Experiences and possibilities of ALS based forest inventory in Finland

Article (PDF Available)with 246 Reads
Cite this publication
Abstract
During last ten years the research concerning airborne laser scanning based forest inventory applications has been very active in different parts of the world. In Finland, both basic approaches, single tree detection and area based modeling have been widely examined. In the following the results of the ALS based forest inventory experiments and further possibilities in Finland are reviewed and discussed. A short review of Finnish forestry in relation to possibilities of ALS based forest inventories is included as well. Finally, some examples concerning the comparison of single tree detection and area based modeling and the usability of spatial information provided by ALS data are presented.
EXPERIENCES AND POSSIBILITIES OF ALS BASED
FOREST INVENTORY IN FINLAND
Maltamo, M.a, Packalén, P. a, Peuhkurinen, J.a, Suvanto, A. a, Pesonen, A.a & Hyyppä, J.b
a) University of Joensuu, Faculty of Forest Sciences, P.O. Box 111, FI-80101 Joensuu, Finland. Tel.+358 13 2513615. E-mail:
matti.maltamo@joensuu.fi, petteri.packalen@joensuu.fi, jussi.peuhkurinen@joensuu.fi, aki.suvanto@joensuu.fi,
annukka.pesonen@joensuu.fi
b) Finnish Geodetic Institute, Geodeetinrinne 2, P.O. Box 15, FI-02431 Masala, Finland, E-mail: juha.hyyppa@fgi.fi
KEY WORDS: remote sensing, forestry, inventory, modelling, prediction, recognition, lidar, accuracy
ABSTRACT
During last ten years the research concerning airborne laser scanning based forest inventory applications has been very active in
different parts of the world. In Finland, both basic approaches, single tree detection and area based modeling have been widely
examined. In the following the results of the ALS based forest inventory experiments and further possibilities in Finland are reviewed
and discussed. A short review of Finnish forestry in relation to possibilities of ALS based forest inventories is included as well.
Finally, some examples concerning the comparison of single tree detection and area based modeling and the usability of spatial
information provided by ALS data are presented.
1. INTRODUCTION
During last ten years the research concerning airborne laser
scanning (ALS) based forest inventory applications has been
very active in different parts of the world (e.g. Næsset, 1997;
Magnussen & Boudewyn, 1998; Hyyppä et al., 2001; Persson
et al., 2002; McCombs et al., 2003; Takahashi et al., 2005;
Tickle et al., 2006; Koch et al., 2006). Most of the studies
have been conducted by using discrete return small footprint
systems but there are also large footprint lidar applications as
well (Drake et al., 2002).
In Finland, both basic approaches to utilize ALS data, single
tree detection (Hyyppä & Inkinen, 1999; Hyyppä et al, 2001;
Maltamo et al., 2004a; Yu et al., 2006; Korpela, 2007;
Peuhkurinen et al., 2007) and area based modeling (Suvanto
et al., 2005; Maltamo et al., 2006a; c; Packalen & Maltamo,
2007) have been examined. In the following, the results of
the ALS based forest inventory experiments and further
possibilities in Finland are reviewed and discussed. Forest
inventories are usually multipurpose but here we concentrate
mainly on prediction methods of living tree stock. Some
examples concerning the comparison of single tree detection
and area based modeling as well as spatial information
provided by ALS data are also presented. However, to
understand the growing conditions and forest inventory
traditions in Finland a short review of Finnish forestry in
relation to possibilities of ALS based forest inventories is
also included.
2. FOREST INVENTORIES IN FINLAND
The forests of Finland are located in boreal vegetation zone.
The number of existing tree species is rather low, including
coniferous pine (Scots pine [Pinus sylvestris L.]) and spruce
species (Norway spruce [Picea abies L. karst]), and as
deciduous birches (Silver birch [Betula pendula Roth],
downy birch [Betula pubescens Ehrh.]), alders (grey alder [
Alnus incana], red alder [Alnus glutinosa]) and aspen
(European aspen [Populus tremula]) species. At stand level,
most of the forests are dominated either by pine or spruce.
There is no pure plantation forestry in Finland. Although
considerable proportion of regenerated stands have been
planted by one tree species the rotation age is so long that
other species usually naturally regenerate to stand. As a
result, a considerable proportion of stands are mixed at least
in some level. Only the least fertile stands, usually located in
northern Finland consist of pine only. Of course, silvicultural
treatments may also favour certain tree species. One specific
phenomenon in boreal forests is also high stand density by
means of number of stems whereas trees are rather small. For
example, in managed forests of Matalansalo test area, used in
several ALS studies, the average stand density is about 1500
stems per hectare and in mature stand strata still over 1250
stems per hectare (Suvanto et al., 2005). As a comparison, in
the study data by Heurich and Weinacker (2004) the stand
density in southeastern Germany on temperate forests was on
average 540 stems per hectare.
In Finland forest inventory is carried out on two levels:
National forest inventory (NFI) and forest management
planning. NFI is based on systematic cluster sampling of
field plots and covers whole country (Tomppo, 2006a). This
data is used for calculation of national and regional forest
resources and for national level planning. In addition, satellite
images are used in the multi-source National Inventory as an
auxiliary material (Tomppo, 2006b).
ALS data based sample plots is not a realistic alternative for
replacing field measured NFI sample plots. This is due to the
fact that one of the main requirements of NFI is that forest
resource results should be unbiased and this cannot be
quaranteed by using ALS data. Furthermore, information
concerning tree stock is only a minor part of the
measurements in these plots. Some other data needs consist
e.g. of forest health, biodiversity and forest soil variables and
most of these cannot be remote sensed (the head of the NFI
of Finland, Dr. Kari T. Korhonen, personal comm.). On the
other hand, large scale ALS data such as National Laser
Scanning (e.g. Artuso et al., 2003) could provide auxiliary
information for multi-source inventory in the form of digital
270
ISPRS Workshop on Laser Scanning 2007 and SilviLaser 2007, Espoo, September 12-14, 2007, Finland
terrain model (DTM), canopy height model (CHM) or stand
characteristics interpretation on systematic grid.
Forest management planning in private forests is usually
based on information collected by forest compartments
(stands) (Poso, 1983). A conventional inventory by
compartment includes expensive field work, but the number
of assessments per stand is typically small, resulting in low
precision of estimated stand variables. All assessments are
made on tree species level already in the field and as a final
product tree species specific timber sortiments are calculated.
The accuracy of prediction of stand total volume achieved in
compartment inventory usually varies between 15 and 30%
(Haara & Korhonen, 2004). For tree species the results are
even considerably worse. In fact, during recent years, the
costs and accuracy of conventional field work based small
area forest inventories have been at unsatisfactory levels
(e.g., Kangas & Maltamo, 2002). There is, therefore, an
increasing pressure to improve methods of carrying out field
inventories in small areas. The main approaches to
developing a compartment inventory have been the
modification of field measurements and the application of
remote sensing methods to support, or even replace, field
measurements. For the purpose of replacing current
inventory, ALS based methods have a very high potential.
Other forest inventory applications in Finland include
specific inventories for detailed purposes, such as wood
procurement planning or forest protection survey. These
inventories should produce very fine grained information of
variables of interest. In addition, the area they cover can vary
from one marked stand to large scale level. The usability of
ALS data in these inventories varies. For wood procurement
planning ALS based methods could provide detailed
information but due to the small area of target stands the cost
efficiency of data may not be sufficient. Concerning
characteristics of forest protection, some of them may be
mapped by ALS data and the others may be almost
impossible to recognise. In general, ecological information
can be obtained from ALS data (Hill et al., 2003; Hashimoto
et al., 2004)
All in all the characteristics of Finnish forests (low number of
tree species but usually more than one in stand, high stand
density, no fast growing plantations) and inventory output
needed (stand variables by tree species) define the
possibilities for ALS data to be applied. These possibilities
are further discussed later in this paper.
3. ALS EXPERIENCES OF FOREST
INVENTORY IN FINLAND
3.1 General
The first ALS based forest inventory studies in Finland were
based on single tree detection and high pulse density data
(Hyyppä & Inkinen, 1999; Hyyppä et al., 2001). In fact,
Hyyppä and Inkinen (1999) were among the first ones to
apply single tree detection with ALS data. The accuracy was
found to be superior already in these first studies the standard
error (without bias) being about 10% for stand volume. In the
studies by Hyyppä and Inkinen (1999) and Maltamo et al.
(2004b) the proportion of detected trees was only about 40%.
This was due to the multilayered and unmanaged stand
structure of the study area. Detailed information concerning
ALS studies in Finland before 2004 can be found from
review by Hyyppä et al. 2003. More recent developments in
Nordic countries and in boreal forests in general have been
reported e.g., by Næsset et al. (2006) and Hyyppä et al
(2007). In Finland, there has also been active research going
on concerning the quality of DTM construction in forested
areas (e.g. Hyyppä et al., 2005; Korpela & Välimäki, 2007).
In addition, the TerraScan software (by Arttu Soininen) from
Terrasolid Oy is assumed to be the global market leading
software concerning laser scanning processing.
3.2 Single tree detection
More recently the research of single tree approach has
concentrated on detection algorithms, recovery of undetected
trees, height growth and change detection. In addition, crown
height estimation and, especially, tree species recognition are
under growing research interest in Finland.
Concerning detection algorithms one problem on raster
canopy height models is handling of tree crowns of different
sizes. On laser scanner data one size attribute, height, is
directly available. This gives possibilities to develop
processing methods that adapt to the object size. In the study
by Pitkänen et al. (2004) three adaptive methods were
developed and tested for individual tree detection on CHM.
In the first method, the CHM was smoothed with canopy
height based selection of degree of smoothing and local
maxima on the smoothed CHM were considered as tree
locations. In the second and third methods, crown diameter
predicted from tree height was utilised. The second method
used elimination of candidate tree locations based on the
predicted crown diameter and distance and valley depth
between two locations studied. The third method was
modified from scale-space method used for blob detection.
Instead of automatic scale selection of the scale-space
method, the scale for Laplacian filtering, used in blob
detection, was determined according to the predicted crown
diameter.
Possibility to characterize suppressed trees that cannot be
detected has also been of interest. Maltamo et al. (2004a)
combined theoretical distribution functions and laser
scanning data to describe small and suppressed trees, which
tree crown segmentation methods was not able to detect. The
use of original point clouds instead of digital surface models
(DSM) or CHMs also gives possibilities for detection of
small trees. Since, some of the laser pulses will penetrate
under the dominant tree layer, it is also possible to analyze
multilayered stands. In Maltamo et al. (2005), the existence
and number of suppressed trees was examined. The results
showed that multilayered stand structures can be recognised
and quantified using quantiles of laser scanner height
distribution data. However, the accuracy of the results is
dependent on the density of the dominant tree layer.
Correspondingly, Mehtätalo (2006) used theoretical approach
to describe small trees. The probability of a tree being
observed was related to its height and was equal to the
proportion of the forest area not covered by taller trees.
Mehtätalo (2006) presented mathematical formula which was
based on the following assumptions: (i) trees are randomly
located within the stand and crown diameters within a stand
are uncorrelated, (ii) tree height increases as a function of
crown diameter, (iii) the tree crown forms a circle around the
tree tip, and (iv) a tree is invisible if the tree tip locates within
the crown of a taller tree. Furthermore, different approaches
were proposed for the correction of the censoring effect upon
271
IAPRS Volume XXXVI, Part 3 / W52, 2007
the observed distribution of crown areas. The used approach
provided theoretically accurate estimates for the distribution
of crown areas and the number of stems.
Yu et al. (2004) demonstrated the applicability of airborne
laser scanners in estimating height growth and monitoring
fallen or cut trees. Out of 83 field-checked fallen or cut trees,
61 were detected automatically and correctly. All the mature
cut trees were detected; it was mainly the smaller trees that
were not. Height growth was demonstrated at plot and stand
levels using an object-oriented tree-to-tree matching
algorithm and statistical analysis. In Yu et al. (2006) the
potential of measuring individual tree height growth of Scots
pine in boreal forest was analysed. Three different types of
variables were extracted from the point clouds representing
each tree: (i) the difference of highest z value, (ii) difference
between DSMs of tree tops and, (iii) difference of 85, 90 and
95% quartiles of the height histograms corresponding to a
crown. The results indicate that it is possible to measure the
height growth of an individual tree with multi-temporal laser
surveys.
Maltamo et al. (2006b) compared the results of the prediction
of crown height characteristics using ALS data and intensive
field measurements. Crown height models were constructed
both at the tree and plot level for Scots pine, Norway spruce
and birches. The ALS based models included independent
variables of tree levels, such as tree height, crown area and
independent plot-level variables, i.e. canopy height and
density quantiles and proportion of vegetation hits. The
results indicated that the ALS-based crown height models
were more accurate than the field-measurement-based models
when plot-level information was used as independent
variables. However, the field-measurement-based tree-level
models for Scots pine and Norway spruce were more
accurate than the ALS-based models. Even so, the accuracy
of the different models was very similar.
Related to wood procurement planning Peuhkurinen et al.
(2007) applied ALS data and field measurements to
characterize timber sortiments of two pure Norway spruce
marked stands. Pre-harvest measurement was realized by
using different methods as follows: (i) lidar-based individual
tree detection (see Pitkänen et al., 2004) and local
constructed dbh model (tree height as predictor), (ii) lidar-
based individual tree detection and existing regional dbh
model for spruce presented by Kalliovirta and Tokola (2005),
(iii) lidar-based individual tree detection and existing
regional dbh model (both tree height and maximum crown
diameter as predictors), (iv) systematic field plot sampling
data, (v) field inventory by compartments and, (vi) area
based canopy height distribution approach. The mean stand
variables were predicted with the models presented by
Suvanto et al. (2005). As a ground truth data harvester
measurements were used and the comparison of the methods
was based on bucking simulations.
Diameter distributions
0
20
40
60
80
100
120
8
12
16
20
24
28
32
36
40
44
48
52
56
60
Diameter class, cm
Number of trees
harvester data (reference)
individual tree detection from laser canopy height model
inventory b y comparme nt s
Fig. 1. Comparison of diameter distributions of single tree
detection, compartment inventory and harvester reference
data in a marked stand (Peuhkurinen et al. 2007).
The results of Peuhkurinen et al. (2007) illustrated
considerable advantage of lidar-based single tree detection
procedure compared to other studied methods in producing
pre-harvest measurement information. Single tree detection
with local dbh model (method (i)) was the most accurate
method by means of error index of diameter distribution
(Reynolds et al., 1988), saw wood and pulp wood volumes
and apportionment indexes used in relation with distribution
of logs. In fact, single tree detection found 2561 trees
whereas harvester data included 2638 trees, corresponding
figures for saw wood volumes were 1262 m3 and 1267 m3,
respectively. Predicted tree diameters were even able to
produce bi-modal shape of diameter distribution (Fig. 1).
Though, it must be noticed that the study by Peuhkurinen et
al. (2007) was done using two marked stands only and, thus,
has the nature of a case study.
Pyysalo (2006) developed 3D vector models of single trees
from ALS data in order to derive geometry features. The
vector model construction included four stages: (i) laser point
classification, (ii) DTM construction, (iii) extraction of points
from each individual tree and, (iv) vector model creation. The
extracted features were tree height, crown height, trunk
location, and crown profile. According to the derived results
tree shape is underestimated in vector models in both vertical
and horizontal direction Tree location were extracted with an
accuracy of 2 m and tree heights with an accuracy of 1.5 m
(Pyysalo, 2006).
Säynäjoki (2007) examined tree species classification
between aspen and other deciduous trees by using single tree
recognition of ALS data. Watershed segmentation was used
to create crown segments on the smoothed CHM (Pitkänen et
al., 2004). Crown segments of deciduous trees were used to
classify trees to aspen or other deciduous trees using linear
discriminant analysis. Classification accuracy between aspen
and other deciduous trees was as its best 79.1%. Predictors in
this classification function were proportion of vegetation hits,
standard deviation of pulse heights, accumulated intensity on
90th percentile and relation of proportions of laser points
reflected on 95th and 40th height percentiles. In addition to the
study by Säynäjoki (2007) there is a lot of research interest
going on in Finland to recognise tree species from single tree
detected ALS data. In Liang et al. (2007), it was shown that
the difference between first and last pulse is a valuable
feature for trees species classification. It gives reliable (89%
272
ISPRS Workshop on Laser Scanning 2007 and SilviLaser 2007, Espoo, September 12-14, 2007, Finland
accuracy) classification between coniferous and deciduous
trees under leaf-off conditions.
Finally, international EuroSDR/ISPRS Tree Extraction
project is coordinated by Finland (Hyyppä & Kaartinen
2006). The project includes twelve partners and the study
area is located in southern Finland. The aims of the project
are: (i) to compare different algorithms in tree extraction, (ii)
to study the effect of pulse density and (iii) to improve results
by combining ALS data and aerial images. The
characteristics to be compared are tree and tree species
detection and tree height estimation. The results clearly
showed that the variability of tree location accuracy is small
as a function of pulse density and it mainly changes as a
function of the provider. With the best models for all the
trees, the mean location error was less than 1 m and the
difference with 2, 4 and 8 pulses per m2 was negligible. With
trees over 20 m, the accuracy of tree location of 0.5 m was
obtained. Tree height quality analysis using selected 70
reference trees, the reference height was known with
accuracy of 10 cm, showed again that the variability of the
pulse density was negligible compared to method variability.
With best models RMSE of 50 to 80 cm was obtained for tree
height. Even the 2 pulses per m2 seemed to be feasible for
individual tree detection. Percentage of the found trees by
partners showed that the best algorithms found 90% of those
trees that were found at least by one of the partners. There
was again higher variation with the method used rather than
pulse density. The results of the test showed that the methods
of individual tree detection vary significantly and that the
method itself is more significant for individual tree based
inventories rather than the applied pulse density (Harri
Kaartinen and Juha Hyyppä, personal comm.).
3.3 Area based modelling
Research concerning area based modelling by using canopy
height distribution approach and low pulse density ALS data
started year 2004 in Finland (Suvanto et al., 2005; Maltamo
et al., 2006a). First studies confirmed the corresponding
accuracy observed in other Nordic countries (e.g. Næsset,
2002; 2004; Holmgren, 2004; Næsset et al., 2004). Juntunen
(2006) also made cost plus loss comparisons between ALS
based stand variables and characteristics of conventional
inventory by compartments (see e.g., Eid et al., 2004). When
compared to optical sensors canopy height distribution
approach was found to be more suitable alternative for the
next generation method for compartment inventory in
Finland (Uuttera et al., 2006). Instead of using regression
models in construction of stand variable models k-MSN
model was used by Maltamo et al. (2006c). The k-MSN
method is a non-parametric method, which uses canonical
correlation analysis to produce a weighting matrix used in the
selection of k Most Similar Neighbors from reference data.
Most Similar Neighbors are observations that according to
predictor variables are similar to the target of prediction.
When using k-MSN model the accuracy of stand volume was
improved when compared to regression models (Maltamo et
al., 2006). Additional information of aerial photographs or
stand register data further slightly improved the accuracy.
When constructing area based forest inventory application a
ground truth sample of accurately measured field plots is
needed. One possibility for reducing the costs lies in the use
of existing field plots for ground truth purposes. The most
obvious alternative in Finland is to use truncated angle count
sample plots of the National Forest Inventory. Due to the
lack of suitable angle count ground truth data and
corresponding laser data, Maltamo et al. (2007a) tested this
possibility using data on fixed area sample plots, in which
tree locations were simulated. The trees for a truncated angle
count sample plot were then chosen and the resulting data
together with the characteristics of an ALS -based canopy
height distribution were used to construct regression models
to predict stem volume, basal area, stem number, basal area
median diameter and the height. The accuracy of the stand
attributes was found to be almost as good as in the case of
models of fixed area plots. However, one drawback of this
study was that there were no field plots which were located
on stand edge. Such plots are typical for systematic sampling
based forest inventory applications, such as NFI of Finland.
Närhi (2007) tested the usability of area based canopy height
distribution approach to define the need and timing of
silvicultural treatment on Norway spruce sapling stands. Two
approaches were used: (i) ALS characteristics were directly
used to classify sapling stands according to treatment need by
using discriminant analysis and, (ii) regression models were
constructed for mean height and stand density
correspondingly as Næsset and Bjerknes (2001). After that,
the need and timing of silvicultural treatment was classified
according to these predicted characteristics. The results
indicated that overall accuracy of about 70% was achieved in
classification. The stands where there is a need for treatment
were found more accurately than those who did not have
need for that.
Basicly, area based canopy height distribution approach
produces stand variables, usually stand volume, stem
number, basal area, basal area median diameter and tree
height. However, ALS data can also be used to predict
parameters of a theoretical diameter distribution model of a
stand (Gobakken & Næsset, 2004; 2005). In Finland,
Maltamo et al. (2006a) compared prediction of diameter
percentiles and the use of predicted stand characteristics to
further predict Weibull parameters. More flexible and local
percentile based distribution was able to better describe
diameter distribution of heterogeneous stands.
In Maltamo et al. (2007b) the accuracy of ALS-based stem
frequency and basal area diameter distribution models by
using Weibull distribution were compared. Furthermore, the
usability of calibration estimation (see, e.g., Kangas &
Maltamo, 2002) to adjust the predicted distributions to be
compatible with the ALS based estimated stand volume was
presented. As a main result, the authors state that when
diameter distributions are predicted using ALS data, basal
area diameter distributions may not be needed. This
represents a considerable improvement in the inventory
system, since basal area is not in itself an interesting end-
product variable. When stem frequency distributions are
directly usable, this would provide a more realistic
description of the stand structure and generate simulations for
the further development of the tree stock.
Pesonen et al. (2007) analysed the potential of ALS data for
estimating coarse wood debris (CWD) volumes in
conservation area of the Koli National Park. The accuracy of
the ALS data proved adequate for predicting the downed
dead wood volume (RMSE 51.6%), whereas the standing
dead wood volume estimates were somewhat poorer (RMSE
78.8%). The downed dead wood volume estimates were
found to be substantially more accurate than traditional
predictions based on field measurements. Correspondingly,
273
IAPRS Volume XXXVI, Part 3 / W52, 2007
Kotamaa (2007) analysed the potential of ALS data for
estimating downed dead wood volumes in managed forests in
Juuka, eastern Finland. The accuracy was found to be
considerably worse. However, ALS data was able to
satisfactory classify plots whether they included downed
dead wood or not.
In all abovementioned studies (excluding Säynäjoki, 2007
and Liang et al., 2007) tree species have been ignored and
total tree stock has been considered. However, species-
specific stand characteristics are essential in Finland. To
solve this problem Packalén and Maltamo (2006) combined
information from ALS data with digital aerial photographs to
predict stand volume by tree species. Furthermore, Packalén
and Maltamo (2006; 2007) applied the non-parametric k-
MSN method to predict species-specific forest variables
volume, stem number, basal area, basal area median diameter
and tree height simultaneously for Scots pine, Norway spruce
and deciduous trees as well as total characteristics as sums of
the species-specific estimates. The combination of ALS data
and aerial photographs was used in these studies. The
predictor variables derived from the ALS data were based on
the height distribution of vegetation hits, whereas spectral
values and texture features were employed in the case of the
aerial photographs. The results showed that this approach can
be used to predict species-specific forest variables at least as
accurately as from the current stand-level field inventory for
Finland.
4. CALCULATION EXPERIMENTS
4.1 Comparison of single tree detection and canopy height
distribution approaches
Area based canopy height distribution and single tree based
approaches to utilise ALS data have been compared and
discussed in some reviews (Næsset et al., 2004; Hyyppä et
al., 2007). However, reliability figures presented earlier have
been based on different reliability characteristics and study
areas as well. Peuhkurinen et al. (2007) observed the better
accuracy of single tree detection in pre harvest measurement
case study. The example stand had rather low stand density
(465 stems per hectare). However, for forest inventory
purposes, comprehensive forest resource estimate should be
provided in relation to area to be considered, not just for
mature stands.
In this paper we theoretically compare these two approaches
in Matalansalo test area. This area has been earlier used in
several ALS studies (Suvanto et al., 2005; Maltamo et al.,
2006c; 2007a; b; Packalén & Maltamo, 2006; 2007). The
total size of the area is about 1200 hectares. There are a total
of 472 field sample plots (radius 9 meters) located in the area
and ALS campaign was conducted in summer 2004 using an
Optech ALTM 1233 laser scanning system operating at an
altitude of 1500 m above ground level. Sampling density of
the data was about 0.7 measurements per one square metre.
The pulse density does not allow individual tree detection
and, therefore, we simulated single tree approach as follows:
(i) it was expected that all trees were found, i.e. true tee
heights of all field measured trees were used, (ii) tree species
recognition produced 100% accuracy, i.e., tree species
recorded for each field measured tree was used and, (iii) tree
diameter was predicted with the help of tree height (h) either
by using existing, i.e. no calibration, regional regression
models by Kalliovirta and Tokola (2005) or from sample tree
material, i.e. calibrated by using about 1200 measurements,
constructed local tree diameter models. The model forms
were for Scots pine )( hfdbh = and for Norway spruce and
deciduous tree species )(hfdbh =. After that we calculated
stand volumes by using volume functions of Laasasenaho
(1982). As a result RMSE’s of 25.3% and 22.9% for plot
level volumes were obtained for regional and local models,
respectively. When compared these figures to canopy height
distribution approach based estimates of regression models
19.9% (Suvanto et al., 2005), k-MSN estimate 15.6%
(Maltamo et al., 2006c), species-specific k-MSN estimates
summed to plot level 20.5% (Packalén & Maltamo, 2007)
and diameter distribution based plot volume estimate 20.6%
(Maltamo et al., 2007b) it can be seen that the accuracy is
slight worse although it was expected that tree and tree
species detection totally succeeded. This is due to the fact
that the relationship between tree height and diameter is far
from deterministic. Allometric relationship between tree
diameter and height defines only certain limits for the
variation of these to variables, but characteristics such as
stand density, stand silvicultural history, genetic factors of
tree seed, tree position in a stand, site fertility, height above
sea level, distance from sea, mineral soil/peatland and stand
development class effect considerably to this relationship
In real world applications all trees and tree species would not
be detected, tree groups would cause some problems and tree
heights would be underestimates (e.g. Maltamo et al., 2004b),
but, the errors obtained here might not be increased
considerably due to the correlations between different errors
(Kangas, 1999). On the other hand, our simulation was not
able to take into consideration other variables produced by
single tree detection, usually tree crown area or diameter
(Hyyppä et al., 2001; Persson et al., 2002; Maltamo et al.,
2004a). In the study by Kalliovirta and Tokola (2005) the
increase in accuracy when maximum width of tree crown was
added to dbh/h model was 2.5 %-units in tree diameter
prediction. Furthermore, in Finnish conditions the effect of
tree crown area or diameter to increase accuracy of volume
prediction has been found to be 5-7 %-units (Maltamo et al.,
2004a; Villikka et al., 2007). Furthermore, the use of height
and density distributions of 3D point cloud of each detected
tree would additionally slightly improve the accuracy as
suggested by Villikka et al. (2007). Also some of the
characteristics mentioned in previous paragraph could be
used in dbh/h models or in the stratification of the data.
In fact, it is obvious in statistical manner that if the target
variable is stand volume, direct prediction model, as in the
case of canopy height distribution, is the most accurate
alternative to predict it. To further compare these two
approaches by using diameter distribution estimates we also
calculated error index presented by Reynolds et al. (1988):
=
=
K
ii
iffe 1
(1)
where and f
i
fi are the predicted and true frequency of
diameter class i, respectively, and K is the number of
diameter classes. The error index was calculated in 1-cm-
diameter classes for stem numbers at the plot level. Thus, the
error index of a given plot was the sum of the absolute
differences between the actual and predicted stem
frequencies of the diameter classes. Diameter distribution
274
ISPRS Workshop on Laser Scanning 2007 and SilviLaser 2007, Espoo, September 12-14, 2007, Finland
estimate was in the case of canopy height distribution
approach based on Weibull distribution (Maltamo et al.,
2007b). Correspondingly, diameter estimates of local
diameter models were used in the case of single tree
detection. Tree species recognition was not used in these
comparisons, i.e. distributions described tree total stock. As a
result, the average values of error indexes were 30.1 for
Weibull distribution based estimates and 30.8 for single tree
detection. In 239 plots Weibull distribution was more
accurate and in 211 plots single tree detection, in the rest (n=
22) these methods were as accurate by means of Reynold’s
error index.
Finally, RMSE figures for volumes of saw wood sized trees
(dbh>17 cm) were calculated. This characteristic
approximates saw wood proportion, an important variable
when deriving value or final cut decision of the stand. In the
case of Weibull distribution the RMSE was 32.2% and for
detected single trees, which diameters were predicted with
local tree diameter models, the RMSE was 40.3%.
At least when only tree height is used to predict tree diameter
single tree detection is not able to produce more accurate
common forest resource estimates than area based methods.
This is true for the stand volume as well as derived diameter
distribution or certain detailed part of tree stock. Single tree
detection does directly measure physical dimension of a tree
but tree height in itself is not an interesting variable in most
of the forest inventory applications. When considering
biodiversity aspects or certain habitat requirements stand
vertical structure is of primary interest but usually diameter
distribution and end products derived from it together with
tree height and tree species information are most important
output variables of forest inventory.
The accuracy of single tree detection could be improved by
calibrating tree diameter estimates at stand level. This would,
however, need field visit and measurement of GPS mapped
tree(s) and their height/diameter -relationship almost in each
target stand, i.e. extensive and very expensive reference data
for calibration is needed. Alternatively, single tree detection
based estimates, such as number of trees and stand volume,
could also be calibrated by using corresponding area based
ALS estimates. For general forest resource information field
calibration would not be cost-efficient, since such field visits
could take as much time as current compartment inventory in
Finland (on average 10 minutes per stand). Without
calibration there is, however, a high possibility of large errors
in operational inventories. For certain purposes, such as pre
harvest measurement of a marked stand in Finland single tree
detection could still be very interesting alternative expecting
that tree and tree species detection algorithms are highly
successful.
4.2 Usability of spatial information provided by ALS data
Both basic approaches to process ALS data for forest
information purposes also include some spatial information
which has not yet been utilised as much as possible. In the
case of single tree detection location and height of each
detected trees are obtained as well as same characteristics of
neighbouring trees. This allows us to calculate height based
competition indexes. In the following we calculate additive
competition index based on elevation angle sums (see Miina
& Pukkala, 2000):
)
)*8.0(
tan(
1
=
=
n
i
neighbour
dist
hh
aCI (2)
where hneighbour is height of neighbour tree, n is number of
neighbouring trees and dist is distance between target and
neighbour trees, maximum distance of 8 meters was taken
into consideration in the calculations.
This index was calculated for trees of two pine dominated
mapped field sample plots and corresponding individual tree
recognised ALS data including point density of 3.88 pulses
per square meter. ALS data was acquired in summer 2005
using an Optech ALTM 3100 scanner operating at a mean
altitude of 900 m above ground level. Example sample plots
were located on Koli National park and algorithms of
Pitkänen et al. (2004) were used for tree detection. Tree
coordinates and heights were needed in the calculations.
In the case of sparse density sample plot almost all trees were
detected and distributions CI-indexes were quite close to each
other (Fig. 2). All in all, ALS data was capable of producing
realistic estimates of competition indexes. On the other hand,
in dense sample plot less than 50% of trees were detected and
ALS data based CI estimates are not realistic (Fig. 3). In the
case of small trees there are only a few ALS detected trees
and also for larger trees underestimates are obtained since
there are too few neighbouring trees taken into consideration
in these estimates.
Further use of spatial indexes lies, e.g., in situation where tree
level distance dependent growth models are constructed (e.g.
Miina & Pukkala, 2000). Additional information that ALS
data could provide for such models would be spatial indexes
as described here, past height growth estimates from
multitemporal ALS data either at single tree level or plot
level as suggested by Yu et al. (2004, 2006) and information
included in laser based DTM, e.g. slope. Field measurements
including at least time series of two measurements are, of
course, needed for model construction, but when applying
such models a considerable amount of independent variable
information could come from ALS data.
0
0.05
0.1
0.15
0.2
0.25
0.3
0 5 10 15 20 25 30
Tree h eight, m
Comp etition inde x
ALS
field
Fig. 2. An example plot of spatial competition indexes
calculated from field measurements (n=44) and ALS detected
trees (n=39). Stand variables: basal area=24.5 m2ha-1, mean
height=24.7 m and number of stems = 489 per hectare.
275
IAPRS Volume XXXVI, Part 3 / W52, 2007
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
0 5 10 15 20 25 30
Tree height, m
Comp etition ind ex
ALS
field
Fig. 3. An example plot of spatial competition indexes
calculated from field measurements (n=151) and ALS
detected trees (n=67). Stand variables: basal area=31.7
m2ha-1, mean height= 24 m and number of stems= 1677 per
hectare
In the case of canopy height distribution approach spatial
information can be considered by using within stand
information provided by ALS based grid cells. In our
example two stands in Juuka test area are used. ALS data
were collected during summer 2005 using an Optech ALTM
3100 scanner operating at an altitude of 2000 m above
ground level resulting point density of about 0.6 pulses per
square meter. Field ground truth consists of systematic
sample (30 m distance) of angle count sample plots measured
originally for stand delineation purposes (Mr. Jukka
Mustonen, personal comm.). ALS based stand variable
estimation was based on the principle presented by Packalén
and Maltamo (2007). The size of the systematic grid cell was
16 m * 16 m which was close to original field sample plot
(radius 9 m).
0
5
10
15
20
25
2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40
Basal area, m2ha-1
Relativ e propor tion, %
Field measurement
ALS estimate
Fig. 4. Within stand distributions of field measured and ALS
based basal area estimates. Field measurements: mean basal
area 18 m2ha-1 and number of field measured angle count
plots 44. ALS: mean basal area 15.6 m2ha-1 and number of
grid cells 153. Area of the stand is 3.9 hectares.
As shown in the example figures (4 and 5) ALS data can
reproduce realistic estimates of within stand variation of
basal area. The averaging effect of models can be seen on
both ends of the distribution since extreme values are not
predicted. Compared to current field estimate of compartment
inventory which is only average value of basal area in the
stand this kind spatial information is of primary interest. As
within stand variation can be described, the need for
silvicultural operations such as thinnings can be more
accurately timed and spatial pattern at stand level can also be
defined. Of course, information presented in Figures 4 and 5
can also be produced by using single tree detection if as a
result of it high proportion of trees are detected and accurate
calibrated tree diameter model is used.
0
5
10
15
20
25
2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40
Basal area, m2ha-1
Relative proportion, %
Field measured
ALS estimate
Fig. 5. Within stand distributions of field measured and ALS
based basal area estimates. Field measurements: mean 24
m2ha-1 and number of field measured angle count plots 87.
ALS: mean basal area 22.9 m2ha-1 and number of grid cells
327. Area of the stand is 8.4 hectares.
5. CONCLUSIONS
In Finland there are numerous research activities going on
concerning the utilisation of ALS data in forest applications.
First commercial ALS applications for forest inventory
purposes were also introduced in Finland year 2006. This
paper reviewed and discussed most of research works
concentrating especially on forest inventory purposes. Some
further utilisation possibilities, such as spatiality and
biodiversity aspects in terms of CWD and large aspens, were
also considered and proposed. There are also numerous other
research topics, such as canopy cover, tree quality, forest
condition, stand delineation, forest planning, site
classification and forest structure which are currently
examined in Finland by using ALS data. From the point of
forest research both area based and single tree approaches
have a very high potential to be further developed and used in
novel applications. One possible application would also be a
combination of area based and single tree detection methods.
Also the rapid technological development of laser technology
gives new possibilities all the time.
A lot of interest is currently being shown especially in remote
sensing-based forest inventories in Finland, the driving force
being the possibility for reducing costs, although the potential
for improved accuracy is also important. Species-specific
stand characteristics are essential in Finland, because they are
used as an input to forest management planning. The
accuracies achieved in the study by Packalén and Maltamo
(2007) in the estimation of species-specific characteristics
were at least as good as those achieved with the current
inventory practise, but more testing must be carried out in
different types of forests with varying species compositions,
different geographical locations and different age
distributions before it is fully justified to conclude that the
combined use of ALS and aerial photographs prove superior
to a conventional inventory by compartments (Packalén,
276
ISPRS Workshop on Laser Scanning 2007 and SilviLaser 2007, Espoo, September 12-14, 2007, Finland
2006). The inclusion of young forests to inventory chain is
also one research topic to be further examined.
ALS-based tree-level forest inventories may become a
realistic alternative in the near future. Tree-level inventories
require denser ALS data but technological development will
mean that costs will decrease rapidly. An approach in which
aerial photographs are not needed for species recognition
would also be interesting at the individual tree level, but
more development work must still be done in the fields of
individual tree recognition, tree species classification,
modelling tree variables, especially tree diameter and the
inventory chain as a whole before tree-level inventories can
be valid operationally.
6. REFERENCES
Artuso, R., Bovet, S. & Streilein, A., 2003. Practical Methods
for the Verification of Countrywide Terrain and Surface
Models. ISPRS WG III/3 Workshop '3-D reconstruction from
airborne laser scanner and InSAR data', Dresden, Germany 8-
10 October 2003. The International Archives of
Photogrammetry, Remote Sensing and Spatial Information
Sciences, Vol. XXXIV, part 3/WG13.
Drake, J. B., Dubayah, R. O., Knox, R. G., Clark, D .B. &
Blair, J. B. 2002. Sensitivity of large-footprint lidar to canopy
structure and biomass in a neotropical rainforest. Remote
Sensing of Environment, 81, pp. 378–392.
Eid, T., Gobakken, T. & Næsset, E., 2004. Comparing stand
inventories based on photo interpretation and laser scanning
by means of cost-plus-loss analyses. Scandinavian Journal of
Forest Research, 19, pp. 512-523.
Gobakken, T. & Næsset, E., 2004. Estimation of diameter
and basal area distributions in coniferous forest by means of
airborne laser scanner data. Scandinavian Journal of Forest
Research, 19, pp. 529-542.
Gobakken, T. & Næsset, E., 2005.Weibull and percentile
models for lidar-based estimation of basal area distribution.
Scandinavian Journal of Forest Research, 20, pp. 490-502.
Haara, A. & Korhonen, K.T., 2004. Kuvioittaisen arvioinnin
luotettavuus. Metsätieteen aikakauskirja, 4/2004, pp. 489-
508. (In Finnish).
Hashimoto, H., Imanishi, J., Hagiwara, A., Morimoto, Y. &
Kitada, K., 2004. Estimating forest structure indices for
evaluation of forest bird habitats by an airborne laser-
scanner. International Archives of Photogrammetry, Remote
Sensing and Spatial Information Sciences, Vol. XXXVI, Part
8/W2, pp. 254–258.
Heurich, M. & Weinacker, H., 2004. Automated tree
detection and measurement in temperate forests of Central
Europe using laser-scamnning data. International Archives of
Photogrammetry, Remote Sensing and Spatial Information
Sciences, VOL. XXXVI(8/W2), pp. 198-203.
Hill, R.A., Hinsley, S.A., Gaveau, D.L.A. & Bellamy, P.E.,
2004. Predicting habitat quality for Great Tits (Parus major)
with airborne laser scanning data. International Journal of
Remote Sensing, 25, pp. 4851-4855.
Holmgren, J. 2004. Prediction of Tree Height, Basal Area
and Stem Volume in Forest Stands Using Airborne Laser
Scanning. Scandinavian Journal of Forest Research, 19, pp.
543-553.
Hyyppä, H., Yu, X., Hyyppä, J., Kaartinen, H., Honkavaara,
E. & Rönnholm, P., 2005. Factors affecting the quality of
DTM generation in forested areas. In Proceedings of ISPRS
Workshop Laser scanning 2005, September 12-14, 2005,
Enschede, Netherlands, (Netherlands: GITC bv), XXXVI
(Part 3/W19), pp. 85-90.
Hyyppä, J. & Inkinen, M., 1999. Detecting and estimating
attributes for single trees using laser scanner. The
Photogrammetric Journal of Finland, 16, pp. 27-42.
Hyyppä, J., Kelle, O., Lehikoinen, M., & Inkinen, M., 2001.
A segmentation-based method to retrieve stem volume
estimates from 3-dimensional tree height models produced by
laser scanner. IEEE Transactions on geoscience and remote
sensing, 39, pp. 969-975.
Hyyppä, J., Hyyppä, H., Maltamo, M., Yu, X., Ahokas, E. &
Pyysalo, U., 2003. Laser Scanning of Forest Resources -
Some of the Finnish Experience. In Proceedings of the
ScandLaser Scientific Workshop on Airborne Laser Scanning
of Forests, September 3-4, 2003, Umeå, Sweden, pp. 53-59.
Hyyppä, J. & Kaartinen, H., 2006. EuroSDR/ISPRS test on
tree extraction. Proceedings of the international concerence
Silvilaser 2006. November 7-10,2006. Ehime Prefectual
Convention Hall Matsuyama, Japan, pp. 75-79.
Hyyppä, J., Hyyppä, H., Leckie, D., Gougeon, F., Yu, X &
Maltamo, M., 2007. Methods of small-footprint airborne
laser scanning for extracting forest inventory data.
International Journal of Remote Sensing, In press.
Juntunen, R., 2006. Puustotiedon laadun vaikutus
metsänkäsittelyn optimoinnin tuloksiin – UPM Metsän
laserkeilausaineiston ja kuviotiedon vertailu. University of
Helsinki, Department of Forest Economics, Helsinki,
Finland. M.Sc. thesis, 94 p. (In Finnish).
Kalliovirta, J. & Tokola., T., 2005. Functions for estimating
stem diameter and tree age using tree height, crown width
and existing stand database information, Silva Fennica, 39,
pp. 227-248.
Kangas, A., 1999. Methods for assessing the uncertainty of
growth and yield predictions, Canadian Journal of Forest
Research, 292, pp. 1357-1364.
Kangas, A. & Maltamo, M., 2002. Anticipating the variance
of predicted stand volume and timber assortments with
respect to stand characteristics and field measurements. Silva
Fennica, 36, pp. 799-811.
Koch, B., Heyder, U. & Weinacker, H., 2006, Detection of
indidual tree crowns in airborne lidar data. Photogrammetric
Engineering & Remote Sensing, 72, pp. 1339-1348.
Korpela I., 2007. 3D treetop positioning by multiple image
matching of aerial images in a 3D search volume bounded by
lidar surface models. Photogrammetrie, Ferkundung,
Geoinformation, 1/2007, pp. 35-44.
277
IAPRS Volume XXXVI, Part 3 / W52, 2007
Korpela, I. & Välimäki, E., 2007. Talousmetsän
maanpintamallinnus arkistoilmakuvilta ja laserkeilauksella.
Maanmittaus, 82. (In Finnish).
Kotamaa, E., 2007. Lahopuun määrän ennustaminen
laserkeilauksella talousmetsissä. University of Joensuu,
Faculty of Forest Sciences, candidate’s thesis in in Forest
Planning and Economics. (In Finnish).
Laasasenaho, J., 1982. Taper curve and volume function for
pine, spruce and birch. Communicationes Instituti Forestalis
Fenniae, 108, pp. 1-74.
Liang, X., Hyyppä, J. & Matikainen, L., 2007. Deciduous-
coniferous tree classification using difference between first
and last pulse laser signatures. ISPRS Workshop on Laser
Scanning 2007 – Silvilaser 2007.
Magnussen, S., & Boudewyn, P. 1998. Derivations of stand
heights from airborne laser scanner data with canopy-based
quantile estimators. Canadian Journal of Forest Research,
28, pp. 1016-1031.
Maltamo, M., Eerikäinen, K., Pitkänen J., Hyyppä, J. &
Vehmas, M., 2004a. Estimation of timber volume and stem
density based on scanning laser altimetry and expected tree
size distribution functions. Remote Sensing of Environment,
90, pp. 319-330.
Maltamo, M., Mustonen, K., Hyyppä, J., Pitkänen, J. & Yu,
X., 2004b. The accuracy of estimating individual tree
variables with airborne laser scanning in a boreal nature
reserve. Canadian Journal of Forest Research, 34, pp. 1791-
1801.
Maltamo, M., Packalén, P., Yu, X., Eerikäinen, K., Hyyppä,
J. & Pitkänen, J., 2005. Identifying and quantifying structural
characteristics of heterogeneous boreal forests using laser
scanner data. Forest Ecology and Management, 216, pp. 41-
50.
Maltamo, M., Eerikäinen, K., Packalén, P. & Hyyppä, J.,
2006a. Estimation of stem volume using laser scanning based
canopy height metrics. Forestry, 79, pp. 217-229.
Maltamo, M., Hyyppä, J. & Malinen, J., 2006b. A
comparative study of the use of laser scanner data and field
measurements in the prediction of crown height in boreal
forests. Scandinavian Journal of Forest Research, 21, pp.
231 - 238
Maltamo, M., Malinen, J., Packalén, P., Suvanto, A. &
Kangas, J., 2006c. Non-parametric estimation of stem
volume using laser scanning, aerial photography and stand
register data. Canadian Journal of Forest Research, 36, pp.
426-436.
Maltamo, M., Korhonen, K.T., Packalén, P., Mehtätalo, L. &
Suvanto A., 2007a. A test on the usability of truncated angle
count sample plots as ground truth in airborne laser scanning
based forest inventory. Forestry, 80, pp. 73-81.
Maltamo, M., Suvanto, A. & Packalén, P., 2007b.
Comparison of basal area and stem frequency diameter
distribution modelling using airborne laser scanner data and
calibration estimation. Forest Ecology and Management, In
press.
McCombs, J.W., Roberts, S.D. & Evans, D.L., 2003.
Influence of fusing lidar and multiscpectral imagery on
remotely sensed estimates of stand density and mean tree
height in a managed loblolly pine plantation. Forest Science,
49, pp. 457-466.
Mehtätalo, L., 2006. Eliminating the effect of overlapping
crowns from aerial inventory estimates. Canadian Journal of
Forest Research, 36, pp. 1649-1660.
Miina, J. & Pukkala, T., 2000. Using numerical optimization
for specifying individual tree competituion models. Forest
Science, 46, pp. 277-283.
Närhi, M., 2007. Kuusen taimikoiden inventointi ja
taimikonhoidon kiireellisyyden määrittäminen
laserkeilauksen avulla. University of Joensuu, Faculty of
Forest Sciences, master’s thesis in Management of Forest
Ecosystems. (In Finnish).
Næsset, E., 1997. Estimating timber volume of forest stands
using airborne laser scanner data. Remote Sensing of
Environment, 51, pp. 246-253.
Næsset, E., 2002. Predicting forest stand characteristics with
airborne scanning laser using a practical two-stage procedure
and field data. Remote Sensing of Environment, 80, pp. 88-
99.
Næsset, E., 2004. Practical large-scale forest stand inventory
using a small airborne scanning laser. Scandinavian Journal
of Forest Research, 19, pp. 164-179.
Næsset, E & Bjerknes, K-O., 2001. Estimating tree heights
and number of stems in young forest stands using airborne
laser scanner data. Remote Sensing of Environment, 78, pp.
328–340
Næsset, E., Gobakken, T., Holmgren, J., Hyyppä, J., Hyyppä,
H., Maltamo, M., Nilsson, M., Olsson, H., Persson, Å. &
Söderman, U., 2004. Laser scanning of forest resources: the
Scandinavian experience. Scandinavian Journal of Forest
Research, 19, pp. 482-499.
Næsset, E., Gobakken, T., Kangas, A., Maltamo, M.
Bollandsås, O.M., Aasland, T. & Solberg, S., 2006.
Extending and improving methods for operational stand-wise
forest inventories utilizing multi-resolution airborne laser
scanner data. In Proceedings of the international concerence
Silvilaser 2006. November 7-10,2006. Ehime Prefectual
Convention Hall Matsuyama, Japan, pp. 75-79.
Packalén, P., 2006. Using airborne laser scanning data and
digital aerial photographs to estimate forest growing stock.
Licenciate thesis. University of Joensuu.
Packalén, P. & Maltamo, M., 2006. Predicting the volume by
tree species using airborne laser scanning and aerial
photographs. Forest Science, 52, pp. 611-622.
Packalén, P. & Maltamo, M., 2007. The k-MSN method in
the prediction of species specific stand attributes using
airborne laser scanning and aerial photographs. Remote
Sensing of Environment, 109, pp. 328-341.
278
ISPRS Workshop on Laser Scanning 2007 and SilviLaser 2007, Espoo, September 12-14, 2007, Finland
Persson, Å., Holmgren, J. & Söderman, U., 2002. Detecting
and measuring individual trees using an airborne laser
scanner. Photogrametric Engineering & Remote Sensing, 68,
pp. 925-932.
Pesonen, A, Maltamo, M., Eerikäinen, K. & Packalèn, P.,
2007. Airborne laser scanning-based prediction of coarse
woody debris volumes in a conservation area. Submitted
manuscript.
Peuhkurinen, J., Maltamo, M., Malinen, J., Pitkänen, J. &
Packalén, P., 2007. Pre-harvest measurement of marked stand
using airborne laser scanning. Forest Science, In press.
Pitkänen, J., Maltamo, M., Hyyppä, J. & Yu, X., 2004.
Adaptive methods for invidual tree detection on airborne
laser based height model. International Archives of
Photogrammetry, Remote Sensing and Spatial Information
Sciences, XXXVI(8/W2), pp. 187-191.
Poso, S., 1983. Kuvioittaisen arvioimismenetelmän
perusteita. English summary: Basic features of forest
inventory by compartments. Silva Fennica, 17, pp. 313-349.
Pyysalo, U., 2006. Single tree feature extraction from
airborne laser scanner data applying vector models. Thesis
for degree of Licentiate of Science In Technology.
Department of Surveying at Helsinki University of
Technology.
Reynolds, M.R. JR., T.E. Burk, & Huang, W.-C., 1988.
Goodness-of-fit tests and model selection procedures for
diameter distribution models. Forest Science, 34, pp. 373–
399.
Säynäjoki, R., 2007. Classification of aspen and other
deciduous trees using high resolution remote sensing data.
University of Joensuu, Faculty of Forest Sciences, master’s
thesis in Forest Planning and Economics.
Suvanto, A., Maltamo, M., Packalén, P. & Kangas, J., 2005.
Kuviokohtaisten puustotunnusten ennustaminen
laserkeilauksella. Metsätieteen aikakauskirja, 4/2005, pp.
413-428. (In Finnish).
Takahashi, T., Yamamoto, K., Senda, Y. & Tsuzuku, M.,
2005. Predicting individual stem volumes of sugi
(Cryptomeria japonica D. Don) plantations in mountainous
areas using small-footprint airborne LiDAR. Journal of
Forest Research, 10, pp. 305–312.
Tickle, P.K., Lee, A., Lucas, R.M., Austin, J. & Witte C.,
2006. Quantifying Australian forest floristics and structure
using small footprint LiDAR and large scale aerial
photography. Forest Ecology and Management, 223, pp. 379-
394.
Tomppo, E., 2006a. The Finnish National Forest Inventory.
In: Kangas, A. & Maltamo, M. (eds.). Forest inventory.
Methodology and applications. Managing Forest Ecosystems,
Vol 10, Springer, Dordrecht, pp. 179-194.
Tomppo, E., 2006b. The Finnish multi-source National
Forest Inventory - small area estimation and map production.
In: Kangas, A. & Maltamo, M. (eds.). Forest inventory.
Methodology and applications. Managing Forest Ecosystems,
Vol 10, Springer, Dordrecht, pp. 195-224.
Uuttera, J, Anttila, P., Suvanto, A. & Maltamo, M., 2006.
Yksityismetsien metsävaratiedon keruuseen soveltuvilla
kaukokartoitusmenetelmillä estimoitujen puustotunnusten
luotettavuus. Metsätieteen aikakauskirja, 4/2006, pp. 507-
519. (In Finnish).
Villikka, M., Maltamo, M., Packalén, P., Vehmas, M. &
Hyyppä, J., 2007 Alternatives for predicting tree-level stem
volume of Norway spruce using airborne laser scanner data.
Submitted manuscript.
Yu, X., Hyyppä, J., Kaartinen, H. & Maltamo, M., 2004.
Automatic Detection of Harvested Trees and Determination
of Forest Growth using Airborne Laser Scanning. Remote
Sensing of Environment, 90, pp. 451-462.
Yu, X., Hyyppä, J., Kukko, A., Maltamo, M. & Kaartinen,
H., 2006. Change detection techniques for canopy height
growth measurements using airborne laser scanner data.
Photogrammetric Engineering & Remote Sensing, 72, pp.
1339-1348
279
IAPRS Volume XXXVI, Part 3 / W52, 2007
  • Article
    Full-text available
    Aim of study: To test the use of LiDAR data from a single acquisition in order to estimate volume overbark variations ina 5-yr period of Pinus radiata D. Don. Area of study: Province of Bizkaia in the Autonomous Community of the Basque Country (Spain). Material and methods: Two field plot measurements were made in 2011 and 2015 and two wood volume models (one for each year) were fitted using the metric variables of the 2012 LiDAR points cloud. The models were applied to a 26.59 m raster covering the study area and the increase in volume at each pixel was calculated by subtraction. Main results: The increase in estimated wood volume, when added to the volume of timber extracted in the area during the 5-yr period under consideration, yielded an average increase of 13.74 m3 ha-1 yr-1, which corresponds to the average growth of the P. radiata in that area. The harvest area estimated using this procedure largely coincides with the actual harvest area in the same period. The value of R2 (85%) of the wood volume model for 2011 is similar to that obtained in other studies. However, as expected, the one obtained for the wood volume model for 2015 (80%) is significantly lower. Research highlights: The increase in wood volume can be estimated using a single LiDAR flight and field data from the 5-yr period provided that data from plots subjected to this kind of harvest is included in the models.
  • Article
    Airborne Light Detection and Ranging (LiDAR) and Landsat data were evaluated as auxiliary information with the intent to increase the precision of growing stock volume estimates in field-based forest inventories. The aim of the study was to efficiently utilize both wall-to-wall Landsat data and a sample of LiDAR data using model-assisted estimation. Variables derived from the Landsat 7 ETM + satellite image were spectral values of blue, green, red, near infra-red (IR), and two shortwave IR (SWIR) bands. From the LiDAR data twenty-six height and density based metrics were extracted. Field plots were measured according to a design similar to the 10th Finnish National Forest Inventory, although with an increased number of plots per area unit. The study was performed in a 30 000 ha area of Kuortane, Western Finland. Three regression models based on different combinations of auxiliary data were developed, analysed, and applied in the model-assisted estimators. Our results show that adding auxiliary Landsat and LiDAR data improves estimates of growing stock volume. Very precise results were obtained for the case where wall-to-wall Landsat data, LiDAR strip samples, and field plots were combined; for simple random sampling of LiDAR strips the relative standard errors (RSE) were in the range of 1–4%, depending on the size of the sample. With only LiDAR and field data the RSEs ranged from 4% to 25%. We also showed that probability-proportional-to-size sampling of LiDAR strips (utilizing predicted volume from Landsat data as the size variable) led to more precise results than simple random sampling.
  • Chapter
    Dieses Kapitel gibt einen Überblick über Methoden zur Segmentierung und Klassifikation von Objekten, die mit modernen flugzeuggestützten LiDAR-Systemen erfasst werden. LiDAR-Systeme der neuesten Generation weisen eine hohe Punktmessrate auf und können die überflogenen Objekte mit hoher Punktdichte abtasten. Gleichzeitig zeichnen sie den ausgesendeten Laserimpuls und das reflektierte Echosignal mit hoher geometrischer Auflösung auf. Die daraus resultierende dichte Objektrepräsentation ermöglicht die erfolgreiche Anwendung von speziellen Methoden aus dem Bereich der Computer Vision und des maschinellen Lernens. Insbesondere bei Anwendungen im Forstbereich eröffnen sich dadurch neue Möglichkeiten, Waldobjekte zu detektieren und in ihren Eigenschaften zu charakterisieren.
  • Chapter
    Dieses Kapitel gibt einen Überblick über Methoden zur Segmentierung und Klassifikation von Objekten, die mit modernen flugzeuggestützten LiDAR-Systemen erfasst werden. LiDAR-Systeme der neuesten Generation weisen eine hohe Punktmessrate auf und können die überflogenen Objekte mit hoher Punktdichte abtasten. Gleichzeitig zeichnen sie den ausgesendeten Laserimpuls und das reflektierte Echosignal mit hoher geometrischer Auflösung auf. Die daraus resultierende dichte Objektrepräsentation ermöglicht die erfolgreiche Anwendung von speziellen Methoden aus dem Bereich der Computer Vision und des maschinellen Lernens. Insbesondere bei Anwendungen im Forstbereich eröffnen sich dadurch neue Möglichkeiten, Waldobjekte zu detektieren und in ihren Eigenschaften zu charakterisieren.
  • Article
    The objective of research is to find DBH prediction models that: use variables derived from remote sensing, field mensuration and previous forest inventory data; can be used in STRS methods; are suitable for Latvian forest conditions. In paper different tree DBH predicting models from field and remote sensing data were researched. The study site is a forest in middle of Latvia at Jelgava district (56°39' N, 23°47' E). The area consists of mixed coniferous and deciduous forest with different age, high density, complex structure, various components, composition and soil conditions. Aerial photography camera (ADS 40) and laser scanner (ALS 50 II) was used to capture the data. LiDAR resolution is 9p/m2 (500 m altitude). The image data is RGB, NIR and PAN spectrum with 20 cm pixel resolution. Image processing was made using Fourier transform, frequency filtering and reverse Fourier transform. LiDAR data processing methods was based on canopy height model, Gaussian mask and local maxima. Field measurements are tree coordinates, species, height, diameter at breast height, crown width. Totally seven different linear models were developed, using data collected. General linear model that predicts DBH includes a tree height, effective crown area, soil type and age factors. It showed strongest relationship between predicted and measured DBH (R2 = 0,872). Summary results show that the models predict DBH reasonably well and factors included in all models are significant. Using combined LiDAR and optical imagery data is able to detect at least 63 % of all trees and about 85% of the dominant trees. Not identified trees at 82% of cases diameter at breast height was less than 20 cm and 88% of cases height was less than 20 m. Relationship between the Lidar detected height and observed total height shows showed strong relationship (R2 = 0,986), also between Lidar and aerial photography detected and observed tree crown is strong relationship (R 2 = 0,869).
  • Thesis
    Full-text available
    The importance of single-tree-based information for forest management and related industries in countries like Sweden, which is covered in approximately 65% by forest, is the motivation for developing algorithms for tree detection and species identification in this study. Most of the previous studies in this field are carried out based on aerial and spectral images and less attention has been paid on detecting trees and identifying their species using laser points and clustering methods. In the first part of this study, two main approaches of clustering (hierarchical and K-means) are compared qualitatively in detecting 3-D ALS points that pertain to individual tree clusters. Further tests are performed on test sites using the supervised k-means algorithm in which the initial clustering points are defined as seed points. These points, which represent the top point of each tree are detected from the cross section analysis of the test area. Comparing those three methods (hierarchical, ordinary K-means and supervised K-means), the supervised K-means approach shows the best result for clustering single tree points. An average accuracy of 90% is achieved in detecting trees. Comparing the result of the thesis algorithms with results from the DPM software, developed by the Visimind Company for analysing LiDAR data, shows more than 85% match in detecting trees. Identification of trees is the second issue of this thesis work. For this analysis, 118 trees are extracted as reference trees with three species of spruce, pine and birch, which are the dominating species in Swedish forests. Totally six methods, including best fitted 3-D shapes (cone, sphere and cylinder) based on least squares method, point density, hull ratio and slope changes of tree outer surface are developed for identifying those species. The methods are applied on all extracted reference trees individually. For aggregating the results of all those methods, a fuzzy logic system is used because of its good reputation in combining fuzzy sets with no distinct boundaries. The best-obtained model from the fuzzy system provides 73%, 87% and 71% accuracies in identifying the birch, spruce and pine trees, respectively. The overall obtained accuracy in species categorization of trees is 77%, and this percentage is increased dealing with only coniferous and deciduous types classification. Classifying spruce and pine as coniferous versus birch as deciduous species, yielded to 84% accuracy.
  • Article
    Full-text available
    The study develops equations for the prediction of stem volume of Milicia excelsa (Welw C.C. Berg.) in some selected institutions in Ibadan, Oyo State, Nigeria. Sequel to the relationship between stem volume (sv), diameter at base (db) and at breast height (dbh) from enumeration of 61 trees in selected institutions, equations were developed for estimating tree volumes of M. excelsa. Of all the equations developed, logarithmic for volume was the best fit equation containing the db and dbh as the predictors. The equation is lnSV = 1.5924 + 1.4915lndb + 0.8600lndbh with coefficient of determination (R 2) and standard error of estimate being 0.9011 and SEE = 0.3485 respectively. Residual analysis revealed that the assumption of independence of residuals is valid, and there is no evidence of an outlier. Validation of the equation was done by testing for significant difference between the predicted stem volume (PSV) and observed stem volume (OSV). The study showed that stem volume of M. excelsa can be predicted from db and dbh by using this equation with reasonable precision. The prediction equation developed in this work would be very useful and applicable where tree dimensions such as diameter at middle and top as well as the height of M. excelsa is difficult to assess and there is need to reduce the cost of inventory of such species.
  • Thesis
    Full-text available
    Over the past decades it has been shown that remotely sensed auxiliary data have a potential to increase the precision of key estimators in sample-based forest surveys. This thesis was motivated by the increasing availability of remotely sensed data, and the objectives were to investigate how this type of auxiliary data can be used for improving both the design and the estimators in sample-based surveys. Two different modes of inference were studied: model-based inference and design-based inference. Empirical data for the studies were acquired from a boreal forest area in the Kuortane region of western Finland. The data comprised a combination of auxiliary information derived from airborne LiDAR and Landsat data, and field sample plot data collected using a modification of the 10th Finnish National Forest Inventory. The studied forest attribute was growing stock volume. In Paper I, remotely sensed data were applied at the design stage, using a newly developed design which spreads the sample efficiently in the space of auxiliary data. The analysis was carried out through Monte Carlo sampling simulation using a simulated population developed by way of a copula technique utilizing empirical data from Kuortane. The results of the study showed that the new design resulted in a higher precision when compared to a traditional design where the samples were spread only in the space of geographical data. In Paper II, remotely sensed auxiliary data were applied in connection with model-assisted estimation. The auxiliary data were used mainly in the estimation stage, but also in the design stage through probability-proportional-to-size sampling utilizing Landsat data. The results showed that LiDAR auxiliary data considerably improved the precision compared to estimation based only on field samples. Additionally, in spite of their low correlation with growing stock volume, adding Landsat data as auxiliary data further improved the precision of the estimators. In Paper III, the focus was set on model-based inference and the influence of the use of different models on the precision of estimators. For this study, a second simulated population was developed utilizing the empirical data, including only non-zero growing stock volume observations. The results revealed that the choice of model form in model-based inference had minor to moderate effects on the precision of the estimators. Furthermore, as expected, it was found that model-based prediction and model-assisted estimation performed almost equally well. In Paper IV, the precision of model-based prediction and model-assisted estimation was compared in a case where field and remotely sensed data were geographically mismatched. The same simulated population as used in Paper III was employed in this study. The results showed that the precision in most cases decreased considerably, and more so when LiDAR auxiliary data were applied, compared to when Landsat auxiliary data were used. As for the choice of inferential framework, it was revealed that model-based inference in this case had some advantages compared to design-based inference through model-assisted estimators. The results of this thesis are important for the development of forest inventories to meet the requirements which stem from an increasing number of international commitments and agreements related to forests. Keywords: design-based, Landsat, LiDAR, model-based, multivariate probability distribution, sampling.
  • Article
    Full-text available
    Since it is impossible to measure diameters at breast height (dbh) directly from aerial photographs, existence of reliable dbh estimation models is crucial for the application of photogrammetric method in forest stands measurements. Research of relationships and creation of mathematic models for correlations between diameter at breast height and tree variables measured on aerial photographs (crown diameter, tree height, tree number etc.) was therefore the object of numerous scientific studies. Main goal of this paper was to create the regression models for main tree species (Sessile oak, Common beech, European hornbeam and Black alder) dbh estimation in “Donja Kupčina – Pisarovina” forest management unit of uneven-aged, privately owned, forests located at hilly regions. These models would serve as a prerequisite for the application of photogrammetric method in forest stands measurements, by using contemporary tools and digital photogrammetry techniques. Based on the former studies, and keeping in mind the heterogeneous structure of the researched stands, in dbh modelling were used two independent variables. First model (dM1) used crown diameter and tree height, while second model (dM2) used crown projection area and tree height for the above mentioned variables. Field measurements of stands’ structural elements (diameter at breast height, crown diameter and tree height) needed for creating regression models was conducted on the sample of 383 trees in total (103 Sessile oak trees, 103 Common beech trees, 127 European hornbeam trees and 50 Black alder trees), distributed through 6 chosen compartments (16 to 21) at “Donja Kupčina – Pisarovina” forest management unit. Conducted partial correlation confirmed the statistical significance of all independent variables (crown diameter, crown area and tree height) planned for model creation. Multiple regression analysis confirmed the statistical significance of all created models – both model types (dM1 and dM2) and all tree species (Sessile oak, Common beech, European hornbeam and Black alder). Modelling results have shown that independent variables crown diameter and tree height from the first model (dM1), as well as the crown projection area and tree height from the second model (dM2) explain the variability of diameter at breast height with high values of determination coefficients (R2 > 0,76). By comparing the results based on tree species between different model types, it was determined that the first model for Sessile oak and Common beech shows better results – 4% higher values of determination coefficient, and lesser error values in dbh estimation, expressed through root mean square error (RMSE), were obtained. For European hornbeam and Black alder both models produced almost identical results regarding R2 and RMSE values. Since crown diameter and tree height variables in Common beech model (dM1) increase the explained dbh variability by only 3%, using the simpler model with crown diameter as the only independent variable is recommended for estimating Common beech diameter at breast height, especially if tree height is not measured during the photogrammetric measuring. Based on the obtained results and regression analysis parameters for each individual model, as well as the results from graphic and analytic testing of each individual model, we can conclude that created regression models can be used for dbh estimation in “Donja Kupčina – Pisarovina” forest management unit of uneven-aged, privately owned forests located at hilly regions, and also in forest stands with similar characteristics. In order to confirm the possibility of practical application for created regression models it is necessary to conduct a photogrammetric measurement of forest stands in “Donja Kupčina – Pisarovina” forest management unit, and to compare the obtained results and the costs of its application to terestrial measurements. Keywords: diameter at breast height, crown diameter, crown projection area, tree height, modelling, multiple regression analysis
  • Chapter
    Airborne laser scanning (ALS) has emerged as one of the most promising remote sensing technologies to provide data for research and operational applications in a wide range of disciplines related to management of forest ecosystems. This chapter starts with a brief historical overview of the early forest-related research on airborne Light Detection and Ranging which was first mentioned in the literature in the mid-1960s. The early applications of ALS in the mid-1990s are also reviewed. The two fundamental approaches to use of ALS in forestry applications are presented – the area-based approach and the single-tree approach. Many of the remaining chapters rest upon this basic description of these two approaches. Finally, a brief introduction to the broad range of forestry applications of ALS is given and references are provided to individual chapters that treat the different topics in more depth. Most chapters include detailed reviews of previous research and the state-of-the-art in the various topic areas. Thus, this book provides a unique collection of in-depth reviews and overviews of the research and application of ALS in a broad range of forest-related disciplines.
  • Article
    High-resolution airborne laser scanner data offer the possibility to detect and measure individual trees. In this study, an algorithm which estimated position, height, and crown diameter of individual trees was validated with field measurements. Because all the trees in this study were measured on the ground with high accuracy, their positions could be linked with laser measurements, making validation on an individual tree basis possible. In total, 71 percent of the trees were correctly detected using laser scanner data. Because a large portion of the undetected trees had a small stem diameter, 91 percent of the total stem volume was detected. Height and crown diameter of detected trees could be estimated with a root-mean-square error (RMSE) of 0.63 m and 0.61 m, respectively. Stem diameter was estimated, using laser measured tree height and crown diameter, with an RMSE of 3.8 cm. Different laser beam diameters (0.26 to 3.68 m) were also tested, the smallest beam size showing a better detection rate in dense forest. However, estimates of tree height and crown diameter were not affected much by different beam size.
  • Article
    Full-text available
    Laser scanning provides a good means to collect information on forest stands. This paper presents an approach to delineate single trees automatically in small footprint light detection and ranging (lidar) data in deciduous and mixed temperate forests. In rasterized laser data possible tree tops are detected with a local maximum filter. Afterwards the crowns are delineated with a combination of a pouring algorithm, knowledge-based assumptions on the shape of trees, and a final detection of the crown-edges by searching vectors starting from the trees’ tops. The segmentation results are assessed by comparison with terrestrial measured crown projections and with photogram-metrically delineated trees. The segmentation algorithm works well for coniferous stands. However, the implemented method tends to merge crowns in dense stands of deciduous trees.
  • Article
    This paper analyzes the potential of airborne laser scanner data for measuring individual tree height growth in a boreal forest using 82 sample trees of Scots pine. Point clouds (10 points/m2, beam size 40 cm) illuminating 50 percent of the treetops were acquired in September 1998 and May 2003 with the Toposys 83 kHz lidar system. The reference height and height growth of pines were measured with a tacheometer in the field. Three different types of features were extracted from the point clouds representing each tree; they were the difference between the highest z values, the difference between the DSMs of the tree crown, and the differences between the 85th, 90th and 95th percentiles of the canopy height histograms corresponding to the crown. The best correspondence with the field measurements was achieved with an R2 value of 0.68 and a RMSE of 43 cm. The results indicate that it is possible to measure the growth of an individual tree with multi-temporal laser surveys. We also demonstrated a new algorithm for tree-to-tree matching. It is needed in operational growth estimation based on individual trees, especially in dense spruce forests. The method is based on minimizing the distances between treetops in the N-dimensional data space. The experiments showed that the use of the location (derived from laser data) and height of the trees were together adequate to provide reliable tree-to-tree matching. In the future, a fourth dimension (the crown area) should also be included in the matching.
  • Article
    In this paper, a deciduous-coniferous tree classification mechanism is proposed, tested and analyzed using solely laser scanner data. The data were acquired under leaf-off conditions by Toposys II system. Under such circumstance, sources of last pulse hits of deciduous and coniferous are different, which allows concise discrimination between these two species. Tree positions were located from first pulse DSM, species were identified by the difference between two pulse data and field measurements were used for validation. The classification results demonstrate that first-last pulse laser data, under leaf-off condition, is ideal for deciduous and coniferous trees classification; and also indicate that the data collected for high accuracy DEM production is also suitable for forest investigation. 1.
  • Article
    Good estimates of the precision and accuracy of stand growth and yield predictions are needed in the decision-making process. Future growth and yield are often projected with a complex simulation system, so that these assessments are not easy. The precision of long-term growth and yield predictions has often been estimated through Monte Carlo simulation, combining several error sources with different variances. A simple method to assess the uncertainty of stand growth and yield predictions is to model the observed (past) errors obtained by comparing the observed stand characteristic with the characteristic predicted using the simulation system. Using this model, the uncertainty of future predictions can then be anticipated. Another possibility is to form an elementary model for stand growth (or yield) and use its variance as an assessment of the uncertainty for the simulation system. In this paper, the above-mentioned three methods were used for assessing the uncertainty of a simulation system. These assessments were then compared with the empirical estimates of uncertainty. The three methods were also compared with respect to their data requirements and the capabilities of the methods.
  • Article
    Mean tree height, dominant height, mean diameter, stem number, basal area and timber volume of 116 georeferenced field sample plots were estimated from various canopy height and canopy density metrics derived by means of a small-footprint laser scanner over young and mature forest stands using regression analysis. The sample plots were distributed systematically throughout a 6500 ha study area, and the size of each plot was 232.9 m 2 . Regressions for coniferous forest explained 60-97% of the variability in ground reference values of the six studied characteristics. A proposed practical two-phase procedure for prediction of corresponding characteristics of entire forest stands was tested. Fifty-seven test plots within the study area with a size of approximately 3740 m 2 each were divided into 232.9 m 2 regular grid cells. The six examined characteristics were predicted for each grid cell from the corresponding laser data using the estimated regression equations. Average values for each test plot were computed and compared with ground-based estimates measured over the entire plot. The bias and standard deviations of the differences between predicted and ground reference values (in parentheses) of mean height, dominant height, mean diameter, stem number, basal area and volume were −0.58 to −0.85 m (0.64-1.01 m), −0.60 to −0.99 m (0.67-0.84 m), 0.15-0.74 cm (1.33-2.42 cm), 34-108 ha −1 (97-466 ha −1 ), 0.43-2.51 m 2 ha −1 (1.83-3.94 m 2 ha −1 ) and 5.9-16.1 m 3 ha −1 (15.1-35.1 m 3 ha −1 ), respectively.