ìeijing 2008 
27 
ACCURACY OF CROWN SEGMENTATION AND ESTIMATION OF SELECTED TREES 
AND FOREST STAND PARAMETERS IN ORDER TO RESOLUTION OF USED DSM 
AND NDSM MODELS GENERATED FROM DENSE SMALL FOOTPRINT LIDAR DATA 
K. Sterenczak 3 , K. B?dkowski a , H. Weinacker 
department of Forest Management, Geomatics and Economics, Warsaw University of Life Sciences - SGGW, 
159 Nowoursynowska str., PL-02-776 Warszawa, (e-mail: - Krzysztof.Sterenczak@wl.sggw.pl) 
department of Remote Sensing and Land Information Systems, University of Freiburg, Tennenbacher StraBe 4 D- 
79106 Freiburg, Germany. 
Commission VI, WG VI/5 
KEY WORDS: Airborne Scanner System, Segmentation, Aerial photogrammetry, Comparative analysis, Forestry 
ABSTRACT: 
The purpose of this study is to categorize how different surface model resolutions impact the estimation of trees and forest stands 
parameters: number of recognized trees, crown area, treetop location and height of single trees. For forest analysis the following 
models - DSM (Digital Surface Model) and nDSM (normalized Digital Surface Model) were taken into account. For each model 
resolutions 0.25; 0.5 and 1.0 m were used. The studies were cared out in 1000 ha area of Forest Experimental Station in Gluchow, 
owned by Warsaw University of Life Sciences - SGGW, central Poland. Research was based on 34 sample plots measured mainly 
by airborne laser scanning (LIDAR) and by stereo-photogrammetric observation. Forest structure is mixed, with one layer Scotch 
pine (Pinus silverstris L.) and Common oak (Quercus robur L.) stands, as we as with multilayer, rebuild stands. In this paper 
discrimination between coniferous and deciduous was not made. Stereo photogrammetric measurements were used as a reference 
data to comparisons to results from LIDAR data processing. System Falcon II (TopoSys GmbH, Biberach Germany) was used for 
LIDAR data acquisition. Just first echo (FE) cloud point was used in processing. In the presented study the main findings were that 
for raster resolution 0.25 and 0.5 m number of detected trees was the largest, about 80 % of the reference value and was no 
statistician significant difference between these two resolutions. The number of extracted trees based on nDSM was slightly larger 
comparing to DSM. Results show for all 3 raster resolutions tree height estimation were close to the reference data and did not vary 
significantly between models. 
1. INTRODUCTION 
One form of output generated form airborne LIDAR data is that 
of the surface model. Expectation is that each of these models 
will properly describe a presented surface. Based on surface 
models from LIDAR data of forested areas it is possible to 
capture many different tree and forest parameters (Olsson, 2004, 
Hyyppa et al., 2004) with very accurate results. LIDAR data 
density and raster resolution models have a principal influence 
on the accuracy of estimated trees location, crowns 
delimitations (segmentation), trees heights and resulted forest 
stands characteristics. 
As a reference to the LIDAR data, accuracy assessment data 
from field surveying was also used. The most important 
problem connected with using LIDAR data is that measuring 
trees from ground level can cause errors (Coops et al., 2004; 
Maltamo et al., 2004). Another difficulty is the accurate 
measurement of tree position. These factors have fundamental 
influence on latter analysis because they can distort results 
based on LIDAR data study. 
The results presented in this study mainly deal with just the first 
level of the forest canopy, where the stereo photogrammetric 
method was used for reference data acquisition. Research on 
using stereoscopic measurement of Polish forests started in 
early 1970’s (Piekarski, 1972). There exists a range of literature 
suggesting that 3D measurements based on pars images gives 
reasonable and accurate information about the firs forest crown 
layer (Abraham, Adolt, 2006; Adler, 2001; Akça, 1984; 
Bçdkowski, 2005; Duvenhorst, 1998; Feldkôtter et al., 1995; 
Huss et al., 1984; Mauser, 1990). 
There were two main goals of this study. Preliminary 
investigations were made to locate any differences between 
DSM and nDSM in number of delineated trees. Resolutions of 
0.25 m; 0.5 m and 1.0 m were used. Based on the three 
resolutions of nDSM we examined how crown area, tree height 
and tree top position is changing. Does resolution of used 
surface model have any influence on trees parameters or not? 
The selection of the model resolution is a compromise between 
expected accuracy and time of data processing. 
2. MATERIAL 
2.1 Study area 
The study plot used is a 1000 ha area of forest experimental 
station in Gluchow, owned by Warsaw University of Life 
Sciences - SGGW, central Poland. Different forest types exist 
in the forested region, from one layer stands of Scotch pine 
{Pinus silverstris L.) and Common oak {Quercus robur L.) to 
multilayer and mixed stands with Birch (Betula), Alder {Alnus), 
European beech {Fagus sylvatica L.) and Hornbeam {Carpinus 
betulus L.) Age of analyzed stands vary between 30 and 120