Climate science has a near-universal rule for northern ecosystems: warming means carbon loss. Higher temperatures accelerate microbial decomposition, releasing stored carbon, creating a positive feedback loop that makes the warming worse. Boreal forests lose carbon. Thawing tundra loses carbon. The sign of the relationship — warming negative, carbon negative — is treated as a physical law.
Feng Xiaojuan and colleagues broke the rule for a specific ecosystem (Nature Ecology & Evolution, 2026). Boreal Sphagnum peatlands gain carbon under warming. Across 735 paired observations from 93 studies and 16-year field experiments in Finland, the relationship reverses sign. Warming makes these peatlands store more carbon, not less.
The mechanism has three layers, each counter-intuitive. First, warming accelerates Sphagnum growth — the moss produces more biomass. That part is expected. Second, the growing Sphagnum produces more antimicrobial secondary metabolites, which suppress the microbial decomposition that warming should accelerate. The plant defends its own carbon against the organisms that warming empowers. Third — the most interesting part — Sphagnum is what the authors call a “rust engineer.” Its growth promotes the accumulation of reactive iron hydroxides in the soil, which physically bind to organic carbon molecules and protect them from decomposition. The plant manufactures its own preservative.
The third mechanism is the one that reframes the problem. Iron-mediated carbon protection was known but wasn't part of the standard warming-decomposition model because the models don't track iron cycling. When you model warming as “temperature goes up, decomposition speeds up,” the iron shield is invisible. The variable that reverses the relationship isn't in the equation.
The scale matters. Sphagnum peatlands cover roughly 20% of the boreal biome but store 40% of its carbon. The warming-induced carbon gain in these systems could offset nearly half of the boreal forest's carbon decline or the tundra's increased respiration. A biome that was assumed to be part of the problem is part of the solution, and the assumption held because nobody checked the iron.
The pattern generalizes in an uncomfortable direction. Any system-level prediction built from component-level rules will be wrong when a major component has been omitted. The component doesn't need to be exotic — iron in peat soil is mundane chemistry. It just needs to be absent from the model. The rust didn't hide. The equations didn't include it.