Calibration of capacitive soil moisture sensor based on random forest

Abstract

Objective Soil moisture information crucial for various applications, such as natural ecological restoration, farmland irrigation management, and soil engineering construction. One of the main sensors used to obtain this information is the capacitive soil moisture sensor. Methods To accurately calibrate the soil water content observation data of the 5TE capacitive soil water sensor, soil dielectric permittivity, electric conductivity and temperature observation experiments were carried out under different temperature, salt content and soil water content conditions. A soil water content estimation model based on the random forest machine learning method was established. Results The results showed that: ① The soil dielectric permittivity was significantly affected by varying salinity and temperature with constant soil water content. The traditional soil water content estimation model based only on soil dielectric permittivity became invalid, ② The soil water content estimation model based on the random forest method could effectively improve the soil water content estimation with the soil dielectric permittivity, electric conductivity and temperature data as input. Random forest method obtained soil moisture estimation with RMSE=0.05 m3/m3 and R2=0.77, while RMSE=0.07 m3/m3 and R2=0.54 were obtained by the modified Topp equation, and ③ The soil electric conductivity was the most important factor for soil water content estimation, followed by the dielectric permittivity and temperature. Nevertheless, the importance of the dielectric permittivity and temperature did not reach a negligible level. Conclusion This study provides a way to support the successful application of capacitive soil moisture sensors in areas with variable temperature and salinity.

Keywords

• soil water content
• dielectric permittivity
• electric conductivity
• temperature
• random forest
• soil moisture sensor