Introduction

The polarization orientation angle (POA) refers to the angle of rotation of the polarization ellipse of the electromagnetic wave about the radar line of sight direction (Koeniguer 2014; Lee et al. 2018). The POA changes are induced by the surfaces with non-zero azimuth slopes and by the man-made targets which are not aligned in the radar azimuth direction (J. S. Lee et al. 2000). But the range slope also causes a shift in the POA and results in higher cross-polarization intensity in the HV and VH channels making the reflection asymmetrical (Gupta and Kumar 2016). Since the model-based polarimetric decomposition techniques were developed based on the assumption of reflection symmetry (Verma 2012), the shift in polarization orientation angle causes these decomposition techniques to produce wrong results which leads to the misinterpretation of the ground targets (Khati et al. 2013; Lee et al. 2011; Shukla and Kumar 2018).

The primary objective of this study is to analyse the effect of POA shift in the ground target characterization using polarimetric SAR (PolSAR) datasets. For this purpose, the UAVSAR L-band PolSAR datasets were used, and the model-based Yamaguchi 3-component (Cui et al. 2012) and eigenvector-based HA/A/Alpha decomposition (Cloude and Pottier 1996) techniques were implemented on it.

Study Area and Dataset

Study Area

The Fujiyama mountain of Japan was selected for this study. The optical image of 31st December 2013 from Google Earth is shown in Fig. 1.

Fig. 1
figure 1

Google earth image of 31st December 2013

Full size image

From the Google Earth image, it can be seen that the study area comprises of both man-made and natural targets. The Fujiyama volcano cone is barren without the presence of any vegetation. The area surrounding the Fujiyama volcano does not have any human activities due to the threat of the possible  volcanic eruptions and is filled with thick vegetation. Surrounding this vegetation there is the presence of agricultural fields and man-made structures. The study area comprising this kind of heterogeneous ground targets is best suitable for this study because of the possibility to get all kinds of scattering mechanisms. The hilly terrain of the study area also helps to analyse the effect of terrain slope on the POA shift.

Dataset

The airborne UAVSAR L-band quad-pol dataset (Rosen et al. 2006) acquired over Fujiyama mountain, Japan was considered for this study. The dataset was acquired in quad-pol mode and the details are shown in Table 1.

Table 1 Metadata of the datasets
Full size table

Methodology

The methodology adopted for this study is shown in Fig. 2. The polarization orientation angle (POA) estimation and compensation method using the circular polarization technique (Lee et al. 2000) was implemented for this study. The phase difference induced between the right circular transmitted–right circular received (\(S_{RR}\)) polarization and left circular transmitted–left circular received (\(S_{LL}\)) polarization (Izumi et al. 2017) due to the POA shift was utilized here to estimate the POA shift angle. The \(S_{RR}\) and \(S_{LL}\) polarizations were generated from the scattering matrix as follows:

$$S_{RR} = \frac{{S_{HH} - S_{VV} + i2S_{HV} }}{2}$$
(1)
$$S_{LL} = \frac{{S_{VV} - S_{HH} + i2S_{HV} }}{2}$$
(2)

The degree of polarization orientation angle (POA) shift was then estimated as follows (Lee et al. 2003):

$$\eta = \frac{{\left[ {{\text{Arg}}\left( {S_{RR} S_{LL}^{ *} } \right) + \pi } \right]}}{4}$$
(3)

\({\text{Arg}}\left( {S_{RR} S_{LL}^{ *} } \right)\) was estimated using Eqs. [(1), (2)] as (Lee et al. 2003):

$${\text{Arg}}\left( {S_{RR} S_{LL}^{*} } \right) = \tan^{ - 1} \left( {\frac{{ - 4\text{Re} \left( {\left( {S_{HH} - S_{VV} } \right)S_{HV}^{*} } \right)}}{{ - \left| {S_{HH} - S_{VV} } \right|^{2} + 4\left| {S_{HV} } \right|^{2} }}} \right)$$
(4)
$$\theta = \left\{ {\begin{array}{*{20}l} {\eta ,} \hfill & {{\text{if}}\; \eta \le \frac{\pi }{4}} \hfill \\ {\eta - \frac{\pi }{4},} \hfill & {{\text{if}}\; \eta > \frac{\pi }{4}} \hfill \\ \end{array} } \right\}$$
(5)

where \(\theta\) is the POA shift angle in degrees.

Fig. 2
figure 2

Flowchart of methodology

Full size image

After estimating the POA shift angle, the new coherency matrix (T3) after compensating for the POA shift was estimated as shown below (Lee et al. 2003):

$$T3^{{{\text{POA}}\; {\text{Compensated}}}} = \left[ { \left[ U \right]} \right]\left[ {T3} \right]\left[ {U^{T} } \right]$$
(6)

where

$$\left[ U \right] = \left[ {\begin{array}{*{20}c} 1 & 0 & 0 \\ 0 & {\cos 2\theta } & {\sin 2\theta } \\ 0 & { - \sin 2\theta } & {\cos 2\theta } \\ \end{array} } \right]$$

and \(T3^{{{\text{POA }}\;{\text{Compensated}}}}\) is the new POA compensated coherency matrix (T3).

The Yamaguchi 3-component model-based polarimetric decomposition and H/A/Alpha eigenvector-based polarimetric decomposition techniques were implemented with the original T3 matrix and the T3 matrix generated after POA shift compensation to analyse the effect of POA shift on the polarimetric decomposition.

Results

The polarimetric decomposition results before and after POA shift compensation were analysed in this section. Figure 3 shows the Yamaguchi's decomposition-based RGB composite generated from the coherency matrix (T3) before POA compensation. The red band represents the double-bounce scattering component, the green band represents volume scattering component and blue band represents the surface scattering component. The entire Fujiyama volcano cone is appearing in blue colour indicating the dominance of surface scattering behaviour. The vegetated areas are appearing in white colour indicating the presence of all the three types of scattering mechanisms. By comparing Fig. 3 with the Google Earth image of the same time is shown in Fig. 1, it can be observed that some of the non-vegetated agricultural fields and barren lands were showing volume scattering behaviour appearing in green colour which can be due to the POA shift in the dataset causing the Yamaguchi 3-component decomposition technique to produce wrong output.

Fig. 3
figure 3

RGB composite of Yamaguchi 3-component decomposition before POA shift compensation

Full size image

The POA angle shift was then estimated using the methodology explained in the previous section, and it was observed that there was a considerable amount of POA angle shift in the dataset. The POA shift image is shown in Fig. 4.

Fig. 4
figure 4

POA shift image generated with circular polarization technique

Full size image

From the POA shift angle, it can be seen that the slope of the volcano towards the north side is showing positive values of POA shift, while the slope of the volcano towards the south side is showing negative values of POA shift. This pattern indicates that the POA angle shift is strongly dependent on the slope of the terrain in the radar azimuth direction. The vegetated regions around the volcano are showing random values of positive and negative values of POA shift which can be due to the random orientation of the canopy structure. The flat terrain barren lands were showing positive values of POA shift. Overall there was a mean POA shift of − 4.236° in the dataset.

Figure 5 shows the RGB composite of Yamaguchi 3-component decomposition after POA shift compensation. By comparing with Fig. 3, it can be observed that there is a considerable reduction in volume scattering component indicated by the reduction in green colour. The regions which were green in colour before POA shift compensation turned to the red colour indicating the increase in volume scattering.

Fig. 5
figure 5

RGB composite of Yamaguchi 3-component decomposition after POA shift compensation

Full size image

Figure 6 is the subset of fig showing a non-vegetated agricultural field. Figure 6a is the Google Earth image of the area acquired on 31st December 2013, from the image it can be seen that there is no vegetation present in the agricultural field. Figure 6b shows the RGB composite of the Yamaguchi's 3-component decomposition model before POA shift compensation, it can be observed that the non-vegetated agricultural lands were appearing in a green colour representing volume scattering. Figure 6c shows the RGB composite of Yamaguchi 3-component decomposition after POA shift compensation, by analysing the image it can be observed that the volume scattering behaviour shown by the non-vegetated agricultural fields was considerably reduced, and those regions were now appearing in the red colour representing double-bounce scattering behaviour.

Fig. 6
figure 6

Comparison of Yamaguchi's 3-component decomposition-based RGB composite, a Google Earth image of 31st December 2013, b before POA shift compensation, c after POA shift compensation

Full size image

Figure 7 shows the graph of the volume scattering component of the same agricultural field shown in Fig. 6. The volume scattering magnitude of 70 pixels was considered for the plot, the black colour line indicates the volume scattering magnitude before POA shift compensation, and the green colour line indicates the volume scattering magnitude after POA shift compensation. By analysing the graph, it can be clearly understood that the volume scattering magnitude is considerably reduced after POA shift compensation. Even though the volume scattering magnitude is considerably reduced, the trend of both the lines remains the same.

Fig. 7
figure 7

Volume scattering component analysis

Full size image

Figure 8 shows the graph of the double-bounce scattering component of the same agricultural field shown in Fig. 6. The double-bounce scattering magnitude of 70 pixels was considered for the plot, the black colour line indicates the double-bounce scattering magnitude before POA shift compensation, and the red colour line indicates the double-bounce scattering magnitude after POA shift compensation. By analysing the graph, it can be clearly understood that the double-bounce scattering magnitude is considerably increased after POA shift compensation. The double-bounce scattering component can be due to the result of ploughing of the agricultural land as seen in the Google Earth image in Fig. 6a. The ploughing causes the land to become irregular which increases the probability of double-bounce scattering.

Fig. 8
figure 8

Double-bounce scattering component analysis

Full size image

Figure 9 shows the graph of the surface scattering component of the same agricultural field shown in Fig. 6. The surface scattering magnitude of 70 pixels was considered for the plot, the black colour line indicates the surface scattering magnitude before POA shift compensation, and the blue colour line indicates the surface scattering magnitude after POA shift compensation. By analysing the graph, it can be understood that there is no considerable change in the surface scattering behaviour before and after POA shift compensation. This is due to the insensitivity of the surface scattering component to the orientation angle of the electromagnetic wave known as the roll invariance property.

Fig. 9
figure 9

Surface scattering component analysis

Full size image

Figure 10 shows the alpha parameter analysis for the same agricultural field. The alpha value ranges from 0 to 90, the values close to 0 indicate surface scattering, the values close to 45 indicate volume scattering, and the values close to 90° indicate double-bounce scattering. Figure 10b shows the alpha image before POA shift compensation and Fig. 10c shows the alpha image after POA shift compensation. By analysing both the images, it can be observed that there is no change in the alpha image before and after POA compensation which indicates the roll invariance nature of the alpha parameter (Touzi 2007). The non-vegetated agricultural lands were showing alpha values higher than 60° indicating double-bounce scattering which validates the result obtained with Yamaguchi 3-component decomposition after POA compensation.

Fig. 10
figure 10

Alpha parameter analysis, a Google Earth image, b before POA correction, c after POA correction

Full size image

Conclusion

The POA shift induced in the UAVSAR dataset seriously affected the model-based decomposition techniques which resulted in the overestimation of volume scattering component and underestimation of double-bounce scattering component. So to retrieve exact scattering mechanisms from the PolSAR datasets, the POA shift compensation should be considered as an essential requirement before implementing any polarimetric decomposition techniques. Also, the models used for deriving the volume scattering and double-bounce scattering components in the model-based polarimetric decomposition techniques require to be modified to take in account of the effects caused by the POA shift in the scattering matrix.