Keywords: CRL/NASDA PI-SAR, Full polarimetric, Land cover, Classification, Hokkaido.
Abstract CRL and NASDA have collaborated to develop an airborne Polarimetric and
Interferometric SAR (CRL/NASDA PI-SAR). The PI-SAR is a high-resolution, dual-frequency
(L/X-band), full-polarimetric and interferometric SAR instrument. A field investigation for
evaluating the potential of PI-SAR in monitoring land cover classes was conducted on a
experiment forest area in Tomakomai, Hokkaido, from 12 July to 17 July 1999.
The potentiality of PI-SAR for land cover classification was evaluated. As a result, a resultant
thematic layer containing seven representative land cover types in the study area was generated.
1. Introduction
The Communications Research Laboratory (CRL) of the Ministry of Posts and
Telecommunications of Japan and the National Space Development Agency of Japan (NASDA)
have collaborated to develop an airborne Polarimetric and Interferometric SAR (CRL/NASDA
PI-SAR) and mounted it on an airplane from 1996. The PI-SAR is a high-resolution,
dual-frequency (L/X-band), full-polarimetric and interferometric SAR instrument. NASDA
developed the L-band SAR in order to calibrate and validate a satellite SAR system in the near
future (Wakabayashi, 1999). The sensor to be calibrated is the Phase Array type L-band
Synthetic Aperture Radar (PALSAR) on board the Advanced Land Observing Satellite (ALOS),
which is scheduled to launch in 2002 (Shimada, 1999; Wakabayashi, 1999).
We conducted a joint CRL and NASDA flight experiment on 14 July 1999 in Hokkaido, Japan.
The purposes of this study are to investigate the backscattering from different land cover targets,
to evaluate the effectiveness of PI-SAR in classifying land cover classes, and to generate a land
cover map of the test area using PI-SAR data. Additionally, we expect that the results from this
work may be used to calibrate and validate PALSAR in the near future.
2. Study Area and the Experiment
The study area was the Tomakomai region located the southern part of Hokkaido, Japan. This is
an area of coniferous and broad-leaved forest including mixed forest plantations. It is a flat
region excluding a small hilly area in the western parts, ranging from 7 to 60 m above sea level.
Several land cover types were distributed in the study area including water, wetland along the
river, bare land including ground and paved highway, forest and urban area.
Field measurements had been taken close to the SAR data acquisition on 14 July. During the
field campaign, it rained from 13 to 16 July. Hence, soil moisture content of the land surface
increased continuously during the campaign (Tadono et al., 1999). Water accumulated in some
flat places. Three trihedral corner reflectors with 80 cm sides and three with 60 cm sides were
deployed parallel to the flight line on the bare land and inside the forest for radiometric
calibration check.
3. Data Processing
A total of four polarization PI-SAR images were acquired. The delivered original image is a
Multi-look Ground-range Amplitude (MGA) image and had a pixel spacing of 2.5 m (4 looks).
Speckle noise was reduced using the Gamma-Map Filter with a 7-by-7 pixel moving window.
Distortion was not removed because the test site is relatively flat.
The intensity of each polarization was converted to Normalized Radar Cross Section (NRCS).
Sigma°HH,VH=20log10(DN)-35.6 (1)
Sigma°HV, VV =20log(DN)-35.0 (2)
where DN is the intensity (Digital Number) on the SAR image.
4. Methodology
A color composite image (5 km by 5 km) based on the polarization of PI-SAR data was then
generated to identify the training signatures for further study. The NRCSs are computed by
averaging the NRCS over all pixels belonging to the signatures contained within the 25-by-25
pixel window. The unsupervised clustering algorithm was first used to produce an adequate land
cover classes to be described hereinafter (Fig. 1). Next, a supervised classification was applied.
A textural algorithm was then applied to extract the urban area. Finally, the correctly classified
classes were gathered in order to obtain the final classification result.
Fig. 1 Analytical procedure for PI-SAR data classification
