ACCURACY ASSESSMENT OF LINEAR SPECTRAL MIXTURE MODEL(LSMM) IN
ESTIMATING VEGETATION ABUNDANCE IN MOUNTAIN AREA
WANG Tianxing\CHEN Songlin',MA Ya 2
(1 College of Geographical Sciences ,Fujian Normal University , Fuzhou 350007 ,China;
2 Guangzhou Institute of Geochemistry, Chinese Academy of Sciences Guangzhou 510640,China)
KEYWORDS: Linear Spectral Mixture Model —LSMM; Terrain Undulation; Correction; Vegetation Abundance; NDVI; MVI;
Regression Analysis
ABSTRACT:
Linear Spectral Mixture Model—LSMM which is prevailing presently is one of pixel unmixing models,the unmixing-accuracy of
which is restricted by kinds of factors,but now the research about LSMM is mostly focused on appraisement of linear hypothesis
relating to itself and techniques used to select endmembers, nevertheless, the environment conditions of study area which could sway
the unmixing-accuracy such as atmosphere reflectance or scatteration and terrain undulation are not studied.This paper probes
emphatically into the accuracy uncertainty of LSMM resulting from the terrain undulation with reference to unmixing vegetation
abundance under LSMM. ASTER data set was chosen and the C terrain correction was applied to it. Based on this, vegetation
abundances were extracted from both pre-C corrected and post-C terrain illumination corrected ASTER using LSMM, then the
regression analysis between vegetation abundance and vegetation indices(NDVI and MVI) was further conducted to assess the
unmixing accuracy which quantitatively measure the impact of terrain illumination on LSMM. The results indicate that terrain
undulation could dramatically constrain the application of LSMM in inversion of vegetation abundance. A improved unmixing
accuracy of 17.6% and 18.6% for R2 was achieved in regression against to NDVI and MVI respectively because of the removing
terrain undulation by C correction method. Especially, effective removal or minimization of terrain effects is essential in
mountainous areas. This study can also provide new theory basis for LSMM applications in mountainous areas. Though we took
vegetation abundance as a case study, it should be envisioned that the similar result for other endmember types (water, barren soil,
impervious area and so on ) could be achieve because of the same impact mechanism of terrain undulation and the identical unmixing
procedure with LSMM.
1. INTRODUCTION
Information of land surface targets observed by remote sensing
is measured spatially by pixels. Because of the the heterogeneity
of ground features and relatively coarse spatial resolution of the
satellite borne imagery characterized by
TM,ETM+,MODIS,NOAA/AVHRR and so on., it is common
that mixture spectra are generated when the pixel is occupied by
more than one land-cover class (Ichoku &Kamieli, 1996). The
effective information interpreted from mixture pixels is limit for
quantitatively analyzing the characteristics of the targets. So a
large number of sub-pixel models are developed, such as Linear
or non-linear Spectral Mixture Model, Probabilistic model,
Geometric-optical
2. TEST SITE AND DATA
Considering the computational burden and representation, we
choose the rectangle-shaped suburb area of Fuzhou(26° 10 ' N,
116 ° 21 ' E at center) where the land covers are
abundant(Fig. 1). The site is covered by cropland, water, forest
vegetation which is dominant, urban cover which mostly locate
at SW of the site and bare soils. The terrain condition within the
area is relatively undulate with elevation from 50m to 1000m, it
is appropriate to investigate the impact of terrain undulation on
LSMM.
As one of the recent developments in remote sensing
technology , Advanced Spacebome Thermal Emission and
Reflection Radiometer (ASTER) provide substantial
improvements over the traditional multispectral sensor, such as
Landsat thematic mapper (TM), in spatial, spectral and
radiometric resolutions, and has become a vital data source for
earth science researchers (Fang Qiu et al,2006). ASTER is a
multispectral scanner that produces images of high spatial
resolution launched on July, 1999 aboard on the first platform
of NASA's Earth Observing System -Terra. The instrument
has three bands in the visible and near-infrared (VNIR) spectral
range (0.5-0.9 pm) with 15-m spatial resolution, six bands in the
shortwave infrared (SWIR) spectral range (1.6-2.4 pm) with
30-m spatial resolution, and five bands in the thermal-infrared
(TIR) spectral range (8-12 pm), with 90-m resolution (Kahle et
al.,1991 ;Abrams,2000). An additional backward-viewing
telescope with a single band duplicating VNIR band 3 could
provide the capability for same-orbit stereogram metric data at
15m spatial resolution.
Another appealing aspect of ASTER data is the open
availability of its data and even the on-demand standard
products for research use are at very low cost (Rowan & Mars,
2003).
With the advantage of ASTER data described above, we
collected ASTER data in Fuzhou area attempting to make use of
the improved spatial and spectral information of ASTER data.
WANG Tianxing, E-mail: watixi@163. com. TEL: 13459194134.