Neural Models to Predict Irrigation Needs of a Potato Plantation

Abstract

Reducing water consumption is an important target required for a sustainable farming. In order to do that, the actual water needs of different crops must be known and irrigation scheduling must be adjusted to satisfy them. This is a complex task as the phenology of plants and its water demand vary with soil properties and weather conditions. To address such problem, present paper proposes the application of time-series neural networks in order to predict the soil water content in a potato field crop, in which a soil humidity probe was installed. More precisely, Non-linear Input-Output, Non-linear Autoregressive and Non-linear Autoregressive with Exogenous Input models are applied. They are benchmarked, together with different interpolation methods in order to find the best combination for accurately predicting water needs. Promising results have been obtained, supporting the proposed models and their viability when predicting the real humidity level in the soil.