Context.

Accurate prediction of within-field crop yield in response to spatial and temporal variability provides essential information for farm managers to improve productivity and ensure optimal use of inputs. Understanding yield spatial and temporal variability cannot be solely addressed by crop modelling or remote sensing but by integrating the instantaneous spatial information from remote sensing and the temporal information from crop modelling. Sequential data assimilation techniques allow wheat and soil observations to be assimilated into the crop model while it evolves and evaluate model and observational uncertainties to improve the accuracy of crop monitoring and yield prediction.

Objectives

The objective of this study was to comprehensively explore the potential yield estimation improvement by assimilating observations of all prognostic wheat and soil states, including various repeat intervals and accuracy, allowing recommendations on implementation to be made.

Methods

This study develops an Ensemble Kalman filter (EnKF) data assimilation framework for the APSIM-Wheat model and illustrates potential improvements in wheat yield estimation through a synthetic study. Through several scenarios, assimilation of wheat and soil observations into APSIM was explored, by assimilating these variables solely or collectively, and in various phenological stages.

Results and conclusions.

The results showed, under the specific weather and soil conditions assumed in this study, that while open-loop (no data assimilation) provided a yield estimation with a bias of 10.1%, assimilation in the flowering to end of grain filling stage reduced the bias to 1.4%, 2.9%, 4.4%, and 1.0% when constraining with leaf area index, leaf weight, stem weight, surface soil moisture observations, respectively. When assimilating in the floral initiation to the flowering stage, the yield estimation bias was reduced to 7.1%, 9.8%, 1.1%, and 1.2% when constraining with leaf nitrogen, stem nitrogen, top-layer soil ammonium‑nitrogen and nitrate‑nitrogen, respectively. Leaf area index, biomass and surface soil moisture are recommended for data assimilation especially with observations from remote sensing.

Significance

This study developed a data assimilation framework for the APSIM-Wheat model and can be extended to over 20 crop modules integrated with APSIM. This synthetic study provided a exhaustive data assimilation experiment for wheat and soil states that are measurable by current techniques with a rigorous justification on uncertainties. It thus provides a guide for future agricultral data assimilation practices in choosing crop and soil states for assimilation, and for planning the timing and frequency of data collection. It should also inspire researchers to develop new techniques for measuring wheat states.

中文翻译：

---

小麦和土壤状态的同化以改进产量预测：APSIM-EnKF 框架

语境。

响应空间和时间变化，准确预测田间作物产量，为农场管理者提高生产力和确保最佳使用投入提供了重要信息。理解产量的空间和时间变化不能仅仅通过作物建模或遥感来解决，而是通过整合来自遥感的瞬时空间信息和来自作物建模的时间信息。序列数据同化技术允许将小麦和土壤观测数据同化到作物模型中，同时改进和评估模型和观测的不确定性，以提高作物监测和产量预测的准确性。

目标

本研究的目的是通过同化所有预测小麦和土壤状态的观察结果，包括各种重复间隔和准确性，全面探索潜在的产量估计改进，从而提出实施建议。

方法

本研究为 APSIM-Wheat 模型开发了集成卡尔曼滤波器 (EnKF) 数据同化框架，并通过综合研究说明了小麦产量估计的潜在改进。通过几个情景，通过单独或集体同化这些变量，以及在不同的物候阶段，探索了将小麦和土壤观测同化到 APSIM 中。

结果和结论。

结果表明，在本研究假设的特定天气和土壤条件下，虽然开环（无数据同化）提供的产量估计偏差为 10.1%，但在开花期至灌浆期末期的同化降低了偏差分别以叶面积指数、叶重、茎重、表层土壤水分观测值约束时为 1.4%、2.9%、4.4% 和 1.0%。在花开始同化至开花期同化时，在叶氮、茎杆氮、表层土壤铵-氮和硝酸盐-氮的约束下，产量估计偏差分别降低到7.1%、9.8%、1.1%和1.2%。 ， 分别。叶面积指数、生物量和地表土壤水分被推荐用于数据同化，特别是通过遥感观测。

意义

本研究为 APSIM-Wheat 模型开发了一个数据同化框架，并且可以扩展到与 APSIM 集成的 20 多个作物模块。这项综合研究为小麦和土壤状态提供了详尽的数据同化实验，这些实验可通过当前技术进行测量，并对不确定性进行了严格的论证。因此，它为未来农业数据同化实践在选择作物和土壤状态进行同化以及规划数据收集的时间和频率方面提供了指南。它还应该激发研究人员开发测量小麦状态的新技术。