*2.3. Sensors Supported*

As there are low-cost sensors that are able to provide measurements of high accuracy with a careful calibration [36] or by using deep-learning-based sensor modelling [37], more than 80 different sensors were tested and evaluated to select the ones with the highest accuracy and durability. In the case of the ones that passed these functional tests, in some cases (pH, temperature, and turbidity sensors), modifications were made to increase their accuracy and make them waterproof. Waterproofing was achieved by potting the sensitive electrical/electronic parts, wiring, and connections of the aforementioned sensors using epoxy resin. Moreover, as the majority of the low-cost sensors were OEM-branded operating using circuits developed from multiple manufacturers (e.g., TDS, pH, and Ultrasonic level sensor), new circuits were developed and embedded into the IoT node for ensuring the proper functionality of the low-cost sensors as well as their measurements' accuracy. The sensors that have been supported to date by the IoT node are:


#### *2.4. Actuation*

To enable remote control and automation, the communication between the node and the cloud was bidirectional, and the actuation could be achieved by remote control through a website in which the user can:


#### *2.5. Field Trials*

The ability of the system to efficiently manage different water sources for automated irrigation scheduling was evaluated at 3 different pilot sites developed for the needs of the HYDROUSA project. More specifically, the evaluation procedure was held at Ano Mera, Mykonos, Greece (37◦26 51.4 N 25◦24 15.7 E), at Agios Fokas, Tinos, Greece (37◦31 59.1 N 25◦10 44.0 E), and at an eco-tourist facility in Tinos Greece (37◦33 56.7 N 25◦12 55.5 E). Field trial tests included the evaluation of: (i) low-cost sensors' accuracy, (ii) system's monitoring and water management capabilities, (iii) automated irrigation scheduling efficiency, and (iv) the system's energy autonomy.

### **3. Results**

#### *3.1. Evaluation of the Low-Cost Sensors' Accurancy*

The water quality sensors were tested and evaluated at the pilot site of Agios Fokas, Tinos, Greece, by comparing their measurements with industrial type sensors that were installed in parallel in closed tanks used for water storing. Both low-cost (Temperature: DS18B20, AGENSO, Athens, Greece; pH: H-101, HAO SHI, Taiwan) and industrial sensors (Temperature and pH: Sensolyt 700 IQ, YSI, Yellow Springs, OH, USA) were calibrated before their installation. The measurement rate was 1 h for the low-cost sensors and 15 min for the industrial sensors. To compare their results, the average daily values of each sensor were calculated. For pH measurements, the maximum difference recorded between the low-cost and the industrial sensor was 0.22 with a mean difference at 0.08 and R2 = 0.8392 (Figures <sup>3</sup> and 4), with the low-cost sensor having an accuracy of ±0.1 at 25 ◦<sup>C</sup> and the industrial one ±0.05 (from 0 ◦C to 60 ◦C). For water temperature measurements, the maximum difference recorded was 1.78 ◦C with a mean difference of 0.75 ◦C and R2 = 0.9914 (Figures <sup>5</sup> and 6), with both sensors having an accuracy of ±0.5 ◦C (low cost: from −10 ◦C to +85 ◦C; industrial: from 0 ◦C to 60 ◦C). The average values per day and their differences are shown in Table 1.

**Figure 3.** Comparison of industrial and low-cost pH sensors' measurements.

In order to perform a robust comparison and further determination of the statistically significant differences between the obtained measurements, an analysis of variance was performed by conducting a one-way ANOVA using a Fisher's least significance difference (LSD) test at a 95% confidence level (*p* < 0.05). As pH industrial sensors have a temperature compensation function to correct the measured pH value according to the water temperature, the accuracy of the low-cost pH sensor had to be improved. For this reason, a firmware update of the IoT node was developed and will be tested in summer 2022, to regulate the pH results according to the water temperature, for increasing the accuracy of the low-cost pH sensor. The low-cost temperature sensor that was used showed a very high level of accuracy (R<sup>2</sup> = 0.9914), which was proved also during the initial lab tests that were conducted.

**Figure 4.** Correlation of industrial and low-cost pH sensors' measurements.

**Figure 5.** Comparison of industrial and low-cost temperature sensors' measurements.

As no low-cost sensors for measuring electrical conductivity on water exist, the measurements of a low-cost TDS sensor (TDS-1000, AGENSO, Athens, Greece) were evaluated in comparison with the measurements of an industrial sensor for measuring electrical conductivity. Thus, the TDS and electrical conductivity (EC) on water are correlated. As shown in Figure 7, the results show a high correlation between the TDS and EC measurements at an EC up to 3 mS/cm. After that point, the correlation was lower as the TDS sensor reached its maximum range. The general rule for the salinity hazard of irrigation water based upon conductivity is that EC over 3 mS/cm creates severe damage to crops [38,39]. The low-cost TDS sensor can be used to evaluate the quality of the water and its properness for irrigation, or to select crops that are tolerant to saline water.

In addition, as weather parameters are the most important factors in decision making in agriculture, the selected low-cost that station (MeteoIoT 2100S, AGENSO, Athens, Greece) was evaluated in an experiment which run at the Municipality of Trikala, Greece, where the data of the station were compared with the data of a high-end weather station (Vantage Pro 2, Davis, Hayward, CA, USA) used by the municipality. This high-end station was connected to the network of weather stations of the National Observatory of Athens, which

is the largest network of weather stations in Greece, used for weather monitoring and forecasting. Both stations were installed in open places within the municipality and their distance in a straight line was about 400 m. Figure 8 presents the comparison of the daily average temperature and the total rain recorded using the low-cost and the high-end weather stations for a 30-day period. The average temperatures recorded using the low-cost and high-end weather stations were 10.33 ◦C and 10.37 ◦C, respectively, while the total rain recorded was 142.50 mm for the low-cost weather station and 145.80 mm for the high-end weather station, proving the reliability of the measurements retrieved with the low-cost weather station (Table 2).

**Figure 6.** Correlation of industrial and low-cost temperature sensors' measurements.

**Figure 7.** Comparison between the low-cost TDS and industrial EC sensors to determine water salinity for irrigation needs.


**Table 1.** Comparison of the measurements of the low-cost and industrial sensors. The different letters accompanying daily means and monthly average values of each distinct measurement type (pH and temperature) for each set of the industrial and low-cost sensors indicate a significant difference between the measurements, based on a Fisher's least significance difference (LSD) test (*p* < 0.05).


**Table 2.** Comparison of measurements between the low-cost and high-end weather stations. Different letters accompanying monthly means and monthly sum values of each distinct measurement type (temperature and total rain) for each set of the industrial and low-cost sensors indicate a significant difference between the measurements, based on a Fisher's least significance difference test (*p* < 0.05).

Daily measurements, meaning daily average temperature and sum of total daily rain, were obtained from the open source access www.meteo.gr [40], supported by the National Observatory of Athens, as single point measurements; thus, any further analysis of variance between the measurements was not applicable due to the lack of access to the hourly data from which the means and sums were generated. As a result, a statistical analysis was performed at monthly level, using the available data, for assessing and determining the statistically significant differences between the monthly values, by conducting a one-way ANOVA using a Fisher's least significance difference test at a 95% confidence level (*p* < 0.05). The results indicate the lack of statistical differences between the operation of the low-cost and industrial components for both average temperatures and total rain, indicating the sufficient function of both components in the long term.
