Mapping and monitoring of vegetation using airborne laser scanning

• Main Author: Nyström, Mattias
• Format: Doctoral or Postdoctoral Thesis
• Language: English
• Published: 2014
• Subjects: Remote Sensing; Forest Science; Abisko; Torp; Northern Sweden; Tundra
• Online Access: https://pub.epsilon.slu.se/10982/; https://pub.epsilon.slu.se/10982/1/nystrom_m_140121.pdf

In this thesis, the utility of airborne laser scanning (ALS) for monitoring vegetation of relevance for the environmental sector was investigated. The vegetation characteristics studied include measurements of biomass, biomass change and vegetation classification in the forest-tundra ecotone; afforestation of grasslands; and detection of windthrown trees. Prediction of tree biomass for mountain birch (Betula pubescens ssp. czerepanovii) using sparse (1.4 points/m²) and dense (6.1 points/m²) ALS data was compared for a site at the forest-tundra ecotone near Abisko in northern Sweden (Lat. 68° N, Long. 19° E). The predictions using the sparse ALS data provided almost as good results (RMSE 21.2%) as the results from the dense ALS data (18.7%) despite the large difference in point densities. A new algorithm was developed to compensate for uneven distribution of the laser points without decimating the data; use of this algorithm reduced the RMSE for biomass prediction from 19.9% to 18.7% for the dense ALS data. Additional information about vegetation height and density from ALS data improved a satellite data classification of alpine vegetation, in particular for the willow and mountain birch classes. Histogram matching was shown to be effective for relative calibration of metrics from two ALS acquisitions collected over the same area using different scanners and flight parameters. Thus the difference between histogram-matched ALS metrics from different data acquisitions can be used to locate areas with unusual development of the vegetation. The height of small trees (0.3–2.6 m tall) in former pasture land near the Remnings¬torp test site in southern Sweden (Lat. 58° N, Long. 13° E) could be measured with high precision (standard deviation 0.3 m) using high point density ALS data (54 points/m2). When classifying trees taller than 1 m into the two classes of changed and unchanged, the overall classification accuracy was 88%. A new method to automatically detect windthrown trees in forested areas was developed and evaluated at the Remningstorp test site. The overall detection rate was 38% on tree-level, but when aggregating to 40 m square grid cells, at least one windthrown tree was detected in 77% of the cells that according to field data contained windthrown trees. In summary, this thesis has shown the high potential for ALS to be a future tool to map and monitor vegetation for several applications of interest for the environmental sector.