|
In recent years, the Kurdo-Zagrosian mountains in western Iran and northern Iraq have faced numerous wildfire fires. Mapping forest fire susceptibility is crucial for several reasons, including its role in prevention and mitigation, resource allocation, ecological conservation, early warning systems, policy development, insurance and risk management, and wildfire risk mapping. Machine Learning (ML) has found numerous applications in remote sensing, including fire detection, severity assessment, fuel moisture content estimation, fire spread prediction, fire susceptibility mapping, smoke plume detection, air quality monitoring, post-fire recovery monitoring, and decision support systems for fire management. This study employs a new approach to leveraging Non-negative Matrix Factorization (NMF) for detecting fire-susceptible areas in the Kurdo-Zagrosian forests of Marivan and Sarvabad in Kurdistan Province, western Iran. The NMF is a ML method used for dimensionality reduction and feature extraction. NMF differs from traditional matrix factorization methods by enforcing non-negativity constraints on the factor matrices, making the resulting factors interpretable and often more suitable for real-world data analysis. Sentinel-2 satellite imagery, elevation, distance to the road network, and Zagros Grass Index (ZGI) have been used as the primary inputs of the model, combined with in situ data for verifying and interpreting the resulting maps. The results showed that, besides providing useful information in extracting fire susceptible areas, NMF handles wide study areas efficiently, especially for tasks like feature extraction from large-scale datasets such as satellite images or multispectral data. The results especially revealed that ZGI has specifically demonstrated improved accuracy and reliability. The resulting map also showed a very close overlap between the fired area provided by Sentinel imagery from 2021 to 2023 and the areas labeled as highly susceptible regions in 2020, especially when ZGI has been regarded between the input variables.
|
