Detecting changes due to forest fire by using satellite images with different resolution: The example of Antalya, Kumluca

Forest fires are one of the major disasters that affect our planet and biological life. It affects not only the living life but also the economy, especially the region where it is located. In this study, it is aimed to determine the affected areas at the end of the fire by using different resolution satellite images with different change detection techniques. For this purpose, satellite images of IKONOS, Worldview-2, ASTER, Landsat-7 and Landsat-8 were used. The forest fire that occurred on 28 June 2014 in the Adrasan region of Kumluca district of Antalya was used in the change detection analyses carried out in this study. In the study, images with different spatial resolutions, Landsat-7 acquired on November 3, 2004, Landsat-8 acquired on July 3, 2014, IKONOS acquired on October 14, 2004, WorldView-2 acquired on July 5, 2014, ASTER acquired on November 4, 2004 and August 12, 2014 were used. In order to detect fire-induced changes in regions with the same area over a 10-year period, satellite data with different spatial resolutions were analyzed using various image analysis techniques and the results were comparatively investigated. Pre-fire and post-fire images were calibrated geometrically and radiometrically to each other. Then, difference image, image ratio, NDVI (normalized vegetation index) difference image, NDVI image ratio, PCA (Principal Component Analysis) and MNF (Minimum Noise Fraction) techniques were applied to near infrared (NIR) bands. Within the scope of the study, the data obtained from the reference satellite images and the images created as a result of median filtering were analyzed, and accuracy assessments were made by determining the median filtering value corresponding to the highest accuracy value and kappa coefficient. Within the scope of the study, the median filtering value corresponding to the highest accuracy value and kappa coefficient was determined by analyzing the data obtained from the reference satellite images and the images created as a result of median filtering. When all the analysis results were compared, it was seen that the method with the highest accuracy value and kappa coefficient was PCA. By PCA technique, the accuracy value corresponding to the median 3x3 kernel size of Landsat-7 and Landsat-8 satellite images is 96.5% kappa coefficient 0.73, the accuracy value corresponding to the median 97x97 kernel size of IKONOS and WV-2 satellite images is 94.7% kappa coefficient 0.73, the median 21x21 of the ASTER images The accuracy value corresponding to the kernel size was found as 99.1% kappa coefficient of 0.94. With the ASTER sensor with the highest kappa coefficient and accuracy, the fire area was calculated as 129.5 hectares.