Using UAV Multispectral Imagery to Predict Leaf SPAD Dynamics During Maize Growth Under Different Plant Densities

Abstract

Chlorophyll content represents a key growth indicator for maize. The traditional SPAD (Soil and Plant Analyzer Development) method, though easy to operate, is inefficient, destructive, and unsuitable for high-throughput field monitoring. UAV (Unmanned Aerial Vehicle) remote sensing technology is highly efficient and detects abundant indicators, enabling large-scale SPAD measurement. In this study, 18 vegetation indices and eight texture features were selected as the indicator system by combining prior knowledge and experimental analysis. In a two-year maize density experiment, multispectral images were collected in the growth period. The correlations among SPAD values, multispectral indices and texture features were analyzed using Pearson correlation coefficients. Then the detection accuracies of three algorithms, i.e., RF (Random Forest), PLSR (Partial Least Squares Regression), and SVR (Support Vector Regression), were compared under this indicator system. Compared with models constructed using single vegetation indices or single texture features, the estimation accuracy of the indicator system at the jointing stage was improved by 0.13 and 0.22, respectively. The results showed that SVR achieved the highest estimation accuracy among the three algorithms, with determination coefficients (R2) of 0.73, 0.77and 0.70 at the jointing, silking, and grain-filling stages, respectively. This study established a non-destructive monitoring framework for chlorophyll content during the entire maize growth stage based on UAV data.

Plant phenotyping relevance

UAVマルチスペクトル画像と回帰モデルにより、トウモロコシの葉緑素量（SPAD）を非破壊推定する手法を構築し、複数アルゴリズムの精度比較も行っており、表現型取得手法が中心である。

Code and data availability

The paper describes a UAV multispectral maize SPAD dataset (SPADmaize_ahau) and SVR/RF/PLSR models, but no block contains a public data or code availability statement, repository deposit, or authors' URL for the imagery, SPAD measurements, dataset, or trained models. The only URL present is the article DOI itself.

No evidence-backed public reproduction asset is currently recorded.