Rice Mapping Using RADARSAT-2 Dual- and Quad-Pol Data in a Complex Land-Use Watershed: Cau River Basin (Vietnam)

Abstract

The strong temporal backscatter signature of rice growing above the water's surface allows for the use of synthetic aperture radar (SAR) for paddy rice crop mapping in Southern Vietnam (Mekong Delta). In Northern Vietnam (Red River Delta), rice mapping using SAR is a challenge and is rarely performed because of the complex land-use/land-cover. Nevertheless, information about rice fields is needed for hydrological simulations in river basins such as the Cau River basin. The objective of this research is to investigate the potential of RADARSAT-2 band- C in identifying rice fields over a large and fragmented land-use area. Two methods are proposed, one for each data type, adapted to the land-use/land-cover of the study area. The thresholding technique, with a statistical analysis of the temporal variation of rice backscattering, was applied to the HH like-polarized ratio of dual-pol data. The support vector machine (SVM) algorithm was applied to the full quad-pol and a single HH-polarization calculated from polarimetric data. This study demonstrates that RADARSAT-2 dual- and quad-pol data can be successfully used to identify cultivated rice fields. However, the dual-pol data seems less efficient than the quad-pol data and the SVM classification is more flexible than the thresholding technique. Between the full quad-pol and a single polarization, the overall classification accuracy shows that the results derived from the single HH polarization are 3 to 10% less accurate than those derived from the classification of full quad-pol data. The results show the usefulness of polarimetric C-band data for the identification of rice fields in Northern Vietnam.

Keywords

• Cau river basin
• RADARSAT-2
• rice identification
• support vector machine (SVM)
• thresholding