A Machine Learning-Based Probabilistic Approach for Irrigation Scheduling

Accurate prediction of irrigation requirements ensures that water is applied only when necessary, reducing wastage and conserving this precious resource. This study provides a probabilistic framework for determining the irrigation requirements of crops, referred to as the Irrigation Factor (IF). IF was calculated based on three indicators - soil moisture (SM), leaf area index (LAI), and evapotranspiration (ET). Irrigation requirement is determined based on a three-step methodology. First, relevant variables for each indicator are identified using a Random Forest regressor, followed by the development of a Long Short-Term Memory (LSTM) model to predict the three indicators. Second, errors in the simulation are calculated for each indicator by comparing the predicted and actual values in the historical time period, which are then used to calculate the error weights (normalized) of the three indicators for each month to also capture the seasonal variations. Third, we calculate the lower and upper limits by adding the error values (5th and 95th percentiles) to a simulated value. Using these values, we determine the mean, minimum, and maximum levels of irrigation requirement based on the levels’ threshold values. To determine the final levels of irrigation requirement at a daily time scale, we multiply the calculated levels (mean, minimum, and maximum) for each indicator by their respective weights. The outcome derived from the test case indicated that while certain variables exhibited no demand for water, there was a necessity for irrigation in other cases, and vice versa. This holistic approach to irrigation scheduling helps to ensure that plants receive adequate water while minimizing water wastage and promoting sustainability. It is especially valuable for agricultural operations, where optimizing water usage is essential economically and environmentally.