Planned ESA interval selection
The ESA recording supplies much longer observations at lower ground speeds than the original archive cohort. Its selection policy must reflect that geometry. The existing 53 stretches and their criteria remain unchanged.
This is a development proposal, not a frozen selection or a finding. It is based on receiver trajectory, source completeness and the survey report, before any ESA Earth-model fitting. Reported survey-line boundaries describe the survey and do not automatically define eligible intervals.
Proposed initial cohort
| Requirement | Proposed rule | Purpose |
|---|---|---|
| Source completeness | Intact full-file hash, supported format, finite values and strictly increasing timestamps | Establish usable observations without repairing them. |
| Timing continuity | Split at IMU gaps above 1.5 nominal periods, GNSS gaps above 0.75 s, or configuration/time discontinuities | Avoid bridging unsupported observations. |
| Duration | At least 20 continuous minutes after motion exclusions | Use the longer records to compensate for slower travel. |
| Ground speed | Every supported receiver sample at least 180 km/h | Focus on sustained flight rather than ground operations. |
| Traveled distance | At least 60 km, calculated from the independent receiver trajectory | Preserve substantial accumulated change of local level. |
| Low maneuvering | Ten-second trajectory windows: median absolute vertical speed at most 1 m/s, absolute course-rate at most 0.02°/s | Screen climbs and aircraft turns using receiver data. |
| Maneuver margin | Exclude 30 seconds before and after failed motion windows | Avoid fitting through poorly isolated maneuver transitions. |
| Receiver quality | Finite positive position covariance diagonals; horizontal standard deviations at most 5 m, vertical at most 10 m, PDOP at most 4, at least 8 reported satellites | Reject unsupported positions using recorded quality fields. |
At 180 km/h, twenty minutes covers 60 km. The distance is somewhat greater than the minimum 46.7 km covered by the original four-minute, 700 km/h requirement. This is a geometric design comparison, not evidence derived from the gyro. GNSS coordinates and their geodetic construction must remain explicit in later model comparisons.
Use complete retained stretches, with exact source hash, GPS week/time, UTC bounds, duration, speed, distance, height and heading summaries in a durable inventory. Preserve all rejected intervals and reasons. Do not treat subwindows of one recording as independent flights. Do not use the gyro's agreement with a physical model, or the supplied fused navigation answer, to choose intervals.
Required qualification before use
Implement cheap deterministic controls for gaps, time ordering, heading wrap, nonuniform timestamps, motion margins, quality failures and boundary inclusivity. Compute course changes from receiver velocity with actual timestamps; do not interpolate measurements. Define window edge coverage and behavior at zero speed explicitly. Freeze one version of the policy and its inputs before evaluating it. Publish its results even if no intervals qualify; changed rules require a new inventory version rather than overwriting an unfavorable outcome.
The first selected recording has 8.49 hours of IMU observations and a GNSS-only PPP export described by the survey report. Its longest reported line is 57.4 minutes at about 263 km/h but changes ellipsoid height by about 390 m. Whether the motion rules retain that line or split it must be determined by the declared trajectory calculation.
Selection alone authorizes no model-fitting campaign. The IMU logs engineering-unit rates and accelerations in reported aircraft axes, rather than Applanix integer increments. The observation adapter, integration clock, antenna lever arm, processing assumptions and physical error bounds need separate qualification. No existing decision threshold or sensor-performance assumption transfers automatically to ESA or Wuhan recordings.