Abstract
Soil moisture exhibits high spatio-temporal variability that necessitates dense monitoring networks, yet the cost of commercial sensors often limits widespread deployment. Despite the mass production of low-cost capacitive soil moisture sensors driven by IoT applications, significant gaps remain in their robust characterisation and in the availability of open-source, reproducible monitoring systems. This study pursues two primary objectives: (1) to develop an open-source, low-cost, off-grid soil moisture monitoring station (picoSMMS) and (2) to conduct a sensor-unit-specific calibration of a popular low-cost capacitive soil moisture sensor (LCSMS; DFRobot SEN0193) by relating its raw output to bulk static relative dielectric permittivity (ϵs), with the additional aim of transferring technological gains from consumer electronics to hydrological monitoring while fostering community-driven improvements. The picoSMMS was built using readily available consumer electronics and programmed in MicroPython. Laboratory calibration followed standardised protocols using reference media spanning permittivities from 1.0 (air) to approximately 80.0 (water) under non-conducting, non-relaxing conditions at 25 ± 1 °C with temperature-dependency characterisation. Models were developed relating the sensor's output and temperature to ϵs. Within the target permittivity range (2.5-35.5), the LCSMS achieved a mean absolute error of 1.29 ± 1.07, corresponding to an absolute error of 0.02 ± 0.01 in volumetric water content (VWC). Benchmarking revealed that the LCSMS is competitive with the ML2 ThetaProbe, and outperforms the PR2/6 ProfileProbe, but is less accurate than the SMT100. Notably, applying the air-water normalisation procedure to benchmark sensors significantly improved their performance, particularly for the ML2 ThetaProbe and PR2/6 ProfileProbe. A brief field deployment demonstrated the picoSMMS's ability to closely track co-located HydraProbe sensors. Important limitations include the following: inter-sensor variability assessment was limited by the small sensor ensemble (only two units), and with a larger sample size, the LCSMS may exhibit greater variability, potentially resulting in larger prediction errors; the characterisation was conducted under non-saline conditions and may not apply to peat or high-clay soils; the calibration is best suited for the target permittivity range (2.5-35.5) typical of mineral soils; and the brief field deployment was insufficient for long-term validation. Future work should assess inter-sensor variability across larger sensor populations, characterise the LCSMS under varying salinity, and conduct long-term field validation.