Calibration Of CS655 Soil Moisture Sensors Under Sahelian Conditions: Effects Of Moisture And Temperature

Long-term in situ observations of soil moisture are essential to understand eco-hydrological processes, in the vadose zone and to provide ground reference for remote sensing, especially in Sahelian Africa where such datasets are poorly available. Since 2019, a dense network of time domain reflectometry sensors (CS655 model, Campbell Scientific) has been continuously monitoring soil moisture at the “Faidherbia Flux” experimental site in Sob, Senegal. The system records high-resolution data across multiple locations and depths, from 10 cm down to 480 cm.

However, these data face particular quality issues representative of sandy soils in semi-arid agroecosystems. The main challenges stem from (1) the limited accuracy of the standard Topp calibration under a narrow range of soil water content dominated by dry soil conditions (2) the influence of strong diurnal thermal fluctuations on dielectric measurement near the soil surface. High accuracy is particularly required when it is expected to process reliable modelling based on retention curves, very steep in the case of sandy soils.

To address these questions, we designed an experimental protocol combining in situ and laboratory calibrations. In situ calibration was performed during three distinct hydrological periods—dry (June), intermediate (January), and wet (October) to cover the full range of soil water natural conditions. The results revealed a strong correlation between CS655 readings and gravimetric moisture values (R² = 0.97), but also a consistent underestimation of actual soil moisture by CS655 sensor.

In the laboratory, undisturbed soil samples were collected from two depths (20 cm and 80 cm), chosen based on contrasting bulk densities likely to influence sensor response and potentially require distinct correction relationships. These samples were subjected to controlled temperature variations (from 25 °C to 45 °C) and progressive moisture levels (from 17% to 0%).  At a reference temperature of 25 °C, a relationship between the sensor readings and the actual soil moisture was first established, resulting in a correction coefficient for water content. This relationship confirmed the underestimation of soil water content by CS655 observed in the field. Then, for each moisture level, the slope of the sensor response to temperature was calculated. The average of these slopes defined a temperature correction coefficient.

Based on this two-step approach, we developed a three-variable calibration model, linking measured soil moisture, actual soil moisture, and soil temperature variations. Applying these corrections to field data significantly improved the accuracy and robustness of the CS655 readings. The systematic underestimation bias was corrected, and temperature-driven fluctuations were substantially reduced, allowing a more reliable interpretation of daily and seasonal moisture dynamics.

These findings highlight the importance of sensor calibration protocols for long-term soil moisture monitoring in our ecosystem type. Our work contributes to global efforts aimed at improving in situ networks and supporting satellite validation and hydrological modeling in arid and semi-arid regions.