Abstract Assimilating remote sensing data with crop growth model is a promising method to estimate crop yields over a large area. However, the method is always subject to the problems with biases in remote sensing products and assimilation weights in practical applications. In this study, we demonstrated the robustness of a ‘spatial assimilation’ method in dealing with the biases in three different remote sensing leaf area index (LAI) products. We further explored assimilation strategies for determining the assimilation weights when using ‘spatial assimilation’ method. Three different remote sensing LAI products were assimilated with MCWLA-Wheat model in the North China Plain during 2008–2015. The results demonstrated that the ‘spatial assimilation’ method was robust in mitigating the influences of biased LAI values and easily coupled with various LAI products based on different sources and retrieving algorithms. Furthermore, we found that the historical experiences of the optimal assimilation weights were not suitable to directly drive data assimilation in the coming seasons. Thus data-assimilation strategies to estimate crop yields without prior knowledge on the optimal assimilation weights were investigated. Two ensemble-mean-based assimilation strategies were recommended, which could reach 84 ~ 98% of yield estimation accuracy using the optimal assimilation weights. This study provides reliable and promising solutions for yield estimation over a large area using data assimilation without being limited by the biased state variables in remote sensing products and the lack of prior knowledge on the optimal assimilation weights. The ‘spatial assimilation’ method and the proposed ensemble-mean-based assimilation strategies have great potentials for wide applications, laying solid foundations for developing crop growth monitoring and yield forecasting system.

中文翻译：

---

使用空间同化算法和集合同化策略提高大面积基于遥感数据同化的作物产量估算的实用性

摘要 利用作物生长模型同化遥感数据是一种大面积估算作物产量的有效方法。然而，该方法在实际应用中始终存在遥感产品偏差和同化权重等问题。在这项研究中，我们证明了“空间同化”方法在处理三种不同遥感叶面积指数 (LAI) 产品中的偏差方面的稳健性。我们进一步探索了在使用“空间同化”方法时确定同化权重的同化策略。2008-2015 年华北平原 MCWLA-Wheat 模型同化了三种不同的遥感 LAI 产品。结果表明，“空间同化”方法在减轻有偏差的 LAI 值的影响方面是稳健的，并且很容易与基于不同来源和检索算法的各种 LAI 产品耦合。此外，我们发现最佳同化权重的历史经验不适合直接驱动未来季节的数据同化。因此，研究了在没有关于最佳同化权重的先验知识的情况下估计作物产量的数据同化策略。推荐了两种基于整体均值的同化策略，使用最优同化权重可以达到84~98%的产量估计精度。本研究为使用数据同化进行大面积产量估算提供了可靠且有前景的解决方案，而不受遥感产品中状态变量的偏差和缺乏关于最佳同化权重的先验知识的限制。“空间同化”方法和提出的基于集合均值的同化策略具有广泛的应用潜力，为开发作物生长监测和产量预测系统奠定了坚实的基础。