An Interactive GEE Application for Burned Area Mapping Using ΔNBR Thresholding and Machine Learning: A Case Study from the 2025 Bilecik–Sakarya Wildfire, Türkiye

Accurate and timely burned-area mapping is essential for supporting post-fire assessment, ecological monitoring, and disaster response. In this study, an interactive Google Earth Engine (GEE) application was developed to provide a user-friendly, no-code environment for detecting burned areas using multiple approaches, including ΔNBR (Difference Normalized Burn Ratio) thresholding, unsupervised classification, and a Random Forest (RF) supervised model. The 2025 Bilecik–Sakarya wildfire in northwestern Türkiye was selected as a case study to evaluate the performance of these methods and demonstrate the operational capabilities of the application. Sentinel-2 pre- and post-fire composites were processed within the application to compute ΔNBR, train spectral classifiers, and generate burned-area maps. A comprehensive accuracy assessment using 5,000 stratified reference samples showed notable variation among the methods. The ΔNBR Threshold of 0.20 emerged as the best-performing approach, achieving the highest accuracy with an Overall Accuracy of 0.917, Precision of 0.930, Recall of 0.902, F1-score of 0.916, and Kappa of 0.834. Unsupervised methods (Otsu and K-means) performed comparably, while the RF classifier, despite high precision, underestimated burned pixels due to lower recall. Burned-area extent comparisons further revealed the strong sensitivity of ΔNBR results to threshold selection, underscoring the importance of flexible and visually supported method selection. The proposed GEE application integrates data preprocessing, ΔNBR computation, classification, accuracy assessment, and export functionality within a single guided interface, making advanced remote-sensing techniques accessible to non-expert users. The tool proved effective for rapid post-fire mapping and offers a scalable framework for future developments, including radar integration, additional machine-learning algorithms, and multi-temporal fire severity analysis. Overall, the application represents a practical and accessible solution for operational burned-area monitoring, with strong potential for use in wildfire-prone regions globally