Enhancing yield and protein content estimation in dry peas through multispectral data integration and ensemble learning model

Abstract Accurately predicting yield and protein content in dry peas ( Pisum sativum L.) is essential for improving agricultural productivity, sustainability, and nutritional outcomes. Traditional methods for estimating these traits are labor‐intensive and limited in scope, while remote sensing technologies and machine learning offer scalable, nondestructive alternatives. This study addresses the challenge of integrating multispectral data and advanced machine learning models, specifically stacked meta‐models, to enhance prediction accuracy for yield and protein content in dry peas. Data were collected from 860 genotypes across two North Dakota locations over three growing seasons using uncrewed aerial system‐mounted multispectral sensors. The results revealed significant variability in yield and protein content due to location and year, with the meta‐model outperforming individual machine learning models for both traits. The meta‐model achieved a root mean squared error (RMSE) of 396.28 kg ha −1 and an coefficient of determination ( R 2 ) of 0.77 for yield, demonstrating strong predictive performance. However, protein content predictions were moderate, with an RMSE of 1.28% and an R 2 of 0.54, highlighting limitations in the correlation between spectral features and protein synthesis. Growth stage analysis emphasized the critical importance of the maturity stages to predict the yield and protein content in dry peas. These findings support the potential of integrating multispectral data and meta‐modeling for accurate agricultural predictions. Future research should focus on incorporating additional data sources, refining feature engineering techniques, and tailoring models to specific growth stages to further improve predictive accuracy. This study provides a foundation for advancing precision agriculture practices in pulse crops.