Irrigation demand model for open field vegetable crops based on artificial neural networks

Open field vegetable production requires effective irrigation management. The challenges of climate change and both economic and environmental constraints require optimized irrigation models, which precisely estimate the crops' water demand. This is faced with demands for user-friendliness that are mandatory for wide-ranging implementation. Minimizing the complexity of irrigation models, i.e., by reducing the number of input parameters, offers a solution. Machine learning algorithms are a powerful tool to estimate water demand based on a small number of parameters and to map the nonlinear processes of the water balance. We parameterized an artificial neural network (ANN), which models the daily soil moisture of three soil layers from 0 to 60 cm depth, and the inferred irrigation demand. Reference values for soil moisture, calculated from tensiometer data, are obtained from two-year irrigation trials comprising four crop sets of spinach, grown on sandy loam in Geisenheim, Germany. The ANN was initially trained with temperature, net irradiation at the crop surface, relative humidity, wind speed, precipitation, and cumulative temperature and irradiation as input parameters. Additional information was the water supply by the Geisenheim irrigation scheduling, and a starting value of soil moisture, which corresponded to the measured value of the previous day. The Olden algorithm identified the main driving input parameters for each model. Model complexity was reduced by limiting the number of input parameters to these identified variables. Results revealed that the ANN can accurately predict soil moisture for all three soil depths. ANNs are suitable for an effective water demand analysis of open field spinach. Implementing this ANN in an application that automatically retrieves weather data may contribute to a sustainable, user-friendly, and demand-based irrigation management in open field vegetable production.