{"title":"Joint prediction of potato yield and nitrogen status at row closure using multi-task deep learning and UAV-derived LiDAR–multispectral data","authors":"Ehsan Chatraei Azizabadi, Nasem Badreldin","doi":"10.1016/j.compag.2026.111893","DOIUrl":null,"url":null,"abstract":"<div><div>Accurate prediction of crop yield and nitrogen (N) status from remote sensing data remains challenging due to fundamental differences in how cumulative and physiological traits are expressed at the canopy level. This study investigated a multi-task learning (MT) framework based on deep learning (DL) for the joint prediction of potato tuber yield and petiole nitrate–N (PNN) using fused unmanned aerial vehicle (UAV)-derived multispectral (MS) and LiDAR data acquired at row closure across two growing seasons. Model performance was evaluated against single-task learning (ST) across multiple network architectures using independent test data. Joint optimisation substantially improved yield prediction relative to yield-only models, increasing test R<sup>2</sup> from ≤ 0.26 under ST to 0.46–0.60 across MT architectures, whereas benefits for PNN were limited and architecture dependent. Attribution analysis indicated that yield prediction relied on a broader combination of spectral and three-dimensional canopy features, with LiDAR-derived height, volumetric, and heterogeneity metrics contributing consistently but with lower relative importance than spectral indices. In contrast, PNN prediction was dominated by chlorophyll-sensitive spectral indices, consistent with its short-term physiological basis. These results show that shared-learning frameworks in potato systems are most useful when the sensing inputs and model structure are matched to the physiological timescale of the target trait.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"250 ","pages":"Article 111893"},"PeriodicalIF":10.3000,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Computers and Electronics in Agriculture","FirstCategoryId":"97","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0168169926004886","RegionNum":1,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"2026/5/23 0:00:00","PubModel":"Epub","JCR":"Q1","JCRName":"AGRICULTURE, MULTIDISCIPLINARY","Score":null,"Total":0}
引用次数: 0
Abstract
Accurate prediction of crop yield and nitrogen (N) status from remote sensing data remains challenging due to fundamental differences in how cumulative and physiological traits are expressed at the canopy level. This study investigated a multi-task learning (MT) framework based on deep learning (DL) for the joint prediction of potato tuber yield and petiole nitrate–N (PNN) using fused unmanned aerial vehicle (UAV)-derived multispectral (MS) and LiDAR data acquired at row closure across two growing seasons. Model performance was evaluated against single-task learning (ST) across multiple network architectures using independent test data. Joint optimisation substantially improved yield prediction relative to yield-only models, increasing test R2 from ≤ 0.26 under ST to 0.46–0.60 across MT architectures, whereas benefits for PNN were limited and architecture dependent. Attribution analysis indicated that yield prediction relied on a broader combination of spectral and three-dimensional canopy features, with LiDAR-derived height, volumetric, and heterogeneity metrics contributing consistently but with lower relative importance than spectral indices. In contrast, PNN prediction was dominated by chlorophyll-sensitive spectral indices, consistent with its short-term physiological basis. These results show that shared-learning frameworks in potato systems are most useful when the sensing inputs and model structure are matched to the physiological timescale of the target trait.
由于累积和生理性状在冠层水平上的表达方式存在根本差异,利用遥感数据准确预测作物产量和氮状态仍然具有挑战性。本研究研究了一个基于深度学习(DL)的多任务学习(MT)框架,利用融合无人机(UAV)衍生的多光谱(MS)和激光雷达(LiDAR)数据,在两个生长季节的行合集采集数据,对马铃薯块茎产量和叶柄硝酸盐- n (PNN)进行联合预测。使用独立的测试数据对跨多个网络架构的单任务学习(ST)模型性能进行评估。联合优化大大提高了相对于仅产量模型的产量预测,将测试R2从ST下的≤0.26提高到跨MT架构的0.46-0.60,而PNN的收益有限且依赖于架构。归因分析表明,产量预测依赖于光谱和三维冠层特征的更广泛组合,lidar衍生的高度、体积和异质性指标的贡献一致,但相对重要性低于光谱指数。而PNN预测以叶绿素敏感的光谱指标为主,符合其短期生理基础。这些结果表明,当传感输入和模型结构与目标性状的生理时间尺度相匹配时,马铃薯系统中的共享学习框架最有用。
期刊介绍:
Computers and Electronics in Agriculture provides international coverage of advancements in computer hardware, software, electronic instrumentation, and control systems applied to agricultural challenges. Encompassing agronomy, horticulture, forestry, aquaculture, and animal farming, the journal publishes original papers, reviews, and applications notes. It explores the use of computers and electronics in plant or animal agricultural production, covering topics like agricultural soils, water, pests, controlled environments, and waste. The scope extends to on-farm post-harvest operations and relevant technologies, including artificial intelligence, sensors, machine vision, robotics, networking, and simulation modeling. Its companion journal, Smart Agricultural Technology, continues the focus on smart applications in production agriculture.