Revised 26 September 2026. The June version of this essay quoted analyses run at a grid cell 44 km to the south-east of La Saladita — the wrong town. On 19 September 2026 the site’s coordinates were corrected to 17.8375°N, 101.7606°W and the datasets rebuilt; two of the essay’s three legs moved. The wave-power leg weakened: the annual-peak trend is now +3.8 kW/m per decade at p=0.27 — positive, inside the noise — and the “24 → 81 kW/m” contrast is gone (1979’s hardest hour was 40 kW/m; the record is 232 kW/m in 1993, not 81 in 2024). The rainfall leg changed shape: it rests on wet-run clustering and a shortening longest wet spell; where the June text said “drier dry stretches”, it was wrong. The storm-intensity leg comes from HURDAT2, not from the grid cell. The argument now stands on two legs; the third shows no trend. The paragraphs below have been revised to say so; the essay is otherwise as written.
Revised 27 September 2026. The storm leg quoted the “90th percentile of decade-level peak intensity” as 120 knots in the 1950s and 185 in the 2010s. Those numbers are the decade maximum, the single strongest storm of each decade (1959 and 2015): the script’s percentile index landed on the last value when a decade has ten or fewer years. Re-run on the current HURDAT2 table with a proper interpolated percentile, the 90th percentile rises too, from 80 to 131 knots (p=0.005), and the maximum’s rise holds (p=0.032). Decade spread (standard deviation, IQR) does not trend, so the storm leg is about the top of the distribution moving up, not the distribution widening. The storm paragraphs below now say so.
Revised 29 September 2026. The rainfall leg was checked against two independent daily rain products, the CPC gauge analysis and MERRA-2. Wet runs rise in all three and only ERA5 is significant. The shorter longest wet spell (−3.26 days per decade in ERA5) is not reproduced: CPC gives −0.44 with an interval that spans zero, and MERRA-2 +1.97. No rainfall-structure trend is robust across products. The storm leg is unchanged. The rainfall paragraphs below now say so; the essay is otherwise as written.
On the morning of June 8, 2026, the surf forecast for La Saladita said small, clean, dropping. The reality was a sky the color of bruised pewter and a wind blowing sideways off the headland with enough water in it to make the eyes sting.
Tropical Storm Boris — first of the Eastern Pacific season — sat 280 kilometers off the coast. The forecast had not seen him.
That missed forecast is the small story. The bigger story is what we found when we went back and looked at every storm that had visited this stretch of Pacific in the seventy-five years on record. And at the rain. And at the waves. The same odd shape kept showing up.
What averages are good at hiding
When climate scientists describe what is changing in the ocean, they tend to lead with averages. Average sea-surface temperature. Average storm intensity. Average annual rainfall. Averages are easy to communicate. They are hard to argue with. And they are beloved of policymakers because they describe a world that is changing slowly enough to be planned for.
Averages, it turns out, are often the wrong number.
Consider what an average can hold steady. Take a hundred days of rain in a year. Spread them evenly across three months and you have a wet season. Stack ninety of them into two weeks and leave the rest scattered — same total, very different country. The mean does not flinch. The land does.
Storms
Here is the polite story about storms at this coast. The mean peak intensity of hurricanes that have come within five hundred kilometers of Saladita has risen about +4.8 knots per decade since 1949. A modest slope. The kind of trend a graph can hold without looking alarmed.
Here is the dangerous story. The strongest storm to come within five hundred kilometers in a decade — the decade maximum, one storm — was 120 knots in the 1950s. In the 2010s it was 185 knots, Category-5 territory; the 2000s and 2020s maxima are Category 5 as well. The strong-storm years a notch below the top moved too: the decade 90th percentile went from about 80 knots to 131. In seventy years, the average storm got a little stronger. The strongest storm got a lot stronger.
Five statistical methods agree on this, for the maximum and the 90th percentile alike — Pearson correlation, Mann–Kendall, permutation, rolling-window Pearson, rolling-window Mann–Kendall — though none of the decade-level tests clears a strict multiple-testing bar, and the decade spread does not trend. The rise at the top survives restriction to the satellite era, so it is not a record-keeping artifact. The same chorus loop that produced this finding caught two other plausible-looking signals as catalog artifacts and walked them back. This one held.
Rainfall
A meteorologist looking at the rainfall record for this stretch of Guerrero coast would tell you, accurately, that annual totals have not changed in any way you could prove. The same amount of water falls now as fell forty years ago.
Anyone who has farmed this stretch of coast for the past forty years has also noticed the change — and would also be right.
Both can be right. The annual mean is flat. In ERA5, what has shifted is how the water is delivered. The number of distinct bursts of consecutive wet days — wet runs — has risen by about one run per decade (p=0.015), and the longest wet spell has shortened by about three days per decade (p=0.0008). The water comes in more, shorter bursts. The heaviest single day of the year trends up about 12 mm per decade, but at p=0.058 that is marginal, not a finding. Two other daily rain products, the CPC gauge analysis and MERRA-2, were run with the same definitions. Wet runs rise in both, by less, and inside the noise. Neither shows the shorter wet spell: CPC gives −0.44 days per decade with an interval that spans zero, MERRA-2 +1.97. The taps are the same; on the timing, the records do not agree.
Waves
The pattern a third time, in a signal that has nothing structurally to do with the first two — or so the June version of this essay said. Mean annual wave power at the nearest ERA5-Wave grid point (17.5°N, 102.0°W, about 45 km offshore to the south-west of La Saladita) — offshore forcing, not conditions at the break: flat across forty-seven years of record. That part holds. Peak wave power per year, at that offshore node: a slope of +3.8 kW/m per decade at p=0.27, no trend detected. 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. The average swell is the average swell. Whether the big day is bigger, this record cannot say. The third leg does not bear weight.
Three signals, one shape
Three signals. Same coast. The storm record shows the shape; rainfall shows it in one product of three; wave power shows no trend. The everyday is steady. The rare is louder.
You could plausibly look at any one of these by itself and tell yourself a story about local variability. Maybe the storm record looks like that because of the satellite era's improved detection. We checked; it does not. Maybe the rainfall pattern is ENSO. Across the windows tested, it persists across both phases in ERA5; it is the product, not the phase, that it does not survive. Maybe wave power is being moved by some Pacific basin oscillation we do not understand. At the offshore node we cannot rule that out — the wave trend is the one leg that does not clear significance, and a single hour in 1993 is enough to flip its verdict either way.
What is harder to wave off is that the two signals that clear the bar in their own records are doing the same thing. One of them, rainfall, clears it in a single product.
The individual signals each appear in regional climate literature — what this site is doing is showing all three operating simultaneously at one specific coastal point. That cross-signal convergence at a single location is the thing worth pausing over, rather than any of the components on their own.
Why the tail is the part that matters
The reason this matters comes down to a basic fact about how things break. Coastal infrastructure, fishing operations, beach communities, the families running the turtle camps at Petatillo and Ayotlcalli, the palapas a hundred meters from the lagoon — none of them are sized to the mean. They are sized to the extremes.
A road designed for the average storm is a road that fails in the strong-storm year. A village provisioned for the average wet season is caught out when the same water arrives in fewer, harder bursts. A house built for forty years of average swell is the house that washes out on the one day the average never described.
Climate communication has spent a generation telling the world that the averages are moving. The averages are moving. But the part of the world that breaks things is not moving the way the averages are moving. It is moving differently. It is moving faster.
The forecast that missed
On June 8, when the forecast missed Tropical Storm Boris, the small failure was a tropical system too close, too soon, in a model that wasn't quite watching the right window. The deeper failure — the one we have been making for longer, and the one this essay is about — is reading this coast through the mean of its weather when the news has been in the tail.
The mean tells you what to expect. The tail tells you what can erase you.
The two have not been moving in unison. They have not been moving in unison for some time.