# Prior work: calibration and skill of seasonal Atlantic hurricane forecasts

Collected 2026-10-02. All claims below were read from the cited source (provenance: web_source). Where a
local copy exists it is in this directory. Treat blog items as unreviewed.

## (a) Do NOAA's "70% likely" ranges contain the outcome?

1. **Blake, Pasch & Bell (2010), "Verification of 12 years of NOAA seasonal hurricane forecasts",
   AMS 29th Conf. on Hurricanes and Tropical Meteorology, paper 3A.1.**
   https://ams.confex.com/ams/pdfpapers/167939.pdf (local: noaa_12yr_verification_ams.pdf/.txt)
   - May 1999-2009 and Aug 1998-2009, scored by range midpoint vs a 5-yr running-mean benchmark. May outlooks:
     MAE only slightly below the 5-yr mean; only May ACE improvement significant (>95%). August outlooks:
     MSE improvement 32-62% over the 5-yr mean for all parameters; cuts TS and hurricane errors by ~1 per season.
   - Calibration: "the May outlooks have not been verifying within the outlook ranges at the target rate of 70%"
     (their Fig. 5; exact percentages only in the figure). May and August range widths were "nearly the same",
     and they recommend wider May ranges. NOAA named-storm forecasts "too low on average".
   - NOAA's May forecasts were comparable to CSU's and better than TSR's. In August, NOAA was more skilful
     than CSU or TSR, with the largest gap in named storms.

2. **Bell, G. (CPC), "NOAA Updated 2015 Atlantic Hurricane Season Outlook", NRCC webinar slides, 27 Aug 2015.**
   https://www.nrcc.cornell.edu/workshops/webinars/2015/08/present_3.pdf (local: nrcc_2015_present3.pdf/.txt)
   - A forecast counts as correct if the observation falls in the range. May outlooks are verified against Jun-Nov
     and August outlooks against Aug-Nov. August outlooks were more accurate than May ones. "Since 2008, for both
     the May and August outlooks approximately 70% of forecasted parameters have verified", and both improved
     substantially after CFS was introduced in 2008. The bar values for 2001-2007 are only in the chart.

3. **Olson, R. (19 May 2026), "NOAA's hurricane outlook misses one way: it underforecasts big seasons" (blog).**
   https://www.randalolson.com/2026/05/19/noaa-hurricane-outlook-vs-reality/
   - Covers May named-storm ranges only, 2003-2025: 14 of 23 hits (~61%). 8 of the 9 misses were
     underforecasts; 2006 was the only overforecast. The page says the chart was built end-to-end by an AI agent.
   - Similar unreviewed tally (WUWT, May 2026; climate-skeptic blog): ~17/25 (~68%) for May named storms.
     https://wattsupwiththat.com/2026/05/13/scorecard-how-well-does-noaas-hurricane-outlook-actually-perform-26-years-of-forecasts-vs-reality/

## (b) Skill comparisons across agencies, and UKMO

4. **Klotzbach, P. et al. (2019), "Seasonal Tropical Cyclone Forecasting", Tropical Cyclone Research and Review
   8(3):134-149, doi:10.6057/2019TCRR03.03.** https://repository.library.noaa.gov/view/noaa/63904
   (local: klotzbach2019_tcrr.pdf/.txt)
   - CSU/TSR/NOAA real-time, 2003-2018: r ~0-0.2 in early April, ~0.2-0.4 in early June, and ~0.5-0.8 in early
     August. NOAA named storms: r = 0.39 for the mid-May outlook and 0.83 for the early-August outlook.
     TSR is best for ACE; NOAA is best for named storms in August.
   - **UKMO: "Verification of real-time TC frequency forecasts issued by the Met Office between 2007 and 2018
     ... Over the 12 years of forecasts, 10 have verified well with observed values falling within the range
     predicted"** (Fig. 5c; Jul-Nov for 2007-10, Jun-Nov for 2011-18). That is 83% against a nominal 70%.
   - For comparison, Hong Kong Observatory range forecasts: 60% correct in 2009-2018.

5. **Klotzbach, Saunders, Bell & Blake (2017), "North Atlantic Seasonal Hurricane Prediction", AGU Geophysical
   Monograph 226 (Climate Extremes), ch. 19.** https://tropical.colostate.edu/Publications/bookchapters/klotzbach_et_al_2017.pdf
   (local: klotzbach2017_chapter.pdf/.txt)
   - Real-time CSU/NOAA/TSR 2003-2014, scored with Spearman rank r and MSSS, using the NOAA range midpoint.
     Skill is moderate-to-good in August, low-to-moderate in June, and lower in April. TSR is best for ACE and
     NOAA is best for August named storms. NOAA ranges have been explicitly ~70% since May 2003.

6. **Camp, J. et al. (2015), "Seasonal forecasting of tropical storms using the Met Office GloSea5 seasonal
   forecast system", QJRMS 141:2206-2219, doi:10.1002/qj.2516.** https://rmets.onlinelibrary.wiley.com/doi/10.1002/qj.2516
   - Hindcasts (1996-2009) of Jun-Nov North Atlantic TS numbers: ensemble-mean correlation r = 0.51 (significant).
     ACE is also significant. Caribbean landfall shows skill; US landfall shows none.

7. **TSR forecast documents (Saunders/Lea; e.g. April 2023, December 2025 updates).**
   https://www.tropicalstormrisk.com/docs/TSRATLForecastApril2023.pdf (local: tsr_april2023.pdf/.txt);
   https://www.tropicalstormrisk.com/docs/TSRATLForecastDecember2025.pdf
   - Self-verification uses MSSS against a running 10-yr norm for publicly issued 2003-2022 forecasts. Skill is
     0% at the December/April lead for ACE, intense hurricanes, hurricanes and TS, rising by August (their Fig. 3
     compares TSR/CSU/NOAA). The December 2025 doc gives 4%/3%/0%/0% for 2003-2024.
   - Lea & Saunders (2004) hindcast August ACE MSSS was 37% for 1950-2003. US ACE August replicated real-time
     r = 0.54 for 1980-2022 (Saunders & Lea 2005, Nature 434:1005).

## Not obtained / leads
- Takaya et al. (2023), TCRR 12(3):182-199, update of #4: https://www.sciencedirect.com/science/article/pii/S2225603223000413
- Owens & Landsea (2003), Wea. Forecasting: skill of the Gray/CSU forecasts relative to climatology.
- Yale Climate Connections graphic on TSR/CSU/NOAA accuracy: https://yaleclimateconnections.org/2024/04/forecasters-predict-an-extremely-active-2024-atlantic-hurricane-season/tsr-csu-noaa-accuracy/
- No source found that scores UKMO 70% ranges for hurricanes, major hurricanes or ACE, or any UKMO year after 2018.