Radiometry and data integrity

Raw thermal evidence before colour imagery

EmberScope detection claims should trace to raw or radiometric frames, detector state, calibration references, and replayable alert records, not to false-colour video alone.

Read data-integrity story Review contrast model Review calibration

Evidence basis

Detection claims must trace back to pre-AGC thermal frames.

14-bit+ Radiometric, raw-count, or pre-AGC source frames are the minimum acceptable evidence path.

No AGC False-colour video and JPEGs can help operators, but they cannot be the quantitative detection evidence.

2 refs Ambient and warm references anchor each calibration segment during field flights.

Replay Every alert needs raw-frame identity, detector state, calibration state, and threshold provenance.

How the pieces connect

Radiometry is a flight-data contract, not an assumed detector feature.

The requirements separate quantitative thermal evidence from operator imagery. A flight can still show useful colour video, but validation metrics need raw or radiometric frames with checksums, detector settings, and calibration provenance.

The first error gates are provisional. They let detector integration and field testing proceed while the small-hot-target radiometry and bandpass models mature.

Radiometry map

Use this page as the main path, then open the detailed model, bandpass, calibration, and pipeline pages.

Contrast model

Radiometry model tests subpixel target contrast with NETD, target size, and background temperature.

Bandpass trade

Bandpass trade compares LWIR, MWIR, and dual-band options for the first payload.

Fire radiometry

Fire radiometry literature explains mixed pixels and why FRP retrieval is a later evidence layer, not the first alert gate.

Calibration

Calibration strategy defines warm and ambient references, drift checks, and NUC state handling.

Pipeline

Onboard pipeline keeps raw frames, geotags, alert provenance, and review labels tied to each detection.

Calibration cadence

Reference passes are required flight events.

Stage	Required action	Failure gate
Bench characterization	Measure warm-up, shutter/NUC timing, bad-pixel state, window offset, and two-reference response.	No validation flights until detector behavior is known.
Pre-flight stabilization	Use 20 minutes as the conservative warm-up rule until bench data supports less.	Pre-stabilization frames are diagnostic only.
Launch reference pass	Capture ambient and warm references in raw or radiometric mode before scoring survey detections.	No launch reference means no benchmark-grade detections.
In-flight repeats	Repeat at least every 20 minutes, after major NUC behavior change, after more than 3 deg C payload-temperature shift, or after window contamination.	Stale references downgrade later alerts.
Post-flight closeout	Capture final references, inspect the window, record payload temperature, and close the manifest.	Missing closeout blocks benchmark reporting unless manually justified.

Temperature-error gates

Weak-target work has explicit benchmark, advisory, and exclusion classes.

Benchmark-grade

Reference residual within 2 deg C, target-background repeatability within 1 deg C, and references no older than 20 minutes or 3 deg C payload-temperature shift.

Advisory-grade

Reference residual within 5 deg C or incomplete repeat cadence. Useful for operator review and debugging, not headline performance metrics.

Uncalibrated

Missing references, unknown shutter/NUC state, unexplained residuals, or contaminated window state. Save for review but exclude from detection claims.

What remains to prove

The final error budget waits on target radiance, detector noise, atmosphere, bandpass, and end-to-end simulation closure.

Record contract

Every alert needs enough thermal provenance to be replayed.

Alert packets should carry raw-frame ID and checksum, calibrated and uncorrected crop values, local background estimate, target-background contrast, calibration segment, reference age, residual, and quality class.

Frame indexes should preserve detector settings, shutter/NUC state, bad-pixel map, dropped-frame state, payload temperature, window state, weather fields, geotag state, and threshold profile.

Dataset exports should state source flights, excluded frames, detector branch, optics profile, calibration quality, split membership, labels, checksums, and replay configuration.