EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Machine learning reveals controls on heavy metal distribution in war-affected soils of Eastern Ukraine

J Mandal, O Bonchkovskyi, P Ostapenko, S M Hutchinson

Environmental Pollution · 2026-12 · 411 · 129416

DOI: 10.1016/j.envpol.2026.129416 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Environmental Pollution, 2026-12. Verfasst von J Mandal und 3 weiteren Autor:innen. Beteiligte Einrichtung: School of Science, Engineering and Environment, University of Salford, Salford, M5 4WT, UK. Hauptthemen laut MEDLINE: Metals, Heavy, Soil Pollutants, Machine Learning, Environmental Monitoring. Die Arbeit nennt 48 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Armed conflict causes severe and lasting soil contamination, yet the relative contributions of military activities, geogenic factors and pre-war anthropogenic influences on heavy metal distributions in war-affected soils remain poorly understood. This study applies generalized additive models (GAM) and Random Forest (RF) algorithms to characterise the multi-factor controls on Cu, Pb, Ni and Zn concentrations in soils from eleven sites across Kharkiv Oblast, eastern Ukraine, affected by hostilities between March and November 2022. A total of 114 soil samples were collected from explosion craters, weapon-damaged fields and sites impacted by specific weapons. Predictor variables encompassed site disturbance category, ammunition type, soil type, land use, pH and soil organic carbon (SOC). Measured pseudo-total concentrations spanned wide ranges Cu 5.2-71.5, Pb 6.1-24.6, Ni 10.1-46.0 and Zn 0.43-114.2 mg kg-1 with the highest values confined to specific disturbance categories. Under repeated cross-validation, Random Forest performed comparably to GAM for Cu, Ni and Zn (RF R2 ≈ 0.82, 0.79 and 0.81, respectively) and significantly better only for Pb, improving mean R2 from 0.56 to 0.67 and reducing root mean square error (RMSE) from 2.60 to 2.06 mg kg-1; its overall advantage was modest and metal-specific. Two contrasting behavioral groups were identified: Cu and Zn were predominantly associated with indicators of military activity, with ammunition type and helicopter crash site category as the dominant predictors, indicating localized contamination from military activity; Pb and Ni were primarily soil-controlled, governed by pH, organic carbon and soil type rather than disturbance intensity. Random Forest resolves the Cu/Zn point-source signal (spatially random residuals) while residual structure remains for Pb and Ni. These findings demonstrate that non-linear, multi-factor modelling approaches are essential for disentangling war-related geochemical signals from background variability, and that contamination patterns in active conflict zones are markedly more heterogeneous than those reported from military training grounds. The results have important implications for post-war land assessment and soil reclamation strategies in Ukraine.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Metals, HeavySoil PollutantsMachine LearningEnvironmental MonitoringUkraineSoilRandom Forest

Stichwörter der Autor:innen

Generalized additive modelsHeavy metalsMachine learningRandom forestSoil contaminationUkraineWar-affected soils

Substanzen

Metals, HeavySoil PollutantsSoil

Autor:innen und Institutionen

  • School of Science, Engineering and Environment, University of Salford, Salford, M5 4WT, UK.
  • NGO "Ukrainian Researchers Society", Kyiv, Ukraine; Institute of Geography of the National Academy of Sciences of Ukraine, Kyiv, Ukraine.
  • NGO "Ukrainian Researchers Society", Kyiv, Ukraine; Taras Shevchenko National University of Kyiv, Kyiv, Ukraine.
  • School of Science, Engineering and Environment, University of Salford, Salford, M5 4WT, UK. Electronic address: S.M.Hutchinson@salford.ac.uk.

Zitieren

APA

Mandal, J., Bonchkovskyi, O., Ostapenko, P., & Hutchinson, S. M. (2026). Machine learning reveals controls on heavy metal distribution in war-affected soils of Eastern Ukraine. Environmental Pollution, 411, 129416. https://doi.org/10.1016/j.envpol.2026.129416

Vancouver

Mandal J, Bonchkovskyi O, Ostapenko P, Hutchinson SM. Machine learning reveals controls on heavy metal distribution in war-affected soils of Eastern Ukraine. Environmental Pollution. 2026;411:129416. doi:10.1016/j.envpol.2026.129416

BibTeX
@article{Mandal2026,
  author = {Mandal, J and Bonchkovskyi, O and Ostapenko, P and Hutchinson, S M},
  title = {Machine learning reveals controls on heavy metal distribution in war-affected soils of Eastern Ukraine},
  journal = {Environmental Pollution},
  year = {2026},
  volume = {411},
  pages = {129416},
  doi = {10.1016/j.envpol.2026.129416}
}
Kostenlos starten →