Accurate soil data is crucial for precision agriculture.  While existing optical methods can correlate soil health to the gasses emitted from the field, in-soil electronic sensors enable real-time measurements of soil conditions at the effective root zone of a crop. Unfortunately, modern soil sensor systems are limited in what signals they can measure and are generally too expensive to reasonably distribute the sensors in the density required for spatially accurate feedback.  In this work, we determined the optimal placement of soil sensors in various agricultural fields using a genetic algorithm informed by geostatistical data.  From this, we designed and fabricated low-cost, passive RFID wireless sensor nodes for measuring nitrate content in soil. Finally, we characterized the performance of the wireless sensor nodes in both greenhouse and field experiments.

First, we built a genetic algorithm to determine the optimal placement of soil sensors in an agricultural field.  This algorithm takes a given field geometry - given as a simple geometric shape, boundary coordinates, or a satellite photo – and prescribes geographic coordinates for the placement of agricultural sensors in that field. When optimizing only to minimize the number of devices needed to achieve spatially accurate data for a given field geometry, we found that a brute force method was fast and sufficient for sensor placement.  However, when optimizing for additional objectives, such as for minimizing drone flight paths to read the sensors or obstacle avoidance, we found that a genetic algorithm was required for such complexity.

Second, we designed and fabricated wireless sensor nodes featuring a fully printed nitrate-selective potentiometric sensor paired with a RFID chip and a printed antenna.  We used print-based methods for fabricating the wireless sensor node because printing is a low-cost and scalable method which will ultimately allow us to achieve the large number of sensors required for spatially accurate data at a reasonable cost.  The antenna and nitrate sensor electrodes were screen-printed in three consecutive layers on the same substrate and encapsulated. Next, the RFID chip was attached using a pick-and-place machine. Afterwards ion-selective membrane and salt membrane layers were drop cast onto the ion-selective and reference electrodes, respectively.  Device design and fabrication choices are discussed in detail and the features of the as-fabricated device are characterized.

Finally, the overall device performance in greenhouse and field experiments was reported.  We demonstrated the wireless readout of the wireless sensor nodes using a RFID interrogator antenna. The reported soil nitrate concentration was compared against commercial sensors as ground-truth.  We propose a control scheme to operate an agricultural field with our wireless sensor network that seamlessly integrates into existing center-pivot, variable-rate irrigation systems.