Soil water status and water table depth modelling using electromagnetic surveys for precision irrigation scheduling

Electromagnetic surveys have been used to quantify soil variability with respect to soil water storage in an irrigated maize field. A fluctuating water table sub-irrigates the crop in some places, and a wireless sensor network simultaneously monitors real-time depth of water table and soil moisture content, with large differences in soil moisture measured at any one time in these uniformly textured sands. These large differences justify assessment of the spatio-temporal variability of soil hydraulic properties when aiming for precision management of the resource. Regression models were used to spatially predict water table depth and moisture content at 50 cm using EM38 survey data, a rainfall time series and a wetness index extracted from a digital elevation model. A multiple linear regression modelling (MLM) approach was compared with a data‐mining approach using a random forest model (RF). The RF model implements a more thorough interrogation of the data using classification trees with subsequent regression of the data and provided the best prediction of soil moisture (R2 = 0.94; RMSE = 0.03 m3 m− 3 using RF; R2 = 0.77; RMSE = 0.06 m3 m− 3 using MLM) and water table depth (R2 = 0.91; RMSE = 7.17 cm using RF; R2 = 0.71; RMSE = 12.48 cm using MLM).

Figure optionsDownload as PowerPoint slideHighlights
► Using a wireless sensor network to acquire soil moisture and water table depth data
► Spatio-temporal modelling using EM38 data, a wetness index and rainfall
► Random forest model outperforms multiple linear regression model
► Accurate prediction of soil moisture at 50 cm and water table depth
► The model is available for high resolution precision irrigation scheduling