Innovation-Space Error Diagnostics API
This module implements the Desroziers innovation‑space diagnostic for
estimating observation‑error variance and background‑error contributions
using only the innovations (OMB and OMA). It is a lightweight tool for
monitoring the statistical consistency of the assumed observation-error
variance R in UFS DA workflows.
The diagnostic is implemented in:
ufs_da_diagnostics/obs/innovation_br_check.py
and is designed to operate on the same YAML configuration and data
structures used by the standard obs_diag utilities.
ATMS‑only innovation-based B/R diagnostics using OMB/OMA and obs_diag.yaml.
- Outputs per channel:
Sd = E[OMB^2] (innovation variance)
R_est = E[OMA * OMB] (Desroziers estimate of R_true)
Sd/R = Sd divided by assumed R (chi-square proxy)
R_est/R = R_est divided by assumed R (variance scaling indicator)
HBH^T = Sd - R_est (background error contribution)
HBH^T/R = (Sd - R_est) / R (background-to-observation ratio)
scale_R = R_est / R (variance multiplier from Desroziers)
infl_chi = sqrt((Sd/R) / chi_target) (std-dev inflation to target chi^2)
- Usage:
ufsda-obs-br-check –yaml obs_diag.yaml