Figure 1.
Endmember extraction by 2-D scatter plots method.
(a–c) 2-D scatter plots of MNF showing the locations of potential endmembers; (d) Spectral reflectance curves of the endmember pixels selected via MNF 2-D scatter plots are plotted against the original bands.
Figure 2.
Comparison of high albedo endmembers selected from different 2-D MNF scatter plots of MNF.
(a) Overlaying those endmembers derived from MNF 2&3 (b) with those from MNF 1&2, reveals mismatched cases: green points, representing the MNF 1&2 induced high albedo pixels that are not present in MNF 2&3, and red points, representing the opposite cases. The yellow points represent the endmember pixels existing on both MNF 1&2 and MNF 2&3; (c) The overlay of high albedo endmembers is zoomed up in the red rectangular box.
Figure 3.
Selected endmembers by the PPI method.
(a) endmembers displayed in different colors (red - high albedo, green - vegetation, blue - low albedo, and yellow - soil); (b) their spectral reflectance characteristics as summarized from their corresponding image pixels.
Figure 4.
Flowchart of the tetrahedron-based endmember selection approach.
Figure 5.
A 3-D scatter plot of MNF transformed pixels viewed from different angles (a shape of distribution approximates to a tetrahedron, with most pixels being enclosed inside the solid).
Figure 6.
Study area located in the urban core of Shanghai and shown on the false-color composite image of Landsat ETM+ multispectral data, acquired on 3 July 2001 (The red box delineates the coverage of the color-infrared aerial photographs).
Figure 7.
Pareto optimal solutions found by the proposed algorithm.
Figure 8.
The distribution of ETM+ pixels in a 3-D MNF transformed space.
(a) their optimal tetrahedron with vertices in red; (b) the spatial location of the outlying pixels circled by the red ellipse; (c and d) compared to the original ETM+ image and the high-resolution image from Google Earth for their physical identities.
Figure 9.
The locations and spectral reflectance characteristics of potential endmembers in the 3-D MNF scatter plot.
(a) the vertices of the sub tetrahedron were visually identified and denoted as red hollow points, and pixels belonging to each endmember were confined within a sub tetrahedrons at the vertices of the optimal tetrahedron, displayed as green (vegetation), red (high albedo impervious surface), pink (low albedo impervious surface) and yellow (soil) solid points; (b) spectral reflectance characteristics of the selected endmembers were charted for further analysis.
Figure 10.
Endmember pixels obtained through 2-D scatter plots are displayed in 3-D space (green – vegetation, red – high albedo, pink – low albedo, yellow – soil).
Some pixels in the sub tetrahedron were not selected as endmember pixels in 2-D space (see red circle), whereas some endmember pixels determined by 2-D scatter plots were outside the sub tetrahedron (see green circle).
Figure 11.
Four endmember fraction maps resulting from LSMA results with endmembers identified from the tetrahedron-based endmember selection approach (fraction values range from 0 to 1, with the lowest values in blue and the highest values in red).
Figure 12.
Distribution of Shanghai impervious surface.
(a) the original Landsat ETM+ imagery; (b) the fraction map derived with the V-H-L-S model.
Figure 13.
Error distribution along the value range of impervious surface fraction maps generated using different endmember derivation methods.
Figure 14.
Comparisons of impervious surface estimation accuracy with endmembers derived from 2-D scatter plots (a), PPI (b), and 3-D scatter plot (c).
Table 1.
A comparison of model performance for impervious surface estimation.
Figure 15.
Spatial patterns of four typical land use types in Shanghai (color-infrared aerial photographs (above) and impervious surface fraction maps (below)) (a) medium-intensity residential; (b) high-intensity residential; (c) very high-intensity residential; (d) commercial.