Remote Sensing of Grassland
Biodiversity
This section will be split into three broad categories. The first is
based on the spectral variation hypothesis (SVH). This the most common
method for mapping plant biodiversity and is centred on the premise that
individual plant species absorb and reflect sunlight in unique ways,
creating a distinct spectral signature (Figure 3). Where there are many
distinct species in a grassland, the spectral diversity (SD) recorded by
the remote sensing instrument will be greater than in areas with fewer
species (Rocchini et al., 2004). This type of analysis can be performed
with both multi- and hyper-spectral instruments, with measures of SD
ranging from simple standard deviations of spectral bands to convex hull
volume of the principal components of hundreds of hyperspectral bands
and more. Studies utilising the SVH approach are the focus of 18 of the
37 biodiversity papers in this section, representing refinement of the
methodology, application in different environments, as well as
exploration of mediating factors and limitations.
The second biodiversity section encompasses studies with a focus on
machine learning. As in many scientific fields, remote sensing of
grassland biodiversity has experienced and accelerated uptake in the use
of machine learning in the last few years. Here, they account for 12 of
the 38 papers presented.
The third section will explore studies that focus on neither the SVH nor
machine learning (although they form small parts of some studies) but
include approaches from manual identification of species from UAV
imagery to interdisciplinary research.