Small-scale ocean current errors may undermine El Niño forecasts, study finds
A differentiable ocean AI model combined with the CNOP method traces El Niño forecast uncertainty to mesoscale current errors in initial ocean states, according to a new arXiv paper.
In plain words
El Niño forecasts have long been limited by large-scale errors in how models set up initial ocean conditions. This study argues those errors may start much smaller: in mesoscale ocean currents, eddies tens to hundreds of kilometres wide. Using an AI-based ocean model, the authors identify which small current errors grow into large forecast failures.
What was announced
The paper, posted as a cross-listed arXiv abstract, applies the conditional nonlinear optimal perturbation (CNOP) method to a differentiable ocean AI model to expose what the authors call a hidden source of El Niño-Southern Oscillation (ENSO) forecast uncertainty. They report that mesoscale errors in initial oceanic states, common in current models and ocean observation systems, may ultimately give rise to the slowly evolving large-scale errors in heat content and surface wind stress that have historically been linked to ENSO predictability.
The most influential current errors differ by event, the authors find. For the 1997/98 El Niño, initial current errors in the eastern Pacific produced the largest forecast error by locally suppressing subsurface warming, while for the 2015/16 event the dominant error came from the western Pacific, reorganizing upper-ocean temperature structure and reaching the eastern Pacific through the equatorial waveguide. The current errors perturbed mesoscale eddy count and eddy kinetic energy, triggering large-scale temperature errors that hindered El Niño development. The timing of maximum error growth aligned with the spring predictability barrier of ENSO, and the error structures appeared in simulated analysis errors caused by unresolved mesoscale currents. The authors say the findings provide a physical basis for targeted ocean observations and data assimilation.
What it does not show
- The evidence is the paper's abstract only, so the full article's methods, results and peer-review status cannot be verified from the supplied passage.
- The claims rest on the authors' own model experiments and are not independently confirmed by the supplied sources.
For security and IT teams
- Scope
- The supplied source is an arXiv abstract describing a modelling study; it covers ENSO predictability research and contains no operational or product information.
Sources
Source-support review completed 17 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.
Coverage: arXiv: abstract.
Updates and corrections
No corrections so far.
First published 17 Nov 2026
New to the subject? AI from the Ground Up starts from the beginning.