AI for Ocean Currents & Tsunami Early-Warning

Course Description

AI for Ocean Currents & Tsunami Early-Warning is a three-week, hands-on workshop that bridges coastal physics with modern machine learning techniques to turn ocean data into actionable early-warning alerts. Participants will work with buoy, HF-radar, and satellite altimetry data, applying data quality control, assimilation, and forecasting models such as PINNs, FNOs, and encoder–decoders. By the end of the course, participants will be able to detect tsunami-like anomalies, estimate ETAs with uncertainty bands, and create a mini operational pipeline for ingesting, processing, and generating calibrated alerts. This course will prepare participants to handle real-time ocean observation data for effective tsunami prediction and response.

Aim

Equip participants to design an end-to-end, physics-aware AI workflow that converts ocean observations into reliable, uncertainty-calibrated forecasts and tsunami early-warning alerts. The course covers everything from data ingestion and quality control (QC) to machine learning forecasting (PINNs, FNOs, transformers), with an emphasis on operational alert design and uncertainty quantification.

Course Objectives

By the end of the course, participants will be able to:

• Understand core coastal and tsunami physics, including shallow-water dynamics and bathymetry.

• Ingest, QC, and align ocean observation data from sources like buoys, HF-radar, and satellite altimetry.

• Benchmark traditional baselines (persistence, AR) against machine learning models like PINNs and FNOs.

• Apply lightweight data assimilation methods (e.g., EnKF, 3D-Var) to improve forecasts.

• Produce short-horizon forecasts with calibrated uncertainty (P10/P50/P90 outputs).

• Detect tsunami-like anomalies and estimate ETA with uncertainty bands.

• Evaluate model performance using metrics like CRPS, PR/ROC, and cost–loss analysis.

• Build and operationalize a mini pipeline for tsunami forecasting and alert generation, complete with a one-page situation report.

Course Structure

Day 1 – Introduction to Ocean Currents & Sensing with AI

• Ocean Observations 101: Buoys/DART, HF radar, satellite altimetry (SSH)

• Data Formats & Access: NetCDF/Zarr, coordinates, and gridding for coastal domains

• Physical Basics: Shallow-water dynamics, bathymetry effects, and boundary conditions

• ML Landscape: Baselines (e.g., persistence, AR), PINNs vs FNOs (Neural Operators)

• Hands-on: Access and QC small buoy, HF radar, and altimetry samples; clean, align, and grid data

• Hands-on: Train a tiny PINN on synthetic wave probes and run a minimal FNO for 6-hour surface-current forecasts vs persistence

Day 2 – Forecasting, Data Assimilation & Uncertainty Management

• Data Assimilation Essentials: EnKF/3D-Var basics, increments, observation error handling

• Sequence Models for Marine Time Series: Encoder–decoder transformers, SSMs; handling missing data and masking

• Uncertainty & Calibration: Ensembles, heteroscedastic outputs, and conformal prediction for uncertainty quantification

• Hands-on: Assimilate an HF-radar snapshot into a previous state and visualize the assimilation increments

• Hands-on: Build an encoder–decoder model for buoy data (24-hour horizon) and add conformal bands; verify coverage and uncertainty

Day 3 – Tsunami Detection, ETA Estimation & Operational Alerts

• Tsunami Detection from DART/Tide Gauges: Detrending, anomaly scoring, robustness to noise/clock drift

• ETA Estimation & Reporting: Estimate tsunami ETA with uncertainty bands and communicate the results

• Operational Workflow: Ingest → QC → assimilation → forecast → uncertainty quantification → post-process → publish

• Alert Design: Precision–recall, ROC, cost–loss analysis, alert tiering (Advisory, Watch, Warning)

• Hands-on: Detect tsunami-like anomalies, estimate ETA with uncertainty, and generate a one-page situation report

• Hands-on: Wire a mini pipeline (configurations/CLI) to produce alerts and set thresholds using PR curves

Who Should Enrol

• Coastal & ocean modelers, met-ocean analysts, and early-warning/disaster risk management (DRM) teams

• Data scientists/ML engineers working in public agencies, NGOs, or blue-economy startups

• Graduate and advanced undergraduate students in Oceanography, Climate Science, Electrical Engineering (EE), Computer Science (CS), and Mechanical Engineering (ME)

• Researchers working with ocean observation data from buoys, HF-radar, satellite altimetry, or tide gauges