Our change detection methods, combined with the readily
available
NLCDs
and Landsat Analysis Ready Data, can be di-
rectly applied to other fast-growing urban regions in the U.S.
to derive annual urban changes. The annual urban change
maps contribute to
NLCD
impacts. More importantly, the high
temporal urban change information could potentially improve
our understanding of urban development patterns/trends.
For example, the linkage between urban development and
population dynamics (or other socioeconomic factors) could
be examined in a more continuous modeling framework.
The main limitation for our change detection algorithms is
their poor performance in dealing with subgroup 22 (pixels
changed from low intensity to medium-high intensity urban).
However, we note that the number of pixels in the subgroup
22 pixels is much smaller than those of other urban change
categories, thus our change detection methods would main-
tain relatively high overall accuracies when all change pixels
are included for accuracy assessment. Future study needs to
focus on improving change detection accuracy for subgroup
22. Potential solution may include exploring other spectral
and spatial indices in addition to the commonly used
NDVI
time-series. Newly developed change detection algorithms,
Figure 7. Frequency distribution of annual urban change.
Figure 6. Urban change years: pixels wer
ation purpose.
722
October 2019
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
