1 2004 
———— 
March 
on of 
ultural 
toring, 
1471- 
metric 
among 
nment, 
hange- 
learing 
metric 
er, pp. 
luation 
rolina. 
vol.59, 
textual 
remote 
vol.15, 
t MSS 
torical 
metric 
108. 
tection 
metric 
RU 
ymated 
niques. 
vol.57, 
using 
ensing, 
1990. 
lement 
zement 
ntre la 
system 
(PRS), 
erdam, 
  
HIGHFOREST - FOREST PARAMETER ESTIMATION FROM 
HIGH RESOLUTION REMOTE SENSING DATA 
Heikki Astola, Catherine Bounsaythip, Jussi Ahola, Tuomas Häme, Eija Parmes, Laura Sirro, Brita Veikkanen 
  
Technical Research Centre of Finland, Information Technology, Information Systems, 
P.O. Box 1201, FIN-02044 VTT, Finland, http://www.vtt.fi/tte/ - 
heikki.astola@vtt.fi. (firsthname.lastname)@vtt.fi 
Commission VII, WG VII/3 
KEY WORDS: remote sensing, forestry, automation, estimation, inventory, high resolution 
ABSTRACT: 
The aim of the study was to develop a tool for the estimation of forest variables using high-resolution satellite data. The tool 
included modular operative software. The image analysis methodology focused on the reduction of the known problems of the 
previous satellite image based methods, i.e. the saturation of the estimates at higher biomass levels and uncertainty in tree species 
estimation. Modern contextual image analysis methods were combined with the spectral information of the imagery. In the test 
application the tool used images from the Ikonos satellite with a ground resolution of one and four meters. 
The developed Forestime software estimated the forest variables by segmenting the imagery to ‘micro-stands’, by computing stand- 
wise image feature vectors for the stands from the input satellite image, and by combining ground reference data with clusters from 
an unsupervised clustering stage. The estimates are produced as weighted sums of the input sample class probabilities. The target 
variables in the study were stem volume, average stem diameter, stem number and tree species proportions. The RMSE% for total 
stem volume was 37.4 % (% of mean), for average stem diameter 23.4 %, for stem number 87 %, for pine percentage 111 %, for 
spruce percentage 47 %, and for broad-leaved tree percentage 137 %. 
1. INTRODUCTION The suitable application areas are all operative forest 
inventories as well as wood procurement mapping. Strategic 
Forest owners and wood procurement organisations have a planning could also utilize the information whenever 
continuous need to acquire updated information on forests. The information on species is required. Significant cost savings can 
traditional stand-wise ground inventory is, however, expensive be achieved if estimates applicable in operative forestry can be 
and its frequency is too slow. There is also a need to reduce computed using imagery with a resolution of one to few meters 
stand sizes in order to delineate stands more flexibly when instead of using airborne data with a resolution of tens of 
cuttings and silvicultural measures are planned, and in order to centimeters. By making the method for forest mapping of large 
simulate future development of the forest more reliably. ^ forest areas fully automatic, fresh information for wood 
Satellite imagery with a ground resolution of twenty to thirty procurement purposes will be rapidly available. 
meters has been tested and is partly used even operatively for 
forest inventories. Although the estimates for stand 
characteristics are reasonable good using such data there are 2. DATA AND PREPROCESSING 
two major limitations that have prevented a wider operative use. 
The first limitation is saturation of the reflectance at higher 2.1 Satellite data 
levels of the growing stock volume. The saturation limit in 
Finnish conditions is at approximately 200 m’/ha. Another The acquisition of one Ikonos image was initiated mid June 
limitation is the difficulty to estimate tree species proportions 2003. A geo-referenced and orthorectified Ikonos image of size 
since one pixel of a dimension of e.g. 30 meter includes often 10 km x 13 km was acquired from the specified coordinates in 
several tree species. Suonenjoki, Finland, dated 5.9.2003 (see Figure I). Due to 
excess cloud cover in the eastern side of the image, it was 
The spatial resolution of remotely sensed optical imagery has extended three km towards west without additional charge. 
increased from 30 m (Landsat Thematic mapper) to 1 m Fortunately the cloud cover did not extend over the study area 
panchromatic and 4 m multispectral (Ikonos, Quickbird) in the located in the middle of the ordered image. The Ikonos images 
. . ave fer «nectr £ E > "^e o 
last 20 years. At VTT's remote sensing group various methods — have four spectral channels: blue (450 - 520 nm), green (520 - 
> > . . . . : Z3 € n 
for forestry applications using optical and radar satellite 600 nm), red (630 - 690 nm) and NIR (760 - 900 nm), with a 
imagery have been developed within the same time span, and ground resolution of four meters, and one panchromatic channel 
are in operative use as well. One goal of the HighForest study (450 - 900 nm), with ground resolution of one meter. A smaller 
Was to combine some of the existing methods into a new image covering the ground data area was extracted from the 
operative software tool utilizing the new high resolution — Whole Suonenjoki image for testing (see Figure 2). 
satellite data. 
335