ve
id
d.
he
es
>
le
image bands which make in evidence the abundance of specific
categories of land cover types such as vegetation, built-up area,
wet lands, etc, in order to reduce the effects of environment
and then the “noise” due to the mixture between different
classes. These procedures also provide subtle spectral-
reflectance or colour differences between surface materials that
are often difficult to detect in a standard image. Linear Spectral
Unmixing (LSU) technique has been used to determine the
relative abundance of specific categories of coverage of the
soil, according to the spectral characteristics of the materials,
through the use of the original multispectral images composed
of six bands (blue, green, red, near infrared and two middle
infrared bands) The reflectance of a pixel in the image is
considered as a linear combination of the reflectance of each
land cover class present within the pixel itself. The larger the
pixel, the more the mixture of materials occurs. Since the
spectrum of a pixel a weighted average of the quantity of a
material for the spectrum of the material, the use of this
technique allows to generate new images in which a specific
category of use of soil is discriminated and highlighted with
high brightness. For this investigation, LSU it has been applied
twice, based on two different spectral libraries each one based
on three categories: urban, industrial, and soil (Figure 2). The
LSU process provides a total of six layers.
=
5
xe
H
D
ao
sj
T
ai
4
H
E
1
31
i
1
3
Figure 2. Spectral libraries used to make Linear Spectral
Unmixing
Tasseled cap vegetation index, for Landsat TM 5 data, was
employed to perform an orthogonal transformation of the
original data into three factors: Brightness, Greenness, and
Third. Brightness consists of the soil brightness index (SBI),
Greenness is the green vegetation index (GVI), while the third
component is related to soil features, including moisture status.
The concept of tasseled cap transformation is a useful tool for
compressing spectral data into a few bands associated with
physical scene characteristics (Crist and Cicone 1984). In order
to transform multispectral data into a single image band which
enhances the vegetation distribution, the NDVI (Normalized
Difference Vegetation Index), the Atmospherically Resistant
Vegetation Index (ARVI), and the Enhanced Vegetation Index
(EVI) were used. The NDVI transformation, obtained as
Band4-Band3/Band4+Band3, indicates the amount of green
vegetation present in a pixel. Higher NDVI values indicate
more green vegetation. ARVI is an enhancement to the NDVI
that is relatively resistant to atmospheric factors (for example,
aerosol). It uses the reflectance in blue to correct the red
reflectance for atmospheric scattering. The value of this index
ranges from -1 to 1. The common range for green vegetation is
0.2 to 0.8 (Kaufman and Tanre, 1996). EVI was developed to
improve the NDVI by optimizing the vegetation signal in Leaf
Area Index (LAI) regions by using the blue reflectance to
correct for soil background signals and reduce atmospheric
influences, including aerosol scattering. This VI is therefore
most useful in LAI regions, where the NDVI may saturate. The
value of this index ranges from -1 to 1. The common range for
green vegetation is 0.2 to 0.8 (Huete et al. 1997). The
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B7, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
Normalized Difference Soil Index (NDSI) was used because
gives a more reliable estimation in a case of exposed soil
conditions. This index is formulated to portray the characteristic
of responses from soil other than vegetation or water (Kasimu
and Tateishi 2010) and can be helpful to discriminate deciduous
broad-leaved forest and dry land with sparse crop. A simple
logic expression by combining 4-3 was used to enhance
residential areas information, since this composite index allows
an ideal effect to extract urban built-up land (XU H. Q. 2005).
We found out useful another simple expression such as TM4-
TM3 in order to discriminate between Built-up area, soils and
vegetation. The Soil Adjusted Vegetation Index (SAVI) is
generally used to minimize the effects of soil background.
Similarly to the Normalized Difference Vegetation Index, the
near infrared and red bands are used in the calculation, but with
the addition of an adjustment factor (L), which varies between
zero and one (Crocetto and Tarantino 2009). According to Qi et
al. (1994) for this work we have used the Modified SAVI
(MSAVI) which appears to be a more sensitive indicator of
vegetation amount by raising the vegetation signal and
simultaneously lowering soil-induced variations. It also were
used two normalized water indices: the Normalized Difference
Water Index (NDWI) and the Modified NDWI. The NDWI
(McFeeters 1996) works in the same manner as the NDVI
transformation used to map vegetation. This index produces a
single gray-scale image where water is brighter. While the
NDWI works with bands 2 and 4, the modified NDWI
(MNDWI) takes into account band 2 and band 5 (Xu 2006).
The computation of the MNDWI will produce three results: (1)
water will have greater positive values than in the NDWI as it
absorbs more MIR light than NIR light; (2) built-up land will
have negative values as mentioned above; and (3) soil and
vegetation will still have negative values as soil reflects MIR
light more than NIR light (Jensen 2004). In order to improve
the discrimination between impervious land cover categories,
and bare soils, it appears to be really useful the use of the
texture analysis. Taking into account that, within an image are
present different regions characterized by a variation of
brightness, the texture analysis refers to the spatial variation of
the brightness and as a function of the scale. In order that a
given area can be discriminated for different characteristics of
texture, the gray levels within the area must have a high level of
homogeneity between them. Co-occurrence-based texture filter
was used in this study, which provides mean, variance,
homogeneity, contrast, dissimilarity, entropy, second moment,
and correlation. Co-occurrence measures use a gray-tone spatial
dependence matrix to calculate texture values. This is a matrix
of relative frequencies with which pixel values occur in two
neighbouring processing windows separated by a specified
distance and direction. It shows the number of occurrences of
the relationship between a pixel and its specified neighbour’. It
is also taken into account the orographic component of the
territory through the use of the Digital Elevation Model (DEM)
at 30m resolution, and the slope measured in degrees and
calculated on the DEM. It was so obtained a final set of 28
layers also by using the three infra red bands of Landsat: Band
4, band 5, band 7, LSU 1 and LSU 2, Tasselled Cap, NDVI,
ARVI, EVI, NDSI, 4-3, MSAVI, NDWI, MNDWI, Texture
LSU 2, Slope, DEM. To increase the divisibility between
different group objects, all the indices were stretched to the
range from 0 to 255 (Lin et al. 2010), which appears to be also
advantageous to visualize the different behaviours of the
categories for different indices.
" ENVI help guidelines