Examples
Each example opens a large ocean dataset as a virtual Icechunk store and answers one question with it in a few seconds or minutes. You download nothing in advance, even when the dataset is hundreds of terabytes or thousands of files. Use them to see how a store opens, how much a read costs, and what kind of analysis becomes easy. Each one has an Open in Colab button.
| Dataset | What it shows | Read time | |
|---|---|---|---|
| 1 | HYCOM reanalysis, 306 TB | Bering Sea bottom temperature on the snow crab grounds | seconds |
| 2 | PacIOOS ROMS Hawaiʻi reanalysis | How deep the seasons reach; water masses in T–S | ~15 s |
| 3 | NCEI ocean acidification indicators | The aragonite saturation horizon | seconds |
| 4 | NOAA OISST, ~16,500 daily files | Daily world SST, one line per year since 1981 | ~40 s |
| 5 | NOAA OISST, every daily file | Marine heatwave days in the Gulf of Maine | ~3 min |
| 6 | GOBAI HR oxygen, weekly 1993–2025 | The eastern tropical Pacific oxygen minimum zone | seconds |
Times were measured on the workshop’s JupyterHub; elsewhere they can be longer.
Try your own questions
The examples are starting points. Each store holds far more than one notebook uses. Here are some questions to try:
- HYCOM: how does the Bering Sea cold pool’s extent change from year to year? How do currents through the Aleutian passes vary?
- PacIOOS: how big is the tide at Honolulu? Can you see the afternoon surface warming in calm summer weeks?
- OA indicators: where are pH and aragonite saturation lowest at the surface? Which regions have the fewest observations behind the map?
- OISST: in which year was your favourite stretch of coast warmest? How does the Niño 3.4 index line up with the world SST chart?
- GOBAI: is the oxygen minimum zone of the Arabian Sea or off Peru changing too? Where does nitrate track oxygen?
Ask your agent
You do not need to write the code. Give your agent the store’s URL and a question, and ask it to tell you what it will read before it reads it.
| Store | URL | README |
|---|---|---|
| HYCOM GOFS 3.1 reanalysis | https://data.source.coop/ocean-icechunks/hycom/hycom-gofs-3pt1-reanalysis |
README |
| PacIOOS ROMS Hawaiʻi reanalysis | https://data.source.coop/ocean-icechunks/pacioos/roms_hiig_reanalysis/icechunk |
README |
| NCEI OA indicators | https://data.source.coop/ocean-icechunks/oa-indicators/climatology |
README |
| NOAA OISST v2.1 | https://data.source.coop/ocean-icechunks/noaa-oisst/oisst.icechunk |
README |
| GOBAI HR | gs://noaa-oar-gobai/icechunk/v202606 |
page |
A sample prompt:
Open the virtual Icechunk store at
https://data.source.coop/ocean-icechunks/noaa-oisst/oisst.icechunkanonymously. Its data are virtual references to NOAA’s files, so authorize anonymous reads of the store’s virtual chunk containers. List its groups and variables, and explain how the arrays are chunked.Then answer this question: in which year was the sea surface off Cape Cod warmest, and on which day? Before you read any data, tell me how many chunks and roughly how many bytes your plan reads. If there is a cheaper way, say so. Then write a short notebook with one figure, and run it.
Two parts of that prompt do most of the work:
- “Tell me what you will read first.” A careless read of a big store can pull gigabytes; a good plan for the same question might pull megabytes.
- “Explain the chunking.” How a store is chunked decides which questions are cheap. A map at one time step and a long time series at one point can differ in cost by a thousand times.
If the agent gets stuck opening a store, give it the store’s README link from the table above: each README has working code to open that store.