Epsilon Archive for Student Projects

Abstract

Prescribed burning is a widely used management tool to reduce wildfire hazard, restore (fire-adapted) ecology and enhance ecosystem resilience. Detailed insight into vegetation fuels is crucial for safe and successful controlled burning, but also an important part for understanding ecological processes. After prescribed burning, careful monitoring of ecosystems gives better insight into (post-fire) ecosystem restoration and allows for more direct management. Scientists recognise the potential of remote sensing, and especially LiDAR, for large scale research and monitoring, but empirical comparisons with traditional field-based methods are scarce. This study tried to answer the question whether classification through LiDAR scanning is a reliable and efficient method for mapping fuel composition. Through an experimental setup, point cloud data and orthophotos were gathered from 2 Swedish forests. These were used to train models to classify green fuel coverage and deadwood presence, which were then evaluated based on their reliability and efficiency. The resulting models performed relatively poorly (F1: 33-53%, R2: 21%, RMSE: 12.84), but the process proved to be efficient. Additionally, a comparison of four other methods (fixed plot, planar intersect, photoload and below-canopy stereophotogrammetry) proved that LiDAR has potential to be the most reliable and efficient method in the right context. Implementing hybrids of methods increases the reliability further, since different methods can compensate for each other’s limitations. Overall, this study encourages future studies to test LiDAR with the given methodology and more (diverse) test areas, and highlights the potential of combining LiDAR and field-based approaches for ecological research and monitoring.