hone.
247-
1resti,
on of
ıltural
tional
AVD.
li, S.,
zlades
alized
vi par
"nder
(réd.).
ue et
té du
indsat
Urban
C.F.,
à and
in the
nsing,
rsum,
Iceous
1984.
d'une
andsat
J.-M.
ection
hique.
uébec,
T.R.,
| Lake
metric
A COMPARATIVE ANALYSIS OF VERY HIGH RESOLUTION MULTI SPECTRAL SENSOR SYSTEMS WITH
MULTI STAGE SENSOR SYSTEMS DATA IN FEATURE EXTRACTION FOR MOUNTAINOUS TERRAIN.
H. Hugh L. Bloemer, Professor, Ohio University, Athens, Ohio, USA
James O. Brumfield, Professor, Marshall University, Huntington, West Virginia, USA
Janette C. Gervin, Instrument Manager, NASA/Goddard Space Flight Center, Greenbelt, Maryland, USA
Joseph A. Langdon, Information Specialist, NASA/HQ, Washington, DC, USA
Charles Yuill, Professor, West Virginia University, Morgantown, West Virginia, USA
Commission VII, Working Group 10
KEY WORDS: Forestry, Simulation, Multi-spectral, Spatial, Spectral, Infrared Photography, Mountainous, Terrain
ABSTRACT
Mountainous terrains have been largely ignored because they present numerous difficulties from a remote sensing
perspective in that process and change often occur on a much smaller spatial scale. Higher spatial frequency
variabilities require higher resolution spatial analysis over similar spectral bands to extract comparable features.
Multistage sampling, involving field studies, and aerial sensor measurements of the Spruce Knob area forest of the
Appalachian Mountains illustrates a comparative level of information that may be extracted at different spatial resolutions.
This research investigates forest species and forest associations discrimination by optimization of spatial resolution or
instantaneous field of view(IFOV 0.1 m - 6.0 m) for selected spectral bands and selected sensor systems. Feature
extraction in pattern recognition is affected by natural and artificial spatial frequencies. These features include forest
species identification in vegetation associations, hydrologic expression, folded strata, joints, fractures and low order
drainage in mountainous terrain. This differs from the artificial periodicities of the electro-optical imaging systems. These
vegetation association features are evaluated by field techniques, analysis of variance, cluster and discriminate analysis,
in a geobiophysical modeling system environment.
The research has been conducted including via primary data collection, preprocessing, and evaluation of remotely sensed
data and feature extraction. The results utilize multistage sampling with 1993 and 1995 digitally derived data from Color
Infrared Photography; simulating multi spectral data to provide a viable alternative for remotely sensed derivations of
vegetation associations in mountainous terrain. Utilizing digitized CIR photography leads to adjusted variables for more
reliable modeling of geobiophysical data for forestation cycles, geologic, climatic and hydrologic processes for long term
forest ecological assessment, management practices and global change impact evaluation.
INTRODUCTION This research investigates forest species and forest
associations discrimination that may eventually be
The mountainous regions of the world are a major source correlated with geologic lower order drainage, lineaments,
of the flood and estuarine coastal plains. These plains are soils and rock outcrops through optimization of spatial
home to the larger proportion of earth's population. The resolution or instantaneous field of view (IFOV 0.1 m -
condition of the mountains with regard to hydrology, 11.0 m) for selected spectral bands from various
vegetation, lithologic outcrops, and their effects upon photographic and electro-optic sensor systems (Levin,
regional and, ultimately, global cycles impacting the forests 1978) . The resulting data are used for feature extraction
are linked(Comins and Noble, 1985). The mountains have affecting pattern recognition in species identification in
been largely ignored due to the complexity of the terrain forest associations that can be evaluated by field analysis
interwoven intricately through geologic, hydrologic, climatic of variance (Mills, et al, 1963). Further, the discrimination
and biologic processes in the generation and rejuvenation of naturally varying spatial frequencies resulting from
of the forests. Mountainous terrain presents numerous vegetation, hydrologic low order drainage, folded strata,
difficulties from a remote sensing perspective in that joints, and fractures with slope, aspect and elevation
process and change often occur on a much smaller spatial effecting microclimate in mountainous terrain, from
scale than typically observed on large plains. Higher artificial periodicities of the recording systems are
spatial frequency variabilities require higher resolution evaluated by a field analysis of variance, cluster and
spatial analysis over similar spectral bands to extract discriminate analysis, and Fourier Analysis (Boyd, et al,
comparable features (Brumfield, et al, 1983). Multi-stage 1982, 1983; Oberly and Brumfield, 1991).
sampling for computerized geobiophysical model of the
Spruce Knob mountainous area forested ecosystem of the Preliminary results by Oberly and Brumfield, 1991,
Appalachian Mountains illustrates a comparative level Bloemer and Brumfield, 1992, and others involving
forestry and related geologic information that may be EOS/TM simulator and orbital data sampling of different
extracted at different spatial resolutions (Bloemer, et al, dates and scenes involving dissected plateau forested
1994; Wriggley, et al, 1985). mountains, indicate that both high and low spatial
frequencies, due to natural and instrumentation
periodicities, can affect feature extraction in pattern
59
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996