Measurements and methods

Observations

Stations report hourly, many of them more often. Not every station carries every sensor: each link below opens the chart at a station that reports the measurement.

Directional spectra

Buoys that measure direction report, for each frequency band, the spectral density and four coefficients describing how that energy spreads over direction. NDBC publishes them in five files.

Spectral density .data_spec m²/Hz

Energy per unit frequency in each band. Each record also gives NDBC's separation frequency, in hertz, between swell and wind sea. Where it is left blank, as it is at the buoys CDIP runs, it is found from NDBC's swell and wind-sea heights for the hour: the frequency below which the spectrum holds the swell's share of the energy.

α₁, mean direction .swdir °T, from

The mean direction the waves in each band come from.

α₂, principal direction .swdir2 °T, from

The principal direction of each band.

r₁ .swr1 0 to 1

The first normalised polar Fourier coefficient: near 1 when a band's energy comes from a single direction.

r₂ .swr2 0 to 1

The second normalised polar Fourier coefficient.

Band layouts depend on the buoy's payload. Bluegraph currently sees three: 46 bands from 0.033 to 0.485 Hz, 64 bands from 0.025 to 0.58 Hz, and 98 bands from 0.025 to 0.96 Hz.

From the four coefficients NDBC reconstructs each band's distribution over direction as

D(f, θ) = (1/π) [ ½ + r₁ cos(θ − α₁) + r₂ cos 2(θ − α₂) ]

which can dip below zero and blurs separate peaks together. Bluegraph estimates the distribution with the maximum entropy method instead (Lygre and Krogstad, 1986): it reproduces the same four coefficients with a distribution that stays positive and resolves peaks more sharply.

The sea on each station page is simulated from the measured spectrum: many simple waves added together, each with the energy, period and direction the buoy measured, their wavelengths set by the water depth. Only their starting positions are chosen at random, so the heights, periods and directions are real but where each crest falls is not.

Wave Intelligence

Wave Intelligence analyses the stored spectra of the last few days at each station, in these steps.

Wave systems

The estimated spectrum, on 36 directions of 10°, is smoothed and split into systems by following each cell uphill to its peak. Two systems merge when the valley between them is at least 0.6 of the lower peak, or when their peaks lie within 0.02 Hz and 40°; systems holding under 5% of the energy, or under 0.15 m, are absorbed by a neighbour. At most five are kept, and systems of wind sea are combined into one, since one wind raises one sea (Hanson and Phillips, 2001).

Swell or wind sea

A system is wind sea when the wind, raised to 10 m, outruns its waves: 1.5 · U₁₀ · cos Δθ exceeds the phase speed at the system's peak, Δθ being the angle between wind and waves. Otherwise it is swell, and groundswell at 13 s or longer. Without a wind reading, systems above NDBC's separation frequency are wind sea.

Tracks

Systems are followed from spectrum to spectrum by period and direction, across gaps of up to 3 hours.

Storm sources

Swell disperses as it travels: waves of frequency f move at g/(4πf), so a swell from a storm a distance D away arrives with its frequency rising at g/(4πD) (Munk and others, 1963). A line fitted to that rise gives the distance and the time the waves left; the storm lies along the direction the first of them came from. A source is shown only when the fit is good (R² at least 0.7, its slope at least four standard errors), the distance is 300 to 7,000 nautical miles with a 95% range no wider than a factor of two, the storm lies within 65° of the equator, and it and the great-circle path to it, beyond the first 100 nautical miles, are at sea. Positions are good to a few hundred miles.

Storms placed by several buoys

One storm's swell reaches many buoys, and each places it on its own. Placements are kept once two analyses have found them, and grouped each half hour over the last month: a set of buoys is one storm when a single place and a single departure time explain each buoy's distance, bearing and departure within 2.5 of its own standard deviations (the distance's from its 95% range, the bearing's taken as 12°, the departure's the distance's share of the swell's travel time, at least 3 hours), and when their own placements lie within a fifth of their distance from it, or 300 nautical miles. The place is the point that best fits every buoy's distance and bearing together; the departure, their weighted mean. Each buoy counts once in a storm. A storm's swell counts as still arriving while one of its buoys has measured it in the last 12 hours; the Storms page lists those storms first. A storm's replay runs from three hours before its waves left to six hours after its swell was last measured. On a storm's page, swell fronts are drawn at the deep-water group speed of their period along great circles from the storm, and stop where land would stop them; they show where the swell could have reached, not measurements between the buoys.

Crossing seas

Flagged when the two largest systems together make a sea of at least 1.5 m, the smaller stands at least 0.5 m and holds at least 30% of the larger's energy, and they meet at 40° or more: the combination linked with steeper, less regular seas than the height alone suggests (Onorato and others, 2006).

Crest length

A system's directional spread is √(2(1 − r)) radians, r being the mean resultant length of its energy over direction (Kuik and others, 1988). Under 25° its crests are long, as in groundswell; over 40°, short and confused; between, moderate.

Surface currents

Near many coasts a station also carries the surface current: how fast the top metre or two of the sea is moving, and where to. It is measured not at the station but by the national network of high-frequency (HF) radars on the shore, which the U.S. Integrated Ocean Observing System coordinates and NDBC serves as hourly grids. The network reaches the U.S. West, East and Gulf coasts, the Gulf of Alaska, Hawaii, Puerto Rico and the Virgin Islands, and parts of the Great Lakes.

How a radar measures

A radar on the shore transmits at 4 to 50 MHz and hears its strongest echo from the ocean waves half its radio wavelength long (Bragg scattering; Crombie, 1955). Those waves travel at a speed their length fixes; whatever the echo's Doppler shift adds to that is the water carrying them, along the radar's line of sight (Barrick and others, 1977). One radar measures only that part of the current, toward or away from it, averaged over the top half metre to two metres or so of the sea, deeper at lower frequencies (Stewart and Joy, 1974).

Combining radars

Where two or more radars see the same patch of sea, their line-of-sight speeds are combined into the one current that best explains them: a total vector each hour, on grids of 6 km cells and, nearer shore, 2 km cells. (Grids of 1 km and 500 m cover a few bays; Bluegraph does not use them yet.) Where the lines of sight cross at a shallow angle, small errors in each become a large error across them; the vector's horizontal dilution of precision (HDOP) measures this. Bluegraph keeps only vectors from two or more radars with an HDOP of 2 or less.

A station's cell

Each station within coverage is given one cell: in the finest grid that covers it, the nearest cell with a good vector in at least half of the last three days' hours, within 2.5 nautical miles on the 2 km grid or 4 on the 6 km grid. The daily audit chooses again, as radars go down and come back. The chart names the cell and its distance from the station, usually one to two and a half nautical miles.

Timing

Readings are hourly. Each hour's grid fills in over three or four hours as the radars' data arrives, so Bluegraph reads the last eight hours each hour, and the newest reading is a few hours behind the station's others. The radar server keeps about 92 days, which filled each station's record when it was first covered. The station page shows the reading nearest the time chosen, up to six hours off, and says when it was.

Direction

A current is named by where the water goes, in degrees true; wind and waves are named by where they come from. A current flowing toward where the swell comes from runs against it: the waves shorten and steepen, the more so the stronger the current (Longuet-Higgins and Stewart, 1961).

What it is not

It is the current in a cell a mile or two from the station, averaged over the cell and the hour, not at the buoy itself. It is the surface layer, driven by the wind as well as the tide, and the water below can move differently. Near shore and in bays the tidal current dominates, turning with the tide.

On the 3D sea

The current is drawn as chevrons drifting with the water, pointing where it goes. They move faster than the water, by the factor the legend gives (about 100 times on a kilometre-wide patch), so that a given current crosses the screen at the same pace at every station; a faster current draws more, larger and brighter chevrons, with longer trails. Their placement is illustrative; the speed and direction are the measurement.

Sea level along the coast

A tide gauge's level, less the tide predicted for it, is the residual: what the weather and the sea add to the tide. Over hours it is mostly wind and air pressure, a storm's surge among them. Over weeks it is the slow rise and fall of the coastal sea itself, higher and warmer than usual or lower and cooler, and the coastally trapped waves that carry such swings along a coast.

The sea level page follows that slow part along two coasts, at their open-coast gauges: the Pacific from San Diego to Kodiak (NOAA's in the United States and the Canadian Hydrographic Service's in British Columbia), and the Gulf and Atlantic from Eastport, Maine, round Florida to Port Isabel, Texas. Gauges up rivers and inside long bays are left out: river flow and the basins' own setup swamp it there. Each coast also has its islands' gauges, one an island: Hawaii, Midway, Wake, Kwajalein, Guam and American Samoa on the Pacific; Puerto Rico, the Virgin Islands and Bermuda on the Atlantic. The Pacific also has five gauges from the equator to Central America, where a Kelvin wave reaches the Americas first: La Libertad in Ecuador, Santa Cruz in the Galápagos, Cocos Island, Quepos and Acajutla. They are the University of Hawaii Sea Level Center's fast delivery data, daily means it has already filtered the tides from, released a month or two after they were measured; each day is set against the gauge's average over the past year, as the others are. For each gauge and each UTC day, the observed level and the prediction are each averaged by the hour, matched hour by hour, and averaged over the day; a day needs 18 matched hours. The daily residual is averaged over a week centred on the day (the last three days lean on fewer days), and the gauge's own average over the past year is taken away. That last step matters: the predictions are made against a datum two decades old, so the raw residual also carries the sea-level rise since.

On the globe a ribbon lies on the sea along the coast. The coast is each gauge's average, and the ribbon reaches out to sea as far as the level stands from it, warm where higher and blue where lower: a width drawn on the map, not a height, so it stays on the coast however the globe is tilted. At the whole-coast view 25 cm reaches about 350 km out; closer in, the ribbon narrows with the zoom. Each gauge's reading is a tick ending in a dot; between gauges the ribbon is interpolated by the distance along the coast, and it is not drawn across a gap of more than 600 km. The path follows the shore round its capes, so distances along it are close to the coast's. An island's gauge is a disc on the sea round it, as wide as the ribbon would reach.

The timeline also shows the last ten days hour by hour: each hour's level less its prediction, with nothing else taken away, which is how a storm's surge is measured and what a week's mean would smooth away. There the ribbon reaches about 700 km out for a metre, and stops growing at 1.2 m.

On the Pacific, how fast the swings travel is read from the gauges themselves, along the U.S. West Coast, where they are clearest; north of Washington the gauges' swings are mostly the weather's. Each gauge's daily residual is band-passed, a three-day mean less a 31-day mean, which keeps swings lasting from a few days to a few weeks. Each gauge more than 300 km up the coast is compared with San Diego at delays of up to 15 days, over the last five months; the delay at which the two agree best is how much later it saw the same swings. The speed is the distance along the coast over the delay, fitted through the gauges that agree well (a correlation of 0.3 or more). Distances are measured gauge to gauge in straight lines, which run a little short round headlands, so the speed is approximate.

Offshore, faintly, the page draws the satellites' sea surface height from altimetry (NOAA CoastWatch's blended near-real-time product, every third day, at 1°), averaged over about 300 km to smooth away eddies: over the Pacific from the western Pacific to the Americas and 15°S to 62°N, and over the Gulf, the Caribbean and the western Atlantic from 7°N to 46°N. The product carries values over land too, which the page leaves out by the coastlines the globe draws. Against the satellites' own long-term mean nearly everything stands high, because the sea has risen since, so each place is set against its own average over the past year, as the gauges are (read from every third day of the six months shown and every ninth of the months before). Unlike the gauges' residual, it still carries the seasons' rise and fall; it is there to show where along the equator and the coasts the sea stands high.

Below the Pacific, a longer record: San Diego's monthly mean sea level from NOAA, back to 1906, each month set against the usual for its time of year. The long-term rise, a straight line fitted to 1950 through last year (about 2.4 mm a year), is taken away, and then each calendar month's average departure from it over the same years, the seasons. This season, June into the following spring, is drawn beside 1982–83, 1997–98 and 2015–16, the strongest El Niños of the past half-century. NOAA verifies a month some weeks after it ends; each complete month since is averaged from its preliminary six-minute levels, needing nine in ten of them, and drawn hollow.

The page shows these swings as they were measured. It does not separate a Kelvin wave from the wind's and the pressure's share, and it does not forecast.

Charts and data

• Times are UTC throughout. Station pages also give the station's local time.
• Charts of up to 7 days show every reading. Longer ranges show the last reading in each interval: 20 minutes over 2 weeks, 30 minutes over 30 days, an hour over 3 months, 2 hours over 6 months and 3 hours over a year, so a peak between readings can be missed. The CSV and JSON downloads on each station page hold every reading.
• Readings NDBC marks missing are gaps in the charts, never zeros.
• Bluegraph reads NDBC's realtime files every 30 minutes and keeps what it reads: observations since April 2025, directional spectra since August 2026.
• Realtime data are provisional: NDBC has not yet quality-controlled them. NDBC's historical archive holds the quality-controlled record.

Data files

The readings charted on the station pages are published as Parquet files: for each table, a file for each of the last five months and one for the months before, sorted by station and time, with each column's unit in its name. The manifest lists every file with its rows and time span. Query them here, in your browser, on the Query page; download them; or read them where they are with any tool that reads Parquet over HTTP. A query fetches only the parts of a file it needs.

SELECT station_id, max(wvht_m) AS highest_m
FROM 'https://bluegraph.io/data/files/observations/2026-09-….parquet'
GROUP BY station_id
ORDER BY highest_m DESC
LIMIT 10;

• The files are rewritten every six hours; a month's file changes name when its contents do, so take its current name from the manifest.
• Readings are as the sources reported them, and provisional, as on the station pages. Readings older than 90 days are thinned to about one an hour.
• Each station's owner and the terms its data are shared under are in the stations table; the terms say more.

Units

Values are stored as NDBC reports them and converted for display.