he
en
ain
1e-
the
er-
ifi-
ich
or-
ale
OS-
rall
le-
ult
ich
hic
re-
OT
yuld
nro-
e at
\T-7
zure
Munro, Duncan
43
Figure 1. Example of LANDSAT-7 Band 8 (left) and SPOT PAN after filtering and stretching (scale approx. 1:55,000)
32. Quantitative Analysis
3.21. Land Cover Classification
The results of the correlation analysis led to the selection of Channels 3, 4 and 5 for both original and merged products
for the classification processing. The selection of these bands was based upon the proportion of the overall variance of
the data set they represent which is a consequence of their sensitivity to the reflectance spectrum minima and maxima for
vegetation. Thus the correlation matrices indicate that these three bands show the least degree of correlation and thereby
encompass the majority of the variance within the data sets.
Representative training samples were selected and a numerical description of the spectral attributes of each land cover
type identified from the land cover category map of the Spanish Ministry of the Environment were developed. These
samples are then used by the maximum likelihood classification algorithm during the comparison of spectral signatures
of all pixels as a tool to derive the probability of category membership and assignment of each pixel to a unique class.
The overall results are presented - from best to worst - in the following table:
LANDSAT-7 LANDSAT-7 LANDSAT-5 LANDSAT-5 LANDSAT-S (97) LANDSAT-S (99)
(+ band 8) (97) (99) SPOT pan (97) SPOT pan (99)
Overall
accuracy 84.46% 82.12% 79.52% 74.98% 74.27% 74.09%
Overall
K stat. 0.8221 0.7957 0.7659 0.7147 0.708 0.7057
It is essential to be aware that these results may incorporate a given degree of error due to the fact that it is impossible a
priori to establish and prove that the training samples were definitely designed to equally/statistically identify - with the
same degree of likelihood - the same categories in the different data sets. Assuming however that an error had been
introduced in a homogeneous way the results indicated that the classifications obtained by processing the LANDSAT-7
(both pure and merged) reached the best overall accuracy especially when comparing the scores of the merged products.
3.2.2 Thermal Analysis
The pre-established goal of comparing the performances of the TM and ETM+ in mapping absolute ground temperature
could not be achieved at the time of writing this document, due to the difficulty in obtaining meteorological data. How-
ever the presence in the reservoir of a warmer discharge at the level of the plant is evident. The temperature gradients
indicate that the industrial discharge follows a counter clockwise path. The circulation is guided by a series of barriers
within the reservoir that are identifiable in the SPOT imagery.
Analysis of the variability of temperature within the reservoir adjacent to the power plant and the variability of the
Embalse de Valdecafias to the SE shows that the mean temperatures of the reservoir adjacent to the power station vary to
a much greater degree than those observed in the Embalse de Valdecañas (see Figure 2). This conclusion drawn from the
relative temperature measurements indicates clearly that changes in industrial activity can be monitored readily with
LANDSAT TM and ETM+ data. However a much greater challenge not met by the present study is to link quantitatively
the changes in water surface temperature with precise measures of power generation.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 937