Precision nitrogen management for maize based on crop modeling, remote sensing and machine learning

The simultaneous improvement of crop yield and nitrogen (N) use efficiency (NUE) can be potentially achieved through precision N management. In this study, crop modeling, remote sensing, and machine learning were combined to develop and assess a precision N recommendation strategy (MSM-PNM) for maize (Zea mays L.) based on twenty-eight site-years of field experiments conducted in Northeast China involving various N rates and planting densities. For the MSM-PNM strategy, a crop growth model was used at the eight-leaf stage of maize to initially predict the in-season optimal side-dressing N rate (EOSN) by combining current season's available weather data prior to side-dressing with weather data from previous years for the remaining growing period. Maize N status was then estimated using machine learning models based on data collected with an active canopy sensor along with weather, soil, and management information. Finally, the crop model predicted EOSN was adjusted using the estimated maize N status. The prediction accuracy of EOSN (R² = 0.70, root mean squared error (RMSE) = 18.60 kg ha⁻¹) based on this integrated MSM-PNM strategy was higher than using crop model only (R² = 0.65, RMSE = 20.10 kg ha⁻¹). The precision maize management system based on the integrated MSM-PNM strategy decreased N rates by 6–16 % and increased NUE by 8–18 % over farmer practice applying 250 kg N ha⁻¹ as basal N fertilizer without side-dress N application and regional optimal management practice applying split N at fixed rate and timing, while maintaining high grain yield and marginal net return. It is concluded that this new integrated precision N management strategy combining the advantages of crop modeling, remote sensing, and machine learning can significantly increase maize NUE while maintaining high crop yield, thus contributing to food security and agricultural sustainability.