Regime-Aware Conformal Uncertainty Quantification for Irregular ADS-B Trajectories in Terminal Airspace
- 1School of Graduate Studies, Embry-Riddle Aeronautical University, Daytona Beach, FL 32114, USA
- 2Aviation Institute, College of Public Affairs and Community Service, University of Nebraska at Omaha, Omaha, NE 68182, USA
Abstract
Automatic Dependent Surveillance-Broadcast (ADS-B) receiver streams contain irregular updates and within-trajectory gaps that may reflect forecast uncertainty. We evaluate reception-conditioned residual calibration using established global and Mondrian split conformal methods. The KLAF archive contains 99,736,221 observations from 5161 aircraft over 463 observed dates; repeated experiments use 1,001,738 observations from 276 complete aircraft–days. Training-selected reported-state constant velocity achieves a test mean horizontal distance error (MDE) of 0.848 km and root-mean-square error of 1.385 km on 220,061 common predictor cases. A finite-adjacent-heading calculation labels 99.97% of these cases; stable and maneuvering MDEs are 0.578 and 1.993 km. On 220,040 identical raw forecast cases, causal selective preprocessing increases MDE by 0.0054 km (95% day-cluster interval 0.0045–0.0061). At 90% target coverage, nominal and degraded reception radii are 2.974 and 3.666 km, with a degraded-to-nominal ratio of 1.233 (95% interval 1.143–1.284). At 95%, the ratio is 1.175 (1.108–1.209). Degraded empirical coverage is 89.49% and 93.79%, respectively. Wider degraded radii persist across segmentation, reception-threshold, and motion-stratified analyses. These are empirical out-of-time results for eligible, matched forecasts from one selected terminal benchmark; chronological splitting and cluster resampling do not establish formal conformal validity under the observed dependence.
Related
- Conformal Uncertainty Quantification Guarantees for Neural Operators
- A Unified Perspective on Conformal Prediction and Wasserstein Distributionally Robust Optimization for Uncertainty Quantification
- Uncertainty Quantification of State Variables Trajectories in the Context of Inverse Problems: An Approach from Bayesian Inference and FDA
- A Combinatorial Formula and a Hilbert Scheme Realization of a Rank-One Virasoro Irregular Conformal Block
- SPACR: Single-Pass Adaptive Training of Uncertainty-Aware Conformal Regressors
BibTeX
@article{yang2026regimeaware,
title = {{Regime-Aware Conformal Uncertainty Quantification for Irregular ADS-B Trajectories in Terminal Airspace}},
author = {Chuyang Yang and Chenyu Huang and Haoruo Fu},
journal = {Mathematics},
year = {2026},
month = oct,
doi = {10.3390/math14203660},
url = {https://www.mdpi.com/2227-7390/14/20/3660},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Regime-Aware Conformal Uncertainty Quantification for Irregular ADS-B Trajectories in Terminal Airspace
AU - Chuyang Yang
AU - Chenyu Huang
AU - Haoruo Fu
PY - 2026
DA - 2026/10/09
JO - Mathematics
DO - 10.3390/math14203660
UR - https://www.mdpi.com/2227-7390/14/20/3660
AB - Automatic Dependent Surveillance-Broadcast (ADS-B) receiver streams contain irregular updates and within-trajectory gaps that may reflect forecast uncertainty. We evaluate reception-conditioned residual calibration using established global and Mondrian split conformal methods. The KLAF archive contains 99,736,221 observations from 5161 aircraft over 463 observed dates; repeated experiments use 1,001,738 observations from 276 complete aircraft–days. Training-selected reported-state constant velocity achieves a test mean horizontal distance error (MDE) of 0.848 km and root-mean-square error of 1.385 km on 220,061 common predictor cases. A finite-adjacent-heading calculation labels 99.97% of these cases; stable and maneuvering MDEs are 0.578 and 1.993 km. On 220,040 identical raw forecast cases, causal selective preprocessing increases MDE by 0.0054 km (95% day-cluster interval 0.0045–0.0061). At 90% target coverage, nominal and degraded reception radii are 2.974 and 3.666 km, with a degraded-to-nominal ratio of 1.233 (95% interval 1.143–1.284). At 95%, the ratio is 1.175 (1.108–1.209). Degraded empirical coverage is 89.49% and 93.79%, respectively. Wider degraded radii persist across segmentation, reception-threshold, and motion-stratified analyses. These are empirical out-of-time results for eligible, matched forecasts from one selected terminal benchmark; chronological splitting and cluster resampling do not establish formal conformal validity under the observed dependence.
SN - 2227-7390
ER -