Estimation of Ramie Key Phenotypic Traits Based on UAV Remote Sensing

Abstract

UAV-based phenotyping enables efficient high-throughput measurement of field crops. Phenotypic monitoring of ramie is critical for its cultivation management and variety breeding. However, ramie exhibits characteristics including multiple annual harvests, short growth cycles and rapid dynamic growth change, all of which increase the difficulty of growth monitoring and yield estimation. This study aims to utilize UAV-based multispectral remote sensing to estimate ramie plant height (PH), leaf area index (LAI), and above-ground biomass (AGB) over multiple time series, and to assess the influence of seasonal effects and different data processing strategies on the accuracy of ramie digital phenotyping. Over three ramie growth cycles, a total of 15 UAV flights were conducted over an experimental field consisting of 72 plots. The structure from motion (SfM) algorithm was applied to estimate PH. Remote sensing features derived from UAV imagery were used with background segmentation and machine learning to estimate LAI. The AGB was estimated by combining remote sensing-derived PH, LAI, and climate data. The results showed that the estimated and measured phenotypes were highly correlated, with optimal coefficients of determination of 0.961 for PH and 0.873 for LAI. Background segmentation improved LAI accuracy. Integrating climate data, remote sensing-derived PH and LAI significantly enhanced the accuracy of AGB estimation. In conclusion, this study provides a feasible method for extracting ramie phenotypes from UAV remote sensing imagery, providing methodological support for large-scale management of the crop industry and intelligent, precise monitoring of crop growth.

Plant phenotyping relevance

UAVマルチスペクトル画像、SfM、背景分割、機械学習を用いてラムーの草丈・LAI・地上部バイオマスを推定し、精度評価まで行う手法研究であり、表現型取得・抽出が中心である。

Code and data availability

The paper reports UAV multispectral phenotyping of ramie (PH, LAI, AGB) with ground-truth measurements, orthomosaics/DSM imagery, and RF/SVR/PLSR models. No public repository, dataset deposit, or author code URL is provided; the Data Availability Statement directs inquiries to the corresponding authors. The only public

No evidence-backed public reproduction asset is currently recorded.