*2.1. Experimental Setup*

The study was conducted in February 2020 at the CREA-IT institute (45◦31 21.9" N 9◦33 54.9" E, Treviglio, Bergamo, Italy). The liquid digestate used was gathered from an anaerobic digestion plant, stored in a tank and filtered using a hydrocyclone filter (Alfaturbo Hydrocyclone Sand separator 2"). The hydrocyclone filter was placed between a storage tank and the operating tank. Both tanks had a 1 m3 maximum volume. After the filtration, the filtered liquid was shaken once per day before the beginning of the tests (see below) by gently shaking the tank using a forklift truck. The storage of the liquid digestate before using it at the irrigation setup lasted for 8 days.

The characteristic of the liquid digestate before the filtration are depicted in Table 1. The digestate liquid fraction was kindly provided by the Società Agricola Pallavicina S.R.L. (Via Fara—24047 Treviglio, IT), which also undergoes quality analyses, and did not display the presence of any pathogen nor pollutants according to the Italian laws.


**Table 1.** Composition of the digestate liquid fraction used for the test. Values are means ± s.d. (3 subsamples) provided by the management office of the digester. The analyses were made on the digestate liquid fraction hydrocyclone filtering.

A water pump of 0.75 kW power (Pedrollo company, model: JSWm 2CX, San Bonifacio, Verona, Italy) was used to pump the digestate liquid fraction in the irrigation system. The operating pressure was set to 0.2 MPa. Samples were taken from the non-filtered digestate liquid fraction (DLF), from the filter outlet and from the filtered DLF to determine dry matter content and pH.

The experimental design consisted of two factors: filtrate dilution (FD) × time of the sampling within the irrigation cycle (TIME). The FD factor had four treatments: One control using freshwater and three filtrate dilutions (10%, 25%, and 50% of filtrate in freshwater). Each irrigation cycle lasted 8 h and samples were taken once per hour.

One irrigation cycle per day was performed, the irrigation cycles began with the tap water and continued with each increasing concentration of the digestate to avoid contamination. Each water-HF-DLF mixture was prepared mixing the relevant amount of tap water and HF-DLF in an operating tank and reflushing it several times with the pump. The water pump used was set at 0.2 MPa operating pressure, and the irrigation tank was filled with 400 L of DLF. Before starting each test, three samples were collected from the irrigation tank to measure the dry matter and the pH of the solution, following the methodology described above. In addition, flushing with tap water was performed by 15 min at the end of each cycle to allow for the cleaning the system. A pre-flushing was also made before the beginning of the first experiment with tap water. Within each irrigation cycle, the pump recycled part of the water or diluted HF-DLF into the tank to keep it mixed and avoid particle deposition. The irrigation system was organized by three polyethylene 1-m long dripper tubes (Stocker company N◦26085, Bozen, Italy), as replicates, with three emitters each. The dripper tubes used had 0.016 m of diameter (maximum design pressure 0.4 MPa) and were spaced 0.33 m each other. Water flow declared by the manufacturer was2Lh<sup>−</sup>1. Emitters were not changed from each cycle to the following one.

#### *2.2. Measurements and Analyses*

During the tests, the water or filtrate dropping from the tubes was collected in plastic flagons placed underneath each emitter (Figure 1). The flagons were weighted once per hour with a portable scale (RADWAG WLC6/C1/R, Radom, Poland, used with 0.1 g sensitivity) to calculate the water flow (g h−1). At the time of weighting the turbidity and the temperature of the liquid were measured. The temperature of the water or DLF were measured in ◦C using the DS18B20 digital thermometer (Maxim IC, San Jose, CA, USA). The turbidity of water or DLF were measured by using the turbidity sensor SKU:SEN0189 (Arduino, Ivrea, Italia), which was used as an indirect measurement of filtrate and water quality. The turbidity sensor SKU:SEN0189 uses light to detect suspended particles in water by measuring the light transmittance and scattering rate, which depends on the concentration of the total suspended solids (TSS) in the solution/dispersion. In particular, the sensor provide an output expressed in mV, which should be calibrated to the corresponding Nephelometric Turbidity Units (NTU). The output slightly and linearly decreases at increasing

temperatures. In addition, the relationship between output and NTU value is not linear (for a brief description see https://wiki.dfrobot.com/Turbidity\_sensor\_SKU\_\_SEN0189). In particular, the higher the sensor output, the lower the liquid NTU value. The manufacturer provide an output for pure water of 4.1 ± 0.3*V* when temperature span from 10 ◦C to 50 ◦C. Integration of the temperature and turbidity systems was made according to [21]. In the present work, the output of sensor was provided along with a direct measurement of the total suspended solids.

**Figure 1.** Design of the irrigation tests.

Dry matter and moisture content of the samples were assessed by oven drying at 105 ◦C until constant weight [22]. The pH of the samples was measured with no dilution by using a CRISON GLP21 pH-meter (Hach Lange Spain, S.L.U., Barcelona, Spain).

Then, the samples of each emitter per line were mixed and a random composite subsample of 500 mL of liquid was taken. In total, 24 sample per irrigation test were obtained (8 sampling moments × 3 irrigation lines). Each irrigation test consisted of the injection of a DLF dilution in an 8-h irrigation cycle. Thus 72 total samples of DLF released by each line were obtained. Dry matter of each subsample and its pH, following the methodology described above, were measured. For the control test (100% water), turbidity and pH were measured only before starting the test and no samples were collected during the test. This was done since these variables did not change by the time in the control from previous tests (data not shown). To monitor the air temperature and humidity of the indoor environment where the test was performed a sensor DHT22 (Guangzhou Aosong Electronics Co., Ltd., Guangzhou, China) was used. The Waterproof DS18B20 Digital Temperature Sensor was used to read the liquid temperature.

### *2.3. Computations and Statistical Analysis*

The amounts of OH− ions per ton of solution released by the emitters were computed by using the pH and used as a proxy of the potential of the irrigation with the DLF to increase the pH of a soil compared to the tap water used as a control. The analysis of variance was performed according to the statistical design by means of a general linear mixed model (Glimmix procedure in SAS/STAT 9.2 statistical package; SAS Institute Inc., Cary, NC, USA). The model used was similar to that shown in Saia et al. [23] (see the supplementary material in [23] for both a description of the procedure and the SAS package model applied) in which TIME was modelled as a repeated measurement [24]. Unbiased estimates of variance and covariance parameters were estimated by restricted maximum likelihood (REML). Repeated measurement analysis was modelled by applying a random statement with a first-order autoregressive covariance structure. In particular, the subject of reference was the emitter for the data related to the amount of liquid released, its turbidity and temperature, and it was the line for date on the pH. Denominator degrees of freedom of each error were estimated by Kenward–Roger approximation and least square means (LSmeans, see below for a definition) of the treatment distributions were computed. Data were provided both as LSmeans in figures and arithmetic means in supplemental materials, along with each standard error estimation or computation, respectively. Differences among means were compared by applying t-grouping at the 5% probability level to the LSMEANS *p*-differences. Time-sliced significance was also computed.

When the effect of time was significant, variation per unit time was modelled. Variation by time per each variable and treatment significantly varying by time was fitted to a linear distribution function and significance of the regression models were computed using the Slide Write Plus for Windows version 7.01 (Advanced Graphics Software, Inc., Encinitas, CA, USA).
