Geography Reference
In-Depth Information
There is a variety of problems that confuse the detection of variations in the
reflected EMR: (1) low irregular precipitation and high potential ETP allows only
spatially-limited low vegetation cover by the available moisture. As a result, the
greater part of the area-averaged reflectance of a pixel is for the soil substrate
(Smith et al.
1990a
). Associated problems in these regions include the low organic
components of the soils, which therefore tend to be bright. These issues join to
negate, or reduce, the vegetation signal present within an individual pixel (Huete
et al.
1985
); (2) the variability of soils (light, dark, etc.), and their spectral
responses, over the ecosystem of the study area and over the resulting image also
cause problems to the detection of vegetation.
Existing remote sensing algorithms allow the application of LULC-change
detection in moderate areas of the world (Berberoglu and Akin
2009
). However
these algorithms are less able to be applied in the Mediterranean environment
because: (1) the high temporal variability of the spectral responses of major LC
causes large inter-class spectral variability; (2) the complex mixed spatial fre-
quency of the landscape; and (3) the similar reflectance responses of major LC
makes spectral separation hard (e.g., the bright toned, often calcareous soil can
have alike reflectance responses to urban areas and alike near-infrared reflectance
to a crop canopy) (Berberoglu et al.
2000
). Therefore, the observation of land
cover change is complicated in Mediterranean environments.
Before mapping LULC-change detection using optical sensors data in arid and/
or semi-arid areas, we have to answer this question: at which scale is green
vegetation detectable and how can we best distinguish it? Siegel and Goetz (
1977
)
demonstrated that major changes in the reflectance characteristics need a vege-
tation cover of more than 10 %, and that a vegetation signal has a tendency to be
more significant than the soil signal when vegetation coverage is more than 30 %.
Hill (
2000
) argued that this does not mean that vegetation coverage of less than
30 % is not detectable by remote sensing, but affirms that ratio based vegetation
indices do not offer the best approximation. Vegetation approximation under the
spectral un-mixing concept offers better approximation of the true vegetation
coverage (Hurcom and Harrison
1998
).
A number of change detection studies, such as (Ray
1995
; Kwarteng and
Chavez
1998
; Ram and Chauhan
2009
) rely on the clear difference between
agricultural fields or urban areas, and the neighboring arid environment, in order to
detect LULC-change. However, for example, the detection of vegetative change
(within the same LULC-category) within arid areas is significantly more difficult.
Image differencing, especially the vegetation index differencing, is one of the most
familiar vegetation change detection approaches, because of its simplicity (Singh
1989
; Lu et al.
2003a
). Pilon et al. (
1988
) favored the use of the visible red spectral
band information to detect changes for their semi-arid study area. Chavez and
Mackinnon (
1994
) established that the red band differencing process presented
improved information about vegetation change rather than NDVI in an arid
environment. Lyon et al. (
1998
) accomplished that the NDVI-vegetation index
differencing technique achieved the best when comparing several vegetation
indices for change detection.
Search WWH ::
Custom Search