Nitrogen removal from summer to winter in a eld pilot-scale multistage constructed wetland-pond system

Single-stage constructed wetlands (CWs) has a single ecological service function and is greatly affected by temperature, which are general in removal of total nitrogen. Multistage hybrid CWs were proven to capable of enhancing removal of nitrogen. Therefore, this study aimed to explore the variation in nitrogen removal in the combined CWs-pond process from summer to winter and the contribution of plant harvesting and the functions of bacteria to nitrogen removal. A pilot-scale multistage constructed wetland-pond system (MCWP) with the process of "the pre-ecological oxidation pond + the two-level horizontal subsurface ow constructed wetland (HSCW) + the surface ow constructed wetland (SFCW) and the submerged plant pond (SPP)" was used to treat actual polluted river water in the eld. During the 124 days of operation, the nitrogen concentrations in the units inuent and euent of the system were measured every two days, and the plant height in HSCWs and SFCW was measured once per month. When the system operated stably to the 72nd day, the substrates in the CWs were sampled to analyze the bacterial community structure and composition.


Introduction
In recent years, with the development of the national social economy, the contradiction between the construction of material civilization and the development of the water environment has become increasingly prominent. The deterioration of the water environment and ecological environment has become more severe. Surface waters have become increasingly polluted by nitrate nitrogen (NO 3 − -N) and ammonia nitrogen (NH 4 + -N) to varying degrees (Wilbers et al., 2014). According to monitoring data from recent decades, nitrogen concentrations in rivers throughout the world are increasing annually (Domangue and Mortazavi, 2018). Nitrogen pollution in aquatic environments is a major international concern due to its toxicity, tendency to cause eutrophication, and associated loss of oxygen and subsequent sh kills, among other effects (Haas et al., 2017;Paredes et al., 2019). Furthermore, nitrogen compounds in drinking water may transform into nitrites, which will lead to an increase in the risk of various diseases (including cancers) to humans (Sajedi-Hosseini et al., 2018). Therefore, the problem of nitrogen pollution in water, whether from the perspective of ecological balance or the impact on human health, merits more attention Smith et al., 2015).
As a near-natural ecological treatment technology, constructed wetlands (CWs) have been widely used in wastewater treatment due to their advantages of simplicity, ecological economy, high e ciency, low consumption and strong hydraulic load resistance. The NH 4 + -N and total nitrogen (TN) removal e ciencies in river water contaminated by CWs reached 90% and 50%, respectively (Zheng et al., 2014;Huang et al., 2011). The near-natural ecological treatment technology represented by CWs is an effective way to solve the problem of nitrogen removal from polluted river water.
However, the CWs approach has some defects, including limited adsorption saturation, clogging, and a single ecological service function.
These aspects limit the promotion and application of CWs to some extent. The NH 4 + -N removal e ciency prevails over TN, which may not meet the demand of puri ed water quality (Vymazal, 2013). At the same time, temperature is an important factor limiting the puri cation of water quality by CWs (Lu et al., 2006). There are a variety of nitrogen removal mechanisms in CWs, and microbial denitri cation is widely considered the most important mechanism. Plant uptake, substrate adsorption, and ammonia volatilization are generally considered second-level contributions to nitrogen removal (Green, 1996;Vymazal, 2007). In view of the physiological and biochemical characteristics of denitrifying bacteria, 34-37 °C is a suitable temperature range to ensure that the denitri cation reaction is carried out stably, which can ensure the growth activity of denitrifying bacteria. When the temperature was 15 °C, the denitri cation reaction rate decreased signi cantly, and the reaction almost stopped at 5 °C (Lu et al., 2006). The biomass of heterotrophic bacteria showed obvious seasonal changes (Gao, 1998). The nitrogen removal e ciency in CWs in summer was signi cantly higher than that in winter (Reddy, 1985;Nie et al., 2006). Therefore, it is particularly necessary to ensure the removal effect in CWs and improve the overall temperature resistance in the process.
Multistage CWs can effectively decrease the disadvantages of single-stage CWs, such as the low removal e ciency of nitrogen and unstable e ciency caused by season (temperature). When the hydraulic load rate was 0.053 m 3 /(m 2 ·d), the annual average removal e ciencies of NH 4 + -N and TN for river wastewater by surface-vertical ow constructed wetlands were 54.0 ± 6.3% and 53.9 ± 6.0%, respectively, and these values were signi cantly better than those obtained for single-stage CWs (Zheng et al., 2014). There may be an inherent quantitative relationship between temperature and removal e ciency. At the same time, the removal effect in each unit in relation to puri cation of the actual water body is worth studying. Moreover, there is limited information regarding the testing of bacterial community structure changes in different units of multistage CWs.
Therefore, a eld pilot-scale multistage constructed wetland-pond system (MCWP) was constructed to investigate the variation in nitrogen removal performance from summer to winter. The contribution of plant harvesting to nitrogen removal was also discussed. The following assumptions were made: (1) there was a quantitative relationship between nitrogen removal e ciency and temperature in the MCWP; (2) plants played an important role in the process of nitrogen removal. The purposes of this paper were to explore the relationship between nitrogen removal e ciency and temperature, quantify the correlation by formula tting and make it possible to predict the TN removal e ciency according to the daily highest temperature from summer to winter in a eld CW.

Materials And Methods
Construction of the MCWP An MCWP system including a raw water tank (RWT), ecological pond (EP), oxidation pond (OP), rst-level horizontal subsurface ow constructed wetland (FHSCW), second-level horizontal subsurface ow constructed wetland (SHSCW), surface ow constructed wetland (SFCW) and submerged plant pond (SPP) was designed. The process is shown in Fig. 1. Each unit was welded by 1-cm-thick UPVC boards, reinforced by stainless steel, and connected by 20-mm PVC pipes. The size of each unit, effective water depth, hydraulic retention time (HRT) and water inlet and outlet modes are shown in Table 1. The substrate in the horizontal subsurface ow constructed wetland (HSCW) was designed in two layers, and the thickness of the lower layer was 1000 mm. The lower layer was designed in three sections. The middle part was 840 mm thick. A mixture of gravel and steel slag of φ10-20 mm with a mass ratio of 4:1 was used as the ller layer for microbial lm hanging; both the inlet and the outlet sides were 80 mm thick, and φ20-40 mm gravel was laid to intercept residual particulate pollutants in the water body and reduce the clogging of the main part. The upper layer was 200 mm thick, and φ5-8 mm gravel was laid for the rooting and xation of emergent plants. The substrate in the SFCW was designed in two layers. The lower layer was 100 mm thick, and φ10-20 mm gravel was used as the supporting layer; the upper layer was a 300-mm-thick soil layer, which was used for the rooting and xation of emergent plants. A schematic diagram of the HSCW and SFCW is shown in Fig. 2.
The plants used in the experiment were transported from the local river. The plants were carefully dug out, washed clean and stored in tap water for 2 weeks. Nymphaea L. (1 plant) was planted in the EP, and Myriophyllum verticillatum L. was planted in the OP and SPP at densities of 26 plants/m 2 and 33 plants/m 2 , respectively. In the HSCW and SFCW, Iris pseudacorus L. was planted at densities of 5 plants/m 2 and 35 plants/m 2 , respectively. The experiment was operated from Jul. 29 to Nov. 30, 2019, totaling 124 days. The experimental water was taken from the Xiaoyi River, which is located in Baoding city, Hebei Province (coordinates: 115° 51' 36" E, 38° 46' 12" N). The experimental site was located in Rongcheng county, Baoding city, Hebei Province (coordinates: 115° 52' 12" E, 39° 3' N). Speci cally, 300 L of river water was taken from the river into the RWT at 9 a.m. every day, and the ow rate was 160 mL/min and obtained using a peristaltic pump. The river water owed from the EP to the SFCW based on the height difference. The water owed from the SFCW into the storage water tank, and then it was pumped into the SPP. Then, the water was discharged to the municipal pipe. To increase the oxygen content in the OP and SFCW, the mode of drop water was adopted. Samples were taken once every two days, and pH was measured immediately. All samples were analyzed within 24 hours. The in uent quality was as follows: permanganate index (COD Mn ) was 6.82±4.70 mg/L; total phosphorous (TP) was 0.1±0.041 mg/L; TN was 3.46±1.03 mg/L; NH 4 + -N was 0.72± 0.51 mg/L; suspended solids (SS) concentration was 17.62±12.44 mg/L; and pH was 8.31±0.21.
Methods of determination and analysis TN was determined by the alkaline potassium persulfate digestion UV spectrophotometric method (HJ636-2012); NH 4 + -N was determined by Nessler's reagent spectrophotometry (HJ535-2009). The calculation of nitrogen content in Iris pseudacorus L. was based on previous research (Luo et al., 2009).
In the HSCW and SFCW, three representative plants of high, medium and low were selected, and the plant height was measured once per month. At the end of the experiment, the plant growth was recorded, and the total amount of plant absorption during the experiment was as follows: where M is the uptake of nitrogen per unit area of wetland, g/m 2 ; M1 and M2 are the parts of plant height growth and new shoots at the end of the experiment, respectively, g/m 2 ; and P is the nitrogen content in plants, mg/g. The calculation formula of wetland removal rate per unit area is as follows: where F is the amount of nitrogen removal per unit area in the wetland, g/m 2 ; C 1 and C 2 are the nitrogen concentrations in the units in uent and e uent, respectively, mg/L; Q is the in uent ow rate of wetlands, L/d; t is the operation time of wetlands, d; and S is the surface area of wetlands, m 2 .
When the system operated stably to the 72nd day (Oct. 9), the substrates in the CWs were sampled to analyze the bacterial community structure and composition. The sampling points of steel slag and gravel mixed substrates were selected at the position of 500 mm from the bottom of the vertical line of the substrate layer in the FHSCW and SHSCW, and the soil sample was taken at the position 200 mm from the bottom of the vertical line of the substrate layer in the SFCW. The samples were stored in an ultralow temperature refrigerator at -80 ℃. Speci c methods were applied according to previous studies (Li, 2015;Tong et al., 2019). The V3-V4 region of the bacterial 16S rRNA gene was ampli ed according to Hai et al. (2014). The removal of low-quality sequences and the screening of high-quality sequences with a 97 % identity threshold followed the methods of previous studies (Liu et al., 2019;Liu et al., 2020;. Temperature data came from the web of weather (www.Tianqi.com). All statistical analyses were performed using Origin 2019 and SPSS version 26.0 software and were considered signi cant at the p< 0.05 level.

TN removal
TN removal effect in the EP and OP Because the experimental wastewater was collected from actual river water, the in uent TN concentration uctuated greatly (Zhang, 2016). EP and OP were pretreatment devices for this constructed wetland system and were mainly designed for removing SS from the river water . In fact, these devices did play important roles in removing SS (data not presented in this study). During the experiment, the in uent TN concentration was 1.78-7.18 mg/L, the average concentration was 3.46 mg/L, and the water quality was inferior V water (GB3838-2002). The e uent concentration and the removal e ciency in EP were 3.18 ± 1.10 mg/L and 6.93 ± 24.46% (P > 0.05), respectively.
EP was the rst treatment unit of the system, and the nitrogen in the in uent tended to be removed by denitri cation in the aerobic/anoxic environment formed by the different dissolved oxygen (DO) concentration distributions in the pond, which may be one of the reasons why EP had a better TN removal effect in the early stage of operation (the average removal e ciency was 26.36%), as shown in Fig. 3. In addition, the uptake of plants in EP played an important role in TN removal. In the later period of operation, as the air temperature decreased, the activity of microorganisms and the growth of plants were inhibited, and the TN removal rate in EP tended to generally decline. At the same time, the TN concentration in the e uent was higher than that in the in uent. This result may also be related to the secondary release of nitrogen from plant residues due to the disturbance of ow.
The TN concentrations in the in uent and e uent of OP were 3.18 ± 1.1 mg/L and 2.99 ± 0.98 mg/L, respectively. The TN removal e ciency was 6.93 ± 24.46% (P > 0.05). As shown in Fig. 4, in the early period, the puri cation effect in OP was poor, and the e uent TN concentration increased, which may have been related to the release of nitrogen into water after the death and decay of Myriophyllum verticillatum L. in OP. Within one month after Aug. 10, the TN removal rate was maintained at a relatively high level, up to 56%, which was mainly dependent on the microbial action in OP. With OP running to the later stage, the TN removal rate showed a downward trend due to the in uence of temperature change.

TN removal effect in the FHSCW and SHSCW
The TN removal e ciency in the FHSCW uctuated little. The TN concentrations in the in uent and e uent of the FHSCW were 2.99 ± 0.98 mg/L and 2.80 ± 1.00 mg/L, respectively. As shown in Fig. 5, the overall removal effect was good. The removal e ciency was 4.60 ± 22.1% (P < 0.05). Regardless of the negative value, the removal e ciency was 18.20 ± 12.50%. This result is because the interior of the FHSCW, especially the non-rhizosphere zone, was often in an anoxic state; thus, the removal effect of TN caused by denitri cation in the wetland was obvious. In the later period, the removal effect was relatively less affected by temperature changes, indicating that the FHSCW had a certain thermal insulation effect.
The TN concentrations in the in uent and e uent of SHSCW were 2.80 ± 1.00 mg/L and 2.60 ± 0.91 mg/L, respectively. There was a signi cant difference in the in uent and e uent concentrations of the SHSCW (P < 0.05). The removal e ciency was 2.40 ± 33.45%. Regardless of the negative value, the removal e ciency was 20.27 ± 15.10%. The pH value in the e uent of the SHSCW was higher than 9 in Aug., which may have inhibited the growth and reproduction of denitrifying bacteria and led to the low removal rate of wetlands in the early period ( Fig. 6). In general, the optimal pH range for denitri cation was 7-8, and the denitri cation rate was the highest when the pH was approximately 7.5. When the pH was more than 9, the denitri cation rate decreased signi cantly (Lu et al., 2006), which was consistent with the results of the experiment. Therefore, ammonia volatilization was an important way to remove nitrogen in Aug. From the beginning of Sep., the pH value of the e uent in the SHSCW dropped below 9. By the middle of Sep., the microorganisms in the SHSCW gradually adapted to the wetland environment and grew rapidly. In addition, after the e uent in the FHSCW entered the SHSCW, the DO concentration was further reduced. The anoxic/anaerobic environment in the wetland promoted the denitri cation effect. The highest TN removal e ciency was 63.49% at the end of Sep. With the SHSCW running until Oct., the TN removal rate showed a downward trend due to the in uence of air temperature change.
TN removal effect in the SFCW and SPP As shown in Fig. 7, the change rule of the removal e ciency in the SFCW was similar to that in the SHSCW. The TN concentrations in the in uent and e uent of the SFCW were 2.60 ± 0.91 mg/L and 2.42 ± 1.03 mg/L, respectively. The removal e ciency was 2.40 ± 33.45% (P < 0.05). Regardless of the negative value, the removal e ciency was 22.28 ± 14.79%. In general, the TN removal rate in the SFCW was good in Aug. and Sep. (the average removal rate was 27.78%), but as the operation period entered Oct., the TN removal e ciency decreased continuously (the average removal e ciency was 15.44%) due to the in uence of air temperature changes.
The TN concentrations in the in uent and e uent of the SPP were 2.42 ± 1.03 mg/L and 2.04 ± 1.17 mg/L, respectively. The removal e ciency was 16.08 ± 26.92% (P < 0.05). Regardless of the negative value, the removal e ciency was 28.99 ± 19.75%. As shown in Fig. 8, the TN removal in the SPP could be divided into three stages: the rst stage was from the beginning of the experiment (Jul. 31) to Aug. 19, which was the stage in which the removal effect increased, with the highest removal e ciency of 59.37%. Since the plants in the pond grew gradually and the microbial action increased gradually, the removal e ciency showed an upward trend. The second stage was from Aug. 19 to Oct. 5, which was the stable period in the SPP. The average removal rate of the pond was 34.49%, with a maximum of 76.69%, due to the high temperature. The microorganism activity was strong at this stage, and the denitri cation effect was obvious. On the other hand, plant growth in the pond made a certain contribution to the uptake of nutrients. In the third stage, from Oct. 5 to the end of the experiment (Nov. 30), the microbial activity and plant growth in the pond were inhibited by the change in temperature; thus, the overall removal rate of the pond showed a downward trend.

TN removal effect in the MCWP
To investigate the overall removal rate in the MCWP for raw river water, Fig. 9 was drawn. The TN concentrations of the in uent and e uent in the MCWP were 3.46 ± 1.03 mg/L and 2.04 ± 1.17 mg/L, respectively. The removal rate was 40.74 ± 28.84% (P < 0.05). The overall change in TN in the e uent could be divided into three stages. The initial stage was the increasing stage, which was mainly due to the release of nutrients from the plant residues in the pond and the increase in the pH value in the e uent of the HSCW, resulting in the low activity of microorganisms; thus, the total TN removal e ciency of the system was relatively low. However, as the microorganisms gradually adapted to the wetland environment, denitri cation was enhanced. The middle stage was a stable period (Aug. 5-Oct. 5). In this period, the temperature was relatively high, the microorganisms grew vigorously, the average removal e ciency was 60.93%, and the total e uent concentration met the class V water standard. In the later period, the removal rate showed a declining trend from summer to winter. In early and mid-Nov., the removal rate slightly increased twice, which may have been related to the slight increase in temperature.
The TN concentration in the e uent of each unit showed a similar uctuating trend as that in the in uent. There was a signi cant correlation between the TN concentrations in the in uent and e uent (P < 0.01). The TN concentration was gradually reduced in the system, and the nitrogen removal performance was mainly re ected in the back of the system. The later the unit was, the higher the TN removal e ciency was. This result was because more DO was consumed in the degradation of organic matter in the front of the system, and with the gradual deepening of the in uent, the environment in the system gradually changed from an aerobic environment to an anoxic/anaerobic environment, which was more conducive to denitri cation reactions. In addition to DO, denitri cation was affected by factors such as organic matter concentration and temperature. In general, the optimal C/N (carbon as COD Mn ) for denitri cation is 4-5 (Hanaki et al., 1992). The average accumulation rate of NO 3 − -N was 3.3 g/(m 3 ·d) in tide ow constructed wetlands (TFCWs) at 15 °C, and the C/N ratio was 4 (Zhi and Ji., 2014). In this experiment, the in ow C/N was 1.9, which indicated that the reason for the decreased TN removal e ciency in Oct. was not only the temperature but also the organic concentration.
According to Vymazal's classi cation of hybrid CWs (Vymazal, 2013), the FHSCW-SHSCW-SFCW in this experimental device belongs to the hybrid systems that include the SFCW. Various types of CWs provide different redox conditions that are suitable for enhancing the removal of TN (Vymazal, 2011). It was necessary to compare removal e ciency in terms of removed loadings. Vymazal's evaluation of the treatment performance of 60 hybrid CWs indicated that the most e cient system for the removal of TN was the hybrid CW, which included an SFCW stage in the treatment line and was able to remove an average of 4.24-5.12 g TN m − 2 d − 1 . Linear tting results revealed that the removed load of TN accounted for approximately 77% of the in uent load. In the FHSCW-SHSCW-SFCW, the in uent TN load and removed load of the hybrid CW were 2.65 g TN m − 2 d − 1 and 1.95 g TN m − 2 d − 1 , respectively. The removed load accounted for 74% of the in uent load, which was consistent with the results of Vymazal. The difference between single horizontal ow CWs and hybrid systems with the SFCW stage was statistically signi cant (p < 0.05). This result was because in the SFCW, radial oxygen loss of emergent plants and atmospheric reaeration provided aerobic conditions for nitri cation, while root exudates and plant litter provided additional carbon source conditions for denitri cation.
NH 4 + -N removal Figure 10 shows that the in uent NH 4 + -N was 0.27-3.18 mg/L during the experiment, and the average concentration was 0.72 mg/L. After Aug. 5, the e uent in the MCWP was stabilized in Class III. The NH 4 + -N removal effect in each unit of the system differed greatly. The NH 4 + -N concentrations and removal rates in the EP, OP, PHSFCW, SHSFCW, SFCW and SPP were 0.58 ± 0.42 mg/L and 24.11 ± 17.63%; 0.54 ± 0.39 mg/L and 7.88 ± 14.84%; 0.56 ± 0.52 mg/L and 9.44 ± 20.62%; 0.51 ± 0.44 mg/L and 9.65 ± 21.26%; 0.56 ± 0.51 mg/L and − 3.43 ± 14.39%; and 0.57 ± 0.50 mg/L and 0.18 ± 21.24%, respectively. In general, the removal e ciency of NH 4 + -N in the EP was the best, while that in the SFCW and SPP was the worst, even though the e uent TN concentration was higher than that of the in uent, which may have been related to the release of pollutants from the dead plant residues into the water. The total removal e ciency of NH 4 + -N in the system was 22.02 ± 25.97%, and the removal effect was mainly in the front of the system. The later the unit was, the lower the NH 4 + -N removal e ciency was. The EP was the main unit of NH 4 + -N removal because the content of DO in the front of the system was higher, which promoted the conversion of nitrogen in the form of NH 4 + -N into NO 3 − -N through nitri cation, thereby reducing the concentration of NH 4 + -N.
As more DO was consumed in the front of the system by Organic matter degradation, the system environment gradually changed from an aerobic environment to an anoxic/anaerobic environment, which was not conducive to nitri cation. The removal e ciency decreased and the concentration increased when the e uent in the OP entered the FHSCW. The reason may have been that the DO concentration of the HSCW was lower than that of the pond and the SFCW. The removal rate of NH 4 + -N in the SHSCW was improved, probably because of ammonia volatilization and Anammox. In general, the SFCW had a better reoxygenation ability than the HSCW, which could have provided a good living environment for nitrifying bacteria (Fu et al., 2012). Therefore, it had a better effect on NH 4 + -N removal, which was inconsistent with the results of this study. This may have been due to the poor hygienic conditions in the SFCW being prone to breeding mosquitoes and ies. In the form of water retention, the algae on the surface layer propagated in large quantities. The DO in the water decreased, resulting in an increase in the NH 4 + -N concentration in the e uent of the SFCW. Vymazal and Kropfelova (2008) calculated that the average removal load of ammonia nitrogen by 85 single vertical ow CWs was 2.07 ± 2.03 g NH 4 + -N m − 2 d − 1 . Free-drain vertical ow CWs are aerobic due to intermittent feeding, which allows for oxygen diffusion into the ltration bed (Vymazal, 2007). The characteristics of vertical ow CWs greatly enhance the removal e ciency of NH 4 + -N. Vymazal's results (Vymazal, 2013) indicated that there was no signi cant difference between single vertical ow CWs and hybrid CWs with the SFCW (2.34 ± 3.31 g NH 4 + -N m − 2 d − 1 ) (P > 0.05). However, in this study, the in uent N load and removal load of the FHSCW-SHSCW-SFCW hybrid CW were 0.48 g NH 4 + -N m − 2 d − 1 and 0.32 g NH 4 + -N m − 2 d − 1 , respectively. The removed load accounted for 66% of the in uent load, slightly lower than that of Vymazal (78%). The removal load of NH 4 + -N by the HSCW was signi cantly weaker than that of other wetland types. Therefore, the weak reoxygenation capacity of the SFCW in this study led to a lower removal load of the FHSCW-SHSCW-SFCW hybrid CW.
Plant height growth Figure 11 shows that the plant height in the HSCW and SFCW increased signi cantly throughout the experimental process. The plant height in the planted emerged plant unit showed that SFCW > FHSCW > SHSCW. Compared with the HSCW, the average plant height in the SFCW was higher. On the one hand, this result may have been due to the high contents of organic matter, nitrogen and phosphorus in the upper soil of the SFCW, which was more conducive to the growth of plants; in contrast, all nutrients required by plants in the HSCW came from the polluted water body and lacked a certain amount of elements necessary for plant growth. On the other hand, due to the in uence of steel slag in the HSCW, the pH in the e uent of the FHSCW and SHSCW was higher than 8.73 and 8.23 and 9.13 and 8.37 in Aug. and Sep., respectively, which may have been a main reason for the slow growth of plants in this period. Water hyacinths were inhibited from growing under highly alkaline conditions, with yellowing leaves, relatively small plants, and some plant deaths (Zhang et al., 2013). With increasing pH, the plant height, leaf width and biomass of water hyacinth decreased signi cantly (Wang et al., 1996). The plant height increased obviously in Oct., which indicated that Iris pseudacorus L. had a strong ability to adapt to low temperatures. In Nov., with a further decrease in temperature, plant metabolism and enzyme activity decreased, plant growth stopped, and some leaves began to turn yellow and wither.

Effect of temperature on nitrogen removal
Because the in uent of this experiment was the actual river water, the TN concentration in the in uent uctuated greatly. Fig. 12 shows that the TN removal e ciency was positively correlated with air temperature (P<0.01). The change trend of removal e ciency was divided into three stages. The early period was a period of rapid growth (Jul. 29-Aug. 5). This stage was the reproduction and growth stage of the wetland microbial community, and the removal e ciency was 16.80-65.20 %. The daily average lowest temperature was 23 ℃, and the daily average highest temperature was 30 ℃. The middle period was the stable period (Aug. 5-Oct. 2), with a removal e ciency of 37.70-84.00 %, a daily average lowest temperature of 18 ℃, and a daily average highest temperature of 30 ℃. Due to the high temperature and vigorous growth of microorganisms in this period, the TN removal e ciency was maintained at a high level. The later period was the declining period (Oct. 2-Nov. 28), with a removal e ciency of 5-54.2 %, a daily average lowest temperature of 3 ℃, and a daily average highest temperature of 14 ℃. With decreasing temperature, the denitri cation effect in the system fell accordingly due to the weakening of microbial activity. The total removal e ciency of TN and NH 4 + -N was set as group A, and the lowest temperature and the highest temperature were set as group B. Then, canonical correlation analysis was performed. The correlation coe cient was 0.808 (P<0.01), and A= -0.930 TN total removal e ciency+0.126 NH 4 + -N total removal e ciency, and B= -0.296 lowest temperature-0.724 highest temperature.
The maximum temperature was the main factor affecting the total removal e ciency of TN. From the discussion in Section 3.1, it can be seen that the TN removal e ciency was higher and the impact of air temperature change was greater in the later units of the system. The nitrogen removal performance began to be affected by the temperature drop around Oct. 2 (the daily lowest temperature was 15 ℃, and the overall temperature showed a downward trend starting on this day). The TN removal e ciency in the FHSCW and subsequent units all showed an asynchronous decline, indicating that the denitri cation bacteria, as the main nitrogen removal mechanism in these units, were signi cantly affected by temperature changes (P<0.05). Huang et al. (2000) obtained the rate constant (K T ) formula related to air temperature, , and predicted the total Kjeldahl nitrogen concentration in wetland e uent using the formula. Because the NO 3 --N concentration was very low, the formula could be used to predict the e uent TN concentration. However, the limitation of this formula was that it was suitable for predicting only the e uent TN concentration under different HRTs at a certain temperature. In this paper, under an HRT of 5.27 d, the in uence of temperature change on TN removal e ciency was explored. The nonlinear tting equation is shown in Fig. 13, and the relationship between the highest temperature and TN removal e ciency was obtained as follows: where x is the highest daily temperature, °C; y is the total TN removal rate in the MFCW, %.
The temperature change has a strong impact on denitri cation. The TN removal e ciency in the experimental study of CWs for wastewater treatment increased with increasing temperature in the range of 22-32 ℃ (Yang et al., 1991). Low temperature affected the proliferation rate and activity of denitrifying bacteria. The denitri cation rate decreased by 9 % per 1 °C decrease in temperature. The denitri cation speed was very slow when the temperature was lower than 5 ℃ (Yang et al., 2009;Lu et al., 2006). At the same time, it further showed that plant uptake had little effect on TN removal because it Oct. was a rapid period of plant growth but the TN removal e ciency was reduced by temperature. The inhibition of denitri cation would lead to the accumulation of NO 3 --N in the TFCWs (Chang et al., 2014;Hu et al., 2014;Li et al., 2015). When the temperature increased from 4 °C to 12 °C, the accumulation rate of NO 3 --N decreased, indicating that the lower temperature restricted the denitri cation reaction and promoted the accumulation of NO 3 --N (Pang et al., 2015). In the experiment, the TN removal e ciency in the system decreased after Oct. (i.e., denitri cation was inhibited). When the removal e ciency of NH 4 + -N had an upward trend (i.e., the system still had good nitri cation performance), the accumulation of NO 3 --N was promoted.
There was a signi cant negative correlation between the NH 4 + -N total removal e ciency and temperature (P<0.01). The removal of NH 4 + -N is mainly completed by two groups of chemoautotrophic bacterium through two processes. The rst step is to strictly oxidize NH 4 + -N to nitrite nitrogen (NO 2 --N) by strict aerobe. Such bacteria have been identi ed as N. europaea in fresh water and soil, as well as N. briensis, Nitrosovibrio (N. tenuis) and N. multiformis in soil. The second step is to oxidize NO 2 --N to NO 3 --N by facultative aerobe. Only one Nitriteoxidizing bacteria (NOB), i.e., Nitrobacter winogradskyi, was found in soil and fresh water, and Nitrospira gracilis was found in seawater. After Oct., the CWs still had good nitri cation performance, which may have been due to the average water temperature being higher than the lowest temperature required for nitri cation (5 ℃) (Lu et al., 2009). When the temperature was lower than 15 °C, the metabolism of amonifying bacteria was inhibited, and the growth activity was weak , which reduced the ammoni cation e ciency of organic nitrogen, and nitri cation was greater than ammoni cation; thus, the concentration of NH 4 + -N was reduced. Pang et al. (2015) found that at 4 ℃, 8 ℃ and 12 ℃, the TFCW had a high and stable removal e ciency of NH 4 + -N (93-96 %). In addition, the conversion rate of NH 4 + -N showed little change in the TFCWs (P<0.05), indicating that temperature had little effect on the conversion of NH 4 + -N.

Contribution of plant harvesting to nitrogen removal
The growth of aquatic plants in the MCWP played an important role in nitrogen removal. On the one hand, nitrogen taken up by plants was used for their own growth and metabolic needs and converted into plant protein and organic nitrogen. Then, nitrogen was removed from the system through harvesting (Brix, 1997;Wu, 1995). Because the role of plant nitrogen uptake accounts for a small part of N removal in wetlands, the effect of nitrogen removal by plant harvesting was not obvious (Robert and Kadec, 2003;Kem and Idler, 1999). On the other hand, the soil volume touched by the developed roots of emergent plants in wetlands varied depending on the depth and extension patterns of roots (Zhao et al., 2006) and formed a well-diffused and large-scale physical environment in the rhizosphere to support the survival and reproduction of a large microbial community. In addition, root exudates increased the organic matter content in the substrate of the rhizosphere, which promoted the growth and reproduction of microorganisms. Plant roots can transport oxygen to the substrate, forming aerobic and anoxic microenvironments around the rhizosphere, which is conducive to nitri cation and denitri cation.
Because of the periodicity of plant growth, some people advocate regular harvesting of plants to prevent pollutant accumulation in the plant caused by the decomposition of plant residues, resulting in another release and causing secondary pollution (Jiang et al., 2013). Others believe that the organic matter released in the process of plant decomposition enters the water body. After that, it provides a certain carbon source for the anaerobic bacteria in the water body and sediment and promotes the denitri cation effect (Fennessy et al., 2008;Chimney and Pietro, 2006). At the end of this experiment, the concentration of TN in the e uent was occasionally higher than that in the in uent, which was caused by the release of NH 4 + -N into the environment after the decomposition of plant roots, in addition to the denitri cation limited by temperature. Qi and Liu. (2016) showed that the decomposition of Phragmites australis could increase the concentration of NH 4 + -N in the e uent. The NH 4 + -N concentration increased with increasing residue amount. Therefore, appropriate harvesting frequency, methods and timing are very important. During the experiment, the in uent nitrogen load rate in the FHSCW was 321.95 g/m 2 .
The nitrogen mass balance results showed that the TN removal in the FHSCW, SHSCW and SFCW was 17.08 g/m 2 , 23.57 g/m 2 and 6.29 g/m 2 , respectively. The TN removal e ciency in the HSCW was 2.7-3.7 times that in the SFCW. In the MCWP, the HSCW played a leading role in TN removal, which was mainly caused by different substrates. In the HSCW, the substrates were gravel and steel slag, while the substrate of the SFCW was soil. Compared with soil, gravel and steel slag have larger speci c surface areas and stronger adsorption capacities (Lu et al., 2016) and are more conducive to the growth and reproduction of microorganisms (Sun et al., 2020). Therefore, the TN removal e ciency in the HSCW was better than that of the SFCW. The results showed that the substrate played an important role in the puri cation of polluted water by CWs.
Plant uptake played an important role in TN removal in the CWs. Based on previous research, the nitrogen content of Iris pseudacorus L.
leaves was 20.11 mg/g. The TN uptake of Iris pseudacorus L. was 2.66 g/m 2 (FHSCW), 2.76 g/m 2 (SHSCW) and 3.43 g/m 2 (SFCW), accounting for 15.57 %, 11.71 % and 54.53 % of the total removal, respectively. In the CWs used for surface water pretreatment (i.e., the HSCW in this study), the amount of nutrients removed by harvesting plants was not signi cant. In the SFCW used for advanced treatment, harvesting plants played a greater role in nitrogen removal. The results of Vymazal et al. (1998) were consistent with those of this study, i.e., the amount of nitrogen removed by harvesting in CWs for secondary treatment accounted for 10 %-16 % of the total N removal under optimal conditions. The amount of nitrogen removed by harvesting in CWs for advanced treatment may play a more important role. Therefore, in the study of the SFCW, a series of problems urgently need to be solved, such as harvesting time, harvesting frequency, and the relationship between the release of pollutants and environmental factors in CWs, which are of great signi cance to improve the puri cation effect of the SFCW.
The puri cation effect of the SFCW directly depends on the retention time of pollutants (Sabokrouhiyeh et al., 2016). The HRT in the SFCW increased with increasing plant density (Jaddhav and Buchberger., 1995). The planting density was positively correlated with the HRT. Properly increasing planting density could improve mixing e ciency in CWs. When the planting density reaches a certain threshold, the HRT of CWs increases slowly (Sabokrouhiyeh et al., 2016). Luo (2005) showed that increasing planting density could improve the removal effect of NH 4 + -N and TN. However, when the planting density was too high, further increasing planting density had little effect on the improvement of nitrogen removal, even though it was not conducive to nitrogen removal. Therefore, when the SFCW was used for advanced treatment, a suitable high-density planting method could effectively increase the proportion of nutrients removed by plant harvesting and enhance the level of pollutant removal.

Analysis of bacterial community structure and composition in CWs
Alpha diversity analysis The functions and behaviors of bacteria in uence the performance of CWs. Therefore, the bacterial community diversities and phylogenetic structures were analyzed through Illumina pyrosequencing in the two-level HSCW and SFCW, which are the core units of the MCWP. Table 3 summarizes the richness and diversity indices of the bacterial community at the genus level in the CWs when the system was operated on the 59th day. The alpha diversity indices were calculated based on operational taxonomic units (OTUs) at a 97 % cutoff, with 2491, 2612 and 2975 OTUs identi ed in the two-level HSCW and SFCW, respectively. The number of OTUs increased gradually in the CWs along the MCWP. The results showed that ACE, Shannon, Chao1 and Sobs were greater in the SFCW, and coverage and Simpson were smaller in the SFCW than in the HSCW. The ACE and Chao1 indices indicate bacterial community richness, and Sobs is the actual observation of richness. The Shannon and Simpson indices re ect bacterial community diversity . Therefore, compared with the HSCW results, the bacterial community richness and diversity were higher in the SFCW. This difference might be because of the radial oxygen loss and root exudates in the SFCW with high-density planting, which had a signi cant impact on rhizosphere bacterial activity, community richness, diversity and function (Chen et al., 2015;Liang et al., 2016;Chen et al., 2019). Plants might enhance nitri cation e ciency by releasing oxygen from roots, supporting the growth of nitri ers (Li et al., 2018). Denitri cation could also be promoted by organic compounds in plant exudates, which provide a carbon source for denitri ers (Hou et al., 2018). Therefore, plant roots create favorable environments for the growth and reproduction of speci c bacterial communities in the rhizosphere, especially ammonia-oxidizing bacteria (AOB) and NOB (Faulwetter et al., 2013). It was found that AOB and NOB in the rhizosphere were 17.9 % and 26.8 % higher, respectively, than those in the non-rhizosphere by the most-probable-number method (Hua et al., 2017). Therefore, the SFCW showed greater bacterial diversity and was more effective in removing nitrogen from actual polluted river water.

Bacterial species composition analysis
As shown in Figs. 14 and 15, the main bacteria detected in the two-level HSCW and SFCW were Saprospiraceae, Gemmatimonadaceae, Gammaproteobacteria and Alphaproteobacteria. At the genus level, Thiobacillus, Terrimonas, Hygromobium and Nitrospira were the dominant bacteria in the two-level HSCW, and Hydrogenophaga was the dominant bacteria in the SFCW.
Proteobacteria is one of the main dominant bacteria in wastewater treatment plants. The change in the sludge bacterial community in the actual printing and dyeing of wastewater by biological treatment systems also shows that Alphaproteobacteria and Gammaproteobacteria are the main dominant bacteria (Yang, et al., 2012). Gammaproteobacteria can metabolize glucose to produce acid, reduce nitrate to nitrite, use ammonium as a nitrogen source and glucose as a carbon source, and grow at 10-43 ℃. Saprospiraceae is the main bacteria family causing lamentous bulking of sewage treatment plant sludge. It is a strictly aerobic chemoorganotrophic bacterium whose optimum living environment temperature is 30-37 °C, and when the pH is approximately 7, it can metabolize glucose, galactose, acetate, etc. The nitrifying bacteria in the activated sludge of sewage treatment plants are mainly Nitrospira, which can oxidize nitrite into nitrate in CW media, and Nitrospira are the main nitrogen removal functional bacteria in the FHSCW. So far, Nitrospira have been reported in many papers. NOB are dominant in the process of biological nitrogen removal. As autotrophic bacteria, when coexisting with heterotrophic bacteria, they are easily restrained due to their long growth cycle and lack of competitive advantage in space and time. Therefore, the species abundance was low in the CWs of this study. Gemmatimonadaceae has strong resistance to extreme environments. Thiobacillus is a colorless, rod-shaped and facultative sulfur-oxidizing bacterium that mainly exists in sediment and wet soil. Among them, Thiobacillus denitri cans is the most common genus of Thiobacillus in sewage treatment and is a kind of autotrophic denitrifying bacteria. It can obtain energy through sul de oxidation in a low carbon source environment and take nitrate as an electronic receptor to generate nitrogen, in which nitrite and nitric oxide accumulate as intermediate products. The Hydrogenophaga detected in the SFCW are heterotrophic nitri cation-aerobic denitri cation (HN-AD) bacteria. In recent years, the discovery of HN-AD bacteria has altered the traditional theory that nitri cation can be performed only by autotrophic bacteria and denitri cation can be carried out only under anaerobic conditions. The theory of HN-AD has become a hot topic at home and abroad. The HN-AD bacteria isolated from the sediment of the seabed were identi ed as Klebsiella sp. It has been proven that the bacteria have high HN-AD e ciency (Sun et al., 2016). The bacteria selected from the activated sludge of land ll leachate were identi ed as Pseudomonas. The study showed that the bacteria had high nitrogen removal characteristics and could realize simultaneous nitri cation and denitri cation . Compared with traditional denitrifying bacteria, HN-AD bacteria have greater advantages in nitrogen removal and organic matter removal. The carbon source, C/N and DO are the main factors affecting nitrogen removal. The free surface of the SFCW promoted air reoxygenation, and the high-density planting mode enhanced radial oxygen loss and improved the aerobic conditions for the survival of HN-AD bacteria. Therefore, it is of great signi cance to study the natural restoration of polluted water by HN-AD bacteria.

Bacterial species diversity analysis
According to the bacterial community richness data obtained, a signi cance test of intergroup differences was carried out. Bacterial species with different abundances in different wetland units (samples) were detected by strict statistical methods, and the signi cance of the observed differences was evaluated by hypothesis tests. Fisher's exact test and FDR multiple test were selected as test methods.
As shown in Fig. 16, Saprospiraceae, the dominant species in the FHSCW, increased signi cantly in the SHSCW (P<0.01). This nding was consistent with our previous research (i.e., among units, the SHSCW has the best COD Mn removal effect, data not presented in this study).
Saprospiraceae are strictly aerobic chemoorganotrophic bacteria. The large speci c surface area of the gravel and steel slag in the HSCW provides a place for the growth and reproduction of bacteria. The genus Terrimonas is a known type of bacteria that can undergo aerobic denitri cation. Its species abundance decreased signi cantly in the SHSCW because when the DO in the in uent was consumed by the FHSCW and then pushed into the SHSCW, aerobic chemoorganotrophic bacteria, such as Saprospiraceae, further consumed the DO and organic matter in the water, which was not conducive to the growth and reproduction of Terrimonas. It was also found that the relative abundance of autotrophic bacteria such as Thiobacillus and Nitrospira decreased signi cantly, which may have been due to the survival competition of heterotrophic spirochetes.
As shown in Fig. 17, the species abundance of Saprospiraceae, the dominant species in the HSCW, was signi cantly decreased in the SFCW because the COD concentration in the e uent after the raw water owed through the two-level HSCW was reduced to the lowest level of the system (data not presented in this study), which also limited the growth of heterotrophic bacteria. Hydrogenophaga are HN-AD bacteria, and the species abundance of these new denitrifying functional bacteria increased signi cantly in the SFCW, which indicated that the root exudates and radial oxygen loss in the SFCW with high-density planting improved the rhizosphere microenvironment. At the same time, the species richness of Alphaproteobacteria and Sphingomonas increased signi cantly in the SFCW. Alphaproteobacteria (root nodule bacteria) have been shown to be capable of denitri cation under autotrophic and heterotrophic conditions in liquid media (Vilar-Sanz et al., 2018). Sphingomonas widely exists in plant rhizosphere soil and has good biological xation of nitrogen and biodegradation functions (Hansen et al., 2020;Ali et al., 2019). A Sphingomonas sp. strain isolated and screened from a shrimp culture pond e ciently removed both ammonia nitrogen and nitrite (Yun et al., 2019). Therefore, the bacteria completed the nitrogen cycle in the SFCW with high-density planting through a variety of nitrogen removal pathways.

Conclusions
An MCWP was constructed to investigate the in uence of seasonal uctuations in temperature on nitrogen removal in the system. It was found that the TN concentration in the MCWP gradually decreased. The average removal e ciency in the MCWP was approximately 40.74%. The removal e ciency in the SPP was the best among all units, with a value up to 16.08%. It was found that the TN total removal e ciency was signi cantly positively affected by the daily high temperature. A formula between the total removal e ciency and the highest temperature was obtained by nonlinear tting, and the tting result R 2 reached 0.7192. The TN removal load rate per unit area in the HSCW was 2.7-3.7 times that in the SFCW. The bacteria completed the nitrogen cycle in the SFCW, which had high-density planting (35 plants/m 2 ), through a variety of nitrogen removal pathways. Furthermore, the TN uptake by Iris pseudacorus L. accounted for 54.53% of the total TN removal in the SFCW. The SFCW, as an advanced treatment unit, could effectively increase the proportion of nutrients removed by plant harvesting and enhance the removal level of pollutants.

Declarations
Ethics approval and consent to participate Not applicable.

Consent for publication
Not applicable.

Availability of data and materials
Page 12/24 The data are available from the corresponding author upon reasonable request.

Competing interests
The authors declare that they have no competing interests.

Contributions
Lu H B conceived and designed the experiments. Wang T, Li J X, Xiao L P and Zhao X L performed the experiments. Wang T analyzed and interpreted the data. Wang T and Lu H B were major contributors in writing the manuscript. Lu S Y and Guo X C funded the work.
Zhao L F, Zhu W, Zhao J. Nitrogen removal mechanism in const-ructed wetland used for treating polluted river water with lower ratio of carbon to nitrogen. Acta Scientiae Circumstantiae. 2006;26: 1821-1827. 10.13671/j. hjkxxb. 2006 Zheng Y C, Wang X C, Xiong J Q, Liu Y J, Zhao Y Q. Hybrid constructed wetlands for highly polluted river water treatment and comparison of surface-and subsurface-ow cells. Journal of Environmental Sciences. 2014; 26: 35-42. https://doi.org/10.1016/S1001-0742(13)60482-9 Zhi W, Ji G D. Quantitative response relationships between nit-rogen transformation rates and nitrogen functional genes in a tidal ow constructed wetland under C/N ratio constraints.   Bacterial species diversity analysis of the two-level HSCW Figure 17 Bacterial species diversity analysis of the SHSCW and SFCW