184
Keywords Biodiversity monitoring • Drone • GLCM • LAS • Namibia • NDVI •
Point cloud • Spectral discrimination
Introduction
Classifying and mapping individual trees is increasingly applied in forestry, urban
management and nature conservation. According to a review by Fassnacht et al.
(2016), since the year 2000 and particularly from 2010 onwards, there has been a
dramatic increase in the number of studies that compare the suitability of different
datasets, classifiers, and sensor platforms. However, most studies on the classification of tree species use expensive technology to capture data, e.g. hyperspectral or
LiDAR sensors, with only a few studies applying relatively cheap solutions such as
UAVs carrying consumer-grade cameras that provide Red-Green-Blue (RGB) or
Near-Infrared (NIR) imagery. Furthermore, most studies to date have focussed on
temperate or boreal forest ecosystems while Savannah ecosystems, which are relatively rich in tree species, remain understudied.
This chapter evaluates the suitability of image parameters derived from low-cost,
UAV-borne, consumer-grade cameras for classifying tree species in a savannah ecosystem. In particular, we aim to test (a) whether savannah tree species can be discriminated successfully with very high resolution UAV imagery, (b) whether RGB
or NIR spectral indices perform better, and (c) if a canopy height model can significantly improve the classification. Finally, we discuss the role of a species abundance
for it´s potential to be accurately mapped.
Background
Mapping the distribution of tree species using remote sensing means producing a
vector or raster layer that contains the information on locations of tree species either
at the stand-level or single-stem level. These maps or data sets are valuable in nature
conservation, particularly in a biodiversity monitoring context. However, until
recently, the most commonly used image data for mapping tree species were from
hyperspectral and LiDAR sensors (Fassnacht et al. 2016) which are costly, difficult
to preprocess, and require expert knowledge in their analysis. The recent advent of
drones, also called Unmanned Aerial Vehicles (UAVs), provides new tools and the
opportunity to obtain more spatial detail for tree species mapping, and the use of
UAVs for mapping tree species is becoming increasingly popular (Singh et al. 2015;
Lisein et al. 2015). UAVs have several advantages over satellite or airborne data.
They are extremely flexible in usage, can be scheduled in a very short time interval
(e.g. daily or weekly), are easily carried to diverse locations and, unlike satellites,
are not limited by clouded skies. The main drawbacks are the limited spatial
J. Oldeland et al.
Keywords Biodiversity monitoring • Drone • GLCM • LAS • Namibia • NDVI •
Point cloud • Spectral discrimination
Introduction
Classifying and mapping individual trees is increasingly applied in forestry, urban
management and nature conservation. According to a review by Fassnacht et al.
(2016), since the year 2000 and particularly from 2010 onwards, there has been a
dramatic increase in the number of studies that compare the suitability of different
datasets, classifiers, and sensor platforms. However, most studies on the classification of tree species use expensive technology to capture data, e.g. hyperspectral or
LiDAR sensors, with only a few studies applying relatively cheap solutions such as
UAVs carrying consumer-grade cameras that provide Red-Green-Blue (RGB) or
Near-Infrared (NIR) imagery. Furthermore, most studies to date have focussed on
temperate or boreal forest ecosystems while Savannah ecosystems, which are relatively rich in tree species, remain understudied.
This chapter evaluates the suitability of image parameters derived from low-cost,
UAV-borne, consumer-grade cameras for classifying tree species in a savannah ecosystem. In particular, we aim to test (a) whether savannah tree species can be discriminated successfully with very high resolution UAV imagery, (b) whether RGB
or NIR spectral indices perform better, and (c) if a canopy height model can significantly improve the classification. Finally, we discuss the role of a species abundance
for it´s potential to be accurately mapped.
Background
Mapping the distribution of tree species using remote sensing means producing a
vector or raster layer that contains the information on locations of tree species either
at the stand-level or single-stem level. These maps or data sets are valuable in nature
conservation, particularly in a biodiversity monitoring context. However, until
recently, the most commonly used image data for mapping tree species were from
hyperspectral and LiDAR sensors (Fassnacht et al. 2016) which are costly, difficult
to preprocess, and require expert knowledge in their analysis. The recent advent of
drones, also called Unmanned Aerial Vehicles (UAVs), provides new tools and the
opportunity to obtain more spatial detail for tree species mapping, and the use of
UAVs for mapping tree species is becoming increasingly popular (Singh et al. 2015;
Lisein et al. 2015). UAVs have several advantages over satellite or airborne data.
They are extremely flexible in usage, can be scheduled in a very short time interval
(e.g. daily or weekly), are easily carried to diverse locations and, unlike satellites,
are not limited by clouded skies. The main drawbacks are the limited spatial
J. Oldeland et al.