Scientific and Technical Aerospace Reports Volume 38 July 28, 2000

20000064525 Brazil Univ., Inst. of Geosciences, Rio de Janeiro, Brazil
Use of AVIRIS Data for Mineralogical Mapping in Tropical Soils, in the District of Sao Joao D’Alianca, Goias
deMello Baptista, Gustavo Macedo, Brazil Univ., Brazil; deSouza Martins, Eder, Brazil Univ., Brazil; deCarvalho, Osmar Abilio,
Jr., Brazil Univ., Brazil; Meneses, Paulo Roberto, Brazil Univ., Brazil; daSilva Madeira Netto, Jose, Centro de Pesquisas Agropecuaria
dos Cerrados, Brazil; Summaries of the Seventh JPL Airborne Earth Science Workshop January 12-16, 1998; Dec. 19,
1998; Volume 1, pp. 33-42; In English; See also 20000064520; No Copyright; Avail: CASI; A02, Hardcopy; A04, Microfiche
Soil cartography in Central Brazil is available only in large scales. The increased agriculture pressure over these lands require
basic soil knowledge obtainable in more detailed surveys. Data collected by hyperspectral sensors may represent a valuable source
of information for soil scientists, mainly in the distinction of mineralogical classes. Previous studies have shown the possibility
of using indices developed from high-resolution diffuse reflectance spectra obtained in laboratory to estimate hematite content,
and the Ki ratio (molecular ratio of SiO, and Al2O3). In this work we use AVIRIS data to verify the possibilities offered by hyperspectral
data to differentiate tropical soils mineralogy. AVIRIS data from Sao Joao D’Alianca district, Goias state (950816L2
scene 3) after atmospheric correction and reflectance transformation were used. Field work was conducted to sample the main
soil units. The location of the sampled points was obtained with a GPS, which allowed the precise plotting in the image. Samples
were air-dried and sieved to 2 mm before being used for laboratory determination of 400 to 2500 nm spectral reflectance. X ray
diffractions of these samples were also obtained. Most of the scenes considered were covered by crop residues, pasture and native
Cerrado vegetation with only a few hundred hectares of bare soils. The bare soil areas were considered in this work. Kaolinite,
goethite and quartz were present in all samples. Gibbsite and hematite were also present in some sampled soils. Features attributable
to kaolinite, gibbsite, hematite and goethite were clearly detected in the spectra obtained in the laboratory and in the AVIRIS
data. The occurrence of these minerals was confirmed by the X-ray diffraction data.
Derived from text
Minerals; Soil Mapping; Iron Oxides; Remote Sensing; Spectrum Analysis
20000064526 Colorado Univ., Cooperative Inst. for Research in Environmental Sciences, Boulder, CO USA
Incorporating Endmember Variability into Spectral Mixture Analysis Through Endmember Bundles
Bateson, C. Ann, Colorado Univ., USA; Asner, Gregory P., Colorado Univ., USA; Wessman, Carol A., Colorado Univ., USA;
Summaries of the Seventh JPL Airborne Earth Science Workshop January 12-16, 1998; Dec. 19, 1998; Volume 1, pp. 43-52; In
English; See also 20000064520
Contract(s)/Grant(s): NAGw-4689; NAGw-2662; No Copyright; Avail: CASI; A02, Hardcopy; A04, Microfiche
Variation in canopy structure and biochemistry induces a concomitant variation in the top-of-canopy spectral reflectance of
a vegetation type. Hence, the use of a single endmember spectrum to track the fractional abundance of a given vegetation cover
in a hyperspectral image may result in fractions with considerable error. One solution to the problem of endmember variability
is to increase the number of endmembers used in a spectral mixture analysis of the image. For example, there could be several
tree endmembers in the analysis because of differences in leaf area index (LAI) and multiple scatterings between leaves and stems.
However, it is often difficult in terms of computer or human interaction time to select more than six or seven endmembers and
any non-removable noise, as well as the number of uncorrelated bands in the image, limits the number of endmembers that can
be discriminated. Moreover, as endmembers proliferate, their interpretation becomes increasingly difficult and often applications
simply need the aerial fractions of a few land cover components which comprise most of the scene. In order to incorporate endmember
variability into spectral mixture analysis, we propose representing a landscape component type not with one endmember
spectrum but with a set or bundle of spectra, each of which is feasible as the spectrum of an instance of the component (e.g., in
the case of a tree component, each spectrum could reasonably be the spectral reflectance of a tree canopy). These endmember
bundles can be used with nonlinear optimization algorithms to find upper and lower bounds on endmember fractions. This
approach to endmember variability naturally evolved from previous work in deriving endmembers from the data itself by fitting
a triangle, tetrahedron or, more generally, a simplex to the data cloud reduced in dimension by a principal component analysis.
Conceptually, endmember variability could make it difficult to find a simplex that both surrounds the data cloud and has vertices
that are realistic endmember spectra with reflectances between 0 and 1. In this paper, we create endmember bundles and bounding
fraction images for an AVIRIS subscene simulated with a plant canopy radiative transfer model. The simulated subscene is spatially
patterned after a subscene from the AVIRIS image acquired August, 1993 over La Copita, Texas. In addition, for comparison,
we performed a traditional unmixing with image endmembers.
Derived from text
Spectral Reflectance; Spectrum Analysis; Principal Components Analysis; Remote Sensing; Canopies (Vegetation); Mathematical
Models
102