Aircraft Calibration

How HyPlan’s aircraft performance models are derived from in-situ field-campaign data.

What “calibrated” means

For each calibrated platform, the following parameters in hyplan/aircraft/_models.py come from per-altitude-bin medians of real flight data rather than manufacturer brochures:

• climb_profile and descent_profile (VerticalProfile) — active-VS bin medians (5-kft bins, n≥30/bin) from sustained- climb and sustained-descent fixes. “Active” means vertical_rate * sign ≥ ACTIVE_VS_THR_FPM, where the threshold is per-aircraft (1500 fpm for jets and most turboprops, 1000 fpm for the B-200, 500 fpm for the Twin Otter).

• climb_schedule / cruise_schedule / descent_schedule (TasSchedule) — per-phase TAS-vs-altitude medians at the same 5-kft binning, anchored at the typical rotation TAS at SL and picked at aircraft-specific target altitudes.

• turn_model.max_bank_deg — usually max(AFM normal-ops 30°, data p90) of |roll_deg| over fixes with |roll| > 5°.

• approach_speed — median TAS in the last ~500 ft AGL with vertical_rate < -200 fpm.

• service_ceiling — operational p99 of per-sortie peak altitudes (not the certified service ceiling at MTOW).

• sources (SourceRecord) — campaign / archive citation + number of sorties + confidence.

We don’t enforce monotonicity on climb_profile / descent_profile. Jets and turboprops typically peak ROC near FL050-FL100 (limited below by 250-KCAS ATC procedures); descent VS typically peaks around FL150-FL200 (descent at VMO in CAS) and then declines in the upper levels (Mach-limited descent at constant M). Clamping these to monotone shapes pushes bins outside their IQRs and obscures the real envelope, so the calibration step ships the raw bin medians and lets reviewers see the shape directly.

Calibration status

Each aircraft class exposes its provenance via the Aircraft.calibration_status attribute, with values "calibrated", "inferred", or "uncalibrated" corresponding to the Status column below.

Aircraft class	Status	Source	Sorties	Notebook
NASA_ER2	calibrated	NASA AFRC IWG1	cache-dependent	er2/calibration.ipynb
NASA_GIII	calibrated	NASA ASP archive IWG1	~153	giii/calibration.ipynb
NASA_GV	calibrated	NASA ASP archive IWG1	~101	gv/calibration.ipynb
NASA_WB57	calibrated	NASA ASP archive IWG1	~100	wb57/calibration.ipynb
C130	calibrated	NASA ASP archive IWG1 (ACT-America)	~87	c130/calibration.ipynb
NASA_P3	calibrated	NASA ASP archive IWG1	~252	p3/calibration.ipynb
KingAirB200	calibrated	NASA ICARTT (ACTAMERICA, DISCOVER-AQ, KORUS-AQ, LMOS)	250	b200/calibration.ipynb
TwinOtter	calibrated	NASA / NOAA ICARTT (FIREX-AQ N48RF)	~17	twin_otter/calibration.ipynb
NCAR_GV	inferred	Mirrored from NASA_GV (same airframe class)	—	(deferred — see below)
NASA_C20A	inferred	Mirrored from NASA_GIII (same type certificate)	—	(deferred — see below)
NASA_GIV	uncalibrated	Manufacturer brochure	—	—
NASA_B777	uncalibrated	Manufacturer brochure	—	—
KingAirA90	uncalibrated	Manufacturer brochure	—	(deferred)

Sortie counts may shift as data deliveries refresh. See each notebook’s §1 summary table for the current count.

Methodology

Each calibration.ipynb notebook follows the same recipe (the ER-2 notebook predates this builder pattern but the structure is the same):

1. Load every IWG1 / ICARTT file from the campaign directory, trim ground taxi (or fall back to altitude-only airborne detection when groundspeed is missing).

2. Filter sorties on duration (60–600/900 min, per-aircraft) and peak altitude (per-aircraft floor and ceiling), drop sorties with no valid altitude or vertical rate.

3. Phase-label each fix as climb / cruise / descent from vertical_rate against ±300 fpm gates.

4. Bin climb / descent fixes by 5-kft altitude bin and compute median, p25, p75 of vertical rate.

5. Bin TAS the same way for each phase.

6. Pick breakpoints at aircraft-specific target altitudes; anchor SL at typical rotation / approach TAS so the SL point isn’t contaminated by takeoff-roll or pattern fixes.

7. Validate by overlaying the proposed VerticalProfile on the per-bin IQR + median plot.

8. Emit a paste-ready cell with the calibrated constants, SourceRecord, and labeled service_ceiling / approach_speed so it’s clear which numbers are operational vs aircraft- intrinsic.

Shared helpers live in notebooks/calibration/_common.py: label_phases, apply_sortie_filters, per_bin, tas_per_bin, schedule_pts, evaluate_profile, summary_table. The aircraft-specific knobs (active-VS threshold, target altitudes, rotation TAS, hold bands for the ER-2) stay in the per-aircraft builder.

Operational vs aircraft-intrinsic numbers

Several reported metrics describe operational behavior across the sortie set rather than aircraft-intrinsic performance:

• Wall-clock time-to-cruise (TOC) includes pre-cruise level-offs, ATC routing, and weight-management step climbs. Reviewers should not read TOC variance as a performance bound; the planner’s modeled TOC comes from integrating the calibrated climb_profile, which excludes those operational delays.

• service_ceiling as shipped is the operational p99 of per-sortie peak altitudes for the mission mix. This is below the airframe service ceiling under MTOW (e.g., B-200 ships 30000 ft vs 35000 ft brochure; Twin Otter ships ~15000 ft vs 25000 ft brochure). Each calibration notebook explicitly labels this in its paste-ready cell.

• approach_speed is the median final-approach TAS for the mission mix. The Twin Otter notebook’s 99 kt is on the high side of the DHC-6’s 80–90 kt typical AFM approach, reflecting the NOAA mission profile.

Deferred calibrations

Three aircraft classes remain on inferred or brochure values because their public data sources are auth-walled or unavailable:

• NCAR_GV (HIAPER, N677F) — sources at https://data.eol.ucar.edu/ (HIPPO / SOCRATES / ORCAS / ATTREX / WE-CAN), accessible via NCAR EOL ORDER request. The generic ICARTT loader recognizes HIAPER LRT variable names via hyplan/aircraft/eol_ncar.py.

• NASA_C20A (NASA 502, AFRC G-III variant) — currently inferred from NASA_GIII (same airframe + type certificate). No public ICARTT/IWG1 nav data; AFRC mission ops contact required for NASDAT housekeeping logs.

• KingAirA90 — no public IWG1-grade A-90 data; the only available files are ADS-B-grade tracks (n94s_alltracks.csv) with no TAS / Roll / IAS. A wind-derivation fallback (GPS groundspeed + MERRA-2 wind field) is the long-term path.

Order of effort if access becomes available: HIAPER (loader is closest to ready, ~50–100 sorties available across HIPPO / SOCRATES / ORCAS) → C-20A (only meaningful if AFRC ops shares data) → A-90 (wind-derivation last resort).

Conventions

• Active-VS thresholds (per-aircraft):

  • Jets and most turboprops: 1500 fpm

  • King Air B-200: 1000 fpm (climb rates drop below 1500 fpm above FL150; the lower threshold extends bin coverage to FL250-FL280)

  • Twin Otter: 500 fpm (slow climber; 1500 fpm loses everything above FL050, 800 fpm cuts off above FL100)

• 5-kft bin width is universal; smaller bins thin the per-bin sample below the n=30 floor at high altitude.

• n≥30/bin floor for VS bins; n≥50–200/bin for TAS bins (per-aircraft, set in the builder).

• n≥30/bin for active-VS keeps every bin that survives contamination filtering above autopilot precision.

• The ER-2 notebook excludes weight-management hold bands (FL220–FL260, FL240–FL280, FL336–FL376) when computing active-climb medians so sortie-specific level-offs and holds do not depress the aircraft-intrinsic climb profile.

• Default confidence=0.85 for data-fit calibrations, confidence=0.7 for inferred-from-sibling values.

• max_bank_deg ships as max(AFM normal-ops 30°, data p90) so the planner uses a bank that the aircraft is willing to fly, not the typical-mix median or a steep-turn / emergency value.