he
om-
2 MSS
le-
d.
en was
Kuhmo
vari-
ver-
r, but
ns. The
multi-
for
"tral
ichr.
- 1781 -
The aircraft scanner data were acquired by operating a leased ll-
channel scanner on board the twin-engine airplane .of the Finnish
National Board of Survey. Taken at two altitudes and in very
clear weather conditions the data provide good basis for studying
the effect of spectral and spatial resolution on vegetation
classification. The parameters of this data are summarized in
Appendix 1.
3. DATA PROCESSING AND ANALYSIS METHODS
3.1 Numerical interpretation and coordinate transformation of
digital multispectral data
A supervised pattern recognition program installed in a general
purpose computer is being used for the numerical interpretation
of digital multispectral data. The color enhancement and certain
interactive image processing phases are done in a dedicated mini-
computer and color display system. The use of this system is
expanding along with the software development. The color display
of the classification results is done by a color jet plotter and,
in the near future, by a film drum scanner.
In order to present the resulting thematic maps in the Finnish
topographic map coordinate system, an affine coordinate transfor-
mation was developed for Landsat-2-data. The underlying poly-
nomial transformation assumes the form
3
+ B,u + C:,Vv + D u? + E,uv + F y? + Hu +
x = A, i 1 1 1 1 1 al
2 2 3
y A, + B,v + Cou ih Dv + Ejuv * Fou + Hv "ul.
The coefficients are determined on the basis of control points
known in both coordinate systems involved. A least square fitting
is calculated for the control points,and the coefficients resulted
are used for coordinate transformation. An example of the Land-
sat data transformation is shown in Fig. 1. It is a special case
for presenting classification results as line printer output at
a scale of 1:10 000. The interpretation of classification values
among transferred pixels was done by nearest neighbor rule.
3.2 Timber volume estimation
Timber volume functions are being derived by regression analysis
where the responses, or their transformations, of the Landsat MSS
channels are used as independet variables. The functions give
the estimates of timber volume in each pixel. In the procedure,
water bodies are eliminated by using channel 7, and open land
areas by using channel 5. The remaining timber lands comprise
mainly pine stands which eliminates most of the variance caused
by different tree species, thus simplifying the problem. Ina
sample survey, channel 4 was highly correlated with the volume.
Quite a few functions derived explain the bulk of the variance
of volume. A common problem is the overestimation of the volume
in seedling and young thinning stands. The best function to handle