Barker and Knorr. 10.1073/pnas.0708494104. |
Fig. 5. GISP2 d18O (1) with the defocusing function (numbers indicate canonical D-O events; ref. 2). The D-O picking algorithm allows identification of incomplete D-O transitions. These tend also to correspond with non-canonical D-O transitions, not originally identified from the d18O record (highlighted in blue). Note the symmetric appearance of the D-O oscillations defined by this procedure as compared with the asymmetric form implied by the temperature records.
Fig. 6. (A) 14C ages for MD98-2181 (1) define a linear sedimentation rate for the upper 18m of this core. (B) Linear sedimentation rate applied to the whole core to obtain an age model for comparing the WEP record with Greenland and Antarctic temperature records. Note similarity with published model.
Fig. 7. Lead/lag correlations for the Hulu Cave d18O record (1) versus the Greenland and Antarctic temperature records for the period 60-30 kyr ago. Left hand panel includes orbital variability, right hand panel is hi-pass filtered at 7,000 yr (all data are lo-pass smoothed at 200 yr).
Fig. 8. Lead/lag correlations of (A) Byrd d18O (1) versus benthic foraminiferal d18O from the deep North Atlantic (core MD95-2042; ref. 2) for the interval 60-30 kyr with and without orbital filtering at 7,000 yr (all data are lo-pass smoothed at 30 yr; for each comparison both records are placed on a common time scale; either GISP2 or GRIP); (B) Greenland versus Byrd d18O and planktonic versus benthic d18O from MD95-2042 on the GRIP and GISP2 time scales (all records are lo-pass smoothed at 30 yr and orbitally filtered). The relationship between the "northern" and "southern" signal is shifted in time in the marine records relative to the ice-core records (B). This is reflected by a lag of benthic d18O behind Antarctic temperature as shown in A. See also main text Fig. 4.
SI Text
Algorithm for picking D-O transitions and de-jumping the GISP2 dust record
A MATLAB routine was designed to pick D-O transitions and produce a de-focusing function which is combined with the GISP2 [Ca] data to produce a de-jumped dust record. All calculations were made on the log10 [Ca] record (denoted by [Ca]*), which allows identification of interstadial variability in the dust record. The monotonic nature of this operation ensures that maxima and minima in the raw [Ca] record are preserved. The de-jumped record is the sum of [Ca]* and the defocusing function. This is equivalent to multiplying the raw [Ca] record by a constant factor during stadial and interstadial periods. The interstadial value of DF (DFIS) is set equal to zero (equivalent to multiplying the raw [Ca] record by one). The stadial value of DF (DFS) is set to achieve maximum contiguity of the de-jumped record. In this case DFS is set to log10(1/7). This is equivalent to dividing the raw [Ca] record by 7 during stadial periods. A similar result is obtained if a factor of 6 is used. Factors of 5 or below preserve some of the "D-O" structure of the raw dust record while factors greater than 7 tend to overcompensate for smaller jumps in the raw record.
The algorithm calculates the first time differential of [Ca]* (i.e., d[Ca]*/dt), then identifies D-O transitions based on a threshold limit which is set to the standard deviation of d[Ca]*/dt multiplied by ± 1.15 (Fig. 2). This ensures all canonical D-O events (1) are identified in addition to a few others (see below). A benefit of using the dust record is the similarity in abruptness of stadial-to-interstadial and interstadial-to-stadial transitions. This means that, unlike the d18O record, equal thresholds (of opposite sign) can be used for both types of transition. If d[Ca]*/dt crosses a threshold value of the same sign two or more times consecutively, the crossing with the largest mean value of d[Ca]*/dt (for the duration while d[Ca]*/dt exceeds the threshold value) is selected as the D-O transition. This ensures that the defocusing function alternates between two states without veering off to some other state. It also highlights a complexity of D-O climate evolution during the Last Glacial Maximum (LGM, gray shaded region in Fig. 2). Here the dust record clearly shows two consecutive and significant abrupt increases (downward maxima in d[Ca]*/dt) the second of which is marked by a yellow circle in Fig. 2. This feature possibly represents a third 'state' of dustiness during the LGM and demands a higher degree of complexity in the approach to picking D-O events. However, since we are here concerned primarily with D-O variability during MIS 3 the LGM is excluded from further discussion with the acknowledgment of its increased complexity.
The age of any particular transition is defined as the mean age of all consecutive values of d[Ca]*/dt which exceed the threshold value. The routine then constructs the defocusing function which alternates between stadial and interstadial values of the defocus factor, DF (DFS and DFIS respectively) with the same rate of change and timing as the dust data. The defocusing function constructs each D-O transition starting from the mid-point between DFS and DFIS. DF is extrapolated forwards and backwards in time along the maximum value of d[Ca]*/dt for that transition until it intercepts with DFS and DFIS. This results in each transition being slightly different and allows for contiguity of the resulting de-jumped record. This approach also allows for the recognition of "incomplete" D-O transitions (in either direction) (Fig. 5). These are defined as transitions where the rate of change of dust is too slow to allow a full transition to the other state before returning to the previous state. Interestingly, non-canonical transitions defined by this algorithm tend to be incomplete. This would explain why they also tend to be rather unpronounced in the d18O record and perhaps why there were omitted from the original selection of D-O events. All figures show the de-jumped record raised to the power 10 for comparison with the raw [Ca] record.
Lead/Lag correlations and data filtering
For all lead/lag correlations, original datasets were re-sampled with a 10 yr interval and filtered (for lo- and hi-pass) using the MATLAB "filtfilt" algorithm to avoid introducing phase lags or loss of signal amplitude. Lead/lag correlations were performed using the MATLAB "tstool" GUI.
Comparison of Hulu Cave and WEP records with Greenland and Antarctica
To compare the marine planktonic d18O record from the West Equatorial Pacific (WEP) (2) with the ice-core temperature records it is first necessary to define a reasonably objective age model for this record. The original time scale for the core (MD98-2181) was derived from a number of 14C dates (ranging from ~7 to 30 kyr) with older age tie-points obtained by visual tuning of the record to the Greenland temperature record (2). For the present study the linear sedimentation rate obtained from the 14C ages is extrapolated to the full length of the core (Fig. 6). The difference between the revised (linear) age model and the original is generally less than 2,000 yr.
The age model for the Hulu Cave d18O record (3) is derived from U/Th dates and is therefore independent from all ice-core records. A similar lead/lag analysis as shown in Fig. 2 (for the WEP) reveals significantly better correlation between the Hulu Cave d18O record and Antarctic records than between Hulu and Greenland (Fig. 7). However, the situation is made more complicated by the fact that North Atlantic Heinrich events are recorded as pronounced maxima in the Hulu record (see main text).
Phase relationships between marine and ice-core records
Shackleton et al. (4) demonstrated that the record of benthic foraminiferal d18O from a deep North Atlantic sediment core (MD95-2042) revealed Antarctic-style fluctuations when placed on the ice-core time scale by tuning of the planktonic d18O record with Greenland temperature (Fig. 4). Fig. 8A shows a lead/lag correlation between the tuned benthic d18O record from MD95-2042 and the Byrd temperature record. The analysis reveals a time lag of up to several hundred years between the Antarctic signal and the benthic record irrespective of whether the GRIP or GISP temperature record is used as a tuning target. A similar result it attained by comparing the relative phasing between the planktonic and benthic d18O records from the marine core with that between northern and southern ice core temperature records (Fig. 8B). The latter approach is insensitive to the ultimate success of tuning the marine core to Greenland.
2. Stott L, Poulsen C, Lund S, Thunell R (2002) Science 297:222-226.
3. Wang YJ, Cheng H, Edwards RL, An ZS, Wu JY, Shen CC, Dorale JA (2001) Science 294:2345-2348.
4. Shackleton NJ, Hall MA, Vincent E (2000) Paleoceanography 15:565-569.