GIDGeoscientific Instrumentation, Methods and Data Systems DiscussionsGIDGeosci. Instrum. Method. Data Syst. Discuss.2193-0872Copernicus GmbHGöttingen, Germany10.5194/gid-5-549-2015Estimation of forest parameters using airborne laser scanning dataCohenJ.juval.cohen@fmi.fiFinnish Meteorological Institute, P.O. BOX 503, 00101 Helsinki, FinlandJ. Cohen (juval.cohen@fmi.fi)21December2015525495764December20158December2015This work is licensed under a Creative Commons Attribution 3.0 Unported License. To view a copy of this license, visit http://creativecommons.org/licenses/by/3.0/This article is available from https://gi.copernicus.org/preprints/5/549/2015/gid-5-549-2015.htmlThe full text article is available as a PDF file from https://gi.copernicus.org/preprints/5/549/2015/gid-5-549-2015.pdf
Methods for the estimation of forest characteristics by airborne
laser scanning (ALS) data have been introduced by several
authors. Tree height (TH) and canopy closure (CC) describing the
forest properties can be used in forest, construction and industry
applications, as well as research and decision making. The National
Land Survey has been collecting ALS data from Finland since 2008 to
generate a nationwide high resolution digital elevation
model. Although this data has been collected in leaf-off conditions,
it still has the potential to be utilized in forest mapping.
A method where this data is used for the estimation of CC and TH in
the boreal forest region is presented in this paper. Evaluation was
conducted in eight test areas across Finland by comparing the
results with corresponding Multi-Source National Forest Inventory
(MS-NFI) datasets. The ALS based CC and TH maps were generally in
a good agreement with the MS-NFI data. As expected, deciduous
forests caused some underestimation in CC and TH, but the effect was
not major in any of the test areas. The processing chain has been
fully automated enabling fast generation of forest maps for
different areas.
Introduction
Numerous methods for estimating forest characteristics by airborne
laser scanning (ALS) data have been introduced in the last
decades. Several authors have published summaries about the use of ALS
in forest applications, such as Naesset et al. (2004), Hyyppä
et al. (2009), Holopainen et al. (2010, 2011) and
Kaartinen et al. (2012). There are two main approaches in forest
parameter retrieval using ALS (Hyyppä et al., 2008), which are
area-based approaches (ABAs) and individual/single-tree detection
approaches (ITDs). The ABA methods are based on the statistical
relationship between the estimated forest parameter and point cloud
features. In the ITD methods individual trees are recognized and
tree-level variables such as height and width are measured. Therefore
the IDT approach requires higher point density than the ABA.
Tree height (TH) and canopy closure (CC) are important forest
parameters which describe the characteristics of the forest. These
parameters can be used for instance in forest planning and managing,
construction planning, wood industry, and other decision making in
municipal and national level. They are also essential information in
optical and microwave remote sensing applications and in ecological
and biological research. To understand the meaning of TH and CC forest
parameters, it is first necessary to know their definition
(Gschwantner et al., 2009). The definition for forest TH is the
average height of the trees in a certain area, and it is expressed in
meters. CC is usually expressed in percent units (%), and is
defined by the relation between surface area obscured by tree canopy
and the total surface area, when looking vertically from ground
surface level upwards or to the opposite direction (Jennings et al.,
1999).
The National Land Survey (NLS) of Finland has been collecting ALS data
from Finland since 2008, in order to produce a nationwide digital
elevation model (DEM) in 2 m resolution. To reduce the effect
of vegetation on the laser beam, the scanning has been performed
annually in early spring leaf-off conditions. The data is hence not
ideal for vegetation detection, but it can still provide a valuable
source in forest mapping, as it is planned to cover all Finland by the
year 2019, and because most of the forests in Finland are coniferous
dominated, preserving their needles all year round. NLS ALS data has
been previously tested for above ground biomass estimation (Kankare
et al., 2014).
The main goal of this work was to assess the applicability of this
LiDAR point cloud data collected in early spring for generating CC and
TH maps in the boreal forests region of Finland. A method for the
retrieval of CC and TH maps in spatial resolution of 10 m from
the ALS point cloud data was tested and validated. Validation of the
method was carried out in eight test areas across Finland in various
forest conditions by comparing the results with the corresponding
Multi-Source National Forest Inventory (MS-NFI) data generated by the
Natural Resources Institute Finland (LUKE), in 2012. Special
attention was given to the effect of deciduous forest to the results,
because at the time of the laser scanning, leaves were not present on
the trees, which was expected to cause some underestimation in both CC
and TH.
Data and test areas
Point density of the input ALS data is at least
0.5 pointsm-2, which is equivalent to a distance of
approximately 1.4 m between the points. The mean vertical and
horizontal errors of the LiDAR points are up to 15 and 60 cm
respectively. The flight altitude was 2000 m, which creates
a footprint area of 50 cm on the ground. The scanning in the
test areas was performed between the years 2008 and 2014, during early
spring when deciduous trees were still leafless. MS-NFI CC and TH maps
in a spatial resolution of 20m×20m were
used as a reference data for validation. These data have been
developed by employing ground information, satellite images such as
Landsat and Spot, as well as other numerical GIS data to estimate
different forest characteristics (Tomppo et al., 2008). Corine Land
Cover 2006 was used to mask water and urban areas from the generated
forest maps and to separate deciduous forests when analyzing only
deciduous areas.
The CC and TH maps were generated for eight different sites across
Finland, in order to test the method for various forest
types. Saariselkä, Kittilä and Sodankylä from Northern
Finland, Pudasjärvi and Kajaani from Central Finland, and Evo,
Miehikkälä and Tuusula from Southern Finland were chosen. The
locations of the test areas can be seen in Fig. 1. Due to colder
winters and shorter summers in Northern Finland, the trees are
generally shorter and more sparsely distributed than in Southern
Finland. Hence, in principal, the more south is the area, the denser
and higher are the forests. Saariselkä represents the most arctic
environment and the site has the most elevation differences. The high
elevated parts are treeless tundra and the lower parts are forests.
Kittilä and Sodankylä are characterized by mostly coniferous,
relatively short and sparse forests, Pudasjärvi and Kajaani have
somewhat higher and denser, mostly coniferous forests, whereas Evo,
Miehikkälä and Tuusula represent the denser and higher boreal
forest areas of Finland with mostly coniferous trees. Tuusula was
chosen because it has relatively large areas covered by deciduous
forests. Figure 2 shows the CC of all trees, the CC of only deciduous
trees and the TH in whole Finland.
Methods
The LIDAR data was first processed using LAStools, a toolbox for LiDAR
processing created by Martin Isenburg, and then in SAGA
GIS software (Conrad et al., 2015). In the validation, SAGA and Matlab were used. The
production of the CC and TH maps was done in 2 steps. In the first
step, a 1 and 2 m resolution vegetation height
(VH) raster map was produced from the LIDAR point cloud data using
LAStools and SAGA GIS. In the second step, 10 m resolution CC
and TH maps were generated from 1 and the 2 m VH
maps respectively, using SAGA GIS. There is no clear definition for
minimum TH in the literature (Gschwantner et al., 2009), but several
sources have specified outlines for a tree, such as a minimum height
of 3 m (Delijska and Manoilov, 2004), minimum diameter of
12.7 cm and height of 4.6 m (Helms, 1998), or
a minimum height of 10 m (Allaby, 1998). According to
international definitions (FAO, 1998, 2004; UNECE/FAO, 2000) the height
of shrubs (in maturity) is generally between 0.5 and 5 m. In
this work, the minimum TH was set to 1.5 m, because especially
in the northern boreal forests a typical tree can in many cases be
very short. Vegetation under this height was considered low vegetation
such as shrubs, bushes, grass etc.
The whole process was automated using Python, including all LAStools
and SAGA GIS processing steps, in order to enable faster processing
for different areas. The script also enables running the process
repeatedly while changing and optimizing the input parameters, such as
the minimum TH, output resolution and calculation methods. In the next
sub-sections, a detailed description of the processing steps (VH, CC
and TH maps) is given.
Vegetation height
In the generation of the VH maps, the data was first processed in
LAStools and then in SAGA GIS. In LAStools the files were unzipped,
ground level was retrieved, height of each point above the ground
level was calculated, and points where the scanning angle (SA) was
more than 20∘ were removed. Points that were more than
0.5 m below the ground surface were ignored, because there is
a high probability that these points are incorrect due to the double
bounce effect, in which the laser beam is reflected more than once,
and therefore the target is interpreted as being further away than it
really is. Also points that were more than 30 m a.g.l. (above
ground level) were ignored, because they were most probably
representing objects such as high antennas or birds.
VH grids in 1 and 2 m resolution, and scan angle grids in
2 m resolution were then created from the point clouds in SAGA
GIS. The 1 m VH grids were used for the generation of the CC
maps, and the 2 m VH grid was used in the TH
generation. Table 1 lists all the processing steps for the generation
of the 1 and 2 m VH maps.
When converting point clouds to grids, maximum value gridding was
preferred, because mean value gridding caused underestimation of the
TH. The SA information was used to correct the effect of high
inclination scanning to the retrieved CC maps. For a detailed
explanation about the SA correction (see Sect. 3.2). Nodata areas in
the point clouds usually indicated water areas due to the specular
reflection of the laser beams over water surface causing very small
signal coming back to the sensor. The pixel value of these nodata
(water) areas was set to -1.
Canopy closure
The 10 m resolution tree CC map was derived from the
1 m VH grid. The retrieved CC inside a 10m×10m cell area was the relation (%) between the amount
of 1 m tree pixels (above 1.5 m) and the total number
of data pixels inside the 10 m cell. The pixel value was set
to zero if the majority of the 1 m pixels inside
a 10 m cell area were -1. It was noticed that CC was
overestimated in areas where the SA was high, which can be explained
by more laser beams hitting the canopy instead of the ground surface
when coming in a more shallow angle towards the ground surface. The SA
correction was done by using a pixel wise cosine correction:
CCSA corrected=CCcosx(SA).
Different values for the power x were tested, and the best results
were achieved when x was 4. Two methods were used to find the best
value for x, a visual test, and an analytical test. In the visual
test CC maps were generated for different values of x, and the maps
were compared against each other. The best value for x was decided
based on the appearance of stripes in the CC maps. As the x was
increased, the CC in areas under high SA was decreased. In the
analytical test, CC maps were also generated for different values of
x, but low SA areas were removed. The areas with high SA were
compared with the reference MS-NFI CC data, and the best value for x
was decided based on the correlation between the LiDAR CC and the
MS-NFI CC maps (only in high scan angle areas). Table 2 gives
a detailed description of the processing steps in the CC retrieval.
When the 2 m VH maps were used as the input data for the
retrieval, CC was notably overestimated. The reason for this
overestimation was that when gridding the VH maps, some pixels were
always partly over trees and partly over bare ground. These pixels
were usually classified as trees, because in the maximum value
gridding method the pixel value is determined by the highest point
inside a grid cell. The result was that after the gridding, the trees
became wider than in reality. Therefore, to decrease this error, it
was decided to grid the point clouds to 1 m resolution. This
was expected to decrease the width of the trees and bring it closer to
their real width, which would eventually decrease the overall CC.
Figure 3 shows an example of several trees which are seen from above
by the laser scanner, and how the pixels are classified to trees or no
trees when gridding to 2 and 1 m resolutions. From Fig. 3 it
can be seen that the 2 m map would give a higher estimate for
CC, because more area is classified as trees compared to the
1 m map. When gridding to 1 m resolution, more pixels
are assigned nodata values, because no points are found within their
area. However, these nodata pixels are scattered almost uniformly all
over the area and they are ignored when calculating the relation
between tree pixels and bare ground pixels, and therefore they do not
distort the calculated CC.
Tree height
The 10 m resolution TH map was derived from the 2 m VH
grid. The pixel value of a 10m×10m grid
cell was set to zero if the majority of the 2 m pixels inside
a 10 m grid cell were -1 (water). The pixel value was set to
the mean value of the 2 m “low vegetation pixels” (between
-0.05 and 1.5 m) if the 10 m grid cell did not
contain any tree pixels (over 1.5 m), and water pixels were
not the majority. The pixel value was set to the mean value of the
“tree pixels” (above 1.5 m) if at least one 2 m
pixel inside the 10 m grid cell area was higher than
1.5 m (a tree), and water pixels were not the
majority. Table 3 gives a detailed description of the processing steps
in the TH retrieval.
When no filtering was applied on the TH grid, an underestimation of TH
occurred, because the laser beams often hit the lower canopy of the
trees thus lowering the average TH. The underestimation problem was
corrected by filtering the 2 m VH map before creating the
actual TH map (Table 3). Two different filtering methods were tested,
Laplace (2nd derivative) filtering, and dilation (maximum value)
filtering. The aim in the Laplace filtering method was to detect the
individual trees from the VH map using a 2nd derivative filter and to
find the maximum value (height) for each tree. This was done as
follows:
applying the circle shaped Laplace filter (radius of 3 pixels) on the 2 m VH
grid;
separating trees and bare ground by using a threshold value;
converting the grid to a vector image such that each tree is a separate
polygon,
picking the maximum value (height) inside each polygon;
converting back to the original grid, but assigning the maximum value to all
pixels inside a tree polygon.
In the dilation filtering method a circle shaped dilation (maximum)
filter with a radius of 2 pixels (4 m) was applied on the
2 m VH map. This method was much simpler and faster than the
Laplace filtering method. The disadvantages of the Laplace method
compared to the dilation filtering method were the long processing
time and the failure to separate between individual trees in dense
forests (CC over 40), because trees were too close to each other. Yet,
the advantage of the Laplace method was that in sparse forests it was
handling each tree separately and thus was expected to give more
accurate estimates for the real TH. Therefore, it was decided to try
combining the two approaches. In the combined method, for CC over 40,
the dilation filtering method was used and for CC under 40, the
Laplace filtering method was used. The results of the combined method
were compared with the results of using only the dilation filtering
method to all CC values. The comparison showed very small difference
between the two methods, and therefore the Laplace method was
completely abandoned due to the considerably longer processing time.
Results
In Fig. 4 an example of a transformation from point cloud to raster
image is presented. Three main land cover types; water, low vegetation
and forests can be recognized from the images. On the right side,
a river having nodata values in the point cloud image is seen, which
were set to -1 in the raster image. Low vegetation areas are next to
the river and in the upper side of the images, and forested areas are
seen in the rest of the image.
From the point cloud image it can be seen that the distance between
points is not always equal. In Fig. 5 the density of the points can be
seen. The image shows the number of points inside 2m×2m grid cells. The point density varies much and can be
between 0 to more than 10 points per grid cell. The overlapping flight
line areas are recognized easily from this image, because the point
density in them is much higher.
In Fig. 6, an example of a VH map from Saariselkä is presented in
1 and 2 m resolution. In the 1 m VH map the width of
the trees is more accurate than in the 2 m VH map, but due to
the distance between points in the LiDAR data, which is often more
than 1 m, part of the pixels are left with nodata
values. However, as said, this does not affect the CC results as the
nodata gaps inside 10m×10m grid cells are
mostly equally distributed. The gridding to 2 m also left some
one cell nodata gaps which were then filled based on the adjacent
pixel values.
The influence of the SA correction is quite significant in areas of
high SA. A difference image in Fig. 7 shows the difference between CC
before the SA correction and CC after the SA correction in
Sodankylä. In Fig. 8, the produced CC maps are presented for
Sodankylä, Kajaani, Miehikkälä and Tuusula, and in Fig. 9
the produced TH maps are presented for the same test areas.
Validation
The produced CC and TH maps were validated against corresponding
MS-NFI datasets. The MS-NFI data were in 20 m resolution, and
therefore for validation purposes, the CC and TH maps were processed
to the same 20 m resolution grid. Validation was performed for
eight different sites across Finland; Saariselkä, Kittilä and
Sodankylä in the North, Pudasjärvi and Kajaani in Central, and
Evo, Miehikkälä and Tuusula in the South of Finland. The
location of the test areas can be seen in Fig. 1. Figure 10 shows the
validation results of the CC maps, and Fig. 11 the validation results
of the TH maps for the eight test areas. Water and urban areas were
masked from the final products by using Corine 2006. To validate only
deciduous forests, the Tuusula area was chosen, because compared to
other areas in Finland, it has a large proportion of deciduous
forests. Still, only 14 % from the total forested areas were
deciduous forests according to CLC2006. Coniferous forests constituted
37 %, mixed forests 30 % and sparse forests 19 % of the
total forested areas in Tuusula test area. Nevertheless, the amount of
deciduous forest pixels was 36 622 (in the 20 m product), which
is enough for conducting a proper statistical analysis.
Generally, the LiDAR based CC corresponded well with the MS-NFI CC
data. In Northern and Central Finland the RMSE between LiDAR based and
MS-NFI CC was between 11.5 and 15.4 percent points (PP), and the bias
was between 0.74 and 2.73 PP, not including Kittilä. In Southern
Finland the RMSE was between 19.8 and 20.07 PP and the bias between
-4.94 and -1.15 PP. If only deciduous dominant forests in Tuusula
were included, RMSE was 21.5 and bias -8.56 PP. In some of the
test areas a tendency of underestimation in low CC or/and
overestimation in high CC areas was observed, such as in
Saariselkä, Miehikkälä, Evo, Tuusula, and somewhat in
Kajaani and Sodankylä. In Kittilä, an overestimation occurred
in most CC values increasing the total bias to 8.65 and RMSE to
17.1 PP. As expected, the CC in deciduous dominated forests of
Tuusula was underestimated, but not more than ∼10 PP, and only
in denser forests.
The best similarity between the LiDAR based and the MS-NFI TH was
achieved in the Northern and Central Finland test areas;
Saariselkä, Kittilä, Sodankylä, Pudasjärvi and
Kajaani. There, the RMSE between LiDAR based and MS-NFI TH was between
3.09 and 4.25 m, and the bias between -0.52 and
1.14 m. In Southern Finland the RMSE was between 5.90 and
6.91 m, and bias between 0.40 and 0.98 m. The TH was
usually underestimated for the highest trees of the different test
areas. In Southern Finland, and also somewhat in Kajaani, the TH was
typically overestimated in low TH and underestimated in high TH
areas. In the deciduous dominant forests of Tuusula this effect was
more prominent. The weaker precision of LiDAR based CC and TH results
(larger RMSE) in Southern Finland compared to Northern and Central
Finland can also be explained by the larger variation in CC and TH.
Conclusions
A method for generating 10 m resolution CC and TH maps from
LiDAR point cloud data collected by NLS in early spring has been
presented in this report. The processing chain has been fully
automated for easier and faster processing. In this method, 1 and
2 m VH grids were derived from the point cloud data and used
in the retrieval of CC and TH maps respectively. Appropriate
filtering and the use of scanning angle information from the point
clouds were utilized to gain better accuracy. The method was evaluated
in eight test areas across Finland having different forest
characteristics, by comparing the retrieved CC and TH maps with the
corresponding MS-NFI datasets generated by LUKE in 2012.
Generally, this method shows good agreement between the LiDAR based
and the MS-NFI datasets. Results were generally better in Northern and
Central Finland than in Southern Finland. Early spring laser scanning
in leaf-off conditions is not optimal for deciduous vegetation
detection and was expected to cause underestimation of CC and TH. When
only deciduous dominated forest areas were selected (in Tuusula area)
the CC was indeed underestimated in high CC areas, and the
underestimation of TH in high TH areas was more prominent. The effect
of deciduous forests can hence be recognized, but in practice, the
effect is not major because vast majority of the Finnish forests are
coniferous dominated.
For a more reliable accuracy assessment, the results should be
validated against ground true data in the future, as some
discrepancies between the LiDAR and the MS-NFI data may also result
from inaccuracies in the MS-NFI data.
Acknowledgements
This work was supported by the Academy of Finland Center of Excellence in
Atmosphere Science (CoE ATM). National Land Survey of Finland is acknowledged
for providing the LiDAR data, and Natural Resources Institute Finland for
providing the MS-NFI forest data.
References Allaby, M.: A dictionary of plant sciences, Oxford
University Press, 508 pp., 1998.Conrad, O., Bechtel, B., Bock, M., Dietrich, H., Fischer, E., Gerlitz, L.,
Wehberg, J., Wichmann, V., and Böhner, J.: System for Automated Geoscientific
Analyses (SAGA) v. 2.1.4, Geosci. Model Dev., 8, 1991–2007, 10.5194/gmd-8-1991-2015, 2015. Delijska, B. and Manoilov, P.: Elsevier's dictionary of
forestry, Elsevier, Amsterdam, 1051 pp., 2004. FAO: The Global Forest Resources Assessment 2000 (FRA 2000):
Terms and definitions, FRA Working Paper 1, FAO Forestry Department, Rome, 1998. FAO: Global Forest Resources Assessment update 2005 (FRA 2005):
Terms and definitions, FRA Working Paper 83/E, FAO Forestry Department, Rome, 34 pp., 2004. Gschwantner, T., Schadauer, K., Vidal, C., Lanz, A.,
Tomppo, E., di Cosmo, L., Robert, N., Duursma, D. E., and
Lawrence, M.: Common tree definitions for national forest
inventories in Europe, Silva Fenn., 43, 303–321, 2009. Helms, J. A.: The dictionary of forestry, CAB
International: the Society of American Foresters, Wallingford, 210 pp., 1998. Holopainen, M., Hyyppä, J., Vaario, L.-M., and
Yrjälä, K.: Implications of technologial development to
forestry, invited book chapter, in: Forest and Society – Responding
to Global Drivers of Change, edited by: Mery, G., Katila, P.,
Galloway, G., Alfaro, R. I., Kanninen, M., Lobovikov, M., and
Varjo, J., IUFRO-World Series Vol. 25, Convening lead authors of
the chapter: Hetemäki, L. and Mery, G., International Union of Forest
Research Organizations (IUFRO), Vienna, 157–182, 2010. Holopainen, M., Hyyppä, J., Vastaranta, M., and
Hyyppä, H.: Laserkeilaus metsävarojen hallinnassa, The
Photogrammetric Journal of Finland – Special issue of the 80 years
anniversary of the Finnish Society of Photogrammetry and Remote
Sensing, 22, 128–149, 2011. Hyyppä, J., Hyyppä, H., Leckie, D., Gougeon, F.,
Yu, X., and Maltamo, M.: Review of methods of small-footprint
airborne laser scanning for extracting forest inventory data in
boreal forests, Int. J. Remote Sens., 29, 1339–1366, 2008. Hyyppä, J., Hyyppä, H., Yu, X., Kaartinen, H.,
Kukko, A., and Holopainen, M.: Forest inventory using
small-footprint airborne lidar, in: Topographic Laser Ranging and
Scanning: Principles and Processing, edited by: Shan, J. and Toth, C.,
CRC Press, Taylor & Fracis, London, 335–370, 2009. Jennings, S. B., Brown, N. D., and Sheil, D.: Assessing
forest canopies and understorey illumination: canopy closure, canopy
cover and other measures, Forestry, 72, 59–74, 1999. Kaartinen, H., Hyyppä, J., Yu, X., Vastaranta, M.,
Hyyppä, H., Kukko, A., Holopainen, M., Heipke, C.,
Hirschmugl, M., Morsdorf, F., Næsset, E., Pitkänen, J.,
Popescu, S., Solberg, S., Wolf, B. M., and Wu, J.-C.: An
international comparison of individual tree detection and extraction
using airborne laser scanning, Remote Sens., 4, 950–974, 2012. Kankare, V., Vauhkonen, J., Holopainen, M.,
Vastaranta, M., Hyyppä, J., Hyyppä, H., and Alho, P.: The
Accuracy of National ALS Data in Biomass Mapping in Finland,
European Biomass Conference and Exhibition proceedings, ETA-Florence
Renewable Energies, 22nd European Biomass Conference and Exhibition,
Hamburg, Germany, 2014. Naesset, E., Gobakken, T., Holmgren, J., Hyyppä, H.,
Hyyppä, J., Maltamo, M., Nilsson, M., Olsson, H., Persson,
Å., and Söderman, U.: Laser scanning of forest resources:
the Nordic experience, Scand. J. Forest Res., 19, 482–499, 2004. Tomppo, E., Haakana, M., Katila, M., and
Peräsaari, J.: Multi-Source National Forest Inventory: Methods
and Applications, Springer Netherlands, Dordrecht, 2008. UNECE/FAO: Forest resources of Europe, CIS, North
America, Australia, Japan and New Zealand, Geneva Timber and Forest
study papers 17, United Nations, New York and Geneva, 2000.
Processing steps for generating the 1 and 2 m VH maps.
Vegetation height, 2 m resolutionVegetation height, 1 mresolutionLAStools processing: 1.Decompress LASZIPfiles2.Remove high scanningangleAreas with scanning angle higher than 20 ∘ were removed3.Extracting groundpointsParameters: – Forest or hills – Default step size (5 m)4.Calculating heightabove ground levelHeight above the ground level is calculated for each point. Points 0.05 m below and 30 ma.g.l. were ignored.SAGA processing: 5.Import LAS point filesto SAGAImport point height and scan angle attributes6.Converting point filesto gridsPixel values: – Point height grid: highestpoint in a 2 m grid cell – Scan angle grid: mean valueof the absolute scan angle ina 2 m grid cellPixel values: – Point height grid:highest point in a1 m grid cell7.Closing one-cell gapsOne-cell gaps are filled with a pixelvalue which is the mean value of thesurrounding pixels.No closing gaps8.Setting a value forwaterSetting nodata to -1
Processing steps in the retrieval of CC.
Processing step Details1.Create two grids from theoriginal vegetation heightgridOutput grids are in 1 m resolution – Trees grid: low vegetation pixels (between 0 to 1.5)are set to zero, water (-1) is set to nodata and treepixels (above 1.5 m) are set to 100 – Water grid: all other values except water (-1) are setto zero.2.Trees grid and water grid areresampled to 10 m– Trees grid: new value is the mean value of the 1 mpixels – Water grid: new value is the majority value from the1 m pixels3.Reclassify zero values tonodata from the water grid4.Resample Scan Angle (SA)grid to 10 m resolutionThe output is the absolute scan angle (values between 0and 20 ∘), where the value in each 10 grid cell is the meanvalue of the 2 m pixels5.Reclassify SA gridWater areas in scan angle grid are reclassified to 06.Calculate SA correction to CCgridCCSAcorrected=CC⋅(cos(SA))4 where SA is the scan angle.7.Merge the 10 m resolutiongrids (water, trees)In case of overlapping cells, the value from the first grid isused
Processing steps in the retrieval of TH.
Processing step Details1.Create two grids from the originalvegetation height gridOutput grids are in 2 m resolution – Tree height grid: low vegetation (between 0to 1.5) and water (-1) are set to nodata. – Water grid: all other values except water (-1)are set to zero.2.FilteringA circle shaped dilation (maximum) filter with a radiusof 2 pixels is applied on the tree height grid3.Create three 10 m grids byresampling the 2 m grids– Tree height grid: new value is the mean valueof the 2 m pixels of the tree height grid – Low vegetation grid: new value is the meanvalue of the 2 m pixels of the vegetationheight grid – Water grid: new value is the majority value ofthe 2 m pixels of the water grid4.Reclassify zero values to nodatafrom the water-grid5.Merge the 10 m resolution grids(water, tree height, low vegetation)In case of overlapping cells, the value from the firstgrid is used
The test areas are marked with red color. Saariselkä,
Kittilä and Sodankylä in Northern Finland, Pudasjärvi
and Kajaani in Central Finland and Evo, Miehikkälä and
Tuusula in Southern Finland. The size of the red squares is
proportional to the size of the test areas in reality.
CC of all trees, CC of only deciduous trees and average TH in
whole Finland. Source: Paikkatietoikkuna, MS-NFI (LUKE, 2015).
Gridded 2 m VH on the left side and 1 m VH on
the right side. Red color indicates trees, blue color low vegetation
or bare ground and white is nodata. The total area covered by the
canopy is larger in the 2 m image, because pixels that are
partly over trees and partly over ground are usually classified as
trees, because the pixel value is determined by the higher point
inside a grid cell. Therefore using 1 m VH maps as input to
CC retrieval was preferred. UTM coordinates and length in m is shown
in the outline.
An example area where the VH is shown as a point cloud (upper
panel) and as a 2 m resolution grid (lower panel). On the
right side there is a river, which has nodata in the point cloud but
in the gridded image it has been set to -1. Next to the river and
on the upper side of the image there are low vegetation areas, and
the rest of the areas are forested. UTM coordinates and length in m
is shown in the outline.
Points per 2 m grid cell. The density of the point
clouds varies between 0 to more than 10 points per 2 m grid
cell. The white areas are grid cells where no points were found
(river). One cell gaps were filled using the surrounding grid
values, and larger gaps were considered water and marked as
-1. Point density under the overlapping flightline areas is higher
than in other areas. UTM coordinates and length in m is shown in the
outline.
VH map in 1 m (upper panel) and 2 m (lower panel)
resolution from Saariselkä. When gridding the point cloud to
1 m resolution, more nodata cells are left in the image, but
the width of the trees is closer to reality than when gridding to
2 m resolution. UTM coordinates and length in m is shown in
the outline.
A difference image showing the effect of SA correction in
Sodankylä. The SA correction can lead to a difference of more
than 10 percentage points in the retrieved CC in high scanning angle
areas. UTM coordinates and length in m is shown in the outline.
CC in Sodankylä (upper left panel), Kajaani (upper right panel),
Miehikkälä (lower left panel) and Tuusula (lower right panel). UTM
coordinates and length in m is shown in the outline.
TH in Sodankylä (upper left panel), Kajaani (upper right panel),
Miehikkälä (lower left panel) and Tuusula (lower right panel). UTM
coordinates and length in m is shown in the outline.
Validation of LiDAR CC against corresponding MS-NFI data from
Saariselkä, Kittilä, Sodankylä, Pudasjärvi, Kajaani,
Evo, Miehikkälä, Tuusula and only deciduous dominant forest
in Tuusula.
Validation of LiDAR TH against corresponding MS-NFI data from
Saariselkä, Kittilä, Sodankylä, Pudasjärvi, Kajaani,
Evo, Miehikkälä, Tuusula and only deciduous dominant forest
in Tuusula.