Within-field yield variability affects crop production and management decisions. To understand and manage this variability, different techniques have been deployed to measure and monitor the crops (and soils) at various spatial scales, including manual measurements, harvester-mounted yield monitors, proximal and remote sensing and crop simulation modelling. The value of this increasing data availability to enhance process understanding and on-ground management is unclear. This study aimed to investigate the value of the increasingly available spatial data from different sources to understand important soil-plant processes amenable to improvement in both simulation modelling and for better management decisions for dryland cropping. We collected three types of measurement data (manual sampling, sensed data from satellite and drone, and yield maps) over a 10 ha field and conducted simulations using the process-based soil-plant model APSIM at different spatial scales (varied from 1 m up to 10 ha). We assessed the agreement between ground measurements and yield maps, analysed the potential to use remotely sensed vegetation indices to estimate yield, and the scale at which process-based modelling could be reliable. Wheat yield extracted from yield map at 1 m spatial resolution only explained 30% of the variation in yield measured from 1 m manual sampling, with better agreement when data was aggregated to 1 ha strip-scale (R = 0.66, NRMSE = 9.1%). Remotely sensed vegetation indices (VI) were better correlated with the yield map when aggregating images to coarse spatial resolution (> 50 m × 50 m), while high-resolution drone VI increased the correlation at finer scales. However, the relationship and the timing of the highest correlation differed between years. APSIM simulated point-based yield measured from manual samples with NRMSE of 19.4%, but it was difficult to capture spatial variation in yield due largely to uncertainties in input data. However, APSIM simulations captured the average crop growth dynamics and yield well at 1 ha strip- and 10 ha whole field scales. The results highlight the need for caution when using yield maps and remote sensing data to quantify spatial variability and inform spatially explicit management decisions at a fine resolution (e.g., 1 m). In our case, remote sensing data and yield maps only became consistent and process-based modelling became skilful at scales larger than a 1 ha strip. Despite an increasing amount of high-resolution spatial data, the usefulness at fine resolution needs further investigation, particularly under heterogeneous field conditions. Such data need to be analysed in conjunction with the landscape, soil and climate data to understand the drivers of spatial variability and inform process understanding and modelling. This further implies potential problems in developing spatial management practices at finer scales using such data.

中文翻译：

---

在不同空间尺度上测量和模拟作物生长和产量的技术的有用性

田间产量变异性影响作物生产和管理决策。为了了解和管理这种变异性，人们采用了不同的技术来测量和监测不同空间尺度的作物（和土壤），包括手动测量、收割机上安装的产量监测器、近端和遥感以及作物模拟建模。不断增加的数据可用性对于增强流程理解和现场管理的价值尚不清楚。本研究旨在调查来自不同来源的日益可用的空间数据的价值，以了解重要的土壤-植物过程，这些过程有助于改进模拟模型和更好的旱地种植管理决策。我们在 10 公顷的田地中收集了三种类型的测量数据（手动采样、卫星和无人机传感数据以及产量图），并使用基于过程的土壤植物模型 APSIM 在不同空间尺度（从 1 m 到 1 m 不等）进行模拟。至 10 公顷）。我们评估了地面测量值和产量图之间的一致性，分析了使用遥感植被指数来估计产量的潜力，以及基于过程的建模的可靠规模。从 1 m 空间分辨率的产量图提取的小麦产量只能解释 1 m 手动采样测量的产量变化的 30%，当数据汇总到 1 ha 带状尺度时，一致性更好（R = 0.66，NRMSE = 9.1%） 。当将图像聚合到粗空间分辨率（> 50 m × 50 m）时，遥感植被指数（VI）与产量图具有更好的相关性，而高分辨率无人机 VI 则在更精细的尺度上增加了相关性。然而，不同年份之间的关系和最高相关性的时间有所不同。 APSIM 模拟了根据手动样本测量的基于点的产量，NRMSE 为 19.4%，但由于输入数据的不确定性，很难捕获产量的空间变化。然而，APSIM 模拟很好地捕捉了 1 公顷田地和 10 公顷整田规模的平均作物生长动态和产量。结果强调，在使用产量图和遥感数据量化空间变异性并以精细分辨率（例如 1 m）为空间明确的管理决策提供信息时需要谨慎。在我们的案例中，遥感数据和产量图仅变得一致，并且基于过程的建模在大于 1 公顷地带的规模上变得熟练。尽管高分辨率空间数据的数量不断增加，但精细分辨率的有用性还需要进一步研究，特别是在异质现场条件下。这些数据需要与景观、土壤和气候数据结合起来进行分析，以了解空间变异的驱动因素并为过程理解和建模提供信息。这进一步暗示了使用此类数据开发更精细尺度的空间管理实践的潜在问题。