# Literature / provenance notes for scoring CPC/IRI official ENSO outlooks (2013-08 to 2025-04)

Compiled 2026-10-03 by a research subagent. Read levels: **FULL** = I read the page or PDF text myself; **PARTIAL** = I extracted specific sentences with a script; **SNIPPET** = only a search-engine summary, not verified.

## 1. Spot-check of parsed IRI quick-look probabilities

Independent sources:
- (a) CPC ENSO Diagnostic Discussion (DD) text from `https://www.cpc.ncep.noaa.gov/products/analysis_monitoring/enso_disc_<mon><yyyy>/ensodisc.html` (or `.shtml`). PARTIAL: the synopsis and summary sentences were pulled by script. Raw files are in `/work/enso-outlooks/lit/dd/`.
- (b) For Nov 2020 onward, CPC's archived **"ENSO Strengths" table**, `.../enso_disc_<mon><yyyy>/strengths/index.php`. It gives P(Niño3.4 <= -0.5) and P(>= +0.5) for all 9 seasons, so it is a full independent copy of the La Niña and El Niño probabilities. FULL. Script: `/work/enso-outlooks/lit/fetch_strengths.py`.

| Issue | DD headline statement (source a) | Parsed IRI value(s) | Result |
|---|---|---|---|
| 2014-03 | "about a 50% chance of El Niño developing during the summer or fall" | Eln JAS 49, ASO 50, SON 51 | match |
| 2016-02 | Qualitative only: neutral likely by late spring/early summer, possible La Niña in fall. climate.gov Feb 2016 post: La Niña "slightly favored" by SON (SNIPPET, https://www.climate.gov/news-features/blogs/enso/february-2016-el-ni%C3%B1o-update-q-%E2%80%A6and-some-thursday-morning-quarterbacking) | MJJ Neu 46 > Eln 44; SON Lan 50 | consistent (no number to check) |
| 2016-11 | "slightly favored to persist (~55% chance) through winter 2016-17" | DJF Lan 55 | match |
| 2017-01 | Qualitative only: transition to neutral by Feb 2017, neutral through first half of 2017 | JFM Neu 70, FMA-JJA Neu 54-74 | consistent (no number to check) |
| 2020-01 | neutral "~60%" spring, "~50%" summer | FMA/MAM/AMJ Neu 60/61/60; JJA/JAS 52/49 | match |
| 2020-05 | ~65% neutral in summer; 45-50% in autumn; La Niña ~45% at end of year; ~10% El Niño | JJA Neu 64; SON/OND 48/45; NDJ/DJF Lan 46; Eln 10-12 | match |
| 2021-02 | ~60% transition to neutral in AMJ; fall ~50% La Niña | AMJ Neu 60; ASO/SON Lan 49/52 | match. Strengths table (b): all 9 seasons exact |
| 2022-04 | 59% La Niña JJA; 50-55% through fall | JJA 59; ASO-NDJ 51-54 | match. (b): exact |
| 2023-01 | 82% neutral in MAM | MAM Neu 82 | match. (b): exact |
| 2023-07 | >90% El Niño through winter | NDJ/DJF Eln 96/93 | match. (b): exact (97,96,96,96,96,96,93,90,85) |
| 2023-10 | 80% El Niño in MAM 2024 | MAM Eln 80 | match. (b): exact |
| 2023-11 | 62% El Niño in AMJ 2024 | AMJ Eln 62 | match. (b): exact |

**Full cross-check (source b):** I compared 54 issue months (2020-11 to 2025-04; 486 season cells) against `official_probs_raw.csv`. 485/486 cells match exactly. The one exception is 2021-01 ASO: IRI gives Lan 44 / Neu 41 / Eln 14 (sums to 99), while CPC gives Eln 15. That is a one-point rounding or transcription difference. Run `/work/enso-outlooks/lit/build_cpc_archive.py` to reproduce.

**Is the IRI quick-look table the official consensus, not IRI's model-based forecast?** Yes, in all three eras.
- **2013-2015 and 2016-2019.** Each quick-look page puts the parsed table in the first section, which reproduces the CPC DD text. The separate "CCSR/IRI Technical ENSO Update and Model-Based Probabilistic ENSO Forecast" heading (published around the 19th-21st) comes after the table. The page text says: "Figures 1 (the official CPC ENSO probability forecast) and 3 (the objective model-based CCSR/IRI ENSO probability forecast)". I checked the saved pages for 2014-03, 2016-02, 2017-01 and 2019-06 in `/work/enso-outlooks/raw/iri_ql/` (PARTIAL). The headline numbers match the CPC DD for 2014-03 and 2016-11, and DDs from this era link "CPC/IRI consensus forecast" to the IRI page. The 2019-06 quick-look also says "The official CPC/IRI outlook ... calls for a 66% chance of El Niño ... Jun-Aug", and the parsed table is in the same section.
- **2020-2025.** Headline matches, plus the exact 9-season match with CPC's own strengths archive. The 2025-05 quick-look still says "The official CPC ENSO probability forecast, based on a consensus of CPC forecasters" (https://iri.columbia.edu/our-expertise/climate/forecasts/enso/2025-may-quick-look/, PARTIAL).
- **Labelling change.** DDs through about 2022 say "issued by CLIMATE PREDICTION CENTER/NCEP/NWS and the International Research Institute for Climate and Society". The Jan 2023 DD onward says only "CLIMATE PREDICTION CENTER/NCEP/NWS" (https://www.cpc.ncep.noaa.gov/products/analysis_monitoring/enso_disc_jan2023/ensodisc.html). From 2023 the product is formally CPC-only, though IRI still reposts it.

## 2. Definitions: what is verified

- **CPC DD, Nov 2016** (PARTIAL): "La Niña conditions (3-month average Niño-3.4 index less than or equal to -0.5°C)". https://www.cpc.ncep.noaa.gov/products/analysis_monitoring/enso_disc_nov2016/ensodisc.shtml
- **CPC DD, Jan 2017** (PARTIAL): "ENSO-neutral (3-month average Niño-3.4 index between -0.5°C and 0.5°C)". https://www.cpc.ncep.noaa.gov/products/analysis_monitoring/enso_disc_jan2017/ensodisc.shtml
- So each season's category depends only on that season's 3-month value. No 5-season persistence is needed.
- **L'Heureux et al. 2019** (FULL PDF) defines the categories as La Niña Niño-3.4 <= -0.5, neutral -0.5 < x < +0.5, El Niño >= +0.5. It verifies against ERSSTv5 seasonal Niño-3.4, "the basis for NOAA's official ENSO index, the oceanic Niño index", with "departures ... computed based on rolling 30-yr" base periods. https://repository.library.noaa.gov/view/noaa/20555/noaa_20555_DS1.pdf
- **CPC ONI page** (PARTIAL): "threshold of +/- 0.5 °C for the ONI [3 month running mean of ERSST.v6 SST anomalies in the Niño 3.4 region (5N-5S, 120-170W)], based on centered 30-year base periods updated every 5 years. For historical purposes, periods ... are colored ... when the threshold is met for a minimum of 5 consecutive overlapping seasons". The 5-season rule therefore applies to historical *episodes*, not to the per-season probabilities. https://www.cpc.ncep.noaa.gov/products/analysis_monitoring/enso/oni/v6/
- **CAUTION: index changes.**
  - The old ONI_v5 page now says "This page has moved". The ERSSTv5 ONI table has moved to https://www.cpc.ncep.noaa.gov/products/analysis_monitoring/enso/oni/v5/, and the current ONI is ERSSTv6. A search snippet (SNIPPET only) said NCEI is discontinuing ERSSTv5 (around Aug 2026). Pin the verification ONI version (v5 vs v6) and the download date.
  - Per NWS Public Information Statement 26-05, CPC switched official monitoring and prediction to the **Relative ONI (RONI)** in Feb 2026, using a 1991-2020 base. https://www.cpc.ncep.noaa.gov/products/analysis_monitoring/enso/roni/announcement.php (FULL) and https://www.cpc.ncep.noaa.gov/products/analysis_monitoring/enso/roni/probabilities/ (FULL), which says "Probabilities are verified using the RONI, with El Niño and La Niña defined using -/+ 0.5 °C thresholds". This does not affect forecasts issued 2013-2025, which were made against ONI. It does matter if forecasts issued from Feb 2026 are added.

## 3. Prior verification of the official probabilities

- **L'Heureux, Tippett, Takahashi, Barnston, Becker et al. 2019, "Strength Outlooks for the ENSO", Wea. Forecasting 34, 165-175** (FULL PDF). https://doi.org/10.1175/WAF-D-18-0126.1 and https://repository.library.noaa.gov/view/noaa/20555/noaa_20555_DS1.pdf
  - **Method.** The official outlook ("CPCoff") is the average of the individual forecasters' 3-category probabilities. "Forecasters have been providing three-category ENSO probabilities since January 2012."
  - **Scope.** It verifies 41 forecasts, Jun 2015 to Sep 2018, using RPSS and LSS by target season and lead (-2 to 7). Numbers appear only in figures; there is no table.
  - **Main finding: the forecasters were underconfident.** CPCoff probabilities were "more conservative", and "forecasters tend to imagine the spread is wider than it really is".
  - **Ranking.** CPCcalib (forecaster mean Niño-3.4 plus a Gaussian with historically calibrated spread) has higher mean and median RPSS/LSS than CPCoff. NMME-calibrated forecasts have a higher median but large busts, notably summer 2016 (rapid La Niña) and mid-2017.
  - **Skill swings.** All methods had negative skill at all leads after the 2016 El Niño decay.
  - Supplementary material compares Jan 2012 onward (not read).
- **Process change, about Nov 2020** (inferred plus observed). The CPC DD first links "A probabilistic strength forecast is available here" in Nov 2020 (https://www.cpc.ncep.noaa.gov/products/analysis_monitoring/enso_disc_nov2020/ensodisc.shtml). The Oct 2020 DD has no such link. The strengths page cites L'Heureux et al. 2019 as the basis for its values. Its P(>= 0.5) and P(<= -0.5) equal the official 3-category probabilities in 485/486 cells. This strongly suggests the official 3-category probabilities have been generated by the CPCcalib method since about Nov 2020, replacing the averaged subjective probabilities (INFERRED, not stated outright). A climate.gov Apr 2023 post reportedly says "we've only been using it for a couple of years" (SNIPPET, https://www.climate.gov/news-features/blogs/enso/april-2023-enso-update-el-nino-watch). The 2019 climate.gov post "ENSO forecast mash-ups" describes forecasters supplying Niño-3.4 means (SNIPPET, https://www.climate.gov/news-features/blogs/enso/enso-forecast-mash-ups-what%E2%80%99s-best-way-combine-human-expertise-models).
  - **Implication for scoring.** Treat 2013-08 to 2020-10 (subjective averaging) and 2020-11 onward (calibrated) as two regimes when assessing reliability.
- **Table format change** (observed). Through 2023-10 the strengths table has 6 columns (±0.5, ±1.0, ±1.5). From 2023-11 it has 8 columns, adding ±2.0.
- **climate.gov, M. L'Heureux comment, 2014** (SNIPPET): the CPC/IRI joint probabilistic forecasts started in early 2012, and the record was "too short to quantify our reliability scores". I found no climate.gov post that verifies the official 3-category probabilities. Related posts:
  - "How Good Have ENSO Forecasts Been Lately?" (Barnston; plume, 2012-14). https://www.climate.gov/news-features/blogs/enso/how-good-have-enso-forecasts-been-lately (SNIPPET)
  - "How good have models been at predicting ENSO in the 21st century?" (on Ehsan et al. 2024). https://www.climate.gov/news-features/blogs/enso/how-good-have-models-been-predicting-enso-21st-century (SNIPPET)
- **Ehsan, L'Heureux, Tippett, Robertson, Turmelle 2024, npj Clim. Atmos. Sci.** "Real-time ENSO forecast skill evaluated over the last two decades..." covers 247 real-time IRI-plume Niño3.4 forecasts, Feb 2002 to Aug 2022. It is a deterministic model average, not the official probabilities. Reported: La Niña onset hit rate 40-60% at 2-season lead, near zero at longer leads. https://www.nature.com/articles/s41612-024-00845-5 (SNIPPET)
- **Tippett, Ranganathan, L'Heureux, Barnston, DelSole 2019, Clim. Dyn.**: NMME 3/5/7-category ENSO probabilities 1982-2015, RPSS vs LSS. Model-based, not official. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6934243/ (SNIPPET)
- **Tippett, L'Heureux, Becker, Kumar 2020, GRL, "Excessive momentum and false alarms in late-spring ENSO forecasts"**: NMME. https://repository.library.noaa.gov/view/noaa/54881 (SNIPPET)
- **Barnston, Tippett, L'Heureux, Li, DeWitt 2012, BAMS 93, 631-651**, "Skill of real-time seasonal ENSO model predictions during 2002-11": IRI plume skill. https://doi.org/10.1175/BAMS-D-11-00111.1 (cited in L'Heureux 2019; not read)

## 4. Machine-readable archives beyond the IRI tables

- **2020-11 to 2026-04, CPC per-month strengths pages** (HTML tables, parseable): `https://www.cpc.ncep.noaa.gov/products/analysis_monitoring/enso_disc_<mon><yyyy>/strengths/index.php`. Scraped to `/work/enso-outlooks/data/cpc_strengths_lan_eln.csv`, where p_neu = 100 - p_lan - p_eln. These cover **May 2025 to Jan 2026**, which the IRI table lacks. They do not exist for 2020-09/10 or earlier, or for 2026-05 onward.
- **2026-04 onward, CPC RONI probabilities archive** (HTML table with La Niña/Neutral/El Niño columns): `https://www.cpc.ncep.noaa.gov/products/analysis_monitoring/enso/roni/archives/?type=probabilities&month=MM&year=2026`. The current issue is at `.../enso/roni/probabilities/`. Months 2026-04 to 2026-09 are appended to the same CSV, and the 2026-04 values are identical in both sources. These forecasts verify against RONI.
- **2002-2013.** The CPC/IRI official consensus only started in early 2012 (L'Heureux et al. 2019). The new-style IRI quick-look URLs return 404 for 2011 to 2013-07 (checked 2011-06, 2012-06, 2013-02, 2013-07).
  - The 2012 DDs link the consensus to `http://iri.columbia.edu/climate/ENSO/currentinfo/figure1.html`. Wayback Machine snapshots exist for the old `iri.columbia.edu/climate/ENSO/currentinfo/QuickLook.html` monthly from 2002 (CDX query worked; content not inspected; the figure1 CDX query got "Temporarily Offline").
  - Before 2012 those pages carry IRI's own probabilistic forecast, not the CPC consensus.
  - For 2012-01 to 2013-07, Wayback snapshots of figure1.html or QuickLook.html are the likely source (UNVERIFIED).
  - I found no CSV or JSON archive from CPC or IRI.