Arduinos in the wild: a novel, low-cost sensor network for high-resolution microclimate monitoring in remote ecosystems.
The importance of microclimate conditions is becoming increasingly recognised in ecological research, especially as they can differ widely from macro- and mesoclimate. Effective measurement of microclimate variability in heterogeneous field conditions requires measurements from a large number of sensors, ideally sampling across both small and large scales, which necessitates the development of cost-effective sensor networks. Here, we develop an environmental microcontroller (EMU) sensor network for measuring soil moisture and temperature with high spatiotemporal coverage. The system can easily be implemented in larger or smaller setups, though our study consisted of 40 plots distributed across 4 sites in northern Lapland. Using 40 EMUs, each equipped with 10 soil temperature and 10 soil moisture sensors, we collected roughly 3.5 million unique soil temperature and moisture records over the course of about 2 months. We then compared these measurements to those from commercial temperature sensors (iButtons) and TDR soil moisture probes. We show that our sensor network is able to successfully characterise microclimate variation at the site, plot and within-plot scales with high accuracy and good reliability. Moreover, even including development costs, the total price per EMU unit was less than €100, yielding a much cheaper and higher resolution system than is currently available from commercial producers. For both temperature and soil moisture, we found a strong relationship between the sensor network measurement and those obtained from commercial systems. Roughly 10% of the overall spatial variation in temperature and moisture occurred within plots-that is. over the scale of 25-50 cm. Our results show that accurate measurements of microclimate soil temperature and soil moisture can be obtained using low-cost sensor networks. We also show that microclimate variability at small scales can make up a considerable fraction of total environmental variability. To this end, we discuss some of the difficulties encountered during the development process and suggest possible improvements for further studies. While the sensor network described here can be applied without any modifications for quantifying topsoil conditions, we note that with minor changes to our code and hardware, the system can be customised for other kinds of measurements.