Peer-Reviewed Articles
1265 Unsupervised Spectral Characterization of Shallow Lagoon Waters by the Use of Landsat TM and EMT+ Data
Fabio Maselli, Luca Massi, Chiara Melillo, and Mario Immamorati
Abstract Download Full Article
Assessing the conditions of coastal and shallow lagoon
waters is a top priority among environmental monitoring
activities, due the high ecological and economic importance
of relevant resources. Satellite remote sensing offers great
potential for this scope, although the interpretation of the
spectral responses of shallow areas is complicated by the
mixed signal coming from the water column and the bottom.
This is particularly the case when using satellite data taken
in few wide spectral channels, such as those of the Landsat
TM and ETM+ sensors, which were not specifically designed
for marine applications. Relying on the hypothesis that these
data are anyway informative on shallow water conditions,
an unsupervised procedure was developed to separate the
spectral contributions of seawater and bottom on the basis
of simple approximations. The procedure is based on the
simulation of different water/bottom multispectral configurations up to find out that which best fits the observed data. The validation of the procedure was carried out by its
application first to synthetic images and next to two TM and
one ETM+ scenes taken over the Orbetello Lagoon in Central
Italy. The outputs produced in the latter case were evaluated
by comparison to existing ground references. In particular,
correlation analyses were performed between the original
and decomposed spectral signatures and the concentrations
of optically active water constituents (pigments and yellow
substance) measured “in situ” at dates close to those of the
satellite data acquisitions. These analyses demonstrated the
potential of the methodology, while also highlighting some
limitations which could be overcome through the use of
imagery taken by sensors with enhanced spectral and
radiometric features.
1275 Urban Classification Using Full Spectral Information of Landsat EMT+ Imagery in Marion County, Indiana
Dengsheng Lu and Qihao Weng
Abstract Download Full Article
This paper compares different image processing routines to
identify suitable remote sensing variables for urban classification in the Marion County, Indiana, USA, using a Landsat
7 Enhanced Thematic Mapper Plus (ETM+) image. The
ETM+ multispectral, panchromatic, and thermal images
are used. Incorporation of spectral signature, texture, and
surface temperature is examined, as well as data fusion
techniques for combining a higher spatial resolution image
with lower spatial resolution multispectral images. Results
indicate that incorporation of texture from lower spatial
resolution images or of a temperature image cannot improve
classification accuracies. However, incorporation of textures
derived from a higher spatial resolution panchromatic image
improves the classification accuracy. In particular, use of
data fusion result and texture image yields the best classification accuracy with an overall accuracy of 78 percent and
a kappa index of 0.73 for eleven land use and land cover
classes.
1285 Assessment of Very High Spatial Resolution Satellite Image Segmentations
A.P. Carleer, O. Debeir, and E. Wolff
Abstract Download Full Article
Since 1999, very high spatial resolution satellite data
represent the surface of the Earth with more detail. However, information extraction by per pixel multispectral
classification techniques proves to be very complex owing
to the internal variability increase in land-cover units
and to the weakness of spectral resolution. Image segmentation before classification was proposed as an alternative
approach, but a large variety of segmentation algorithms
were developed during the last 20 years, and a comparison
of their implementation on very high spatial resolution
images is necessary. In this study, four algorithms from the
two main groups of segmentation algorithms (boundary-based and region-based) were evaluated and compared. In
order to compare the algorithms, an evaluation of each
algorithm was carried out with empirical discrepancy
evaluation methods. This evaluation is carried out with a
visual segmentation of Ikonos panchromatic images. The
results show that the choice of parameters is very important
and has a great influence on the segmentation results. The
selected boundary-based algorithms are sensitive to the
noise or texture. Better results are obtained with region-based algorithms, but a problem with the transition zones
between the contrasted objects can be present.
1295 Geometric Accuracy Evaluation of the DEM Generated by the Russian TK-350 Stereo Scenes Using the SRTM X- and C- band Interferometric DEMs
Gurcan Buyuksalih, Guven Kocak, and Murat Orue
Abstract Download Full Article
TK-350 stereo-scenes covering 200 km × 300 km on the
ground with a base-to-height-ratio of 0.52 have been analysed
on Zonguldak testfield in the northwest of Turkey. The pixel
size on the ground is 10 m. Control points digitised from
1:25 000 scale topographic maps have been used in the test.
The sensor orientation was executed by the PCI Geomatica® V8.2 software package. TK-350 stereo-images can yield 3D
geopositioning to an accuracy of about 10 m horizontally
and 17 m vertically. Based on this orientation, DEM with
40 m cell size was generated by the related module of PCI
system. For the validation of extracted DEM, matched data
was checked against the interferometric DEMs from SRTM
X- and C-band SAR data. Based on this comparison, the
RMSE of Z values was found to be in the range of 25.6 to
36.9 m and 28.7 to 38.7 m outside and inside the forest area,
respectively. However, accuracy results obtained against
the SRTM C-band DEM are more representative than those of
X-band since the coverage of C-band DEM on the interest area
is larger than the X-band. There are some systematic shifts of
the TK-350 DEM against the SRTM DEMs which lie between the
3.7 m to 6.2 m which is probably due to the different sensor
orientation of TK-350 and SRTM datasets. Height discrepancies are also analysed as a function of terrain slope. It was
found that slope depending components were always larger
in the case of C-band DEM because of its larger cell spacing.
In the forest areas, more dependency upon the slope was
observed against the open areas.
1303 Agreement Assessment of Spatially Explicit Regression-derived Forest Cover and Traditional Forest Industry Stand Type Maps
Jacob W. Metzler and Steven A. Sader
Abstract Download Full Article
Forest regeneration assessment is an important forest
management goal that requires accurate data about site-specific forest type and stand density. In this study, a
methodology was developed to convert regression model
output to maps of predicted softwood and hardwood percent cover at the scale of a Landsat ETM+ pixel. These maps
provide forest type and percent cover at higher spatial
scale (0.09 ha) than traditional GIS forest stand databases
employ (e.g., 2 to 4 ha minimum mapping units). A modified accuracy assessment was performed between the
Landsat regression derived maps and GIS type maps to
evaluate their relative agreement. Two variations of the
traditional error matrix were examined. The first was a“plus-one” matrix, where values next to the diagonal were
included in the agreement calculations. The second variation, considered most appropriate for this study, included
the use of “fuzzy logic” where the off-diagonal values were
weighted for a better approximation of the GIS forest mapping criteria and forest type composition of the northern
New England forest. The fuzzy logic error matrix indicated
strong agreement between the regression derived and GIS
forest type maps with an overall agreement ranging from
76 percent to 79 percent. Producer’s agreement from the
fuzzy-logic error matrices ranged from 89 percent to 97
percent for softwood classes and 72 percent to 77 percent
for hardwood. User’s agreement for softwood ranged from
71 percent to 82 percent and 80 percent to 87 percent for
hardwood. These results suggest that the Landsat-derived
maps can provide objective and reliable site-specific forest
type and percent cover information that is not dependent on
subjective photo interpretation methods. These maps will be
evaluated in future studies to demonstrate practical forest
regeneration management applications.
1311 Classifying and Mapping Wildfire Severity: A Comparison of Methods
C. Kenneth Brewer, J. Chris Winne, Roland L. Redmond, David W. Opitz, and Mark V. Mangrich
Abstract Download Full Article
This study evaluates six different approaches to classifying
and mapping fire severity using multi-temporal Landsat
Thematic Mapper data. The six approaches tested include:
two based on temporal image differencing and ratioing
between pre-fire and post-fire images, two based on principal component analysis of pre- and post-fire imagery, and
two based on artificial neural networks, one using just post-fire imagery and the other both pre- and post-fire imagery.
Our results demonstrated the potential value for any of
these methods to provide quantitative fire severity maps, but
one of the image differencing methods (ND4/7) provided a
flexible, robust, and analytically simple approach that could
be applied anywhere in the Continental U.S.
Based on the results of this test, the ND4/7 was implemented operationally to classify and map fire severity over
1.2 million hectares burned in the Northern Rocky Mountains and Northern Great Plains during the 2000 fire season,
as well as the 2001 fire season (Gmelin and Brewer, 2002).
Approximately the same procedure was adopted in 2001 by
the USDA Forest Service, Remote Sensing Applications Center
to produce Burned Area Reflectance Classifications for
national-level support of Burned Area Emergency Rehabilitation activities (Orlemann, 2002).
1321 Adaptive Patch Projection for the Generation of Orthophotos from Satellite Images
Liang-Chien Chen, Tee-Ann Teo, and Jiann-Yeou Rau
Abstract Download Full Article
In this paper, we describe an “Adaptive Patch Projection” scheme that can accelerate the orthorectification for satellite
images without losing accuracy. The proposed scheme is
comprised of two major components: (a) orbit modeling, and
(b) image orthorectification. In orbit modeling, we provide a
collocation procedure to determine the precise orbits. In
image orthorectification, the area of interest is sequentially
subdivided into four quadrate tiles until a specified threshold for terrain variations is met. The threshold of maximum
terrain variation in a tile will be optimized according to the
computational efficiency and the accuracy requirements.
Once the ground tiles are determined, we perform adaptive
patch projection to the corresponding image pixels. Test
images from SPOT5 Supermode and QuickBird satellites are
included. The experimental results show that this algorithm
can minimize the orthorectification computation time, while
the modeling error is insignificant.