Author(s): Hassan A Al-Mahrouqi

Abstract: Integrated nutrient management (INM) improves potato productivity and nutrient-use efficiency, yet its field-level optimization remains difficult because crop responses depend on non-linear interactions among soil fertility, climate and organic-inorganic nutrient combinations. This study developed AI-driven prediction models to optimize INM for potato using a harmonized multi-environment dataset (288 plot observations from 24 site-year combinations across six agro-ecologies). Predictors included detailed organic and mineral nutrient inputs, soil properties, and in-season weather indices, while outcomes comprised total and marketable tuber yield, quality traits, net returns and nitrogen-use efficiency metrics. Conventional linear mixed-effects regression was benchmarked against random forest, gradient boosting, XGBoost and deep neural networks under nested spatial-temporal cross-validation. On an independent test set, AI models substantially improved yield prediction relative to the baseline model (RMSE reduced from 3.9 to 2.6 t ha?¹; R² increased from 0.62 to 0.86). Simulation-optimization using the best model identified Pareto-optimal INM strategies that increased predicted mean yield (33.5 t ha?¹) and net returns (1, 640 USD ha?¹) while reducing mineral N input (?165 kg ha?¹) and N surplus (?75 kg ha?¹) compared with recommended fertilizer dose and farmer practice. Overall, AI-guided INM offers a scalable pathway for site-specific, climate-robust nutrient recommendations that enhance profitability while limiting nitrogen losses in potato systems.

DOI: 10.33545/27076571.2025.v6.i2d.210

Pages: 308-315 | Views: 28 | Downloads: 17

Download Full Article: Click Here