In-Season Biomass Estimation of Oilseed Rape (Brassica napus
L.) Using Fully Polarimetric SAR Imagery
Hao Yang
1,2,3
• Guijun Yang
1,
* • Rachel Gaulton
2
• Chunjiang Zhao
1,
* • Zhenhong Li
2
• James
Taylor
4
• Daniel Wicks
5
• Andrea Minchella
5
• Erxue Chen
3
• Xinting Yang
1
Abstract Accurate estimation of crop biophysical and biochemical parameters during crop
growing seasons is essential for improving site-specific management and yield estimation. The
potential ability of fully polarimetric SAR (Synthetic Aperture Radar) data in estimating above-
ground biomass of oilseed rape was investigated in this study. The temporal profile of different
scattering intensity and polarimetric features during the entire growing season was identified with
ground measurements. A polarimetric feature, relying on the polarimetric decomposition method,
was put forward to estimate the biomass of oilseed rape. Validation results revealed great potential
with a determination coefficient (R
2
) of 0.85, Root Mean Squared Error (RMSE) of 41.6 g/m
2
, and
Relative Error (RE) of 28.5% for dry biomass, and an R
2
of 0.76, RMSE of 527.4 g/m
2
and RE of
28.6% for fresh biomass. Moreover, the use of full polarization SAR data was compared with
single and dual polarization SAR data. The results suggest that when full polarization SAR data is
available, a simpler model, higher saturation point and better accuracy can be achieved in biomass
estimation of oilseed rape, which highlights the importance and value of polarimetry information
in quantitative crop monitoring. This study provides guidelines for in-season monitoring of crop
growth parameters with SAR data, which further improves crop monitoring capability in adverse
weather conditions.
Keywords: RADARSAT-2; biomass; polarimetric features; polarimetric decomposition; rapeseed
Contact: Guijun Yang, Chunjiang Zhao
yanggj@nercita.org.cn, zhaocj@nercita.org.cn
1
Beijing Research Center for Information Technology in Agriculture, Beijing Academy of
Agriculture and Forestry Sciences, Beijing 100097, China
2
School of Civil Engineering and Geosciences, Newcastle University, Newcastle upon Tyne, NE1
7RU, UK
3
Institute of Forest Resource Information Techniques, Chinese Academy of Forestry,
Beijing 100091, China
4
School of Agriculture, Food and Rural Development, Newcastle University, Newcastle upon Tyne,
NE1 7RU, UK
5
Satellite Applications Catapult, Didcot, Oxfordshire, OX11 0QR, UK
Introduction
Oilseed rape (Brassica napus L.) is produced as a source of edible vegetable oils, biodiesel and animal
feeds (Ren et al. 2015) and is the most important oil crop in China. The Food and Agriculture
Organization of the United Nations (FAO) reports that China is the largest oilseed rape cultivated region
in the world, in terms of both planting area and yield (FAO 2013). Accurate estimation of crop
biophysical and biochemical parameters during the growing season is essential for improving site-
specific management and yield estimation (Jin et al. 2015), among which above-ground biomass is the
most important indicator for crop yield (Tilly et al. 2015). Therefore, quantitative monitoring of oilseed
rape biomass across large areas will not only provide information for crop growth and precision
agriculture, but also support production forecasts and encourage competition in the international edible
oil market.
Crop biomass is traditionally estimated through destructive and time-consuming in situ methods,
which are difficult to conduct when crops cover large regions (Jin et al. 2015). Remote sensing methods
offer several advantages for monitoring biomass in a timely within-season manner and a number of
studies have developed and validated methods to estimate biomass using optical imagery (Hosseini et
al. 2015, Kross et al. 2015, Bendig et al. 2014, Hunt et al. 2005, Ahamed et al. 2011). A review of remote
sensing methods for assessing biomass is given by Ahamed et al. (2011). Although these studies have
demonstrated the successful retrieval of biomass, almost all of them are based on optical remote sensing,
which faces the problem of frequent cloud or haze cover. Bad weather limits the acquisition of high
quality optical remote sensing data. It further hampers the frequency of data update, which is vital given
that crop vegetation grows and changes rapidly during certain growth stages. In addition, saturation is a
common problem in optical data for dense crop coverage, because the vegetation indices lose sensitivity
when the leaf area index (LAI) exceeds 2-5 (Gao et al. 2013, Ahamed et al. 2011).
Synthetic aperture radar (SAR) imagery provides a complementary and unique characterization of
vegetation relative to that achieved with optical data, making its use of particular significance and
interest for crop monitoring (McNairn et al. 2014, 2009, McNairn and Brisco 2004). Due to wavelengths
being in the microwave region and are actively generated by the sensor itself, SAR systems are
unaffected by the presence of clouds/hazes and are therefore able to collect earth surface data irrespective
of weather or light conditions. More importantly, due to its penetration ability, SAR can penetrate crop
canopies and has potential in overcoming the saturation limitations of optical data for dense vegetation
measurement. These advantages, along with the sensitivity of microwave scattering to soil and
vegetation characteristics, such as structure, moisture content and biomass, have led researchers to
investigate the application of SAR for agriculture monitoring (Yang et al. 2015, McNairn and Brisco
2004, Lopez-Sanchez and Ballester-Berman, 2009). Despite these advantages, SAR images are affected
by the sensors imaging geometry and radiation mechanism as well as speckle effects which present
challenges in data interpretation.
The potential of radar backscatter to estimate crop growth parameters has been studied for many
crop types, including paddy rice (Inoue et al. 2014, Lopez-Sanchez et al 2014), wheat (Kim et al. 2014,
Jin et al. 2015), barley (Fontanelli et al. 2013, Cable et al. 2014), sugarcane (Baghdadi et al. 2009), maize
and soybean (Gao et al. 2013, Jiao et al. 2011), using airborne and/or space-borne sensors (Tanase et al.
2014). A number of retrieval strategies have been adopted including empirical models (Wiseman et al
2014, Jin et al. 2015, Palosica 2002) that related radar backscatter to crop biophysical parameters, semi-
empirical approaches (Beriaux et al. 2015, 2013, Lin et al. 2009), which have frequently been based on
the water cloud model (Attema and Ulaby, 1978), and more recently, numerical models (Zhang et al.
2014, Jia et al. 2014, 2013) that are based on machine learning algorithms or complex vegetation
scattering models. Although a consensus on the best models for biomass retrieval is yet to emerge, most
of these models make use of SAR scattering intensity information, which is based on single polarization
or dual polarization channels.
Polarimetric information is a unique property of SAR observations that has not received deserved
attention in quantitative crop monitoring. Fully or quad polarimetric (quad-pol) SAR systems have the
advantage of enabling a complete description of the scattering process, thus providing information on
the entire scattering matrix that can be decomposed into surface, volume, and double-bounce
contributions (Cloude 2009, Lee and Pottier, 2009). The quad-pol system, alternatively transmitting two
orthogonal polarizations and receiving in both polarizations simultaneously, has many advantages over
single polarization SAR systems. With increased access to polarimetric SAR data, i.e., RADARSAT-
2(Canada, launch 2007), TerraSAR-X(Germany, launch 2007), ALOS PALSAR(Japan, launch
2006/2014), Sentinel-1(ESA, launch 2014), and GF-3(China, launch 2016), the use of polarimetric
information becomes feasible for crop monitoring. Most polarimetric decomposition studies have been
undertaken for classification purposes (McNairn et al. 2014; Qi et al. 2015) rather than direct biomass
estimation (Tanase et al. 2014). To the best knowledge of the authors, the use of specific scattering
components for model parameterization and direct biomass estimation has only been investigated
recently with great uncertainty (Wiseman et al 2014, Tanase et al. 2014, Zhang et al. 2012). Quantitative
crop biomass estimation from fully polarimetric SAR data remains to be explored.
Therefore, the objectives of this study were to: (1) evaluate whether fully polarimetric data is
necessary for quantitative crop monitoring, as opposed to single or dual polarimetric data; (2) develop
an operational method for the estimation of biomass of oilseed rape (Brassica napus L.); and (3)
investigate the use of polarimetric target decomposition components for direct model parameterization.
Relevant issues regarding crop biomass retrieval error are also discussed.
Materials and methods
Study Area
The study region is located in the Shangkuli agricultural area (50°17′ to 50°23′ lat. N; 120°46′ to
120°53′ long. E), Hailar City, Inner Mongolia, China (Fig. 1). This area, typical of agricultural land use
for Northeast China, lies on the transition zone between forest and prairie (northwest of the Greater
Khingan Mountains and north of the Hulunbuir steppe). It has a cold temperate continental monsoon
climate, i.e., cold and dry in the long winter and warm and wet in the short summer, supporting one
arable harvest per year. The soil is classified as Leached Chernozem with sand clay silt texture (sand =
5.86%, clay = 42.08%, silt = 52.06%), based on three soil samples collected in the study area.
Topographic variations are minimal with slopes generally less than 1% in arable areas. Approximately
90% of the farmland is used for annual crop cultivation, principally oilseed rape and spring wheat. This
farm is managed by a single entity. Crops are usually sown from May to June, reaching maturity in the
middle of August, and harvested in September. The arable area extends over 3000 ha, with simple sowing
patterns and homogeneous field parcels. A set of 88 oilseed rape fields was selected for this study, all sown
with a near constant east-west row direction. The area of fields ranges from 3.3 to 47.0 ha, with a median
of 18.6 ha.
Fig. 1 Location of the Shangkuli agricultural area and the oilseed rape fields; Background: Pauli-
basis RGB image of RADARSAT-2, acquired on 16 June 2013.
SAR data and ground truth data
Canada’s next-generation commercial radar satellite, RADARSAT-2 is an Earth observation satellite,
launched in December 2007 for the Canadian Space Agency by Starsem. RADARSAT-2 carries a C-
band (~5.3 GHz) SAR sensor with multiple polarization modes, including a fully polarimetric mode in
which HH, HV, VV and VH polarized data are acquired (H-Horizontal, V-Vertical, i.e., HV denotes
transmitting a Horizontal linear polarization and receiving a Vertical linear polarization). RADARSAT-
2 is a follow-on to RADARSAT-1. It has the same orbit (798 km altitude sun-synchronous orbit) and 24
days repeat cycle.
Five consecutive fully polarimetric RADARSAT-2 images were acquired in repeat pass with 24-day
intervals during the growing season in 2013. All of them were acquired with the same imaging mode,
including beam mode, incidence angle and orbit direction, in order to build a time series in the most
consistent way (Table 1). Their acquisition periods cover the most critical growth stages of the crop,
from sowing to harvest, as shown in Table 2. Original RADARSAT-2 images were provided in single
look complex (SLC) format with pixel spacing of 4.73 m and 4.96 m in azimuth and slant range
directions, respectively.
Table 1 Main parameters of RADARSAT-2 images.
Parameter
Values
Imaging Mode
Fine Quad Polarization
Center frequency
5.405 GHz
Incidence angle
37.4°–38.8°
Resolution
7.6m × 5.2m (diffrent from
pixel spacing)
Swath width
25 km
Orbit direction
Ascending
Beam mode
FQ18
Acquired Time
UTC 09:47:33
Table 2 Acquisition dates of five RADARSAT-2 images and corresponding phenological stages
of oilseed rape.
Acquisition Dates
Principal Growth Stage
23 May 2013
Germination (0)
16 June 2013
Leaf development (1) and formation of
side shoots (2)
10 July 2013
Stem elongation (3), inflorescence
emergence (5), and flowering (6)
03 August 2013
Development of fruit (7)
27 August 2013
Ripening (8) and senescence (9)
During the seeding period in 2013, the sowing dates of all the 88 fields in this study site were
recorded. The sowing period lasted from 8 May to 31 May 2013, with 17 different sowing dates and 24
days duration. Five synchronous ground measurement campaigns were carried out within ±24 h of
each RADARSAT-2 acquisition time. In each campaign, the same 11~14 representative oilseed rape
fields were surveyed from the 88 fields. Quantitative agronomic and biophysical parameters were
measured, including leaf area index (LAI), plant height, surface soil moisture, above-ground biomass
(fresh and dry weight per square meter) and vegetation water content. LAI (m
2
/m
2
) was randomly
measured at three sample sites within each field using a LI-COR LAI-2200 instrument
(https://www.licor.com/env/products/leaf_area/LAI-2200C ). The top 76 mm soil volumetric water
content (m
3
/m
3
) was also randomly measured at three sample sites within each field using the FieldScout
TDR 300 (https://www.specmeters.com/soil-and-water/soil-moisture/fieldscout-tdr-meters/ ) Soil
Moisture Meter in High Clay mode (since the clay content in this site is high) after calibration. At each
of the three sample sites, five soil moisture readings were taken resulting in 15 moisture readings per
field. One representative biomass sample was collected within the field for each field. All the above-
ground vegetation in a square of 0.5 m × 2 rows was harvested and weighed. The row distance was
measured to obtain the sample area. All biomass samples were dried in an oven at 95 °C for at least 48h
and then reweighed, providing dry biomass (g/m
2
) as well as plant water content ((wet−dry)/wet, g/g).
During the ground surveys, photographs of the crop development were taken as a reference relative to
the key growth stages for oilseed rape (Fig. 2). Daily precipitation, air temperature, humidity and wind
data, and other meteorological parameters were recorded by an automatic meteorological station
(MidWest, WPH1-PH-6, http://detector.midwest-g.com/ WPH1-PH-6.html).
