Estimating Key Phenological Dates of Multiple Rice Accessions Using Unmanned Aerial Vehicle-Based Plant Height Dynamics for Breeding  

作　　者：HONG Weiyuan LI Ziqiu FENG Xiangqian QIN Jinhua WANG Aidong JIN Shichao WANG Danying CHEN Song 

机构地区：[1]State Key Laboratory of Rice Biology and Breeding,China National Rice Research Institute,Hangzhou 311400,China [2]College of Computer Science and Technology,Zhejiang Sci-Tech University,Hangzhou 310018,China [3]College of Agriculture,Yangtze University,Jingzhou 434025,China [4]Plant Phenomics Research Centre,Academy for Advanced Interdisciplinary Studies,Nanjing Agricultural University,Nanjing 210095,China

出　　处：《Rice science》2024年第5期617-628,I0066-I0070,共17页水稻科学（英文版）

基　　金：supported by the National Key Research and Development Program of China (Grant No.2022YFD2300700);the Open Project Program of the State Key Laboratory of Rice Biology and Breeding, China National Rice Research Institute (Grant No.2023ZZKT20402);the Agricultural Science and Technology Innovation Program, the Central Public-Interest Scientific Institution Basal Research Fund, China (Grant No.CPSIBRF-CNRRI-202119);the Zhejiang ‘Ten Thousand Talents’ Plan Science and Technology Innovation Leading Talent Project, China (Grant No.2020R52035)。

摘　　要：Efficient and high-quality estimation of key phenological dates in rice is of great significance in breeding work. Plant height(PH) dynamics are valuable for estimating phenological dates. However, research on estimating the key phenological dates of multiple rice accessions based on PH dynamics has been limited. In 2022, field traits were collected using unmanned aerial vehicle(UAV)-based images across 435 plots, including 364 rice varieties. PH, dates of initial heading(IH) and full heading(FH), and panicle initiation(PI), and growth period after transplanting(GPAT) were collected during the rice growth stage. PHs were extracted using a digital surface model(DSM) and fitted using Fourier and logistic models. Machine learning algorithms, including multiple linear regression, random forest(RF), support vector regression, least absolute shrinkage and selection operator, and elastic net regression, were employed to estimate phenological dates. Results indicated that the optimal percentile of the DSM for extracting rice PH was the 95th(R^(2) = 0.934, RMSE = 0.056 m). The Fourier model provided a better fit for PH dynamics compared with the logistic models. Additionally, curve features(CF) and GPAT were significantly associated with PI, IH, and FH. The combination of CF and GPAT outperformed the use of CF alone, with RF demonstrating the best performance among the algorithms. Specifically, the combination of CF extracted from the logistic models, GPAT, and RF yielded the best performance for estimating PI(R^(2) = 0.834, RMSE = 4.344 d), IH(R^(2) = 0.877, RMSE = 2.721 d), and FH(R^(2) = 0.883, RMSE = 2.694 d). Overall, UAV-based rice PH dynamics combined with machine learning effectively estimated the key phenological dates of multiple rice accessions, providing a novel approach for investigating key phenological dates in breeding work.

关 键 词：phenological date plant height unmanned aerial vehicle machine learning rice breeding 

分 类 号：S511[农业科学—作物学]