Tide Prediction Methodology
What is computed, which rules are applied, and what this batch leaves out.
Where the predictions come from
Tide times and heights are NOAA CO-OPS tide predictions for reference stations, datum MLLW, heights in feet, times in local standard or daylight time as NOAA returns them. The station list and the prediction files are retrieved when the site is built and then cached. A later build reuses the cache for the same dates.
NOAA's wording on reuse is kept in the project sources: the information on government servers is in the public domain unless annotated otherwise, and it may be used freely. It must not be modified and then presented as official government material. These pages are not official NOAA pages.
The metadata field NOAA calls tideType was empty for the stations in the retrieved list. The tide-type words on a station page are a TideHour rule, not a NOAA label.
What is computed
For each published station the build keeps 395 days of high and low predictions starting on the build date, and 30 days of the six-minute series. The station summary, the daylight-minus count, the king tides, and the spring-tide lag use the first 365 days. The 30-day table uses the first 30 days.
A daily rise is the higher high minus the lower low on dates that have both. The average daily rise is the mean of those daily rises. Days missing a high or a low are left out of that mean.
Sunrise and sunset are computed on the station's coordinates with the astral library and the IANA time zone that matches NOAA's time-zone offset. They are not a NOAA product. Moon phase names are computed the same way and appear only as a column and in the spring-tide comparison.
Tide-type rule
A date counts as a one-high date when the predictions list one high water or none. If at least 40 percent of the dates are one-high dates, the label is mostly diurnal. Otherwise, if at least 75 percent of the dates have two high waters and the median gap between the two crests is at most 30 percent of the taller crest, the label is semidiurnal. Every remaining station is labeled mixed.
Those three cutoffs are TideHour rules for this batch. They are not a percentile of all 1,171 reference stations, because this batch does not hold that full set. They are not a harmonic form factor.
Daylight minus tides and king tides
A daylight minus tide is a predicted low water strictly below 0.0 ft MLLW that occurs at or after sunrise and at or before sunset, in the station's local time. A low at exactly 0.0 ft does not count. A low before sunrise or after sunset does not count. The count on each page is the number of dates, not the number of lows.
A king tide is not a NOAA term. On this site it means one of the three highest predicted high waters in the prediction window at that station. The page does not say that water will flood a road or a building.
The spring lag is the median number of days from a new moon or a full moon to the highest high water in the four days that follow that moon. Neap is not given a separate index.
What this batch leaves out
Subordinate stations do not get pages. Nearby comparisons use other stations on TideHour so far. A state percentile against every NOAA station in the state is not stated, because the batch does not include every station. There are no pages for a single date, a single month, a ZIP code, sunrise by itself, or fishing times.
If two published stations within 5 km have a median high-water time gap of 10 minutes or less, a median height gap of 0.2 ft or less, and the same tide-type label, only one of them stays in the sitemap. The other keeps its URL and points its canonical link at the page that remains.
Station pages whose wording is too close, after place names and numbers are removed, are marked noindex and left out of the sitemap. The thresholds are a median cosine under 0.90, a maximum cosine under 0.95, and a maximum 5-word Jaccard under 0.50.
Automation
The tables, the counts, and the sentences are produced by the build. A sentence is chosen by the bin the numbers fall in. The bins and the sentence text are in the repository. The build does not ask a language model to write a harbor description.
Prediction files record the time they were retrieved. The privacy page has its own review date. Those dates are not the same field.
Stations on TideHour so far
Three published stations, linked from this page: