Improving Deterministic Air Quality Forecasts Using Supervised Machine Learning: A Feasibility Study

Lech Łobocki

doi:10.2478/oszn-2025-0018

.blurhash-client-img { display: none !important; }

Improving Deterministic Air Quality Forecasts Using Supervised Machine Learning: A Feasibility Study

Environmental Protection and Natural Resources

Volume 36 (2025): Issue 3 (September 2025)

By: Lech Łobocki

Open Access

|Dec 2025

Figures & Tables

Location of the air-quality monitoring stations used in this study. Model grid is marked with grey lines. Station codes are explained in Table 1

GEM-AQ predictions vs. observations at six air quality monitoring stations in Warsaw, during the test period in 2024

The hybrid deterministic-SML predictions vs. observations at six air quality monitoring stations in Warsaw, during the test period in 2024. All the features listed in Section II were used for training

Residuals (errors) histograms in the hybrid deterministic-SML predictions at six air-quality monitoring stations in Warsaw during the test period in 2024

Residuals (errors) histograms in the deterministic (GEM-AQ) predictions at six air-quality monitoring stations in Warsaw during the test period in 2024

Quantile-Quantile comparison of distributions of the hybrid deterministic-SML predictions vs. observations at six air quality monitoring stations in Warsaw, during the test period in 2024

Quantile-Quantile comparison of distributions of the deterministic GEM-AQ predictions vs. observations at six air quality monitoring stations in Warsaw, during the test period in 2024

Observed and predicted PM2.5 concentrations during the first 300 hours of the test period. OBS, observations; P-ML, prediction using hybrid deterministic-SML; P-D, deterministic forecast only

Location of air quality monitoring stations used in this study*

Station Code	Address	Geographic Coordinates (WGS84)
Station Code	Address	Longitude [°E]	Latitude [°N]
PL0140A	Warszawa, Al. Niepodległości 227/233	21.004724	52.219298
PL0141A	Warszawa, ul. Wokalna 1	21.033819	52.160772
PL0143A	Warszawa, ul. Kondratowicza 8	21.042458	52.290864
PL0308A	Warszawa, ul. Tołstoja 2	20.933018	52.285073
PL0717A	Warszawa, ul. Bajkowa 17/21	21.176233	52.188474
PL0739A	Warszawa, ul. Chróścickiego 16/18	20.906073	52.207742

Performance metrics of statistical forecasts using RF regression model, without consideration of meteorological variables in RF*

Station ID	Location	RMSE	MAE	CoD (R²)	PCC (r)
PL0140A	Al. Niepodległości	0.0592	0.0099	0.99996	0.999982
PL0141A	ul. Wokalna	0.0397	0.0066	0.99998	0.999991
PL0143A	ul. Kondratowicza	0.0287	0.0059	0.99999	0.999995
PL0308A	ul. Tołstoja	0.0382	0.0089	0.99999	0.999995
PL0717A	ul. Bajkowa	0.0903	0.0179	0.99995	0.999975
PL0739A	ul. Chróścickiego	0.2305	0.0161	0.99937	0.999693

Performance metrics of deterministic-statistical forecasts using the GB regression model*

Station ID	Location	RMSE	MAE	CoD (R²)	PCC (r)
PL0140A	Al. Niepodległości	0.133	0.085	0.9998	0.9999
PL0141A	ul. Wokalna	0.115	0.069	0.9998	0.9999
PL0143A	ul. Kondratowicza	0.111	0.070	0.9999	0.9999
PL0308A	ul. Tołstoja	0.148	0.099	0.9999	0.9999
PL0717A	ul. Bajkowa	0.155	0.098	0.9999	0.9999
PL0739A	ul. Chróścickiego	0.186	0.080	0.9996	0.9998

Performance metrics of statistical forecasts using RF regression model, based upon the deterministic meteorological forecast (without consideration of the model-predicted PM2_5 concentration)*

Station ID	Location	RMSE	MAE	CoD (R²)	PCC (r)
PL0140A	Al. Niepodległości	9.12	6.70	0.154	0.461
PL0141A	ul. Wokalna	8.29	6.19	0.198	0.497
PL0143A	ul. Kondratowicza	8.91	6.40	0.070	0.471
PL0308A	ul. Tołstoja	11.64	8.59	0.129	0.464
PL0717A	ul. Bajkowa	11.39	7.90	0.198	0.547
PL0739A	ul. Chróścickiego	8.890	6.43	0.066	0.496

Performance metrics of the deterministic forecasts done using GEM-AQ*

Station ID	Location	RMSE	MAE	CoD (R²)	PCC (r)
PL0140A	Al. Niepodległości	9.15	6.19	0.0585	0.5152
PL0141A	ul. Wokalna	8.39	5.86	0.0855	0.5236
PL0143A	ul. Kondratowicza	8.92	5.94	0.0325	0.5497
PL0308A	ul. Tołstoja	11.5	8.11	0.0978	0.5343
PL0717A	ul. Bajkowa	9.11	5.94	0.2719	0.6071
PL0739A	ul. Chróścickiego	8.30	5.54	0.0205	0.6077

Performance metrics of deterministic-statistical forecasts using the XGBoost regression model*

Station ID	Location	RMSE	MAE	CoD (R²)	PCC (r)
PL0140A	Al. Niepodległości	0.432	0.137	0.9981	0.9991
PL0141A	ul. Wokalna	0.395	0.122	0.9982	0.9991
PL0143A	ul. Kondratowicza	0.472	0.117	0.9974	0.9987
PL0308A	ul. Tołstoja	0.444	0.151	0.9987	0.9994
PL0717A	ul. Bajkowa	0.951	0.295	0.9944	0.9972
PL0739A	ul. Chróścickiego	0.510	0.189	0.9969	0.9985

Performance metrics of deterministic-statistical forecasts using the RF regression model*

Station ID	Location	RMSE	MAE	CoD (R²)	PCC (r)
PL0140A	Al. Niepodległości	0.0712	0.0123	0.99995	0.99997
PL0141A	ul. Wokalna	0.0574	0.0096	0.99996	0.99998
PL0143A	ul. Kondratowicza	0.0362	0.0079	0.99998	0.99999
PL0308A	ul. Tołstoja	0.0534	0.0132	0.99998	0.99999
PL0717A	ul. Bajkowa	0.0936	0.0216	0.99995	0.99997
PL0739A	ul. Chróścickiego	0.2685	0.0199	0.99915	0.99958

Performance metrics of deterministic-statistical forecasts using the SVR model*

Station ID	Location	RMSE	MAE	CoD (R²)	PCC (r)
PL0140A	Al. Niepodległości	1.339	0.471	0.9818	0.9913
PL0141A	ul. Wokalna	1.445	0.503	0.9755	0.9885
PL0143A	ul. Kondratowicza	1.219	0.432	0.9826	0.9914
PL0308A	ul. Tołstoja	2.060	0.675	0.9727	0.9867
PL0717A	ul. Bajkowa	1.791	0.660	0.9802	0.9909
PL0739A	ul. Chróścickiego	1.614	0.525	0.9692	0.9853

References

Authors

Metrics

Articles in this issue

DOI: https://doi.org/10.2478/oszn-2025-0018 | Journal eISSN: 2353-8589 | Journal ISSN: 1230-7831

Journal RSS Feed

Language: English

Page range: 15 - 29

Published on: Dec 31, 2025

Published by: National Research Institute, Institute of Environmental Protection

In partnership with: Paradigm Publishing Services

Publication frequency: 4 issues per year

Keywords:

air quality,

air quality prediction,

supervised machine learning,

deterministic-statistical forecast

Related subjects:

Life sciences,

Ecology

© 2025 Lech Łobocki, published by National Research Institute, Institute of Environmental Protection
This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 3.0 License.

Volume 36 (2025): Issue 3 (September 2025)

Improving Deterministic Air Quality Forecasts Using Supervised Machine Learning: A Feasibility Study

Figures & Tables

Figure 1.

Figure 2.

Figure 3.

Figure 4.

Figure 5.

Figure 6.

Figure 7.

Figure 8.

Location of air quality monitoring stations used in this study*

Performance metrics of statistical forecasts using RF regression model, without consideration of meteorological variables in RF*

Performance metrics of deterministic-statistical forecasts using the GB regression model*

Performance metrics of statistical forecasts using RF regression model, based upon the deterministic meteorological forecast (without consideration of the model-predicted PM2_5 concentration)*

Performance metrics of the deterministic forecasts done using GEM-AQ*

Performance metrics of deterministic-statistical forecasts using the XGBoost regression model*

Performance metrics of deterministic-statistical forecasts using the RF regression model*

Performance metrics of deterministic-statistical forecasts using the SVR model*

Paradigm

My account