Deep learning-based estimation of plant functional traits from canopy spectral reflectance.

Abstract

Large-scale canopy-level plant trait quantification enhances crop yield and quality assessment, supports sustainable forestry economic development, and improves ecosystem monitoring globally. However, traditional methods relying on vegetation spectral libraries and machine learning models often face challenges in capturing the nonlinear and multivariate characteristics of canopy spectral responses. To overcome these challenges, we propose a novel deep learning framework, the Canopy-level Plant Functional Trait Retrieval Network (CTRN), which integrates Kolmogorov-Arnold Networks (KAN), Transformer, and Convolutional Neural Networks (CNN) to effectively extract informative representations from high-dimensional hyperspectral reflectance. The model is trained and evaluated using a comprehensive spectral-trait dataset, covering various plant species, different sensors, and multiple continents, and focuses on ten key canopy functional traits. Experimental results show that CTRN consistently outperforms other models, achieving R 2 values greater than 0.82 across all traits. Furthermore, even with just 44 spectral bands at a 40 nm resolution, CTRN demonstrates commendable accuracy in estimating LMA and C, with R 2 values approaching 0.80. These findings highlight the robust ability of the model to characterize complex associations between canopy spectra and plant functional traits, supporting accurate parameter retrieval in ecological and agricultural applications.

Plant phenotyping relevance

キャノピーのハイパースペクトル反射から植物機能形質を推定する深層学習手法を開発・評価しており、形質取得法が研究の中心である。

Code and data availability

公開本文の所在を確認できませんでした。非公開または購読が必要な可能性があります。

No evidence-backed public reproduction asset is currently recorded.