Full text of DOI:10.1080/01431160120213

This article was downloaded by: [Columbia University]
On: 09 December 2014, At: 00:23
Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number:
1072954 Registered office: Mortimer House, 37-41 Mortimer Street,
London W1T 3JH, UK
International Journal of
Remote Sensing
Publication details, including instructions for
authors and subscription information:
http://www.tandfonline.com/loi/tres20
Thinning-caused change
in reflectance of ground
vegetation in boreal forest
T. Nilson , H. Olsson , J. Anniste , T. LÜkk & J.
Praks
Published online: 25 Nov 2010.
To cite this article: T. Nilson , H. Olsson , J. Anniste , T. LÜkk & J. Praks (2001)
Thinning-caused change in reflectance of ground vegetation in boreal forest,
International Journal of Remote Sensing, 22:14, 2763-2776
To link to this article: http://dx.doi.org/10.1080/01431160120213
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all
the information (the “Content”) contained in the publications on our
platform. However, Taylor & Francis, our agents, and our licensors
make no representations or warranties whatsoever as to the accuracy,
completeness, or suitability for any purpose of the Content. Any opinions
and views expressed in this publication are the opinions and views of
the authors, and are not the views of or endorsed by Taylor & Francis.
The accuracy of the Content should not be relied upon and should be
independently verified with primary sources of information. Taylor and
Francis shall not be liable for any losses, actions, claims, proceedings,
demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in
relation to or arising out of the use of the Content.

This article may be used for research, teaching, and private study
purposes. Any substantial or systematic reproduction, redistribution,
reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access
and use can be found at http://www.tandfonline.com/page/terms-and-
conditions

int. j. remote sensing, 2001, vol. 22, no. 14, 2763–2776
Thinning-caused change in re ectance of ground vegetation in boreal
forest
¨
T. NILSON†‡, H. OLSSON§, J. ANNISTE¶, T. LUKK† and
J. PRAKS*
†Tartu Observatory, 61602 Toravere, Estonia
˜
§Remote Sensing Laboratory, Swedish University of Agricultural Sciences,
S-90183 UmeaÊ, Sweden
¶Estonian Forest Survey Centre, Iva 12, 12618 Tallinn, Estonia
*Helsinki University of Technology, Otakaari 5A, 02150 Espoo, Finland
(Received 28 April 1998; in Ž nal form 2 June 2000)
Abstract. Ground re ectances were measured in the blue, green, red and near
infrared (NIR) regions of the spectrum in a set of recently thinned pine- and
spruce-dominated stands near UmeaÊ, Sweden. Compared with the untouched
reference stands, the change in ground re ectance of the thinned stands was
approximately linearly related to the thinning grade and to the coverage of the
cutting waste left on the ground. Typically, thinning resulted in a re ectance
increase in the red and decrease in the NIR band. The major eVects of the presence
of cutting waste on the ground re ectance can be simulated following a rather
simple theoretical analysis. It appeared to be more diYcult to quantitatively
describe the eVects on re ectance caused by the successional changes in the
ground and Ž eld layer vegetation.
1. Introduction
The detection of areas subject to recent cuttings by use of remote sensing
methods is of interest to, among others, the National Forest Inventories (NFIs) in
Scandinavian countries. Since one important task of the NFIs is to estimate changes
in timber volume, prior knowledge of potentially changed areas would help to direct
a portion of the ground sample studies directly to these areas.
Clearfelling of pine-dominated or mixed pine and spruce stands in the boreal
forests of northern Scandinavia causes changes in the values of the stand re ectance
factor from approximately 0.02 to 0.05 in the visible and NIR spectral bands, and
even larger changes in the short wave infrared bands (Olsson 1994). It has long been
known that these changes are easily detectable when multitemporal satellite images
are compared (Colwell and Weber 1981, Wastenson et al. 1981). On the other hand,
normal thinning cuttings in these forests cause changes in re ectance from only
0.0015 to 0.015 and the respective change for pure spruce stands might be even less
(Olsson 1994). Still, on limited test sites in pine-dominated areas, thinning cut areas
have been automatically delineated in satellite images with an accuracy of more than
‡e-mail: nilson@aai.ee
International Journal of Remote Sensing
ISSN 0143-1161 print/ISSN 1366-5901 online © 2001 Taylor & Francis Ltd
http://www.tandf.co.uk/journals
DOI: 10.1080/01431160010007079

2764
T . Nilson et al.
80% (Olsson et al. 1994). It is, however, not known whether these results can be
extrapolated to areas of less pine.
The summary eVect of re ectance change due to a thinning is caused by the
direct removal of trees, resulting in a change in proportions of scene components
(sunlit and shaded canopy and background) as well as by a change in background
re ectance and illumination. By simulation with a forest re ectance model, Nilson
and Olsson (1995) tried to separate the cutting-caused re ectance changes into
contributions from the canopy and background. They assumed that the ground
re ectance was a linear function of the thinning grade, which was obtained by
interpolation between ground re ectance in undisturbed forest and the re ectance
for clear-cut areas. The re ectances were determined by a parameter Ž tting and from
the satellite images respectively. The simulations showed that the change in the
background re ectance essentially contributed to the total stand re ectance change.
Any kind of thinning or clear cutting in a managed forest introduces a disturb-
ance in the forest ecosystem. In addition to the direct removal of overstorey trees,
the disturbance mainly concerns the understorey vegetation. From the point of view
of remote sensing, at least three diVerent co-eVects of thinning on the understorey
should be considered:
(i) mechanical damage caused by the cutting machinery resulting in partial or
full destruction of the ground and Ž eld layer vegetation and, in some cases,
the appearance of bare soil;
(ii) appearance of cutting waste on the ground;
(iii) successional changes in the ground and Ž eld layer vegetation due to changed
micrometeorological conditions (irradiance at the ground level, regimes of
water and nutrition) resulting in changes of abundance and composition of
dominating species.
The aim of the present paper is to study the eVects of thinning on background
re ectance and to derive quantitative relations between the background re ectance
change and thinning grade. Ground re ectance values have also been sought by
those concerned with modelling forest stand re ectance. The measured values of
ground re ectance of the studied stand type as presented below can thus also be
used in modelling stand re ectance.
2. Materials and methods
Measurements were taken in thinning cuttings near UmeaÊ, Sweden (latitude
64 N, longitude 20 E). Cuttings were made between the summers of 1994 and 1995
°
mostly by harvesters. Only cuttings from operational forestry, without any experi-
°
mental treatment, were used.
Two groups of forest stands were chosen for detailed investigation, i.e. stands
dominated by Scots pine (Pinus silvestris) and Norway spruce (Picea abies (L.) Karst).
For the ground re ectance measurements, two or three representative stands were
chosen from the classes of unthinned reference, light thinning, heavy thinning,
extremely thinned seed tree and clear-cut stands. According to the management
practice for Norway spruce stands, no representatives of the heavily thinned and
seed tree stands could be found, since extremely sparse spruce stands are not wind-
resistant. In most of the stands, the ground layer vegetation was of mesic moss type
with a Ž eld layer dominated by blueberry (Vaccinium myrtillus).
In each of the stands, a minimum of six sample plots with a 50 m separation

T hinning-caused ground re ectance change
2765
were established. At each plot forestry parameters were measured within a radius of
10 m. The NFI measurement standard (Anonymous 1995) was followed, but only a
subset of the variables was collected. Basal areas and vegetation types for the stands
are summarized in table 1.
Breast height diameters (DBH, measured in centimetres) of all trees on each plot
were measured and the respective mean basal areas of all stands calculated. For
thinned and clear-cut stands, diameters of all stumps within the plots were also
measured. The stump diameter (D
regression functions computed from approximately 7000 sample trees from the region,
, in centimetres) was related to DBH by
STUMP
available in the NFI database (DBH 0.201 0.8176 D for pine and DBH
=
=-
+
for spruce). Using these functions, the basal area of the stand
STUMP
0.223 0.7955 D
+
before thinning could be estimated. The intensity of thinning was characterized by
STUMP
the thinning grade (THGR) deŽ ned as the basal area of the trees removed divided
by the basal area before thinning. Thus, for extreme cases THGR 0 for unchanged
=
reference stands and THGR 1 (100%) for clear-cut stands.
=
Radiometric measurements were taken at the same sample plots as the forestry
measurements. A series of radiometric measurements of ground re ectance was made
from 12 to 23 July 1995. The measurements were repeated on the same plots the
Table 1. Basal area of the studied stands before and after thinning, thinning grade (THGR)
and a description of the ground and Ž eld layer vegetation. Codes for ground layer
vegetation: 2, mosses of genus Sphagnum, rich in lichens; 3, lichen-rich type; 4, mosses
of genus Sphagnum; 5, swamp moss type; 6, mesic moss type. Codes for Ž eld layer
vegetation: 2, tall herbs with Vaccinium myrtillus; 3, tall herbs with Vaccinium vitis-
idaea; 4, low herbs without dwarf shrubs; 5, low herbs with Vaccinium myrtillus; 9, thin
grasses, 11, low Carex; 13, Vaccinium myrtillus; 14, Vaccinium vitis-idaea; 15, predomi-
nantly Calluna and Empetrum; 16, predominantly Vaccinium uliginosum and L edum
palustre.
2
1
Basal area (m haÕ )
2
1
Basal area (m haÕ )
before thinning
after thinning
THGR Ground
(%)
Field
layer
Stand id
Pine Spruce Total
Pine Spruce Total
layer
Pine-dominated stands
1101
1102
1103
2102
2103
3102
3103
4101
4102
5101
5102
21.1
16.3
15.2
16.8
19.8
15.2
11.0
17.0
16.9
15.6
15.1
0.1
4.1
0.2
0.1
2.7
22.8
20.5
15.4
17.1
24.6
16.3
19.7
18.2
19.6
20.1
21.7
21.1
16.3
15.2
13.0
15.7
4.2
2.8
3.3
3.9
0.1
4.1
0.2
0.1
1.7
22.8
20.5
15.4
13.3
17.6
4.9
3.0
4.0
3.9
0
0
0
22.2
28.5
69.9
84.8
78.0
80.1
6 (2)
6 (3)
6 (3,4) 14 (13)
13
14
6
6
6
6
6
6
6
6
14,13
13
15 (14,13)
13
13 (14)
13 (14)
13 (14,16)
13 (14)
6.7
0.8
2.5
4.4
6.4
0.6
0.1
0
0.2
100
99.1
Spruce-dominated stands
1201
1202
2201
2202
5201
5202
2.1
2.0
12.0
4.8
12.8
1.5
24.4
20.8
17.0
18.3
13.8
26.4
28.5
25.5
37.9
28.2
26.6
28.8
2.1
2.0
7.4
4.8
0.4
24.4
20.8
11.0
16.5
28.5
25.5
22.2
23.9
0.4
0
0
41.4
15.2
98.5
99.0
4 (5,6) 5, 13 (11,2)
6 (5,4) 5 (13,4)
6
6
6
6
13 (9)
13 (5,9)
13
0.3
13

2766
T . Nilson et al.
following year, from 25 July to 6 August 1996. A four-channel Ž eld radiometer
KFM-4M designed at the Tartu Observatory was used. The radiometric head was
mounted with a Cardan joint at one end of a horizontal rod Ž xed on a photographic
tripod. The radiometer had four spectral bands corresponding to blue (maximum
transmission wavelength of the Ž lter, l
max
14 nm), red (l
487 nm, transmission half-width, Dl
=
=
0.5
9 nm) and
=
10 nm ), green (l
556 nm, Dl
650 nm, Dl
=
=
=
max
0.5
11 nm) regions of the spectrum. These bands corre-
=
max
0.5
NIR (l
791 nm, Dl
=
max
0.5
sponded approximately to Landsat Thematic Mapper (TM) bands TM1, TM2, TM3
and TM4 respectively. Unfortunately, many of the measurements in the blue band
appeared be unreliable because of technical problems with the sensor. In all the
spectral bands, the radiometer measured irradiance at a height of approximately
1.5 m from the ground level and radiance from the nadir direction with 13 Ž eld
of view.
°
Radiometric measurements were made at Ž ve points on each 10 m radius sample
plot: at the plot centre and 5 m to the north, east, south and west from the centre.
When choosing the timing of measurements, we avoided rapid changes of illumina-
tion from full sunlight to shade or vice versa caused by movement of clouds. During
a measurement sequence that lasted 50 s, the radiometer was manually scanned by
rotating the radiometric head on the tripod along a circle of approximately 1 m
radius. As a result, at each measurement point, an average of ground radiance over
approximately 2 m and irradiance along a circular line 3 m in length was recorded.
2
Measurement of ground surface re ectance factors under the forest canopy is a
complicated task because of the high variability of irradiance at the forest  oor.
Despite the averaging resulting from the scanning technique used, it is still possible
to obtain biased and unrealistic re ectance factors at some of the measurement
points. A typical problematic case is when the ground surface in the Ž eld-of-view of
the radiometer occurs in the sun  eck, but the radiometric head itself at 1.5 m above
the ground is in the shade or vice versa. To reduce the in uence of these extreme
situations, the radiances and irradiances were averaged over all Ž ve measurement
points on each sample plot before calculating the re ectance factors. Thus, on each
sample plot the measured re ectance factors represent averages over 10 m of
ground area.
2
The ground re ectance factors for each of the sample plots were calculated as
the ratio of the average measured radiance to the average irradiance. The radiometer
was calibrated using measurements of a reference panel in an open area taken before
and after the measurement sequence. The re ectance factors of the panel were
determined earlier in the laboratory.
As a result of thinning, the ground surface was partly covered by cutting waste.
The relative ground coverage of the cutting waste was visually estimated in the Ž eld-
of-view of the radiometer at each measurement location. The visual estimate was
classiŽ ed into one of Ž ve classes (0, 25, 50, 75 or 100%). In addition, the average
cutting waste coverage for each stand was determined on linear transects through
the plot centres.
3. Results and discussion
3.1. Empirical relations between ground re ectance and thinning grade
Figure 1 shows the stand average ground re ectances in the green, red and NIR
bands plotted against the thinning grade of the stand for pine- and spruce-dominated
stands. Unfortunately, in the set of stands chosen we have almost no representatives

T hinning-caused ground re ectance change
2767
Figure 1. Measured and simulated (solid lines) relations between ground re ectance and
thinning grade of the studied pine-dominated and spruce-dominated stands in green,
red and NIR bands. Error bars show the standard deviation of re ectance on a single
s
10 m radius sample plot. Measurements: pine-dominated , 1995, , 1996; spruce-
d
n
dominated , 1995, , 1996.
m

2768
T . Nilson et al.
of the medium thinning grade (50%). Most measurements from 1995 are supplied
with error bars to show the standard deviation of re ectance among the six plots
within the same stand. Note that the standard deviation of the mean ground
re ectance for a stand is by
of horizontal variability of ground re ectance as well as the variability caused by
instrumental and methodological (e.g. limited averaging) eVects.
6
2.45 times less. The errors indicate the magnitude
ã #
In an analysis of Ž gure 1, considerable variability in the measured ground
re ectance can be noted. However, some systematic trends are also evident. For the
pine-dominated stands in the green band, the ground re ectance is relatively insensit-
ive to the thinning grade. However, an increase in re ectance in the red band and a
decrease in the NIR can be noted as the thinning grade increases. For the spruce-
dominated stands, surprisingly, the red band re ectance shows little dependence on
the thinning grade. A decreasing trend, similar to that in the pine stands, appears in
the NIR band. As a Ž rst approximation, all of these relations can be considered as
linear (table 2). Thus, the linear assumption made by Nilson and Olsson (1995)
seems to be realistic. The values of average ground re ectances for the extreme
cases—untouched reference stands and clear-cut stands—are shown in table 3.
For the pine-dominated stands in Ž gure 1, the diVerences in ground re ectance
between the reference and clear-cut stands are generally as expected. Adding cutting
waste and decreasing the green leaf area and chlorophyll content of leaves should
just result in some re ectance decrease in the green and NIR regions and a concurrent
increase in the red band. The only problematic case is the diVerence in the ground
re ectance of clear-cut stands when compared to visually similar seed trees or heavily
thinned stands (red and green bands). For spruce-dominated stands, the measured
=
Table 2. Parameters of linear regression R bTHGR 100 a of ground re ectance factors
(R) on visual estimates of the thinning grade (THGR); s, standard error of the estimate;
/
+
2
R , coeYcient of determination.
2
R (%)
Spectral band
b
a
s
Pine, 1995
Blue
Green
Red
NIR
0.0120
0.0017
0.0284
0.0875
0.0104
0.0641
0.0518
0.2535
0.0124
0.0088
0.0093
0.0289
14.1
0.7
62.3
61.9
-
-
Spruce, 1995
Blue
Green
Red
NIR
0.0129
0.0209
0.0056
0.1041
0.0123
0.0758
0.0756
0.2632
0.0041
0.0060
0.0068
0.0303
73.0
76.6
15.1
76.1
-
-
Pine, 1996
Blue
Green
Red
NIR
0.0198
0.0149
0.0101
0.0849
0.0105
0.0724
0.0576
0.2543
0.0056
0.0051
0.0069
0.0296
69.0
77.2
27.4
59.6
-
-
Spruce, 1996
Blue
Green
Red
NIR
0.0176
0.0183
0.0061
0.1058
0.0112
0.0724
0.0753
0.2560
0.0040
0.0051
0.0081
0.0231
83.3
77.2
13.0
84.8
-
-
-

T hinning-caused ground re ectance change
2769
Table 3. The average (m) and standard deviation (s) of ground re ectances in the reference
and clear-cut stands, and on mechanically undisturbed subplots in heavily thinned
=
stands (an average over all plots having CW 0, without cutting waste, in heavily
thinned, seed tree and clear-cut stands).
1995
Blue Green
1996
Blue Green
Stand type
Red
NIR
Red
NIR
Pine
Reference
m
s
m
s
m
s
0.009 0.065 0.051 0.264
0.005 0.009 0.006 0.020
0.034 0.064 0.085 0.191
0.008 0.002 0.002 0.004
0.018 0.058 0.065 0.243
0.004 0.007 0.012 0.028
0.011 0.071 0.059 0.263
0.003 0.011 0.004 0.034
0.037 0.060 0.078 0.163
0.003 0.002 0.004 0.003
0.019 0.056 0.060 0.203
0.006 0.006 0.007 0.015
Clear-cut
=
CW 0,
Heavy thin
Spruce
Reference
m
s
m
s
m
s
0.016 0.073 0.080 0.244
0.002 0.006 0.006 0.010
0.026 0.055 0.072 0.159
0.000 0.001 0.002 0.003
0.021 0.053 0.065 0.174
0.006 0.013 0.017 0.040
0.013 0.070 0.076 0.248
0.003 0.005 0.010 0.027
0.029 0.053 0.069 0.150
0.003 0.001 0.003 0.006
0.022 0.049 0.059 0.184
0.006 0.004 0.003 0.012
Clear-cut
=
CW 0,
Heavy thin
re ectance behaviour in the red and green bands is diVerent from that observed in
the pine stands. The ground re ectance in the green and red bands is surprisingly
high among the studied reference and light thinning stands, and the respective
average re ectance in clear-cut stands is even smaller than in the reference stands.
This eVect could possibly be due to some diVerences in Ž eld (and ground) layer
vegetation type (site type) between the reference and cut stands (see table 1). If we
could compare the ground vegetation re ectance changes in more similar sites, the
resulting eVects would, hopefully, be more alike those in pine-dominated stands.
Comparing the ground re ectances of the same plots on the Ž rst (1995) and
second (1996) year after cutting, contrary to our expectation, no signiŽ cant changes
can be noted. In heavily thinned, seed tree and clear-cut pine-dominated stands,
some decrease in the red band and a minor decrease in the NIR band is apparent.
Changes could be expected due to changes in the appearance of the pine cutting
waste on the second year after cutting (reddish needles having been dropped from
cut branches) and by a decrease in the Leaf Area Index (LAI) of dominant Ž eld-
layer species (mostly Vaccinium myrtillus). Soil scariŽ cation, carried out on most of
the clear-cuts in summer 1996, has certainly had an eVect on the 1996 re ectances.
Some diVerences of ground vegetation re ectance could also be caused by possible
minor diVerences in the phenological time during the measurement series of these
two years.
In addition to the measured values, simulated curves are presented in Ž gure 1.
Details of the simulation are explained below. Here, it is important to note that the
simulated curves are not linear, while the main non-linearity eVects are present at
lower thinning grades.
The empirical (linear) relations between the ground re ectance and thinning
grade that were derived from our datasets (table 2) could be suYcient for several
practical applications in interpretation and simulation of cutting eVects on stand

2770
T . Nilson et al.
re ectance in boreal forests of the same type. However, in the following sections we
will try to apply some quantitative theoretical analysis of the in uence of the main
factors responsible for the ground re ectance change. First of all we consider the
in uence of cutting waste. If the cutting waste exposed on the ground was the only
cutting-caused reason for the change of ground and Ž eld layer vegetation re ectance,
then it would be better to express the re ectance and its change in terms of cutting
waste coverage rather than in terms of the thinning grade. Note that the thinning
grade is a relative measure of cutting intensity, however, the total amount (coverage)
of thinning waste should be closely related to the amount (area) of branches of
cut trees.
3.2. Relations between cutting waste coverage and thinning grade
More trees cut implies more cutting waste covering the ground after the thinning.
It was possible to derive an empirical relation between thinning grade and cutting
waste coverage. This relation appeared to be non-linear—at higher thinning grades
(more than 50%) cutting waste coverage became saturated (Ž gure 2). According to
our data, it was possible to approximate the relation between the thinning grade
(THGR, in %) and cutting waste coverage (CW, in %) by means of the formula
CW 55.4[1 exp( 0.0360 THGR)],
-
s 5.13, R 95.3%
= 2=
(1)
=
-
where s is the standard deviation of the estimate and R is the coeYcient of determina-
2
tion of the non-linear regression.
Let us now try to derive theoretical formulae to predict cutting waste coverage.
DiVerent management practices result in diVerent eVects of cutting on the ground
vegetation. They determine how much of the ground and Ž eld layer vegetation is
mechanically damaged and how the cutting waste is left on the ground (either
randomly or in piles on/at the cutting roads).
Cutting waste is mainly composed of the branches of cut trees. The simplest case
is if the cutting waste is left in the form of distinct and thick piles of branches, while
the rest of the ground surface remains unin uenced. Then we simply need to know
Figure 2. Relation of the coverage of cutting waste on the ground surface of thinned or cut
stands with the thinning grade: *, measured data on the linear transects the Ž rst year
s
after cutting; , simulated values; ———, empirical relation, equation (1).

T hinning-caused ground re ectance change
2771
the number and dimensions of typical piles. However, in the majority of the thinned
or cut stands we studied, the cutting waste was distributed more or less randomly
on the ground. In this case it is possible to derive a theoretical estimate for the
cutting waste coverage (CW) in the form
CW
1
exp( 2kBAI /p)
-
(2)
= -
cut
where BAI is the branch (woody part) area index of the cut trees deŽ ned as half
of the total surface area of branches per unit ground area, the coeYcient k is
cut
introduced to account for the eVects of the angular distribution of branches laid on
the ground and their clustering (mutual overlapping) eVects and the term 2/p takes
into account the relation between the projected and one-sided branch area.
Supposing that all branches are horizontal, i.e. parallel to the ground and ran-
domly dispersed, then k 1. For instance, if, before cutting, a pine stand had a one-
=
sided BAI equal to 1.0, then the cutting waste coverage for the respective clear-cut
stand would be 0.47. For a clustered dispersion of branches on the ground, k
and the cutting waste coverage is less than in the case of a random dispersion.
1
<
We tried to produce a rough estimate of the BAI of the removed trees by using
the regression of branch area against DBH (cm) and tree height h (m) of trees and
cut
density of the removed trees (N , trees m ) in the following form:
Õ2
cut
1.355N exp[12.1095DBH/(DBH 7)
BAI
=
=
+
cut
cut
0.0413h 1.565 ln(h) 3.4781] for pine
+
-
-
BAI
cut
0.792N exp[9.7809DBH/(DBH 12)
+
cut
0.4873 ln(h) 1.8551] for spruce
(3)
-
-
These equations originate from Marklund’s (1988) regressions of the tree needle
mass on DBH and height. The needle mass was converted to needle area and the
area of branches was assumed to be proportional to needle area with coeYcients of
0.18 for pine and 0.12 for spruce. Equations (1) or (2) together with (3) enable us
to estimate the cutting waste coverage if information about the parameters of cut
trees is available.
We estimated the cutting waste coverage of our study stands by means of
equations (2) and (3). When we compared the simulated estimates of the cutting
waste coverage to the measured values, it appeared that these formulae clearly
underestimated the cutting waste coverage for the pine-dominated stands. This
problem was most probably caused by the pine needles that remained on cut branches
for the Ž rst year after cutting. In visual estimates of the cutting waste coverage, the
needles on cut pine branches made an important contribution. We thus added the
contribution of the needles on cut branches of pine as a factor of 2, i.e. the projected
needle area index was twice the area of cut branches. As a result, a reasonable
agreement between the measured and simulated cutting waste coverage was achieved
(Ž gure 2).
3.3. Empirical relations between ground re ectance and cutting waste coverage
Since visual estimates of cutting waste coverage were made on each subplot, it
was possible to relate the measured ground re ectance factors to the coverage of
cutting waste on the same subplot. The re ectance factors were averaged over all
subplots with no (0%), 25%, 50%, 75% and 100% of cutting waste respectively.

2772
T . Nilson et al.
The results obtained (Ž gure 3) are rather similar to those in Ž gure 1. For pine-
dominated stands, a statistically signiŽ cant re ectance increase in the red band and
a decrease in the NIR band could be noted with an increase in the cutting waste
coverage, while in the green band the re ectance was found to be almost insensitive
to the presence of cutting waste. Hence, typical cutting waste should have higher
re ectance than mechanically undisturbed ground vegetation in the red region and
lower in the NIR. In the green band the re ectances of cutting waste and typical
ground vegetation should be similar.
For the spruce stands studied, a decrease of re ectance in the NIR band was
found. However, in the green and red bands the dependence of the re ectance factor
on the cutting waste coverage was small. Here, again, the problem could be that on
the plots with no cutting waste in the spruce-dominated reference and clear-cut stands
there are some diVerences in the Ž eld and ground layer vegetation type (table 1).
As a Ž rst approximation, the relation between the ground re ectance factors and
cutting waste coverage can be described by linear functions. On the basis of Ž gure 3,
the re ectances of typical cutting waste (100% cutting waste in the Ž eld-of-view of
the radiometer) of the Ž rst and second year can also be compared. With the exception
of spruce in 1996, the cutting waste re ectances for pine and spruce were nearly the
same in the green and red spectral bands. For the NIR band, evidently, the presence
of pine needles in 1-year old (1995) cut branches had an eVect on NIR re ectance,
even though they had turned reddish. The diVerences between pine and spruce
cutting waste in the second year was also considerably diminished in the NIR band.
3.4. Simulated cutting waste coverage eVect on ground re ectance
In order to create a simulation, Ž rst a certain optical model of the medium
(consisting of undisturbed ground and Ž eld layer vegetation as the background and
pine or spruce branches forming the ‘upper’ layer) is needed. It is relatively simple
to simulate the situation when the cutting waste is left in distinct piles. While the
undisturbed surface and a pile of cut branches generally have diVerent spectral
signatures, a linear relation is expected between the ground surface re ectance and
coverage of the cutting waste. The linear approximation would clearly work rather
well at approximating the measured data shown in Ž gure 3.
Another possibility is to simulate the layer of cut branches as more or less
randomly dispersed above the ground and Ž eld layer vegetation. For this kind of
optical medium, plant canopy re ectance models can be applied. We used the Nilson–
Kuusk (1989) canopy re ectance model. The input parameters needed to run the
model were determined in the following way.
d
For background re ectance, the average measured values on plots having no
cutting waste were used.
d
Hemispherical re ectance of a canopy element (branch bark in our case) was
adjusted so that the simulated re ectance factor of an inŽ nitely thick layer of
Figure 3. Measured and simulated dependence of ground re ectance in the green, red
and NIR bands on the coverage of cutting waste on a subplot. For measurements in
1995 on pine-dominated stands, in addition to the average values, 95% conŽ dence
limits of the mean are given. Simulated curves: . . . . ., pine-dominated stands (1995);
———, pine-dominated stands (1996); – – – –, spruce-dominated stands (1995).
s
Measurements: pine-dominated , 1995; , 1996; spruce-dominated , 1995; , 1996.
d
n
m

T hinning-caused ground re ectance change
2773

2774
T . Nilson et al.
branches would coincide with the measured average re ectance on the plots
of 100% cutting waste coverage.
Transmittance of an element (branch) was set to zero.
d
d
The element size needed in the re ectance model and medium total height
ratio was assumed to be 0.3 (the average diameter of cut branches is one third
of the height of the ‘layer’ of branches).
d
d
The two parameters of the elliptical distribution describing the angular dist-
ribution of canopy elements were Ž xed at h
(eccentricity), which means that the cut branches are preferably horizontal.
0 (modal angle) and e 0.9
=
=
m
The one-sided (cutting waste) area index ( BAI ) was calculated from the
=
given coverage of cutting waste (CW) by equation (2).
cut
With the latter method relations between the ground re ectance and cutting
waste coverage were simulated and the respective curves added to the measured
points on Ž gure 3. For the pine-dominated stands, simulations in the NIR band were
made separately for the Ž rst year (1995) and second year (1996, when the branches
were without needles). We see that the simulated relations are non-linear. On average,
the random canopy model seems to agree better with the measured values than the
simple linear model. The random canopy model somewhat underestimates the ground
re ectance at medium and high cutting waste coverages (CW 75%). However, with
=
the present experimental material, it is diYcult to judge which of the theoretical
simulations, the simple linear model or the random canopy model, is preferable.
An alternative scenario to a random distribution of cutting waste is the sometimes
used technique where cutting waste is left in piles on the cutting roads. The model
could then be used on cutting roads, which are well exposed in the nadir view
direction and almost fully covered with the cutting waste, and intermediate thinned
areas in which the ground surface remains almost unin uenced by the cutting waste.
3.5. Simulation of re ectance and thinning grade relation
Another series of calculations was made to simulate the relation between ground
re ectance and thinning grade (Ž gure 1). For this purpose, Ž rst a set of thinning
grades (0, 10, 20, ..., 100%) was chosen and equation (1) was used to calculate the
respective cutting waste coverage. From formula (2) the cut branch area index BAI
was calculated and further, the ground re ectances were calculated by means of the
cut
Nilson–Kuusk (1989) canopy re ectance model. The same set of optical parameters
was used as above when the relation between the cutting waste coverage and ground
re ectance was simulated. However, in addition, it was assumed that the background
re ectance linearly changed with increasing thinning grade. Thus, we also tried to
simulate the eVect of successional change in the ground and Ž eld layer vegetation
on ground re ectance.
From these simulations, the main conclusion that can be drawn is that the most
remarkable re ectance changes should be expected at lower thinning grades. At
higher thinning grades, the saturation of the level of cutting waste coverage causes
a smaller ground re ectance dependence on the thinning grade. Comparing the
simulated curves with the measured points (Ž gure 1), a reasonable agreement can
be noted. The only eVect that cannot be simulated with the present model is the
signiŽ cant diVerence in the ground re ectance between the clear-cut and heavily
thinned/seed tree stands in the red band.

T hinning-caused ground re ectance change
2775
Is it possible to separate the eVects of cutting waste and bare soil appearance on
the ground re ectance from the eVect of successional change? Detailed measurements
of the species composition and LAI change were not made. However, on the Ž rst
summer after cutting the dominant species in the Ž eld layer, blueberry, had turned
brownish-reddish in heavily thinned, seed tree and clear-cut stands. During the next
summer, the abundance of blueberry was considerably reduced on these plots. With
the present available data, it is rather diYcult to quantify the possible eVects of
successional changes in the ground and Ž eld layer vegetation on re ectance. Here,
we need a re ectance model for the ground vegetation, which can be developed on
the basis of existing canopy re ectance models, such as the Nilson-Kuusk (1989)
model or its multispectral analogue (Kuusk 1994). However, the most diYcult
problem would be the introduction of quantitative overstorey–understorey relations,
e.g. expressing ground vegetation green LAI as a function of overstorey canopy
closure, LAI, etc. In addition, the transition processes of the degradation of blue-
berry—the changes in leaf pigmentation (destruction of chlorophyll) and the corres-
ponding changes in leaf re ectance in the visible region—would need to be simulated.
We tried to compare the re ectances of mechanically undisturbed plots from
heavily thinned, seed tree and clear-cut stands to those from the reference stands
(table 3). A re ectance decrease in the NIR band and the concurrent increase in the
red band in pine-dominated stands could well be a result of a decrease of abundance
of Vaccinium myrtillus. However, there were too few mechanically undisturbed plots
among the studied plots for the radiometric measurements in heavily thinned stands
to draw Ž nal conclusions. Also, at present we have not quantiŽ ed the eVect of
mechanical damage of ground and Ž eld layer vegetation by cutting machinery,
neither have we presented reliable re ectance factors of the bare soil.
By comparing the re ectances on plots with no cutting waste, one can conclude
that there are systematic diVerences between the pine-dominated and spruce-
dominated stands in the red and green bands: in the spruce stands the ground
vegetation re ectance is higher than in the pine stands.
4. Conclusions
The change in ground re ectance caused by thinning is important for the elabora-
tion of methods for optical detection and monitoring of all types of cutting in forests.
The following conclusions can be drawn from this particular study.
1. An increase of ground re ectance in the red and a decrease in the NIR regions
of the spectrum were found to be concurrent with recent thinning in Vaccinium-
type pine-dominated boreal forest stands. The higher the thinning grade, the
larger the change in ground re ectance. As a Ž rst approximation, the re ect-
ance change can be considered to be linear with respect to the thinning grade.
No considerable diVerences between ground re ectances during the Ž rst and
second year after cutting were found.
2. The coverage of cutting waste left on the ground surface appeared to be the
most important factor determining the cutting-caused re ectance change.
3. A simple theoretical formula was proposed for calculating the cutting waste
coverage, provided that the stem density and some dimensions of the removed
trees are known and the cutting waste has been more or less randomly
dispersed on the ground surface.

2776
T hinning-caused ground re ectance change
4. Canopy re ectance models can be used to simulate the eVect of cutting waste
on ground re ectance. Thus, together with the theoretical estimates of cutting
waste coverage, the expected cutting-caused change in ground re ectance can
be simulated.
5. Further studies are needed to quantify the successional eVects in the ground
re ectance change as well as more systematic measurements of ground
re ectance in forests of diVerent site types.
Acknowledgments
We thank the Ž eld crew headed by H. BaÊaÊth who supervised the necessary stand
measurements. Many thanks also go to Mr Steve Joyce for revising the English of
the paper. This work was supported by Estonian Science Foundation Grant No.
2626, by the Swedish Council for Forestry and Agricultural Research, and by the
Swedish Remote Sensing for the Environment (RESE) program.
References
Anonymous, 1995, Instruktion for faltarbetet vid Riksskogstaxeringen aÊr 1995. Swedish
¨
¨
University of Agricultural Sciences, Department of Forest Survey, UmeaÊ (in Swedish).
Colwell, J. E., and Weber, F. P., 1981, Forest change detection. Proceedings of the 11th
International Symposium of Remote Sensing of Environment, Ann Arbor, Michigan, USA,
11–15 May 1981 (Ann Arbor, MI: Environmental Research Institute of Michigan),
pp. 839–859.
Kuusk, A., 1994, A multispectral canopy re ectance model. Remote Sensing of Environment,
50, 75–82.
Marklund, L. G., 1988, Biomass functions for pine, spruce and birch in Sweden. Report 45,
Department of Forest Survey, Swedish University of Agricultural Sciences, UmeaÊ.
Nilson, T., and Kuusk, A., 1989, A re ectance model for the plant canopy and its inversion.
Remote Sensing of Environment, 27, 157–167.
Nilson, T., and Olsson, H., 1995, EVect of thinning cutting on boreal forest re ectance: a
comparison of simulations and Landsat TM estimates. International Journal of Remote
Sensing, 16, 2963–2968.
Olsson, H., 1994, Changes in satellite measured re ectances caused by thinning cuttings in
boreal forest. Remote Sensing of Environment, 50, 221–230.
Olsson, H., Nasholm, B., and Hagner, O., 1994, Detection of cuttings using segmentation of
¨
multitemporal SPOT-satellite images. Monitoring of local re ectance changes in boreal
forests using satellite data, edited by H. Olsson, Report 7, Remote Sensing Laboratory,
Swedish University of Agricultural Sciences, UmeaÊ.
Wastenson, L., Arnberg, W., Boresjo, L., and Ihse, M., 1981, Computer analysis of multitem-
¨
poral Landsat TM data for mapping of land-use, forest clearcuts, and mires.
Methododological studies. GeograŽ ska Annaler A, 63, 325–337.