Marine Forecast does not reliably return data

[mwood77]

Hi all, I'm trying to create a dynamic marine forecast display, but I'm encountering some difficulties with Open Meteo's Marine Forecast API. The issues seem to stem from the automatic model selection.

For example, let's build a request around a know working source, Rotterdam.

1. Initially I won't specify the model:
   https://marine-api.open-meteo.com/v1/marine?latitude=51.9225&longitude=4.4792&hourly=swell_wave_height,swell_wave_direction,swell_wave_period&timezone=auto&forecast_days=1
   • This generates a 200 with an hourly response filled with null values
2. Let's now specify the model as &models=best_match: https://marine-api.open-meteo.com/v1/marine?latitude=51.9225&longitude=4.4792&hourly=swell_wave_height,swell_wave_direction,swell_wave_period&timezone=auto&forecast_days=1&models=best_match
   • This also generates a 200 with an hourly response filled with null values
3. Lastly, let's specify the model as &models=ncep_gfswave025: https://marine-api.open-meteo.com/v1/marine?latitude=51.9225&longitude=4.4792&hourly=swell_wave_height,swell_wave_direction,swell_wave_period&models=ncep_gfswave025&timezone=auto&forecast_days=1
   • Now we get a 200 and data!

It seems like there is some inconsistency with how the models are being scanned. Generally if data isn't available, I expect to see an error response.

For example:

• https://marine-api.open-meteo.com/v1/marine?latitude=51.9225&longitude=4.4792&hourly=swell_wave_height,swell_wave_direction,swell_wave_period&models=ewam&timezone=auto&forecast_days=1
  • this returns a 400 (seems like this should actually be a 5xx code) with an appropriate message: { "error": true, "reason": "No data is available for this location" }

Can you please provide any insight? Thanks!

[devendra1497]

It seems like the automatic model selection isn’t effectively resolving to a model with available data, at least in this location (Rotterdam). Ideally, either:

best_match should fall back to the nearest valid dataset, or

The API should return a proper 4xx/5xx error (like it does with models=ewam) instead of a successful response with unusable data.
Manually specifying ncep_gfswave025 works reliably for now, but this behavior might lead to confusion for other users relying on auto-selection.

Would be great to hear from the maintainers if this is expected behavior or something that can be improved.

@mwood77

[mwood77]

I worked around it by querying all available models for a given lat/lon and checking for data validity. Feels excessive, but it is what it is...at least until &models=best_match is fixed.

[patrick-zippenfenig]

Hi,
it seems the location is too far from the coast to retrieve a valid data point. The search radius is limited, so locations inland won't return wave data. I'll revisit the search radius next week, but adjusting it isn't straightforward.

[devendra1497]

Patrick — that makes sense. I figured it might be related to inland distance and the radius constraint, but it’s helpful to have that confirmed.

Expanding the search radius (even slightly) could definitely improve best_match behavior in borderline coastal locations like Rotterdam, though I understand the complexity involved. Until then, the workaround shared by @mwood77 (querying all models and checking validity) seems like a solid interim solution — albeit not ideal from an efficiency standpoint.

Looking forward to any future improvements on this — and happy to test if needed.