Abstract. Wildfires, intensified by climate change and anthropogenic factors, represent a major threat to ecosystems, infrastructure, and human life. Early detection is critical for effective prevention and mitigation. This study presents a Convolutional Neural Network (CNN)-based system designed to detect wildfires from Sentinel-2 satellite imagery. The methodology integrates advanced preprocessing techniques—including atmospheric correction, cloud masking, and spatial normalization—with a tailored CNN architecture optimized for spectral–spatial feature extraction. Training utilized a curated dataset of wildfire events in Türkiye (2016–2025), annotated at the pixel level to distinguish fire-affected and unaffected regions. The CNN model, employing LeakyReLU activation and AdamW optimization, achieved robust performance, with precision, recall, and F1-scores all averaging 0.97, and an overall accuracy of 97%. High fire-class precision (1.00) minimized false alarms, while strong recall (0.93) ensured reliable detection. The model’s lightweight design allows deployment on cloud platforms and edge devices, enabling real-time monitoring across forestry, agriculture, and urban planning applications. Analysis of Turkish wildfire statistics (1988–2023) revealed that over 90% of fires stem from human negligence or arson, underscoring the need for public education, stricter enforcement, and technology-driven surveillance. Long-term data further indicate a rising severity in fire events, strengthening the case for AI-enhanced early warning systems. This research demonstrates that CNNs applied to Sentinel-2 imagery offer a scalable, accurate, and cost-effective solution for wildfire detection. By bridging remote sensing and deep learning, the proposed system supports both immediate disaster response and long-term policy planning for wildfire risk reduction.
Karaş et al. (Tue,) studied this question.