《1. Introduction》

1. Introduction

Excessive nitrogen (N) from agricultural ecosystems and sewage is a severe problem in agricultural and urban systems around the world [1], causing eutrophication, algal blooms, fish kills, and potential damage to ecosystems. Moreover, high N concentrations in surface and ground waters potentially increase the risk of methemoglobinemia [2], mutagenicity, teratogenicity, and birth defects in humans [3]. To solve this problem, denitrifying bioreactors (DNBRs), also known as denitrifying beds or denitrifying walls, have been considered one of the most sustainable and costeffective approaches in the past two decades [4] due to their long lifespan (10–15 years) [5,6] and lower investment cost [7]. A field DNBR is usually constructed by filling up an organic carbon (C) source (typically sawdust or woodchips) in a barrier below the water table to intercept the groundwater flow [8,9]. Available C released from the media supports denitrification, in which nitrate (NO3– ) is reduced to N2 through a series of intermediate nitrogen oxide products.

The types of substrates used to improve DNBR function have been a topic of intensive study. In the initial stages of DNBR development, woody materials, such as tree bark, woodchips, and sawdust, were used as major sources of C [10–14]. Later, agricultural wastes-including maize cobs [15,16], wheat straw [17,18], rice [19], cotton [20,21], cornstalks [22], corn stover [16], conifers [23], barley straw [24,25], leaves [23], and green waste [26,27] were used as natural carbon (NC) sources. Nevertheless, woody materials are still preferred due to their sustainability and longevity [9].

Many studies have focused on improving woodchip-derived DNBRs (Woodchip+) by adding auxiliary materials, such as sand [13], steel byproducts [28], biochar [29,30], seashells [31], acetate [32], fly ash pellets [33], mire flora, and potatoes [34]. In addition, non-natural carbon (NNC) sources including synthetic C, biochar [35,36], and cardboard [24,25] have been investigated in the laboratory, achieving relatively high nitrate removal rates (NRRs; N removal (g∙m–3 ∙d–1)). Types of synthetic C include polycaprolactone (PCL) [37–39], polylactic acid (PLA), polybutylene succinate (PBS), poly-3-hydroxybutyric acid (PHB) [37,39], and poly-3- hydroxybutyrate-co-hydroxy valerate (PHBV). These materials can increase the NRR to around 400 g∙m–3 ∙d–1 [40]. Studies have also used a mixture of multiple materials (MMs), including various NCs or other industrial and agricultural waste. For example, combinations of woodchips, sulfur (S) particles, and zeolite [41] or of woodchips, corncobs, and modified coconut coir [32] have generated acceptable NRRs.

The nitrate removal mechanism is critical to the performance of an DNBR. Most studies have focused on heterotrophic denitrification (HD), which converts nitrate into nitrogen gas by means of denitrifying bacteria [42–45]. However, other mechanisms may be involved depending on substrate type, such as autotrophic denitrification (AD) [46,47], dissimilatory nitrate reduction to ammonium (DNRA) [48], anaerobic ammonium oxidation (anammox) combining nitrite and ammonium into nitrogen gas [49], and chemical denitrification (chemodenitrification; CD) [50]. The dominant nitrate reduction mechanism is a function of the substrate type. However, few studies have attempted to obtain a generalized pattern for the role of the substrate in nitrate removal.

In addition to the nitrate removal performance of various substrates, the cost of the substrate plays a critical role in real-world applications. The cost of DNBR installation generally ranges from 6 940 to 11 820 USD in the Midwestern United States [51]. In general, the N removal cost efficiency ranges from 0.54 to 48.00 USD∙ (kg∙a)–1 , depending on the bioreactor size, type, substrate, and location [52]. However, most cost reports have focused on woodchips rather than on other substrates.

In this review, we compare the performance and removal mechanisms of various substrates with the aim of helping practitioners and researchers choose optimal materials as media in their DNBR. Overall, even though these substrates have been partly reviewed in previous studies [53], their performance has not been quantitatively evaluated. Here, we review the nitrate removal performance of various substrates in DNBRs. Our specific objectives are to develop a taxonomy of substrates, assess the NRR and NRE of different substrates, summarize the nitrate removal mechanisms, and summarize the environmental and economic impacts of using DNBRs.

《2. Materials and methods》

2. Materials and methods

《2.1. Data compilation》

2.1. Data compilation

Data were compiled from published articles following the method described in the flow chart in Fig. 1(a). We found 14 articles related to ‘‘denitrification bed,” 36 related to ‘‘denitrification wall,” 83 related to ‘‘denitrifying bioreactor,” and 69 related to ‘‘permeable reactive barrier”. There was a steady increase in the number of publications related to these topics during the past two decades (Fig. 1(b)). The majority of the published articles were from the United States, followed by China, and then New Zealand. Research data used in this review were extracted from the articles using GetData Graph Digitizer software (Germany). In some cases, absent information was indirectly calculated. In total, we collected 10 179 sets of raw data from 63 peer-reviewed articles and compiled them in Microsoft Excel. When compiling the information, we included the following for each wall, bed, or laboratory column bioreactor unit: author(s), year of publication, article title, journal title, site, number of measures (n), medium type, dimension, flow rate, medium age, hydraulic retention time (HRT), influent nitrate concentration (Cin), temperature, dissolved organic carbon (DOC) concentration, dissolved oxygen (DO), pH, effective porosity, NRR, NRE, and the associated standard deviations (SDs). Batch experiments were not included in the database due to the lack of a water flow component.

《Fig. 1》

Fig. 1. (a) Flow chart of the data compilation; (b) the number of the published DNBR articles over the year. PRB: permeable reactive barrier; VT: volume of influent water; T: temperature; NRE: nitrate removal efficiency; NRR: nitrate removal rate;ρ: effective porosity; HRT: hydraulic retention time; DO: dissolved oxygen; DOC: dissolved organic carbon.

《2.2. Bioreactor types》

2.2. Bioreactor types

DNBRs are also known as denitrification beds, denitrification walls, or permeable reactive barriers (PRBs) [26]. Conceptual designs of DNBRs are provided in Fig. 2, including lab columns (Fig. 2(a)), lab bioreactors (Fig. 2(b)), field beds (Fig. 2(c)), and field walls (Fig. 2(d)). Lab columns and bioreactors are widely adopted for laboratory-scale studies with synthetic influents containing nitrate [4,54]. Field beds are generally lined and placed into the upper 1–2 m of shallow groundwater in a trench on either side of a tile drain container to control the water level, with a bypass pipe to remove excess water flow [9]. Field walls are installed perpendicular to the groundwater flow path to intercept nitrate-laden groundwater [8].

《Fig. 2》

Fig. 2. Conceptual scheme of four denitrifying bioreactors, including (a) lab columns, (b) lab bioreactors, (c) field beds, and (d) field walls.

《2.3. Bioreactor substrates》

2.3. Bioreactor substrates

Fig. 3 shows the classification system used to define and distinguish among the different substrate types. The system categorized the various DNBR substrates into four main groups: NC, NNC, inorganic material (IM), and MM. The proportion of the materials used in the DNBRs was 61.2% NC, 4.1% NNC, 0.5% IM, 32.9% MM, and 1.3% soil as control group. The NC group consisted of materials from agricultural systems, including woodchips, wheat straw, rice straw, barley straw, corncobs, corn stalks, corn stover, cotton, green waste, leaves, lodge pine needles, and macrophyte residues. The use of woodchips made up the highest percentage (45.8%) in the NC group, which was further categorized into hardwood and softwood groups. The NNC group, which was mainly composed of artificial chemical and physical synthases, included biodegradable polymers (e.g., PHB, PCL, and PBS) and wastes, and other byproducts (e.g., cardboard and biochar). IMs constituted only a tiny portion (0.5%) of the DNBR substrates and included pyrite, sand, and amorphous iron oxides. The MM group was subgrouped into Woodchip+ (27.9%) and corncob-derived DNBR (Corncob+) (4.6%), as most of the MMs were woodchip or corncob derived. Soil was used in most studies as an additive to the substrate; however, in the classification system, soil was considered to be an independent material or a control category.

《Fig. 3》

Fig. 3. A taxonomy of substrates in DNBRs based on a review of 219 bioreactor studies from 63 peer-reviewed articles. Substrates are classified into five main groups: NC, NNC, IM, MM, and soil, respectively. In the NC group, the single substrate of woodchips can be categorized as hardwood or softwood. The MM groups include Woodchips+, Corncobs+, and others. W: woodchips; CC: corncobs; Woodchip+: woodchips plus an auxiliary carbon source; Corncobs+: corncobs plus an auxiliary carbon source. Soil was used as a control group. The numbers in the bracket represent the percentage of involved studies.

《2.4. Meta-analysis of substrate effect》

2.4. Meta-analysis of substrate effect

A meta-analysis was conducted to analyze the nitrate removal response of NRR and NRE among various substrates. NRR [55] and NRE [19] were calculated as follows:

where VT is the volume of influent water (m3 ), Nin and Nout are the inflow and outflow nitrate concentrations (mg∙L–1), VBR is the total volume of the DNBR (m3 ), and t represents the operation time (d).

A meta-analysis is a widely used tool that provides an indication of the comparable magnitude of differences in variables. In this study, the NRR and NRE from different studies were investigated and compared according to their effect size [8]. The effect size reflects the magnitude and direction of the treatment effect from each study [56], including the odds ratio, relative risk, risk difference, weighted mean difference (WMD), and standardized mean difference (SMD) [57,58]. In this study, we chose WMD as the effect size indicator, because our data (NRR and NRE) were continuous variables with common units (g∙m–3 ∙d–1 or percentage). Thus, it was easier to interpret the findings using WMD (with units) compared with a unitless term such as SMD [56]. The WMD is based on three parameters: ① mean value (mean NRR and NRE); ② SD; and ③ number of measurements (n). The above information may also be obtained for control treatments (i.e., NRR and NRE without a substrate in a DNBR). However, the data collected from published articles did not always include control treatments, so we set the value of our control NRR (0.1 g∙m–3 ∙d–1; SD: 0.01) according to the value set by Addy et al. [8]. The control NRE was set at 1% and the SD at 0.01.

Each DNBR (wall, bed, and laboratory column) was defined as one unit. Based on the database we established, we calculated the mean NRR and NRE values, the associated SD values, and the n of each DNBR as a unit to perform a meta-analysis. As shown in Table 1, we then assigned substrate categories (main groups: NNC, NC, auxiliary, Woodchip+, wood source, scales, and carbon to nitrogen (C/N) ratio). The C/N ratio is the ratio of the mass of carbon to the mass of nitrogen in a substrate. Previous studies have reported a wide range of C/N ratios for substrate, so we roughly classified the media C/N ratio into two categories: > 100 and < 100. To account for scaling effects on the NRR and NRE, the DNBR systems were divided into lab- and field-scale categories; the resulting groups and subgroups are summarized in Table 1. In addition, the lignocellulose index (LCI) was used as an indicator to describe the C quality in natural C sources (e.g., woodchips). The LCI was calculated as follows [16]:

where the P means the percentage of lignin or cellulose dry weight.

《Table 1》

Table 1 Ranges of data within categories for the parameters assessed by the meta-analysis.

We used a random-effects model, which accounts for the variation in methods between studies and the sampling error within individual studies. The 95% confidence intervals (95% CIs) were corrected for bias and calculated by means of bootstrapping (i.e., sampling with replacement), with 999 iterations on the mean effect size. All meta-analysis was done using STATA (version 16; StataCorp, USA) software. Forest plots of our meta-analysis results were constructed in Microsoft Excel. The details in categories of meta-analysis by the effect size of NRR and NRE can be seen in Tables S1–S5 in Appendix A.

Many of the included studies did not report the age of the materials in their bioreactors, so there was insufficient data to distinguish the nitrate removal effect of different substrates according to certain abiotic factors, including the scale and duration of the experiments. Therefore, the average values of the meta-analysis should not necessarily be used for design purposes but can rather be used as a relative comparison between different substrates.

《2.5. Cost analysis》

2.5. Cost analysis

To evaluate the substrate’s cost-effectiveness, we used the substrate price (USD∙m–3 ) divided by the NRR as an economic analysis of the substrate:

where NRRi represents the ith substrate in NRR (g∙m–3 ∙d–1) and Pricei represents the market price of the ith substrate in USD∙m–3 . SubEco represents the substrate economic index in N (g∙d–1∙USD–1). A greater SubEco means that that substrate is more cost-effective as media. Because the substrate price varies between locations and with time, we used a range of costs in the United States in 2021, obtained by searching the online market (Table S6 in Appendix A). This statistic could be recalculated according to the local market.

《3. Results and discussion》

3. Results and discussion

《3.1. Performance of different substrate classes》

3.1. Performance of different substrate classes

As shown in Fig. 4, the NRR effect sizes were ranked in order of magnitude as follows: NNC (369.9 g∙m–3 ∙d–1) > MM (11.1 g∙m–3 ∙d–1) > NC (8.1 g∙m–3 ∙d–1) > IM (2.7 g∙m–3 ∙d–1) > soil (0.3 g∙m–3 ∙d–1) (Fig. 4(a)). For the NRE, the ranking order was NNC (79%) > NC (58%) = MM (58%) > IM (54%) > soil (37%) (Fig. 4(b)). The overall mean NRR and NRE for the substrate groups were 9.6 g∙m–3 ∙d–1 and 58%, respectively.

《Fig. 4》

Fig. 4. Mean NRR by the categories of (a) main groups, (c) NC, (e) NNC, (g) auxiliary, and (i) woodchip+. Mean NRE by the categories of (b) main groups, (d) NC, (f) NNC, (h) auxiliary, and (j) woodchip+. The bracketed number represents the number of DNBR units in that category that were used in the analysis. The error bar represents the 95% CI. The grey dashed line represents the mean value of the effect size in the main groups.

The NNC media showed the greatest NRR and NRE, because NNC contains biodegradable polymers and is likely easier to decompose than the C in the NC group, which contains a large number of recalcitrant C substances. In addition, biodegradable polymers were often used to treat concentrated nitrate-laden water, such as recirculated aquaculture systems [37] and wastewater [39]. A higher NRR generally indicates the need for more frequent substrate replacement-that is, when there is a faster supply of C from the substrate to the microbes, the C runs out more quickly. Therefore, biodegradable polymers (NNC) showed a much greater NRR than the other substrates. Conversely, soil had the lowest NRR and NRE values. In most cases, the soil in DNBRs is a mixture of material with natural organic C, containing a large number of microorganisms, which can be treated as a microbial inoculation. Even though the NNC had a greater NRR than the other groups, it had a large variance, which was mainly attributed to the reported biodegradable polymers with a high value and the cardboard with a lower NRR. 

3.1.1. Nitrate removal by means of NC

In the NC group, the NRR effect sizes decreased in the following order: corncobs (19.9 g∙m–3 ∙d–1) > corn stover (13.6 g∙m–3 ∙d–1) > macrophyte residue (12.7 g∙m–3 ∙d–1) > green waste (12.2 g∙m–3 ∙d–1) > wheat straw (8.3 g∙m–3 ∙d–1) > woodchips (7.1 g∙m–3 ∙d–1) > barley straw (5.3 g∙m–3 ∙d–1) > rice straw (4.8 g∙m–3 ∙d–1) > pine needles (1.0 g∙m–3 ∙d–1) (Fig. 4(c)). The NRE effect sizes were ranked in decreasing order: pine needles (98%) > leaves (91%) > macrophyte residue (81%) > barley straw (71%) > woodchips (59%) > corncobs (52%) > wheat straw (47%) = green waste (47%) > rice straw (38%) > corn stover (28%) (Fig. 4(d)). These NC sources are popular as substrates due to their accessibility and low cost [9]. Many studies have investigated the performance of NC materials in removing nitrate from agricultural drainage water. Healy et al. [25] recorded the NRE for NC sources including lodge pine needles (75%), barley straw (74%), and lodge pine woodchips (70%). Cameron and Schipper [26] compared several NCs and ranked the NRR from highest to lowest: corncobs (15.00–19.8 g∙m–3 ∙d–1) > green waste (7.8– 10.5 g∙m–3 ∙d–1) > wheat straw (5.8–7.8 g∙m–3 ∙d–1) > softwood (3.0–4.9 g∙m–3 ∙d–1) > hardwood (3.3–4.4 g∙m–3 ∙d–1). The above NRR and NRE results are consistent with our results, with the exception of pine needles and corn stover. The performance of these NCs was significantly different in our evaluations, possibly due to differences in HRTs, N loading rates, and experiment duration.

Woodchips are the most reported NC material, with 95 NRR and 86 NRE citations. Woodchips generate intermediate NRR and NRE values, which may be associated with low C release (0.13– 0.53 mg∙(g∙L)–1) [59] and low surface area (2.7 m2 ∙g–1 ) [25], compared with the other NCs (Figs. 4(c) and (d)). Most NCs showed a greater NRR than woodchips, but woody media is preferable due to its stability and sustainability [8], as well as its cost, conductivity, longevity, and C/N ratios [9]. As shown in Fig. 5, the NRR effect size of softwood was greater than that of hardwood (p < 0.05) (Fig. 5(a)). However, there was no significant difference in the NRE effect sizes of hardwood and softwood (Fig. 5(b)). The average NRR effect sizes of softwood and hardwood were 10.5 and 3.9 g∙m–3 ∙d–1, respectively (Fig. 5(a)). Previous studies showed that softwood had a greater NRR than hardwood in the first ten months of a DNBR [8,26]. This result was probably due to the indirect effects of wood density on C decomposition [8]. Even though the higher density of the hardwood is expected to result in a greater C supply than that of softwood, C is more easily released from softwood because of the low density and greater oxygen infusion during decomposition [60].

《Fig. 5》

Fig. 5. Mean NRR effect size by the categories of (a) wood type, (c) C/N ratio, and (e) scale. Mean NRE by the categories of (b) wood type, (d) C/N ratio, and (f) scale. The bracketed number represents the number of DNBR units in that category that were used in the analysis. The error bar represents the 95% CI. The grey dashed line represents the mean value of the effect size in the main groups.

3.1.2. Nitrate removal by means of NNC

In terms of NNC, the biodegradable polymers presented a much higher NRR than cardboard (Fig. 4(e)). In contrast, cardboard showed a greater NRE than the biodegradable polymers (Fig. 4(f); p < 0.05). A possible reason why a polymer substrate generates a greater NRR and lower NRE is that the DNBR studies in which such substrates appear use higher nitrate loading rates (38.8–95.2 mg∙L–1) than those used in other studies [37,38,40]. The NRR commonly increases with the influent nitrate concentration [61]; however, as the Michalis–Menten model showed, there is an upper limit to the NRR, which is likely controlled by the C supply from the substrate [62]. In contrast, the NRE always decreases with increasing influent nitrate concentration [63]. In addition, the studied polymers can treat both nitrate and ammonium through simultaneous nitrification and denitrification in laboratory-scale bioreactors [39,64]. In combination, these processes generate higher NRR values than when traditional C sources are used (e.g., woodchips), which rely on HD alone. Cardboard has been studied as a substrate in DNBRs, primarily due to its relatively high surface area and C content [25,65,66]. Although lower NRR values for cardboard have been reported [24], this substrate generates greater NRE values relative to woodchips and barley straw [25]. In contrast to biodegradable polymers, cardboard has a lower NRR performance. However, due to its low cost and recycling capacity, cardboard may hold potential to become a substrate in real-world applications.

Other NNC sources, such as biochar [65], are uncommon as the only substrate in a DNBR. Nevertheless, the performance of biochar has been associated with an improved microorganism habitat [66] and increased nutrient retention capacity [35], so biochar is widely used as an auxiliary material in Woodchip+ for improving the NRR [29,36].

3.1.3. Nitrate removal by means of IMs

Few IMs have been reported as DNBR substrates. We were only able to find a report on the use of pyrite as a sole substrate to remove nitrate by means of AD [46]. As shown in Fig. 4(a), IMs had the lowest NRR and NRE compared with NNC, MMs, and NC. Most IMs play an auxiliary role in the bioreactor. For example, Hua et al. [28] reported that adding steel byproduct into woodchip-based bioreactors removed both nitrate and phosphate. Li et al. [41] added S and zeolite together into a woodchip-based bioreactor, which achieved up to 86.6% NRE and a half nitrate removal was contributed to AD. Thus, IMs may be appropriate as an auxiliary substrate when included in NC-based bioreactors to increase the resilience of nitrate removal under complicated conditions, rather than relying on heterotrophic removal alone.

3.1.4. Nitrate removal by means of MMs

Our study showed that adding auxiliary materials into a Woodchip+ increased the NRR, except for Corncob+ (Fig. 4(g), p > 0.05). While Fig. 4(h) showed that adding auxiliary materials didn’t increase the NRE. In addition, adding other NCs into a woodchipderived DNBR (W + NCs) showed a greater NRR than the addition of NNCs and IMs (Fig. 4(i)). However, adding auxiliary materials into a Woodchip+ and Corncob+ did not increase the NRE (Fig. 4(j)).

The different effects of the addition of auxiliary materials on woodchip and corncob NRR values is probably because a corncob substrate has a greater available organic C than a woodchip substrate [16]. It is expected that NRR will increase with C content and lability, but this is not a linear relationship [67]. Thus, adding an extra C source to a Corncob+ (which has a higher C content and lability) may not have much influence on the NRR and NRE, compared with adding the source to a Woodchip+, which has a lower C content and lability. Furthermore, even though a Woodchip+ has a lower NRR than a Corncob+, the Woodchip+ has been verified as a sustainable and stable DNBR by several previous studies [5,9,68]. 

The woodchip-based DNBR has been extensively studied [4,12,13,16,32,34,59,68,69]; however, there is no consensus about which auxiliary material is most effective for improving the NRR and NRE in a woodchip DNBR. Our study indicates that adding NCs into a woodchip DNBR results in a greater NRR, relative to the addition of NNCs and IMs (Fig. 4(i)). We infer that this result is probably due to the greater availability of C from the addition of NCs. A study has reported that adding NNCs (e.g., biochar) did not increase the NRR [4], while another study reported that adding IMs (e.g., steel byproduct) removes phosphate rather than nitrate [28]. We also studied the performance obtained by adding auxiliary materials to a Woodchip+. The NRR values obtained by adding each auxiliary material in descending order were as follows: sodium acetate (30.7 g∙m–3 ∙d–1) > potato (20.5 g∙m–3 ∙d–1) > fly ash pellet (19.1 g∙m–3 ∙d–1) > corncobs (18.0 g∙m–3 ∙d–1) > mire flora (15.2 g∙m–3 ∙d–1) > steel (11.3 g∙m–3 ∙d–1) > biochar (10.3 g∙m–3 ∙d–1) > seashells (1.8 g∙m–3 ∙d–1) > flaxseed cake (1.6 g∙m–3 ∙d–1) > sand (1.5 g∙m–3 ∙d–1) > filtralite (1.4 g∙m–3 ∙d–1) (Fig. S1(a) in Appendix A). In descending order, the obtained NRE values were: steel (90%) > potato (86%) > sodium acetate (84%) > fly ash pellet (73%) > mire flora (69%) > sand (54%) > biochar (53%) > corncobs (38%) (Fig. S1(a)).

These results imply that the content and lability of the C in the auxiliary sources and the blending choice have a significant influence on the NRR of an MM DNBR. However, we found that the application of MMs in field experiments was very rare. It is expected that MMs can be customized according to the local environment and the desired NRR and NRE. To achieve these goals, more diverse substrates should be considered in order to make up for the deficiencies of each component material. For example, adding biochar and silage leachate together into a woodchipbased bioreactor might be a win–win strategy. As an industrial byproduct, biochar can provide a favorable habitat for microbes with its higher surface area. In addition, silage leachate is an agricultural waste that contains plenty of C sources. This combination could improve the nitrate removal performance of woodchip bioreactors. Therefore, combining various substrates holds great potential for increasing the NRR under a wide range of environmental conditions and may be an approach for maintaining a balance between longevity and nitrate removal. Overall, the MM-derived DNBR is the optimal choice, as it is able to operate in the long term, increases nitrate removal, and reduces environmental risks.

《3.2. The mechanisms of nitrate removal associated with each substrate》

3.2. The mechanisms of nitrate removal associated with each substrate

3.2.1. Nitrate removal by means of media C/N ratio

In general, woodchips and cardboard materials have a high C/N ratio (> 100), while the C/N ratio of single substrates, such as barley straw, wheat straw, and corn stalks, is below 100. As shown in Fig. 5(c), the materials with C/N ratios below 100 had a greater NRR than the substrates with higher C/N ratios (> 100). Fig. 5(d) shows that the C/N ratio has no significant influence on the NRE. More specifically, Fig. 6(a) shows that if aiming to optimize the NRR, then the substrate C/N ratio should be below 100 or above 300. In contrast, the optimal C/N ratio for the NRE is between 100 and 200 (Fig. 6(b)). Thus far, it is not possible to determine an optimal C/N ratio. The observed decrease in NRR for materials with a higher C/N ratios is probably due to the influence of woodchips and cardboards; these materials have lower decomposition rates because the C source is mainly made of cellulose, hemicellulose, and lignin [70]. Lignin is more recalcitrant than hemicellulose and cellulose [71], so its LCI is greater. As shown in Table 2 [25,71– 76], woodchips (both hardwood and softwood) and cardboard have a greater LCI and C/N ratio than other plant materials. This may explain why such materials with a higher C/N ratio generated lower NRRs. Moreover, it has been observed that the C/N ratio decreases and the LCI increases with time [72]. It has also been suggested that the C/N ratio is critical for the longevity of a DNBR, and that the LCI is a C quality indicator [16,72].

《Fig. 6》

Fig. 6. The (a) NRR effect size and (b) NRE effect size correspond to the C/N ratio of the substrates. The red line is the polynomial curve-fitting for the NRR or NRE effect size with the C/N ratio of the substrates. y represents the NRR or NRE effect size; x represents the C/N ratio of the substrates; R2 is the coefficient of determination.

《Table 2》

Table 2 Ranges of data within categories for the parameters assessed by the meta-analysis. For the list of agricultural waste chemical properties and C/N ratios, the data comes from Refs. [25,71–76].

n.a.: not available.

3.2.2. Nitrate removal by experiment scale

As shown in Figs. 5(e) and (f), the lab- and field-scale NRR effect sizes were 10.6 and 4.1 g∙m–3 ∙d–1 and the NRE effect sizes were 57% and 66%, respectively. This significant difference of NRR may be attributed to the greater control in a lab over experimental variables, such as stable higher temperatures (20–25 °C) and higher nitrate loading (> 10 mg∙L–1), than is possible in natural agricultural drainage systems [9]. In addition, lab studies generally use fresher material with more degradable C and run for a shorter period of time [8]. Furthermore, the activity of the dominant denitrifying bacteria can vary greatly with long-term exposure to synthetic wastewater, due to their acclimation to stable wastewater properties [3]. Moreover, the influent water in the field comprises solutions with variable and complicated compositions compared with synthetic laboratory solutions. It has also been reported that higher salinity inhibits nitrate removal [77]. Thus, our evaluation suggests that it is inadvisable to compare the performance of DNBRs at different experimental scales, and that the design of field-scale bioreactors should not use data derived only from short lab-based studies.

3.2.3. The relationship between NRE and NRR

In general, for the NC and MM categories, the NRR increased with increasing NRE, as shown in Fig. 7. A small number of NNC and IM samples showed a high NRR when the NRE was in the range of 20%–80%. However, all the substrates showed a greater uncertainty when the NRE ranged from 80% to 100%. This could be because the denitrification was limited by the nitrate concentration of the influent water, or because the HRT was too high, resulting in a low NRR value. More specifically, the influent nitrate concentration has been widely reported to be a potentially limiting factor for denitrifying bacteria [62,78]. Thus, a lower nitrate loading will limit the denitrification process. In addition, the longer the HRT of a DNBR, the lower the treated water volume but the longer the denitrification reaction time. Thus, most researchers have reported that the NRR decreased with increasing HRT [79], while the NRE increased with increasing HRT [79,80]. To date, the NRR has been used more frequently than the NRE when reporting DNBR studies, because the NRR is best for design purposes in field studies, whereas the NRE is dependent on the load and substrate. Although the NRR has been advocated as a replacement for the NRE in the past few years [8], we believe that the two indicators (NRR and NRE) can be used to evaluate the performance of a DNBR from different perspectives. Therefore, both indicators should be taken into account in future research.

《Fig. 7》

Fig. 7. NRR vs NRE for different substrate groups: (a) NC, (b) NNC, (c) IM, and (d) MM. The red line represents the average value of each substrate group.

3.2.4. Possible nitrate removal mechanisms in various substrates

The possible nitrate removal mechanisms that occur in DNBRs are summarized in Fig. 8; they include HD, AD, DNRA, and anammox. The HD and AD mechanisms are the most commonly observed processes, while DNRA, anammox, and CD are uncommon. More specifically, HD is more likely to occur in NC, NNC, and MMs. These organic compounds release C as the major electron donor for nitrate removal. Then, the HD reaction will occur as follows (where CH2O represents organic C):

The AD mechanism differs from the HD mechanism in that inorganic compounds act as electron donors. These inorganic compounds include pyrite [99] (Eq. (6)), elemental S (Eq. (7)), and zerovalent iron (ZVI) [59] (Eq. (8)).

《Fig. 8》

Fig. 8. Schematic of the mechanisms for various substrates (NC, NNC, MM, IM in a DNBR). These mechanisms include HD, AD, CD, DNRA, and anammox.

The DNRA mechanism has been reported to occur in DNBRs, with reported substrates including woodchips [48], mulch, coniferous matter, willow, compost, leaves, mixtures [23], and woodchips-S [41]. The DNRA reaction is as follows [81]:

This mechanism is competitive with denitrification, especially at higher concentrations of electron donors [82] or limited nitrate [83]. Thus, DNRA is favored when the C dosage is high and is likely to occur in DNBRs with MMs, NC, or NNC due to the extra C dosage.

CD is the reduction of NO3– via a chemical process (e.g., Fe(II) oxidation) [84]:

Based on our review of the literature, CD has not been reported in DNBR studies. This is partially due to the fact that CD requires a catalyst. For example, it was found that nitrate cannot be directly reduced by Fe(II) without a catalyst [85], such as Cu2+, iron oxides, and hydroxides, or microbial surfaces [86,87].

Compared with the above mechanisms, the most economical approach for nitrate removal is anammox [88]. Its benefits include reductions in aeration consumption (60%), organic C requirement (100%), sludge production (90%), and greenhouse gas (i.e., N2O and CO2) emissions [89]. Anammox bacteria have been confirmed to be a type of chemoautotrophic bacteria, which use CO2 as the C source [90]:

Although some studies have reported anammox in natural and artificial environments [89], very little research has been conducted on this topic in connection with DNBRs. To the best of our present understanding, Rambags et al. [49] were the first to report that anammox occurred in a DNBR that received ammonium and nitrate in the inflow to woodchip, coconut husk, and gravel-fill bioreactors. A prerequisite of anammox is the presence of both NO3– and NH4+ . However, in most cases, DNBRs are used to treat agricultural drainage water without NH4+ . Therefore, anammox is not used in bioreactors receiving agricultural runoff but is generally used for industrial wastewater treatment. Overall, further studies are needed to investigate the advantages and disadvantages of the different substrates for efficient nitrate removal in DNBRs.

《3.3. Pollution swapping》

3.3. Pollution swapping

Previous studies of DNBRs have revealed the production of many parameters, including chemical oxygen demand (COD) [63] and many substances, such as ammonium (NH4+ ), phosphate (PO43–) [63], methane (CH4) [91], CO2 [92], and N2O [91]. For most of the NC group, including woodchips, the DOC and soluble phosphorus (P) are of concern, especially during the start-up phase [15,63]. Sharrer et al. [15] reported that approximately 20–30 mg P per kilogram was released during DNBR operation, and that the bulk of this was released within the first 24 h of operation [15]. Moreover, Rivas et al. [63] reported that elevated DOC and dissolved P occurred in the effluent during the start-up phase. Another issue is the incomplete denitrification of NO3– and the production of N2O, which is attributed to HD [93]. Some researchers have reported that adding biochar can reduce N2O losses [30]. Also, other NC sources such as corncobs will lead to increased C and N2O in the effluent, as well as substantial C consumption by non-denitrifiers [44].

NNC materials can be high-risk materials that may cause secondary pollution, thereby posing environmental risks. For example, cardboard was reported to emit the greenhouse gas CH4 (1 g CH4 brings the greenhouse gas effect equal to 296 g∙m–2 ∙d–1 CO2) [24]. In addition, biodegradable polymers have the problem of producing color in the effluent [44].

Among IMs, elemental S serves as an electron donor for the AD process. It is water-insoluble, non-toxic, and stable under normal conditions. The primary shortcoming of S oxidation is the production of sulfate and acid [41]:

The acidity produced in this process may be alleviated by adding limestone, but doing so increases the hardness of the treated water [94]. ZVI can also serve as an electron donor for the AD process. In this case, the undesirable pollutant is ammonium, and the abiotic reaction is illustrated in Eq. (8) [95].

The MM group of substrates may present all of the issues described above. It is concerning that the addition of extra C sources may excessively increase the release of organic C from bioreactors, while the addition of activated C, which is produced by the pyrolysis of organic materials of plant origin, substantially reduces organic C losses while maintaining the denitrification [96].

《3.4. Cost analysis》

3.4. Cost analysis

Our meta-analysis showed that NNC and MMs have greater NRR and NRE than NC (Figs. 4(a) and (b)). However, the cost of NNC limits its application in the real world. For example, biodegradable polymers (PBS, PLA, and PCL) cost from 3 to 10 USD∙kg–1 [64], making them not worth using in agricultural practices. So far, most field DNBR cost analyses have focused on woodchips, with few reports of other substrates. Christianson et al. [52] recently reviewed DNBR design, installation, and removal cost. The average design cost was around 7500 USD, installation costs ranged from 6 940 to 11 820 USD in the Midwestern United States, and the N removal cost efficiency of woodchips was in the range of 2.40– 15.20 USD∙kg–1 . In contrast, the cost efficiencies of other substrates, such as corncobs and wheat straw, were insufficiently reported in previous studies.

Substrate costs contribute to about 13%–55% of the total cost of establishing a DNBR in the field [97]. Therefore, the cost of the substrate is an important factor to be considered in DNBR construction. However, it is difficult to evaluate the cost of substrates, due to the large variation in costs in different locations. In this review, in order to compare the cost efficiency of different substrates, we collected a rough price range from the online market price (USD∙m–3 ). Therefore, we were able to compare the cost of different substrates using the SubEco index (g∙d–1∙USD–1), as shown in Fig. 9, which indicates that woodchips (softwood and hardwood) and corncobs are more cost-effective than other substrates. Biodegradable polymers and pyrite had the lowest value on the index, showing that they are less feasible for adoption by practitioners. This is probably why woodchip-based and corncobbased bioreactors have been commonly studied in the past decades. However, this index may have a large variance due to the range of locations and market fluctuations.

《Fig. 9》

Fig. 9. An economic analysis of the substrates. All prices varied with time and location; the price data used here is according to the online market in the United States in October 2021 (Table S6 in Appendix A).

Thus, in terms of economic evaluation, NC is a cost-effective substrate when comparing the NNC, NC, IM, and MM categories, because NC can almost always be easily obtained from nearby croplands or forest lands. Some researchers have reported that organic electron donors (e.g., methanol and acetic acid) result in a higher cost than inorganic electron donors (e.g., pyrite and sulfite) [98]. This is because methanol and acetic acid, the organic C sources, are commercial chemicals that are more expensive than the NC sources. Inorganic electron donors for nitrate removal, including elemental S, sulfides, ZVI, pyrite, and thiocyanate, have a relatively lower cost [16].

From our perspective, we suggest that MMs are the optimum media because there is a wide choice of auxiliary materials to reduce cost and achieve a circular economy by considering agricultural and industrial wastes. So far, the use of agricultural solid waste in DNBRs has been well documented [25,44,99,100]. However, little is known about the use of industrial waste. Recent studies have indicated that utilizing industrial waste in DNBRs not only contributed to a circular economy but also enhanced the nitrogen removal [34]. In fact, this strategy has been investigated by a few studies, such as those using biochar [35,36,101] or fly ash pellets [33] as auxiliary materials. Our meta-analysis also demonstrated that adding biochar or fly ash pellets into a Woodchips+ can significantly increase the NRR and NRE relative to woodchips alone (Fig. S1(a) and (b)). Furthermore, the use of organic liquid hazardous waste in DNBRs was recently proposed to reduce nitrate and dispose of organic liquid wastes simultaneously, resulting in lower environmental stress and the requirement for fewer nonrenewable inputs than the traditional process [102]. Yet, little progress has been made in using organic liquid waste in DNBRs. Therefore, we suggest that reutilizing industrial and agricultural waste in DNBRs holds potential to be a win–win strategy, and therefore needs further testing to improve the nitrate removal and reduce the environmental risks.

《4. Conclusions》

4. Conclusions

This review categorizes 41 types of bioreactor substrates into four main groups: NC, NNC, IMs, and MMs. Using a metaanalysis, we conducted a comparative assessment of the substrates and found that the measures of nitrate removal (NRE and NRR), in descending order, were as follows: NNC > MM > NC > IM. We showed that nitrate removal was associated with media type, C/N ratio, and the study scale. Importantly, the ranking order of the nitrate removal performance according to the NRR and NRE was not the same for each substrate. Thus, both indicators should be considered in future studies.

For homogeneous substrates, our study indicates that a C/N ratio of below 100 or above 300 may result in a greater NRR. In addition, our cost-effective index (SubEco) showed that woodchips and corncobs are more economical than other NCs or NNCs. Furthermore, among woodchip types, softwood had a greater NRR and lower costs than hardwood. From a holistic perspective, we suggest MMs as the optimum material for nitrate removal due to their low cost and excellent potential for making up deficiencies in the media. When using MMs, adding auxiliary materials to a woodchip-based or corncob-based DNBR may be a promising choice for enhancing denitrification and maintaining long-term operation. Again, the average values given by the meta-analysis should not necessarily be used for design purposes but should rather be used as a relative comparison between different substrates. It is still necessary for future studies to put more effort into coming up with various reasonable substrates on a field scale.

《Acknowledgment》

Acknowledgment

This work was partially supported by the Chinese Government Scholarship, Chinese Scholarship Council (201706350282), which is funding Yuchuan Fan. We thank Wen Liu for her help with data collection.

《Compliance with ethics guidelines》

Compliance with ethics guidelines

Yuchuan Fan, Jie Zhuang, Michael Essington, Sindhu Jagadamma, John Schwartz, and Jaehoon Lee declare that they have no conflict of interest or financial conflicts to disclose.

《Appendix A. Supplementary material》

Appendix A. Supplementary material

Supplementary material to this article can be found online at https://doi.org/10.1016/j.eng.2022.08.017.