4.6 CONCLUSIONS
This review starts with a discussion of the requirements for mapping urban materials,
land cover, and land use and their relationships. This examination allows us to
understand what spectral property should and should not be considered in the imaging
and mapping. The level of detail in classification categories is essentially modulated
by spectral resolution. But, another important consideration in urban remote sensing is
spatial resolution. Its interaction with the fabric of urban landscape and the problem of
mixed pixels are the subject of the debate between the pixel and subpixel approaches.
Research has yet to clearly reveal the linkage between the categorical scale and spatial
resolution.
Scale influences the examination of landscape patterns (Liu and Weng, 2009). The
change of scale is relevant to the issues of data aggregation, information transfer, and
the identification of appropriate scales for analysis (Krönert et al., 2001; Wu and
Hobbs, 2002). Sections 4.4 and 4.5 relate to this theme of study. The patterns of land
surface temperature at different aggregation levels were assessed in order to find the
operational scale/optimal scale through two case studies. The issues of data aggregation and information transfer were addressed by reviewing the concept of scale
dependency and by discussing LST variability and residential population estimation
modeling across multiple census levels.
Remote sensing technology has been evolving rapidly in the twenty-first
century. New frontiers such as very high resolution sensing, hyperspectral sensing,
lidar and their synergy with existing technologies and advances in image processing
techniques (such as object-oriented image analysis, data fusion, artificial neural
networks) are changing the image information content we obtain and the way we
handle the image processing. Both aspects will change our understanding of this
basic but pivotal issue—scale and therefore future research should be warranted in
these aspects. For example, an increase in spatial resolution may not lead to a better
observation since objects may be oversampled and their features may vary and be
confusing (Hsieh et al., 2001; Aplin and Atkinson, 2004). Moreover, imagery
with too fine resolution for specific purpose can be degraded in the process of
image resampling (Ju et al., 2005). In the object-based image analysis, the
extraction, representation, modeling, and analyses of image objects at multiple
scales have become common concerns of many researchers (Hay et al. 2002a,b;
Tzotsos et al., 2011).
REFERENCES
Alker, H. R. 1969. A typology of ecological fallacies. In M. Dogan and S. Rokkan (Eds.),
Quantitative Ecological Analysis in the Social Sciences. Cambridge, MA: MIT Press,
pp. 69–86.
Aplin, P., and Atkinson, P. M. 2004. Predicting missing field boundaries to increase
per-field classification accuracy. Photogrammetric Engineering & Remote Sensing
70(1):141–149.
REFERENCES
73
This review starts with a discussion of the requirements for mapping urban materials,
land cover, and land use and their relationships. This examination allows us to
understand what spectral property should and should not be considered in the imaging
and mapping. The level of detail in classification categories is essentially modulated
by spectral resolution. But, another important consideration in urban remote sensing is
spatial resolution. Its interaction with the fabric of urban landscape and the problem of
mixed pixels are the subject of the debate between the pixel and subpixel approaches.
Research has yet to clearly reveal the linkage between the categorical scale and spatial
resolution.
Scale influences the examination of landscape patterns (Liu and Weng, 2009). The
change of scale is relevant to the issues of data aggregation, information transfer, and
the identification of appropriate scales for analysis (Krönert et al., 2001; Wu and
Hobbs, 2002). Sections 4.4 and 4.5 relate to this theme of study. The patterns of land
surface temperature at different aggregation levels were assessed in order to find the
operational scale/optimal scale through two case studies. The issues of data aggregation and information transfer were addressed by reviewing the concept of scale
dependency and by discussing LST variability and residential population estimation
modeling across multiple census levels.
Remote sensing technology has been evolving rapidly in the twenty-first
century. New frontiers such as very high resolution sensing, hyperspectral sensing,
lidar and their synergy with existing technologies and advances in image processing
techniques (such as object-oriented image analysis, data fusion, artificial neural
networks) are changing the image information content we obtain and the way we
handle the image processing. Both aspects will change our understanding of this
basic but pivotal issue—scale and therefore future research should be warranted in
these aspects. For example, an increase in spatial resolution may not lead to a better
observation since objects may be oversampled and their features may vary and be
confusing (Hsieh et al., 2001; Aplin and Atkinson, 2004). Moreover, imagery
with too fine resolution for specific purpose can be degraded in the process of
image resampling (Ju et al., 2005). In the object-based image analysis, the
extraction, representation, modeling, and analyses of image objects at multiple
scales have become common concerns of many researchers (Hay et al. 2002a,b;
Tzotsos et al., 2011).
REFERENCES
Alker, H. R. 1969. A typology of ecological fallacies. In M. Dogan and S. Rokkan (Eds.),
Quantitative Ecological Analysis in the Social Sciences. Cambridge, MA: MIT Press,
pp. 69–86.
Aplin, P., and Atkinson, P. M. 2004. Predicting missing field boundaries to increase
per-field classification accuracy. Photogrammetric Engineering & Remote Sensing
70(1):141–149.
REFERENCES
73