Modeling lowland wheat yield and water productivity using aquacrop and machine learning with IoT-based precision irrigation in Southern Ethiopia

Efficient irrigation plays a critical role in sustaining wheat production by improving yield and water productivity (WP), particularly in semi-arid lowlands where water availability and irrigation infrastructure are limited. The integration of conventional and machine learning modeling with IoT-enabled soil moisture monitoring has the potential to significantly enhance irrigation efficiency, water conservation, and agricultural productivity. Nevertheless, research employing such integrated approaches is scarce. This study addresses the integration of crop modeling, machine learning (ML), and IoT-based soil moisture monitoring for optimized irrigation water application management and improving wheat yield. In this study, AquaCrop outputs, IoT soil moisture, and climate data were combined to train and evaluate seven ML models for Ogolcho (OWV) and Kakava (KWV), wheat varieties under multiple irrigation water level applications. The AquaCrop performance was revealed to be very high performed with a coefficient of correlation R² varying from 0.97 to 0.99, and ML models (Random Forest, Gradient Boosting, and Support Vector Regression) achieved high predictive accuracy of R² = 0.95–0.99 and RMSE ≤ 0.06, enhancing yield and Water productivity estimation. These results revealed that OWV consistently outperformed KWV, with optimal performance at 100 and 125% ETc, while deficit (50% ETc) and over-irrigation (150% ETc) reduced productivity due to water stress or surplus. This integrated AquaCrop-ML-IoT approach provides a robust framework for accurate yield prediction, irrigation optimization, and improved water use efficiency in water and infrastructure-limited wheat production systems. However, the study is limited to the features considered across five irrigation levels, two wheat varieties, two growing seasons, and semi-arid environments.