A semi-empirical model to predict the EM38 electromagnetic induction measurements of soils from basic ground

Abstract

Electromagnetic induction (EMI) measurements (σb*) are widely used for the survey of several soil attributes, among which basic properties such as salinity (σe),water content (θw), clay (wc), organic matter (wom) and bulk density (ρb) standout. In usual practice, purely empirical models relating one of these properties to σb* are calibrated at selected sites. However, this calibration is site and time specific and has to be repeated time and again. In order to understand where the variability of the EMI empirical models comes from, it is necessary to know how the different soil properties contribute to them and, with this aim, a more physically based relationship between σb* and, at least, σe, θw, wc, wom and ρb was developed in this work, additionally including soil temperature (t). It was calibrated and cross-validated with the data from one survey carried out in a wide agricultural irrigation area in SE Spain, taking σb* measurements with the Geonics EM38 in the horizontal and vertical dipole modes and at various heights over the ground. Then, it was externally validated with the data from a second survey carried out 4 years later in the same area but in a different season. In the calibration, R2 and root mean square error (RMSE) were 0.84 and 0.18 dS m−1(41%), respectively, forthe vertical dipole orientation and 0.90 and 0.11 dS m−1(39%) for the horizontalone. In the external validation, R2 and RMSE were 0.80 and 0.24 dS m−1(44%),respectively, for the vertical dipole orientation and 0.90 and 0.13 dS m−1(38%) for the horizontal one. Therefore, because the performance of the model barely worsened as time passed by, it can be considered to represent the underlying physicalprocess and, therefore, to increase our understanding of how the soil EMI signalsare generated, with potential benefits for the planning and comparability of EMI soil measurements, specifically with the EM38, among different areas.