ADR-0060: Pseudo-BARC256 fixture derivation
Date: 2026-09-09
Status: accepted for test artifacts, 2026-09-09.
Context and decision
The owner requested pseudo-BARC256 maps from the two existing Arizona dNBR
fixtures. Those source samples encode dNBR × 1000. Derive byte samples as
floor(clip((source + 275) / 5, 0, 255)), retaining source grid and finite valid
support. Use an internal mask rather than reserving valid byte values for NoData.
No production model or source fixture changes; no field-validated SBS claim.
Rationale and alternatives
The USGS BAER description provides the inverse scale relation. Explicit clipping prevents byte wraparound; floor defines deterministic quantization. Per-image min/max stretching would lose the common BARC scale. Reserving 0 or 255 would discard valid observations. Four-class severity thresholds and field calibration are outside this request.
Decision provenance
- Venue: repository task conversation, 2026-09-09, America/Los_Angeles.
- Participants: requesting repository owner and Codex.
- Owner: requesting user, authorizing pseudo-BARC256 fixture creation.
- Implementer: Codex; quantization/mask choices are documented engineering choices.
- Change: two new derived test rasters; no previous conversion is replaced.
Evidence, risk and rollback
Fixture derivation and checks record source/output hashes, clipping and readback validation. These derivatives share dNBR ancestry and cannot count as independent observed SBS evidence. Remove the derivatives to roll back; originals remain unchanged.