The exact text we saved

This is the version exactly as it was saved, character for character. The SHA-256 hash below is a 64-character code worked out from that text. Run the same text through SHA-256 and you get the same code; change a single letter and the code comes out completely different. Each version also stores the previous version’s hash, which chains them together so no version can be swapped out quietly.

TRUVACE RECORD VERSION
record: TRV-2026-2623
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-11-12T08:05:14.830020Z
status: published
lens: g_space
sector: climate
headline: Atmospheric Predictability beyond 30 Days with Machine Learning
dek: Abstract Atmospheric predictability research has long held that rapid error growth at small spatial scales imposes an intrinsic limit of roughly 2 weeks on deterministic weather forecast skill. We challenge this limit using GraphCast, a machine learning weather model, by optimizing initial conditions for twice-daily forecasts spanning 2020. This approach yields an average error reduction of 86% at 10 days relative to control forecasts from reanalysis initial conditions, with skill lasting beyond 30 days. Mean op…
gain_title: In a study of 732 twice-daily hindcasts for 2020, optimizing starting conditions in the GraphCast AI weather model cut 10-day error by 86% against control forecasts and kept significant skill to 33 days, more than double the usual two-week limit.
problem_title: (none)
trace_subject: (none)
gain_reading: In a study of 732 twice-daily hindcasts for 2020, optimizing starting conditions in the GraphCast AI weather model cut 10-day error by 86% against control forecasts and kept significant skill to 33 days, more than double the usual two-week limit.
gain_evidence: This approach yields an average error reduction of 86% at 10 days relative to control forecasts from reanalysis initial conditions, with skill lasting beyond 30 days. | we achieve skillful deterministic hindcasts to at least 33 days—more than twice the conventional limit
problem_reading: (none)
problem_evidence: (none)
quick_read: Researchers at the University of Washington and the Air Force Institute of Technology used the GraphCast machine-learning weather model to tune the starting conditions of 732 forecasts, run at 00Z and 12Z for every day of 2020. Ten days out, error fell by 86% on average compared with control forecasts started from ERA5 reanalysis, and skill lasted beyond 30 days.

The result challenges the long-held view that rapid small-scale error growth limits deterministic forecasts to about two weeks. The authors read it as evidence that the atmosphere may be predictable at large scales for longer than assumed, with predictability perhaps governed mostly by large-scale dynamics.

The work is a hindcast study, not an operational forecast. The optimization needs knowledge of the future trajectory, so the authors say that finding such initial conditions in real time remains an open question and that no operational gains are yet shown.
limitation: The optimization uses knowledge of the verifying trajectory, so the authors say it does not yet identify such initial conditions in real time, which would be required before any operational forecasting gain. The 1.0° model also does not resolve mesoscale processes, so the findings apply only to synoptic and planetary scales.
tag: Evidence-backed gain
key_points: Every one of the 732 optimized forecasts improved at 10 days, with reductions ranging from 77% to 91% and no failures. | A 61-forecast subset (every sixth day of 2020) was optimized in double precision to 32 days; anomaly correlation for 500 hPa geopotential stayed statistically different from control to 33 days. | Practical skill, defined as an anomaly correlation of 0.6, lasted to 27.5 days. | Running the GraphCast-optimized conditions in the Pangu-Weather model gave a smaller 21% average error reduction at 4 days, versus 62% for GraphCast, suggesting the corrections partly fix model-specific error. | Average corrections were largest in the tropics and subtropics and broadly amounted to a stronger Hadley circulation. | Each optimization took about 4 hours on an NVIDIA A100 GPU, and the 32-day window was limited by its 80 GB of memory. | A check on 12 cases using the MERRA-2 reanalysis gave equal or slightly larger error reductions, suggesting the gains do not come from verifying against the data GraphCast was trained on.
rundown: The team used the small version of GraphCast, which runs on a 1.0° grid and has 36.7 million parameters. They used gradient descent through the model to adjust the two input atmospheric states, widening the optimization window gradually from 2 days in 3-day steps. Single-precision runs covered 14 days; double-precision runs reached 32 days.

Error growth in the optimized forecasts followed a doubling time of 5.8 days from day 2.5 to day 14. After the optimization window ended, error returned toward the control and merged with it near 30 days in single precision and around day 37 in double precision. The authors say this rate should not be read as the atmosphere's intrinsic error doubling time.

To rule out artificial skill from smoothing, the authors measured information error, noise error and forecast activity. Forecast activity stayed close to ERA5 for most variables, which they say shows the improvement comes from better agreement with the verifying field. They also discuss whether the method acts like an adversarial attack on the network and argue it does not, since it minimizes loss and yields physically interpretable large-scale patterns.

The authors say future work could test higher-resolution and coupled atmosphere–ocean models, add regularization, and try the optimized conditions in physics-based models.
sources:
- peer_reviewed | Artificial Intelligence for the Earth Systems | https://doi.org/10.1175/aies-d-26-0009.1 | 2026-06-03
prev: 0000000000000000000000000000000000000000000000000000000000000000

SHA-256 hash

503d476862041b84d3483d1fb0d0e4f4245f0d0446e44e14cadc08d864d951ae

Previous version’s hash

0000000000000000000000000000000000000000000000000000000000000000