The Potential of Sentinel-1 SAR Parameters in Monitoring Rice Paddy Phenological Stages in Gimhae, South Korea

Abstract

Synthetic Aperture Radar (SAR) at C-band is an ideal remote sensing system for crop monitoring owing to its short wavelength, which interacts with the upper parts of the crop canopy. This study evaluated the potential of dual polarimetric Sentinel-1 at C-band for monitoring rice phenology. Rice phenological variations occur in a short period. Hence, the short revisit time of Sentinel-1 SAR system can facilitate the tracking of short-term temporal morphological variations in rice crop growth. The sensitivity of SAR backscattering coefficients, backscattering ratio, and polarimetric decomposition parameters on rice phenological stages were investigated through a time-series analysis of 33 Sentinel-1 Single Look Complex images collected from 10th April to 25th October 2020 in Gimhae, South Korea. Based on the observed temporal variations in SAR parameters, we could identify and distinguish the phenological stages of the Gimhae rice growth cycle. The backscattering coefficient in VH polarisation and polarimetric decomposition parameters showed high sensitivity to rice growth. However, amongst SAR parameters estimated in this study, the VH backscattering coefficient realistically identifies all phenological stages, and its temporal variation patterns are preserved in both Sentinel-1A (S1A) and Sentinel-1B (S1B). Polarimetric decomposition parameters exhibited some offsets in successive acquisitions from S1A and S1B. Further studies with data collected from various incidence angles are crucial to determine the impact of different incidence angles on polarimetric decomposition parameters in rice paddy fields.

1. Introduction

As the world’s population has been rising, the demand for agricultural products has increased, triggering the need for reliable and precise crop cultivation practices for sufficient agricultural production to meet the demand. Most importantly, rice is a staple food crop consumed by half of the world’s population. Rice crop monitoring is economically important for optimal rice production. Furthermore, it is essential to ensure effective agricultural practices such as sowing, applying fertilisers, checking crop health and growth status, predicting production, and thereby taking necessary actions to optimise yield (Lopez-Sanchez et al., 2012). Conventional crop monitoring in the field provides accurate measurements of crop biophysical parameters and yield estimates. However, it is time-consuming, labour-intensive, and expensive. The remote sensing technique is an alternative for monitoring crop vegetation fields on a large scale within a reasonable time. The use of optical satellite imagery from multispectral and hyperspectral sensors in crop monitoring is limited because of their inability to capture images in all weather conditions and during the night (Toan et al., 1997).

On the contrary, active microwave sensors, such as Synthetic Aperture Radar (SAR), have observational capabilities regardless of weather conditions and time. SAR instrument transmit electromagnetic pulses in the microwave range to the Earth’s surface and receive the echoes of the backscattered signal from targets in its line of sight. The radar backscattering intensity provides useful information about the geometric and physical characteristics of the targets (Inoue et al., 2002). For crop fields, SAR sensors are sensitive to large crop structural variables, such as height, shape, leaves, and stem (Soria-Ruiz et al., 2007). Microwave remote sensing has been widely applied in monitoring large vegetation fields. It has demonstrated high performance in tracking temporal and seasonal changes during the crop growth period, in crop height and biomass estimation, and in mapping the extent of vegetation fields (Ndikumana et al., 2018; Nguyen and Wagner, 2017; Xie et al., 2021; Yuzugullu et al., 2017). Previous studies have found a strong correlation between radar backscattering intensity and biophysical features of rice, such as plant height, leaf area index (LAI), and biomass (Liu et al., 2016; Ndikumana et al., 2018; Zhang et al., 2014). As the rice plant grows, significant variations in SAR parameters are expected to emerge from the changes in the structure, geometry, and dielectric properties (soil and plant water content) of the rice paddy (Wang et al., 2005).

Besides the backscattering coefficients, SAR polarimetric decomposition parameters provide useful information about the scattering medium and have shown significant correlation with crop biophysical parameters such as LAI (Jiao et al., 2014), crop height, and dry biomass (Betbeder et al., 2016). Previous studies have identified essential phonological stages of wheat, barley, and canola based on the temporal variations of polarimetric decomposition parameters during the crop growth cycle (Canisius et al., 2018; Harfenmeister et al., 2021). Polarimetric SAR parameters depend also on the incidence angle, frequency, and polarisation of the transmitted microwave signal (Jiao et al., 2014). Therefore, it is important to choose the appropriate sensor configuration (frequency and polarisation) for a particular study.

The frequency of the SAR signal mainly governs the signal penetration in the target, i.e., the longer the wavelength (lower frequency), the deeper the penetration, and vice versa. Previous studies have proven that the relatively shorter wavelength (higher frequency) C-band (8-15 cm) SAR data are suitable for rice crop monitoring as they mainly interact with the upper part of the canopy segment (Liu et al., 2016; Ndikumana et al., 2018; Zhang et al., 2014). Combining cross-polarisation and co-polarisation radar backscattering provides additional information on the target’s structural characteristics and scattering behaviour (Nasirzadehdizaji et al., 2019). Given the aforementioned sensor configurations, Sentinel-1 is a potential SAR data set for rice crop monitoring because of its C-band high frequency, VH and VV polarisation channels, spatial resolution, and six-day short revisit time owing to the constellation of two satellites, Sentinel-1A (S1A) and Sentinel-1B (S1B).

The short revisit time of Sentinel-1 can allow the detection of instant morphological variations in crop growth, as observable phonological changes take a short time (Lopez-Sanchez et al., 2012). Multiple studies have successfully applied Sentinel-1 data for crop monitoring (Nguyen and Wagner, 2017; Veloso et al., 2017; Ndikumana et al., 2018; Van Tricht et al., 2018; Sharifi and Hosseingholizadeh, 2020).

To date, the previous studies have mainly focused on the application of Sentinel-1 SAR backscattering intensities in rice crop monitoring. This study not only examined the potential of Sentinel-1 backscattering coefficients in rice crop monitoring, it also investigated the sensitivity of Sentinel-1 polarimetric decomposition parameters on rice growth; additionally, their implications in monitoring rice paddy phenology have been discussed. A time-series analysis of SAR polarimetric parameters was conducted using 33 Sentinel-1 Single Look Complex (SLC) Interferometric Wide Swath (IW) images acquired in both the descending and ascending pass directions at slightly different incidence angles. Those images were acquired during the entire rice-growing season in Gimhae, South Korea.

2. Materials and Methods

1) Test Site

The Gimhae Plain lies in the southeast of the Korean Peninsula, close to the Busan city. Rice is a major food crop cultivated in Gimhae. The study area terrain is mostly flat with a very gentle slope. The weather of the area is characterised by the subtropical monsoon climate. In Gimhae, rice is grown in paddy fields, and sowing is mostly done in the spring season in late April or early May, while the harvest period is in mid or late October. Rice cultivation in Gimhae is an important source of rice for the populations of Busan and South Gyeongsang provinces. Polygons of Gimhae rice paddy parcels were obtained from the National Geographic Information Institute (NGII) of South Korea, and 120 parcels were selected for analysis in this study. The test area location and the shapefile of the selected rice paddy parcels are shown in Fig. 1.

Fig. 1. Google satellite image overlay by the shapefile of rice paddy sample parcels in Gimhae rice paddy field. South Korea map is on the right top corner. The red rectangle represents the test site.

2) Sentinel-1 data sets

A series of 33 dual-polarisation (VV and VH) Sentinel-1 SLC IW images acquired over Gimhae with six-day interval from 10th April to 25 October 2020 was used in this study.

The Sentinel-1 mission is a constellation of two SAR satellites, S1A and S1B, equipped with C-band imaging radar sensors. The Sentinel-1 instrument acquires data in four observational modes of different coverages and resolutions: IW, extra wide swath (EW), strip-map (SM), and wave (WV) modes. The IW is the primary operational imaging mode for Sentinel-1 satellites. It uses Terrain Observation by Progressive Scans (TOPSAR) to cover a wide swath width of250 km with 50 m × 20 m ground geometric resolution (Torres et al., 2012). The Sentinel-1 IW mode enables the acquisition of long time-series data with equidistant time intervals: 12 days for each satellite and 6 days for the constellation of both satellites (Geudtner et al., 2014). Sentinel-1 data were freely accessible from the Copernicus Open Access Hub (Open Access Hub, n.d.). The detailed specifications and satellite sensing dates of the Sentinel-1 dataset are listed in Table 1.

Table 1. Specification of Sentinel-1 SLC IW dual-polarisation (VH and VV) data used in this study. “De” and “Asc” stand for descending and ascending pass directions, respectively. The incidence angle is in degrees

3) Rice Phenology

This section briefly describes the phenological stages of the rice growth cycle. Different rice phenological stages are associated with different SAR backscattering mechanisms. Therefore, it is essential to understand the morphological variations that occur during the rice growth cycle.

According to the general BBCH scale (from Biologische Bundesanstalt, Bundessortenamt und CHemische Industrie), the rice growth cycle lasts for 110-150 days, depending on the rice type, weather conditions, soil type, and cultivation practices. It comprises three main phases: vegetative, reproductive, maturation, and ripening (Liu et al., 2016; Lopez- Sanchez et al., 2012). Each stage corresponds to the development or growth of particular biophysical features of the rice paddy (see Table 2). The vegetative phase lasts for almost half of the cycle, approximately 46-100 days, and comprises five stages: germination, leaf development, tillering, stem elongation, and booting. An increase in plant height by stem elongation, an increase in the number of tillers, and leaf development are the typical changes observed during this phase. In the reproductive phase (19-25 days), a slight or insignificant increase in plant height results from the formation of headings where panicles emerge from sheaths and the development of reproductive organs and flowers at the top part of the plant.

Table 2. BBCH Scale of the rice growth cycle

The maturation phase lasts for approximately 25-35 days and defines the last phase of the cycle, during which grains develop and change to a milky colour, leaf and stem moisture content decrease, plants dry, and the number of leaves decreases.

3) Sentinel-1 SLC IW Polarimetric SAR Data Processing

Sentinel-1 images were processed using Sentinel Application Platform (SNAP), an open-source software of the European Space Agency (ESA) (SNAP –STEP, n.d.), as shown in the workflow in Fig. 2. Data processing is subdivided into three main parts: the generation of backscattering coefficients, formation of dual polarimetric decomposition parameters, and extraction of SAR parameters of rice paddy parcels. Images from S1A and S1B were separately processed owing to the different orbit acquisitions. The preprocessed S1A and S1B data were combined at the final step of the data analysis. The SLC images consist of three sub-swath images that have a series of bursts processed as separate SLC images (Torres et al., 2012). Thus, each image was split to generate the sub-swath image that includes our test site. A precise orbit file was applied to all images to update the orbit state vectors in the abstract metadata, and images were then calibrated to generate pixel values directly corresponding to the radar backscatter of the target. Backscattering coefficients were retrieved by sigma naught calibration, followed by debursting and geocoding by the Doppler range radar correction with the SRTM (Shuttle Radar Topography Mission) 1Sec HGT DEM. A multi-temporal stack was generated, and a multi-temporal refined Lee speckle filter was subsequently applied to reduce the inherent speckle noise of the SAR images.

Fig. 2. Sentinel-1 Data Processing Workflow. Processes indicated in brown boxes were performed in the SNAP Software.

For polarimetric decomposition, calibrated images were saved in a complex format to allow the generation of a polarimetric matrix. The bursts of each calibrated image were merged into single SLC images (debursting). For each complex format-calibrated image, a 2 × 2 polarimetric covariance C₂ matrix was formed as:

\( \begin{aligned} \mathrm{C}_{2}=&\left[\begin{array}{ll} \mathrm{C}_{11} & \mathrm{C}_{12} \\ \mathrm{C}_{21} & \mathrm{C}_{22} \end{array}\right]=\left[\begin{array}{lll} <\mathrm{S}_{\mathrm{VV}} & \mathrm{S}_{\mathrm{v}}^{*}> & <\mathrm{S}_{\mathrm{Vv}} & \mathrm{S}_{\mathrm{vH}}^{*}> \\ <\mathrm{S}_{\mathrm{VH}} & \mathrm{S}_{\mathrm{Vv}}^{>}> & <\mathrm{S}_{\mathrm{VH}} & \mathrm{S}_{\mathrm{VH}}^{>} \end{array}\right] \\ & \stackrel{\text { Polarimetric Decomposition }}{\longrightarrow} \mathrm{H}-\mathrm{A}-\alpha \end{aligned} \)

where C₁₁= VV backscattering coefficient, C₂₂= VH backscattering coefficient, and C₂₁and C₂₂are complex numbers. The covariance matrix describes the mean polarimetric information for a set of distributed targets and is formed by the second-order statistics of the scattering matrix, which provides information about the features of a deterministic target (Xie et al., 2021). A refined Lee speckle filter was applied to the polarimetric covariance. Polarimetric features such as entropy(H), alpha(α), and anisotropy(A) were extracted from the despeckled covariance matrices using Claude and Pottier’s dual polarimetric decomposition (Cloude and Pottier, 1996). The polarimetric decomposition images were geocoded using the SRTM with 1 arc second resolution, followed by the creation of a time series stack.

Ultimately, the pre-processed multi-temporal stacks were masked using the sampled rice paddy parcel shapefile. The sample rice paddy parcel vector data were extracted from the Gimhae land parcel data obtained from the NGII using the Quantum GIS (QGIS) software. The data analysis was carried out on the mean pixel values of 120 rice paddy parcels.

5. Results

This study assessed the sensitivity of Sentinel-1 dual-polarimetric SAR data to rice paddy growth stages in Gimhae, South Korea. Data analysis was carried out for 120 rice paddy parcels covering an area of approximately 1 km². The sample parcels were selected from the shapefile of the Gimhae rice paddy field obtained from the NGII (see Fig. 1). The temporal variations of the mentioned SAR parameters were reviewed concerning different rice phenological stages to understand their correlation with rice growth stages and the implication of using Sentinel-1 SAR in monitoring the rice crop.

Visual interpretation of multi-temporal images of VH and VV backscattering power showed the sensitivity of C-band Sentinel-1 SAR to rice paddy phenological stages in our test site (see Fig. 3 and Fig. 4). In each image, the backscattering power is indicated by the contrast from dark grey to white. The bright pixels in all the images are greenhouses. By examining the images through the sequence of their acquisition dates, one can note that the backscattering intensity of rice paddy parcels varies throughout the growth cycle. There was a lower backscattering intensity in May and June, but it increased in July and August. The images corresponding to 1st September are very bright compared to the images from previous months; it implies increase in backscattering power. Meanwhile, a decline in backscattering power was observed in October.

Fig. 3. Multi-temporal VH backscattering power. The red polygon represents the area of rice paddy sample parcels in Gimhae as shown in Fig. 1.

Fig. 4. Multi-temporal VV backscattering power. The red polygon represents the area of rice paddy sample parcels in Gimhae as shown in Fig. 1.

In this study, precise evaluation of the sensitivity of Sentinel-1 SAR parameters to rice growth involved a time-series analysis of the mean values and standard deviations of SAR parameters extracted from 120 rice paddy parcels. To investigate time-series analysis with a higher temporal baseline, SAR parameters measured from S1A and S1B images were combined in one plot but marked with different colours. The temporal variation of measured SAR parameters showed five distinct periods that corresponds to different scattering mechanisms (see Fig. 5 and Fig. 6). The periods P-1, P-2, P-3, P-4, and P-5 are outlined for better presentation of results.

Fig. 5. Time-Series plots of backscattering coefficients and the backscattering ratio. Grey vertical bars separate periods with significant changes. Solid black vertical bars represent standard deviations. P-1, P-2, P-3, P-4, and P-5 represents distinct periods identified based on temporal variation patterns.

Fig. 6. Temporal variations of polarimetric decomposition parameters: Entropy, anisotropy, and alpha throughout the rice growing season in Gimhae South, Korea, in the year 2020. Grey vertical bars separate periods with significant changes. Solid black vertical bars represent standard deviations. P-1, P-2, P-3, P-4, and P-5 represents distinct periods identified based on temporal variation patterns.

1) Backscattering Coefficients and Backscattering Ratio

The temporal observations of cross-polarisation VH, co-polarisation VV backscattering coefficients, and backscattering coefficient ratio are shown in Fig. 5. The time-series profiles of both the VV and VH intensities exhibit almost similar patterns. The backscattering parameters measured from S1A and S1B images were generally consistent. However, some offsets were detected before 9th June and after 1st September.

The backscattering intensity of VH is characterized by an increasing trend from -19.5 to -15.0 dB in P-1 and lower values around -22.0 dB in P-2. At the beginning of P-3, VH increases from -21.0 to -18.0 dB and slightly decreases from -17.0 and to -16.0 dB. A remarkable increase in VH backscattering intensity was observed on 1st September. Afterward, VH exhibited a gradual and regular decreasing trend from -16.0 to -20.0 dB during P-4 and P-5. Standard deviations in the VH backscattering intensity are higher in the early stages and slightly decrease throughout the rest of the cycle. However, a high standard deviation (3.5) is observed on 13th October.

The backscattering response in the co-polarised channel VV also initiates with an increasing trend, with values higher than -10.0 dB observed in P-1. The backscattering intensity of VV is also lower in P-2, with values less than -15.0 dB. Similar to the VH backscattering coefficient, the VV backscattering coefficient also significantly increases at the beginning of P-3. It increased from -15.0 to approximately -11.0 dB on 21st June but exhibited an opposite slightly decreasing trend from -11.0 to -12.6 dB between 21st June and August 26th in P-3. In the late periods P-4 and P-5, VV backscattering values measured from S1B images were higher than those of S1A. The backscattering values of VV also exhibited a significant increment on 1st September, as did the VH values. Moreover, a high standard deviation of 3.7 in the VV backscattering intensity was observed in the late stage, on 13th October.

The polarisation intensity ratio VH/VV showed sensitivity during P-3, as indicated by an increasing trend from 0.16 to around 0.30. The combination of S1A and S1B on the VH/VV intensity ratio exhibited insignificant patterns for P-1, P-4, and P-5. Higher values of the standard deviation in the polarisation ratio were observed in all periods.

2) Polarimetric Decomposition Parameters

This section describes the temporal variation of the polarimetric decomposition parameters, namely entropy, alpha, and anisotropy, throughout the rice growth cycle. Entropy and alpha measured from S1B images were generally higher than those measured from S1A images, while anisotropy showed the opposite pattern. Despite the differences in the values of polarimetric decomposition parameters from S1A and S1B images, their temporal variations followed almost identical patterns (see Fig. 6). It should be also noted that the polarimetric decomposition parameters of S1B have lower standard deviations than those of S1A. The entropy and alpha angle parameters had a positive correlation with the rice growth stages. In contrast, anisotropy had a negative correlation with the rice phenological stages in Gimhae.

Entropy values measured from S1B exhibited an increasing trend and reached relatively high values of approximately 0.78 in P-2, followed by a decreasing trend to lower values of around 0.60.

During P-1 and P-2, the S1A entropy slightly changed and was within the range of 0.58-0.62. In P-3, the entropy from both S1A and S1B increased to its maximum and dropped from 1st September.

The anisotropy from S1A was higher than that from S1B. The pattern started with higher values around 0.7 for both S1A and S1B. For S1A, the anisotropy reduced to 0.51 on 28th May in P-2 and increased to approximately 0.72 at the end of P-2. On the other hand, no significant variation was detected in the anisotropy from S1B in P-1 and P-2 acquisitions. The anisotropy decreased down to approximately 0.53 during P-3 and finally increased to approximately 0.6 until 13th October.

The alpha from S1A experienced an increasing trend to reach relatively high values around 26° and subsequently decreased to approximately 16° from 9th to 27th June. As the entropy, alpha from S1A and SIB also tended to increase to its maximum around 28° during P-3 and decreased after 1st September to values around 22°.

4. Discussion

This study conducted a time-series analysis of Sentinel-1 SAR polarimetric parameters during the rice growing season in Gimhae, South Korea. Our results demonstrated their sensitivity to phenological stages of the rice crop. We also investigated rice phenological stages using the temporal variation profiles of VH and VV backscattering coefficients, VH/VV ratio, and polarimetric decomposition parameters presented in Fig. 5 and Fig. 6. Based on the temporal variation profiles of measured SAR features throughout the rice growing season, we identified five distinct periods. The backscattering intensity observed in each period resulted from different scattering mechanisms that arise from the change in the structure, geometry and dielectric properties of the rice paddy during the rice growth cycle. Hence, we distinguished the rice phenological stages (see Table 2) based on the interpretation of the backscattering intensity patterns in each period.

Before rice plantation, the surface roughness and moisture content of rice paddies mainly affect the backscattering power. The rougher the surface, the stronger the backscattering power, and the higher the moisture content, the lower the backscattering power. After rice plantation, the backscattering signal is influenced largely by the change in the rice paddy canopy caused by the growth of rice paddy biophysical attributes and the change in the dielectric properties of underlying ground surface.

The higher VH and VV backscattering intensity observed in P-1 from 22nd April to 10th May can be interpreted to result from reflections from bare rough surfaces created by farmers’ practices of land preparation before rice planting. The lower VH and VV backscattering power in P-2 indicates the planting period where the rice paddy is flooded and completely submerged in water. Therefore, the radar reflections originate from a smooth water surface that reflects the incident radar pulses away from the radar receiver, resulting in lower backscattering values.

The increase in VH and VV intensities that emerges from 9th June to 9th July can be inferred to result from the germination and leaf development, and the drying of the underlying ground surface of the rice paddy, causing double-bounce reflections from the interaction between the emerging rice plants and the underlying ground surface. This period can be associated with the early vegetative phase. From 9th July to the end of P-3, only VH backscattering power shows an increasing trend as expected. For instance, an increase in backscattering power is expected to result from the growth of leaves, increase in the number of tillers, and stem elongation. In contrast, VV backscattering power showed an opposite trend. Therefore, P-3 can be attributed to the vegetative phase and it seems to have lasted for approximately 76 days from 9th June to 26th August. The decreasing trend in VH backscattering intensity observed in P-4 and P-5 might have been driven by the reduction in the homogeneity of the crop canopy by plant dryness in the reproductive and maturation stages. Although the VV backscattering power also showed a decreasing trend in P-4 and P-5, its values measured from S1B are higher than those from S1A during P-4 and P-5.

According to VH backscattering behaviour found in P-3, P-4, and P-5, these three successive periods correspond to the vegetative, reproduction and maturation stages respectively. The backscattering power in the VH polarization channel showed high sensitivity to rice growth in all phonological stages. Meanwhile, VV backscattering indicates only the early vegetative phase and showed a slight decreasing pattern during P-3 from 9th July to 26th August, where an increasing trend was expected. During P-4 and P-5, VV backscattering intensity measured from S1B are higher than those measured from S1A resulting in uneven patterns in the variation of VV backscattering coefficients for consecutive S1A and S1B acquisitions.

The standard deviation of backscattering coefficients is also sensitive to the rice phonological stages. The heterogeneity in the surface roughness before rice plantation resulted in larger standard deviations during P-1. The decrease in standard deviation in the course of the rice paddy growth cycle implies that as the plants grow, the rice crop canopy becomes more homogeneous, leading to the nearly uniform scattering mechanisms. The higher standard deviations of 3.5 dB for VH and 3.7 dB for VV observed on 13th October can potentially indicate the harvest period, since in practice the harvest is not performed on the same day for all parcels, the harvested parcels exhibit higher backscattering power due to reflections from rough surfaces as compared to non-harvested parcels with relatively lower backscattering power. The polarisation VH/VV is seemingly sensitive during the vegetative period, with a gradually and slightly increasing trend.

Entropy, anisotropy, and alpha parameters also showed sensitivity to the rice phenological stages in Gimhae. However, the observed temporal variations of polarimetric decomposition parameters measured from Sentinel-1 data collected at different incidence angles present offsets for consecutive acquisitions of different pass directions, which seemingly stem from the influence of different viewing directions and acquisition times of S1A and S1B. Although polarimetric decomposition parameters also showed the potential to detect and monitor rice paddy phenological stages, the temporal variation of polarimetric decomposition parameters exhibit small offsets for successive acquisitions of S1A and S1B images. The offsets might have been driven by the slightly different incidence angles of S1A (41.4°) and S1B (37.4°) images and their different acquisition time.

5. Conclusion

The sensitivity of Sentinel-1 SAR backscattering coefficients and dual polarimetric decomposition parameters to rice paddy phenological stages was studied in the Gimhae rice paddy field in South Korea. All the polarimetric SAR parameters showed sensitivity to rice phenological stages, except for VV backscattering power. The VH backscattering power can be used for identification of almost all rice phenological stages. Moreover, the backscattering response in VH polarisation is less affected by the difference in incidence angles compared to other parameters. Our result show that the polarisation intensity ratio is sensitive to the vegetative phase only. The offsets between S1A and S1B emerge in the polarimetric decomposition parameters, but the temporal variation patterns are preserved for both S1A and S1B. Based on the observations from this study, the combination of polarimetric decomposition parameters measured from images acquired at different incidence angles in monitoring rice phenological stages might be misleading. Therefore, further studies with the use of data acquired from various incidence angles are required in order to identify the impact of different incidence angles on polarimetric decomposition parameters in rice paddy fields.