Ocean Temperature Proxies Disagree on Last Interglacial Warming

The Last Interglacial, roughly 130,000 to 115,000 years ago, is the closest natural analogue we have for a world with CO2 near 280 parts per million. Proxy records disagree on how warm it was by several degrees Celsius. That spread, not any single number, is the honest state of the evidence.

The 125,000-Year-Old Temperature Puzzle

Marine Isotope Stage 5e sits inside the Last Interglacial and lasted a few thousand years around its peak. Atmospheric CO2 was about 280 ppm, close to preindustrial values. Global mean temperature was warmer than the late Holocene, but the magnitude is genuinely contested. Some proxy compilations put the anomaly near 2 degrees Celsius above preindustrial. Others land closer to 5 degrees at high latitudes.

The disagreement is not about whether the Last Interglacial was warm. It is about how warm, where, and in which season. That distinction matters because the Last Interglacial is used as a test case for how sensitive the climate system is to CO2. If the true anomaly was modest, the implied sensitivity is lower. If it was large, the opposite follows.

Ice cores, marine sediment cores, and fossil coral reefs all contribute to the picture. Each records a different slice of the ocean-atmosphere system. None records a global annual mean directly. The work of comparing them is where the controversy lives.

How Proxies Read Ancient Oceans

Foraminifera, single-celled organisms with calcium carbonate shells, incorporate magnesium in place of calcium as a function of temperature. The Mg/Ca ratio in fossil shells from marine sediment cores is a standard sea surface temperature proxy. The relationship is calibrated against modern core-top samples, which is a strength and a weakness.

Alkenones are long-chain organic molecules produced by haptophyte algae. Their degree of unsaturation tracks growth temperature. Because these algae live in the surface mixed layer, alkenone records are often read as annual mean sea surface temperature, though the seasonality of production varies by region.

Coral Sr/Ca ratios and ice-core deuterium provide independent constraints. Corals give sub-annual resolution but only at tropical reef sites. Ice-core deuterium reflects condensation temperature over the polar plateau, which is not the same as sea surface temperature. Each proxy has a distinct seasonal bias, and that bias propagates into any reconstruction.

Beyond these, the TEX86 proxy, based on membrane lipids of marine archaea, offers another window, though its applicability to the Last Interglacial remains debated because the archaeal community structure may have differed. The point is that no single proxy is a perfect thermometer; each carries assumptions that may not hold in a warmer world.

Where the Records Clash

Tropical sea surface temperature estimates for the Last Interglacial spread across roughly 1 to 2 degrees Celsius in some compilations and 3 to 4 degrees in others. That gap is larger than the signal many studies are trying to detect. High-latitude amplification is even less constrained, because polar proxy coverage is sparse and seasonally skewed.

Age models differ by thousands of years between cores. A marine core dated by tuning to a neighboring record can shift the timing of peak warmth relative to an ice core by a full precession cycle. When records are stacked into a global mean, those offsets blur the peak and flatten the amplitude.

Ice-volume effects contaminate the signal. A proxy that responds to both temperature and salinity will misread meltwater events as cooling. This site has argued in a related piece that provenance decisions can flip a result, and the same logic applies to proxy calibration choices.

Named Studies and Their Numbers

Hoffman and colleagues, writing in 2017, reviewed the Last Interglacial evidence and concluded that global mean temperature was roughly 0.5 to 2 degrees Celsius above preindustrial. Turney and Jones, in 2010, estimated a global anomaly of about 1.5 to 2 degrees Celsius. The CAPE-Last Interglacial project, a multi-proxy effort, reported regional anomalies in the 2 to 3 degree range.

Shakun and colleagues, in 2015, used a global compilation of proxy records and arrived at roughly 1.5 to 2.5 degrees Celsius above preindustrial. These numbers are not wildly inconsistent with one another. The problem is that each carries a different seasonal and spatial weighting, so they are not strictly comparable.

A related commentary on this site noted that effect sizes often shrink under larger samples, and paleoclimate compilations show a similar pattern when new cores are added to an existing stack.

Why the Disagreement Persists

Proxy calibration depends on modern analogues that may not exist. The tropical Pacific of the Last Interglacial had no exact modern counterpart, so transferring a modern Mg/Ca-temperature relationship into that world assumes a stability that is difficult to verify. When the assumption fails, the error is systematic, not random.

Seasonal bias skews annual means. A proxy that records summer growth will report warmer values than one that integrates the year. If different proxies in a compilation peak in different seasons, the stack inherits a seasonal artifact that looks like spatial variability.

Chronology errors amplify at 125,000 years. Radiocarbon dating fails beyond about 50,000 years, so Last Interglacial age control relies on orbital tuning, uranium-series dating of corals, and geomagnetic markers. Each method carries its own uncertainty, and those uncertainties compound when records are aligned.

Another factor is diagenesis: the chemical alteration of fossils after deposition. Foraminiferal shells can recrystallize, altering their Mg/Ca ratios. Corals can experience secondary aragonite cementation that shifts Sr/Ca values. These processes are often invisible in the final data but can bias reconstructions by amounts comparable to the signals being studied.

Deep-Time Analogues and Their Limits

The Mid-Pliocene Warm Period, about 3 million years ago, offers another test case. CO2 was around 400 ppm, similar to today, and global temperatures were 2 to 3 degrees Celsius above preindustrial. Proxy coverage is better than for the Last Interglacial because more cores have been recovered, but the same calibration issues apply. The Pliocene has been used to argue for high climate sensitivity, yet the spread among proxies remains wide. The lesson is that even with better coverage, the range persists.

The Last Glacial Maximum, about 20,000 years ago, provides a colder analogue. Proxy compilations there agree more tightly, partly because the signal is larger and partly because the ice sheets were extensive, leaving more geological evidence. The contrast is instructive: when the signal-to-noise ratio is high, proxies converge; when it is low, as in the Last Interglacial, they diverge.

Reading Past Warmth for Future Risk

Compare proxy spread rather than quoting a single value. If the range is 1 to 4 degrees Celsius, report the range. A point estimate hides the very uncertainty that matters for climate sensitivity.

Weight records by seasonal coverage. A compilation that mixes summer-only and annual-mean proxies will produce a biased stack. Check whether the studies you cite report seasonality explicitly.

Check age-model uncertainty before comparing records. Two cores that appear to disagree may simply be misaligned by a few thousand years. The apparent temperature conflict can dissolve once the chronologies are reconciled.

Cross-validate with independent proxies. Mg/Ca, alkenones, coral Sr/Ca, and ice-core deuterium respond to different physical variables. Agreement across them is stronger evidence than agreement within one proxy family.

Treat the Last Interglacial as an analogue, not a forecast. It was a different world with different ice sheets and different orbital forcing. Its value is in testing how the climate system responds to CO2, not in predicting a specific future temperature.