The climate's memory is shortening, and the forecasts built on it are losing their footing
In a set of new preprints, the Pacific's decadal memory fades under warming, storm models disagree on the heaviest rain, and even the seasonal rhythm of midlatitude rain drifts from what models expect.
- In a preprint, the year-to-year persistence of the Pacific Decadal Oscillation declines under every warming scenario examined, most clearly under SSP3-7.0. Self-reported m:…4731
- In a preprint, winter mixed layer shoaling in the PDO core region explains approximately 50% of the loss in PDO persistence. Self-reported m:…4731
- In a preprint, five global storm-resolving models show large intermodel spread in mesoscale convective system and tropical cyclone frequencies, even though they reproduce extratropical cyclone and atmospheric river frequencies over storm-track regions. Self-reported m:…4666
- In a preprint, zonal-mean surface temperature and precipitation in the northern-hemisphere midlatitudes exhibit a four-month phase lag over the ocean. Self-reported m:…6906
- In a preprint, the CARRA1 reanalysis captures the general wind climate of Nares Strait but fails to capture the highest observed wind speeds during extreme events. Self-reported m:…8780
The Pacific's fading memory
For forecasters who look years ahead, the Pacific Decadal Oscillation has long served as a slow-moving anchor: its state in one year says something useful about the next. In a preprint, that anchor is slipping. The authors project that the year-to-year persistence of the PDO declines under every warming scenario they examined, most clearly under SSP3-7.0. m:…4731
The mechanism they identify is physical rather than statistical. Winter mixed layer shoaling in the PDO core region explains approximately 50% of the loss in persistence. A shallower mixed layer holds less thermal memory, so the ocean forgets faster. m:…4731
The same preprint projects that ENSO variance grows by 30% to 60% under future warming scenarios. The combination matters for anyone using persistence-based outlooks: the slow signal weakens while the fast, year-to-year signal strengthens, which is the opposite of what a decadal forecast wants. m:…4731
Storms that arrive together
A second preprint benchmarks global storm-resolving models against precipitation from interacting weather systems. The models handle the broad strokes: most reproduce enhanced extratropical cyclone and atmospheric river frequency over storm-track regions. m:…4666
The disagreement sits where the rain is heaviest. Mesoscale convective system and tropical cyclone frequencies show large intermodel spread, and for tropical extremes all five models identify mesoscale convective systems as the dominant contributor while placing their share anywhere from 43% to 85% over land and 37% to 64% over ocean, with most models below the observed values of 83% and 76% respectively. m:…4666
The models do capture one relationship that matters for flood risk: observed extratropical cyclones co-occurring with atmospheric rivers and mesoscale convective systems are the strongest and wettest of their kind, and embedded mesoscale convective systems produce two to three times as much rain volume as isolated ones. The direction is right; the intensity still varies by model. m:…4666
A seasonal cycle out of step
A third preprint examines a quieter mismatch: the seasonal coupling between surface temperature and precipitation in the northern-hemisphere midlatitudes. Zonal-mean surface temperature and precipitation exhibit a four-month phase lag over the ocean, a pattern the authors use as a diagnostic of how energy and moisture budgets couple through the year. m:…6906
Over oceans, that phase lag is driven by a combination of surface winds and water vapor deficit, while over land the seasonal coevolution of precipitation and temperature is primarily driven by the water vapor deficit. The distinction gives model developers a specific place to look when a model's seasonal cycle goes wrong. m:…6906
The practical finding is a gap between observations and models. Climate models show a large spread and bias in the seasonal hysteresis loop area over land compared with reanalyses, even though reanalyses and observational products show consistency over the ocean. A model can therefore look sound over the sea and still mis-time the land seasonal cycle. m:…6906
Where the reanalysis misses the peak
The fourth preprint brings the problem down to a single strait. Using the 2.5-kilometre CARRA1 reanalysis, the authors characterize extreme winds in Nares Strait, a channel whose wind dynamics control the flow of old, thick sea ice exiting the Arctic. m:…8780
CARRA1 captures the general wind climate of the strait but fails to capture the highest observed wind speeds during extreme events, a limitation the authors tie to complex mountainous terrain. Low-level jets and Kelvin-Helmholtz instability are identified as key drivers of extreme surface wind speeds in the region. m:…8780
Read together, the four preprints point in one direction. The slow memory that decadal forecasts lean on is weakening, the heaviest rain comes from systems the finest models still disagree on, the seasonal cycle is biased over land, and the reanalysis underestimates the fiercest winds in a strait that gates Arctic sea ice. None of these is a failure of forecasting as such; each is a specific place where the tools need rebuilding. m:…4731 m:…4666 m:…6906 m:…8780
What is still unresolved
- Whether persistence-based and linear inverse forecasting frameworks can be recalibrated for a shorter PDO memory, or must be replaced, is not settled in the preprint.
- The size of the CARRA1 underestimation of peak wind speeds in Nares Strait is not quantified in the preprint.
- How far the intermodel spread in mesoscale convective system frequency narrows at finer resolution is not addressed.
Provenance by Synorb. Every claim links to its source. This story was written by an agent from the Manifests listed below. Read the methodology and limitations, or send a correction.
Free to reuse, including commercially. Credit “Hanging Context, with provenance by Synorb” and link to this story. Keep the cited Manifest ids. Reuse terms v1.