Nomination-favoured opinion pool for optical-SAR-synergistic rice mapping in face of weakened flooding signals

• Published in: ISPRS journal of photogrammetry and remote sensing Vol. 155; pp. 187 - 205
• Main Authors: Guo, Yiqing, Jia, Xiuping, Paull, David, Benediktsson, Jón Atli
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.09.2019
• Subjects: algorithms; Consensus theory; crops; Data fusion; direct seeding; growers; growing season; monitoring; New South Wales; Opinion pool; Optical; Oryza sativa; planting; remote sensing; rice; Rice mapping; Rice variety; Sowing method; summer; synthetic aperture radar; Synthetic-Aperture Radar (SAR); time series analysis; water conservation; water utilization; New South Wales; Synthetic-Aperture Radar (SAR); Sowing method; Optical; Data fusion; Opinion pool; Rice mapping; Rice variety; Consensus theory
• DOI: 10.1016/j.isprsjprs.2019.07.008
• ISSN: 0924-2716

Using satellite image data it is possible to distinguish paddy rice from other crops by detecting its unique signals during the initial flooding period for crop mapping and growth monitoring. In recent years Australian rice growers have applied direct drilling sowing method to reduce water usage. This effort, however, has challenged existing algorithms for paddy rice planting area mapping, as the flooding signals have become weaker, especially for the newly introduced water-saving variety YRM 70. In order to alleviate this problem, an optical-SAR-synergistic rice mapping approach is proposed in the time domain in this study. Time-series images from both sensor types were collected in the rice growing seasons. Optical indices and SAR (Synthetic-Aperture Radar) features were generated and analysed. Pixel dependent multiple features are developed to reveal the strongest remote sensing signatures. As the information provided by optical and SAR data are complementary, a novel nomination-favoured opinion pool, NF-OP, is constructed where a pixel is identified as rice if either the optical or SAR data (or both) provide a relatively positive such opinion. In the research, mapping experiments were conducted for the Riverina region of Coleambally, New South Wales, Australia, during the 2016–2017 summer season. The experimental results suggest the following: (1) rice crops sown with the direct drilling method show weakened flooding signals due to the practice of flush floodings and the adoption of water-saving management strategies, especially for directly drilled YRM 70 rice; (2) conventional rice mapping algorithms using only optical or SAR data tend to generate high mis-detection rates (i.e., low true positive rates) for direct drilling rice as a result of weakened flooding signals; (3) by fusing the complementary optical and SAR data with the proposed NF-OP, the mis-detection problem is effectively alleviated, with the true positive rate for directly drilled YRM 70 rice being improved to 82.5% as compared to 54.5% or 44.0% obtained with optical or SAR data only; (4) an overall rice mapping accuracy of 94.7% is achieved with the proposed optical-SAR-synergistic approach, which is 9.5% and 11.5% higher than the respective optical-only and SAR-only algorithms.