wind, Rev. Geophys., 32, , Reader, M. C., I. Fung, and N. McFarlane, The mineral dust aerosol cycle during the Last Glacial Maximum, J. Geophys. Res., 104, , K. Fuhrer, Ionworks, 2742 Bolsover Street, Suite 255, Houston, TX Sloan, L. C., K. Taylor, and E. Small, Last Glacial Maximum climate S.J. Johnsen, Geophysical Institute, University of Copenhagen, Juliane modelled with GENESIS v.l.02a, including computed sea surface Maries Vej 30, DK2100 Copenhagen 0, Denmark. (sigfus gfy.ku.dk) temperatures, Eos Trans. AGU, 76 (46), Fall Meet. Suppl., F284, E.W. Wolff, British Antarctic Survey, Natural Environment Research Sokolik, I. N., and O. B. Toon, Direct radiative forcing by anthropogenic Council, High Cross, Madingley Road, Cambridge CB3 0ET, England, UK. airborne mineral aerosols, Nature, 381, , (e.wolff bas.ac.uk) Steffensen, J.P., Microparticles and chemical impurities in ice cores from Dye 3, south Greenland and their interpretation palaeoclimatic (Received January 28, 1999; revised August 6, 1999; reconstructions, Ph.D. thesis, Univ. of Copenhagen, Copenhagen, accepted August 12, 1999.)