Rice crop mapping using radar imagery: comparison of classification accuracy of different Envisat ASAR modes and classifiers

Abstract

Food security has become a key global issue due to rapid population growth in many parts of Asia, as well as the effect of climate change. For this reason, there is a need to develop spatio-temporal monitoring system that can accurately assess rice area planted. In the Mekong Delta, Vietnam, the changes in cultural practices have been gradually adopted in the last ten years. These changes have impacts on remote sensing methods developed for rice monitoring, particularly on the accuracy of the resulting classified image. Thus, the aim of this study was to compare the accuracy obtained by different Envisat ASAR modes (APP and WS) and the classifiers used. Using Envisat ASAR APP data, the study showed that the radar backscattering behaviour is much different from that of the traditional rice previously reported in the literature, due to changes brought by modern cultural practices. The polarisation ratio (HH, VV) of rice fields at a single date during a long period of the rice season could be used to derive the rice/non-rice mapping algorithm. The results of this thresholding algorithm achieved higher and consistent accuracies between seasons and districts (i.e. maximum accuracy of 99% and 98%, respectively) across the study site when compared to other classifiers, such as the minimum-distance-to-means, maximum likelihood, spectral angle mapper (SAM), ISODATA and K-Means. Regarding the ASAR modes of data acquisition, the ASAR APP data yielded higher and consistent classification accuracy. However, the ASAR WS product proved to be a potential data for rice mapping at regional scale.