Identifying paddy fields with dual-polarization ALOS/PALSAR data

Abstract

Accurate documentation of paddy-rice cultivation areas is valuable in estimating rice production for food security and in assessing the environmental impacts of rice ecosystems such as water consumption, soil degradation, and river eutrophication. This study explores the feasibility of dual-polarization L-band advanced land observing satellite/phased array-type L-band synthetic aperture radar (ALOS/PALSAR) imagery acquired during three growing stages in delineating paddy fields from other agricultural land uses. The study area was in the Yangtze River Delta of east China, a rapidly developing region where land has been intensively used. Among a set of arithmetic outputs of PALSAR backscatter, the amplitude ratio (HH/HV) and product (HH×HV) in the rice-transplanting and -heading stages significantly enhanced the backscatter difference between rice and nonrice fields. With these outputs, a segmentation-based decision-tree classifier successfully extracted paddy-rice fields from other land uses in the study area. Both random-point accuracy assessment and area comparison with a high-resolution Quickbird image showed that the paddy-rice map reached accuracies higher than 90%. The simplicity and generality of the approach in this study indicated that it may serve as an efficient tool for rice mapping in the highly fragmented agricultural region in southeast China.