AI for Ocean Currents & Tsunami Early-Warning
From waves to warnings: building AI pipelines for ocean risk.
Skills you will gain:
About Program:
AI for Ocean Currents & Tsunami Early-Warning is a 3-day, hands-on workshop that blends coastal physics with modern ML to turn ocean data into actionable alerts. You’ll clean and grid buoy/HF-radar/altimetry data, run data assimilation, train compact models (PINNs, FNOs, encoder–decoders) for short-term forecasts, and quantify uncertainty with ensembles and conformal prediction. By the end, you’ll detect tsunami-like anomalies, estimate ETAs with confidence bands, and wire a mini operational pipeline—from ingest and QC to calibrated alerts and a one-page situation report.
Aim: Equip participants to build an end-to-end, physics-aware AI workflow that converts ocean observations into reliable, uncertainty-calibrated forecasts and tsunami early-warning alerts—covering data ingest/QC, data assimilation, ML forecasting (PINNs/FNOs/transformers), and operational alert design.
Program Objectives:
-
Grasp core coastal/tsunami physics (shallow-water, bathymetry).
-
Ingest, QC, align, and grid buoy/HF-radar/altimetry data.
-
Benchmark baselines vs ML (PINNs, FNOs, encoder–decoders/SSMs).
-
Apply lightweight data assimilation and read analysis increments.
-
Produce short-horizon forecasts with calibrated uncertainty.
-
Detect tsunami-like anomalies and estimate ETA with bands.
-
Evaluate skill (CRPS, PR/ROC, cost–loss) and set alert thresholds.
-
Wire an ops pipeline and generate a one-page situation report.
What you will learn?
📅 Day 1 – Introduction to AI for Ocean Currents & Sensing
- Ocean Observations 101: Buoys/DART, HF radar, satellite altimetry (SSH)
- Data Formats & Access: NetCDF/Zarr, coordinates, gridding to coastal domains
- Physical Basics: shallow-water intuition, bathymetry, boundaries
- ML Landscape: baselines (persistence/AR), PINNs vs Neural Operators (FNO)
- Hands-on: Access small buoy/HF radar/altimetry samples; run quick QC (spike/flatline/time); align & grid into a tidy dataset
- Hands-on: Train a tiny PINN on synthetic wave probes; run a minimal FNO for +6 h surface-current forecast vs persistence
📅 Day 2 – Forecasting, Data Assimilation & Uncertainty
- Data Assimilation Essentials: EnKF/3D-Var intuition, increments, observation error
- Sequence Models for Marine Time Series: encoder–decoder transformers, SSMs; masking & missing data
- Uncertainty & Calibration: ensembles, heteroscedastic outputs, conformal prediction; coverage & CRPS
- Hands-on: Assimilate an HF-radar snapshot into yesterday’s state and visualize analysis increments
- Hands-on: Build a compact encoder–decoder for buoy streams (+24 h horizon) and add conformal bands; verify nominal coverage
📅 Day 3 – Tsunami Detection, ETA & Operational Alerts
- Tsunami Detection from DART/Tide Gauges: detrending, anomaly scoring, robustness to noise/clock drift
- ETA Estimation & Communication: uncertainty bands and practical reporting
- Operational Workflow: ingest → QC → assimilation → forecast → UQ → post-process → publish
- Alert Design: precision–recall vs ROC, class imbalance, cost–loss analysis, tiering (Advisory/Watch/Warning)
- Hands-on: Detect tsunami-like anomalies and estimate ETA with uncertainty
- Hands-on: Wire a mini pipeline (configs/CLI) to produce alerts; pick thresholds from PR curves and export a one-page situation report (maps, bands, alert level)
Mentor Profile
Fee Plan
Get an e-Certificate of Participation!
Intended For :
-
Coastal & ocean modelers, met-ocean analysts, and early-warning/DRM teams
-
Data scientists/ML engineers in public agencies, NGOs, or blue-economy startups
-
Graduate/advanced undergraduate students in Ocean/Climate/EE/CS/ME
-
Researchers working with buoy, HF-radar, satellite altimetry, or tide-gauge data
Career Supporting Skills
Program Outcomes
-
QC, align, and grid buoy/HF-radar/altimetry data (NetCDF/Zarr).
-
Train baselines + physics-aware ML (PINNs, FNOs, encoder–decoders).
-
Assimilate observations (EnKF intuition) and visualize increments.
-
Deliver +6 h surface-current & +24 h buoy forecasts with calibrated uncertainty.
-
Detect tsunami-like anomalies and estimate ETA with confidence bands.
-
Evaluate skill (CRPS, PR/ROC, cost–loss) and set tiered alert thresholds.
-
Ship a minimal ops pipeline (ingest→QC→DA→forecast→UQ→publish) and a one-page situation report.