PREDICTION OF SOIL MOISTURE DATA BY VARIOUS REGRESSION TECHNIQUES

ABSTRACT

Measuring soil moisture it is usually not performed on a daily time step due to financial costs, time-consuming measurements and variability of the weather. However, since this information is an important part of many hydrological and water management studies, the author tries to solve this situation by proposing a suitable interpolation method. The goal is to predict daily moisture values so that their conformity with the test data is as high as possible. The calculation is based on the use of soil moisture data obtained by derivation from satellite images ©EUMESAT and data on temperatures and rainfall from climate database ECA&D. The article examines the use of linear and data-mining methods and uses multiple algorithms such as simple linear regression, Radom Forest, Support Vector Machines. The highest degree of precision was achieved using the SVM model with a correlation with measured data of 0.83. This indicates the suitability of using the SVM method for the prediction of soil moisture.

Keywords: soil moisture, remote sensing, data mining, support vector machines