Unmanned aerial vehicle (UAV) remote sensing and machine learning have emerged as a practical approach with ultra-high temporal and spatial resolutions to overcome the limitations of ground-based sampling for continuous crop monitoring. However, little is known on the suitability of distinct sensing indices for different crop management and distinct crop development phases. In this study, we assessed the potential of the UAV-based modeling to monitor field-scale crop growth under different water and nutrient supply considering distinct phenological phases of maize. UAV multispectral observations were deployed over two long-term experimental sites in three growing seasons. Calibration and validation of the random forest model took place at the Nutrient Balance Experimental Site (NBES) and the Water Nitrogen Crop Relation Site (WNCR), respectively. Leaf area index, leaf chlorophyll concentration, and aboveground dry matter were measured at the jointing, heading, and grain filling phases of maize in 2018–2020. Our results revealed that the suitability of sensing indicators differed at distinct maize phenological phases. Overall, red edge, red edge reflectance ratio, and chlorophyll index green are the most appropriate UAV indicators for estimating maize growth variables. The random forest model developed and calibrated at NBES with nutrient supply detected the signal of nitrogen × irrigation interactions at the other experimental site (WNCR) in different development phases and years very well, suggesting that random forest models developed by UAV images of same spatial and spectral attributes could be transferred across sites with the same cultivar while different irrigation and fertilizer management. We conclude that the selected number of UAV detected indicators processed with a random forest model could be used for robustly estimating environment × management (fertilizer and irrigation) interactions on maize growth variables.

无人机（UAV）遥感和机器学习已经成为一种具有超高时间和空间分辨率的实用方法，以克服基于地面采样的连续作物监测的局限性。然而，对于不同的传感指数对不同作物管理和不同作物发育阶段的适用性知之甚少。在这项研究中，我们评估了基于无人机的建模在考虑玉米不同物候阶段的情况下监测不同水和养分供应下大田规模作物生长的潜力。无人机多光谱观测部署在三个生长季节的两个长期实验地点。随机森林模型的校准和验证分别在营养平衡实验站点 (NBES) 和水氮作物关系站点 (WNCR) 进行。对2018-2020年玉米拔节、抽穗和灌浆阶段的叶面积指数、叶片叶绿素浓度和地上干物质进行测定。我们的研究结果表明，传感指标的适用性在不同的玉米物候阶段有所不同。总体而言，红边、红边反射率和叶绿素指数绿色是估计玉米生长变量最合适的无人机指标。在 NBES 开发和校准的随机森林模型，在不同的发展阶段和年份很好地检测到了其他实验地点 (WNCR) 的氮×灌溉相互作用的信号，表明由具有相同空间和光谱属性的无人机图像开发的随机森林模型可以在具有相同栽培品种而不同灌溉和肥料管理的地点之间转移。我们得出结论，使用随机森林模型处理的选定数量的无人机检测指标可用于稳健地估计环境×管理（肥料和灌溉）对玉米生长变量的相互作用。