Environmental Engineering Reference
In-Depth Information
Fig. 16.4
Predicted Landsat surface reflectance (
bottom row
) from daily MODIS reflectance
imagery (
top row
) and Landsat/MODIS imagery pairs (in
blue rectangles
) on the BOREAS
Southern Study Area (54
N, 104
W). In this example, Landsat observations were not available
on 6/4 (
t
2
) and 7/4 (
t
3
) in 2001 - the Landsat-resolution images at
t
2
and
t
3
are entirely
synthesized from the simultaneous MODIS data and bracketing Landsat images using the
STARFM algorithm
Figure
16.4
shows the STARFM predicted Landsat surface reflectance on 6/4
and 7/4 in 2001 for the BOREAS Southern Study Area using same-day MODIS
images and ETM+/MODIS image pairs on 5/24 and 7/11 in 2001. In this example,
Landsat observations are not available on 6/4 (
t
2
) and 7/4 (
t
3
). This process may also
be thought of as using daily MODIS information to make a “time correction” to
infrequent Landsat data. The predicted images (bottom row) capture rapid seasonal
changes from MODIS data while retaining the Landsat spatial details. Clear land
and water boundaries can be predicted. Linear objects such as roads can be
observed in the predicted images.
Initial validation studies show that both mean differences and absolute
differences of surface reflectance between predictions (i.e., synthesized Landsat-
resolution data) and real Landsat observations are small. STARFM can preserve the
high spatial resolution of Landsat and high temporal resolution of MODIS if “pure”
coarse-resolution neighbor pixels can be found within the moving window. For
complex mixtures of different land cover types, performance degrades somewhat,
and the adjustment of algorithm parameters may be needed to improve prediction.
Search WWH ::
Custom Search