Two-dimensional time-lapse imaging of soil wetting and drying cycle using EM38 data across a flood irrigation cotton field

Understanding the spatiotemporal distribution of soil volumetric water content (θ, m3 m−3) at field level is required to maximise water-use efficiency in irrigated agriculture. Several commercial sensors are available; however, they only provide point-information. To value-add to this soil data, mathematical models can be used in conjunction with proximal sensed data, such as soil apparent electrical conductivity (ECa, mS m-1) or inverted ECa (σ, mS m-1). In this research, we determine if ECa from an electromagnetic (EM) instrument (EM38) at various heights (0, 0.2, 0.4, 0.6, 0.8 and 1.0 m) or σ estimated from ECa can be used to value add to limited θ at four depths (i.e., 0.15, 0.45, 0.75 and 1.35 m). Moreover, we compare which mathematical (i.e. multiple linear regression (MLR), random forest (RF), Cubist, support vector machine (SVM) and Artificial Neural Networks (ANN)) model can best be used to predict θ from σ. We also determine the number of calibration sites required along a uniform heavy-clay transect used for furrow irrigated cotton. In terms of a leave-one-out cross validation, the best Lin’s concordance between measured and predicted θ was achieved using SVM (0.91) when estimates of σ and depth used to model θ . We showed that satisfactory results could be achieved using a single calibration site. Considering the results at day 10 when permanent wilting point was evident, irrigation scheduling could be recommended based on the use of the EM38h0 (80 mS m-1) and EM38v0 (100 mS m-1) reaching critical measurements.