Greenhouse gas measurements along polar ice cores provide important insight into the former composition of the atmosphere, its natural variations, and the responses to fast climatic changes in the past. We present high-resolution nitrous oxide (N2O) and methane (CH4) records measured along two ice cores from central Greenland covering part of Marine Isotope Stages 3 and 4 in the last glacial epoch. The N2O data confirm the hypothesis that N2O shows variations in phase to fast climatic changes observed in the Northern Hemisphere, the so-called Dansgaard-Oeschger (D-O) events. Variations exist not only for events with a long duration (1500 years and more) but also for the shorter ones. The comparison with CH4 unveils interesting differences between the response of CH4 and N2O to D-O events. While the average amplitudes of CH4 oscillations associated with D-O events are similar to those of the Northern Hemisphere summer insolation, the magnitude of the N2O concentration change instead correlates with the duration of the D-O events. The records give further insight into the timing of concentration changes at the beginning of D-O events. They show that for long-lasting events the N2O concentration starts to increase before both the sharp increase in the CH4 concentration and the temperature reconstructed for Greenland.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Methane (CH4) and nitrous oxide (N2O) are the most important greenhouse gases after water vapor and carbon dioxide (CO2). In the past, they all underwent significant variations on glacial-interglacial timescales [Flückiger et al., 1999; Petit et al., 1999] as well as on millennial timescales during the last glacial epoch [Flückiger et al., 1999; Indermühle et al., 2000; Blunier and Brook, 2001; Sowers et al., 2003]. N2O and CH4 variations in the last glacial period show a close relation to the fast climatic changes in the Northern Hemisphere, the so-called Dansgaard-Oeschger (D-O) events (Figure 1). This is in contrast to the CO2 concentration, which shows variations rather in phase with the temperature reconstructed for Antarctica. The origin of these concentration variations is thought to be caused mainly by variations in the different sources of the greenhouse gases, which are the ocean for CO2, wetlands for CH4, and for N2O terrestrial soils (about two thirds of the total source in preanthropogenic times) as well as the ocean (about one third of the total source in preanthropogenic times).
 Understanding the causes, mechanisms, and potential feedbacks of fast climatic changes in the past is of great interest for a better understanding of the climate system. High-resolution N2O and CH4 records can provide such insight as well as important information about global biogeochemical cycles. Variations of the N2O and the CH4 concentration in the past had only a minor effect on the direct greenhouse effect. However, knowledge about the natural variability of greenhouse gas concentrations and their relation to fast climatic changes is also of interest in terms of present climate change. Over the last 250 years, atmospheric N2O, CH4, and CO2 contents have increased dramatically mainly due to anthropogenic emissions (see Stauffer et al.  for an overview). By 1998, respective levels had reached global means of 314 parts per billion by volume (ppbv), 1745 ppbv [Intergovernmental Panel on Climate Change (IPCC), 2001], and 367 parts per million by volume (ppmv) [Keeling and Whorf, 2000]. Ongoing global warming could enhance future greenhouse gas emissions from natural sources.
2. Method and Data Evaluation
 Samples of about 40 g of polar ice are used at Bern for the CH4 and N2O measurements. The air trapped in the ice is extracted with a melt-refreezing method, and the extracted gas is then analyzed for N2 + O2 + Ar, N2O, and CH4 by gas chromatography [Chappellaz et al., 1997; Flückiger et al., 2002]. Tests with bubble-free ice (air-free ice produced by a repeatedly applied zone melting technique starting with highly purified water) and two standard gases (201 ppbv N2O/408 CH4, 304 ppbv N2O/1050 ppbv CH4) have shown that the melt-refreezing extraction developed for CH4 can be used for N2O despite the high solubility of N2O in water. However, the extraction causes a loss of (2.32 ± 0.47) × 10−8 mL of N2O (at standard temperature and pressure) corresponding to a correction of about 6 ± 1.2 ppbv with a typical sample size of 40 g of ice. The given analytical uncertainty σ of the N2O results is a combination (σ2 = σe2 + σd2) of the uncertainty from the extraction process σe, which is determined by measuring bubble-free ice samples with standard gas, and the actual precision of the detectors σd. The latter is taken into account by the scatter of three individual measurements per extracted ice core sample multiplied by the corresponding t-value of the t-distribution due to the small number of injections. The mean analytical uncertainty (1σ) for the N2O data presented in this study is 3.7 ppbv. The reproducibility of neighboring ice samples from artifact-free ice like the Holocene part of the Dome C ice core [Flückiger et al., 2002] is within this analytical uncertainty. For CH4 the analytical uncertainty (1σ) has been estimated at 10 ppbv [Chappellaz et al., 1997]. Further details about the method, the tests performed, and the calibration of the standard gases used are given by Flückiger et al.  and Chappellaz et al. .
 Here we present high-resolution records of CH4 and N2O covering two time periods in Marine Isotope Stage (MIS) 3 and 4 in the last glacial epoch. Measurements were performed along the North Greenland Ice Core Project (NGRIP) ice core (75°06′N, 42°20′W) at 158 and 164 different depth levels for N2O and CH4, respectively. These data cover the time period of D-O events 9 to 12 (Figure 2). The second data set was measured along the Greenland Ice Core Project (GRIP) ice core (72°34′N, 37°37′W). Ninety-six and 107 samples from different depth levels were measured for N2O and CH4, respectively, covering the time period of D-O events 19 and 20 (Figure 3). The two data sets yield excellent resolution. The GRIP record achieves a mean time resolution of 97 years and 77 years for N2O and CH4, respectively, reaching a resolution of about 30 years for both records at the beginning of D-O event 19. The resolution of the NGRIP records is about 63 years with highest resolutions of up to 35 years at the beginning of D-O events. Neither data set is corrected for gravitational fractionation, which is small for CH4 (depletion in the order of 4.1 ± 0.8 ppbv for the GRIP record, 4.5 ± 0.5 ppbv for the NGRIP record) and N2O (enrichment in the order of 2.0 ± 0.3 ppbv for the GRIP record, 2.2 ± 0.3 ppbv for the NGRIP record). The GRIP and the NGRIP N2O and CH4 records are available at http://www.ngdc.noaa.gov/paleo/data.html.
2.2. N2O Artifacts
 The N2O records from the two ice cores are disturbed by artifacts that episodically show concentrations that are too high. This is similar to what was reported in earlier studies for various ice cores from Greenland and Antarctica [Flückiger et al., 1999; Sowers, 2001; Stauffer et al., 2003]. No indications are found that results were biased toward concentrations that are too low. Therefore we assume that artifacts are caused by a production of N2O in situ and that a depletion of N2O is unlikely. While very high values caused by artifacts can easily be distinguished from the general trend of the record, it is very difficult to discriminate with certainty all measurements that are biased by a smaller surplus of N2O. To get an atmospheric N2O record that can be interpreted in terms of climatic changes, it is crucial to exclude the artifacts from the data sets. The method used is described in Appendix A. Resulting data sets for D-O events 9 to 12 and D-O events 19 and 20 are shown in Figures 2 and 3, respectively. They show the artifacts as well as the remaining record, which is thought to represent the atmospheric variations of the N2O concentration.
 To find the cause of the N2O artifacts, further investigations were conducted. The surplus of N2O must be produced in the ice prior to handling and not during the extraction of the air because different extraction techniques yield the same high N2O outliers [Kawamura, 2000; Sowers, 2001]. Two mechanisms could lead to the observed surplus of N2O in distinct depth intervals: chemical reactions and microbiological activity in the ice. Indications for a production by bacteria were previously found in the Vostok ice core in ice from the glacial maximum preceding the Eemian [Sowers, 2001]. However, a production by chemical reactions that involves, for example, ammonium (NH4−) and nitrate (NO3−) cannot be excluded. Less than 1.6% of the present NH4− or less than 0.3% of the NO3− has to be transformed to N2O to explain the high N2O artifacts.
 The question arises whether the general trend of the glacial N2O record is an artifact due to changing concentrations of impurities in the ice. This can be excluded due to the observed depth shift between the climatic signals recorded in the ice (e.g., variations of chemical impurities) and the N2O concentration recorded in the gas enclosed in the ice (Figure 4). This depth shift accounts for the age difference between ice and enclosed air [Schwander et al., 1997].
 Both the NGRIP and GRIP data sets yield artifacts that are mainly located in ice representing transitions from stadials to interstadials and vice versa. To a minor extent, they can be found in ice from cold periods. This is illustrated for the NGRIP measurements in Figure 4. Cold periods are characterized by higher concentrations of dust [Ruth et al., 2003] and chemical impurities, for example, calcium (Ca2+) [Fuhrer et al., 1999], sodium (Na+), and NO3− [Mayewski et al., 1997]. The transitions between stadials and interstadials are marked by changing impurity concentrations and relative compositions of the impurities in the ice. A connection between impurities in the ice and N2O artifacts is further supported by the fact that no N2O artifacts were found in ice from the Holocene epoch [Flückiger et al., 2002], which is characterized by much lower impurity concentrations compared to glacial ice. However, the comparison of surplus N2O in the NGRIP record and high-resolution chemical records of Ca2+, dust, Na+, NO3−, and NH4− measured at the corresponding depths (M. Bigler and R. Röthlisberger, unpublished data, 2003) do not yield a correlation at all [Flückiger, 2003]. This indicates that N2O artifacts do not depend on a single component in the ice but that a combination of different parameters have to be fulfilled for a production of surplus N2O. Further investigations are needed to understand the causes leading to a production of N2O in the ice.
 An overview of the data is given in Figure 1. The N2O and CH4 records covering D-O events 9 to 12, and 19 and 20, are shown in detail in Figures 2 and 3, respectively. The CH4 records compare well with earlier published data sets from Greenland as well as from Antarctica when taking the interpolar difference into account [Brook et al., 1999; Blunier and Brook, 2001]. Indeed, the new records unveil variations in much greater detail due to the high time resolution. Previously published N2O data that overlap partly the records presented in this study are only available for the beginning of D-O event 8 [Flückiger et al., 1999]. Over this time period the NGRIP record is in good agreement with the measurements from Greenland and Antarctica.
 The main characteristics of the N2O records are concentration variations in phase with all investigated D-O events. N2O increases at the beginning of the D-O events and decreases at the end. No large variations or trends are found during the mild phase of the D-O events except for event 12 where the concentration starts to decrease exceptionally early at about 45,800 years before present (45.8 kyr BP), approximately 1000 years before the main CH4 decrease at the end of the D-O event. This N2O decrease is interrupted by an unusual plateau (45 to 44 kyr BP) before the concentration reaches the stadial level between D-O events 12 and 11. The amplitude of the N2O variations is largest for the long-lasting D-O events 12, 19, and 20. For the shorter D-O events 9, 10, and 11 the N2O amplitude decreases as the duration of the events decreases.
 CH4 variations associated with D-O events 9 to 12 are characterized by a sharp increase at the beginning and a relatively sharp decrease at the end of the warm events. This observation precludes smoothing of atmospheric variations due to the process of gas trapping as an explanation for the decreasing N2O amplitudes. The general evolution of the CH4 concentration during the mild phases differs from event to event with either gradually increasing, decreasing, or stable concentrations, except for D-O events 19 and 20. The CH4 increase at the beginning of these events is 110 and 120 ppbv, respectively, which is comparable to others, for example, event 12, but these peak concentrations are followed by an unusually fast decrease to much lower values that characterize the main part of the two events.
 A detailed comparison of CH4 with δ18O illustrates that their general trends during the mild phase of D-O events do not show the same evolution. Even so, in addition to the sharp increases/decreases at the beginning/end of D-O events, some submillennial variations in CH4 and δ18O coincide within the dating uncertainty of the gas and the ice age scale during D-O events 9 to 12. Other submillennial changes, however, cannot be found in both records. A decoupling of δ18O and CH4 is even more pronounced during D-O events 19 and 20. The comparison of CH4 and δ18O could help to determine whether submillennial temperature variations were of local, regional or hemispheric extent and shed light on the possibility that changes of the vapor source regions of the precipitation influenced at least a part of the submillennial changes seen in δ18O.
 Important insight into the processes during fast climatic variations and the response of different ecosystems to climatic changes during the last glacial epoch are obtained from the comparison of the N2O and CH4 records. We will focus our discussion on two main points. First, the long-term variation of the average amplitudes of the N2O and CH4 concentration changes across D-O events will be discussed (section 4.1.). Second, time lags between δ18O, CH4, and N2O are observed at the beginning of long-lasting D-O events. The interpretation of these is considered in section 4.2.
4.1. Response of N2O and CH4 to D-O Events on Long-Term Scales
 The most pronounced difference between the N2O and the CH4 record is the variation in the magnitude of the response of the two greenhouse gases to different D-O events over time. To investigate this different behavior in detail, we calculate the average amplitude for N2O and CH4 over the different D-O events in two independent ways. First, we determined the differences between the mean of the stadial levels before and after each individual event and the corresponding averaged interstadial level. Second, for events with a high time resolution, the average amplitude was calculated as the height of equilateral trapezoids fitted to the data of each individual D-O event and the stadials before and after the event. The equilateral trapezoids were optimized to the smallest possible squared deviation from the data by adjusting the baseline, the starting time, and the angle of the increase as well as the heights and the length of each individual event. The results of the two methods are in good agreement and show only slight differences for CH4 as well as for N2O (Figure 5).
 The average amplitudes are thought to be a measure of the average change of the climatic factors that controlled the N2O and CH4 concentrations during D-O events. The question arises whether the long-term trends in the average amplitudes correlate with variations in the baselines of the records, which are defined by the different stadial levels. To investigate this for CH4 the method developed for excluding N2O artifacts (see Appendix A) was applied to a CH4 record composed of time-synchronized CH4 data from GRIP, GISP2 [Blunier and Brook, 2001], and NGRIP. For the calculation a cutoff period of 20 kyr and an offset value of 15 ppbv were used. The calculated baseline for the Greenland CH4 record shows small variations around an average value of 450 ppbv between 77 and 47 kyr BP. The baseline then decreases linearly toward the Last Glacial Maximum (LGM), which shows a mean concentration of 360 ppbv (Figure 1). It is apparent that the variations in this baseline and the average amplitude of CH4 are not related. Further, the long-term trend in the amplitude is mainly determined by changes in the mean concentration during D-O events rather than changes in the baseline. The same is assumed for N2O. Owing to the incomplete glacial record of N2O and the low resolution of the CH4 record over part of the glacial period, the following discussion focuses only on the average amplitude of the two greenhouse gas variations.
 The average amplitude of CH4 during D-O events in the last glacial epoch shows a strong imprint of the precession cycle of the Milankovitch forcing, similar to that seen for the tropical and mid-northern latitude summer insolation (Figure 5) [Brook et al., 1996]. That implies that the amount or the seasonal distribution of solar radiation influences mechanisms controlling the response of CH4 to D-O events. The average amplitude of N2O does not show such a relation. This is, for example, documented in the high N2O amplitudes for D-O events 19 and 20, which coincide with very small CH4 amplitudes and small tropical and mid-northern latitude summer insolation. The average amplitude of N2O rather correlates with the duration of the D-O events (Figure 6).
 For N2O the question arises whether the close relation of N2O amplitude and event duration is due to slowly changing sources, which do not have the time to reach their full interstadial level during short events, and/or the long lifetime of N2O of 120 years [Minschwaner et al., 1998], which smoothes the atmospheric signal. The latter is only significant for very short D-O events such as D-O event 9. For events that last longer than 500 years, the smoothing due to the long lifetime of N2O is less than 5% (Figure 7) and, therefore, not the main reason for small N2O amplitudes. To investigate the response of the total N2O source to different D-O events the evolution of the total N2O source was calculated from the measured N2O data under the assumption of a constant lifetime. The evolution of the calculated total source indicates that the amplitude of the source change during the short D-O events is much smaller than during long D-O events although there would be enough time for the total source to reach high interstadial levels even during short D-O events. This means that the climatic parameters that caused the N2O source changes evolved differently for D-O events of different durations. The events can be divided into two groups characterized by small and large N2O amplitudes, respectively (Figure 6). They coincide with two different types of D-O events. Large N2O amplitudes coincide with long D-O events, which are preceded by a significant warming in the Antarctic [Stocker, 1998; Blunier and Brook, 2001] (D-O events 8, 12, 19, and 20). Shorter events, which have no significant counterpart in the south, are characterized by lower N2O amplitudes (D-O events 9, 10, and 11). It is hypothesized that the former follow a state with no North Atlantic deepwater formation and start with a full recovery of the thermohaline circulation, while the latter are preceded by a weak deepwater formation state [Stocker and Marchal, 2000; Ganopolski and Rahmstorf, 2001]. In the following, we will investigate the different sinks and sources of CH4 and N2O and possible mechanisms leading to the different observed long-term trends in the average amplitudes.
 Three different mechanisms could be responsible for the CH4 concentration increase during D-O events and variations in the average amplitude of the CH4 response to D-O events: (1) changes in the CH4 sink due to changing OH concentrations in the atmosphere; (2) changes in the extent and/or productivity of global wetland areas; and/or (3) releases of CH4 from methane hydrates in marine sediments.
 The main sink for atmospheric CH4 is its oxidation by OH radicals within the troposphere. Atmospheric chemical models indicate that changes in the sink account only for a small part of the atmospheric CH4 concentration change during the last glacial-interglacial transition [Crutzen and Brühl, 1993; Thompson et al., 1993; Martinerie et al., 1995]. Therefore we focus on the source intensity as an explanation for the CH4 variations during D-O events.
 The main CH4 sources in pre-anthropogenic times are wetlands [Chappellaz et al., 1993; Brook et al., 1996], which are predominantly located in tropical regions and in the northern extra-tropics. The southern extra-tropics are of minor importance due to the small land surface area in these latitudes [Walter et al., 2001; Kaplan, 2002]. CH4 emissions from wetlands depend mainly on precipitation and temperature with wetter and warmer conditions enhancing emissions [Matthews, 2000]. Changes in precipitation patterns and temperature are therefore likely to have caused variations in the extent and productivity of wetland ecosystems.
 Although the uncertainty of source distribution calculations from interpolar CH4 gradients are large, they indicate that the CH4 increase during D-O events may have its origin mainly in sources located north of 30°N [Brook et al., 1999; Dällenbach et al., 2000]. A higher CH4 productivity of these wetlands is in principle consistent with the large temperature increases in the north associated with D-O events. The long-term trend in the average amplitude of CH4, however, cannot be explained by long-term variations in the temperature alone. The temperature in Greenland and probably in the Northern Hemispheric CH4 source regions increased significantly more at the beginning of D-O event 19 (with a small CH4 amplitude) than during D-O event 8 (with a large CH4 amplitude) [Jouzel, 1999; Lang et al., 1999]. Possibly a combination of temperature and humidity change can explain the long-term trend in the CH4 amplitude. This would imply that D-O events with a high temperature increase, like event 19, were relatively dry, leading to a smaller CH4 production increase than during other events. Changes in the extent of ice sheets, which modulated the ice free area may additionally have contributed to the long-term variation in the CH4 amplitude [Brook et al., 1996].
 Further indications come from the comparison of the long-term variation of the average CH4 amplitude and the mean insolation of the summer months June, July, and August, which correspond to the important season for CH4 production in northern latitudes [Fung et al., 1991; Christensen et al., 2003]. The strong imprint of the precession cycle of the orbital forcing in the CH4 amplitudes correlates with the summer insolation in mid-northern latitudes, while summer insolation in high northern latitudes is dominated rather by the obliquity cycle (Figure 5). In glacial epochs, wetlands possibly migrated south to mid-northern latitudes where they may have undergone changes in response to the local summer insolation. A strong imprint of the precession cycle can also be found in the tropics, which are the latitudes with the highest CH4 emissions at present and during the last glacial epoch. This points toward the hypothesis that the tropics are another key region for the variations in the last glacial epoch.
 Palaeo-records from tropical and subtropical Asia [Schulz et al., 1998; Wang et al., 2001] and tropical Africa [Gasse, 2000; Stager et al., 2002] show variations associated with D-O events. They indicate higher southwest monsoon activity during D-O events than during stadial periods implying enhanced precipitation and CH4 emissions in these regions during D-O events. The picture of long-term variations in the monsoon intensity is not homogeneous in records from different sites [Leuschner and Sirocko, 2000; Wang et al., 2001], but at least some studies [Clemens et al., 1991; Gasse, 2000] indicate a correlation to low northern latitude summer insolation and therefore to the variations in the average amplitude of CH4.
 Records from the Amazon show high-frequency variations, which may be related to D-O events [Maslin and Burns, 2000; van der Hammen and Hooghiemstra, 2003]. On the basis of these records it is assumed that the moisture budget in the Amazon Basin changed in phase with D-O events. Precipitation changes further have a strong imprint of the precession cycle [Hooghiemstra and van der Hammen, 1998]. In summary, it is not possible to draw definitive conclusions as to whether changes in the northern extra-tropics or in the tropical regions are mainly responsible for the observed CH4 oscillations associated with D-O events, as well as for the average CH4 amplitude changes on long-term scales.
 Another hypothesis involves the release of CH4 from methane hydrates in marine sediments to explain the CH4 variations in phase with D-O events [Kennett et al., 2003]. However, emissions to the atmosphere via catastrophic events are unlikely given the long durations of D-O events 9 through 12, 19, and 20. The high resolution of the records implies that short-lived (thus catastrophic) events did not occur during these intervals [Brook et al., 2000]. However, we do not exclude the possibility that more regular emissions from CH4 hydrates may cause a small contribution to the CH4 variations in phase with the D-O events.
 Possible causes for concentration fluctuations of N2O associated with D-O events are terrestrial sources located mainly in the tropics and to a minor extent in temperate regions. Emission rates from soils depend essentially on precipitation and temperature and are therefore closely linked to climatic changes [Bouwman et al., 1993; Potter et al., 1996; Bollmann and Conrad, 1998]. Further candidates are changes in the open ocean nitrification [Nevison et al., 1995; Suntharalingam and Sarmiento, 2000] and variations in the denitrification in oceanic upwelling areas, for example, in the Arabian Sea and the eastern tropical North Pacific [Dore et al., 1998; Bange et al., 2001]. The main sinks of N2O are photodissociation (90%) and reactions with excited oxygen (10%) in the stratosphere. Chemical models of the atmosphere have shown that the relative change of the stratospheric N2O sink is small between the LGM and the preindustrial period [Crutzen and Brühl, 1993]. We therefore assume that changes in the N2O sources of up to 1.6 Tg N yr−1 were responsible for the atmospheric concentration variations during the last glacial epoch rather than variations in the destruction processes of N2O and focus our further discussion on the different sources.
 Today's terrestrial N2O sources in high northern latitudes are small [Bouwman et al., 1993; Potter et al., 1996]. It is assumed that these sources were of even minor importance during glacial times. Temperature and precipitation changes in the moderate northern latitudes and especially in the tropics, however, influence the source strength in these regions and likely have led to enhanced emissions during D-O events.
 Other important sources of N2O are denitrification and nitrification in the ocean. These sources could be partly responsible for the N2O variations in phase with D-O events and the different response of N2O and CH4 to these events on long-term scales.
 In general, the denitrification in the world's ocean was strongly reduced during glacial times [Ganeshram et al., 2002] in line with the generally lower atmospheric N2O concentrations during these time periods. During D-O events, however, important upwelling areas like the Arabian Sea [Schulz et al., 1998; Suthhof et al., 2001] and the eastern tropical North Pacific [Emmer and Thunell, 2000; I. L. Hendy et al., The intermittent existence of a southern California upwelling cell during submillennial climate change of the last 60 kyr, submitted to Paleoceanography, 2003] (hereinafter referred to as Hendy et al., submitted manuscript, 2003) show high denitrification rates. These are documented on the California margin as close correspondence between sea-surface temperature changes and the δ15N in bulk sediments (Hendy et al., submitted manuscript, 2003). Warming events, as recorded by lighter values in the δ18O of planktonic foraminifera, are associated with heavier δ15N values implying most intense denitrification in sediments deposited on the margin during D-O events. Inferred increases in denitrification in the Arabian Sea during D-O events are very similar. These high denitrification rates have been linked to high export production associated with enhanced upwelling. Both effects are related to episodic shifts in the tropical Pacific Ocean/atmosphere system [Stott et al., 2002].
 Denitrification rates in the Arabian Sea during D-O events are believed to be similar to those observed today and low or even zero during stadials. Modern N2O emissions from the Arabian Sea are 0.21 to 0.50 Tg N yr−1 [Bange et al., 2001]. A source change of this magnitude corresponds to a concentration change of 5 to 12 ppbv when assuming a constant lifetime of 120 years [Minschwaner et al., 1998]. Changes in the denitrification rates in the important upwelling regions such as the Arabian Sea and the tropical North Pacific could therefore explain a significant part of the atmospheric N2O variations during D-O events. The most pronounced changes in the extent of denitrification implied by the Arabian Sea δ15N data are found during the long D-O events. These correspond to the events that have the largest N2O amplitudes while shorter variations in the Greenland δ18O record are associated with a much smaller or even negligible change in the marine denitrification record and smaller N2O amplitudes.
 Open-ocean nitrification is dependent on stratification of the upper water masses. Increased stratification leads to decreased upwelling of nutrients into the euphotic zone as well as a physical reduction of N2O emissions. Model results show that the increased stratification that results from a reduction in deep water formation during the Younger Dryas event indeed probably led to a reduction in the N2O emissions. Solubility effects due to changing sea surface temperatures and increased storage of N2O in the ocean were of minor importance [Goldstein et al., 2003]. The same impact of stratification is assumed to have been active during the last glacial epoch, causing smaller N2O emissions during stadials and higher emissions during interstadials, and thereby contributing to the N2O concentration changes associated with D-O events.
 Knowing the main sources of N2O, it is likely that a combination of oceanic and terrestrial sources have led to the observed concentration variation during D-O events. This is in line with findings from an isotope study of N2O covering D-O event 1 during the last glacial-interglacial transition [Sowers et al., 2003].
4.1.3. Comparison of N2O and CH4
 What causes the different response of N2O and CH4 to D-O events over time? Four hypotheses are possible. (1) Changes in the sink of one or both greenhouse gases are responsible for the different response of CH4 and N2O to D-O events over time. (2) Contrasts between the two records point to partly different terrestrial source regions, for example, different latitudinal bands, which are responsible for the concentration variations of the two greenhouse gases during D-O events. (3) The terrestrial sources are located in the same regions, but wetlands producing CH4 are more sensitive to different interstadial levels of precipitation and temperature than the soils producing N2O. (4) The ocean is the important key player for N2O.
 We exclude the hypothesis that one or both records are dominated by changes in their sinks. In both cases the sinks seem to be of minor importance for the difference in the atmospheric concentration between the LGM and the preindustrial time period [Crutzen and Brühl, 1993; Thompson et al., 1993; Martinerie et al., 1995]. It can therefore be assumed that the influence was also small during climatic variations of the last glacial period. Still, we note that such a conclusion remains based on very few modeling experiments that were applied to glacial conditions and were based on major assumptions on, for example, the fluxes of other trace species (nitrogen oxides, carbon monoxide, ozone) and the troposphere/stratosphere exchange which are still unconstrained at that point.
 N2O and CH4 both have important soil sources. Does the different response to D-O events over time support the hypothesis that different regions, for example, different latitudinal bands were responsible for the concentration fluctuations associated with D-O events? This is not compelling, because different soil conditions favor N2O or CH4 emissions. While CH4 is produced in soils with 100% water saturation, the production of N2O has its optimum at 60 to 90% water-filled pore space in the soil and ceases as the water-filled pore space approaches 100% [Davidson, 1991]. CH4 could therefore be more sensitive to changes in moisture content in the soil due to changing temperature and precipitation than N2O. Such a situation is indicated for the first part of the Holocene. The evolution of N2O does not follow the drying in the tropics as reported in the CH4 record [Flückiger et al., 2002], implying that the terrestrial N2O sources in the tropics were probably not affected in the same way as the wetlands. Temperature and precipitation changes in phase with D-O events thus could lead to changes in the terrestrial N2O and CH4 sources in the same region but with a higher sensitivity of CH4 to the interstadial moisture status. Therefore it cannot a priori be excluded that the causes for the two different long-term trends in the response to D-O events are located in the same latitudes. Nevertheless, we cannot exclude differences in the latitudinal distribution of the terrestrial N2O and CH4 sources responsible for the concentration variations. While the changing terrestrial source of N2O is located in tropical and subtropical regions, the tropical and high northern latitudes could be both responsible for the CH4 variations.
 An unlikely hypothesis is that the terrestrial sources of N2O did not show oscillations associated with D-O events and the N2O concentration predominantly reflects changing oceanic sources. The CH4 variations document that soils underwent dramatic shifts in the production due to temperature and precipitation changes. These climatic changes certainly influenced the N2O production in soils as well. Therefore a combination of terrestrial and oceanic sources is more likely responsible for the N2O variations during the last glacial epoch. Oceanic sources, however, could be responsible for the contrast in the N2O amplitudes of D-O events of different duration. Support for this hypothesis is provided by the fact that CH4, which is probably mainly a terrestrial product, shows no such pattern.
4.2. Start of Dansgaard-Oeschger Events
 The high-resolution N2O and CH4 measurements performed on the same samples offer the opportunity to compare the two records without any relative time uncertainty, whereas the comparison with the δ18O record measured along the same core is only possible within the uncertainty of delta age, i.e., the time difference between the ice and the enclosed air [Schwander et al., 1997]. Additional information about the timing of the temperature increase recorded in Greenland ice cores and the CH4 record at the beginning of D-O events comes from measurements of δ15N of N2 in the bubbles enclosed in the ice, which is another proxy for temperature. For the transitions from the late glacial epoch to the Bølling-Allerød and from the Younger Dryas to the Holocene, it was found that the start of the CH4 increase lags the beginning of the temperature increase by 0 to 30 years, and that the increase lasts several decades longer than the temperature increase [Severinghaus et al., 1998; Severinghaus and Brook, 1999]. The comparison of δ15N [Lang et al., 1999] and CH4 measurements performed on the GRIP core excludes a lag of CH4 to δ15N at the beginning of D-O event 19. Rather, the data suggest that the beginning of the CH4 increase was synchronous with the start of the warming (Figure 8). As for the other events, the CH4 increase lasts several decades longer than the temperature increase revealed in the δ15N record. According to these results, we assume that the CH4 increase at the beginning of D-O events is synchronous with the temperature increase, within a few decades.
 The comparison of N2O to CH4 shows that two types of events must be distinguished (Figure 9). During the long D-O events (8, 12, 19, and 20) the N2O concentration starts to increase continuously several hundred years before the main CH4 rise, and thus also leads the beginning of the sharp temperature increase reconstructed for Greenland. For the shorter D-O events presented in this study the data are less clear. However, they indicate that within the resolution of the data, and, at least for D-O events 10 and 11, the start of the N2O increase did not precede but was rather synchronous with the beginning of the CH4 rise.
 The long D-O events (8, 12, 19, and 20) are preceded by a significant slow warming in Antarctica, which starts about 1500 to 3000 years before the sharp temperature increase of the corresponding D-O events observed in the north [Blunier and Brook, 2001]. A similar warming has been found for at least part of the Southern Hemisphere outside Antarctica [Voelker and Workshop Participants, 2002], for example, for the Southern Ocean surface temperature [Charles et al., 1996; Ninnemann et al., 1999] and in the South Atlantic [Vidal et al., 1999], and has also been suggested by models [Stocker, 1998; Vellinga and Wood, 2002]. A close relation between the early N2O rise and the warming in the south can be excluded due to the very different timing of the two records. The N2O concentration starts to increase more than 1000 years later than the southern warming in all the cases.
 It is further known that D-O events 8 and 12 are preceded by massive iceberg discharges, Heinrich events H4 and H5, respectively [Hemming, 2004]. In MIS 4, for example, before D-O events 19 and 20, distinct events are missing in the deep-sea records [Hemming, 2004]. A direct link of the early increase of N2O to Heinrich events can therefore be excluded.
 The missing link of the early rise of N2O to known climatic changes indicates that a picture of North Atlantic instability causing the D-O events may be too simple. Instead, it points to changes in other parts of the Earth's climate system that start before the abrupt warming event recorded in Greenland. Such changes could be an early warming in the tropics, or early shifts in some parts of the ocean, that cause the ocean to approach the threshold where deepwater formation through convection abruptly starts. Early changes are, for example, seen in the intermediate waters of the Santa Barbara basin [Hendy and Kennett, 2003]. Finding the cause of the early start of the rise in N2O concentration is a crucial task for future research, because it will certainly yield further insight into complex global processes that are associated with climatic variations on millennial time scales in glacial epochs and could give insight into a precursor of D-O events.
 Although N2O starts to rise earlier than CH4, at least at the beginning of long D-O events, the concentration increase lasts significantly longer than for CH4. While the main increases of CH4 at the beginning of the D-O events discussed here are completed within less than 150 years, interstadial levels of N2O are reached 200 to 400 years later. This longer-lasting increase can mainly be explained by the longer lifetime of N2O (120 years [Minschwaner et al., 1998], versus 10 years for CH4 [Martinerie et al., 1995; Chappellaz et al., 1997]) (Figure 7).
 CH4 and N2O both show concentration variations in phase with D-O events in the last glacial epoch. However, significant differences exist in their response to D-O events over time. While the average amplitude of CH4 shows a close relation to the precession component of tropical and mid-northern latitude summer insolation, N2O amplitudes correlate instead with the duration of the D-O events. The arrangement of the D-O events into two groups of large and small N2O amplitudes supports the view of two types of D-O events with different extent and, perhaps, different underlying process. The long-lasting D-O events are preceded by a significant warming in the south and have larger N2O amplitudes than the shorter events with no significant southern-warming counterparts. Further, models suggest that the long events follow an ocean state without North Atlantic deep water formation, while the shorter events are associated with only a shift and amplification of the deepwater formation [Stocker and Marchal, 2000; Ganopolski and Rahmstorf, 2001]. For the longer D-O events, N2O starts to increase several hundred years before the main CH4 and temperature increases recorded in Greenland but more than 1000 years after the beginning of the warming in Antarctica. This indicates that changes related to D-O events in some parts of the climate system preceded the abrupt warming observed in Greenland.
 Additional work is necessary if the assessment and attribution of gas sources is to be improved for individual events. In particular, emphasis needs to be placed on a better reconstruction and understanding of precipitation and temperature variations in different latitudes and their associated impacts on terrestrial CH4 and N2O sources. Furthermore, there is an obvious need to gain more insight into the oceanic N2O source and its variation in the last glacial period. In addition to the understanding that can be gained by terrestrial and oceanic models, through which the different responses of CH4 and N2O sources to climatic changes in the past can be investigated, additional high-resolution ice core data should provide insight into the responsible mechanisms. For example, CH4 gradient and source-distribution calculations provide a powerful means to probe the changing response of CH4 sources to different D-O events. However, this is not an option for N2O due to the small interhemispheric gradient (only about 0.8 ppbv today [IPCC, 2001]). Improved attribution of sources of CH4 and N2O can also be expected from measurements of their isotopic composition and from the intramolecular site preference in N2O isotopomers. In addition a better understanding of fractionation processes and the isotopic signature associated with each individual source is required especially for N2O to reduce uncertainties in isotope calculations.
Appendix A:: Method to Exclude N2O Artifacts
 A smoothed spline is calculated through the full N2O data set. Data points with a positive offset of more than a certain value (hereinafter referred to as the “offset value”) to the spline are considered as artifacts and are excluded from the record. This procedure is applied iteratively until none of the individual points in the data set exceeds the spline by more than the offset value. The important parameters to be determined are the offset value and the degree of smoothing by the spline function. The offset value is set to 8 ppbv, which is on the order of the mean 2-sigma uncertainty of each individual data point. The smoothing of the spline function excludes all periods lower than a certain cutoff period [Enting, 1987]. To estimate the cutoff period to be applied to the data set, a suite of variations of the total N2O source that have a triangular or a trapezoidal shape is simulated, resulting in atmospheric variations of 14 to 49 ppbv. The spline method is then applied to the individual signals adjusting the cutoff period to the highest value where none of the data points are excluded. The simulations start from a steady state situation with a source of 0 Tg N yr−1. The source is then increased in the range of 2.04 Tg N yr−1 to 2.85 Tg N yr−1 in 1 to 50 years. This is equivalent to 50 to 70 ppbv at steady state, corresponding to the largest increase observed at the beginning of a D-O event and the glacial-interglacial transition, respectively. The source is then decreased at a rate of 0.005 Tg N yr−2, which roughly corresponds to the fastest concentration decrease in the record at the end of D-O event 19. Further, to exclude an underestimation, a decrease is simulated at a rate of 0.02 Tg N yr−2. The duration of the entire source perturbation is varied from 100 to 600 years. The range of calculated cutoff periods from these simulations is 400 to 750 years. To exclude artifacts from the GRIP and NGRIP N2O data sets, a cutoff period of 600 years is used, which is within this range. The choice of this cutoff period is supported by the lack of overshooting of the spline in any part of the GRIP and the NGRIP record. Furthermore, small variations in the data sets as a result of the measurement uncertainty are cancelled out while variations related to D-O events are still nicely reported in the spline. Small changes in the offset parameter and the cutoff period do not have a significant influence on the N2O data sets except from the depth interval of 2176 to 2184 m in the NGRIP core (about 43.8 to 44.6 kyr BP). In this interval, the determination of artifacts is very sensitive to small changes in the timescale and the cutoff period. Two additional data points in this depth interval are excluded and are highlighted in the figures. This shows the limitation of the used artifact identification algorithm in depth intervals with a little bit lower time resolution and a scattering of the data, which is slightly higher than on average. However, all the conclusions of the paper do not depend on individual data points and are therefore not sensitive to small changes in the two parameters of our artifact identification method.
 We thank T. F. Pedersen and R. Knutti for fruitful discussions and helpful comments on the manuscript. Comments of two anonymous reviewers further improved the paper. This work, as part of the NorthGRIP project, was supported by the University of Bern and the Swiss National Science Foundation. The NorthGRIP project is directed and organized by the Department of Geophysics at the Niels Bohr Institute for Astronomy, Physics and Geophysics, University of Copenhagen. It is being supported by Funding Agencies in Denmark (SNF), Belgium (FNRS-CFB), France (IFRTP and INSU/CNRS), Germany (AWI), Iceland (RannIs), Japan (MEXT), Sweden (SPRS), Switzerland (SNF), and the United States of America (NSF). Partial support through contract EVK2-2000-22067 (POP) by the Swiss Federal Office of Science and Education is acknowledged.