IMPACT OF POST EMERGENCE HERBICIDES ON WEED DYNAMICS, YIELD AND NUTRIENT UPTAKE OF MAIZE: A COMPARATIVE EVALUATION OF LINEAR AND MACHINE LEARNING APPROACH Authors: Rayala Srinivasa Rao, Jambamma, B., Danam Sravanthi, Srinivasa Reddy, I.V. Journal: Journal of Animal and Plant Sciences (JAPS) ISSN: 1018-7081 (Print), 2309-8694 (Online) Volume: 37 Issue: 1 Year: 2027 DOI: https://doi.org/10.36899/JAPS.2027.1.0005 URL: https://doi.org/https://doi.org/10.36899/JAPS.2027.1.0005 Publisher: Pakistan Agricultural Scientists Forum Abstract:

Effective weed management is essential for improving maize (Zea mays L.) productivity through the identification of selective and efficient post-emergence herbicides. Field experiments were conducted during 2019–2021 at the farmer field to evaluate weed control efficacy, weed dynamics, and maize yield under four herbicide treatments i.e. T1: Weedy check, T2: Tembotrione (42% S.C.) 287.5 mL ha-1 [25 DAS (days after sowing)], T3: Topramezone (33.6% S.C.)  100 mL ha-1 (25 DAS) and T4: Halosulfuron Methyl (75% WG) 90 g ha-1 (25 DAS) with data averaged across five locations for four treatments each year. The resulting dataset comprised 12 treatments-year observations with 11 agronomic variables. The machine-learning analyses were conducted as an initial exploratory assessment. Using chronological validation, 2019–2020 observations were used for training and 2021 observations for independent testing (66:34 split). Among the evaluated treatments, application of tembotrione proved the greatest broad-spectrum weed suppression and produced the highest grain yield (8,936 kg ha⁻¹), compared with the untreated control (3,897–4,324 kg ha⁻¹). Spearman correlation analysis revealed strong positive associations between grain yield and nutrient uptake of maize (ρ = 0.916) and strong negative associations with the weed index (ρ = −0.909) and broadleaf weed density (ρ = −0.825). These predictive modeling using linear regression (LR), random forest (RF), and XGBoost was evaluated using the coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE). Within the constraints of the small dataset, LR achieved the best predictive performance (R² = 0.949, RMSE = 379.58, MAE = 327.96). Ridge Regression with SHAP analysis identified total nutrient input (total NPK; +408.95) as the strongest predictor of maize yield while year had the lowest contribution (+192.78). These findings demonstrate that effective broad-spectrum weed management enhances maize productivity, while the modelling provides preliminary insights into weed control, soil fertility and yield relationships.

Keywords: Linear regression, maize yield, random forest, SHAP analysis, weed density, weed management, XGBoost