328
or 3-D representation of the tree and with the help of classification techniques
(Reitberger et al. 2008). While differentiation between deciduous and coniferous
trees is highly accurate (>80%, up to ca. 97%), differentiation within these classes
is more difficult and leads to a higher classification error. Moreover, it is possible to
distinguish between living trees, standing dead trees, and snags (Yao et al. 2012)
and to map dead trees at the plot or stand level. However, 3-D LiDAR has its limitations in differentiating between trees species and dead trees when not combined
with multispectral optical data. One drawback of the individual-tree approach is that
the LiDAR beam loses some transmission on its way through the canopy and is
therefore not always suitable for smaller understory trees, which results in their
underestimation. To overcome this problem, methods have been developed to predict diameter distributions of forest stands based on detectable trees in the upper
canopy and LiDAR-derived information on the vertical forest structure and density
(Lefsky et al. 2002).
In addition to the traditional parameters related to forestry, a multitude of traits
that describe the ecological conditions of the forest can be estimated with LiDAR
sensors. One key element for assessing plant diversity and vegetation structure is
canopy cover, which is defined as the projection of the tree crowns onto the ground
divided by ground surface area. This parameter can be easily obtained from LiDAR
data by dividing the number of returns measured above a certain height threshold by
the total number of returns. Many studies have proven the strong (R
2
> 0.7) relationship between this LiDAR metric and ground measurements. By using hemispherical
images or other ground-based instruments for calibration, leaf area index (LAI) and
solar radiation can also be derived from LiDAR data with a high precision over large
areas (Moeser et al. 2014). Because canopy metrics are affected by sensor and flight
characteristics, it is recommended that each campaign be calibrated to obtain highquality results. However, it has been shown that even without calibration, fairly
reliable results can be obtained.
Vertical vegetation structure is highly relevant for the description of forest and
vegetation heterogeneity and highly important for biodiversity studies. A widely
used LiDAR metric for representing vertical canopy complexity is the coefficient of
variation. High coefficient values correspond to more diverse multilayer stands,
whereas low values represent single-layer stands. The coefficient of variation can be
applied at point clouds, the digital crown model, or individual trees. Zimble et al.
(2003) applied this principle and classified vegetation types according to stand
structure with an overall accuracy of 97%.
Another approach is the partitioning of the vertical structure into different height
layers in relation to ecological importance. Latifi et al. (2015) divided the canopy
into height layers according to phytosociological mapping standards and found a
strong relationship to various LiDAR metrics in regression models. Similar
approaches were used by Ewald et al. (2014) to represent understory offering protection for birds and deer and to detect forest regeneration. A more recent study
applied a 3-D segmentation algorithm to estimate regeneration cover with an accuracy of 70%. LiDAR-derived information about the vertical structure is also used
A. Lausch et al.
or 3-D representation of the tree and with the help of classification techniques
(Reitberger et al. 2008). While differentiation between deciduous and coniferous
trees is highly accurate (>80%, up to ca. 97%), differentiation within these classes
is more difficult and leads to a higher classification error. Moreover, it is possible to
distinguish between living trees, standing dead trees, and snags (Yao et al. 2012)
and to map dead trees at the plot or stand level. However, 3-D LiDAR has its limitations in differentiating between trees species and dead trees when not combined
with multispectral optical data. One drawback of the individual-tree approach is that
the LiDAR beam loses some transmission on its way through the canopy and is
therefore not always suitable for smaller understory trees, which results in their
underestimation. To overcome this problem, methods have been developed to predict diameter distributions of forest stands based on detectable trees in the upper
canopy and LiDAR-derived information on the vertical forest structure and density
(Lefsky et al. 2002).
In addition to the traditional parameters related to forestry, a multitude of traits
that describe the ecological conditions of the forest can be estimated with LiDAR
sensors. One key element for assessing plant diversity and vegetation structure is
canopy cover, which is defined as the projection of the tree crowns onto the ground
divided by ground surface area. This parameter can be easily obtained from LiDAR
data by dividing the number of returns measured above a certain height threshold by
the total number of returns. Many studies have proven the strong (R
2
> 0.7) relationship between this LiDAR metric and ground measurements. By using hemispherical
images or other ground-based instruments for calibration, leaf area index (LAI) and
solar radiation can also be derived from LiDAR data with a high precision over large
areas (Moeser et al. 2014). Because canopy metrics are affected by sensor and flight
characteristics, it is recommended that each campaign be calibrated to obtain highquality results. However, it has been shown that even without calibration, fairly
reliable results can be obtained.
Vertical vegetation structure is highly relevant for the description of forest and
vegetation heterogeneity and highly important for biodiversity studies. A widely
used LiDAR metric for representing vertical canopy complexity is the coefficient of
variation. High coefficient values correspond to more diverse multilayer stands,
whereas low values represent single-layer stands. The coefficient of variation can be
applied at point clouds, the digital crown model, or individual trees. Zimble et al.
(2003) applied this principle and classified vegetation types according to stand
structure with an overall accuracy of 97%.
Another approach is the partitioning of the vertical structure into different height
layers in relation to ecological importance. Latifi et al. (2015) divided the canopy
into height layers according to phytosociological mapping standards and found a
strong relationship to various LiDAR metrics in regression models. Similar
approaches were used by Ewald et al. (2014) to represent understory offering protection for birds and deer and to detect forest regeneration. A more recent study
applied a 3-D segmentation algorithm to estimate regeneration cover with an accuracy of 70%. LiDAR-derived information about the vertical structure is also used
A. Lausch et al.