850 
ion 
Spectral response from thinning cuttings 
measured by multitemporal satellitedata. 
HSkan Olsson 
Dept, of Biometry and Forest Management, Remote Sensing Lab., 
Swedish University of Agricultural Sciences,S-901 83 UmeS. 
Phone +46 90 166596, telefax +46 90 165925. 
ABSTRACT 
Spectral mean values for thinning cuttings and seed tree stands were measured 
before and after cutting, using a time series of six digital Landsat TM 
acquisitions. The studied area, 30 km by 40 km, was situated in boreal forest 
at 65° N,20° E. The type of forest and degree of cutting were surveyed in 
field using objective sampling methods. Dominant species were Scots Pine 
(Pinus Silvestris), Norway Spruce (Picea Abies) and Birch (Betula Spp.). 
Linear regression functions for prediction of the DN values at the 
following year were computed for each of the first five Landsat acquisitions. 
This was done by using a systematic sample of pixels taken under a forest 
mask. Differences between the actually measured stand mean values and the 
predicted were obtained for the thinned stands. Thinning caused a minor 
increase in spectral radiance in the visual channels TM 1, TM 2 and TM 3, and 
a more significant increase in the middle infrared channels TM 5 and TM 7. A 
decrease in radiance was observed for the near infrared TM 4. Between 19 % 
and 51 % of the basal area was cut in the thinnings. The variations in 
spectral response among the thinnings could hardly be explained as a function 
of thinning grade. The most important factor found was the remaining 
proportion pine forest after cutting. The change in proportion of deciduous 
forest had a strong influence on TM 4. By multispectral discriminant analysis 
81 % of the thinnings was detected. The spectral radiance starts to increase 
when more than 40 % of the basal area is cut. This is shown by measurements 
from five seed tree stands. The study is of basic interest for the 
development of standwise change detection in a GIS environment. The results 
are also of interest for the evaluation of Remote Sensing for the Swedish 
National Forest Inventory. 
KEYWORDS: Landsat TM, change-detection, thinning-cuttings. 
1. INTRODUCTION 
The life cycle of a forest stand in Sweden is 80- 
100 years. One to three thinning cuttings are 
normally carried out before the final cutting. The 
aim of the thinnings is to give space for the 
most valuable stems to grow to timber, the 
thinnings will also give pulp wood for sale. 
On the National level the forest resources in 
Sweden are monitored by the National Forest 
Inventory (NFI). The NFI covers yearly the whole 
of Sweden in a ground based sample plot design 
(Ranneby et. al. 1987). At present the possi 
bilities to incorporate satellite data analysis in 
the sampling design of the NFI are investigated 
(Thomas and Ranneby 1990). One of the questions 
raised is then how reliably thinning cuttings can 
be detected by satellite data analysis. 
The present study is part of a larger project at 
the Remote Sensing Laboratory in Umea. The overall 
aim of this project is to develop general methods 
for standwise change detection in a GIS 
environment. The most important features to 
detect with this technique are probably forest 
damages (Olsson 1989, 1990). The spectral changes 
to be detected are equally small in the cases of 
both thinnings and damages. The development of 
procedures for calibration and discrimination 
might therefore be carried out on thinnings, for 
future applications on damages. 
Earlier work on detection of thinning cuttings in 
Scandinavian forests has been carried out in 
Finland (Häme 1986). Häme found a mean increase in 
TM 3 and a decrease in TM 4. He considered 
thinnings as difficult to detect. 
In this paper the first results from a Swedish 
study are presented. It is investigated how 
spectrally separable thinning cuttings are from 
unchanged forest, using a multitemporal analysis. 
The spectral response in different channels as a 
function of thinning grade, type of forest and 
changes in tree species combinations are studied. 
It is also of interest to find linear combinations 
of the satellite channels that are effective in 
discriminating newly thinned stands from the 
unthinned stands. 
The approach to change-detection used in this 
study is calibration with regression functions. It 
has the advantages that no absolute calibration of 
the data is necessary, the regressors used can 
also be extended. The approach has successfully 
been tested by the author (Olsson 1990), as well 
as by others (Sing 1989) . 
Presented at the ISPRS commission VII Mid-Term Symposiom, 
Victoria, Canada. September 17-21 1990.