OceanTACO

OceanTACO is a multi-source sea surface variable dataset with cloud-native dataloaders for machine learning workflows.

---

OceanTACO provides co-located observations of sea surface height (SSH), sea surface temperature (SST), sea surface salinity (SSS), ocean currents, wind, and Argo float profiles — organized as regional NetCDF tiles and hosted on HuggingFace.

Key features:

• Quickstart workflow: install, query a regional tile, and plot your first map in Getting Started

• Data access workflow: browse remote files, subset by region/date, and work cloud-first via Dataset Workflows

• Machine learning workflow: build training/evaluation query sets and use OceanTACODataset with PyTorch in Dataset ML Loader

• Data generation workflow: reproduce formatting, tiling, and statistics with the Dataset Generation Pipeline

• Hands-on tutorials: run complete examples for mapping, coupling, and case studies in Tutorials

Contents

• Getting Started
  • Prerequisites
  • Installation
  • Choose the Right Workflow
  • Quick Start: Retrieval (Direct xarray Access)
  • Quick Start: ML Loader (OceanTACODataset)
  • Next Steps
• OceanTACO Dataset Workflows
  • 1) Install Profiles
  • 2) Main Retrieval APIs
  • 3) End-to-End Retrieval Example
  • 4) Retrieval Pattern: One Exact Tile
  • 5) Retrieval Pattern: Stream Records
  • 6) Retrieval Pattern: Local Full Snapshot
  • 7) Retrieval Pattern: Multiple Sources, Same Region, Date Range
  • 8) Choosing Retrieval Mode
  • 9) Next Step: ML Loader Workflows
• OceanTACO ML Loader Workflows
  • 1) Why OceanTACODataset for ML
  • 2) Local and Remote Usage
  • 3) End-to-End Minimal ML Example
  • 4) Query Design: Training vs Evaluation
  • 5) Patch Size Guidance
  • 6) Query Set Persistence
  • 7) Practical Checklist
  • 8) Related Pages
• Dataset Description
  • Dataset Versions and Coverage
  • Processing Levels and Sensor Semantics
  • Data Sources
  • Variables
  • Ocean Regions
  • Spatial and Temporal Indexing
  • Data Format
  • Processing Workflow and Known Limitations
  • HuggingFace Access
  • Licenses
• Dataset Generation Pipeline
  • Prerequisites
  • Directory Flow
  • Quick End-to-End Commands
  • Step 1: Download Raw Data (download.py)
  • Step 2: Format and Regionalize Data (format.py)
  • Step 3: Build TACO (build_taco.py)
  • Recommended Workflow Patterns
  • Troubleshooting
• Tutorials
  • OceanTACO Data Retrieval Workflows
  • OceanTACO Dataset & Query API
  • Spatio-Temporal Query Generation Intuition
  • Hurricane Milton Wind and SSH Figure
  • SSH Cross-Product Comparison: Hurricane Milton
• API Reference
  • Dataset
  • Queries
  • Remote I/O