trend (1979–2025)
trend (1979–2025)
(peak >25 kW/m, 1993–2023)
(1979–2025, all months)
The question isn’t how many swells arrived. It’s how much energy they carried when they did. Wave power integrates wave height squared and period together — a 1.8-meter, 14-second south swell delivers about 18 times the energy of a 0.6-meter, 7-second chop. This analysis runs 47 years of hourly ERA5-Wave reanalysis through the deepwater power approximation P ≈ 0.5 × Hs² × Tp at the nearest ERA5-Wave grid point (17.5°N, 102.0°W, about 45 km offshore to the south-west) — offshore forcing, not conditions at the break — then asks whether the number changed over time.
Four metrics, four regression tests. None passes cleanly at p<0.05. One is marginal (the April–October season mean, p=0.096), one is flat (annual mean power), and peak annual power is positive but inside the noise (p=0.27). Energetic days show no trend once a step in the model record is allowed for: about 15 a year before 1992 and 24 after 1993, with no trend within 1993–2023 (−0.15 a decade, p=0.94). The step lines up with ERA5 starting to assimilate satellite wave heights in 1991–93. No systematic change is detected in the mean wave energy a surfer experiences in any given week over 47 years. Whether the upper tail — the hardest days, the heaviest swells — is changing, this record cannot yet say.
Revised 26 September 2026: rerun with the corrected site coordinates (17.8375°N, 101.7606°W); wave values come from the nearest ERA5-Wave node, 17.5°N 102.0°W. The June 2026 version of this page used a site coordinate 44 km to the south-east (17.5897°N, 101.4317°W — a different town), which resolved to a different ERA5-Wave node, and reported a peak-power trend of +4.1 kW/m per decade at p=0.032, which the page then headlined as “peak power climbing”. At the corrected node the slope is +3.8 kW/m per decade at p=0.27. Mean power reads 11.2 kW/m rather than 9.5, September 14.6 kW/m rather than 12.7, the average year sees 21 energetic days rather than 12, and the record hour is 232 kW/m in 1993, not 81 kW/m in 2024. Every number below is from the artifact generated 19 September 2026.
Revised 27 September 2026: the 1993 record hour is now named (Hurricane Calvin, 129 km offshore, 6–7 July 1993), and a new section, What makes a peak year, matches every annual peak to the hurricane best track. No trend number above changed.
No signal clears p<0.05. One is marginal.
OLS regression of each annual metric vs year. 47 data points; p-values approximate (autocorrelation is possible at annual scale).
Peak annual power: +3.8 kW/m per decade (p=0.27, R²=0.03). Positive, but inside the noise.
In 1979 the single hardest hour of the year was modeled at about 40 kW/m. The record hour is 232 kW/m in 1993 — Hurricane Calvin passing 129 km offshore on 6–7 July, a single tropical-cyclone hour that dwarfs everything else — followed by 2009 (111 kW/m), 2024 (103), 1992 (99) and 2018 (96). Years above 80 kW/m fall in both halves of the record, and the mean annual peak is 60.2 kW/m for 1979–2001 against 60.7 kW/m for 2002–2025. The fitted slope is upward, but R²=0.03 means the trend explains almost none of the year-to-year variance, and the June headline that the hardest hours are getting harder is not supported at this cell.
Mean annual wave power: effectively flat at 11.2 kW/m (p=0.85).
The average week at Saladita is indistinguishable energetically from the average week in 1979. If anything has changed for a regular, it is not the baseline. The April–October season mean is marginal (p=0.096). *The full-record energetic-day slope comes from a step between 1979–91 (about 15 a year) and 1993–2023 (24 a year), with no trend within 1993–2023 (−0.15 a decade, p=0.94); the step lines up with ERA5 starting to assimilate satellite wave heights in 1991–93.
The hardest day of the year is usually a storm passing close.
Each year’s peak day was matched to the nearest NHC HURDAT2 best-track fix within 1,000 km of the break and one day of the peak. Figures here use the 0.49 coefficient of the matching script, so they run 2% below the 0.50 figures elsewhere on this page (the 1993 record reads 227.7 here, 232.4 above); the page’s own figures are in brackets.
34 of 47 annual peaks were cyclone swell, 10 were groundswell, 3 were mixed.
Cyclone swell here means a storm within 400 km of the break, or within 1,000 km with the swell direction pointing within 60° of it. Groundswell means no storm fix within 1,000 km and a day either side. Mixed means a storm in range but the swell coming from somewhere else. The share barely moves between halves of the record: 19 cyclone and 4 groundswell in 1979–2001, 15 cyclone, 6 groundswell and 3 mixed in 2002–2025.
Cyclone peaks are short-period. Groundswell peaks are long-period and from the south-south-west.
Every cyclone-matched peak from a storm within 530 km had a wave period of 10.6 s or less. The three exceptions were distant storms: Lane 2000 (732 km, 11.1 s), Ileana 1994 (647 km, 11.9 s) and Alma 2002 (944 km, 13.2 s). The ten groundswell peaks ran 13.1–16.2 s from 194–210°. The period here is ERA5’s wave_period. Alma’s 13.2 s overlaps the groundswell floor of 13.1 s, so period alone does not separate the two classes at the margin.
With storm days removed, the annual peak rises +2.24 kW/m per decade (p=0.023, n=47).
Take each year’s hardest day among the days with no storm fix within 1,000 km and one day either side. That groundswell-only annual peak averages 38.3 kW/m and rises +2.24 kW/m per decade. The storm-day peak rises +3.7 kW/m per decade at p=0.28, inside the noise. Four caveats, stated plainly. The split was built after looking at the data, so treat the p-value as closer to 0.05–0.07 than 0.023. ERA5 began assimilating satellite altimeter wave heights in 1991–93, which could put a step into the record that reads as a trend. The regression is OLS with no autocorrelation correction. The 1,000 km, ±1 day band is a choice; a different band would move days between the classes. The same 1991–93 step that explains the energetic-day count could produce this slope, so it is not a finding.
September is the hardest month.
Mean and 90th-percentile wave power by calendar month, pooled across all 47 years. SH groundswell season (April–October) highlighted.
The season opens in April, peaks in September, and drops sharply after October.
September averages 14.6 kW/m mean power and a 90th percentile of 24.6 kW/m — about twice December's average of 7.1 kW/m. June (14.1 kW/m) and May (13.8) follow, holding above August (13.2) and July (13.1). April and October (both 12.0) drop off but remain above the annual mean of 11.2. The window April–October accounts for the bulk of all meaningful surf energy at Saladita.
Consistent energy, ceiling unproven.
Wave power touches two things at Saladita: surf quality and the reef below it.
The everyday session hasn't changed. Whether the occasional big day has is an open question.
Mean energy is flat, so the typical chest-high groundswell that makes Saladita rideable is arriving with the same average frequency and size it did in the 1980s. The upper-tail events — the days when it closes out, when the current runs hard, when midlength equipment makes more sense than a longboard — show no trend: the count of energetic days steps up between 1979–91 and 1993–2023, when ERA5 began assimilating satellite wave heights, and does not trend within 1993–2023 (−0.15 a decade, p=0.94). The peak-power trend does not clear significance either. No change is detected in how often the “lots of water” days that call for different equipment arrive.
Extreme events drive coral stress. The wave trend is not established.
Coral structural integrity at reef-flat depths depends partly on wave-induced mechanical stress during extreme events. No trend in energetic days is detected once the 1991–93 step in the model record is allowed for, so there is no evidence here of more hours of high shear stress on coral per year. The documented axis of change is thermal: surface warming of +0.13°C per decade from the companion climate analysis. This analysis cannot quantify biological impact; it documents that the mechanical driver has not demonstrably moved while the thermal one has.
ERA5-Wave is a reanalysis, not a measurement. The wave field is derived from a global atmospheric model forced with ERA5 winds, then validated against buoy observations globally, at the nearest 0.5° ERA5-Wave grid node (17.5°N, 102.0°W) — an offshore point in deep water, about 45 km south-west of La Saladita, not validated against any local instrument. Saladita’s particular bathymetry, headland shadow, and kelp/reef bottom interact with incoming swells in ways this offshore node cannot resolve. Local wave heights at the break can differ meaningfully from the model output.
The deepwater power formula is an approximation. P ≈ 0.5 × Hs² × Tp assumes fully developed deepwater conditions. In the transitional depth range where Saladita breaks, wave energy is modified by shoaling and refraction. The formula gives a useful relative proxy but not an absolute physical energy at the break.
The peak-power trend is not significant, and it is hostage to one hour. R²=0.03 means about 97% of the year-to-year variance in peak power is explained by factors other than the long-term trend — ENSO state, tropical cyclone proximity, north Pacific storm tracks. The 232 kW/m hour in 1993 — Hurricane Calvin, 129 km offshore, 6–7 July 1993 — sits early in the record and pulls the fitted line flat; drop that one year and the slope is +5.7 kW/m per decade at p=0.004. We report the all-years result from the artifact as the finding. That the verdict flips on a single hour is exactly why it should not be headlined in either direction.
The Open-Meteo marine API requires models=era5_ocean for pre-2022 data. Without that parameter, the default endpoint only returns data from October 2021. All values here use the full ERA5-Wave reanalysis; the script is reproducible and the parameter is documented.
The ocean at the ERA5-Wave offshore node delivers about the same average energy it did in 1979. Whether the hardest hours — the ones that determine what boards you bring and how the reef ages — are getting harder is an open question: no trend is detected.Synthesis · ERA5-Wave, 412,008 hourly observations · Revised 26 September 2026
Sources & method
Open-Meteo marine archive, ERA5-Wave reanalysis (models=era5_ocean), requested at 17.8375°N, 101.7606°W (the corrected site coordinate); the API resolves this to the nearest 0.5° grid node, 17.5°N, 102.0°W, January 1979–December 2025. Variables: wave_height (Hs, significant wave height in m), wave_period (Tp, peak period in s). Wave power computed as P ≈ 0.5 × Hs² × Tp (deepwater approximation; P in kW/m). 412,008 valid hourly records.
Annual metrics aggregated per calendar year. Linear regression with scipy.stats.linregress (OLS). p-values approximate; temporal autocorrelation in annual series is possible. Monthly climatology pools all 47 years of each calendar month. Energetic days were also compared across 1979–91 and 1993–2023 (Welch p=0.0008) and fit within 1993–2023; whether the step comes from ERA5 or the ocean is untested.
Corrected 28 September 2026: energetic days are now read as a step around 1991–93 in the model record, with no trend within 1993–2023, rather than as a marginal rise.
Script: scripts/analyze_wave_power_trend.py → functions/api/_findings_wave_power.js. Artifact JSON served at /api/findings-wave-power. Artifact generated 19 September 2026 with the corrected site coordinates; wave values come from the nearest ERA5-Wave node, 17.5°N 102.0°W. The June 2026 numbers on this page came from the 17.5897°N, 101.4317°W site coordinate, which resolved to a different ERA5-Wave node, and are superseded.