ADR-0066: Staley scalar estimates on valid support
Date: 2026-09-11
Status and provenance
Accepted direction, 2026-09-11; numerical implementation pending. Venue: repository owner/Codex conversation, America/Los_Angeles; exact clock time not recorded. Participants: repository owner and Codex. Decision owner: repository owner. Implementer: Codex (documentation and subsequent authorized work).
Decision
Compute scalar catchment estimates over non-masked usable data and communicate valid coverage. Publish the exact valid-support GeoTIFF through normal project files and a download link. Defer mapped reports and the GL dashboard. Missing observations are excluded, never assigned zero. Zero usable support remains unavailable. Explicit numerical aggregation/schema details are ratified before scientific implementation; the current increment wires UI and model RQ tasks.
Initial M3 source eligibility is effective [general] cellsize = 10 and
dem_db = "ned13/2022"; full integration checks actual raster alignment too.
M1 gains no 10 m restriction. Keep POLARIS/RUSLE automatic enablement for both
models; only M1 requires the user to prepare K through RUSLE.
Change and rationale
Previously unknown M1 slope/SBS intersections produced bounds and no point T, and incomplete K also suppressed point estimates. A large real basin had only 114 unknown intersections among 4,311,420 cells yet no probabilities. The owner prefers useful scalar estimates with explicit spatial support. This supersedes the production point-availability direction in ADR-0058/0059, without altering Horn, the 23-degree threshold, the raw three-state tool outputs or coefficients. Existing outputs retain their original semantics; do not silently reinterpret them.
Alternatives
Probability bounds were proposed and rejected in favor of scalar valid-support estimates. Imputation, silent renormalization without coverage, zero-filling and mandatory map dashboards are not the selected workflow.
Evidence, risk and rollback
See selection contract, SSURGO assessment and the addicted-reservist published predictor diagnostics. Masking changes representativeness; valid coverage must accompany results. No arbitrary minimum coverage cutoff is introduced by this decision. Keep accepted versions and attempts observable; rollback changes future execution policy, never deletes or rewrites prior scientific results. Exact M3 support/geometry treatment remains part of the scientific integration checkpoint.
Analysis-support ratification (2026-09-14)
The owner accepted one common spatial support per model: M1 uses determined slope/SBS intersection plus valid normalized dNBR and K; M3 uses valid SBS and usable thickness after the approved source-priority process. Include valid unburned cells. M3 ruggedness retains full contributing-basin relief and area; masking observations does not redefine basin geometry. Study-area warnings keep full watershed area. This prevents predictor averages describing different observed portions of the basin without fabricating terrain changes.
Venue: owner/Codex conversation, America/Los_Angeles, 2026-09-14; exact clock time not recorded. Decision owner: repository owner; recorder: Codex. The production integration contract records the accepted support and soil-builder noninterference requirement. Soil material/component/fallback details and exact output schema remain subject to their checkpoint; numerical implementation remains pending.