Exploration of the flow inside the roughness sublayer often suffers from sub-sampling of its complex three-dimensional and non-homogeneous flow fields. Based on detailed particle image velocimetry within a randomly-ordered canopy model, we analyze the potential differences between single-location flow statistics and their spatially-averaged values. Overall, higher variability exists inside the canopy than above it, and is two to four times higher than found inside similar, however ordered, canopy arrangements. The local mean absolute percentage error (MAPE), vertically averaged within three different regions (below, above, and at canopy height), provides a measure for quantifying and characterizing the spatial distribution of errors for various flow properties (mean velocity and stresses). We calculated the value of MAPE at predefined farthest-locations based only on geometric considerations (i.e., farther away from surrounding roughness elements), as commonly done in the field. Interestingly, most of the vertical profiles at the farthest locations lie within the interquartile range of the measured spatial variability for all studied flow and turbulent properties. Additionally, our results show that, for at least 23% of the total canopy plan area, the double-averaged streamwise velocity component and its variance inside the canopy can be reproduced from a single measured profile for which the value of MAPE does not exceed 25%. These regions also constitute most of the farthest locations. The property that exhibits the highest MAPE value inside the canopy is the Reynolds stress (up to \(130\%\)); however, these errors are dramatically reduced in the upper half of the canopy. Furthermore, at the canopy interface and above it, the errors rarely exceed \(20\%\). The variability is also manifested in the computed integral length scales. The single-point velocity autocorrelation always underestimates the length scales obtained from the two-point statistics. These findings have implications for canopy flow and transport modelling inside the roughness sublayer and can help explain and evaluate the source of discrepancies between measurements and transport models.

对粗糙子层内部流动的探索常常受到对其复杂的三维和非均匀流场的二次采样的困扰。基于随机排序的冠层模型中详细的粒子图像测速技术，我们分析了单位置流量统计数据及其空间平均值之间的潜在差异。总体而言，顶篷内部存在比其上方更高的可变性，并且比相似但有序的顶篷布置内部的可变性高出二到四倍。局部平均绝对百分比误差（MAPE）在三个不同区域（下方，上方和顶棚高度）内进行垂直平均，可提供一种量化和表征各种流量特性（平均速度和应力）误差的空间分布的措施。我们计算了MAPE的价值仅基于几何考虑（即，距离周围的粗糙度元素更远），在预定义的最远位置进行定位，这在现场是很常见的。有趣的是，对于所有研究的流量和湍流特性，最远位置的大多数垂直剖面都位于所测量的空间变异性的四分位数范围内。此外，我们的结果表明，对于至少23％的总冠层计划面积，可以从单个测得的轮廓中复制出双倍平均水流速度分量及其在冠层内部的方差，而MAPE值不超过25 ％。这些区域也构成了最远的位置。在顶篷内部表现出最高MAPE值的特性是雷诺应力（最高\（130 \％\））; 但是，这些错误在顶篷的上半部大大减少了。此外，在顶篷界面及其上方，错误很少超过\（20 \％\）。可变性还体现在计算出的积分长度尺度上。单点速度自相关始终会低估从两点统计信息中获得的长度比例。这些发现对于粗糙度子层内部的冠层流动和传输模型具有重要意义，并且可以帮助解释和评估测量和传输模型之间差异的根源。