Theoretical Modeling and Measurement Comparison of Season-long Rice Field Monitoring

The development of a theoretical model to describe the scattering mechanisms involved in the remote sensing of rice crops is essential, as it ensures correct application of remote sensing data for rice monitoring. The theoretical model used in this study is based on the radiative transfer theory applied on a layered dense discrete random medium. The dense medium phase and amplitude correction theory (DM-PACT), which considers the coherent effects of the scatterers, is incorporated in the development of the phase matrices of the scatterers, which are modeled after the physical geometry of the plants. Ground truth measurements of rice fields were acquired at Sungai Burung, Selangor, Malaysia for an entire season. These measurements are used in the theoretical model to calculate the backscattering coefficients of rice fields. The results are then compared to those obtained from RADARSAT images to test the validity of the model. Comparisons show promising results, but further research is required to improve on the current model. Introduction In recent years, there has been a lot of international interest in the use of microwave remote sensing for rice field monitoring and yield prediction applications. Initial studies [1-3] have shown that earth observation satellites such as ERS-1 and RADARSAT can be used to classify rice covered areas from non-rice areas due to the high temporal variations in the backscattering coefficient of rice fields. In addition to that, the backscattering information allows the growth stage of the rice plants to be determined. However, the actual interaction between electromagnetic waves and rice crops still remains relatively unknown. There is therefore a need to develop a theoretical model that will enable us to understand the scattering mechanisms involved when electromagnetic waves interact with rice crop canopies. This theoretical model will ensure correct application of remote sensing data, as well as allow the retrieval of physical parameters of rice crops using inversion algorithms. Ground Truth Measurements Ground truth measurements of rice fields were obtained at regular 12 day intervals between 27 August 2004 to 1 December 2004 at Sungai Burung, Selangor, Malaysia. These measurements were acquired from 6 different test fields in the region. Parameters that were measured include plant geometry (such as plant height, leaf length, leaf width, leaf thickness and leaf inclination angle), plant density, plant gravimetric water content and plant biomass. These measured parameters were then used to calculate and obtain other parameters for the theoretical model. RADARSAT images were acquired on the 27 of August, 20 of September, 14 of October and 6 of November of 2004, which coincide with four ground truth measurements. The RADARSAT operates at 5.3 GHz (C-Band), and all the images were obtained using Fine Mode 2. Theoretical Modeling In this study, the theoretical model is developed based on the radiative transfer theory [4], which describes the change in intensity of an electromagnetic wave due to scattering and absorption as it travels through an inhomogeneous medium, and is given by: cos θ dĪ dz = −κeĪ + ∫ P Ī dΩ (1) where Ī is the Stokes vector, κe and P are the extinction matrix and phase matrix of the medium respectively. This equation is solved iteratively up to second order in both the upward and downward directions. The phase 26 Progress In Electromagnetics Research Symposium 2005, Hangzhou, China, August 22-26 matrices of the rice canopies are developed using the generalized Rayleigh-Gans approximation to obtain the scattered fields from needle shaped and cylindrical scatterers [5] with Fresnel phase corrections being considered by including the higher order terms in the expression of the scattered fields [6]. The dense medium phase and amplitude correction theory (DM-PACT) [7, 8] has also been incorporated to include the coherent effects of closely packed scatterers, by multiplying an array phase correction factor to the Stokes matrix to obtain the phase matrix of the medium. The phase matrix is thus given by: P (θs, φs; θi, φi) =< |ψ| >n ·S(θs, φs; θi, φi) (2) where S is the Stokes matrix and < |ψ2| >n is the array phase correction factor given by: < |ψ| >n= 1− e−k2 siσ2 d3 + e−k 2 siσ 2 d3 ∞ ∑ q−1 (k siσ ) q! [( √ π q ( l d ))exp( −k2 sil 4q )− a(kx)a(ky)a(kz)] (3) where: ksi = |ks − ki| and a(kr) = √ π q ( l d )exp( −k2 r l 4q )Re{erf( (qd/l) + jkrl 2 √ q )} ki and ks are the propagation vectors in the incident and scattering directions, l is the array correlation length, d denotes the average distance between scatterers and σ is the standard deviation of scatterers from their mean positions. The rice canopy is modeled as either a single layer or multilayer dense discrete random medium, depending on its growth stage, over a smooth water surface. Table 1 shows the different models used for the different growth stages of the rice crops corresponding to its age and dates of RADARSAT image acquisition. In its early vegetative stage, corresponding to the RADARSAT image obtained on the 20 of September, the rice model consists of a single layer of needle-shaped scatterers, in the consideration of the uniform orientation distribution of the rice leaves. For the image obtained on the 14 of October, the rice plants are now in their late vegetative stage, and the canopy is modeled as a double-layer medium. The upper layer consists of needle shaped leaves, while the lower layer is a combination of needle shaped leaves and cylindrical stems. During the reproductive stage, tiny cylinders are added to the upper layer of the model to simulate grains. This corresponds to the RADARSAT image acquired on the 6 of November. The RADARSAT image obtained on the 27 of August will not be included in this study as the seeds have just been broadcasted and the only source of backscattering is the soil. Test fields 2 and 3 have also been omitted due to incomplete data collection as a result of heavy rains and partial destruction of rice fields respectively. The list of parameters used in the model is shown in Table 2. The water surface is assumed to be flat and smooth. The dielectric constants of water and rice plants are calculated from the equations given in [9]. The standard deviation of scatterers from their mean positions is chosen to be 0.5d, where d is the average distance Table 1: Various models used for the different growth stages of rice plants corresponding to plant age and date of RADARSAT image acquisition Date Test Field Age (days) Growth stage Model Scatterers 20/9/04 1 27 early vegetative single layer needles 4 26 early vegetative single layer needles 5 29 early vegetative single layer needles 6 21 early vegetative single layer needles 14/10/04 1 51 late vegetative double layer needles, stem cylinders 4 50 late vegetative double layer needles, stem cylinders 5 53 late vegetative double layer needles, stem cylinders 6 45 late vegetative double layer needles, stem cylinders 6-11-04 1 75 early reproductive double layer needles, stem cylinders,grain cylinders 4 74 early reproductive double layer needles, stem cylinders,grain cylinders 5 77 early reproductive double layer needles, stem cylinders,grain cylinders 6 69 early reproductive double layer needles, stem cylinders,grain cylinders Progress In Electromagnetics Research Symposium 2005, Hangzhou, China, August 22-26 27 between the scatterers. The angle distribution parameters for the leaves, stems and grains are based on the equation in [10]. Table 2: Model input parameters Model Input Parameters Values array standard deviation of scatterers 0.5d array correlation length 1.0d radius and length of leaves, stems and grains according to test field measurements volume fraction of leaves, stems and grains according to test field measurements layer heights according to test field measurements plant dielectric constant (at 5.3 GHz) according to test field measurements leaf, stem and grain angle distribution parameters according to test field measurements Results and Comparisons -14 -12 -10 -8 -6 -4 -2 0 20 30 40 50 60 70 80 Test Field 1 RADARSAT model B a c k s c a tt e ri n g C o e ff ic ie n t (d B ) Plant Age (Days) -14 -12 -10 -8 -6 -4 -2 0 20 30 40 50 60 70 80 Test Field 4 RADARSAT Model B a c k s c a tt e ri n g C o e ff ic ie n t (d B )