Assessment of forest fire effects based on automated processing of Earth remote sensing imager

The article presents a study of the automated detection specifics within forest-covered areas traversed by fires based on the different time satellite imagery from the Sentinel-2A and Sentinel-2B using the differential normalised burn ratio index (dNBR) for the pre-fire and post-fire periods the calculation. The studies carried out on the research topic are given and a review of the currently functioning forest fire monitoring systems has been implemented. The urgency of the development and testing of an automated system for assessing the forest fire consequences using open source software and Earth remote sensing data has been substantiated. It has been established that the differential index dNBR, calculated from the Sentinel-2A and Sentinel-2B satellite images captured on different dates makes it possible to effectively detect burned-out areas. It is shown that the Python ecosystem makes it possible to successfully create systems for automated processing of Earth remote sensing data. A prototype of a system for the automated detection of forest-covered areas traversed by fires has been developed, based on the materials of different dates satellite imagery from Sentinel-2A and Sentinel-2B spacecraft. The flowchart of the algorithm of processing Earth remote sensing data using the proposed system was presented. For the Sentinel-2 satellite images for the dates before and after the fire, the differential index dNBR was calculated, the analysis of the results of which showed a close correlation of the dNBR index with the degree of burnout of the territory. A schematic map of the areas affected by the fire has been drawn up and the accuracy of identifying burnedout areas has been assessed by calculating the confusion matrix. An assessment of the effectiveness of the automated system for identifying areas affected by forest fires, ways of its modernisation and improvement, as well as the prospects for implementation in production has been carried out. It is noted that the results of the created system have high reliability indicators. At the same time, the need was revealed to increase the sensitivity of the system when identifying areas that have undergone partial burnout. A variant of improving the algorithms used in the work by introducing the multilevel Otsu’s method, intended to significantly increase the sensitivity of the system, has been proposed.