Prediction of symbiotic nitrogen fixation in common bean ( Phaseolus vulgaris L.) using unmanned aerial system remote sensing

Abstract

Abstract Common bean ( Phaseolus vulgaris L.) can fix atmospheric nitrogen (N) through symbiosis with Rhizobia species. This trait is often underutilized by growers and overlooked by breeders due to the laborious and costly evaluation techniques involved. There is a critical need for the development of new screening tools to enhance nitrogen fixation efficiency. Remote sensing techniques utilizing unmanned aerial systems offer a potential solution to this challenge, providing a high‐throughput phenotyping method for trait evaluation. In this study, we investigated the use of vegetation indices and machine learning (ML) methods in estimating symbiotic nitrogen fixation (SNF). Forty‐two black bean breeding lines from the Dry Bean Breeding Program at Michigan State University were grown and compared under both high and low N conditions. A random forest model developed to predict percent nitrogen derived from the atmosphere (%Ndfa) using remote sensing (RS) data resulted in an average accuracy of R 2 = 0.86. A 3‐year evaluation of these trials in Michigan demonstrated how seed yield under unfertilized conditions could be used as an indirect indicator of SNF ability. Two accurate prediction models for yield were developed using stepwise general linear modeling (StepwiseGLM) and Bayesian regularized artificial neural network (BRNeural Network) (stepwise general linear model r = 0.64; Bayesian regularized neural network r = 0.65). These results suggest that seed yield and RS data coupled with ML offer a promising tool to efficiently implement indirect selection for SNF in common bean.

Plant phenotyping relevance

UASリモートセンシングと機械学習により、共生窒素固定という植物形質を推定するスクリーニング手法を開発・評価しており、表現型取得と予測モデルが研究の中心である。

Code and data availability

The paper's data availability statement explicitly says the code and methodologies used in this study are available in the authors' public GitHub repository (msudrybeanbreeding). No phenotype dataset, imagery, or model checkpoint deposit is stated in the supplied blocks.