Geography Reference
In-Depth Information
Table 19.2
Number of pixels selected for different types of LULC change
03/21-04/14
04/14-05/08
05/08-06/01
06/01-06/25
06/25-07/19
07/19-09/29
BL-UB
4,075
1,543
7,514
1,202
1,443
8,134
BL-V
0
1,484
5,301
4,059
1,337
9,009
BL-W
5,090
1,331
10,227
2,538
5,035
545
BL-P
6,526
5,365
333
0
0
10,946
V- B L
10,296
2,709
0
2,988
9,624
7,250
V- U B
256
549
360
2,037
0
3,290
V- W
4,546
1,434
2,407
494
437
1,995
W-BL
5,729
7,748
6,344
11,510
2,655
3,986
W-V
1,018
472
1,585
415
228
5,362
W-P
0
3,890
435
0
0
5,619
horizontal (H) and vertical (V) polarizations depending on the selected mode. Each
scattering element (HH, VV, HV, and VH) has varying sensitivities to different
surface characteristics and properties, thereby helping improve the discrimination
among LULC types.
Field investigations were carried out simultaneously with the acquisitions of the
images to collect ground truth. In the field investigations, field plots were selected
across the typical LULC classes using a clustered sampling approach (McCoy
2005
). In a terrain with poor access, this sampling approach enables the use of most
of the accessible sites. GPS was used to record the coordinates of these field plots.
LULC classes in the study area can be summarized into five categories: urban/built-
up (UB), vegetation (V), paddy (P), water (W), and barren land (BL). On the
basis of the experience with multinomial distribution (Congalton and Green
2009
),
we collected a minimum of 50 samples for each LULC category. The sampling
size per field plot in the images ranged from 39 to 603 pixels. This range was
determined using the ground coverage in the photos taken during the fieldwork.
To evaluate change detection and change type determination results, samples of
different types of LULC change were also selected through visual interpretation and
field investigations. Visual interpretation was first conducted to identify changed
areas from the entire coverage of images. The selected changed areas were then
validated in field investigations. The number of plots of different types of LULC
change was determined in the light of actual change. More samples were selected
for types with more changes, whereas fewer samples were selected for types with
fewer changes. Table
19.2
shows the number of pixels selected for each type of
LULC change.
19.3
Methodology
This study utilized the time series of RADARSAT-2 PolSAR images to extract
short-interval LULC changes. The repeat cycle of RADARSAT-2 is 24 days and
thus monthly LULC changes were extracted by using two successive images.
Search WWH ::
Custom Search