Machine Learning based Soil VWC and Field Capacity Estimation Using Low Cost Sensors

The amount of water present in soil is measured in terms of a parameter commonly referred to as Volumetric Water Content (VWC) and is used for determining the field capacity of any soil. It is an important parameter accounting for ensuring proper irrigation at plantation sites for farming as well as for afforestation activities. The current work is an extension to already going on research in the area of wireless underground sensor networks (WUSNs). Sensor nodes equipped with Decagon 5TM volumetric water content (VWC) and temperature sensor are deployed underground to understand the properties of soil for agricultural activities. The major hindrances in the deployment of such networks over a large field are the cost of VWC sensors and the credibility of the data being collected by these sensors. In this paper, we analyze the use of low-cost moisture and temperature sensors that can either be used to estimate the VWC values and field capacity or cross-validate the data of expensive VWC sensors before the actual deployment. Machine learning algorithms, namely Neural Networks and Random Forests are investigated for predicting the VWC value from low-cost moisture sensors. Several field experiments are carried out to examine the proposed hypothesis. The results showed that low-cost moisture sensors can assist in estimating the VWC and field capacity with a minor trade-off.