Why the upper mantle’s carbon isotopes needed a rewrite
A plain-language explainer of our 2026 Geochemical Perspectives Letters paper revising the carbon isotope composition of the convecting upper mantle — what we measured, why it differs from textbook values, and what it means for the long-term carbon cycle.
The paper: Shea JJ, Maclennan J, Edmonds M, Hughes EC, Hartley M, Mikhail S, Perfit M, Shorttle O. (2026). A revised carbon isotope composition of the convecting upper mantle. Geochemical Perspectives Letters, accepted.
TL;DR
The textbook value for the carbon isotope composition of mid-ocean-ridge basalt (MORB) source mantle — roughly δ¹³C ≈ −5 ‰ — is, it turns out, a little off. Using a new high-precision SIMS protocol on olivine-hosted melt inclusions, we find the convecting upper mantle is isotopically lighter than previously thought, with implications for how much subducted carbon recycles back, and how we balance the long-term carbon cycle.
Why it matters
Carbon is the volatile element that most directly couples Earth’s interior to its climate. Volcanic CO₂ outgassing balances silicate weathering on timescales of ~10⁵–10⁸ years, and the long-term δ¹³C composition of the atmosphere–ocean system reflects an exchange between biological, surface, and mantle reservoirs. If we don’t know the mantle endmember properly, we can’t close the budget.
The “−5 ‰ mantle” is a number that’s been propagating through the literature for decades. It comes mostly from older bulk-sample CO₂ extractions from MORB glasses and vesicles, with real and well-known issues around degassing, contamination, and atmospheric exchange.
What we did
Two things, in combination:
Better samples. We targeted olivine-hosted melt inclusions from undegassed MORB. Inclusions trapped at depth preserve primary volatile contents, and (importantly for carbon) they avoid the runaway degassing that compromises pillow-rim glass.
Better measurements. We used the SIMS protocol we developed in 2025 to measure carbon concentration and δ¹³C concurrently in the same spot, with precision tight enough to resolve a few permil — the scale of mantle-source variability.
The reference materials we developed for that 2025 method paper are what made this possible. Without well-characterised glasses spanning a range of carbon contents and known δ¹³C, you can’t correct for instrumental mass fractionation properly, and δ¹³C results just drift.
What we found
The convecting upper mantle, as sampled by undegassed MORB melt inclusions, has a δ¹³C lighter than the conventional value. (The exact number, with uncertainties, lives in the paper — I’ll update this post once the proofs are out.)
Two things follow:
- The recycled-carbon budget changes. The mass-balance between the mantle, subducted carbonates (heavy δ¹³C), and subducted organic carbon (light δ¹³C) gives a different flux ratio with the new mantle endmember. The first-order conclusion: less of the carbon outgassed at ridges is recycled organic.
- Hotspot–MORB contrasts get sharper. A separate, in-prep paper on Iceland shows the primordial reservoir there is heavier than the upper mantle — and that contrast is larger once you’ve corrected the upper mantle baseline. More on that one later.
How to read the figures
The headline plot is a histogram of δ¹³C from MORB melt inclusions, with the published bulk-glass values overlain for context. The shift between the two distributions is the result. The key thing to look at is the mode, not the mean — a few mixed samples in the bulk-glass dataset pull the average around in ways that aren’t representative of the source.
What’s next
A natural follow-up is plagioclase-hosted inclusions and gabbroic cumulate samples — the work I started in January 2026. Spreading-ridge cumulates redistribute volatiles in ways that we don’t have a good handle on, and getting δ¹³C in those phases will close another loop.
If you work on carbon-cycle modelling and want the numbers in a usable form, email me — happy to share the dataset and the isotope corrections.
This post is a plain-language summary. For methods, uncertainties, and full context, please see the paper.