[turban]

First step of CLIM-1575: land cover groups per district, computed from the class dataset rather than precomputed at ingest.

• remap_classes(data, groups): merges a categorical raster's classes into named groups, numbered in the order given, and records the groups as the cube's CF flags. OCS core's aggregate_spatial_classes reads its classes from those flags, so it reports one share, total or majority per group, labelled with the group names, unchanged. A remap through the standard apply and array_element keeps the original flags, so the aggregation would still look for the original classes.
• esacci_landcover_groups workflow: OCS core's esacci_landcover_class_yearly in the ten groups of the fractions dataset (the IPCC land categories, with shrubland, sparse vegetation and bare land kept apart).
• README section with a process graph, and the reason the merge goes before the aggregation: majority is otherwise wrong.

The fractions dataset it replaces is removed in #29. Sending a per-district group share to DHIS2 still needs a change in OCS core's export (step 2 of CLIM-1575).

Tests: codes to groups and flags, a group's share equal to the sum of its classes' shares, majority picking the group that covers most when no single class does, malformed groups refused, discovery through the entry point, and the workflow's groups covering every class of the core dataset exactly once.

Run live against the Nepal test instance's 2022 ESA CCI store: the workflow followed by aggregate_spatial_classes gave ten labelled shares per district, adding up to 1, in under two seconds for five districts.

[Copilot]

🟡 Changes recommended

Unbounded lookup allocation can exhaust worker memory, and malformed class codes can silently produce incorrect mappings.

2 open issues

• Prevent memory exhaustion from sparse high-valued class codes · New
• Reject fractional and boolean class codes before conversion · New

What changed in this PR

Adds categorical land-cover grouping before spatial aggregation, allowing OCS core to report group shares and correctly determine the majority group without precomputed fractions.

Changes:

• Adds remap_classes to remap raster codes and replace their CF classification flags.
• Adds an ESA CCI workflow defining ten land-cover groups.
• Documents usage and adds tests for remapping, aggregation, validation, discovery, and class coverage.

File	Description
tests/​test_remap_classes.py	Tests remapping, aggregation, validation, and workflow integration.
README.md	Documents grouping and its placement before aggregation.
ocs_plugins/​workflows/​esacci_landcover_groups.json	Defines the ten-group ESA CCI workflow.
ocs_plugins/​processes/​remap_classes.py	Implements class remapping and replacement classification metadata.
ocs_plugins/​__init__.py	Updates the package docstring to mention workflows.

_{Review effort: Balanced}