Wildfire-Induced Burned Area Mapping in Dohezar Forest (Northern Iran) Using Sentinel-۲ Imagery and U-Net Deep Learning Model

Wildfires pose a significant threat to forest ecosystems, especially in ecologically sensitive areas such as the Dohezar Forest in northern Iran. Rapid and accurate detection of burned areas is essential for effective environmental monitoring and resource management. The primary objective of this study was to develop and evaluate a deep learning-based framework for accurately mapping burned areas in the Dohezar Forest using Sentinel-۲ imagery and the U-Net model. In this study, Sentinel-۲ satellite imagery with a ۱۰-meter spatial resolution was used to develop a labeled dataset of fire-affected regions before and after a major wildfire event. The U-Net deep learning architecture was employed for semantic segmentation of burned areas, with binary masks generated to highlight fire-impacted zones. The results revealed that the total burned area decreased from ۳۷,۷۹۳.۰۷ hectares in August ۲۰۲۳ to ۳۷,۴۹۸.۵۸ hectares in May ۲۰۲۴, indicating a net reduction of ۲۹۴.۴۹ hectares. The U-Net model successfully produced accurate and spatially coherent segmentation outputs, demonstrating its robustness in detecting fire boundaries across complex landscapes. This study confirms the potential of deep learning-based remote sensing approaches for monitoring wildfire dynamics and supporting data-driven forest management efforts in high-risk regions.

کلیدواژه ها:

Wildfire Detection ، U-Net ، Remote Sensing ، Deep Learning ، Sentinel-۲ ، Dohezar Forest