Abstract

Modeling the water content in soil is important for the development of agricultural information systems. Various data are necessary for such modelling. In this paper the authors are proposing a methodology for a frequent situation, i.e., when the modeler is facing a problem due to the lack of available data. Soil water prediction, e.g., for irrigation planning, should be performed with a daily time step. Unfortunately, past measurements of soil moisture, which are necessary for the calibration of a model, are often not available at such a frequency. In the case study presented the soil moisture data were acquired every two weeks. The authors have tested a model utilizing the Random Forests (RF) algorithm, which was used for the conversion of the original data to data with a daily time step. The accuracy of the application of RF to this task is compared with a neural network-based model. The testing accomplished shows that the RF algorithm performs with a higher degree of accuracy and is more suitable for this task.