The core insight of the new work is that when only a partial view of a complex system is available, the observed variables become history-dependent. In the language of dynamical systems, the SST-only evolution is effectively non-Markovian: knowing the temperature field today is not sufficient to determine the field next month, because hidden variables—the atmospheric winds, the deeper ocean, unresolved eddies—carry memory that leaks into the surface record. Traditional approaches either ignore this memory or approximate it with time-delay coordinates, the classic Takens embedding, in which past snapshots are stacked into a vector. Sugiura and colleagues take a different and arguably more faithful route: instead of collapsing history into an unordered vector, they treat each year of SST data as an ordered path, a continuous trajectory through a high-dimensional space that preserves the temporal sequence of monthly anomalies within the year.

Representing dynamics as paths, however, creates a new mathematical challenge. Koopman analysis, the framework the authors adopt, offers a way out of nonlinearity by shifting attention from states to observables. First proposed by Bernard Koopman in 1931, the Koopman operator is a linear operator that acts on scalar-valued functions of the state, propagating them forward in time. The remarkable property is that even when the underlying dynamics are strongly nonlinear, the evolution of observables can be exactly linear—an infinite-dimensional linearity, but one that can be approximated in finite dimensions. Extended dynamic mode decomposition, or EDMD, and its kernelized variant kEDMD have become the workhorse tools for this approximation, projecting dynamics onto a chosen function space or a reproducing kernel Hilbert space. The question is which function space to use when the states themselves are entire annual trajectories.

This is where path signatures enter the picture. Rooted in the theory of rough paths developed by Terry Lyons, the signature of a path is a canonical collection of iterated integrals that captures the temporal order, area, and higher-order interactions along a trajectory. Signatures form a feature representation so expressive that sufficiently regular functionals of paths can be approximated by linear functionals of these features. Computing signatures explicitly, however, becomes prohibitive for high-dimensional data—the number of features grows combinatorially with truncation depth—and SST fields, with their vast spatial grids, are nothing if not high-dimensional. The authors sidestep this bottleneck by kernelizing the signature: following work by Kiraly and Oberhauser, they use a signature kernel that implicitly compares two paths through their iterated-integral features, computing inner products in an RKHS without ever constructing the features themselves.

The resulting pipeline is a single, coherent procedure. Monthly SST fields are first converted into anomalies using a strictly past-only rolling climatology, ensuring that no future information contaminates the preprocessing—a crucial discipline for honest out-of-sample testing. Twelve consecutive monthly anomalies are then grouped into an annual segment and embedded as a piecewise-linear path via cumulative summation, so that the path’s increments are precisely the monthly anomalies. The one-year shift operator—the map that carries one annual path to the next—is learned by applying kEDMD to Gram and cross-Gram matrices built from the truncated signature kernel, with hyperparameters tuned through a kernel-alignment objective that measures the normalized similarity between predicted and true paths. The output is a finite-dimensional Koopman matrix whose spectrum and iterates serve double duty: multiplying it forward yields multiyear forecasts, while its eigendecomposition yields oscillatory modes with well-defined periods and amplitudes.

Before confronting real ocean data, the team validated the approach on a controlled synthetic benchmark: the two-scale Lorenz–96 system, a standard testbed in which slow variables are coupled to fast ones, mimicking exactly the partial-observability situation of SST. Only the slow variables were used as observables and prediction targets; the fast variables acted as unresolved stochastic-like forcing, formally inducing memory and random forcing in the reduced slow-only description, in the spirit of the Mori–Zwanzig formalism. Across three coupling regimes, the signature-kernel method matched an explicit truncated-signature EDMD almost perfectly—confirming the kernel implementation—and outperformed baselines that used either block-mean states or conventional snapshot DMD, particularly at intermediate and longer lead times. The message was clear: the ordered structure within trajectory segments carries predictive information that averages and instantaneous snapshots destroy.

Applied to observed SST, the method delivered on its promise. Forecast experiments were conducted under two rigorous time-ordered protocols: leave-future-out evaluation for forecasting skill and leave-start-out splits for spectral diagnostics, both ensuring that model training never touched data from the verification period. Against a climatology baseline that simply repeats the anchor-year month-of-year climatology, the learned Koopman operator improved out-of-sample multiyear forecast skill, with performance measured both by area-weighted RMSE on anomaly fields and by a novel kernel-based path correlation, the kPC, which scores the similarity between predicted and true annual paths in the signature-kernel space. Equally important, the eigenspectrum of the learned one-year operator revealed coherent spectral modes organized in band-like structures in period–amplitude space, with representative modes at interannual and longer periods—the kind of structured oscillatory signal that climate scientists associate with phenomena such as the El Niño–Southern Oscillation and other basin-scale variability.

The contrast with the other dominant paradigm in modern geophysical forecasting—neural operators—is instructive. Architectures such as Fourier neural operators, and systems like OceanNet, which demonstrates competitive seasonal prediction for regional ocean dynamics, learn powerful nonlinear mappings between input and output fields and transfer well across grid resolutions. But their predictions emerge from compositions of linear and nonlinear layers, so no native eigenvalues, eigenfunctions, or Koopman modes fall out of the model; spectral diagnostics require additional post-processing or are simply unavailable. The signature-kernel kEDMD approach, by contrast, yields an explicit linear operator from which forecasting and spectral analysis follow from the same object. For scientists seeking not just predictions but physical interpretation—the identification of oscillatory patterns, their decay rates, their amplitudes in physical temperature units—this linearity is a decisive advantage.

The method also highlights a subtle conceptual shift in how memory is handled. Where delay embeddings represent the past as a static vector of coordinates, the path representation retains explicit order structure: the signature distinguishes a year in which temperature anomalies rose sharply in spring and plateaued in autumn from one in which they drifted gradually upward, even if the set of monthly values were identical. Higher-order signature terms are sensitive to increment magnitude as well as order, which is why the authors normalize cumulative paths by a data-dependent scale parameter before feature construction, ensuring that the kernel reflects relative temporal structure rather than absolute amplitude. In the SST experiments, the base kernel on the spatial field is a radial basis function with a bandwidth fixed by the dataset itself, tying the geometry of path space to the physical statistics of the ocean.

What emerges from this study is a template that could extend well beyond sea surface temperature. Any climate or geophysical variable observed only partially—ocean heat content, sea ice extent, atmospheric composition—suffers from the same effective non-Markovianity that has limited purely snapshot-based statistical models. By lifting the state along the time axis into path space and then lifting nonlinearity away through the Koopman operator, the signature-kernel pipeline offers a way to learn linear dynamics from data that is simultaneously memory-robust, scalable to high dimensions, and transparent to spectral interpretation. The authors’ demonstration that a single learned operator can both beat climatology at multiyear horizons and expose coherent modes of variability suggests that the trajectory-based view of climate dynamics may be more than a mathematical curiosity—it may be the form that the ocean’s memory actually takes.

The South Indian Ocean Dipole is characterized by alternating warm and cool sea surface temperature anomalies on opposite sides of the southern Indian Ocean. These thermal variations do not merely affect regional waters; they can cross the equator, thereby exerting significant influence on the East Asian monsoon system and modulating rainfall patterns across vast expanses of China. The complex dynamics of SIOD events have, until now, limited forecasting capabilities, with conventional numerical models relying heavily on physical simulations that often fall short of capturing intricate temporal and spatial oceanic variations well in advance.

The breakthrough, detailed in a publication in Atmospheric and Oceanic Science Letters, stems from a sophisticated deep learning architecture that assimilates vast datasets of sea surface temperatures alongside upper ocean heat content anomalies down to 300 meters. By integrating these multi-dimensional inputs, the model taps into the latent patterns embedded within the thermodynamic state of the ocean, which are essential precursors and indicators of SIOD development and intensification. This method transcends previous limitations by allowing the model to decipher complex, nonlinear interactions in the climate system that classical models might oversimplify or overlook.

Central to this innovation is the deployment of a multi-temporal convolutional neural network augmented with an attention mechanism. This hybrid architecture enables the model not only to process time-series data with fine temporal resolution but also to dynamically prioritize key features and spatial regions that hold predictive weight at different lead times. This attention-driven focus allows the system to adaptively shift its emphasis from localized oceanic signals when making short-term forecasts to more distant, teleconnected climatic drivers for longer-range predictions.

As demonstrated by the research team led by Dr. Meng Xu, the model’s behavior reveals a fascinating transition in the underlying physical drivers of SIOD predictability. For forecasts within a 2 to 3-month window, the model predominantly relies on local ocean-atmosphere feedback mechanisms within the southern Indian Ocean itself. These include interactions where wind patterns influence sea surface temperatures, which in turn modulate atmospheric circulation locally—a classic example of ocean memory effects that sustain short-term predictability.

However, when the lead time extends toward seven months, the model’s attention shifts markedly towards the central eastern equatorial Pacific Ocean, a region renowned as the epicenter of the El Niño–Southern Oscillation (ENSO). This finding underscores the critical role of remote climate teleconnections threaded through atmospheric bridges that link the Pacific and Indian Oceans. The model effectively harnesses these large-scale, cross-basin interactions, which have a profound effect on SIOD evolution by altering wind, temperature, and pressure patterns over vast geographical distances.

The emergent understanding of these temporal dependencies not only enhances forecasting skill but also provides valuable physical insights, serving as a novel example of how AI can be used to reveal climate system processes that are otherwise challenging to isolate in conventional models. Through attention analysis and sensitivity experiments, the researchers have validated the model’s interpretability, assuring that it respects established climate dynamics while benefiting from the predictive prowess of deep learning.

Another salient discovery pertains to the asymmetry between positive and negative SIOD events. The model reveals that positive SIOD occurrences are closely tied to remote La Niña conditions in the Pacific, which tend to enhance the dipole’s intensity and associated atmospheric patterns. Conversely, negative SIOD events are influenced not just by El Niño phases but also by a secondary mid-term forcing originating in the South Atlantic Ocean. This additional signal contributes to the excitation of eastward-propagating atmospheric Rossby waves, which eventually impact the southern Indian Ocean region—underscoring the complexity and interconnectedness of global climate drivers.

The implications for climate prediction are profound. By extending effective SIOD forecasts into a seven-month timeframe and elucidating the physical mechanisms behind these predictions, this study opens new avenues for improving seasonal and interannual weather forecasts across the Indian Ocean rim countries. The ability to anticipate SIOD variability with greater lead time could inform water resource management, agriculture, disaster preparedness, and other climate-sensitive sectors throughout Asia and Africa.

Moreover, this research exemplifies the transformative potential of combining state-of-the-art artificial intelligence techniques with foundational physical knowledge in climate science. Unlike black-box AI models that sacrifice interpretability for accuracy, this approach integrates the strengths of machine learning with mechanistic understanding, yielding not only better predictions but also enhanced scientific comprehension of ocean-atmosphere coupling.

Looking ahead, the research team suggests that further refinement of this deep learning framework, including the incorporation of additional climate variables and exploring multi-model ensemble integrations, could yield even more robust prediction systems. Such advancements would contribute to a new generation of climate forecasting tools that are both scientifically transparent and operationally reliable.

Ultimately, the work by Dr. Xu and colleagues represents a milestone in the evolving dialogue between artificial intelligence and climatology. By demonstrating how deep learning can be applied to capture the subtle and interconnected signals governing oceanic climate phenomena like the SIOD, this study charts a promising path toward more accurate and insightful climate prediction paradigms that are crucial in an era of intensifying climate variability and global change.

Subject of Research: South Indian Ocean Dipole (SIOD) prediction using deep learning methods

Article Title: A multi-temporal convolutional attention network for South Indian Ocean Dipole prediction

News Publication Date: 22-May-2026

Web References:
https://doi.org/10.1016/j.aosl.2026.100862

Image Credits: Meng Xu

Keywords: Deep learning, South Indian Ocean Dipole, climate prediction, ocean-atmosphere interaction, El Niño–Southern Oscillation, teleconnections, convolutional neural network, attention mechanism, sea surface temperature, ocean heat content, Rossby waves, climate modeling