Multitemporal radar backscattering measurement of wheat fields using multifrequency (L, S, C, and X) and full-polarization

Authors


Abstract

[1] This paper presents the measurements of the backscattering coefficients over the wheat fields using an L-, S-, C-, and X-band scatterometer system during a whole period of wheat growth. The wheat field located at Qionglai County of China was measured during a wheat growing season from November 2010 to May 2011. Twelve experimental acquisitions (ground and radar data) were measured over the wheat field in a flat area at VV-, VH-, HV-, and HH-polarizations of L-, S-, C-, and X-bands, with the incidence angles ranging from 0° to 80°. Wheat biomass, canopy structure, leaf area index (LAI), soil moisture, and eco-physiological canopy variables were also collected to investigate the radar sensitivity. It shows that the HH measurements for wheat are higher than VV at each band after jointing, and the value of HH/VV is larger for lower frequency. The temporal variations of each band at selected incidence angles were compared with the wheat biomass, canopy height, LAI, and soil moisture. The correlations were analyzed between backscattering coefficients and wheat crop variables at 23°, 38°, 53°, and 68°, respectively. The results show that the backscattering coefficient has a strong correlation with biomass and LAI, especially for the HH- and cross-polarizations of L-band, and the radar backscatter signatures at high frequency (X- and C-bands) are sensitive to detect newly transplanted thin wheat seedlings at the high incident angle. These data can provide experimental evidence to allow for determining the growth status and/or condition of wheat by use of measured radar backscatters.

1 Introduction

[2] Wheat is one of the major food crops in China and even over the world. To monitor changes and cultivation intensity in the wheat production area, satellite remote sensing data constitute a unique tool which can provide timely, consistent spatial and temporal coverage needed at regional to global scales. Microwave sensors may be more effective in monitoring wheat growth than optical sensors, since an electromagnetic wave with a longer wavelength is less affected by clouds and precipitation events. There are a lot of studies that observed and analyzed the temporal variation of synthetic aperture radar (SAR) measurements over wheat fields, by which useful information can be retrieved from satellite SAR data.

[3] In the previous studies, the backscattering measurements of wheat-growing areas were carried out using ground-based scatterometers [Bouman and van Kasteren, 1990; Ulaby et al., 1986; Brown et al., 2003; Mattia et al., 2003] and SAR [Macelloni et al., 2001; Ferrazzoli et al., 1997; McNaim et al., 2004; Satalino et al., 2009]. For example, the X-band radar backscattering of wheat was investigated through the whole growing season at VV- and HH-polarizations at the incidence angles from 10° to 80° [Bouman and van Kasteren, 1990]. Ground-based radar observations showed that the geometrical architecture of the crop canopy was a major factor that influenced the X-band radar backscattering of wheat. The ground-based SAR can be used to measure the wheat canopies [Brown et al., 2003], in which the scattering indicated a pronounced layered structure, with strong returns from the soil and the flag leaves, and in some cases a second leaf layer. In particular, C-band ground-based scatterometer can be used to analyze the relevance of C-band measured backscatter coefficient to wheat biomass and the underlying soil moisture content [Mattia et al., 2003]. On the other hand, C- and L-bands airborne and satellite SAR data were obtained [Macelloni et al., 2001], which proved that the relations between the backscattering of crops and the vegetation biomass varying with plant types, and the trends for narrow and broad leaf crops were different. Furthermore, Ferrazzoli and others [Ferrazzoli et al., 1997] analyzed the ability of monitoring different vegetation types and biomass by P-, L-, and C-band observation. McNaim and others [McNaim et al., 2004] used the polarimetric radar data to investigate the sensitivity of wheat backscattering to crop yield. However, Satalino and others [Satalino et al., 2009] achieved wheat classification by applying an optimal threshold to the HH/VV-copolarized backscatter ratio acquired during the peak growing stage.

[4] Although these scatterometers and SAR data were useful in assessing the sensitivity of radar backscatter to wheat growth, they cannot be used to understand the multifrequency and full-polarimetric responses of wheat fields at various incidence angles. In this study, we aimed to obtain as many backscatter measurements as possible over a wheat field using our scatterometer system. The system had multifrequency, full-polarization, and multiangular sensor configurations. A part of the results was already published at reference [Jia et al., 2012], and more details will be discussed in this paper, including system description, experimental process, wheat growth parameters, data analysis, and conclusion. Our aims are to investigate the sensitivity of backscattering coefficients to wheat growth parameters and to provide more clearly defined understanding on the interaction of microwave backscatter signatures with the wheat biophysical information, including the growth condition (biomass, leaf area index (LAI), and height) and the underlying soil moisture as a function of wheat phenological stages.

2 Study Area and Field Measurements

[5] The study area is located at a pilot field of the University of Electronic Science and Technology of China (UESTC) field (103°32′24″E, 30°24′11″N) in Qionglai County, Chengdu, Sichuan Province, China, where wheat is a major planted crop in winter and spring. The winter wheat is seeded in early winter, with the growing season from early November to next middle May. We carried out 12 times field measurements during the whole period of wheat growth. The wheat growth period is about 200 days, which is usually divided into 11 growth stages, including emergence, two leaves, tillering beginning, advanced tillering, jointing, flag leaf fully emerged, boot, head emergence, flowering, maturity beginning, and full maturity. Figure 1 shows the growth state of the wheat phonological stage in each of the field measurements.

Figure 1.

The wheat growth chart of different periods: ((1) emergence, (2) two leaves, (3) tillering begins, (4) advanced tillering, (5) jointing, (6) flag leaf fully emerged, (7) boot, (8) head emergence, (g) flowering, (10) maturity begins, (11) full Maturity, and (12) after harvest).

[6] During the period of wheat growth, the ground truth data of the wheat field were collected in each experiment. The collected data included fresh weight biomass, LAI, canopy height, stalk parameters (length, diameter, density, and moisture), leaf parameters (major axis, minor axis, thickness, number of leaves per plant, Eulerian angles, and moisture), ear parameters (length and moisture), soil parameters (moisture, root mean square (RMS) and correlation length). Table 1 shows the detailed parameters of 11 experiments. The twelfth experiment was conducted after the wheat harvest, so wheat parameters were not included. In order to obtain representative data, at least 20 sampling points were randomly selected in the wheat field of each experiment. The data shown in Table 1 are the average of the random sampling parameters. Figure 2 shows the temporal variation of main sampling parameters and their fitting curve, including biomass, wheat high, LAI, and stem density.

Table 1. The Wheat Growth Parameters Collected in the Field
 Number of Experiments
 1234567891012
Age (days)16365778110126141155169182201
Fresh weight biomass (g)24367133524752093410769275446475836692064
LAI (m2/m2)0.10.70.60.92.84.95.95.55.34.34.5
Canopy height (cm)5.414.616.628.138.755.962.183.384.28681.7
Stalk length (cm)4.35.67.79.416.128.745.557.265.86764.6
Stalk diameter (mm)1.02.12.235.55.75.565.25.15.5
Stalk density (m−2)100211247237250266298344347348344
Stalk moisture (%)78.388.284.38489.387.589.782.478.371.221.1
Leaf major axis (cm)5.17.617.215.617.222.722.924.424.120.619.9
Leaf minor axis (mm)3.57.20.910.19.91817.520.918.216.812.6
Leaf thickness (mm)0.410.260.260.530.410.310.70.330.450.360.51
No. of leaves per plant24444554444
Leaf moisture (%)78.388.284.383.374.58288.384.480.762.312.5
Ear length (cm)-------1210.714.79.9
Ear moisture (%)-------81.277.463.122.7
Soil moisture (%)25.820.419.127.932.026.127.426.227.522.923.8
Leaf Eulerian angles β (deg.), parameters of lognormal normal distribution (μ σ)--(3.7 1.6)(2.9 1.3)(2.9 1.2)-(2.7 1.2)(3.7 1.1)(3.7 1.1)(3.2 0.8)(4.5 0.8)
Figure 2.

Temporal variation of the main wheat parameters, including wheat biomass, wheat high, LAI, and stem density.

[7] Soil moisture content was measured as volumetric moisture content, since there was a difference of approximately 13% for the moisture during the wheat growing season. Soil roughness was also measured at the first three and last experiments in the field, because it could not be measured at other times when wheat is too high to block the device for roughness measurements [Lu et al., 2008]. The average profile RMS of wheat height in the first three times was 1.3 cm, and the correlation length was 5.7 cm. The last time profile RMS of height was also 1.3 cm, and the correlation length was 6.0 cm. The data indicate that the roughness was a little bit changed in the wheat growing season.

[8] The shape of wheat leaves is similar with ellipse, and the angle of wheat leaves is randomly distributed with the Eulerian (α, β, γ), but the angle is usually different in different growing phase. In order to obtain the angle of wheat leaves probability density function (PDF) to describe the spatial distribution of different growing phases, more than 1000 leaf angles were measured in each experiment by scattering measurements. Based on these data, the lognormal normal distribution was used to describe the PDF of Beta.

display math(1)

[9] Figure 3a shows the probability distribution histogram of 1000 leaf angles in the last experiment, which is the curve of the lognormal normal distribution based on the curvilinear regression analysis. Figure 3b gives the parameters of the lognormal normal distribution (μ, σ) and a data-fit curve (μ = 0.0002x2 − 0.047x + 5.569 and σ = − 0.0049x + 1.8208), by which the parameters of the lognormal normal distribution (μ, σ) were also listed in Table 1. However, the Alpha angles were supposed to have a uniform distribution in azimuth (α = 0°–360°), and the value of Gamma angles was zero obviously (γ = ).

Figure 3.

(a) PDF curve of wheat leaves of eighth; (b) Parameters of lognormal normal distribution each time.

3 The System for Measurements

[10] The ground-based radar scatterometer (GBRS) is a frequency-modulated continuous-wave radar system, which is used to measure full-polarization in our field measurements [Jia and Tong, 2011]. The GBRS system can measure the backscattering coefficients between −45 dB and 20 dB with a target-distance from 10 m to 100 m. The GBRS is equipped with four operating frequencies at L-band (2 GHz), S-band (3.1 GHz), C-band (5.3 GHz), and X-band (10 GHz), respectively. The system was constructed at the UESTC, with four different pairs of parabolic antennas. Each pair of antennas has a transmitting channel illuminating the soil with a linear polarized (h or v) electromagnetic wave and a receiving channel detecting the linear polarized (h or v) wave backscattered from the target. Main parameters of the GBRS system are summarized in Table 2. It has a tilt and azimuth scanning capability of the land-based radar platform, which can be operated at the incidence angle or pitch angle range from 0° to 90° and the azimuth angle range from 0° to 360°. A computer connects with the scatterometer via a standard serial interface RS232 communication cable with the length of 15 m. Figure 4 shows the scatterometer under the measuring state. A software interface provides the functions of backscattering coefficient-versus-angle curves, each of the backscattering coefficients, distance and temperature parameters, state of antenna azimuth angle, and incidence angle. It also supplies a variety of function control keys. The software can achieve continuous and automatic measurements at different incidence angles. All measured data can be automatically stored in the database.

Table 2. The Scatterometer Parameters
ParameterL-BandS-BandC-BandX-Band
Center frequency2.0 GHz3.1 GHz5.3 GHz10 GHz
Bandwidth0.6 GHz0.8 GHz0.8 GHz0.8 GHz
Measurement error0.4 dB0.4 dB0.4 dB0.5 dB
Range resolution0.38 m0.35 m0.38 m0.35 m
Dynamic range65 dB65 dB65 dB60 dB
Transmit power16 dBm21 dBm20 dBm20 dBm
Beam width11.5°/13°9.0°/11.8°6.8°/8.0°4.5°/5.7°
Antenna typeParabolic antennaParabolic antennaParabolic antennaParabolic antenna
Polarizations modesHH,HV, VV,VHHH,HV, VV,VHHH,HV, VV,VHHH,HV, VV,VH
Figure 4.

The picture of the eleventh measurement.

[11] In the period of wheat growth, the scatterometer was located at the same site, with the same vertical height of 12.5 m for each experiment. The measurements had been conducted at full-polarization (VV, VH, HV, and HH) with the incidence angle (from 0° to 80°) at a certain band of L-, S-, C-, and X-bands. Figure 4 shows an example of the site of the eleventh measurement.

[12] The antennas were installed at a pivoting system to measure several different independent samples of wheat targets, by which these measurements are latterly averaged to reduce the speckle. To ensure a 90% confidence interval of 1 dB, at least 36 independent samples are required to be averaged [Ulaby et al., 1982]. The number of independent samples (of wheat) contained in a footprint is the same as the number of resolution cells that fall in the illuminated area. It can be calculated with the following expression:

display math(2)

Where, Ns is the number of independent samples, h is the height from antenna to ground, βν is the beam width, Br is the modulation band of the scatterometer, βν and Br are different bands listed in Table 2, and θ is the antenna incidence. For example, at a 12.5 m height and an incidence angle of 45°, the C-band scatterometer footprint contains about 11.2 resolution cells. So, the required independent samples can be achieved by acquiring 20 unoverlapping footprints, by pivoting the antenna around a vertical axis. In this case, the total number of independent samples is more than 220 resolution cells. As a result, the number of independent samples is from 60 (at the incidence angle of 20°) to 5000 (at the incidence angle of 80°) at C-band, for example. With the assumption that the calibration and statistical errors are independent, the overall system error of the measurement can be estimated as ±1 dB.

4 Data Analysis

[13] In this section, the backscattering coefficients were analyzed with the variation of angular, temporal, and wheat growth parameters. First, full-polarization backscattering coefficients were analyzed with the changes of the incidence angle at each band. To better understand the scattering characteristics, the wheat growth season is divided into three separate periods. The temporal variations of the full-polarization backscattering coefficients were then analyzed at the four selected incidence angles of 23°, 38°, 53°, and 68°, respectively. The backscattering coefficients were finally analyzed and compared with the temporal variation of wheat biomass, LAI, canopy height, and soil moisture. Here, the cross-polarizations (HV and VH) were averaged for all of the measurements, since the cross-polarization backscattering coefficients are the same in both theory and experiments.

4.1 Angular Variation of Wheat Backscatter

[14] Figures 5a, 5b, 5c, and 5d show the backscattering coefficients of full-polarization with angle variations in three selected experiments of the first, sixth, and ninth. During the course of the wheat growing season, the backscattering coefficient gradually decreases with the increasing of incidence angle for all bands. The difference is that higher frequency varies more gently with the incidence angle. At the 0° to 20° range of incidence angles, the angular behavior of each band observed over the wheat field is similar to the exponential decay. At the range of 20° to 60° of incidence angles, the backscattering coefficient is similar to logarithmic decay because the dominant mechanism changes to canopy scattering at each band in sixth and ninth [Mattia et al., 2003]. The backscattering coefficients of each band have different dynamic ranges from 20° to 60°: L-band at 15–10 dB, S-band at 13–8 dB, C-band at 12–6 dB, and X-band at 10–4 dB. When the incidence angle draws on 70°, the antenna radiation at the edge of wheat fields construct a corner reflection effect, so the backscattering coefficient is increasing rapidly. On the other hand, the trend of cross-polarization is similar to the HH-polarization as shown in Figure 5. Thus, the backscattering coefficients of different polarizations show very distinct behaviors in different wheat stages.

Figure 5.

Full-polarization backscatter values versus incidence angles for data referring to first, sixth, and ninth. (a) L-band, (b) S-band, (c) C-band, and (d) X-band.

[15] In the first experiment, the stem height was only about 5 cm, and fresh weight biomass was less than 30 g. The physical structure of the wheat plants which consist of mainly short vertical leaves and stems has little contribution to the total backscattering. So the backscatter values versus incidence angles are close to the backscatter from bare soil with the same soil moisture content and surface roughness. The VV-polarization backscattering coefficient is higher than the HH-polarization for the L- and S-bands, but HH-polarization is little higher than the VV-polarization for the C- and X-bands. As can be seen from the size of the HH/VV difference value for soil backscatter, the HH/VV is greater at the longer wavelength.

[16] In the sixth experiment, the wheat was at the flag leaf fully emerged stage. The stem height was about 56 cm, and fresh weight biomass was more than 4000 g. The main scattering is from wheat canopy. The backscattering coefficient for HH-polarization is higher about 3–5 dB than VV-polarizations at a wide range of incidence angles (20° to 60°) for each band. This is referred to the difference in the HH and VV attenuation. Due to the vertical structure of plant stems, the attenuation for VV is higher for HH. As a result, the soil contribution to the total backscatter at HH is significantly higher than at VV. This phenomenon is also present in other wheat stages, such as the heading and ripening stage.

[17] In the ninth experiment, the wheat was at the flowering stage. At the incidence angles from 20° to 45° for each band, the backscattering coefficient for HH-polarization was about 3–5 dB higher than that for VV-polarization. The dominant mechanism changes to canopy scattering for VV after heading at higher incidence angles (45° to 70°). These results are in substantial agreement with a recent experiment reported in Mattia et al. [2003].

4.2 Temporal Variation of the Wheat Backscatter

[18] For the growth of wheat canopy and the change of soil parameters, the backscattering coefficients have different characteristics at each band. In this study, the growing season is divided into three separate periods to show the distinct behaviors by the wheat age (see Figure 1). First Period (Age 0–78): from sowing to advanced tillering; Second Period (Age 78–169): from jointing to flowering; and Third Period (Age 169–201): from maturity begins to before harvest. Figure 6 shows the temporal variation of HH-, VV-, and cross-polarizations at 23°, 38°, 53°, and 68° at L-band, S-band, C-band, and X-band.

  1. [19] In the first period, the temporal variation curves of L-band and S-band have similar trends as shown in Figures 6a and 6b. At VV-polarization, the backscatter decreases about 3–5 dB, and the backscatter first decreases, then stabilizes at HH-polarization. The cross-polarization backscatter is similar with the HH-polarization backscatter, but the backscatter value is smaller about 8 10 dB. Each of the polarization backscatter decreases significantly in the first period, due to the reducing of soil moisture which could decrease the backscatter, and the growing of wheat which could increase the attenuation of soil backscatter (Table 1). From earlier to later growth stages, the VV-polarization backscatter is higher than the HH-polarization backscatter, which is consistent with the simulation results of integral equation method [Fung, 1994]. But the differences were gradually reduced with the growing of wheat. As the vertical structure of the plant stems makes the attenuation gradually higher for VV than for HH, the backscatter values of HH and VV are consistent at the end of this period. Similarly, each polarization backscatter of C-band is also gradually reduced. However, the backscatter of HH-polarization was slightly larger than VV-polarization. In comparison, the backscatter of X-band was relatively stable, and the HH-polarization even remained unchanged in this period. The VV-polarization first decreased and then increased, so the polarization difference of X-band was alternating. Therefore, these features indicate that for the shorter wavelength (C- and X-bands), the wheat growth parameters have a strong impact on the backscatter, but for the longer wavelength (L- and S-bands), soil moisture has a strong impact on the backscatter just in the early growth period.

  2. [20] In the Second Period, the HH measurements for wheat were higher than VV at each band. These differences were changing with the phenological stage, band, and incidence angle. The largest difference of each band was stated as follow: 5–6 dB for L-band (booting stage for 38° and 53°), 4–6 dB for S-band (booting stage for 23°, 38°, and 53°), 2–4 dB for C-band (flag leaf fully emerged stage for 23°, 38°, and 53°), and 3–5 dB for X-band (jointing and booting stage for 38° and 53°), respectively. Following the wheat growth, the attenuation is higher for VV than for HH, but the backscatter of wheat canopy is higher for HH than for VV. As a consequence, the soil contribution to the total backscatter was significantly lower for HH-polarization than for VV-polarization. In comparison, the contribution of wheat canopy to the total backscatter was significantly higher for HH- than for VV-polarization. This is the reason that the VV-polarization backscatter continues to decrease, while that for HH-polarization gradually increases, except for at X-band. This phenomenon is more important at middle incidence angles than at low and high incidence angles, because the attenuation is smaller at low incidence angle, and the canopy backscatter is relatively larger at high incidence angle for VV-polarization. In addition, the volume scattering from the canopy is less sensitive to a large range of incidence angle from 30° to 60° (see Figure 1) [Stiles et al., 2000]. So, the difference between VV- and HH-polarizations is smaller at both low and high incidence angles.

  3. [21] In the third Period, canopy (including stalk and leaves) water content decreases, and wheat leaves begin to wither. Each band had different trends in different polarizations and at different incident angles, as these conditions changed with the various backscatter contribution of soil and wheat canopy. For example, for L-band, the backscatters of HH-, VV-, and cross-polarizations first decreased, then increased at low incidence angles (23°, 38°), but gradually decreased at high incidence angles (53°, 68°). As the canopy contribution decreased with the lower water content, when soil contribution increased with the decrease of canopy attenuation, if the incidence angle is higher, the backscatter is lower. For S-band, the backscatter of each polarization was first stable, then increased at low incidence angles (23°, 38°), and finally decreased at high incidence angles (53°, 68°). The reason for the above differences with L-band is that the soil backscatter is higher at S-band than L-band. For C-band, the backscatter of HH-polarization increased at low and middle incidence angles (23°, 38°, and 53°), but VV- and cross-polarizations were first increased, then decreased at middle and high incidence angles (38°, 53°, and 68°), and the cause of the difference is similar to that at S-band. For X-band, the backscatter of HH- and VV-polarizations first decreased, then increased or stabilized at low and middle incidence angles (23°, 38°, and 53°). The change of canopy backscatter was more impacted on X-band that has the shortest wavelength, so the backscatter decreased with the lower water content. While wheat leaves begin to wither, the soil backscatter is still large at low incidence angles.

Figure 6.

Multitemporal backscatter values acquired at HH-, VV-, and cross-polarizations and at 23°, 38°, 53°, and 68° incidence angles. (a) L-band, (b) S-band, (c) C-band, and (d) X-band.

[22] Throughout the wheat growing season, for L- and X-bands, the overall trend was reduced at each polarization, which is very clear at low incidence angles. The difference between maximum and minimum of the backscattering coefficient is about 5–8 dB (L-band) and 8–12 dB (X-band), respectively. For S- and C-bands, the overall trend first decreased, then increased, and the backscattering coefficient between emergence and maturity stage was consistent at low incident angles. The difference between the maximum and minimum is about 6–8 dB (S-band) and 8–10 dB (C-band). For S-band, the minimum of VV-polarization appears in the heading emergence stage, and the minimum of HH- and cross-polarizations appears in advanced tillering. For C-band, the minimum of each polarization appears in flag leaf fully emerged stage.

4.3 Backscatter and Wheat Growth Parameters

[23] The temporal variations of each band at selected incidence angles have been compared with the temporal variation of wheat biomass, canopy height, LAI, and soil moisture (see Table 1), which refer to the total fresh weight (kg/m2), canopy height (cm), LAI (m2/m2), and volumetric water content of soil (%), respectively. Figure 7 shows the temporal variation of backscatter at each band of HH/VV at 45° together with those of wheat growth parameters and their correlation coefficients, (a) biomass, (b) canopy height, (c) LAI, and (d) soil moisture. Figures 7a, 7b, and 7c show that the temporal change of HH/VV at 45° follows the biomass, canopy height, and LAI variation well at L-band and S-band, but poorly agrees with C-band and X-band. Figure 7d plots HH/VV at 45° against soil moisture, in which they are negative consistent at each band, especially at C- and X-bands.

Figure 7.

Multitemporal values of HH:VV ratio at 45° incidence angle compared to the wheat growth parameters values at different bands. (a) Biomass, (b) canopy height, (c) LAI, and (d) soil moisture.

[24] Here, a polarization discrimination ratio (PDR) is defined as [Singh, 2006]:

display math(3)

[25] In our experiments on wheat, however, the backscattering coefficient at high-frequency bands (C- and X-bands) was poorly correlated with LAI (less than 0.50). This discrepancy is probably because of the difference in wheat structure (i.e., leaf size or stem diameter), background situation (i.e., surface roughness and soil moisture), and analysis method (i.e., selected stages in the entire wheat season). Correlations were analyzed between backscattering coefficients and wheat crop variables at 23°, 38°, 53°, and 68°, respectively. Figure 8 shows the results of the correlations coefficient for biomass, canopy height, LAI, and soil moisture.

Figure 8.

The correlation coefficient between the wheat growth parameters (biomass, canopy height, LAI, and soil moisture) and multitemporal values of different polarizations (HH, VV, cross, HH/VV, and PDR) at 23°, 38°, 53°, and 68° incidence angles in each band.

[26] All of the correlations between wheat growth parameters and PDR are negative, but almost positive with HH/VV. The HH/VV at L- and S-bands is highly correlated with LAI, and the highest correlation is 0.96 at L-band at 23°. VV-polarization is negatively correlated with growth parameters, but the best correlation is at L- and X-bands with the canopy height. HH- and cross-polarizations are positive at S-band, but almost negative at other bands. However, only cross-polarization of L-band has a strong correlation with the soil moisture.

[27] For L-band, HH-polarization was correlated better with canopy height and LAI at the incidence angle of 53° than other angles, but was poorly correlated with the biomass at each angle. VV-polarization is strongly correlated with canopy height and LAI at low incidence angles (23°, 38°) and well correlated with biomass at the incidence angle of 68°. HH/VV and PDR are highly correlated with biomass, canopy height, and LAI at each incidence angle, except the high incidence angle of 68° with LAI. The correlation coefficient is higher than VV-polarization with LAI, but lower than VV-polarization with canopy height. Moreover, there are slightly higher correlations of HH- and cross-polarizations with soil moisture than other polarizations.

[28] For S-band, it is very close to L-band, HH/VV and PDR are also highly correlated with the biomass, canopy height, and LAI at each incidence angle like L-band. HH-polarization is strongly correlated with the three growth parameters of wheat (biomass, canopy height and LAI) at the middle incidence of 53°, as well as soil moisture at high incidence of 68°. Here, cross-polarization is similar with HH-polarization, but the correlation is weak. In addition, HH-polarization and cross-polarization are poorly correlated with the soil moisture at the low-middle incidence angles (23°, 38°, and 53°) with a relevance of less than 0.20. Most of VV-polarizations were negatively correlated with the three growth parameters, but appeared as large negative correlations at the incidence angle of 23°. The overall correlations are very poor, compared with L-band, indicating that VV-polarization at L-band is more sensitive to volume scattering.

[29] For C-band, each polarization is almost negatively correlated with wheat parameters except HH/VV. The overall correlations were much smaller than those at the L- and S-bands. Even though the correlations with wheat parameters of HH- and cross-polarizations at C-band are similar to those at L-band, the high correlation is only observed at the high incidence angle. VV-polarization has little correlation with the three growth parameters except at 53°. HH/VV and PDR are less correlated with the growth parameters than VV-polarization, except for the canopy height. All polarizations are poorly correlated with the soil moisture, indicating that the microwave penetration of wheat canopy at C-band is not strong.

[30] For X-band, HH-, VV-, and cross-polarizations have higher correlations with the three growth parameters than those at C-band, especially at the low incidence angle. This indicates that X-band with the short wavelength is more sensitive to the change of wheat growth parameters. The backscattering coefficient of X-band is poorly correlated with soil moisture, less than 0.10, because the X-band can hardly penetrate the wheat canopy. HH/VV is highly correlated with the growth parameters at the high incidence angle, but less correlated at the low incidence angle. They are more highly correlated with soil moisture at the high incidence angles than those at C-band. The correlation of PDR with the soil moisture is only slightly higher than that with other parameters at the middle incidence angle.

[31] The correlations between backscatter and wheat growth parameters give a direct reference to develop scattering model and inversion method. If the absolute values of the correlations are greater than 0.7, we can establish empirical or semiempirical model to invert the wheat growth parameters, such as HH/VV (or PDR) and biomass at L- and S-bands, VV and canopy height at L- and X-bands, PDR and LAI at L- and S-bands, HH and soil moisture of high incidence angle at S- and C-bands, and so on (see Figure 8).

5 Conclusions

[32] The paper presents data analysis of the wheat field measurements using a multitemporal ground-based radar scatterometer system at the Qionglai pilot site in Southwest China, during the wheat growing season from November 2010 to May 2011. A data set was collected, which included the backscattering coefficients at four frequencies (L, S, C, and X), full-polarization (VV, VH, HV, and HH), and incidence angles from 0° to 70° for the entire wheat cycle. The data set also included a wide range of wheat crop variables, such as wheat biomass, canopy structure, LAI, soil parameters, and eco-physiological canopy variables. In addition, over 1000 leaf angles were measured in each experiment of scattering measurement. The lognormal normal distribution was first used to describe the PDF of wheat leaf angles, and the parameters of the lognormal normal distribution (μ, σ) were fitted by a polynomial curve in the wheat growing season.

[33] The full-polarization backscattering measurements at four bands have been analyzed as a function of the incidence angle. The backscattering coefficient of each band has a different dynamic range from 20° to 60°, the shorter the wavelength, the smaller the dynamic range. The backscattering coefficients of the different polarization show distinct behaviors in different wheat stages. For example, the backscattering coefficient for HH-polarization is about 3–5 dB higher than that for VV-polarization at a wide range of incidence angles (20° to 60°) for each band at the flag leaf fully emerged stage. The temporal variations of full-polarization at four selected incidence angles (23°, 38°, 53°, and 68°) have been analyzed by three separate phenological stages. The HH measurements for wheat are higher than VV at each band after jointing, and the value of HH/VV is larger for lower frequency. However, the difference between maximum and minimum of the backscattering coefficient in the wheat growing season is larger for higher frequency. It indicates that multipolarimetric data may be useful for low frequency to determine the wheat growth status, but multitemporal data may be useful for high frequency. They were also compared with the temporal variations of wheat biomass, LAI, canopy height, and soil moisture. The sensitivity of full-polarization radar signals at each band, to biophysical parameters of wheat had also been analyzed and compared. It is found that the relationship of radar backscattering with biophysical measurements is highly related to the wheat canopy structure during the wheat phenological stages. Additionally, a great difference in the sensitivity of backscatter to wheat biophysical exists before and after the heading stage.

[34] Our analysis clearly indicates some new findings of the interaction between radar backscatter and wheat variables. One is that the backscattering coefficient is very sensitive to the changes of biomass in the early wheat growing period, which just reached the saturation point with the increase of biomass, then backscattering of wheat declined near their maturity, significantly at L- and S-bands. The backscattering coefficient has a strong correlation with biomass and LAI, especially for the HH- and cross-polarizations of L-band. It was also found that the radar backscatter signatures at high frequency (X- and C-bands) are sensitive to detect newly transplanted thin wheat seedlings at the high incident angle.

[35] In summary, our study collected detailed radar backscattering coefficients and phenological parameters during the entire wheat growing period. The obtained results clearly show that backscattering coefficients are highly correlated with some wheat growing parameters.. The results also show this useful information can interpret the backscattering signatures over the wheat area. In addition, the numerical modeling methods [Cookmartin et al., 2000; Prasad, 2009] were applied in our data set to compare with the wheat cropping areas monitoring by SAR image [McNairn and Brisco, 2004; Satalino et al., 2009]. Thus, our data set is helpful for the research on the relationship between microwave radar backscatter and wheat geometrical and functional variables, including absorbed photosynthetically active radiation of wheat canopy. This would be useful to apply the geometrical models to the data set for such investigations in the near future.

Acknowledgments

[36] This work is supported by the National Natural Science Foundation of China (41071222 and 41271434), and Advance Research Program of Civil Aerospace Technology of “12th Five-Year,” and also partially supported by Hong Kong General Research Fund (GRF) of Research Grants Council (RGC) (CUHK 457212), Hong Kong Innovation Technology Find (GHP/002/11GD), and the National Key Technologies R&D Program in the 12th Five Year Plan of China (Applied Remote Sensing Monitoring System for Water Quality and Quantity in Guangdong, Hong Kong and Macau, 2012BAH32B03).

Ancillary

Advertisement