<i>Model-Predictive Real-Time Fertilization and Irrigation Decision-Making Using RZWQM</i>

Abstract. Agriculture productivity and impact is dependent on fertilization and irrigation decisions. There exists a tradeoff between yield and application cost: Low application can compromise yield, while high application may be costly without improving yield, while also polluting the environment. Farmers generally use their experience and general guidelines to decide amount and time of irrigation and fertilizer application. For instance, high yielding corn requires 20 to 25 inches of water and around 150 to 200 kg N fertilizer per hectare. Currently, no good real-time decision-making tool exists that factors into the current conditions, together with the future weather forecasts. In this regard, we propose a model-predictive real-time decision-making framework. The recommended amounts are determined by running forward simulation of a calibrated RZWQM agriculture model each day with different combination of irrigation and fertilization amounts. Optimization is done to maximize the net profit, which is the yield minus the input application cost. The optimization steps are repeated each day, and the recommendations for only the current application period are applied if the gain in yield due to the application is greater than a certain threshold fraction. We compare our above model-predictive real-time decision-making strategy with the case of an off-line decision-making that is based on guidelines or manual interventions, as in the case of the USDA field setup in Greeley, Colorado, from where we obtained the field data for RZWQM calibration.