Tropical corals are extraordinary natural archives. As they grow, sometimes for centuries, they lay down annual layers of calcium carbonate, quietly recording the temperature of the surrounding seawater in their skeleton chemistry. By measuring the ratio of strontium to calcium (Sr/Ca) or the composition of oxygen isotopes (δ¹⁸O) along a coral core, researchers can reconstruct tropical sea surface temperatures at monthly resolution, stretching back far beyond the era of thermometers.

But there has long been a nagging problem. Coral records consistently suggest that tropical temperatures varied far more from decade to decade than either climate models or instrumental data would imply. The debate has run for years: are climate models systematically underestimating natural variability, or is something going on inside the corals themselves? Our new study suggests it is largely the latter.

Pairing Up Corals to Find the Truth

Our key insight was conceptually simple. Two coral colonies growing on the same or nearby reefs, just metres or at most a few kilometres apart, experience essentially the same ocean temperatures. Any genuine climate signal should appear in both records. Biological quirks, slow drifts in a coral's own physiology, would be unique to each colony. By analysing pairs of closely located corals and asking what they share versus what they don't, we can mathematically separate the climate signal from the noise.

We applied this approach to the PAGES CoralHydro2k database, a major international compilation of coral records from across the tropics and subtropics. From this we identified 58 pairs of Sr/Ca records and 31 pairs of δ¹⁸O records from colonies growing close enough together that their local temperatures should be nearly identical at inter-annual and longer timescales. We then analysed the frequency content of the records by estimating power spectra of the signal and the noise, essentially asking: at each timescale, from seasons to decades to centuries, how much of what is recorded in the coral is shared between colonies, and how much is unique to each individual?

A Noise Problem That Grows With Timescale

The results were striking. Across both chemical tracers, and across all timescales longer than the annual cycle, individual coral records contain a large non-climate noise component. Crucially, this noise is not random from year to year, it is strongly autocorrelated, meaning it drifts slowly over time. The coral does not simply add white noise on top of the temperature signal. Instead, it introduces slow, wandering variations that look, superficially, exactly like long-term climate change.

Once we subtract the estimated noise from the raw coral records, the power spectra of the remaining climate signal align well with spectra of instrumental sea surface temperature records. The uncorrected coral variance, however, was inflated by a factor of two to seven compared to this corrected estimate, meaning individual coral records have been overstating the amplitude of decadal to centennial temperature swings by factors of 1.4 to 2.6.

What Is Causing the Noise?

We can't say definitively, but we think the most plausible explanation involves what coral biologists call "vital effects", the ways in which a coral's own physiology influences the chemistry of its skeleton. Corals are living animals, and the biochemistry of skeleton formation is not a simple passive recording of seawater temperature. Growth rate, the composition of the coral's symbiotic algae, and responses to thermal stress events can all subtly shift the chemical signal. If these biological factors drift slowly over the lifetime of a colony, what we might call "wandering vital effects", they would produce exactly the kind of slowly-varying, non-climatic noise we find.

The Fix: Replicate

The good news is that a clear path forward exists. Although the noise is large within a single coral colony, it appears to be largely independent between different colonies. Average two nearby records together, and much of it cancels out. The climate signal survives; the biological quirks largely do not. Even stacking just two replicates is sufficient to dramatically improve the reliability of temperature reconstructions.

This reinforces calls, made by others before us, to prioritise replication when collecting new coral records. Where possible, new cores should be taken near existing ones so paired analyses can be performed. Existing databases like CoralHydro2k, where multiple records from the same sites already exist, offer an opportunity to revisit past conclusions with this correction applied.

Corals remain irreplaceable archives of tropical climate history. But like any witness, they are most reliable when their testimony can be cross-checked.

The paper "Strontium to calcium ratio and oxygen isotopic coral records can exaggerate past decadal tropical climate variability" is published in Communications Earth and Environment. This paper was produced as part of the Priority Programme “Tropical Climate Variability & Coral Reefs” (SPP 2299), funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation). The CoralHydro2k database used in this study was compiled by the PAGES CoralHydro2k working group.