13
A SAMPLING PROCEDURE TO GUIDE THE COLLECTION OF
NARROW-BAND, HIGH-RESOLUTION SPATIALLY AND
SPECTRALLY REPRESENTATIVE REFLECTANCE DATA*
Richard R. Brand John L. Barker
Geographic Information Earth Resources Branch
Systems of Virginia NASA-Goddard Space
P.0. Box 583 Flight Center
Sterling, VA 22170 Greenbelt, MD 20771
ABSTRACT
Described is a multi-stage sampling procedure to guide the
collection of representative, high resolution spectra sup
porting the prioritization of spectral band characteristics
for future satellite sensors. Since Multi-Linear Array
(MLA) sensors will produce spectral data in the order of 10
nanometers (nm) at spatial resolutions of 10 to 15 meters
(m) , pre-design characterization experiments must obtain
data more refined in spectral and spatial terms. A quanti
tatively based sampling design is proposed to derive a rep
resentative set of discrete reflectance targets. The proce
dure is general and can be adapted to characterize areas as
small as minor watersheds, as large as multi-state regions,
or even the entire United States. Beginning with a user de
termined study area, successive reductions in size and spec
tral variation are performed using image analysis techniques
on data from the Multispectral Scanner, orbital and simulat
ed Thematic Mapper, low altitude photography synchronized
with the simulator, and associated digital data. An inte
grated image-based geographical information system supports
processing requirements. The sampling design produces a po
sitionally accurate, geographically referenced field guide
to spectrally representative 5 m minimum polygons of land
cover to guide the collection of field data with a spectral
reflectometer imaging from 350 to 2500 nm. Interband corre
lation studies can then be initiated in search of optimum
spectral bands for sensor characterization.
SPATIAL AND SPECTRAL INFORMATION REQUIREMENTS
Narrow-band multispectral data with sensitivity less than 5
nm and a spectral range of 350 to 2500 nm are expected from
future MLA sensors. Accordingly, characterization experi
ments to prioritize spectral band locations and widths
should be based on empirically derived data with at least 5
nm resolution. Similarly, spatial requirements are a func
tion of both the potential sensor capability and the infor
mation content of its spatial resolution. A spatial resolu
tion of the order of 5 m is needed for defining sensor
* Synthesis of a report to the Earth Resources Branch, NASA-
Goddard Space Flight Center in support of a project enti
tled "Spectral Band Selection for Satellite Sensor Sys
tems" .
