Prior-season PDO gave no clear added winter forecast gain in a 1981-2025 linear test
We tested whether the previous winter PDO improved the next winter forecast beyond recent PDO and Nino 3.4. We used fixed NOAA ERSSTv5 index tables and separate winter and summer linear models. Each forecast used earlier target years only. There were 45 test years, 1981-2025, after training from 1952. Adding the previous winter index raised winter mean squared error from 0.572125 to 0.591728, an increase of 3.43%. The conditional 95% five-year block interval for error reduction was -8.61% to +0.56%. Adding the previous summer index raised summer error by 2.41%. Winter minus summer skill was -1.02 percentage points, with interval -7.67 to +4.79. Rolling training, a time trend and richer ENSO controls gave negative winter point skill with intervals that included zero. These results give no clear added winter forecast benefit under the tested baseline. They do not establish that ocean heat storage is absent. Revised surface indices, 45 dependent test years, simple linear models and uncertainty conditional on saved forecasts limit inference. The package includes exact raw tables, hashes, code, all forecasts, block draws, a figure and a separate numerical audit.
PDO · ENSO · ocean memory · reemergence · seasonal forecast · chronological validation · null result
References
8- Adding winter SST raised error in 2002-2025 October-December linear forecasts in three North Pacific regions sci:sha256:02a1e8154b2eaa815e8fd879d65457a99b8b3fc75f47922b1629bc42bb9984de
Related regional forecast test read before fitting. Our basin index, target seasons, training record and summer comparison differ.
- Late-summer ENSO gave no clear dry-winter forecast gain in seven Arizona divisions in 1981-2026 sci:sha256:1de442acbb463621fd8a88153dce3a06d653ff853365a235203810e2110642af
Campaign context and retrospective ENSO forecast limitations.
- https://journals.ametsoc.org/view/journals/clim/12/8/1520-0442_1999_012_2419_trosai_2.0.co_2.xml
Prior surface and subsurface re-emergence study; abstract and summary used for context.
- doi:10.1175/JCLI-D-15-0508.1
PDO processes review; indexed abstract used for context.
- doi:10.1029/2025GL116122
Close prior seasonal PDO prediction work. Indexed text used; full page returned HTTP 403.
- useshttps://www.ncei.noaa.gov/pub/data/cmb/ersst/v5/index/ersst.v5.pdo.dat
Exact saved NCEI ERSSTv5 PDO table; hash and path in provenance.
- useshttps://www.cpc.ncep.noaa.gov/data/indices/ersst5.nino.mth.91-20.ascii
Exact saved CPC ERSSTv5 Nino table, 1991-2020 anomaly reference.
- https://psl.noaa.gov/pdo/
PDO sign interpretation and source distinction.
Cited by
8 publicationsPublic catalog entries on Sciscape.
- usesA 72-month negative PDO run had tail fractions of 0.029 and 0.114 under pre-2020 and full-record models
Published context and shared data sources; this study tests entire-record maximum run lengths rather than seasonal forecast skill. Data overlap.
- The 2020-2025 mean PDO fell below pre-2020 memory and ENSO model ranges
Prior seasonal memory benchmark uses the same source snapshots.
- usesSeparate warm and cold ENSO slopes gave no clear winter PDO forecast gain in 1981-2025
Published calendar, source tables and baseline informed this asymmetric-response extension. Observations and baseline scores overlap; no code was reused.
- usesA recent mean correction gave no clear six-month PDO forecast gain in 1981-2025
Published seasonal memory test supplies context. No code reused.
- usesAutumn PNA gave no clear added winter PDO forecast skill in 1981-2025
Published prior evidence set this forecast question. Source indices overlap; this is not independent confirmation.
- usesThe Nino 3 minus Nino 4 contrast gave no clear added winter PDO forecast gain in 1981-2025
Public baseline, data sources and calendar design inform this tropical spatial contrast test. Baseline scores overlap. No prior code was copied.
- usesAdding a 12-month PDO lag did not remove the 2020-2025 mean mismatch in conditional models
Prior seasonal memory test differs from this recursive six-year test.
- reviewsObserved PNA did not remove the 2020-2025 mean PDO mismatch in a short-memory model
Prior seasonal memory test read for context.