Munro, Duncan
sons the comparison to evaluate the performances of the new ETM+ was planned with a LANDSAT-5 TM together with
a SPOT PAN.
In order to achieve the proposed goals, the comparison process had been split in two parts: (1) Qualitative analysis: the
two sets of original and fused images were compared in terms of subjective interpretation / extraction of information.
Single feature analysis (e.g. single anthropic feature, landscape element, etc.) and geological interpretation were used as
examples to estimate the improvements/benefits to be derived from using LANDSAT-7. (2) Quantitative analysis: the two
sets of original and fused images were compared according to their performance in classifying land cover classes. A land
cover map produced by the Spanish Ministry of Environment was used to define training sample and ground truths for
the classification process. Additionally in the frame of the quantitative analysis, particular attention was paid to the ther-
mal bands of the TM and ETM+ sensors to evaluate the plant activities, mainly focusing on its discharge into the neigh-
bouring reservoir.
1.2 Study Area
The area is covered by the series M7814 sheets 13-25, 14-25, 13-26, 14-26 (1:50,000) of the Spanish Military Carto-
graphic Institute. It covers approximately 2100 Km? of the province of Extremadura, including the small towns of Alma-
raz, Navalmoral de la Mata, Lagartera and Valdeverdeja. The landscape is characterized on the N side of the Tajo river
(cutting the area from E to W) mainly by crops, pasture / grassland and dehesa type cover classes. The S side is mainly
characterised by woods covering mountainous formations (peaks about 800 m a.s.l.). Three reservoirs are present in the
area. The Almaraz nuclear power plant is adjacent to the westernmost reservoir, Embalse de Arrocampo-Almaraz, close
to the town of Almaraz. The two reactor domes are clearly visible on the eastern bank of the reservoir.
2 METHODS
2.1 Data Acquisition and Preparation
The imagery acquisition strategy followed certain priorities determined by the objectives of the study. For the LAND-
SAT-7 data the earliest available scene of 1999 (27 August) was acquired to meet the land cover classification objective.
The aim was to capture as much as possible of the variation in vegetation vigour in order to differentiate between land
cover categories characterised by different phenological curves (note that LANDSAT-7 data only became available in
late summer). Once the LANDSAT-7 scene had been identified, the data search focused on the selection of a 1999
LANDSAT.5 scene acquired under the most similar conditions. In order to meet the objective of comparing image fusion
products, a SPOT PAN scene with a low incidence angle and similar acquisition date was also ordered. Additional
LANDSAT-5 and SPOT PAN scenes from similar dates in previous years were requested to serve as reference data. Due
to the excellent weather conditions characterising the area, cloud cover did not represent an obstacle for the objectives.
The topographic maps were used to georeference the panchromatic band of LANDSAT-7. This was then used as unique
reference to geometrically correct the remaining imagery. After geometric correction, all images were subset using a
frame matching the bounding coordinates of the study area. The Spanish Ministry of Environment provided a digital land
cover file of the study area with a full description of the cover classes (Ministerio de Medio Ambiente, 1998). Four geo-
logical maps (1:50,000) (corresponding to the layout of the topographic maps) produced by the Spanish Technological
Geo-mining Institute were also acquired, scanned and georeferenced.
Various image processing techniques were applied to the data sets considering both the multi-spectral and panchromatic
components. As the study is focused on a comparison of data sets, the same processing was susbsequently applied to
each set of images (LANDSAT-7 and LANDSAT-5 + SPOT PAN). For the thermal analysis only the thermal band of
each LANDSAT scene was considered.
2.2 Image Fusion
Image fusion is achieved by an initial stage of geometric registration between images and a subsequent manipulation, via
a range of algorithms, of their DN values to produce new output images. The technique used in this study manipulates
panchromatic and multi-spectral images with the goal of combining the advantages of smaller pixels within the panchro-
matic data and spectral information from the multi-spectral data. The assumption underlying this combination is that the
level of spatial detail within the panchromatic data is higher than that within the multi-spectral data, whilst the discrimi-
nation between objects based on their reflectance characteristics is preserved in the output data.
932 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000.
geor
tion
Con:
rese:
of tk
com
soft
the
valu
emp
proc
23.
ont
ima
para
the :
anal
Con
disp
grec
ima
cov
23.
LA]
upo
that
PA]
that
Att
pris
ciat