Ocean temperature forecasting,
at the edge of what's possible.
An AI-driven forecasting system for short- and long-horizon temperature prediction across the South China Sea and the Yellow Sea. Three sites. Four depths. One-hour to thirty-day horizons. Evaluated on reanalysis hindcasts and cross-dataset out-of-sample tests.
Pick an entry point below — live forecasts, regional case studies, or a paste-your-own-data backtest.
Malaysian Stations
Forecast temperature at PETRONAS offshore fields — Bintulu (D35) and Dulang (Terengganu).
- Surface, 30 m, 50 m, bottom
- Horizons: 1 h, 24 h, 72 h, 30 days
- Hourly HYCOM, 1992–2012
Penglai Offshore Field
Forecast temperature at the Penglai field in the Bohai Sea, with a stronger seasonal signal.
- 4 m and 20 m depths
- Horizons: 72 h, 30 days
- 3-hourly HYCOM, 1995–2015
Backtest Tool
Drop in a measurement column and see all four depth models forecast against it, side by side.
- Click any timestamp to issue a forecast
- M1–M4 compared in one chart
- Per-model RMSE / MAE / MAPE / Max Err
- 1-day · 3-day · 30-day horizons
The approach
Built on a state-of-the-art Transformer architecture — the same family that powers today's most advanced AI systems — purpose-tuned for ocean temperature. Each site and each depth gets its own dedicated model, so forecasts reflect the physics of where it matters. One unified service serves every station and every horizon through a single, station-aware endpoint.
The experience
Pick a site, pick a depth, pick a horizon — get a forecast in seconds. Every prediction is held up against observed measurements where available, so you see not just the forecast but how much to trust it. Built for ocean engineers, planners, and decision-makers who need answers now, not after a batch run.