Analyzing Post-fire Vegetation Dynamics with Ultra-high Resolution Remote Sensing Data

Abstract. Monitoring post-fire vegetation dynamics is essential for understanding forest recovery processes and informing management strategies. UAV-based ultra-high resolution multi-temporal imagery, combined with the Structure-from-Motion andMulti-View Stereo (SfM-MVS) workflow, provides a cost-effective and scalable solution for forest monitoring. However, challenges remain in co-aligning multi-temporal datasets, segmenting individual trees in dense canopies, and ensuring classification accuracy. This study presents a comprehensive workflow for post-fire forest monitoring using UAV imagery, covering data acquisition, co-alignment, tree segmentation, species classification, and biophysical parameter estimation using growth models. The workflow was tested on three sites in Central Yakutia, with varying post-fire regeneration scenarios. Co-alignment was applied to multi-temporal UAV datasets, and tree segmentation was performed using the algorithms developed for Airborne Laser Scanning (ASL) forest point clouds. Tree species classification relied on statistical spatial variables of point clouds, and growth models were used to estimate parameters such as tree height, age, canopy area, above-ground biomass, and net primary productivity. The results demonstrated that co-alignment enabled consistent multi-temporal analysis, but performance was sensitive to flight planning consistency and lighting conditions. Tree segmentation accuracy was high in open-canopy areas but decreased in dense canopies. The classification of larch and birch species achieved relatively high precision and recall values, while dead trees showed lower classification accuracy due to challenging lighting conditions. Growth models successfully estimated biophysical parameters, but further validation using dendrochronological methods is required. This study highlights the potential of UAV-based multi-temporal monitoring for post-fire forest assessment. Future research should focus on improving tree segmentation of SfM-MVS point clouds in dense canopies, optimizing co-alignment under varying environmental conditions, and integrating additional point cloud classification methods to improve accuracy in areas with complex species distribution.