多云多雨区水稻熟制信息高精度提取方法

Current applications on rice paddy mapping use either high temporal resolution data that has low spatial resolution or high resolution spatial data with limited temporal information. However, these applications are not sufficient for areas that are affected by clouds and fragmentation. Therefore, this project aims to integrate observations from multiple sensors using the probabilistic approach and iterative Bayesian approach on multiple high resolution remote sensing data to map rice cropping systems at cloud-prone areas, including Landsat8, Sentinel-1/2, and Gaofen-1/3. First, we employ the random forest classifier on the integration of multiple sensor information by evaluating the importance of the optical vegetation indices and the synthetic aperture radar (SAR) signatures, in order to identify rice distribution and its spatial pattern. Furthermore, we combine multiple optical and SAR satellites using a probabilistic approach to develop the probability density function of sensor-specific harvesting to improve the near real-time monitoring. Lastly, we use iterative Bayesian updating to detect the cropping intensity. The method developed in this project is a novel approach integrating multiple sensor information that increases the accuracy of mapping rice cropping system. The results of this method can provide accurate cropping intensity data at cloud-prone areas for effective crop monitoring and planning.

我国南方水稻种植区多云多雨、地块破碎，现有的水稻熟制信息提取方法大多采用单一高时间或高空间分辨率数据，难以准确获取水稻熟制信息，无法满足应用需求。针对这一问题，本研究拟通过融合多个光学数据(Landsat8、Sentinel-2、GF-1)和SAR数据(Sentinel-1、GF-3)，开展基于概率度量和概率迭代优化的多云多雨区水稻熟制信息提取的方法研究。首先，对多源数据进行特征提取与优化，结合随机森林分类进行水稻空间分布制图；其次，利用概率度量，重构高时间分辨率准实时对地观测数据，对水稻收获特征进行概率表达；最后，通过概率迭代优化的方法，量化不同熟制水稻收获频数，准确提取水稻熟制信息。本研究可为多云多雨区水稻熟制信息的提取提供精度可靠的方法，为水稻面积监测和农作物种植结构调整提供更准确的信息。

我国南方水稻种植区多云多雨、地块破碎，现有的水稻熟制信息提取方法大多采用单一高时间或高空间分辨率数据，难以准确获取水稻熟制信息，无法满足应用需求。针对这一问题，本项目充分利用多种光学数据和SAR数据，进行关键算法和技术创新，提出了农作物时相信息和空间信息特征优化方法，建立了特征选择度量，构建了高时间高空间遥感观测数据集，提出了多云多雨区水稻种植信息提取的方法，进行水稻等作物种植区域判定和种植频率提取，量化不同熟制水稻收获频数，破解了云雨频发区域水稻精准识别的技术难题，为多云多雨区水稻熟制信息提取提供精度可靠的方法。结果表明，本项目基于Landsat、Sentinel、GF等多个光学数据和SAR数据的融合应用能够很好的获取水稻熟制信息。和传统的分类方法相比，本项目方法能够更有效利用多源遥感数据，提升水稻熟制信息提取精度，可为国家种植结构调整和政策制定提供数据支撑。在本项目的支持下，发表论文6篇（项目主持人第一作者或通讯作者发表论文4篇），其中SCI论文4篇；获得授权专利1项。