Abstract
The use of sensors to estimate soil water content is of great importance for managing water resources and making decisions on its multiple uses. However, commercial platforms are still very expensive, and the development of more accessible systems is essential. This study was to identify the insights in the calibration of low-cost capacitive sensors v1.2 to estimate the water content in the soil using the Arduino platform. Undeformed samples of two different soils (Rhodic Paleudalf and Rhodic Hapludox) were collected at two different depths (0 to 10cm and 20 to 30cm) and different sample volumes (196.35cm³, 785.4cm³, 1767.15cm³). The mass difference data due to sample drainage were collected every five minutes together with the reading values of the soil sensors. To evaluate the obtained results, statistical resources were used, such as Person Correlation Analysis (r), simple linear regression, second-order polynomial regression, Root-Mean-Square Error (RMSE), Willmott Index (IW) and Performance Index (ID). It was possible to verify that factors such as soil type, sample volume and the time interval between collections affect the performance of the sensors. The shortest time intervals between each reading of the sensors showed that soils with sandy texture should be calibrated with second order equations in soil samples, at least, greater than 785cm³ and with reading intervals not exceeding 24 hours.