遥感与作物模型多变量数据同化的区域冬小麦生长模拟研究

There are several limitations in the field of remotely sensed and crop models assimilation, such as using single state variable, scale dismatch between remotely sensed observation and models simulation and imperfect data assimilation algorithms. This study try to improve the regional winter wheat growth simulation by using multivariate data assimilation method. We firstly explore the coupled mechanism of Fraction of absorbed Photosynthetically Active Radiation(FPAR), Leaf Area Index(LAI) and Evapotranspiration(ET) retrieved from remotely sensed data and sub-modules of crop growth model based on the calibration of soil water atmosphere plant(SWAP) model. Multi-objective merit function vector sets have been constructed dependent on the analyzing the global sensitivity of SWAP model and remotely sensed observation errors in different winter wheat growing stages. Achieving the consistency of derived from remotely sensed data and simulated from SWAP model through developing scale transformation method. In order to achieve the goal of improving the ability of reginoal crop growth simulation, we are going to develope an innovative multivariate data assimilation strategy by integrating multi-objective optimization and Ensemble Kalman Filter(EnKF) to achieve crop parameters update and state variables assimilation in the temporal dimension, which help us understand the response ralationship between FPAR, LAI, ET in growing season. We are going to carry out field campaigns and validate the optimized crop growth model simulation and the new multivariate data assimilation method. This project aims to further promote the theory and methodology of assimilating remotely sensed data into crop growth model, and improve quantitative remote sensing applications in agricultural condition monitoring.

针对目前遥感与作物模型同化中存在的同化变量单一、遥感观测与模型模拟时空尺度不匹配以及数据同化算法不够完善等关键问题,本研究采用遥感与作物模型多变量数据同化方法，致力于改进区域冬小麦生长模拟精度。探索遥感反演的光合有效辐射吸收比率、叶面积指数和蒸散发与土壤-水分-大气-植被耦合模型（SWAP）光热水各子模块的耦合机制；在SWAP模型全局敏感性和生育期遥感观测误差分析的基础上，确定多目标优化的代价函数及其参数设置；通过发展尺度转换模型，实现多变量在同化系统中时空尺度的一致；重点开发内外循环相互嵌套的多目标扰动的集合卡尔曼滤波多变量同化算法，实现模型参数优化与状态变量更新,明晰同化过程中多个状态变量在作物生育期的互响应规律，进而提高区域冬小麦生长模拟能力；深入开展地面同步观测实验，以期达到对模型模拟和同化结果的验证和评价。项目旨在进一步完善定量遥感与数据同化技术在农情监测的深入应用与发展。

遥感与作物模型的数据同化是定量遥感与精准农业的重要研究内容，也是提高区域作物生长过程模拟最具潜力的技术方法。本项目以河北南部冬小麦为研究对象，针对目前遥感与作物模型同化中存在的同化变量单一、遥感观测与模型模拟时空尺度不匹配以及数据同化算法不够完善等关键问题,提出采用遥感与作物模型多变量数据同化方法，致力于改进区域冬小麦生长模拟精度。主要研究内容：（1）针对遥感像元与作物模型模拟单元的空间尺度不匹配问题，提出LAI的二级尺度转换模型，通过融合时间序列MODIS LAI产品的高时间频率信息和Landsat TM5的空间细节信息，通过构建最小化四维变分的代价函数，逐格网单元重新初始化模型的不确定参数，显著提高作物模型的区域模拟能力；（2）创新性地提出了采用卡尔曼滤波（Kalman filter:KF）融合多时相中等分辨率TM LAI与时间序列MODIS LAI生成合成LAI的方法，随后采用集合卡尔曼滤波同化KF合成的LAI到WOFOST模型，优化全生育期LAI轨迹，减少作物模型的累计误差，提高作物模型的区域模拟能力；（3）开展叶面积指数和蒸散发双变量数据同化的研究，利用了MODIS LAI产品和MODIS ET产品，分别构建同化趋势信息的代价函数和四维变分代价函数，对比分析了水分胁迫模式下单独同化蒸散发ET、叶面积指数LAI、ET和LAI双变量的三种同化策略的模型产量估测精度。结果表明，联合同化ET和LAI的双变量取得同化单变量LAI或ET更高的精度。（4）开展反射率同化的研究，通过WOFOST 与PROSAIL模型的耦合，构建遥感反射率与WOFOST-PROSAIL耦合模型模拟反射率的四维变分的代价函数，逐格网单元优化WOFOST-PROSAIL耦合模型的敏感参数，将优化后的参数代入WOFOST模型，结果表明， 同化反射率的精度略低于同化LAI的精度，但是这种策略避免了区域LAI反演带来的误差，为数据同化的业务化运行提供了可行的技术途径。本项目对农业数据同化领域中，同化变量的选择、尺度效应与转化模型和最优同化算法等关键问题进行系统和深入地探讨，为今后开展遥感与作物模型数据同化研究提供了借鉴价值，促进了农业遥感学科的发展。