SES
and
ition
sed,
nage
and
s of
ated.
nnel
have
sults
s in
nent
mera
that
' for
ning
near
are
apid
ands
| as
itive
sson
rom
ised
n is
able
ame
rent
e to
real
rent
2. MATERIAL AND METHODS
The test site was located in Hyrynsalmi, referred H, in
Eastern Finland (centre Long. 28°30’ E, Lat. 64°30’N). The
forest in the study area is typical Boreal Forest dominated by
coniferous species (Pinus sylvestris and Picea abies). An
operational forest stand map based on aerial photo
interpretation was combined with multitemporal images, and
these forest stands were applied as observation units. Only
mineral soil stands were used in this work. The untreated
observations totalling 390 stands were used in calibration and
the change analyses was based on all observations totalling
496 stands (Table 1). The mean area of a stand was 6.0 ha.
Table 1. Number of observations in Hyrynsalmi data.
Name Obs. Description
Untreated (Unt) 390 only normal growth exist
Uncommercial thin- 13 clearing or thinning of
ning (Unc. Thinn.) young stand
Commercial thinning 15 about 30% of basal area is
(C.thinn) removed
Preparatory cut (Prep 26 thinning of mature stand
cut)
Regeneration cut 13 only seed trees remain
(Reg.cut)
Clear cut 28 all trees removed
Soil preparation (Soil 11 harrowing for regeneration
prep.)
There were three Landsat TM images available for the study
area. In addition two acquisition from Varjo’s (1996) study
were used (Table 2). The details of the Nurmes data, referred
N, used for comparing image pairs form different locations
are presented by Varjo (1996).
Table 2. Landsat TM images used, H=hyrynsalmi data,
N=nurmes data.
Date Track Row
H 21.06.1990 188 15
H 10.06.1992 188 15
H 31.07.1993 188 15
N 08.06.1988* 187 15
N 23.06.1990* 186 15
*From Varjo’s (1996) study
Robust regression was used to make an earlier image
radiometrically comparable with a later one within one image
pair. Regression coefficients between stand means were
estimated using unmanaged forest stands. For Landsat TM
channels 3,4 and 6 multiple regression was used (Equations
1,2 and 3) and for other channels simple regression was
selected (Equation 4). For the image pair 187,186/15 band to
band regression, such as in equation 4, was used for all the
TM channels (Varjo 1996).
v n n7 i-n n^ i-n
yas UD 2 ft * B1 xoa t Boxers +E (1)
-- n n ^ í-n n^ -n
Yona (n) = Bg t Bi xaa * B5xaa *€ (2)
"t —i —t-
Yos) D. a B xw kB xat, D
- n n -n
Yay) V0) = A" + Bi XchG) + € (4)
B, = parameters, pe {0,1,2}
y (i) = mean intensity of a stand on channel 1 at the
moment t
X ch(i) = mean intensity of a stand on channel ie {1,2,5,7} at
the moment t-n
n = interval between image acquisitions ne {1,2,3}
E — error term.
The inverse of within stand variance on the independent
channel of the earlier image was used as weight in parameter
estimation. The calibration parameters were estimated twice.
After the first estimation the outliers and leverage points were
detected and excluded from calibration and the final
parameters were estimated without those observations (Varjo
1996). An observation was considered to be an outlier if the
difference of the residual from zero was statistically
significant according T-test at 5% risk. Similarly, an
observation was considered to be a leverage point if the
Cook's distance was greater than 1 (Rousseew & Leroy 1986,
Varjo 1996).
After calibration, the differences of stand means (Equation 5)
and standard deviations (Equation 6) were formed for each
channel between the calibrated earlier image and the original
later image.
At-n
i. ai
AM s 7 Yi - Yi (5)
^ t—n
t
ASD = SDy; — SD y; (6)
AM,,, = difference of stand intensity means expressed in dn
on channel i
ASd, = difference of within stand standard deviation
expressed in dn on channel 1
y; = original stand mean intensity on channel i at the
moment t
An
y i = stand mean intensity on channel i at the moment
t-n calibrated to the intensity level of moment t
SDy; = standard deviation of stand intensities on channel i
at the moment t
^ t-n
SD y; = standard deviation of stand intensities on
channel i at the moment t-n calibrated to the
intensity level of moment t.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996