Three-month Nino 3.4 forecasts had lower May skill but no clear May error increase over November in 1981-2025
We compared three-month linear Nino 3.4 forecasts for May and November targets in 1981-2025. Fits used earlier target years only in a fixed NOAA CPC ERSSTv5 monthly record. May and November mean squared errors were 0.2564 and 0.2529 C squared. Their paired difference was 0.0034 C squared, with a 95% five-year block interval of -0.1322 to 0.1293. Thus, a higher May absolute error is uncertain. Skill against the training mean was 36.31% and 82.07%. The difference was -45.76 percentage points, with an interval of -68.09 to -26.98. November training-mean error was 3.50 times May error, so close absolute errors give different relative scores. Rolling training and a year-trend check retained the relative skill contrast. A six-month lead did not. This is a retrospective score audit of one linear surface-index model. It does not discover the spring barrier, identify a physical cause, or measure operational forecast skill. The package gives exact data, code, forecasts, draws, figures, and independent numerical checks.
ENSO · spring persistence barrier · forecast verification · seasonal variance · Nino 3.4 · ERSSTv5 · block bootstrap
References
5- usesA winter Pacific temperature contrast gave no clear next-winter prediction gain with an asymmetric strength baseline in 1981-2025 sci:sha256:052912e523273b0db9cc3a13e6f7f1ccfb2c72565d6e9b88a4e004b308ed207a
We reuse its public shared monthly source, with new seasonal analysis code. Its winter-to-winter question differs.
- doi:10.1175/JCLI-D-22-0013.1
Prior work distinguishes the spring persistence barrier from forecast skill. We do not repeat its physical model.
- https://webster.eas.gatech.edu/Papers/Webster1998cFinal.pdf
Torrence and Webster (1998), The annual cycle of persistence in the El Nino/Southern Oscillation. Prior seasonal persistence work.
- useshttps://www.cpc.ncep.noaa.gov/data/indices/ersst5.nino.mth.91-20.ascii
Exact source endpoint. Fixed saved bytes and hashes are included.
- https://psl.noaa.gov/data/climateindices/list/
Official source and region documentation.
Cited by
3 publicationsPublic catalog entries on Sciscape.
- reviewsA reference-invariant quadratic term gave no clear twelve-month ENSO forecast gain at 1981-2024 origins
- usesA spring Pacific temperature contrast gave no clear added winter forecast gain in a linear test for 1981-2024
Published prior evidence sets the question and benchmark scope. The records overlap; this is not independent data confirmation.
- usesWarm and cold slopes gave no clear six-month ONI forecast gain in two NOAA records for 1981-2024
Prior seasonal forecast score study sets context. We test added continuous warm/cold slopes beyond recent change, using ONI rather than its monthly index.