Taguchi-Based Optimization and Machine-Learning Prediction of Forage Yield and Nutritive Value in Annual Ryegrass Across Multiple Environments

Kökten, Kağan, Aasim, Muhammad, Akçura, Mevlüt, Kaplan, Mahmut, Ali, Seyid Amjad, Ali, Amjad, Sürmen, Mustafa, Yılmaz, Mehmet Fatih, Çankaya, Necda, Özdemir, Selim, Ekmekçi, Muammer, Alay, Fatih, Uçar, Rıdvan
International Journal of Plant Production · Springer Nature · 2026-11-06 · DOI: 10.1007/s42106-026-00476-1

Abstract

Annual ryegrass ( Lolium multiflorum Lam.) is a high-value forage crop whose yield and nutritive performance are strongly influenced by environment, cultivar response, and genotype × environment interaction. This study evaluated 28 cultivars across nine contrasting environments in Türkiye to quantify variation in forage biomass, nutritive quality, fiber composition, and feed value and to develop an integrated framework for assessing cultivar performance within the tested environments. Plant height, green and dry herbage yield, crude protein, crude ash, crude oil, acid detergent fiber, neutral detergent fiber, acid detergent lignin, dry matter digestibility, dry matter intake, and relative feed value were analyzed using analysis of variance, Taguchi signal-to-noise (S/N) analysis, machine-learning models, and multivariate clustering. Trait means varied among environments and cultivars, with environmental effects generally exceeding over cultivars. Green herbage yield ranged from 19,200 to 74,290 kg ha⁻¹, whereas crude protein ranged from 13.6 to 16.5%. Taguchi S/N analysis revealed trait-specific rankings and identified environment as the dominant factor for yield, nutritional traits, and fiber-related quality. Random Forest and LightGBM showed broadly comparable predictive performance for most traits, however, generalization to independent environments remains unverified. For green herbage yield, both models achieved an R2 score of 0.990, with RF showing slighly lower RMSe and MAE than LightGBM. Principal component analysis with K-means clustering explored multivariate patterns in yield and forage quality among cultivars and environments. This integrated approach provides exploratory evidence for cultivar assessment within the tested environments, with broader recommendations requiring further validation.

Read at publisher →

Similar articles