多源高分光学遥感数据与作物模型同化的不确定性研究

Assimilation of optical remote sensing data into crop model is an important way to implement the regional crop growth monitoring and yield estimation. However, assimilation is complex and always restrained by several uncertainty factors, leading to unreliable results in regional application. Among these factors, insufficient spatial resolution in the observation is the most evident issue. In recent years, China and other countries have launched several optical remote sensing satellites with high-spatial-resolution sensors, which provides the possibility for reducing assimilation uncertainties. The high spatial resolution multi-spectral remote sensing data will be used in this program. The uncertainties in high resolution remote sensing observation, temporal and spatial scales in the remote sensing data and assimilation algorithms, and their effects on assimilation results will be explored by using a well-tuned winter wheat crop model. Observation errors will be analyzed through canopy LAI which is an assimilation variable retrieved from two different assimilation algorithms using remote sensing data. The differences in assimilation results from high, middle, and low spatial resolution data will be investigated to understand uncertainties from different spatial scales of assimilation data. The uncertainty from temporal scales caused by investigation step, wheat growth stages and their combination will be also analyzed. By adjusting the systematic parameters of the two algorithms, uncertainties and work efficiency will be analyzed and compared. Finally, based on the above-mentioned analysis, we put forward an optimal assimilation scheme for regional precision crop monitoring and yield estimation and validate it with observations, through which a way to reduce uncertainty in the assimilation process will be discussed.

光学遥感数据与作物模型同化是实现区域化作物长势监测和产量评估的重要途径。然而，同化过程复杂且受制于一些不确定因素，导致区域应用精度较低。其中，遥感对地观测空间分辨能力不足是限制同化模拟精度的重要原因之一。近年来，国内外连续发射了多颗兼顾高时空分辨率的光学遥感卫星，为降低观测误差提供了数据支持。本项目拟应用多源高分多光谱遥感数据，结合严格定标的冬小麦模型，研究高分遥感观测、时空尺度和同化方法的不确定性及其对同化的影响。通过在像元尺度上搭建两种同化法的同化模拟平台，以高分遥感反演的作物生长参数为同化变量，比较高分辨率数据与中、低分辨率数据的同化差异，并从观测步长、所处生育阶段的组合等方面共同分析时空尺度上的不确定性。再调整两种同化法各自的系统参数，分析和比较不同算法下同化的不确定性和运行效率。最后，提出并验证适用于精细化作物监测和产量评估的最优同化方案，进一步探讨减少同化不确定性的途径。

光学遥感数据与作物模型同化是实现区域化作物长势监测和产量评估的重要途径。然而，同化过程复杂且受制于一些不确定因素，导致区域应用精度较低。项目应用包括高分遥感数据在内的多源遥感数据，提出并验证适用于精细化作物监测和产量评估的最优同化方案，进一步探讨减少同化不确定性的途径。项目建立了基于Sobol全局敏感性分析和改进的马尔科夫链蒙特卡洛方法的作物模型参数率定方法，标定后的WOFOST模型对作物生长和产量的模拟精度均有明显提高。完成了高分遥感影像的大气校正，构建了基于NDVI影像的空间异质性指标，定量解析了不同空间分辨率遥感数据由空间尺度变化带来的不确定性，给出了数据同化中遥感影像最优空间分辨率选择依据。解析了EnKF顺序同化与SCE-UA优化同化两种同化算法用于研究区产量预测的不确定性与效率，利用趋势法和四维变分算法构造了三种同化代价函数，解析了使用三种连接变量时同化的不确定性，提出研究区的最优同化方案为采用SCE-UA算法优化同化LAI和ET双变量。在实现最优同化模型的基础上，通过Sentinel-2高分遥感数据与MODIS遥感数据结合运用，开展了基于高分数据改进的精细化冬小麦生长模拟。项目通过找到减少遥感数据与作物模型同化不确定性的途径，优化了遥感与作物模型融合估产技术，实现了精细化冬小麦生长模拟，可用于作物生长监测和产量评估等农情动态监测业务，可提供具有较强机理性的作物生长定量评价产品和估产产品，为农业管理部门提供决策支持，指导田间科学管理，促进粮食增产增收。