Over 1949–2025 the first storm within 1,500 km arrives earlier (−2.3 days a decade) and storms are more frequent, but both trends come from the pre-satellite years. From 1971 or 1978, when the catalogue is consistent, no change in first arrival, storm count or storm-days is detected (p 0.13–0.97). Peak wind within 500 km keeps a similar slope (+5 to +6 kt a decade) but is not significant on the satellite era (p≈0.07–0.09). Over the full record the storm-count rise appears within El Niño, La Niña and neutral years, so ENSO does not explain it; the change in detection before satellites does.
Revised 27 September 2026. Re-run on the storm table rebuilt 19 September 2026 from hurdat2-nepac-1949-2025-091426.txt (the June run used an earlier download) and on NOAA’s current El Niño index. First arrival moved from −2.45 to −2.30 days per decade (p 0.004 → 0.008); peak wind from +4.6 to +4.8 kt per decade. The current index re-classifies several years (El Niño 17 → 15, La Niña 19 → 17, neutral 40 → 44), and with it the first-arrival trend in neutral years drops from p=0.046 to 0.066, so the June claim that the earlier start holds within every phase is withdrawn; the storm-count trend still holds in all three. The decade p90 column now uses a linearly interpolated 90th percentile.
No change detected in when the season starts, or in how many storms it brings, since the satellite era.
The earlier, busier season in the 1949–2025 record comes from the pre-satellite years. Peak wind keeps a similar slope from 1971 or 1978 but is not significant (p≈0.07–0.09).
Four measures, full record against the satellite era.
Annual trend by linear regression; change per ten years. The last column refits from 1978, when geostationary coverage makes the catalogue consistent; p-values allow for year-to-year correlation.
How to read the first row: across 1949–2025 the line through the first-storm dates moves about 2.3 days earlier each decade. Refit from 1971 (+0.62 days, p=0.59) or 1978 (+0.55 days, p=0.72), it shows no detectable change, so the long-record slope comes from storms the pre-satellite catalogue missed early in the season.
Over 1949–2025 the storm-count rise appears in every climate phase.
If the rise were an ENSO artifact it would disappear within each climate phase. It does not, so ENSO is not the cause. These fits use the full record, including the pre-satellite years that produce the rise.
Warmer Pacific
Cooler Pacific
Neither warm nor cool
In this corridor, neutral years (13.7 storms/yr) outpace La Niña (11.7) and El Niño (11.1) — El Niño, which raises basin-wide activity, brings the fewest here. First arrival within each phase: El Niño −3.9 days/decade (p=0.18), La Niña −1.3 (p=0.36), neutral −2.0 (p=0.066) — the same direction everywhere, significant nowhere on its own. El Niño may push tracks westward toward Hawaii, reducing Mexican coastal exposure while raising basin totals. Observation from data; mechanism unconfirmed.
Storm winds near Saladita by decade.
Maximum sustained wind speed (in knots; 1 knot ≈ 1.15 mph) among storms passing within 500 km of Saladita, grouped by decade. The rise is significant over 1949–2025 (+4.8 kt a decade, p=0.002); from 1971 or 1978 the slope is similar (+4.8 to +6.1 kt) but not significant (p≈0.07–0.09).
Early records are less reliable. Before 1966 (17 years of the record), storms were tracked without satellites — some storms were missed entirely, and the apparent rise in storm counts partly reflects better detection, not more storms. Dropping those years removes the earlier start and the rise in counts and storm-days: from 1971 or 1978 none is detected (p 0.13–0.97).
Geostationary satellite coverage improved further in the late 1970s, catching more short-lived and weaker systems. Some of the count increase after 1978 may reflect this surveillance improvement rather than real activity change.
The p-values are approximations. The test assumes each year is independent, but storm counts in adjacent years are weakly correlated. The true effective sample size is probably 30–50 years, not 77. The p-values in the table allow for that correlation; the satellite-era fits are not significant with or without it.
The ENSO classification is a simplification. Each year is classified by its July–September climate state, the peak storm quarter. Other climate indices (different months, different measurement methods) would shift some years between categories and could change the edge-case results.
This is a description of the data, not an explanation. Over the full record the catalogue trends are clear; on the satellite era only peak wind keeps a similar slope, and it is not significant. Possible causes of long-term change — warming sea-surface temperatures, changing wind patterns, expansion of the tropical belt — are active research topics with peer-reviewed findings. This analysis does not assert a mechanism; it describes what changed.
Since the satellite era the date of the season's first storm shows no detectable trend. The earlier start in the 77-year line is built on storms the early catalogue missed.Synthesis · revised September 2026
Sources
NHC HURDAT2 1949–2025, file hurdat2-nepac-1949-2025-091426.txt (storm table built 19 September 2026). NOAA CPC ONI index, current file (JAS average; ≥+0.5°C = El Niño, ≤−0.5°C = La Niña). OLS regression with p-values adjusted for lag-1 autocorrelation, cross-checked with Mann–Kendall. Over 1949–2025 all four p-values clear the four-test Bonferroni threshold (p<0.0125); from 1971 or 1978 none does. *Decade p90 is a linearly interpolated 90th percentile of storm years (years with no storm inside 500 km excluded); with about ten years per decade it is coarse. Decade maxima are on the intensity finding. Script: scripts/analyze_hurricane_phenology.py → functions/api/_findings_hurricane_phenology.js.
Corrected 28 September 2026: the headline now rests on the satellite era, where no change in first arrival, storm count or storm-days is detected.