Integrating Landsat with MODIS Products for Vegetation Monitoring - Satellite-based Applications on Climate Change - page 244

Fig. 16.3 A simulation test for changing reflectance and linear objects. The MODIS-like resolu-
tion data ( d - f ) are aggregated from the Landsat-like data ( a - c ), respectively. Imagery ( g )is
predicted from fine-resolution imageries ( a ) and ( c ) and coarse-resolution imageries ( d - f ). Imag-
ery ( h ) is the absolute difference between prediction ( g ) and truth ( b ). In this test, the STARFM
algorithm predicts constant reflectance, changing reflectance, and linear segment ( within circle )
well. Linear segments outside circle are predicted but not perfect value
images with background changing from 0.1 (Fig. 16.3a ) to 0.2 (Fig. 16.3b ) and then
0.4 (Fig. 16.3c ). The simulated water body (circle) has a constant reflectance of
0.05. The simulated road (over background) and bridge (over water) have a constant
reflectance of 0.5. The MODIS-like data (Fig. 16.3d-f ) are aggregated from the
Landsat-like data. Roads (over background) are still visible in Fig. 16.3d, e ,butnot
in Fig. 16.3f due to the smaller contrast of roads and background in (Fig. 16.3c ).
Figure 16.3g is a predicted version of Fig. 16.3b using fine-resolution images
(Fig. 16.3a, c ) and coarse-resolution imageries (Fig. 16.3d-f ). Figure 16.3h shows
the absolute difference between prediction (Fig. 16.3g ) and truth (Fig. 16.3b ). In this
test, constant reflectance, changing reflectance, and the bridge are predicted per-
fectly. Roads (segment outside circle) are predicted but not perfect value.