Seasonally biased proxies and the Holocene temperature conundrum

Last year, Marcott et al (2013) published a synthesis of Holocene proxy temperature records showing a warm early Holocene followed by a slight cooling in the mid and late Holocene. Liu et al, in a paper to be published in PNAS, challenge this reconstruction, showing climate model runs that show a progressive warming throughout the Holocene as ice sheets melted and greenhouse gasses increased. Liu et al argue that part of the divergence is due to seasonal biases in the proxies used by Marcott et al and show that the global Holocene trend in Marcott et al can be recreated by averaging the temperature at each proxy location using  seasonal temperature that each proxy is sensitive to rather than the annual mean.

Liu et al Fig 1. Global surface temperature over the last 22 ka:
reconstructions of Marcott et al (blue) after 11.3 ka and by Shakun et al. (cyan) before 6.5 ka, model annual mean temperature averaged over the global grid points (black), and model seasonally biased temperature averaged over the proxy sites (red). Models are CCSM3, FAMOUS (□) , and LOVECLIM (○), with the ensemble mean shown by heavy lines and individual members shown by thin lines. Temperature curves are aligned at 1 ka. The ensemble mean model annual temperature averaged over proxy sites is also shown (yellow), whose similarity to the model grid average demonstrates the insensitivity of the temperature trend to the average scheme.

When Marcott et al was published, it got a lot of attention. Tamino showed that the uptick at the end of the record was due to proxy-dropout (not all proxies records continue till the end of the reconstruction). Steven McIntyre made a lot of noise about essentially irrelevant problems with the age-depth models used by Marcott et al (they had had to make new age-depth models for their Monte Carlo methods and had made the assumption that the top of the core was 1950 unless otherwise stated. This is a commonly used assumption, but is not always appropriate).  McIntyre’s concerns about 20^th Century patterns in the alkenone records used in Marcott et al are more interesting.

I spent so much time discussing the trivial issues about the age-depth models, I had no energy for considering other potential problems with Marcott et al. The seasonality of the proxies is an obvious candidate, and which Marcott et al give due consideration. In this post, I’m going to first show how these seasonal biases in proxies can occur and then examine how Marcott et al investigated this potential problem.

Many temperature sensitive palaeoceanographic proxies have been used in the Norwegian Sea: diatoms, foraminifera and radiolarian assemblages, alkenones (biomarkers from coccolithophorids), and trace element and isotope geochemistry of planktonic foraminifera. All these proxies are predominantly produced in the spring-summer months as there is very little biological production during the cold dark winter, so we might expect all of the proxies to record a common signal. However, the different reconstructions are very different, as shown by Andersson et al 2010.

Fig. 6. (A) Mean July temperature based on pollen from Tsuolbmajavri, Finland (Seppa and Birks, 2001). Other proxies are from core MD952-11 from the Vøring Plateau (B) Alkenone derived sea-surface temperatures (Calvo et al., 2002; Jansen et al., 2008). (C) Diatom-based sea-surface August temperature estimates (Birks et al., 2002). (D) Foraminiferal-based transfer function sea-surface temperature estimates for summer (JAS) (Andersson et al., 2003; Risebrobakken et al., 2003). (E) Radiolarian-based transfer function sea-surface summer temperature estimates (Dolven et al., 2002). The alkenone and radiolarian reconstructions are included in Marcott et al.

The diatom and alkenone (U₃₇^k) reconstructions resemble one another, and share the downward trend since the Holocene thermal maximum with pollen-inferred summer temperature and the decline in summer insolation at these high latitudes as the Earth’s orbit changed. The foraminifera and radiolarian transfer function-based reconstructions have weak but opposing trends to the diatoms and alkenones. Not shown here are the isotope and trace element analyses on the foraminifera that share the same trend as the foraminiferal assemblages.

The critical difference between the proxies reconstructing a declining Holocene temperature and those reconstructing an increasing temperature is the depth habitat of the organisms. Diatoms and the coccolithophorids that produce alkenones are photosynthetic. They are constrained to live in the photic zone, the upper 50m of the water column where there is sufficient light. The foraminifera and radiolaria live over a range of depths. The dominant foraminifera in this area live between about 75 and 250 m.

The depth habitat of the organisms is important because of the seasonal pattern of stratification in the Norwegian Sea. All the organisms are living within the current of warm Atlantic water that flows into the Nordic Seas. In summer, the surface layers of this water mass absorb sunlight and become warmer and less dense, separated from the subsurface water by a thermocline. Consequently the subsurface warms little in summer. In autumn, the surface cools and stormy weather force mixing over the winter months. The temperature of the surface in winter sets the temperature in the subsurface year round. So although all the organisms are growing in summer, the temperature of the water they are living in is set by either the summer or winter.

Over the Holocene, the amount of summer insolation declined, summers became cooler, and this is reflected by the surface proxies. Winter insolation increases slightly over the Holocene. Hence we would not expect surface and subsurface proxies to show the same trends in this region. In other regions the seasonal biases attached to proxies may differ.

Marcott et al were well aware of the potential problems with seasonality and discuss it in the supplementary material. They examine the difference between alike and unalike proxies from the same 5° grid cells and find no difference in the mean difference, so assume that  “that if a seasonal bias exists between proxies, it adds no more uncertainty than that associated with proxy-temperature calibrations.” This is optimistic. First, that any bias is no larger than the variance does not imply that the result will be unbiased. Second, the test is weak as unalike proxies may be sensitive sensitive to the same season, so the number of genuine contrasts may be smaller than it appears.

Marcott et al also run an analysis akin to that in Liu et al, comparing climate model output averaged across the core sites for the annual mean temperature and the seasonal temperatures that the proxies are most sensitive to. The result is surprisingly similar to that obtained by Liu et al – the temperature increases throughout the Holocene – but the difference between the two analyses is smaller, so Marcott et al argue that with the inclusion of proxies with different seasonal biases, the biases have cancelled out.

The final test is to re-make the global Holocene reconstruction with just the proxies defined as annual proxies by the original authors. Here the reconstruction shows the same trend, but with a higher amplitude than the inclusive reconstruction, the opposite of what is expected if seasonal biases are important. This analysis depends on the original authors assignment being correct – some are dubious.

That the seasonality (and perhaps other factors) have proved to be a problem for Marcott et al does not surprise me greatly. Earlier this year I discussed Hessler et al who found that proxies from the mid-Holocene were difficult to reconcile with each other and with models. I warned that if this result was correct, Marcott et al’s reconstruction would be questionable.  

Liu et al is certainly not the last word. Even though they show that by accounting for seasonal biases in the proxies the global mean temperature trends in the model and proxies can be reconciled, the correlation between the trends in the model and individual proxies is near zero. Part of the problem may be the depth assignment of proxies – all reconstructions are assumed by Liu et al to be surface reconstructions. I showed above that depth matters, different Holocene temperature trends can occur at the same locality at different depths.