Automated Forest Inventory: Comparing Drone-based Approaches for Determining Tree Species and Tree Volumes

Abstract

This thesis investigates the accuracy and practical viability of drone-based approaches for automated Forest Inventory, with special emphasis on tree species classification and wood volume estimation. High-resolution aerial imagery is captured across three distinct Austrian forest sample areas using a UAV equipped with an RGB camera and RTK positioning system. The acquired imagery is processed through photogrammetry to generate detailed 3D point clouds, which form the foundation for subsequent analysis. A semi-automated multi-stage processing pipeline is developed and implemented, incorporating semantic and instance segmentation for individual trees, tree species classification using the DetailView model, and volume estimation through a novel hybrid approach that combines geometric modelling with empirical volume-estimation equations. This integrated methodology enables end-to-end processing from raw aerial imagery to quantitative forest inventory metrics. The experimental evaluation demonstrates that drone-based method accuracy is critically dependent on point cloud quality, which varies significantly based on flight parameters, environmental conditions, and underlying forest structure complexity. Tree species classification achieves accuracies ranging from 40 % to 83.3 % across different forest types, while the hybrid volume estimation approach yields mean errors between 15.6 % and 25 % when validated against ground-truth measurements. The research establishes that drone-based photogrammetry represents a promising and cost-effective alternative to conventional forest inventory methods, though complete automation remains constrained by requirements for manual intervention during data acquisition and quality assurance. While the developed pipeline provides a robust foundational framework, the thesis identifies that enhanced algorithm robustness – particularly for dense and structurally complex forest environments – constitutes a critical area for future research to achieve precision levels comparable to established ground-based inventory techniques.

Date of Award	2025
Original language	English
Supervisor	Herwig Mayr (Supervisor)