USE OF DIGITAL SATELLITE DATA FOR STAND DELINEATION AND ESTIMATION OF 
STAND VARIABLES BY REGRESSION ANALYSIS AND FIELD INVENTORY 
01le Hagner 
Dept, of Biometry and Forest Management, Remote Sensing Laboratory, 
Swedish University of Agricultural Sciences, S-901 83 Umea, Sweden 
ABSTRACT 
The method described is based on integrated use of digital satellite data, National Forest Inventory (NFI) 
sample plots, map information, and subjective field inventory. Forest stands are delineated by using a 
digital region growing technique called "t-ratio segmentation" and SPOT satellite data. The segmentation is 
guided by digitized map information on landuse and administrative regions. Stand characteristics are 
estimated for each stand by combining estimates obtained from both satellite data and field inventory. NFI 
sample plots and corresponding spectral signatures are used to construct regression functions for esti 
mation of stand variables. 
The method was evaluated at five separate test sites in northern Sweden. The t-ratio segmentation method 
produced results similar to visual interpretation of aerial photos and field checking. The accuracy of 
stand data estimation was comparable to subjective field inventory. A substantial improvement in estimate 
precision was obtained when combined estimates were calculated from both satellite data and field inven 
tory. 
Key Words: SPOT, Landsat TM, segmentation, region growing, stand delineation, forest inventory, estimation 
of stand characteristics. 
1 INTRODUCTION 
1.1 Background 
1.1.1 Stand delineation and inventory methods 
used in practice. The basic unit in Swedish forest 
management planning is the forest stand, which is 
a homogeneous region of about 1-20 hectares in 
size. Several adjacent stands with similar charac 
teristics may be grouped into treatment units, or 
compartments. Stands are delineated by means of 
visual interpretation of black and white aerial 
photos, usually at the scale of 1:30 000 or 
1:20 000. The result is checked and corrected 
during field inventory. Stand characteristics are 
normally estimated by means of ocular field 
methods, supported by a few relascope measurements 
at subjectively selected spots within each stand. 
After field inventory, the final delineation is 
transferred to a forestry map with orthophoto 
background and, in some cases, the result is 
digitized and stored in a GIS. Although field 
inventory constitutes the major part of forest 
mapping costs today, very few additional sources 
of information are used in the estimation of stand 
characteristics. 
1.1.2 The thematic classification approach. 
Several attempts to use the "classic" satellite 
remote sensing approach for forest mapping, i.e. 
supervised thematic classification, have been made 
in the Nordic countries. The results, however, 
have usually been of no or limited value to 
foresters for management planning. One of the 
main reasons is that the information needed is 
standwise estimates of several stand character 
istics rather than discrete pixel by pixel class 
information. Most of the important stand variables 
are continuous, e.g., volume/hectare, mean age, 
mean diameter, tree species mixture etc. This 
information can not be expressed by a few discrete 
classes, without large approximations. Some other 
limitations of the classic approach are: (1) The 
procedure of selecting appropriate training samp 
les is laborious and often subjective. (2) Dis 
crete class definitions are difficult to match 
with other sources of information. 
1.1.3 The Finnish approach. Several studies in 
Finland have proposed the use of digital satellite 
data for standwise estimation of continuous stand 
variables. Poso et. al. (1987) used satellite 
data, digitized stand boundaries, and clustering 
techniques in a stratified two-phase sampling 
design. Tomppo (1986) used National Forest Inven 
tory (NFI) sample plots and corresponding spectral 
signatures from Landsat TM to construct regression 
functions. They also used digital satellite data 
for delineation of homogeneous regions. The 
results were very promising and indicated that 
stand characteristics could be estimated almost as 
accurately with satellite data, as with manual 
ground-only methods. They also demonstrated the 
importance of using map information in the analys 
is. The segmentation technique used, called "di 
rected trees" (Narrendra & Goldberg, 1985), was 
considered very promising, although some improve 
ments were needed. 
1.2 Objectives 
Inspired by the Finnish results, a project was 
started in 1986 at the Remote Sensing Laboratory 
in Umea. The aim was to develop a new integrated 
inventory method combining, via statistical tech 
niques the potentials of satellite remote sensing, 
field inventory methods and geographic information 
systems (GIS). The objectives of this study were 
to define the method and to test the main parts; 
the computer delineation of stands and the esti 
mation of stand variables. 
2 MATERIAL AND METHODS 
2.1 Outline of the integrated inventory design 
The basic idea behind the method is to enhance 
existing inventory methods by using computer 
support to derive new information from digital 
satellite data and other sources of information. 
The methods should be implementable on mobile 
workstations and capable of being operated by 
field inventory personnel. 
2.1.1 Stand delineation. Segmentation tech 
niques and high resolution satellite data are 
used (Tomppo 1986), (Hagner 1989) for delineation 
of homogeneous stands of approximately 0.5-5 
hectares in size. The delineation is guided by map 
393