0007 Visualization and result statistics go through pyramids¶

Status¶

Superseded by 0018

Date¶

2026-09-28

Context¶

moraine is made for data larger than memory. zarr reads whole chunks, so strided sampling of a large array still reads every chunk. moraine already has pyramids (ras-pyramid, pc-pyramid) and plots that read the level matching the view (ras_plot, pc_plot, holoviews).

Decision¶

moraine info shows metadata only for plain arrays (no data read).
For pyramids it adds statistics from the finest level of at most 64 MiB (a regular decimation of the scene) and warnings for all-nan, infinite and constant values; point cloud pyramids skip empty cells.
moraine quicklook only accepts pyramids and draws them with ras_plot / pc_plot, saved with the holoviews matplotlib backend (no browser needed).
pc_plot builds its rtree lazily so overviews do not read all coordinates.

Consequences¶

Results worth checking get a pyramid step in the pipelines.
Checking a result costs the same for any data size.

Do not¶

Do not write custom sampling, statistics over full arrays or custom plotting code for checks.
Do not read whole arrays to inspect results.