One of the hidden assumptions of proxy reconstructions, as carried out by IPCC authors, is that each “proxy” has a linear relationship to temperature plus relatively low-order red noise. Under such circumstances, the noise will cancel out in a linear combination of proxies (reconstruction) and a “signal” will emerge. However, I’ve never seen any author discuss the validity of this assumption, let alone establish the validity.
In today’s post, I’m going to look at low-latitude South American d18O isotope series mainly from Peru, including three proxies from Neukom. Tropical ice core d18O series (especially Quelccaya, but also Huascaran and Sajama) have been a staple of temperature reconstructions. During the past few years, d18O series have also been obtained from speleothems and lake sediments.
In my opinion, before one can begin thinking about temperature reconstructions using many different types of proxies, some of which are singletons, it makes sense to see if one can make sense of something as simple as d18O series within one relatively circumscribed region.
Holocene Scale
I am increasingly looking at proxies on a Holocene scale prior to considering them on a medieval scale. By showing values from the LGM through the Holocene Optimum to the present, one can see which way is up – a proxy property that Mann, Neukom and others all too often neglect.
All but one of the proxies in the figure below is from the Peru area: the top panel shows a speleothem (Tigre Perdido); the second panel shows the Pumacocha lake sediment series (used in Neukom 2014); the third panel shows Lonnie Thompson’s Huascaran series (slightly to the north of Quelccaya); the bottom panel shows two tropical ocean d18O series: Galapagos and Guinea. I’ve shown a series from Guinea because the Atlantic is the source of Andean precipitation, because it is high resolution (and I was unable to locate a comparable high resolution series from the South American side of the Atlantic).
I find it somewhat reassuring that the three land d18O series all have “major” features in common, at least on a Milankovitch scale. In a general sense, the Huascaran, Pumacocha and Tigre Perdido series all show increasing d18O values going from the LGM to the Holocene, with a gradual decrease through the Holocene. The LIA in the Pumacocha series is rather pronounced.
A first obvious observation, but worth keeping in mind: d18O isotope values go in opposite directions between tropical oceans and tropical land. Land d18O values go up between the LGM and the Holocene (more sharply at higher altitudes), while maring d18O values become increasingly depleted. The increasing depletion of marine d18O values from the LGM moving towards the Holocene presumably reflects the melting of continental ice sheets: the ice sheets trapped precipitation with very depleted isotopes and the melting of the ice sheets returned them to the oceans.
Precipitation in the Andes comes from the Atlantic and, like the Himalayas, is governed by monsoons: more depleted precipitation comes in the summer from rainout. The vast majority of specialists (all except Lonnie Thompson) interpret d18O as an “amount” effect: lower d18O values mean more precipitation.
In a general sense, 20th century d18O values are in between LGM and Holocene Optimum values – readers will recall that Marcott et al had reported that 20th century values were exceptional: that Elvis was leaving the building, so to speak. If there was an overall pattern of 20th century exceptionalism, then one would expect the proxy results to be swamping Milankowitch patterns, but that does not seem to be the case in this region at least. (I’ve examined many other regions with similar results.)
Medieval Scale
Neukom et al included three Peru-area d18O series in their network (plotted below from original data). The top panel shows a speleothem series (Cascayunga); the middle is the Pumacocha lake sediment once again; bottom – Quelccaya ice core. These are plotted in absolute data, not the anomalies that too often remove interesting information. Cascayunga and Quelccaya are both higher resolution than Tigre Perdido and Huascaran, though the latter are much longer. One of the three proxies is screened out of the Neukom reconstruction: readers are invited to ponder which one – I’ll tell below.
Figure 2. d18O: top – Cascayunga speleothem; middle – Pumacocha lake sediment; bottom – Quelccaya ice core.
On this scale, there is much greater high-frequency variability in the high-altitude Quelccaya ice core series than in the lower-altitude speleothem (note however that the scales are different).
In each of the three series, one can perceive d18O values increasing from the LIA to the modern period. However, the precise timing of changes do not reconcile all that well. When similar comparisons are made between proxies on a regional scales, IPCC authors (e.g. Mann, Crowley) have routinely interpret such inconsistency as evidence that the LIA and/or MWP do not reflect globally organized events, but are regional to the North Atlantic. However, here, the three d18O series are in very close proximity.
On the other hand, there is some reassurance that all three series show depleted d18O values in the LIA. In specialist (non-multiproxy articles), the depleted d18O values at Quelccaya during the LIA have been compared to Ti values at Cariaco, leading to a hypothesis that the changing precipitation patterns are due to the ITCZ moving north and south. In past CA posts, I’ve suggested that N-S movements of the ITCZ is an alternative and simpler explanation for Kim Cobb’s Palymyra coral (which joins YAD061 on the pantheon, her medieval coral is the “most influential” coral in the world) than the Mannian hypothesis of a cool medieval Pacific.
One would be hard-pressed to develop a theory under which climate change at Pumacocha and Quelccaya proceeded on radically different schedules. It seems far more plausible to me that (1) the proxies integrate d18O values from precipitation in different and complicated ways; (2) the dating of the proxies is not consistent and the errors are cumulative, not gaussian.
If one reflects back on the reconstruction project, consider trying to derive (forward model) each of these three d18O series from a “true” regional O18 series (not even trying to get to temperature just yet). It takes only a moment of reflection to realize that the residuals will be highly correlated to each of the proxy series themselves and will not be low-order red noise.
PS – Neukom screening excluded the Cascayunga series.
