Why Wave Period Comparisons Are Problematic for Three of the Four Models

Published 21-Oct-2026

Wave-period skill looked suspiciously poor for 3 of the 4 models (MeteoFrance Wave, DWD GWAM, ECMWF WAM 0.25) at several buoys, while GFS Wave looked fine everywhere. The honest conclusion, after digging in, is that there is no direct, consistent link between what those three models report as "wave period" and anything NDBC buoys actually measure -- the comparison is structurally approximate for them, not a simple pairing mistake to fix once and be done with.

What NDBC gives us, and what Open-Meteo says it gives us

NDBC buoys report two different period statistics: APD (average wave period, a simple mean across the whole measured spectrum) and DPD (dominant/peak period, the period at the spectrum's peak energy, reported only to the nearest whole second). Open-Meteo's own documentation describes its wave_period field identically for every model -- "period... of mean, wind and swell waves" -- implying a mean period, comparable to APD, across the board.

That description doesn't hold up model-by-model.

GFS Wave: consistent, already handled

GFS Wave's wave_period behaves like NDBC's DPD everywhere we checked, despite Open-Meteo's generic "mean" label -- likely because NOAA's native WaveWatch III output field underneath it is itself peak-period-like. This one has a clean fix: GFS Wave is compared against DPD instead of APD (see the buoy-type-skill article's sibling fix for period).

The other three models: no single answer

MeteoFrance Wave, DWD GWAM, and ECMWF WAM 0.25 are all WAM-family models, and their wave_period output doesn't consistently match either NDBC column. Testing both pairings at every active buoy (14-day skill, all 3 models) turns up a clear geographic split -- but a split, not a single rule:

Sea-state regime	Example buoys	Better NDBC match
Enclosed / wind-sea-dominated (Gulf of Mexico, Caribbean, nearshore Atlantic shelf)	41013, 42035, 42036, 42058, 42084, 42092, 44008	APD (46-100% within 1.5s)
Open-ocean / swell-dominated (US West Coast, open Atlantic/Pacific)	41004, 41040, 46011, 46028, 46054, 46059, 46069, 46215, 46259, 51000	DPD (30-58% within 1.5s, vs. 2-28% against APD)
Small Pacific islands (ambiguous)	American Samoa, Palau, Pohnpei	Splits by model -- no clear winner

The likely mechanism: these three models probably report a spectral-moment mean period. In a narrow, swell-dominated spectrum that statistic numerically converges toward the peak period (so it reads like DPD); in a broader, locally wind-driven spectrum it stays close to a simple average (so it reads like APD). There's no way to know, from the outside, which regime a given forecast is really in at the moment it's issued -- we're inferring the right comparison from each buoy's typical conditions, not from anything the model or Open-Meteo tells us directly.

What we did about it

Rather than pick one NDBC column and accept a wrong answer at half the roster, the 10 open-ocean/swell-dominated buoys above are compared against DPD for these three models; the rest default to APD. The 3 ambiguous small-island buoys were deliberately left on the APD default rather than guessed, since the data itself didn't point either way.

One more wrinkle worth knowing: NDBC's DPD is quantized to whole seconds (as few as 15 distinct values over a 2-week window, versus 50+ for APD), so any DPD-based comparison carries more built-in noise than an APD-based one -- on top of, not instead of, the pairing uncertainty above.

The takeaway