A calibration method of detecting soil water content based on the information-sharing in wireless sensor network

In agricultural production, there is contradiction between the cost and accuracy of detection during the course of acquiring soil water content (θ) online. This conflict is one of the core issues of automatic water-saving irrigation technology in agriculture. At the same time, capacitive soil moisture sensor (CSMS) has received considerable attention, for it can acquire θ with low cost and high precision, and meet the application requirements of wireless sensor network (WSN). But CSMS is vulnerable to the soil temperature (Ts) and salinity (Ss) in the measurement process. Therefore, this study took EC-5 sensor for example to establish water detection calibration models of soil temperature and salinity for single sensor, using Least Squares Support Vector Machines on MatLAB (LS-SVMlab) as the tool. On this basis, we explored the spatial variability of Ts and Ss, and then a method, which could be used to calibrate the output signals of sensors in multi-point network, was proposed based on the information-sharing (Ts or Ss) technology of WSN. Through laboratory experiment, we effectively reduced the impact of soil temperature and salinity on the single sensor. In example analysis, we investigated the detection precision and costs under different information-sharing radiuses (r). And the results indicated that the method we proposed based on the information-sharing technology of WSN could successfully calibrate the influence of soil temperature and salinity on sensors in multi-point network, and it was an efficacious approach to determine the balance between the calibration accuracy of moisture sensor and the investment of agricultural production. For example, while the calibration precision of soil temperature and salinity is respectively 1%, the costs can be reduced by 30%.

Research highlights
► A method based on the information-sharing is proposed for soil water content detection in agriculture of WSN.
► In this paper, geostatistics is taken as the tool for the method, which should work on the premise that calibration models for the single sensor EC-5 were established using Least Squares Support Vector Machines on MatLAB (LS-SVMlab).
► Laboratory experiments and example analysis show that this method can effectively solve the contradiction between the detection precision and costs for water detection in wireless sensor network used in farmland, that is to reduce the detection costs and improve the accuracy on the premise of not increasing the cost of a single sensor.