Extending Aircraft Coverage of the Poll Schumann Model Using Public Data

Gabriel Jarry, Ramon Dalmau, Philippe VeryarXiv 2026 · arXiv

TL;DR

Researchers have developed a robust method to reconstruct missing aircraft performance parameters for the Poll–Schumann (PS) model using only publicly available technical data. By combining geometric calculations with machine learning for efficiency estimation, they expanded the model's coverage to 90 aircraft types—including 18 previously unrepresented ones—with fuel prediction accuracy comparable to existing "gold standard" tables.

What Problem Does This Paper Address?

To accurately predict how much fuel a plane burns or whether it will leave a contrail, scientists use physics-based models like the Poll–Schumann (PS) model. However, these models are "parameter-hungry." They require specific dimensionless coefficients describing an aircraft's drag, engine efficiency, and design limits.

Currently, these coefficients are only published for a fraction of the global fleet. If a researcher wants to model a flight for an aircraft not in the "official" tables, they have to either substitute a similar aircraft (a "synonym") or exclude the flight entirely. This creates significant gaps and potential biases in large-scale aviation environmental impact studies. The authors set out to see if public records—like certification documents and engine databases—could be used to "reverse-engineer" these missing parameters.

Methods

The authors developed a hybrid reconstruction workflow that blends classical aerodynamics with statistical learning:

  1. Geometric Reconstruction: They gathered airframe dimensions (wing area, fuselage length, etc.) from manufacturer manuals and type-certificate data sheets. These were plugged into established aerodynamic equations to calculate parameters like the zero-lift drag scale ($\psi_0$).
  2. Engine Data Integration: They pulled engine-specific data (bypass ratios, pressure ratios, rated thrust) from the ICAO Aircraft Engine Emissions Databank.
  3. Efficiency Estimation via Regression: Since propulsion efficiency ($\eta_1$) isn't directly available in public docs, they used Ridge Regression. They trained this model on known aircraft types to predict efficiency based on four public predictors: take-off fuel flow, pressure ratio, the ratio of fuel flow to thrust, and wing area.
  4. Iterative Optimization: They used an iterative process to estimate the design-optimum Mach number, ensuring the reconstructed parameters were physically consistent.

The overall workflow is summarized in Figure 1: Fig.

  1. Inputs to the reconstruction. Solid arrows show the calculations and their inputs or outputs. Dashed arrows carry published values used for fitting or comparison. The regression estimates the efficiency parameter η1 from public predictors, using published efficiency values as training labels.

![Fig.

  1. Inputs to the reconstruction. Solid arrows show the calculations and their inputs or outputs. Dashed arrows carry published values used for fitting or comparison. The regression estimates the efficiency parameter η1 from public predictors, using published efficiency values as training labels.](/figures/art_2f46c85b220a4e26aaaee0e0f0d66be2)

Datasets

  • Public Technical Records: Manufacturer airport-planning manuals, Type Certificate Data Sheets (TCDS), and version 32 of the ICAO Aircraft Engine Emissions Databank.
  • Reference Tables: The "pinned" Poll–Schumann parameter tables from the pycontrails library (a standard open-source tool for contrail modeling).
  • Flight Data: 100 Quick Access Recorder (QAR) flight logs from eight different Airbus A320-214 aircraft, used to validate fuel flow predictions against real-world measurements.

Key Insights and Ideas

  • The "Good Enough" Threshold: The paper demonstrates that while public data is often fragmented or approximate, it contains enough "physical DNA" to reconstruct performance models that rival proprietary datasets.
  • Statistical Efficiency: The most surprising finding was the effectiveness of the $\eta_1$ regression. Even when a whole aircraft family (like the entire A320 family) was excluded from the training set, the model could still predict its efficiency with a Mean Absolute Percentage Error (MAPE) of only 3.91%.
  • Error Compensation: Interestingly, individual reconstructed coefficients sometimes differed from published ones by 7-8%, yet the final fuel burn predictions remained highly accurate. This suggests that in complex physics models, small errors in drag and efficiency can sometimes cancel each other out, leading to robust system-level performance.

Key Results

  • Expanded Inventory: The study produced a new inventory for 90 ICAO type designators (detailed in Table III). This includes 22 types missing from the standard pycontrails table, 18 of which previously had no substitute at all.
  • Coefficient Agreement: Across 68 types, the reconstructed parameters showed strong agreement with published values: 7.38% MAPE for drag and 3.14% for design Mach.
  • Flight Validation: On the 100 A320 test flights, the reconstructed parameters actually slightly outperformed the published ones in cruise fuel flow prediction, achieving a MAPE of 4.50% compared to the published set's 4.68%.
  • Inventory Summary: The scale of the reconstruction is captured in Table II: Extracted table from page 3 showing inventory counts
Extracted table from page 3 showing inventory counts

Limitations and Future Work

  • Validation Breadth: While the reconstruction covers 90 types, the authors only had high-fidelity flight data (QAR) to validate one type (the A320). Future work needs to test these parameters against flight data for a wider variety of airframes and engines.
  • Phase-Specific Accuracy: The model uses a "clean configuration" assumption, meaning it is most accurate during cruise. Accuracy drops during climb and descent where flaps, landing gear, and speed brakes—which aren't explicitly modeled here—are deployed.
  • Hybrid Modeling: The authors suggest using these reconstructed coefficients as a physical baseline for Neural Ordinary Differential Equation (Neural ODE) models, which could better account for operational uncertainties and non-standard flight conditions.

Summary updated

Abstract

The Poll Schumann model requires airframe and engine parameters, but published tables cover only part of the fleet. Applications must obtain missing parameters, substitute another aircraft type or exclude unrepresented types. We reconstruct parameters using geometry from certification documents and manufacturer manuals, together with ICAO engine records. Published equations supply the physical parameters, regression estimates the propulsion efficiency scale. The inventory covers 90 type designators, including 22 absent from the direct pycontrails table and 18 without a listed substitute. Across 68 types, mean absolute percentage error (MAPE) against pycontrails was 7.38% for drag and 3.14% for design Mach. Across 66 types in 21 families, efficiency regression achieved 3.91% MAPE with the evaluated family excluded from input selection and fitting. On 100 recorded A320 flights, a type already represented in the published tables, cruise fuel flow MAPE for reconstructed and published parameter sets was 4.50% versus 4.68%, while mean absolute percentage error in integrated fuel over covered airborne intervals was 7.19% versus 6.82%. Public technical records can therefore extend Poll Schumann parameter coverage, with quantified agreement against published coefficients and comparable fuel prediction errors on the evaluated A320 flights.
BibTeX
@misc{jarry2026extending,
  author = {Gabriel Jarry and Ramon Dalmau and Philippe Very},
  title = {{Extending Aircraft Coverage of the Poll Schumann Model Using Public Data}},
  year = {2026},
  eprint = {2611.02880},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.02880},
}
Figure