HEALPix in Zarr on S3

This section explains the key storage and access decisions behind Waterpark.

Waterpark combines three ideas: a common HEALPix grid, chunked Zarr datasets, and S3-compatible object storage. Together, these choices make climate and Earth observation data easier to compare, easier to stream, and easier to use in interactive analysis or machine-learning workflows.

We first explain why Waterpark remaps datasets to HEALPix, a hierarchical equal-area grid that provides a common geometry across very different source datasets. We then describe why the data is stored as Zarr pyramids and why this format works especially well when served from an S3 object store.

flowchart TB
        A["Source datasets<br/>ERA5 · ICON · CMIP6 · Observations"]
        B["HEALPix grid<br/>common equal-area geometry"]
        C["Zarr pyramid<br/>chunked multi-resolution arrays"]

    A -->|"remap once"| B
    B -->|"coarsen into levels"| C