Machine learning-based forest fire susceptibility mapping of Southern Mizoram, a part of Indo-Burma Biodiversity Hotspot
- 18 June 2025
- journal article
- research article
- Published by Springer Nature in Environmental Science and Pollution Research
- Vol. 32 (59) , 31433-31454
Abstract
Forest fires are a significant global environmental hazard, causing widespread economic losses and ecological damage to natural habitats. Biodiversity-rich regions like Mizoram, a northeastern Indian state known for its lush forests and a part of Indo-Burma Biodiversity Hotspot, are particularly vulnerable to these fires. Between 2012 and 2021, Mizoram incurred losses amounting to approximately $8,910,000 USD due to wildfires. This study addresses the urgent need for high-resolution forest fire susceptibility mapping for southern Mizoram (Lunglei, Lawngtlai, Serchhip, and Tlabung), highlighting the region’s ecological fragility and vulnerability. We employed six machine learning (ML) algorithms—AdaBoost, Decision Tree, Gaussian Process, K-Nearest Neighbor, Random Forest, and Support Vector Machine and analyzed ten wildfire conditioning factors. These factors include topographical elements (DEM, slope, aspect, curvature, TWI), vegetation indices (pre-fire EVI, pre-fire VARI), anthropogenic factors (LULC), and solar radiation. A forest fire inventory was created using high-resolution satellite images from April 2021 through visual manual interpretation. Feature importance analysis using Gini Impurity revealed that pre-fire NDMI, EVI, DEM, aspect, and solar radiation were the most significant contributors. Performance metrics such as average accuracy, precision, recall, F1-score, area under the curve (AUC), and G-mean were used to evaluate the ML algorithms. AUC values ranged from 0.84 to 0.91, with accuracy scores between 0.74 and 0.81. Among the models, the Random Forest algorithm demonstrated the best performance across all metrics. Lawngtlai exhibited the highest susceptibility (64%, 869.66 km2), followed by Tlabung (38%, 956.09 km2), Lunglei (27%, 556.57 km2), and Serchhip (21%, 21.72 km2). Overall, 37.01% (2677.21 km2) of the study area was classified as highly susceptible. Our analysis further indicates that lower elevations and specific aspect orientations—namely East, Southeast, Southwest, and South—substantially influence forest fire susceptibility. Finally, the forest fire susceptibility map was validated using high-resolution Planet images. This study demonstrates that ML-based susceptibility estimation can be used to implement effective natural resource management and proactive measures to mitigate the environmental impact of forest fires.Keywords
This publication has 70 references indexed in Scilit:
- Demi-decadal land use land cover change analysis of Mizoram, India, with topographic correction using machine learning algorithmEnvironmental Science and Pollution Research, 2024
- Machine Learning with scikit-learn2023
- A performance comparison of machine learning models for wildfire occurrence risk prediction in the Brazilian Federal District regionEnvironment Systems and Decisions, 2023
- Understanding Forest Fire Patterns and Risk Assessment to Local Biodiversity of Selected Districts of Mizoram Using GIS and Remote SensingProceedings of the Zoological Society, 2023
- Flood, landslides, forest fire, and earthquake susceptibility maps using machine learning techniques and their combinationNatural Hazards, 2023
- Machine learning based forest fire susceptibility assessment of Manavgat district (Antalya), TurkeyEarth Science Informatics, 2023
- Forest Fire Susceptibility Modeling Using a Convolutional Neural Network for Yunnan Province of ChinaInternational Journal of Disaster Risk Science, 2019
- Modeling forest fire risk in the northeast of Iran using remote sensing and GIS techniquesNatural Hazards, 2012
- Assessment of multifaceted environmental issues and model development of an Indo-Burma hotspot regionEnvironmental Monitoring and Assessment, 2011
- The Nature of Statistical Learning Theory1995