Images from unmanned aircraft systems for surveying aquatic and riparian vegetation
Aquatic and riparian vegetation in lakes, streams, and wetlands has important ecological and regulatory functions and should be monitored to detect ecosystem changes. Field surveys are often tedious and in countries with numerous lakes and streams a nationwide assessment is difficult to achieve. Rem...
| Main Author: | |
|---|---|
| Format: | Doctoral or Postdoctoral Thesis |
| Language: | English Swedish |
| Published: |
2016
|
| Subjects: | |
| Online Access: | https://pub.epsilon.slu.se/13816/ https://pub.epsilon.slu.se/13816/1/husson_e_161103.pdf https://pub.epsilon.slu.se/13816/11/husson_e_errata_161201.pdf |
| Summary: | Aquatic and riparian vegetation in lakes, streams, and wetlands has important ecological and regulatory functions and should be monitored to detect ecosystem changes. Field surveys are often tedious and in countries with numerous lakes and streams a nationwide assessment is difficult to achieve. Remote sensing with unmanned aircraft systems (UASs) provides aerial images with high spatial resolution and offers a potential data source for detailed vegetation surveys. The overall objective of this thesis was to evaluate the potential of sub-decimetre resolution true-colour digital images acquired with a UAS for surveying non-submerged (i.e., floating-leaved and emergent) aquatic and riparian vegetation at a high level of thematic detail. At two streams and three lakes in northern Sweden we applied several image analysis methods: Visual interpretation, manual mapping, manual mapping in combination with GPS-based field surveys, and automated object-based image analysis and classification of both 2D images and 3D point data. The UAS-images allowed for high taxonomic resolution, mostly at the species level, with high taxa identification accuracy (>80%) also in mixed-taxa stands. UAS-images in combination with ground-based vegetation surveys allowed for the extrapolation of field sampling results, like biomass measurement, to areas larger than the sampled sites. In automatically produced vegetation maps some fine-scale information detectable with visual interpretation was lost, but time-efficiency increased which is important when larger areas need to be covered. Based on spectral and textural features and height data the automated classification accuracy of non-submerged aquatic vegetation was ~80% for all test sites at the growth-form level and for four out of five test sites at the dominant-taxon level. The results indicate good potential of UAS-images for operative mapping and monitoring of aquatic, riparian, and wetland vegetation. More case studies are needed to fully assess the added value of UAS-technology in terms of invested labour and costs compared to other survey methods. Especially the rapid technical development of multi- and hyperspectral lightweight sensors needs to be taken into account. |
|---|