PaperlayerView exact revision on arXiv

Source-linked AI summary

3-D Emissions Mapping and Social Cost Estimation for US Domestic Aviation at West Coast Hubs

Hesam Shafiei Nia, Don MacKenzie

arXiv:2609.31686v1cs.LGphysics.data-an

TL;DR

Missing or incomplete ADS-B trajectories limit high-resolution aviation emissions and impact assessment. The paper reconstructs trajectories and estimates altitude-based emissions and costs for west coast hubs; its results report a small NOx climate credit alongside substantial NOx air-quality and health costs.

  • Problem

    ADS-B coverage gaps and missing observations can make emissions inventories, social-cost estimates, and health assessments inaccurate.

    Source: INTRODUCTION · p. 3

  • Method

    The paper combines autoencoder-based trajectory reconstruction with altitude-based emissions estimation and social-cost assessment for US west coast hubs.

    Source: INTRODUCTION · p. 4

  • Results

    NOx emissions provide a small net climate credit of $1.35 million but contribute $34.6 million to air-quality and health impacts.

    Source: DISCUSSION AND CONCLUSION · p. 16

  • Takeaways & Limitations

    The results aim to inform operational enhancements and policy introduction to reduce aviation’s climate, air-quality, and health impacts.

    Source: INTRODUCTION · p. 4

    The framework assumes constant aircraft weights using BADA-defined values, and downstream estimates depend on physically flyable reconstructed altitude and speed profiles.

    Source: Training objectives · p. 8, Emissions estimation framework · p. 11

Abstract

from arXiv · show

Existing aviation emissions inventories lack accurate trajectory data for high-resolution social cost and health impact assessment. This paper develops a 3-D emissions map by reconstructing flight trajectories for US west coast hubs to estimate regional environmental and near-airport health impacts. A physics-informed autoencoder (AE) is applied to ADS-B trajectory records for January 2025 covering US west coast hubs. The encoder combines a Convolutional Neural Network (CNN), a Bi-GRU, and a 3-D CNN with skip connection; the decoder is a Temporal Convolutional Network (TCN). It is benchmarked against a baseline-AE and cubic spline interpolation. Emissions are mapped via EUROCONTROL Base of Aircraft Data (BADA) performance tables and ICAO Engine Emissions Databank (EEDB) emission indices, with altitude corrections via Boeing Fuel Flow Method 2 (BFFM2). Social costs are quantified for all flight phases, with health impacts assessed for Landing and Takeoff cycles within 50 km of each hub. The proposed AE model outperforms both a TCN-AE and cubic spline interpolation across 5% to 50% missing rates. Monetizing the emissions inventory shows NOx produces a small net cooling effect in direct climate forcing, while accounting for 99.7% of monetized air-quality and health cost despite being under 0.4% of CO2 by mass, making it the dominant health-cost driver. To our knowledge, this is among the first studies combining AE-based trajectory reconstruction with separate spatial-temporal feature encoding and altitude-based emissions modeling to produce a regional aviation emissions inventory for air quality, climate impact and population exposure. The resulting emissions map and social cost estimates provide quantitative context for environmental impact assessment and near-airport health policy evaluation for US domestic aviation.

INTRODUCTION

Aviation emissions and their impacts vary by altitude and location, but missing ADS-B observations and conventional interpolation limit high-resolution assessment. The paper proposes trajectory reconstruction and altitude-based emissions estimation for west coast hubs to support climate and near-airport health-cost analysis.

INTRODUCTION

  • ADS-B coverage gaps and interpolation-based reconstructions constrain accurate emissions inventories, social-cost estimates, and health assessments.ADS-B gaps can result from signal loss, receiver dropout, and incomplete network coverage; smooth interpolation may miss nonlinear altitude, speed, and thrust relationships.Source: INTRODUCTION · p. 3
  • High-resolution social- and health-cost estimates require an emissions inventory that resolves where and at what altitude aviation emissions occur.Aviation emissions affect air quality and climate, with impacts varying across locations and altitudes.Source: INTRODUCTION · pp. 3–4

METHODOLOGY

  • The proposed framework estimates altitude-based emissions for one month of ADS-B trajectories to and from US west coast hubs.It uses trajectory imputation, BADA aircraft-performance data, ICAO EEDB emission indices, and BFFM2 altitude corrections to estimate emissions and associated costs.Source: INTRODUCTION · p. 4, METHODOLOGY · p. 4

INTRODUCTION

  • The paper combines autoencoder-based trajectory reconstruction with emissions inventory estimation to assess climate and near-airport health costs together.Prior studies often addressed climate-scale impacts and near-airport health effects independently.Source: INTRODUCTION · p. 4
  • The study aims to provide decision-makers with results they can use to reduce aviation’s climate, air-quality, and health impacts through operational enhancements and policy introduction.The methodology focuses on US domestic flights in the west coast region.Source: INTRODUCTION · p. 4, METHODOLOGY · p. 5

Data and processing

The study processes January 2025 ADS-B trajectories for three west coast hubs and imputes incomplete flights before emissions estimation. Its autoencoder encodes spatial, temporal, and joint spatial-temporal patterns, then fuses them for reconstruction.

Data and processing

  • 20,134 January 2025 trajectories were identified for KSEA, KLAX, and KSFO, and 53.4% required imputation under the 36-second completeness threshold.Trajectories were treated as complete only when their largest gap was at most 36 seconds.Source: Data and processing · p. 5

Trajectory imputation

  • The imputation model learns patterns from complete flights and checks reconstructions against physical constraints to produce plausible incomplete trajectories.This is intended to support accurate and realistic downstream emissions and social-cost estimates.Source: Trajectory imputation · p. 5

Proposed architecture (CBiG3D-AE)

Encoder

  • A CNN branch encodes spatial features while a Bi-GRU reads temporal features in both directions, capturing local route shape and motion dynamics.The bidirectional recurrent body can use evidence from either side of a gap when inferring missing values.Source: Encoder · p. 6
  • A 3D-CNN branch encodes spatial-temporal evolution by processing aircraft-position maps as a timestamped stack.It complements the route-shape and temporal-dynamics features encoded separately in the first branch.Source: Encoder · p. 6
  • The encoder fuses outputs from its three bodies into a fixed-size latent vector, while skip connections preserve timestamp-aligned detail for decoding.The skip path carries input information forward, and dilated kernels project observed-neighbor information into gap positions.Source: Encoder · pp. 6–7

Decoder

The decoder uses a TCN to reconstruct full trajectories, while training combines masked-region reconstruction with physics-informed regularizers. Training corruptions expose the model to varied missing-data patterns and implausible altitude artifacts.

Decoder

  • The TCN decoder reconstructs long gaps using dilated convolution blocks with expanding receptive fields and information from both sides of each gap.Its sequence-wide parallel processing is given as the reason for choosing it over LSTM or GRU decoding.Source: Decoder · p. 7

Training objectives

  • Training uses 27,500 complete historical trajectories without manual labels, while the January 2025 west-coast dataset is reserved for evaluation.The training trajectories use coordinates and kinematic features to teach the model what a flyable trajectory looks like.Source: Training objectives · p. 7
  • Three randomly selected corruption modes simulate sparse dropouts, contiguous ADS-B gaps, and altitude spikes; contiguous-gap lengths expand from 15% to 50% over training.Corruption is resampled each epoch, and validation uses a fixed 20% contiguous gap for consistent checkpoint selection.Source: Training objectives · p. 8
  • The total loss combines masked-timestep reconstruction with latent, vertical-rate, smoothness, altitude-bound, and groundspeed regularizers.Reconstructed trajectories must be physically flyable because the downstream emissions inventory directly relies on plausible altitude and speed profiles.Source: Training objectives · p. 8
  • The physics penalties constrain vertical-rate consistency, stepwise vertical-rate and groundspeed smoothness, and altitude to an envelope bounded by terrain elevation and an approximately 13,000 m service ceiling.The groundspeed smoothness term uses an approximately 350 m/s normalization ceiling; fixed weights are set for the physics attributes.Source: Training objectives · pp. 8–9

Data imputation

The model fills missing points while preserving observed records, then combines imputed and complete flights into a full-resolution dataset for emissions estimation.

Data imputation

  • 10,751 of 20,134 January 2025 west-coast flights (about 53.4%) required imputation at the 36-second gap threshold, with a 13.5% average within-flight missing fraction.Each flight is processed multiple times; observed points remain unchanged, and only missing points or blocks are filled before merging with complete trajectories for emissions estimation.Source: Data imputation · p. 9
  • Observed points are kept exactly as recorded, while the model fills only missing points and blocks.The imputed flights are merged with complete flights into a full-resolution dataset passed to emissions estimation.Source: Data imputation · p. 9
  • At a 36-second gap threshold, 10,751 of 20,134 west-coast flights in January 2025 required imputation.The average within-flight missing fraction among the reported flights was 13.5%.Source: Data imputation · p. 9

Baseline AE model

The evaluation compares the proposed autoencoder with a deep-learning baseline and cubic spline interpolation, described as representing current practice.

Baseline AE model

  • The deep-learning comparison is described as lacking the proposed model’s separate spatial-temporal contribution and added skip connection.No model architecture details or performance results are given here.Source: Baseline AE model · p. 9
  • The evaluation uses two approach types: a deep-learning model and conventional cubic spline interpolation.The passages identify the comparison categories but do not specify which approach performs better.Source: Baseline AE model · pp. 9–10
  • The comparison includes cubic spline interpolation as a conventional method representing current practice.The passage does not report comparative performance results.Source: Baseline AE model · p. 10

TCN-AE

The TCN-AE serves as a control for assessing the proposed encoder’s separate spatial-temporal feature design.

TCN-AE

  • The TCN-AE replaces the proposed two-branch encoder with a TCN encoder that processes all spatial-temporal features together.It retains a TCN decoder.Source: TCN-AE · p. 10
  • The TCN-AE uses the same training procedure and self-supervised masking approach as the proposed model.This keeps those aspects consistent in the comparison.Source: TCN-AE · p. 10
  • Comparable parameter counts are intended to keep model capacity from confounding the comparison.The authors attribute performance differences between the models specifically to the proposed encoder’s two-branch spatial-temporal design.Source: TCN-AE · p. 10

Interpolation methods

Cubic spline interpolation provides smooth gap-filling but can produce physically implausible paths. Evaluation uses held-out trajectories with contiguous synthetic gaps and measures feature errors and overall trajectory shape.

Interpolation methods

  • Cubic spline interpolation fits smooth piecewise-cubic curves but may produce physically implausible paths or out-of-range altitudes.Such errors can affect emissions estimates.Source: Interpolation methods · p. 10

Evaluation

  • Methods are tested on held-out trajectories with one contiguous synthetic gap of 5%–20% of trajectory length, matching block-shaped ADS-B dropouts.Scattered single-point dropouts and spike injection are not reported.Source: Evaluation · p. 10
  • Evaluation uses MAE and RMSE for each trajectory feature and dynamic-time-distance position error for overall trajectory shape, computed over masked timesteps.The reconstructed value is compared with the true observed value at each masked timestep.Source: Evaluation · pp. 10–11

Emissions estimation framework

The framework estimates emissions by combining aircraft performance data with altitude-adjusted emission indices, then monetizes climate and near-airport air-quality and health damages. It distinguishes pollutant effects by flight phase.

Emissions estimation framework

  • BADA performance tables and trajectory altitude, calibrated airspeed, and vertical rate inform fuel flow, while ICAO EEDB emission indices are adjusted to flight conditions using BFFM2.The correction is applied to both LTO and non-LTO phases; aircraft weights are assumed constant at BADA-defined values.Source: Emissions estimation framework · p. 11
  • Flight-total emissions mass is calculated by summing each species’ altitude-based emission index multiplied by fuel flow and timestep duration.The framework estimates CO2, NOx, HC, and CO across the west coast, and nvPM for health effects near hubs.Source: Emissions estimation framework · p. 11

Social cost and health impact framework

  • Climate, air-quality, and health damages are monetized from the estimated emissions map, using $190 per metric ton of CO2 for climate costs.The unit damage costs are presented in Table 1.Source: Social cost and health impact framework · p. 12, Table 1 · p. 12
  • NOx has a small net cooling effect on direct climate forcing, while its air-quality and health cost is two orders of magnitude larger and dominates its total impact.Phase-specific marginal damage costs are applied separately to LTO and non-LTO emissions.Source: Social cost and health impact framework · p. 12

RESULTS

The proposed autoencoder has the best overall trajectory-reconstruction accuracy among the compared approaches and performs particularly well on altitude and vertical rate. Its overall MAE and RMSE lead persists across the reported missing-rate scenarios.

RESULTS

  • Aggregate MAE and RMSE are unweighted summaries across eight normalized features, so altitude and vertical-rate errors are reported separately as most relevant to fuel flow and emissions calculations.The aggregate is approximate rather than physically weighted.Source: Imputation accuracy · p. 13

Imputation accuracy

  • The proposed AE achieves the best overall trajectory-reconstruction accuracy over baseline-AE and cubic spline interpolation, with stronger altitude and vertical-rate performance.Cubic spline performs better on raw position and ground speed in the studied short and moderate flights.Source: Imputation accuracy · p. 13, Table 2 · p. 13
  • Across missing-rate scenarios, the proposed model leads in overall MAE and RMSE; it also outperforms cubic spline interpolation at higher missing rates.Table 3 reports synthetic and real missing-data-rate analysis.Source: Imputation accuracy · pp. 13–14, Table 3 · p. 14

Emissions inventory

NOx emissions shift from hub-centered patterns at lower altitudes to broad corridor distributions at cruise, while cruise’s NOx mass share remains below its fuel share. Monetized costs show NOx’s small net climate cooling alongside its dominant share of air-quality and health costs.

Emissions inventory

  • At cruise, NOx contributes 56.5% of its mass but 64.8% of fuel burn; its share exceeds fuel share at every lower altitude band.The passage attributes this pattern to emission indices peaking during climb and reaching their lowest levels at cruise.Source: Emissions inventory · p. 14, Table 4 · p. 16
  • The passage cautions that LTO-phase CO and HC values may be underestimated, but says this does not meaningfully change the main results because their monetized air-quality costs are small.Possible reasons given are missed taxi and idle operations in the LTO-cycle filter or lower BFFM2 accuracy at low thrust.Source: Emissions inventory · p. 14
  • At lower altitudes, NOx is sparse and hub-centered, whereas cruise-altitude emissions form dense corridor-shaped patterns.The passage links lower-altitude hub emissions to local air quality and cruise emissions to competing ozone-formation and methane-destruction pathways with a small net cooling effect.Source: Emissions inventory · p. 14, Figure 3 · p. 15
  • Across latitudes, cruise-altitude NOx forms a broad concentration band, while lower-altitude emissions concentrate near hubs at 33°–48°N.Figure 4 complements the longitude-latitude emissions map in Figure 3.Source: Emissions inventory · p. 15, Figure 4 · p. 15

Social costs and health impact

  • NOx accounts for 99.7% of monetized air-quality and health costs, while its direct climate effect is a small net cooling.The monetized totals are $73.5 million (67.9%) for climate impacts and $34.7 million (32.1%) for air-quality and health impacts.Source: Social costs and health impact · pp. 15–16, Table 5 · p. 16

DISCUSSION AND CONCLUSION

For west coast domestic aviation, the study presents a 3-D emissions map and monetized impacts based on ADS-B trajectories reconstructed with its hybrid autoencoder. CO2 contributes $74.9 million in climate damage, while NOx provides a $1.35 million net climate credit and contributes $34.6 million to air-quality and health impacts; the authors identify broader geographic coverage and altitude-specific species costs as future work.

DISCUSSION AND CONCLUSION

  • The study presents a 3-D emissions map using imputed ADS-B trajectories to map social and health costs for domestic flights in the US west coast area.The authors propose future work on altitude-based species costs and a larger US study area.Source: DISCUSSION AND CONCLUSION · pp. 16–17
  • CO2 emissions contribute $74.9 million to climate damage, while NOx provides a $1.35 million net climate credit and contributes $34.6 million to air-quality and health impacts.The majority of NOx air-quality and health impacts comes from cruise-phase emissions, making NOx the largest driver of monetized health cost in the inventory.Source: DISCUSSION AND CONCLUSION · p. 16
  • The proposed CBiG3D-AE outperforms both a vanilla TCN-AE and cubic-spline interpolation, and the authors state that imputation accuracy directly affects inventory and cost estimates.The model combines CNN-Bi-GRU and 3D-CNN components with a TCN autoencoder decoder.Source: DISCUSSION AND CONCLUSION · p. 17
Loading 2609.31686v1…