Calibration of Soil Water Content Data from EnviroSCAN System Using Artificial Neural Network

Abstract

Irrigation is one of the essential issues in agriculture in developing countries. Usually, in the developing countries, traditional farmers are likely to use more water than the required for crop production, thus wasting water. Hence, soil water sensors are typically needed in such situations to alert the farmer when the field needs irrigation and when it does not. One of these sensors is the EnviroScan system. It has the potential to monitor and estimate the soil water content continuously at various soil depths. Calibration is important to obtain accurate results. In this study, the volumetric soil water content and scaled frequencies from the EnviroScan system were recorded in a 60- cm soil profile. An artificial neural network (ANN) was used to calibrate the soil water content compared with a regression analysis using field data at different soil depths in sandy clay loam soil. Several ANN architectures were employed in order to determine the optimum architecture. The coefficients of determination (R2) of a regression calibration equation of scaled frequency against the gravimetric soil water content were 0.9225, 0.9623, and 0.9593 for 0–20 cm, 20–30 cm, and 30–60 cm soil depths. The R2 between gravimetric soil water content and the estimated by ANN model was 0.9928 for a 0–20 cm soil depth, 0.9809 for a 20–30 cm soil depth, and 0.9878 for a 30–60 cm soil depth. Using the data set for the entire 60-cm soil profile for calibration by ANN model, the R2 value was 0.9715.