By Andy May
In previous posts (here, here and here), we have shown reconstructions for the Antarctic, Southern Hemisphere mid-latitudes, the tropics, the Northern Hemisphere mid-latitudes, and the Arctic. Here we combine them into a simple global temperature reconstruction. The five regional reconstructions are shown in figure 1. The R code to map the proxy locations, the references and metadata for the proxies, and the global reconstruction spreadsheet can be downloaded here. For a description of the proxies and methods used, see part 1, here.
Figure 1A, all proxies except TN057-17 on the Antarctic Polar Front
Figure 1B, the proxies used for the reconstructions
It is interesting that the Northern Hemisphere is the odd reconstruction. This was also true for the Marcott et al. (2013) Northern Hemisphere reconstruction from 30°N to 60°N, see figure S10f, in their supplementary materials. The Northern Hemisphere has the greatest temperature variation of the five regions and a clearly different trend. Is this because it contains most of the land? Perhaps so. It may be, in part, the impact of the melting continental glaciers from the last glacial advance. Certainly, the high Northern Hemisphere insolation, early in the Holocene due to orbital precession and obliquity played a significant role (see figure 2 in part 1, also shown for convenience as figure 2 below). In the figure, the colored curves are the seasonal changes due to precession and the background color is insolation by latitude due to obliquity changes. The black curve is the Greenland NGRIP temperature reconstruction, note that the end of the last glacial period is when both orbital obliquity and precession hit their peak insolation in the Northern Hemisphere. The labels on the curves indicate Northern Hemisphere as “N” and Southern Hemisphere as “S.” The letters after that are the first letters of the months of the year. At the beginning of the Holocene, the Northern Hemisphere summer had maximal insolation due to precession and the higher latitudes (poles) had greater insolation, due to obliquity, at the expense of the tropics. Thus, both the precession cycle and the obliquity cycle were in their warmest phases for the Northern Hemisphere mid and high latitudes. This changed a few thousand years later and the climatic equator (the Intertropical Convergence Zone) shifted and the long Neoglacial cooling period began (see figure 12, in part 2).
Figure 2 (Source: Javier, see his post for a detailed explanation of the figure.)
The Southern Hemisphere is also a bit anomalous, with a dip in the period of the HCO, corresponding with a dip in winter insolation in the Southern Hemisphere. The other interesting thing about the reconstructions is that the Northern Hemisphere has a higher and longer Holocene Climatic Optimum. The Northern Hemisphere was affected much more by the last glacial advance due to the large continental ice masses there. The Southern Hemisphere ice was mostly sea ice which, presumably, melts at a steadier rate with less dramatic effect.
The Arctic and Antarctic each cover 6.7% of the globe, the southern and northern mid-latitudes cover 18.3% each and the tropics covers 50%. If we weight each reconstruction by the area of its region we get the reconstruction in figure 3. Figure 3A uses all proxies, except for TN057-17, which was removed in part 2. Figure 3B also eliminates ODP-658C, KY07-04-01 and OCE326-GGC26. The removal of the latter three proxies are discussed in part 2 and part 3. The two reconstructions only differ in detail.
Figure 3A, all proxies
Figure 3B, three additional proxies removed
We will discuss the reconstruction in figure 3B since we prefer it. In this reconstruction, the depth of the Little Ice Age (LIA) occurs in 1610 AD. The apparent Medieval Warm Period (MWP) is smeared over several hundred years and occurs from around 510 AD to 1050 AD which does not fit the historical record. Oddly, only the Southern Hemisphere and the tropics show a distinct Medieval Warm Period (MWP) in its historical time. This is despite abundant historical evidence of a Northern Hemisphere MWP from around 900 AD to 1200 AD. The Antarctic reconstruction shows several warm spikes during the period, but nothing very distinct. The reason for the lack of a distinct MWP signature in the northern reconstructions is not known. In part 3 we looked at the individual proxies for the Northern Hemisphere and saw that they disagree on the presence and timing of the MWP.
The Roman Warm Period (RWP) shows up well in the reconstruction, at about the right time. The “collapse of civilization” at the end of the Bronze Age is clearly seen. The 4.2 kiloyear event that led to the collapse of the Akkadian empire in 4170 BP can be seen (deMenocal, 2001). The 5.9 kyr event that occurred as the Sahara was turning into a desert, causing a great migration to the Nile valley that ultimately resulted in the Egyptian Old Kingdom is clearly seen. The LIA is the most significant climatic event of the Holocene without question, but the second most severe climatic event may well be the 8.2 kyr event. This event ended the Pre-Pottery Neolithic B culture and was when the Black Sea was catastrophically connected to the Mediterranean in an event that may be remembered as Noah’s great flood (Ryan and Pittman). The 10.3 kiloyear event takes place about the time the Pre-Pottery Neolithic period began. For more details on human history and climate change see “Climate and Human Civilization over the last 18,000 years” here. The historical climatic events match this reconstruction well, except for the MWP.
The details of the regional areas are in Table 1. This table is different from the one presented in part 1 of this series because after part 1 was put up we dropped ODP-658C from the tropics reconstruction and KY07-04-01 and OCE326-GGC26 from the Northern Hemisphere reconstruction. Marcott, et al. (2013) used 73 proxies for their reconstruction and our first pass retained 31 of these and added the Rosenthal et al. (2013) Indonesian proxy for a total of 32. As the study progressed we dropped three more proxies and ended with 29. Fifty-five percent of the proxies are north of 30°N and only 21% are south of 30°S.
If we simply average the 5 reconstructions with no weighting, we get the reconstruction in figure 4.
Figure 4, Straight average, no weighting, final proxy set
The two reconstructions are not very different. In this reconstruction, the depth of the Little Ice Age (LIA) occurs between 1530 AD and 1670 AD and the temperature anomaly is -0.84°C. The Holocene Climatic Optimum (HCO) runs from 10500 BP to 4500 BP and has numerous peaks between 0.35°C and 0.48°C. Figure 3B is similar, with a slightly larger temperature range. The average temperature difference then, in these reconstructions, is between 1.2°C and 1.4°C. This compares well to the geological and biological evidence presented in Javier, 2017.
A word about error
There are many sources of potential error in these reconstructions. In this series of posts, we have emphasized those sources we thought were most important and significant. Specifically, we focused on the geographic distribution of the proxies, proxy selection, the choice of the mean used to generate the temperature anomalies, the effects of proxy dropout, proxy resolution, and the impact of local conditions on the proxies. The latter problem relates to how applicable the proxy is to regional climate as opposed to local climate. Examples of inappropriate proxies due to local conditions are TN057-17 and ODP-658C which are discussed in part 2.
Dating the proxy samples can be problematic. Marcott, et al. (2013) emphasize potential dating errors in their paper and supplementary materials. They consider dating errors to be the largest source of error. Marcott, et al. (2013) also provide a very detailed discussion of proxy-to-temperature calibration uncertainty in their supplementary materials. Generally, they assume one standard deviation (normally distributed) to be the error inherent in the proxy-to-temperature conversion, otherwise they follow the proxy author’s recommendations.
Marcott, et al. assumed a fundamental dating error of 120 to 150 years for most cases and accounted for it using a Monte Carlo procedure (1,000 realizations) which is detailed in their supplementary materials. For the layer counted Antarctic ice-core records they assumed a ±2% uncertainty and for Greenland cores they assumed a ±1% error. All radiocarbon dates were recalibrated using IntCal09. Our reconstructions use the original published dates and not the recalibrated dates.
Dating errors and proxy-to-temperature errors are undoubtedly important and Marcott et al. (2013) provide a good discussion of these problems and their supplementary database contains estimates for these sources of uncertainty. They also considered that some of the proxies may have a seasonal bias and attempted to account for this source of error in their Monte Carlo procedure. They do not believe that seasonal bias is an important source of error. We have nothing to add to their work on these uncertainties and the interested reader is referred to their paper. They do present an interesting figure in their supplementary materials displaying the 1,000 Monte Carlo realizations that result from their study of error due to dating and proxy-to-temperature conversion. It suggests that error due to these factors is roughly ±0.5°C. We show their figure as our figure 5:
Figure 5 (Source: Marcott, et al., 2013 supplementary material)
Marcott, et al. (2013) also provide their own latitudinal temperature reconstructions and display them in their supplementary figure S10, not reproduced here. Their regional reconstructions are different in detail than ours because they use more proxies, but their 30°N to 60°N reconstruction for the Holocene is the same big outlier we see in our figures 1A and 1B. They also note, as others have, that computer simulations of Holocene climate do not agree with the proxy reconstructions, the so-called Holocene temperature conundrum. The largest difference between the simulation results and the proxy reconstructions occurs in the mid-high latitude Northern Hemisphere, which suggests that the models are missing some key component of Northern Hemisphere climate. They suggest that the models may not be modeling north Atlantic Ocean circulation properly, we agree. The global climate models also have other problems, for a discussion see here.
We believe the greater source of error in these reconstructions is in the proxy selection. As documented in this series, some of the original 73 proxies are affected by resolution issues that hide significant climatic events and some are affected by local conditions that have no regional or global significance. Others cover short time spans that do not cover the two most important climatic features of the Holocene, the Little Ice Age and the Holocene Climatic Optimum.
We’ve tried to address the criticism of the Marcott et al. (2013) global temperature reconstruction. Steve McIntyre, Grant Foster and others contested their adjustments of the published proxy dates, their inclusion of some inconsistent proxies, and not compensating very well for proxy drop out. Javier has pointed out that their proxy reconstruction does not reflect abundant geological and biological evidence that the average sea surface temperatures were at least one degree Celsius warmer during the Holocene Climatic Optimum than during the Little Ice Age. In addition, the use of proxies that do not cover the interval from the LIA to the HCO is problematic since these are the two best defined temperature extremes in the period. Further, we are using temperature anomalies from the mean to build these reconstructions and prefer to get the mean from the period 9000 BP to 500BP so that the mean represents both the high temperatures of HCO and low temperatures of the LIA. This is not possible if the proxy does not cover this interval.
We also avoided proxies with long sample intervals (greater than 130 years) because they tend to reduce the resolution of the reconstruction and they dampen (“average out”) important details. The smallest climate cycle is roughly 61 to 64 years, the so-called “stadium wave,” and we want to try and get close to seeing its influence. In this simple reconstruction, we have tried to address these issues.
The reconstructions show a difference of 1.2°C to 1.4°C between the LIA and the HCO. This suggests that the underlying data support this temperature difference. These reconstructions also show more detail. The additional detail appears to correspond to known climatic events. While the LIA, HCO, Roman Warm Period, Minoan Warm Period and other historical events show up well in the reconstructions, the Medieval Warm Period does not, it appears dampened and offset in time from historical records. The reasons for this are unclear. As discussed in part 3, some Northern Hemisphere proxies show an MWP and some do not. The proxies may be wrong or perhaps the MWP occurred in different times or in different intensity in different places, smearing it on a global reconstruction. Either way proxy choice determines the MWP intensity and timing, which is disappointing. More work and better proxies are needed to improve our Holocene temperature record.
An accurate Holocene temperature reconstruction is not possible, even measuring the potential error in a reconstruction this long is incredibly difficult. Marcott, et al. (2013) did a good job of estimating dating error and proxy-to-temperature error, in our opinion. But, they do not address the other issues, such as proxy selection, that may be more important. But, even without a viable error calculation, a generally accepted estimate of Holocene temperature trends is greatly desired. To understand the present, we must know the past. This is a very simple reconstruction and it is not meant to be definitive, but we present it as a starting point for future work. It is a presentation of the data and some useful tools needed to work the data.
To improve the reconstruction, I think we need to compare it and the component proxies to other data. In particular, historical records, archeological records, glacial advance histories, biological and geological data. This “outside data” can be used to select proxies and guide the reconstruction.
The R code to map the proxy locations, the references and metadata for the proxies, and the global reconstruction spreadsheet can be downloaded here.
I am very grateful to Javier who has read this post and made many very helpful suggestions. Any errors are the author’s alone.