Estimating Volumetric Water Content in Soil for IoUT Contexts by Exploiting RSSI-Based Augmented Sensors via Machine Learning.

This paper aims at proposing an augmented sensing method for estimating volumetric water content (VWC) in soil for Internet of Underground Things (IoUT) applications. The system exploits an IoUT sensor node embedding a low-cost, low-precision soil moisture sensor and a long-range wide-area network (LoRaWAN) transceiver sending relative measurements within LoRaWAN packets. The VWC estimation is achieved by means of machine learning (ML) algorithms combining the readings provided by the soil moisture sensor with the received signal strength indicator (RSSI) values measured at the LoRaWAN gateway side during broadcasting. A dataset containing such measurements was especially collected in the laboratory by burying the IoUT sensor node within a plastic case filled with sand, while several VWCs were artificially created by progressively adding water. The adopted ML algorithms are trained and tested using three different techniques for estimating VWC. Firstly, the low-cost, low-precision soil moisture sensor is calibrated by resorting to an ML model exploiting only its raw readings to estimate VWC. Secondly, a virtual VWC sensor is shown, where no real sensor readings are used because only LoRaWAN RSSIs are exploited. Lastly, an augmented VWC sensing method relying on the combination of RSSIs and soil moisture sensor readings is presented. The findings of this paper demonstrate that the augmented sensor outperforms both the virtual sensor and the calibrated real soil moisture sensor. The latter provides a root mean square error (RMSE) of 3.33%, a virtual sensor of 8.67%, and an augmented sensor of 1.84%, which improves down to 1.53% if filtered in post-processing.