Integrating Machine Learning and Field Experiments to Model Crop Coefficient and Yield Response of Lowland Irrigated Wheat in Southern Ethiopia

Precision irrigation is critical for enhancing water productivity and maximizing crop yield in water‐scarce areas. Crop‐specific constants, such as the crop coefficient ( K c ) and yield response factor ( K y ), are crucial in determining water requirements and productivity. This study aims to determine the K c and K y for lowland irrigated wheat in the semiarid region of southern Ethiopia, using an integration of machine learning and experimental approaches. The K c was experimentally determined across different growth stages in two consecutive years (2024 and 2025) and then compared with FAO standards and literature. Crop K y was also assessed under irrigation regimes of 75%, 50%, 125%, and 150% of the full irrigation schedule. Eight machine learning models were trained and evaluated to predict K c and K y , with the best‐performing model selected based on standard performance metrics. Experimentally, the K c increased from 0.40 initially to 1.28 midstage (Ogolcho) and 0.45 to 1.33 (Kakava), with late‐stage values of 0.77 and 0.81, respectively. Crop K y values under 75% ETc were 0.6 (Ogolcho) and 2.2 (Kakava); under 50% ETc, they were 1.5 and 1.6, respectively. Among evaluated models, gradient boosting yielded the highest performance for K c prediction ( R 2 = 1 for Ogolcho, 0.99 for Kakava). Ridge and Lasso regression best predicted crop yield response factor at various crop water requirement levels, with correlation coefficient values from 0.87 to 0.98. Model robustness was validated through cross‐validation and residual analysis and resulting in better performance. This study suggests the potential of machine learning in predicting irrigation parameters to support data‐driven irrigation scheduling and improved water management.