Saruulzaya et al. (2026) Mapping of peatland in Mongolia using machine learning

Identification

• Journal: Ecological Indicators
• Year: 2026
• Date: 2026-09-23
• Authors: Adiya Saruulzaya, Ariunbold Maralmaa, Yondonrentsen Purevdulam, Kurt Fesenmyer, Sara Leavitt, Avidsuren Erdenechimeg, Xiaodong Wu, Tonghua Wu, Gankhuyag Nemekhbayar, Ulgiichimeg Ganzorig, Byambadorj Dolgor
• DOI: 10.1016/j.ecolind.2026.115550

Research Groups

• Institute of Geography and Geoecology, Mongolian Academy of Sciences
• University of Mongolian Academy of Sciences
• The Nature Conservancy, Arlington, Virginia, USA
• The Nature Conservancy, Mongolia
• Cryosphere Research Station on the Qinghai-Tibet Plateau, Key Laboratory of Cryospheric Science and Frozen Soil Engineering, Northwest Institute of Eco-Environment and Resources, Chinese Academy of Sciences
• Department of Environment and Forest Engineering, School of Engineering and Technology, National University of Mongolia

Short Summary

This study aims to develop a high-resolution peatland distribution map for Mongolia using machine learning techniques and field survey data. The research predicts soil organic carbon content (SOCC) distribution using a random forest model, achieving an accuracy of 89% with the most important predictors being temperature, silt content, vegetation productivity, and moisture availability.

Objective

• Identify key environmental indicators controlling SOCC distribution
• Predict the SOCC distribution using the RF model
• Create a detailed map of peatland distribution based on threshold values of peat soil
• Quantify the uncertainty map to improve the reliability and robustness of peatland distribution estimates

Study Configuration

• Spatial Scale: National scale, with a spatial resolution of 250 m
• Temporal Scale: Long-term data from 1970 to 2023 were used for climate and environmental covariates, while field survey data was collected in 2022 and 2023

Methodology and Data

• Models used: Random Forest (RF) model with Recursive Feature Elimination (RFE)
• Data sources:
  • Field soil observations from 5,400 samples at 1,242 sites
  • Satellite data from Landsat 8/OLI imagery for thermal infrared band (Band 10), optical bands, and vegetation indices
  • Climate data from WorldClim 2.1 monthly dataset (1970-2000)
  • Terrain parameters derived from Digital Surface Model (DSM) from ALOS World 3D datasets

Main Results

• The RF model achieved an accuracy of 89% in predicting SOCC distribution, with the most important predictors being temperature, silt content, vegetation productivity, and moisture availability.
• The study identified 260 sites as peat soils and 982 sites as non-peat soils based on threshold values for peat soils of ≥120 g kg−1 SOCC (12%).
• The uncertainty map was generated using Quantile Regression Forest (QRF), capturing approximately 80% of the SOCC prediction intervals.

Contributions

• This study provides a high-resolution peatland distribution map for Mongolia, addressing key gaps in previous studies.
• The research integrates machine learning techniques and field survey data to improve the reliability and robustness of peatland distribution estimates.
• The study contributes to the development of robust ecological indicators for monitoring, conservation planning, and climate change mitigation assessments.

Funding

• This research was funded by [list projects, programs, and reference codes that funded this research]

Citation

@article{Saruulzaya2026Mapping,
  author = {Saruulzaya, Adiya and Maralmaa, Ariunbold and Purevdulam, Yondonrentsen and Fesenmyer, Kurt and Leavitt, Sara and Erdenechimeg, Avidsuren and Wu, Xiaodong and Wu, Tonghua and Nemekhbayar, Gankhuyag and Ganzorig, Ulgiichimeg and Dolgor, Byambadorj},
  title = {Mapping of peatland in Mongolia using machine learning},
  journal = {Ecological Indicators},
  year = {2026},
  doi = {10.1016/j.ecolind.2026.115550},
  url = {https://doi.org/10.1016/j.ecolind.2026.115550}
}