information and digital elevation models (DEM). 
Alternatively digitized aerial photos can be used 
instead of satellite data. The result of segmen 
tation is checked and edited interactively on a 
workstation monitor. When the operator is satis 
fied with the result, a hardcopy is printed for 
field use. 
2.1.2 Stand data estimation. Estimates of 
stand characteristics are obtained by applying 
regression functions to the spectral signatures 
within each delineated stand (Tomppo 1986),(Hagner 
1989). The regression functions are derived from 
NFI-plot data and corresponding multispectral 
satellite data. Techniques developed by Holmgren 
(1990) are used to model site type parameters from 
soil map information and digital elevation models. 
2.1.3 Field inventory. A simplified version of 
the subjective field inventory method used in 
Sweden today is used to collect stand data. The 
surveyor describes all variables frcm one spot 
within each stand. Only a few supporting measure 
ments are made. The surveyor concentrates on 
variables that can not be described by remote 
sensing, such as: Treatment recommendations, 
environmental considerations, etc. Hand-held field 
computers are used for data entering and checking 
of consistency and completeness. During field 
inventory errors and corrections of the delinea 
tion are noted. A few (10-15) stands are selected 
for an objective reference survey, in order to 
determine correction factors for personal bias of 
the ground crew. 
2.1.4 Combined estimation. The results of field 
inventory are used together with the estimates 
obtained from NFI- and satellite data, to calcu 
late new estimates for each stand. If random 
errors are uncorrelated and if field and satellite 
estimates are weighted according to the inverse of 
their variances, then combined estimates will have 
higher precision than any of the separate input 
sources. Also other sources of information, such 
as aerial photo interpretation, previous field 
inventory data, etc., can be included in the 
combined estimates. 
2.1.5 Map production. Once the combined esti 
mates of stand data have been calculated, a second 
automatic merging pass is started. Adjacent stands 
with minor differences are joined into larger 
units, according to criteria for desired size of 
final stands and acceptable within-stand varia 
tion. Also the result of this second merging pass 
is checked and edited by the operator. Finally the 
result is transferred to a GIS for storage, analy 
sis, and presentation. 
2.2 Test sites 
Five test sites were used. They are located near 
the city of Umea in Northern Sweden (63°.9 N, 
20°. 1 E) (Figure 1). The sites are representative 
of the forest types found in most parts of North 
ern Sweden, dominated by natural stands of Scots 
pine and Spruce (Pinus sylvestris, Picea abies) in 
various mixtures with broadleafs, mainly birch 
(Betula pubescens and B. verrucosa). The sites are 
owned by private owners (site 1) and forest com 
panies (sites 2-5). The relative proportions of 
age and cutting classes are fairly normal for 
Northern Sweden, with approximately 50% being old 
mature stands. 
2.2.1 Test site no. 1. The main objective for 
the used site no. 1 was to evaluate the perfor 
mance of various methods for stand delineation. 
The specific location was selected because of the 
diversity of forest types, age, and cutting 
classes represented in the area. Also a wide range 
of site types are found, with several types of 
moraine and sediment soils. The topography is 
rather flat, with an altitude ranging from 45 up 
to 120 meters. 
A dense grid of 1468 sample plots (2 plots/- 
hectare) were surveyed. The variables measured on 
each plot were: wood volume/hectare, mean tree 
diameter, tree species mixture, and various site 
type variables, such as: soil type, moisture, 
vegetation, type etc. A manual stand delineation 
of the test site was made by means of visual 
interpretation of black and white aerial photos, 
at the scale of 1:30 000. The delineation was 
checked and corrected in situ. 
2.2.2 Test sites no. 2-5. These test sites were 
used to evaluate the precision of stand character 
istics estimated frcm satellite data and NFI- 
sample plots. A total of 80 reference stands, 
were located in 4 separate sites (Figure 1). The 
characteristics of each stand were determined frcm 
20 circular (radius: 10 meters) sample plots, 
distributed according to a systematic square grid. 
Also, subjective estimates of stand variables were 
obtained for each stand by professional surveyors, 
using the traditional inventory methods, see 
section 1.1.1. 
TEST SITE 
CLUSTER OF 
8 NFI - 
PLOTS 
Figure 1. location of test sites and National 
Forest Inventory plots used in the study. Each 
dot represents a cluster of eight NFI sample 
plots. The area shewn corresponds to the Landsat 5 
quarter scene acquired 21 06 89. 
2.3 National Forest Inventory sample plots 
The Swedish NFI is a continuous inventory. Approx 
imately 18 500 plots are surveyed each year. 
Of these, 40% are permanent and revisited every 5 
years. More than 200 stand and site variables are 
recorded for each plot (Ranneby et al. 1987 ). 
A total of 740 permanent NFI-sample plots, sur 
veyed during the period of 1983-1987, and corre 
sponding spectral signatures frcm a satellite 
acquisition (Landsat 5 TM, 1989) were used to 
construct regression functions for estimation of 
stand characteristics frcm spectral signatures. 
The location of NFI-plots is shown in Figure 1. 
394