Developing a New Method to Identify Flowering Dynamics of Rapeseed Using Landsat 8 and Sentinel-1/2

: Identifying the rapeseed ( Brassica napus L.) ﬂowering dates are important for planting area estimation, growth monitoring, and yield estimation. However, there is currently a lack of data on rapeseed ﬂowering dates at the parcel scale. In this study, a new spectral index (Normalized Rapeseed Flowering Index, NRFI) is proposed to detect rapeseed ﬂowering dates from time series data generated from Landsat 8 OLI and Sentinel-2 sensors. This study also analyzed the feasibility of using the backscattering coefﬁcients (VV, VH, and VV/VH) of Sentinel-1 to detect the ﬂowering dates of rapeseed at the parcel scale. Based on the spectral and polarization characteristics of 718 rapeseed parcels collected in 2018, we developed a method to automatically identify peak ﬂowering dates by the local maximum of NRFI series and the local minimum of VH and VV, along with the maximum of VV/VH. The results show that most of the peak ﬂowering dates derived from Sentinel-1 and Sentinel-2 can be conﬁrmed by the in-situ phenological observations at the Deutscher Wetterdienst (DWD) stations in Germany. The NRFI outperforms the Normalized Difference Yellow Index (NDYI) in identifying the peak ﬂowering dates from Landsat 8. The derived medians of peak ﬂowering dates by NRFI, NDYI (Sentinel-2), and VH are similar, while a systematic delay is observed by NDYI (Landsat 8). The method with the spectrum and backscattering coefﬁcients will be a potential tool to identify crop ﬂowering dynamics and map crop planting area.


Introduction
Rapeseed (Brassica napus L.) is a staple oil crop used for cooking, high-quality animal feed, etc. [1]. It is cultivated widely in the world (e.g., Europe, Canada, China, India, and Australia) [2][3][4][5]. The global rapeseed production has grown rapidly in recent years [6]. The timing and intensity of flowering are closely related to the final yield of rapeseed [2,[7][8][9] and highly sensitive to temperature [2,10], as other crops [11][12][13]. For instance, sclerotinia stem rot can cause a 50% yield loss when the rapeseed is in bloom [14]. Additionally, the accurate flowering dates are key information to map rapeseed planting areas [15,16]. The peak flowering period is highly informative for monitoring growth status and predicting yield [2,17]. Therefore, it is of vital importance to accurately identify the flowering dates of rapeseed. Given the high spatial heterogeneities of cultivars, farmers' management practices, climate, and soil conditions, it is still challenging to detect the peak flowering dates of rapeseed timely and soundly [18].
Remote sensing data (e.g., optical and radar images) have become a powerful tool to derive biophysical parameters including crop phenological information in the last decades [18][19][20][21][22][23]. Among other things, optical data have prevailed to analyze the relationship between the photosynthetic and properties of crops, mostly via the conventional Normalized Difference Vegetation Index (NDVI) and the Enhanced Vegetation Index (EVI) when an unprecedented amount of free satellite data become available [24,25].
ogische Bundesantalt, Bundessortenamt, and Chemische Industrie) scale [40]. For rapeseed, the start (i.e., BBCH61) and end (i.e., BBCH69) of flowering dates are available, while the peak flowering dates (i.e., BBCH65) are not. Each record associated with a DWD station is considered to represent a crop within a maximum distance of five kilometers from its location [2,40]. A total of 268 rapeseed observation sites with complete records were obtained from the DWD phenology database in 2018. We analyzed the flowering start dates, end dates, and flowering period recorded at all these stations ( Figure 1). The flowering of rapeseed starts usually around the 110th day and ends mainly around the 135th day of the year in Germany. The flowering period lasts for mostly 20~30 days-a considerable portion of the growing season [18]. Among these observation stations, 22 stations were selected for further analyses (Figure 2a) because five L8 images and five S2 images were available from April to June, which covered all these 22 stations.

Identifying the Rapeseed Parcels
Compared with other land-use types, rapeseed has unique spectral characteristics during the flowering period. It appears green-yellow or yellow in the image of a combination of red, green, and blue bands. Thus, the visual interpretation of rapeseed parcels is based on the specific spectral feature [3,32,35,41,42]. Rapeseed flowering dates vary by location due to differences in climate, cultivars, and agricultural management [2] ( Figure  1). We extracted 718 parcels from a region (51°17′00″-50°56′43"N, 14°34′13″N-15°8′10″E) where the image quality of Sentinel-1/2 was good during the rapeseed growing season, and rapeseed was widely cultivated as a winter crop (Figure 2b,c). The visual interpretation was based on high-resolution Google earth images and Sentinel-2 images.

Identifying the Rapeseed Parcels
Compared with other land-use types, rapeseed has unique spectral characteristics during the flowering period. It appears green-yellow or yellow in the image of a combination of red, green, and blue bands. Thus, the visual interpretation of rapeseed parcels is based on the specific spectral feature [3,32,35,41,42]. Rapeseed flowering dates vary by location due to differences in climate, cultivars, and agricultural management [2] (Figure 1). We extracted 718 parcels from a region (51 • 17 00 -50 • 56 43 N, 14 • 34 13 N-15 • 8 10 E) where the image quality of Sentinel-1/2 was good during the rapeseed growing season, and rapeseed was widely cultivated as a winter crop (Figure 2b,c). The visual interpretation was based on high-resolution Google earth images and Sentinel-2 images.

Optical Satellite
All the available optical satellite data from Landsat 8 OLI and Sentinel-2 sensors at the Google Earth Engine (GEE) platform in 2018 were used in this study. The spatial resolutions of L8 true-color images and S2 images were 30 and 10 m, respectively. The S2 top-of-atmosphere reflectance archived in the GEE was processed by radiometric and geometric corrections including orthorectification and spatial registration on a global reference system with sub-pixel accuracy [43][44][45]. The GEE asset addresses of L8 and S2 are LANDSAT/LT08/C01/T1_SR and COPNERICUS/S2, respectively. We used a cloud-score algorithm to remove the clouds, cloud shadows, and snow/ice pixels [2,46].

Optical Satellite
All the available optical satellite data from Landsat 8 OLI and Sentinel-2 sensors at the Google Earth Engine (GEE) platform in 2018 were used in this study. The spatial resolutions of L8 true-color images and S2 images were 30 and 10 m, respectively. The S2 top-of-atmosphere reflectance archived in the GEE was processed by radiometric and geometric corrections including orthorectification and spatial registration on a global reference system with sub-pixel accuracy [43][44][45]. The GEE asset addresses of L8 and S2 are LANDSAT/LT08/C01/T1_SR and COPNERICUS/S2, respectively. We used a cloud-score algorithm to remove the clouds, cloud shadows, and snow/ice pixels [2,46].

SAR Satellite
We collected the Sentinel-1A/B C-band Synthetic Aperture Radar (SAR) data. Sentinel-1A/B provides a series of SAR images. The period of S1 data ranges from 2 to 12 days due to the differences in data types and regions all around the world [47]. For a denser time series, the ascending and descending orbit observations were combined [2,47]. The temporal resolution of S1 data in 2018 was two days, with a high spatial resolution (10× 10 m). The interferometric Wide Swath (IW) instrument mode with VV and VH was also used in the study [47]. The multi-temporal speckle filter at the GEE platform was applied to remove the noise of the backscatter S1 images [48]. Finally, we analyzed the temporal backscatter values of VV, VH, and VV/VH polarization over the rapeseed field during different phenological stages of rapeseed.

SAR Satellite
We collected the Sentinel-1A/B C-band Synthetic Aperture Radar (SAR) data. Sentinel-1A/B provides a series of SAR images. The period of S1 data ranges from 2 to 12 days due to the differences in data types and regions all around the world [47]. For a denser time series, the ascending and descending orbit observations were combined [2,47]. The temporal resolution of S1 data in 2018 was two days, with a high spatial resolution (10× 10 m). The interferometric Wide Swath (IW) instrument mode with VV and VH was also used in the study [47]. The multi-temporal speckle filter at the GEE platform was applied to remove the noise of the backscatter S1 images [48]. Finally, we analyzed the temporal backscatter values of VV, VH, and VV/VH polarization over the rapeseed field during different phenological stages of rapeseed.

Developing a New Index-NRFI to Catch Flowering Dynamics
Reflectance values of green and red bands change considerably during flowering stages of rapeseed according to previous studies [18,38]. The large changes in the two bands are useful to catch their dynamics. We believe the green/red bands from L8 and S2 provide valuable information to develop the NRFI. Most previous studies on rapeseed phenology substantiated the values of visible light bands [2,3,10,14,15,17,18,34,38,42,49], very few have focused on the rapeseed phenology characteristics of the short-wave infrared band (SWIR). The SWIR is less influenced by the atmosphere and more sensitive to water content in soil or plant [50]. Considering the potential different responses between green/red and SWIR2 bands to the flowering processes of rapeseed, this paper constructed a novel index to capture the flowering intensity of rapeseed-Normalized Difference Rapeseed Flower Index (NRFIr/NRFIg)-see Equations (1) and (2). The NRFIr and NRFIg are based on one visible band (green/red) and one invisible band (SWIR2).

Detecting Peak Flowering Stages
The peak flowering dates of rapeseed were identified based on a time series of remote sensing data. This study developed an approach to calculate the extrema of NRFI (Normalized Difference Rapeseed Flowering Index), NDYI, and backscatter values. Different smoothing methods could lead to large uncertainty in the retrieval of phenology [2,51]. The Savitzky-Golay (S-G) filter were used to smooth the time series of VV, VH, and VV/VH for each rapeseed parcel [52,53]. However, we didn't apply the S-G filter to smooth L8 and S2 time series due to their coarser temporal resolutions, and the filtering algorithm could ignore local features and cause large uncertainties [52,53]. The peak flowering period was then obtained through the local maximum of NDYI, NRFI, and VV/VH along with the local minimum of VV and VH for all parcels during the expected flowering period (from 21 April to 30 May). The window size of the S-G filter was three and it was implemented using the TIMESAT 3.1 software. For the specific implementation process, please refer to reference [54].

Evaluating the Peak Flowering Stages Derived from the New Method
Although the starts and ends of the flowering stage were recorded in the DWD observations, the peak flowering stages were not [39]. We verified the detected peak flowering dates and the accuracy through two steps ( Figure 3). First, we tested whether the flowering peak dates identified by our proposed method were between the start dates and the end dates recorded by the DWD stations. Second, the peak flowering dates identified by the two NRFIs from L8 and S2, and VH and VV/VH from S1 were compared with those detected by the two NDYIs from L8 and S2, and VV from S1. The previous study proved the accuracy of S2 NDYI and S1 VV by identifying the peak flowering of rapeseed based on a large amount of field survey data [2]. We used the Pearson correlation coefficient (r) (Equation (4)) and symmetric index of agreement (λ) (Equation (5)) to quantitatively assess the consistency of the peak flowering dates computed by L8, S1, and S2 time series [55,56].
The advantage of the symmetric index of the agreement is that the index indicates how similar the values of the series are in magnitude [38,56].
where X and Y denote the mean values of X and Y, respectively. X  and Y  represent the standard deviations of X and Y, respectively. The metric r is dimensionless, with a range from −1 to 1. A value of zero indicates there is no linear dependence between the two variables, while a value close to 1 indicates significantly positive.
The advantage of the symmetric index of the agreement is that the index indicates how similar the values of the series are in magnitude [38,56].

Spectral Properties and NDVI Phenological Characteristics of Rapeseed
Three different canopy morphologies are in the sequence indicated with the development of rapeseed, which is characterized by leaves, yellow flowers, and pods, respectively [35]. Rapeseed will appear yellowish-green on the image when it begins flowering. The color becomes yellow when rapeseed is approaching peak flowering (Figure 4a-e). Corresponding to the true color images above, the time series of spectral reflectance is shown in Figure 4f,g, which represents the mean values of 718 rapeseed parcels in four bands including blue, green, red, and SWIR2. The red and green bands increase as rapeseed begins flowering, peak at the medium flowering dates, and then decrease until the end of flowering for both L8 (Figure 4f) and S2 (Figure 4g). Moreover, their green bands display the highest reflectance, and the blue bands the lowest, with the red and SWIR2 bands in the middle, during the flowering period (Figure 4f,g). Reflectance values of the green and red bands in S2 changed from 0.11 and 0.079 (on Julian day 99) to 0.154 and 0.13 (on Julian day 109), respectively. While reflectance values of the green and red bands in L8 changed from 0.0695 and 0.05031 (on Julian day 99) to 0.1338 and 0.1177 (on Julian day 109), respectively. Previous research indicates that yellow flowers could increase red reflectance [57]. For rapeseed, the most distinguishable spectral difference between flowering and non-flowering can be identified by the green and red bands, which is consistent with previous studies [2,18,35,38,49,57]. The yellow rapeseed petals (e.g., carotene) increase the reflectivity of the red and green bands [17,18]. Thus, the red and green bands of rapeseed reach a local maximum during the peak flowering period (Figure 4f,g).

NRFI Better Characterize the Flowering Intensity
At peak flowering stages, the reflectance values of the green and red bands have a considerable increase, however, the values of SWIR2 are lower than those of the two visible bands (Figure 4f,g), which can provide valuable information to identify the rapeseed flowering period. The time series of spectral indices are extracted and averaged for the Comparing with the "inverse V" curves of green and red values from L8 and S2 (Figure 4f,g), their NDVI values show a completely different "V" shape ( Figure 4h,i). The NDVI values are high at the start and end of the flowering period, but low at the peak flowering stage, and reach the minimum at ripening [10]. NDVI values of L8 and S2 reduce from 0.78 and 0.66 to about 0.64 and 0.59 at the peak flowering stage, respectively, and then increase to the maximums (0.83 and 0.75) at the end of flowering. During the ripening stage, both NDVI decline to around 0.25. The NDVI dynamics are consistent with many previous studies [2,3,5,38,49]. For example, Pan et al. [35] indicated a decrease in NDVI values during the flowering period (BBCH65) up to 0.35, which was caused by the increase in red reflectance. For other crops such as wheat and corn, NDVI values do not change significantly during the flowering period [14]. Therefore, the unique spectral characteristics of rapeseed during the flowering period would provide a strong signal to identify and monitor the rapeseed flowering stage [5,14,38].

NRFI Better Characterize the Flowering Intensity
At peak flowering stages, the reflectance values of the green and red bands have a considerable increase, however, the values of SWIR2 are lower than those of the two visible bands (Figure 4f,g), which can provide valuable information to identify the rapeseed flowering period. The time series of spectral indices are extracted and averaged for the 718 rapeseed parcels ( Figure 5). The flowering intensity signals are captured well by NRFIr from L8 ( Figure 5a) and S2 (Figure 5b), with NRFIr showing a typical "reverse V" shape during the flowering period. Moreover, such a "reverse V" signature is also characterized by NDYI of S2, with NDYI showing a peak between the start and end of flowering stages recorded at DWD stations. However, two similar flattened peaks of NDYI and NRFIg from L8 ( Figure 5a) and one lagged peak of NRFIg (Figure 5b) were also observed, which suggests the peak signals for flowering identified by the indexes have fallen beyond their actual flowering stages. Nevertheless, all values beyond the flowering stages were consistently lower than those within flowering stages during the whole growing period of rapeseed. Moreover, compared with the other two indexes, NRFIr performed best to capture the real signal of flowering dynamics from L8. The good performance of NRFIr is consistent with NDYI from S2, substantiated by several previous studies focused on field survey data [2,17]. The different curves of the indexes between L8 and S2 may ascribe to their different wavelengths [58,59]. Based on the comprehensive analysis of 718 rapeseed parcels, NRFIr is a reliable indicator to monitor the flowering intensity of rapeseed. Therefore, we will focus on NRFIr in the analyses hereafter, and all NRFI hereafter means NRFIr. capture the real signal of flowering dynamics from L8. The good performance of NRFIr is consistent with NDYI from S2, substantiated by several previous studies focused on field survey data [2,17]. The different curves of the indexes between L8 and S2 may ascribe to their different wavelengths [58,59]. Based on the comprehensive analysis of 718 rapeseed parcels, NRFIr is a reliable indicator to monitor the flowering intensity of rapeseed. Therefore, we will focus on NRFIr in the analyses hereafter, and all NRFI hereafter means NRFIr.

Radar Polarization Characteristics of Rapeseed
The canopy structure changed dramatically as the rapeseed developed buds, flowers, and formed pods [32,35]. The backscatter and randomness in scattering changed with the processes of plant growth and was indicated morphologically by the change in canopy

Radar Polarization Characteristics of Rapeseed
The canopy structure changed dramatically as the rapeseed developed buds, flowers, and formed pods [32,35]. The backscatter and randomness in scattering changed with the processes of plant growth and was indicated morphologically by the change in canopy structure [32]. Thus, the time series of VV, VH, and VV/VH will inform us more on rapeseed growth states ( Figure 6). VV backscatter values decrease sharply from the start of flowering, then reach a minimum in the middle of flowering, and increase sharply again till the end of flowering (Figure 6a). VH behaved similarly with VV but more gently, which was indicated by a more flat "V" curve ( Figure 6b). This finding is supported by Wiseman et al. [60] based on RADARSAT-2 data [60]. As for S1, the backscatter values of the VV and VH changed from 0.11 and 0.079 (on Julian day 99) to 0.154 and 0.13 (on Julian day 109), respectively. Our findings are consistent with in-situ observations conducted by d'Andrimont et al. [2], which showed that the VV of S1 reached a local minimum during the peak flowering period of rapeseed. Moreover, the specific canopy structure of rapeseed might also contribute to the "V" shape during the flowering period. 1) The leaves and pods from neighboring plants are often intertwined to form a dense, randomly oriented canopy structure [14,15,31,32]. 2) As a strong back-scatter signal, stems, and pods of rapeseed, especially leaves, have characteristic sizes that are in the same order as the incident microwave wavelength [2,15]. 3) The smaller structured buds and flowers create a temporary layer of less effective scatters, which partly impedes the scattering of the underlying canopy, resulting in a decrease in VV scattering [2,14,60,61].
Remote Sens. 2021, 13, x FOR PEER REVIEW 10 of 19 [14]. Additionally, the VV/VH of the flowering period is the maximum during the whole growth period, while VH and VV are the local minimums. Therefore, VV/VH might be a more suitable indicator than VH or VV backscatter for monitoring the flowering period of rapeseed.

Comparing the Different Indexes for Monitoring Peak Flowering
To determine whether the peak flowering dates identified from satellites are within the ranges from the start to end of flowering observed at DWD stations, we further summarized the results in Figure 7 for comparisons. The flowering peak dates obtained from NRFIr, VV, VH, and NDYI (S2) fall within the ranges, but 59% of them from NDYI (L8) Furthermore, VV/VH increases as rapeseed enter the flowering stage and reach the maximum at the full flowering stage (Figure 6c). Considering their different responses during the stage (a sharp "V" in Figure 6a vs. a flat "V" in Figure 6b), rapeseed is randomly organized without a strong vertical structure during the flowering period. Furthermore, another advantage of selecting the ratio VV/VH as an index is that it can reduce the doublebounce effect and probably reduce errors from the acquisition system (e.g., due to the radiometric stability) or environmental factors (e.g., due to variations of soil moisture) [14]. Additionally, the VV/VH of the flowering period is the maximum during the whole growth period, while VH and VV are the local minimums. Therefore, VV/VH might be a more suitable indicator than VH or VV backscatter for monitoring the flowering period of rapeseed.

Comparing the Different Indexes for Monitoring Peak Flowering
To determine whether the peak flowering dates identified from satellites are within the ranges from the start to end of flowering observed at DWD stations, we further summarized the results in Figure 7 for comparisons. The flowering peak dates obtained from NRFIr, VV, VH, and NDYI (S2) fall within the ranges, but 59% of them from NDYI (L8) are beyond the range (Figure 7). Also, the spatial distribution of the peak flowering dates of rapeseed pixels based on NDYI (L8) and NRFIr (L8) shows that NRFIr (L8) is more accurate (Figures A1 and A2). The same median dates of peak flowering are obtained by NRFIr (L8 and S2), NDYI (S2), and VH (S1), but with a systematic delay of one to two days by VV (S1) than NDYI (S2) [2]. Therefore, in identifying the peak flowering dates, NRFIr performs better than NDYI in terms of spectral monitoring (L8), and VH does better than VV in terms of morphological monitoring (S1). Since the peak flowering dates were not recorded at DWD stations, a strict validation of them retrieved from satellites is impossible [39,62]. Alternatively, the r (Pearson correlation coefficient) and λ (symmetry consistency index) are used to assess the new indexes (NRFIr, VH, VV/VH) by comparing them with the indexes (NDYI and VV) used in a previous study [2] (Figure 8). The L8 NDYI has the most insignificant relationship with other indexes, further confirming its lower ability for identifying peak flowering (Figure 8a). Both NRFIr from L8 and S2 have significant relationships with S2 NDYI and S1 VV. The relationships among both morphological indexes are significant (r ≥ 0.7) (Figure 8a). λ has a very similar pattern to that of r (Figure 8b). For example, there is a quite weak agreement between L8 NDYI and other indexes as indicated by λ, but the highest agreement between S2 NRFIr and S2 NDYI (λ = 0.9). The λ for all morphological indexes (VH, VV/VH, and VV) is >0.7. Thus, NDYI from L8 does not perform as well as NRFIr from L8. Furthermore, VH and VV/VH also can be used to detect the rapeseed peak flowering period.
From the differences in each combination of two indexes for identifying peak flowering dates, we found the largest difference is caused by two NDYI from L8 and S2, with  Since the peak flowering dates were not recorded at DWD stations, a strict validation of them retrieved from satellites is impossible [39,62]. Alternatively, the r (Pearson correlation coefficient) and λ (symmetry consistency index) are used to assess the new indexes (NRFIr, VH, VV/VH) by comparing them with the indexes (NDYI and VV) used in a previous study [2] (Figure 8). The L8 NDYI has the most insignificant relationship with other indexes, further confirming its lower ability for identifying peak flowering (Figure 8a). Both NRFIr from L8 and S2 have significant relationships with S2 NDYI and S1 VV. The relationships among both morphological indexes are significant (r ≥ 0.7) (Figure 8a).
λ has a very similar pattern to that of r (Figure 8b). For example, there is a quite weak agreement between L8 NDYI and other indexes as indicated by λ, but the highest agreement between S2 NRFIr and S2 NDYI (λ = 0.9). The λ for all morphological indexes (VH, VV/VH, and VV) is >0.7. Thus, NDYI from L8 does not perform as well as NRFIr from L8. Furthermore, VH and VV/VH also can be used to detect the rapeseed peak flowering period.
From the differences in each combination of two indexes for identifying peak flowering dates, we found the largest difference is caused by two NDYI from L8 and S2, with L8 NDYI around 13 days later than that of S2 NDYI (Figure 8c). Also, compared with VV, the peak flowering dates derived from VH is earlier, but those derived from VV/VH have an approximate zero to seven day delay.

The Good Performance of NRFI for Identifying Flowering Stages of Rapeseed
In this work, we constructed a novel spectral index-NRFI, which not only shows high sensitivity to rapeseed petals during a vegetation canopy but also easily catches the peak flowering dates. The results show that L8 NRFI outperforms the L8 NDYI in detecting the flowering stages. NRFI is defined as the combination of red band and short-wave infrared band-2 (SWIR2), while NDYI is the combination of green and blue bands. The blue waveband is sensitive to atmospheric contamination and increasing uncertainty in detecting flowering phenology of rapeseed [17]. Moreover, decreasing chlorophyll content of rapeseed petals induces increased reflection in the red waveband [18,35,57]. A decrease in water contents in soil and rapeseed during the flowering period will inevitably be observed by SWIR2 [63][64][65]. Thus, the NRFI developed in our study can unsurprisingly capture such features signature during the time series of the flowering period of rapeseed.

The Good Performance of NRFI for Identifying Flowering Stages of Rapeseed
In this work, we constructed a novel spectral index-NRFI, which not only shows high sensitivity to rapeseed petals during a vegetation canopy but also easily catches the peak flowering dates. The results show that L8 NRFI outperforms the L8 NDYI in detecting the flowering stages. NRFI is defined as the combination of red band and short-wave infrared band-2 (SWIR2), while NDYI is the combination of green and blue bands. The blue waveband is sensitive to atmospheric contamination and increasing uncertainty in detecting flowering phenology of rapeseed [17]. Moreover, decreasing chlorophyll content of rapeseed petals induces increased reflection in the red waveband [18,35,57]. A decrease in water contents in soil and rapeseed during the flowering period will inevitably be observed by SWIR2 [63][64][65]. Thus, the NRFI developed in our study can unsurprisingly capture such features signature during the time series of the flowering period of rapeseed.
However, previous studies only focused on the visible light waveband (blue, green, and red) and near-infrared waveband. Ignoring the characteristics of short-wave infrared changes (e.g., SWIR2) during rapeseed growth [2,3,5,10,15,17,18,34,38,49]. For example, Siulk et al. [18] designed an index (including green and blue), similar to NDYI, to directly exploit variation in flowering [18]. Plant photosynthesis decreases during rapeseed flowering. They interpreted that the yellowness of rapeseed petals is caused by the carotenoid pigments absorbing the blue light, but reflect a mix of green and red light [17,18].
Besides the above advantages, the NRFI can be further applied to extract the plots cultivated by rapeseed. It has been reported that the phenological features of the crop could be used to identify the planting areas, which has been applied in previous studies although the accuracy varies by locations and methods [20,66]. We analyzed the pixels of NRFIr in the studied areas and find that the NRFIr values of rapeseed during the flowering period are distinctly different from those of other land use types (Figure 9). Comparing with other methods (e.g., visual interpreting of high-resolution images and supervised classifying by machine learning) to extract planting areas, the method based on NRFIr will be a more potential and powerful way since others are either time-consuming or strongly depend on training data collected in fields [2,3,26,27,67]. Therefore, NRFIr has the potential for rapeseed planting area mapping.
Remote Sens. 2021, 13, x FOR PEER REVIEW 13 of 19 although the accuracy varies by locations and methods [20,66]. We analyzed the pixels of NRFIr in the studied areas and find that the NRFIr values of rapeseed during the flowering period are distinctly different from those of other land use types (Figure 9). Comparing with other methods (e.g., visual interpreting of high-resolution images and supervised classifying by machine learning) to extract planting areas, the method based on NRFIr will be a more potential and powerful way since others are either time-consuming or strongly depend on training data collected in fields [2,3,26,27,67]. Therefore, NRFIr has the potential for rapeseed planting area mapping. (1)-(4) are partial drawings enlarged.

The Ancillary Function of Morphological Indexes for Identifying Flowering Stages of Rapeseed
Compared to optical images, SAR images have two advantages: no interruption from clouds and a shorter revisiting cycle. The "V" curves identified by VV and VH from S1

The Ancillary Function of Morphological Indexes for Identifying Flowering Stages of Rapeseed
Compared to optical images, SAR images have two advantages: no interruption from clouds and a shorter revisiting cycle. The "V" curves identified by VV and VH from S1 are shown during flowering periods (Figure 6), simultaneously companying with the "reverse V" shapes from optical images ( Figure 5). Such changes in both VV and VH are mainly affected by soil and canopy [2,16,68]. For example, VV is particularly sensitive to the water content of vegetation [69]. Rapeseeds' vertical transformation from a thick rosette of leaves to a flowering stalk will attenuate VV polarization [34]. Moreover, VV and VH increase from the development of fruit to the beginning of ripening because of the rise in the number and the length of stems per plant [2,14,32,34]. Therefore, such changes identified by SAR will be beneficial for characterizing crop phenology features on the other side of optical images [14,19,31]. The "V" changes of VV and VH during the flowering period provide opportunities for an automatic, continuous, and accurate monitoring rapeseed flowering period.
A typical "reverse V" is identified by VV/VH during the rapeseed flowering period (Figure 6), and such changes in VV/VH may be correlated more with fresh biomass than photosynthetic activity [14,30,34]. Additionally, some previous studies have shown that the fresh biomass of wheat and corn is associated with VV/VH [14,34]. Due to the typical changes of VV/VH with the flowering processes, the VV/VH ratio will provide a potential way for applying it for biophysical parameters retrieval into other crops and other regions in the future.
Our findings highlight a highly potential way to detect the flowering stage of rapeseed using SAR data, particularly for VV/VH ratio, which has been rarely investigated in previous studies. Moreover, the characteristic variations in scattering coefficients during rapeseed growth are useful for mapping high-resolution cultivated areas of rapeseed. For example, Singha et al. [68] drew a high-resolution rice map in cloud-prone Bangladesh and Northeast India based on Sentinel-1 SAR backscatter coefficients during rice growth. Additionally, such radar data may also be applied to fill eventual gaps in the optical data series. Following the free availability of Sentinel data, it will eventually support near real-time crop monitoring in the future [14]. Combining SAR with optical data should be more accurate for monitoring the flowering period and planting areas of rapeseed [2,34], which can also be applied to other crops and other regions.

Limitations and Perspectives
Despite the above findings, the revisit frequency of optical satellite surveys and the number of high-quality images should be considered in characterizing the peak flowering period. Cloud is one of the main limitations in the consistent and accurate detection of flowering phenology of rapeseed. The new method with the indexes proposed in our study will have great potential at high latitudes because of the increasing availability of high-quality images [44,70], especially from Sentinel-1/2 satellites of Europe [58]. Further research should be focused on standardizing a classifier based on the optical and morphological indexes to map in more detail the phenology, growth, and planting areas for rapeseed at a large spatial scale.

Conclusions
The Landsat satellite has longer time series historical images and is the main data source for monitoring historical rapeseed flowering phenology. Sentinel-1/2 data have higher temporal and spatial resolutions, which is conducive to monitoring rapeseed phenology accurately. In this study, we integrated Landsat 8 and Sentinel-1/2 data to assess the potential of satellites to detect rapeseed flowering dates. The time series of rapeseed's reflectance and polarized backscattering coefficients (VH, VV, and VV/VH) were analyzed. The observations collected at different DWD stations were applied. We proposed L8 NRFI and found that it outperforms L8 NDYI in monitoring rapeseed peak flowering. The correlation (r) between L8 NDYI and other indices observed peak flowering dates less than 0.3. L8 NRFIr and other indices were higher correlated (r ≥ 0.5). Also, the results show that VH and VV/VH have the potential to monitor the peak flowering stage of rapeseed.
The relationships among both SAR indices are significant (r ≥ 0.7). The VV/VH signals are more pronounced whereas detection by VV/VH is slightly delayed (zero to seven days) relative to VV and VH. The symmetry consistency index (λ) between different indices has a very similar pattern to that of r. The λ for all SAR indexes is ≥0.7. Our study showed that a better understanding of temporal behaviors of spectral indices and backscatters during rapeseed flowering season will benefit not only biophysical parameters monitoring but also dynamic crop growth monitoring and planting area mapping.

Conflicts of Interest:
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. a very similar pattern to that of r. The λ for all SAR indexes is ≥0.7. Our study showed that a better understanding of temporal behaviors of spectral indices and backscatters during rapeseed flowering season will benefit not only biophysical parameters monitoring but also dynamic crop growth monitoring and planting area mapping. Funding: This study was supported by the National Science Foundation of China (41977405).

Conflicts of Interest:
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.