New insights on measuring soil water content with low-cost sensor Open Access

There can be no food security without water. Effective global water management is crucial, as water is a fundamental natural resource for humanity. However, only 2.5% of the world's freshwater is suitable for human consumption, and most of it is either frozen or stored underground, with less than 1% being readily accessible (Mishra 2023). Discussions about water management are important because water resources directly impact economic activities, such as agricultural production (Testezlaf 2017; Rodina 2019; Daoudy et al. 2022), which must meet the food demands of a projected global population of approximately 9.7 billion by 2050 (Béné et al. 2015). Therefore, improving techniques that enhance water efficiency in agricultural practices can boost production by reducing water deficits and minimizing unnecessary irrigation costs (Levidow et al. 2014; Hara & Gonçalves 2019).

It is, therefore, necessary to monitor variations in soil water content to enable proper management of its use (Berger et al. 2020; Braz et al. 2020). At the river basin management level, several important hydrological processes, such as surface runoff, infiltration, erosion, and groundwater recharge, must be continuously monitored (He et al. 2019; Reichert et al. 2020) to provide early warnings for disasters like flash floods, which rank first among all natural hazards worldwide due to their high frequency, severity, and mortality (Varlas et al. 2019; Zhai et al. 2021). Hydrological modeling is an effective tool for simulating a basin's response to intense rainfall (Zhang et al. 2020). However, to make this feasible, it is crucial to determine when and under what conditions a given monitoring area will be affected by a flood. This is only possible when factors such as precipitation and antecedent soil moisture are continuously monitored, as these variables directly influence water runoff (Crow et al. 2017; Zhai et al. 2018).

There are several methods to measure soil water content (Romano 2014; Chen et al. 2021). These measurements can be obtained directly, such as through gravimetric moisture determination, or indirectly using non-destructive techniques (Topp et al. 1980; Fares & Polyakov 2006), such as the use of various types of sensors (Batista et al. 2016; He et al. 2021). Modern indirect methods include soil resistance-type sensors, tensiometers, and soil capacitance techniques, such as time domain reflectometry (TDR), frequency domain reflectometry (FDR), capacitive sensors, and soil moisture thermal flux sensors (Noborio et al. 1996; Mittelbach et al. 2012; Su et al. 2014). Among these, capacitive sensors are the most widely used due to their low manufacturing and operating costs. This technique involves placing a soil sample in contact with a dielectric material, which conducts little or no electrical current (Dean et al. 1987; Neves 2001). Researchers in countries with limited investment in research and development (R&D), particularly in underdeveloped regions, seek economically viable alternatives to conduct their studies.

Capacitive sensors function by passing a current through the soil and measuring resistance proportional to the soil's moisture level. These sensors are powered by a 3.3 volt battery, simplifying their integration with other devices. Additionally, their availability and affordable price make them suitable for widespread use (Kevin et al. 2020; Pavan et al. 2023; Tumpa et al. 2023).

The use of capacitive sensors, combined with a platform that facilitates practical measurements over time and space, enables dynamic water resource management even with limited budgets (Bégué et al. 2018; Spinelli et al. 2019). This technique involves placing a soil sample in contact with a dielectric material, where the measured capacitance value is affected by changes in the dielectric constant due to variations in soil moisture (Scudiero et al. 2012; Deng et al. 2020). Real-time data acquisition can be further optimized with Internet of Things (IoT) technology. The Arduino platform, for instance, is an effective tool for real-time monitoring of soil water content, generating a decision-support system for determining irrigation depths (Abba et al. 2019; Suresh et al. 2022). However, different soil types affect calibration equations differently, making calibration essential for evaluating the performance of soil sensors (Jiménez et al. 2019).

This study aims to estimate the impact of three different soil sample volumes on sensor calibration, evaluate the sensor's performance in two soil types, and determine the appropriate time interval between readings to enhance the performance of the capacitive sensor v1.2.

The choice of these soils was based on their significant agricultural importance in southern Brazil. Their granulometric compositions are presented in Table 1.

The experiment followed the methodology described by RoTimi Ojo et al. (2015), Muzdrikah et al. (2018), and Pereira et al. (2022). Once the soil reached a moisture level equivalent to field capacity, the first soil weighing was performed using a scale with an accuracy of ±0.2 g, and soil moisture readings were taken using a capacitive soil sensor v1.2. Sensor readings and sample mass measurements were recorded every 5 min, with ambient temperatures maintained between 18 and 23 °C. Small temperature variations do not significantly impact sensor performance compared with the influence of soil composition (Nagahage et al. 2019). Data collection continued until the sample masses stabilized. Subsequently, the samples were placed in an oven to determine their dry mass.

To assess the reliability of the equations, the performance index test (I_d), proposed by Camargo & Sentelhas (1997), was used. This index is the product of the Willmott agreement index and the correlation coefficient. The results were interpreted using the criteria in Table 2.

After processing the obtained data, it was possible to assess the degree of correlation between the study variables, that is, the percentage of water contained in each soil sample and its respective reading by the capacitive sensor. These numbers, presented in Table 3, denote a strong negative correlation for most of the sets, indicating that as one value decreases, the other must increase proportionally, confirming the results found by Bogena et al. (2007), which show that sensor readings should decrease with increasing soil water content. However, sets V1D1R2, V1D1R3, and V1D2R1 of Soil 1 and sets V1D1R1, V1D1R2, and V1D2R2 of Soil 2 showed weak negative correlation values (Lepsch 2002; Figueiredo & Silva 2009).

The values presented in sets V1D1R2 and V1D1R3 of Soil 1, and V1D1R1 of Soil 2, exhibit significant discrepancies compared with the other sets for this diameter. The readings performed by the sensor show the same value for quite different percentages of water volume. For example, set V1D1R2 in Soil 1, where a value of 340 can represent a range of soil moisture from 10 to approximately 27% of the sample volume. The same occurs with set V1D1R1 of Soil 2, where the sensor reading range, with values ranging from 655 to 675, can represent the same soil moisture value.

Table 4 shows that increasing the interval between readings slightly affects the coefficient of determination, which may be higher or lower compared with readings taken at 5-min intervals. However, overall, all time intervals exhibit similar values, with some variations attributed more to the small sample size than to the actual reading time intervals. The Willmott concordance index, which compares the actual measured values with those obtained by the calibration equation, decreased as the interval between readings increased.

Thus, increasing the time interval between sensor calibration collections reduces its reliability, as the data resolution becomes lower. This is evident in sample 5, where the coefficients of determination obtained through linear regression analysis showed R² = 0.20 for a 5-min interval, R² = 0.29 for daily readings, and R² = 0.90 for readings taken every 72 h. Therefore, time intervals greater than 24 h are not recommended for the calibration of capacitive sensor v1.2.

Table 5 presents the results obtained from the coefficients of determination, linear calibration equations, RMSE, Willmott concordance index, and performance index for the 5-min time interval between collections for each sample, both for Rhodic Paleudalf (Soil 1) and Rhodic Hapludox (Soil 2).

The data show that samples with a 50 mm diameter present a higher RMSE compared with larger samples. However, when comparing data from 100 mm diameter samples with 150 mm diameter samples, there is only a slight difference in their RMSE. Therefore, it can be inferred that the primary difference between treatments (different volumes) was the direction in which the sensor was inserted into the soil for the 50 mm diameter samples.

It is confirmed that vertically inserted sensors increase the RMSE values of water moisture content in soil samples, directly affecting the reliability of calibrations performed with capacitive sensors placed in this manner. The analysis proposed by Camargo & Sentelhas (1997) allows for a qualitative classification of the calibration results obtained. Thus, for the Rhodic Paleudalf samples, it was concluded that those with a 50 mm diameter had an average performance index of 0.56 (poor), while those with a 100 mm diameter presented 0.80 (very good), and the 150 mm diameter samples showed an average of 0.78 (very good). The overall average was 0.719, considered to be good. However, larger volume samples demonstrated high-performance index values in some cases, such as sample 16, which reached a value of 0.98, classified as excellent. The slightly higher average for the 100 mm diameter samples compared with the 150 mm diameter samples can be attributed to occasional errors in readings from samples 13 (Soil 1) and 15 (Soil 2).

The results indicate that sensor v1.2 can be effectively used to estimate soil moisture. Polynomial regression analysis shows that this method better explains the variables involved in sensor calibration for estimating soil water content (Baumhardt et al. 2000; Evett et al. 2006; Chen et al. 2019). However, achieving a good calibration equation requires careful consideration of all aspects discussed in this paper. Several studies evaluating capacitive sensors for estimating soil water content have analyzed variables such as operating voltage, temperature, and electrical circuits, concluding that these factors affect the performance of capacitive sensors, though they still yield satisfactory results (Bogena et al. 2007; Chen et al. 2019; Tomar & Patidar 2019; Segovia-Cardozo et al. 2021). These studies also address the influence of sample sizes, textural composition, and reading intervals for calibration purposes.

Regarding time intervals, it is essential that sensors are calibrated using the same intervals that will be applied in the field (in situ). The variation in performance indices for linear equations suggests that soils with higher sand content require larger sample volumes. This is consistent with the findings of González-Teruel et al. (2019), who state that soil texture affects the calibration process of sensors using capacitance techniques. Therefore, for soils with more than 60% sand, it is recommended to use samples with a diameter of at least 150 mm and a height of 100 mm. Additionally, second-order equations are necessary to adequately explain the collected data, as samples with higher sand content drain faster than clayey soils, due to their water retention curve (Gubiani et al. 2009; Reichert et al. 2021).

The results from simple linear regression analysis showed R² values of 0.68, 0.76, and 0.75 for Sets 13, 14, and 15, respectively, for Rhodic Paleudalf. When the same data were analyzed using second-order polynomial regression, R² values increased to 0.86, 0.94, and 0.92. Applying the same method to larger sample volumes, regardless of soil type, allowed for most of the obtained data to be explained, resulting in high performance indices.

Based on the results obtained, it is recommended to use soil samples with a minimum diameter of 150 mm and a height of 100 mm to ensure greater accuracy in calibrating v1.2 capacitive sensors, particularly in sandy soils. For clayey soils, linear calibrations are suitable, while for sandy soils, the use of second-degree polynomial regressions is advised due to the increased variability in water loss. Additionally, it is important that the time interval between data collection does not exceed 24 h, particularly for soils with higher drainage capacity, in order to prevent loss of measurement precision. The horizontal insertion of the sensor in smaller samples (50 mm) helps to reduce result variability, improving the reliability of measurements under different soil and climatic conditions.

In conclusion, increasing the volume of soil samples enhances the performance of both linear and polynomial regression equations. Samples with a diameter of 150 mm and a height of 100 mm provided better results compared with smaller volumes. Therefore, it is recommended to use at least these dimensions for calibrating capacitive sensors v1.2. Additionally, soil type plays a significant role in sensor performance. Clayey soils exhibited higher performance levels and stronger linear correlation coefficients, while sandy soils, which experience faster water loss, benefit from the use of second-degree regression equations. During the calibration process, it is crucial to minimize the time intervals between data collection, ensuring they do not exceed 24 h.

The present work was carried out with the support of CNPq, the National Development Council Scientific and Technological – Brazil.

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.