A frequency-response permittivity sensor for simultaneous measurement of multiple soil properties: Part II. Calibration model tests

A frequency-response permittivity sensor was tested in soil for simultaneous measurement of multiple soil properties. Frequency responses between 200 Hz and 100 MHz were measured at two penetration depths (10 and 25 mm) in soil samples with combinations of three soil textures (very fine sandy loam, silt loam, and silty clay loam), five gravimetric water contents (12% to 22%), five salinities (1.90 to 11.15 dS m(-1)), and five bulk densities (1.05 to 1.30 g cm(-3)). Accurate calibration models for measuring soil bulk electrical conductivity from the frequency-response data obtained at 1 kHz were established at the two penetration depths. The R-2 value achieved by the models was 0.999. Soil permittivity was obtained from the magnitude and phase parts of the frequency-response data after the geometric factor of the sensor was removed. The frequency-response data were analyzed using the locally weighted partial least squares (LWPLS) method. For the training data set, the LWPLS models for measuring volumetric water content (VWC) at combinations of all gravimetric water contents (GWC), salinities, and bulk densities achieved R-2 values of 0.950, 0.948, and 0.943 (with root mean square (RMS) errors of 0.010, 0.010, and 0.010 m(3) m(-3)) for the very fine sandy loam, silt loam, and silty clay loam soil materials, respectively. The R-2 values achieved for the validation data set were 0.885, 0.916, and 0.959 (with RMS errors of 0.016, 0.012, and 0.009 m(3) m(-3)) for the very fine sandy loam, silt loam, and silty clay loam soil materials, respectively. Across the three soil textures, the R-2 values were 0.910 and 0.913 (with the same RMS error of 0.013 m(3) m(-3)) for training and validation, respectively. In measuring salinity, the highest R-2 values achieved by the LWPLS training models were 0.797, 0.702, and 0.443 (with RMS errors of 1.097, 1.044, and 1.282 dS m(-1)) for the very fine sandy loam, silt loam, and silty clay loam soil materials, respectively. For validation, the R-2 values were 0.726, 0.559, and 0.517 (with RMS errors of 1.314, 1.375, and 1.313 dS m(-1)), respectively. Across the three soils, the R-2 values were 0.731 and 0.719 (with RMS error of 1.113 and 1.176 dS m(-1)) for training and validation, respectively. In. general, as the clay content of the soil materials increased, the R-2 values achieved by the models for predicting VWC increased, but for salinity decreased.