Development and Validation of a Neural Network Model for Soil Water Content Prediction With Comparison to Regression Techniques

A set of neural networks, each of which predicts soil water content at a given depth as a function of soiltemperature, has been developed. Networks were developed to predict at depths of 0.15 m, 0.3 m, 0.6 m, and 1.2 m. Inputdata consists of a coefficient describing soil type and soil temperatures measured at two depths above and two depths belowthe desired depth. The networks were trained and tested using data from two sites in Oklahoma, and data from a third sitewere used for additional testing. The networks were generally able to predict the variations in water content over time.Accuracy, as assessed by root mean square error in predicted value, ranged from 0.0142 to 0.0221 m3 m3. Networkperformance generally improved as deeper depths were considered. For comparison purposes, multiple linear regressionmodels were also estimated. They did not perform well, particularly in following the trends in the data over time.