Study On Rice Mapping And Rice Yield Estimation Based On ASAR Data

Rice is a staple food crop in the world and accounts for 15% of the world's total cultivated area. In Asia where 94% of the world rice is produced, rice is also an important source of income. However, as population continues growing dramatically in most Asia countries, like in China and India, there is an ever-increasing need in ensuring rice production to be sufficient for all our need. Hence, there needs to be an effective means for rice monitoring and production estimation with high-accuracy and low-cost. Since remote sensing data could be acquired from different sensors, there has been an increasing amount of international interest in rice monitoring through satellite. However, rice crop is mainly cultivated in warm tropical climates with plentiful rainfall and dense cloud cover throughout its growing season. Hence, such research activities entail use of microwave remote sensing, for microwave can penetrate through clouds and have all-weather capabilities. This allows for a more reliable and consistent rice monitoring and yield prediction in terms of radar sensor data in Asian countries.However, the principle and method to detecting land surface objects by radar sensor are quite different from that by optical sensors. Information contained in the radar images mainly reflects the object's dielectric characteristic, surface roughness, geometric and directional features, which also related to the different configuration of radar systems, like polarization, wave length, incidence angle and resolution. In addition, SAR images contain strong speckle noises that increase the difficulty in quantitative and qualitative analysis. Radar satellites, such as ERS-1/2 and RadarSat have been used in rice monitoring for decades, but a large amount of reach activities have been focused on rice mapping, rice parameters inversion and study on rice backscattering mechanism, few reports have been related to the rice production estimation by radar data.In this paper, the ASAR data from ENVISAT satellite is employed with the measurements from a field campaign in Xinghua study area in 2006 to provide a deep analysis and study on rice mapping, rice parameters inversion and rice yield estimation. For rice mapping, the principle is studied in order to bring forward the optimal temporal parameters for rice mapping with multi-temporal and single-temporal ASAR data respectively. Validation is also conducted while comparing two different rice mapping methods. For rice parameters inversion, two different types of rice cloud model are described and compared and one of them is used to analyze the relationship between modeled rice backscattering coefficients and rice growth parameters, such as wet biomass, LAI, and canopy water content. The relationship between HH/VV and rice growth parameters is also analyzed to examine the efficiency of rice parameters inversion by ASAR data. At last, a scheme for rice yield estimation and mapping rice yield by coupling ASAR data with rice crop model is proposed and validated by field measurements. In conclusion, the main development and results as follows:1. For the situation of rice mapping with multi-temporal ASAR data, the optimal temporal parameters are the one in the early stage and the other one in the middle stage during the rice growth period, with which the rice mapping accuracy can reach more than 85%. However, while the rice mapping with single-temporal ASAR data, the optimal temporal parameter is the one in the middle stage during the rice growth period.2. In the rice mapping, both of the rice mapping methods, the supervised classification method based on Maximum Likelihood and the Decision Tree classification method has similar rice mapping efficiency and accuracy. However, SAR image suffers from speckle noises, a better rice mapping efficiency and higher rice mapping accuracy can be reached by the rice mapping method of the supervised classification based on Maximum Likelihood since the probability distribution of the image filtered by speckle noise filters runs towards the normal distribution which is more suitable for the classification method based on Maximum Likelihood.3. The result for rice parameters inversion by rice cloud model shows that the semi-empirical rice cloud model calibrated in HH polarization is better than that in VV polarization, since HH polarization is more sensitive to the change of rice biomass and LAI and rice canopy water content. Meanwhile, HH/VV shows a better relationship with the temporal change of rice biomass and LAI, but it also shows a significant relation with rice plant height and rice plant water content before rice heading stage.4. A scheme for rice yield estimation and mapping rice yield by assimilating ASAR data into rice crop model is presented in this paper and validated by observations from a field campaign in Xinghua study area in 2006. Due to the potential rice growth condition were assumed in this study, the result shows that the obtained rice yield map generally overestimates the actual rice production situation, with an accuracy of 1133 kg/ha on validation sites, but the tendency of rice growth status and spatial variation of the rice yield are well predicted and highly consistent with the actual production variation, which proves that the scheme described in this study is a promising technique to apply multi-temporal and multi-polarization radar data and rice crop models for regional rice production estimation, when no accurate in-situ information is available and/or optical data are hampered by heavy clouds during the rice season. However, further validation of the presented scheme at different rice planting areas and with different radar configurations (e.g. incidence angle, polarization) is needed.