Control-Oriented Forecasting for Soil Moisture

The challenge of increasing irrigation efficiency to meet the demands of a growing population while protecting natural resources requires the contributions of multiple disciplines, including engineering, agronomical, horticultural, and environmental sciences. Specifically, automatic control can play a pivotal role in improving irrigation scheduling. In this context, incorporating real-time soil moisture (SM) forecasting in irrigation can potentially improve the efficiency of crop water management. However, the complexity of the analytical models that describe soil-water dynamics limits the development of practical and accurate solutions that include SM forecasting in decision-making. Currently, irrigation decisions are based on present and past SM data. This approach can be enhanced if, in addition to those, future or SM forecasting is incorporated. We formulated an SM model-based moving horizon estimation (MHE) and prediction strategy. For this, we propose a parametrizable blue SM control-oriented prediction model (SMCOPM) that obeys a soil-water balance. The SMCOPM is periodically parametrized using a proposed MHE approach, which provides adaptability, guarantees optimality, prevents overfitting, and ensures the water balance fulfillment and stability of the SMCOPM. The SM forecasting is performed by solving the parametrized SMCOPM as a function of rain, irrigation, and temperature forecasts. We evaluated the MHE and prediction strategy using, as a case study, observed data from a commercial sweetcorn field using subsurface irrigation in South Florida. The results show that by using this strategy, the SM can be predicted three days in advance with an average SM prediction error and a dispersion that significantly improves as the SMCOPM adapts over time, demonstrating convergence toward an error less than 2% and dispersion less than 3%. Consequently, the results corroborate the SMCOPM suitability, the proposed estimation strategy’s quality, and the SM behavior’s predictability. The proposed strategy has the potential for use in formulating predictive control approaches toward automating the irrigation process or scheduling irrigation actions.