327
straightforward methodology in which the height distribution of the LiDAR beam
reflections is analyzed for a given area. In the first step, plenty of different “LiDAR
metrics,” e.g., maximum height or fractional cover, are calculated for each area. The
second step is model calibration, where these metrics are compared to on-theground survey data such as plant species richness, aboveground biomass (AGB), or
vertical and horizontal vegetation structure. In the final step, the models are used to
estimate the selected biodiversity indicators for large areas using square grid cells.
Such an analysis is generally conducted using a priori stratification of structural
vegetation types and plant species. In the years that followed, this methodology was
proven to be able to determine key biophysical vegetation variables on a larger
scale. To date, this method has been shown to deliver a precision of 4–8% for height,
6–12% for mean diameter, 9–12% for basal area, 17–22% for stem number, and
11–14% for volume estimations of boreal forests (Maltamo et al. 2006; Næsset
2002, 2007). Because of the highly accurate estimation of important vegetation
structural parameters, the area-based approach was further developed and adapted
to operational forest inventories in boreal forests of Scandinavia. Similar accuracies
have also been achieved for the temperate zone, although the more complex
vegetation structures in this zone, especially the higher number of tree and plant
species and higher amount of biomass, led to less accurate estimations and more
effort in stratification and ground measurement to obtain species-specific results
(Heurich and Thoma 2008; Latifi et al. 2010, 2015).
The second methodology is the individual-tree approach, which has the objective
of extracting data on single trees and modeling the tree properties. The procedure
consists of four steps. In the first step, individual trees are delineated by dividing
each crown into segments with techniques originally used for raster analysis, such
as watershed analysis and local maxima detection (Heurich 2008; Persson et al.
2002). However, these techniques do not take advantage of the full information of
the 3-D point cloud, and therefore, trees beneath the crown surface cannot be
detected. For this reason, new methods based on 3-D point clouds have been developed over recent years (Tang et al. 2013; Yao et al. 2012). When these novel techniques are employed, more than 80% of the trees of the upper canopy level can be
detected. Moreover, tree detection in the lower canopy is much improved compared
to 2-D techniques. In the second step, parameters of each tree (e.g., height, species,
and crown parameters) are derived. Tree height can be determined by measuring the
distance of the highest reflection of a LiDAR beam within the tree segment and the
DTM, with an accuracy of less than 2 m and a slight underestimation (Heurich
2008). The third step is the model calibration of the biophysical parameters of the
tree, namely, diameter at breast height (DBH), volume, and biomass, using trees
measured on the ground as a reference. The tree crown can be modeled using convex hulls and alpha shapes. The fourth step involves the application of these models
to predict DBH, volume, and biomass of all trees delineated by LiDAR. Based on
these crown representations, basic attributes reflecting tree health can be derived,
e.g., total volume, crown length, crown area, and crown base height (Yao et al.
2012). The extracted parameters of individual trees also form the basis for identifying the tree species by calculating point cloud and waveform features within the 2-D
13 A Range of Earth Observation Techniques for Assessing Plant Diversity
straightforward methodology in which the height distribution of the LiDAR beam
reflections is analyzed for a given area. In the first step, plenty of different “LiDAR
metrics,” e.g., maximum height or fractional cover, are calculated for each area. The
second step is model calibration, where these metrics are compared to on-theground survey data such as plant species richness, aboveground biomass (AGB), or
vertical and horizontal vegetation structure. In the final step, the models are used to
estimate the selected biodiversity indicators for large areas using square grid cells.
Such an analysis is generally conducted using a priori stratification of structural
vegetation types and plant species. In the years that followed, this methodology was
proven to be able to determine key biophysical vegetation variables on a larger
scale. To date, this method has been shown to deliver a precision of 4–8% for height,
6–12% for mean diameter, 9–12% for basal area, 17–22% for stem number, and
11–14% for volume estimations of boreal forests (Maltamo et al. 2006; Næsset
2002, 2007). Because of the highly accurate estimation of important vegetation
structural parameters, the area-based approach was further developed and adapted
to operational forest inventories in boreal forests of Scandinavia. Similar accuracies
have also been achieved for the temperate zone, although the more complex
vegetation structures in this zone, especially the higher number of tree and plant
species and higher amount of biomass, led to less accurate estimations and more
effort in stratification and ground measurement to obtain species-specific results
(Heurich and Thoma 2008; Latifi et al. 2010, 2015).
The second methodology is the individual-tree approach, which has the objective
of extracting data on single trees and modeling the tree properties. The procedure
consists of four steps. In the first step, individual trees are delineated by dividing
each crown into segments with techniques originally used for raster analysis, such
as watershed analysis and local maxima detection (Heurich 2008; Persson et al.
2002). However, these techniques do not take advantage of the full information of
the 3-D point cloud, and therefore, trees beneath the crown surface cannot be
detected. For this reason, new methods based on 3-D point clouds have been developed over recent years (Tang et al. 2013; Yao et al. 2012). When these novel techniques are employed, more than 80% of the trees of the upper canopy level can be
detected. Moreover, tree detection in the lower canopy is much improved compared
to 2-D techniques. In the second step, parameters of each tree (e.g., height, species,
and crown parameters) are derived. Tree height can be determined by measuring the
distance of the highest reflection of a LiDAR beam within the tree segment and the
DTM, with an accuracy of less than 2 m and a slight underestimation (Heurich
2008). The third step is the model calibration of the biophysical parameters of the
tree, namely, diameter at breast height (DBH), volume, and biomass, using trees
measured on the ground as a reference. The tree crown can be modeled using convex hulls and alpha shapes. The fourth step involves the application of these models
to predict DBH, volume, and biomass of all trees delineated by LiDAR. Based on
these crown representations, basic attributes reflecting tree health can be derived,
e.g., total volume, crown length, crown area, and crown base height (Yao et al.
2012). The extracted parameters of individual trees also form the basis for identifying the tree species by calculating point cloud and waveform features within the 2-D
13 A Range of Earth Observation Techniques for Assessing Plant Diversity