The iron fertilization hypothesis is simple and elegant: phytoplankton in the Southern Ocean are iron-limited; melting Antarctic ice delivers iron; more iron means more phytoplankton, more carbon uptake, a negative feedback on warming. The hypothesis has driven decades of research, several geoengineering proposals, and a persistent hope that the ocean might partially buffer climate change on its own.
Struve and Winckler analyzed a sediment core spanning 500,000 years from the Pacific sector of the Southern Ocean. Iron input peaked during warm intervals, exactly when you'd expect if icebergs were delivering it. But algae productivity did not increase. In fact, the correlation runs the opposite direction — more iceberg-delivered iron, less carbon uptake.
The resolution is mineralogical. Beneath the West Antarctic Ice Sheet sits geologically ancient, highly weathered bedrock. When icebergs calve and scrape across this substrate, they transport minerals that are iron-rich in total but iron-poor in bioavailable form. The weathering has already converted the soluble iron into insoluble oxides. The ocean gets flooded with iron it cannot use.
This is a sign error created by measuring the wrong variable. Total iron and bioavailable iron have opposite relationships to warming in this system: warming increases total iron delivery (more calving, more iceberg transport) while simultaneously ensuring that the delivered iron is geochemically useless (the source rock is more weathered in warmer conditions). The quantity goes up. The quality goes down. The product — which is what the phytoplankton actually respond to — goes down. Models built on total iron predict the wrong sign.
The pattern appears elsewhere. E. coli chemosensing was assumed to be limited by the external molecule arrival rate (the Berg-Purcell limit); the actual bottleneck is internal signal transduction noise, which operates at a hundred times lower capacity. Cleaner wrasse mirror tests measured habituation time and called it cognitive development time. Morphological taxonomy counts species at a resolution that is exactly half the genetic species resolution, consistently, across all vertebrate groups.
In each case, the obvious variable is easier to measure than the correct one. Total iron is measurable by mass spectrometry on sediment cores. Bioavailable iron requires chemical speciation analysis under conditions that match the actual ocean environment. The easier measurement is more publishable, generates more data, and builds a larger literature — which makes it harder to question.
The iron fertilization geoengineering proposals were built on the total-iron assumption. If the wrong-sign feedback holds, dumping iron into the Southern Ocean won't increase carbon uptake — it depends on the form of iron, not the amount. The variable selection bias doesn't just produce wrong academic conclusions. It produces wrong interventions.