Sadra et al. (2026) Machine learning analysis of Iran’s wildfire landscape and anthropogenic influences

Identification

• Journal: Scientific Reports
• Year: 2026
• Date: 2026-01-06
• Authors: Nasim Sadra, Mohammad Reza Nikoo, Rouzbeh Nazari, Maryam Karimi, Md Galal Uddin, Amir H. Gandomi
• DOI: 10.1038/s41598-025-22387-3

Research Groups

• College of Sciences, School of Mathematical and Computational Sciences, Massey University, Palmerston North, New Zealand
• College of Engineering, Department of Civil and Architectural Engineering, Sultan Qaboos University, Muscat, Oman
• Department of Civil, Construction and Environmental Engineering, The University of Memphis, Memphis, United States
• School of Public Health, University of Memphis, Memphis, United States
• Department of Engineering and I.T., University of Technology Sydney, Ultimo, NSW, Australia
• Research and Innovation Center (EKIK), Óbuda University, Budapest, Hungary
• College of Science and Engineering, University of Galway, Galway, Ireland
• Department of Computer Science, Khazar University, Baku, Azerbaijan

Short Summary

This study analyzes Iran's wildfire occurrences from 2001 to 2022 using machine learning, revealing a 17-fold increase in frequency primarily driven by anthropogenic factors (CO₂ emissions) rather than climate variability (temperature), with significant spatial heterogeneity across the country.

Objective

• To provide fundamental information on wildfire trends during the study period in Iran.
• To identify which factor, human activities or climate variability, has a stronger influence on the observed changes in wildfire dynamics.

Study Configuration

• Spatial Scale: Entire country of Iran (approximately 1.64 million square kilometers).
• Temporal Scale: 2001 to 2022 (22 years).

Methodology and Data

• Models used:
  • K-means clustering (for delineating fire zones).
  • Random Forest regression (for analyzing relationships between drivers and wildfire occurrence).
  • Distance correlation (for lagged dependence analysis).
• Data sources:
  • NASA FIRMS’ active fire detections MCD14DL (Moderate Resolution Imaging Spectroradiometer (MODIS) sensor, 1 km spatial resolution).
  • World Bank (annual CO₂ emissions, average yearly temperature).
  • VIIRS (Visible Infrared Imaging Radiometer Suite) for multi-sensor cross-validation.
  • GADM database of Global Administrative Areas (national boundaries).
  • MODIS/Terra Vegetation Indices 16-Day L3 Global 500 m (MOD13A1) V061 (for Normalized Difference Vegetation Index (NDVI)).

Main Results

• Wildfire frequency in Iran escalated significantly from 2001 to 2022, increasing over 17 times (from 1,818 fires in 2001 to 31,112 in 2017).
• The period 2015–2018 was identified as critical, marked by heightened wildfire activity and rapid annual fluctuations, including a 73.89% increase from 2015 to 2016.
• Wildfire intensity (maximum brightness) showed a more moderate increase of approximately 25.9% (from 400.5 K in 2001 to 502.8 K in 2013).
• K-means clustering delineated ten fire zones; Zone 05 (southwest) was the smallest but most affected, recording 162,734 fires and the highest maximum brightness of 502.8 K.
• Random Forest regression models demonstrated a stronger correlation between CO₂ emissions and wildfire activity (R² = 0.85, RMSE = 2,079.66) compared to average temperature (R² = 0.73, RMSE = 2,858.60).
• Lagged dependence analysis indicated a peak distance correlation for CO₂ emissions at a +2-year lag (approximately 0.49) and for mean temperature at a +1-year lag (approximately 0.45).
• The predictive capability of the CO₂ model was observed to decline when emissions exceeded approximately 7 metric tons per capita.

Contributions

• Provides a comprehensive spatiotemporal analysis of wildfires in Iran over two decades, addressing a significant research gap in an understudied region.
• Quantitatively demonstrates that the wildfire crisis in Iran is primarily driven by an exponential increase in ignition frequency (anthropogenic factors) rather than intensity or climate variability alone.
• Introduces a multi-sensor cross-validation framework for satellite-based fire data in data-scarce regions, enhancing data reliability.
• Identifies specific fire-prone zones and critical periods, offering insights for targeted, zone-specific management strategies.
• Highlights the importance of considering CO₂ emission patterns and their lagged effects in wildfire management and planning.
• Challenges the common misconception that vegetation density is the most important cause of wildfires, emphasizing the role of humidity, human land use, and the rural-urban interface.

Funding

• Open access funding provided by Óbuda University.

Citation

@article{Sadra2026Machine,
  author = {Sadra, Nasim and Nikoo, Mohammad Reza and Nazari, Rouzbeh and Karimi, Maryam and Uddin, Md Galal and Gandomi, Amir H.},
  title = {Machine learning analysis of Iran’s wildfire landscape and anthropogenic influences},
  journal = {Scientific Reports},
  year = {2026},
  doi = {10.1038/s41598-025-22387-3},
  url = {https://doi.org/10.1038/s41598-025-22387-3}
}