University of Twente Student Theses


Drones for conservation: integrating UAVs with field methods to classify satellite imagery to map plant communities – A case study of Drentsche AA, The Netherlands

Chib, Rhea Singh (2021) Drones for conservation: integrating UAVs with field methods to classify satellite imagery to map plant communities – A case study of Drentsche AA, The Netherlands.

[img] PDF
Link to full-text:
(only accessible for UT students and staff)
Abstract:Plant Communities are the rudimentary unit of natural habitats and have high ecological importance. With the increasing rate of biodiversity loss, vulnerable plant communities and associated species are at higher risk. Hence, mapping their extent becomes important to assess the conservation status of an area. The traditional method of ground survey is the most precise, but laborious and expensive, restricting it to smaller areas. On the other hand, satellite imagery can cover larger areas, but the spatial resolution of freely available imageries does not allow detailed mapping. UAVs have several advantages over these methods like very high spatial resolution, being more cost-efficient and allowing flexible data collection, but limitations of lower spatial coverage and restrictive flying regulations. Hence, all these three methods have their advantages and disadvantages but integrating them could result in a promising approach to precisely map plant communities. This study proposes a method to integrate UAVs with ground surveys to improve the classification results of the satellite imagery. It aims to combine the benefits of high spatial resolution of the UAV imagery with the high spatial coverage of satellite imagery, to produce detailed maps on a larger scale. The National Park of Drentsche Aa, the Netherlands is used as a case study area to test this approach. The results show that OBIA efficaciously classifies the UAV imagery and produces detailed maps, with an accuracy of 87% for the classes of interest i.e., plant communities (different from the overall accuracy). The UAV imagery was then used to create additional samples to train RF to classify the satellite imagery; SuperView-1 and Sentinel-2 were used for the analysis. It produced better results in comparison to using only field samples (64% for Sentinel-2 and 67% for SuperView-1). The results did not have high accuracy as the spatial resolution of the satellite imageries did not allow clear separation of some classes. However, to make the classified maps more comparable to the broader scale of the existing vegetation map, the classes were systematically merged. It resulted in an increase in the accuracy (92% for Sentinel-2 and 93% for SuperView-1), though the level of detail was reduced. Nevertheless, this approach with some further improvements can emerge as a promising technique to precisely map plant communities on a larger scale for conservation.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page