15.3.4 Importance of Observation Scale in Land Cover Extraction from
Remote Sensing
The most stable and reliable source of land cover information that can be obtained on
the ground and be linked between remote sensing studies obtained at different scales is
the spectral response of the cover (McCoy, 2004). Field spectroradiometry and spectral
library data have demonstrated already that there is a lot to offer. For example, Rao
et al. (2007) showed the application of field spectroradiometry data of agricultural
crops in India, at the canopy and pixel scales, for classification on Hyperion imagery.
Promising results of 86.5 and 88.8% overall classification accuracy for both scales
respectively were achieved. Nevertheless, the similarity and variability of the spectral
properties of land cover inevitably render high mapping uncertainty in pixel-based
analysis of heterogeneous and cover fragmented landscapes, present in many
geographic regions (Xie et al., 2008).
The spectral signature of vegetation is primarily driven by the different biochemical content and pigment quantity, plant architecture, and growing factors. A
certain degree of variation is thus naturally expected to occur in the reflectance values
between different plants. Field spectroradiometry takes the benefit of the ground scale
in detecting such variation because it is naturally contained in the spectral information
recorded on the field and can indeed be used in enhancing the discrimination between
different covers. It also has potential for identification of spectral bands that are
correlated with land cover properties other than type, such as vegetation health status
or leaf chemical compounds, with high accuracy, though it remains challenging
(Abdel-Rahman et al., 2009; Thenkabail et al., 2004b). Furthermore, discriminant
analysis of field spectral data offers an important understanding of the cover
discrimination potential prior to image classification (McCoy, 2004). If the differences of the reflectance between the land covers being compared in certain wavelengths are significant in statistical means, the results can then provide a sound basis
for future remote sensing mapping. In addition, field hyperspectral data contain
information about the land cover internal properties in its natural environment as
observed from the highly elevated remote sensors. Such data can be additionally used
in digital imagery for proper selection of training pixels for land cover mapping or for
pixel unmixing in more detailed analysis of a subscene (Borengasser, 2007).
Therefore, the advances in sensor technology in higher spatial and spectral resolution
over broader spectral regions impose the need for more accurate and representative
spectral signatures for solving the land cover variability on imagery data (Salisbury,
1998). Both field discrimination and image classification face the need for distinct
spectral signatures obtained on the field scale, especially for economically and
ecologically important land covers such as vegetation.
Of special importance in land cover information extraction and analysis on the
ground scale is the process of planning and executing of fieldwork in remote sensing
studies. The spectral reflectance is an inherent property of an object, independent of
time, location, illumination intensity, and atmospheric conditions (Peddle et al.,
2001). However, this property is strongly affected by the local environment and the
behavior of the field equipment. The benefit of a systematic approach to planning and
310
HYPERSPECTRAL REMOTE SENSING WITH EMPHASIS ON LAND COVER MAPPING
Remote Sensing
The most stable and reliable source of land cover information that can be obtained on
the ground and be linked between remote sensing studies obtained at different scales is
the spectral response of the cover (McCoy, 2004). Field spectroradiometry and spectral
library data have demonstrated already that there is a lot to offer. For example, Rao
et al. (2007) showed the application of field spectroradiometry data of agricultural
crops in India, at the canopy and pixel scales, for classification on Hyperion imagery.
Promising results of 86.5 and 88.8% overall classification accuracy for both scales
respectively were achieved. Nevertheless, the similarity and variability of the spectral
properties of land cover inevitably render high mapping uncertainty in pixel-based
analysis of heterogeneous and cover fragmented landscapes, present in many
geographic regions (Xie et al., 2008).
The spectral signature of vegetation is primarily driven by the different biochemical content and pigment quantity, plant architecture, and growing factors. A
certain degree of variation is thus naturally expected to occur in the reflectance values
between different plants. Field spectroradiometry takes the benefit of the ground scale
in detecting such variation because it is naturally contained in the spectral information
recorded on the field and can indeed be used in enhancing the discrimination between
different covers. It also has potential for identification of spectral bands that are
correlated with land cover properties other than type, such as vegetation health status
or leaf chemical compounds, with high accuracy, though it remains challenging
(Abdel-Rahman et al., 2009; Thenkabail et al., 2004b). Furthermore, discriminant
analysis of field spectral data offers an important understanding of the cover
discrimination potential prior to image classification (McCoy, 2004). If the differences of the reflectance between the land covers being compared in certain wavelengths are significant in statistical means, the results can then provide a sound basis
for future remote sensing mapping. In addition, field hyperspectral data contain
information about the land cover internal properties in its natural environment as
observed from the highly elevated remote sensors. Such data can be additionally used
in digital imagery for proper selection of training pixels for land cover mapping or for
pixel unmixing in more detailed analysis of a subscene (Borengasser, 2007).
Therefore, the advances in sensor technology in higher spatial and spectral resolution
over broader spectral regions impose the need for more accurate and representative
spectral signatures for solving the land cover variability on imagery data (Salisbury,
1998). Both field discrimination and image classification face the need for distinct
spectral signatures obtained on the field scale, especially for economically and
ecologically important land covers such as vegetation.
Of special importance in land cover information extraction and analysis on the
ground scale is the process of planning and executing of fieldwork in remote sensing
studies. The spectral reflectance is an inherent property of an object, independent of
time, location, illumination intensity, and atmospheric conditions (Peddle et al.,
2001). However, this property is strongly affected by the local environment and the
behavior of the field equipment. The benefit of a systematic approach to planning and
310
HYPERSPECTRAL REMOTE SENSING WITH EMPHASIS ON LAND COVER MAPPING