BADA Model: Aircraft Performance & Trajectory

• Base of Aircraft Data (BADA) model is a physics-based framework that predicts aircraft trajectories, speeds, thrust, and fuel flow using energy balance equations.
• It integrates type-specific lookup tables with data-driven corrections to enhance accuracy in climb, descent, and fuel estimation.
• BADA supports stochastic generative modeling and uncertainty quantification, enabling scalable airspace simulation and environmental analysis.

The Base of Aircraft Data (BADA) model is a comprehensive, physics-based aircraft performance modeling framework designed for predicting the trajectories, speeds, thrust, drag, and fuel flow of fixed-wing aircraft. BADA is widely used in airspace simulation, trajectory prediction, air traffic management (ATM), and as a canonical baseline for data-driven hybrid models. Its core is built on energy-balance principles, parameterized by type-specific lookup tables and analytic schedules.

1. Physical Foundations and Core Equations

BADA models the aircraft as a point mass with performance governed by the total energy equation, encapsulating the interplay between thrust, aerodynamic drag, altitude, and airspeed. The fundamental climb/descent law is

$$\frac{dh}{dt} = \frac{T_{HR}(h) - D(h, V_{TAS})}{m\,g_0}\;V_{TAS}\;f(M(h, V_{TAS})),$$

where $h$ is geopotential altitude, $T_{HR}$ is thrust resolved parallel to velocity, $D$ is aerodynamic drag, $m$ is mass, $g_0$ is standard gravity, $V_{TAS}$ is true airspeed, and $f(M)$ is the energy-share factor switching between constant calibrated airspeed (CAS) and constant Mach regimes. The core “total energy” balance also yields an implicit equation connecting rate-of-climb/descent (ROCD) and performance functions: $$(T_{HR} - D)\,V_{TAS} = m\left( g_0\,\dot h + V_{TAS}\,\dot V_{TAS} \right).$$ Thrust schedules $T_{HR}(h)$, drag polars (e.g., $h$0), wing parameters, and CAS–TAS conversions are type- and phase-dependent, and all are systematically provided in BADA’s lookup tables and documentation (Hodgkin et al., 6 Jan 2026, Pepper et al., 2023, Hodgkin et al., 3 Apr 2025, Jarry et al., 2024).

2. BADA as a Basis for Physics-Informed and Hybrid Modeling

Contemporary research increasingly uses BADA as a backbone for physics-informed machine learning (PIML), in which low-dimensional corrections—learned from real trajectory data—are inserted into the base BADA equations to account for operational randomness, meteorological effects, and operator procedures. For example, altitude-dependent thrust corrections are parameterized as

$h$1

with $h$2 orthonormal basis functions from fPCA and weights $h$3 sampled probabilistically, enabling full generative models of ensemble trajectories. Extensions also fit altitude-dependent drag and airspeed profiles, learning $h$4 and $h$5 from radar-derived state inversion, then using functional expansions and data-driven latent-variable priors (Pepper et al., 2023, Hodgkin et al., 3 Apr 2025).

Data-driven corrections respect physical plausibility by enforcing monotonic climb/descent, bounded speed, and realistic fuel flows. Constraints are maintained either directly via BADA’s envelope checks or through rejection sampling of unphysical functional samples.

3. Stochastic Generative Models and Uncertainty Quantification

BADA’s analytic structure enables efficient embedding of stochastic or probabilistic corrections. For trajectories, this entails:

• Sampling latent thrust and/or speed function weights (e.g., $h$6) from a learned conditional distribution $h$7, conditioned on meteorological and contextual features (wind, temperature deviation, operator identity, flight phase).
• Running forward BADA integration for each sample, producing an ensemble $h$8, yielding empirical and distributional uncertainty quantification.
• For confidence bounds, analytic maximization in the functional-PCA weight space produces upper and lower trajectory envelopes with minimal computational expense.

Probabilistic approaches incorporating meteorological conditioning have demonstrated a 20–22.5% improvement in skilfulness metrics (RMSE, CRPS for time-to-climb, ROC, CAS) relative to baseline models that ignore context (Hodgkin et al., 6 Jan 2026). Data-driven thrust corrections in BADA-based climbs have reduced mean absolute error in arrival time by 66.3% compared to vanilla BADA (Pepper et al., 2023), while hybrid descent simulation achieves a tenfold reduction in time-to-descent MAE (Hodgkin et al., 3 Apr 2025).

4. Application to Fuel Flow and Environmental Modeling

BADA contains explicit formulae for fuel flow: $h$9 with thrust required $T_{HR}$0 and drag $T_{HR}$1 modeled via analytic or tabulated polars. These relationships are systematically used both for fuel-burn prediction and as a baseline for machine learning tasks. In deep learning for fuel flow estimation, BADA's outputs act as ground-truth labels, and input features are concatenated with BADA’s aircraft and engine type parameters. Models thus trained generalize to unseen aircraft types with mean absolute percentage error (MAPE) between 2–10% for types close in parameter space, and <1% for types seen in training data (Jarry et al., 2024).

BADA underpins scalable and universal fuel-flow estimation frameworks by encoding domain-specific aircraft performance physics, enabling transfer beyond data-rich aircraft types.

5. Model Calibration, Performance, and Data-driven Extensions

BADA-based models are routinely evaluated on large Mode S radar datasets, with data spanning thousands of aircraft and more than 100 aircraft types. Processing includes:

• Extraction of continuous trajectory segments (e.g., climb segments where $T_{HR}$2 ft/min or descent with $T_{HR}$3 ft/min).
• Functional data analysis (fPCA) to parameterize variability in thrust, drag, or speed as low-order expansions.
• Statistical fitting of latent weight distributions: Gaussian, mixtures (GMM), or normalizing flows, enabling rich multimodality in operational performance.
• Preprocessing features: meteorological (wind, Δ-ISA), context (operator, city-pair, flight type), embedded as encodings or features for conditional modeling.

Physical plausibility is rigorously enforced throughout, e.g., via monotonicity checks on climb rate and bounding generated drag/speed functions within observed envelopes (Hodgkin et al., 6 Jan 2026, Hodgkin et al., 3 Apr 2025).

Quantitative comparisons consistently demonstrate that hybrid models—those coupling BADA with probabilistic, data-driven corrections—outperform vanilla BADA across all trajectory and performance benchmarks, achieving realistic distributional coverage, significant reductions in standard errors, and analytic (not sample-based) uncertainty bounds.

6. Limitations and Prospects for Future Research

Current BADA-based hybrid models exhibit several limitations:

• Applicability is phase-constrained: most results focus on en-route climbs and descents, with terminal maneuvering and managed-descent scenarios requiring new phases, features, or adaptive bases.
• Fuel-flow prediction often leverages BADA’s tabulated relations; direct learning of fuel-burn residuals for emissions or environmental modeling remains open.
• Functional basis truncation (e.g., to 80% explained variance) may miss localized or abrupt changes; richer functional embeddings (e.g., VAE or adaptive bases) could be employed.
• Real-time operational deployment may require light-weight or surrogate BADA emulators for rapid feature computation and trajectory synthesis.
• Regulatory certification of PIML approaches is promising due to BADA’s retention of interpretability and built-in safety margins; however, comprehensive certification requires further study (Hodgkin et al., 6 Jan 2026).

A plausible implication is that, as coverage and fidelity of Mode S or QAR data increases, and as models improve in expressiveness, BADA will remain an indispensable core for physically constrained, generalizable, and certifiable ATM and trajectory prediction frameworks, providing the physical backbone for continued advances in hybrid physics–machine learning methods.