Geography Reference
In-Depth Information
alarms were mainly caused by the growth of paddies. Paddies in the study area
have two growth cycles a year. The first growth cycle is usually from early April
to late July, and the second growth cycle is from early August to early October.
Paddy fields usually turn into barren land covered with stubble and weeds during
the other time periods. The study found that paddy fields showed very different
backscattering characteristics in PolSAR images with the growth of paddies. The
main radar return from paddy fields was paddy-water backscatter (double-bounce
scattering) on June 25, but the double-bounce scattering from the paddy fields
significantly reduced on July 19 because the water beneath the paddies was almost
dried up. Meanwhile, with the growth of paddies, crowns of the paddies became
dense, and volume scattering from paddy crowns increased. The main radar return
from the paddy fields was dominated by the volume scattering on July 19, and it was
hard to distinguish between paddies and other vegetation on that time. Therefore,
the change of backscattering characteristics of the paddies fields caused many false
alarms in the change detection on July 19. The second growth cycle of paddies
started as soon as the end of the first growth cycle. A similar growth process was
found in the second growth cycle. The double-bounce scattering from the paddy
fields increased again. For the same reason, many false alarms were found in the
paddy fields in the change detection on September 29.
The accuracy evaluation for determining types of changes was also carried out
based on the confusion matrix that was determined using the validation samples
of different types of LULC change. We calculated four statistics for the validation:
overall accuracy (OA), estimate of kappa (Kappa), producer's accuracy (PA), and
user's accuracy (UA) (Story and Congalton
1986
; Congalton and Green
2009
). As
showninFig.
19.6
, the average overall accuracy and average kappa value were
72.44 % and 0.68 % respectively. The accuracies were different at different dates,
and the overall accuracies and kappa values on September 29 and July 19 were the
lowest. Paddy fields were easily confused with vegetation on July 19 because of
the similar scattering mechanism, but they became easier to be distinguished from
vegetation on September 29. Therefore, many false alarms caused by the growth
of paddies were mistakenly identified as changes from vegetation to paddies on
September 29. The growth of paddies also caused many false alarms that were
misclassified as changes from paddy to vegetation on July 19. Moreover, some
paddies were harvested on July 19 and then planted again. These paddies were
almost ripe and prone to be confused with vegetation on September 29. This caused
some false alarms to changes from barren land to vegetation. The overall accuracies
and kappa values on May 08 and June 01 were also low. Some paddies were planted
in early April and grew fast from April 14 to May 08. The paddy fields were easily
confused with urban/built-up areas due to the strong double-bounce scattering from
paddy-water surface on May 08. The fast growth of paddies produced many false
alarms to changes from paddy fields to urban/built-up areas in the change detection
on May 08. There were also some paddies planted in late April, which were easily
confused with barren land or water because of sparse vegetation and irrigation. This
created many false alarms to changes from barren land to paddy fields and that from
water to paddy fields.
Search WWH ::
Custom Search