A Holocene Temperature Reconstruction Part 3: The NH and Arctic

By Andy May

In the last post (see here) we reexamined the Marcott, et al. (2013) proxies for the Southern Hemisphere mid-latitudes and the tropics. In this post, we will present two more reconstructions using their proxies, these are for the Northern Hemisphere mid-latitudes (30°N to 60°N) and for the Arctic region (60°N to 90°N). These two regions contain over half of the proxies used in this study. The next post will present a global area-weighted composite temperature reconstruction. As we did in the previous two posts, we will examine each proxy and reject any that have an average time step greater than 130 years or if it does not cover at least part of the Little Ice Age (LIA) and the Holocene Climatic Optimum (HCO). We are looking for coverage from 9000 BP to 500 BP or very close to these values. Only simple statistical techniques that are easy to explain will be used.

Northern hemisphere mid-latitudes

There are 10 proxies that meet our basic criteria for the Northern Hemisphere reconstruction, although two of them are combined into one record. The final reconstruction is shown in figure 1. Figure 1A includes all proxies that meet our basic criteria, figure 1B excludes two anomalous proxies and trims the early data from two more to avoid spikes caused by proxy drop out. The R code, input and output datasets can be downloaded here.

Figure 1A, all proxies that meet the basic criteria (resolution and span)

Figure 1B, excludes KY07-04-01 and OCE326-GGC26

If all proxies are included, as in figure 1A, this reconstruction shows a very flat Holocene Climatic Optimum (HCO) from 10000 BP to 6800 BP and then a steady decline in temperatures to the Little Ice Age (LIA) around 240 years ago (180 BP or about 1770 AD). The range of Holocene temperatures in both reconstructions is 4°C, this is the largest range of any region, including the Arctic. We generally prefer the reconstruction in figure 1B and will discuss the features of this reconstruction here. Since the temperature change in this reconstruction exceeds that seen in the Antarctic and Arctic reconstructions, it calls into question the concept of “Polar Amplification.” We cannot say polar amplification does not exist, but we do not see evidence of it in these proxies. Excluding the two anomalous proxies the coldest portion of the LIA was around 1610 AD.

The 17th and 18th centuries were a time of intense cold weather in Europe, Asia and North America, these centuries were the worst part of the LIA. The early 18th century saw lakes freeze solid in Italy and ice skating took place in Venice. Ships were frozen into ice in New England in 1740. More stories of the severe cold in the Northern Hemisphere in the 18th century can be seen here. The 17th century, if anything, was worse. The 17th century revolutions, droughts, famines, wars and other calamities are detailed in Geoffrey Parker’s book Global Crisis.

In Parker’s book, we see historical records of unusually cold and devastating winters that occurred in Europe and the Middle East in 1620, the United States between 1640 and 1644, China in 1640, Hungary between 1638 and 1641. 1641 remains the coldest year ever in Scandinavia. In the Balkans, in 1654, wine and olive oil froze in jars. In Egypt, in the 1670’s, a country where furs were unknown, was so cold that the citizens started wearing fur coats. Crop yields plunged in Guangxi and Guangdong (Hong Kong area) in southern China due to very cold weather in 1633 and 1634. Icebergs floated down the Thames River in January of 1649 as Charles Stuart was beheaded. In 1698 it was reported, in London, by John Evelyn that the weather was colder than anyone could remember. Harvests failed in Scotland every year between 1688 and 1698 mainly due to cold. And the stories go on and on.

The highest Medieval Warm Period (MWP) peak is at 890 AD. The Medieval Warm Period is very tepid in this reconstruction. Some of the proxies show a bump near the historical MWP and some do not. Below are plots of each set, figure 2 is the set with a visible MWP and figure 3 is the set without.

The proxies with an apparent MWP in figure 2, reach their peaks at different times and they do not line up well, this spreads out the MWP in a reconstruction and dampens the amplitude. The only two that line up are Flarken Lake (Sweden) and D13822 (Portugal). The MWP peak in the MD01-2421 composite from Japan occurs a little later it does in the Newfoundland proxy OCE326-GGC26. MD95-2015 (southwest of Iceland) is a very anomalous proxy with peaks at 1110 AD and 760 AD. In short, in this reconstruction, while it appears the LIA is well defined, the MWP is not. The historical warming from around 760 AD to 1200 AD shows up in these proxies, but not as a single well-defined event.

Figure 2

The Northern Hemisphere proxies in figure 3 do not have a noticeable temperature anomaly in the MWP. KY07-04-01 is in the East China Sea, south of Japan. CH07-98-GGC19 is off the US east coast near Washington, DC; it shows a minor bump around 1060 AD to 900 AD. OCE326-GGC30 is near Nova Scotia, Canada; it shows no response at all. The IOW merged dataset is from the Baltic Sea near Sweden and it also shows no MWP response. These proxies run counter to historical records for this time period.

Figure 3

The Roman Warm Period peak is at 90 BC (figures 1A and 1B) and very noticeable in the reconstruction. So, we see the LIA and the Roman Warm Period here, but the MWP not so clearly. This could be because the proxies are erroneous or because the MWP occurred at different times in different areas and was dampened by averaging. The MWP exists, it is a matter of historical record, but it does not show up well in these proxies.

All nine proxy records are shown in figure 4A.

Figure 4A, all proxies

Figure 4B, proxies used

The anomalous records in figure 4A are OCE326-GGC26 (Sachs 2007, near Newfoundland), KY07-04-01 (Kubota et al., 2010, just south of Japan) and Flarken Lake (Seppa et al., 2005, in Sweden). Flarken Lake is probably being affected by meltwater from glaciers that remained in the area long after the last glacial maximum. The retreating ice delayed the Holocene Climatic Optimum in many northern areas (Bender, 2013). We do not think Flarken Lake was a problem and retained the proxy.


This proxy is just south of Newfoundland and near the Grand Banks. See the location map in figure 5. This proxy record is plotted alongside its neighbor, OCE326-30GGC, in figure 6. Both proxies agree well from 8000 BP to 0 BP, then OCE326-GGC30 flattens out like most of the Northern Hemisphere proxies and OCE326-GGC26 makes a large jump in temperature ending with a 7°C anomaly 11410 BP. This proxy is problematic and was excluded from the reconstruction.

Figure 5 (source: Sachs, 2007)

Figure 6


This proxy was also excluded from the reconstruction for being anomalous. The core is from the East China Sea near the southern tip of Japan. See the map in figure 7.

Figure 7 (Source: Kubota, et al., 2010)

Figure 8

The KY07-04-01 proxy is plotted in figure 8. The proxy is very flat from the present day to 10000 BP, with minor fluctuations up and down. This is a Mg/Ca proxy and the core is located near the mouth of the Changjiang (or Yangtze) River. This river is the largest in China, and has its mouth north of Shanghai. The water here is a varying mixture of fresh water from the river and sea water from the East China Sea. The composition of the water varies with monsoon intensity. The river discharge has also varied during the Holocene as inland glaciers melted. Finally, as noted in Kubota, et al. (2010), this temperature proxy does not compare well with other proxies in the area. The other proxies show normal cooling during the Holocene, see the lower portion of figure 9, which is taken from Kubota, et al., 2010. We chose to exclude this proxy from the reconstruction.

Figure 9 (source: Kubota, et al., 2010)

A map of all the Northern Hemisphere proxy locations can be seen in figure 10. For this region, we have a widespread set of proxies.

Figure 10

The Northern Hemisphere proxies represent a larger range of temperatures and a larger range of temperature anomalies than the other regions. For this reason, proxy drop out at the beginning of the proxy records, about 12000 BP, causes larger than normal temperature fluctuations. This is easily seen in figure 4A between 12000 BP and 10000 BP. Even after excluding KY07-04-01 and OCE326-GGC26, unrealistic fluctuations appeared as proxies ended in the early time and dropped out. To avoid this, we deleted the earliest 3 samples of OCE326-GGC30 and the earliest 20 samples of CH07-98-GGC19. By comparing the right side of figures 4A and 4B you can see what was eliminated.

Proxy drop out at both ends of the reconstructions is a problem. We are using anomalies from the mean for these reconstructions which helps, since the anomalies tend to have similar ranges. But, in the case of the Northern Hemisphere, even the anomalies have widely different values and depending upon the order in which they drop out, they can cause strange spikes at the beginning and the end of each reconstruction. The earliest few values and the last few values for the Northern Hemisphere proxies are shown in figure 11 as an example.

Figure 11

At 0 BP (1950 AD, the upper panel) we have no values for four of the proxies, “NA” means no value. The three remaining proxies have values of -1.698, -3.422, and -2.323, that average to -2.481. At -20 BP (1970) we only have two values as GGC30A has dropped out, so the average is a very different -1.86. Compare this to the situation at very early time where the MD012421 proxy is about -5.5 and the GGC30 proxy is 1.84 and you can see the problem with drop out in the Northern Hemisphere. This problem is much less pronounced in the other regions which show less variability. We only trimmed excessive values like this in the Northern Hemisphere.

Arctic reconstruction

The Arctic reconstruction is shown in figure 11. The R code and input and output datasets can be downloaded here.

Figure 11

The lowest point in the LIA in this reconstruction, occurs at 1850 AD. There is no well-defined Medieval Warm Period, but there are peaks at 850 AD and 1070 AD. The Roman Warm Period is seen from 270 BC to 50 BC. Figure 12 plots the nine component proxies.

Figure 12

A map showing the proxy locations is presented in figure 13. The proxies are all in the north Atlantic area, but widespread.

Figure 13

Only JR51GC-35 and GIK23258-2 look a little anomalous, but not severely so. GIK23258-2 (Sarnthein, et al., 2003) is the most northerly proxy with a latitude of 75°N. This may explain the slightly anomalous warm anomalies at about 9000 BP and between 2500 BP and 1000 BP. The Iceland proxy JR51GC-35 is quite spiky. The location of the JR51GC-35 is shown in figure 14. It is in an area where multiple currents can influence the temperature quite dramatically, which probably explains the spiky nature of the curve.

Figure 14


These regions have the most data and are probably well represented by the proxies. The Northern Hemisphere mid-latitude reconstruction is quite different from the other regions. Most of the regions show a Holocene temperature variability of ±1°C whereas the Northern hemisphere reconstruction shows a temperature variability of ±4°C.

Considering the abundant historical evidence from the Northern Hemisphere for the Medieval Warm Period, it is odd that this climatic event does not show up well in the Northern Hemisphere reconstruction. It is possible that this warming event took place in different areas at different times and this smeared and dampened the record. The variation in the Northern Hemisphere proxies suggests that the climatic history of the Northern Hemisphere was very complex during the Holocene, relative to the other regions.

We see no evidence of polar amplification in these reconstructions. The Northern Hemisphere mid-latitudes shows a larger range of temperatures than either the Arctic or the Antarctic.

The Northern Hemisphere reconstruction illustrates the problem with proxy selection and with temperature proxies in general. Proxies are not thermometers, they do not measure temperature directly. They react to temperature in the present day in a certain way and we assume they react in the same way in the distant past. How accurate are the temperature estimates? Further, we assume that burial and time have had no effect, or a predictable effect on the quantities measured. We assume that we have measured the age and depth or height of each sample accurately. Finally, we assume that each proxy represents the surface temperature of a very large area with no local distortions. So, how do we choose which proxies to include and which to reject? Our basic requirements of a span of 9000 BP to 500 BP and a resolution better than 130 years are reasonable. Was it reasonable to reject KY07-04-01 and OCE326-GGC26? Perhaps, but it is hard to tell, the decision was mostly subjective.

In the next post, we will present a global reconstruction. We will also discuss the various sources of error in the proxies.

The R code and input and output datasets for the Arctic reconstruction 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.