Treatment of Winery Wastewater with a Multistage Constructed Wetland System for Irrigation Reuse

: This paper reports a study on the performance of a multistage constructed wetland (CW) system adopted for winery wastewater and on the analysis of its suitability for irrigation reuse. The CW system treats about 3 m 3 · day − 1 of wastewater produced by a small winery located in Sicily (insular Italy). Wastewater samples were collected at the CW inlet and outlet for physical–chemical and microbiological quality characterization. CW e ﬃ ciency was evaluated on the basis of water quality improvement and of the achievement of Italian and EU irrigation reuse regulation limits. The CW system showed Chemical Oxygen Demand (COD) and Total Suspended Solids (TSS) mean removal rates of about 81% and 69%, and a maximum removal of about 99% (for both COD and TSS) occurred during grape harvest phase. The CW removal e ﬃ ciencies for nutrients were 56% for TN and 38% for PO 4 -P, considering their low average concentrations at CW inlet. The CW system evidenced an e ﬄ uent average quality compatible with the limits imposed by the Italian regulation and EU proposal regulation on the minimum requirement for water reuse. The CW vegetated area showed regular growth and vegetative development; phytotoxicity phenomena were not detected. The results of the study suggest the important role of CW systems in the treatment of winery wastewater and for their subsequent reuse in agriculture.


Introduction
The wine sector occupies a prominent position in the Italian agri-food industry, being the most important in the beverage sector. The latest estimates by the International Organization of Vine and Wine indicate that Italy is the world's largest wine producer with a production of 46.6 million hectoliters of the 263 million hectoliters of world wine production [1]. Considering that the vinification process generates wastewater volumes higher than wine production (from 1 L of wine, about 1.6-2.0 L of wastewater are produced) [2], the annual quantities of winery wastewater produced in Italy reach 10 million cubic meters. Therefore, there is a need to improve the sustainability of winery wastewater disposal; in fact, if not controlled, such disposal, although it contains mainly non-toxic organic polluting waste, can have a negative impact on natural ecosystems, with their consequent alteration [3][4][5]. Winery wastewaters are characterized by high concentrations of organic matter, with Chemical Oxygen Demand (COD) varying between about 300 and 50,000 mg·L −1 [4], Biological Oxygen Demand (BOD 5 ) between about 40% and 90% of the COD value [5,6], and Total Suspended Solids (TSS) ranging between 190 to 18,000 mg·L −1 [6,7].

The Experimental Multistage CW System
The research was conducted in a multistage pilot CW system, located in the semi-arid southeastern Sicily (latitude 36 • 45' N, longitude 15 • 01' E, altitude 60 m), which is used for the treatment of a portion of the wastewater produced by the Marabino winery (production capacity of approximately 1500 hL·year −1 ). The wastewater from the winery is mixed with the sewage produced by the toilets used by employees and visitors.
Winery wastewater (about 3 m 3 ·day −1 ) is subjected to a pre-treatment phase (coarse screening) followed by an Imhoff tank, an equalization tank (5 m 3 ), and a multistage CW ( Figure 1); then the effluent is disposed on the ground through a sub-irrigation system. The CW has a total area of about 230 m 2 and consists of a vertical subsurface flow bed (VF) (140 m 2 ) followed by a horizontal subsurface flow bed (HF) (60 m 2 ) and a free surface flow unit (FSF) (30 m 2 ). The treated wastewater volume was measured with a flow meter.
The CW beds were excavated and covered with a 4 mm thick bentonite layer to avoid leaks due to infiltration. The banks of the CW beds had been covered with jute nets to facilitate the growth of vegetation.
The VF bed was filled, from the bottom to the surface, with gravel of the following sizes: 25-40 mm for a depth of 0.40 m, 10-15 mm for a depth of 0.15 m, 5-10 mm for a depth of 0.15 m, and 2-5 mm for a depth of 0.15 m; washed sand was used for a depth of 0.10 m followed by gravel size of 2-5 mm for a depth of 0.05 m ( Figure 2). The equalization tank was equipped with a pump activated by an electronic timer for intermittent loading of the VF (five minutes every four hours and about 0.5 m 3 for each loading). Wastewater was distributed over the surface of the bed by a network of pressurised perforated pipes fed through a manifold by the pump. Wastewater was collected in a drainage pipes system placed at the bottom of the bed. Passive aeration is encouraged with four perforated pipes placed along the bottom and sides and extended approximately 0.50 m over the surface of the bed. The VF CW bed was planted with Phragmites australis L. at a density of four rhizome m −2 . The CW beds were excavated and covered with a 4 mm thick bentonite layer to avoid leaks due to infiltration. The banks of the CW beds had been covered with jute nets to facilitate the growth of vegetation.
The VF bed was filled, from the bottom to the surface, with gravel of the following sizes: 25-40 mm for a depth of 0.40 m, 10-15 mm for a depth of 0.15 m, 5-10 mm for a depth of 0.15 m, and 2-5 mm for a depth of 0.15 m; washed sand was used for a depth of 0.10 m followed by gravel size of 2-5 mm for a depth of 0.05 m ( Figure 2). The equalization tank was equipped with a pump activated by an electronic timer for intermittent loading of the VF (five minutes every four hours and about 0.5 m 3 for each loading). Wastewater was distributed over the surface of the bed by a network of pressurised perforated pipes fed through a manifold by the pump. Wastewater was collected in a drainage pipes system placed at the bottom of the bed. Passive aeration is encouraged with four perforated pipes placed along the bottom and sides and extended approximately 0.50 m over the surface of the bed. The VF CW bed was planted with Phragmites australis L. at a density of four rhizome m −2 .   The CW beds were excavated and covered with a 4 mm thick bentonite layer to avoid leaks due to infiltration. The banks of the CW beds had been covered with jute nets to facilitate the growth of vegetation.
The VF bed was filled, from the bottom to the surface, with gravel of the following sizes: 25-40 mm for a depth of 0.40 m, 10-15 mm for a depth of 0.15 m, 5-10 mm for a depth of 0.15 m, and 2-5 mm for a depth of 0.15 m; washed sand was used for a depth of 0.10 m followed by gravel size of 2-5 mm for a depth of 0.05 m ( Figure 2). The equalization tank was equipped with a pump activated by an electronic timer for intermittent loading of the VF (five minutes every four hours and about 0.5 m 3 for each loading). Wastewater was distributed over the surface of the bed by a network of pressurised perforated pipes fed through a manifold by the pump. Wastewater was collected in a drainage pipes system placed at the bottom of the bed. Passive aeration is encouraged with four perforated pipes placed along the bottom and sides and extended approximately 0.50 m over the surface of the bed. The VF CW bed was planted with Phragmites australis L. at a density of four rhizome m −2 .  The HF CW bed has a nominal hydraulic retention time (HRT) of about 110 h and was filled, to an average height of 0.60 m, with calcareous gravel (diameter 8-10 mm) and, in the initial and terminal section, with 80-100 mm of coarse volcanic gravel to prevent clogging and improve homogenous flow distribution ( Figure 3). The first half of the bed was planted with Cyperus Papyrus var. Siculus The HF CW bed has a nominal hydraulic retention time (HRT) of about 110 h and was filled, to an average height of 0.60 m, with calcareous gravel (diameter 8-10 mm) and, in the initial and terminal section, with 80-100 mm of coarse volcanic gravel to prevent clogging and improve homogenous flow distribution ( Figure 3). The first half of the bed was planted with Cyperus Papyrus var. Siculus   The planting density was determined according to further previous experiences carried out in similar environments [18,19], and adopting higher density values for the species with lower propagation capacity [20].
After plantation of vegetation (October 2013), the VF and HF units were maintained in saturated conditions until February 2014 to allow the rooting and growth of seedlings. In the first year of operation, only the Phragmites australis showed stunted growth due to the low nutrient concentrations in the treated wastewater. For this reason, at the beginning of the second growth season (March 2015), about 9.9 g·m −2 of urea fertilizer (N 46%) was applied in the VF CW, determining a complete plant cover in June 2015.

Wastewater Quality Characterization
Wastewater quality parameters were monitored from March 2014 to June 2018 with a frequency of 1 to 3 months according to the operations carried out in the winery. Monthly monitoring frequency was chosen during the harvest periods, between August and October.
A total of 104 wastewater samples were taken at the inlet and outlet of the CW beds to analyse the following physico-chemical parameters according to standard methods [21]: pH, Electrical   The planting density was determined according to further previous experiences carried out in similar environments [18,19], and adopting higher density values for the species with lower propagation capacity [20].
After plantation of vegetation (October 2013), the VF and HF units were maintained in saturated conditions until February 2014 to allow the rooting and growth of seedlings. In the first year of operation, only the Phragmites australis showed stunted growth due to the low nutrient concentrations in the treated wastewater. For this reason, at the beginning of the second growth season (March 2015), about 9.9 g·m −2 of urea fertilizer (N 46%) was applied in the VF CW, determining a complete plant cover in June 2015.

Wastewater Quality Characterization
Wastewater quality parameters were monitored from March 2014 to June 2018 with a frequency of 1 to 3 months according to the operations carried out in the winery. Monthly monitoring frequency was chosen during the harvest periods, between August and October.
A total of 104 wastewater samples were taken at the inlet and outlet of the CW beds to analyse the following physico-chemical parameters according to standard methods [21]: pH, Electrical The planting density was determined according to further previous experiences carried out in similar environments [18,19], and adopting higher density values for the species with lower propagation capacity [20].
After plantation of vegetation (October 2013), the VF and HF units were maintained in saturated conditions until February 2014 to allow the rooting and growth of seedlings. In the first year of operation, only the Phragmites australis showed stunted growth due to the low nutrient concentrations in the treated wastewater. For this reason, at the beginning of the second growth season (March 2015), about 9.9 g·m −2 of urea fertilizer (N 46%) was applied in the VF CW, determining a complete plant cover in June 2015.

Wastewater Quality Characterization
Wastewater quality parameters were monitored from March 2014 to June 2018 with a frequency of 1 to 3 months according to the operations carried out in the winery. Monthly monitoring frequency was chosen during the harvest periods, between August and October.
A total of 104 wastewater samples were taken at the inlet and outlet of the CW beds to analyse the following physico-chemical parameters according to standard methods [21]: pH, Electrical Conductivity (EC), TSS at 105 • C, BOD 5 , COD, Total Nitrogen (TN), Ammonia Nitrogen (NH 4 -N), Nitrite Nitrogen (NO 2 -N), Nitrate Nitrogen (NO 3 -N), Organic Nitrogen (N org ), and Orthophosphates (PO 4 -P). Microbiological analyses were conducted according to Standard Methods for the Examination of Water and Wastewater [22] and following the method described by Russo et al. [23]. Faecal indicator bacteria (Escherichia coli, total coliforms, and enterococci) were enumerated according to the ISO procedures [24], and results were expressed as log 10 colony-forming units (CFU) per unit of volume.
For the physico-chemical parameters, the removal efficiency (R, %) was calculated as follows: Water 2020, 12, 1260 where C in and C out are, respectively, the concentrations at inlet and outlet (mg·L −1 ). Two important parameters for the empirical design of CW systems, such as Surface Loading Rate (SLR) and Surface Removal Rate (SRR), were determined. These parameters define the amount of pollutant that is being treated or removed per unit surface. The relationship between SLR and SRR is important for determining the wetland surface area needed to meet the discharge target. SLR and SRR were calculated as follows [5]: where SLR and SRR were in g·m −2 ·day −1 , Q is the influent flow rate (m 3 ·day −1 ) to the CW system, C in is the inlet concentration (g·m −3 ), A is the CW's area (m 2 ), and C out is the outlet concentration (g·m −3 ). Microbiological data were analysed by ANOVA (One-way Analysis of Variance) using Tukey's post-hoc test in order to assess the overall differences among samples. Statistical analysis was performed using XLSTAT PRO 5.7 (Addinsoft, New York, NY, USA), and the reference level of significance was 0.05 in all the assays.
The quality of the effluent of the CW system was also assessed in terms of achieving the standards defined by Italian law [17] and EU proposal regulation on the minimum requirement for water reuse [25].

Plant Measurements
For every December in the period 2014-2017, plant density and aboveground biomass of Phragmites australis, Cyperus papyrus, and Canna indica were measured on three 1 m 2 sampling areas selected at the opposite ends and in the middle of each area occupied by the plant species. Accordingly, three sampling areas were identified in VF (Phragmites australis) and six areas in HF (Cyperus papyrus and Canna indica). The bio-agronomic parameters were measured in the whole sampling areas. Dry biomass was evaluated by drying the fresh biomass samples in a thermo-ventilated oven at 65 • C until a constant weight was reached.
The bio-agronomic results were analysed using ANOVA and Student-Newman-Keuls tests after verifying the homogeneity of the variances using the Bartlett test.
The possible symptoms of plant toxicity in the vegetated areas of the CW beds were identified by a careful visual inspection carried out during the study period.

Weather Data at the Experimental Site
The meteorological data for the study site were recorded by a nearby meteorological station of the Sicilian Agrometeorological Information Service (SIAS), located in Noto (latitude 36 • 50' N, longitude 15 • 03' E, altitude 30 m). From the climatic observations, it is possible to observe that the study area is characterized by a typical Mediterranean climate (Csa under the Koppen-Geiger classification) with hot summers and rainfall not well distributed throughout the year; the spring/summer period is the driest (occasional rains), while the autumn/winter period is the rainiest. Figure 5 shows daily temperature extremes and rainfall from January 2014 to June 2018. Annual rainfall ranged between 468 mm (in 2014) and 654 mm (in 2015), with 1220 days of absence of rainfall during the 1643 total days of experimental study; average rainfall during the grape harvest season (August and September) was 84.3 mm. Daily air temperature values ranged between −1.9 • C (in January 2017) and 42.3 • C (in July 2017). The annual average temperatures were comparable among the different years of trial. rainfall ranged between 468 mm (in 2014) and 654 mm (in 2015), with 1220 days of absence of rainfall during the 1643 total days of experimental study; average rainfall during the grape harvest season (August and September) was 84.3 mm. Daily air temperature values ranged between −1.9 °C (in January 2017) and 42.3 °C (in July 2017). The annual average temperatures were comparable among the different years of trial.  Table 1 shows the concentrations of the physico-chemical parameters monitored at the CW beds inlet and outlet and the relative removal efficiency percentages. The pH of the influent varied between 4.67 and 11.06, with the lowest values observed during the harvest periods and the maximum during the winery washing operations with caustic detergents (sodium hydroxide). After the transition to the multistages (CWs) of the treatment, a reduction in the pH variation interval was observed, with standard deviations (SD) of 1.34 (first stage inlet) and 0.33 (last stage outlet). Similar results have been obtained by other authors [6,[26][27][28], who report pH values of the CW effluent nearby neutrality in the presence of acid wastewater and, in particular, during the harvest. The electrical conductivity of the CW influent varied from 561 to 2930 μS·cm −1 , with average value of 1086 μS·cm −1 , slightly reduced, of about 15%, after the passage through the multistage CW.    Table 1 shows the concentrations of the physico-chemical parameters monitored at the CW beds inlet and outlet and the relative removal efficiency percentages. The pH of the influent varied between 4.67 and 11.06, with the lowest values observed during the harvest periods and the maximum during the winery washing operations with caustic detergents (sodium hydroxide). After the transition to the multistages (CWs) of the treatment, a reduction in the pH variation interval was observed, with standard deviations (SD) of 1.34 (first stage inlet) and 0.33 (last stage outlet). Similar results have been obtained by other authors [6,[26][27][28], who report pH values of the CW effluent nearby neutrality in the presence of acid wastewater and, in particular, during the harvest. The electrical conductivity of the CW influent varied from 561 to 2930 µS·cm −1 , with average value of 1086 µS·cm −1 , slightly reduced, of about 15%, after the passage through the multistage CW. As expected for the winery effluent, the fluctuations in the concentrations of TSS (average value of 86 ± 130 mg·L −1 ), BOD 5 (average value of 316 ± 418 mg·L −1 ) and COD (average value of 587 ± 730 mg·L −1 ) were observed, with peaks during the various washing operations of fermentation tanks, barrels, and the equipment used for pressing the grapes.

Wastewater Physico-Chemical Parameters
However, the ranges of variations were lower (8-630 for TSS mg·L −1 , 4-1243 mg·L −1 for BOD 5 , and 6-2020 mg·L −1 for COD) compared to those reported in the literature [2,5,6]; these values are the direct consequence of the high volume of washing water used at the Marabino winery (4-5 L of washing water vs. 1 L of wine).
The results revealed a clear correlation, for each CW unit, between TSS surface loading rate (SLR) and TSS surface removal rate (SRR) (Figure 6), with correlation factors (R 2 ) ranging from 0.8499 (FSF) to 0.9815 (VF). This is in accordance with other research carried out in other wineries [5,29]. 42% ± 25% for VF, 34% ± 20% for HF, and 24% ± 28% for FSF, with the maximum values corresponding to the peaks of TSS in the influent (VF 78%, HF 75%, and FSF 94%).
The results revealed a clear correlation, for each CW unit, between TSS surface loading rate (SLR) and TSS surface removal rate (SRR) (Figure 6), with correlation factors (R 2 ) ranging from 0.8499 (FSF) to 0.9815 (VF). This is in accordance with other research carried out in other wineries [5,29]. COD values showed a high correlation between SLR and SRR with decreasing R 2 values between the first and the last CW unit (0.965 for VF, 0.949 for HF, and 0.941 for FSF). The organic loading rate in the VF ranged between 0.08 to 60.38 g COD m −2 ·day −1 , with an average of 15.74 g COD m −2 ·day −1 , comparable with results shown by Masi et al. [30] in small Italian wineries, but quite lower than the 164 g COD m −2 ·day −1 reported by Masi et al. [2] in their overview of the experiences at worldwide level during the last 15 years.
Despite these significant differences, the average CW removal efficiencies of BOD5 (78% ± 27%) and COD (81% ± 25%) resulted as similar than those reported by Masi et al. [2]. The lower removal of the organic matter shown by FSF (29% for BOD5 and 40% for COD) is probably due to the low concentrations at inlet (average BOD5 and COD of about 17 and 28 mg·L −1 , respectively), associated with the algal bloom phenomenon, only partially removed by filtration at the CW outlet.
The total nitrogen SLR ranged from 0.05 to 3.98 g·m −2 ·day −1 , and for VF and HF beds, it seems related to SRR, which has an average value of about 0.19 and 0.28 g·m −2 ·day −1 , respectively.
Serrano et al. [5] report similar relationships between SLR and SRR for Total Kjeldhal Nitrogen (TKN) and ammonium in a multistage CW (VF + HF) treating a mixed effluent derived from a winery and a tourist establishment in Spain (R 2 = 0.9691 for TKN and R 2 = 0.9007 for NH3-N). The organic nitrogen was the major fraction (mean value 8.4 ± 8.1 mg·L −1 ) of TN load (mean value 13.3 ± 11.9 mg·L −1 ) in the CW system, followed by ammonia (mean value 3.5 ± 3.7 mg·L −1 ), with a removal efficiency of 57%.
There were no significant differences between the average removal rate of NH4-N in VF (31%) and HF (28%), probably due to the low rate of nitrification in VF, which can result from an insufficient supply of NH4-N in the influent (mean concentration of 3.5 ± 3.7 mg·L −1 ).
As evidenced by the TN and PO4-P, the relationship between SRR and SLR was lower (maximum R 2 = 0.8133 in VF) than that shown for TSS and COD, but higher than that reported by Serrano et al. [5] (R 2 = 0.4017), with a similar concentration of PO4-P in the CW influent (about 2.3 ± 2.1 mg·L −1 ).
The average nutrient removal efficiency was slightly lower in FSF (11% for PO4-P and 15% for TN) compared to the other two CW stages. This is probably due to the low values of the average COD values showed a high correlation between SLR and SRR with decreasing R 2 values between the first and the last CW unit (0.965 for VF, 0.949 for HF, and 0.941 for FSF). The organic loading rate in the VF ranged between 0.08 to 60.38 g COD m −2 ·day −1 , with an average of 15.74 g COD m −2 ·day −1 , comparable with results shown by Masi et al. [30] in small Italian wineries, but quite lower than the 164 g COD m −2 ·day −1 reported by Masi et al. [2] in their overview of the experiences at worldwide level during the last 15 years.
Despite these significant differences, the average CW removal efficiencies of BOD 5 (78% ± 27%) and COD (81% ± 25%) resulted as similar than those reported by Masi et al. [2]. The lower removal of the organic matter shown by FSF (29% for BOD 5 and 40% for COD) is probably due to the low concentrations at inlet (average BOD 5 and COD of about 17 and 28 mg·L −1 , respectively), associated with the algal bloom phenomenon, only partially removed by filtration at the CW outlet.
The total nitrogen SLR ranged from 0.05 to 3.98 g·m −2 ·day −1 , and for VF and HF beds, it seems related to SRR, which has an average value of about 0.19 and 0.28 g·m −2 ·day −1 , respectively.
Serrano et al. [5] report similar relationships between SLR and SRR for Total Kjeldhal Nitrogen (TKN) and ammonium in a multistage CW (VF + HF) treating a mixed effluent derived from a winery and a tourist establishment in Spain (R 2 = 0.9691 for TKN and R 2 = 0.9007 for NH 3 -N). The organic nitrogen was the major fraction (mean value 8.4 ± 8.1 mg·L −1 ) of TN load (mean value 13.3 ± 11.9 mg·L −1 ) in the CW system, followed by ammonia (mean value 3.5 ± 3.7 mg·L −1 ), with a removal efficiency of 57%.
There were no significant differences between the average removal rate of NH 4 -N in VF (31%) and HF (28%), probably due to the low rate of nitrification in VF, which can result from an insufficient supply of NH 4 -N in the influent (mean concentration of 3.5 ± 3.7 mg·L −1 ).
As evidenced by the TN and PO 4 -P, the relationship between SRR and SLR was lower (maximum R 2 = 0.8133 in VF) than that shown for TSS and COD, but higher than that reported by Serrano et al. [5] (R 2 = 0.4017), with a similar concentration of PO 4 -P in the CW influent (about 2.3 ± 2.1 mg·L −1 ).
The average nutrient removal efficiency was slightly lower in FSF (11% for PO 4 -P and 15% for TN) compared to the other two CW stages. This is probably due to the low values of the average concentrations in the influent (3.0 mg·L −1 for PO 4 -P and 5.3 mg·L −1 for TN), which were very close to the background concentration [31].

Microbiological Parameters
Microbiological counts of E.coli, total coliform, and Enterococcus spp., detected at the inlet and at the outlet of each CW unit, during both harvest (3 months) and non-harvest periods (9 months), are reported in Figure 7.
the optimal growth pH range of faecal bacteria, which is between 5.5 to 7.5 [33].
The presence of E.coli contamination in winery wastewater is related to the presence of domestic constituents (i.e., toilet use by workers). Generally, faecal bacteria are present in very small quantities in the winery wastewater [34]. However, also Rozema et al. [35] report similar E.coli concentrations in the influent of CW treating winery process water and domestic sewage at a winery in Niagara region of Ontario (Canada).  Table 2 shows the percentages of wastewater samples that have been below the limits of Italian legislation (M.D. 185/2003) and EU proposal regulation on the minimum requirement for water reuse.

Winery Wastewater Reuse Option
In particular, COD concentrations have always been below the limit imposed by Italian legislation for wastewater reuse. The FSF system has proven to be indispensable to meet this legal requirement; in fact, at the HF outlet, the percentage of samples that met the legal limit for COD was 86% and increased to 100% after the FSF treatment.
Only 34% and 18% of the wastewater samples at the CW outlet exceeded the concentrations of TSS (10 mg·L −1 ) and BOD5 (20 mg·L −1 ), respectively, set by M.D. 185/2003. This could be explained by the algae growth that occurs in the FSF stage, which increased the TSS and, consequently, the BOD5 concentrations in the effluent.
The TN concentrations were always below the Italian reuse limit (35 mg·L −1 ) already at the VF stage, demonstrating the low concentrations of nutrients that characterize wastewater from wineries.
Even though the multistage CW treatment was able to set a reduction of the microbial population, the E. coli cell density in the 60% of samples collected at the FSF effluent was higher than Overall, total coliform and Enterococcus spp. did not show statistically significant differences in the VF influent during the monitoring periods. Different results were obtained for E.coli, which exhibited a higher count during the harvest period. Evaluating the reduction of the considered microbiological indicators, the highest removal efficiency was observed during the harvest period. In fact, the CW treatment determined a mean reduction of 2.3 log units for all the microbial groups under study. During the remaining periods, the total coliform count exhibited a statistically significant reduction of 1.26 log units. The highest removal efficiencies detected during the harvest period were probably related to (i) the higher loading rate, which, as reported in numerous published studies [32], appears to be positively related to the efficiency; (ii) wastewater's pH values lower than the optimal growth pH range of faecal bacteria, which is between 5.5 to 7.5 [33].
The presence of E.coli contamination in winery wastewater is related to the presence of domestic constituents (i.e., toilet use by workers). Generally, faecal bacteria are present in very small quantities in the winery wastewater [34]. However, also Rozema et al. [35] report similar E.coli concentrations in the influent of CW treating winery process water and domestic sewage at a winery in Niagara region of Ontario (Canada). Table 2 shows the percentages of wastewater samples that have been below the limits of Italian legislation (M.D. 185/2003) and EU proposal regulation on the minimum requirement for water reuse.

Winery Wastewater Reuse Option
In particular, COD concentrations have always been below the limit imposed by Italian legislation for wastewater reuse. The FSF system has proven to be indispensable to meet this legal requirement; in fact, at the HF outlet, the percentage of samples that met the legal limit for COD was 86% and increased to 100% after the FSF treatment.
Only 34% and 18% of the wastewater samples at the CW outlet exceeded the concentrations of TSS (10 mg·L −1 ) and BOD 5 (20 mg·L −1 ), respectively, set by M.D. 185/2003. This could be explained by the algae growth that occurs in the FSF stage, which increased the TSS and, consequently, the BOD 5 concentrations in the effluent.
The TN concentrations were always below the Italian reuse limit (35 mg·L −1 ) already at the VF stage, demonstrating the low concentrations of nutrients that characterize wastewater from wineries.
Even though the multistage CW treatment was able to set a reduction of the microbial population, the E. coli cell density in the 60% of samples collected at the FSF effluent was higher than the threshold fixed by the M.D. 185/2003 (80% equal to 10 CFU·100 mL −1 with a maximum admitted value equal to 100 CFU·100 mL −1 ). a Classes of reclaimed water quality and allowed agricultural use and irrigation method. Class A: All food crops, including root crops consumed raw and food crops where the edible part is in direct contact with reclaimed water. All irrigation methods allowed. Class B: Food crops consumed raw where the edible part is produced above ground and is not in direct contact with reclaimed water; processed food crops and non-food crops including crops to feed milk-or meat-producing animals. All irrigation methods allowed. Class C: The same crop category irrigable with water quality of Class B. Only drip irrigation or other methods that avoid direct contact with the edible part of the crop are admitted. Class D: Industrial, energy, and seeded crops. All irrigation methods allowed. b 80% of samples. c Max value.
Considering the EU minimum quality requirements for water reuse, treated wastewater herein analysed can be included in class C (characterized by TSS and E.coli limits equal to 35 mg·L −1 and 1000 CFU·100 mL −1 , respectively), except for 14% of samples that exceeded the concentration of 25 mg·L −1 fixed for the BOD 5 . However, the observed values of this latter parameter (mean of 17 mg·L −1 with a maximum of 65 mg·L −1 ) may not be considered a limiting factor for CW effluent reuse for irrigation, which can represent a source of organic matter with high agronomic potential [36][37][38].

Characteristics of Vegetated Areas in the Multistage CW
Average wetland plant density was affected by the age of plantation with a significant increase from 2014 to 2016 (Figure 8). This plant feature remained pretty steady in the two last years of trial, with the exception of Phragmites australis, which significantly decreased from 284 ± 28 plant·m −2 to 178 ± 41 plant·m −2 . This reduction in the plant density (about 38%) caused a decrease of Phragmites australis biomass dry yield (66% in 2017) ( Figure 8). However, despite this reduction, Phragmites australis showed the highest values of dry biomass, with a total mean value of about 2600 g·m −2 , followed by Cyperus papyrus and Canna indica.
Water 2020, 11, x FOR PEER REVIEW 11 of 14

Characteristics of Vegetated Areas in the Multistage CW
Average wetland plant density was affected by the age of plantation with a significant increase from 2014 to 2016 (Figure 8). This plant feature remained pretty steady in the two last years of trial, with the exception of Phragmites australis, which significantly decreased from 284 ± 28 plant·m −2 to 178 ± 41 plant·m −2 . This reduction in the plant density (about 38%) caused a decrease of Phragmites australis biomass dry yield (66% in 2017) ( Figure 8). However, despite this reduction, Phragmites australis showed the highest values of dry biomass, with a total mean value of about 2600 g·m −2 , followed by Cyperus papyrus and Canna indica. The dry yield of Phragmites australis was lower than that measured in other studies conducted in similar climatic conditions, but in line with CWs treating municipal wastewater, characterized by lower COD and higher nutrient concentrations [39,40]. Arienzo et al. [8] reported that winery wastewater did not produce phytotoxic effects on Phragmites australis planted in microcosm wetland at pollutant concentrations comparable to those of our wastewater. Therefore, differences in biomass yield performance shown in our study may be linked to the highest organic matter which is able to modify nutrient availability to plants via complexation, chelation, and ion exchange [41] and is associated with the low concentrations of nutrients easily assimilable (average NH4 value in VF influent of 3.5±3.7 mg·L −1 ). This has been proven by the fertilization carried out in March 2015 that was able to significantly increase the plants' density and, consequently, the biomass production, which reached the maximum value during the following year.
The Cyperus papyrus showed less sensitivity to nutrient deficiency, highlighting a constant biomass yield after the first growing season and reaching an average plant height of about 2.5 m. The high growth of Cyperus papyrus caused a partial shading of the area occupied by the Canna indica which has significantly reduced its biomass.
The green areas close to the multistage CW showed regular growth and vegetative development with the absence of symptoms of chlorosis or necrosis related to the phytotoxic effects of irrigation with wastewater from the winery. This result was also favoured by the sandy loam characteristics of the CW substrate and by the high irrigation volumes (average value of about 900 mm/year) which, despite the long period of absence of rainfall, prevented the accumulation of soluble salts in the root zone by leaching. The dry yield of Phragmites australis was lower than that measured in other studies conducted in similar climatic conditions, but in line with CWs treating municipal wastewater, characterized by lower COD and higher nutrient concentrations [39,40]. Arienzo et al. [8] reported that winery wastewater did not produce phytotoxic effects on Phragmites australis planted in microcosm wetland at pollutant concentrations comparable to those of our wastewater. Therefore, differences in biomass yield performance shown in our study may be linked to the highest organic matter which is able to modify nutrient availability to plants via complexation, chelation, and ion exchange [41] and is associated with the low concentrations of nutrients easily assimilable (average NH 4 value in VF influent of 3.5 ± 3.7 mg·L −1 ). This has been proven by the fertilization carried out in March 2015 that was able to significantly increase the plants' density and, consequently, the biomass production, which reached the maximum value during the following year.

Conclusions
The Cyperus papyrus showed less sensitivity to nutrient deficiency, highlighting a constant biomass yield after the first growing season and reaching an average plant height of about 2.5 m. The high growth of Cyperus papyrus caused a partial shading of the area occupied by the Canna indica which has significantly reduced its biomass.
The green areas close to the multistage CW showed regular growth and vegetative development with the absence of symptoms of chlorosis or necrosis related to the phytotoxic effects of irrigation with wastewater from the winery. This result was also favoured by the sandy loam characteristics of the CW substrate and by the high irrigation volumes (average value of about 900 mm/year) which, despite the long period of absence of rainfall, prevented the accumulation of soluble salts in the root zone by leaching.

Conclusions
The multistage CW system, used for the treatment of wastewater from a small winery located in southern Italy, has a strong buffer capacity able to mitigate the large fluctuations of concentrations in the influent, with an average efficiency of removal of about 69% for TSS, 78% for COD, and 81% for BOD 5 . The average nutrient removal efficiency was low. This was probably due to the low influent nutrient values, already close to the background concentration. Regarding the option of wastewater reuse for irrigation, the study showed that the CW effluent meets the limit of the Italian legislation for COD, while, due to the algae growth process, TSS and BOD 5 were found higher than the legal threshold for 34% and 18% of the analysed samples, respectively. A complete compliance with the legal limits could be obtained by increasing the filtering surface in the terminal section of FSF. A gravel material with a diameter of 10-25 mm, less than the 80-100 mm used in the CW pilot plant, could be adopted.
Despite the high levels of E.coli (up to 4.8 log units) of the CW influent, following the sewage component of the wastewater, the bacteriological quality of the effluent (outlet of FSF unit) meets the C category standard of the EU regulation proposal for water reuse.
The low concentrations of nutrients in the CW influent determined the need to use fertilizers to promote the growth of Phragmites australis in the first CW stage. The fertilizer can be applied in raw wastewater or on the surface of the CW beds at the beginning of each growing season to prevent macrophyte growth problems.
The plant species present in the green area close to the multistage CW showed regular vegetative growth and no phytotoxicity phenomena related to the use of treated wastewater. However, more reliable results on the impact of treated wastewater on the green area will only be achieved after a further irrigation season [42,43].
To conclude, the results of the study confirm the high potential and efficiency of the multistage CW system for the treatment of wineries wastewater and subsequent reuse for irrigation. CWs are particularly suitable for medium-small wineries, where the construction and O&M costs of conventional wastewater treatment options may not be economically sustainable for the owner.