The scientific problem at the heart of the effort is deceptively simple to state and notoriously difficult to solve. Aerosols — microscopic particles released by sources such as wildfire smoke, industrial activity, and sea spray — serve as the seeds around which cloud droplets form. When more aerosols enter a cloud, the available moisture is distributed across a larger number of smaller droplets, which can make the cloud brighter and more reflective. Smaller droplets can also suppress rainfall, allowing the cloud to persist longer before it dissipates. Both effects change how much sunlight the cloud reflects away from the planet, and both remain among the largest sources of uncertainty in projections of Earth’s future climate.

“Small changes in these clouds can have a surprisingly large effect on Earth’s energy balance and climate,” said Jian Wang, a professor of energy, environmental and chemical engineering in the McKelvey School of Engineering and director of the Center for Aerosol Science and Engineering at WashU. Wang, the principal investigator for the project, noted that scientists still do not fully understand how aerosols change marine low clouds. The difficulty lies in separating the genuine fingerprint of aerosols from the noise of natural meteorological variability — winds, temperature, humidity, and large-scale circulation patterns that shape clouds independently of any pollution or smoke influence.

That separation problem is precisely where the new AI-driven approach comes in. The WashU team, which includes co-principal investigators Bo Li and Maxine Yu of the Department of Statistics and Data Science in WashU Arts & Sciences, plans to develop three physics-informed generative AI models trained on real satellite observations and field measurements. The data will capture episodic smoke plumes drifting from North American wildfires over the ocean, providing natural experiments in which a known aerosol source interacts with marine cloud systems. Rather than treating the atmosphere as a snapshot at a single place and time, the models will trace the full history of an air mass as it travels, incorporating the emissions and meteorological conditions it encounters along its path.

The first two models will predict what aerosol concentrations and cloud properties would plausibly look like in the absence of wildfire influence — a statistical counterfactual. The third model will then quantify the probability distributions of aerosol concentration and cloud properties under actual wildfire-smoke influence. Comparing the two worlds allows the researchers to isolate the causal effect of the smoke. “The AI creates the counterfactual: what would the aerosols and clouds likely have looked like if the wildfire had not occurred?” Yu said. “This allows us to separate the aerosol effect from the many meteorological factors that also influence clouds.”

This counterfactual strategy represents a meaningful departure from conventional approaches to aerosol-cloud research. Traditional statistical studies often correlate cloud properties with aerosol loading at a given location and moment, an approach vulnerable to confounding: meteorology influences both where smoke goes and how clouds behave, so simple correlations can misattribute causes. By following air-mass histories and explicitly modeling the world with and without the smoke, the WashU framework borrows ideas from causal inference — a branch of statistics concerned with identifying genuine cause-and-effect relationships rather than mere associations. Li and Yu contribute expertise in AI, statistical learning, environmental and spatial statistics, causal inference, and uncertainty quantification, complementing Wang’s strengths in atmospheric aerosols, clouds, and field observations.

“Bringing these perspectives together within WashU allows us to develop AI methods that are not only technically advanced, but also physically meaningful and statistically rigorous,” Li said. That emphasis on physical meaning matters because purely data-driven models can learn spurious patterns that fail when applied to new conditions. By embedding physical constraints into the generative models, the researchers aim to ensure that the AI’s outputs remain consistent with the known behavior of aerosols, droplets, and radiation, while the statistical machinery quantifies how confident scientists can be in each prediction. The result, the team says, should be a tool that improves understanding of aerosol effects on marine low clouds and strengthens scientists’ capability to predict the environment.

The stakes extend well beyond academic curiosity. According to the NSF, which is funding the work, the research could help American communities prepare for extreme weather, protect public health, and guide sound investments in infrastructure and risk management. “Even for people living far from the ocean, including here in the Midwest, understanding these processes matters because they affect our ability to predict future warming, precipitation and other climate changes,” Wang said. Marine low clouds influence regional and global temperature trends, and better constraints on how aerosols modify them could narrow the range of warming projected for the coming decades — a refinement with direct consequences for policy, planning, and climate adaptation worldwide.

The project also carries a significant workforce dimension. The researchers will train students and researchers in advanced AI methods, supporting the NSF’s priority of building a national AI-ready scientific workforce. “Training is an important part of the project, because the next generation of scientists increasingly needs to understand both the physical sciences and modern AI and statistics,” Li said. The team’s goal is to give students hands-on experience applying sophisticated machine learning to consequential real-world scientific problems, bridging the gap between atmospheric science and data science that often separates the two fields.

Beyond its immediate scientific targets, the framework is designed for reuse. The researchers plan to develop a human-in-the-loop agentic AI framework that can automate parts of the scientific workflow — from preparing and screening data to training models and analyzing results — while keeping scientists in charge of the critical decisions. “Our goal is therefore not just one scientific result, but a reusable tool for tackling similar problems in earth science,” Wang said. The AI and statistical machinery being developed could, in principle, be adapted to disentangle the effects of other environmental drivers from background weather and climate variability, offering earth scientists a general-purpose instrument for one of their field’s most persistent challenges: hearing a clear signal through a noisy atmosphere.

Subject of Research: Aerosol effects on marine low clouds and climate prediction using AI and statistics

Article Title: In the clouds

Article References: In the clouds. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: aerosols, marine low clouds, wildfire smoke, artificial intelligence, climate modeling, causal inference, generative AI, cloud physics, National Science Foundation, uncertainty quantification, climate prediction, Washington University in St. Louis

Seasonal forecasting sits in an awkward middle ground between weather prediction and climate projection. Weather forecasts lose most of their useful skill after about two weeks, because the atmosphere is chaotic and small errors grow rapidly. Climate projections, meanwhile, look decades ahead and are shaped mainly by greenhouse gas emissions scenarios. In between lies a window of months in which slow variations in the ocean and land surface — patterns that evolve over seasons rather than days — can nudge the atmosphere in predictable ways. The El Niño-Southern Oscillation, the periodic warming and cooling of the tropical Pacific, is the best-known of these memory sources, and it is precisely the kind of signal that a well-designed seasonal model can exploit.

What makes the new work distinctive is the way it dissects where spring solar predictability actually comes from. Rather than treating forecast skill as a single number, the researchers identify two leading sources of solar predictability. The first and dominant component is linked directly to primary El Niño-Southern Oscillation conditions, meaning the canonical state of the tropical Pacific during the season in question. When the Pacific is in a pronounced El Niño or La Niña phase, it reshapes atmospheric circulation patterns far beyond the tropics, altering cloudiness and hence the amount of sunlight reaching solar panels in the American Southwest.

The second component is subtler and, in many respects, more scientifically interesting. It is associated with the diversity of El Niño events themselves — not all El Niños look alike, and their different flavors have different atmospheric consequences — together with variability in the tropical Atlantic. El Niño diversity has become a major topic in climate science precisely because conventional indices can mask important differences between events centered in the eastern versus central Pacific. By separating this secondary component, the study shows that the spatial structure of forecast skill across the United States, and the physical drivers behind it, can be traced to identifiable and interpretable climate phenomena rather than appearing as an opaque statistical byproduct.

The practical implications of that interpretability are considerable. A grid operator who knows that a forecast of above- or below-normal spring sunshine is tied to a specific, monitored pattern in the tropical Pacific has some basis for judging how much confidence to place in it, and for watching whether that pattern is strengthening or weakening as the season approaches. A forecast that is merely a correlation with no physical story behind it offers no such guidance. In this sense, the study provides what the authors describe as a physically interpretable foundation for integrating seasonal forecasts into wind-solar energy planning.

The team did not stop at regional averages. They evaluated their forecasts at the level of individual states, a scale that matters enormously for real-world energy decisions because electricity markets, transmission planning, and renewable portfolio standards are largely organized along state lines. The results show robust solar prediction skill in high-resource states — those with large installed solar capacity and abundant insolation. For states whose economies and grids increasingly hinge on the sun, the finding suggests that seasonal outlooks could move from an academic curiosity toward a genuine operational tool, complementing the day-ahead and week-ahead forecasts that already guide dispatch decisions.

Perhaps the most consequential part of the research concerns what happens when wind and solar resources fluctuate together. Because solar and wind generation are both weather-driven, they can decline simultaneously, producing what energy analysts call energy droughts, or surge together, producing surpluses that stress storage and transmission. The historical co-fluctuations of wind and solar output are not random; they are shaped by the same large-scale climate patterns that govern seasonal averages. The study’s forecast system captures this compound behavior of wind-solar variability, meaning it can in principle anticipate seasons in which the two resources rise and fall in tandem — the situations that pose the greatest risk to a grid increasingly reliant on variable renewables.

Compound risk management has become a central theme in climate adaptation literature, and the energy sector is one of its most demanding arenas. A single-resource shortfall can often be covered by another generator or by storage, but a compound event — a spring in which weak winds and cloudy skies coincide across a broad region — can exhaust those buffers. By demonstrating that a seasonal model can reproduce the historical co-fluctuations associated with such events, the research opens a path toward seasonal-scale warnings of compound energy risk, giving planners months rather than days to arrange reserve capacity, adjust maintenance schedules, or coordinate fuel purchases.

The study arrives at a moment when the stakes for forecast-informed planning are rising quickly. The United States has added solar and wind capacity at a record pace in recent years, and the technical challenges of balancing a weather-dependent grid scale up with every gigawatt installed. Batteries help smooth out hour-to-hour swings, but they are not yet an economical solution to a season-long resource deficit. Hydro resources, which traditionally provide flexible seasonal storage in the western grid, are themselves climate-sensitive. In that context, even a modest seasonal signal — for example, an enhanced likelihood of a cloudy spring in the Southwest during certain Pacific states — carries real economic and engineering value if it can be trusted.

Yang and his colleagues, all based at NOAA’s Geophysical Fluid Dynamics Laboratory in Princeton apart from co-author Colleen McHugh, worked with a forecast system of the kind that underpins operational seasonal prediction worldwide: coupled ocean-atmosphere models initialized from observed conditions and run forward in ensembles, so that the spread among ensemble members provides a measure of forecast uncertainty. The work was supported through base funding from NOAA to the Geophysical Fluid Dynamics Laboratory, and the authors report no competing interests. The article was received in April 2026, accepted in September, and published in open-access form on October 3, 2026.

The broader significance of the paper lies in its framing of renewable energy forecasting as a climate science problem. Much of the current effort in renewable meteorology focuses on short horizons — nowcasting cloud fields, predicting wind ramps over hours, refining day-ahead dispatch models. The new results argue that the seasonal horizon deserves equal attention, and that the tools to fill it already exist. Two decades of investment in El Niño prediction, tropical Atlantic monitoring, and coupled model development have quietly produced a capability that the energy transition now needs.

There are, inevitably, limits to what any seasonal forecast can deliver. Skill varies by region, by season, and by the state of the climate system in a given year; years without a strong Pacific or Atlantic signal will remain difficult to anticipate. The study itself makes clear that predictability is structured, not uniform: it is the two identified components, tied to El Niño conditions and their diversity plus tropical Atlantic variability, that together explain the spatial pattern of skill. That honesty about the boundaries of predictability is part of what makes the work usable — operators can know when a forecast is likely to be informative and when it is likely to be a coin flip.

For the rapidly growing community of grid planners, commodity analysts, and climate scientists now working at the intersection of weather and energy, the message of the study is encouraging: the atmosphere’s slow ocean memory, harnessed by modern seasonal prediction systems, can illuminate spring solar resources in one of the nation’s most important solar regions months in advance, and can capture the joint behavior of wind and sun that determines how risky a given season will be for a renewable-powered grid. As the energy transition accelerates, forecasts of this kind — physically grounded, regionally validated, and transparent about their sources of skill — are likely to become as familiar to grid operators as today’s seven-day weather outlook is to the rest of us.

Subject of Research: Seasonal predictability of spring solar energy resources and compound wind-solar variability in the United States

Article Title: Seasonal predictability of spring solar resources and wind-solar variability across the United States

Article References: Yang, X., Delworth, T. L., Jia, L., Johnson, N. C., McHugh, C. E., Cooke, W. F., & Wittenberg, A. T. (2026). Seasonal predictability of spring solar resources and wind-solar variability across the United States. Communications Earth & Environment. https://doi.org/10.1038/s43247-026-04114-6

Image Credits: AI Generated

DOI: 10.1038/s43247-026-04114-6

Keywords: seasonal forecasting, solar energy, wind energy, El Niño-Southern Oscillation, renewable energy grid, energy droughts, tropical Atlantic variability, southwestern United States, climate prediction, compound variability, NOAA Geophysical Fluid Dynamics Laboratory, Communications Earth & Environment

The achievement, published in the journal Geophysical Research Letters by project researcher Shuhei Matsugishi and colleagues Masaki Satoh and Tomoki Ohno of the Atmosphere and Ocean Research Institute, is being described as the world’s first demonstration of a global large-eddy simulation, or GLES. Large-eddy simulation is a technique long used in small-scale fluid dynamics studies to resolve the largest turbulent eddies explicitly rather than modeling them statistically. Applying it to the whole planet means that the fine internal anatomy of deep convective clouds, including the storm-bringing cumulonimbus systems that deliver much of the world’s heaviest rainfall, can be represented directly. Individual updrafts and downdrafts, the vertical conveyor belts of rising and sinking air that drive a storm’s growth, become visible features of the simulation rather than assumed quantities.

To appreciate why this matters, it helps to understand the hierarchy of global models. Conventional global climate models, or GCMs, typically operate with grid cells spanning tens to hundreds of kilometers. At those scales, an entire thunderstorm fits comfortably inside a single cell, so the model must guess at its behavior using relationships derived from observations and theory. More recent global storm-resolving models, or GSRMs, narrow the spacing to between one and ten kilometers, which allows them to resolve the largest storm systems explicitly and to capture how those systems interact with larger-scale atmospheric and energy flows. Even at kilometer resolution, however, the fine-grained turbulence inside a convective cloud remains below the resolvable limit, and parameterization schemes still shape the model’s behavior in ways that can distort rainfall patterns.

Popcornlike rain is one of the clearest symptoms of that distortion. In current GSRMs, precipitation tends to appear as scattered, unrealistically intense, highly localized bursts, resembling kernels of popcorn popping across the map. These phantom downpours arise because the parameterization schemes that trigger convection at sub-kilometer scales do not communicate properly with their neighbors, so rain forms in isolated cells rather than in the broader, more organized structures seen in nature. Because the bias is systematic rather than random, it can skew a model’s estimates of where and how heavily rain falls, which in turn affects projections of flooding risk, water resources, and the atmospheric energy budget. In the Tokyo team’s 220-meter simulation, precipitation emerged in a far more realistic pattern, without those spurious localized bursts, because convection no longer needed to be parameterized at all.

The finer resolution also changes what scientists can study. As Matsugishi explained, representing the internal structure of convective clouds explicitly opens the possibility of investigating not only individual clouds but also how they interact with one another and with the larger-scale atmospheric circulation. Convection is not a purely local phenomenon: clusters of thunderstorms organize into larger systems, those systems influence the circulation around them, and the circulation in turn determines where new storms form. A global simulation that resolves each link in that chain provides a virtual laboratory in which these feedbacks can be observed directly, something that has previously been possible only in limited regional domains or in heavily simplified idealized setups.

The demonstration came at a staggering computational price. The simulation contained nearly one trillion three-dimensional grid points, each representing a data location across the globe and through the depth of the atmosphere. To simulate a single eight-hour period, specifically August 5, 2016, the team had to occupy more than half of the supercomputer Fugaku simultaneously, one of the most powerful machines in the world. The energy consumed was equivalent to roughly forty years of electricity use by an average Japanese household. The chosen date was not arbitrary: it allowed the researchers to validate their simulation against a real tropical weather system, Typhoon Omais, and to compare how the storm’s precipitation rate and location appeared at kilometer-scale versus 200-meter grid spacing, with the finer version showing markedly more realistic structure.

Despite the success, the researchers are careful about what the result does and does not imply for forecasting. At present, a global simulation at 220-meter resolution is far too computationally expensive to replace operational weather prediction systems, which must deliver forecasts within hours on fixed computing budgets. For now, Matsugishi noted, these simulations are better suited to research experiments. One promising application is the detailed study of tropical convection and heavy rainfall, processes that remain among the largest sources of uncertainty in climate science. Another is the use of the high-resolution output as a reference benchmark: because the GLES resolves processes that coarser models must parameterize, its results can serve as ground truth against which the parameterization schemes in kilometer-scale and hundred-kilometer-scale models can be evaluated and improved.

Indeed, the new model is not bias-free. While the higher resolution eliminated the popcornlike rain artifact, other biases related to the distribution of cloud cover persisted, a reminder that resolving more of the atmosphere does not automatically resolve every problem. Cloud microphysics, the microscopic processes by which water vapor condenses into droplets and ice crystals, still must be represented through simplified schemes even in a large-eddy simulation, and turbulence near the smallest scales remains only partially resolved. The team intends to use the model to investigate the characteristics and properties of convective clouds in much greater detail, with the goal of understanding how turbulence, cloud microphysics, and other unresolved processes should be represented as global models march from kilometer scale to several-hundred-meter and eventually tens-of-meters resolution.

The timing of the work is significant. This year has brought record-breaking weather across multiple continents, with intense storms battering Europe, exceptional August heat, and devastating flooding in parts of Asia and Africa. Climate change is expected to make strong and unpredictable storms more common, raising the stakes for precise climate modeling, weather forecasting, and early warning systems. Better representation of convection is central to that effort, because convective storms deliver extreme rainfall and are poorly captured by coarse models. Ultimately, the researchers argue, the kind of explicit simulation demonstrated here could better represent extreme weather and reduce uncertainties in future weather and climate predictions, turning what is today a research experiment running on one of the world’s largest supercomputers into the foundation of tomorrow’s forecasting systems.

The study, titled Resolution Dependence in a Global Atmospheric Simulation from km to 220 m Grid Spacing, was supported by JSPS KAKENHI grants and by the Core-to-Core Program, with simulations performed on the Fugaku supercomputer under a series of project proposals. It stands as a proof of concept for a new tier in the hierarchy of atmospheric models, one in which the planet’s storms are no longer approximated but resolved, and in which the errors born of approximation, like the phantom popcorn rain that has plagued global simulations for years, can finally be left behind.

Subject of Research: A 220-meter-resolution global large-eddy atmospheric simulation that eliminates the popcornlike rain bias in global storm-resolving models

Article Title: New 220-meter global atmospheric model erases “popcornlike” rain

Article References: New 220-meter global atmospheric model erases “popcornlike” rain. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: global large-eddy simulation, atmospheric modeling, convection, cumulonimbus clouds, popcornlike rain bias, grid resolution, supercomputer Fugaku, Typhoon Omais, climate prediction, weather forecasting, cloud microphysics, Geophysical Research Letters

The team’s central finding is quantitative and unambiguous. Before the early 1980s, only 51.8 percent of mature eastern-pole cooling events went on to develop into a full, two-poled dipole pattern. After the early 1980s, that success rate jumped to 82.1 percent during the period from 1981 to 2023. In other words, over the past four decades, the eastern and western halves of the Indian Ocean have become far more tightly coupled, so that a cooling anomaly near Sumatra is now much more likely to be mirrored by warming near Africa. This is not a trivial bookkeeping detail. The spatial shape of the IOD determines where its climatic fingerprints land, from East African long rains to East Asian summer monsoon rainfall, so knowing whether a dipole will fully form is essential for seasonal prediction.

To understand the mechanism behind this shift, the researchers turned to a mixed-layer heat budget analysis, a diagnostic technique that tracks every major term contributing to temperature change in the upper ocean. Their analysis singled out one player as decisive: the strength of zonal wind anomalies along the equatorial Indian Ocean. When easterly wind anomalies blow from east to west along the equator, they do two things at once. They drive the local Bjerknes feedback that amplifies the dipole by reinforcing the east-west temperature contrast, and, crucially, they trigger downwelling oceanic Rossby waves that propagate westward toward the thermocline ridge region of the western Indian Ocean. These waves deepen the local thermocline, trap warm water in the subsurface, and set the stage for the warm western pole that completes the dipole.

The logic of the mechanism is elegant. Eastern-pole cooling, often initiated by coastal upwelling off Sumatra driven by southeasterly winds, is only the first act of the drama. The second act depends on whether the equatorial easterlies are strong enough to launch the Rossby waves that warm the west. In the decades before the early 1980s, those winds were frequently too weak, so the eastern cooling fizzled into a monopole rather than a dipole. Since the early 1980s, stronger and more persistent easterly anomalies have made the full two-poled pattern the norm rather than the exception. The dipole, in this view, is not a self-contained Indian Ocean phenomenon but the end product of a chain of interactions that spans the entire tropical belt.

So what strengthened the winds? The study points to two remote drivers operating through the tropical atmosphere. The first is the El Niño-Southern Oscillation, the dominant mode of Pacific climate variability. During El Niño events, shifts in deep convection over the Pacific excite atmospheric circulation anomalies that extend across the Maritime Continent into the Indian Ocean basin, a pathway long described in the literature as the atmospheric bridge. After the early 1980s, the linkage between ENSO and the IOD strengthened, meaning that El Niño events more effectively excite the easterly wind anomalies over the equatorial Indian Ocean that are needed to warm the western subsurface. A stronger ENSO handshake, in effect, has been pulling the Indian Ocean’s two poles into tighter synchronization.

The second driver comes from an ocean basin that most people would not suspect: the tropical Atlantic. The researchers found that sea surface temperature anomalies in the tropical North Atlantic influence the winds over the equatorial Indian Ocean through Gill-Matsuno-type atmospheric responses, in which heating or cooling over the Atlantic generates Rossby and Kelvin wave patterns in the atmosphere that reach across Africa and the Indian Ocean. Before the early 1980s, cold sea surface temperature anomalies in the tropical North Atlantic worked in the opposite direction to El Niño, suppressing the easterly wind anomalies over the equatorial Indian Ocean. This suppression inhibited the subsurface warming in the western Indian Ocean and impeded dipole development. After the early 1980s, with stronger ENSO forcing no longer being undercut by Atlantic cooling, the easterlies could do their work unimpeded, and full dipoles became far more common.

The evidence base for these conclusions is unusually broad. The team analyzed multiple independent sea surface temperature datasets, including the NOAA Extended Reconstructed Sea Surface Temperature version 5 and the Met Office Hadley Centre’s HadISST, together with atmospheric reanalyses from the European Centre for Medium-Range Weather Forecasts, including ERA-20C and ERA5, and ocean reanalysis products such as SODA and GODAS. Triangulating across datasets matters enormously in this field, because the Indian Ocean’s observational record before the satellite era is sparse, and spurious trends in any single product can masquerade as interdecadal change. The authors also conducted model simulations using the GFDL-CM4 coupled climate model to test the causal chain, complementing the observational diagnostics with controlled numerical experiments.

The findings add an important nuance to a lively scientific debate. Previous studies have documented changes in the ENSO-IOD relationship over recent decades, with some analyses suggesting a weakening of the coupling and others pointing to enhanced ENSO influence on the dipole under greenhouse warming. This study reframes the question: what matters for the dipole’s spatial completeness is not simply whether ENSO and the IOD co-occur, but whether the cross-basin forcing is strong enough to activate the oceanic pathway, the Rossby waves and thermocline deepening, that builds the western pole. It also echoes and extends earlier work showing that the influence of north tropical Atlantic sea surface temperatures on the IOD strengthened after the mid-1980s, placing that result within a unified cross-basin framework that includes the Pacific as well.

There are practical consequences for climate prediction. Seasonal forecast systems and the coupled models used for climate projection often struggle to reproduce the IOD’s amplitude and spatial pattern, and a recent companion study by overlapping authors traced some of those biases to the models’ representation of dipole dynamics. If the probability that an eastern-pole event becomes a full dipole depends on interdecadal states of the Pacific and Atlantic, then forecast systems need to capture those cross-basin links to predict the dipole’s shape, not just its existence. For the roughly two billion people living around the Indian Ocean rim, whose agriculture, fisheries and flood defenses respond to IOD-driven rainfall anomalies, the difference between a monopole and a full dipole can be the difference between a manageable season and a disaster.

The study also carries a cautionary message for the future. As greenhouse warming continues to alter the mean states of all three tropical oceans, the cross-basin interactions identified here will not remain static. Research has already suggested that the frequency of extreme positive IOD events may increase under warming, while other work indicates that long-term projections of dipole variability remain deeply uncertain. What this new analysis makes clear is that the Indian Ocean Dipole cannot be understood, modeled, or predicted in isolation. It is a node in a planetary network of tropical ocean-atmosphere interactions, and its behavior over the past four decades has been quietly rewritten by forces originating thousands of kilometers away in the Pacific and Atlantic. Recognizing that connectivity, the authors argue, is a key step toward improving the predictability of the IOD’s spatial pattern and, by extension, the climate of the regions that depend on it.

Subject of Research: Interdecadal strengthening of the coupling between the eastern and western poles of the Indian Ocean Dipole through tropical cross-basin interactions with ENSO and the tropical Atlantic

Article Title: Strengthened coupling between the eastern and western poles of the Indian Ocean Dipole since the early 1980s: roles of tropical cross-basin interactions

Article References: Ju, X., Chen, S., Chen, L., Cao, X., Wang, Z., Wu, R., & Chen, W. (2026). Strengthened coupling between the eastern and western poles of the Indian Ocean Dipole since the early 1980s: roles of tropical cross-basin interactions. Climate Dynamics, 64(11), Article 452. https://doi.org/10.1007/s00382-026-08412-9

Image Credits: AI Generated

DOI: 10.1007/s00382-026-08412-9

Keywords: Indian Ocean Dipole, ENSO, tropical Atlantic, cross-basin interactions, sea surface temperature, interdecadal variability, oceanic Rossby waves, zonal wind anomalies, climate prediction, Climate Dynamics, thermocline, atmospheric bridge

In a study published in SCIENCE CHINA Earth Sciences, the team introduced a method called the SFA-based Climate Prediction approach, or SCP, which is built on a mathematical technique known as slow feature analysis. Rather than relying on a hand-picked list of climate predictors such as sea surface temperatures or atmospheric oscillation indices, the method extracts slowly varying signals directly from the observational precipitation record itself. The idea is that within the noisy, day-to-day fluctuations of rainfall data, there exist underlying components that change gradually over time—and it is precisely these slow components that carry the predictability needed for seasonal forecasting.

Slow feature analysis is an unsupervised learning technique originally developed in computational neuroscience to describe how biological vision systems extract stable information from rapidly changing sensory input. When applied to climate data, the algorithm searches for the slowest-varying signals embedded in a high-dimensional dataset, effectively separating the sluggish, large-scale drivers of precipitation variability from the fast, essentially unpredictable weather noise. By identifying these slow features and extrapolating their evolution forward in time, the SCP method constructs forecasts of precipitation anomalies without requiring forecasters to nominate specific physical mechanisms in advance.

This is a meaningful departure from conventional statistical seasonal prediction. Traditional approaches typically depend on a limited set of preselected predictors—indices of the El Niño–Southern Oscillation, the Indian Ocean Dipole, or snow cover anomalies, for example—chosen on the basis of physical reasoning or statistical screening. The problem is that the influence of any single predictor can shift from year to year and from decade to decade, a phenomenon climate scientists call predictor stationarity. When a previously reliable teleconnection weakens or reverses, a statistical model built around it can fail abruptly. By letting the data itself reveal which slow signals matter, the SCP framework sidesteps some of this fragility.

To test the method, the researchers applied it to summer precipitation over South China, a densely populated region where seasonal rainfall anomalies carry serious consequences for agriculture, flooding, and water resource management. South China’s summer climate is shaped by the East Asian monsoon, whose interannual variability is notoriously difficult to capture. Dynamical coupled models, which simulate the full physics of the ocean–atmosphere system, have historically shown limited skill for this region, and hybrid dynamical–statistical systems that blend model output with statistical correction have improved results only modestly.

The evaluation was designed to be rigorous. The team performed rolling independent forecasts for the twelve summers from 2011 through 2022, with a lead time of roughly three months—meaning the forecasts were issued before the summer season began, using no information from the period being predicted. This independent validation is the gold standard for assessing whether a prediction method has genuine skill or has merely been tuned to fit historical data. The results showed that the SCP method achieved a regional mean temporal correlation coefficient of 0.36 between predicted and observed precipitation anomalies across the validation period.

That number becomes more meaningful in comparison. The hybrid dynamical–statistical prediction system FODAS, one of the operational approaches used in China, achieved a regional mean correlation of 0.19 over the same period. The dynamical model BCC-CSM1.1, developed at the Beijing Climate Center, reached 0.13, and the American CFSv2 model managed 0.08. In other words, the purely data-driven SCP method roughly doubled the skill of the best competing system and outperformed the dynamical models by a wide margin. When the analysis focused specifically on the regional mean precipitation anomaly—the single most policy-relevant quantity for the region—the gap widened further: SCP achieved a correlation of 0.62, compared with 0.45 for FODAS, 0.28 for BCC-CSM1.1, and 0.17 for CFSv2.

Correlation coefficients capture the fine texture of forecast performance, but for decision-makers a simpler question often matters most: will the coming season be wetter or drier than normal? On this binary test, SCP also delivered. The method correctly predicted the sign of the regional-mean summer precipitation anomaly—above or below the climatological normal—in ten of the twelve independent validation summers. A comprehensive evaluation using multiple verification metrics confirmed that SCP consistently outperformed all three competing prediction systems across the test period, suggesting the advantage is not an artifact of any single scoring choice.

The implications extend beyond South China. Because the SCP framework does not prescribe specific climate factors, it is in principle portable: the same slow feature analysis could be trained on observational precipitation data from other monsoon regions, mid-latitude domains, or even other climate variables where slow-varying predictable components exist. The authors suggest the approach could provide technical support for flood and drought risk management and for climate-related decision-making, areas where even modest gains in seasonal forecast skill translate into substantial practical value. A three-month lead time, in particular, is long enough for farmers to adjust planting decisions and for water authorities to prepare reservoir operations.

At the same time, the study is a reminder of how much room for improvement remains. A temporal correlation of 0.36, while the best among the systems tested, still leaves the majority of precipitation variance unexplained, and twelve independent validation summers represent a limited sample in a climate system influenced by slowly evolving decadal modes. Whether the SCP method’s advantage holds over longer periods, under different climate backgrounds such as strong El Niño or La Niña years, and across other regions will be the critical test of its broader utility. The work, published as Wang, Guo, and He, ‘A new method for predicting summer precipitation over South China based on slow feature analysis,’ in Science China Earth Sciences, nonetheless offers what the field has long sought: a new perspective on identifying potentially predictable signals, and evidence that the observational data record itself still holds untapped predictive information waiting to be extracted.

Subject of Research: Seasonal prediction of summer precipitation over South China using slow feature analysis of observational data

Article Title: Extracting predictable signals from observational precipitation data: A new approach to seasonal precipitation prediction

Article References: Extracting predictable signals from observational precipitation data: A new approach to seasonal precipitation prediction. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: seasonal prediction, precipitation, slow feature analysis, South China, East Asian monsoon, climate prediction, statistical forecasting, dynamical models, FODAS, BCC-CSM1.1, CFSv2, Science China Earth Sciences

The paper grew out of The Fifth Forum on Artificial Intelligence Oceanography, held earlier this year in Jinan, in China’s Shandong Province. The forum convened more than 300 experts, scholars, and students drawn from more than 80 institutions, making it one of the most comprehensive gatherings to date at the intersection of marine science and machine learning. Across talks and panel discussions, a clear consensus crystallized among participants: the next stage of AI oceanography will increasingly depend on models that fuse the flexibility of learning algorithms with the governing principles of ocean physics. The authors of the paper distilled those discussions into a forward-looking assessment of where the field stands and where it must go.

At the heart of that assessment lies the concept of physics-constrained artificial intelligence. In conventional machine learning, algorithms sift through enormous volumes of historical observations and model simulations, extracting statistical patterns without any explicit understanding of the dynamical laws that produce them. That approach can yield impressive results when training data are abundant and conditions resemble the past. But the ocean is an unforgiving test bed. Observations beneath the surface are sparse and expensive to collect, and environmental conditions can shift in ways that leave a purely statistical model extrapolating blindly. Physics-constrained approaches counter this weakness by embedding physical knowledge, dynamical equations, and process-based relationships directly into AI learning and evaluation, so that every prediction the system makes must remain compatible with the known behavior of rotating, stratified, wind-driven fluids.

The practical payoff, the authors argue, is threefold: improved reliability, improved interpretability, and improved performance under precisely the conditions where conventional AI falters. When observations are scarce, physical constraints act as a form of regularization, narrowing the space of plausible solutions to those the real ocean could actually occupy. When environmental conditions diverge from the training distribution, embedded dynamics give the model a scaffold to lean on, reducing the risk of physically nonsensical outputs. And when scientists interrogate a prediction, a model organized around meaningful physical relationships is far easier to diagnose and trust than an opaque network whose internal representations may correspond to nothing in the real ocean.

Drawing on the forum’s presentations, the authors organized the field’s most promising applications into three complementary classes of problems. The first is reconstructing subsurface ocean states from surface information. Satellites can observe sea surface temperature, sea surface height, and surface salinity with global coverage, but the three-dimensional structure of the ocean beneath remains largely hidden. AI systems trained on historical observations and simulations can infer subsurface temperature and salinity fields from what is visible at the surface, and studies presented at the forum showed that incorporating physical knowledge into these reconstructions makes them more reliable and more useful for research and operations alike.

The second class is long-range climate prediction, exemplified by forecasting the El Niño Southern Oscillation. ENSO emerges from coupled interactions between the tropical Pacific Ocean and the atmosphere, and its swings reshape rainfall, temperature, and storm patterns across the globe. Machine learning models have recently matched or exceeded traditional dynamical forecast systems in some ENSO prediction tasks, but the forum’s participants emphasized that learning physically meaningful relationships among ocean and atmosphere processes is what elevates these forecasts from statistical tricks to genuine scientific tools. When an AI system captures the underlying coupled dynamics, its predictions become interpretable, its errors diagnosable, and its skill more likely to persist as the climate changes.

The third class is operational wave forecasting, a domain with immediate consequences for shipping, offshore engineering, coastal management, and marine hazard warning. Waves respond rapidly to winds and currents, and forecasting them demands models that are both fast and physically sound. AI-based wave prediction systems presented at the forum demonstrated that embedding wave dynamics and process relationships into learning frameworks can deliver the speed of machine learning without sacrificing the physical consistency that operational users require. In each of the three problem classes, the pattern was the same: physical knowledge was not a constraint on AI’s power but the source of its trustworthiness.

The authors are explicit about what comes next. In their view, the field must move beyond isolated demonstrations toward AI systems that systematically combine physical constraints, uncertainty estimation, and real ocean observations. Uncertainty estimation deserves particular emphasis, because a forecast that comes with an honest measure of its own confidence is vastly more valuable to decision-makers than one that does not. A shipping company rerouting around a storm, a coastal community preparing for a marine hazard, and a climate negotiator weighing long-term risks all need to know not just what a model predicts but how much that prediction can be trusted. Building calibrated uncertainty into physics-constrained AI is therefore a central challenge for the coming years.

The longer-term ambition articulated in the paper is equally clear. The authors state that the ultimate goal is to make AI a trustworthy complement to numerical ocean models, one capable of supporting faster and more reliable ocean forecasting, climate prediction, marine hazard warning, and scientific understanding. This framing matters. Rather than positioning machine learning as a replacement for the general circulation models and dynamical simulation systems that oceanographers have refined over decades, the authors envision a hybrid landscape in which AI accelerates computation, fills observational gaps, and extracts structure from data, while numerical models and physical theory supply the dynamical foundation that keeps learning algorithms anchored to reality.

The implications extend well beyond oceanography. The tension the paper addresses, between the pattern-finding power of data-driven models and the explanatory rigor of physics-based ones, is one of the defining questions of modern Earth system science. The ocean, with its sparse observations, slow memory, and outsized influence on climate, is arguably the harshest environment in which to resolve that tension. If physics-constrained AI proves itself in the deep blue, the lessons will carry directly into atmospheric science, hydrology, cryosphere research, and the broader effort to model a planet in flux. For now, the message from more than 300 researchers gathered in Jinan is unambiguous: the future of AI in ocean science belongs not to algorithms that merely memorize the sea, but to those that learn the motion of the ocean itself.

Subject of Research: Physics-constrained artificial intelligence for ocean forecasting and climate prediction

Article Title: AI needs to learn the motion of the ocean, researchers report

Article References: AI needs to learn the motion of the ocean, researchers report. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: artificial intelligence, oceanography, physics-constrained machine learning, El Nino Southern Oscillation, wave forecasting, ocean modeling, climate prediction, subsurface ocean reconstruction, uncertainty estimation, numerical ocean models, marine hazards, AI oceanography forum