半干旱区玉米估产的AquaCrop模型与遥感数据双变量同化方法研究

It is essentially significant to know the spatiotemporal variation of the growth status at the key crop growing stage, timely manage field crops, and then ensure stable crop production for country's food security. In the previous studies, many researchers have focused on the assimilation of one state variable, while the double state variables have seldom been reported. In addition, little was known that the water-driven model was carried out to estimate corn yield using the data assimilation in the semi-arid region. Based on the field experiments data of corn growing status, the main objectives of this proposal were to: (1) determine the sensitive parameters in the AquaCrop model using extended Fourier amplitude sensitivity test method, combine with prior knowledge of agronomy, select the appropriate machine learning optimization algorithms for automatic parameters tuning, localize the parameters of AquaCrop model, and then simulate accurately canopy cover, biomass and yield of corn; (2) establish vegetation indices to estimate corn canopy cover and biomass using hyperspectral data and satellite images; (3) carry out the assimilation of AquaCrop model and remote sensing data based on the combined point of canopy cover and biomass, accurately monitor the continuous dynamic changes of canopy cover and biomass, and ultimately estimate corn yield. The potential results of this study will provide technical support for timely diagnosis of corn growing status, yield estimation, corn cultivation management, and structural adjustment in the semi-arid region.

掌握作物关键生育阶段的长势时空变化规律，及时进行作物田间管理，确保作物稳产增产，对国家的粮食安全有着重要的意义。当前同化研究主要以单变量为主，双变量同化研究相对比较缺乏。此外，鲜见以水驱动模型开展对半干旱地区的玉米估产同化研究。因此，本项目拟以玉米为研究对象，基于田间实验观测，使用扩展傅立叶振幅敏感度测试方法确定AquaCrop模型的敏感参数，结合农学先验知识，选择适合的机器学习优化算法，实现参数自动调试，开展AquaCrop模型本地化研究，进而对冠层覆盖度、生物量和产量等的准确模拟；利用冠层高光谱数据和卫星影像数据，构建植被指数估算冠层覆盖度和生物量；以冠层覆盖度和生物量为结合点，开展AquaCrop模型和遥感数据的同化方法研究，精准地监测冠层覆盖度和生物量的连续动态变化，最终实现产量估算。本研究结果将为半干旱地区玉米生长适时诊断和估产，玉米栽培管理和结构调整提供决策依据。

数据同化方法将作物模型和遥感数据各自的优势进行整合，近些年来成为国内外的研究热点问题。当前的研究主要以单变量同化研究为主，双变量的同化研究相对比较缺乏。此外，当前同化研究涉及的作物模型主要以光能驱动或碳驱动模型为主，鲜见以水驱动模型开展对玉米长势监测和产量估算同化研究。本项目通过研究解决以下问题：（1）如何更好地开展AquaCrop模型的参数本地化和模拟研究？（2）选择什么样的数据同化算法和代价函数开展AquaCrop模型与遥感数据的双变量同化研究？主要研究内容包括：（1）基于田间试验获得的玉米冠层覆盖度、生物量和产量等测量结果，利用EFAST方法对AquaCrop模型的作物参数进行全局敏感性分析，确定AquaCrop模型在研究区域的敏感作物参数；（2）基于田间试验获得的玉米冠层覆盖度、生物量和冠层高光谱数据，分析高光谱数据波谱特征信息与玉米冠层覆盖度和生物量之间的回归关系，构建玉米冠层覆盖度和生物量的遥感定量估算模型；（3）以玉米冠层覆盖度和生物量两个动态变量作为结合点，比较分析不同同化算法的效率，筛选最优的同化策略，并构建双变量的代价函数，进行AquaCrop作物生长模型和遥感数据的双变量同化研究。按照项目研究计划和目标，本课题完成了上述各项研究任务并开展了广泛深入的国内外合作交流，基于研究成果，已发表学术论文11篇（其中SCI论文11篇）；项目负责人当选为Agronomy Journal副主编，培养博士研究生1名，硕士研究生3名。