0004 — RW and SF are read in their HDF5 edition

Status: accepted (2026.10.1)

Context

Up to 2026.10.0 RW and SF were read in DWD's legacy binary edition (.bz2, a national 900x900 spherical grid), with a reader extracted from wradlib (radar/radolan.py, radar/georef.py). DWD announced that the binary edition would be switched off; the same analyses are published in an HDF5 edition (raa01-<p>_10000-<yymmddHHMM>-dwd---bin.hdf5), ODIM_H5 on the DE1200 grid that RS, RV and HymecNG already use.

Decision

Read RW and SF in the HDF5 edition, through the same read_odim_composite_cell and grid lookup as the RADVOR products. The binary reader was removed with its tests and fixtures; git history has them if the binary edition is ever needed again.

Consequences

The two editions are the same analysis on different grids, not a cell-for-cell copy. Checked on live files (2026-10-12): area means over ~1100 locations agree to 1 % (SF 0.717 vs 0.725 mm), while a location's own value differs in ~65-80 % of wet cases (median 0.1 mm, up to a few mm in sharp gradients), and is found in the old cell's 3x3 neighbourhood only ~60 % of the time. Users see small value changes at the switch; nothing else changes.

Timing: RW's HDF5 file appears ~25.7 min after its window ends, like the binary one; SF's ~29.3 min, ~2.5 min later than its binary one, hence SF's RELEASE_DELAY of 31 min.