Tracing Uncertainties of Extreme El Niño Prediction to Mesoscale Ocean Currents

Bo Qin, Zeyun Yang, Qiaofei Xue, Yanan Ding, Jinyu Wang, Chaopeng Ji, Shijin Yuan, Bin Mu, Wansuo Duan, Guokun Dai, Yi DengarXiv 2026 · arXiv

TL;DR

Researchers have identified mesoscale ocean currents—small-scale eddies and flows—as a critical "hidden" source of uncertainty in predicting extreme El Niño events. By using a differentiable AI model, the team demonstrated that tiny initial errors in these currents can cascade into massive forecast failures, potentially explaining the notorious "spring predictability barrier" in climate modeling.

What Problem Does This Paper Address?

Predicting the El Niño-Southern Oscillation (ENSO) is vital for global disaster preparedness, yet even our best models often stumble. Historically, scientists blamed these failures on large-scale errors in initial conditions, such as inaccurate measurements of total ocean heat content or surface wind stress.

However, there is a persistent "Spring Predictability Barrier" (SPB)—a period during the boreal spring when forecast accuracy drops sharply. This paper investigates whether the root of this uncertainty actually lies in much smaller, "mesoscale" oceanic processes (features spanning 10–100 km, like eddies) that are often poorly resolved in standard climate models. The challenge has been that traditional dynamical models are too computationally expensive to run the complex optimization needed to trace these small-scale errors back to their source.

Methods

The authors bridged the gap between high-resolution oceanography and climate prediction using two key tools:

  1. Differentiable Ocean AI Model: They built a data-driven forecasting model using an Encoder-Processor-Decoder architecture featuring Swin-Transformer blocks. Because the model is "differentiable" (built using deep learning frameworks), they could calculate gradients efficiently.
  2. Conditional Nonlinear Optimal Perturbation (CNOP): This mathematical method identifies the specific initial "nudge" (perturbation) that leads to the largest possible forecast error at a later time.

By coupling these, the researchers could perform "sensitivity analysis on steroids," finding exactly which tiny current errors in December would most effectively ruin a forecast for the following winter.

Datasets

The study relied on the GLORYS (Global Ocean Reanalysis and Simulations) high-resolution product.

  • Resolution: 1/12° horizontal resolution with 9 vertical levels in the upper 100 meters.
  • Scope: Tropical-subtropical Pacific Ocean (45°S – 45°N).
  • Timeframe: Training on data from 1993–2020; evaluation on subsequent years.
  • Variables: Sea temperature ($T$), zonal current ($U$), and meridional current ($V$).

Key Insights and Ideas

  • The Butterfly Effect is Event-Specific: The "worst" errors aren't universal; they depend on the specific El Niño event. For the 1997/98 event, the forecast was most sensitive to current errors in the Eastern Pacific. For the 2015/16 event, the sensitivity shifted to the Western Pacific.
  • Eddy-Driven Cascades: Small current errors don't just stay small. They change the count and kinetic energy of mesoscale eddies. Specifically, an increase in cyclonic (cold) eddies can suppress subsurface warming, eventually triggering large-scale temperature errors that stall El Niño development.
  • Mapping to the Barrier: The growth rate of these mesoscale errors peaks exactly during the spring months, providing a physical mechanism for the "Spring Predictability Barrier."
Figure 1. ENSO forecast skill evaluated by Niño indices. (a)-(c) show 𝐴𝐶𝐶 of Niño 3.4, Niño 4, and

Key Results

  • High Precision: The AI model maintained effective forecast skill (ACC > 0.5) for the Niño 3.4 index for up to 81 pentads (over 13 months), rivaling or exceeding several major operational forecast systems.
  • Small Error, Big Impact: The researchers found that an initial current error with an amplitude of only 5% of the mean anomaly was enough to transform a predicted extreme El Niño into a neutral state in the model.
  • Regional Sensitivity:
    • 1997/98: Errors in the East region ($5^\circ N–14^\circ N, 124^\circ W–96^\circ W$) suppressed local subsurface warming.
  • 2015/16: Errors in the West region ($4^\circ N–10^\circ N, 136^\circ E–168^\circ E$) reorganized the upper ocean structure and sent "error signals" eastward via the equatorial waveguide.
Figure 2. Spatial patterns of the most influential initial current errors (CNOP) for the two events. Note the distinct East-West contrast in where the kinetic energy errors are concentrated.

Limitations and Future Work

  • Air-Sea Coupling: The current model focuses on the ocean. It does not yet fully incorporate the complex feedback loops between the ocean and the atmosphere (like wind-stress response to SST changes).
  • Sample Size: The CNOP experiments focused on two specific extreme events. The authors acknowledge that larger ensembles are needed to see if these sensitivities hold across all types of ENSO events (e.g., Central Pacific vs. Eastern Pacific types).
  • Observation Strategy: These findings suggest that future ocean observing systems should prioritize high-resolution current measurements in the identified "sensitive regions" to improve long-lead climate forecasts.

Summary updated

Abstract

The predictability of the El Niño-Southern Oscillation (ENSO) has historically been linked to large-scale errors in models' initial conditions such as heat content and surface wind stress. Here we demonstrate that these slowly evolving errors may ultimately originate from mesoscale errors in initial oceanic states that are common in current models and ocean observation systems. This hidden source of ENSO forecast uncertainty, in terms of mesoscale ocean-current errors, is convincingly exposed by the conditional nonlinear optimal perturbation (CNOP) method applied to a differentiable ocean AI model. The CNOP-identified most influential current errors are strongly event-dependent. For the 1997/98 El Niño, the initial current errors in the eastern Pacific results in the largest forecast error through locally suppressing original subsurface warming. For the 2015/16 event, the dominate error originates from the western Pacific, which reorganizes upper ocean temperature structure and affects the eastern Pacific through the equatorial waveguide. Further analysis reveals that these current errors perturb mesoscale eddy count and eddy kinetic energy triggering large-scale temperature errors that hinder El Niño development. The timing of the maximum growth rate of the obtained CNOP error is well aligned with the spring predictability barrier of ENSO, and the error structures are reflected in simulated analysis errors caused by unresolved mesoscale currents. These findings extend the potential source of ENSO predictability beyond large-scale thermal signals, reveal for the first time an uncharted route of ENSO forecast failure, and provide an important physical basis for targeted ocean observations and data assimilation in the future.
BibTeX
@misc{qin2026tracing,
  author = {Bo Qin and Zeyun Yang and Qiaofei Xue and Yanan Ding and Jinyu Wang and Chaopeng Ji and Shijin Yuan and Bin Mu and Wansuo Duan and Guokun Dai and Mu Mu and Yi Deng},
  title = {{Tracing Uncertainties of Extreme El Niño Prediction to Mesoscale Ocean Currents}},
  year = {2026},
  eprint = {2611.17854},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.17854},
}
Figure