La Saladita forecast.

The data layer.

We pull live marine and atmospheric data from Open-Meteo's free public APIs every 30 minutes — significant wave height, swell period, swell direction, wind speed, wind direction, temperature, precipitation — at La Saladita's coordinates (17.5897°N, 101.4317°W).

The interpretation layer.

The raw data is run through a rules engine encoding local knowledge of the wave. Specifically:

• Swell direction quality — south swells (185°–200°) produce the cleanest peeling rides; the band 175°–215° is productive; west-heavy swells (>215°) close out the inside reform; northerly swells aren't the productive swell at Saladita.
• Size translation — Open-Meteo's significant wave height (Hs) is converted to typical face-height at the point using a local multiplier (~3.5×). Hs is the average of the largest one-third of waves; the actual face you ride is bigger than the height number suggests.
• Wind microclimate — Saladita's reliable daily pattern is offshore E in the morning (5–9 AM), onshore SW by midday (10 AM–2 PM), with evening glass-off potential after 5 PM. The engine grades each hour by that pattern.
• Tide-phase reform behavior — the inside-section sandbar reform connects to the close-out on south swells during a dropping mid-tide; closes out on west-heavy swells; softens on high tides. The engine flags this per hour.
• Board recommendation — face-height range maps to suggested boards (logs in small days; mid-lengths and performance longboards in shoulder-high; shortboards only on the biggest days). Saladita is a longboard wave; recommendations weight accordingly.
• Crowd estimate — calibrated from quality score, day-of-week (weekend bump), and time-of-day (morning peak, midday dip, evening return). Not from data; from local observation.
• Quality verdict (1–10) — a synthesis of swell direction, period, size, and tide. Above 7 is a session worth driving for. Below 4 is a rest-day signal.

The learned layer.

On top of the rules engine, we run a logistic-regression classifier trained on 82 days where local surfers described conditions in their own social-media captions — phrases like "fast and short," "lots of water," "amazing afternoon," "a morning of slow waves." Each caption became a label; each day's marine and wind data became the features. The model learns which combinations of swell direction, period, wind angle, and afternoon chop predict the captions that get written when the wave is good. We surface its output as the "fit to local archetype" badge — a number between 0% and 100% saying how structurally similar a given forecast day is to historical good days at this break. Full methodology →

The energy and percentile layer.

Wave power per unit crest length (kW/m) is computed from significant wave height and period using the deep-water energy-flux formula. The percentile ribbon ranks each day against the full 5+ years of Saladita marine data — a top-2% energy day in May means more than it would in October.

What this forecast isn't.

This isn't a substitute for checking conditions yourself before paddling out. The rules engine reflects what we've observed at the wave over time. The model has modest skill (cross-validated AUC ≈ 0.60) and helps rank days against one another more than it calibrates any single day's probability. Use this as a planning tool, not as gospel. And if you're here, walk down to the point before the first session and watch a set break with your own eyes.