Evaluating the effect of the incidence angle of ALOS-2 PALSAR-2 on detecting aquaculture facilities for sustainable use of coastal space and resources

Study area and aquaculture rafts

The selected study site was the inner part of Ago Bay, located in the central Japanese prefecture of Mie (Fig. 1). This area has been subject to the Nankai Trough Megathrust Earthquake and tsunamis, with historical accounts of earthquake and tsunami damages recorded along the Nankai Trough in 1498, 1361, 1096, 887, and 684 CE, coupled with the prediction of future occurrences (Fujino et al., 2018). Therefore, recording the aquaculture facilities’ type, number, and location in advance has been considered important. Notably, aquaculture rafts used here were constructed using Hinoki cypress (Chamaecyparis obtusa) wood poles and buoys, as shown in the presented example image observed by drone Mavic 2 Zoom (DJI, Nanshan, SZ, China) on August 29, 2022 (Fig. 2A), with results showing that they connected several 5.5 m × 6.4 m rafts, each of which has varying overall raft sizes (Sawada, 1965). These have been used to cultivate the Japanese pearl oyster, Pinctada fucata (Gould, 1850), and the oyster, Magallana gigas (Thunberg, 1793).

ALOS-2 PALSAR-2 data and data analysis

We subsequently analyzed ALOS-2 PALSAR-2 data obtained under the ultra-fine beam mode with HH polarization in 2021. The spatial resolution used was 3 m, and the pixel resolution was 2.5 m. To this end, we first studied the data differences between the satellite’s ascending (Asc.) and descending (Desc.) orbits, after which two datasets with the same incidence angles were used to confirm whether the data were peculiar or not at different seasons (Table 1). Next, we examined the differences due to incidence angles, using four datasets with different off-nadir angles under the Desc. orbit (Table 2). Since the incidence angle increased outward from a near to far range in a SAR image, and the difference in the incidence angle between the near and far range sides of the study area in the observation image was less than 0.1, we adopted the center incidence angle for the results and discussion. Then, the incidence angles of each ALOS-2 PALSAR-2 L1.1 data were collected using the ESA SNAP 8.0 software (flowchart is shown in Fig. 3). The results indicated in the applied datasets, as shown in Table 2. Afterward, we calculated the sigma zero ( ${\sigma }^0$) (also called the backscattering coefficient) for analysis, following conversion with the following equation:

(1) $${\sigma }^0=10\cdot lo{g}_{10}⟨D{N}^2⟩+CF$$where DN is the digital number of data and CF is the calibration factor, defined as a constant value of −83 dB (JAXA, 2017).

First, we mapped 50 aquaculture rafts (shown in red polygons in Fig. 1C), after which we randomly selected 50 other areas without aquaculture rafts (sea surface) through visual interpretation based on a Google Earth image collected on March 15, 2021. In second, we analyzed the ALOS-2 PALSAR-2 L2.1 data and collected DN using the ENVI 5.5 (Harris Geospatial, USA) software. The aquaculture rafts direction to satellite orbit direction was also collected using ENVI 5.5 software. Finally, we launched the drone Mavic Mini (DJI, Nanshan, SZ, China) on December 16, 2021, with results confirming that the 50 aquaculture rafts were still in place (a part shown in Fig. 2B).

Differences in sigma zero between the Asc. and Desc. orbits

The results under the Asc. orbit with an incidence angle of 40.1°, and the results under the Desc. orbit with an incidence angle of 44.0°, are shown in Figs. 4 and 5, and Table 3. The overlap was observed between the aquaculture raft and sea surface ${\sigma }^0$ range in the two data collected on November 24, 2021, and March 23, 2021. Result of the data from same orbit at different seasons showed median of ${\sigma }^0$ difference in aquaculture rafts or the sea surface within 0.5–1.8 dB. However, it was difficult to assess whether these data were peculiar or not only from this result.

Differences in sigma zero due to different incidence angles under the Desc. orbit

The results under the Desc. orbit with different incidence angles are shown in Figs. 6, 7 and Table 4. We observed a difference in the median of ${\sigma }^0$ between the incidence angles of the aquaculture rafts. Notably, while the incidence angle was between 33.8° to 45.1°, and the median of ${\sigma }^0$ was between −12.7 dB to −13.5 dB, the incidence angle was 52.1°, and the median of ${\sigma }^0$ was −15.1 dB. However, in two of the four datasets, overlap in the ${\sigma }^0$ range between the aquaculture raft and the sea surface was observed. Still, the ${\sigma }^0$ range was not overlapped under the incidence angles 33.8° and 45.1°, with distinct differences in the median of ${\sigma }^0$ being observed between the aquaculture rafts and the sea surface. While the incidence angle was between 33.8° to 45.1°, and the ${\sigma }^0$ difference was 6.0 dB to 8.1 dB, the incidence angle was 52.1°, and the ${\sigma }^0$ difference was 2.6 dB.

Differences in sigma zero due to the direction of the aquaculture rafts

The results of ${\sigma }^0$ due to the direction of the aquaculture rafts relative to the satellite orbit direction are plotted in Fig. 8. Notably, high values of ${\sigma }^0$ were found when the direction was close to 90°, with the three highest ${\sigma }^0$ of the aquaculture raft’s direction being 93.1°, 100.1°, and 102.1°, and the incidence angle being at 33.8°. Investigations also revealed that seven plotted results had ${\sigma }^0$ higher than −5 dB, with the incidence angles being four at 33.8°, two at 44.0°, one at 45.1°, and none at 52.1°. Moreover, the two plots at incidence angles of 33.8° and 45.1° were the same aquaculture raft, whose direction was 102.1°.

The effect of satellite orbit’s direction difference

Mahdavi, Amani & Maghsoudi (2019) analyzed the backscatter of a tree, ground vegetation, and water area using space-borne C-band SAR RADARSAT-2 full polarimetric images of Asc. and Desc. orbits whose incidence angles were both between 22.1° (lower edge) and 24.1° (higher edge). They reported differences in backscatter between the Asc. and Desc. orbits. The image observed in Asc. orbit was generally higher than the image acquired in the Desc. orbit, and the average difference was more than 1 dB, except for the water area.

In this study, however, the results differed from the previous study. Our investigations showed that the median of ${\sigma }^0$ under the Desc. orbit, aquaculture rafts and sea surfaces were maximum 2.2 dB and 2.6 dB higher than those under the Asc. orbit, respectively. However, it was difficult to conclude that the difference in ${\sigma }^0$ between the Asc. and Desc. orbits was caused only by orbital differences due to the incidence angle differed from 40.1° in the Asc. orbit and 44.0° in the Desc. orbit. Therefore, we tried to analyze data at similar incidence angles in the Asc. and Desc. orbits to clarify this gap. Still, there were no data on them on this study site.

The effect of aquaculture raft direction, satellite orbit direction, and Bragg resonance scattering appearance on aquaculture rafts

Since the ${\sigma }^0$ from aquaculture rafts are caused by a reflection from the wood poles or buoys that are to construct these aquaculture rafts and the sea surface inside them, differences in the ${\sigma }^0$ of aquaculture rafts may be attributed to these differences. The outstandingly high value of ${\sigma }^0$ for several aquaculture rafts compared to the median of ${\sigma }^0$ were found when the aquaculture raft direction was close to the perpendicular (90°) of the satellite orbit direction, and those incidence angle was 33.8°, 44.0°, and 45.1°. We consider this mechanism as Bragg scattering. In previous studies, Rosenqvist (1999) and Ouchi, Niiuchi & Mohri (1999) reported Bragg scattering from rice plants using JERS-1 L-band SAR images of mechanically planted rice fields. Their results showed strong radar backscattering if the planting orientation, incidence angle, and distance between the bunches of rice plants matched the Bragg resonance condition. The Bragg resonance condition was defined by Rosenqvist (1999), as shown in Eq. (2). Ouchi et al. (2006) later reported the mechanism of Bragg resonance scattering, which arises from the double-bounce scattering between water surfaces and the regularly spaced bunches of rice plants.

(2) $$d_{max}=n\gamma {\left(2sin{\theta }_i\right)}^{-1}$$where d is the plant spacing, $\gamma$ is the radar wavelength, and ${\theta }_i$ is the local incidence angle. In this equation, d could replace the distance of wood poles that make up the aquaculture raft, and $\gamma$ = 22.9 cm for ALOS-2 PALSAR-2. Furthermore, the incidence angle was 33.8°, 40.1°, 44.0°, 45.1°, and 52.1°. For example, when n = 3 in Eq. (2), d = 0.62 m, 0.53 m, 0.49 m, 0.48 m, and 0.44 m, respectively. In this case, the incidence angle of 33.8° was close to 0.64 m, the distance of wood poles. Since the aquaculture rafts were made by hand, there were slight differences in the distance of wood poles. In addition, it has an effect from the sea surface. Based on this finding, we propose that the roughness of the sea surface changed the reflection and effect to appearance of the Bragg resonance scattering on the aquaculture raft. In this study, the ${\sigma }^0$ of the aquaculture rafts were considered includes the effect of Bragg resonance scattering that means double-bounce scattering occurred between the aquaculture raft and the sea surface, resulting in a high value of ${\sigma }^0$ (Fig. 9). Therefore, Bragg resonance scattering was considered to make detecting the aquaculture rafts from the sea surface easier. Moreover, if the distance between wood poles is known, the theoretical incidence angle can be calculated to obtain the Bragg resonance scattering.

The effect of incidence angles on detecting aquaculture rafts

Skrunes et al. (2018) studied the backscatter of clean sea surfaces at different incidence angles using UAV-based SAR. They reported that the clean sea backscatter decreased as the incidence angle increased in HH polarization. Studies on sea ice using SAR have also reported the incidence angle dependence of backscatter. For example, Mahmud et al. (2018) used L-band SAR ALOS PALSAR and RADARSAT-2 and investigated the incidence angle dependence of backscatter over Arctic sea ice. They analyzed Arctic first-year ice and multiyear sea ice and reported mean ice type-specific incidence angle dependencies were −0.21 dB/1° and −0.30 dB/1°, respectively for ALOS, and −0.22 dB/1° and −0.16 dB/1°, respectively for RADARSAT-2. They concluded that it must normalize SAR images calculated by incidence angle dependence to improve sea ice classification. In this study, the aim is to detect aquaculture rafts daily and immediately after the disaster. Therefore, incidence angle normalization is not necessary, but it is necessary to recognize the ${\sigma }^0$ value depends on the incidence angle and the effects of the ground surface. Singha, Johansson & Doulgeris (2020) used L-band SAR ALOS-2 PALSAR-2 full polarimetric images with a range of incidence angles and during different environmental conditions for artificial neural network-based sea ice type classification. The highest incidence angle (39.05°) they used has values deviating from the otherwise consistent trend, including previous studies (Casey et al., 2016; Wakabayashi et al., 2004; Mahmud et al., 2018) with decreasing backscattering value with the incidence angle. This was considered since rough surfaces were more prominent at higher incidence angles due to different imaging geometries and different proportions of ice types in those scenes.

Alternatively, this study showed that although the median of ${\sigma }^0$ on the sea surface was almost the same (−20.7 dB and −20.6 dB), even when the incidence angles were 33.8° and 45.1° (11.3° difference), respectively, ${\sigma }^0$ did not decrease with an increasing incidence angle, as reported in previous studies. Furthermore, the median of ${\sigma }^0$ on the sea surface was −17.7 dB when the incidence angle was 52.1°, higher than the incidence angle of 33.8° and 45.1°. Therefore, we propose that this finding was due to the roughness of the sea surface, confirming the results of Singha, Johansson & Doulgeris (2020), which indicated that when the sea surface is rough, ${\sigma }^0$ indicates high value prominence at higher incidence angles, making it difficult to distinguish them from aquaculture rafts.

We observed that when the incidence angle was 44.0° on September 7, 2021, the ${\sigma }^0$ range of the aquaculture raft and sea surface was overlapped. However, when the incidence angle was 44.0° on December 28, 2021, the ${\sigma }^0$ range did not overlap. The incidence angle of 44.0° on September 7, 2021, was attributed to the roughness of sea surfaces, such as rain, wave, and other movements. Furthermore, the median of ${\sigma }^0$ on aquaculture raft at an incidence angle 52.1° was obviously lower than that at 33.8°–45.1°. Therefore, an incidence angle between 33.8° and 45.1° was considered optimum for detecting aquaculture rafts with high ${\sigma }^0$ values, minimizing the effects of sea surface roughness. However, since there are no data on incidence angles <33.8°, further data and analysis are still required.