Soil Health Monitoring with IoT and Machine Learning

Abstract

Soil degradation is detrimental to sustaining agricultural productivity and to the ecosystem as a whole. Manual sampling and laboratory analyses, soil monitoring methods have been used for decades, but these methods are costly, lack real time measures, and are very inefficient. In this paper, we describe the first integrated soil health monitoring system that automates the soil moisture, pH, NPKs, and temperature monitoring through a combination of low cost and low power IoT devices and machine learning (ML). A functioning soil health monitoring system should have real time soil data collection, soil data analyses, and soil health in functioning ML systems should predict state data (soil health and/or nutrient deficiencies), and determine the best times for irrigation and/or fertilization. A prototype of the system implemented in a variety of agricultural fields documented a 35% gain in water used, 25% less fertilizer use, and soil salinity was detected 85% correctly. Early salinity detection is crucial and of particular importance to agriculture; thus, the ML system used to predict soil state is beneficial. ML and IoT in combination captured the two primary soil monitoring systems: we now have precision agriculture. The system was designed with the end user in mind; thus, a mobile system with a soil monitoring application was designed to give users the functional ease to enact real time changes in monitoring and management.