~/aixsci
weekly updates

climate-weather/ai in a supporting role/Toxics 2025 · v2

Machine learning used to predict fine particle emissions at a Greek regional airport

Researchers calculated monthly aircraft emissions at Alexandroupolis Regional Airport for 2019 and 2020, then fitted five regression algorithms to predict fine particle levels from the other pollutants, traffic and weather, and to rank which inputs mattered most.

1. Collect air traffic, fleet and meteorological data2. Compute pollutant emissions from LTO emission factors3. Describe and correlate emission and weather variables4. Fit regression models for PM2.55. Evaluate and compare models on error metrics6. Rank feature importance of the best model

spectrum · one line per step, placed by what the step does · bright lines used AI

Assessing the Impact of Aviation Emissions on Air Quality at a Regional Greek Airport Using Machine Learning
Toxics, 2025

doi:10.3390/toxics13030217 · record aix-00293 v2 · checked 2026-11-05

ai-supportingrole of AI
AI was for
Property prediction
Model family
Linear model, Gradient-boosted trees, Random forest, Multilayer perceptron
Checked by
Held-out
Code
not reported

AI processed or interpreted data, but the main finding does not rest on it.

read as

The science is explained before the AI appears. Switch to field specialist to go straight to the method.

Assumes the discipline and goes straight to the method.

The diagram, the record and what the paper did not report are identical in both modes. Only the framing changes, never the evidence.

Introduction by AIxSci · plain language

What this research was about

Aircraft burn most of their fuel in the air, but a share of their pollution is released at ground level, during the taxiing, take-off, climb and landing that make up what emissions accounting calls the landing and take-off cycle. That part matters for the air people nearby actually breathe. It is hard to pin down because nobody can meter the exhaust of every aeroplane. Instead, analysts count movements, note which aircraft types flew, and multiply by standard per-aircraft emission factors published in official inventory guidebooks. Weather then shapes what happens next, since temperature, rain, sunshine and wind all affect how pollutants disperse or linger.

The team looked at a single regional airport in Greece, Alexandroupolis, across 2019 and 2020. They took air traffic and fleet records from the Hellenic Civil Aviation Authority, applied the standard landing and take-off emission factor method to estimate monthly carbon dioxide, nitrogen oxides, carbon monoxide, hydrocarbons, sulphur oxides, fine particles (PM2.5) and fuel use, and joined these to monthly weather averages from a nearby station, giving a dataset of 168 measurements. Emissions were lower in 2020, with carbon dioxide down 28.1 per cent, nitrogen oxides down 26.5 per cent and PM2.5 down 35.4 per cent, which the authors attribute to COVID-19 travel restrictions.

Where AI came in

The emissions figures themselves came from the conventional inventory calculation, not from a learned model. Machine learning was applied on top of that table. Five supervised regression algorithms, meaning methods trained on examples to predict a number, were fitted in Microsoft Azure Machine Learning Studio Classic to predict PM2.5 from the remaining pollutants, fuel consumption, traffic volume and weather variables: Bayesian Linear Regression, Linear Regression, Boosted Decision Tree Regression, Decision Forest Regression and Neural Network Regression. They were compared using standard error measures and the coefficient of determination, a score for how much of the variation a model accounts for. Bayesian Linear Regression and Linear Regression reported coefficients of 0.96 and 0.97.

The best-performing model was then examined with permutation importance, a technique that shuffles one input variable at a time and watches how much the predictions worsen. A large drop suggests the model leaned heavily on that variable. This produced a ranking of how much each pollutant, activity and weather measure contributed to the PM2.5 predictions. The models stood in for an explicit statistical relationship between the airport's activity, the weather and its fine particle output, one that would otherwise have to be written down by hand.

Written by AIxSci from the checked record below, to give context for readers outside the field. It is not part of the record.

The work

Technical · from the record

Monthly aviation emissions at Alexandroupolis Regional Airport for 2019 and 2020 were calculated from Civil Aviation Authority traffic records using the standard LTO emission-factor methodology, and combined with monthly meteorological averages into a dataset of 168 measurements. Five regression algorithms were then fitted to predict PM2.5 from the other pollutants, fuel consumption, traffic volume and weather variables; Bayesian Linear Regression and Linear Regression reported coefficients of determination of 0.96 and 0.97 respectively. Emissions were lower in 2020 than in 2019, with CO2 down 28.1%, NOx down 26.5% and PM2.5 down 35.4%, which the authors attribute to COVID-19 travel restrictions. Carbon dioxide accounted for 75.5% of total emissions in 2019 and fuel for 24%.

How AI was used

After emissions were computed by a conventional inventory calculation, machine learning was applied to the resulting table of monthly emissions, fuel consumption, total traffic and meteorological averages. Five supervised regression algorithms (Bayesian Linear Regression, Linear Regression, Boosted Decision Tree Regression, Decision Forest Regression and Neural Network Regression) were fitted in Microsoft Azure Machine Learning Studio Classic to predict PM2.5 from the remaining pollutant, activity and weather variables. Models were compared using mean absolute error, root mean square error, relative absolute error, relative squared error and the coefficient of determination, and the Permutation Importance method was then applied to the best-performing model to rank the contribution of each input variable to its PM2.5 predictions.

The shape of the work

Structural · the record, drawn

ACQUISITIONPREPARATIONINTERPRETATIONTRAININGVALIDATIONINTERPRETATION123456AIAICollect airtraffic, fleetand meteorologic…Compute pollutantemissions fromLTO emission fac…Describe andcorrelateemission and wea…Fit regressionmodels for PM2.5Evaluate andcompare models onerror metricsRank featureimportance of thebest model
AI stepNo AI↤ what the AI stood in for

Collect air traffic, fleet and meteorological data

Obtaining raw data, whether by measurement, download or retrieval.

The Hellenic Civil Aviation Authority (CAA) collected all air traffic and fleet composition data.where the paper describes this · verbatim
in the paper

Compute pollutant emissions from LTO emission factors

Cleaning, filtering, normalising or labelling data already obtained.

The emissions have been calculated using each aircraft’s emission factors, following the standard LTO emission factor methodologywhere the paper describes this · verbatim
in the paper

Describe and correlate emission and weather variables

Extracting understanding from model behaviour.

Descriptive statistics and the Pearson correlation coefficient were applied to the meteorological and pollutant variables at a 0.01 confidence levelwhere the paper describes this · verbatim
in the paper

Fit regression models for PM2.5

Fitting model parameters, including fine-tuning an existing model.

an attempt was made to predict the emissions levels of PM2.5 using various machine learning methodswhere the paper describes this · verbatim
in the paper

Evaluate and compare models on error metrics

Testing outputs against ground truth.

Coefficient of Determination (R), were used to assess all the machine learning methodswhere the paper describes this · verbatim
in the paper

Rank feature importance of the best model

Extracting understanding from model behaviour.

the Permutation Importance method (PIM) was used to interpret the best machine learning model that will resultwhere the paper describes this · verbatim
in the paper

What the record says

Technical · every part carries its own basis

+ in the paper~ our reading− not reported

How to read the quotations. A quotation shows where the paper describes something. It does not quote every value beside it: one passage locates a part of the work, and values without their own quotation are our reading of that passage.

~Role of AI
AI in a supporting roleour reading

Emissions themselves were computed with the EMEP/EEA LTO emission-factor methodology, not with a learned model; machine learning was applied on top of that dataset to predict PM2.5 and to rank feature importance. The reported emission reductions and correlations do not depend on the models, although the model-comparison metrics are themselves a reported result.

+What the AI was for
Five basic algorithms, Bayesian Linear Regression, Boosted Decision Tree, Linear Regression, Decision Forest Regression, and Neural Network Regression, were used for regression analysiswhere the paper describes this · verbatim
+How it was taught
Supervisedin the paper
+Models named
Bayesian Linear Regression · Trained from scratchLinear Regression · Trained from scratchBoosted Decision Tree Regression · Trained from scratchDecision Forest Regression · Trained from scratchNeural Network Regression · Trained from scratchin the paper
~How results were checked
Held-outour reading
+Code · weights · data
code not reportedweights not reporteddata not reportedin the paper
The data presented in this study are available on request from the corresponding author.where the paper describes this · verbatim
+Compute
not reportedin the paper

What this paper did not report

Technical · absence is published deliberately

Reported as not stated: 12 items
  • CodeWhether the code is available is not stated.
  • Trained model weightsWhether the trained model is available is not stated.
  • DataWhether the data are available is not stated.
  • ComputeThe hardware or time used is not stated.
  • How many were testedThe paper gives no count of what was tested.
  • Version of Bayesian Linear RegressionWhich version of the model was used is not stated.
  • Version of Linear RegressionWhich version of the model was used is not stated.
  • Version of Boosted Decision Tree RegressionWhich version of the model was used is not stated.
  • Version of Decision Forest RegressionWhich version of the model was used is not stated.
  • Version of Neural Network RegressionWhich version of the model was used is not stated.
  • What step 4 replacedThe paper gives no basis for what the AI stood in for.
  • What step 6 replacedThe paper gives no basis for what the AI stood in for.

About this article

Record aix-00293, version 2, checked by a person on 2026-11-05. The record describes the paper; it does not assess whether the paper's findings are right. The paper is published under CC-BY; quotations are at most 25 words. How we work · Report an error

Cite this record: AIxSci (2026). Machine learning used to predict fine particle emissions at a Greek regional airport. Record aix-00293, version 2, checked 2026-11-05. https://aixsci.org/articles/aix-00293/