It started as an ML project a few years ago for my home break. Now a Python pipeline pulls NOAA's GFS-Wave model and samples it at every spot. Each spot's facing and exposure come from the coastline, so swell blocked by land doesn't count. Each hour gets a 0–100 score based on surf height, swell period and wind, and days are rated only on daylight hours. Tides come from NOAA stations, with the FES2022 tide model as a fallback. The web app is Go and SQLite.
The global model is coarse (1°) and doesn't account for local reefs or the seafloor, so sheltered spots can be rated wrong. Days 10+ are trends, not exact numbers.
Honestly not bad, best I’ve seen so far. The kelvin wave callout is nice. Surfline still easier though + cams. What’s your edge / improvement over their forecast?
Hi HN, I built OpenSwells, a free and open-source surf forecast: https://github.com/open-swells/open-swells-app
It started as an ML project a few years ago for my home break. Now a Python pipeline pulls NOAA's GFS-Wave model and samples it at every spot. Each spot's facing and exposure come from the coastline, so swell blocked by land doesn't count. Each hour gets a 0–100 score based on surf height, swell period and wind, and days are rated only on daylight hours. Tides come from NOAA stations, with the FES2022 tide model as a fallback. The web app is Go and SQLite.
The global model is coarse (1°) and doesn't account for local reefs or the seafloor, so sheltered spots can be rated wrong. Days 10+ are trends, not exact numbers.
Now I'm back to experimenting with ML weather models toward a wave model of our own. Looking at FourCast and GraphCast https://deepmind.google/blog/graphcast-ai-model-for-faster-a... https://arxiv.org/abs/2202.11214
Feedback is welcome, especially if your local spot is rated way off.
Honestly not bad, best I’ve seen so far. The kelvin wave callout is nice. Surfline still easier though + cams. What’s your edge / improvement over their forecast?