Remote sensing imagery has been a key data source for precision agriculture. However, high-resolution and/or hyperspectral imagery have typically been favored for their greater information content. This study aims to demonstrate the capability of medium-resolution imagery in precision agriculture by developing an example of canola yield mapping using Sentinel-2 data in central Alberta. Two simple empirical models for mapping precision canola yield are tested: one using random forest regression and a second using functional linear regression. Both take as input freely-available Sentinel-2 time series images and use these to predict precision yield gathered by a yield monitor. The models were able to predict crop yield to within 12–16% accuracy of the reference yield. These results also demonstrate that a time series of medium-resolution multispectral imagery can capture small-scale variation in crop yields. The proposed methods can be applied to other areas or cropping systems to improve understanding of crop growth at both the field-level and regional-level.

遥感影像一直是精准农业的关键数据源。然而，高分辨率和/或高光谱图像通常因其更多的信息内容而受到青睐。本研究旨在通过使用艾伯塔省中部的 Sentinel-2 数据开发油菜产量绘图示例，展示精准农业中的中等分辨率图像的能力。测试了用于绘制精确油菜产量的两个简单经验模型：一个使用随机森林回归，另一个使用函数线性回归。两者都将免费提供的 Sentinel-2 时间序列图像作为输入，并使用这些图像来预测产量监视器收集的精确产量。这些模型能够以参考产量的 12-16% 的准确度预测作物产量。这些结果还表明，中等分辨率多光谱图像的时间序列可以捕捉到作物产量的小尺度变化。所提出的方法可以应用于其他领域或种植系统，以提高对田间和区域层面作物生长的理解。