Retrieval of crop parameters by microwave remote sensing in a conventional manner is done by using one-dimensional data sets. Referring to the work of Lopez-Sanchez & Ballester-Berman (2009) it is evident that higher dimensionality SAR data is needed to
=
* 2 * 22
VV VV HH VV HH HHS
S
S
S
S
S
C
33 describe the complex nature. For this purpose higher dimensionality can be achieved by acquiring multi-frequency, multi-polarization, multi-angle or multi-temporal data. Multi-dimensionality increases the number of data layers that can be used for relation to crop parameters (McNairn & Brisco 2004).
2.2.3.1 Rice Crop Monitoring*
As a cereal grain, rice is the most important staple food for a large part of the world. For this reason, monitoring its biophysical variables is valuable for agricultural management and yield prediction. In 2009, about 1.61 million km2 of the earth’s surface was used for rice cultivation with a global production estimated at 679 million tons (FAOSTAT 2013). Beside its function as source of food, it is also important as a source of income. As population increases in most of the Asian countries, there is a great demand for effective rice monitoring with high reliability (IRRI 2011).
Information extracted from remotely sensed data can assist in estimating key plant
growth parameters such as biomass, crop height and leaf area index (LAI). In the past, optical satellite data have been successfully used for rice plant parameter
estimation (Tennakoon et al. 1992). For mapping rice cultivation in Asia, time series of vegetation indices (e.g., NDVI) derived from different sensors such as MODIS (Peng et al. 2011) were applied. However, operational crop monitoring and yield prediction based on optical remote sensing is hindered by unfavourable atmospheric conditions, which can lead to data gaps especially during critical growth stages.
As compared to optical sensors, spaceborne Synthetic Aperture Radar (SAR) instruments can overcome inherent limitations of optical systems owing to its all- weather, day and night acquisition capabilities. This allows a more reliable and consistent rice monitoring during the growing season. Especially short wavelength SAR (X- and C-band) interacts with the upper part of the crop canopy, thus offering the potential to retrieve crop biophysical parameters (Ulaby et al. 1984). Compared to this, longer wavelength (L-band) provides a deeper penetration into the vegetation and hence a higher sensitivity to overall plant biomass (Brisco & Brown 1998). To benefit
34 from both optical and SAR data, there are investigations that use complementary information from both systems, e.g., for crop type mapping (Blaes et al. 2005) or crop condition estimation (Koppe et al. 2010a).
A considerable number of research projects have been set up to investigate the capability of microwave data for agricultural monitoring since the first SAR satellites have been available for scientific and commercial use. Across all frequencies and crop types, the following aspects have been addressed:
• soil moisture retrieval (Gherboudj et al. 2011, Koyama et al. 2010)
• SAR backscatter analysis as a function of crop biophysical parameters and their temporal change (Bouvet et al. 2009, Shao et al. 2001)
• theoretical modelling of backscatter to support interpretation of the observations (Chen et al. 2005)
• development of methods for crop type mapping (Ribbes & Le Toan 1999a, Zhang et al. 2009)
• crop parameter estimation (Karjalainen et al. 2008, Jinsong et al. 2007)
• integration of SAR data in crop growth model for yield estimation (Shen et al. 2009, Ribbes & Le Toan 1999b)
Results of the mentioned studies confirm that microwave backscatter is highly sensitive to different crop types and to changes in the crop canopy due to increasing biomass during the growing cycle. The degree of sensitivity is strongly dependent on the applied polarizations, as identified by quadpol analysis (Wu et al. 2011). Despite good results in crop monitoring, it has to be considered that the recorded SAR backscatter from a vegetated surface is a function of several physical properties. These are crop type, surface roughness, soil moisture, vegetation structure and plant moisture content as well as sensor configuration (e.g., frequency, polarization and incidence angle). Furthermore, the different cultivation practices are important for rice monitoring. Lam-Dao et al. (2009) reported various backscatter behaviours for direct sowing of rice into wet soil in comparison to traditional transplanting techniques. Besides parameter estimation based on direct inversion from the recorded signal or
35 integrating SAR into growth modelling, there also have been promising results by using repeat-pass SAR interferometric coherence with one day offset for vegetation biomass estimation (Blaes & Defourny 2003). Reasonable results have been already achieved using Polarimetric SAR Interferometry (POLInSAR) for rice biophysical parameter retrieval with indoor wide-band polarimetric measurements (Ballester-Berman et al. 2005). In the near future, Polarimetric SAR Interferometry for crop monitoring with single pass will be demonstrated by the TanDEM-X mission (Hajnsek et al. 2010). The scattering process and penetration depth into the canopy is highly dependent on the wavelength and the incidence angle (Lim et al. 2007). Inoue et al. (2002) identified typical multi-temporal backscatter signatures of rice for frequencies at around 35, 15, 10, 5 and 1 GHz and at different incidence angles. In terms of electromagnetic interaction between microwaves and canopy, the received radar backscatter is a sum of three main components, including volume scattering, the double bounce scattering from the vegetation–surface interaction and the contribution from the surface itself. At the X-band, experiments conducted by Kim et al. (2000) using ground-mounted scatterometer data have demonstrated that the co-polarised backscatter from a paddy rice field at the beginning of the growing season is dominated by double bounce scattering from the stem–surface (water) interaction. With increasing plant density, the double bounce scattering is replaced by a random scattering from the upper canopy. Inoue et al. (2002) mentioned a typical dual-peak trend for higher frequencies; the first peak at the maximum of double bounce scattering and the second peak with appearance of the top leaf and the heads in top layer of the canopy.
For rice crops, the temporal backscattering behaviour has been extensively reported and understood in a number of studies based on spaceborne C-band data mentioned above. Comparatively to C-band data, much less effort has been put on the use of spaceborne X-band data in rice application. This is mainly due to lack of spaceborne X- band systems in the last decades. With the launch of TerraSAR-X and Cosmo Skymed in 2007, X-band data gained interest for rice monitoring. Lopez-Sanchez et al. (2010) adapted an electromagnetic model to simulate X-band backscatter from rice field. It was used for interpretation of dual-polarised TerraSAR-X images over rice fields in
36 Spain. Suga & Konoshi (2008) investigated the temporal change of SAR backscatter during the rice growing cycle.
2.2.3.2 Winter Wheat Crop Monitoring*
China cereal acreage and production is one of the most important in the world, with a crop area of about 88 million ha and production estimated at 483 million tonnes in 2009, accounting for ca. 22% of total global production (Fao 2013). The North China Plain is one of the most important cereal production regions in China, accounting for almost 50% of China's winter wheat cultivation (National Bureau of Statistics of China, 2010). In agricultural issues, timely monitoring of crop growth status at an early stage is important for in-season site specific crop management, detection of plant vitality as well as assessment of seasonal production at local and regional level (Miao et al. 2009, Laudien & Bareth 2006).
Since the amount of energy backscattered towards the sensor strongly depends on dielectric properties and surface roughness, it is reasonable that SAR can be used for crop type classification, growth stage mapping and crop condition monitoring (McNairn & Brisco 2004). For different applications, knowledge of the interaction of the surface characteristics with sensor configurations such as resolution, frequency, incidence angle and polarization is of importance (Inoue et al. 2002). For C-Band SAR measurements, many studies dealt successfully with prediction of crop and soil parameters such as biomass, crop height and soil moisture (Brisco & Brown 1998), but the interpretation of the SAR backscatter has proven to be complicated.
In the past, quite a few experiments have been performed on wheat fields, either based on spaceborne SAR sensors or on ground-based scatterometers. Satalino et al. (2009) and Brown et al. (2003) acquired C-band spaceborne and scatterometer data over wheat fields and found that wheat biomass is strongly related to HH/VV backscatter during the whole growing season. The good performance of the HH/VV ratio is due to the differently attenuated vertically and horizontally polarised waves that propagate through a mainly vertical medium of wheat (Picard et al. 2003).
37 McNairn et al. (2004) differentiated zones of productivity of wheat fields also using scatterometer data. They reasoned that zones of higher productivity had higher backscatter for linear polarizations, with the greatest contrast for HV.
There are also many investigations on wheat’s crop parameter retrieval and crop classification based on spaceborne C-band sensors (Baghdadi et al. 2010, Mattia et al. 2003). The results from these studies showed that the backscattering of crops is a complex combination of acquisition parameters (polarization, incidence angle) as well as crop and cultivation characteristics (crop geometry, density, canopy and soil moisture). The combination of these parameters controls the interaction of the incoming electromagnetic wave with the crop canopy and the underlying soil layer. Because the backscatter is a function of SAR and crop related physical properties, the crop parameter estimation is an inversion of the recorded signal with many unknown parameters. Under consideration of these specifics of interaction, crop parameters related to crop growth could be reproduced by SAR backscatter, such as crop height (Chakraborty 2005), standing biomass (Liu et al. 2006) and LAI (Lin et al. 2009).