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 study evaluated NOAA’s Seamless System for Prediction and Earth System Research, known as SPEAR, and the National Centers for Environmental Prediction’s Climate Forecast System version 2, or CFSv2, across the contiguous United States from 1991 to 2022. SPEAR is the newer of the two, developed at NOAA’s Geophysical Fluid Dynamics Laboratory with an atmospheric resolution of roughly 50 kilometers, and it runs a 15-member ensemble initialized from an ensemble coupled data assimilation system that blends in subsurface ocean temperature and salinity observations. CFSv2, operational since 2011, uses a coarser atmospheric grid of about 0.94 degrees and a 24-member lagged ensemble built from forecasts launched at four different times of day. Both systems were verified against the PRISM observational dataset, a high-resolution 4-kilometer gridded precipitation product built from quality-controlled weather stations, with sea surface temperatures checked against NOAA’s OISST and circulation fields against the ERA5 reanalysis.

The researchers deployed a battery of verification metrics designed to probe different aspects of forecast quality. Percentage bias quantified how much the models systematically over- or under-predicted rainfall, while the anomaly correlation coefficient measured whether forecasts tracked the year-to-year ups and downs of real precipitation. For the first time in this context, they also applied quantile-based categorical metrics, the critical success index, probability of detection, and false alarm ratio, to assess how well the models captured wet extremes of seasonal-mean precipitation, defined as seasons exceeding the 80th and 90th percentile thresholds of the climatological distribution. Statistical significance of model differences was tested with bootstrap resampling over 10,000 iterations, ensuring that the reported advantages of one model over the other were not artifacts of a small sample of hindcast years.

The headline finding is a stark seasonal asymmetry. During spring, both models exhibit modest but regionally coherent skill, with positive anomaly correlations across the West Coast, the north-central states, and parts of the Southwest. SPEAR’s CONUS-average spring correlations ranged from about 0.25 to 0.30, compared with roughly 0.18 to 0.28 for CFSv2, and the bootstrap test showed that SPEAR’s advantage was statistically significant only at the shortest one-month lead time. By summer, skill collapsed in both systems. SPEAR’s correlations fell to between 0.2 and 0.3 while CFSv2 hovered around 0.15, and over the South-Central United States, a case-study region encompassing Kansas, Oklahoma, and Texas, summer correlations dropped to nearly zero, exposing a profound inability of either system to anticipate summer rainfall anomalies in a region where drought and deluge both carry enormous economic stakes.

The bias analysis added further texture to the picture. In spring, both models were too wet over much of the central and eastern United States, but the problem was far worse in CFSv2, whose area-averaged gridded spring bias over the full continent reached 62 to 78 percent, compared with 31 to 33 percent for SPEAR. Over the South-Central region specifically, CFSv2’s spring wet bias climbed as high as 73 percent while SPEAR remained nearly neutral. Summer flipped the sign: SPEAR showed a robust dry bias over the continent and a severe 38 to 43 percent underestimation over the South-Central region, while CFSv2’s local wet and dry errors partially cancelled in the regional mean. The authors emphasize that how bias is summarized matters enormously, because compensating local errors can hide behind a seemingly benign regional average.

Perhaps the most practically troubling results concern wet extremes. When the researchers asked whether the models could identify seasons whose precipitation landed in the top 20 or top 10 percent of the historical distribution, the answer was largely no, especially in summer. Probability of detection declined sharply from the 80th to the 90th percentile threshold, while false alarm ratios climbed toward very high values across much of the country during June through August. Crucially, the spatial patterns of detection and false alarms mirrored the bias maps: in wet-biased regions the models trigger too many threshold exceedances, inflating both hits and false alarms simultaneously, whereas dry-biased regions suppress detections altogether. High detection rates, in other words, do not guarantee reliable forecasts when they come bundled with a flood of false alarms.

To understand why the models struggle, the team turned to the physics of the El Niño-Southern Oscillation, the dominant source of seasonal predictability for North American rainfall. An empirical orthogonal function analysis confirmed that the leading mode of tropical Pacific sea surface temperature variability during the study period was the canonical eastern-Pacific El Niño pattern, explaining 31.1 percent of SST variance and correlating at 0.95 with the standard Niño-3.4 index. Observations showed that the link between this tropical signal and South-Central U.S. rainfall is strongly seasonal: robust during winter and early spring, weakening through late spring and summer as regional convection, land-atmosphere feedbacks, and the Great Plains Low-Level Jet, which funnels Gulf of Mexico moisture into the continent’s interior, take center stage. Both models broadly reproduced this seasonal ebb and flow, though CFSv2 tended to overstate the SST-precipitation correlation in many seasons.

The centerpiece of the diagnostic work was a singular value decomposition analysis coupling tropical Pacific SST to CONUS precipitation. In spring, the observed leading mode explained 34.2 percent of the squared covariance and traced a coherent pathway: an El Niño-like SST pattern driving a Pacific-North American ridge-trough couplet in the mid-troposphere, with an anomalous ridge over the northeastern Pacific and a trough over the southeastern United States that steers Gulf moisture northward. Both models captured the qualitative structure of this spring teleconnection, but their 500-hectopascal geopotential height anomalies were only 40 to 60 percent of the observed amplitude. Strikingly, the ensemble-mean versions of the models showed squared covariance fractions of 70.5 percent for SPEAR and 87.7 percent for CFSv2, far above observations, yet when the decomposition was repeated for individual ensemble members the values fell back near the observed level. The inflation, the authors show, is a statistical artifact of ensemble averaging, which filters out chaotic internal atmospheric noise and isolates the forced SST signal rather than evidence that the models are genuinely over-dominated by El Niño.

In summer, the teleconnection machinery visibly breaks down in the models. The observed leading coupled mode shifts away from the canonical equatorial El Niño pattern toward western and off-equatorial Pacific anomalies, and the associated circulation features a deep anomalous trough over the central and eastern United States paired with rising motion that feeds continental rainfall. SPEAR retained some spatial fidelity, with a pattern correlation of 0.24 against the observed height regression, but CFSv2’s pattern correlation collapsed to 0.01, indicating a fundamental displacement of the summer circulation response. Vertical-velocity cross sections told the same story: the models produced weaker, smoother, and less vertically coherent ascent and descent anomalies than observations, consistent with known deficiencies in convective parameterizations and the simulation of Rossby wave propagation. An El Niño-like SST signal alone, the study concludes, is simply not enough; the atmospheric bridge that translates ocean warmth into rain must also be simulated faithfully.

The implications reach well beyond model scorecards. Seasonal forecasts underpin drought preparedness, reservoir operations, and agricultural planning, and the South-Central United States, a transition zone between the humid Southeast and the arid Southwest, sits squarely where those decisions hurt most when forecasts fail. The authors argue that future gains will require more than better ocean prediction: improved representation of SST-forced teleconnections, regional atmospheric variability, and additional warm-season drivers such as land-surface memory, soil moisture feedbacks, Gulf moisture transport, and modes of variability beyond ENSO. Until then, the study offers forecast users a candid calibration of what to trust: spring rainfall anomalies over the southern plains deserve guarded confidence, while summer outlooks, particularly for unusually wet seasons, should be treated with the skepticism that three decades of hindcast evidence now firmly justify.

Subject of Research: Seasonal forecast skill of the SPEAR and CFSv2 climate models for warm-season precipitation over the contiguous United States and its relation to ENSO teleconnections

Article Title: Relative skills of SPEAR and CFSv2 in foretelling warm-season precipitation anomalies over the conterminous United States and potential sources of errors

Article References: Nguyen, H., Zhang, Y., Bakhtar, A., Lu, F., & Fernando, N. (2026). Relative skills of SPEAR and CFSv2 in foretelling warm-season precipitation anomalies over the conterminous United States and potential sources of errors. Climate Dynamics, 64(9), Article 406. https://doi.org/10.1007/s00382-026-08357-z

Image Credits: AI Generated

DOI: 10.1007/s00382-026-08357-z

Keywords: seasonal forecasting, SPEAR, CFSv2, precipitation, ENSO, teleconnections, climate models, drought, Great Plains, sea surface temperature, forecast verification, Climate Dynamics

A new study published in Climate Dynamics by Caihong Wen, Wanqiu Wang, Arun Kumar, and Michelle L’Heureux of NOAA’s Climate Prediction Center examines what happened next, and the results are a striking cautionary tale about how a change in input data can propagate through an entire forecast enterprise. The researchers found that the switch from OISSTv2 to NSST coincided with a sharp discontinuity in the sea surface temperatures recorded in CFSR. Before 2020, the reanalysis SST tracked the OISSTv2.1 reference dataset almost perfectly, with negligible differences. After the switch, a large cold bias appeared across much of the global ocean, as if the ocean’s skin had suddenly been cooled by an invisible hand.

The geography of the bias is telling. The largest discrepancies, exceeding minus one degree Celsius, emerged along the energetic western boundary currents such as the Gulf Stream and Kuroshio extensions, and along the Antarctic Circumpolar Current, regions where the ocean is dynamically active and where different SST products can diverge substantially. In the tropical upwelling zones, where winds drive cold water to the surface, the differences were more moderate but still significant, ranging from about minus 0.2 to minus 0.5 degrees Celsius. These are precisely the regions that matter most for the El Niño-Southern Oscillation, the planet’s dominant source of seasonal climate variability, which swings between warm El Niño and cold La Niña states and reshuffles weather patterns from California to Kenya.

What makes the finding particularly consequential is how these biases behaved once forecasts were launched from the reanalysis. The researchers examined two seasonal forecast systems that rely on CFSR for their ocean initial conditions: NCEP’s Climate Forecast System version 2, known as CFSv2, and the Community Climate System Model version 4, or CCSM4, which contributes to the North American Multi-Model Ensemble. Both models ingested the cold anomalies at initialization almost undamped. At zero-month lead time, meaning at the very start of the forecast, both systems exhibited cold anomalies in the eastern tropical Pacific. Rather than fading away as the coupled atmosphere-ocean system evolved, these anomalies intensified and spread westward toward the dateline within four months, contaminating the forecast trajectory itself.

The consequences for ENSO prediction were systematic and measurable. By constructing composites of forecast behavior around historical El Niño and La Niña events, the team showed that forecasts initialized after 2020 carried a persistent cold bias during boreal fall and winter, the seasons when ENSO events typically reach peak strength and when forecast skill is usually highest. In practical terms, the models were systematically underestimating ocean temperatures in the tropical Pacific during the most important window of the ENSO cycle, affecting the predicted amplitude of both El Niño warm events and La Niña cold events. A forecast that starts with a cold ocean and then cools it further will inevitably lean too far toward La Niña-like outcomes.

The 2024 La Niña episode provided a vivid real-world demonstration. Even during seasons when ENSO forecasts historically perform well, the post-2020 forecast systems produced overly cold La Niña predictions, exaggerating the intensity of the event. For forecasters, agricultural planners, disaster managers, and governments that depend on seasonal outlooks for drought, flood, and temperature guidance, such a bias is not an academic curiosity. It shifts probability distributions that inform billions of dollars in decisions, from crop planting to water reservoir management to famine early warning systems across the tropics.

There is also a deeper historical irony embedded in the findings. The CFSR has long been known to carry a warm bias in tropical Pacific SST forecasts, a discontinuity that emerged around 1999 and was documented in earlier studies of the system’s nonstationary errors. The 2020 switch in nudging datasets effectively reversed that sign, replacing a warm bias with a cold one. The result is a forecast system whose errors are nonstationary in time, meaning that bias corrections calibrated on one era of the reanalysis become invalid in another. Statistical post-processing, machine learning corrections, and forecaster intuition all implicitly assume that a model’s errors are reasonably stable; a sudden sign flip violates that assumption and degrades every downstream correction built on it.

The technical root of the problem lies in the subtle difference between what the two SST products actually measure. The OISST analysis targets a foundation SST, broadly representative of the temperature near the ocean surface free of diurnal variability, while the NSST scheme represents the temperature at the very top of the ocean model, including near-surface gradients that can differ from bulk or foundation temperatures. When a reanalysis nudges its top ocean layer toward a product defined differently from what the model layer represents, systematic offsets can be introduced. Because the nudging is strong, the ocean model has little freedom to correct those offsets through its own physics, and because the coupled forecast models are initialized from the reanalysis, the offsets ride along into every prediction. The study’s authors emphasize that although different SST analyses are routinely adopted in operational ocean reanalysis systems around the world, the impacts of switching between them have rarely been examined with this level of rigor.

The broader lesson extends well beyond NCEP. Ocean reanalysis systems operated by centers in Europe, Japan, and the United States all face similar choices about which SST product to assimilate, and operational systems inevitably evolve as datasets are updated or replaced. The Wen and colleagues analysis demonstrates that such transitions deserve the same scrutiny as changes to model physics or observation networks, because a change in the anchoring dataset is, in effect, a change to the climate the reanalysis describes. The researchers suggest that monitoring reanalysis products against independent references, quantifying discontinuities at transition points, and assessing how initialization biases propagate through coupled forecasts should become standard practice whenever operational data streams change. Otherwise, the quiet swap of one temperature dataset for another can silently recalibrate the world’s seasonal forecasts, turning a well-understood prediction system into one that is systematically, and invisibly, off balance.

Subject of Research: Impact of switching the sea surface temperature nudging dataset in the CFSR ocean reanalysis on seasonal ENSO forecast biases.

Article Title: Impact of SST nudging on CFSR analysis and seasonal ENSO forecasts

Article References: Wen, C., Wang, W., Kumar, A., & L’Heureux, M. (2026). Impact of SST nudging on CFSR analysis and seasonal ENSO forecasts. Climate Dynamics, 64(10), Article 418. https://doi.org/10.1007/s00382-026-08375-x

Image Credits: AI Generated

DOI: 10.1007/s00382-026-08375-x

Keywords: ENSO prediction, ocean reanalysis, CFSR, sea surface temperature, SST nudging, seasonal forecasting, CFSv2, CCSM4, NMME, La Niña, forecast bias, Climate Dynamics