The Shuttle Imaging Radar-C/X-band SAR)SIR-C/X-SAR) and GlobeSARdata are used in the study. They were acquired in April 1994 and November 1993. Table 1 shows their system parameters. For the GlobeSAR data, we performed antenna pattern correction for the data to remove the artificial for the data to remove the artificial brightening in range due to side looking illumination. The digital elevation model (DEM) generated by digitzing a 1:50,000 topographical map was used for geometric correction of the data and to reduce

The terrain effect. Combining C and x band with HH, HV and VV polarization SAR data produced the false color image. Filter and image enhancement techniques were used to improve the quality of image. For the SIR-C/X-SAR data, we did geometric rectification and speckle filtering prior to generating the false color composite images.

Table1. The system parameters of SIR-C/X-SAR and GlobeSAR data

System Parameter	SIR-C/X-SAR	GlobeSAR
Wavelength (cm)	L(23.5),C(5.8), X (3.1)	C(5.66),X(3.24)
Plarization	HH,HV,VH,VV	HH, HV, VH, VV
Azimuth	25x 25	6 x 6
Resolution(m) Range	13x26	6 x 6
Resolution(m) Imaging swath (km)	15~90L and C), 15~60(X)	22
Pixel space (m)	12.5 x 12.5	6 x6
Look angle(°)	20~55
Flight altitude	225km	5400m
Platform	Shuttle	Airplane

Effect of Multiparamters SAR on Forest Type Discrimination and Classification
The SIR-C/X-SAR data were used to study the effect of forest type discrimination and classification. In the test site for three north shelter forest, the false color frequency-polarization combinations of L-HH, L-HV, C-HH, CH-HV, and X-VV were generated. In the false color composite image of L-HV (Blue), coniferous forest appears as yellow, mixed forest as red and white, and deciduous forest as blue; therefore three forest types can easily be discriminated by using color (Figure 1,2).


Figure 1. Multifrequency and multipolarization false color SIR -C/X-SAR image of three north shelter forests in the Yichuan area, Shanxi Province.

Figure 2. Classification map of three north shelter forests in the Yichuan area, in the Yichuan area, Shanixi Province based on multifrequency and multipolarization SIR-C/X-SAR data.
Five to seven training areas were selected from the image based on the 1:500,000 "Forest Distribution Map in Loess Plateau Area of China". A supervised classification, maximum likelihood method was employed using L and C bands, HH and HV polarization. Coniferous, deciduous, and mixed forest, river course plain or floodplain, and radar shadow can be discriminated on the image. The accuracy of classification results was analyzed, and the results showed that coniferous forest had the best classification accuracy at 79.7%, mixed forest had an accuracy of 68%, and deciduous forest had an accuracy of 60.2%. The average classification accuracy was 75.10%, and the overall accuracy was 70.4%. Therefore, the SIR-C/X-SR data were quite successful for discrimination of three north shelter forest types.

Figure 1. Multifrequency and multipolarization false color SIR -C/X-SAR image of three north shelter forests in the Yichuan area, Shanxi Province. Figure 2. Classification map of three north shelter forests in the Yichuan area, in the Yichuan area, Shanixi Province based on multifrequency and multipolarization SIR-C/X-SAR data. The radar backscatter coefficient can demonstrate the interactive relationship between radar beam and ground targets. In the study, it was used to analyze and discuss the effect of multiparameters SAR data in forest discriminiation and classification. We extracted the backscatter coefficient (Table 2) of the different frequency and plarization SIR-C/X SAR data using the calibrated results (Freeman et al., 1995). For the single band and single polarization data, the backscatter coefficient values of coniferous, deciduous and mixed forest is very similar. It suggests that is very difficult to identify and separate these three types of forest in the single band and single polarization image.

Table 2. The radar backscatter coeffiecients (dB) of three types of forest in Yichuan area, Shanxi Province

Forest Type	Coniferous	Deciduous	Mixed forest
L-HH	-4.68	-5.75	-4.58
L-HV	-7.28	-7.96	-7.62
C-HH	-4.67	-4.86	-3.56
C-HV	-6.50	-6.19	-5.84
X-VV	-3.64	-3.37	-3.46

When the wavelength is difficult, the backscatter coefficient values of same polarization of these three types of forest have no much variable, such as -3.65~5.75dB for L-HH and C-HH, -5.84~7.96dB for L-HV and C-HV, and -3.37~-3.64dB for X-VV. However at the different polarization, the values of same frequency have large change. The result shows that it is more efficient for polarization than wavelength to discriminate the forest type. This reason is that the wavelength mainly determines that the radar penetrates the ability of forest canopy, and the polarization mainly delineates geometric features of canopy, so it is available for different types of forest discrimination. So the forest types can be efficiently discriminated with the combination of multifrequency and multipolarization SAR data.

Forest Volume Estimation Based Relating Radar Backscatter to Forest Stand Parameters

It is important to study the relationship between radar backscatter and forest parameters. It may improve understanding of the relationship between radar backscatter and phonological variables, and improve radar backscatter models of tree canopy properties, and develop a radar-based scheme for monitoring forest phonological changes.

The GlobeSAR data were uncalibrated. We utilized the radar backscatter intensity instead of the radar backscatter coefficient. To compute the mean of SAR data for a stand, the stand on the SAR imagery was located, and the largest possible window within the stand was extracted. For each stand, at least 200 image pixels were averaged. In the study, the mean values of radar vackscatter intensity from C and x bands, HH and VV polarizations were extracted.

Correction between the SAR data and forest stand parameters has been analyzed. The correlation coefficients between the SAR data and the forest stand parameters are shown in Table 3. The forest stand parameters are shown in Table 3. The forest stand parameters consist of stand basal area, density, mean height and dbh. It can be seen that the correlation coefficients are larger in the mean stand height and dbh than in the stand basal area and density. As the mean dbh and height of trees in the stands increase, the C-HH, C-VV, X-HH and X-VV backscatter increase. However, there is almost no relationship between the radar backscatter and the stand density, and the backscatter from the C-HH, C-VV, X-HH and X-VV also show almost no trend as the stand basal areas vary.

Table 3. The correlation coefficients between BlobeSAR data and forest stand parameters

	Basal Area	Stand density	Tree Height	dbh
C-HH	0.29	-0.44	0.93	0.93
C-VV	0.09	-0.42	0.90	0.91
X-HH	0.70	-0.49	0.97	0.97
X-VV	0.53	-0.29	0.87	0.87