Microwave Remote Sensing for Land cover identification
Objective
The aim of the preliminary study is to investigate and to assess the potential application of using digital
JERS-1 SAR imagery for land cover identification.
Study Area
The study area is located in the province of Khon Kaen, north- east Thailand (Figure1). It lies between
latitude 16°29’ and 16°45’N, and between longitude 102°45’and 103°1' E covering the area of
approximately 825 sq.km. The topography of the area covers river terrace, flood plain, low to middle and
high terrace, undulating features. The flood plain extends along Pong river. There are villages setting
along both sides of river bank. The amount of annual rainfall is approximately 1,200 mm. The major
cultivations in this area are paddy and upland crops (i.e, cassava, sugarcane). For mixed orchard and
forest, they are found in the higher area. The typical orchard produce is papaya, mango, tamarind and
eucalyptus. Vegetation cultivation, particularly cabbages, is carried out on the flood plains beside Pong
river. Paddy fields consist of rainfed paddy fields and large area irrigated paddy fields under Nong Wai
irrigation project. However, the second rice and vegetables are cultivated in this irrigated area in the dry
season.
Data and Materials
- JERS-1 SAR data
The three scenes of digital multi-temporal SAR imagery are available on 5 January 1998, 3 April 1998
and 17 May 1998 with path 124 and row 273. (Figure 1)
- Topographic map at a scale of 1:50,000
- Image processing analysis system of PCI EASY/PACE software.
Methodology
Geometric Correction
These three digital SAR imageries were geocoded against topographic map and registered to a
Universal Transverse Mercator (UTM) projection. The SAR data coded on 16 bits for each scene have
only a single panchromatic band and have Digital Number (DN) values ranging from 0 to 65,000. This
poses a problem to display of the data. Computer monitor generally displays a monochrome image in
256 shades of grey level. Very few details can be made from the original data. Therefore, the SAR data
obtained on 16 bits were compressed to 8 bits in order to obtain 256 values of intensity. Then the JERS-1
SAR data were displayed and examined.
Filtering of SAR Data
The objective is to reduce speckle noise of the SAR image. A Frost filter using at 5x5 window size was
employed as a spatial filter to the SAR data to remove the speckle noise while retaining the image
sharpness.
Image Enhancement
All images were digitally enhanced using histogram equalization in order to increase the contrast of the
image.
Figure 1. Study area and JERS-1 SAR original data images
Visual Interpretation
Composite color images were generated for visual analysis and for a selection of profiles in
representative areas for land cover type classification. The color composite SAR images contained data
on 3 April 1998 in red, 17 May 1998 in green and 5 January 1998 in blue. Visual interpretation was
investigated on the composite color image, general categories were identified and discussed based on
topographic maps at a scale of 1:50,000 made by the Royal Thai Survey Department and other existing
data.
Ground Truth Survey
Ground information was collected from field observation and made possible to acquire necessary ground
truth for image interpretation, analysis and image classification.
Training Area and Statistical Analysis
The test sites or training areas were selected to get a representative of various land cover types and to
have a good ground truth (polygons) available on the images. The statistics of pixel values(unitless) for
all categories of training areas were obtained and discussed by mean and standard deviation. The.statistical characteristics of training area for various land cover derived from JERS-1 SAR data are
shown in Table 1.
Table 1 Statistical characteristics of pixels values of SAR images for land cover classification.
| Land cover type |
Image of 3 April 1998 |
Image of 17 May 1998 |
Image of 5 Jan 1998 |
| |
Mean |
Deviation |
Mean |
Deviation |
Mean |
Deviation |
| Paddy |
18.375 |
4.529 |
12.816 |
2.947 |
16.159 |
4.367 |
| Upland crops |
12.153 |
3.232 |
11.927 |
3.371 |
14.212 |
3.816 |
| Forest |
20.770 |
3.190 |
20.264 |
3.53 |
23.032 |
3.909 |
| Riparian forest |
19.769 |
4.808 |
15.166 |
3.256 |
21.252 |
5.060 |
| Village |
45.023 |
21.840 |
34.873 |
15.365 |
46.305 |
21.849 |
| Water bodies |
7.598 |
1.348 |
5.605 |
1.289 |
8.635 |
1.785 |
Supervised Classification
Some statistics were calculated for each land cover, for all images as illustrated in Table 1. These three
sets of data were used to classify land cover using the Maximum Likelihood method. Digital image
processing technique was employed to identify and performed on PCI EASI/PACE software. The
procedure of image processing is shown in Figure 2.
Figure 2 General methodology of the digital image processing
